Measurement and Communication Techniques for ...

Measurement and Communication Techniques for Remote Condition Monitoring of High Voltage Transformer Bushings D. Cornforth and G. Ledwich Department of Electrical and Computer Engineering University of Newcastle Newcastle New South Wales

Abstract The insulation used in high voltage transformer bushings experiences an aging process, which reduces its effectiveness and ultimately may lead to catastrophic breakdown, which is dangerous and expensive. The progress of aging can be estimated by measuring various parameters of the insulator material. The most ubiquitous and most easily obtainable parameters are electrical, and this fact has led to the introduction of commercial automated measurement systems suitable for condition monitoring of various types of high voltage bushings. This paper describes the development phase of a system using fibre optic communications to construct a network based on standard PCs serving as the instrumentation at the point of measurement. Such an approach allows all measurements to be digitised and transmitted over a rugged interference-resistant medium. The use of standard PCs running Windows NT or Windows 95 provides networking services, including error detection and correction. The flexibility of this platform is conducive to multitasking software, allowing multiple measurement functions, data logging and data processing at the remote node. The system is designed to measure tan δ, partial discharge and harmonics. It is expected that the system can measure relative phase angle to within 0.1 mRadians. All nodes across the network are synchronised using a dedicated fibre carrying timing pulses. The software design uses a modular approach to maintain flexibility. A supervisory system comprises a Graphical User Interface communicating with network software using Windows Dynamic Data Exchange (DDE). The remote nodes communicate with this system using Remote Procedure Calls (RPC), which is a multi-platform protocol, enabling other operating systems to be integrated if desired. A flexible storage design at the remote site allows data to be collected in the absence of any communication

from the supervisory system. The data can be saved to disk, or selectively expunged according to programmed rules, then forwarded on request.

1. Introduction Efficient generation and distribution of power requires transformers at the power station and in the switchyard. The high voltages required for transmission of power lead to dependence on bushings, for the support of conductors away from ground potentials. The insulating material used in the bushings has a finite life due to chemical and physical changes mediated by localised breakdowns of insulation under the electrical stresses commonly experienced 1, 9. Various parameters may be used to indicate the progress of aging, but electrical parameters measured at the bushing itself are non-intrusive, have been proven over years to be effective, and can be applied to many different types of insulation material [2]. The following discussion is limited to insulators used in bushings as part of a system for the supply and distribution of three phase alternating current at 50Hz frequency. It is desirable to make these measurements without recourse to interruption of power, so all measurements should be made at normal operating voltage. The time scale of aging, and the geographical separation of equipment, lead to the development of continuous remote condition monitoring systems. The development of such a system must address the particular measurement and communication issues of the environment.

2. Electrical test parameters The common parameters associated with insulator aging are ascertained using Dielectric

Measurement and Communication Techniques for Remote Condition Monitoring of High Voltage Transformer Bushings 1

Leakage Analysis (DLA), and Partial Discharge (PD) analysis [3,4,7]. Both analysis methods can be performed on measurements taken from a diagnostic point provided on high voltage equipment, known as the DDF point. Both methods rely on measurements made in the time domain and translated to the phase domain.

typically rely on a bandwidth limited front end to perform the necessary integration, followed by a current to voltage conversion [6]. The magnitude of the resulting voltage spike is used to provide a pulse height and pulse frequency distribution, related to phase [5,8].

3. Specifications of a measurement system DLA usually focuses on the measurement of tan δ, where δ is the difference in phase angle between voltage and current. Indications of insulation condition are obtained by comparison of this quantity with tan δ from a similar device of known condition, and by observing changes in tan δ over time. DLA requires measurement of the instantaneous 50Hz waveform of the leakage current through the insulator. This waveform is filtered to remove noise, and convolved with a reference 50Hz waveform, to produce the magnitude and phase angle of the 50Hz component of the leakage current. The reference waveform is ideally the voltage across the insulator, but practical considerations usually exclude this. Owing to the nature of electrical equipment and the typical layout, it is usually necessary to measure the leakage current and voltage at points geographically distant from one another. Also, voltage changes little with respect to distance, so it is convenient to measure voltage at one point only in the system. This can be measured at a voltage transformer (VT). Data from the leakage current measurement point is combined with the voltage measurement at a central point within the network, and this enables the phase angle difference to be calculated. The successful combination of data measured from different points in the network requires accurate time and phase synchronisation between all nodes, requiring a system-wide timing reference signal. PD measurement can also be made from the instantaneous leakage current through the insulator. Once the 50Hz component is removed by filtering, PDs are manifest as spikes lasting a fraction of a cycle of 50Hz. Since a discharge event represents the movement of a charge q within the insulator, the magnitude of the PD is proportional to the integral of the current measured. Existing systems for PD analysis

