Download file - CO2 - CATO

Advances in Model-Based Design of Flexible and Prompt Energy Systems The CO2 Capture Plant at the Buggenum IGCC Power Station as a Test Case

Advances in Model-Based Design of Flexible and Prompt Energy Systems The CO2 Capture Plant at the Buggenum IGCC Power Station as a Test Case

Proefschrift

ter verkrijging van de graad van doctor aan de Technische Universiteit Delft, op gezag van de Rector Magnificus prof. ir. K. C. A. M. Luyben, voorzitter van het College voor Promoties, in het openbaar te verdedigen op maandag 23 juni 2014 om 12.30 uur door

Carsten TRAPP

Diplom-Ingenieur f¨ ur Luft- und Raumfahrttechnik (Universit¨ at Stuttgart) geboren te Radebeul, Duitsland

Dit proefschrift is goedgekeurd door de promotor: Prof. dr. ir. P. Colonna Samenstelling promotiecommissie: Rector Magnificus Prof. dr. ir. P. Colonna Prof. Dr.-Ing. J. Groß Prof. Dr.-Ing. H. Spliethoff Prof. ir. J. Grievink Dr. D. Bhattacharyya Dr. F. Casella Dr. K. Damen Prof. ir. J. P. van Buijtenen

Voorzitter Technische Universiteit Delft, promotor Universit¨ at Stuttgart, Duitsland Technische Universität M¨ unchen, Duitsland Technische Universiteit Delft West Virginia University, Verenigde Staten Politecnico di Milano, Italië Vattenfall Technische Universiteit Delft, reservelid

The work documented in this thesis has been performed within the CO2 Catchup R&D programme aimed at demonstrating and optimising pre-combustion CO2 capture technology for the energy sector. This programme is executed in a consortium of Vattenfall, the Delft University of Technology and the Energy research Centre of the Netherlands. This project has been carried out with subsidy from the Ministry of Economic Affairs, EOS Unieke Kansen Regeling (Projectnumber: UKR05003). This research is also part of CATO-2, the Dutch national programme on CO2 capture, transport and storage. ISBN 978-94-6259-222-3 c 2014 by Carsten Trapp Copyright All rights reserved. No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without the prior permission of the author. An electronic version of this thesis is available at http://www.library.tudelft.nl Published by Carsten Trapp, Delft Printed by Ipskamp Drukkers in the Netherlands

Dedicated to my love Yshya, to my parents, Christine and Detlef, together with my sisters, Michaela and Janina

Table of Contents

1 Introduction 1.1 Challenges in the energy sector . . . . . 1.2 CCS technologies . . . . . . . . . . . . . 1.2.1 CO2 capture . . . . . . . . . . . 1.2.2 CO2 transport . . . . . . . . . . 1.2.3 CO2 storage . . . . . . . . . . . . 1.3 Pre-combustion capture for IGCC plants 1.4 Research motivation . . . . . . . . . . . 1.5 Thesis outline . . . . . . . . . . . . . . . Nomenclature . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

1 2 4 4 9 9 9 11 13 15 16

2 Steady-state modelling and validation of a pre-combustion CO2 capture pilot plant 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Process description . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Model development . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Process models . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Thermodynamic models of the process fluids . . . . . . . . 2.4 Validation methodology . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Validation against design data . . . . . . . . . . . . . . . . 2.4.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Data acquisition and analysis . . . . . . . . . . . . . . . . . 2.4.4 Data reconciliation and parameter estimation . . . . . . . . 2.5 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19 20 22 23 23 25 25 26 26 27 28 32 39 41 42

3 Design optimization for flexible operation of a pre-combustion CO2 capture plant embedding experimental knowledge 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Process description and process models . . . . . . . . . . . . . . . .

45 46 47 49

i

3.3.1 Water-gas shift unit . . . . 3.3.2 H2 S and CO2 removal unit 3.4 Pilot plant experiments . . . . . . 3.5 Result analysis and discussion . . . 3.5.1 Global design decision . . . 3.5.2 Local design decision . . . . 3.6 Conclusions . . . . . . . . . . . . . Nomenclature . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

51 53 55 58 58 64 68 70 71

4 Efficiency improvement in pre-combustion CO2 removal units with a waste-heat recovery ORC power plant 75 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.2 CO2 capture process configuration and waste heat recovery possibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.3 Waste-heat recovery ORC power plants and their configurations . . 80 4.4 Analysis and optimization methodology . . . . . . . . . . . . . . . 81 4.5 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.5.1 The base-case: ORC system composed of standard power modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.5.2 Optimized subcritical ORC power plant . . . . . . . . . . . 88 4.5.3 Optimized supercritical ORC power plant . . . . . . . . . . 94 4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 5 Dynamic modelling and validation of a pre-combustion CO2 capture plant for control design 105 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 5.2 Model development . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 5.2.1 Modelling approach . . . . . . . . . . . . . . . . . . . . . . 107 5.2.2 Thermophysical properties . . . . . . . . . . . . . . . . . . . 108 5.2.3 Development of component models . . . . . . . . . . . . . . 108 5.3 Modelica-FluidProp interface . . . . . . . . . . . . . . . . . . . . . 110 5.3.1 Library architecture . . . . . . . . . . . . . . . . . . . . . . 111 5.3.2 Lessons learned . . . . . . . . . . . . . . . . . . . . . . . . . 113 5.3.2.1 Choice of state variables . . . . . . . . . . . . . . 113 5.3.2.2 Developing index-1 models . . . . . . . . . . . . . 115 5.3.2.3 Improvement of computational time . . . . . . . . 117 5.3.3 Recommendations for a future interface . . . . . . . . . . . 120 5.4 Dynamic model validation . . . . . . . . . . . . . . . . . . . . . . . 120 5.4.1 Validation approach, experiments and results . . . . . . . . 120 5.5 Process analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 ii

Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 6 Dynamic system model of the absorption section of pre-combustion CO2 capture plants 141 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 6.2 Model development . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 6.2.1 Process description . . . . . . . . . . . . . . . . . . . . . . . 143 6.2.2 Modelling approach . . . . . . . . . . . . . . . . . . . . . . 143 6.2.3 Dynamic absorber model . . . . . . . . . . . . . . . . . . . 144 6.2.4 Additional process models and control . . . . . . . . . . . . 148 6.3 Dynamic validation . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 6.3.1 Absorber model validation . . . . . . . . . . . . . . . . . . 149 6.3.2 Absorption and solvent regeneration section model validation 158 6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 7 Conclusions and Perspectives 7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

167 168 172 174

Summary

175

Samenvatting

179

Acknowledgements

183

Selected publications

187

About the author

189

iii

“Niemand hat die Absicht, eine Mauer zu errichten.” “No one has any intention of building a wall.”

Walter Ulbricht, Chairman of the State Council of the German Democratic Republic, press conference, East-Berlin, June 15, 1961

1

Introduction The work documented in this thesis investigates the design of pre-combustion CO2 capture systems for integrated gasification combined cycle (IGCC) power plants. The motivation of this research stems from the need for design competence and validated design tools enabling future large-scale implementation of this technology in the power sector. First, recent developments in the energy sector are described concerning the need for reduction in emission intensity and the demand for more flexible and prompt operation of fossil-fuelled power plants. Then a general introduction to carbon capture and storage (CCS) technologies is given. The different capture processes and their technical challenges are addressed, with an emphasis on the key challenges in the design of pre-combustion CO2 capture plants. The research motivation is discussed in detail and scientific questions addressed in this thesis are formulated. The chapter concludes with a thesis outline.

Chapter 1

1.1

Challenges in the energy sector

The rapid growth of the world population and the emergence of developing economies might lead to the continuation or even to the increase of the use of fossil fuels in the coming decades in order to meet the growing world energy demand [1]. In particular, coal remains an attractive source of energy because it is still cheap, abundant, and easily accessible. However, coal is also the most carbon-intensive fossil fuel in comparison to other primary energy sources, and thus its consumption contributes to a large extent to environmental pollution and the increasing level of carbon dioxide in the atmosphere. It is expected that constraints on carbon emissions will be more strict in future, in order to mitigate the effects of global warming attributed to the high atmospheric concentration of greenhouse gases. The generation of power and heat, with its major use of fossil fuels, was identified as the largest producer of CO2 emissions, being responsible for a share of more than 40 % of the global CO2 emissions in 2010 [2]. Reduction of the emission intensity in the energy sector can be achieved by implementation of mitigation measures, such as the improvement of power plant and end-use technology efficiency, the increase of the share of non-emitting sources, i.e., renewables and possibly nuclear fuel, the use of more biofuels and carbon capture and storage. Not to mention, that the most crucial measure for CO2 abatement is electricity saving. The International Energy Agency (IEA) estimated that CCS technologies applied to fossil-fuelled power plants is a potentially beneficial technology, because it might contribute to the required reduction of CO2 emissions for as much as 17 % by year 2035 [3]. This is based on the so called 450 scenario, whereby the longterm temperature increase is limited to 2 ◦ C in comparison to the pre-industrial level. CCS technologies might therefore play an important role in a foreseeable near-future energy scenario, in which a carbon-constrained transition period leads to electricity primarily supplied by renewable energy resources. An increasing number of countries are adopting climate targets in order to promote the transition to a more sustainable electricity generation, resulting in significant changes in the energy sector. In 2009 the European Union (EU) set the objective that 20 % of its final energy consumption1 is provided by renewable sources by the year 2020 [4]. Stimulated by this so-called Renewables Directive, wind and solar energy technologies have exhibited the largest growth rates among all the technologies for the conversion of renewable energy sources. In 2013 they accounted for almost 90 % of newly installed power capacity, as far as conversion of renewable energy is concerned. The share of wind power with respect to the total installed power capacity increased from 2.4 % in 2000 to 13 % in 2013, resulting in a gross power capacity of 117.3 GW. The share of installed photovoltaic (PV) power capacity was marginal in 2000, and increased to 9 % in 2013, which corresponds to a power capacity of 80 GW. As a result, wind power plants can cover about 1 The

final energy consumptions refers to the total energy consumed by the end user, such as private households, agriculture, industry and transport.

2

Introduction

8 % of the EU electricity demand, while PV power plants about 3 % [5, 6]. It is expected that the growth of the wind and solar power installations will continue, though the growth rates might be more modest than in the last few years due to cutbacks in governmental subsidies for renewable energy. The major drawback of renewable energy sources, such as wind and solar, is their naturally fluctuating availability, which can vary considerably, and even at different time-scales: hourly, daily or seasonally. This results in intermittent electricity generation. Due to the rapidly increasing share of the electricity converted from the wind or solar energy, balancing energy demand and supply becomes more challenging [7, 8]. One of the prospective technical measures against gridunbalance is the use of so-called smart grid technology. A smart grid is a network that integrates the actions of generators and consumers by means of bi-directional communication in order to improve the efficiency and reliability of the electricity supply. For example, consumers can optimize their energy use as they receive accurate information on electricity prices, and, as a consequence, the number of power plants for peak demand can be reduced. Another solution in order to match generated and consumed power is the implementation of electricity storage like, for example, batteries, pumped-storage plants, or production of fuels. Currently, energy storage capacity is insufficient in Europa, and smart grids are at an early stage of development. However, the increasingly intermittent characteristic of electricity generation needs to be matched to the inherently variable demand for energy. Thus conventional power plants must be made capable of sustaining a much higher level of flexible operation. It is also worth noting that wind and solar power plants are prioritized in the power generation scheduling of all European countries, and therefore the number of hours in which conventional power plants are operated at full load is decreasing rapidly, while their part-load operation is increasing. The reduction in the utilization rate of conventional power plants is currently also amplified by the excess of installed capacity, due to the lower energy consumption at times of economic crisis. In order for fossil-fuelled power plants to remain competitive in this rapidly transforming electricity market, improvements regarding plant flexibility are essential. These are in general the reduction of the minimum load limit and the enhancement of the dynamic performance of power plants, in particular the so-called ramp rate [9]. The rapid growth of renewable energies but also the liberalization of the European electricity market have thus significantly changed the power sector. Until recently, electricity supply relied on large base-load power plants, such as nuclear or fossil-fuel plants, and on dynamically operated gas turbines in order to cover peak demand. Nowadays however, the evolution of the electricity market demands for more prompt and flexible operation also of fossil-fuel power plants. As it is expected that constraints on carbon emissions will require fossil-fuel power plants to be equipped with CO2 capture units, consequently these gas-processing plants have to be able to follow the very dynamic operation of power plants, and at the same time meet environmental targets in terms of CO2 removal. 3

Chapter 1

1.2

CCS technologies

CCS entails the capture of CO2 emissions from large, stationary point sources, such as fossil-fuelled power plants, but also from industrial sites, like refineries, cement production and steel making plants, and the transport of the concentrated and liquefied CO2 to its permanent storage location in deep geological formations, such as depleted oil and gas fields, saline aquifers or unmineable coal seams. Figure 1.1 depicts the entire CCS process chain for fossil-fuelled power plants. In the following, the different technologies to perform carbon capture, transport and storage are introduced. In addition, the technical challenges are discussed, with primary focus on the power sector due to its role as the major contributor to the global emissions of CO2 .

1.2.1

CO2 capture

Currently, three general processes for the removal of CO2 applied to fossil-fuelled power plants are investigated [11]: • Removal of CO2 from synthetic gas prior its combustion — pre-combustion capture, • Separation of CO2 from flue gas after combustion — post-combustion capture, • Fossil fuel combustion in nearly pure oxygen to yield pure CO2 as combustion product — oxyfuel combustion. Figure 1.2 gives an overview of the different capture systems, which are explained in more detail below. In general, different separation technologies can be adopted for these capture processes, namely absorption, adsorption, membranes, and cryogenics. The choice of the most suitable technology depends among others on the condition of the gas stream to be treated (such as temperature, pressure, and concentration of CO2 ) and the desired purity level of the CO2 product stream [13]. Some of these technologies are already applied commercially for separation of CO2 in the chemical industry. Examples are: process gas treatment (chemical absorption), natural gas sweetening (physical absorption or polymer membranes), and hydrogen purification (chemical absorption, adsorption) [14]. However, the scale, process conditions and requirements that would apply to CO2 removal in power plants are largely different from CO2 capture technologies currently adopted (higher flow rates, possibly low pressure, presence of different impurities). As a consequence available technologies are primarily not cost-effective [13]. Physical and chemical absorption are presently the most mature and viable technologies for the removal of CO2 from the syngas or flue gas of fossil-fuelled power plants, while other technologies require further development before commercial-scale implementation can occur. 4

Introduction

Figure 1.1: Graphical representation of the entire CCS process chain applied to fossilfuelled power plants [10].

Figure 1.2: Overview of the different capture processes [12].

5

Chapter 1

In general, CO2 capture is the most energy intensive step of the entire CSS process, resulting in a power plant efficiency loss in the range of 6.4 − 11 %-points depending on the employed technology and the type of power plant [15]. Moreover, CO2 separation accounts for about 70 % of the costs related to the implementation of CCS [16]. Pre-combustion capture Pre-combustion CO2 capture is suitable for integrated gasification power plants, i.e., combined cycle power stations whereby the gaseous fuel for the gas turbine is obtained by gasification of fossil fuels or biomass at elevated pressure to yield a synthetic gas (syngas) predominantly containing CO and H2 . In case of the addition of a pre-combustion CO2 capture process plant, steam is added to the syngas to produce CO2 and H2 in catalytic shift reactors. Finally, CO2 is separated from the syngas typically by means of physical absorption, which is most effective at high partial pressure. Thereafter, the CO2 recovered of the loaded solvent is further compressed for sequestration. The resulting H2 -rich gas is fed to the gas turbine of the combined cycle power plant. Figure 1.3 illustrates a simplified IGCC power plant with a pre-combustion CO2 capture unit. A more detailed discussion of the entire process is given in Section 1.3. The main processes of pre-combustion capture, such as the water-gas shift and the CO2 absorption, have already been used in the chemical industry for a long time, hence these technologies are well proven. However, the application of this technology to IGCC power plants implies a much larger scale, and continuous and dynamic operation, which introduces remarkable differences and challenges if compared to the CO2 capture process plants of the chemical industry. Disadvantages of pre-combustion capture are related to the complexity of the fuel treatment process and issues related to the burning of hydrogen-rich syngas in the gas turbine [11]. Post-combustion capture Figure 1.4 depicts a simplified scheme of a power plant with post-combustion capture. Chemical absorption with aqueous amine solutions is the most mature CO2 separation technology for post-combustion capture commercially deployed for process gas treatment. CO2 is absorbed from the flue gas by the amine solvent, which is thereafter regenerated at elevated temperature (about 120 ◦ C), and continuously recycled. The distinct advantage of the post-combustion capture process is that it can be implemented as retrofit to existing power plants without significant modifications to the power generation system. The disadvantages of this technology are i) the high demand of energy required by the solvent regeneration, and ii) the relatively large size of the equipment due to the low partial pressure of CO2 . Challenges are related to operation under varying plant conditions, and to the scale-up of the technology, which is required in order to treat the entire flue gases emitted 6

Introduction

Figure 1.3: Simplified scheme of an IGCC power plant with integrated pre-combustion capture unit [17].

Figure 1.4: Simplified scheme of a power plant with post-combustion capture unit [17].

7

Chapter 1

from a power plant, so that approximately 90 % capture rate can be achieved. Further improvements of post-combustion capture performance can be obtained by a higher degree of process integration with the power plant [18].

Oxyfuel combustion Figure 1.5 visualizes a simplified configuration of a power plant based on oxyfuel combustion technology. Oxyfuel combustion is a process whereby the fossil fuel is combusted by means of nearly pure oxygen, yielding exhaust gases primarily containing CO2 and water vapour, which can easily be separated to produce CO2 with high purity. One of the technical difficulties inherent in the oxyfuel combustion process is the flue gas cleaning due to the higher concentration of impurities, such as SO2 and NOx , in comparison to those in the flue gas of an air-coal combustion plant. This makes this technology less suitable for combustion of low quality fuels [19, 20]. The oxygen required for combustion is produced by means of air separation based on cryogenic distillation. This process can be considered as a mature technology, however it is very energy intensive. A part of the flue gases is recycled in order to lower the high flame temperatures related to combustion with pure oxygen. Current research is directed, for example, at the investigation of the oxyfuel combustion processes, boiler design and optimization of air separation in order to reduce the energy consumption [11, 21].

Figure 1.5: Simplified scheme of a power plant based on oxyfuel combustion [17].

8

Introduction

1.2.2

CO2 transport

After removal from the fuel or flue gas, the CO2 is liquefied by compression to pressures between 110 and 150 bar, and transported to storage sites. Arguably, if CCS will become mainstream, the only way of transporting and distributing large quantities of CO2 will be by means of pipelines. Possible transport options for smaller quantities and short distances are by truck, rail or ship. CO2 delivery by various means already occurs since few decades. Examples can be found in the food industry, and in the oil extraction sector, whereby compressed CO2 is utilized for enhanced oil recovery. The required technology and operational experience is therefore available. However, CCS deployment would require an extremely large new network, including hubs in order to redistribute CO2 collected from various power plants to individual storage sites. In this context, more specific health and safety regulations need to be considered.

1.2.3

CO2 storage

Storage involves the injection of pressurized CO2 into geological formations deep underground, where the CO2 is retained in the pore space of sedimentary rocks. Suitable storage sites are oil and gas reservoirs, unmineable coal seams and saline formations (aquifers). Aquifers are estimated to provided the largest storage volumes worldwide. The storage of CO2 in oil and gas reservoirs is utilized since various decades for enhanced oil recovery. Current activities concern the exploration of suitable storage sites, the demonstration of CO2 storage in aquifers in order to determine actual available capacity, and the long-term monitoring of injected CO2 in order to ensure that the gas cannot escape from the reservoir. To summarize, most of the individual component technologies for capture, transport and storage are already proven in industry, sometimes in different configuration and/or with other purposes. However, the largest challenge for CCS deployment is the integration of these individual technologies and its implementation at large-scale in the power sector [22]. This entails continuous and sustained technology development in order to reduce the power plant efficiency loss due to CO2 removal. Improvements in overall conversion efficiency can be obtained by reducing the energy consumption of the CO2 capture process, by making transport and storage less energy demanding, and by better integrating all CO2 capturerelated processes into the power conversion system.

1.3

Pre-combustion capture for IGCC plants

The integrated gasification combined cycle is a concept for complex energy conversion systems, which combines solid fuel gasification technology with a highly efficient power generation system, the combined cycle power plant. The combined cycle configuration comprises a gas turbine (GT) and a heat recovery steam gener9

Chapter 1

Air

Air Separation Unit (ASU)

Raw fuel gas Gasification

Coal

Coal Preparation

Steam (from gas coolers)

Steam Turbine

Syngas Scrubber/ COS Hydrolysis

Steam

Steam

Stack Gas To ASU

HRSG

Electricity Generation

Sulphur Removal

Sulphur Recovery

Water-gas shift

Sulphur (By-product)

CO2 Removal

CO2 Compression

Hydrogen

Gas Turbine

CO2 to Storage

N2 from ASU Air

Figure 1.6: Process flow diagram of an IGCC power station integrating a CO2 removal plant [23].

ator (HRSG) powering a steam turbine (ST). Figure 1.6 shows a simplified scheme of an IGCC power plant with integrated pre-combustion CO2 capture unit. The fuel, in most cases coal, first undergoes a preparation process depending on the employed gasification technology (e.g., milling and drying in case of dryfeed gasifiers), before it is fed together with oxygen, produced by an air separation unit, to the gasifier. Under conditions of high temperature (1400 − 1600 ◦ C) and pressure (30 − 40 bar) a synthetic gas is produced predominately containing H2 and CO. Thermal energy is recovered from the syngas leaving the gasifier by using it in order to generate high-pressure steam which is fed to the ST. Thereafter, the syngas is cleaned by means of cyclones, filters and water scrubbing in order to remove the remaining fly ashes and HCl. The carbonyl sulphide (COS) present in the syngas is converted into H2 S during the COS hydrolysis. The sulphur is thereafter removed in the H2 S removal unit, and sent to the Claus plant whose final product is elementary sulphur. After all these treatments, the syngas enters the pre-combustion CO2 capture unit. As described in Section 1.2, first the CO present in the syngas is converted into CO2 and H2 by means of a staged water-gas shift process, which requires adding steam to the syngas. Thereafter, the CO2 is removed by means of absorption and compressed for sequestration. The resulting H2 -rich syngas is fed to the gas turbine of the combined cycle in order to produce power. The thermal energy of the gas turbine exhaust is recovered in the HRSG, and the obtained high-pressure steam is thereafter expanded in the ST. Many different plant configurations are possible, depending on the type of gasification technology, the degree 10

Introduction

of process integration and the choice of capture technology. The process description given here concerns therefore only a general IGCC plant configuration with CO2 capture, but many variants are possible, if more details are considered. To conclude, integrated gasification combined cycle power plants are a promising technical solution if electricity production must integrate carbon capture, because CO2 can effectively be removed at high partial pressures, and the plant net energy efficiency is estimated to be higher than that of conventional pulverized coal (PC) steam power plants [15]. The U.S. National Energy Technology Laboratory (NETL) predicted that IGCC power plants with 90 % CO2 capture can reach net plant efficiencies between 31.2 and 32.6 %, depending on the employed gasification technology. The efficiencies are based on the higher heating value (HHV). Notably, the estimated efficiency of supercritical PC steam power plants with CO2 capture is 28.4 %. Moreover, gasification allows for i) lower emission levels of regulated pollutants resulting from effective syngas cleaning, ii) greater fuel flexibility (the fuel can be any kind of coal, and biomass, even in combination with coal), and iii) the integrated generation of different products, such as electricity, fuels and chemicals. However, a number of challenges must be overcome in order bring this technology to commercial scale and spread its adoption. These are the high energy penalty associated with CO2 capture, compression, transport and storage, the increase in system complexity, the process availability, and the high capital investment. Moreover, the integration of the capture unit into the very complex gasification process and combined cycle power plant leads to outstanding technical problems as far as dynamic operation is concerned. As outlined in Section 1.1, transient performance of power plants is becoming extremely relevant, due to recent developments in the electricity market, namely the liberalization (in the European countries), and the increase of the share of electricity obtained from renewable energy sources. As a consequence, the IGCC power plant and the integrated capture unit have to be able to follow frequent and fast load changes in order to balance the intermittent nature of the conversion of wind and/or solar radiation. Apart from the technological problems, the competitiveness and economic viability of CCS technologies in the power sector will depend on future policies and regulations. Recent activities regarding exploration of storage sites demonstrated that public acceptability plays also a key role.

1.4

Research motivation

The technical obstacles related to large-scale implementation of pre-combustion CO2 capture plants are addressed in an increasing amount of scientific literature which documents performance analysis by means of steady-state modelling and simulation [23–28]. These studies compare different technologies and evaluate the impact of several operating parameters on the energy efficiency penalty due to the integration of a CO2 capture plant into a power station. However, sophisticated 11

Chapter 1

process optimization maximizing efficiency or power output by means of modelbased design is addressed in only a few studies [29, 30] and demands, in particular for the CO2 capture process, further research. The challenges related to transient operation and control of CO2 capture systems integrated with IGCC power plants have been treated so far only by few researchers, due to the complexity of the required modelling and simulation work [31–33]. For the majority of the documented models used for process analysis, model validation could not be performed due to lack of experimental or industrial data. Data for validation were lacking in case of both steady-state and dynamic operation. Validation of the models against measurements is essential to improve the accuracy of the simulation results leading to reliable and predictive design tools. Therefore, comprehensive experimental investigations accompanied by modelling activities to develop detailed and accurate steady-state as well as dynamic models for process design were required. The work presented here was part of a larger research project involving the utility company Vattenfall, the Energy research Centre of the Netherlands (ECN) and the Delft University of Technology aimed at the development of pre-combustion CO2 capture technology to be applied in a future commercial-scale IGCC power plant. A unique, fully instrumented CO2 capture pilot plant was realized at the Buggenum IGCC power station in the Netherlands in order to demonstrate the technology and investigate its performance [34]. The general objective of the work documented in this thesis is to generate knowledge on pre-combustion CO2 capture systems for IGCC power plants utilizing the pilot facility in order to provide design competence and validated design tools for future large-scale implementation of this technology in the power sector. This overall goal translates into original research questions which aim to identify important design variables of the IGCC pre-combustion CO2 capture system, such as the most relevant process parameters, as well as environmental and operational limits and/or targets, and their impact on the CO2 removal efficiency penalty, and on the optimal operating conditions. This investigation is approached by means of model-based process simulation and design optimization targeting reduction in energy consumption, and through design and execution of experiments to evaluate the impact of different parameters on the process performance throughout the operating window of the pilot plant. Furthermore, this research project targeted the investigation of the capabilities of a capture system to follow prompt load variations, and the study of control strategies that enhance the responsiveness of the plant. The most relevant research objectives are to improve and develop tools and methodologies which i) facilitate detailed steady-state performance analysis and sophisticated optimization of process design and operating conditions, and ii) enable studies on process dynamics already during the early design phase in order to support the choice of equipment and control strategies aiming at the improvement of transient performance. The tools and methods are developed for the case-specific analysis of the pre12

Introduction

combustion CO2 capture plant at the Buggenum IGCC power station. However, the aim of this work is to generalize as much as possible aspects of the adopted novel system engineering techniques and tools, which would then be applicable to the design of a larger class of chemical and energy conversion systems. The title of this thesis has therefore been chosen so as to represent this more general objective, whereby the application to the specific case of the pre-combustion capture plant is highlighted by the subtitle.

1.5

Thesis outline

This thesis covers two aspects of the design of prompt and flexible energy systems, demonstrated by analysis of a pre-combustion CO2 capture plant. The studies related to steady-state modelling, simulation and optimization are discussed in Chapter 2 to 4, and the investigations targeting dynamic performance are summarized in Chapter 5 and 6. Chapter 2 documents the steady-state modelling and simulation of the precombustion CO2 capture pilot plant built at the Buggenum IGCC power station comprising a water-gas shift and an absorption and solvent regeneration process. Comprehensive model validation is demonstrated for the water-gas shift unit utilizing 20 experimental data sets recorded at the pilot facility by applying a procedure of simultaneous data reconciliation and parameter estimation including gross error detection based on the contaminated Normal estimator. The chapter concludes with an analysis of the reconciled measurements to evaluate the accuracy of the developed model and to identify biased measurements. Chapter 3 presents the design optimization of a large-scale pre-combustion CO2 capture plant following a two-phase approach suited to the use of process simulator environments. In the first phase, global design decisions at plant level are evaluated, targeting the minimization of the energy consumption due to CO2 capture. These are the extent of CO conversion in the water-gas shift unit and the percentage of CO2 capture in the removal unit. An optimization of both global design variables is presented considering i) flexible operation in terms of overall carbon capture target, ii) deactivation of catalyst activity throughout the catalyst life, and iii) different operational limits of the steam/CO ratio in the water-gas shift unit. In the second design optimization phase, local design decisions at unit level are evaluated. Two studies are presented focusing on: 1) the design of the solvent regeneration and CO2 compression section, and 2) the impact of the solvent temperature on energy efficiency penalty and equipment cost of the removal unit. Chapter 4 illustrates the recovery of low-grade thermal energy from the precombustion CO2 capture process by means of organic Rankine cycle (ORC) turbo13

Chapter 1

generators. Differently from other conventional ORC power system applications, the thermal energy source in this case is a syngas-water mixture, which is cooled from a temperature of approximately 140 ◦C, and partly condenses due to the heat transfer to the ORC primary heat exchanger. The performance of the three categories of systems, depending on working fluid and cycle configuration, i.e., systems based on (i) commercially available units, (ii) tailor-designed subcritical cycle, (iii) tailor-designed supercritical cycle, is analysed in terms of net power output, second law efficiency and component-based exergy efficiencies. In this study, particular attention is focused on the semi-empirical optimization approach, in order to avoid unnecessary computations, and general guidelines are provided. Chapter 5 discusses the development and implementation of dynamic models of the pre-combustion CO2 capture process into an open source software library by means of the object-oriented, equation-based Modelica language. Moreover, the content includes the description of the development of an interface prototype in order to enable the computation of fluid properties with accurate thermodynamic models available within external property packages. Comprehensive dynamic model validation is demonstrated at component, sub-system and system level by comparison against experimental measurements obtained from various open- and closed loop transient tests at the CO2 capture pilot plant. Finally, a simulation-based control design study is presented, whereby a control strategy involving feed-forward, feed-back and cascade control has been implemented and tested with the aim of improving the dynamic performance of the capture unit. Chapter 6 provides a detailed treatment of the development of the dynamic model of the absorption and solvent regeneration unit as part of the pre-combustion CO2 capture system, following the equilibrium-based approach for modelling of the physical absorption process. The accuracy of the model predictions is evaluated by comparison against measurements obtained during two transient tests monitoring the system response to step changes in syngas and solvent mass flow rate. Chapter 7 concludes this thesis by summarizing the main results and discussing possible evolutions of this work.

14

Introduction

Nomenclature Acronyms CCS ECN EU GT HHV HRSG IEA IGCC NETL ORC PC PV ST

= = = = = = = = = = = = =

Carbon capture and storage Energy research Centre of the Netherlands European Union Gas turbine Higher heating value Heat recovery steam generator International Energy Agency Integrated gasification combined cycle U.S. National Energy Technology Laboratory Organic Rankine cycle Pulverized coal Photovoltaic Steam turbine

15

Chapter 1

References [1] Massachusetts Institute of Technology, 2007. The future of coal. Tech. rep. [2] International Energy Agency (IEA), 2012. CO2 emission from fuel combustion. Tech. rep. [3] International Energy Agency (IEA), 2012. World energy outlook 2012. Tech. rep. [4] The European Parliament and the Council of the European Union, 2009. “Directive 2009/28/ec on the promotion of the use of energy from renewable sources”. Official Journal of the European Union. [5] European Wind Energy Association (EWEA), 2014. Wind in power - 2013 European statistics. Tech. rep. [6] European Photovoltaic Industry Association (EPIA), 2014. Market Report 2013. Tech. rep. [7] Ziems, C., Meinke, S., Nocke, J., Weber, H., and Hassel, E., 2012. Auswirkungen von fluktuierender Windenergieeinspeisung auf das regel- und thermodynamische Betriebsverhalten konventioneller Kraftwerke in Deutschland, Teil II - Auswirkungen großer Windeinspeisungen auf den zuk¨ unftigen Kraftwerkspark und dessen Tageseins. Tech. rep., VGB PowerTech e.V. Projektnummer 333. [8] Saint-Drenan, Y.-M., von Oehsen, A., Gerhardt, N., Sterner, M., Bofinger, S., and Rohrig, K., 2009. Dynamische Simulation der Stromversorgung in Deutschland nach dem Ausbauszenario der Erneuerbaren-Energien-Branche. Tech. rep., Fraunhofer Institut fr Windenergie und Energiesystemtechnik (IWES). [9] Meinke, S., 2012. “Modellierung thermischer Kraftwerke vor dem Hintergrund steigender Dynamikanforderungen aufgrund zunehmender Windenergie- und Photovoltaikeinspeisung”. PhD thesis, Universit¨ at Rostock. [10] Scottish Centre for Carbon Storage, 2014. CCS process. www.geos.ed.ac.uk/sccs, April 6. [11] Rackley, S. A., 2010. Heinemann/Elsevier.

Carbon Capture and Storage.

Butterworth-

[12] Intergovernmental Panel on Climate Change (IPCC), 2005. Special Report on Carbon Dioxide Capture and Storage. Tech. rep. [13] Zaman, M., and Lee, J., 2013. “Carbon capture from stationary power generation sources: A review of the current status of the technologies”. Korean Journal of Chemical Engineering, 30(8), pp. 1497–1526. [14] Wong, S., and Bioletti, R., 2002. Carbon Dioxide Separation Technologies. Tech. rep., Alberta Research Council. [15] National Energy Technology Laboratory (NETL), September 2013. Cost and Performance Baseline for Fossil Energy Plants Volume 1: Bituminous Coal and Natural Gas to Electricity. Tech. rep. DOE/NETL-2010/1397, Revision 2a.

16

Introduction

[16] Haszeldine, R. S., 2009. “Carbon capture and storage: how green can black be?”. Science, 325(5948), pp. 1647–1652. [17] Vattenfall, 2014. www.vattenfall.com, April 6. [18] Cifre, P., Brechtel, K., Hoch, S., Garc´ıa, H., Asprion, N., Hasse, H., and Scheffknecht, G., 2009. “Integration of a chemical process model in a power plant modelling tool for the simulation of an amine based CO2 scrubber”. Fuel, 88(12), pp. 2481–2488. [19] Li, H., Yan, J., Yan, J., and Anheden, M., 2009. “Impurity impacts on the purification process in oxy-fuel combustion based CO2 capture and storage system”. Applied Energy, 86(2), pp. 202–213. [20] Liu, H., and Shao, Y., 2010. “Predictions of the impurities in the CO2 stream of an oxy-coal combustion plant”. Applied Energy, 87(10), pp. 3162–3170. [21] Toftegaard, M., Brix, J., Jensen, P., Glarborg, P., and Jensen, A., 2010. “Oxyfuel combustion of solid fuels”. Progress in Energy and Combustion Science, 36(5), pp. 581–625. [22] International Energy Agency (IEA), 2009. Technological Roadmap Carbon Capture and Storage. Tech. rep. [23] Huang, Y., Rezvani, S., McIlveen-Wright, D., Minchener, A., and Hewitt, N., 2008. “Techno-economic study of CO2 capture and storage in coal fired oxygen fed entrained flow IGCC power plants”. Fuel Processing Technology, 89(9), pp. 916–925. [24] Descamps, C., Bouallou, C., and Kanniche, M., 2008. “Efficiency of an Integrated Gasification Combined Cycle (IGCC) power plant including CO2 removal”. Energy, 33(6), pp. 874–881. [25] Kunze, C., and Spliethoff, H., 2010. “Modelling of an IGCC plant with carbon capture for 2020”. Fuel Processing Technology, 91(8), pp. 934–941. [26] Kanniche, M., and Bouallou, C., 2007. “CO2 capture study in advanced integrated gasification combined cycle”. Applied Thermal Engineering, 27(16 SPEC. ISS.), pp. 2693–2702. [27] Martelli, E., Kreutz, T., and Consonni, S., 2009. “Comparison of coal IGCC with and without CO2 capture and storage: Shell gasification with standard vs. partial water quench”. Energy Procedia, 1(1), pp. 607–614. [28] Gr¨ abner, M., Morstein, O., Rappold, D., G¨ unster, W., Beysel, G., and Meyer, B., 2010. “Constructability study on a german reference IGCC power plant with and without CO2 -capture for hard coal and lignite”. Energy Conversion and Management, 51(11), pp. 2179–2187. [29] Bhattacharyya, D., Turton, R., and Zitney, S., 2011. “Steady-state simulation and optimization of an integrated gasification combined cycle power plant with CO2 capture”. Industrial and Engineering Chemistry Research, 50(3), pp. 1674–1690.

17

Chapter 1

[30] Martelli, E., Kreutz, T., Carbo, M., Consonni, S., and Jansen, D., 2011. “Shell coal IGCCS with carbon capture: Conventional gas quench vs. innovative configurations”. Applied Energy, 88(11), pp. 3978–3989. [31] Robinson, P., and Luyben, W., 2010. “Integrated gasification combined cycle dynamic model: H2 S absorption/stripping, water-gas shift reactors, and CO2 absorption/stripping”. Industrial and Engineering Chemistry Research, 49(10), pp. 4766– 4781. [32] Bhattacharyya, D., Turton, R., and Zitney, S., 2012. “Control system design for maintaining CO2 capture in IGCC power plants while load-following”. In Proceedings of the 29th Annual International Pittsburgh Coal Conference, Pittsburgh, PA, October 15-18, Vol. 3, pp. 2160–2173. [33] Zitney, S., Liese, E., Mahapatra, P., Turton, R., Bhattacharyya, D., and Provost, G. “AVESTAR Center: Dynamic simulation-based collaboration toward achieving operational excellence for IGCC plants with carbon capture”. In Proceedings of the 29th Annual International Pittsburgh Coal Conference 2012, Pittsburgh, United States, 15-18 October 2012, Vol. 3, pp. 2093–2147. [34] Damen, K., Gnutek, R., Kaptein, J., Nannan, N. R., Oyarzun, B., Trapp, C., Colonna, P., van Dijk, E., Gross, J., and Bardow, A., 2011. “Developments in the pre-combustion CO2 capture pilot plant at the Buggenum IGCC”. Energy Procedia, 4(0), pp. 1214–1221.

18

“All free men, wherever they may live, are citizens of Berlin, and, therefore, as a free man, I take pride in the words ‘Ich bin ein Berliner!’ ” John F. Kennedy, President of the United States, public speech, West-Berlin, Juni 26, 1963

2

Steady-state modelling and validation of a pre-combustion CO2 capture pilot plant Model validation plays an important role during the development of reliable process models in order to demonstrate that the obtained model can predict the process performance with sufficient accuracy with respect to the modelling purpose. Comprehensive model validation requires process measurements from industrial or laboratory facilities. These measurements are subject to random and gross errors which must be eliminated from the validating data in order to successfully perform model validation. This chapter documents the steadystate modelling and simulation of the pre-combustion CO2 capture pilot plant built at the Buggenum integrated gasification combined cycle (IGCC) power station, which comprises a water-gas shift section and an absorption and solvent regeneration section. Model validation is demonstrated for the water-gas shift (WGS) section utilizing 20 experimental data sets which were recorded at the pilot facility. A procedure of simultaneous data reconciliation and parameter estimation including gross error detection was applied using the contaminated Normal estimator. It can be concluded that the steady-state model of the water-gas shift section predicts the process performance throughout the entire operating range with sufficient accuracy. The model therefore serves as a reliable foundation for the development of commercial-scale models of precombustion CO2 capture plants, which are essential for process analysis and design optimization.

Chapter 2

2.1

Introduction

Steady-state process models of energy conversion and chemical systems are essential tools during the early process design phase up to commissioning and plant operation. They can be used for purposes of performance verification or analysis, and for process optimization considering design as well as optimal operating conditions. Quantitative validation of the process models, not only against design data but in particular against experimental measurements, plays an important role during the model development process in order to obtain reliable and accurate performance predictions. The aim of the work documented here is to promote the development of precombustion CO2 removal technology for future large-scale IGCC power plants [1] by means of modelling and simulation of a small-scale CO2 capture plant. This pilot plant was realized at the Buggenum IGCC power station in the Netherlands by the utility company Vattenfall for technology demonstration and experimental studies. Measurements covering on- and off-design operation were used for the validation of the pilot plant models, using data reconciliation and parameter estimation as discussed in this chapter. Ultimately, the validated models are used as basis to develop models of the full-scale CO2 capture system in order to support process design. In the following, the choice of the validation methodology is motivated by considering some fundamental aspects of the use of experimental measurements. Experimental data obtained from industrial processes or laboratory analyses are subject to different errors and possibly process variability. Measurements therefore will not satisfy the conservation laws and constraints which are used to mathematically describe the process. The errors can be classified as random errors – randomly distributed with expected average value of zero – caused by the inaccuracy of the measurement device, and gross errors – considered to be nonrandom – which can occur due to instrument malfunction, miscalibration or poor sampling among others. Gross errors can be further distinguished between outliers and biases, whereby the first describe abnormal measurement values caused, for instance, by malfunction, and the second define measurement values which are systematically higher or lower than the true process value [2]. Data reconciliation is the process of adjusting the measurements by minimization of the residual between the corrected and recorded process value in order to satisfy material and energy balance constraints. Reconciliation can only be performed if redundant measurements are available. The result of the reconciliation procedure is quality process data which can be used for performance analysis or model validation. In the presence of gross errors, which do not follow the statistical distribution of the sampled data, the procedure of data reconciliation might lead to significantly biased estimates. Any performance analysis based on poorly reconciled data will provide non-reliable results. Gross errors therefore need to be identified during the reconciliation process, and either replaced by corrected measurements or the 20

Steady-state modelling and validation of a pre-combustion CO2 capture pilot plant

measurement must be discarded from the data set. Various approaches to gross error detection and elimination have been proposed, ranging from statistical measurement tests [3] performed in a sequential manner to simultaneous gross error detection and data reconciliation methods. Good reviews on the gradual advancement of data reconciliation procedures are authored by Crowe [4], Romagnoli and Sanchez [5] and Narasimhan and Jordache [6]. Among the statistical tests, the modified iterative measurement test, a method of serial elimination of the measurements that are most likely affected by large errors, performs best in terms of computational speed and efficiency in comparison to similar algorithms [7]. This test is based on the assumption of normal distribution of the measurements and makes use of the weighted least squares (WLS) as maximum likelihood estimator. The presence of gross errors violates this assumption, resulting in biased estimates, and therefore an iterative detection and elimination procedure is required, which comes at the cost of computational effort. Instead of using the weighted least squares estimator, Tjoa and Biegler [8] proposed the contaminated Normal as the maximum likelihood estimator, which is a bivariate distribution function accounting both for contributions of random and gross errors. They demonstrated a method of simultaneous gross error detection and data reconciliation. Based on robust statistics, additional estimators which are applicable to data reconciliation problems in presence of gross errors have been proposed, such as Lorenzian, Fair or Hampel estimator [9–12]. The corresponding studies demonstrated that robust estimators show a low sensitivity to gross errors present in measured data, and, as a consequence, robust estimators provide less biased estimates of reconciled process data. At the same time, the iterative sequential procedures for gross error detection and elimination, as required for statistical approaches of gross error detection, are avoided. Among the several classes of robust estimators, M-estimators, which are based on the maximum likelihood principle, are the most important ones for problems of data reconciliation. Prata et al. [2] provides a comprehensive overview of studies on robust es¨ timators applied to simulated as well as industrial data. Ozyurt and Pike [13] published one of the only comparative studies, in which six different methods derived from robust statistics were applied to literature and industrial process cases. It is concluded that robust estimators based on Hampel, Cauchy and Logistic distribution achieve promising performance for simultaneous data reconciliation and gross error detection. In case process models contain unknown parameters, an additional step of parameter estimation, in which the bias-free reconciled estimates are used to fit the values of the model parameters, has to be applied. The two-step method is the common approach to solve the two non-linear programming problems (NLP) in a sequential manner. First, a simultaneous data reconciliation and gross error detection is performed generating a set of measurements with only random errors by gross error elimination or correction. Second, the measurement set with ran21

Chapter 2

dom errors is used for simultaneous data reconciliation and parameter estimation (DRPE). Advanced methods formulate this problem as simultaneous gross error detection, reconciliation of measurements to obey conservation laws and estimation of parameters, by solving one single non-linear programming problem (onestep methods) [8, 12]. Chen et al. [14] performed a comparison between these methods concluding that with both approaches accurate parameter estimates and reconciled process values are obtained. The two-step method showed a better performance at the cost of 82 % higher computational time. Instead of using NLP algorithms for DRPE problems, also non-deterministic methods have been suggested, such as particle swarm optimization [2] or genetic algorithms. For these methods the determination of the derivatives of the problem is not required and the implementation is rather straightforward. Due to the global search character, non-convex optimization problems can be solved effectively, though the computational effort is typically higher due to a larger number of function evaluations. To conclude, for the analysis of the measurement data and validation of the CO2 capture pilot plant model the joint data reconciliation, gross error detection and parameter estimation procedure using robust estimators is deemed most suitable. This chapter is organized as follows: Section 2.2 gives a brief process description of the CO2 capture pilot plant, while Section 2.3 introduces the corresponding process and fluid thermodynamic models. The validation methodology consisting of experimental tests design and implementation, data acquisition and analysis, data reconciliation and parameter estimation is outlined in Section 2.4. The results of the validation are discussed in Section 2.5. The concluding remarks of Section 2.6 complete this chapter.

2.2

Process description

The process flow diagram of the CO2 capture pilot plant built at the site of the Buggenum IGCC power station in the Netherlands is depicted in Figure 2.1. This plant is a simplified, smaller version of a foreseen large-scale capture plant, equipped with sensors and analysers allowing for extensive performance measurements [1]. The syngas from the gasifier, which contains about 55 − 60 mol % CO and 2 − 6 mol % CO2 , enters the water-gas shift section of the CO2 capture plant at process conditions of 21 bar and 40 ◦ C and is mixed with process water in order to obtain a pre-set steam/CO ratio. The syngas-water mixture is fully evaporated and superheated by means of electrical heaters. Carbon monoxide present in the syngas is converted into hydrogen and carbon dioxide via a three-stage, sweet, high-temperature water-gas shift process with interstage cooling. Table 2.1 gives an overview of the temperature and pressure conditions of the reactors. The partial bypass around the first reactor (gas quench) allows for lower steam consumption, hence substantial energy saving [15, 16]. The excess process water is recovered 22


Parameter [◦ C]

Inlet temperature Outlet temperature [◦ C] Outlet pressure [bar] a

Reactor 1

Reactor 2

Reactor 3

340 495 18.5

340 470 18

340 350a 17.2

Reactor 3 features a lower catalyst activity than expected, probably caused by a slight over-reduction of the catalyst during commissioning or initial operation [17].

