Examples of Interactive and Visualisation Systems for Engineering Design
A comprehensive review by (Takagi 2001) lists a number of approaches to and applications of the IEC.
One of the first preliminary design systems devised by Pham and Yang (1993) allowed the user to view and evaluate solutions produced by the GA. Jo (1998) discovered that adding human interaction to his evolutionary design system allows domain knowledge to be incorporated online; solutions can be independently visualized in a space layout problem and the user was able to modify individual elements of the design. The interaction of a user has also been considered in a multiobjective environment: Fonesca and Fleming (1993) proposed a decision-maker (DM) that controls which objectives have more importance within a non-dominated set of solutions. They suggested the DM could be a human or an expert system. Horn (1997) pointed 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 and Rafiq (Mathews, 2000; Rafiq et al., 2003; Rafiq et al., 2001) developed a Conceptual Building Design (CBD) system using the Structured GA. CBD 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 Figure 1, 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 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. This is also useful if the design brief requires a particular construction material to be used or a designer/client prefers a particular floor system, etc. to be considered.
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. For example by clicking the mouse in a point on the graph, the corresponding details of the concept is shown in the 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.
The system also allows pausing the GA search at any time and changing either the GA parameters (e. g. mutation and crossover probabilities, etc.) or directing the search to specific branches of the design hierarchy.
Fig. 1. Dialogue box representing the design hierarchy.
Fig. 2. Tracing design Evolution during a GA run.
Grierson and Khajehpour (2002) used Pareto optimisation techniques for multi-criteria conceptual design, involving genetic-based stochastic search and colour-filtered graphics. In this investigation, a large set of Pareto non-dominated designs were captured. They then 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 cost – revenue conceptual design of high-rise office buildings, including several examples was presented. The work of Grierson and Khajehpour is an excellent example of knowledge discovery using a simple visualisation tool, and it demonstrates that by using visualisation tools effectively, it possible to discover the interrelationships between design parameters more readily.
Research in Plymouth Engineering Design Centre by Parmee and Bonham (1998) 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 (Parmee et al., 2000). 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 (Abraham and Parmee, 2004; Parmee and Abraham, 2004), that uses the parallel coordinate technique developed by Inselberg and Dimsdale (1994) to visualize many variables at once and compare the distribution of solutions between objectives using statistical analysis and discover overlap between various objectives.
A major objective of the IVCGA, developed by Packham and colleagues (Packam, 2003; Packham and Denham, 2003; Packham et al., 2004; Rafiq et al., 2004), is to assist the designer through interaction to explore a range of feasible and innovative solution that best fit the design brief requirements. This 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.
It is argued that such interaction is fundamental in real life design problems which are multidimensional, multi-criteria 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. This can lead to discovery of new knowledge about the specific regions of the search and solution spaces and more importantly understanding of collaborative design issues.
To this end, in the IVCGA the focus is on user interaction right from the start, allowing the user to freely interact with specific regions of the search space. Relating to the categories of visualisation and interaction discussed earlier, the IVCGA exhibits a number of unique features:
Firstly, in terms of visualisation the user is able to choose a number of high dimensional visualisation techniques which can help in understanding the relationship between design variables (i. e. 2D and 3D scatterplots, scatterplot matrix and parallel coordinates), additionally the data can be displayed in alternative coordinate systems, (i. e. ‘Principal Components’, ‘Independent Components’).
Secondly, a novel clustering technique based on kernel density estimation identifies the clusters in terms of the design variables. Clustering can be performed in alternative coordinate systems such as the principal components that reveal the ‘natural’ clusters in the data. Colour is used to identify different clusters.
Thirdly, in terms of interaction the user is in full control of the process and is either able to ‘zoom in’ in specific regions of the search space, i. e. on those regions containing specific colour coded clusters, enabling a more concentrated and focused search ‘inside’ such regions, or to extend the search ‘outside’ these regions to explore other possibilities which has not been currently discovered by the GA. It is both these cases which allow human experts to use their domain knowledge to guide the search to those areas of the search space which may meet differing, and possibly changing, design requirements. The nature of the IVCGA interaction places it inside the broader definition of IEC.