The measurement system employed must take account of various factors, and these are examined in this section. 3.1 Time scale of measurements The observable effects of aging take place over months. This makes it desirable to provide continuous on-line monitoring of the condition of insulators. 3.2 Geographical distribution As these insulators are physically situated at some distance from power station facilities, and at some distance form each other, a data network is required to enable real-time collection of data, and presentation of the data to a central point. The specific conditions of this environment dictate to some extent the characteristics of such a network. High voltage insulators are normally situated in a switchyard at distances of hundreds of metres apart, and may be several kilometres from the power station. 3.3 Noise The signals being measured are often small, and must be conveyed over large distances in an environment of large electric and magnetic fields. While the 50Hz current produces strong interference, the main noise source is the high frequency discharges in air, or corona discharges. Careful shielding, and the use of optic fibres for all long runs, may reduce this. All data is digitised before transmission. 3.4 Integration There are a number of non-electrical parameters associated with insulator aging. The use of a flexible arrangement of active and passive nodes

Measurement and Communication Techniques for Remote Condition Monitoring of High Voltage Transformer Bushings. 2

ensures that the data from any other transducers may be processed locally in an active node before being sent to the master station.

making decisions on what data to keep if there are storage restrictions. Active nodes also allow the possibility of connection to other transducers and systems

3.5 Accuracy The system described is expected to achieve an accuracy in the order of 0.1m Radians. As this corresponds to a time of 318nS, a mechanism other than the internal clock of a PC must be relied upon. This accuracy may be achieved by the use of a fibre network dedicated to the provision of timing signals. Corrections may be applied for the latency times of the fibres.

4. Network Implementation In order to make measurements comparing phase angle at geographically separate, it is necessary to have a high degree of synchronisation between nodes on the network. The time of transmission of a signal across the network cannot be predicted with sufficient accuracy to allow for clock synchronisation by network. The proposed network uses optic fibres to provide timing signals for the whole network. The proposed network is composed of two systems: a communication network supporting Ethernet running at 10Mbs, and a timing network supporting a proprietary pulse stream. Both networks use fibres between nodes to enhance noise immunity.

The communication network is bi-directional, allowing commands from the master station to reach all nodes, and data to be returned. The use of Ethernet allows commercially available and relatively cheap components to be used. The timing network is unidirectional, as the timing signal is generated at one point and propagated to all nodes on the network. The active nodes run Windows 95 or NT operating system, which offers extensive support for Ethernet, as well as providing a popular platform for the integration of third party software. The passive nodes are proprietary, possess measurement and storage capability, and are capable of responding to simple commands from an active node. Both active and passive nodes have three channels for three phase monitoring, and either type of node can be connected to the equipment being monitored. Measurements can be made simultaneously on any or all nodes across the network, and these measurements can be time stamped then compared cycle by cycle at the master station. An example of such a network is shown in Fig. 1. Fig. 1. Network organisation

Several existing commercial systems of this type depend on passive nodes, that is, nodes where a minimum of data processing is performed, such as A to D conversion. Each node is connected to a central node using star topology, and all data processing is done at a central computer. By contrast active nodes perform data processing and accept commands from the central node, so are able to send data selectively to the central node. While passive nodes are typically cheaper than active nodes, the use of active nodes can minimise network traffic, give more flexibility in the use of network topology and reduce the amount of cabling required. Active nodes can also perform as semi-autonomous measurement units, collecting data in the event of temporary disconnection from the rest of the network, and

Node Node Node

Remote station

Timing network Data network

Timing and data network Remote station

Node Node Remote station

The timing signals are generated at the master station, and are synchronised with the local mains using a phase locked loop. The timing signals include phase position information. This allows time domain measurements to be easily


Master station

converted to the phase domain at the remote station. The use of active and passive nodes to form a network provides a flexible and reliable system, as well as the capability of balancing cost per node against functionality at the point of measurement.