Table 2.1: Water-gas shift reactor conditions at reference state operation.

from the shifted syngas through condensation and recycled. Then the shifted syngas, which contains about 35 − 40 mol % of CO2 , enters the CO2 absorption and solvent regeneration section. Carbon dioxide is removed from the syngas in a packed column by means of physical absorption utilizing the solvent dimethylether of polyethylene glycol (DEPEG) at process condition of 40 − 45 ◦ C and 21.5 − 22.5 bar. The resulting H2 -rich syngas is fed to the gas turbine of the combined cycle power plant and the CO2 is recovered by threestage depressurization of the loaded solvent (flash pressures: 7.5 bar, 2.9 bar and 1.3 bar). The lean solvent is recycled to the absorber, while the CO2 -rich product stream is compressed and, in the case of the pilot plant, mixed with the H2 -rich syngas. Typically, 80 − 85 % of the CO2 present in the shifted syngas is removed. A more detailed process description is given by Damen et al. [1]. The large-scale CO2 capture plant process is very similar to the described process of the pilot plant, with the main difference being thermal energy recovery, or so-called heat integration within the WGS section: Electrical heaters and coolers are replaced by feed-effluent and feed-steam heat exchangers, whereby the steam is drawn from the heat recovery steam generator of the combined cycle power plant. In the absorption and solvent regeneration section, the gas recovered from the first flash vessel (also called H2 recovery vessel), which primarily contains co-absorbed hydrogen, is recompressed and recycled to the absorber column. This way the combustible H2 is not lost with the CO2 product.

2.3 2.3.1

Model development Process models

The model of the pre-combustion CO2 capture process is implemented into a commercial software tool [18], which is widely used in academia and industry for the modelling of chemical processes. The system model is assembled from models available in its process component library. The main components are the three water-gas shift reactors and the absorber column, which are briefly described in the following. The water-gas shift reactors are modelled as adiabatic equilibrium reactors 23

Chapter 2

Knock-out Drum Electric Heater 2

Electric Heater 1 Waste water

Reaction Water

Syngas Make-up Water

Electric Heater 3 Gas Quench

Feed Splitting Vessel

Process water

Rectifier Water-gas shift Cooler section Separator Booster Rectifier Compressor

1st Shift Reactor 2nd Shift Reactor

Compressor Cooler

Absorption & solvent regeneration section

CO2 Absorber

Airblown Cooler 1

H2-rich Syngas CO2 Compressor

3rd Shift Reactor Airblown Cooler 2 Condensate Pump

Flash 1

Flash 2 Flash 3

Product CO2

Solvent Heater Solvent Cooler Solvent Pump

Figure 2.1: Process flow diagram of the CO2 capture pilot plant built at the site of the Buggenum IGCC power station.

based on minimization of the Gibbs free energy of the reacting species. The reactor model allows specifications for restricted equilibrium in case the system does not reach complete equilibrium [18]. The corresponding parameter, the temperature approach to equilibrium, was set to 0 K. This assumption is addressed again during the discussion on data reconciliation and parameter estimation (see Subsection 2.4.4). The moderately exothermic reaction is given by CO(g) + H2 O(g) −→ CO2(g) + H2(g) ,

∆Hr = −41.1 kJ/mol.

(2.1)

The reactor inlet temperatures are maintained at a constant value, which is 340 ◦ C at reference state operation. This is achieved in the system model by choosing the outlet temperature of electrical heater 3 and air-blown cooler 1 as input (see Figure 2.1) and by providing the required temperature value. The temperature at the inlet of reactor 2 is maintained by a design specification which manipulates the flow rate of the gas quench. The absorber column is represented by a rigorous, rate-based model for simulating all types of multi-stage, vapour-liquid fractionation operations under steadystate operating conditions [18]. Model parameters are related to the packing specifications, which are summarized in Table 2.2 for the random and structured packing tested at the pilot plant. The correlation of Billet and Schultes [19] was selected to compute the mass transfer coefficients and the interfacial area. The solvent flow rate is maintained by a design specification manipulating the flow rate of solvent make-up added to the system. In the process model the following chemical components are accounted for: Ar, CO, CO2 , COS, DEPEG, H2 , H2 O, H2 S and N2 . Other trace components such as HCN, NH3 or CH4 are neglected. 24


Parameter

Random packing

Structured packing

Packing type

Raschig Super-Ring 0.6 215 0.98

Raschig Super-Pack 250Y 250 0.98

Specific surface area [m2 m−3 ] Void fraction [m3 m−3 ]

Table 2.2: Absorber column packing specifications.

2.3.2

Thermodynamic models of the process fluids

The thermophysical properties of the two-phase multi-component syngas-water and syngas-DEPEG mixture are calculated with the perturbed chain - statistical associating fluid theory (PC-SAFT) equation of state (EoS) [20] due to its success in predicting vapour-liquid equilibria of complex fluids and mixtures for a broad range of conditions. Moreover, due to the strong physical background and the small number of pure-component parameters, the PC-SAFT EoS is robust, consistent and extrapolative [20] even if calibrated on the basis of a limited amount of experimental thermodynamic property data. The solvent DEPEG is employed in industry under the commercial names of SelexolTM or Genosorb 1753TM , whereby the latter one was tested at the pilot plant. For simplicity the solvent, which is a blend of glymes, is represented as a pseudo pure fluid in the thermodynamic model. The required pure-component parameters were obtained by fitting the EoS to published vapour pressure and liquid density data which is available for the lighter compounds of the blend [21]. The parameters of the heavier pseudo glyme were estimated by extrapolation following a method demonstrated by Nannan et al. In order to determine the binary interaction parameters, the PC-SAFT EoS was applied to experimental vapour-liquid equilibrium data provided by the solvent vendor, resulting in good agreement for binary mixtures between DEPEG and gases such as CO, CO2 , H2 , N2 and water, if for the latter mixture cross-association interactions are considered [21]. The accuracy of the resulting thermodynamic model has proved to be suitable for engineering purposes. With regard to the transport properties, the liquid and vapour viscosity are predicted with the default models suggested for use together with the PC-SAFT EoS, see Ref. [18].

2.4

Validation methodology

The study documented here was aimed at the validation of the WGS section model by comparison of simulation results with design data and by evaluation of model estimates obtained during data reconciliation and parameter estimation using experimental data. The reconciled process measurements of the WGS section are further used for validation of a detailed reactor model which is based on kinetics. The validation of the absorption and solvent regeneration section was beyond 25

Chapter 2

the scope of this study. The focus of the absorption section validation is on the tuning of the mass transfer parameters in the absorber column using composition measurements at the absorber top.

2.4.1

Validation against design data

First, the developed process model of the WGS section of the pilot plant was compared to plant design data, which are consistent and free of measurement errors. The comparison of the design simulation results with the model predictions of the main mass flows, temperatures and compositions shows satisfactory agreement with average deviations smaller than 2 %. Larger deviations were observed for some composition estimates, which might be related to the choice of the EoS which is used to compute the thermodynamic properties of the process fluids. The detailed results of the comparison are not presented for the sake of conciseness.

2.4.2

Experiments

For the purpose of quantitative model validation but also individual component performance analysis, various parametric tests have been designed and executed. In order to increase the reliability of the validated models and its parameters, the experimental data used for validation should cover the entire operational range of the process. Therefore, the most sensitive system variables have been varied between their lower and upper operational limit. The resulting four different test runs for the water-gas shift section cover changes in 1. the individual reactor inlet temperatures, 2. the syngas inlet composition, 3. the syngas inlet mass flow rate and 4. the steam to CO ratio (overall as well as reactor specific). From the performed experiments 20 individual data sets were selected for model validation based on criteria discussed in Subsection 2.4.3, data acquisition and analysis. An overview of the data sets is given in Figure 2.2 in terms of measured values for temperatures, flows and compositions in the water-gas shift section. The variations in reactor 1 and 2 inlet temperature are depicted in Figure 2.2(a) together with the variations in syngas and reaction water flow rate. The inlet temperature of reactor 1 was varied from 315 ◦ C to 355 ◦ C , while that of reactor 2 from 335 ◦ C to 355 ◦ C. For the syngas inlet flow rate the experiments covered the range from 840 to 1235 kg/h, and for the reaction water flow rate from 990 to 1480 kg/h. Some of the observed changes in the reaction water flow are related to performed variations in the overall steam/CO ratio (measured at the inlet of the WGS section) and applied changes in the reactor 2 steam/CO ratio (measured at the inlet of reactor 2), which are shown in Figure 2.2(c). Among all data sets, 26

0.4

1400

0.35

370

1200

360

1000

350

800

340

600

330 320 310 300 0

2

4

6

Reactor 1 inlet temp. 400 Reactor 2 inlet temp. Syngas inlet flow 200 Reaction water flow 0 8 10 12 14 16 18 20 Data set

0.7

CO CO

2

H2 N2

0.3

0.65

0.25 0.2

0.6

0.15 0.1

0.55

CO [mole fration]

1600

380

CO2, H2, N2 [mole fraction]

390

Mass flow rate [kg/h]

Temperature [°C]


0.05 0 0

2

4

6

8

(a)

10 12 14 16 18 20 Data set

0.5

(b) 5

Molar steam/CO ratio [−]

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0

Reactor 1 inlet Reactor 2 inlet Overall (inlet WGS section) 2

4

6

8

10 12 Data set

14

16

18

20

(c)

Figure 2.2: Overview of the data sets used for the validation of the water-gas shift section model. a) Variations in the inlet temperature of reactor 1 and 2, syngas and reaction water flow rate. b) Variations in the syngas inlet composition. c) Variations in the molar steam/CO ratio in front of reactor 1, reactor 2 and overall (inlet WGS section).

the overall molar steam/CO ratio varies from 1.07 to 1.55, whereas the individual molar steam/CO ratio at the inlet of reactor 1 ranges from 3.51 to 4.57 and that of reactor 2 from 1.96 to 3.45. Figure 2.2(b) visualizes the variations in the syngas inlet composition. The CO mole fraction varies from 0.54 to 0.62, the CO2 content from 0.01 to 0.06 and the H2 content from 0.27 to 0.33.

2.4.3

Data acquisition and analysis

The process measurements obtained from the distributed control system include temperature, pressure, flow rate, level and composition measurements as well as control variables such as valve position, heater duty and cooler fan-speed. Measurements are recorded when changes in variable values exceed a threshold which was set for most of the variables to 0.1 % of the individual measurement range. 27

Chapter 2

The on-line storage, display and analysis of the recorded experimental data is managed by a commercial software [22]. Selected measurement data were transferred to suitable data processing tools for off-line analysis. First, the periods of steady-state operation were determined from the raw experimental data of the test runs via visual inspection of the main process variables, such as mass flow rates, compositions, temperatures and pressures. Constant values of these variables signify steady operation of the pilot plant. A minimum period length of 3 hours was considered to ensure a sufficient number of recorded data points, which applies especially to the discrete composition measurements (normal gas chromatograph (GC) analysis mode: one measurement every 15 minutes at each location). Outliers in composition measurements were removed from the steady-state period on a heuristic basis as these errors are rather easy to identify. For further data analysis the mean and the relative standard deviation of all variables were determined for each identified steady-state period and compared to other data sets. In case the relative standard deviations of the individual process variables were comparable in terms of their absolute value to the ones observed in other tests, the quality of the data was considered satisfactory for further analysis. Coriolis and vortex flow meters are used for mass flow measurements in the pilot plant. The coriolis meters measure mass flow rates which are directly recorded. The vortex meters measure volumetric flow rates which are converted into mass flow rates using stream dependent density conversion factors and then recorded. The data analysis is based on mass flow rates, and therefore coriolis measurements can be used straightforward, whereas recorded vortex measurements need to be corrected according to the actual density, which changes during operation based on variations in pressure, temperature and composition of the measured stream. The density calculations were performed with the PC-SAFT EoS. GC measurement analysis indicated that the recorded wet compositions are unreliable due to steam condensation within the sampling and/or analysis system. Using dry gas composition however resulted in closing elementary balances over the reactors within generally 2 %. Dry gas compositions were therefore used throughout data analysis as measures to prevent steam condensations proved not successful in fully eliminating condensation effects.

2.4.4

Data reconciliation and parameter estimation

Problem definition Commonly, multiple sets of independent data, whereby all measurements change with each data set, are used in order to obtain reliable parameter estimates (for the presented case: multiple data sets from independent test runs). In case the model parameters do not change throughout the measurement sets and assuming all measurements are subject to errors, then the individual data sets are coupled through the parameter estimates. The resulting simultaneous data reconciliation and parameters estimation problem is described as an errors-in-variables measured problem (EVM) and can be formulated in general terms as 28


J

minimize zi ,ui ,p

∑

i=1

f

zi − zM i σi

J

= ∑ f (εi ) i=1

subject to gi (zi , ui , p) = 0 hi (zi , ui , p) ≤ 0

(2.2)

zL ≤ zi ≤ zU uL ≤ ui ≤ uU pL ≤ p ≤ pU where the data reconciliation problem is described by f (generalized maximum likelihood objective function proposed by Huber [9]), a function of the standard residual εi which expresses the deviation between the measured variables zM i and the reconciled variables zi weighted by the standard deviation σi related to the i-th data set. The set of parameter estimates is represented by pi and the unmeasured variables by ui . hi corresponds to the equality and gi to the inequality constraints, which can represent either simple mass and material balances or non-linear process equations. In case of Gaussian error distribution the function f is formulated as the weighted least squares 1 f = ε2i . (2.3) 2 For the data reconciliation presented in this chapter the contaminated Normal estimator was used, that is 2 pCN ε2 ε exp − 2i . f = ln (1 − pCN ) exp − i + 2 bCN 2bCN

(2.4)

The contaminated Normal is constructed based on the maximum likelihood principle accounting both for contributions from random and gross errors in the measurements. The parameter pCN is defined as the probability of a gross error in the measurements and bCN as the ratio of the standard deviation of the gross error to that of the random error. These values were set to 0.05 and 10 respectively [8]. The influence function, which represents the influence of a measurement on the reconciled estimates, can be used to judge the robustness of such algorithms. For the Gaussian distribution the influence function increases linearly with the size of the error and is therefore of unbounded nature (Figure 2.3). Large gross errors in the measurements will be assigned a higher weight during the reconciliation leading to biased estimates. The influence function of the contaminated Normal distribution has the same trend as the WLS for small errors εi < 2 as the distribution function for random errors dominates. For larger errors the gross error distribution dominates, resulting in the decrease of the influence function for errors 2 < εi < 4, and then in the proportional increase at a low rate. The contaminated Normal is most effective for moderate-size gross errors (3 − 30 σ) due to the unbounded nature for errors above 50 σ [14]. 29

Chapter 2

4

Weighted least squares Contaminated Normal

Influence function

3.5 3 2.5 2 1.5 1 0.5 0 0

2

4 6 Standard residual

8

10

Figure 2.3: Influence function for weighted least squares and contaminated Normal distribution.

Gross errors can be identified through analysis of the obtained residuals resulting from the application of the DRPE procedure. For robust estimators, explicit cut points can be defined over the roots of the first and second derivative of the influence function, which represent optima or inflection points. The cut points of the contaminated Normal distribution are determined at 2.131 σ and 3.34 σ [13]. Data reconciliation framework The process model of the WGS section contains 3 main parameters, the approach to equilibrium of the water-gas shift reactors, 30 independent variables (input variables), which determine the state of the process, and a large number of dependent variables (output variables), whereby 32 of them are used to evaluate the performance of the system. The distributed control system provided measurements of all dependent and independent variables (62 measurements) including 7 flow rates, 17 temperature, 8 pressure and 30 composition (components counted individually) measurements. This provides sufficient redundancy of the measurements and allows to perform data reconciliation. A sensitivity analysis aiming at reducing the complexity of the optimization problem was conducted in order to identify the independent variables whose value uncertainty would have minor impact on the system performance. From the analysis it can be concluded that, for 20 of the independent variables, the measured value can be taken as model input because possible measurement errors of these variables have negligible impact on the output variables. This is equivalent to the assumption that these variables are error-free and no reconciliation is required. Further, a preliminary analysis was performed in order to investigate if the reactor model parameters are time-dependent and therefore dependent on the data set, namely on the operating condition. A preliminary simultaneous data reconciliation and parameter estimation was conducted based on the WLS problem formulation of each individual data set utilizing the internal optimization structure 30


and algorithms of the process simulation tool. The estimates of the temperature approach parameters of reactor 1 and reactor 2 were constant with a value of 0 K throughout all data sets. This observation has been confirmed during detailed reactor model validation [17]. The parameter value of 0 K indicates that the equilibrium temperature is reached and no corrections have to be applied. Both parameters are therefore removed from the optimization problem and assumed constant with a value of 0 K for the considered measurement period. Reactor 3 features a lower catalyst activity than expected, probably caused by a slight overreduction of the catalyst during commissioning or initial operation, and is thus not representative of common catalyst performance [17]. This resulted in performance variations during the test period as often equilibrium was not reached in reactor 3. Consequently, the value of the model parameter of reactor 3 is different for each data set, ranging from 5 to 60 K. To summarize, the number of independent variables can be reduced from 30 to 10 variables, the model parameters of reactor 1 and 2 can be considered constant with a value of 0 K for the considered measurement period, and the model parameter of reactor 3 is dependent on the operating condition. The optimization problem can therefore be reduced to a DRPE which can be performed for each data set individually, as no common parameters couple the individual data sets. The problem can be simplified as minimize x,y,p

∑ f (εxi ) + f

y

εi

(2.5)

subject to g(x, y, p) = 0 where x corresponds to the independent variables and y to the dependent variables. The objective function is formulated in terms of the contaminated Normal distribution, and was implemented using a common programming language and software tool [23]. The non-linear equality process constraints are represented by the process model, which is interfaced to the programming environment. Inequality constraints were not involved for this specific optimization problem. The approach of using different tools for the process modelling and the optimization of the objective function requires typically the use of non-deterministic optimization methods. Different numerical multi-variable, derivative-free algorithms for global optimization of non-linear problems have been evaluated for this specific problem. A good comparison of such algorithms has been published by Rios and Sahinidis [24]. An open source implementation of the DIRECT algorithm first introduced by Jones et al. [25] was used. The DIRECT algorithm (Dividing RECTangles) is a sampling algorithm for difficult global optimization problems with bounded constraints which requires no knowledge of the objective function gradient. Global search algorithms for optimization problems with a non-convex objective functions feature several advantages as the algorithm aims to converge to the global minimum avoiding local minima. The disadvantage of the global convergence is the typically higher number of function evaluations related to the extensive search 31

Chapter 2

over the domain, which results in a more computationally intensive solution. The reduction of the search domain will therefore contribute to a significantly faster convergence. In order to provide good initial guesses for the optimization variables and to reduce the search domain of the global search, initially a simultaneous DRPE based on the WLS distribution was performed for each data set. This optimization was carried out within the process modelling tool, which supports data reconciliation based on WLS. The process simulation is solved in the so-called equation-oriented (EO) mode, which is based on the strategy to simultaneously solve the entire flowsheet avoiding nested convergence loops. This allows to solve much larger problems in comparisons to the sequential solution method with the same computational effort. The numerical method available with the process simulator is a solver which implements a variant of the successive quadratic programming (SQP) algorithm making use of analytical derivatives. This makes the simulations efficient and fast. The result of this optimization are reconciled data sets with estimated parameters which might be biased due to gross errors present in the measurements. The obtained estimates for the independent variables and the estimate for the parameter are used as starting value for the DIRECT algorithm and to determine the bounds of the search domain. Based on experience of multiple preliminary optimization tests, the number of iterations for the DIRECT algorithm has been set to 2000. To summarize, the DIRECT algorithm, initialized with estimates obtained from a WLS reconciliation, minimizes the value of the objective function, which is formulated as contaminated Normal distribution, by iteratively calling the process model with a set of updated optimization variables.

2.5

Results and discussion

The results of the simultaneous DRPE in terms of the standardized residuals1 of the main process variables (mass flow rates, temperatures and compositions) within the water-gas shift section are depicted in Figure 2.4 and Figure 2.5. From the total 920 measurements, which were reconciled for 20 data sets, 31 mass flow rate, 23 temperature and 53 composition measurements were detected as gross errors using the inflection point of the influence function of the contaminated Normal estimator as cut-off point. As the process model was only validated against design data, first the accuracy and validity of the model is investigated by comparing the simulation results obtained from the DRPE against the experimental measurements. This analysis shall help to verify that the detected gross errors are not the result of inaccuracies related to the process or thermodynamic model, but related to measurement 1 The

standardized residual is defined as the deviation between the reconciled and the measured value (residual) divided by the corresponding standard deviation.

32


Measurement Mass flows Reaction water Syngas inlet Quench Reactor 1 inlet Shifted syngas Make-up water Temperatures Quench Knock-out drum Reactor 1 outlet Reactor 2 outlet

Average deviation [%] 1.92 3.11 1.19 0.09 0.58 1.63 0.12 0.03 0.12 0.76

Table 2.3: Average deviation of model predictions with respect to experimental measurements.

errors. The results of this comparison, which are summarized in terms of average deviations in Table 2.3, demonstrate that a good agreement between the model predictions and the measurements for the main mass flows and temperatures in the WGS section is achieved, with deviations smaller than 3.2 %. Considering that similar values were obtained during the validation against design data, it can be concluded that the process and employed fluid thermodynamic model are valid and sufficiently accurate to predict the pilot plant performance and can be used for data reconciliation and parameter estimation. Figure 2.4(a) shows the standardized residuals of the mass flow rates. Relatively large errors are mainly observed in the syngas and quench flow measurements. Considering the corresponding absolute residuals, see Figure 2.4(b), it can be concluded that syngas measurements are systematically too high by approximately 35 kg/h. For the majority of the measured values of the quench flow and the reactor 1 inlet flow the deviations are very small. Measured values of the make-up water and shifted syngas are slightly too high, and the reaction water mass flow is systematically too low by approximately 25 kg/h. The systematic errors are most likely attributable to miscalibration of the measurement instruments. In general, except for a few outliers, in particular present in the quench flow, the mass flow residuals are randomly distributed. Figure 2.4(c) and Figure 2.4(d) depict the standardized and absolute residuals of the temperatures. The deviations of the quench and knock-out drum temperature are small and distributed around zero. For the outlet temperature of reactor 1 the model predicts on average 0.6 K lower values than the measurements. Only in case of data set 4 and 18 the difference is somewhat higher, being 2.5 K. Overall, as far as reactor 1 is concerned, these results show good agreement between measured data and predictions of process simulations. 33

Chapter 2

50

Syngas inlet Reaction water Quench Reactor 1 inlet Shifted syngas Make−up water

15 10

Absolute residual of flow rate [kg/h]

Standardized residual of flow rate

20

5 0 −5 −10 −15 −20 0

2

4

6

8

10 12 Data set

14

16

18

40 30 20 10 0 −10 −20 −30 −40 −50 0

20

2

4

6

8

20

Quench Knock−out drum Reactor 1 outlet Reactor 2 outlet

15 10 5 0 −5 −10 −15 −20 0

2

4

6

8

14

16

18

20

14

16

18

20

(b)

Absolute residual of temperature [K]

Standardized residual of temperature

(a)

10 12 Data set

10 12 Data set

14

16

18

20

6 5 4 3 2 1 0 −1 −2 −3 −4 −5 −6 0

2

4

6

8

(c)

10 12 Data set

(d)

Standardized residual (syngas inlet)

10 8

CO CO2

6

H2

4

N2

2 0 −2 −4 −6 −8 −10 0

2

4

6

8

10 12 Data set

14

16

18

20

(e)

Figure 2.4: Results of simultaneous DRPE for the water-gas shift section in terms of standardized and absolute residuals of mass flow rates, temperatures and the syngas inlet composition.

34


The absolute residuals of the outlet temperature of reactor 2 are in the range of 2 to 5.5 K, which are consequently detected as gross errors (deviations larger than 3.34 σ). However, this mismatch is most likely unrelated to a systematic measurement error as other axial thermocouples in reactor 2 indicated similar temperatures at the end of the catalyst bed. It is therefore more likely that the observed deviations are due to model inaccuracies resulting in the underestimation of the outlet temperature of reactor 2. The standardized residual of the syngas composition at the inlet of the WGS section is shown in Figure 2.4(e). The standardized residuals of the syngas composition at the inlet and outlet of the three reactors are represented in Figure 2.5. The majority of the residuals range between ± 2 σ. Larger biases and outliers are mainly observed in reactor outlet composition measurements. All outlet streams are analysed by the same gas chromatograph, which might explain similar trends in terms of the residuals. For reactor 2 larger deviations in the outlet composition might also be related to the previous observation that the model underestimates the adiabatic temperature rise. Measurement values for CO and H2 are indicated to be systematically biased at most of the measurement locations. Values for CO2 and N2 are randomly distributed around zero at most locations. An overview of the locations where systematic biases in the composition measurements are observed is given in Table 2.4. The inlet composition of reactor 2 and reactor 3 are two independent measurements of the same process stream, and therefore similar results in terms of standardized residuals are expected. The residuals of both measurement locations are comparable with slightly higher deviations for the outlet composition of reactor 2, see Figure 2.5(d) and Figure 2.5(e). Outliers can be identified by comparing the results of both streams. The outliers therefore are: the constituents H2 and CO2 measured at reactor 2 outlet of data set 10, 14, 15 and 20, and the constituent CO recorded at reactor 3 inlet of data set 4 and 17. To conclude, the measurements of CO and H2 are systematically biased indicating that the measurement devices are not optimally calibrated for these two constituents. In addition, the quality of the measurements at the reactor outlet is lower than those at the reactor inlet, and therefore calibration of the respective gas chromatograph is advisable for future measurements. In the following, the data reconciliation results of the four individual test runs as described in Subsection 2.4.2 are discussed in detail, in order to validate the predictive capabilities of the model throughout the entire operational range. Data set 1 to 4 correspond to variations in the inlet temperature of reactor 1, a decrease from 325 to 315 ◦ C, whereas the inlet temperature of reactor 2 remained constant at 335 ◦ C. Data sets 5 and 6 correspond to an operation, whereby the inlet temperature of all reactors was increased from 345 to 355 ◦ C (see Figure 2.2). Figure 2.4(d) suggests that for the outlet temperature of reactor 1 the absolute deviation between measured and reconciled values is rather constant at 0.5 K for inlet temperatures in the range 325 − 355 ◦ C (data set 1, 2, 5, 6), and increases 35

10 8

CO CO2

6

H

4

N2

2

2 0 −2 −4 −6 −8 −10 0

2

4

6

8

10 12 Data set

14

16

18

Standardized residual (reactor 1 outlet)

Standardized residual (reactor 1 inlet)

Chapter 2

20

20

CO CO

2

15

H

2

N

10

2

5 0 −5 −10 −15 −20 0

2

4

6

10 8

CO CO2

6

H2

4

N2

2 0 −2 −4 −6 −8 −10 0

2

4

6

8

10 12 Data set

14

16

18

20

6 4 2 0 −2 −4

CO CO2

−6

H2

−8

N2

6

8

10 12 Data set

(e)

14

16

18

20

Standardized residual (reactor 3 outlet)


8

4

16

18

20

CO CO2

15

H2

10

N2

5 0 −5 −10 −15 −20 0

2

4

6

8

10 12 Data set

14

16

18

20

(d)

10

2

14

20

(c)

−10 0

10 12 Data set

(b) Standardized residual (reactor 2 outlet)


(a)

8

20 15 10 5 0 −5 −10

CO CO2

−15

H2 N2

−20 0

2

4

6

8

10 12 Data set

14

16

18

20

(f)

Figure 2.5: Results of simultaneous DRPE for the water-gas shift section in terms of standardized residuals for the inlet and outlet composition of reactor 1, 2 and 3.

36


Component

Measurement values systematically too low

Measurement values systematically too high

CO

R1 inlet, R1 outlet, R2 outlet, R3 inlet, R3 outlet R1 inlet, R2 inlet Syngas inlet

R2 inlet

CO2 H2

R1 inlet R1 outlet, R2 outlet, R3 inlet, R3 outlet

N2

Table 2.4: Overview of composition measurements which are either systematically too high or too low in comparison to the reconciled estimates.

up to 2.5 K when lowering the inlet temperature of reactor 1 to 315 ◦ C (data set 3). When reducing the reactor inlet temperature a boost in CO conversion and an increase in adiabatic temperature rise is expected, assuming that the reaction front remains within the catalyst bed and thermodynamic equilibrium is reached, which was the case during the tests. The higher deviations at lower inlet temperatures of the reactor are therefore not measurement errors but related to a lower accuracy of the model estimates. For changes in the inlet temperature of reactor 2 (335 − 355 ◦ C) no clear tendency of the absolute error is observed, and the model underestimates the temperature in a range between 2 to 4 K as discussed previously. The deviations in terms of standardized error in the mass flows, quench and knock-out drum temperature and reactor compositions are comparable among these 6 data sets, except for data set 3 and 6, in which larger absolute errors are observed in the quench mass flow. It can be concluded that in the tested operational range of reactor 1 (315 − 355 ◦ C) the model predictions are in satisfactory agreement with the measurements, with maximum absolute deviations of 2.5 K at operation with high adiabatic temperature rise, and that for the tested range of reactor 2 the model systematically underpredicts the temperature. Data set 7 to 10 represent the variation in syngas inlet composition, see Figure 2.2. This was achieved by recycling H2 product or CO2 product in different amounts. The standardized residuals of most of the measured variables are comparable to the residuals observed in other data sets. Large errors are observed though in the H2 and CO content of the syngas inlet stream for data set 7, 8 and 10, see Figure 2.4(e). Throughout these three data sets a maximum amount of hydrogen was added to the syngas at close distance upstream of the sampling point of the gas chromatograph which is measuring the syngas inlet composition. It is likely that the hydrogen enriched syngas stream was not perfectly mixed when it reached the GC, and therefore the recorded measurements are biased. The actual hydrogen content is higher and the carbon monoxide content is lower than measured. For data set 10 large errors of H2 and CO2 are present in all reactor outlet com37

Chapter 2

position measurements. The comparison of measurements at the outlet of reactor 2 and inlet of reactor 3 (same stream but different GC) leads to the conclusion that these errors are outliers, which are caused by malfunction of the measurement device (all outlet compositions measured with the same GC). Data set 11 to 15, 19 and 20 represent operation at different steam/CO ratio, see Figure 2.2. The standardized residuals of the mass flows are comparable to the ones observed in the other data sets. Only for the make-up water flow of data set 19 a different error is indicated. The test represented by data set 20 was performed without the syngas compressor in operation, and therefore no mass flow measurement of the shifted syngas was available. For the temperature of the quench, the knock-out drum and the reactor 1 outlet the deviations between the measured and reconciled values are comparable to the other data set and show no influence due to changes in steam/CO ratio. The outlet temperature of reactor 2 is underpredicted by the model as discussed before. This mismatch is also observed throughout the variations of steam/CO ratio. For data set 11 to 15 an underprediction of 4 K can be observed and for data set 19, 20 the underprediction is of 5.5 K. Data set 13 to 15, 19 and 20 represent operation at low steam/CO ratio, in particular of reactor 2 (see Figure 2.2). No clear increasing trend of the underprediction of the outlet temperature of reactor 2 due to a decrease in steam/CO ratio can be observed. It can therefore be concluded that the accuracy of the model predictions for reactor 2 is the same for operation at different steam/CO ratios. Large errors for CO2 and H2 measurements are indicated for data set 14 and 15 in almost all reactor composition measurements (expect reactor 3 inlet). Similarly, slightly smaller errors are observed for data set 20 in all reactor inlet composition measurements. The question arises, if these deviations are related to gross errors or to inaccuracy of model predictions for operation at low steam/CO ratio. Data set 19 and 13 correspond to similar operation at low steam/CO ratio. For both data sets a good agreement in composition measurements with the model predictions is observed. In addition, data set 14, 15 and 20 were recorded almost in sequence during the same test run, whereby the other data sets correspond to tests performed at different times. It seems likely that the observed deviations are therefore related to measurement gross errors, which are possibly caused by the specific operational conditions at that time. Data set 17 represents operation at low syngas mass flow and data set 18 at high syngas mass flow, see Figure 2.2. The residuals of the mass flows for both data sets show comparable deviations as observed in other data sets, only for the case of high syngas flow a higher error is indicated in the shifted syngas. Considering that all other performance variables (mass flows, temperatures and compositions) show good agreement with the model predictions, it is likely that the syngas outlet mass flow measurement is biased due to decreasing accuracy of the mass flow measurement device. Comparing the absolute residuals of the outlet temperature of reactor 1 of both data sets, see Figure 2.4(d), then good agreement is observed for the case of high 38


Value of objective function

140

WLS Contaminated Normal

120 100 80 60 40 20 0 0

2

4

6

8

10 12 Data set

14

16

18

20

Figure 2.6: Comparison of objective function values using the WLS distribution and the contaminated Normal distribution.

syngas flow, while for low syngas flows the temperature is underpredicted by 2.5 K. A decrease in throughput does not influence the conversion rate nor the adiabatic temperature rise, and therefore it cannot explain the observed difference. During another test at low mass flow operation, which is not documented in this study, a similar underprediction of reactor 2 outlet temperature was observed, which indicates that the reactor heat loss which is assumed constant and equal to 2 kW is flow-dependent, decreasing with lower mass flow rates. Figure 2.6 summarizes the results of the objective function values of the DRPE using the WLS distribution and the contaminated Normal distribution for the 20 measured data sets. Clear improvements of the functional values between the WLS solution and the robust estimator can be observed, typically larger for cases where the WLS result is high indicating the presence of more or larger gross errors. The data sets 4, 10, 14, 15, 17, 20 show in comparison to the other data sets the least good agreement between the measured and reconciled variable values. Reasons for this result have been discussed above. All other data sets represent similar good agreement of the measurements and the model predictions. Hence, it can be concluded that the model is capable of predicting the WGS section performance with satisfactory accuracy throughout the entire operational range.

2.6

Conclusions

This chapter discusses the steady-state modelling and simulation of a pre-combustion CO2 capture plant comprising a water-gas shift and an absorption and solvent regeneration section. Model validation using experimental data obtained from the pilot plant operated at the Buggenum IGCC power station has been demonstrated for the WGS section model by performing simultaneous data reconciliation and parameter estimation including gross error detection. The contaminated Normal 39

Chapter 2

distribution has been identified to be most suitable to formulate the objective function of the combined minimization problem consisting of 11 optimization variables (10 independent process variables and 1 model parameter). In order to perform the simultaneous DRPE the process simulator was interfaced to an optimization tool and an implementation of the DIRECT algorithm was used to perform the global optimization. Experimental measurements corresponding to 20 data sets have been selected covering the entire operational range of the WGS section, including variations of the inlet temperature of reactor 1 and reactor 2, of the syngas mass flow, of the syngas composition and of the steam/CO ratio. The following conclusions can be drawn from the comparison of the reconciled and measured data: 1. The model predictions of mass flows, temperatures and compositions show good agreement with the measured values and 90 % of the reconciled estimates (independent and dependent variable values) are within a deviation of ±3.34 σ (gross error cut point). 2. The syngas and reaction water mass flow rate are measured with a systematic bias most likely caused by miscalibration of the measurement device. The experimental values are on average 35 kg/h too high for the syngas flow and 25 kg/h too low for the reaction water flow. 3. The predictions of the adiabatic temperature rise of reactor 1 is in good agreement with the measurements. For operation with high adiabatic temperature rise (low reactor inlet temperatures) the model underestimates the temperature with maximum deviation of 2.5 K. The adiabatic temperature rise of reactor 2 is systematically underpredicted by 4 K throughout the entire operational range. 4. Regarding the composition measurements, systematic biases affect most of the measurements of CO concentrations (measured values are too low), and of H2 concentrations (measured values are too high). Thus, it can be concluded that the measurement devices are not optimally calibrated as far as CO and H2 are concerned. The gas chromatograph measuring the reactor inlet compositions is more accurate than the one measuring the reactor outlet compositions. The validated pilot plant model provides a reliable basis for the development of large-scale CO2 capture plant models. This would require few adaptations of the process scheme like, for example, replacing pilot plant specific components, such as electrical heaters and coolers with components utilized in a large-scale CO2 capture plant. The resulting models are powerful tools for further performance analysis and process optimization considering plant design and optimal operating conditions with the objective to maximize plant efficiency or power output. Moreover, steadystate process simulation results for on- and off-design operation are useful to ease initialization of dynamic simulations. 40


Nomenclature bCN Hr p pCN u z

= = = = = =

Parameter of the contaminated Normal estimator Heat of reaction, J mol−1 Process parameter Parameter of the contaminated Normal estimator Unmeasured variable Process variable

= =

Standard residual Standard deviation

=

Data set

= = = = =

Lower bound Measured Upper bound Independent variable Dependent variable

= = = = = = = = = = = = =

Dimethylether of polyethylene glycol Dividing rectangles Data reconciliation and parameter estimation Equation-oriented Equation of state Errors-in-variables Gas chromatograph Integrated gasification combined cycle Non-linear programming Perturbed chain - statistical associating fluid theory Succesive quadratic programming Water-gas shift Weighted least squares

Greek symbols ε σ Subscripts i Superscript L M U x y Acronyms DEPEG DIRECT DRPE EO EoS EVM GC IGCC NLP PC-SAFT SQP WGS WLS

41

Chapter 2

References [1] Damen, K., Gnutek, R., Kaptein, J., Nannan, N. R., Oyarzun, B., Trapp, C., Colonna, P., van Dijk, E., Gross, J., and Bardow, A., 2011. “Developments in the pre-combustion CO2 capture pilot plant at the Buggenum IGCC”. Energy Procedia, 4(0), pp. 1214–1221. [2] Martinez Prata, D., Schwaab, M., Luis Lima, E., and Carlos Pinto, J., 2010. “Simultaneous robust data reconciliation and gross error detection through particle swarm optimization for an industrial polypropylene reactor”. Chemical Engineering Science, 65(17), pp. 4943–4954. [3] Tamhane, A. C., and Mah, R. S., 1985. “Data reconciliation and gross error detection in chemical process networks”. Technometrics, 27(4), pp. 409–422. [4] Crowe, C., 1996. “Data reconciliation - Progress and challenges”. Journal of Process Control, 6(2-3 SPEC. ISS.), pp. 89–98. [5] Romagnoli, J., and Sanchez, M., 2000. Data Processing and Reconciliation for Chemical Process Operations. Academic Press. [6] Narasimhan, S., and Jordache, C., 1999. Data Reconciliation and Gross Error Detection: An Intelligent Use of Process Data. Gulf Professional Publishing. [7] Serth, R., Valero, C., and Heenan, W., 1987. “Detection of gross errors in nonlinearly constrained data: A case study”. Chemical Engineering Communications, 51, pp. 89–104. [8] Tjoa, I., and Biegler, L., 1991. “Simultaneous strategies for data reconciliation and gross error detection of nonlinear systems”. Computers and Chemical Engineering, 15(10), pp. 679–690. [9] Huber, P., 1981. Robust Statistics. John Wiley & Sons, New York. [10] Johnston, L., and Kramer, M., 1995. “Maximum likelihood data rectification: Steady-state systems”. AIChE Journal, 41(11), pp. 2415–2426. [11] Albuquerque, J., and Biegler, L., 1996. “Data Reconciliation and Gross-Error Detection for Dynamic Systems”. AIChE Journal, 42(10), pp. 2841–2856. [12] Arora, N., and Biegler, L., 2001. “Redescending estimators for data reconciliation and parameter estimation”. Computers and Chemical Engineering, 25(11-12), pp. 1585–1599. ¨ [13] Ozyurt, D., and Pike, R., 2004. “Theory and practice of simultaneous data reconciliation and gross error detection for chemical processes”. Computers and Chemical Engineering, 28(3), pp. 381–402. [14] Chen, X., Pike, R., Hertwig, T., and Hopper, J., 1998. “Optimal implementation of on-line optimization”. Computers and Chemical Engineering, 22(SUPPL.1), pp. S435–S442.

42


[15] Martelli, E., Kreutz, T., and Consonni, S., 2009. “Comparison of coal IGCC with and without CO2 capture and storage: Shell gasification with standard vs. partial water quench”. Energy Procedia, 1(1), pp. 607–614. [16] Carbo, M., Boon, J., Jansen, D., van Dijk, H., Dijkstra, J., van den Brink, R., and Verkooijen, A., 2009. “Steam demand reduction of water-gas shift reaction in IGCC power plants with pre-combustion CO2 capture”. International Journal of Greenhouse Gas Control, 3(6), pp. 712–719. [17] van Dijk, H. A. J., Cohen, D., Hakeem, A. A., Makkee, M., and Damen, K., 2014. “Validation of a water-gas shift reactor model based on a commercial FeCr catalyst for pre-combustion CO2 capture in an IGCC power plant”. Chemical Engineering Journal. submitted for publication. [18] Aspen Technology, Inc., 2013. Aspen Plus V7.3. www.aspentech.com. [19] Billet, R., and Schultes, M., 1993. “Predicting mass transfer in packed columns”. Chemical Engineering and Technology, 16(1), pp. 1–9. [20] Gross, J., and Sadowski, G., 2001. “Perturbed-Chain SAFT: An Equation of State Based on a Perturbation Theory for Chain Molecules”. Industrial and Engineering Chemistry Research, 40, pp. 1244–1260. [21] Nannan, N. R., de Servi, C. M., van der Stelt, T., Colonna, P., , and Bardow, A., 2013. “An Equation of State Based on PC-SAFT for Physical Solvents Composed of Polyethylene Glycol Dimethylethers”. Industrial and Engineering Chemistry Research, 52, pp. 18401–18412. [22] ABB Ltd, 2014. Power Generation Information Manager (PGIM). www.abb.com. [23] The MathWorks, Inc., 2010. MATLAB R2010b. www.mathworks.com. [24] Rios, L., and Sahinidis, N., 2013. “Derivative-free optimization: A review of algorithms and comparison of software implementations”. Journal of Global Optimization, 56(3), pp. 1247–1293. [25] Jones, D., Perttunen, C., and Stuckman, B., 1993. “Lipschitzian optimization without the Lipschitz constant”. Journal of Optimization Theory and Applications, 79(1), pp. 157–181.

43

“There is one sign the Soviets can make that would be unmistakable, that would advance dramatically the cause of freedom and peace. General Secretary Gorbachev, if you seek peace, if you seek prosperity for the Soviet Union and eastern Europe, if you seek liberalization, come here to this gate. Mr. Gorbachev, open this gate. Mr. Gorbachev, tear down this wall!” Ronald Reagan, President of the United States, Berlin Wall Speech, Brandenburg Gate, West-Berlin, Juni 12, 1987

3

Design optimization for flexible operation of a pre-combustion CO2 capture plant embedding experimental knowledge1 The work documented in this chapter is focused on the design optimization of a pre-combustion CO2 capture plant comprising a sweet high-temperature water-gas shift process and an integrated H2 S and CO2 removal process using dimethylether of polyethylene glycol as physical solvent. The steady-state model of the commercialscale plant has been derived from pilot plant models, which have been extensively validated against experimental data obtained from the pre-combustion CO2 capture pilot facility. A two-phase optimization-based design approach suited to the use of process simulator environments has been adopted. In the first phase, global design decisions at plant level are evaluated, targeting the minimization of the energy consumption due to CO2 capture. These are the extent of CO conversion in the water-gas shift unit and the percentage of CO2 capture in the removal unit. An optimization of both global design variables is presented considering i) flexible operation in terms of overall carbon capture target, ii) deactivation of catalyst activity throughout the catalyst life and iii) different operational limits of the steam/CO ratio in the watergas shift unit. Experimental data from the pilot plant has been used to determine the rate of catalyst deactivation and the operational limits of steam/CO ratio. The second phase of the design procedure targets the local decision variables at unit level. Two studies are presented focusing on: 1) the design of the solvent regeneration and CO2 compression section, and 2) the impact of the solvent temperature on the energy consumption and equipment cost of the removal unit. 1 The contents of this chapter are submitted for publication in: Trapp, C., Thomaser, T., van Dijk, H.A.J., and Colonna, P., 2014. Fuel.

