Keywords: soft computing, genetic algorithms, neural networks, cellular automata, particle swarm optimization.

The focus of this review paper is on soft computing methods that more closely model the working of the human mind, particularly its ability to show greater tolerance for imprecision, uncertainty, and approximations in obtaining high-quality engineering solutions. The paper provides an overview of the adaptation of neural networks and fuzzy-logic based tools for modeling, representing imprecise information in a formal design process, and discovery of causal relationships in design data. A survey of the literature shows the use of neural networks in the task of generating surrogate models for use in structural optimization. This is similar to deploying a response surface as a surrogate model for structural analysis, using limited exemplar sets in the training process. The use of fuzzy logic based approximations in analysis has been more limited, and fuzzy set theory has found a greater application in the optimization strategies and algorithms.

The paper also provides an assessment of the state-of-the-art developments in optimization tools for generically difficult optimization problems - genetic algorithms, particle swarm optimization methods, and immune network modeling is reviewed in this context. The focal areas where these heuristic optimization tools have most benefits are the generically difficult problems of optimization involving discontinuous or nonconvex design spaces. The past decade has seen significant activity in enhancing the efficiency of these methods, and many variants of the plain genetic algorithm or the simplistic particle swarm optimization have been proposed. The focus has resided in adding capability for solving problems of practical import, including efficient ways of accounting for design constraints in the problem solution. It is clear from the literature that these challenges will continue to drive new innovations in these search algorithms. The paper also examines the role of soft computing methods in analysis and design algorithms developed for distributed computing environments, most specifically, the decentralized computing approach referred to as the cellular automata. This computational paradigm has not only been used to develop computationally efficient structural analysis tools but also provides a framework for deploying spatially distributed versions of soft computing based optimization algorithms. A coupling of this optimization and analysis provides opportunities for implementing simultaneous analysis and design approaches for structural optimization.

New frontiers of research in soft computing will most likely reside in developing versions that are naturally amenable to a massively parallel computing environment, and to adapting these methods in formulations that overcome the high computational costs of analysis in the design optimization process. Increasing applications of soft computing in model formulation and discovery, and using the inherent flexibility of modeling to include uncertainties in the design process, are also inevitable goals for application of these methods.

purchase the full-text of this chapter (price £20)

go to the previous chapter
go to the next chapter
return to the table of contents
return to the book description
purchase this book (price £88 +P&P)