Evolutionary Computation and Visualisation as Decision ... - CiteSeerX

Prepublication version of the paper to appear in: Innovation in Civil and Structural Engineering, Chapter 3, (invited lecture), Edited by B.H.V. Topping edited, Saxe-Coburg Publications, pp 49-74 ISBN 1-874672-24http://www.saxe-coburg.co.uk/pubs/descrip/sl2005.htm

Evolutionary Computation and Visualisation as Decision Support tools For Conceptual Building Design Yaqub Rafiq, Martin Beck, Ian Packham and Sue Denhan University of Plymouth, UK. {mrafiq,ipackham,mbeck,sdenham}@plymouth.ac.uk

Abstract The conceptual stage of the design is undoubtedly the most innovative stage of the design process during which most decisions that determine the future of the project are taken, and about 70% to 80% of resources are committed. Unfortunately, due to discrete, fuzzy and incomplete nature of information in this stage the design and decision making process is non-algorithmic which is difficult to be modelled in a computer program. In the past a number of attempts have been made, using knowledge based expert systems and classical optimisation tools to involve computers in decision making process, but none of these attempts have been totally successful. Applications of Evolutionary Computation in design opened up a new opportunity in this area. This paper introduces a novel approach and demonstrates that interactive use of evolutionary computation, assisted by visualisation tools, leads to a human-led search that could be the key for solving this deadlock. A system based on an interactive visualisation clustered genetic algorithm is introduced and its application on an example of a multi-disciplinary decision making process is demonstrated. Keywords: Evolutionary Computation, Visualisation. Conceptual Design.

1 Introduction Extensive exploration of the entire search space, in order to increase the possibility of discovering discrete regions of high quality designs, increases the designers’ knowledge of the problem. Human interaction with the computer, in order to conduct a focused and concentrated exploration inside these regions, can lead to the discovery of new knowledge that would not have been possible by other means. Use of interactive visualisation supported by evolutionary computation tools allows designers to conduct a human-led exploration over the entire search space. Combining the search capabilities of the evolutionary computation with the exploration power of interactive visualization tools at the conceptual stage of the design process enables designers to develop a deeper understanding of the search and solution spaces, and assists them in evaluating the merits of alternative designs and in their decision making. Through direct interaction with the visualisation tools and system to conduct a human-led exploration, the designer may discover a new direction as having potential to be explored in the hope of discovering new and innovative solutions.

1


Optimisation is a repetitive process of discovery, evaluation, comparing, and identification and selection of better solutions until no further improvement in the quality of the solution is possible. Classical optimisation methods, which are mainly gradient or heuristic based, are more suited for solving single objective problems rather than multi-objective optimisation problems. Although classical mathematical based optimisation methods have proven to be powerful optimisation tools, they have exhibited certain serious limitations when applied to engineering design [1]. A major difference between classical and evolutionary optimisation techniques is that in the evolutionary optimisation a population of solutions is processed in successive generations. This characteristic of the evolutionary optimisation techniques gives them an advantage in solving multi-objective optimisation problems particularly if the objective is to find multi-optimal solutions (also know as the Pareto front) concurrently. The majority of real word problems are multi-objective optimisation problems where the objectives are typically in conflict with each other (e.g. minimising cost whilst maximising performance and profitability). During a multi-objective optimisation process a number of optimal or near optimal solutions can be found. In this process as trade off between conflicting objectives are essential when selecting a compromised solution [2]. While the main focus of this paper is towards the application of adaptive and evolutionary techniques and visualisation tools as applied to the conceptual stage of structural building design, that assist designers to interactively explore discrete regions of promising designs and assist them in their decision making, the methodologies discussed are equally applicable in any area of science and engineering. Firstly though, this paper will briefly discuss how the early use of KnowledgeBase Expert Systems aided the designer at the initial stage of the design process. Following this, the area of Evolutionary Computation (EC) and its applicability to conceptual design will be outlined. The topic of section 3 will be on the application of EC techniques to structural building design, while all stages of the design will be reviewed, the focus will be on the conceptual stage. Section 4 contrasts search with exploration and examines the role that the designer should play in a evolutionary design tool.

2


Initially section 5 will examine how interactive and visualization techniques have played a role in EC, before discussing how an interactive visualization system can enhance a designer capability. Examples of such systems are reviewed in section 6. Finally section 7 will describe our recent research on interactive visualisation: the Interactive Visualisation and Clustering Genetic Algorithm System (IVCGA). This is as a powerful decision support tool for the conceptual stage of the design process. It allows the user to identify regions of promising designs and permits the designers to use their expert domain knowledge to conduct a more concentrated and focused search ‘inside’ these regions, identified by the EC search system or to extend the search ‘outside’ these regions to explore other possibilities that has not been currently discovered by the EC. A case study in office building layout will be used to illustrate the capabilities of such a tool.

2 Early applications of Artificial Intelligence in Building Design The early 1980s, saw extensive research activity in engineering and architectural design which captured expert knowledge and utilised it in Knowledge-Based Expert Systems (KBESs). Some examples of this research related to building design are: Karakatsanis [3] : FLODER: A Floor Designer Expert System; Lane & Loucopoulas [4]: A Knowledge Based System as a Mechanism for Optimization of Conceptual Design in Civil Engineering Projects. In late 1980s a number practical systems were developed under the supervision of Steven Fenves, some of the most popular one are: Maher ML & Fenves [5]: HI-RISE: An Expert System for Preliminary Structural Design of High-Rise Buildings; Sriram [6]: DESTINY: A Model for Integrated Structural Design; Sriram [7]: ALL-RISE: A Case Study in ConstraintBased Design; Karakatsanis [3]: FLODER: A Floor Designer Expert System. Other relevant research in this are: Harty [8]: An Aid to Preliminary Design; Kumar & Topping [9]: Issues in the Development of a Knowledge-Based System for the Detailed Design of Structures. Rafiq & MacLeod [10]: Automatic Structural Component Definition from a Spatial Geometry Model; Jain et al. [11]: A Formal Approach to Automating Conceptual Structural Design, Part 2: Application to Floor Framing Generation; Harty & Danaher [12]: A Knowledge-Based Approach to Preliminary Design of Buildings and, Najafi [13]: An Integrated Generative Computer-Aided System for Architectural and Structural Design. Due to restrictive and domain specific nature of KBESs they were not able to fully penetrate the realm of conceptual design.

