Adaptation of Evolutionary Search in Topology and Decomposition Based Design Optimization


The study focuses on the use of genetic algorithms in complex and coupled multi-disciplinary design problems. Two special classes of problems are identified for study; these include topology design problems and situations in which a large coupled problem must be solved as a sequence of smaller subproblems. The latter is referred to as a decomposition based design strategy. Binary coded genetic algorithms have been used effectively in the topological design of discrete structural systems. In a majority of such applications, the structural topology is extracted from a pre-defined "structural universe", a set of all permissible joints and elements that can be used in the development of the optimal design. In the presence of a dense "structural universe", the genetic algorithm search process must contend with very long string lengths, with attendant degradation in the effectiveness of the search process. The present study presents a novel approach for handling binary strings that may have variable string lengths and/or varying binary string representations of design variables. Varying string lengths in a population and topological difference of string representation requires a redefinition of the crossover process, and both inter- and intra-species crossover mechanisms are explored in this study. The use of micro genetic algorithms is proposed as an approach to increase the search efficiency in problems involving a large number of candidate topologies. The proposed strategies are implemented in representative algebraic problems, truss topology design, and the layout of a stiffened composite panel. New strategies for the implementation of genetic algorithms in a decomposition based design optimization were examined in this study. In decomposition based approaches, the design problem is decomposed into smaller-sized sub problems, the solutions for which are obtained through co-evolution. The emphasis resides in evaluating methods for exchanging design information relevant to coordinating solutions of temporarily decoupled sub problems. Methods based on modification of genetic makeup through experiential inheritance (exposure to another species), and through inter species migration are investigated in this work. Different forms of design problem coupling are investigated, ranging from coupling through constraints only, to coupled objective and constraint functions. The proposed strategies are validated through implementation in representative algebraic and structural design problems. The strategies for decomposition based design optimization were extended to multicriteria design optimization environment. Among many methods to address multicriteria optimization problems, the weighted sum and the weighted minimax method were used to investigate that experiential inheritance and interspecies migration is applicable to multicriteria optimization problems. The proposed strategies are tested in representative algebraic problems, a truss design, and a stiffened composite panel design.