5. Instrumentation Interface The use of an instrumentation quality current transformer (CT) is desirable for connection to the DDF point of the equipment under test, as electrical isolation is required for safety reasons. The measurement of leakage current with accurate phase angle requires a linear response at 50 Hz, while the measurement of PD signals requires good high frequency response. In order to achieve this, the following points were considered: • A toroid core has low leakage and good coupling. • A magnetic material with high permeability will require low magnetising current, and will exhibit good frequency response, and low losses. • A linear response in terms of current is required. This is indicated by the linearity of the B-H curve for the material. • Changes in temperature may limit the permeability of the material. • Changes in phase shift with frequency can be minimised by using windings having inductance much greater than resistance. • Good electrical isolation is required between primary and secondary The instrumentation CT was designed using a toroid core. The magnetic material was selected on the basis of manufacturer specifications. The core losses can be visualised as a winding series resistance. The permeability can then be expressed as a complex quantity, where the product of real and imaginary parts should be high for low losses. A magnetic material was chosen having a complex permeability product greater than 1000 for frequencies up to 1 MHz. The inductance factor, AL of the material was 10700 nH ± 30%, where inductance, L = N2.AL. The effect of temperature increase on the core

material is to increase permeability, so this will not introduce a frequency limit on the transformer response. Two windings of 100 turns each were made, which is the practical maximum number of turns for the size of the core. This allows a gap between primary and secondary of 10mm, to achieve isolation of up to 5kV in the event of a high voltage appearing on the primary. There is another effect of temperature on the resistance of the winding, which causes a shift in phase angle. This can be minimised by using relatively thick wire, and using sufficient number of turns, as the inductance increases with the square of turns, while resistance increases linearly. To test the frequency response of the designed CTs, the primary was connected to a sine wave generator with sweep capability, and the secondary was shorted. Current probes were connected to both primary and secondary, and a graph was obtained of the ratio of secondary to primary current, against frequency of the source. The resulting graph is shown in Fig. 2. The graph shows the frequency response flat to 500kHz, dropping by 5dB at 1MHz, then rising by 20 dB to 10 MHz. Fig. 2. Frequency response of instrumentation current transformer.

A test for phase shift was conducted using the same equipment, but sweeping the frequency manually, so that individual cycles could be observed on an oscilloscope. A phase shift of about 10 degrees was observed, but this appeared to be constant over the entire frequency range. The CT was heated using a heat gun to 80


degrees C. No temperature related phase shift was detected. These results are impressive for an instrumentation CT, and show that the use of a CT of this type represents a reasonable compromise between performance and the need for circuit isolation. A current to voltage conversion is required to allow the use of an A to D converter. The conversion must not introduce frequency dependant phase shift, and the secondary of the CT should be effectively shorted. A suitable arrangement utilises an operation amplifier (opamp) as shown in Fig. 3. Fig. 3. Active current to voltage converter Leakage current from equipment under test

CT

R

_

output

+ Op-amp earth

Assuming the turns ratio of the CT is 1:1, and as the inverting input of the op-amp is a virtual earth, the current flowing in R must be of the same magnitude as the current in the primary of the CT. This is the leakage current of the equipment under test. The phase difference between leakage current and output voltage will be close to 180o, and this will be constant over the range of frequency, assuming a perfect opamp, and that it is supplied with sufficient operating voltage. In practice the following limitations restrict choice of the op-amp: • Capability of supply of sufficient output current • Provision of a gain of unity over the required bandwidth and maximum output current. • Sufficient voltage swing under the same conditions.