2.1 Evolutionary Computing in Design Evolutionary Computation (EC) is an umbrella term for an approach which draws its inspiration from biological and natural systems. More specifically the fundamental

3


Darwin principles of survival of the fittest, using the processes of selection, recombination and mutation. The field of Evolutionary Computation can be characterised by a set of four algorithms (known collectively as Evolutionary Algorithms - EA): Genetic Algorithms (GA), Holland [14]; Evolutionary Strategies (ES), Rechenberg [15]; Evolutionary Programming (EP), Fogel et al [16] and Genetic Programming (GP), Koza [17]. While each contributes to EC in its own way, they can be distinguished, albeit with over simplification, by their core tenets. Thus the area of GAs has tended to concentrate on a population of fixed length binary strings, the genotype, encoding a problem domain, the phenotype. An individuals probability of being a member of the next generation is proportional to that individuals fitness, and once selected operators such as crossover and mutation are applied to produce new, possibly diverse, individuals. For ES the focus was to use real valued genes which are operated on by primarily mutation (although crossover is used in some implementations). A departure from the GA approach is that these operators are used prior to selection, and the selection itself is deterministic: The best individuals are selected. The direction of EP, on the other hand, was originally developed for the evolution of finite state machines [16], but later extended [18] to encode real numbers. The main operator used is again mutation, with crossover playing no role. Finally, the defining characteristic of GP lies in the structure of the chromosome. Typically the chromosome is represented as a tree structure with mathematical operators as nodes, and operands as leaf nodes. Its main application area is in that of function approximation and curve or surface fitting. The operation of crossover and mutation are adapted so that in the case of the former, entire branches are swapped between individuals and in the case of the latter the values of the nodes are randomly changed. While all these EA’s have been applied to design domains, perhaps the most widely used is that of the GA. However, as stated earlier, an individuals probability of selection is dependant upon its fitness, and for the GA this needs to be a singular value. Unfortunately the majority of real world problems are multi-objective in nature, and the issue now becomes how to combine multiple objectives to obtain a single fitness value. A simple approach is to use a weighted sum technique: Each objective is assigned a weight and the fitness is simply the sum of these weighted objectives. The problem with this approach is, firstly, the assignment of an appropriate set of weights, and secondly, the results will tend to be a single solution rather than a set of good candidate solutions.

4


A more common approach, though, is to use Pareto based techniques [19]. Here the goal is to evolve a set of individuals which not only lie on the Pareto front (as in a two objective case), but are also relatively spread across the front. With a diverse set of equally ‘fit’ solutions the designer is now able to make a selection between them, using domain knowledge and personal experience. Since their inception [20], Pareto based GA techniques have spawned a considerable number of variants (see Deb [2]). While such techniques differ in their implementation, it is observed they are generally all based on a similar premise: The fitness of a individual is now some function of the number of individuals which it dominates or its position on a series of Pareto fronts. The diversity across a front is achieved by some sharing or niching technique (e.g. The Non-Dominated Sorting GA of Srinivas and Deb, [21]). While traditional mathematical programming techniques can be applied to Pareto type problems, such techniques generally only produce one solution on the front or are susceptible to shape of the front. As EAs do not suffer from these weaknesses a significant amount of research has grown in the area known as Multi-Objective Evolutionary Algorithms (MOEA), see Coello [22] for an extensive list of published material in this area. Evolutionary computation has found wide use and acceptance in the design community and while initially the focus was towards its application at the detailed design stage, the role that EC can play at the conceptual stage has become far more evident in recent years. It is at the conceptual stage that, EC techniques, particularly the GA can offer the following advantages: ♦ It possesses a population of solutions in successive generations. This is compatible with the way ideas are generated through divergent reasoning at the conceptual design stage. ♦ A wide exploration of the search space is possible: The stochastic and random nature of the GA allows the search to be conducted over the entire search space and to identify regions of high quality (optimal or near optimal) solutions. ♦ Crossover allows the recombination of different potential solutions, while mutation allows for new and innovative characteristics to be introduced. ♦ It can be adapted to multi-objective design problems. The next section illustrates applicability of EC based techniques across all stages of the design process (detailed, component and conceptual).

3 Evolutionary Computation in Building Design

5


3.1 Detailed Design Applications - Trusses and Frames Amongst the earliest applications based on biological principles were those used to optimise aircraft design. From the mid-1960’s, Rechenberg [16, 17] applied an optimisation technique that determined the optimum configuration of a steel plate, hinged in five sections, for minimum drag. Early proponents of the GA considered structural design applications included Goldberg et al [23]. Schwefel [24] investigated an 18-bar truss optimisation and successfully showed that a GA could achieve a design solution having a weight only 5% short of the optimum solution, noting that a convex linear programming technique had previously only achieved an 6%-from-optimum solution, in 1989. In 1991, Jenkins [25, 26] applied the GA to structural design optimisation problems. Jenkins pioneered the study of various other multi-member structures. These included a trussed beam and a thin-walled section. Shortly thereafter, in 1992, Rajeev & Krishnamoorthy [27] also examined a classical three-bar truss problem, followed by ten-bar, and 25-bar planar trusses, using the GA. In all cases, the prime objective was to obtain minimum-weight solutions, capable of withstanding the prescribed loading. More complex structures require more complex representations and present a greater challenge to the GA. Adeli and Cheng [28] reported their results on the optimisation of space structures using genetic algorithms. Adeli and Cheng [29] introducing an augmented Lagrangian Multiplier Method to determine the minimum weight design of high-rise steel structures and space frames. They found this algorithm to reduce extensive numerical testing to find a suitable value for the penalty function coefficients. In their latter work, [30] successfully applied their algorithm on large structure optimisation using highperformance computers. Cai and Thireauf [31] and Jenkins [32] have studied theoretical aspects for handling very long chromosome structures efficiently, required for modelling complex structures. Dhingra and Lee [33] investigate the application of GA's to single and multiobjective structural optimisation with discrete-continuous variables. Kicinger and Arciszewski [34] have reported the initial results of the applications of generative cellular automata-based representations in evolutionary structural design and Kicinger et al [35] have presented initial results of a study on the application of evolutionary multi-objective optimisation methods in the design of the steel structural systems of tall buildings. Kicinger et al [36] have conducted extensive review of evolutionary computation in the context of structural design. This review briefly gives a general introduction to evolutionary computation and recent developments in this field and then focuses on the field of evolutionary design and its relevance to structural design and presents a comprehensive review of the applications of evolutionary computation in structural design. Finally, a summary of the current research status and a discussion on the most promising paths of future research are also presented.

6


3.2 Component/Detailed Design Applications In 1994, Kousmousis & Georgiou [37] presented GA applications in which the practicality of solutions were of paramount concern. Their research was amongst the first to combine construction factors as direct objectives of the optimisation. A practical steel roof truss arrangement appropriate for industrial / warehouse usage was the goal in a pioneering combinatorial, mixed layout and sizing optimisation problem. In this study objectives were weighted to simulate a desirable level of compromise between separate criteria representing minimum-weight design and buildability. The aim was not necessarily to seek the global optimal solution, but rather to achieve a highly satisfactory near-optimal solution, using an efficient search procedure. Success led Kousmousis & Arsenis [38] to address the detailed design of a twospan continuous RC beam using the same approach. Here the optimal arrangement of reinforcement within the section was based upon criteria representing the threefold aim: - to use the fewest total number of reinforcing bars as possible, to use the fewest different bar sizes, and to provide the minimum amount of reinforcement necessary to comply with design requirements. Layouts were configured to comply with the dimensions particular to a given section. Kousmousis & Arsenis [38] reported the system was efficient at searching the large design space (which contained over 16 million potential combinations) to detail a RC beam, and said that it represented a more viable alternative to rule–processing systems applied previously to similar tasks. In 1995, Rafiq [39] applied similar rationale to the reinforcement detailing of RC columns in bi-axial bending. Again, the goal of the optimisation was to use the fewest number of different sized steel reinforcing bars distributed in an optimal fashion. A difference here, however, was that both layout and sizing activities were incorporated directly into the optimisation algorithm. The objective function used normalised values, representing the degree of satisfaction of each individual design criteria, to give each equal consideration. A real number scheme was also employed and was found to be effective. In 1996, a study by Lucas and Roddis [40] was reported that also followed the work of Kousmousis et al [37, 38] in component design. This study concerned the design of rectangular RC beams. A constraint-based GA technique that combined evolutionary search with constraint satisfaction programming in Common Lisp was employed.