Chapter 3

3.1

Introduction

Integrated gasification combined cycle (IGCC) power plants are a promising technical solution if electricity production must integrate carbon capture, because the pre-combustion CO2 capture is realized at high partial pressure, and the plant net energy efficiency is estimated to be higher than that of conventional pulverized coal (PC) steam power plants [1]. Moreover, gasification features advantages regarding fuel flexibility, low emissions of regulated air pollutants, and generation of different products, such as electricity, fuels and chemicals. However, a number of technical challenges must be overcome in order to bring this technology to commercial scale. These are high energy penalty associated with CO2 capture, compression, transport and storage, the increase in system complexity, and process availability. For the successful implementation of CCS technologies also challenges related to economic viability, public acceptability, uncertainties in policies and regulations need to be overcome. Due to these shortcomings, an increasing amount of scientific literature documents research and development focused on the investigation of IGCC technology with CCS. These studies are commonly based on the use of steady-state modelling and simulation, and are aimed at the analysis of performance and efficiency, [2–4] as well as they often provide supplemental economic evaluation [1, 5–9] at different level of detail. The efficiency penalty due to 85 − 90 % of CO2 capture is estimated in the range of 6.4 to 12 %-points. Differences in the estimation of the efficiency penalty due to CO2 capture and overall plant performance are related to the selected technology (for example, the type of gasifier and the type of CO2 capture technology), to the modelling assumptions, and finally to the level of detail of the models. Various CO2 capture technologies are currently studied, with the aim to reduce cost and energy penalty. These are chemical and physical absorption [5, 10, 11], membrane-based gas separation [12–15] and adsorption [16–18]. Physical absorption is next to amine-based chemical absorption the most mature and commercially proven technology for removal of CO2 , with the advantage of being effective at high partial pressure, as it is the case in pre-combustion capture plants. Employed solvents are methanol and blends composed of glymes, which goes under various commercial names2 . The use of glymes is better in terms of energy consumption of the capture process [10]. The overall process performance and efficiency can also be improved by optimizing the thermal integration of a decarbonised IGCC system with the application of such methods as thermodynamic heuristics and pinch theory [19, 20]. The automated efficiency or power output optimization of IGCC power plants with CO2 capture is addressed in only few studies. Martelli et al. [21] considered a system which employs a novel water-gas shift (WGS) configuration within the CO2 capture plant, which allows for a substantial reduction of the steam consumption, while maintaining the same CO2 capture rate. The design of the novel WGS scheme, which is based on a concept introduced by Carbo et al. [22], is 2 SelexolTM

46

or Genosorb 1753TM

Design optimization for flexible operation of a pre-combustion CO2 capture plant embedding experimental knowledge

optimized and finally analysed in terms of thermodynamic and economic aspects. Bhattacharyya et al. [23] performed an extensive optimization on the design and operating conditions of a pre-combustion CO2 capture unit integrated with an IGCC power plant. Their methodology is based on a three-phase design approach specifically suited for process simulator environments. Apart from design studies based on simulations, it was recognized that an essential step toward the commercial realization of IGCC power plants with CO2 removal is pilot-scale demonstration of the technology [23]. Together with performance verification, an important aspect of such a demonstration project is the generation of measurements for model validation. This allows to overcome one of the pitfalls of previous studies, namely the unknown uncertainty affecting the models on which the design studies are based. In order to demonstrate pre-combustion CO2 capture technology and investigate its performance a unique, fully instrumented CO2 capture pilot plant has been realized at the Buggenum IGCC power station in the Netherlands by the utility company Vattenfall [24]. The design study on a commercial-scale CO2 removal plant for IGCC power stations discussed here is based on the knowledge obtained during the test campaigns at the pilot facility. The novel aspects are: a) The process models of the large-scale CO2 capture plant used in this design study are based on extensively validated models of the capture pilot plant. Moreover, the thermodynamic model of the solvent-syngas mixture is validated against experimental data. b) An optimization-based approach is presented with focus on operational flexibility. The values of important design variables and operational limits applied during the design optimization are determined based on experiments performed at the pilot plant. This chapter is structured as follows: Section 3.2 treats the optimization methodology. Thereafter, the CO2 capture process and the models of the watergas shift and H2 S and CO2 removal unit are described in Section 3.3. Pilot plant experiments relevant for the design optimization are discussed in Section 3.4. The results of the analysis are presented and discussed in Section 3.5. Finally, conclusions are given in Section 3.6.

3.2

Methodology

This design study is based both on steady-state process simulations and measurements obtained from a pilot plant. The experimental data were used for extensive validation of models of the pilot plant process, which then served as the basis for the development of commercial-scale CO2 capture plant models. In addition, the experimental results have been used to identify operational limits relevant for the design optimization. The design of chemical and power plant processes typically requires to determine both the optimal operating conditions and the optimal plant configuration. 47

Chapter 3

Optimum Global Design Decision

considering Operational Flexibility Optimum Local Design Decision

considering

Optimum Operating Conditions Convergence?

No

Yes Final Design

Figure 3.1: Two-phase design approach applied for optimization of a pre-combustion capture plant.

This results in a mixed-integer non-linear programming (MINLP) problem, which can be addressed with a rigorous automated method. Bhattacharyya et al. [23] pointed out that these methods are affected by a number of drawbacks in case process simulators are used: i) General algorithms for solving constrained nonlinear or mixed-integer programming problems are most often not yet available in process simulation tools, and ii) obtaining the solution of process simulations whose equipment configuration includes a number of working fluid recycles, which is the case for pre-combustion CO2 capture plants, can be numerically challenging. Due to these limitations, Bhattacharyya et al. suggested and successfully demonstrated a three-phase, top-down, optimization-based design approach, which was adopted for the design problem treated here, with some modifications. The modifications are: i) the operational flexibility of the plant is considered together with the optimization of the global design decisions (first phase)3 , and ii) the optimum operating conditions are evaluated together with the local design decisions4 . The modified step-wise process design optimization method, whereby a global and a local design optimization phases are performed in sequence, is presented in Figure 3.1. In the first phase, the focus is on global design decisions at plant level, targeting energy efficiency or power output, while considering the operational flexibility of the process in terms of operational and environmental limits and/or targets. For a CO2 capture plant the identified decision variables are the extent of CO conversion in the water-gas shift unit, and the percentage of CO2 capture in the removal unit. Both variables are optimized in order to minimize the energy penalty of the capture 3 Following

the three-phase approach, the operational flexibility is assessed as a final step after the global and local design optimization. 4 Following the three-phase approach, the optimization of the local design decisions and the operating conditions are two separate steps.

48


plant considering i) flexible operation in terms of overall carbon capture target, ii) deactivation of the catalyst activity throughout the catalyst life and iii) different operational limits of the steam/CO ratio in the water-gas shift unit. The values of the operational limits were defined based on the knowledge generated during the experimental campaign. In the second phase, local design decisions involving integer variables are evaluated at unit level for optimized operating conditions. The optimum values of the global decision variables determined in the previous phase are considered as additional constraints for the optimization. The first study in this phase focuses on the design of the solvent regeneration and CO2 compression section by comparing a layout with three and four flash stages in terms of auxiliary power consumption and impact on the operating conditions. In this case, the optimal operating conditions of interest are the pressures of the flash drums. The second study evaluates the impact of the solvent temperature on the energy consumption and equipment cost of the removal unit by comparing a design using chillers (chilled-solvent configuration) with a scheme based on water coolers (cooled-solvent configuration) for solvent temperature conditioning. The final design of the process is obtained by iteration on both design phases until the user defined convergence criteria is met, which, for example, can be defined by the tolerance on the difference between the value of the operating condition of the current and previous iteration step assuming local design decisions remain unchanged [23]. This design method features a high level of flexibility as the optimization of the global and local design phase can be performed by means of sensitivity study, rigorous optimization, or case study (if integer variables are involved), depending on the algorithms implemented into the simulation software.

3.3

Process description and process models

The integrated gasification combined cycle is a concept for complex energy conversion systems, which combines solid fuel gasification technology with a highly efficient power generation system, the combined cycle, comprising a gas turbine (GT) and a heat recovery steam generator (HRSG) powering a steam turbine (ST). Figure 3.2 shows a simplified scheme of the IGCC power plant with pre-combustion CO2 capture as considered in this study. Vattenfall chose the process configuration based on maturity of the technology, with the aim of potential near future large-scale implementation. As for the gasification technology, the Shell Coal Gasification Process was selected. The main characteristics are: dry-feed, pressurized, entrained-flow, slagging gasifier with syngas coolers for heat recovery producing high pressure steam. The Buggenum IGCC power station [25], taken into operation in 1994, is based on the same gasification technology. As for the configuration of the pre-combustion CO2 capture process, the sweet water-gas shift process (sulphur content in syngas < 20 ppm [26]) was adopted, followed by physical CO2 absorption. The H2 S removal unit is 49

Chapter 3

Air

Air Separation Unit (ASU)

Raw fuel gas Gasification

Coal

Coal Preparation

Steam (from gas coolers)

Steam Turbine

Syngas Scrubber/ COS Hydrolysis

Steam

Steam

Stack Gas To ASU

HRSG

Electricity Generation

Sulphur Removal

Sulphur Recovery

Water-gas shift

Sulphur (By-product)

CO2 Removal

CO2 Compression

Hydrogen

Gas Turbine

CO2 to Storage

N2 from ASU Air

Figure 3.2: Process flow diagram of an IGCC power station integrating a CO2 removal plant.

therefore located upstream the water-gas shift in order to remove H2 S according to the catalyst requirements. The separated H2 S is thereafter sent to a Claus plant in order to recover elementary sulphur. The solvent for the integrated H2 S and CO2 removal process is dimethylether of polyethylene glycol (DEPEG). The separated CO2 is compressed for sequestration, while the H2 -rich syngas is routed to the gas turbine of the combined cycle. The main objective of this study is the design and performance evaluation of the pre-combustion CO2 capture system, therefore a detailed model of the capture plant was developed, whereby the integration with the IGCC process is represented in a simplified manner by means of power loss factors. These conversion factors account for the impact of the capture plant on the power plant performance in terms of generated electricity and were determined with a separate model of the combined cycle unit [27]. This model is capable of simulating the steady-state operation of a single shaft layout and consists of a H-class Mitsubishi Heavy Industries (MHI) gas turbine, together with a triple pressure level HRSG and steam turbine with reheat. The decrease of electric power output due to the integration with a CO2 capture plant is due to a) the decrease in lower heating value (LHV) of the syngas entering the gas turbine due to the upstream water-gas shift reaction estimated to be 4.66 MWe per percentage in LHV loss, b) the consumption of intermediate pressure (IP) steam for heating/evaporation of the syngas-water mixture in the WGS unit - estimated to be 760 kJe /kg IP steam, and c) the consumption of low pressure (LP) steam in the reboiler of the H2 S stripper - estimated to be 590 kJe /kg LP steam. The losses due to auxiliary power consumption for H2 S and 50


Rectifier Cooler Rectifier

Shifted Syngas

Knock-out Drum Rectifier Pump

Booster Compressor

Reaction Water Pump

Shift Reactor 1 Gas Quench IP Steam

Treated Syngas Heat Ex. 1 Make-up Water Pump

Heat Ex. 2 Steam Heat Ex. 1

Process water


Shift Reactor 2 IP Steam

Steam Heat Ex. 2

Heat Ex. 4

Shift Reactor 3

Figure 3.3: Process flow diagram of the water-gas shift unit.

CO2 removal, as well as CO2 compression can be determined independently of the IGCC power plant operation. The detailed steady-state model of the pre-combustion CO2 capture process and its implementation into a commercial software tool [28] is discussed in the following. The capture plant is divided into two units: i) water-gas shift unit and ii) H2 S and CO2 removal including CO2 compression.

3.3.1

Water-gas shift unit

The process flow diagram of the water-gas shift unit is depicted in Figure 3.3. The treated unshifted syngas from the upstream H2 S absorber is fed to a booster compressor in order to overcome the pressure drop of the capture plant, and is mixed with process water in order to obtain the desired steam/CO ratio, which is the most important process parameter of the WGS unit. Thereafter, the syngaswater mixture is step-wise heated, evaporated and superheated, thus recovering reactor effluent heat and utilizing thermal power supplied by IP steam. The IP steam is subsequently flashed in order to produce LP steam, which is then recycled to the HRSG unit. Carbon monoxide present in the syngas is converted into hydrogen and carbon dioxide in a series of three high-temperature water-gas shift reactors. The moderately exothermic reaction is given by CO(g) + H2 O(g) −→ CO2(g) + H2(g) ,

∆Hr = −41.1 kJ/mol.

(3.1)

The ratio of water vapour to CO needs to be sufficiently high to avoid carbide formation, which causes catalyst deactivation and reduces the catalyst strength. 51

Chapter 3

Water-gas shift unit Reactor temperature approach to equilibrium Reactor inlet temperature Reactor outlet pressure Operational constraints Reactor outlet temperature Molar steam/CO ratio at reactor inlet H2 S and CO2 removal unit, CO2 compression Water content in rich solvent Chilled-solvent temperature Pinch-point temperature difference in reboiler Pressure of CO2 product gas Product specifications H2 S content in acid gas stream Total sulphur content in CO2 product gas a

10 K 340 ◦ C 38.3 / 37.9 / 37 bar ≤ 520 ◦ C ≥ 2.65 4 mol % 4 ◦C 10 K 110 bar ≥ 25 mol % ≤ 10 ppma

Requirement set by Vattenfall Table 3.1: Modelling assumptions and constraints.

In addition, excess of steam is required in order to enhance the CO conversion by forcing the equilibrium to the right, and in order to limit the adiabatic temperature increase in the reactor. The partial bypass around the first reactor (gas quench) allows to independently operate the first and second reactor at minimum steam/CO ratio, which consequently lowers the total steam consumption. A scheme with one or multiple syngas split streams is advantageous in comparison to a conventional configuration without syngas split, in which the steam/CO ratio can only be manipulated at the inlet of the first reactor, and then increases for each subsequent reactor [22]. For a similar layout of the WGS unit, but with direct steam injection, Martelli et al. [21] demonstrated that optimum performance is achieved by employing only one syngas split, just as in the design presented here. Martelli et al. considered a configuration of three as well as four reactors. The water-gas shift reactors are modelled by assuming that thermodynamic equilibrium, which is calculated by Gibbs energy minimization, is not completely reached. A temperature approach to equilibrium of 10 K is therefore imposed. The reactors’ inlet temperature are maintained at a constant value, which is 340 ◦ C for operation at the reference state. In the considered configuration, the reactor effluent thermal energy is recovered within the WGS unit. Another viable option would be thermal integration with the HRSG unit, as considered by other researchers [21, 23]. The excess process water is recovered from the shifted syngas through condensation, and is then recycled. The modelling assumptions and constraints are summarized in Table 3.1. 52


CO2 absorber Shifted Syngas

Lean Solvent Chiller

HP Lean Solvent Pump

Heat Ex. 6 Product CO22

H22-rich Syngas

CO2 Cooler

CO2 Cooler

Recycle Recycle Cooler Compressor

LP compressor Flash Gas Compressor

HP Flash Drum MP Flash Drum

Syngas Liquids Untreated Syngas Treated Syngas Heat Ex. 7

Semi-Lean Solvent Chiller

LP Flash Drum

Gas Cooler

Semi-Lean Solvent Pump

Reflux Drum MP/HP compressor Reflux Condenser H2S Stripper

Acid Gas Make-up Water

KO Drum Reflux Pump Purge Water

H2S absorber Gas Cooler KO Drum Heat Ex. 5

Rich Solvent Flash Drum

Reboiler

Solvent Make-Up

Syngas Liquids LP Lean Solvent Pump

Figure 3.4: Process flow diagram of the H2 S and CO2 removal including CO2 compression.

3.3.2

H2 S and CO2 removal unit

The configuration of the H2 S and CO2 removal unit, as depicted in Figure 3.4, is tailored to a sweet water-gas shift process. The unit integrates the process of selective removal of H2 S from the untreated syngas and of CO2 from the shifted syngas using DEPEG as physical solvent. The untreated syngas from the upstream gasifier (see Table 3.2 for gas composition and operating conditions) is first sent to the H2 S absorber, where most of the H2 S is removed by the semi-lean solvent, and is then routed to the WGS unit. The rich solvent from the bottom of the H2 S absorber is fed to a heat exchanger, which raises its temperature by recovering thermal energy from the lean solvent leaving the H2 S stripper bottom. Thereafter, the rich solvent is routed to a flash drum. The pressure of the drum is manipulated to maintain the required concentration of H2 S in the acid gas stream fed to the Claus unit. The vapour from the flash drum is recompressed and recycled to the H2 S absorber, while the liquid is fed to the H2 S stripper. The stripper is a distillation column operated at low pressure and with a reboiler at the bottom heated by LP steam from the HRSG unit (5.5 bar and 156 ◦ C). A minimum pinch point of 10 K is assumed for the reboiler, such that it limits the maximum reboiler temperature to 146 ◦ C. Water condensate recovered from the overhead vapour stream, which primarily contains H2 S and CO2 , is recycled to the top of the H2 S stripper to provide reflux. The acid gas stream is sent to the Claus unit. The stripped solvent leaving the H2 S stripper bottom is mixed with 53

Chapter 3

Temperature [◦ C] Pressure [bar] Mass flow rate [kg/s] Composition [mol %] H2 O CO2 H2 S COS H2 CO N2 Ar

40 36 71.22 0.26 5.34 0.27 5 ppm 29.02 56.51 8.51 0.09

Table 3.2: Composition and conditions of untreated syngas.

make-up solvent and is sent to the top of the CO2 absorber. CO2 is removed from the shifted syngas in the CO2 absorber column resulting in a H2 -rich gas, which is fed to the gas turbine of the combined cycle. The CO2 in the loaded solvent at the bottom of the CO2 absorber is recovered by threestage depressurization. The vapour from the first flash drum, also referred to as H2 recovery drum, is compressed and recycled to the top of the absorber column to recover combustible components dissolved in the solvent, primarily H2 . Two objectives should be satisfied with the pressure of this drum. First, the pressure should be adjusted such that the required impurity limit of H2 in the CO2 product stream is met. Current recommendations set the limit at 4 vol. % [29] motivated by the loss of CO2 transport and storage capacity caused by non-condensable gases such as N2 , H2 and Ar in the CO2 product stream. Moreover, the presence of non-condensables in the CO2 product might require higher operating pressures in order to keep the CO2 in the dense phase, resulting in additional compression work. Second, the operating pressure is the result of an energy optimization within the removal unit, aimed at balancing the energy content of the recovered gases against the power requirement of the recycle compressor. This results, for example, in high recovery rate at low pressure, but high power requirement for recompression. The results of this optimization problem are presented in Section 3.5. The semi-lean solvent from the LP flash drum is pressurized and split into a fraction sent to the H2 S absorber and a fraction which is chilled and returned to the CO2 absorber at 2/3 of the column height. The solvent temperature at the outlet of the semi-lean and lean solvent chillers are maintained at 4 ◦ C. A coefficient of performance (COP) of 3.5 is assumed for the chillers, considering the current typical performance of the conventional vapour-compression machine. The CO2 released from the flash drums is compressed in a single-stage LP and a four-stage medium pressure (MP) / high pressure (HP) compressor with interstage cooling to a pressure of 110 bar. Upstream the last compression stage the CO2 is dehydrated to achieve water contents below the recommended target [29] 54


which limits the occurrence of corrosion and hydrate formation. The dehydration unit is represented by a simplified model. The columns of the removal unit are modelled following the equilibrium-based approach, which is a justified assumption for an initial design. In comparison to a rate-based model the absorption efficiency is overestimated for liquid-to-gas ratios smaller than the design value and underestimated for values larger than the design ratio. The thermophysical properties of the two-phase, multi-component syngas-water and syngas-DEPEG mixtures are calculated with the perturbed chain - statistical associating fluid theory (PC-SAFT) equation of state (EoS) [30] due to its success in predicting vapour-liquid equilibria of complex fluids and mixtures for a broad range of conditions. For simplicity, the solvent DEPEG, which is a blend of glymes, is treated as pseudo pure fluid in the thermodynamic model. The molecular parameters of the solvent have been obtained by fitting the EoS to published vapour pressure and liquid density data [31]. The interaction parameters for binary mixtures involving DEPEG have been validated against experimental data provided by the solvent vendor. The accuracy of the resulting thermodynamic model has proved to be suitable for engineering purposes.

3.4

Pilot plant experiments

Various experimental tests were performed at the CO2 capture pilot plant realized at the IGCC power station in Buggenum in order to verify the performance of the selected technology, to obtain operational experience and to record measurement data for model validation. The models of the commercial-scale CO2 capture plant discussed in this chapter are all derived from pilot plant models, which have been extensively validated against experimental data [32]. In the following, two experiments focusing on the performance of the water-gas shift unit are explained in more detail. Their results and the generated knowledge were used to define the capture plant optimization presented in Section 3.5. The moderately exothermic WGS reaction is limited by the thermodynamic equilibrium, which implies that the CO conversion is dependent on the reaction temperature. Low temperatures favour the formation of reaction products, hence the CO conversion [33]. However, higher operating temperatures are required in order to preserve the activity of the catalyst. Typically, due to catalyst deactivation throughout its lifetime, the inlet temperature needs to be increased in order to maintain the reaction front within the catalyst bed. The vendor of the high-temperature shift (HTS) FeCr-based catalyst suggests that the operating temperature shall be maintained within the 325 − 500 ◦ C operating range [34]. Furthermore, the equilibrium can be shifted towards the right by increasing the steam concentration, or by removing reaction products from the reaction mixture. Accordingly, the operational variable is the steam/CO ratio at the reactor inlet. In the first set of experiments, measurements of the axial reactor temperature profiles at reference state operation throughout an operational period of 5840 hours 55

Chapter 3

370

Inlet temperature [◦ C]

360 350 340 330 320 310 300 0

Reactor Reactor Reactor 0.2 0.4 0.6 0.8 1 1.2 Relative operational time [-]

1 2 3 1.4

Figure 3.5: Required minimum reactor feed temperature as function of time.

were used to quantify the rate of deactivation of the catalyst activity. Extrapolation of the quantified stability was then used to predict the catalyst lifetime and the required adjustment of the operating conditions to reach the design specifications throughout its lifetime, namely the reactor inlet temperature, in order to compensate for catalyst deactivation. The aim of such an investigation is therefore to evaluate the changes in reactor performance during lifetime within the framework of the system process analysis. An adiabatic, heterogeneous, one-dimensional reactor model [32] supplemented with validated equations to predict changes in catalyst activity was used to determine the minimum reactor feed temperatures as a function of time (see Figure 3.5). It is assumed that at each state the equilibrium is approached by 10 K at the end of the catalyst bed. For confidentiality reasons the operational time is plotted in non-dimensional terms, whereby the reference time was arbitrarily chosen. Considering reactor 1 and 2, the chart of Figure 3.5 shows that the inlet temperatures need to be steadily increased with time due to catalyst deactivation in order to ensure stable and high-rate conversion of CO. Frequent adjustments according to the predictions ensure the possibility of maintaining the operation of the WGS unit at minimum energy consumption, as the full capacity of each catalyst is used by constantly keeping the reaction front just at the end of the catalyst bed. From a practical point of view, less frequent temperature adjustments might be preferable at the expense of operating the unit at higher temperatures than required for some periods of time, resulting in higher energy consumption. It worth pointing out that the deactivation rate for reactor 1 is larger than for reactor 2 (see Figure 3.5), due to the higher steam content of the syngas entering reactor 1. The difference in deactivation rate between both reactors is addressed in more detail in the publication of van Dijk et al. [32]. It can therefore be inferred that the optimal design procedure must take into account a) the general impact of catalyst deactivation on the performance of the capture plant, depending on the optimal operating conditions and b) if frequent 56


adjustments of the inlet temperatures result in significant reduction of energy consumption. In the staged reactor configuration it is assumed that reactor 3 does not show any significant deactivation, hence a constant inlet temperature can be maintained throughout its lifetime. The main reason is that reactor 3 is operated at the mildest operating conditions in comparison to the other reactors: lowest steam and CO content of the syngas, lowest adiabatic temperature rise. The predictions in Figure 3.5 are based on reference state operation (defined syngas composition and flow), hence some safety margin should be considered in order to account for disturbances in operating conditions. The second set of experiments was targeted to the investigation of the catalyst performance and its stability during reduced steam/CO ratios. Operation at low steam/CO ratio implies that less steam is required in the WGS unit, which might significantly contribute to the reduction of the energy penalty, but can result at the same time in iron-carbide (FeC) formation, the so-called carbiding of the catalyst. Severe carbiding can lead to permanent loss of catalyst activity and selectivity, and even to physical damage of the catalyst pellets. Fe-carbide is the basis for Fischer-Tropsch catalysts, producing methane and higher hydrocarbons in a very exothermal reaction. The carbiding tendency increases at higher total pressure, reduced steam and larger CO content of the syngas feed. The catalyst vendor suggested a safe molar steam/CO ratio of 2.65 for the employed HTS catalyst at the pilot plant operating conditions. During the experimental campaign reduced ratios down to 1.5 were tested, typically by 100 − 150 hours tests. It was observed that the catalyst, which was operated already for more than 5000 hours before the reduced steam/CO ratio testing, was stable at the tested conditions. No indications for progressive carbiding were observed, and during operation following the ratio testing, the catalyst displayed unaffected catalyst activity. The system optimization analysis should therefore also provide insights about the influence of the steam/CO ratio reduction (down to 1.5) on energy efficiency. In addition to the experiments highlighted above, various parametric tests were performed to evaluate the impact on the CO2 removal rate considering process variables such as water content in the solvent, solvent temperature, absorber pressure, liquid-to-gas ratio of the column and flash drum pressure. The measurements were used to validate the process and the fluid thermodynamics models. For reasons of conciseness these experiments are not described extensively here and will be published in an upcoming paper. To conclude, based on the results of the experimental tests, the following aspects of the capture plant performance should be addressed by means of process optimization: a) Investigation of the impact of catalyst deactivation on the capture plant performance and optimal operating conditions; b) analysis of energy reduction potential for operation at reduced steam/CO ratio and the resulting impact on CO conversion, and overall carbon capture rate. 57

Chapter 3

3.5 3.5.1

Result analysis and discussion Global design decision

Global design decisions can be identified by analysing the interaction of different units at plant level and the impact on the overall plant performance, e.g., in terms of efficiency or net power output [23]. The design point of the capture plant is mainly determined by the nominal overall carbon capture rate,5 whereby the main decision variables are the extent of CO conversion in the water-gas shift unit and the extent of CO2 capture in the removal unit.6 There is a trade-off between both variables in order to achieve a given rate of carbon capture, which can be obtained by a low CO conversion and a high CO2 removal or vice versa. The optimum combination of both variables can be determined by optimization targeting minimum energy consumption. Currently, most studies assume an overall capture rate of 90 %, however it is still unclear which CO2 capture rate is required for future large-scale implementations from an environmental point of view, or which capture rate would allow for maximum energy/economic efficiency. Initial process design studies should therefore evaluate the optimal design considering flexible operation of the capture plant in terms of the overall carbon capture rate. An assessment of the operational flexibility should also cover expected variations in operational or environmental limits and uncertainties in process parameters. One important operational limit is the steam/CO ratio in the WGS unit. A study on this aspect is therefore presented, discussing the influence of different limits of the steam/CO ratio on the performance of the capture plant. Degradation occurring throughout the lifetime of the process equipment will influence the process performance and possibly the optimal operating conditions. In case of the capture plant, performance losses are expected due to deactivation of the catalyst activity over time. The evaluation of the influence of the catalyst deactivation on the optimum operating conditions thus enables to estimate the operating parameters for optimal performance. In the following the optimum of the global decision variables and the energy consumption of the capture plant are presented for three test cases, namely, a) flexible operation in terms of carbon capture, b) different states of catalyst activity, and c) different limits of the steam/CO ratio. The resulting optimization problem is formulated as

5 The

overall carbon capture rate is defined as mole flow rate of captured carbon for sequestration divided by mole flow rate of carbon in the untreated syngas. 6 The extent of CO capture in the removal unit is defined as mole flow rate of captured CO 2 2 for sequestration divided by mole flow rate of CO2 in the shifted syngas

58


min

fenergy,

s.t.

g(·) = 0,

x

capture plant

= ELHV

loss + EIP steam + ELP steam + Eauxiliary + ECO2 compression ,

h(·) ≤ 0, (3.2) where x corresponds the set of decision variables, fenergy represents the objective function, g(·) is the set of algebraic equations that describes the system and h(·) is the set of inequality constraints defined by the operating conditions and product specifications, which are summarized in Table 3.1. For the optimization five decision variables were identified, namely, i) the semi-lean solvent flow rate of the CO2 absorber, ii) the lean solvent flow rate, iii) the pressure of the rich solvent flash drum, iv) the flow rate of process water and v) the temperature of the feed splitting vessel. The two latter variables are associated with the WGS unit, while the remaining three with the removal unit. With respect to the inequality constraints, it is often observed during optimization that the optimum is located at the intersection of some inequality constraints. This means that the degree of freedom (i.e., number of variables minus the number of equality constraints) of the optimization problem is reduced by the number of activated inequality constraints. In case of the optimization of the capture process the following inequality constraints are activated (i.e., treated as equality constraints): i) H2 S concentration in the acid gas stream of 25 mol % and ii) total sulphur content in the CO2 product stream of 10 ppm. This results in an optimization problem with two degrees of freedom, considering that the desired rate of overall carbon capture is also treated as equality constraint. The degree of freedom might further reduce during the optimization if other inequality constraints, e.g., the reactor outlet temperature, are approached. The optimization was carried out using the sequential quadratic programming algorithm of Lang and Biegler [35] available within the adopted software tool. Environmental target: Carbon capture rate The results from the first optimization study are presented in Figure 3.6, which shows the optimum combination of CO conversion and CO2 removal as well as energy consumption of the capture plant as function of the carbon capture rate. Overall capture rates of 91 % can be achieved with the considered plant configuration. Operation beyond 91 % capture is unfavourable as further increase of the overall steam/CO ratio by adding process water in the WGS unit has marginal effect on the CO conversion, and the increase in solvent flow rate leads to minimal enhancement of the CO2 removal. As the percentage of carbon capture is decreased by adapting the appropriate process variables it appears beneficial to decrease the CO conversion to a larger extent than the CO2 removal, see Figure 3.6(a), due to the fact that reduction in IP steam utilized in the WGS unit has a larger impact on the total power output than the reduction in solvent flow rate in the CO2 removal unit. 59

Chapter 3

95

3

94

2

93

1

82

84 86 88 Carbon capture [%]

(a)

90

0 92

130

1500

Total energy consumption Energy / m ˙ carbon,captured

1450

120

1400

110

1350

100

1300

90

1250

80 80

82


90

Energy / m ˙ carbon,captured [MJ/t]

4

Total energy consumption [MW]

96

92 80

140

5

CO conversion CO2 removal Reactor 2 steam/CO ratio

Steam/CO ratio [mol/mol]

CO conversion, CO2 removal [%]

97

1200 92

(b)

Figure 3.6: a) Optimum CO conversion, CO2 removal and reactor 2 steam/CO ratio as a function of the carbon capture rate. b) Total and specific energy consumption as a function of the carbon capture rate.

At 87 % carbon capture the overall molar steam/CO ratio in the WGS unit is lowered such that the allowed minimum steam/CO ratio of 2.65 (recommended value by the catalyst vendor) is reached at the inlet of reactor 2, see Figure 3.6(a). According to standard industrial practice, further reduction in overall steam/CO ratio is not possible without compromising the safe operation of the reactor 2 catalyst. For capture rates below 87 % it is therefore beneficial to bypass a part of the unshifted syngas leaving the H2 S absorber, and directly mix it with the treated syngas from the top of the CO2 absorber. The optimum operating conditions in terms of CO conversion and CO2 removal as well as overall and reactor-specific steam/CO ratios remain constant for this operational region, see Figure 3.6(a). The CO conversion and CO2 removal are calculated based on the actual amount of syngas entering the water-gas shift and CO2 removal unit, hence the syngas bypass is not considered. Figure 3.6(b) shows the total7 and specific energy consumption8 per amount of carbon captured. As the percentage of carbon capture is decreased the total energy consumption of the capture plant decreases as expected. The same trend is observed for the specific energy consumption up to the minimum at 87 % carbon capture. For lower capture rates, the specific energy consumption remains almost constant, due to the bypass of syngas and the rather unchanged operating conditions in the water-gas shift and removal unit. The reduction of the total carbon capture rate from 91 % to 87 % corresponds to the decrease of the energy 7 The total energy consumption of the CO capture plant is the sum of i) the decrease in LHV 2 value of the syngas, ii) the consumption of IP steam in the WGS unit, iii) the consumption of LP steam in the reboiler of the H2 S stripper, iv) the auxiliary power consumption of the entire capture plant and v) the power consumption related to CO2 compression. 8 The specific energy consumption is defined as the total energy consumption of the capture plant divided by the amount of carbon captured for sequestration.

60


consumption of approximately 20 %.

Lifetime performance: Reactor deactivation The second design optimization problem presented here is aimed at investigating the influence of catalyst deactivation on the performance of the capture plant. Two cases are compared: 1) operation with fresh catalyst - start of run (SOR), and 2) operation with aged catalyst - end of run (EOR). These cases are modelled assuming 315 ◦ C for the SOR case and 365 ◦ C for the EOR case as temperature at the inlet of both reactor 1 and 2. The inlet temperature of reactor 3 is maintained at 340 ◦ C. These temperature values result from the analysis of the catalyst deactivation rate performed with the detailed reactor model validated with experimental measurements, as described in Section 3.4. Figure 3.7 presents the optimization results covering a range of carbon capture from 85 % to 90 %. For capture rates above 87.5 % the specific energy consumption is approximately 1 % higher for operation at the EOR condition in comparison to that at the SOR condition. Higher inlet temperatures, which are due to the loss in catalyst activity, are less favourable with respect to the thermodynamic equilibrium of the water-gas shift reaction, thus the CO conversion decreases. To maintain a given extent of carbon capture this effect is mainly compensated by increasing the overall steam/CO ratio in the WGS unit, as shown in Figure 3.7(b). This leads as a consequence to the observed increase in energy consumption. The optimum combination of CO conversion and CO2 removal for the SOR and the EOR catalyst conditions are very similar for the different carbon capture rates, thus they are not reported for the sake of conciseness. For capture rates below 86.5 % the situation is reversed, namely, a lower specific energy consumption is estimated for EOR operation. Due to catalyst deactivation at the EOR condition, the CO conversion and therewith the overall carbon capture is decreased in comparison to the SOR condition, assuming a fixed overall steam/CO ratio, see Figure 3.7(b): steam/CO ratio = 1.41, carbon capture = 87.5 % (SOR) and 87.2 % (EOR). This is one reason why the minimum in specific energy consumption for EOR operation is shifted toward a lower carbon capture rate. If the same overall steam/CO ratio is assumed for both the SOR and the EOR conditions, then the steam/CO ratios at the inlet of the reactors are higher in case of EOR, as less CO is converted during the reaction. As a consequence, in case of EOR operation it is possible to further lower the energy consumption by reducing the overall steam/CO ratio, whereas for SOR the limit of the steam/CO ratio is reached already at reactor 2. To conclude, the influence of catalyst deactivation on the performance of the capture plant is rather small. Frequent adjustments of the reactors’ inlet temperature purely for energy efficiency reasons are not advisable. Changes of the inlet temperatures throughout the catalyst lifetime should rather focus on safe and stable reactor operation. 61

Chapter 3

Reactor 1 & 2: Tinlet = 315 ◦ C (SOR) Reactor 1 & 2: Tinlet = 365 ◦ C (EOR)

1400

1350

1300

1250 85

86


90

(a)

1.9 Overall steam/CO ratio [kg/kg]


1450

Reactor 1 & 2: Tinlet = 315 ◦ C (SOR) Reactor 1 & 2: Tinlet = 365 ◦ C (EOR)

1.8 1.7 1.6 1.5 1.4 1.3 85

86

87 88 Carbon capture [%]

89

90

(b)

Figure 3.7: Specific energy consumption (a) and optimum of overall steam/CO ratio (b) as a function of the carbon capture rate for SOR and EOR operation conditions of the catalyst.

Operational limit: Steam/CO ratio The third optimization study focuses on the comparison of the capture plant performance for three different limits of steam/CO ratio (2.65, 2.0, 1.5) at the inlet of the reactors, whereby case 1 corresponds to the value recommended by the vendor, while case 2 and 3 were successfully tested during the experimental campaign (see Section 3.4). Figure 3.8 visualizes the results of the analysis. For capture rates above 87.5 %, the optimum steam/CO ratios at the inlet of the reactors are always greater than 2.65 (limit of case 1). The limit of the steam/CO ratio has therefore no influence and the resulting optimized process variables are the same. At 87.5 % the value of 2.65 is reached at the inlet of reactor 2, whereby the minimum in specific energy consumption is found for case 1. In this case, at lower capture rates the syngas bypass is used, therefore the optimal operating condition in terms of overall and reactor inlet steam/CO ratios, as well as CO conversion and CO2 removal, remains the same. For case 2, instead of using the syngas bypass, it is more energy efficient to further lower the overall steam/CO ratio up to the carbon capture rate of 84 %, where the limit of 2.0 is reached at the inlet of reactor 2. For case 3, the even lower steam/CO limit allows a further reduction of the specific energy consumption, which reaches its minimum at 78 % carbon capture. To conclude, for higher carbon capture rates (above 87.5 % for the presented examples) the allowed minimum steam/CO ratio to avoid carbide formation has no influence on the optimal operating conditions. Considering a wide operational range in terms of carbon capture, then the minimum in specific energy consump62


4

S/COmin = 2.65 mol/mol S/COmin = 2.0 mol/mol S/COmin = 1.5 mol/mol

1350

Reactor 2 steam/CO [mol/mol]


1400

1300 1250 1200 1150 1100 76

78

80 82 84 86 Carbon capture [%]

88

3

2

1

0 76

90


78

80 82 84 86 Carbon capture [%]

(a) 98

90

88

90

(b) 95


94


94 CO2 removal [%]

CO conversion [%]

96

88

92 90 88 86

93

92

84 82 76

78

80


(c)

88

90

91 76

78

80


(d)

Figure 3.8: Optimization results for different limits of steam/CO ratio. a) Specific energy consumption of the capture plant. b) Reactor 2 steam/CO ratio. c) Optimum CO conversion. d) Optimum CO2 removal.

63

Chapter 3

tion of the capture plant is achieved when the steam/CO ratio reaches the defined limit, which is typically first approached at the inlet of reactor 2. For a minimum steam/CO ratio of 1.5 the specific energy consumption can be reduced by 10 % in comparison to a minimum ratio of 2.65, which comes at the expense that the optimal carbon capture rate lowers from 87.5 to 78 %. The optimum of the corresponding CO conversion changes from 93 to 83 %, whereas the optimum of the CO2 removal only shifts from approximately 93 to 92 %.

3.5.2

Local design decision

Local design decisions are identified at unit level by analysing the interaction of different sections, and their impact on the plant performance [23]. Two studies are presented here and they concern the design of the H2 S and CO2 removal unit. The optimization targeting minimum energy consumption is performed by fixing the carbon capture rate at the value of 84 %, which is the optimum capture rate assuming a minimum steam/CO ratio of 2.0 (see Subsection 3.5.1). The corresponding optimum values of CO conversion and CO2 removal are considered as additional constraints. The first study focuses on the design of the solvent regeneration and CO2 compression section. The solvent regeneration in the CO2 removal unit is performed stage-wise in order to reduce compression power by routing part of the released CO2 gas at elevated pressure to the CO2 compressors. The designs with three and four flash stages are compared considering the auxiliary power consumption. The second study compares two configurations with different solvent temperatures, whereby one design uses chillers (chilled-solvent configuration) and the other water coolers (cooled-solvent configuration) for solvent refrigeration or cooling. The optimization at unit level is carried out by means of case studies, as integer variables are involved and rigorous algorithms for solving such a mixed integer nonlinear programming problem were not available. Each case is optimized in terms of its operating conditions represented by the following continuous variables: i) semi-lean solvent flow rate of the CO2 absorber, ii) lean solvent flow rate, iii) pressure in the rich solvent flash drum, iv) MP flash pressure, and v) LP flash pressure. The optimization problem is formulated as min

fenergy,

s.t.

g(·) = 0,

x

removal unit

= ELP

steam + Eauxiliary + ECO2 compression ,

(3.3)

where x corresponds the set of decision variables (integer and continuous), fenergy represents the objective function and g(·) is the set of algebraic equations that describes the system. In comparison to the optimization of the global design decisions here no inequality constraints are present. The reason is that the CO conversion and therewith the operational constraints of the WGS unit are treated as equality constraints. Furthermore, the product specifications of the removal unit are treated as activated inequality constraints. 64

5

32.5 4

32 31.5

3

31

2

30.5 3 stage regeneration 1 LP flash pressure MP flash pressure

30 29.5 10

12

14

16 18 20 22 24 26 H2 recovery pressure [bar]

28

0 30

7

1.6 LHV loss Recycle compressor 1.4 H2 content in CO2 product 1.2

6 5

1

4

0.8 3

0.6

2

0.4

1 0 10

H2 content [mol%]

6

33

Energy consumption, LHV loss [MW]

33.5

Pressure [bar]

Energy consumption removal unit [MW]


0.2 12

(a)

14

16 18 20 22 24 26 H2 recovery pressure [bar]

28

0 30

(b)

Figure 3.9: a) Energy consumption of the removal unit (including CO2 compression) and optimum of the LP and MP pressure as a function of the H2 recovery pressure. b) LHV loss, recycle compressor power requirement and H2 content as a function of the H2 recovery pressure.

Solvent regeneration and CO2 compression The number of flash drums in the solvent regeneration section and their pressure levels influence both the power consumption for CO2 compression and the energy requirement for CO2 removal. For example, if the LP pressure is lowered, then the stripping of CO2 is enhanced, which decreases the solvent flow rate and the cost of solvent refrigeration/cooling, if the CO2 rate is assumed constant. Conversely, if the LP pressure is low, the compression power of the LP compressor becomes larger. The optimization of the operating conditions and evaluation of different design options of the CO2 removal and compression section is therefore important if the design objective is the minimization of the energy consumption of the plant. In the following, firstly, the influence of the pressure levels on the energy consumption is analysed for a solvent regeneration section with three stages, secondly, these results are compared to a design with four flash drums. Figure 3.9(a) shows the energy consumption of the removal unit together with the optimized values of the LP and MP flash pressure as a function of the H2 recovery pressure. The minimum energy consumption is achieved at a H2 recovery pressure of 15 bar. The corresponding values of the LP and MP pressure are 1.3 bar and 4.9 bar, respectively. Considering that the slope of the energy curve is steeper at the left of the minimum, it would be advisable to operate the plant at slightly higher pressure. In this case process disturbances might have a smaller impact on energy consumption. The optimum values of the MP and LP pressure remain almost unchanged for different H2 recovery pressures. Figure 3.9(b) explains in more detailed the influence of the pressure in the H2 recovery drum on the energy consumption. At low recovery pressures, most of the combustible H2 is recycled back to the 65

Chapter 3

Energy consumption [MW]

34

33

32

31

30 3 stage regeneration 4 stage regeneration 29 10 12 14 16 18 20 22 24 26 28 30 H2 recovery pressure [bar]

Figure 3.10: Energy consumption of the removal unit (including CO2 compression) for three- and four-stage solvent regeneration as a function of the H2 recovery pressure.

CO2 absorber, therefore the LHV loss related to H2 in the CO2 product is low. However, the power required for the compression of the recycled H2 is high due to the large pressure difference between the absorber and the recovery drum. If the H2 recovery pressure is increased, then the recycle compression power decreases while the LHV loss is larger. The optimal trade-off between both losses is achieved at a H2 recovery pressure of 15 bar, leading to the minimal value of the energy consumption. The choice of the H2 recovery pressure value and therewith the H2 content in the CO2 product gas is also motivated by technical and economical considerations related to CO2 transport and storage. Currently, no strict targets on H2 are established and the calculated values depicted in Figure 3.9(b) are well below the recommended value of 4 vol. % [29]. The same optimization was performed with four flash stages, and the final results in terms of energy consumption of the removal unit are compared to the three stage design in Figure 3.10. The optimum H2 recovery pressures are very similar, that is 15 and 15.5 bar for the three stage and four stage design, respectively. However, the energy requirement is about 1.3 MW lower if four drums are used for solvent regeneration. This is explained by the fact that a larger part of the released CO2 is routed at elevated pressure to the compression section, thus making compression less energy demanding. A final design decision depends on the techno-economic evaluation in order to assess if the costs of the different and additional equipment required for four-stage regeneration is justified by the energy saving. Such evaluation is outside the scope of this work. Furthermore, it is assumed that for the three-stage regeneration a single-stage LP compressor and a four-stage single-shaft MP/HP compressor can be used, whereas for the four-stage regeneration, a single-stage LP, two-stage MP and two-stage HP compressor is required. The compression ratio in each stage of the individual compressors was kept constant during the optimization. 66


Solvent temperature The absorption of CO2 can be enhanced by adjusting the following operating conditions: a) lowering the temperature of the solvent, b) increasing the solvent circulation rate in order to raise the liquid-to-gas ratio, and c) lowering the pressure of the flash drums, in particular the LP pressure in order to enhance the stripping of the solvent. The options b) and c) are considered and discussed in the previous section. Here the impact of the solvent temperature on the performance of the removal unit and its influence on the optimum operating conditions is treated. Two cases are compared, chilled-solvent configuration and cooled-solvent design. The first case considers a design whereby the solvent is refrigerated down to a temperature of 4 ◦ C by compression chillers. The low solvent temperature allows to decrease the solvent flow rate, hence the power associated with solvent circulation. Less stripping is required, which reduces the cost of CO2 compression. However, this comes at the expense of costs due to energy consumption for solvent chilling. The second design is based on water-cooling of the solvent, which allows to keep the temperature at 35 ◦ C. Solvent cooling requires considerably less power than solvent refrigeration. However, the resulting decrease in CO2 removal requires to increase the liquid-to-gas ratio and to enhance solvent stripping associated with an increase in power for solvent circulation and CO2 compression. The performance of the removal unit is not only influenced by the operating conditions discussed above but also by the height of the columns, which is proportional to the number equilibrium trays defined in the models, e.g., CO2 absorption is enhanced by increasing the column height. A fair comparison of the two designs with different solvent temperatures must consider the impact of the column sizing on the performance, and thus in this case the optimization must include this techno-economic aspect. Moreover, also the sizing of all other components is expected to depend on each of the two considered configurations. For example, the chilled-solvent configuration requires smaller equipment due to lower solvent flow rates, but chillers are more expensive than water coolers. This optimization study therefore includes a preliminary cost evaluation. The number of equilibrium stages, which represents the column height, is a discrete variable in the model. For that reason, the optimization was performed by means of a case study. Multiple design cases, i.e., designs with different values of the stage number, were evaluated for the two configurations. Each design case was first optimized targeting energy consumption, and thereafter the associated equipment cost was evaluated for each solution. A tool for preliminary costing, available with the adopted process simulator, was used to this end [36]. Figure 3.11 presents the comparison of the best solutions of the chilled and cooled-solvent design obtained from this preliminary techno-economic analysis. The comparison shows that the cooled-solvent design is advantageous both in terms of energy consumption (about 12 % lower) and equipment cost (about 15 % 67

Chapter 3

30

Equipment cost [m $]

29 28 27 26 25 24 23 22 30

Chilled-solvent design Cooled-solvent design 31 32 33 34 35 36 Energy consumption removal unit [MW]

37

Figure 3.11: Comparison of chilled- and cooled-solvent configuration in terms of equipment cost and energy consumption of the removal unit (including CO2 compression).

lower for the H2 S and CO2 removal unit). The chilled-solvent design features smaller values of the optimized flow rates and higher pressure levels. It shall be highlighted that these results do not represent the Pareto front of the optimal design solutions, but rather an indication suitable for a preliminary comparison of the two design configurations. The set of Pareto lines corresponding to the optimal design solutions can be obtained by means of a multi-objective optimization targeting energy consumption and capital cost. Alternatively, the design optimization problem can also be reduced to a single-objective optimization whose objective variable is the net present value of the plant. However, both optimizations involve integer variables, the number of trays in the columns, thus a MINLP solver is required. Performing such rigorous but complex optimization was not deemed reasonable considering the scope of this work.