Depending on the magnitude of current likely to be encountered, a suitable op-amp can be chosen. If the current being measured is likely to be beyond the capability of the op-amp, the turns ratio of the transformer may be changed.

6. Software design The design of the software is reviewed in this section. 6.1 General considerations The network infrastructure can be modeled by the industry standard seven layer OSI model: • • • • • • •

Application layer Presentation layer Session layer Transport layer Network layer Data link layer Physical layer

The use of a commercially available computer and operating system provides most of these layers. PCs running Windows 95 or Windows NT were chosen because of ease of use, availability, and provision of network support. Proprietary application software was designed to supply the application and presentation layers. The choice of Remote Procedure Calls (RPCs) was made for the session layer to provide flexibility, ease of use, and the possibility of integrating other operating systems. The operating system and hardware supply the other layers. This includes the use of TCP/IP for the transport layer for the same reasons as the use of RPC. The use of RPC, and the hardware interface, requires the use of multithreaded application software. The design goal of modular software and the very different functionality between the master station network software and the GUI, made it desirable to implement these as separate applications. This choice allowed these applications to be developed separately and in different development environments. This separation required a method of communication


between the two applications. As both will run on the same computer, Dynamic Data Exchange (DDE) was chosen, as this provides a simple and industry standard method.

• •

6.2 Remote station

The master station GUI must be in contact with all remote nodes, requesting data according to schedules, and presenting that data on demand, in text or graphical format. The software must perform the following tasks:

The remote station software is required for active nodes only. Passive nodes have been designed to operate without software. In order to provide a sufficient range of functions, the passive nodes make use of a Field Programmable Gate Array (FPGA). The flexibility of these devices also makes them suitable for performing many hardware tasks in the active node and in the master station. The active node must perform the following tasks: • Windows interface and diagnostics • FPGA programming and interface • Hardware management • Hardware synchronisation • Communication and data processing for all connected passive nodes • Automatic and manual gain control • Averaging, DLA phase angle calculation, PD detection • Data transfer • RPC interface and marshalling • Thread control and message passing • Command parsing and response The network connectivity may be put to further use, by providing the means to reprogram the FPGA remotely, from the master station. 6.3 Master station network software The master station functions as a data concentrator, as well as a user interface. The network software must perform the following tasks: • Windows interface and diagnostics • DDE server • FPGA programming and interface • Network wide synchronisation • Message and data passing between GUI and remote stations

RPC interface and marshalling Thread control and message passing

6.4 Master station GUI

• • • • •

Windows interface and diagnostics DDE client Data and command transfer Schedules User interface

7. Sources of error The accuracy in measuring phase angle between a voltage and a current measured at two geographically separate points in the system depends on the various sources of error, which can be characterised by fixed error or by a random process with zero mean. The requirement of condition monitoring is to detect changes in phase angle of the leakage current, rather than the absolute phase angle. In this respect any errors which are unlikely to change over time give no cause for concern. Fixed errors due to propagation time along copper or fibre paths can be calculated from knowledge of the medium transmission properties. These are: • Voltage phase shift along power transmission lines between voltage and current measurement points. • Delay in timing signals travelling through fibres, including fibre transducers and repeaters. Another source of fixed error is the phase shift in the instrumentation CT and measurement electronics (e.g. filters). This error may vary with each node and with each channel or phase within that node, but can be measured during calibration and compensated by software settings.


The VT used as the reference represents another source of fixed error in phase angle. This cannot be measured to sufficient accuracy, but will affect all insulators equally, so can be ignored. Another class of errors can be considered random with zero mean: • •

ε = ε ×ε = 2

4 N2

N

k =1 l =1

( )

The 95% confidence deviations) is therefore

v(t ) = cos (2πt + ϕ ) 2π vk = cos ( 2πk + ϕ ), ω N = N N  jkω N + φ  vk = Re e   

− jkω N

the angle of the resulting signal is the phase difference between the measured signal and the reference,

2 N − jkω N  vk e  ∑  N k =1 

φ = arg

N=

(2

standard

16δ 2 = 127 3.10 −8

A frequency error in the timing signal may be represented as

kω vk = cos ( kω + ϕ + e ) N N

δk ≤ δ

where ω is the true angular frequency, and ωe is the error in frequency. As ωe is a fixed amount for one cycle, it may be treated like ϕ. An error of 0.1 mRadian is equivalent to a change in period of 318nS for a 50Hz signal. This sets the upper limit of cycle-to-cycle error in the timing signal.