3.3 Conceptual Design of Trusses and Frames Due to non-algorithm nature of the information at the conceptual stage of the design, penetration of computers at this stage has been extremely slow. As stated by Goldberg [41]: The creative processes of engineering design have long been

7


regarded as a black art. While the engine of analysis steamrolls ever forward, our understanding of conceptual design seems locked in a timewarp of platitudes, vague design procedures, and problem-specific design rules. Successful application of evolutionary computation, particularly GAs to solve multi-dimension problem was very promising. Some examples of such application in details and element design of structures were discussed in the previous section. Up to this point, the research that has been mentioned has minimal relevance to conceptual design. However, the study of trusses by Jenkins [25,26] encouraged Grierson & Pak [42] to attempt simultaneous sizing, topological and geometrical optimisation problems, not possible with conventional mathematical programming techniques. In this research, Grierson & Pak approached the combined optimisation problem of planar frames, by first taking a simple example and later extending the approach to a more detailed skeletal structure resembling a building in section. Rajan [43] conducted similar investigations using trusses instead of frames. In addition to member size variation, Rajan studied techniques that simulated the removal of non-essential bracing members from a truss. Grierson and Pak [42] applied an approximation technique to analyse the frames generated by the GA in order to significantly reduce the computational effort required. Jenkins [44] formally reported the need to investigate re-analysis techniques to improve the computational efficiency in structural analysis procedures. Experiments by Grierson & Pak [42] produced unsymmetrical solutions for unsymmetrical load conditions. Jenkins [26] advocated analysing the final state of constraint satisfaction for optimal design solutions produced using the GA. In 1993, Parmee and Bullock [45] described how the GA had been successfully applied to the design of a hydropower scheme, which incorporated the optimal geometric design of a concrete arch dam to resist self-weight and hydrostatic forces. In this work, Parmee and Bullock recognized that radically different geometric forms (i.e. different curvatures) embodied different design concepts that warranted consideration. Parmee and Bullock proposed that an optimisation based on the Structured Genetic Algorithm (SGA), by Dasgupta and McGregor [46] was appropriate for handling multiple different concepts simultaneously. During many years of research using evolutionary computing to aid engineering design, Ian Parmee and the research team in the Plymouth Engineering Design Center, developed a number of useful ideas, algorithms and techniques [47-50] that could be combined in an interactive evolutionary system, particularly useful in addressing conceptual design problems. The research started from general ideas to locate and analyse robust regions of the search space and using directed search to define feasible regions in multiobjective problems [49], and later on they combined some of the developed techniques in an interactive evolutionary system, such as cluster-oriented genetic algorithms (COGAs).

8


The Cluster Oriented Genetic Algorithm (COGA) is perhaps one of the novel ideas that have been developed in the Plymouth Engineering Design Centre [51, 52], which mainly focuses on the conceptual design issues. The purpose of this development was to enable the user to explore complex design search space. The COGA rapidly decomposes the search space into succinct regions of high performance. Successive COGA runs may then investigate specific areas of the search space where novel design directions may be found. Throughout this iterative process the designer continuously gain general knowledge relating the overall design domain, such as parameter sensitivity, constrain violation and more specific information relating to the regions of high performance. The overall objective of the COGA is therefore the identification of high performance regions and the achievement of sufficient regional set-cover (in terms of number of solutions) to allow significant qualitative and quantitative information to be extracted. Using this information and information from successive runs, design expertise and in house knowledge; the search parameters, objectives, constraints etc. may be modified to investigate the feasibility of novel search areas or the further development of the established search direction. From 1997 to the present day, there has been noticeable growth in interest in the application of GAs specifically to conceptual building design. One reason for this is that building design problems are highly multi-dimensional, multi-criteria and multi disciplinary problems with lots of inherently conflicting criteria. Studies by Donald Grierson and his research team in Waterloo, [53,54], Yaqub Rafiq in Plymouth [5559] and John Miles and his research group in Cardiff [60-62] are perhaps the first people who have made particular contribution is this area.

4 Role of Human Computer Interaction in Design While there are many definitions of design and the purpose of each stage which makes up the whole design process (i.e. [63], [64]), in essence design is a human activity. An early categorisation of design as a problem solving process [65] and the subsequent idea that design is an activity of searching the solution space for a solution dominated the direction of Artificial Intelligence in design for some time. It was Smithers [66] who questioned the validity of this approach: Search implies a well defined problem from which a solution can be generated, the design requirements (i.e. the ‘problem’) at the conceptual stage are ill-defined and the process should be viewed as one of exploration. This issue is also eloquently advanced by the work of Gero [67] and Maher (e.g. [68], [69]) who point the way to which conceptual design is both an exploration of the design requirements and the potential solutions to those requirements.

9


It this notion of 'design as exploration' in which human involvement and visualisation techniques play a key role. In a standard, pure machine driven optimization process, either classic or evolutionary, it is usual that only a final ‘best’ solution is being reported and a range good (near optimal) solutions are often obscured from the designer. This is because searching for a single best solution forces the search to a certain direction and latching onto a single solution to the extent of refusing to consider other possibilities [70]. In the field psychology this attitude is called ‘fixation’ [71]. Unfortunately, this not only deprives the designer of seeing alternative solutions, but tends to promote fixation onto a single machine-preferred solution [72]. In optimisation, the fitness evaluation of a potential solution is based on quantifiable criteria encoded in the objective function of the computer based algorithm. While it is extremely difficult and sometimes unnecessary to encode qualitative criteria such as aesthetics, historic consideration, buildability issues (etc.), into the objective function, such criteria may influence the direction the designer takes. However with only quantifiable criteria driving the search process any solutions relating to qualitative criteria are simply not evaluated as such. While at the outset it is possible to constrain the search space to those criteria to which the designer is currently most concerned, thus enabling a design focus, such constraints serves to only restrict the search space with no guarantee of a ‘good’ solution. Design at the conceptual stage implies creative thinking and creative thinking does not fit a prescribed set of serial steps. Unlike analysis process of a single given condition, which is a well defined algorithmic process, design solution partially defines required conditions and, therefore seldom result in only one possible solution. Creative design is a complex, multi-path exploration process of exploring many possible ideas [71]. Interactive user interface and human-driven search goes beyond the predefined algorithmic procedure which aids the designer to freely explore the design search space in a creative way. Unlike analysis tools it is not intended to yield one single solution, but rather to supply the designer with stimulating, plausible directions [71]. This enables the designer to widely explore search and solution spaces, evaluate merits of computer generated solutions by considering non-quantifiable nonencoded criteria in order to derive the search to the right direction.