3.6

Conclusions

This chapter discusses the use of steady-state modelling and simulation for the design optimization of a large-scale pre-combustion CO2 capture plant utilizing validated process models of a capture pilot plant realized at the Buggenum IGCC power station as basis for development. The CO2 capture process has been optimized targeting reduction in energy consumption and considering flexible operation in terms of overall carbon capture rate, different operational limits of steam/CO ratio and deactivation of catalyst activity throughout the catalyst life. For a minimum molar steam/CO ratio of 1.5 the specific energy consumption of the capture plant can be reduced by 10 % in comparison to a molar steam/CO ratio of 2.65, which is the value recommended by the vendor. This comes at the cost that the optimal carbon capture rate is reduced from 87.5 % to 78 %. Moreover, the design of the CO2 removal unit has been evaluated by comparing the optimized 68


design with three- and four-stage solvent regeneration, and also by analysing two configurations with different solvent temperature. The plant configuration comprising compression chillers is advantageous both in terms of energy efficiency and equipment cost. The following conclusions can be drawn from this study: • Pilot plant experiments are an important element during the development and large-scale implementation of CO2 capture technology. Experimental data is essential in order to perform model validation with the aim of improving the reliability and the accuracy of process simulation results used during the early process design phase. Moreover, acquired knowledge and process understanding resulting from experiments allow to identify more easily the most relevant process parameters and operational limits required as input for process analysis and optimization. • Operational flexibility in terms of environmental and operational limits, and/or targets must be correctly address in the preliminary design phase of a pre-combustion CO2 capture plant. Variations of these values might have a large impact on the plant configuration and on the optimal operating conditions. • The two-step design optimization method, comprising of a global and a local design phase, allows to analyse and optimize complex chemical and power plant processes by dividing the design problem into manageable optimization sub-problems. The final plant design is obtained by iteration. Each design optimization problem can be tackled by sensitivity study, rigorous optimization or case study, depending on the availability of suitable algorithms in the adopted process simulation tool. As always, the final design optimization must include the economic evaluation of the equipment, which is outside the scope of this work. If a reliable costing tool can be coupled to the process simulator, the automated global design problem can be multi-objective and target the maximum efficiency and minimum investment, or single objective and minimize the net present value of the plant. This work provides the necessary understanding of the complex aspects that is needed if such a complex numerical problem has to be solved.

69

Chapter 3

Nomenclature E x

= =

Energy consumption, W Design decision variable

= = = = = = = = = = = = = = = = = = = =

Coefficient of performance Dimethylether of polyethylene glycol End of run Equation of state Gas turbine High pressure Heat recovery steam generator High-temperature shift Integrated gasification combined cycle Intermediate pressure Lower heating value Low pressure Mitsubishi Heavy Industries Mixed-integer non-linear programming Medium pressure Pulverized coal Perturbed chain - statistical associating fluid theory Start of run Steam turbine Water-gas shift

Acronyms COP DEPEG EOR EoS GT HP HRSG HTS IGCC IP LHV LP MHI MINLP MP PC PC-SAFT SOR ST WGS

70


References [1] National Energy Technology Laboratory, September 2013. Cost and Performance Baseline for Fossil Energy Plants Volume 1: Bituminous Coal and Natural Gas to Electricity. Tech. rep. DOE/NETL-2010/1397, Revision 2a. [2] Descamps, C., Bouallou, C., and Kanniche, M., 2008. “Efficiency of an Integrated Gasification Combined Cycle (IGCC) power plant including CO2 removal”. Energy, 33(6), pp. 874–881. [3] Kunze, C., and Spliethoff, H., 2010. “Modelling of an IGCC plant with carbon capture for 2020”. Fuel Processing Technology, 91(8), pp. 934–941. [4] Gazzani, M., Macchi, E., and Manzolini, G., 2013. “CO2 capture in integrated gasification combined cycle with SEWGS - Part A: Thermodynamic performances”. Fuel, 105, pp. 206–219. [5] Kanniche, M., and Bouallou, C., 2007. “CO2 capture study in advanced integrated gasification combined cycle”. Applied Thermal Engineering, 27(16 SPEC. ISS.), pp. 2693–2702. [6] Huang, Y., Rezvani, S., McIlveen-Wright, D., Minchener, A., and Hewitt, N., 2008. “Techno-economic study of CO2 capture and storage in coal fired oxygen fed entrained flow IGCC power plants”. Fuel Processing Technology, 89(9), pp. 916–925. [7] Martelli, E., Kreutz, T., and Consonni, S., 2009. “Comparison of coal IGCC with and without CO2 capture and storage: Shell gasification with standard vs. partial water quench”. Energy Procedia, 1(1), pp. 607–614. [8] Gr¨ abner, M., Morstein, O., Rappold, D., G¨ unster, W., Beysel, G., and Meyer, B., 2010. “Constructability study on a german reference IGCC power plant with and without CO2 -capture for hard coal and lignite”. Energy Conversion and Management, 51(11), pp. 2179–2187. [9] Manzolini, G., Macchi, E., and Gazzani, M., 2013. “CO2 capture in Integrated Gasification Combined Cycle with SEWGS - Part B: Economic assessment”. Fuel, 105, pp. 220–227. [10] Cormos, C.-C., 2012. “Integrated assessment of IGCC power generation technology with carbon capture and storage (CCS)”. Energy, 42(1), pp. 434–445. cited By (since 1996)18. [11] Urech, J., Tock, L., Harkin, T., Hoadley, A., and Maréchal, F., 2014. “An assessment of different solvent-based capture technologies within an IGCC-CCS power plant”. Energy, 64, pp. 268–276. [12] Kaldis, S., Skodras, G., and Sakellaropoulos, G., 2004. “Energy and capital cost analysis of CO2 capture in coal IGCC processes via gas separation membranes”. Fuel Processing Technology, 85(5), pp. 337–346. [13] Grainger, D., and H¨ agg, M.-B., 2008. “Techno-economic evaluation of a PVAm CO2 -selective membrane in an IGCC power plant with CO2 capture”. Fuel, 87(1), pp. 14–24.

71

Chapter 3

[14] Krishnan, G., Steele, D., O’Brien, K., Callahan, R., Berchtold, K., and Figueroa, J., 2009. “Simulation of a Process to Capture CO2 From IGCC Syngas Using a High Temperature PBI Membrane”. Energy Procedia, 1(1), pp. 4079–4088. Proceedings of the 9th International Conference on Greenhouse Gas Control Technologies (GHGT9), 16-20 November 2008, Washington DC, USA. [15] Merkel, T. C., Zhou, M., and Baker, R. W., 2012. “Carbon dioxide capture with membranes at an IGCC power plant”. Journal of Membrane Science, 389(0), pp. 441–450. [16] Walspurger, S., Boels, L., Cobden, P., Elzinga, G., Haije, W., and Van Den Brink, R., 2008. “The crucial role of the K+-aluminium oxide interaction in K +-promoted alumina-and hydrotalcite-based materials for CO2 sorption at high temperatures”. ChemSusChem, 1(7), pp. 643–650. [17] Van Selow, E., Cobden, P., Verbraeken, P., Hufton, J., and Van Den Brink, R., 2009. “Carbon capture by sorption-enhanced water-gas shift reaction process using hydrotalcite-based material”. Industrial and Engineering Chemistry Research, 48(9), pp. 4184–4193. [18] Garc´ıa, S., Gil, M., Mart´ın, C., Pis, J., Rubiera, F., and Pevida, C., 2011. “Breakthrough adsorption study of a commercial activated carbon for pre-combustion CO2 capture”. Chemical Engineering Journal, 171(2), pp. 549–556. [19] Emun, F., Gadalla, M., Majozi, T., and Boer, D., 2010. “Integrated gasification combined cycle (IGCC) process simulation and optimization”. Computers and Chemical Engineering, 34(3), pp. 331–338. [20] Ng, K., Lopez, Y., Campbell, G., and Sadhukhan, J., 2010. “Heat integration and analysis of decarbonised IGCC sites”. Chemical Engineering Research and Design, 88(2), pp. 170–188. [21] Martelli, E., Kreutz, T., Carbo, M., Consonni, S., and Jansen, D., 2011. “Shell coal IGCCS with carbon capture: Conventional gas quench vs. innovative configurations”. Applied Energy, 88(11), pp. 3978–3989. [22] Carbo, M., Boon, J., Jansen, D., van Dijk, H., Dijkstra, J., van den Brink, R., and Verkooijen, A., 2009. “Steam demand reduction of water-gas shift reaction in IGCC power plants with pre-combustion CO2 capture”. International Journal of Greenhouse Gas Control, 3(6), pp. 712–719. [23] Bhattacharyya, D., Turton, R., and Zitney, S., 2011. “Steady-state simulation and optimization of an integrated gasification combined cycle power plant with CO2 capture”. Industrial and Engineering Chemistry Research, 50(3), pp. 1674–1690. [24] Damen, K., Gnutek, R., Kaptein, J., Nannan, N. R., Oyarzun, B., Trapp, C., Colonna, P., van Dijk, E., Gross, J., and Bardow, A., 2011. “Developments in the pre-combustion CO2 capture pilot plant at the Buggenum IGCC”. Energy Procedia, 4(0), pp. 1214–1221. [25] Eurlings, J., and Ploeg, J., 1999. “Process performance of the SCGP at Buggenum IGCC”. Gasification Technologies Conference, San Francisco, CA.

72


[26] Twigg, M. V., ed., 1989. Catalyst handbook, 2nd ed. Wolfe Publishing Ltd. [27] Han Raas, Vattenfall, 2014. Private communication, February 13. [28] Aspen Technology, Inc., 2013. Aspen Plus V7.3. www.aspentech.com. [29] de Visser, E., Hendriks, C., Barrio, M., Mølnvik, M. J., de Koeijer, G., Liljemark, S., and Gallo, Y. L., 2008. “Dynamis CO2 quality recommendations”. International Journal of Greenhouse Gas Control, 2(4), pp. 478–484. [30] Gross, J., and Sadowski, G., 2001. “Perturbed-Chain SAFT: An Equation of State Based on a Perturbation Theory for Chain Molecules”. Industrial and Engineering Chemistry Research, 40, pp. 1244–1260. [31] Nannan, N. R., de Servi, C. M., van der Stelt, T., Colonna, P., , and Bardow, A., 2013. “An Equation of State Based on PC-SAFT for Physical Solvents Composed of Polyethylene Glycol Dimethylethers”. Industrial and Engineering Chemistry Research, 52, pp. 18401–18412. [32] van Dijk, H. A. J., Cohen, D., Hakeem, A. A., Makkee, M., and Damen, K., 2014. “Validation of a water-gas shift reactor model based on a commercial FeCr catalyst for pre-combustion CO2 capture in an IGCC power plant”. Chemical Engineering Journal. submitted for publication. [33] Ruettinger, W., and Ilinich, O., 2006. Encyclopedia of Chemical Processing. Taylor & Francis, ch. Water Gas Shift Reaction, pp. 3205–3215. [34] Haldor Topsøe, 2013. High temperature shift catalyst. www.topsoe.com. [35] Lang, Y.-D., and Biegler, L., 1987. “A unified algorithm for flowsheet optimization”. Computers and Chemical Engineering, 11(2), pp. 143–158. [36] Aspen Technology, Inc., 2013. Aspen Process Economic Analyzer. www.aspentech. com.

73

“Die Mauer wird in 50 und auch in 100 Jahren noch bestehen bleiben, wenn ¨ die dazu vorhandenen Grunde nicht beseitigt sind.” “The Wall will be standing in 50 and even in 100 years, if the reasons for it are not removed.”

Erich Honecker, Chairman of the State Council of the German Democratic Republic, speech at a congress, East-Berlin, January 19, 1989

4

Efficiency improvement in pre-combustion CO2 removal units with a waste-heat recovery ORC power plant1 This chapter presents an analysis about recovering low-grade thermal energy from a pre-combustion CO2 capture process by means of organic Rankine cycle (ORC) turbogenerators. First, the application of commercially available ORC units is explored by means of steady-state simulations. The performance of the system composed by available standard ORC turbogenerators is taken then as a benchmark for the simulation of tailor-designed ORC power plants. The working fluid has a major influence on system performance and other technical and economic factors. The effect of selecting a fluid from the hydrocarbon and refrigerant families are therefore investigated, targeting the maximum net power output. In addition to pure fluids, also two-component mixtures are considered. The use of mixtures as working fluids in subcritical heat-recovery ORC systems allows for a better match of the temperature profiles in the primary heat exchanger and the condenser due to the temperature glide associated with phase-transition. In order to further improve the thermal coupling between the cooling heat source and the heating of the working fluid, the supercritical cycle configuration is also studied. The performance of the three categories of systems, depending on working fluid and cycle configuration, i.e., systems based on (i) commercially available units, (ii) tailor-designed subcritical cycle, (iii) tailor-designed supercritical cycle, is analysed in terms of net power output, second law efficiency and component-based exergy efficiencies. In this study, particular attention is focused on the semi-empirical optimization approach. 1 The contents of this chapter appeared in: Trapp, C., and Colonna, P., 2013. Journal of Engineering for Gas Turbines and Power, 135(4).

Chapter 4

4.1

Introduction

This study takes as an example the design of the Magnum integrated gasification combined cycle (IGCC) power plant and its CO2 capture unit [1], in order to investigate the performance of an ORC system recovering thermal energy downstream the water-gas shift reaction section and by inter-cooling the CO2 compression section, see Figure 4.1. An ORC power plant is arguably the only available technology capable of efficiently converting low-grade thermal energy at ≈ 140 ◦ C into electricity in the considered power range (few MWe ). An increasing number of studies on low-grade thermal energy conversion by means of ORC systems have been published in recent years, covering applications ranging from geothermal power, to waste-heat recovery, and to concentrated solar power. The selection of the working fluid affects all the most important design variables, and as such has a large influence on system and components performance and cost [2]. In addition, other important factors, like fluid cost, availability, toxicity, flammability, global warming potential (GWP), ozone depletion potential (ODP), etc., must be accounted for, making for a complex preliminary design optimization problem, requiring diverse competences, and tackled in a semi-empirical manner. Most of the documented research studies include a screening of suitable working fluids, based on different criteria, such as their environmental impact and thermophysical properties, complemented by simulation studies aimed at evaluating the system’s performance, and exhibiting a different level of methodological approach and completeness. Saleh et al. [3] studied geothermal ORC systems applications featuring a reservoir with a temperature of 100 ◦C and estimated the conversion efficiency as a function of 31 pure working fluids belonging to the family of the hydrocarbons and of the refrigerants. The analysis covered both the subcritical and the supercritical cycle configuration. Highest thermal efficiencies are obtained for high-boiling retrograde fluids in subcritical processes, though the largest amount of thermal energy is transferred to a supercritical fluid, leading to the highest net power output. Madhawa Hettiarachchi et al. [4] proposed the ratio of the total heat exchanger area to net power output as a design criterion in order to determine the cost-effective optimum of a geothermal ORC power plant. The evaporation and condensation temperatures, geothermal and cooling water velocities were chosen as variables for the performance optimization obtained with the steepest descent method. For a 10 MWe ORC power plant considering a geothermal water temperature of 90 ◦ C the selection of ammonia as the working fluid results in the minimum of the objective function, while the exergy efficiency of the ammonia cycle is largely compromised. In the area of ORC waste-heat recovery applications, Dai et al. [5] performed optimizations of subcritical and superheated ORC systems in terms of exergy efficiency by means of a genetic algorithm. The turbine inlet temperature (TIT) and pressure were optimized considering 10 different working fluids (wet and dry) 76

Efficiency improvement in pre-combustion CO2 removal units with a waste-heat recovery ORC power plant

under the same given waste heat condition. The simulations show that a thermodynamic cycle using R236ea as the working fluid is optimal and allows reaching an exergy efficiency of 35.43 %. Furthermore, the study concluded that the addition of an internal heat exchanger would not improve the ORC system performance. Tchanche et al. [6] studied low-temperature 2 kWe ORC systems with a microturbine to convert concentrated solar radiation and evaluated 20 refrigerants as working fluids for a saturated subcritical cycle configuration. The system is designed for applications in hot areas where the average monthly temperature is around 28 ◦ C. Thermodynamic, environmental and safety criteria were assessed: none of the considered working fluids allows to satisfy all the stated requirements, namely high efficiencies (thermal and exergetic), reasonable condensation and evaporation pressures, low ODP and GWP, non-toxicity and non-flammability. A ranking of the working fluids was conducted, and R134a was identified as the most suitable fluid for the considered application, followed by R152a, R290, R600 and R600a. In analogy to steam power plants, also in ORC power systems the supercritical cycle configuration can improve on efficiency, with the advantage, compared to steam power plants, that the supercritical pressure in the primary heat exchanger is typically one order of magnitude lower, see e.g. Refs. [2, 7]. To the knowledge of the authors, the first experimental supercritical ORC power plant has been commissioned in 2012 [8]. Schuster et al. [9] evaluated the performance of small-scale systems using various organic working fluids, featuring either subcritical or supercritical conditions in the primary heat exchanger for a heat source of 210 ◦ C. Their study shows that adopting the supercritical cycle configuration yields an efficiency improvement of 8 % in the considered cases. The main thermodynamic advantage of a supercritical Rankine cycle in comparison to the superheated configuration is the decrease of irreversibility in the primary heat exchanger due to a better match of the temperature profiles. Angelino and Colonna [10] first introduced the concept of using non-azeotropic organic-fluid mixtures as working fluids for Rankine power cycles. They showed that the added degree of freedom of the selection of the composition allows for the fine tuning of the working fluid critical point with respect to a given heat source and sink, with consequences on all relevant design variables. The main thermodynamic benefit is due to the non-isothermal phase change, leading to a better match of the temperature profiles within the heat exchanging equipment, at the cost of additional heat transfer surface [11, 12]. Heberle et al. [13] calculated the performance of low-temperature subcritical ORC systems with mixtures of isobutene/isopentane and R227ea/R245fa as working fluids and quantified for heat source temperatures below 120 ◦C an increase of the second law efficiency in the range of 4.3 to 15 % in comparison to pure fluids, due to the decrease of heat transfer irreversibility. Wang and Zhao [14] investigated the performance of subcritical low-temperature solar ORC systems using three different non-azeotropic mixtures composed of R125a and R245fa, corresponding to a dry, wet and isentropic working fluid. The evaporation temperature was fixed to 80 ◦ C and the 77

Chapter 4

condensation temperature to 25 ◦ C. The analysis showed that the ORC cycles employing mixtures require, in comparison to a cycle with the pure fluid R245fa, smaller expander dimensions (lower volumetric expansion ratio and lower volumetric flow rate at turbine inlet) which would result in a reduction of the expander cost. On the other hand, employing a pure working fluid yields the highest thermal efficiency. Chen [15] described the performance comparison between a low-temperature supercritical ORC system using a non-azeotropic mixture of R134/R32 as working fluid and a subcritical ORC plant using pure R134a. A 10 − 30 % increase in thermal efficiency was estimated for the former. The predicted system exergy efficiency, defined as the product of the exergy efficiencies of the heating process, the energy conversion cycle and the condensation process, increased from 24.1 % for the subcritical cycle with a pure fluid to 38.6 % for the supercritical cycle with the non-azeotropic mixture. Pressurized hot water (p = 0.5 MPa, T = 136.8 ◦ C) is the considered heat source, heating the working fluid with a mass flow of 1 kg/s up to 126.8 ◦ C. The energy sources of low-temperature ORC power systems studied in the literature, simulated or actual, are flue gases (direct waste-heat recovery), diathermic oil (indirect waste-heat recovery and concentrated solar power), or pressurized water (geothermal energy conversion). In all these cases the fluid of the thermal energy source does not change phase. The application of low-grade heat recovery by means of an ORC turbogenerator documented here is different, as the thermal energy source, syngas from the water-gas shift reactor, partly condenses while cooling. The optimization of the thermodynamic cycle and of its implementation in an actual system departs therefore from other more conventional cases. This chapter is organized as follows: Section 4.2 briefly describes the CO2 capture plant, and puts into evidence the streams where low-grade thermal energy can be conveniently recovered with an ORC power plant. Section 4.3 presents the configurations of the ORC systems analysed in this study. Section 4.4 illustrates the methodology adopted to perform the optimization of the thermodynamic cycles, considering also the technical constraints. The analysis of the results of the system simulations is discussed in Section 4.5, while the concluding remarks are presented in Section 4.6.

4.2

CO2 capture process configuration and waste heat recovery possibilities

The simplified process flow diagram of the CO2 removal plant designed for the Magnum power station is depicted in Figure 4.1. The syngas from the gasifier entering the CO2 capture unit is mixed with process water in order to obtain a pre-set H2 O:CO ratio, and then it is fully evaporated and superheated. Carbon monoxide present in the syngas is converted into hydrogen and carbon dioxide via a sweet high-temperature water-gas shift reaction. The excess process water is 78


CO2 Absorber

H2-rich Syngas Flash drums CO2 Compressor

CO2 Product H

Separator

p: 21.3, T: 35-40 : 35.7 WGS reactors H

Solvent pump

B

p: 110 T: 40

: 27.3

A

p: 21.6, T: 137

Syngas Syngas compressor

p: 110.3 T: 140

A - Downstream shifting section B - Downstream CO2 compressor

H

Process water H

IP steam

H

p = Pressure [bar] T= Temperature [°C]  = Thermal energy [MWth]

Figure 4.1: Simplified process flow diagram of a pre-combustion CO2 capture island suitable for integration into an IGCC power plant.

recovered from the shifted syngas and recycled. Subsequently, carbon dioxide is removed from the syngas by means of physical absorption resulting in a H2 -rich syngas, which is fed to the gas turbine of the combined cycle power plant. Finally, the captured carbon dioxide is compressed to a state suitable for storage. Within the process, two streams at moderate temperature must be cooled, and are therefore suitable for the recovery of thermal energy. The first is located downstream of the water-gas shift section, indicated with A in Figure 4.1, whereby the syngas must be cooled in order to recover excess water. The second stream, B in Figure 4.1, corresponds to the outlet of the CO2 compressor. For the sake of simplicity the multistage compressor with interstage cooling is depicted as one component. In both cases thermal energy is available at temperatures around 130 − 140 ◦ C. Both streams are sulphur-free, therefore they can be cooled down to approximately 35 − 40 ◦ C, a limit prescribed by the rest of the CO2 capture process. The mass flows are 72 kg/s (A) and 45 kg/s (B), therefore the maximum amount of recoverable energy is approximately 35.7 MWth from the syngas and 27.3 MWth from the compressed CO2 . The fact that (i) the heat recovery opportunity is located within the power station perimeter, that (ii) the IGCC power plant is mostly suitable for base-load operation, and that (iii) water is available for cooling, are almost ideal conditions for the application of ORC technology to low-grade heat recovery. This study is focused on the optimization of the recovery of thermal energy from the syngas stream, as the energy content is higher than for the compressed CO2 stream. In addition, unlike in other more conventional low-temperature ORC applications, whereby the heat source fluid remains single-phase, the syngas in this case must undergo condensation while it cools down in the heat-recovery vapour generator. 79

Chapter 4

Thermal energy source Mixture components (% mole fractions) Pressure Mass flow rate Inlet temperature Minimum outlet temperature Thermal energy sink Inlet temperature (seasonal average)

bar kg/s ◦C ◦C ◦C

Syngas-Water mixture (A in Figure 4.1) H2 (46.0), CO2 (31.3), H2 O (15.3), N2 (4.3), CO (3.1) 21.6 72 137 35 − 40 Cooling water 17

Table 4.1: Constraints on the operating conditions for the optimized design of the ORC heat-recovery power plant powered by the syngas thermal energy (A in Figure 4.1).

The design values of the main process variables characterizing the syngas stream in A and the values characterizing the available cooling water, which are used as constraints for the optimization of the ORC heat recovery system are listed in Table 4.1. The temperature of the cooling water is taken as an average among its seasonal values.

4.3

Waste-heat recovery ORC power plants and their configurations

In this study three ORC power plant configurations are analysed, namely 1. a system formed by commercially available standard small-capacity units connected in parallel to the thermal energy source, implementing the subcritical (slightly superheated) cycle configuration (base case); 2. a tailor-designed system made by an optimized single unit (or possibly two units) implementing the subcritical configuration, and finally 3. an optimally designed system implementing the supercritical configuration. The process flow diagram is the same for all configurations and is depicted in Figure 4.2(a). Note that a slight degree of vapour superheating at the outlet of the primary heat exchanger (T ≈ 2 K) is common practice in order to avoid erosion of the turbine blades due to accidental admission of liquid droplets from the primary heat exchanger. At high reduced evaporation pressure pevap /pc , see, e.g. Figure 4.2(b), 80


the degree of superheating at the turbine inlet must be larger. The constraint adopted in this study is that in all cases the average thermodynamic states of the expanding fluid (line 3 − 4 and 30 − 40 in Figure 4.2(b)) are superheated by at least 2 K with respect to the saturated state at the given pressure. This results in a higher degree of superheating at the turbine inlet for some of the considered working fluids. A more detailed analysis of the fluid expansion might point to the need of a somewhat higher degree of superheating in order to prevent droplet formation, though this does not appreciably influence the overall result of this analysis. The regenerative configuration, i.e., adopting a recuperator in order to internally transfer thermal energy available at the outlet of the turbine to the liquid at the outlet of the feed-pump, is not considered in this work. Experience and previous studies [5, 13] show that in low-temperature heat recovery applications the low degree of regeneration which is possible has a positive effect on the cycle efficiency, while the net power output and the exergetic efficiency of the process is almost unaffected. The merit parameter in this case is the net power output. Figure 4.2(b) shows the working fluid processes occurring in the system of Figure 4.2(a). Note that an effective regenerator would prevent temperature T6 in Figure 4.2(a) from reaching its minimum value, thus reducing the thermal energy input to the ORC system. These considerations are not valid for high-temperature ORC applications, whereby the trade-off between cycle efficiency and thermal energy input must be evaluated. A single ORC power plant instead of multiple units connected in parallel can be beneficial in terms of conversion performance because the higher volumetric flow allows for the design of a more efficient turbine, being all gap and friction losses proportionally smaller. As for the heat exchanging equipment, the possible advantage in terms of performance is less clear, as the pressure losses, and mean temperature differences deriving from the techno-economic optimization of the heat exchanger design are possibly quite similar. Multiple parallel units can be advantageous from a maintenance and reliability point of view, though ORC systems are known to be extremely reliable and reached a very high level of automation.

4.4

Analysis and optimization methodology

The preliminary design optimization of the ORC systems considered in this study is based on steady-state simulations. The typical mass and energy balances describing the Rankine cycle are therefore solved, using as specified variables the minimum temperature difference in the primary heat exchanger and condenser, and the isentropic efficiency of the feed-pump and the turbine. The working fluid mass flow is obtained from the energy balance of the primary heat exchanger, while the water coolant mass flow is obtained from the energy balance of the condenser. Together with the data in Table 4.1, the performance of the ORC system 81

Chapter 4

Cycle-Tempo 5.0 (Build 486)

3 Primary heat exchanger H

5 2

PureCycle - HT 140

6 4

Cooling water

Fluid Subcritical ORC process Supercritical ORC process

3’

Working fluid: R245fa

Turbine

Condenser

Temperature [°C]

Syngas

Pump

d:\work\orc research\orc ct models\utc purecycle ht.gui

Pnet: 243.87 kW 120

Efficiency: 11.6% Tmax: 105.00°C Tmin: 18.33°C Cooling: Water

100

60 40 20 0 1

1

3

80 4’ 2 2’

4

1

1.1

(a)

1.2 1.3 1.4 Entropy [kJ/(kgK)]

1.5

1.6

(b)

Figure 4.2: a) Configuration of ORC power plants. b) Subcritical and supercritical ORC processes in the T − s thermodynamic plane, in the case of pure working fluids (R227ea).

components as listed in Table 4.2 is also fixed. Hence the only operating parameter which can be varied in order to maximize the power output is the cycle’s maximum pressure, which is constrained to be subcritical for superheated cycle configurations and supercritical for supercritical cycle configurations. A standard algorithm for constrained optimization is coupled to the simulation tool, whereby the objective function is the maximum net power output. The other variable in the problem is the working fluid, therefore its thermodynamic properties. The use of a mixture adds a degree of flexibility, in that, besides selecting the compounds, their composition can also be varied. The resulting optimization problem can formulated in general terms as May 17, 2014 23:58:12

Page 1 of 1

min Pnet = Pturb − Ppump , x

s.t.

g(·) = 0,

(4.1)

h(·) ≤ 0, where x corresponds the set of decision variables containing both continuous and integer variables (pevap for subcritical cycles or T IT for supercritical cycles, mixture compounds, binary mixture composition), g(·) is the set of algebraic equations that describes the ORC system and h(·) is the set of inequality constraints. For the subcritical cycle the inequality constraint is related to the evaporation pressure and for the supercritical cycle to the turbine inlet temperature (see Table 4.2). The optimization problem therefore has two degrees of freedom (i.e., number of variables minus the number of constraints) for cycle configurations with a pure working fluid and four degrees of freedom in case mixtures are considered. Pressure drops in pipes and heat transfer equipment as well as heat losses to the environment, and power consumption of the cooling pump are neglected. Pressure 82


drop estimations are available for the commercial unit considered in this study, and their effect on the net power output is a 3 % decrease. The estimation of the pressure drop is largely uncertain for the other considered systems, therefore it has been decided to neglect pressure drops in all cases. The simulations are performed with a proprietary software [16], utilizing a wellknown library for the computation of the working fluid properties, either pure fluids or mixtures [17, 18]. The software for the calculation of fluid properties implements accurate multiparameter equations of states and provides uncertainty estimations, which are well within engineering practice in this field (< 5 % for all properties in case of pure fluids). In case of binary refrigerant mixtures as working fluids, the mixing parameters in the equation of state have been estimated based on the model of Lemmon and Jacobsen [19], which in turn is based on accurately measured data for mixtures of R-32, R-125, R-134a, R-143a and R-152a. The estimated uncertainties for the calculation of the properties of these reference mixtures are 0.1 % for density and 0.5 % for heat capacities. Note that the uncertainties for mixtures R236fa/R134a, R236fa/R152a and R236fa/R365mfc are expected to be larger. The thermodynamic properties of the syngas-water mixture are calculated using the perturbed chain - statistical associating fluid theory (PC-SAFT) equation of state [20], which is also implemented in the property library [17]. Data for the binary interaction parameters and overall equation of state optimization have been performed within this project and the work is documented in Ref. [21]. As for the exergy analysis, the exergetic efficiency of the components of the ORC system are taken as the ratio between the specific flow exergy made available to the component divided by the specific flow exergy which is converted into the useful effect. This results in ηex,pump ≡

∆epump e2 − e1 = , wpump h2 − h1

(4.2)

for the pump, where the specific flow exergy of the working fluid is defined as e ≡ h − h0 − T0 (s − s0 ),

(4.3)

and the dead state is provided by the ambient conditions featuring T0 = 17 ◦ C, p0 = 1.013 bar. Analogously, for the turbine wturb h3 − h4 = , ∆eturb e3 − e4

(4.4)

∆E˙ORCfluid,1HX m˙ ORC (e3 − e2 ) = , ˙ m˙ syngas (ein − eout )syngas ∆Esyngas

(4.5)

ηex,turb ≡ for the primary heat exchanger ηex,1HX ≡ and for the condenser

83

Chapter 4

Parameter pevap,max (subcritical) p1HX (supercritical) Vapour superheating at turbine inlet (subcritical)a Turbine inlet temperature (supercritical) ∆ T pinch point primary heat exchanger ∆ T pinch point condenser Average cooling water temperature Cooling water temperature rise Turbine isentropic efficiency Generator efficiencyb Pump efficiencyc

Unit bar bar K ◦C K K ◦C K % % %

Value 0.9 × pc 1.1 × pc 2 1.1 · Tc < T IT < (Tsyngas,in − ∆Tpinch ) 10 5 17 5 90 96 65

a For simulated ORC power plants with R227ea, R236fa/R152a and R236fa/R134a as working fluids, vapour superheating at turbine inlet is T = 4 K. b Lumped with the turbine mechanical efficiency. ηpump = ηis · ηmech · ηel c

Table 4.2: Constrained operating parameters for the optimization of subcritical and supercritical ORC power plants.

84


ηex,cond ≡

∆E˙water m˙ water (eout − ein )water = . ˙ m˙ ORC (e4 − e1 ) ∆EORCfluid,cond

(4.6)

The exergetic efficiency of the ORC system is therefore calculated as ηex,cycle ≡

Pturb − Ppump . ∆E˙syngas

(4.7)

Screening of the working fluid The selection of the optimal working fluid for an ORC system is currently performed in a semi-empirical manner. Several requirements and merit parameters are involved, namely - acceptable environmental impact, - acceptable safety requirements and low toxicity, - acceptable cost and availability, - thermal stability in contact with stainless steal and other sealing materials at the maximum operating temperature, - conversion efficiency of the theoretical thermodynamic cycle, - performance and cost of the turbine and the pump (sometimes mere feasibility), - cost of the heat exchanging equipment. A short list of working fluids among those belonging to the suitable families of the hydrocarbons and of the refrigerants is compiled by assessing environmental, safety and cost requirements, together with the fluid critical temperature, which must be in the range of temperature of the thermal energy source. A comparatively high critical temperature means a lower critical pressure, therefore a lower density of the fluid throughout the expansion, therefore a higher volume flow, and larger condenser. Molecular complexity, therefore the isobaric specific heat in the ideal gas state, is also an important fluid parameter, because it is correlated to the slope of the dew line in the temperature-entropy diagram of the fluid. Fluid exhibiting a positive or vertical slope of the dew line, therefore formed by comparatively more complex molecules, are preferable because they entail a dry fluid expansion even without superheating. Two important indexes related to the environmental impact of organic fluids are the ozone depletion potential and the global warming potential. The ODP measures the amount of degradation to the ozone layer caused by chemical substances referenced to refrigerant R11 which is taken as ODP = 1. According to the Montreal Protocol most of the fluids with an ODP much higher than zero, 85

Chapter 4

namely chlorofluorocarbons (CFC) and hydrochlorofluorocarbons (HCFC), have been or are to be phased out. The GWP measures the amount of thermal energy that a certain fluid in the gaseous state can trap within the atmosphere relative to the amount trapped by CO2 , and it is measured over a time span. No legislative restrictions on fluids with high GWP are yet in place. The safety level of refrigerants can be evaluated using the ASHRAE Standard 34 which classifies the fluid in terms of toxicity and flammability2 . Potential working fluid candidates should hence have a low ODP and GWP value and a high safety level. Potential working fluid candidates which are screened for the use in the subcritical and supercritical ORC system are summarized in Table 4.3.

4.5 4.5.1

Results and discussion The base-case: ORC system composed of standard power modules

ORC turbogenerators for low-grade heat recovery can be obtained from few manufacturers as standard skid-mounted modules. These models differ in terms of power capacity, temperature operating range, and type of cooling system. This technical solution can be economically attractive because a certain economy of production can be reached. A list of manufacturers and models is reported in Ref. [26]. The list and the analysis of technical data related to units that recently became commercially available show that very few models are suitable for the application at hand. A 280 kWe standard ORC unit suitable for low-temperature applications is selected for the benchmark system considered in this study. The ORC turbogenerator utilizes R245fa as the working fluid, in a slightly superheated thermodynamic cycle, and a radial turbine. In analogy with a geothermal power plant recently commissioned near Beaver, Utah [27], which features forty such units connected in parallel, in this case thirteen of these ORC modules could be powered by the syngas thermal energy source. Based on vendor performance data and estimated values for the turbine and the pump efficiency, pinch temperature difference in the primary heat exchanger and in the condenser, and the nominal evaporation pressure, a net power output of 3406 kWe is calculated, under the assumptions described in Section 4.4. The exergy efficiency of the system is 32.8 %. Figure 4.3 shows the obtained thermodynamic cycle in the T −s diagram. The temperature profiles of the syngas cooling down and of the cooling water heating up are superimposed to serve as an aid to graphically evaluate the quality of the thermodynamic cycle. It is notable that the temperature profile of the syngas is not linear, indicating that the brine undergoes condensation, 2 It

contains a character (A: Lower toxicity; B: Higher toxicity) and a number (1: No flame propagation; 2: Lower flammability; 3: Higher flammability)

86

ASHRAE 34 [22] n.a. A1 A2 A3 A1 A1 A2 A1 A3 n.a. A3 B1 n.a. n.a. A3 A3

Tc [◦ C] 44.13 66.02 78.11 96.74 101.06 101.75 113.26 124.92 134.66 139.29 151.98 154.01 174.42 186.85 187.20 196.55 pc [bar] 58.97 36.18 57.82 42.51 40.59 29.25 45.17 32.00 36.29 35.02 37.96 36.51 39.25 32.66 33.78 33.70 Molecular mass [kg/kmol] 34.0 120.0 52.0 44.1 102.0 170.0 66.1 152.0 58.1 152.0 58.1 134.0 134.0 148.1 72.1 72.1

GWP 100-yr 97b 3500 675 ∼ 20c 1430 3220 124 9810 ∼ 20c 1200b ∼ 20c 1030 693c 794 ∼ 20c ∼ 20c

a

ODP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Table 4.3: Relevant thermophysical properties and environmental impact indexes of selected working fluids, ordered by increasing critical temperature.

From the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC): Climate Change 2007 [23] except where indicated. b IPCC Third Assessment Report: Climate Change 2001 [24]. c Calm and Hourahan [25]

a

Fluid name R41 R125 R32 Propane R134a R227ea R152a R236fa Isobutane R236ea Butane R245fa R245ca R365mfc Isopentane Pentane


87

Chapter 4

160 140

Temperature [°C]

120 100

Fluid Heat source Cooling water Process pheat= 12.0 bar pcond= 1.67 bar

80 60 40 20 0 1

1.2

1.4 1.6 Entropy [kJ/(kgK)]

1.8

2

Figure 4.3: T-s diagram of base case ORC process for working fluid R245fa.

which starts, see Figure 4.3, few degrees below the syngas inlet temperature. The estimated liquid mass fraction at the outlet, at 65.7 ◦ C, is 14.3 %. More detailed results of the simulation are reported in Table 4.6.

4.5.2

Optimized subcritical ORC power plant

Several thermodynamic considerations support the optimization procedure. The reduction of the irreversibilities in the heat exchanging equipment is of main importance. Considering first the primary heat exchanger, the mismatch of the temperature profiles of the heat source and the working fluid is largely affected by the evaporation of the working fluid, as shown in Figure 4.4(a) and Figure 4.4(b). These figures display the temperatures as a function of the exchanged thermal power for exemplary cycle calculations using R245fa and R236fa as working fluids, respectively. The adoption of a working fluid with a lower critical temperature, for a given minimum temperature difference in the heat exchanger, entails a higher reduced pressure in the evaporator pevap /pc , and therefore less energy is transferred to the fluid while it vaporizes at constant temperature. Note also that in this case the pinch point is located at the inlet of the evaporator, the syngas is cooled to a much lower temperature, and therefore approximately 15 % more energy is transferred to the working fluid. In addition, thermal energy is added to the thermodynamic cycle at comparatively higher temperature, thus also positively affecting cycle efficiency. The constraint acting in this case is therefore the limitation on the subcritical evaporation pressure, which is kept sufficiently 88


120

140 Heat source Working fluid

Temperature [°C]

Temperature [°C]

140

100 80 60 40 20 0

1 2 Thermal power [kW]

(a)

3 4 x 10

120

Heat source Working fluid

100 80 60 40 20 0


3 4

x 10

(b)

Figure 4.4: Temperature profile of the primary heat exchanger in a subcritical cycle with the working fluid a) R245fa and b) R236fa.

lower than the critical pressure (i.e., 90 % of pc ) for the control system to be able to prevent supercritical operation of the evaporator. It is worth noting that the reduction of thermodynamic irreversibilities in the primary heat exchanger comes at the cost of additional heat transfer area. In order to verify the outlined thermodynamic considerations, cycle simulations have been carried out for a broader selection of working fluids, covering a critical temperature range from 102 to 187 ◦ C. The saturation lines in the T − s digram for the considered fluids are shown in Figure 4.5. The figure also shows a temperature band dependent on the temperature of the heat source, encompassing the critical temperature of fluids which are expected to be good candidates as working fluids for optimal ORC systems. Figure 4.6 shows the simulated net power output as a function of the evaporation temperature of systems adopting as working fluids those whose saturation lines are displayed in Figure 4.5. The maximum net power output is obtained with working fluids providing an evaporation temperature between 92 and 105 ◦ C. The system simulation results associated with the working fluids providing the highest net power output are summarized in Table 4.4. The critical temperatures of these fluids, namely R236fa, isobutane and R236ea, are indeed close to the temperature of the heat source, Tin,syngas = 137 ◦ C, and hence the observations can be used as guidelines for similar problems. Secondly, a further reduction of thermodynamic losses can be accomplished by improving the heat transfer in the condenser. As shown in Figure 4.7(a), the condensation of a pure fluid by means of cooling water implies a thermodynamic irreversibility which is proportional to the temperature difference between the two fluids. The pinch point temperature difference, located at the end of the working fluid desuperheating, cannot be cost-effectively reduced beyond a certain limit, typically 3 − 5 K, by just adding heat transfer surface. The use of a mixture as the working fluid in ORC systems is treated in Refs. [10, 11]. The isobaric phase tran89

Chapter 4

200 R365mfc

180

R245ca

160 R245fa

Temperature [°C]

140

R236ea Butane R236fa

120 100

R227ea

Isobutane

80 60 40 20 0 1

1.2

1.4

1.6 1.8 2 Entropy [kJ/(kgK)]

2.2

2.4

2.6

Figure 4.5: Saturation line in the T −s diagram of potential working fluids for subcritical ORC systems.

4600

Net power output [kW]

4400 4200

R227ea R236fa Isobutane R236ea Butane R245fa R245ca R365mfc

R236fa

R236ea

Isobutane

4000 3800 3600 3400 3200 70

75

80

85 90 95 100 Evaporation temperature [°C]

105

110

Figure 4.6: Calculated net power output as a function of the evaporation temperature for subcritical ORC power plants using the same working fluids as in Figure 4.5

90


Fluid R236fa Isobutane R236ea

Tc [◦ C] 124.9 134.7 139.3

pevap [bar] 20.0 17.6 14.3

ηth [%] 12.5 12.3 12.4

Pnet [kW] 4415 4176 4251

ηex [%] 39.2 37.6 38.1

Table 4.4: Performance of optimized subcritical ORC systems adopting pure working fluids providing the highest net power output among all those considered (Table 4.3).

sition of non-azeotropic mixtures is not isothermal, therefore the thermodynamic loss in the condenser can be reduced by promoting heat transfer from the working fluid to the coolant under almost constant temperature difference. Note that the heat transfer coefficient of a mixture undergoing phase transition is lower than that of a pure fluid in similar conditions, see e.g. Ref. [28]. The use of mixtures as working fluid requires conversely additional heat transfer surface and a small increase in the coolant pumping power. The optimization of the trade-off between the increase in conversion efficiency and the cost of the additional heat transfer surface can be performed only by designing the heat exchanger, as shown in Ref. [12]. Note that the same reasoning applies to the evaporator, but the thermodynamic advantage in this case is very small, as the portion of the working fluid heating process entailing phase change is rather limited. Optimal cycle configurations are in fact obtained for vaporization pressures close to the critical one.

40

50 Cooling water Working fluid

Temperature [°C]

Temperature [°C]

50

30 20 10 0 0


(a)

3 4 x 10

40

Cooling water Working fluid

30 20 10 0 0


3 4 x 10

(b)

Figure 4.7: Condensing process of a) pure working fluid R236fa and b) non-azeotropic mixture R236fa/R152a.

In the considered application, the water utilized for cooling implies an environmental constraint on the maximum temperature increase that the water can undergo, namely 5 K. The glide of the condensing mixture fluid can be made approximately the same and linear by selecting appropriate components and by 91

Chapter 4

Fluid R236fa, R134a, 0.8, 0.2 R236fa, R152a, 0.6, 0.4 R236fa, R365mfc, 0.9, 0.1

Tc [◦ C] 120.1 120.3 131.1

p1HX [bar] 26.3 27.4 17.0

ηth [%] 12.8 13.0 12.5

Pnet [kW] 4589 4614 4503

ηex [%] 40.5 40.8 39.7

Table 4.5: Estimated performance of optimized subcritical ORC power plants using nonazeotropic mixtures as working fluid, yielding the highest net power output. 160 140

Temperature [°C]

120 100

Fluid Heat source Cooling water Process pheat= 27.4 bar pcond= 3.85 bar

80 60 40 20 0 1

1.2

1.4 1.6 Entropy [kJ/(kgK)]

1.8

2

Figure 4.8: Cycle diagram in the T − s thermodynamic plane of the optimized subcritical configuration in case of a R236fa (0.6) /R152a (0.4) mixture as the working fluid.

varying the composition. Mixture working fluids which fulfil these requirements are listed in Table 4.5, together with the main performance results of cycle calculations. The optimal subcritical cycle utilizing the mixture R236fa/R152a is depicted in Figure 4.8. As expected, the thermodynamic gain deriving from the use of an optimized mixture, therefore of thermal energy rejection at comparatively lower temperature, is appreciable. The temperature profile of a mixture liquefying in a condenser can be linear, as depicted in Figure 4.7(b), or curvilinear, therefore the location of the pinch point is not known a priori. The simulation of systems employing a mixture as a working fluid includes therefore a discretisation of the condensation process in order to correctly set the value of the minimum temperature difference over the heat exchanger. 92


In order to provide a preliminary evaluation of the trade-off between the increase in net power output and the increase in heat transfer surface in the condenser, the calculation of the heat transfer area of a shell-and-tube condenser is performed in both the case of a condensing pure fluid and of a mixture. The increase in heat transfer surface for the mixture condenser is due both to the smaller average temperature difference and to a smaller heat transfer coefficient. Also the mixture evaporator is comparatively larger than its pure-fluid counterpart, but the economic consequence is far less relevant than for the condenser, being the fluid much more compressed, and the portion of the heating process involving phase change very small. The condenser is assumed to be a shell-and-tube heat exchanger with countercurrent flow configuration. The refrigerant flows in the shell and the cooling water inside the tubes. The heat transfer coefficient and area are calculated with a distributed condenser model utilizing simple correlations for each heat transfer phenomenon. The number of intervals is chosen such that the thermodynamic and transport properties change smoothly. For the intervals of the single-phase heat transfer at the inlet of the condenser equal enthalpy differences are assumed. The region of phase transition is divided in sections of equal vapour quality. For the preliminary dimensioning of the shell-and-tube condenser the design guidelines discussed in Ref. [29] have been followed. The tube-side heat transfer coefficient for water can be described with the single-phase turbulent flow correlation from Sieder and Tate [30] 1

αtube = 0.023Re0.8 Pr 3

λ D

0.14

µ

.