Let the error in the measurement of vk be N

k =1

level

Therefore the number of samples per cycle must be greater than 127. The hardware and software have been designed so that this may be set at any arbitrary number up to approximately 1500. Averaging over several cycles may be used to effectively increase this number, so there is no reason why the stated accuracy cannot be obtained.

The error due to A to D quantisation can be modeled as white noise so that

− jkω ∑δ e

4 3N

Assuming as A to D converter of 12 bits, so that δ = 2-12, and phase error no greater than 0.1m Radians,

where k is the sample index, and ϕ is the phase shift with respect to the reference. If the measured signal is convolved with the reference sine wave

2 N

e jlω N

4δ 3N

If this signal is sampled N times in each period,

ε=

− jkω N

l

Ε ε ×ε = ε rms = δ

Assuming v and t are scaled appropriately, then the voltage signal may be represented by

vk = cos ( 2πk + ϕ ) + δ k , N

k

When k ≠ l, Ε(δkδl) = 0: when k = l, Ε(δkδl) = δ2/3, so

Frequency error in the timing signal. Quantisation error in A to D conversion.

e

N

∑∑ δ δ e

N

k

and the second moment can be expressed as 8

Conclusion


High voltage insulator condition monitoring can be approached using a mixture of passive and active measurement nodes, connected by a network based on the Ethernet standard, and using commercially available components. In order to achieve a high degree of accuracy for phase angle measurement, a second, parallel timing network may be used. The proprietary hardware required for measurement may be integrated with the hardware required for interpretation of timing signals.

Science Measurement & Technology, 1995, 142, (1), pp.29-36. 7 NIEMEYER, L.A.: ‘A Generalized Approach to Partial Discharge Modeling’, IEEE Transactions on Dielectrics & Electrical Insulation, 1995, 2, (4), pp. 510-528. 8 GULSKI, E. and KRIVDA, A.: ‘Influence of Aging on Classification of PD in HV components’, IEEE Transactions on Dielectrics & Electrical Insulation, 1995, 2, (4), pp. 676-684. DANIKAS, M.G.: ‘Small partial discharges and their role in insulation deterioration’, IEEE Transactions on Dielectrics & Electrical Insulation, 1997, 4, (6), pp. 863-867. 9

Sources of error in the measurement system have been systematically examined and corrected using the calibration techniques described. Adjustments can be made by setting parameters in software.

9. Acknowledgements This research is supported by an Australian Research Collaborative grant, and by Pacific Power International.

10. References 1 ARONOV, M.A., and KOKURKIN, M.P.: ‘Energy characteristics of polymer insulation dielectric strength under partial discharge stress’, Electrical Technology, 1994, (3):49-61. 2 SIMONS, J.S.: ‘Diagnostic testing of highvoltage machine insulation’, IEE Proceedings B, 1980, 127, (3), pp. 139-154. 3 KEMP, I.J.: ‘Partial discharge plant-monitoring technology: Present and future developments’, IEE Proceedings – Science Measurement & Technology, 1995, 142, (1), pp. 4-10. 4 NATRASS, D.A.: ‘Partial Discharge Measurement and Interpretation’, IEEE Electrical Insulation Magazine, 1988, 4, (3), pp. 10-23. 5 FRUTH, B.A. and GROSS, D.W.: ‘Partial discharge signal generation transmission and acquisition’, IEE Proceedings-A-Science Measurement & Technology, 1995, 142, (1), pp. 2228. 6 PEDERSEN, A. CRICHTON, G.C. and MCALLISTER, I.W.: ‘Partial discharge detection: theoretical and practical aspects’, IEE Proceedings A Measurement and Communication Techniques for Remote Condition Monitoring of High Voltage Transformer Bushings. 8