5 Major Attributes of an Interactive Visualisation System

10


Before examining the attributes of an interactive visualization system, the role that each of the individual components (i.e. ‘interactive’, and ‘visualization’) have played in EC is briefly reviewed. Interactive evolutionary systems can be categorized into two broad groupings: Where the user makes a subjective evaluation at the selection stage, and where the user changes design objective preferences (see also Parmee, [73], for a similar categorization) In the case of the former, the user effectively replaces the fitness function as in the classic Biomorhps of Dawkins [74], in this instance the user is presented with a series of candidate designs and selects a subset for reproduction. (see also Banzhaf, [75], for a review). While such systems have only found limited usage in conceptual building design [76], it has been argued that such systems “can fit naturally into human design thinking and industrial design practice” [77]. In the case of the latter, the user is able to define design objective preferences at the start of the GA run, or to modify individual elements of the design (these systems are reviewed in section 6) In terms of visualisation two broad approaches can be identified. Firstly, it is the result of the search process which is presented to the designer, thus in the single objective case the classic approach has been a simple fitness plot and in the multiobjective case, plotting the Pareto front or Pareto surface (the work of Grierson and Khajehpour, [78], puts this to technique to good effect). Secondly, in more sophisticated systems the focus has been on the visualisation of the GA search process itself: This may involve tracing the ancestry of the evolving individuals or representing the genetic composition of individuals (see Hart and Ross, [79], Collins, [80], and Smith et al,[81]). In all cases though visualisation is performed during or after the GA run has terminated and the user is not able to interact with the system either to refine the search or to move the search to a different area of the search space. A major objective of an interactive visualisation system, developed by the authors, is to assist designer through interaction to explore a range of feasible and innovative solution that best fit the design brief requirements. The process of exploration of search and solution spaces is more useful at the conceptual stage of the design process where design information is ill defined. The interactive visualisation process is fundamentally different from the traditional search and optimisation in: ♦ It opens a human window in which the user freely interact with the system to explore and visualise the search and solution spaces to find better solution that suite the design requirements. The designer can conduct a focused and concentrated search inside in interesting region of the search space or extend the search outside the current focused region. It is a human-driven search in which

11


♦

♦

♦ ♦ ♦

♦

the designer is in full control of the process and the designer can derive the search using heuristics and domain knowledge. More innovative solutions can be generated based on the selective pressure which is the result of the designer’s interaction. It does not solely rely on encoded quantifiable objectives as the search is guided by the human-interaction and new boundaries, constraints and requirements are dynamically introduced. Many conceptual design considerations that are qualitative parameters can be incorporated through direct human-interaction. It more suited with real life design problems that are multi-dimensional, multicriteria and multi-disciplinary in nature. Experts from different disciplines can interact individually with the system to explore their own areas of interest and evaluate the effect of changes that they propose on the overall design. It therefore, enhances the awareness of the project team on various aspects of the project which could lead to a mutually acceptable solution. Visualising multi-dimensional problems and presenting information about the search and solution spaces can lead to discovery of knowledge about the specific regions and understanding of collaborative design issues. It allows wider exploration of new and unknown regions, within the entire search space, in order to increase the possibility of discovering isolated regions of high performance. This increases the designers’ knowledge of the problem. It also allows exploitation, which enable the designer to conduct a refined and concentrated search within identified regions of the solution space in order to discover solutions of greater optimality or explore alternative solutions that might be located in a new region of the solution space. Through direct interaction with the system, the designer may discover a new direction as having potential to be explored in the hope of discovering new and innovative solutions.

6 Examples of Interactive and Visualisation Systems for Engineering Design A comprehensive review by Takagi [82] lists a number of approaches to and applications of the interactive evolutionary computation. One of the first preliminary design systems devised by Pham & Yang [83], allowed the GA to produce configurations for the user to view and evaluate in order to make decisions about further redesign and other additional GA runs if required. Jo [84] discovered that adding human interaction to his evolutionary design system meant domain knowledge could be incorporated online; solutions can be independently visualized in a space layout problem and the user allowed to modify individual elements of the design. The interaction of a user has also been considered in a multi-objective environment: Fonesca CM, Fleming [85] proposed a decisionmaker (DM) that controls which objectives have more importance within a nondominated set of solutions. They suggested the DM could be a human or an expert

12


system. Horn [86] points out that there are three different approaches to decision making in multi-criteria problems: make a multi-criteria decision before search, make a decision after search or integrate the search and decision making. The latter approach would appear to be the most powerful, incorporating iterative search and decision making. Mathews & Rafiq [87-89] developed a Conceptual Building Design (CBD) system using the Structured GA. This system used the objective oriented functionalities of visual C++ that allows the user to manipulate the system more interactively, a powerful GUI was used which included a number of interactive dialogue boxes for effective user interaction. To allow better interaction with the system at run time, the design hierarchy was made available to the designer. The tree control, as shown in Figure1, provides basic functionality for manipulating the nodes (different frame systems) and branches in a hierarchy (different design options). Using this facility allows the designer to include or exclude particular design options from the GA search, at the runtime. This facility is considered to be extremely useful in a number of ways. It allows the user to force the GA search to follow a particular branch of the design hierarchy, which may not be considered by the GA as a best choice. For example, the GA may have found a steel frame to be the best choice for the design, but the user may want to know what happens if concrete is used instead? In this case, by interactively prohibiting the steel frame options to be included in the search, the user can force the GA to consider the concrete option only. This facility is also useful if the design brief requires a particular construction material to be used or a designer prefers a particular floor system, etc. to be considered.

Figure 1: Dialogue box representing the design hierarchy In this limited interaction the system allows the designer to trace the design evolution process during a whole run of a GA operation. This is an important facility which adds transparency to the otherwise ‘black box’ GA operation. The convergence graph view was modified to provide post-processing support by making it possible to examine solutions using cursor picking. By clicking the mouse in a point on the graph, the corresponding details of the concept were shown in the

13


second window. This facility was also made available during a genetic experiment while the GA was paused. An example of the use of this facility is presented in Figure 2. Fitness

0 8 0 6 0 4 2 0

1 1 5 0 Generati

2

2 5

3 0

Concrete Frame, Precast Floor Plan 35m x 20m,

Figure 2: Tracing design Evolution during a GA run Grierson and Khajehpour [78] used Pareto optimisation techniques for the multicriteria conceptual design, which involves genetic-based stochastic search and colour-filtered graphics. In this investigation, a large set Pareto non-dominated designs that are equal-rank optimal in the sense that each is not simultaneously dominated for all objective criteria by any other feasible design, were captured. They than used a computer colour filtering of the Pareto-optimal design set and created a large body of informative graphics that identify trade-off relationships between competing objective criteria, as well as design subsets having particular designer-specified attributes. A detailed illustration of the method for the costrevenue conceptual design of high-rise office buildings, including several examples was presented in [78]. Grierson and Khajehpour work is an excellent example of knowledge discovery using a simple visualisation tool. This research also clearly demonstrated that using visualisation tools effectively, it possible to discover the interrelationships between design parameters more clearly. Research in Plymouth Engineering Design Centre by Parmee and Bonham [90] resulted in development of an Interactive Evolutionary Design System (IEDS) based on a system of iterative redefinition of variable and objective space by a designer. These ideas evolved from many years of research in using evolutionary computing to aid engineering design, starting from general ideas to locate and analyse robust regions of the search space. Development of cluster-oriented genetic algorithms (COGAs) allowed inclusion of user preferences between objectives to direct coevolutionary search [91]. COGA extracts regions of good solutions from a GA run by filtering high performance solutions as the search progresses. More recently work on visualisation of COGA data has included the development of a novel technique called parallel coordinate box plots [92,93], that uses the parallel coordinate technique developed by Inselberg & Dimsdale [94] to visualize many variables at

14


once and compare the distribution of solutions between objectives using statistical analysis and discover overlap between various objectives.