µwall

(4.8)

For the desuperheating region at the inlet of the shell-side the Taborek version of the Delaware design method is used [31], that is αshell,v = j0 cp Gshell Pr

− 23

0.14

µ

.

µwall

(4.9)

The heat transfer coefficient for the condensation outside a tube bundle is given by Kern [32] as αshell,tp = 0.95λl

ρl (ρl − ρv )g µl Γ

1 3

− 16

Nr

.

(4.10)

The Silver-Bell-Ghaly method [28] is used for the correction of the heat transfer coefficient of non-azeotropic mixtures whereby the vapour heat transfer coefficient αv is calculated according to Equation 4.9, which yields 1 1 Zv = + αeff αtp αv and Zv = xcp,v

∆Tglide dTdew ≈ xcp,v . dh ∆hm

(4.11)

(4.12) 93

Chapter 4

Performance results and relevant system parameters for the optimal subcritical cycle ORC systems are summarized in Table 4.6 together with the results of the preliminary sizing of the condenser. Given the large amount of heat transfer surface the plant configuration with two parallel units seems more suitable, with the added benefit of redundancy.

4.5.3

Optimized supercritical ORC power plant

The supercritical cycle configuration allows for an even better match of the temperature profiles of the cooling syngas and of the heating working fluid in the primary heat exchanger. In addition, a higher turbine inlet temperature can be achieved. The decrease in thermodynamic losses, together with an increase of thermal energy input to the cycle provides an increase in the heat recovery performance, if compared to the subcritical cycle configuration. The maximum cycle pressure is considerably higher, but one of the advantages of ORC systems, is that supercritical pressure conditions for the considered organic fluids are well within reach with standard process technology and are an order of magnitude lower than in supercritical steam power cycles. The design of an efficient feed-pump and of the control system might involve some challenges. The control system needs to maintain the pressure supercritical under any operating condition. The thermodynamic cycle optimization is performed by constraining the maximum cycle pressure to be 10 % higher than the critical pressure, while the turbine inlet temperature is constrained such that 1.1 · Tc < T IT < (Tsyngas,in − ∆Tpinch ). Among the fluids listed in Table 4.3, those making the supercritical cycle configuration possible are selected (R41, R125, R32, Propane, R134a, R227ea and R152a) and the resulting cycles optimized. The best performance is achieved with R134a as the working fluid, and the most relevant results of system simulations are listed in Table 4.6. The adoption of a mixture as the working fluid is beneficial also in the case of the supercritical cycle configuration. Similarly to the subcritical cycle calculation, the composition is varied in order to obtain a linear glide over the condensation of approximately 5 K. As shown in Table 4.6, a mixture of 65 % R134a and 35 % R236fa (mole fractions) allows achieving the highest heat recovery performance among all the simulated optimized ORC systems (see also Figure 4.9).

4.6

Conclusions

This chapter presents a study targeted to the optimization of ORC systems recovering low-grade thermal energy from pre-combustion CO2 capture plants. The starting point for this study is the design of a full-scale capture plant for the future Magnum power station. In this plant, a stream of 72 kg/s of syngas must be cooled down from approximately 140 ◦ C, and 45 kg/s of compressed CO2 at the same temperature level must be cooled before delivery for storage. The syngas thermal energy source is considered in detail here, also because the condensation 94

mol/mol ◦C kg/s m3 /s bar ◦C bar ◦C % % kW kW % % % % m2

Standard ORC unit base case R245fa 1 137/65.7 9.8 0.14 7.7 12.0 99.7 1.67 28.3 11.3 32.8 262 (13 units: 3406) 29.1 53.7 56.8 81.5 -

Subcrit. ORC PP pure fluid R236fa 1 137/38.3 188.6 1.19 9.0 20.0 103.5 2.86 26.5 12.5 39.2 4415 1009 24.5 56.0 66.2 90.8 5600

Subcrit. ORC PP fluid mixture R236fa/R152a 0.6/0.4 137/36.6 164.2 0.91 9.2 27.4 108.9 3.85 22.3/26.8 13 40.8 4614 1208 32.1 56.3 65.8 90.5 6700

Supercrit. ORC PP pure fluid R134a 1 137/40.4 156.4 0.68 7.5 44.7 125 6.97 26.6 13.1 40.9 4591 1185 25 59.3 66.4 90.7 5000

Supercrit. ORC PP fluid mixture R134a/R236fa 0.65/0.35 137/35.7 180.5 0.56 11.8 43.1 123 5.13 22.7/26.9 13.2 41.5 4699 1293 32.8 58.5 66.0 90.4 6800

Table 4.6: Comparison of performance and other relevant system parameters for the best optimized ORC systems, adopting different working fluids and cycle configurations.

Working fluid Composition Tsource (in/out) m˙ ORC V˙ORC,turb,in βv p1HX T IT pcond Tcond ηth ηex,cycle Pnet ∆Pnet ηex,cond ηex,1HX ηex,pump ηex,turb Acond

Parameters


95

Chapter 4

160 140

Temperature [°C]

120 100

Fluid Heat source Cooling water Process pheat = 43.1 bar pcond = 5.13 bar

80 60 40 20 0 1

1.1

1.2

1.3

1.4 1.5 1.6 Entropy [kJ/(kgK)]

1.7

1.8

1.9

Figure 4.9: Cycle diagram in the T − s thermodynamic plane of the optimized supercritical configuration in case of a R134a (0.65)/R236fa (0.35) mixture as the working fluid. The almost perfect thermal coupling of the triangular thermodynamic cycle with the thermal energy source and sink are notable.

of the syngas introduces a novelty with respect to standard ORC heat recovery applications. In this comprehensive simulation study three ORC plant configurations have been optimized and analysed, namely a system made of 13 standard ORC turbogenerators of 280 kWe each, which is taken as a benchmark case, an ORC plant also implementing the more conventional subcritical slightly superheated configuration, and an ORC plant implementing the supercritical configuration. The base case units adopts R245fa as the working fluid, while the working fluid for the two other cases has been optimized, by considering both pure fluids and mixtures. The best performance, 4.7 MWe net power output, is achieved by means of a supercritical cycle configuration and adopting a mixture of 0.65 R134a and 0.35 R236fa (mole fractions) as the working fluid. The base-case solution provides 3.4 MWe , while the other optimized power plants provide a benefit in terms of net power output ranging from 29.6 to 38.0 % (subcritical with pure fluid/mixture, supercritical with pure fluid/mixture). The exergy efficiency advantage of adopting a mixture as the working fluid compared to a pure substance is 4.1 % in case of the subcritical cycle configuration and 1.5 % in case of the supercritical cycle configuration. Given that both the thermal energy source and sink vary their temperature while interacting with the ORC system, both the adoption of a supercritical cycle and of a mixture working fluid 96


H2-rich Syngas

CO2 Absorber

ORC power plant A

CO2 Compressor

CO2 Product H

Separator WGS reactors

Flash drums

H

Solvent pump

Syngas Syngas compressor

ORC power plant B H

Process water H

H

IP steam

Figure 4.10: Process flow diagram of the pre-combustion CO2 capture plant showing integrated ORC systems for low-grade waste-heat recovery.

are beneficial from a thermodynamic point of view. The improved thermodynamic performance comes at the cost of additional heat transfer surface in the primary heat exchanger and condenser (20 − 36 % increase), and might imply some additional technical challenge due to supercritical heat transfer and flow in the primary heat exchanger and mixture vapour-liquid transition in the condenser (entrainment of the heavier component). The turbine is well within the limit of current design for these applications, though a somewhat higher volumetric expansion ratio must be accommodated. Several conclusions can be drawn from this study: - Among all the possible applications of low-temperature ORC power systems, the recovery of thermal energy from a pre-combustion CO2 capture power plant stands out in terms of efficiency and viability. The streams to be cooled are sulphur-free therefore their temperature can be lowered at will. The installation would be situated within the perimeter of an existing power plant, therefore many issues related to location, balance of plant, cooling water availability, etc., are inherently solved. The application of CO2 capture technology increases the value of electricity, therefore making the return on investment of a waste heat recovery solution even more attractive. It is estimated that approximately 10 % electric power can be saved over the total power consumption of the CO2 capture power plant if recovering wasteheat both downstream of the shifting section and downstream of the CO2 compressor. Furthermore, it is expected that ORC systems are able to follow load variations in case of dynamic operation of the decarbonised fossil-fuelled power plant. Flexible and prompt operation of IGCC power plants might become more relevant in future in order to balance the intermittent electricity generation from wind and solar power plants. 97

Chapter 4

- A methodological approach is described: general thermodynamic considerations are used to select candidate working fluids for the considered ORC plant configurations, thereby reducing the number of needed simulations in order to obtain optimal solutions. As for subcritical cycles, fluids having the critical temperature close to the thermal energy source are preferable, while also high reduced evaporation pressure improves the performance. In the case of supercritical cycles, fluids having the critical temperature approximately 30 K below the thermal energy source allow achieving the maximum amount of the power output. Mixtures as working fluids add a degree of freedom, allowing to tune the critical point and the condensation trajectory to the best thermodynamic result. - Syngas condensation is beneficial when coupled to an ORC power plant and prevents the typical pinch-point problem affecting the cooling of sensible heat sources. The syngas can be cooled to a low temperature, by means of an appropriate cycle, thus greatly increasing the amount of energy made available to the power converter. One of the challenges in performing investigations related to the applicability of ORC technology is that the optimization problem involves a number of variables of different nature. On the one hand the working fluid and the cycle must be optimized. On the other hand the feasibility strongly depends on the equipment used to implement the cycle. In order to automate further the preliminary design process, simulation tools incorporating preliminary equipment sizing (fluid machinery and heat exchangers) would allow to get faster to more accurate results, possibly encompassing a larger design envelope. As it is the case for other power systems, the ultimate goal of simulation science in this field could be a virtual prototyping tool.

98


Nomenclature A cp D e E˙ g G h j0 m˙ Nr p P Pr Re s T V˙ w x Z

= = = = = = = = = = = = = = = = = = = = =

Area, m2 Heat capacity, J kg−1 K−1 Tube diameter, m Specific exergy, J kg−1 Exergy flow rate, W Nominal gravitational acceleration, m s−1 Mass velocity, kg s−1 m−2 Specific enthalpy, J kg−1 Heat transfer factor Mass flow rate, kg s−1 Average number of tubes in a vertical tube row Pressure, Pa Mechanical power, W Prandtl number Reynolds number Entropy, J kg−1 K−1 Temperature, K Volume flow rate, m3 s−1 Specific work, J kg−1 Vapour quality Parameter

Greek symbols α βv Γ η λ µ ρ Φ

= = = = = = = =

Heat transfer coefficient, W m−1 K−1 Volume ratio Tube loading, kg m−1 s−1 Efficiency Thermal conductivity, W m−1 K−1 Viscosity, Pa s−1 Density, kg m−3 Thermal energy, W

= = = = = = = = = =

Ambient state Thermodynamic states Primary heat exchanger Critical Condenser; Condensation Cycle Dew point Effective Electrical Evaporator

Subscripts 0 1-4 1HX c cond cycle dew eff el evap

99

Chapter 4

ex fluid glide heat in is l m max net out pinch pump shell source syngas th tp tube turb v wall water

= = = = = = = = = = = = = = = = = = = = = = =

Exergy Fluid Glide Heating Inlet Isentropic Liquid phase Mixture Maximum Net Outlet Pinch point Pump Shell Source Syngas Thermal Two-phase Tube Turbine Vapour phase Wall Water

= = = = = = = = = = = =

Carbon capture and storage Chlorofluorocarbons Energy research Centre of the Netherlands Global warming potential Hydrochlorofluorocarbons Integrated gasification combined cycle Intergovernmental Panel on Climate Change Ozone depletion potential Organic Rankine cycle Perturbed chain - statistical associating fluid theory Power plant Turbine inlet temperature

Acronyms CCS CFC ECN GWP HCFC IGCC IPCC ODP ORC PC-SAFT PP TIT

100


References [1] Damen, K., Gnutek, R., Kaptein, J., Nannan, N. R., Oyarzun, B., Trapp, C., Colonna, P., van Dijk, E., Gross, J., and Bardow, A., 2011. “Developments in the pre-combustion CO2 capture pilot plant at the Buggenum IGCC”. Energy Procedia, 4(0), pp. 1214–1221. [2] Angelino, G., Gaia, M., and Macchi, E., 1984. “A review of Italian activity in the field of organic Rankine cycles”. In VDI Berichte - Proceedings of the International VDI Seminar, Vol. 539, VDI Verlag, pp. 465–482. [3] Saleh, B., Koglbauer, G., Wendland, M., and Fischer, J., 2007. “Working fluids for low-temperature organic Rankine cycles”. Energy, 32, pp. 1210–1221. [4] Madhawa Hettiarachchia, H., Golubovica, M., Woreka, W. M., and Ikegamib, Y., 2007. “Optimum design criteria for an Organic Rankine cycle using low-temperature geothermal heat sources”. Energy, 32, pp. 1698–1706. [5] Dai, Y., Wang, J., and Gao, L., 2009. “Parametric optimization and comparative study of organic Rankine cycle (ORC) for low grade waste heat recovery”. Energy Conversion and Management, 50, pp. 576–582. [6] Tchanche, B. F., Papadakis, G., Lambrinos, G., and Frangoudakis, A., 2009. “Fluid selection for a low-temperature solar organic Rankine cycle”. Applied Thermal Engineering, 29, pp. 2468–2476. [7] Colonna, P., Harinck, J., Rebay, S., and Guardone, A., 2008. “Real-gas effects in organic Rankine cycle turbine nozzles”. J. Propul. Power, 24(2), March–April, pp. 282–294. [8] Turboden, 2012. The Company and the Geothermal Applications. http://www. turboden.eu/en/public/downloads/11-COM.P-5-rev.5_GEOTERMIA_ENG_56654.pdf, October 3. [9] Schuster, A., Karellas, S., and Aumann, R., 2010. “Efficiency optimization potential in supercritical Organic Rankine Cycles”. Energ, 35, pp. 1033–1039. [10] Angelino, G., and Colonna, P., 1998. “Multicomponent Working Fluids for Organic Rankine cycles (ORCs)”. Energy, 23(6), pp. 449–463. [11] Angelino, G., and Colonna, P., 2000. “Organic Rankine Cycles for Energy Recovery from Molten Carbonate Fuel Cells”. In 35th Intersociety Energy Conversion Engineering Conference (IECEC), no. 2000-3052, AIAA, pp. 1–11. [12] Angelino, G., and Colonna, P., 2000. “Air cooled siloxane bottoming cycle for molten carbonate fuel cells”. In Fuel Cell Seminar, pp. 667–670. [13] Heberle, F., Preissinger, M., and Br¨ uggeman, D., 2012. “Zeotropic mixtures as working fluids in Organic Rankine Cycles for low-enthalpy geothermal resources”. Renewable Energy, 37, pp. 364–370.

101

Chapter 4

[14] Wang, X., and Zhao, L., 2009. “Analysis of zeotropic mixtures used in lowtemperature solar Rankine cycles for power generation”. Solar Energy, 83, pp. 605– 613. [15] Chen, H., Goswami, D. Y., Rahman, M. M., and Stefanakos, E. K., 2011. “A supercritical Rankine cycle using zeotropic mixture working fluids for the conversion of low-grade heat into power”. Energy, 36, pp. 549–555. [16] van der Stelt, T. P., Woudstra, N., and Colonna, P., 1985-2012. Cycle-Tempo: a program for thermodynamic modeling and optimization of energy conversion systems. Version 5.0. See also URL www.Cycle-Tempo.nl. [17] Colonna, P., van der Stelt, T., and Guardone, A., 2004-2012. FluidProp: a program for the estimation of thermophysical properties of fluids. Version 3.0. software. [18] Lemmon, E. W., McLinden, M. O., and Huber, M. L., 2002. “NIST Standard Reference Database 23: Reference Fluid Thermodynamic and Transport PropertiesREFPROP, Version 8.0”. National Institute of Standards and Technology, Standard Reference Data Program. [19] Lemmon, E. W., and Jacobsen, R. T., 2004. “Equations of State for Mixtures of R-32, R-125, R-134a, R-143a, and R-152a”. J. Phys. Chem. Ref. Data, Vol. 33, No. 2, pp. 593–620. [20] Gross, J., and Sadowski, G., 2001. “Perturbed-Chain SAFT: An Equation of State Based on a Perturbation Theory for Chain Molecules”. Industrial and Engineering Chemistry Research, 40, pp. 1244–1260. [21] Nannan, N. R., de Servi, C. M., van der Stelt, T., Colonna, P., , and Bardow, A., 2013. “An Equation of State Based on PC-SAFT for Physical Solvents Composed of Polyethylene Glycol Dimethylethers”. Industrial and Engineering Chemistry Research, 52, pp. 18401–18412. [22] American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc., 2009. ASHRAE Handbook - Fundamentals. [23] Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K., Tignor, M., Miller, H., and eds., 2007. Climate Change 2007: The Physical Science Basis. Tech. rep., Intergovernmental Panel on Climate Change (IPCC). [24] Intergovernmental Panel on Climate Change (IPCC), Contribution of Working Group I, 2001. Climate Change 2001: The Scientific Basis. Tech. rep. [25] Calm, J., and G.C.Hourahan, 2007. “Refrigerant Data Update”. Heating/Piping/Air Conditioning Engineering, 79(1), pp. 50–64. [26] Schuster, A., Karellas, S., Kakaras, E., and Spliethoff, H., 2009. “Energetic and economic investigation of Organic Rankine Cycle applications”. Applied Thermal Engineering, 29, pp. 1809–1817. [27] Brasz, J. J., 2011. “Keynote lecture: Low temperature / small capacity ORC system development”. In First International Seminar on ORC Power Systems, 22-23 September 2011.

102


[28] Bell, K. J., and Ghaly, M., 1973. “An Approximate Generalized Design Method for Multicomponent/Partial Condenser”. AIChE Symp. Ser. 69, pp. 72–79. [29] Sinnott, R., 2005. Coulson and Richardson’s chemical engineering, fourth ed., Vol. 6. Elsevier Butterworth-Heinemann. [30] Sieder, E. N., and Tate, G. E., 1936. “Heat transfer and pressure drop of liquids in tubes”. Ind. Eng. Chem., 28, p. 1429. [31] Taborek, J., 1983. Shell-and-Tube Heat Exchangers: Single-Phase Flow, Heat Exchanger Design Handbook, Chapter 3.3. Hemisphere, New York. [32] Kern, D. Q., 1950. Process heat transfer. McGraw-Hill Book Company.

103

“Wir sind das Volk” “We are the people”

Residents of Leipzig, Monday demonstrations, 1989/1990

5

Dynamic modelling and validation of a pre-combustion CO2 capture plant for control design1 This chapter documents the development of dynamic component as well as system models of a pre-combustion CO2 capture unit applied to integrated gasification combined cycle (IGCC) systems following an object-oriented modelling approach. The models have been implemented by means of the Modelica language into an open source software library. The fluid properties are computed with accurate thermodynamic models implemented within an in-house property package which is interfaced with the Modelica process models. Comprehensive model validation has been performed at component, sub-system and system level by comparison against experimental measurements obtained from various open- and closed loop transient tests at the CO2 capture pilot plant situated at the Buggenum IGCC power station. Results are in satisfactory agreement, considering the main process mass flow and temperature variables. In order to demonstrate the use of validated, dynamic process models, a control design study is presented whereby a control strategy based on feed-forward, feed-back and cascade control has been implemented and tested with the aim to improve dynamic performance of the capture unit. It can be concluded that prompt syngas load variation is a feasible operating mode for pre-combustion CO2 capture units featuring sophisticated control systems. The software library containing the validated component and system models serves as reliable foundation for the development of large-scale pre-combustion capture unit models. 1 Excerpts of this chapter are submitted for publication in: Trapp, C., Casella, F., and Colonna, P., 2014. Industrial and Engineering Chemistry Research.

Chapter 5

5.1

Introduction

The integration of a pre-combustion CO2 capture unit into the very complex gasification process and combined cycle power plant leads to many technical challenges especially regarding dynamic operation. Nowadays, dynamic performance of fossilfuelled power plants becomes increasingly important as the share of electricity produced by renewable energy sources, which is inherently unsteady, is steadily growing [1, 2]. Therefore, the capture process has to be able to follow frequent and fast load changes without restraining the performance of the IGCC power plant and violating environmental requirements, which are expected to be more strict in future. The desired dynamic performance of the capture unit can be achieved by adequate process, equipment and control system design. The state-of-the-art approach to this type of design problem is by means of dynamic process modelling and simulation, if possible accompanied with an experimental campaign to facilitate model validation. Simulations of transient operational scenarios are indispensable to compare the performance of different process configurations, to test different control strategies, including control parameter tuning, and to perform dynamic process optimization [3]. In order to investigate the transient performance of pre-combustion CO2 capture units detailed dynamic models of the capture process have been developed and validated by comparison with transient experimental data obtained from the pilot plant realized at the Buggenum IGCC power station to subsequently study process and control system performance during load variations. The literature dealing with transient performance and control of IGCC power plants with integrated CO2 capture units is scarce [4–6]. Few studies document model development and simulation of the IGCC processes without CO2 capture [3, 7, 8], and some authors focused only on dynamics of gasifiers [9–14] or auxiliaries, such as the air separation [15, 16] or the fuel system [17]. Heil et al. [18] presented the modelling of a pre-combustion CO2 removal unit utilizing methanol as physical solvent. Due to unavailability of experimental data, both steady-state and dynamics, validation of CO2 capture process models is still to be performed. The lack of validated process models is related to the fact that currently hardly any research project aiming at the demonstration of pre-combustion CO2 capture technology is undertaken, resulting in a small number of pilot plants of this sort, namely Buggenum (Vattenfall) [19], Eagle (J-Power) [20] and Puertollano (Elcogas) [21]. In comparison, research in the field of post-combustion CO2 capture is much more lively with a number of pilot and demonstration plants in operation [22]. However, data acquisition of these campaigns proved to be difficult due to challenges concerning the dynamic operation of post-combustion CO2 absorption plants [23] and, as a consequence, only a few publications document also dynamic model validation [24, 25]. Similar challenges affect also transient operation of the precombustion capture process, and therefore comprehensive experimental investi106

Dynamic modelling and validation of a pre-combustion CO2 capture plant for control design

gations accompanied with modelling activities to develop detailed and accurate dynamic models for process and control system design are still required. The novel aspects of this work are: a) Demonstration of the application of a state-of-the-art, object-oriented modelling paradigm to the pre-combustion CO2 capture process using a non-proprietary modelling language, which is tool independent and which can be used with proprietary or open source simulation environments. The fluid properties are computed with accurate thermodynamic models, which have been implemented within an in-house property package (free for academic use). The model libraries containing the developed process models are made available for academic purposes under an open source license agreement. b) Comprehensive dynamic model validation, whereby component, sub-system and system models are validated by comparison with experimental data obtained from a pilot plant. Various open-loop and closed-loop transient tests were performed, by monitoring the response to step-wise changes of different operational parameters, such as syngas load, syngas composition and solvent mass flow rate. c) The validated models have been used to investigate different control strategies aiming at the improvement of dynamic performance of the capture unit. This chapter is structured as follows: first the model development is explained in Section 5.2, covering modelling approach, utilized tools and a description of the Modelica models. The interface used to integrate the property package with the dynamic models is described in Section 5.3. The dynamic validation of the models including experiments and exemplary results are treated in Section 5.4, while the process analysis focusing on control strategy improvement is presented in Section 5.5. Finally, concluding remarks are given in Section 5.6.

5.2

Model development

The dynamic models were developed for the CO2 capture pilot plant configuration. The process is not described in this chapter for sake of conciseness. The reader is referred to the detailed explanation including a process flow diagram in Section 2.2.

5.2.1

Modelling approach

For prediction of transient process performance, non-linear dynamic models based on first principles were developed following a modular approach in order to master system complexity, e.g., the system is decomposed into suitable component models, which are connected through interfaces representing physical boundaries. Typically, zero-dimensional or one-dimensional component models were considered, which provide a sufficient degree of detail for accurate predictions of the transient system performance. The models were implemented using the object-oriented, equation-based Modelica language [26, 27]. Modelica is a non-proprietary modelling language, which is supported by various proprietary as well as open source simulation tools. Modelling features include, among others, reusability and extensibility, which allows an 107

Chapter 5

easy re-use of models developed by other researchers, or during previous projects, together with adaptation of existing models. Currently, an increasing number of open source and commercial Modelica libraries covering different engineering fields are available.

5.2.2

Thermophysical properties

The thermophysical properties of the two-phase multi-component syngas-water and sygnas-solvent mixtures are calculated with the perturbed chain - statistical associating fluid theory (PC-SAFT) equation of state (EoS) [28]. A detailed description of the thermodynamic model is given in Subsection 2.3.2. This EoS has been implemented, together with fast and robust algorithms, into an in-house property package [29], which is interfaced with the dynamic modelling tool. The use of external fluid property functions in Modelica process models imposes some restrictions to model development. A detailed discussion of these modelling aspects is given in Section 5.3.

5.2.3

Development of component models

The modelling of the CO2 capture process requires various component models. Whenever possible available Modelica library models were reused. For example, basic component models such as sinks, sources, valves, pressure drops, pumps, heat exchange and flow models, are taken from the ThermoPower library, [30, 31] and adapted in terms of their media models, which have been replaced with functional calls to the external property tool. New models were developed and implemented for the following components: • Flash vessel The process of phase separation is modelled under the assumption of thermodynamic equilibrium between the liquid and vapour phase at all times. The model describes the holdup of vapour and liquid with conservation equations applied to control volumes containing the two-phase fluid in thermodynamic equilibrium. Saturated conditions are assumed for the liquid and vapour outlet streams, therefore entrainment of liquid in the vapour flow is neglected. The flash vessel model is implemented as a storage component, hence flowfriction losses are not considered. The static pressure head due to the liquid level in the vessel is accounted for in the algebraic momentum balance. Heat transfer from the fluid (both vapour and liquid phase) to the vessel wall, storage of thermal energy in the wall, as well as thermal energy losses to the environment are neglected. Superficial condensation is thus also assumed to be negligible. • Water-gas shift (WGS) reactor The reaction of carbon monoxide with steam to produce carbon dioxide and hydrogen is described in a lumped-parameter model. The syngas entering 108


and leaving the reactor is an ideal-gas mixture containing CO, CO2 , H2 , H2 O and N2 . Other trace constituents are neglected. The model accounts only for the WGS reaction. Intermediate reactions involving other chemical species are neglected. The reactor model is subdivided into five sub-models: reaction node, mixing gas volume, convective heat transfer, thermal storage and pressure drop. The object diagram of the model is depicted in Figure 5.1.

Thermal Storage Convective Heat Transfer CO

H2

Pressure Drop Reaction Node Mixing Gas Volume

Figure 5.1: Object diagram of reactor component.

The WGS reaction takes place in an infinitesimally small volume (reaction node) representing one finite discretization of the catalyst, and it is assumed that it reaches thermodynamic equilibrium. The storage of mass and energy in the bulk phase of the reactor are modelled in a perfectly mixed volume (mixing gas volume) which receives the reaction products. This control volume exchanges heat with the catalyst by means of convection. The storage model describes the storage of thermal energy in the catalyst. Heat transfer to the environment is neglected. The water-gas shift reactor is discretized in the axial direction by an array of reactor models in order to correctly describe the gradual changes in reactor outlet conditions during transient operation. Changes in the reactor inlet conditions reach the reactor outlet with a delay due to thermal storage in the catalyst, which cannot be represented with a 0-dimensional model due to the high number of transfer units between the gas and the catalyst itself. This one-dimensional discretization does not however represent the actual axial reactor profile as equilibrium conditions are assumed in each reactor model element for simplicity. At steady-state conditions the equilibrium temperature is reached at each discretization of the catalyst, which also determines the temperature-dependent WGS reaction. • Pilot plant specific heater and cooler components Various electrical components for evaporation, superheating, cooling and condensation were in particular developed for the pilot plant and will not be part of a large-scale plant with this specific configuration. The models were typically subdivided, if applicable, in flow models, heat transfer models and thermal storage models. Whenever possible models from the ThermoPower library were used, typically in case the medium was water or ideal gas, or 109

Chapter 5

Storage module Absorption column

Resistive module

sink_va…

Storage module

sourceL…

Resistive module Storage module Resistive module Storage module

P sourceP2

Resistive module

Sump sinkW_li…

Figure 5.2: Model structure of the absorption column and sump.

adapted. The heat transfer coefficients were either computed with specific heat transfer correlations or tuned to experimental data. • Absorption column and sump The model of the packed column for physical absorption (no chemical reactions) is discretized in theoretical stages in the axial direction, and countercurrent flow of the vapour and liquid is assumed. Each stage is modelled by an equivalent tray module (only-storage model), together with a resistive module, which are connected in series to form a column model as shown in Figure 5.2. A detailed description of the absorber column model is given in the next chapter in Section 6.2. The sump model is implemented as a storage component accounting for holdup of liquid. The static pressure head due the liquid level in the sump is modelled with an algebraic momentum balance. Storage of thermal energy in the sump wall and heat losses to the environment are neglected. The sub-system models (e.g., water-gas shift section, absorption and solvent regeneration section) and system model of the capture plant are assembled by connection of individual component models. Component as well as system models are implemented in a Modelica library which is made available for academic use.

5.3

Modelica-FluidProp interface

The main challenge of the dynamic model development is related to the computation of fluid properties, in particular phase equilibria, due to the fact that highly non-ideal, two-phase, multi-component fluids are involved in the capture process. Currently, Modelica medium models are not available for this type of fluids. One possibility is to implement required medium models, just as the process models, in the Modelica language, with the advantage to be able to perform effi110


cient simulations as the code can be optimized, provided the equation of state is written in a declarative way, which might not always be possible. However, the implementation of non-ideal fluid property models is rather time-consuming and not trivial, as dedicated solution algorithms might be required for efficiency and numerical robustness. The other possibility is to make use of available thermophysical property packages and interface these tools with Modelica. Employing external tools for the computation of fluid properties provides some advantages: 1) typically the property software employs dedicated algorithms for fast and robust calculations of the fluid properties, 2) the property package can be interfaced with a wide variety of engineering software tools (e.g., steady-state and dynamic system modelling, component design, computational fluid dynamics, etc.) allowing for the use of the same thermophysical properties, thus eliminating one common source of uncertainty, 3) a wide range of pure fluids and fluid mixtures described with suitable and accurate equation of states are available in the property package. This concept was successfully demonstrated with the ExternalMedia library [32] which supports two-phase, single-substance fluids all compatible to the Modelica.Media interface. However, multi-component fluids as required for the CO2 capture process are not supported, because the interface for multi-phase fluid mixtures is limited to two-phase pure (or pseudo-pure) components. For this reason, a prototype for an interface supporting two-phase, multicomponent fluids was developed and tested for the modelling and simulation of the CO2 capture process. The objective of the work documented in this section is to demonstrate the feasibility of modelling such a chemical process with Modelica by making use of external fluid property code and to indicate limitations concerning the modelling approach as well as to discuss possibilities for the improvement of the computational efficiency. Finally, recommendations shall be drawn for the design of a generic interface to external fluid property code.

5.3.1

Library architecture

The ModelicaFluidProp library provides the functional interface that allows to integrate external fluid property codes into Modelica models. The library contains two parts, the Modelica front-end which makes various functions available for the calculation of different property sets (for instance, ‘AllProps’ or ‘TwoPhaseDeriv’) and the C/C++ back-end, containing C++ objects that carry out the interfacing between the Modelica level and the external software tool. The Modelica library contains a generic package FluidPropMedium. The actual external fluid property code is specified by setting values to constants such as ModelName, which defines the name of the external library, nComp, which specifies the number of fluid constituents and Comp, which defines the name of the individual constituents. The external medium model can be used in any component model and is not extending any medium package from the standard Modelica 111

Chapter 5

library. The implemented set of functions in the FluidPropMedium package mirrors one-to-one property functions available in the external property tool which are interfaced with corresponding C-functions defined in the C interface layer. In the following, the working principle of the library is explained, based on the exemplary code below: import SI = Modelica.SIunits ; package SyngasDEPEG extends ModelicaFluidProp.FluidPropMedium( ModelName=" PC-SAFT " , nComp=6, Comp={" carbon monoxide " ," hydrogen " , " carbon dioxide " ," water " ," nitrogen " , " DEPEG "}) end SyngasDEPEG ; model Example SyngasDEPEG.AllPropsOut prop ; SI.Temperature T; SI.Pressure P; SI.SpecificEnthaly h; SI.Density d; SI.MoleFraction Y[SyngasDEPEG.nComp]= SyngasDEPEG.reference_X ; SI.MoleFraction Yliq[nComp] " Molar fractions of liquid phase "; SI.MoleFraction Yvap[nComp] " Molar fractions of vapour phase "; equation (prop , Yliq , Yvap) = Medium.AllProps(" PT " ,P ,T ,Y); h = prop.h ; d = prop.d ; end Example ;

The AllProps function of the FluidPropMedium package calls the corresponding C function of the interface and passes the specification of the thermodynamic state (‘PT’), the values of pressure, temperature and composition as well as the constants for medium identification. The interface function handles the creation of an object for the external property code and the execution of the solver to compute the required properties. The calculated fluid properties are passed via the prop record to the Modelica code. The AllProps function returns all primary thermodynamic properties such as pressure, temperature, specific volume, enthalpy, entropy, internal energy, etc., which can be computed with hardly any additional computational cost when solving the equation of state. Secondary thermodynamic properties such as heat capacity, speed of sound, various single-phase partial derivatives and transport properties are computed with a separate function as these properties are less often needed and require additional computations. The computationally expensive two-phase partial derivatives are combined in another property function. The arrangement of primary and secondary fluid properties in meaningful functions allows for a flexible use and avoids unnecessary repeated computations. 112


5.3.2

Lessons learned

The use of external fluid property functions in Modelica process models puts some restrictions on the model development. Specific attention requires the formulation of the differential model equations, the choice of state variables and the causality of the system model. These issues are addressed in the following. 5.3.2.1

Choice of state variables

For dynamic modelling of thermo-physical systems not only the choice of the system state variables is of importance but also the selection of the thermodynamic states used to determine fluid properties. In general, the system state variables should allow for an easy computation of the thermodynamic properties required to determine the system performance, and the thermodynamic state variables should unambiguously determine the fluid state. Further, the choice of system state variables can have a significant influence on ease of initialization, numerical robustness and computational speed. In the following, three different possibilities for the choice of state variables and their influence on resulting differential and algebraic equation (DAE) system as well as simulations speed are analysed. As an example, dynamic mass and energy balances for describing storage of vapour and liquid in a volume under the assumption of thermodynamic equilibrium are used. These equations can be found in the flash vessel and absorber tray models. Explicit system state variables M, u, Xi In the most simple way the dynamic mass and energy balance can be formulated as dM dXi Xi + M = win Xin,i − wliq Xliq,i − wvap Xvap,i , dt dt M

du dM +u = win hin − wliq hliq − wvap hvap , dt dt

(5.1)

(5.2)

where M is the total mass, u is the internal energy, Xi is the component mass fraction vector whereby the subscript i ranges from 1 to the number of species in the mixture, w is the mass flow rate and h is the specific enthalpy. The left-hand side of both equations has been developed according to the chain rule in order to obtain separate differential terms for each state variable. A formulation purely based on moles is also possible and might be the preferred choice as conversions between mole and mass are avoided. Assuming M, u and Xi are selected as state variables and no other variables than the states appear under the time derivative the system solution can be obtained straightforward. This applies, for example, to a single flash vessel model where the index of the DAE system is 1. The case of higher index problems, e.g., when 113

Chapter 5

dealing with a model of two flash vessels connected with a zero pressure drop is discussed below. The mass and energy equations represent a compact and very declarative formulation of the conservation equations. The pressure P, the temperature T and the mass composition Xi are chosen as thermodynamic states, therefore prop = prop(P, T, Xi ). For this state selection property computations for two-phase mixtures are most efficient and robust with the used external media library. As the system states and thermodynamic states differ, implicit equations must be solved at each time step in order to determine the thermodynamic state variables. Consequently, with the common solution strategy followed by most Modelica tools, i.e., first turn the DAE’s into ordinary differential equations (ODE’s) and then integrate over time, the computational speed is affected by the iterations on the property function. These property computations are already the most expensive part of the entire simulation. Implicit system state variables P, T , Xi The Modelica language features the option to change states by means of the StateSelect attribute, without changing the declarative formulation of the dynamic mass and energy balance. Hence, pressure, temperature and composition can be selected as preferred states in accordance to the thermodynamic states with the aim to reduce iterations during the solution procedure. However, as the total mass and internal energy remain under the time derivative and not being states, the tool attempts to symbolically differentiate M and u with respect to the states P, T , Xi in order to establish a relationship between the old states, which have been discarded, and the newly selected states. This procedure fails as the external property call cannot be differentiated. Explicit system state variables P, T , Xi du In order to facilitate a feasible solution an explicit expression for dM dt and dt needs dXi dP dT to be provided as a function of the state variable derivatives dt , dt and dt . The needed time derivatives therefore read " # n dM ∂v dP ∂v dT ∂v dXi = −Mρ + + , (5.3) dt ∂P T,X dt ∂T P,X dt ∂Xi P,T,Xj6=i dt

∑

i=1

du = dt

∂u ∂P

T,X

dP + dt

∂u ∂T

P,X

dT + dt

n

∑

i=1

∂u ∂Xi

P,T,Xj6=i

dXi . dt

(5.4)

The partial derivatives of u and v with respect to P, T , Xi need to be provided by the external media library. Two-phase partial derivatives of mixtures, which are based on properties obtained from the phase equilibrium calculations, are commonly not available among the typical thermophysical fluid properties and therefore have 114


been specifically implemented in the framework of this model development. These mixture derivatives are currently computed numerically. Future work might consider the analytical formulation as implemented, for example, in Multiflash [33]. However, also the calculation based on analytical expressions is expected to be computational expensive [34]. In Figure 5.3 the simulation performance of a flash vessel with system states M, u, Xi and state variables P, T , Xi is compared. The simulation of the latter model requires 90 % less functional calls to the external property library promoted by the same choice for system and thermodynamic state variables. However, considering the computational time the latter model only leads to an improvement of 40 %. This is explained by the additional calculation of partial derivatives required when changing the states from M, u, Xi to P, T , Xi . 50 Flash vessel MuXi Flash vessel PTXi

8000

Total simulation time [s]

Number of property calls [−]

10000

6000 4000 2000 0 0

500

1000 Time [s]

1500

2000

(a) Number of property calls.

40

Flash vessel MuXi Flash vessel PTXi

30 20 10 0 0

500

1000 Time [s]

1500

2000

(b) Total simulation time.

Figure 5.3: Comparison of flash vessel models with state variables M, u, Xi and P, T , Xi .

The observed difference in computational speed between both choices of state variables might also be related to the solution strategy employed by the software tool. The adopted tool [35] for this project obtains the solution of the DAE system by iterating on a nested loop, and solving the ODE’s in the outer and the algebraic equations in the inner loop. It might be possible that using a DAE solver directly on the original DAE system in this case might lead to a much smaller difference in computational speed between both solutions. However, a direct solution strategy might entail more trouble during initialization and shorter time steps to achieve convergence might be required. A hybrid strategy where both approaches are applied, dynamically switched, might be interesting to explore. 5.3.2.2

Developing index-1 models

Considering a model containing two flash vessels, component mass and energy balance represented by Equation 5.1 and Equation 5.2, with zero or constant 115

Chapter 5

pressure drop, then the index of the DAE system is larger than 1, assuming M, u and Xi are chosen as state variables. The solution of a DAE system with higher index is commonly obtained by symbolic manipulation of the equations system in order to reduce its index to 1. Current simulation tools implementing the Modelica language employ state-ofthe-art techniques for index reduction. Difficulties during index reduction might arise in case fluid property calculations must be symbolically differentiated. If the fluid correlations or equation of state are implemented as a Modelica media library, possibly accompanied with annotations to compute its time derivatives, symbolic manipulation can be performed by the tool resulting in a successful index reduction. However, in case external media libraries are used, interfaced with the Modelica models, index reduction fails as external functions cannot be manipulated. This can be resolved by either supplying any time derivatives required for index reduction or by developing the Modelica models such that the DAE system remains in the index-1 form. The latter approach has been followed for the dynamic modelling of the CO2 capture pilot plant. Causal versus acausal approach Following the acausal modelling approach, which is fully supported by Modelica, connections of sub-models might not respect the causality leading to a system of DAE’s of higher index (example of two flash vessels described above). Therefore, a mixed approach is applied during the model development. The sub-models and interfaces are defined and developed in an object-oriented manner such that the models can be employed in an acausal context. However, the system has been analysed and decomposed into subsystems following the causal modelling paradigm in order to avoid the occurrence of higher order DAE systems [36]: • Identification of bilaterally coupled variables of the models. • Discretisation of the model in resistive and storage modules, namely solving the conservation laws for flow and potential in different control volumes. • Connect the resistive to the storage modules and vice versa such that potential variables are inputs of the resistive and outputs of the storage modules. Flow variables are inputs to the storage and outputs to the resistive modules. By following a more causal development and arrangement of the models the DAE system can be maintained in the index-1 form which allows for a straightforward use of external property functions. The disadvantage of this approach is that is poses restrictions on the connection of modules. It might be necessary to include dummy modules with no process functionality in order to maintain the resistive-storage structure. 116


5.3.2.3

Improvement of computational time

Based on the current experience obtained from the modelling of the CO2 capture process, the computation of thermodynamic fluid properties, in particular phase equilibria, accounts for the main share of the simulation time (in the order of 95 %). In addition, much more complex and hence computational expensive thermodynamic models are required when dealing with highly non-ideal, multi-component mixtures in comparison to fluid models for water or ideal gas. Therefore, an appropriate and smart use of property functions should be considered during the entire model development process. In the following, suggestions are presented which might significantly contribute to a successful convergence and more efficient simulations. Choice of thermodynamic states For an efficient use of external property functions in Modelica, it is necessary to have knowledge on the property calculations performed in the external tool. The thermodynamic state can be determined for different choices of independent variables, for example ‘PT’, ‘Pv’, ‘Ph’, ‘Pq’, ‘Tv’, ‘Tq’, ‘uv’, etc., where P is the pressure, T is the temperature, v the specific volume, h is the specific enthalpy, q is the vapour quality and u is the specific internal energy. In case of mixtures the component fraction vector Xi is added to set of independent variables. The external tool employs an isothermal (‘PT’) or an isenthalpic (‘Ph’) flash algorithm developed by Michelsen [37] for the PC-SAFT and cubic EoS’s, which are robust and reliable equilibrium calculations based on the minimization of Gibbs free energy. If other thermodynamic states are chosen as input, then the solution is obtained by iteration on either the ‘PT’- or ‘Ph’-flash calculation. Any computations including the vapour quality q as input use a bubble/dew point calculation, which is much more difficult to perform and is far less robust than the flash algorithms. To conclude, ‘PT’ or ‘Ph’ are recommended as thermodynamic states as these inputs allow for fast and robust fluid property calculations. Single and two-phase property calculations The flash algorithm first determines the vapour and liquid composition of the fluid (computation of phase equilibrium), which indicates if the fluid state is in the twophase or the single phase region. Then the vapour and liquid properties such as v, h, s, and u are computed. If applicable two-phase properties are calculated based on the single phase properties using appropriate mixing rules. The initial determination of the liquid and vapour composition is computational very expensive and hence this step should be omitted if it is known a priori that the fluid is present in the single phase. Therefore, the option of skipping the flash calculation and just performing a single phase calculation was implemented by using, for example, ‘PT-1ph’ as input specification. 117

Chapter 5

PID

Vapour pressure sink

p

P

Vapour control valve

Flash vessel 2

PID

Liquid pressure sink

Vapour/liquid flow source

P

Liquid control valve

Figure 5.4: Object diagram of a flash vessel with vapour and liquid control valves. Case Reference (Valves ‘PT’) Valves ‘PT-1ph’ Vapour density transfer Valves ‘PT-1ph’, CallID

Valves AllProps [s]

Vessel AllProps [s]

Vessel TwoPhaseDeriv [s]

Total time [s]

107.6 11.2 7.2 11.1

5.2 5.2 5.2 5.2

21.5 21.5 21.5 4.5

134.3 37.9 33.9 20.8

Table 5.1: Simulation results of model with flash vessel and two control valves simulated for 2000 seconds.

The flash vessel component is one example where this ability to explicitly indicate the fluid phase finds application. For the conditions in the vessel two-phase properties are required whereas in the valves connected to the vapour and liquid outlet single phase properties are sufficient. Experience has shown, that by using single phase property computations where applicable throughout the process the simulation time can be reduced significantly (see Table 5.1). Redundant property calls The model of a simple phase separation as depicted in Figure 5.4 provides another example on how the computational efficiency can be improved by optimizing the property calculations. The flash vessel model contains a property call which determines required primary two-phase, liquid and vapour properties, such as h2ph , d2ph , hliq , dliq , hvap and dvap , prop2ph,liq,vap = prop(P, T, Xi,2ph ).

