Evolutionary and Adaptive Computing in Engineering Design

The utility of the established evolutionary and adaptive search algorithms of the previous chapters, although well-proven on mathematical test functions and upon specific, well-defined real-world problems, can be significantly reduced when they are integrated with engineering design processes. This is largely due to the complexity of the problems encountered, uncertainties relating to initial problem definition and human-centred aspects relating to design preferences, experiential knowledge etc. Many techniques have been developed to handle either constraints, multi-criterion, high dimension and modality, discontinuity or mixed discrete/continuous variable descriptions within an evolutionary/adaptive computing framework. However, when faced with a design problem that contains most, if not all of these characteristics then appropriate strategies specific to the problem at hand must be developed.

The chapter investigates the potential of evolutionary and adaptive search for the decomposition of complex multi-dimensional design spaces into succinct regions of high-performance solutions. Many evolutionary search techniques that attempt to identify and maintain multiple local optima within a multi-modal fitness landscape have been developed within the EC research community and are described in the literature; a selection of these are outlined in the following section. The perceived utility of such techniques when applied in the design domain is that they can return a number of high-performance solutions to the engineer for further off-line evaluation. Such techniques can be of considerable value when applied to specific well-defined problems where some a priori knowledge relating to the nature of the fitness landscape is available.

During the early stages of engineering system design the engineer is generally faced with multiple discrete choices relating to the major elements which define the overall structure of the system. However, in many cases the system cannot be adequately described in terms of discrete decisions alone. Continuous variables that to some extent describe the characteristics of each discrete design configuration need to be included to achieve a meaningful definition. This creates a search/optimisation problem of considerable complexity as each set of continuous variables will be dependent upon the particular configuration that they describe and the constituent variables of these sets may differ from one configuration to another as illustrated in the simple hierarchy of Figure 7.1. The result is a set of differing, dependent, continuous design spaces each of which describes a particular discrete design option/configuration. In this high-level whole-system design (WSD) situation an efficient search strategy is required to provides a multi-level search capability that can negotiate the discrete hierarchy, efficiently sampling the different continuous design spaces in order to identify those configurations that offer best potential. If such a strategy can be developed it would be possible to explore a far greater number of whole-system design solutions than would be possible from a more traditional heuristic approach. The result would be a high-level, decision support tool that may reduce lead times during conceptual/preliminary stages of design whilst allowing a more extensive search of the available design alternatives within budget and time constraints. This will result in the identification of competitive system configurations that may have been overlooked during the problem decomposition processes of heuristic design. The developed strategies would enable the engineer to rapidly survey the potential of diverse regions of the hierarchy within time constraints thereby avoiding a compromise of search space potential by rapidly returning to known regions and prematurely concentrating search effort in a small subset of the complex solution space.