7 An Interactive Visualisation and Clustering Genetic Algorithm System (IVCGA) The Interactive Visualisation and Clustering Genetic Algorithm System (IVCGA) developed by Packham et al. [95-98] is a system that was designed and built to focus on user interaction at the outset and allows the user to freely interact with specific regions of the search space. The work is centred on a user interface that allows the user to ‘zoom in’ to specific regions of the search and solution space, defined by the user, and create new data using further GA runs. Normally, domain knowledge is heavily relied upon within the GA optimization programs to constrain the objective function so that only feasible solutions are guaranteed, but are these results optimal or have better results been ignored because of the constraints imposed on the GA search? Many researchers have realised that these issues can be better addressed by visualising the search and solution space of designs generated by the GA if user interaction is freely allowed. This could possibly lead to the discovery of more creative designs. By including a flexible interactive user interface supported by visualisation tools, such as the IVCGA, this allows the designer to conduct a focused and concentrated search inside regions of high performance solutions or to extend the search outside the boundaries of the current search, it is possible to incorporate non-encoded design requirements into the human-driven search and evaluation process. In the IVCGA, the constraints are removed as much as possible and all the data produced by a short GA run is initially available for visualisation, the user can then choose to keep or filter out information as required interactively. The data is then clustered using a fast, novel technique according to user requirements. The advantage of this method is that data is generated very quickly and low performing, but potentially interesting solutions are not filtered from the results by imposing constraints. The interface allows any combination of 2D or 3D variables to be visualised and compared with other views. Other techniques such as the scatter-plot matrix and parallel coordinates are included which allow visualisation of all or part of the data so that the interaction between variables at the local level is clear. The additional use of colour enables the user to emphasise the clusters that are important to the user whilst keeping other interesting data available. Further techniques, such as comparing the data with the principal and independent component views, can help to reduce the dimensionality of the problem. The authors believe that by combining flexible user interaction and visualization capabilities with the power of evolutionary computation in a human-driven interactive system, it was hypothesised that such a system would deliver invaluable decision support and knowledge discovery, as advocated by [99] and [58]. The design of the system was inspired by the visualisation literature for engineering

15


design, for more details see [100-103], also [104] for guidelines on interactive visualization.

•

•

•

•

IVCGA was designed to offer the following features: The fast generation of diverse data by running a simple GA for a low number of generations to reduce the number of function calls with normal crossover and mutation rates. The diversity of solutions is maintained by applying a high mutation rate when duplicate solutions are identified. An easy to use interface that allows direct manipulation of the data and views. Various high dimensional visualization techniques are supplied to enable understanding of the data from different viewpoints and combination of parameters. An automatic clustering procedure is built into the system that identifies clusters relevant to the problem in hand. Colour is used to highlight important clusters, enhancing perceptual understanding of the data. Extensive interaction is supported allowing the generation of further data with the GA inside or outside regions identified by the user or clustering algorithm. The definition of clusters can be modified by the user or even created manually, ensuring complete freedom of search and human-led exploration of the search space.

7.1 The Application of IVCGA to Office Building Layout The research presented in this section is a preliminary investigation on the interactive use of visualisation in a multi-disciplinary design environment. The exercise has been conducted in close collaboration with an Architect and the Structural Engineer. The main objective of this exercise is to identify essential requirements of the following disciplines which are the main decision makers in the design process. 1. The Architect – Space, functionality and appearance and user comfort. 2. The Structural Engineer – Building stability, strength and safety. 3. The Heat and Ventilation Engineer – Mechanical, electrical, heating cooling and ventilation systems. To satisfy the architectural space requirements for this preliminary study, it is assumed that a major architectural requirement is the amount of available open space with as little columns as possible. This provides a flexible floor area which could be used by various users with different functionality requirements throughout the life of the building. Therefore, there is a need to minimise the floor area lost due to columns and fire escapes. As the existence of columns in the floor makes the area around the columns also restricted for use, it was decided to use a 1m2 unusable area for each column. In this study, a typical floor area of 1000m2 is assumed.

16


Figure 3 shows a population of designs generated after the first 20 generations of the GA. The frame at left hand side shows a relationship between the bay size and total floor required area. Use of colour helps better understanding of the search and solution spaces. In Figure 3, each colour represents a different bay size (in this example only bay sizes 12m, 10m, 8m, 7.5m and 6m have been selected). The facility provided by the system allows the user to view details of each solution by simply right clicking on any design on the left hand frame. An example of this operation is shown on the right hand frame. As the requirements suggests, fewer columns and other restrictive structural elements mean more letable floor area and hence more long term revenue, the system has not only successfully satisfied this criterion, but also presented the designer with much more information about the extent and the quality of various solutions.

Figure 3: Optimum Architectural Layout also showing effect of bay size on the lost area. Colours represent different bay sizes. In this example, only two materials, Concrete and steel, are included. Due to the highly discrete nature of design parameters for all three disciplines, the design search space is also highly discontinuous. Figure 4 clearly demonstrates this. In Figure 4, concepts generated for each material is distinctly separated. Each vertical line represents designs for a specific architectural grid type. This is a clear advantage of the visualisation tools, which allows designers to investigate each alternative material independently and assess the suitability of each alternative against the design/client requirements. Representing design in the objective space has an advantage when design parameters are discrete, as is the case in many engineering design problems, by visualising solutions in the objective space, distinct clusters of potentially good designs are identified in different colours that could be studied in more detail if necessary.

17


Variable Space

Objective Space

Figure 4: Visualising solutions in Variable and Objective Spaces Figure 5 shows 2D representation of the designs generated by the GA, based on the architectural requirements. The left hand frame presents the entire population generated by the GA. No restrictions or filtering of information, to discourage the generation of unfeasible solutions, have been implemented. The rectangle on the left hand frame is interactively drawn by the designer, using the mouse, around the vicinity of 1000m2 floor area, as was specified in the brief. The frame on the right hand side of Figure 5 shows only solutions which are enclosed inside this region. At this stage the system allows the designer to either view current designs generated by the GA within this region or conduct further runs the GA to populate this area with more designs. If this option is selected, the system automatically penalises any designs outside this region if generated by the GA operation. This feature of the system, which interactively and dynamically defines constraints to the problem, is a unique and very useful decision support tool.

Designer restrict the search in the highlighted area

Figure 5: 2D presentation of the most suitable region for the architectural layout

18


As similar investigation could be conducted by each member of the design team to identify regions of suitable designs that best satisfy their individual disciplinary requirements. Finally, these areas are superimposed on each other to identify a mutually inclusive region that partially satisfies the requirements of all disciplines involved in the design. Figure 6 is an example of such region for the three disciplines for this particular example.