(5.5)

In the valve model connected to the vapour outlet of the flash vessel, the vapour density dvap is required to close the set of conservation equations by propvap = prop(P, T, Xi,vap ). 118

(5.6)


It is obvious that, under the assumption of adiabatic operation and no frictional losses, both property computations provide the same result for the vapour density. However, the property calls are different due to the fact that in the vessel the two-phase mixture composition and in the valve the vapour composition is used as input. During the process of translating the model into a set of solvable equations, the compiler will not realize that the second property call is in principle redundant. One solution to overcome this issue is to transfer the required density via additional output and input connectors from the flash vessel to the valve. This solution has been implemented as optional choice, which can be activated by the modeller. Another possibility would be the use of conditional connectors which transfer next to the flow and potential variables possibly an array of all fluid properties. The improvement in computational time is summarized in Table 5.1. In comparison to the use of single phase property calls in the vapour valve the improvement is rather small. However, the aim is not to obtain the most efficient simulations but to demonstrate a modelling approach which might contribute to efficiency. Approximation of two-phase partial derivatives One of the most time-consuming part of the property calculations is the computation of two-phase partial derivatives, which are for example required in the flash vessel component and absorber tray model. The results of the partial derivatives are used in the mass and energy balances. Various tests indicated that if the change in the absolute value of the thermodynamic states during the simulations is small the resulting change in the value of the partial derivatives has hardly any impact on the simulation results. Therefore, the following procedure has been implemented for the models of the CO2 capture process. An additional variable (CallID) has been added to each two-phase component assigning each partial derivative function a unique identification. With this identification it is possible to distinguishably store simulation results in the ModelicaFluidProp interface of the different two-phase derivative calls throughout the modelled process (e.g., LP Vessel CallID = 1, MP Vessel CallID = 2,...). Assuming the derivative function with the ID 1 is executed at a certain simulation time instance, then a check is performed if the difference between the current and the values from the previous time instance (which have been stored in the interface) are below a defined threshold. If that is the case, no property computation for the partial derivatives is performed but the stored results of the derivatives from the previous time instance are returned straight to Modelica. In case the threshold is exceeded, then a normal property calculation is performed with the external tool and the previous results stored in the interface are overwritten. This procedure has the benefit that computational time is saved if the change in the absolute value of the partial derivatives is marginal and thus has no impact on the solution. When modelling complex processes involving two-phase multi-component fluids a signifi119

Chapter 5

cant reduction in computational time can be obtained. Exemplarily a comparison is provided for the model of the flash vessel with two control valves (see Table 5.1). The time spent for derivative computations reduces from 21.5 to 4.5 seconds for such a simple model.

5.3.3

Recommendations for a future interface

The presented model development approach leads to solvable models by making a smart choice for the thermodynamic and system state variables and by manually applying measures in order to keep the system in index-1 form. However, for reasons of numerical robustness, simulation speed and ease of initialization, a different choice of state variables might be more convenient than the one where the differentiated variables are used as states. In order to allow for flexible state variable change and automatic index reduction, partial derivatives of the thermodynamic properties are essential when using external tools, as demonstrated in this chapter. The goal is to design a Modelica library that interfaces to external property packages, whereby Modelica tools can automatically compute the total time derivatives of each variables in a set A (e.g., density, specific energy, specific enthalpy, etc.) with respect to any meaningful subset of variables in a set B (e.g., pressure, temperature, specific enthalpy, etc.) that uniquely identifies the thermodynamic properties of the fluid, including multi-component fluids and two-phase mixtures. This is required to successfully carry out the index reduction and/or state variable change task automatically. A first attempt on how to perform automated state variable change is presented by Wellner et al. [38]. At a higher level, this requires setting up a Modelica infrastructure where annotations point to the appropriate functions to compute all the required derivatives. At a lower level, it has to be ensured that the external property package can compute all required derivatives efficiently, i.e., by avoiding unnecessary duplicate computations.

5.4

Dynamic model validation

In general terms, the dynamic model validation discussed here aims to demonstrate if the developed CO2 capture process models, with their identified relevant phenomena and, assumptions and hypothesis, are a sufficient representation of the actual process in relation to the purpose of the model. The accuracy of the simulated process transients must therefore be sufficient in order to perform control-strategy design, and to improve the dynamic operation of the plant.

5.4.1

Validation approach, experiments and results

The fully instrumented pilot plant was designed for operational flexibility in order to investigate the influence of various operating conditions, both steady-state 120


and dynamic, on component and system performance, and therefore facilitated comprehensive model validation. Moreover, with the support and knowledge of the experienced plant operators, a wide range of different experimental tests could be executed. These possibilities allowed to perform dynamic validation at three different levels: component, sub-system and system level. In a first step, the individual, newly developed component models of the CO2 capture process were validated, such as the water-gas shift reactor, various heater and cooler components and the absorber column. For each component, openloop and closed-loop experiments were performed. During open-loop tests process dynamics are analysed without the control system in operation, whereas in closedloop tests, process and control system performance are assessed for a realistic operational scenario. The open-loop tests (e.g., step responses) are suitable to reveal the inherent dynamics of the process, and make sure that it is correctly captured by the dynamic model. Unfortunately, they cannot always be performed safely, or conveniently, on the real plant. Closed-loop tests are easier to carry out, but only provide relevant dynamic information in the frequency range around the controller’s crossover frequency. On the one hand, this might in fact be better than open-loop tests: if the controller is very fast, the closed-loop behaviour depends on the fast dynamics of the process, which might be hard to discern clearly from open-loop tests, such as step responses. On the other hand, if the controller installed on the plant is very conservative (i.e., slow), then closed-loop tests might fail to reveal the dynamics of interest in case more aggressive controllers are designed, based on the model. In general, a combination of both tests is the best option in order to comprehensively assess the quality of the model. The linear system response was evaluated by applying small perturbations to the input variables starting at steady-state operation, though large enough to obtain an acceptable signal-to-noise ratio. For relatively small changes the expected response of an upward and downward step change is symmetrical in terms of the main dynamic parameters of the transient, such as time and value of maximum overshoot, settling time, presence and damping of oscillations, etc. During transient operation all input variables other than the perturbation variable should be maintained constant. In case this is not possible due to process limitations or other uncontrollable disturbances, e.g., changes in environmental conditions, then these variables are prescribed inputs for the dynamic model. The process measurements are obtained from the distributed control system of the pilot plant and are transferred for off-line data analysis to suitable data processing tools. Systematic biases in the measurements identified during the data reconciliation as presented in Chapter 2 were removed from the experimental data. The individual CO2 capture process components were validated following the outlined procedure. The validation of a superheater component model is briefly described here as an example. 121

355

75

350

70

345

65

340 335 330 0

60

Tmodel Texperiment

55

Duty 15

30 45 Time [min]

60

50

355

80

350

78

345

76 Texperiment

340

74

Duty [%]

80 Outlet temperature [°C]

360

Duty [%]

Outlet temperature [°C]

Chapter 5

Tmodel Dutyexperiment 72

335 330 0

Dutymodel 15

30

(a)

45 60 Time [min]

75

90

70 105

(b)

Figure 5.5: Validation of superheater component, see Figure 2.1: comparison of measurements and simulation results for outlet temperature and duty. a) Openloop test: step decrease in heater duty. b) Closed-loop test: step increase in temperature set point.

Example for component validation: superheater The main phenomena modelled by the electric superheater component, see Figure 2.1, are the storage of thermal energy and the heat transfer. The best manner to demonstrate that these phenomena are modelled with sufficient accuracy is by analysing if the superheater performance, in terms of response in outlet temperature due to a change in heat duty, is satisfactorily reproduced by the model. The open-loop and closed-loop experiments were designed such that the inlet conditions in terms of mass flow, temperature, pressure and composition remained constant during the experiment. This required to operate some upstream control loops in manual mode. During the open-loop experiment a downward step was applied directly to the heater duty without the PI controller in operation. The step in power corresponds to a change in superheater outlet temperature of approximately 10 ◦ C. During the closed-loop experiment the superheater temperature set point was changed stepwise by 10 ◦ C. The temperature controller reacts by adjusting the duty in order to reach the desired set point value. The model validation is performed by comparing measurements of the superheater outlet temperature transient as response to the applied perturbations to simulation results obtained with the superheater component models. Figure 5.5(a) shows the results of the temperature transient for the open-loop step change in superheater duty. A satisfactory match between the measurements and the model predictions is observed considering the main dynamic parameters. The temperature measurements are not smooth and the step-like changes are due to the fact that a measurement is recorded if the variable change in absolute value exceeds a defined and adjustable threshold. In Figure 5.5(b) the closed-loop tran122


Controlled variable

Control variable

Set point

Syngas mass flow rate Outlet temperature of heater Level of 1st vessel Level of 2nd vessel

Opening of syngas control valve Heater duty Opening of water control valve Opening of liquid control valve of 1st vessel Reboiler duty

1100 kg/ha 172 ◦ C 1300 mm 800 mm

Mass flow rate of reactor 1 Inlet temperature of reactor 1 Inlet temperature of reactor 2 Inlet temperature of reactor 3

Superheater duty Set point for opening of quench control valve Fan speed of cooler

Internally calculatedb 340 ◦ C 340 ◦ C 340 ◦ C

a

For the presented test run this control loop was operated in manual mode, hence the PI controller is not depicted in Figure 5.6. b This loop is used to control the ratio of water/syngas (set point) by calculating the expected mass flow rate of reactor 1 based on the measured syngas and quench flow. Table 5.2: Control loops within the WGS section (see Figure 5.6).

sient of the temperature and of the heat duty are compared for a step increase in the temperature set point. For both, the process and the control variable, simulation results and experimental data show good agreement. The measurements of the heat duty are much smoother than the temperature recordings. In a next step, sub-system models were developed based on the validated component models. A sub-system, such as the water-gas shift section or the absorption and solvent regeneration section, cannot be operated without control due to plant safety and stability. Therefore, partial open-loop experiments were designed, whereby only control loops which do not compromise the safe operation were put into manual mode. In the following, the validation of the water-gas shift section model is discussed as an example for sub-system validation. The detailed model validation of the absorption and solvent regeneration section model is demonstrated in Chapter 6. Example for sub-system validation: water-gas shift section During the partial open-loop experiment an upward step change was applied directly to the syngas control valve without the syngas controller in operation. All other controllers of the WGS section were kept in automatic mode and are summarized for a better understanding in Table 5.2. The step in valve opening corresponds to a change in syngas flow of approximately 100 kg/h in absolute and 10 % in relative terms. The scheme of the sub-system model used for the dynamic simulations is depicted in Figure 5.6. 123

Chapter 5

Quench flow control valve

w FP

Syngas flow

Quench flow

Syngas flow control valve

P source…

sensW_sy…

pressDro… ELF20AC1

Water flow

SetPoint_T 1300

Tab_P…

pI…

2nd vessel

sensT

SetPoint_L… 1300

pressD_EL…

pI…

offset…

pressD…

pID2

T

Q

pID1

T

sensT2

sensT_Mix Q

ELF50AA050

340

Reactor 1

sensT1

sensW_R1fl…

FP

SP2 340

SP1

pID

T

durati…

source…

pID…

SP 340

SetPoi…

ELG40AA1

P

w sensW1

ELF31AA…

sensW2

Heater

w

Mode…

Q

Q

P

Q

pressDrop1

w

ELF50AC…

Superheater

Reactor1

sinkP1 Reactor2

Reactor 2

Reactor3

Reactor 3

press…

ELF40BB1

sinkW_liq…

pressD…

ELF20BB1

1st vessel ELF40AA1 SetPoint_L… 800

ELF40AC1

Reboiler

pI…

pressD…

pID…

Reactor 1 flow control

Figure 5.6: Object diagram of the water-gas shift section model (rectifier, separator and booster compressor not included).

The pressures at the syngas inlet, at the reaction water valve inlet, and at the reactor 3 outlet are input variables of the dynamic model. The first two pressures can be set constant for the simulation. The back pressure of reactor 3 varies during the experiment, therefore the actual pressure measurements, which are depicted in Figure 5.7(a), are provided as model input. The perturbation was applied at t = 50 minutes starting from an initial syngas load of 1000 kg/h. Due to the step-opening of the syngas control valve, the mass flow rate of reactor 1 increases almost instantaneously pulling more syngas-water mixture from the downstream components, see Figure 5.7(b). This results in a fast increase in the syngas inlet flow, Figure 5.7(d). Furthermore, the increased mass flow rate of reactor 1, now containing a larger total amount of thermal energy, causes an initial increase of the inlet temperature of reactor 2, see Figure 5.8(c), when mixing at the outlet of reactor 1 with the quench flow, which remained rather unchanged in terms of flow and temperature during this initial transient. The quench flow control opens the quench valve (Figure 5.6) in order to maintain the inlet temperature of reactor 2. This causes an increase in the quench flow and decrease in the mass flow rate of reactor 1. The operating condition of the system starts to fluctuate whereby almost all process variables are influenced. The controllers stabilize the operation such that the oscillations are dampened and the new steady-state is approached at t = 200 minutes. The model predictions for the mass flow rate of reactor 1, of the quench stream and syngas stream compare well with the experimental results (see Figure 5.7), in particular the initial response in terms of rise time and maximum overshoot are predicted accurately. The presence of oscillations is captured by the model, however the damping is overestimated, and this is discussed in more detail in the following. Also for the temperature of the quench and the inlet temperature of the superheater a satisfactory agreement between the simulation results and the measurements is achieved in terms of the main dynamic parameters, see Figure 5.7(e) 124

Back pressure of reactor 3 [bar]

16.9 16.85 16.8 16.75 16.7 16.65 16.6 0

Experiment Model 25

50

Mass flow rate of reactor 1 [kg/h]


1350

1300

1250

1200

1150 0

75 100 125 150 175 200 Time [min]

(a) Back pressure of reactor 3.

1050 1000 950 900

Experiment Model 25

50

Syngas mass flow rate [kg/h]

Quench mass flow rate [kg/h]

75 100 125 150 175 200 Time [min]

1200 1150 1100 1050 1000 950 0

75 100 125 150 175 200 Time [min]

174 173 172 171 170 169

Experiment Model 25

50

75 100 125 150 175 200 Time [min]

(e) Quench flow temperature.

Experiment Model 25

50

75 100 125 150 175 200 Time [min]

(d) Syngas mass flow rate. Inlet temperature of superheater [°C]

(c) Quench mass flow rate.

Quench flow temperature [°C]

50

1250

1100

168 0

25

(b) Mass flow rate of reactor 1.

1150

850 0

Experiment Model

195 194.5 194 193.5 193 192.5 192 0

Experiment Model 25

50

75 100 125 150 175 200 Time [min]

(f) Inlet temperature of superheater.

Figure 5.7: WGS section model validation: comparison of measurements and simulation results for a step increase in syngas control valve opening (part I).

125

Chapter 5

and Figure 5.7(f). Figure 5.8 shows the comparison of model predictions and measurements for the outlet temperature of reactor 1 and reactor 2. Oscillations are observed in the experimental results especially for the outlet temperature of reactor 2, and these are mainly caused by oscillations in the inlet temperature of reactor 2, see Figure 5.8(c). Fluctuations in the composition of reactor 2 have negligible impact. The inlet temperature of reactor 2 is influenced by changes in flow rate and/or temperature of the outlet stream of reactor 1 and the quench stream and is controlled by a master-slave temperature controller damping the oscillations. These oscillations are not inherent to the process dynamics alone, but are rather the result of the interaction between the controller dynamics and the process dynamics, which is therefore captured correctly in the frequency range which is relevant to closed-loop performance. The damping of the oscillations is a bit overestimated by the process model, which means that additional relatively small unmodelled delays (such as, e.g., those due to sensor and actuator dynamics) are present on the real plant. Regarding the outlet temperature of reactor 1, the amplitude of the variations is not predicted accurately by the model. The maximum overshoot is underpredicted by 8 K. However, the general dynamic trend, the presence of oscillations and the settling time can be deemed in good agreement with the experimental data. The results of the comparison are similar for the outlet temperature of reactor 2 with maximum deviations of 10 K. The mismatch in the reactor outlet conditions originates from the use of a simplified reactor model which assumes equilibrium conditions throughout the reactor. In order to correctly predict the reactor performance a more computationally expensive, kinetic-based model would be required. This model would allow the accurate prediction of the axial reactor temperature profile. It shall be kept in mind, that in realistic load change scenarios the perturbations are gradual and not step-wise as in the partial open-loop experiment. Therefore, the system response will be smoother and the amplitude of the variations smaller. As far as the reactors are concerned, this would lead to a better agreement between the actual transients and the simulations results. For the purpose of analysing the overall transient performance of the capture plant and evaluating different control strategies, the use of a simplified reactor component model as part of the entire dynamic model is arguably sufficient. In case the dynamic model should be used to fine-tune the control parameters of the plant control system, a more detailed reactor model might be required. To summarize, the initial and final steady-state values of the main process variables are reproduced with less than 1 % deviation with respect to measured values, an accuracy that can be considered satisfactory. Good steady-state predictions are also achieved for validation at off-design operation (60 % syngas load). This similar validation case is not included here for the sake of conciseness. Considering the main dynamic characteristics of the observed transients, namely, time and value of maximum overshoot, settling time, frequency and damping of oscillations, they are predicted within 20 % of the value that can be obtained from experimental 126

504

Experiment Model

500 496 492 488 484 0

25

50

75 100 125 150 175 200 Time [min]

Inlet temperature of reactor 2 [°C]

(a) Outlet temperature of reactor 1. 355

Outlet temperature of reactor 2 [°C]



480

Experiment Model

476 472 468 464 460 0

25

50

75 100 125 150 175 200 Time [min]

(b) Outlet temperature of reactor 2. Experiment Model

350 345 340 335 330 0

25

50

75 100 125 150 175 200 Time [min]

(c) Inlet temperature of reactor 2.

Figure 5.8: WGS section model validation: comparison of measurements and simulation results for a step increase in syngas control valve opening (part II).

data; larger errors are found in the temperatures of the reactors, and they can be attributed to the less accurate predictions for fast transients by the simplified, equilibrium-based reactor model. Finally, the validated sub-system models were combined to form a system model of the CO2 capture process. For system model validation a closed-loop assessment of the process and control system performance was carried out considering changes in syngas load. In order to avoid repetition, the results are not discussed in detail but presented together with the control system analysis in the next section. Summing up, this step-by-step validation approach from component to subsystem to system level provides a) validated and reliable component models, which can readily be integrated into other processes (adaptation of geometrical data required), b) validated system models including process control, and c) detailed understanding of the capture process in relation to which phenomena in the indi127

Chapter 5

vidual components are relevant and require accurate modelling to serve the purpose of dynamic process analysis and control system design.

5.5

Process analysis

Dynamic performance of fossil-fuelled power plants becomes increasingly important, hence an integrated capture process has to be able to follow relatively fast load variations. It might also be however required to temporarily only reduce the load of the energy-intensive CO2 capture process, while maintaining the gasifier load for an IGCC power plant. For example, this can be the case when the market demands more energy, or it is economically more favourable to produce energy instead of capturing CO2 . The amount of CO2 capturing and hence its energy consumption might also be adjusted for primary or secondary frequency control. The control system of the CO2 capture unit should therefore allow to perform frequent and prompt load variations. The objective of this analysis is to study the improvement of the dynamic performance of the pre-combustion CO2 capture unit by investigating an improved control strategy and demonstrating, by means of dynamic simulation, that this strategy might work if implemented into the actual plant control system. The decentralized control system based on PI controllers as implemented in the pilot plant is used as reference (in the following referred to as reference control ) to evaluate the improved control. It needs to be mentioned that the pilot plant control is possibly far from being optimal and was designed and tuned in order to achieve stable and safe operation during the test campaign. For the investigation of control strategies the validated dynamic system model of the CO2 capture pilot plant was used. The design of the pilot plant and of a foreseen large-scale plant are very similar, with the main difference concerning heat integration in the WGS section. Therefore, it can be assumed that a developed control strategy tested with the pilot plant model can also be applied to a largescale plant with appropriate modifications. Part-load and full-load operation of the CO2 capture unit differs mainly in terms of flow rates, whereas the most of the other process variables and parameters are kept to the same values. In order to allow for fast load variations it is particularly important to apply good set point management for the temperatures in the WGS section, as the thermal inertia of the system is much larger than the mass inertia. Hence, a control strategy based on feed-forward, feed-back and cascade control was implemented into the pilot plant model and tested for virtual plant operation. The improved control strategy and its implementation is explained in more detail in the following. Figure 5.9(a) shows an example of cascade control, as implemented in the dynamic model. The cascade control consists of two control loops. The master control compares the level measurement (process variable) with a given level set point and changes the set point of the slave control (control variable of the master 128


Vapour outlet

Vapour outlet

Syngas

Syngas

Water

Flash vessel FC

Water

LC

LC

+

B

+

A

Liquid outlet

(a)

Flash vessel FC

Syngas set point

Liquid outlet

Feed-forward (b)

Figure 5.9: Flash vessel level control via a) cascade and feed-back control and b) cascade, feed-back and feed-forward control.

control). The slave loop compares the flow measurement (process variable) with the set point provided by the master loop and changes the valve opening (control variable) accordingly. The advantage of cascade control is that it allows the system to be more responsive to disturbances, and it is particularly useful for systems with long dead and lag times [39]. However, this comes at the cost of higher system complexity and requires more process instrumentation. A cascading control scheme has been implemented in the WGS section for the control of the liquid level in both vessels, the syngas mass flow rate and the inlet temperature of reactor 2. Figure 5.9(b) shows the application of feed-forward control to the reaction water loop as a paradigmatic case. The aim of feed-forward control is to measure disturbances upstream the system and compensate for them before the system variables deviate from the set point. In case of the capture unit, one of the main disturbances is the change in syngas load. Hence, the feed-forward control receives the syngas set point as input and determines the set point for the water flow control based on an explicit equation. The feed-back control ensures that the level set point is maintained. The advantage of feed-forward control in comparison to feed-back control is that disturbances do not need to propagate through the process in order to take control actions, hence the set point control is more accurate. However, in order to accurately predict feed-forward control actions accurate measurements and adequate disturbance predictions are required. In case of non-ideal processes accurate prediction might require models consisting of non-linear equations. The feed-forward control logic has been implemented in the dynamic model for all master control loops. The feed-forward action is determined based on the syngas set point (it is currently based on a simple linear correlation) and is sent together with the feed-back action to the slave control loop. The coefficients of the linear equations are determined by considering two steady-state operating points. The implementation can easily be extended: for example, in case real plant data are available, more elaborate correlations can be tuned to cover a wide operating range. 129

Chapter 5

Observation of the results of experimental tests on the pilot plant show that the steady-state reactor performance varies from part-load to full-load, which is revealed by the different values of the reactor outlet temperatures. The reactor performance should thus be kept unchanged in order to avoid long settling times of the temperatures, whenever fast load variations must be performed. The cause of the performance difference is related to changes in the thermodynamic state of the 2nd vessel upstream the reactors, in terms of temperature and pressure, which subsequently leads to changes in the inlet composition of reactor 1 and 2. This applies in case the syngas inlet composition is constant, which is a justified assumption if the same type of fuel is used for gasification, and the gasifier remains at constant load. In the case of the pilot plant, the inlet pressure of the capture unit is constant, hence the actual vessel pressure is a result of the difference between the inlet pressure and the flow-dependent frictional losses. In order to maintain the vessel pressure at different loads, a pressure controller is therefore implemented into the pilot plant model. Furthermore, the vessel conditions at part-load and full-load differ because of changes in process heat losses. In the pilot plant large heat losses occur at the inlet and outlet of the reactors due to the fact that the reactor casing is overdimensioned for the installed amount of catalyst. In a large-scale plant with well-sized reactors these particular heat losses will not be present. In addition, heat losses from piping upstream and downstream the reactors will be much smaller in a large-scale design because of better insulation and a smaller heat transfer area to volume ratio. Hence, the heat losses have been significantly reduced in the system model in order to perform the simulations at large-scale conditions. Both adaptations, the control of the vessel pressure and the heat loss reduction allow to maintain the reactor performance and thus promote the ability of the CO2 capture unit to perform fast load variations. Figure 5.10 (process mass flow rates and shifted syngas composition) and Figure 5.11 (process temperatures) visualize the comparison of simulation results for the WGS section of the CO2 capture plant obtained with the reference and the improved pilot plant control for a load variation from part-load to full-load. The model predictions of the reference control have been validated with experimental data, which is as comparison included in the plots. In the simulation with the improved control, the syngas mass flow rate is ramped from 850 to 1100 kg/h in 10 minutes, whereby in the simulations with the reference control the syngas mass flow set point is changed instantaneously, and the control system takes care of the load change. From the comparison of the reference and the improved control it can observed that it is possible to subject the capture process to prompt load changes. The simulation of the process regulated by the improved control system indicates that, as far as the mass flow rates in the WGS are concerned, more than 95 % of the final steady-state value can be reached within the ramping time. The vessel and reactor temperatures settle within approximately 60 minutes after the beginning of the 130


1400

1100 1050 1000 950 900 Measurements Reference control Improved control

850 800 0

1

2

3

4 5 6 Time [h]

7

8

9

Water mass flow rate [kg/h]


1150

1300 1200 1100

900 0

10

1

2

3

4 5 6 Time [h]

7

8

9

10

(b) Water mass flow rate.

1300

1300

1200 1100 1000 Measurements Reference control Improved control

900 800 0

1

2

3

4 5 6 Time [h]

7

8

9

Quench mass flow rate [kg/h]

Mass flow rate of reactor 1 [kg/h]

(a) Syngas mass flow rate.

1200 1100 1000 900

700 0

10

54.75

54.7 Reference control Improved control 2

3

4 5 6 Time [h]

7

8

(e) CO2 content in shifted syngas.

9

10

H2 content in shifted syngas [mol%]

54.8

1

1

2

3

4 5 6 Time [h]

7

8

9

10

(d) Quench mass flow rate.

54.85

54.65 0

Measurements Reference control Improved control

800

(c) Mass flow rate of reactor 1. CO2 content in shifted syngas [mol%]


1000

38.1 38.05 38 37.95 37.9 37.85 0

Reference control Improved control 1

2

3

4 5 6 Time [h]

7

8

9

10

(f) H2 content in shifted syngas.

Figure 5.10: Comparison of simulation results: reference versus improved control. The objective is to change the operating condition from part-load to full-load (part I).

131

Quench flow temperature [°C]

174 173 172 171 170 Measurements Reference control Improved control

169 168 0

1

2

3

4

5 6 Time [h]

7

8

9

10

Inlet temperature of superheater [°C]

Chapter 5

508


504 500 496 492 488 484 480 476 0

1

2

3

4

5 6 Time [h]

7

8

198 197 196 195 Measurements Reference control Improved control

194 193 0

1

2

3

4

5 6 Time [h]

7

8

9

10

(b) Inlet temperature of superheater.

9

(c) Outlet temperature of reactor 1.

10



(a) Quench flow temperature.

199

492 488 484 480 476 472 468


464 460 0

1

2

3

4

5 6 Time [h]

7

8

9

10

(d) Outlet temperature of reactor 2.

Figure 5.11: Comparison of simulation results: reference versus improved control. The objective is to change the operating condition from part-load to full-load (part II).

132


89 CO2 absorption effiency [%]

Solvent mass flow rate [kg/s]

17 16 15 14 13 12 0

88.8

88.6

88.4

Improved control 1

2

3

4

5 6 Time [h]

7

8

9

10

(a) Solvent mass flow rate. CO2 product mass flow rate [kg/h]

Improved control

88.2 0

1

2

3

4 5 6 Time [h]

7

8

9

10

(b) CO2 absorption efficiency.

1300 1200 1100 1000 900 800 700 600 0

Improved control 1

2

3

4 5 6 Time [h]

7

8

9

10

(c) CO2 product mass flow rate.

Figure 5.12: Simulation results of improved control. The objective is to change the operating condition from part-load to full-load (part III).

perturbation. In addition, with the improved control (cascade and feed-forward) and the measures to maintain the reactor performance, the maximum overshoot during transient can be reduced. In the case of the improved process control, due to the change in heat losses and vessel pressure, the steady-state values of most of the variables are different in comparison to the simulation results obtained with the reference control. Figure 5.12 visualizes the the simulation results related to the absorption section in case the improved control strategy is adopted. The operators of the pilot plant manually adjusted the solvent mass flow rate in order to account for syngas load variations such that the performance in terms of CO2 capture was maintained. No meaningful comparison to the reference control can therefore be provided. The implementation of the feed-forward control suggests that this strategy is a good replacement of the manual operation. A ratio controller is implemented into the dynamic model of the plant in order to maintain the weight-based liquid-to-gas ratio of the absorption column, which 133

Chapter 5

allows to keep the CO2 removal efficiency approximately constant for the operational range of the absorber. Consequently, if the mass flow rate of the shifted syngas changes, the solvent mass flow rate is adapted accordingly. The results indicate that the absorption section quickly responds to fast load variations. As mentioned, feed-forward control requires process measurements in order to predict the control actions. It is therefore important to investigate the impact of possible measurement errors on the performance of the proposed control strategy. Multiple simulations have been performed whereby a relative error of either + or 5 % was applied randomly to each measurement. Figure 5.13 shows the comparison between a simulation unaffected by errors and one in which the errors have been applied in order to produce the largest variation in the main process variables. For conciseness, only four process variables are reported as representative examples for the difference in overall performance. When including measurement errors during the simulation, the overshoot and settling time increases, but only slightly for most process variables. The worst response is observed in the outlet temperature of reactor 1, where the overshoot more than doubles because of the measurement errors. This demonstrates that the performance of the proposed control strategy can decrease if measurement errors are present, however the gain in improvement of the dynamic performance in comparison to the reference control is still significant. To conclude, it has been demonstrated that validated physical-based dynamic process models ease the design and testing of control strategies and that prompt load variations can be performed with a pre-combustion CO2 capture unit. The qualitative results from this investigation related to the pilot plant model can be applied for the design of control strategies of a large-scale capture plant. Ultimately, the goal is to study the transient interaction between the CO2 capture unit and the main power plant during load variations, which is beyond the scope of the work presented here.

5.6

Conclusions

This chapter discusses the dynamic modelling and simulation of pre-combustion CO2 capture plants for the use in control design. The models follow an objectoriented modelling paradigm. The models of the main process components, such as the flash vessel, the water-gas shift reactor and the absorption column are presented briefly. Comprehensive model validation has been performed at component, sub-system and system level by comparing dynamic simulation results to experimental measurements obtained from various open- and closed-loop transient tests performed at the CO2 capture pilot plant operated at the Buggenum IGCC power station. Dynamic simulations with the validated system model have been used to investigate a promising control strategy based on feed-forward, feed-back and cascade control to improve dynamic performance. The following conclusions can be drawn from this study: • Transient performance in terms of prompt syngas load variations is a feasible 134

1400 1300 1200 1100 1000

Syngas w/o meas. errors Syngas with meas. errors Water w/o meas. errors Water with meas. errors

900 800 0

1

2

3

4 5 6 Time [h]

(a)

7

8

9

10

Outlet temperature of reactor 1/2 [°C]

Syngas and water mass flow rate [kg/h]


488 486 484 482 480 478

TR1out w/o meas. errors

476

TR1out with meas. errors

474

470 0

T

w/o meas. errors

T

with meas. errors

R2out

472

R2out

1

2

3

4

5 6 Time [h]

7

8

9

10

(b)

Figure 5.13: Comparison of simulation results: improved control with and without measurement errors. The objective is to change the operating condition from part-load to full-load. a) Syngas and water mass flow rate. b) Outlet temperature of reactor 1 and 2.

operating mode for pre-combustion CO2 capture plants featuring sophisticated control systems. This enables the IGCC power plant to respond to fast changes in the energy demand by reducing the load of the energy-intensive CO2 capture process instead of adjusting the gasifier load. • Physical-based dynamic models can be of manifold use during the design phase of new plants to evaluate different configurations, to support equipment selection and sizing, and to develop and test control strategies with the underlying aim to improve dynamic performance. A demonstration of this capabilities has been given in this chapter with respect to control optimization. Moreover, design and tuning of controllers prior plant construction can save time and trouble during commissioning. • The validated component and pilot plant system models have been implemented in an open source library, which can serve as reliable foundation for the development of large-scale system models of a pre-combustion capture process. This requires the assembly of the required process models, by re-use of existing models and adaptation of equipment sizing along with the implementation and tuning of a control system. The flexibility of the object-oriented modelling approach allows to easily extend existing models and to implement more sophisticated model versions if required. Moreover, the model of the CO2 capture plant can be easily integrated into models of gasification units and combined cycle plants in order to form a system models of the entire IGCC power plant [3], which can ultimately be used to study the interaction of the different units during transient operation. 135

Chapter 5

• Finally, it is worth pointing out that the non-proprietary Modelica language is supported by various software tools, which has the advantage of not being bound to a specific proprietary simulation environment and gives the flexibility to explore the simulation capabilities of different tools. Moreover, it has been demonstrated that Modelica process models can be interfaced with external fluid media libraries for the computation of thermophysical properties. This allows to easily change process fluids as typically a wide range of pure fluids and fluid mixtures described with suitable and accurate equation of states are available in property packages. Future work might tackle, apart from the analysis of the transient performance of decarbonised IGCC power plants, also the dynamic optimization of load change trajectories, whereby the main challenge will be related to the system complexity.

136


Nomenclature h M P q t T u v w X

= = = = = = = = = =

Specific enthalpy, J kg−1 Total mass, kg Pressure, Pa Vapour quality Time, s Temperature, K Internal energy, J kg−1 −1 Specific volume, m3 kg −1 Mass flow rate, kg h Mass fraction

Greek symbols ρ

=

Density, kg m−3

= = = =

2-phase region Mixture component Liquid phase Vapour phase

= = = = = =

Differential and algebraic equation Equation of state Integrated gasification combined cycle Ordinary differential equation Perturbed chain - statistical associating fluid theory Water-gas shift

Subscripts 2ph i,j liq vap Acronyms DAE EoS IGCC ODE PC-SAFT WGS

137

Chapter 5

References [1] Ziems, C., Meinke, S., Nocke, J., Weber, H., and Hassel, E., 2012. Auswirkungen von fluktuierender Windenergieeinspeisung auf das regel- und thermodynamische Betriebsverhalten konventioneller Kraftwerke in Deutschland, Teil II - Auswirkungen großer Windeinspeisungen auf den zuk¨ unftigen Kraftwerkspark und dessen Tageseins. Tech. rep., VGB PowerTech e.V. Projektnummer 333. [2] Saint-Drenan, Y.-M., von Oehsen, A., Gerhardt, N., Sterner, M., Bofinger, S., and Rohrig, K., 2009. Dynamische Simulation der Stromversorgung in Deutschland nach dem Ausbauszenario der Erneuerbaren-Energien-Branche. Tech. rep., Fraunhofer Institut fr Windenergie und Energiesystemtechnik (IWES). [3] Casella, F., and Colonna, P., 2012. “Dynamic modeling of IGCC power plants”. Applied Thermal Engineering, 35, pp. 91–111. [4] Robinson, P., and Luyben, W., 2010. “Integrated gasification combined cycle dynamic model: H2 S absorption/stripping, water-gas shift reactors, and CO2 absorption/stripping”. Industrial and Engineering Chemistry Research, 49(10), pp. 4766– 4781. [5] Bhattacharyya, D., Turton, R., and Zitney, S., 2012. “Control system design for maintaining CO2 capture in IGCC power plants while load-following”. In Proceedings of the 29th Annual International Pittsburgh Coal Conference, Pittsburgh, PA, October 15-18, Vol. 3, pp. 2160–2173. [6] Zitney, S., Liese, E., Mahapatra, P., Turton, R., Bhattacharyya, D., and Provost, G. “AVESTART M Center: Dynamic simulation-based collaboration toward achieving operational excellence for IGCC plants with carbon capture”. In Proceedings of the 29th Annual International Pittsburgh Coal Conference 2012, Pittsburgh, United States, 15-18 October 2012, Vol. 3, pp. 2093–2147. [7] Schoen, P., 1993. “Dynamic modeling and control of integrated coal gasification combined cycle units”. PhD thesis, Delft University of Technology. [8] Robinson, P., and Luyben, W., 2011. “Plantwide control of a hybrid integrated gasification combined cycle/methanol plant”. Industrial and Engineering Chemistry Research, 50(8), pp. 4579–4594. [9] Choi, Y., Li, X., Park, T., Kim, J., and Lee, J., 2001. “Numerical study on the coal gasification characteristics in an entrained flow coal gasifier”. Fuel, 80(15), pp. 2193–2201. [10] Chen, C., Horio, M., and Kojima, T., 2000. “Numerical simulation of entrained flow coal gasifiers. Part I: Modeling of coal gasification in an entrained flow gasifier”. Chemical Engineering Science, 55(18), pp. 3861–3874. [11] Chen, C., Horio, M., and Kojima, T., 2000. “Numerical simulation of entrained flow coal gasifiers. Part II: Effects of operating conditions on gasifier performance”. Chemical Engineering Science, 55(18), pp. 3875–3883.

138


[12] Watanabe, H., and Otaka, M., 2006. “Numerical simulation of coal gasification in entrained flow coal gasifier”. Fuel, 85(12-13), pp. 1935–1943. [13] Robinson, P., and Luyben, W., 2008. “Simple dynamic gasifier model that runs in aspen dynamics”. Industrial and Engineering Chemistry Research, 47(20), pp. 7784– 7792. [14] Kasule, J., Turton, R., Bhattacharyya, D., and Zitney, S., 2012. “Mathematical modeling of a single-stage, downward-firing, entrained-flow gasifier”. Industrial and Engineering Chemistry Research, 51(18), pp. 6429–6440. [15] Seliger, B., Hanke-Rauschenbach, R., Hannemann, F., and Sundmacher, K., 2006. “Modelling and dynamics of an air separation rectification column as part of an IGCC power plant”. Separation and Purification Technology, 49(2), pp. 136–148. [16] Mahapatra, P., and Bequette, B., 2013. “Design and control of an elevated-pressure air separations unit for IGCC power plants in a process simulator environment”. Industrial and Engineering Chemistry Research, 52(9), pp. 3178–3191. [17] Koch, I., Hannemann, F., and Hoffmann, U., 1999. “Dynamic simulation of operating cases and malfunctions of an IGCC power plant fuel system”. Chemical Engineering and Technology, 22(7), pp. 568–570. [18] Heil, S., Brunhuber, C., Link, K., Kittel, J., and Meyer, B. “Dynamic Modelling of CO2 -removal units for an IGCC power plant”. In Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009, pp. 77 – 85. [19] Damen, K., Gnutek, R., Kaptein, J., Nannan, N. R., Oyarzun, B., Trapp, C., Colonna, P., van Dijk, E., Gross, J., and Bardow, A., 2011. “Developments in the pre-combustion CO2 capture pilot plant at the Buggenum IGCC”. Energy Procedia, 4(0), pp. 1214–1221. [20] J-Power, 2011. The Future of Coal-Fired Thermal Power Generation. www.jpower. co.jp. [21] Casero, P., Pe˜ na, F., Coca, P., and Trujillo, J., 2013. “ELCOGAS 14 MWth precombustion carbon dioxide capture pilot. Technical & economical achievements”. Fuel. in Press. [22] Wang, M., Lawal, A., Stephenson, P., Sidders, J., and Ramshaw, C., 2011. “Postcombustion CO2 capture with chemical absorption: A state-of-the-art review”. Chemical Engineering Research and Design, 89(9), pp. 1609–1624. [23] Kvamsdal, H., Chikukwa, A., Hillestad, M., Zakeri, A., and Einbu, A., 2011. “A comparison of different parameter correlation models and the validation of an MEAbased absorber model”. Energy Procedia, 4, pp. 1526–1533. ˚kesson, J., Faber, R., Laird, C., Tummescheit, H., Velut, S., and Zhu, Y., 2011. [24] A “Models of a post-combustion absorption unit for simulation, optimization and nonlinear model predictive control schemes”. In Proceedings 8th International Modelica Conference 2011, Dresden, Germany.

139

Chapter 5

[25] Biliyok, C., Lawal, A., Wang, M., and Seibert, F., 2012. “Dynamic modelling, validation and analysis of post-combustion chemical absorption CO2 capture plant”. International Journal of Greenhouse Gas Control, 9, pp. 428–445. [26] Mattsson, S., Elmqvist, H., and Otter, M., 1998. “Physical system modeling with modelica”. Control Engineering Practice, 6(4), pp. 501–510. [27] Fritzson, P., 2014. Principles of Object-Oriented Modeling and Simulation with Modelica 3.3: A Cyber-Physical Approach. Wiley-IEEE Press. [28] Gross, J., and Sadowski, G., 2001. “Perturbed-Chain SAFT: An Equation of State Based on a Perturbation Theory for Chain Molecules”. Industrial and Engineering Chemistry Research, 40, pp. 1244–1260. [29] Colonna, P., van der Stelt, T., and Guardone, A., 2004-2012. FluidProp: a program for the estimation of thermophysical properties of fluids. Version 3.0. software. [30] Casella, F., and Leva, A., 2006. “Modelling of Thermo-Hydraulic Power Generation Processes Using Modelica”. Math. Comput. Model. Dyn. Syst., 12(1), pp. 19–33. [31] Casella, F., and Leva, A., 2005. “Object-oriented modelling & simulation of power plants with modelica”. In Proceedings of the 44th IEEE Conference on Decision and Control, and the European Control Conference, CDC-ECC ’05, pp. 7597–7602. [32] Casella, F., and Richter, C. C., 2008. “ExternalMedia: a Library for Easy Re-Use of External Fluid Property Code in Modelica”. In Proceedings 6th International Modelica Conference, pp. 157–161. [33] Infochem Computer Services Ltd, 2014. Multiflash - Flow Assurance Software. www. kbcat.com. [34] Richard Szczepanski, Infochem Computer Services Ltd, 2014. Private communication, January 17. [35] Dassault Systèmes, 2013. Dymola. www.3ds.com. [36] Casella, F., Van Putten, J., and Colonna, P., 2008. “Dynamic simulation of a biomass-fired steam power plant: a comparison between causal and a-causal modular modeling”. In ASME International Mechanical Engineering Congress and Exposition, Vol. 6, pp. 205–216. [37] Michelsen, M. L., and Mollerup, J. M., 2007. Thermodynamic Models: Fundamentals & computational aspects, second edition ed. Tie-Line Publications. [38] Wellner, K., Trapp, C., Schmitz, G., and Casella, F., 2014. Interfacing Models for Thermal Separation Processes with Fluid Property Data from External Sources. Submitted to the 10th Modelica Conference, Lund, Sweden, March 10-12,. [39] Luyben, W. L., 1990. Process modeling, simulation, and control for chemical engineers. McGraw-Hill.

140

“Liebe deutschen Landsleute, wir sind zu Ihnen gekommen, um Ihnen ¨ mitzuteilen, dass heute Ihre Ausreise...(‘moglich geworden ist’ ging im Jubel unter).” “Dear fellow Germans, we have come to you to tell you that today, your departure...(‘became possible’ drowned in cheers).”

Hans-Dietrich Genscher, Minister for Foreign Affairs of Germany, announcement to East German refugees, German Embassy Prag, September 30, 1989

6

Dynamic system model of the absorption section of pre-combustion CO2 capture plants 1 This chapter presents the development and validation of the dynamic model of the absorption and solvent regeneration unit as part of a pre-combustion CO2 capture system, with focus on the absorber column model. The models are implemented into an open source software library following an object-oriented, equation-based modelling approach using the Modelica language. The models are validated by comparison against two sets of transient experimental data generated at the CO2 capture pilot plant realized and operated at the Buggenum IGCC power station in the Netherlands. The model predictions satisfactorily reproduce the dynamic performance of the pilot plant considering the main process variables, such as pressures, flow rates and compositions. One of the results of this work is that, in case of physical absorption of CO2 a well-tuned, equilibrium-based absorber model is able to accurately predict transient operation. The validated process models enable process and control system design aiming at improvement of transient performance of the pre-combustion capture plant.

1 The contents of this chapter are submitted for publication in: Trapp, C., de Servi, C., Casella, F., Bardow, A., and Colonna, P., 2014. International Journal of Greenhouse Gas Control.

Chapter 6

6.1

Introduction

The study documented here focuses on the simulation and validation of one of the key components of the pre-combustion CO2 capture plant, the CO2 absorber column with its auxiliaries. The literature treating the dynamic modelling and simulation of physical absorption for pre-combustion CO2 capture is scarce. Heil et al. developed a dynamic model for a pre-combustion CO2 removal unit utilizing methanol as solvent in an equilibrium-based absorber column model. The thermophysical properties of the fluid mixtures involved in the process are modelled with a simplistic approach [1]. A larger number of published studies is related to the dynamic performance of reactive absorption for post-combustion CO2 capture, typically investigating the use of aqueous amine-based solvents [2]. This might be related to the fact that currently various post-combustion demonstration projects are ongoing, and they also include modelling and experimental activities, see, e.g., Ref. [3]. Moreover, the reactive absorption process is more challenging from the modelling point of view in comparison to physical absorption. For adequate model predictions of multi-component fluid processes involving chemical reactions, the more complex rate-based approach is required [4–6]. Different software tools have been identified as suitable for dynamic model development and simulation of chemical absorption for CO2 capture [7–10]. Specific model libraries are available in some commercial tools. Dietl developed an open source library according to a general modelling approach in order to allow for modelling of very different kinds of absorption and distillations processes [11]. Most of the dynamic models of post-combustion capture units were validated by comparison with steady-state experimental data [8–10, 12]. Only few publications present transient measurement data obtained from pilot plant operation, mainly due to challenges in operating the CO2 absorption plant in dynamic mode. Some authors treated the dynamic validation of the standalone CO2 absorber model [13, 14], while others validated the model by comparison with transient measurements related to the entire post-combustion capture unit, therefore including an integrated absorber and desorber column [11, 15, 16]. According to the knowledge of the authors, an experimentally validated model (static or dynamic) of the physical absorption section of a pre-combustion CO2 capture plant is not documented in the literature. The notable aspects of this work are: • Demonstration of an equation-based, object-oriented modelling approach using a non-proprietary modelling language, which is supported by various simulation environments (proprietary and open source) as described in detail in Chapter 5. • Comprehensive dynamic model validation, whereby the absorber model and the model of the absorption and solvent regeneration section are validated by comparison with experimental data obtained from a pilot plant. Two ad hoc 142

Dynamic system model of the absorption section of pre-combustion CO2 capture plants

Knock-out Drum Electric Heater 2

Electric Heater 1 Waste water

Reaction Water

Syngas Make-up Water

Electric Heater 3 Gas Quench


Process water

1st Shift Reactor 2nd Shift Reactor

Rectifier Water-gas shift Cooler section Separator Booster Rectifier Compressor Compressor Cooler

Absorption & solvent regeneration section

CO2 Absorber

Airblown Cooler 1

H2-rich Syngas CO2 Compressor

3rd Shift Reactor Airblown Cooler 2 Condensate Pump

Flash 1

Flash 2 Flash 3

Product CO2

Solvent Heater Solvent Cooler Solvent Pump

Figure 6.1: Process flow diagram of the CO2 capture pilot plant. The absorption and solvent regeneration section is highlighted.

transient tests have been performed, monitoring the response to step-wise changes of the syngas and solvent mass flow rate. This chapter is structured as follows: Section 6.2 describes the model development focusing on the absorber column model. Section 6.3 covers the dynamic model validation of the standalone absorber model and subsequently the model of the absorption and solvent regeneration section. Concluding remarks are summarized in Section 6.4.