Mutually Inclusive

Figure 6: Mutually inclusive region for all three disciplines

8 Conclusions This paper highlighted difficulties associated with the use of computers at the conceptual stages due to the non-algorithm nature of the activities involved at this stage. Various KBESs tried to model some of the activities of the conceptual design, but capture of expert knowledge and rule based characteristics of these system that require pre-programmed set of rules and conditions as instructions for solving a generic problem makes them very difficult for implementation at the conceptual stage of the design process due to incomplete and fuzzy nature of information at this stage. Evolutionary algorithms on the other hand do not rely on predefined rules and the evolution process is normally steered by the fitness function. This characteristic of these algorithms make them a strong candidate for modelling conceptual design problems. The only problem with these systems is the ‘block-box’ nature of the process when finding an optimum solution.

19


Experience has shown that decision process at the conceptual stage of the design is human-led which is based on human intuition, past experiences and heuristics. Therefore, any system that tries to model activities of the conceptual design must also be human-led. This paper demonstrated that the introduction of interactive tools such is Interactive Visualisation and Clustering GA (IVCGA) has shown to be promising for modelling conceptual design activities as the flexibility offered by the system allows the human expert to steer the search to the desired direction. This makes the system to be trusted by the human expert. IVCGA generates a diverse number of alternative solutions that could be assessed by the designer against a set of predefined requirements. Knowledge discovery is a by product of such visualisation system in which during a concentrated and focused search in the regions of high quality designs, new solutions are discovered that would not have been possible by other mean.

References [1]

[2] [3] [4] [5]

[6] [7] [8] [9]

Fryer C, Ceranic B. Optimum Design of Reinforced Concrete Skeletal Structures: A Comparative Study between Volume and Cost Optimisation. In: 2nd World Congress on Structural and Multidisciplinary Optimisation; 1997; Zakopane, Poland; 1997. p. 731-736. Deb K. Multi-Objective Optimization using Evolutionary Algorithms. Chichester: John Wiley and Sons Ltd; 2001. Karakatsanis A. FLODER: A Floor Designer Expert System [MSc Thesis]. Pittsburgh, PA, USA: Carnegie-Mellon University; 1985. Lane VP, Loucopoulas P. A Knowledge Based System as a Mechanism for Optimization of Conceptual Design in Civil Engineering Projects. Optimisation in Computer Aided Design 1985:1-11. Maher ML, Fenves SJ. HI-RISE: An Expert System for Preliminary Structural Design of High-Rise Buildings. In: IFIP Conference on Knowledge Engineering in Computer-Aided Design; 1985; Budapest; 1985. p. 125-46. Sriram D. DESTINY: A Model for Integrated Structural Design. Journal of Artificial Intelligence in Engineering 1986;1(2):109-116. Sriram D. ALL-RISE: A Case Study in Constraint-Based Design. Artificial Intelligence in Engineering 1987;2(4).96-203. Harty N. An Aid to Preliminary Design. In: Sriram D, Adey R, editors. Knowledge Based Expert Systems in Engineering Planning and Design; 1987; Boston, USA: Computational Mechanics Publications; 1987. Kumar B, Topping BHV. Issues in the Development of a Knowledge-Based System for the Detailed Design of Structures. In: Gero J, editor. in Artificial Intelligence in Engineering Design; 1988; Southampton, UK: Computational Mechanics Publications; 1988. p. 295-314.

20


[10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28]

Rafiq MY, MacLeod IA. Automatic Structural Component Definition from a Spatial Geometry Model. Engineering Structures 1988;10:37-40. Jain D, Krawinkler H, Law KH, Luth GP. A Formal Approach to Automating Conceptual Structural Design, Part 2: Application to Floor Framing Generation. Journal of Engineering Computers. 1991:91-107. Harty N, Danaher M. A Knowledge-Based Approach to Preliminary Design of Buildings. Proceeding of the Institution of Civil Engineers 1994(104):135144. Najafi AA. An Integrated Generative Computer-Aided System for Architectural and Structural Design [PhD Thesis]. Glasgow, UK: University of Strathclyde; 1997. Holland JH. Adaptation in Natural and Artificial Systems. Michigan: Ann Arbor: The University of Michigan Press.; 1975. Rechenberg I. Cybernetic Solution Path of an Experimental Problem. Farnborough, Hampshire, UK: Royal Aircraft Establishment; 1965. Report No.: 1122. Fogel L, J., Owens AJ, Walsh MJ. Artificial Intelligence through Simulated Evolution. New York: Wiley; 1966. Koza JR. Genetic Programming: On the programming of Computers by Means of Natural Selection., MIT Press, Cambridge, MA. Cambridge, MA: MIT Press; 1992. Fogel DB. An evolutionary approach to travelling salesman problem. Biological Cybernetics 1988;60(2):139-144. Pareto V. Cours D'Economie Politique. Lausanne, Switzerland: F. Rouge; 1896. Goldberg DE, Samtani MP. Engineering Optimization via the Genetic Algorithm. In: 9th Conf. on Electronic Computation, ASCE; 1987; 1987. Srivanivas N, Deb K. Multi-objective function optimization using nondominated sorting genetic algorithm. Evolutionary Computation 1994;1(2):127-149. Coello, C. A. C, (2005) EMOO web page At http://www.jeo.org/emo/ Goldberg DE, Samtani MP. Engineering Optimization via the Genetic Algorithm. In: 9th Conf. on Electronic Computation, ASCE; 1987; 1987. Schwefel HP. In: Brebbia CA, Hernández S, editors. International Conference on Computer Aided Optimum Design of Structures; 1989: Wessex Institute of Technology; 1989. p. 218-219. Jenkins WM. Towards Structural Optimization via the Genetic Algorithm. International Journal of Computers and Structures. 1991a;40(5):1321-1327. Jenkins WM. Structural Optimisation with the Genetic Algorithm. The Structural Engineer 1991b;69(25):418-422. Rajeev S, Krishnamoorthy CS. Discrete Optimization of Structures Using Genetic Algorithms. The Structural Engineer 1992;118(5):418-422. Adeli H, Cheng NT. Integrated Genetic Algorithms for Optimisation of Space Structures. Journal of Aerospace Engineering 1993;6(4):315-328.

21


[29] [30] [31]

[32] [33]

[34] [35]

[36] [37] [38]

[39]

[40]

[41] [42]