6.2 6.2.1

Model development Process description

This study is about the dynamic modelling of the absorption and solvent regeneration section of pre-combustion CO2 capture units. For the sake of clarity, the simplified process flow diagram of the CO2 capture pilot plant built at the site of the Buggenum IGCC power station is depicted here again, see Figure 6.1, whereby the absorption and solvent regeneration section is highlighted. The detailed process description can be found in Section 2.2.

6.2.2

Modelling approach

The models were developed following the equation-based, object-oriented modelling approach as outlined in the previous chapter in Section 5.2. 143

Chapter 6

6.2.3

Dynamic absorber model

Models of physical absorption of gases into liquids differ in the way the mass transfer is treated [17]. A simple representation of the phenomena is based on the assumption of thermodynamic equilibrium between the vapour and liquid phase. The more rigorous and accurate formulation accounts for the mass and heat transfer resistance between the phases, thus requiring rate equations. However, accuracy of the model predictions comes at the cost of model complexity and computational load. For dynamic process models, which are mathematically more complex than steady-state models, a reduction in degree of detail might be required in order to increase robustness and allow for reasonable simulation times. This applies especially to dynamic models used for plant-wide system analysis, such as it is the case for the absorption and solvent regeneration section model. This model is also intended for integration into the model of the entire CO2 capture plant and ultimately into the full power plant system model. Moreover, the aim of dynamic models is the accurate prediction of the transient performance and deviations in absolute values of process variables have often negligible impact on the prediction of the system dynamics. The packed absorber column was therefore modelled following the equilibriumbased approach. The column is subdivided in theoretical stages assuming equilibrium between the vapour and liquid phase of the working fluid within each volume. Here, the number of equilibrium stages is tuned by comparison with the simulation results of a full rate-based model2 in order to match steady-state performance at nominal operating conditions. When the absorber operation departs far from the design point less accurate steady-state performance estimates are then obtained with the equilibrium-based model. This behaviour is evaluated in the following. Figure 6.2(a) shows the percentage absolute deviation in CO2 absorption efficiency3 between the simulation results of the equilibrium-based and rate-based column model (operating range: syngas mass flow rate 800 − 1600 kg/h and solvent mass flow rate 10 − 18 kg/s). The comparison of the model predictions is also plotted in terms of CO2 absorption efficiency as function of L/G ratio, see Figure 6.2(b). Throughout the considered operating window deviations are smaller than 4.5 % with a maximum at liquid-to-gas (L/G) ratio 22.5 and absorption efficiency 67.3 %. In general, differences are higher at low L/G ratios corresponding to lower absorption efficiencies. Considering the observed differences in CO2 absorption over the wide range of operating conditions, it can be argued that the equilibrium-based model predicts the values of the main process variable with adequate accuracy for the aim of dynamic system simulations, whereby the interest is more on the correct estimation of transients. 2 The performance of the absorber column is obtained by a rigorous, rate-based model for simulating all types of multi-stage, vapour-liquid fractionation operations under steady-state operating conditions [18]. 3 The CO absorption efficiency is defined as the molar flow rate of absorbed CO in the rich 2 2 solvent leaving the absorber divided by the molar flow rate of CO2 in the syngas entering the column.

144


0 1 2 3 4

5 800 1000 1200

1400 1600 18 Syngas flow [kg/h]

(a)

16

14

12

10

Solvent flow [kg/s]

90 86 Design Condition

−1

CO2 absorption efficiency [%]

Absolute deviation [%]

94

82 78 74 70 66 20

Rate−based model Equilibrium−based model 25

30

35 40 45 50 55 Liquid−to−gas ratio [−]

60

65

70

(b)

Figure 6.2: Comparison of steady-state simulation results obtained with the equilibriumand the rate-based models. a) Absolute deviation4 of the prediction of CO2 absorption efficiency. b) CO2 absorption efficiency for different liquid-togas ratios.

It should also be considered that such an absorption and solvent regeneration unit is controlled based on the L/G ratio in order to maintain a constant extent of carbon capture. During transient operation it is therefore expected that the absorption efficiency does not deviate significantly from its design value and hence a well-tuned equilibrium-based model is sufficient to describe the transient performance. Section 6.3 describes the verification of this model assumption by means of model validation against experimental data. In general, a dynamic process model is described by conservation and constitutive equations typically resulting in a set of differential and algebraic equations (DAE’s), which can be classified according to its index [19, 20]. It is numerically difficult to solve high-index DAE systems and therefore current simulation tools implementing the Modelica language employ state-of-the-art techniques for index reduction. However, difficulties related to index reduction operations might arise in case external functions, such as those providing fluid property estimations, are needed within the process models [21]. In order to keep the index of the DAE system as low as possible, the modelling procedure described, e.g., in Refs. [22, 23] has been adopted. It can be synthetically explained as follows: • The model of each component is first discretized into so-called resistive (zero volume approximation) and storage modules (zero potential drop approximation), thus solving the conservation law equations for the generalized 4 The absolute deviation of CO absorption efficiency is defined as the difference between 2 the prediction of the equilibrium-based and rate-based model divided by the estimation of the rate-based model.

145

Chapter 6

flow variable within the resistive module, and for the generalized potential variable within the storage module. • Resistive modules are then connected to storage modules to form components, and such causality is also used to connect components to form the system model. Based on this modelling procedure, the stage-wise discretised packed column with counter-current flow of vapour and liquid can be represented as a series of equivalent tray modules (storage module) and valve or liquid head modules (resistive module). Within the equivalent tray module, pressure, temperature and composition of the liquid and vapour phase are determined by solving the conservation equations for mass and energy, assuming thermodynamic equilibrium between liquid and vapour. The component mass and energy balance read dM dXi Xi + M = mLin XLin,i + mVin XVin,i − mLout XLout,i − mVout XVout,i , dt dt

(6.1)

du dM +u = mLin hLin + mVin hVin − mLout hLout − mVout hVout , (6.2) dt dt where M, u, Xi are the total mass, internal energy and component mass fraction of the fluid within the volume of the tray. For reasons of computational efficiency the vessel pressure p, temperature T and overall mass fraction Xi are selected as state variables. Hence, the mass and energy balance need to be expressed in terms dXi of the state derivatives ddtp , dT dt and dt , yielding M

dM = −Mρ dt

"

du = dt

∂v ∂p

∂u ∂p

T,X

T,X

# n dp ∂v dT dXi ∂v + + , dt ∂T p,X dt ∂Xi p,T,Xj6=i dt

(6.3)

n dp ∂u dT ∂u dXi + + . dt ∂T p,X dt ∂Xi p,T,Xj6=i dt

(6.4)

∑

i=1

∑

i=1

The required thermodynamic properties, including the partial derivatives of fluid thermodynamic properties, are obtained from the external thermophysical property library as prop = prop(p, T, Xi ). The momentum equation pertaining to the resistive module is substituted by empirical correlations to describe the hydrodynamics of the stage predicting the liquid and vapour flow rate as a function of the pressure difference between the stages, the liquid holdup and the packing characteristics. The pressure drop ∆p is calculated with empirical correlations of Billet and Schultes [24], thus FV2 1 ∆p a = ψL , and 3 H (ε − hoL ) 2 K 146

(6.5)


Parameter description

Value

Number of theoretical absorber stages Absorber column diameter d Absorber packing height H Packing type

3 0.76 m 9.4 m Raschig Super-Pack 250Y 250 m2 m−3 0.98 m3 m−3 19.5b (default 0.18a) 0.65a 310 kg m−3 500 J kgK−1 0.76 m 9.3 m 5.6 m3 2.4 m3 2.4 m3

Packing specific surface area a Void fraction ε Constant for pressure drop correlation Cp Constant for holdup correlation Ch Packing density Packing specific heat capacity Absorber sump diameter Absorber sump height 1st flash vessel volume 2nd flash vessel volume 3rd flash vessel volume a b

Provided by packing supplier. Adjusted to foster numerical convergence.

Table 6.1: Model parameters of main pilot plant components.

ε − hoL ε

1.5

0.3

√ exp C1 FrL

13300 . a3/2 (6.6) The pilot plant absorber is operated in the pre-loading region, therefore the liquid holdup equals the holdup at the loading point: hoL = hoL,S . The liquid holdup hoL is given by [24] ψL = Cp

1.8 64 + ReV Re0.08 V

hoL hoL,S

with C1 =

1/3 ah 2/3 1 ηL 2 hoL = 12 wL a , with g ρL a ReL =

wL ρL ah ≥5: = 0.85Ch ReL0.25 aηL a

w2L a g

(6.7)

0.1 .

(6.8)

Storage of thermal energy in the packing is considered, however heat losses to the environment are neglected. Based on the implemented model equations, various model parameters have to be specified related to the column geometry and the characteristics of the packing. These parameters and the related values applicable to the pilot plant absorber are given in Table 6.1. Transport properties such as dynamic viscosity of the liquid and vapour phase are assumed constant and mean values are used based on the absorber operating point. 147

Chapter 6

Process variable

Control variable

Set point

Absorber column pressure 1st flash vessel pressure 2nd flash vessel pressure Absorber sump level 1st flash vessel level 3rd flash vessel level Solvent mass flow rate Lean solvent temperature

Opening Opening Opening Opening Opening Opening Opening Opening

21.7 bar 7.5 bar 2.9 bar 2000 mm 2000 mm 800 mm 15 kg/s 40 ◦ C

H2 -rich gas flow valve 1st flash vessel gas flow valve 2nd flash vessel gas flow valve rich solvent flow valve 1st flash vessel liquid flow valve 2nd flash vessel liquid flow valve lean solvent flow valve cooling water flow valve

Table 6.2: Absorption and solvent regeneration section control loops.

6.2.4

Additional process models and control

For the modelling of the entire absorption and solvent regeneration section, additional process models for the absorber sump, the flash vessels, the solvent cooler and solvent pump are necessary (see Figure 6.1). These models are documented in the previous chapter. The system model is obtained by assembly of the individual component models and adjustment of the geometrical data. Equipment sizing information of the pilot plant components is summarized in Table 6.1. In addition, for dynamic process simulations, an appropriate control scheme is required. That entails in this case the modelling of linear valves and PI controllers. The pilot plant control scheme is thus included in the dynamic system model. The individual control loops are summarized in Table 6.2 and depicted in the object diagrams of the dynamic models presented in Section 6.3. For a large-scale plant, the capture rate is typically maintained by a L/G ratio controller, which was however not implemented in the pilot plant. The basic component models such as sinks, sources, valves, pressure drops and pumps are taken from the ThermoPower library [25, 26] and adapted in terms of their working fluid models which have been replaced with functional calls to the external property tool in order to accurately estimate the property values of the syngas-DEPEG mixture.

6.3

Dynamic validation

Two types of dynamic experiments were performed at the Buggenum CO2 capture pilot plant for model validation purposes. During the first tests (TR-Solvent) the solvent mass flow rate was changed stepwise, while keeping the syngas mass flow rate, the water content in the solvent, the absorption pressure and temperature constant. In the second tests (TR-Syngas) the syngas mass flow rate was perturbed stepwise, while maintaining unchanged the values of the other input variables. Both types of experiments cause transient changes of the liquid-to-gas flow ratio, which has a large impact on the process performance, e.g., on the CO2 148


capture efficiency. These tests are expected to be suitable for the validation of the predictive capabilities of the models. First, the validation of the standalone absorber model is discussed in detail and afterwards the validation of the absorption and solvent regeneration section model is presented.

6.3.1

Absorber model validation

The purpose of the dynamic tests is to obtain data for the qualitative and quantitative validation of the transient performance of the process model. Regarding the absorber column, the holdup in the liquid and vapour phase is the main process variable which determines the dynamic system response. Hence, the experimental data are used to qualitatively validate the correct choice of the holdup correlation, see Equation 6.7 and Equation 6.8. Note that the adopted correlation is commonly used for higher gas capacity factors than observed in the pilot plant [24]. The hydrodynamic coefficient of the holdup correlation Ch can be tuned in order to achieve good agreement between model predictions and measurements. For the test run results presented in this chapter tuning of the hydrodynamic coefficient was not necessary and the default value was used throughout all simulations. The measured process variables used for the quantitative absorber model validation are volumetric flow rate, temperature, CO2 and H2 composition of the H2 -rich gas, absorber pressure and column pressure drop. The object diagram of the system model used for the absorber validation is depicted in Figure 6.3. The model comprises the absorber column and sump, a flow source for the syngas, the lean solvent and the rich solvent, a pressure drop representing frictional losses of the H2 -rich gas in the overhead cooler, a gas tank representing storage of mass in the overhead cooler, the H2 -rich gas control valve including the absorber pressure PI controller, a quadratic pressure drop representing the frictional losses in the piping delivering the H2 -rich gas to the flare and a pressure sink representing the flare. The friction coefficient for the linear pressure drop of the overhead cooler was fitted to experimental data in order to match outlet pressures for on- and off-design steady-state operation. The friction coefficient of the quadratic pressure drop of the piping to the flare was tuned in a similar way in order to match pressures and the valve opening. Inputs of the model which fluctuate or change during the transient and influence the transient response of the system are chosen as prescribed variables. These are: syngas mass flow rate, solvent mass flow rate, solvent inlet temperature and the overhead cooler outlet temperature. The overhead cooler has not been modelled due to lack of measurement data. The other input variables such as syngas temperature, rich solvent pressure and flare pressure remain constant during the transient operation. The process measurements are obtained from the distributed control system of the pilot plant and are transferred for off-line data analysis to suitable data pro149

Chapter 6

ColumnCpressure control

realExpressi… 21.7 Solvent_Tin

Tcooling

AbsorberC column

offset=273.15 SolventMF

prescr…

PID pID

offset=273.15

p P

Mo… FP pressDropLi…

offset=0

LeanCsolvent flowCsource

SyngasMF

SetPoint_…

ELH80AA…

sensP

Plenum

pressDrop_E…

H2-richC gasCvalve

Overhead cooler

sourceW_De…

sinkP

H2-richCgasC pressureCsink (flare)

SumpClevelC control

duration=0

offset=0

sourceW_Sh…

SyngasCflow source

Sump

gain_l…

PID

k=1000

pID_ELF2…

RichCsolvent pressureCsink P

ELH20AA…

sinkP_Loade…

Figure 6.3: Object diagram of the system model used for the absorber validation.

cessing tools. Measurements are recorded when changes in variable values exceed a threshold which was set for most of the variables to 0.1 % of the individual measurement range. All instruments were calibrated either on-site or at the production facility typically at nominal operating conditions. The measurement uncertainty is estimated with about 1 % for most of the instruments. Open-loop tests (i.e., control system not in operation) are most suitable in order to validate that the process dynamics are captured correctly by the dynamic model. However, open-loop tests cannot be performed safely on the absorption and solvent regeneration unit. Partial open-loop test were therefore designed, in which only control loops which do not compromise the safe and stable operation were put into manual mode. The transient of such disturbance rejection experiments is thus the result of the interaction between the process dynamics and the inherent control loop dynamics and primarily provides relevant dynamic information in the frequency range around the controller’s crossover frequency. The evaluation of measured and predicted dynamics is based on the main transient parameters such as time and value of maximum overshoot, settling time, frequency and damping of oscillations. Change in solvent mass flow rate (test run TR-Solvent) The process conditions of the syngas and lean solvent flow are summarized in Table 6.3. During this experiment, the mass flow control valve of the solvent was manually changed without the solvent flow controller in operation. The perturbations to the valve opening correspond to changes in solvent flow of 2, 4 and 5 kg/s. The validation is performed with process data acquired during the test in which the solvent mass flow rate was increased from 10 kg/s to 15 kg/s. The change in solvent mass flow rate is depicted in Figure 6.4(a). The syngas mass flow rate, which is a prescribed variable for the dynamic model, is shown in Figure 6.4(b). The opening of the valve controlling the syngas mass flow is maintained constant during the test in order to keep the syngas flow constant. However, during the change of the absorber column pressure, which 150


Process variable

Shifted syngas

Lean solvent

Pressure [bar] Temperature [◦ C] Mass flow rate [kg/s] Mole fraction CO H2 CO2 H2 O N2 DEPEG

22a 40 0.39b

22a 40b 15b

0.024 0.546 0.375 0.004 0.051 0

5 ppm 30 ppm 0.031 0.106 30 ppm 0.862

a

Initial start value. The actual value is an output of the simulation. b Initial steady-state value. This variable is a prescribed variable for the dynamic model, hence measurements are used as input. Table 6.3: Process condition of syngas and lean solvent stream.

is the same as the back pressure to the syngas control valve, the syngas mass flow varies. Though, the syngas flow changes are relatively small compared to the changes in solvent flow. Considering the absorber pressure response depicted in Figure 6.4(c), the stepwise increase in solvent mass flow rate initially results in a column pressure increase. The pressure control opens the valve in order to maintain the pressure set point. Subsequently, the pressure decreases again and reaches its set point value after a few oscillations. The increase in solvent flow rate enhances CO2 absorption resulting in a lower H2 -rich gas flow, which leads to a decrease of the pressure drop over the overhead cooler. The final steady-state column pressure is therefore slightly lower as the pressure is controlled downstream of the overhead cooler. First, the performance of the pressure control loop is analysed by considering the transient of the pressure at the top of the absorber column (process or controlled variable), the opening of the H2 -rich gas flow valve (control variable) and the H2 -rich volumetric flow rate, which are depicted in Figure 6.4(c), Figure 6.4(d) and Figure 6.4(e) respectively. It can be observed that the frequency and damping of the oscillations, which are determined by the dynamic interaction of the controller and process dynamics, are well matched by the simulation. Thus, the dynamic influence of the control variable on the controlled variable is captured correctly in the frequency range which is relevant for the closed-loop performance, i.e., around the crossover frequency (about 0.016 rad/s). The process response to the applied disturbance on the other hand is overestimated during the initial part of the transient, where the model predicts a much large change than observed in the experimental measurements. For example, the initial pressure peak is predicted to be 3 times larger than measured. Thereafter, 151

Chapter 6



16 15 14 13 12 11 Experiment Model

10 9 0

5

10

15

20 25 Time [s]

30

35

1440 1420 1400

0

28

Valve opening [%]

30

22.4

22 21.8 21.6

21.2 0

5

10

15

20 25 Time [s]

30

35

56 54 52 50

35

40

24 22

5

10

15

20 25 Time [s]

30

35

40

Experiment Model

45 44 43

48 46 0

5

10

15

20 25 Time [s]

30

35

42 0

40

(e) H2 -rich gas volumetric flow rate.

10

15

20 25 Time [s]

30

35

40

5 Experiment Model 5

10

15

20 25 Time [s]

30

35

40

(g) Pressure drop over the 1st packing.

Pressure drop [mbar]

15

10

0 0

5

(f) Gas temperature between packings.

15 Pressure drop [mbar]

30

Experiment Model

46

Experiment Model

58

20 25 Time [s]

(d) Opening of the H2 -rich gas valve.

Temperature [°C]

H2−rich gas flow rate [m 3/s]

60

15

26

18 0

40

(c) Pressure at the top of the absorber.

10

20

Experiment Model

21.4

5

(b) Syngas mass flow rate (model input).

22.6

22.2

Experiment Model

1380

40

(a) Solvent mass flow rate (model input).

Absorber pressure [bar]

1460

10

5

0 0

Experiment Model 5

10

15

20 25 Time [s]

30

35

40

(h) Pressure drop over the 2nd packing.

Figure 6.4: TR-Solvent: Comparison of measurements and simulation results.

152


however, deviations are rather small. This means that the high-frequency response of the model is overestimated, compared to reality. This is probably due to some neglected phenomenon which has a damping action on the effect of the disturbance. At the top of the absorber column, a liquid distributor is located, which can store liquid in the order of 100 litres and thus delay fast transient changes in the solvent flow. This liquid distributor is not included in the dynamic model. The more aggressive variable changes predicted by the simulation might therefore be explained by the fact that storage occurs in the distributor particularly during the initial transient which is not modelled. Another possible explanation might be due to the fact that the absorber is modelled as the connection of a finite number of volumes at thermodynamic equilibrium, while the real process is a continuum of non-equilibrium mass and energy transfer phenomenon, which might react more slowly to this kind of disturbance. As far as steady-state values are concerned, the model underpredicts the initial off-design value of the H2 -rich flow and consequently the opening of the valve. This deviation to the measured flow value, which is in the order of 4 %, is related to the fact that, at L/G ratios much smaller than the ratio at the design point, the equilibrium-based model overpredicts the absorption efficiency (see Subsection 6.2.3). However, the accuracy of the prediction of dynamic indicators is not affected by this deviation. In the following the results of the pressure drop of the absorber packing is discussed. The adopted Billet and Schultes pressure drop correlation includes the friction coefficient Cp , which can be tuned in order to match the measured pressure drop over the column. As the dynamic pressure drop of the two column packings is not measured individually but lumped with the frictional losses of the liquid distributors, a fitting of the measured values to determine the friction coefficient is not meaningful. Moreover, due to numerical instabilities of the simulations in case of small dynamic pressure drop, the friction coefficient was adjusted such that the overall column pressure drop is approximately 20 mbar. Such a relatively high value of the dynamic pressure drop significantly enhances the robustness of the simulations. This value is similar to the total pressure loss measured in the absorber column at nominal operating conditions, which includes the dynamic pressure drop and the distributor losses. In general, the pressure drop does not affect the dynamics of the absorber column model in terms of heat and mass transfer, hence the adjustment of the pressure drop for numerical reasons is justified. The comparison of the time-dependent measurements and model predictions of the pressure drop over the first and the second packing are visualized in Figure 6.4(g) and Figure 6.4(h). The first packing is situated at the upper part of the column and the second below. The transient of the working fluid pressure drop over the first packing is not predicted correctly. The model predictions only depend on the solvent and syngas volumetric flow rate and the densities of the vapour and liquid (see Billet and Schultes pressure drop correlation Equation 6.5 and Equation 6.6), while the pressure loss induced by the liquid distributor situated above the packing is not 153

Chapter 6

modelled. This might be the main reason of the discrepancy between measured and simulated values. The model predicts an initial increase of the pressure drop caused by the step increase in solvent mass flow rate. The pressure drop successively decreases and returns approximately to its initial value, which is explained by the delayed decrease of the H2 -rich gas flow associated with a lower frictional loss due to the increase of CO2 absorption. The model correctly predicts the pressure transient of the working fluid experienced by the second packing. However, the absolute change in pressure drop is underpredicted, which is explained by the fact that the redistributor situated above the second bed is not represented in the model. Inspection of Figure 6.4(h) shows that the pressure drop due to flow friction within the second packing increases with increasing solvent flow rate. Moreover, the gas flow rate variation due to enhanced CO2 absorption is smaller within the second packing than the first. It can therefore be concluded that the pressure drop variation depends mainly on the solvent mass flow rate. Due to the fact that measurements and simulation results of the pressure drop represent different phenomena (dynamic pressure drop and distributor losses versus adjusted dynamic pressure drop), the steady-state values will in principle not match. The friction coefficient of the dynamic model has however been adjusted such that a reasonable agreement is achieved for the final steady-state values at nominal condition. Figure 6.4(f) presents the comparison of measurements and model results for the response in gas temperature between the packings. The gas temperature temporarily increases caused by fluctuations in the lean solvent temperature, which is therefore applied as a prescribed variable for the dynamic model. The transient model predictions show satisfactory agreement with the measurements. Finally, Figure 6.5(a) and Figure 6.5(b) depict the measured H2 and CO2 content in the H2 -rich gas flow as function of time, together with their model-based predictions. The simulation results compare well with the experimental data, except for an initial time delay in the measurements. In this respect, it is important to mention, that the mixture composition is not measured continuously but values are recorded every three minutes. Time delays for sampling and composition analysis are only approximately known (about 300 seconds) and have already been corrected for in the comparison. The initial delay that can be observed in the recorded values cannot therefore fully be explained. The difference between the predicted and measured initial transient might, however, be attributed to the assumption of thermodynamic equilibrium for the absorber model. As far as initial steady-state values are concerned, the model overpredicts the value of H2 content and underpredicts the value of CO2 content. This is attributed to the overprediction of the CO2 absorption efficiency by the equilibrium-based model at this off-design operational point. 154

82 80 78 76 74 0

Experiment Model 5

10

15

20 25 Time [s]

30

35

(a) H2 content in H2 -rich gas flow.

40

CO2 content in H2−rich gas [%]

H2 content in H2−rich gas [%]


16

Experiment Model

14 12 10 8 6 0

5

10

15

20 25 Time [s]

30

35

40

(b) CO2 content in H2 -rich gas flow.

Figure 6.5: TR-Solvent: Comparison of measurements and simulation results.

Change in syngas mass flow rate (test run TR-Syngas) During this test run, manual step changes were applied to the back pressure control valve of the upstream syngas compressor without the pressure control in operation. The validation is performed with the data recorded during the experiment in which the syngas mass flow rate was decreased from 1400 kg/h to 800 kg/h. The change in syngas mass flow rate is depicted in Figure 6.6(b). The set point of the lean solvent mass flow controller was kept constant during the transient in order to maintain the flow rate. However, the solvent mass flow rate fluctuates slightly, by less than 1.5 % (see Figure 6.6(a)). This fluctuations are due to changes in the absorber column pressure during the test. Since the absorber column pressure is the same as the back pressure to the solvent control valve, the solvent flow controller needs to adjust. Therefore, the solvent flow rate is set as prescribed variable for the dynamic model. For the absorber pressure, the step decrease in syngas mass flow rate results in a drop of the column pressure. Subsequently, the pressure controller adjusts the opening of the H2 -rich gas valve to return to the given pressure set point. As a result, the pressure increases again and reaches its final steady-state condition after a few oscillations. The final steady-state value is slightly lower than its initial value. As explained for the validation of the test run TR-Solvent, this off-set is related to the decrease in H2 -rich gas flow rate. The predicted and measured transient of the variables related to the pressure control loop are compared in Figure 6.6(c) – pressure at the top of the absorber column, in Figure 6.6(d) – opening of the H2 -rich gas flow valve, and in Figure 6.6(e) – H2 -rich volumetric flow rate. Excellent agreement is achieved for the closed-loop performance, i.e., the dynamic interaction of the controller and process dynamics, and for the process response to the disturbance. In comparison to the test run TR-Solvent, the model predictions for the pressure control loop are significantly better, in particular considering the initial transient. While during TR-Solvent the damping action of the liquid distributor might 155

Chapter 6



15.3 15.2 15.1 15 14.9 14.8

Experiment Model

14.7 0

5

10

15

20 25 30 Time [s]

35

40

1200 1000

0

Valve opening [%]

21.7 21.6

Experiment Model 5

10

15

20 25 30 Time [s]

35

40

15

20 25 30 Time [s]

(c) Pressure at the top of the absorber. 60

45 40 35

45

Experiment Model

10

5

10

15

20 25 30 Time [s]

45 Temperature [°C]

50

40

35

40

45

(d) Opening of the H2 -rich gas valve.

Experiment Model

55

35

15

5 0

45

Experiment Model

44 43 42

30 25 0

5

10

15

20 25 30 Time [s]

35

40

(e) H2 -rich gas volumetric flow rate. 14

10 8 6 4 2 0 0

5

10

15

20 25 30 Time [s]

35

40

5

10

15

20 25 30 Time [s]

35

40

45

(f) Gas temperature between packings.

Experiment Model

12

41 0

45

45

(g) Pressure drop over the 1st packing.

14 Pressure drop [mbar]

Absorber pressure [bar]

21.8

H2−rich gas flow rate [m 3/s]

10

20

21.9

Pressure drop [mbar]

5

(b) Syngas mass flow rate (model input).

22

21.5 0

800

45

(a) Solvent mass flow rate (model input).

Experiment Model

1400

Experiment Model

12 10 8 6 4 2 0

5

10

15

20 25 30 Time [s]

35

40

45

(h) Pressure drop over the 2nd packing.

Figure 6.6: TR-Syngas: Comparison of measurements and simulation results.

156


have effected the disturbance, this phenomenon is not relevant for TR-Syngas and might therefore explain the much better agreement. It is also worth pointing out that the perturbation of the syngas mass flow rate during TR-Syngas is a ramp with a duration of approximately 600 seconds, whereas the ramp duration of TR-Solvent is roughly 300 seconds. This might indicate that the model predictions are less accurate for high-frequency response, possibly due to the use of a equilibrium-based absorber column model. As far as steady-state values are concerned, the model accurately predicts the initial and final steady-state values of the variables related to the absorber pressure control. The largest deviations are observed for the off-design value of the H2 -rich gas flow. Figure 6.6(g) and Figure 6.6(h) visualize the comparison of the experimental data and model predictions for the pressure drop over the first and second packing. For both packings the transient response, a decrease in pressure drop caused by a decrease in vapour flow, is predicted correctly. The change in vapour flow has two reasons: first the decrease in entering syngas flow and second the decrease in H2 -rich product flow as a result of enhanced CO2 absorption. The absolute values for the initial and final steady-state pressure drops are not predicted correctly. As explained for TR-Solvent, the pressure drop of the liquid distributors above the first and second bed are not modelled and therefore measured losses cannot be compared with model predictions which only represent the dynamic pressure drop. During test run TR-Syngas, the solvent mass flow rate and therewith the pressure drop of the liquid distributors remained almost constant. Hence, the change in pressure drop is mainly related to the variation in vapour flow. Consequently, the absolute change in pressure loss could be used to tune the friction coefficient of the adopted pressure drop correlation in order to match experimental data. However, this fitting was not performed as it was required to adjust Cp in order to improve the numerical robustness of the simulation (see discussion for test run TR-Solvent). Figure 6.6(f) shows the measurements and model predictions for the transient of the gas temperature between the packings. The temporary increase in gas temperature is caused by fluctuations in the lean solvent temperature which is thus defined as prescribed variable. The model predictions show good agreement with the experimental data. The observed deviations in the final steady-state are approximately 0.5 ◦ C. Finally, Figure 6.7(a) and Figure 6.7(b) show the comparison of the experimental data and model predictions for the H2 and CO2 content in the H2 -rich gas flow. With a decrease in syngas mass flow at constant solvent flow rate, the CO2 absorption efficiency increases which leads to a lower CO2 content and higher H2 content in the H2 -rich product flow. The transient response as well as the steady-state values are predicted correctly. A slight time delay is observed in the measurements similar to experiment TR-Solvent. 157

83.5 83 82.5 82 81.5 81 80.5 0

Experiment Model 5

10

15

20 25 30 Time [s]

35

40

(a) H2 content in H2 -rich gas flow.

45

CO2 content in H2−rich gas [%]

H2 content in H2−rich gas [%]

Chapter 6

8.5

Experiment Model

8 7.5 7 6.5 6 5.5 0

5

10

15

20 25 30 Time [s]

35

40

45

(b) CO2 content in H2 -rich gas flow.

Figure 6.7: TR-Syngas: Comparison of measurements and simulation.

To summarize, with respect to the steady-state values of the main process variables, excluding the packing pressure drop, the following conclusions can be drawn: a) at on-design operation (L/G = 38.6) the variable values are reproduced with an error of less than 1 % and b) at off-design operation (L/G = 67.5 & 25.7) the values are reproduced with less than 5 % error. These results can be considered as satisfactory, keeping in mind that the latter deviations are related to the equilibrium assumption (see Figure 6.2). Considering the main dynamic parameters of the observed transients of absorber pressure and temperature, H2 -rich gas flow rate and composition, namely, time and value of maximum overshoot, settling time, frequency and damping of oscillations, then these parameters are predicted with less than 15 % error; larger deviations are observed during the initial transient of TR-Solvent related to the fact that a possible delay in solvent flow variation is not modelled. In general, this agreement between the experimental data and model predictions can be considered as satisfactory. The transient and the absolute values of the column pressure drop over the first and second packing is in most cases not predicted correctly due to the fact that the frictional losses of the distributors are not modelled. In addition, agreement in the absolute values is not achieved because the friction coefficient Cp had to be adjusted in order to enhance numerical robustness of the simulations.

6.3.2

Absorption and solvent regeneration section model validation

After validation of the standalone absorber model, the model of the absorption and solvent regeneration section was validated. The process variables used for the quantitative comparison of the transient performance of the model with the recorded measurements are the volumetric vapour flow rates and pressures of the three flash vessels, and the CO2 and H2 composition of the 3rd flash. The object diagram of the absorption and solvent regeneration section model 158


Solvent_Tin

realExpressi…

AbsorberC column

offset=273415 SolventMF

Tcooling

2147

prescr…

PID offset=273415

p

pID

ColumnCpressure control P

Mo… FP offset=0

pressDropLi… ELH70AC010

sensP

OverheadCcooler

sourceW_De…

LeanCsolvent flowCsource

ELH80AA…

H2-rich gasCvalve

SetPoint_…

SyngasMF SetPoint_…

duration=0

offset=0

SumpClevelC control

pressDrop_E… sinkP_H2Ric…

H2-richCgas pressureCsinkC

1stCflashC pressureC control PID pID_ELH…

p sensP_Flash1

ELH20AA…

PID sourceW_Sh…

SyngasCflow source

pID_ELF2…

SetPoint_…

duration=0

Sump

PID ELH20AA…

2ndCflashC pressureC control

duration=0

pID_ELH…

1stCflash levelC control

duration=0

p

PID

sensP_Flash2

PF3

ELH30AA…

ELH20AA…

3rdCflashC vessel

offset=0

P sinkP_Flash3

ELH40AA…

duration=0 gain_l… k=1000

PID pID_ELH…

LeanCsolvent flowCsink

SolventMF1

SetPoint_…

sinkW_Load…

3rdCflashC levelCcontrol

CO2Cproduct pressureCsink

pID_ELH…

offset=0

1stCflashCvessel 2ndCflashCvessel

SetPoint_…

Figure 6.8: Object diagram of the system model used for validation of the absorption and solvent regeneration section.

used for validation is depicted in Figure 6.8. Here, the model of the absorber (see Figure 6.3) is extended with three flash vessels, two gas control valves with PI pressure controller and two liquid control valves with PI level controller, a CO2 product pressure sink and a lean solvent flow sink. In the solvent regeneration section the pressure of the 3rd flash vessel and the lean solvent flow leaving the 3rd flash are applied as prescribed variables. Change in syngas mass flow rate The validation was performed using the experimental results of test run TRSyngas, in which the syngas mass flow rate was changed stepwise from 1400 kg/h to 800 kg/h, see Figure 6.6(b). As response to the decrease in syngas flow rate, it is expected that the vapour flow rates at the flash vessels decrease. Subsequently, the vessel pressures drop initially and the pressure controllers act in order to maintain the pressure set points. The experimental data and simulation results of the dynamics of the pressures and volumetric vapour flow rates of the three vessels are compared in Figure 6.9. It can be observed that the presence of oscillations is captured correctly, which are primarily the result of the interaction between the controller and the process dynamics. The damping of the oscillations however is slightly underestimated by the model. The transient response of the process to the disturbance is predicted satisfactory for the 1st and 3rd flash vessel, however for the 2nd flash the model estimates a much larger change, in particular in the pressure, than observed in the measurements. Noticeable for the 2nd vessel is also the difference in the initial steady-state value of the volumetric flow rate, which is underpredicted by the model by 10 m3 /s, 159

Chapter 6

see Figure 6.9(d). The difference even increases for off-design. In general, the volumetric vapour flow rate is very sensitive to the temperature and pressure in the vessel. The results indicate that one or both variables are not predicted accurately enough by the simulation in order to achieve good agreement in the volumetric flow rate. As no sufficient temperature measurements were available for the vessels, possible heat losses along the process have not been included in the model and a comparison of vessel temperatures could not be performed. Concerning the pressure, recorded measurements can be subject to biases, hence the actual vessel pressure value might be different than implemented in the model. It is also worth pointing out that saturation is assumed for the vapour flow in the model of the flash vessel. This simplification leads to an underprediction of the vapour flow rate in case the actual vapour quality is smaller than 1. This might in particular occur during fast pressure transients. These possible reasons for the difference in steady-state values of the volumetric flow would also impact the transient of the pressure control loop of the 2nd vessel. The experimental data and the model prediction for the CO2 and H2 content in the 3rd flash vapour flow are depicted in Figure 6.9(g) and Figure 6.9(h). An offset of 300 seconds related to sampling and composition analysis has been removed from the experimental data. Overall, very good agreement is observed in terms of transient response but also considering the absolute values, whereby the H2 content is marginally underpredicted by the model. To conclude, with respect to the values of the main process variables at initial steady-state (data recording stopped before all variables reached final steady-state) it can be observed that most of the variables are reproduced with less than 2 % error, larger errors are found, for example, for the volumetric vapour flow rates in the first and second vessels. These deviations are attributed to the uncertainty in predictions for temperatures and/or pressures in the vessels. The dynamic parameters of the transients are predicted with less than 20 % error, larger errors are observed for pressure and volumetric vapour flow rate of the second vessel, which are most probably also related to the uncertainties mentioned above. The good agreement between simulation results and experimental data for the CO2 and H2 content in the CO2 product confirms that the composition of the rich solvent flow entering the solvent regeneration section is predicted accurately by the absorber model.

6.4

Conclusions

The work documented in this chapter demonstrates that a well-tuned, equilibriumbased dynamic model for physical absorption of CO2 provides sufficiently accurate transient performance predictions for the purpose of dynamic process analysis and control system design. The largest deviations between measurements and simulation results were observed during the initial transient of a test whereby a fast change was applied to the lean solvent mass flow rate. This might indicate 160

1st flash pressure [bar]

8 7.8 7.6 7.4 7.2 7 6.8 0

Experiment Model 5

10

15

20 25 30 Time [s]

35

40

45

1st flash vapour flow rate [m 3/s]


25

15 10 5 0 0

2nd flash pressure [bar]

3 2.95 2.9 2.85 2.8

2.7 0

Experiment Model 5

10

15

20 25 30 Time [s]

35

40

45

1.3 1.28 5

10

15

20 25 30 Time [s]

35

40

45

3rd flash vapour flow rate [m 3/s]

3rd flash pressure [bar]

1.32

1.26 0

90 85

5

10

15

20 25 30 Time [s]

35

40

45

35

Experiment Model

80 60 40 0

5

10

15

20 25 30 Time [s]

550

35

40

45

40

45

(g) CO2 content in 3rd flash vapour flow.

Experiment Model

500 450 400 350 300 250 0

5

10

15

20 25 30 Time [s]

35

40

45

(f) 3rd flash vapour flow rate. H2 in 3rd flash vapour flow [%]

CO2 in 3rd flash vapour flow [%]

Experiment Model

95

80 0

20 25 30 Time [s]

100

(e) 3rd flash pressure (model input). 100

15

(d) 2nd flash vapour flow rate.

Experiment Model

1.34

10

120

(c) 2nd flash pressure. 1.36

5

(b) 1st flash vapour flow rate. 2nd flash vapour flow rate [m 3/s]

(a) 1st flash pressure.

2.75

Experiment Model

20

14 12

Experiment Model

10 8 6 4 2 0 0

5

10

15

20 25 30 Time [s]

35

40

45

(h) H2 content in 3rd flash vapour flow.

Figure 6.9: TR-Syngas: Comparison of measurements and simulation results.

161

Chapter 6

that either some phenomena which have effect on the response have been neglected or that the model predictions are less accurate for high-frequency response due to fact that the absorber is modelled as the connection of a finite number of volumes at thermodynamic equilibrium. The accuracy of the model predictions during initial transient might be improved by using a rate-based absorber model. A ratebased model however is less suitable for analysis of the entire system comprising the power plant and the capture unit due to the increase in model complexity associated with higher computational effort. The validated process models have been implemented in an open source library, which serves as reliable basis for the development of models representing a large-scale CO2 absorption and solvent regeneration unit. The object-oriented modelling approach allows to easily develop new process models by re-use and/or extension of existing models and additional implementation of more sophisticated models if required. The dynamic model of the CO2 absorption and solvent regeneration unit can be easily combined with a model of the water-gas shift unit to obtain a full model of the pre-combustion CO2 capture plant. Ultimately, a model of the entire IGCC power plant can be formed by integrating object-oriented models of the gasification unit, the combined cycle and the CO2 capture plant. Such a system model can be extremely valuable for the design of new plants to support equipment selection and sizing, and to develop and test control strategies. Dynamic simulation are also useful to investigate and ultimately to improve dynamic performance of existing plants.

162


Nomenclature a C d ∆p Fr FV g H hoL K m M p Re T u v w X

= = = = = = = = = = = = = = = = = = =

Specific surface area, m2 m−3 Constant Diameter, m Pressure drop, Pa Froude number Gas/vapour capacity factor, Pa−0.5 Gravitational acceleration, m s−2 Height, m Column holdup, m3 m−3 Wall factor Mass flow rate, kg s−1 Total mass, kg Pressure, Pa Reynolds number Temperature, K Internal energy, J kg−1 Specific volume, m kg−3 Velocity with reference to free column cross-section, m3 m−2 Mass fraction

= = = =

Void fraction, m3 m−3 Dynamic viscosity, kg ms−1 Density, kg m−3 Resistance coefficient

= = = = = = =

Hydraulic Mixture component Inlet Liquid phase Particles Loading point Vapour phase

= = = = = =

Carbon capture and storage Differential and algebraic equation Dimethylether of polyethylene glycol Equation of state Integrated gasification combined cycle Liquid-to-gas ratio

Greek symbols ε η ρ ψ Subscripts h i,j in L P S V Acronyms CCS DAE DEPEG EoS IGCC L/G

163

Chapter 6

PC-SAFT TR

164

= =

Perturbed chain - statistical associating fluid theory Test run


References [1] Heil, S., Brunhuber, C., Link, K., Kittel, J., and Meyer, B. “Dynamic Modelling of CO2 -removal units for an IGCC power plant”. In Proceedings 7th Modelica Conference, Como, Italy, Sep. 20-22, 2009, pp. 77 – 85. [2] Chikukwa, A., Enaasen, N., Kvamsdal, H. M., and Hillestad, M., 2012. “Dynamic Modeling of Post-combustion CO2 Capture Using amines - A Review”. Energy Procedia, 23(0), pp. 82–91. [3] Wang, M., Lawal, A., Stephenson, P., Sidders, J., and Ramshaw, C., 2011. “Postcombustion CO2 capture with chemical absorption: A state-of-the-art review”. Chemical Engineering Research and Design, 89(9), pp. 1609–1624. [4] Kucka, L., M¨ uller, I., Kenig, E., and G´ orak, A., 2003. “On the modelling and simulation of sour gas absorption by aqueous amine solutions”. Chemical Engineering Science, 58(16), pp. 3571–3578. [5] Lawal, A., Wang, M., Stephenson, P., and Yeung, H., 2009. “Dynamic modelling of CO2 absorption for post combustion capture in coal-fired power plants”. Fuel, 88(12), pp. 2455–2462. [6] Kale, C., G´ orak, A., and Schoenmakers, H., 2013. “Modelling of the reactive absorption of CO2 using mono-ethanolamine”. International Journal of Greenhouse Gas Control, 17, pp. 294–308. [7] Ziaii, S., Rochelle, G. T., and Edgar, T. F., 2011. “Optimum design and control of amine scrubbing in response to electricity and CO2 prices”. Energy Procedia, 4(0), pp. 1683–1690. [8] Pr¨ olß, K., Tummescheit, H., Velut, S., and ˚ Akesson, J., 2011. “Dynamic model of a post-combustion absorption unit for use in a non-linear model predictive control scheme”. Energy Procedia, 4, pp. 2620–2627. [9] Harun, N., Nittaya, T., Douglas, P. L., Croiset, E., and Ricardez-Sandoval, L. A., 2012. “Dynamic simulation of MEA absorption process for CO2 capture from power plants”. International Journal of Greenhouse Gas Control, 10(0), pp. 295–309. [10] Jayarathna, S., Lie, B., and Melaaen, M., 2013. “Amine based CO2 capture plant: Dynamic modeling and simulations”. International Journal of Greenhouse Gas Control, 14, pp. 282–290. [11] Dietl, K., 2012. “Equation-Based Object-Oriented Modelling of Dynamic Absorption and Rectification Processes”. PhD thesis, Hamburg University of Technology. [12] Lawal, A., Wang, M., Stephenson, P., Koumpouras, G., and Yeung, H., 2010. “Dynamic modelling and analysis of post-combustion CO2 chemical absorption process for coal-fired power plants”. Fuel, 89(10), pp. 2791–2801. [13] Kvamsdal, H., Chikukwa, A., Hillestad, M., Zakeri, A., and Einbu, A., 2011. “A comparison of different parameter correlation models and the validation of an MEAbased absorber model”. Energy Procedia, 4, pp. 1526–1533.

165

Chapter 6

[14] Posch, S., and Haider, M., 2013. “Dynamic modeling of CO2 absorption from coalfired power plants into an aqueous monoethanolamine solution”. Chemical Engineering Research and Design, 91(6), pp. 977–987. [15] ˚ Akesson, J., Faber, R., Laird, C., Tummescheit, H., Velut, S., and Zhu, Y., 2011. “Models of a post-combustion absorption unit for simulation, optimization and nonlinear model predictive control schemes”. In Proceedings 8th International Modelica Conference 2011, Dresden, Germany. [16] Biliyok, C., Lawal, A., Wang, M., and Seibert, F., 2012. “Dynamic modelling, validation and analysis of post-combustion chemical absorption CO2 capture plant”. International Journal of Greenhouse Gas Control, 9, pp. 428–445. [17] Taylor, R., and Krishna, R., 1993. Multicomponent Mass Transfer. John Wiley and Sons, Inc., New York, USA. [18] Aspen Technology, Inc., 2013. Aspen Plus V7.3. www.aspentech.com. [19] L¨ otstedt, P., and Petzold, L., 1986. “Numerical solution of nonlinear differential equations with algebraic constraints I: Convergence Results for Backward Differentiation Formulas”. Mathematics of Computation, 46(174), pp. 491–516. [20] Unger, J., Kr¨ oner, A., and Marquardt, W., 1995. “Structural analysis of differentialalgebraic equation systems-theory and applications”. Computers and Chemical Engineering, 19(8), pp. 867–882. [21] Trapp, C., Casella, F., van der Stelt, T. P., and Colonna, P., 2014. “Use of External Fluid Property Code in Modelica for Modelling of a Pre-combustion CO2 Capture Process Involving Multi-Component, Two-Phase Fluids”. In Proceedings 10th Modelica Conference, Lund, Sweden, March 10-12. [22] Casella, F., Van Putten, J., and Colonna, P., 2008. “Dynamic simulation of a biomass-fired steam power plant: a comparison between causal and a-causal modular modeling”. In ASME International Mechanical Engineering Congress and Exposition, Vol. 6, pp. 205–216. [23] Casella, F., and Colonna, P., 2012. “Dynamic modeling of IGCC power plants”. Applied Thermal Engineering, 35, pp. 91–111. [24] Billet, R., and Schultes, M., 1999. “Prediction of Mass Transfer Columns with Dumped and Arranged Packings: Updated Summary of the Calculation Method of Billet and Schultes”. Chemical Engineering Research and Design, 77, Issue 6, pp. 498–504. [25] Casella, F., and Leva, A., 2006. “Modelling of Thermo-Hydraulic Power Generation Processes Using Modelica”. Math. Comput. Model. Dyn. Syst., 12(1), pp. 19–33. [26] Casella, F., and Leva, A., 2005. “Object-oriented modelling & simulation of power plants with modelica”. In Proceedings of the 44th IEEE Conference on Decision and Control, and the European Control Conference, CDC-ECC ’05, pp. 7597–7602.

166

¨ “Das tritt nach meiner Kenntnis...ist das sofort, unverzuglich.” “As far as I know...effective immediately, without delay.”

G¨ unter Schabowski, official of the Socialist Unity Party of the German Democratic Republic, press conference, East-Berlin, shortly before 7 pm, November 9, 1989 At 10:45 pm the gates towards West-Berlin were opened.

7

Conclusions and Perspectives This dissertation covers the analysis and design of pre-combustion CO2 capture systems applied to future integrated gasification combined cycle (IGCC) power plants. This concluding chapter summarizes the key contributions of this research to the scientific field of CO2 capture process design. Furthermore, the generic aspects of the adopted system engineering techniques and tools are emphasized. Finally, possible evolutions of this research are discussed.

Chapter 7

7.1

Conclusions

This thesis discusses the design of a flexible and prompt pre-combustion CO2 capture plant adopting a methodology based on first principle process modelling and pilot plant experiments. The motivation for this work stems from recent developments in the power sector related to the commitment to more stringent climate targets in order to reduce emissions and promote the transition to a sustainable electricity generation. It is therefore expected that constraints on carbon emissions require to equip fossil-fuelled power plants, such as future IGCC systems, with CO2 capture units. The removal of CO2 will result in a large efficiency penalty for the thermal power plant. Moreover, transient performance of these decarbonised power plants becomes increasingly important as the share of electricity produced by renewable sources, which is inherently intermittent, is steadily growing. The resulting pre-combustion CO2 capture design problem was divided into two parts, the steady-state design optimization targeting reduction in energy consumption, while considering variations in operating parameters, and the simulationbased improvement of the dynamic performance by process and control system design. Economic design aspects were not included apart from preliminary considerations. Important design variables of the pre-combustion CO2 capture process are the extent of CO conversion in the water-gas shift unit and the extent of CO2 capture in the CO2 and H2 S removal unit, which determine the overall carbon capture rate. The most relevant environmental target is, except for the carbon capture rate, the maximum total sulphur content in CO2 product gas, which directly influences the energy consumption of the removal unit. With respect to operational limits, the steam/CO ratio in the water-gas shift unit is the most important parameter in terms of material constraints but also energy consumption. For the steady-state optimization, up to six decision variables were considered. By activating inequality constraints when appropriate, the degree of freedom of the optimization problem was typically reduced to two. The main uncertainties in the design models are related to the estimation of thermophysical properties, which however feature a satisfactory accuracy in relation to the engineering purpose. In the following, conclusions are drawn for each step of the proposed design method. Experiments were designed and performed at the CO2 capture pilot facility built at the Buggenum IGCC power station in the Netherlands, in order to obtain operational experience and to generate measurement data for comprehensive model validation. It was demonstrated that the validated steady-state pilot plant model predicts the process performance with satisfactory accuracy in relation to its engineering purpose, namely process analysis and design optimization. Moreover, the validation showed that the employed perturbed chain - statistical associating fluid theory (PC-SAFT) equation of state accurately predicts the thermophysical properties of the two-phase multi-component mixtures involved in the process for the considered operating range. Measurement data and acquired knowledge resulting from experiments fur168

Conclusions and Perspectives

thermore allow to identify the most relevant process parameters and values for operational limits. These information are valuable input in order to perform realistic process analyses and to formulate meaningful design optimization problems. During the preliminary design phase of a pre-combustion CO2 capture plant, operational flexibility in terms of environmental and operational limits, and/or targets should be addressed. Variations of these values might have large impact on the plant configuration and the optimal operating conditions. To that end, a preliminary design optimization of a large-scale CO2 capture plant was performed targeting reduction in energy consumption, while considering i) flexible operation in terms of overall carbon capture rate, ii) different operational limits of steam/CO ratio, and iii) deactivation of catalyst activity throughout the catalyst life. From the optimization results it can be observed that there exists an optimum for the overall carbon capture rate, at which the specific energy consumption of the capture unit is minimal. This minimum is at the operational point at which the limit of the steam/CO ratio is reached at the inlet of one of the water-gas shift reactors. The value of the optimal capture rate depends on various modelling choices and assumptions, such as plant configuration or values of process parameters. During the experimental campaign, the catalyst performance and its resistance to iron carbide formation was investigated for reduced steam/CO ratios, which implies that less steam is required in the water-gas shift unit. Molar steam/CO ratios down to 1.5 at the inlet of the reactors were successfully tested. The design optimization revealed that for a minimum molar steam/CO ratio of 1.5 the specific energy consumption of the capture unit can be reduced by 10 % in comparison to a minimum ratio of 2.65, which corresponds to the value recommended by the vendor. However, this comes at the expense that the optimal carbon capture rate lowers from 87.5 to 78 %, and this is related to the decrease in CO conversion due to reduced steam/CO ratio. The influence of catalyst deactivation on the performance of the capture unit in terms of energy consumption and optimal operating conditions is rather small. The adjustment of the reactors’ inlet temperature, which is required due to catalyst deactivation, should therefore not focus primarily on energy efficiency but rather on safe and stable reactor operation. The various optimization studies performed during this work considered primarily the energy consumption as objective, which is justified for a preliminary design analysis. However, the final design optimization must include the economic evaluation. It turned out that already for analysing the impact of the solvent temperature on the performance of the H2 S and CO2 removal unit techno-economic aspects, such as the column sizing, could not be neglected. For this purpose, the comparison of a configuration using chillers with a scheme based on water coolers for solvent temperature conditioning was performed, and it was concluded that the chilled-solvent design is advantageous both in terms of energy efficiency and equipment cost. In general, the steady-state design optimization of the pre-combustion CO2 169

Chapter 7

capture unit demonstrated that the applied two-step method,1 comprising of a global and a local design phase, allows to analyse and optimize complex energy conversion processes by dividing the design problem into manageable optimization sub-problems, whereby the final plant design is obtained by iteration. A decomposition approach was chosen as a rigorous optimization was not possible due to computational limitations. Another viable solution for energy efficiency improvement of the pre-combustion CO2 capture plant is the recovery of low-grade thermal energy by means of an ORC power system. Waste-heat can be recovered both downstream of the water-gas shift unit (syngas stream) and throughout CO2 compression train (CO2 product stream) allowing to save approximately 10 % of the energy consumption of the capture unit. The unique aspect of the recovery of thermal energy from the syngas source is the fact that the syngas undergoes condensation during the cooling, which prevents the typical pinch-point problem in the ORC primary heat exchanger from occurring. Thus, the syngas can be cooled to a much lower temperature, by means of an appropriately designed ORC system, resulting in an increased amount of thermal energy recovered. The second part of this thesis treats the dynamic modelling and simulation of the pre-combustion CO2 capture plant for the use in control design. It was demonstrated that a modular approach together with an object-oriented equation-based modelling language is most suitable to master the complexity of such a system. Extensive model validation was performed at component, sub-system and system level in order to improve the reliability and the accuracy of the model predictions. Satisfactory agreement was achieved between the dynamic simulation results and experimental measurements obtained from various open- and closed-loop step response tests performed at the pilot plant. A more detailed validation is presented for one of the key process components, the absorber column. It was concluded that a well-tuned, equilibrium-based dynamic model for physical absorption of CO2 provides sufficiently accurate performance predictions concerning transient operation. The validated system model of the capture pilot plant was used to perform dynamic simulations in order to investigate a control strategy based on feed-forward, feed-back and cascade control with the aim to improve transient performance. It was preliminarily demonstrated that prompt syngas load variations are a feasible operating mode with the proposed control system. This would enable the IGCC power plant to respond to fast changes in the energy demand by reducing the load of the energy-intensive CO2 capture process instead of adjusting the gasifier load. The qualitative results from this analysis can be applied to the design of control strategies of a large-scale CO2 capture plant, whereby the validated component and pilot plant system models serve as reliable basis for the development of models of commercial-scale plants. One of the challenges encountered during the development of the dynamic 1 Adopted

170

from Bhattacharyya et al. [1] with modifications.


models of the CO2 capture plant was related to the computation of thermophysical properties of the working fluids within the process models. To this purpose, an interface prototype was developed which allows to interface external fluid property code with the dynamic modelling tool. This approach provides the advantage that the process fluids can easily be changed, as typically a wide range of fluid models are available within external property packages. The use of external fluid property functions in the dynamic process models imposes some restrictions to the model development. It is therefore necessary to follow a more causal modelling approach and to choose the system state variables as well as thermodynamic states appropriately. It appeared that for this type of dynamic simulations the computation of thermophysical properties, in particular phase equilibria, accounts for the main share of the simulation time. Various aspects for improvement of the computational efficiency are therefore discussed, such as single versus two-phase property calculations, decrease in redundancy, and approximation of partial derivatives. As concluding remark it is worth to highlight that the developed steady-state and dynamic process models are powerful tools for the analysis and design of flexible and prompt CO2 capture plants, which can be of manifold use during the early design phase up to plant commissioning and operation. With respect to the generic aspects of the adopted system engineering techniques and tools, it is important to emphasize under which conditions these are transferable to the design process of other chemical and energy conversion systems. The approach to model development and validation, to experiment design and execution, to process simulation and design optimization can be considered as transferable with minor changes. The methods, however, have been adapted, to some extent, to the limitations of the specific computational tools. A rigorous steady-state design optimization was not possible with the used process simulator due to computational limitations, and therefore a step-wise approach with a global and local design phase was adopted. Moreover, assigning decision variables to either the global or local design phase will be case specific. The dynamic process models were developed following a more causal modelling approach in order to obtain as model equations a differential algebraic equation (DAE) system of index 1. This was required in order to allow a straightforward use of external fluid property functions in Modelica process models. In case computational tools with improved capabilities become available, then more rigorous methods than the ones proposed could be adopted. In general, the tools used for simulation and optimization are suitable for design of other systems if the nature of the decision variables is the same and degree of freedom is similar. The domain knowledge related to the analysis of pre-combustion CO2 capture plants, i.e., the physical and operational conditions of the system, the process parameters, and the system configuration, might be rather different from other chemical and energy conversion systems. The domain knowledge presented in this thesis is therefore not generally transferable, except possibly for similar CO2 capture plants. Lastly, in the context of generalization it is also important to highlight to which extent the simulation and optimization were tailored to specific non-linear 171

Chapter 7

features of the domain knowledge models. With respect to the steady-state tools, the handling of the inequality process constraints has been tailored to the operational range of the pre-combustion CO2 capture plant. For example, some of the inequality constraints are activated during the optimization in order to reduce the degree of freedom and therewith simplify the optimization. In general, in order to ease initialization and numerical convergence the steady-state and dynamic models are typically initialized with good starting values for the operational variables. Such domain knowledge was especially essential for the dynamic simulations. To sum up, to a large extent generalization can be achieved from the casespecific study of pre-combustion CO2 capture plants, primarily regarding the adopted methods and tools. Generated domain knowledge can be quite different. In general, the presented analysis of CO2 capture can serve as an example on how to handle other design cases.

7.2

Perspectives

The research presented in this thesis has led to a better understanding of the system performance of a pre-combustion CO2 capture plant and is a first step to prepare future large-scale implementation of this technology in the power sector. Further research is however required to support the design of commercial-scale plants in particular in terms of techno-economic process analysis, and studies on the interaction between the CO2 capture unit and the main power plant during dynamic operation. To this purpose, the design optimization should be extended to include the economic evaluation of the equipment. For this step it is essential to obtain reliable cost information which might be available in literature or from dedicated costing tools. The scientific interest however might focus more on tackling the global design problem, which can be multi-objective, targeting maximum energy efficiency and minimum investment, or single-objective, aimed at the minimization of the net present value of the plant. After formulation of the generalized design optimization problem, it has to be carefully evaluated if it would be better to adopt a method whereby the costing tool is coupled to the process simulator, while the optimization is performed by means of an external optimizer, or to follow an approach, whereby the cost information is included into the process simulator allowing to make use of internally available optimization algorithms. The overall energy efficiency can also be improved by optimizing the integration of the decarbonised IGCC power plant. To this end, it is necessary to obtain a complete model of the system using preferably one single simulation environment. Regarding the analysis of transient performance, the next step that would allow simulation-based control strategy design should focus on the development of a large-scale system model of the pre-combustion CO2 capture process, whereby the open source library containing the validated process models serves as reliable foundation. The flexibility of the object-oriented modelling approach allows to 172


easily develop new system models by re-use and/or extension of existing models, or by implementation of more sophisticated models when required. The ultimate aim is to form a system model of the decarbonised power plant, which could be obtained by integrating the model of the CO2 capture plant into models of the gasification unit and combined cycle plant. The latter two models have recently been published by Casella and Colonna [2]. Research can then be conducted to study the transient interaction between the CO2 capture unit and the power generation process during different load change scenarios, with the aim to develop and test control strategies in order to improve dynamic performance. Bhattacharyya et al. [3] published one of the first studies on this topic focusing on IGCC power systems. Moreover, physical-based dynamic models can be valuable during the design of new plants in order to evaluate different configurations and to support equipment selection and sizing. It is worth pointing out, that similar research activities are pursued for the investigation of transient performance of post-combustion capture applied to pulverized coal steam power plants [4]. This project, involving academics as well as partners from industry, adopted the same methodology of using object-oriented modelling [5]. Exchange among the researchers might therefore be worth to consider in order to accelerate the developments and avoid duplication, for example, of process models or other analysis tools. Further activity is also required regarding the interface which integrates fluid property packages into the dynamic modelling tool for the computation of thermophysical properties of two-phase, multi-component fluids. To this purpose, the developments might concentrate on facilitating automated index reduction when using external property functions and making a wide range of partial derivatives available in the property package in a flexible and efficient manner. This can contribute to significant improvements regarding numerical robustness, computational speed and ease of initialization of the dynamic simulation. Such an improved interface would ease the modelling of a large class of engineering systems that require accurate, application-specific media models for two-phase, multi-component fluids. A promising approach to tackle the improvement of transient performance of energy conversion systems is by means of simulation-based dynamic performance optimization. Such optimization allows to find the optimal trajectory for a load change operation targeting, for example, the minimum time required for the load variation. This advanced approach to the improvement of transient performance has been, for example, demonstrated by Casella et al. [6] for the optimization of the start-up process of a combined-cycle power plant using the Modelica modelling language. Increasing efforts have been recently devoted to the improvement of algorithms suitable for such optimization problems [7], however the capabilities of current solvers are still limited, thus substantial model reduction is required in case large and complex energy conversion system are of interest.

173

Chapter 7

References [1] Bhattacharyya, D., Turton, R., and Zitney, S., 2011. “Steady-state simulation and optimization of an integrated gasification combined cycle power plant with CO2 capture”. Industrial and Engineering Chemistry Research, 50(3), pp. 1674–1690. [2] Casella, F., and Colonna, P., 2012. “Dynamic modeling of IGCC power plants”. Applied Thermal Engineering, 35, pp. 91–111. [3] Bhattacharyya, D., Turton, R., and Zitney, S., 2012. “Control system design for maintaining CO2 capture in IGCC power plants while load-following”. In Proceedings of the 29th Annual International Pittsburgh Coal Conference, Pittsburgh, PA, October 15-18, Vol. 3, pp. 2160–2173. [4] Dyncap Project, 2014. http://www.kraftwerkforschung.info/en/ mehr-flexibilitaet-fuer-emissionsarme-kohlekraftwerk, April 6. [5] Brunnemann, J., Gottelt, F., Wellner, K., Renz, A., Th¨ uring, A., Roeder, V., Hasenbein, C., Schulze, C., Schmitz, G., , and Eiden, J. “Status of ClaRaCCS: Modelling and Simulation of Coal-Fired Power Plants with CO2 Capture”. In Proceedings of the 9th International Modelica Conference, September 3-5, 2012, Munich, Germany. [6] Casella, F., Donida, F., and ˚ Akesson, J., 2011. “Object-Oriented Modeling and Optimal Control: A Case Study in Power Plant Start-Up”. In 18th IFAC World Congress, Milano, Italy. ˚kesson, J., ˚ [7] A Arzén, K.-E., G¨ afvert, M., Bergdahl, T., and Tummescheit, H., 2010. “Modeling and optimization with Optimica and JModelica.org-Languages and tools for solving large-scale dynamic optimization problems”. Computers and Chemical Engineering, 34(11), pp. 1737–1749.

174

Summary Advances in Model-Based Design of Flexible and Prompt Energy Systems The CO2 Capture Plant at the Buggenum IGCC Power Station as a Test Case

Carsten Trapp ISBN 978-94-6259-222-3 May, 2014

Pre-combustion CO2 capture applied to integrated gasification combined cycle (IGCC) power plants is a promising technical solution to reduce CO2 emissions due to fossil-fuelled electricity generation in order to meet environmental targets in a carbon-constrained future. The pre-combustion capture process allows to effectively remove CO2 from synthetic gas prior its combustion at high partial pressures. In addition, the net energy efficiency of decarbonised IGCC plants is estimated to be higher than that of conventional pulverized coal steam power plants integrating carbon capture. Moreover, gasification allows i) for lower emission levels of regulated pollutants as a result of effective syngas cleaning, ii) for greater fuel flexibility by, for example, co-gasification of biomass, and ii) for the integrated generation of different products, such as electricity, fuels and chemicals. However, the removal of CO2 leads to a high efficiency penalty for the thermal power plant and an increase in system complexity. Moreover, the integration of carbon capture into the very complex gasification process and combined cycle power plant leads to technical problems as far as dynamic operation is concerned. Transient performance of future IGCC power plants becomes extremely relevant in order to balance the rapidly growing share of electricity converted from inherently intermittent renewable sources, such as wind and solar energy. The state-of-the-art approach to design an energy-efficient, flexible and prompt large-scale pre-combustion CO2 capture plant is by means of process modelling and simulation, both steady-state and dynamic. In this context, the validation of process models against measurements is essential in order to obtain reliable and predictive design tools. Currently, for the majority of the documented steady-state and dynamic process models of pre-combustion capture plants, model validation could not be performed due to unavailability of experimental data. Moreover, transient operation and control of CO2 capture systems integrated with IGCC power plants have been treated so far only by few researchers, due to the com-

Summary

plexity of the required modelling and simulation work. The motivation of this research stems therefore from the need of comprehensive experimental investigations of pre-combustion capture technology accompanied by modelling activities to develop detailed and accurate steady-state as well as dynamic models for process analysis and design. The work documented in this thesis was part of a larger research project involving the utility company Vattenfall, the Energy research Centre of the Netherlands (ECN) and the Delft University of Technology aimed at the development of precombustion CO2 capture technology to be applied in a future commercial-scale IGCC power plant. A unique, fully instrumented CO2 capture pilot plant was realized at the Buggenum IGCC power station in the Netherlands in order to demonstrate the technology and investigate its performance. The pilot plant comprises a water-gas shift unit, in which carbon monoxide in the syngas is reacted to hydrogen and carbon dioxide via a three-stage, sweet, high-temperature shift process, and a CO2 removal unit, in which CO2 is removed from the syngas by means of physical absorption utilizing the solvent dimethylether of polyethylene glycol (DEPEG). The most relevant research objectives of the work are to improve and develop general tools and methodologies which i) facilitate detailed steady-state performance analysis and sophisticated optimization of process design and operating conditions, and ii) enable studies on process dynamics already during the early design phase in order to support the choice of equipment and control strategies aiming at the improvement of transient performance. The tools and methods are developed for the case-specific analysis of the precombustion CO2 capture plant at the Buggenum IGCC power station. With respect to generalization, it is worth to highlight that the adopted system engineering techniques and tools are applicable to the design of a larger class of chemical and energy conversion systems with minor changes. The domain knowledge, however, will be rather different in comparison to other systems and might only be transferable to similar CO2 capture plants. The general objective translates into original research questions which aim to identify important design variables of the IGCC pre-combustion CO2 capture system, such as the most relevant process parameters, as well as environmental and operational limits and/or targets, and their impact on the CO2 removal efficiency penalty, and on the optimal operating conditions. Furthermore, this research targets the investigation of the capabilities of a capture system to follow prompt load variations, and to explore control strategies that enhance the responsiveness of the plant. Answering these research questions requires rigorous steady-state modelling of the pre-combustion CO2 capture pilot plant and validation of the model predictions against experimental data, as documented in Chapter 2. It is found that the accuracy of the model is satisfactory in relation to its engineering purpose, namely process analysis and design optimisation. The model therefore serves as a reliable foundation for the development of commercial-scale models of pre-combustion CO2 176

Summary

capture plants, which is treated in Chapter 3. Moreover, Chapter 3 presents the design optimization of a large-scale capture plant following a two-phase approach suited to the use of process simulator environments. In the first phase, global design decisions at plant level are evaluated, targeting the minimization of the energy consumption due to CO2 capture. These are the extent of CO conversion in the water-gas shift unit and the percentage of CO2 capture in the removal unit. The second phase of the design procedure targets the local decision variables at unit level. Two studies are presented focusing on: 1) the design of the solvent regeneration and CO2 compression section, and 2) the impact of the solvent temperature on the energy consumption and equipment cost of the removal unit. It is found that an optimum exists for the overall carbon capture rate, at which the specific energy consumption of the capture plant is minimal. The value of the optimal capture rate depends on various modelling choices and assumption, such as plant configuration and values of process parameters. Moreover, the design optimization revealed that the specific energy consumption2 of the pre-combustion capture plant can be reduced by 10 % when operating at a minimum molar steam/CO ratio of 1.5, which has been successfully tested at the pilot plant, in comparison to the current vendor suggestion of 2.65. These energy savings come at the cost of a lower optimal carbon capture rate, namely 78 % instead of 87.5 %. Another viable solution for energy efficiency improvement of the pre-combustion CO2 capture plant is the recovery of low-grade thermal energy by means of an ORC power system. Waste-heat can be recovered both downstream of the water-gas shift unit (syngas stream) and throughout CO2 compression train (CO2 product stream) allowing to save approximately 10 % of the energy consumption of the capture plant. Differently from other conventional ORC power system applications, the thermal energy source in this case is a syngas-water mixture, which partly condenses due to the heat transfer to the ORC primary heat exchanger preventing the typical pinch-point problem from occurring. The second part of this thesis treats the dynamic modelling and simulation of the pre-combustion CO2 capture process for control design purposes. Chapter 5 discusses the development and implementation of dynamic models of the capture plant into an open source software library by means of the object-oriented, equation-based Modelica language. Comprehensive dynamic model validation is demonstrated at component, sub-system and system level in order to improve the reliability and accuracy of the model predictions. Satisfactory agreement was achieved between the dynamic simulation results and experimental measurements obtained from various open- and closed-loop step response tests performed at the pilot plant. A more detailed validation is presented for one of the key process components, the absorber column, as presented in Chapter 6. It was concluded that a well-tuned, equilibrium-based dynamic model for physical absorption of CO2 provides sufficiently accurate performance predictions concerning transient operation. 2 The

specific energy consumption is defined as the total energy consumption of the capture plant divided by the amount of carbon captured for sequestration.

177

Summary

The validated system model of the capture pilot plant was used to perform dynamic simulations in order to investigate a control strategy based on feed-forward, feed-back and cascade control with the aim to improve transient performance. It was preliminarily demonstrated that prompt syngas load variations are a feasible operating mode with the control system based on feed-forward. The qualitative results from this analysis can be applied to the design of control strategies of a largescale CO2 capture plant, whereby the validated component and pilot plant system models serve as reliable basis for the development of models of commercial-scale plants. Chapter 7 gives concluding remarks and recommendations for possible future work.

178

Samenvatting Advances in Model-Based Design of Flexible and Prompt Energy Systems The CO2 Capture Plant at the Buggenum IGCC Power Station as a Test Case

Carsten Trapp ISBN 978-94-6259-222-3 May, 2014

Pre-combustion CO2 afvang toegepast op energiecentrales welke een vergassingsproces met een stoom en gasturbine (KV-STEG) combineren is een veelbelovende technische oplossing om de CO2 uitstoot veroorzaakt door opwekking van elektriciteit met fossiele brandstoffen te verminderen. Aldus kunnen aan de milieudoelstellingen van een koolstofarme toekomst worden voldaan. Het proces van pre-combustion maakt het mogelijk om CO2 uit synthetisch gas voor de verbranding bij hoge partiële drukken effectief te verwijderen en de netto energie-efficiëntie van KV-STEG energiecentrales wordt hoger ingeschat dan die van conventionele poederkool gestookte stoomcentrales met ge¨ıntegreerde CO2 afvang. Bovendien, vergassing zorgt voor een lagere uitstoot van de meeste overige emissies als gevolg van effectieve reiniging van het synthetisch gas. Bovendien biedt vergassing de mogelijkheid om verschillende brandstoffen te gebruiken (zoals biomassa) en/of verschillende eindproducten te maken naast elektriciteit, waaronder brandstoffen en chemicali¨ n. De verwijdering van CO2 leidt tot een hoger verlies van het rendement van de thermische centrale en een toename in systeemcomplexiteit. Bovendien leidt de integratie van koolstofafvang in het zeer complexe vergassingsproces en de STEG-centrale tot technische problemen met betrekking tot de dynamische werking, zoals snelle en flexibele bedrijfswisselingen. De dynamische prestatie van toekomstige vergassing centrales wordt uiterst relevant vanwege het snel groeiende aandeel van elektriciteit dat wordt omgezet uit hernieuwbare bronnen, zoals winden zonne-energie, in evenwicht te brengen. De state-of-the-art methodologie voor ontwerp van een energie-efficiënte en flexibele pre-combustion CO2 afvang installatie op industriële schaal is procesmodellering en -simulatie. In deze context is de validatie van procesmodellen met metingen onmisbaar om betrouwbare en voorspellende ontwerphulpmiddelen te verkrijgen. Er is echter een gebrek aan experimentele gegevens, waardoor het merendeel van de gedocumenteerde stationaire en dynamische procesmodellen

Samenvatting

voor CO2 afvang installaties niet zijn gevalideerd. Bovendien zijn de dynamische prestaties en procesregeling van de CO2 afvang systemen ge¨ıntegreerd met KVSTEG centrales tot nu toe slechts door enkele onderzoekers bestudeerd vanwege de complexiteit van het vereiste modelleerwerk en de simulaties. De motivatie van dit onderzoek komt voort uit de behoefte van uitgebreid experimenteel onderzoek van de pre-combustion CO2 afvang technologie, tezamen met de modelleeractiviteiten om gedetailleerde en nauwkeurige stationaire en dynamische modellen te ontwikkelen voor procesanalyse en -ontwerp. Het werk gedocumenteerd in dit proefschrift is deel van een groter onderzoeksproject waarin het energiebedrijf Vattenfall, het Energy research Centre of the Netherlands (ECN) en de Technische Universiteit Delft betrokken zijn. Het doel van het project is het testen en optimaliseren van pre-combustion CO2 afvang technologie voor toekomstige commerciële KV-STEG elektriciteitscentrales. Een uniek, volledig ge¨ınstrumenteerde CO2 afvang proefinstallatie is gerealiseerd bij de KV-STEG energiecentrale in Buggenum in Nederland om de technologie te demonstreren en zijn prestatie te onderzoeken. De proefinstallatie omvat een water-gas shift sectie, waarin koolmonoxide in het syngas naar waterstof en kooldioxide wordt gereageerd via een drie-fase, zoet, hoge-temperatuur shift proces, en een CO2 absorptie sectie, waarin CO2 wordt verwijderd uit het syngas door middel van fysische absorptie, gebruikmakend van het oplosmiddel dimethylether van polyethyleenglycol. De meest relevante onderzoeksdoelstellingen van dit werk zijn de verbetering en ontwikkeling van algemene hulpmiddelen en methodologieën om i) een gedetailleerde stationair prestatie-analyse en optimalisatie van procesontwerp en bedrijfsomstandigheden mogelijk te maken, en om ii) al in de vroege ontwerpfase studies over dynamisch gedrag mogelijk te maken teneinde de dynamisch prestaties te verbeteren door de keuze van apparatuur en procesregeling strategieën. De hulpmiddelen en methodologieën zijn ontwikkeld voor de specifieke analyse van de pre-combustion CO2 afvang installatie bij de energiecentrale in Buggenum. Generaliserend, het is het waard om te benadrukken dat de system engineering technieken en hulpmiddelen met kleine wijzigingen van toepassing zijn voor het ontwerp van een grotere klasse van chemische en energie conversie systemen. De domeinkennis zal nogal verschillend zijn in vergelijking met andere systemen en kan alleen worden overgedragen aan vergelijkbare CO2 afvang installaties. De algemene doelstelling vertaalt zich in onderzoeksvragen die gericht zijn op de identificatie van belangrijke ontwerpvariabelen zoals de meest relevante procesparameters, de milieu- en operationele limieten en/of doelstellingen, hun invloed op het verlies aan rendement door CO2 verwijdering, en over de optimale operationele omstandigheden. Bovendien heeft deze studie tot doel te onderzoeken of een CO2 afvang systeem kan leiden tot een systeem dat snelle bedrijfswisselingen goed kan volgen en of regelstrategieën de respons tijd zouden kunnen verbeteren. Het beantwoorden van deze onderzoeksvragen vereist de stationaire modellering van de CO2 afvang proefinstallatie en validatie van voorspellingen van het model met experimentele gegevens, zoals gedocumenteerd in hoofdstuk 2. Het is 180

Samenvatting

gebleken dat de nauwkeurigheid van het model bevredigend is met betrekking tot zijn technische doel, namelijk procesanalyse en ontwerpoptimalisatie. Het model dient dus als een betrouwbare basis voor de ontwikkeling van modellen van grootschalige pre-combustion CO2 afvang installaties. Dit wordt behandeld in hoofdstuk 3. Hoofdstuk 3 presenteert bovendien de optimalisering van het ontwerp van een grootschalige afvang installatie met behulp van een twee-fasen aanpak die is toegesneden op het gebruik van proces simulatoren. In de eerste fase worden globale ontwerpbeslissingen op systeemniveau geëvalueerd, gericht op het minimaliseren van het energieverbruik als gevolg van CO2 afvang. Dit zijn de mate van CO omzetting in de water-gas shift sectie en het percentage van CO2 afvang in de absorptie sectie. De tweede fase van de ontwerpprocedure richt zich op de lokale ontwerpvariabelen. Twee studies worden gepresenteerd en zijn gericht op: 1) het ontwerp van de regeneratie en CO2 -compressie sectie en 2) het effect van de temperatuur van het oplosmiddel op het energieverbruik en de kosten van de apparatuur in de CO2 verwijderingssectie. Het is gebleken dat er een optimum bestaat voor de totale mate van koolstof afvang, waarbij het specifieke energieverbruik van de afvanginstallatie minimaal is. Uit de optimalisatie van het ontwerp is gebleken dat het specifieke3 energieverbruik van de CO2 afvang installatie kan worden verminderd met 10 % bij bedrijf met een minimale molaire stoom/CO verhouding van 1.5, die met succes is getest bij de proefinstallatie, in vergelijking met de suggestie van de huidige leverancier van 2.65. Deze energiebesparing komt ten koste van een lagere optimale mate van afvang, namelijk 78 % in plaats van 87.5 %. Een andere oplossing voor verbetering van de energierendement van de precombustion CO2 afvang installatie is de terugwinning van laagwaardige warmte door middel van een organic Rankine cycle (ORC) energie systeem. Laagwaardige warmte kan zowel stroomafwaarts van de water-gas shift sectie (syngasstroom) als in de CO2 compressie (CO2 productstroom) teruggevoerd worden, waardoor ongeveer 10 % van het energieverbruik van de CO2 afvang bespaard wordt. Anders dan bij toepassing van conventionele ORC systemen is in dit geval de thermische energiebron een syngas-watermengsel, dat gedeeltelijk condenseert door de warmteoverdracht naar de primaire warmtewisselaar van de ORC systeem, waardoor het typische pinch-point probleem wordt vermeden. Het tweede deel van dit proefschrift behandelt de dynamische modellering en simulatie van het pre-combustion CO2 afvangproces voor het gebruik van het ontwerp van de regeling. Hoofdstuk 5 gaat in op de ontwikkeling en implementatie van dynamische modellen van de CO2 afvanginstallatie in een open source softwarebibliotheek door middel van de object-georiënteerde Modelica taal. Een uitgebreide validatie van het dynamisch model op component-, subsysteem- en systeemniveau wordt gedemonstreerd om de betrouwbaarheid en nauwkeurigheid van de modelvoorspellingen te verbeteren. De bereikte overeenstemming tussen de resultaten van de dynamische simulaties en experimentele metingen uit verschillende open- en closed-loop testen op de proefin3 Het

specifieke energieverbruik wordt gedefinieerd als het totale energieverbruik van de CO2 afvanginstallatie gedeeld door de hoeveelheid koolstof vastgelegd voor opslag.

181

Samenvatting

stallatie is bevredigend. Een gedetailleerde validatie wordt gepresenteerd voor een van de belangrijkste procesonderdelen, de absorptiekolom, zoals beschreven in Hoofdstuk 6. Geconcludeerd werd dat het dynamische model voor fysieke absorptie van CO2 , gebaseerd op thermodynamisch evenwicht, voldoende nauwkeurige prestatievoorspellingen biedt met betrekking tot het dynamisch gedrag. Het gevalideerde model van de CO2 afvang proefinstallatie werd gebruikt om dynamische simulaties uit te voeren, met het doel een regelstrategie gebaseerd op feed-forward, feedback en cascaderegeling te onderzoeken. Het doel van de regelstrategie is de verbetering van de dynamische prestaties. Er is aangetoond dat met het voorgestelde regelsysteem op basis van feed-forward control het snel variëren van de hoeveelheid syngas kan worden verbeterd . De kwalitatieve resultaten van deze analyse kunnen worden toegepast op het ontwerp van regelstrategieën van een CO2 afvang installatie op commerciële schaal, waarbij de gevalideerde systeem modellen van de componenten en de proefinstallatie als betrouwbare basis voor de ontwikkeling van modellen van installaties op commerciële schaal dienen. Hoofdstuk 7 geeft een slotbeschouwing en aanbevelingen voor mogelijke toekomstige werkzaamheden.

182

Acknowledgements This thesis is the result of a 4-year endeavour based on challenging and intense research work and would have not been completed without immense contributions of many kind people. I am deeply indebted to all of them for their intellectual and/or emotional support. Herewith I will make an attempt to express my deepest gratitude knowing, however, that my words are destined to fall short. I am sincerely grateful to my promoter, Prof. Piero Colonna. Piero, I thank you for your continuous guidance, your encouragement and your admirable kindness throughout these years. These helped me together with your scientific knowledge and vast research experience to grow as a scientist but also as a person. I highly appreciate the academic freedom you gave me to explore various research topics in depth. At the same time, I value your keen perception of the broader research picture with which you guided me back on track during times in which I got lost in unnecessary details or deep layers of computer code. I am very thankful for all the time and effort you invested in continuously improving my presentations and persistently correcting the papers and the chapters of this thesis. You are a true inspiration when it comes to presenting and communicating research and I was very fortunate to learn from you. I greatly appreciate the “open door” of your office and to meet with you whenever I deemed it necessary. During these occasions I enjoyed discussing a wide range of topics concerning society and people, politics, economics and life in general as well as relevant scientific matter. Moreover, I admire your contagious enthusiasm and your positive attitude with which you created both a stimulating research environment and a pleasant working atmosphere. I also greatly appreciate that, together with your wife Mirella, you were always concerned about my well-being. I am greatly indebted to Dr. Kay Damen from Vattenfall. Kay, I very much appreciate your effective management of the CO2 Catch-up project and the wellstructured monthly progress meetings, which gave me much comfort in the planning of my PhD research. The fruitful discussions during the regular meetings together with your admirable curiosity and perseverance continuously stimulated my eagerness to tackle the complex technical problems and further helped me to learn how to present new insight in a clear and understandable manner. Without your valuable comments and suggestions to improve my reports and thesis, many important aspects would be missing. Moreover, I greatly appreciate the openminded working atmosphere and the team events/dinners you organized resulting in joyful personal discussions. My thanks are extended to the other R&D staff at

Acknowledgements

Vattenfall, Radoslaw Gnutek, Dr. Richard Faber, Dr. Sebastian Meinke and Han Raas for their just as helpful support and insight into the power generation industry. I have greatly benefited from the invaluable practical advice of the Vattenfall plant operators Perry Brummans, Marc Cuypers and Frans Kornips. With creativity and dedication you performed an enormous amount of experimental tests, even those which seemed impossible. The measurement results form the basis of this thesis. I am deeply grateful to Dr. Francesco Casella from Politecnico di Milano. Francesco, I thank you for introducing me to the wonderful world of Modelica and dynamic modelling. Your scientific insight and your immense modelling experience together with the opportunity to work with you for few weeks in Milan have significantly accelerated my learning process. I highly appreciate your hands-on help with programming and our long discussions on modelling problems, which typically resulted in exciting ideas for further research, of which various were successfully materialized in publications. I have received generous support from many people of the CO2 Catch-up project team, my sincere gratitude goes to all of them. Your skilful help was invaluable and contributed towards a broader scope of this thesis. Prof. André Bardow, I thank you for your contributions to the performance analysis of the CO2 absorber and your advice concerning scientific writing. Prof. Joachim Groß, I am grateful for your insight and support with the modelling of thermophysical properties. Dr. Michiel Makee and Dr. Eric van Dijk, I value your contributions related to the water-gas shift process modelling. Eric, thanks for your clear explanations and your help with the experimental measurements. Dr. Lukas Valenz, I appreciate your practical insight of the CO2 absorption process. Dr. Carlo de Servi, I am grateful for your enormous efforts on the dynamic modelling of absorption column. Carlo, I enjoyed the close collaboration and our scientific discussions which enabled me to sharpen my ideas. Teus van der Stelt, I owe you many thanks for improving speed and robustness of the thermophysical property computations. Without your contributions I would still be waiting for the dynamic simulations to finish. My sincere gratitude is also extended to the fellow PhD researchers for their support and advice, Dariana Hernandez, Marina Stavrou, Abrar Hakeem and Bernardo Oyarzun. I have largely benefited from being surrounded by an inspiring group of researchers which mad the collaboration within the project a pleasure. This thesis would be rather a thin booklet without the contribution of several ambitious graduate students. Keith Gonesh, I am grateful for your extremely hard work in developing the first dynamic models of the pilot plant. Pasquale Condemi, I thank you for extending the models and performing validation. I enjoyed our regular lunches and dinners in Milan where you introduced me to the delicious cuisine of your native Southern Italy. Roberto Raspopov, I owe you many thanks for the implementation of the partial derivatives in FluidProp. Timon Thomaser, I am grateful that you successfully mastered the extremely complex optimization of the capture process. Adam van de Haar, I thank for introducing me to postcombustion capture and conveying the results of your work into a scientific article. 184

Acknowledgements

I would like to express my appreciation to the organizational board of the CATO-2 programme namely Jan Brouwer, Jan Hopman, Sander van Egmond, Marlies Verlinde and Mirjam van Deutekom. I greatly enjoyed participating at the CATO-2 events and PhD excursions which provided an excellent opportunity to exchange ideas with fellow researchers. With your enthusiasm and creativity you made these events outstanding. My thanks are also extended to all CATO-2 PhD students for sharing knowledge and ideas. I would particularly like to thank all my colleagues at Process & Energy and later at Propulsion & Power for filling the coffee breaks with laughter, the lunch times with intriguing and broad discussions, and the hours after work with cheerful drinks. It was truly a stimulating atmosphere to be surrounded by a large number of colleagues from different countries and with diverse backgrounds. Many of them also became close friends. Special thanks go to my offices mates: Martina, Eleonora, Marcin, Jacopo, Ryan, John, Emiliano, Enrico, Mauro, Sebastian, Adam and Salvo. My thanks are extend to: Mattia, Gianluca, Fan Fan, Sowande, Michel, Aditya, Mayte, Albert, Guido, Marloes, Rene and Theo. I am grateful for all the joyful moments together with Alondra, Lucie, Sara, Stephanie, Bernardo, Lawien, Sergio and Tiemo during gatherings, trips, dinners and parties. Alondra, I value our lengthy and heated discussions about society, politics, philosophy, etc., and it was enriching that we almost never agreed. Lucie, thank you for throwing the best parties and the cheerful game evenings. Sara, I enjoyed our coffee breaks whenever you were in Delft and the delicious dinners at your place. Stephanie, thanks for the illuminating discussions and your delivery service for champagne. Bernardo, thanks for organizing the wonderful bike trips and for making sure that my German is not getting rusty. Lawien, I thank you for the hilarious email exchanges after you left Delft and for sharing the same sense of humour. Sergio, thanks for the crazy times with music, food and friends at our shared place. Tiemo, I thank your for sharing all your peculiar stories and for always being ready to grab an after-work beer. Many thanks go also to fellow PhD friends, Alessandro, Michaela, Robert, Jamie, Joolie, Wynand, Norbert, Luaine, Costas, Azadeh, Steef and Hugo, for the fun outside university at gatherings, barbecues and birthday parties. I would also like to thank friends who bring joy to my life since I moved to Leiden, Neta, Alex, Giuliana, Leticia and Angie. Special thanks also to my aerospace engineering friends from student days, Benjamin, Daniel, Florian, Marius and Moritz. I am grateful for our regular meetings to enjoy relaxed days in nature or to celebrate important moments in each other’s lives, even if the distances are far. These thanks are also extended to Andreas and Felix from my student association. My heartfelt thanks and appreciation goes to my parents and my sisters. Mutti and Papi, Michaela and Janina, I am deeply indebted for your encouragement, your loving care and your unconditional support of all my endeavours, even if they take me far away from home. You give me balance and security in life.

185

Acknowledgements

I owe my deepest gratitude to my love Yshya. Thank you, Yshya, for enriching my life with your passion and humour, your strong belief in me, your inspiring enthusiasm, and also for your support through the last year of long working hours.

Carsten Trapp Leiden, May, 2014

186

Selected publications Journal publications • Trapp, C., and Colonna, P., 2013. “Efficiency improvement in pre-combustion CO2 removal units with a waste-heat recovery ORC power plant”. Journal of Engineering for Gas Turbines and Power, 135. • Trapp, C., Casella, F., and Colonna, P., 2014. “Dynamic modelling and validation of a pre-combustion CO2 capture plant for control design”. Industrial and Engineering Chemistry Research, submitted. • Trapp, C., de Servi, C., Casella, F., Bardow, A., and Colonna, P., 2014. “Dynamic modelling and validation of pre-combustion CO2 absorption based on a pilot plant at the Buggenum IGCC power station”. International Journal of Greenhouse Gas Control, submitted. • Trapp, C., Thomaser, T., van Dijk, H. A. J., and Colonna, P., 2014. “Design optimization for flexible operation of a pre-combustion CO2 capture plant embedding experimental knowledge”. Fuel, submitted. Conference proceedings / oral presentations • Trapp, C., and Colonna, P., 2011. “Efficiency improvement in pre-combustion CO2 removal units with a waste-heat recovery ORC power plant”. First International Seminar on ORC Power Systems, September 22-23, Delft, The Netherlands. • Trapp, C., Thomaser, T., and Colonna, P., 2013. “Study of efficiency improvements of pre-combustion CO2 removal units by means of validated steady-state simulations”. Proceedings of the 2013 AIChE Annual Meeting, November 3-8, San Francisco, USA. • Trapp, C., de Servi, C., Casella, F., and Colonna, P., 2013. “Dynamic simulation and model validation of a pre-combustion CO2 capture unit for IGCC power plants”. Proceedings of the 2013 AIChE Annual Meeting, November 3-8, San Francisco, USA. CAST Directors’ Award for best poster presentation. • Trapp, C., Casella, F., van der Stelt, T. P., and Colonna, P., 2014. “Use of External Fluid Property Code in Modelica for Modelling of a Pre-combustion CO2 Capture Process Involving Multi-Component, Two-Phase Fluids”. Proceedings of the 10th Modelica Conference, March 10-12, Lund, Sweden.

About the author Carsten Trapp was born in Radebeul, Germany on July 24th 1982. In 2001 he completed his secondary education at the Gymnasium Luisenstift in Radebeul with distinction, majoring in Mathematics and Physics. In 2003 Carsten started the studies of aerospace engineering at the University of Stuttgart, Germany. In 2007 he received an Erasmus Scholarship and spend one semester at the Royal Institute of Technology in Stockholm, Sweden. As part of the university education Carsten undertook a 6-month internship working for Rolls-Royce in Oberursel, Germany. In 2009 he received his university degree of Diplom-Ingenieur after completing his 6-month graduation assignment on gas turbine compressor modelling at ALSTOM in Baden, Switzerland. In 2009 he was appointed as PhD candidate at the Energy Technology group at the Process & Energy department, Faculty of Mechanical, Maritime and Materials Engineering (3mE), Delft University of Technology under the supervision of Prof. Piero Colonna. His research work was part of the CO2 Catch-up R&D programme performed in cooperation with Vattenfall and the Energy research Centre of the Netherlands (ECN). The results of Carsten’s PhD research are published in peer-reviewed scientific journals and presented at various international conferences in the field of carbon capture and storage (CCS) and process systems engineering. In 2013 Carsten was awarded with the Computing and Systems Technology (CAST) Directors’ Award for the best poster presentation at the 2013 American Institute of Chemical Engineers (AIChE) Annual Meeting. During his time as a PhD researcher he had the opportunity to supervise several students for their Master thesis projects and to be involved in teaching activities.