Adeli H, Cheng NT. Augmented Lagrangian genetic algorithm for structural optimization. Journal of Aerospace Engineering 1994a;7(1):104-118. Adeli H, Cheng NT. Concurrent genetic algorithms for optimization of large structure. Journal of Aerospace Engineering, 1994b;7(3):276-296. Cai J, Thireauf G. Solution of Mixed-Discrete Structural Optimisation with Evolution Strategy. In: Hernández S, Al-Sayed M, Brebbia CA, editors. OPTI IV: 4th Int. Conf. on Computer Aided Optimum Design of Structures; 1995; Miami, FL; 1995. p. 19-26. Jenkins WM. A Space Condensation Heuristic for Combinatorial Optimization. In: Topping BHV, Papadrakakis M, editors. Advances in Structural Optimization, AI CIVIL-COMP 94; 1994; 1994. p. 215-224. Dhingra AK, Lee BH. A genetic algorithm approach to single and multiobjective structural optimization with discrete-continuos variables. International Journal for Numerical Methods in Engineering 1994);37(23):4059-4080. Kicinger R, Arciszewski T. Multiobjective Evolutionary Design of Steel Structures in Tall Buildings. In: AIAA 1st Intelligent Systems Technical Conference; 2004 September; Chicago, Illinois; 2004. p. 20-22. Kicinger R, Arciszewski T. Morphogenesis and structural design: Cellular Automata Representations of Steel Structures in Tall Buildings. In: Proceedings of the Congress on Evolutionary Computation (CEC'2004); 2004; Portland, Oregon; 2004. p. 411-418. Kicinger R, Arciszewski T, De Jong K. Evolutionary Computation and Structural Design: A survey of the state-of-the-art. to appear in Computers and Structures 2005. Kousmousis VK, Georgiou PG. Genetic Algorithms in Discrete Optimization of Steel Truss Roofs. Computing in Civil Engineering 1994a;8(3):309-325. Kousmousis VK, Arsenis J. Genetic Algorithms in a Multi-Criterion Optimal Detailing of Reinforced Concrete Members. In: Topping BHV, Papadrakakis M, editors. 2nd International Conference on Computational Structures Technology; 1994b; Edinburgh, UK: Civil-Comp Press; 1994b. p. 233-240. Rafiq MY. Genetic Algorithms in Optimum Design, Capacity Check and Detailing of Reinforced Concrete Columns. In: Opti 95, Forth International Conference Computing Aided Optimum Design of Structures; 1995 September; Miami, Florida, USA; 1995. p. 161-169. Lucas W, K., Roddis WMK, 1996. Constraint-Based Genetic Algorithm Optimization Applied to Reinforced Concrete Design. In: in Information Processing in Civil and Structural Engineering Design; 1996; Edinburgh, UK: Civil-Comp Press; 1996. p. 171-175. Goldberg DE. Genetic Algorithms as a Computational Theory of Conceptual Design. In: Applications of Artificial Intelligence in Engineering; 1991. p. 316. Grierson DE, Pak WH. Optimal Sizing, Geometrical and Topological Design Using a Genetic Algorithm. Structural Optimization 1993;6(3):151-159.

22


[43] [44] [45] [46] [47] [48]

[49]

[50] [51]

[52] [53]

[54] [55]

[56]

Rajan SD. Sizing, Shape and Topology Design Optimization of Trusses Using Genetic Algorithm. The Structural Engineer 1995:1480-1487. Jenkins WM. A Space Condensation Heuristic for Combinatorial Optimization. In: Topping BHV, Papadrakakis M, editors. Advances in Structural Optimization, AI CIVIL-COMP 94; 1994; 1994. p. 215-224. Parmee IC, Bullock GN. Evolutionary Techniques and their Application to Engineering Design. In: Behesti MR, Zreik K, editors. Advanced Technologies; 1993: Elsevier Science; 1993. p. 32-42. Dasgupta D, McGregor D. A Structured Genetic Algorithm. Research Report I. Glasgow: University of Strathclyde, UK.; 1991. Report No.: IKBS-2-91. Parmee IC. Evolutionary Computing for Conceptual and Detailed Design; Genetic Algorithms and Computer Science, 1997.p 133-152. Parmee IC, Beck MA. An Evolutionary Agent Assisted Strategy for Conceptual Design Space Decomposition. In: al. Ce, editor. Evolutionary Computing: Selected Papers from the 1997 AISB International Workshop; 1997: Springer Verlag; 1997. p. 275-286. Parmee IC, Johnson M, Burt S. Techniques to Aid Global Search in Engineering Design. In: International Conference on Industrial and Engineering Applications of AI and Expert Systems (IEA94AIE); 1994 May31st - June 4th 1994; Austin, Texas.; 1994. Parmee IC, Vekeria HD. Co-operative Evolutionary Strategies for Single Component Design. In: 7th International Conference on Genetic Algorithm; 1997; 1997. p. 529-536. Parmee IC. The Maintenance of Search Diversity for Effective Design Space Decomposition using Cluster Oriented Genetic Algorithms (COGAs) and Multi-Agent Strategies (GAANT). In: Adaptive Computing in Engineering Design and Control, University of Plymouth, UK; 1996a; Plymouth, UK; 1996a. p. 128-138. Parmee IC. Cluster Oriented Genetic Algorithms (COGAs) for the identification of High Performance Regions of Design Spaces. In: EvCA 96; 1996b; Moscow; 1996b. p. 66-75. Grierson DE, Khajehpour S. Multi-Criteria Conceptual Design of Office Buildings Using Adaptive Search. In: 6th Workshop of the European Group of the Structural Engineering Applications of Artificial Intelligence (EGSEA-A) September; 1999 September; Warsaw; 1999. p. 51-74. Grierson DE, Park KW. Optimal Conceptual Topological Design. in Advances in Structural Optimization 1997. Rafiq MY, Bugmann G, Easterbrook DJ. Building Concept Generation Using Genetic Algorithms Integrated with Neural Networks. In: 6th Workshop of the European Group of the Structural Engineering Applications of Artificial Intelligence; 1999 September; Poland; 1999. p. 165-173. Rafiq MY, Mathews JD. An Integrated Approach to the Structural Design of Buildings Using a Structured Genetic Algorithm. International Journal of Integrated Design and Process Science 1998;2(3):20-31.

23


[57] [58]

[59]

[60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72]

Rafiq MY, Mathews JD, Bullock GN. Conceptual Building Design--Evolutionary Approach. Journal of Computing in Civil Engineering 2003;17(3):150-158. Mathews JD, Rafiq MY. Adaptive Search for Decision Support in the Preliminary Design of Structural System. In: Parmee I, editor. Adaptive Computing in Engineering Design and Control, 96, Proceedings of the first Int. Conf.; 1994; Plymouth Engineering Design Centre, University of Plymouth, Plymouth, UK; 1994. Mathews JD, Rafiq MY, Bullock GNB. A Prototype for a Conceptual Structural Building Design System Using Genetic Algorithms. In: Adaptive Computing in Engineering Design and Control, 96, 2nd Int. Conf.; 1996; Plymouth, UK: ED Parmee; 1996. p. pp 287-290. Miles JC, Moore CJ, Price G. Design costing models: An application of heuristic substitution. Computing Systems in Engineering 1995;6(6):521-531. Miles JC, Sisk GM, Moore CJ. The conceptual design of commercial buildings using a genetic algorithm. Computers & Structures 2001;79(17):1583-1592. Sisk GM, Miles JC, Moore CJ. Designer centred development of GA-based DSS for conceptual design of buildings. Journal of Computing in Civil Engineering 2003;17(3):159-166. French M. Conceptual Design for Engineers, (2nd edition). Design Council, London. 1985. Pahl G, and Beitz W. Engineering Design: A Systematic approach SpringerVeralag. 1988. Simon HA. The Sciences of the Artificial MIT Press. 1969. Smithers T. Exploration in design: Discussion. In Gero, JS and Tyugu N. (Eds). Formal Design Methods for Computer Aided Design. North-Holland, Amsterdam. 1993 Gero, JS. Towards a Model of Exploration in Computer-Aided Design. In Gero JS, and Tyugu N. (Eds). Formal Design Methods for Computer Aided Design. North-Holland, Amsterdam. 1993 Maher LM, Poon J and Boulanger S. Formalising Design Exploration as CoEvolution: A combined gene approach. Preprints of the Second W\IPIF WG5.2 Workshop on Formal Design Methods for CAD. 1995. Maher ML. A Model of Co-evolutionary Design. Engineering with Computers 2000;16:195-208. Koberg D and Bagnall J. The Universal Traveller-A Soft-Systems Guide to Creativity, Problem-Solving, and the Process of Reaching Goals. Morgan Kaufmann. 1976. Bentley PJ, Corn BW. Creative Evolutionary Systems: Morgan Kaufman; 2002. Purcell TA and Gero JS. Design and Other Types of Fixation. Design Studies 1996; 17(4): 363-383

24


[73] [74] [75] [76] [77] [78] [79] [80] [81]

[82] [83] [84]

[85] [86] [87] [88]

Parmee I. Towards Interactive Evolutionary Search and Exploration Systems. In: Bird-of-a-feather workshop: Genetic and Evolutionary Conference 2002; New York; 2002. Dawkins R. The Blind Watchmaker. New York: Norton; 1986. Banzhaf W. Interactive Evolution. In Bäck, T. The Handbook of Evolutionary Computation. Oxford: IOP Publishing and Oxford University Press; 1997. Buelow von P. Using Evolutionary Algorithms to Aid Designers of Architectural Structures Systems. Morgan Kaufman. In: Bentley, P. J., and Corn, B. W. (Eds) Creative Evolutionary; 2002. Eckert CM, Kelly I, Stacey MK. Interactive Generative Systems for Conceptual Design: An Empirical Perspective. Artificial Intelligence for Engineering Design, Analysis and Manufacturing 1999;13(4):303-320. Grierson DE, Khajehpour S. Method for conceptual design applied to office buildings. Journal of Computing in Civil Engineering 2002;16(2):83-103. Hart E, Ross P. GAVEL: A new tool for Genetic Algorithm Visualization. IEEE Transactions on Evolutionary Computation 2001;5(4):335-348. Collins TD. Visualizing Evolutionary Computation. In: In Ghosh, A., & Tsutsui, S. (Eds.), Advances in Evolutionary Computation: Springer Verlag; 2002. Smith T, Bird J, Bullock S. Beyond Fitness: Visualising Evolution Workshop Overview. In: The 8th International Conference on the Simulation and Synthesis of Living Systems; 2002; University of New South Wales, Sydney, Australia; 2002. Takagi H. Interactive Evolutionary Computation: Fusion of the Capabilities of EC Optimization and Human Evaluation. Proceedings of the IEEE 2001;89(9):1275-1296. Pham DT, Yang Y. A Genetic Algorithm Based Preliminary Design System. Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering 1993;207(2):127-133. Jo J. Interactive Evolutionary Design System: Process and Knowledge Representation. In: Gero JS, Maher ML, editors. Fourth International Roundtable Conference on Computational Models of Creative Design; 1998; Heron Island, Australia: University of Sydney; 1998. p. 215-224. Fonesca CM, Fleming PJ. Genetic algorithm for multiobjective optimization: Formulation, discussion and generalization. In: Fifth Int. Conf. on Genetic Algorithms; 1993; 1993. p. 416-423. Horn J. Multicriteria Decision Making and Evolutionary Computation, Handbook of Evolutionary Computation. Bristol, UK.: Bäck T., Fogel D.B. & Michalewicz Z. (eds.), Institute of Physics Publishing; 1997. Mathews JD. Optimisation and Decision Support during the Conceptual Stage of Building Design: New techniques based on the genetic algorithm [PhD Thesis]. Plymouth, UK: University of Plymouth; 2000. Rafiq MY, Mathews JD, Jgodzinski P. Interactive Role of the Human Computer Interaction in Design. In: 8th Workshop of the European Group of

25


[89] [90]

[91] [92]

[93]

[94] [95] [96]

[97]

[98] [99]

the Structural Engineering Applications of Artificial Intelligence; 2001; Loughborough, UK; 2001. p. 31-42. Rafiq MY, Mathews JD, Bullock GN. Conceptual Building Design--Evolutionary Approach. Journal of Computing in Civil Engineering 2003;17(3):150-158. Parmee IC, Bonham CR. Supporting Innovative and Creative Design using Interactive Designer / Evolutionary Computing Strategies. In: Gero JS, Maher ML, editors. Fourth International Round-table Conference on Computational Models of Creative Designeds; 1998 December 1998; Heron Island, Australia: University of Sydney; 1998. p. 187-214. Parmee IC, Cvetkovic D, Watson AH, Bonham CR. Multi-objective Optimisation and Conceptual Engineering Design. Journal of Evolutionary Computatio 2000;8(2):197-222. Abraham JAR, Parmee IC. User-centric Evolutionary Design Systems - the Visualisation of Emerging Multi-Objective Design Information. In: Xth International Conference on Computing in Civil and Building Engineering; 2004 June 02-04; Weimar, Germany: American Society of Civil Engineer; 2004. Parmee IC, Abraham JAR. Supporting Implicit Learning via the Visualisation of COGA Multi-objective Data. In: 2004 IEEE Congress on Evolutionary Computation; 2004; Portland, Oregon, USA: IEEE Press; 2004. p. 395-402. Inselberg A, Dimsdale B. Multi-dimensional Lines I: Representation & Multi-dimensional Lines II: Proximity and Applications. SIAM J. Appl. Math. 1994;54(2):559-596. Packham ISJ. An Interactive Visualisation System for Engineering Design using Evolutionary Computing. Plymouth, UK: University of Plymouth.; 2003. Packham ISJ, Denham SL. Visualisation Methods for Supporting the Exploration of High Dimensional Problem Spaces in Engineering Design. In: Roberts JC, editor. IEEE International Conference on Coordinated & Multiple Views in Exploratory Visualization (CMV2003); 2003 15 July 2003, IEEE Computer Society, pp; London, UK: IEEE; 2003. p. 2-13. Packham ISJ, Rafiq MY, Borthwick MF, Denham SL. Interactive Visualisation for Decision Support and Evaluation of Robustness using Evolutionary Computing. In: 11th International Workshop of the European Group for Intelligent Computing in Engineering (EG-ICE); 2004 31st May 1st June 2004; Weimar, Germany; 2004. Rafiq MY, Packham ISJ, Easterbrook DJ, Denham SL. Visualizing Search and Solution Spaces in the Optimum Design of Biaxial Columns. To appear in the Journal of Computing in Civil Engineering, 2005. Parmee IC, Cvetković D, Watson AH, Bonham CR. Multiobjective Satisfaction within an Interactive Evolutionary Design Environment. Evolutionary Computation, 2000;8(2):197-222.

26


[100] [101] [102] [103] [104]

Spence R. Information Visualization. Harlow, England: Addison-Wesley; 2001. Spence, R. The Acquisition of Insight. http://www.ee.ic.ac.uk/research/information/www/Bobs.html. 2003 Swayne DF, Cook D, Buja A, Lang DT. ggobi Manual. In; 2001. Tweedie L, Spence R, Dawkes H, Su H. Externalizing Abstract Mathematical Models. In: CHI96, ACM; 1996; 1996. p. 406-412. Shneiderman B. Designing the User Interface. Reading, Massachusetts: Addison-Wesley; 1998.

27