Inter-GEN: A hybrid approach to engineering design optimization

A large portion of engineering design optimization involves the time-consuming process to iteratively running simulation codes to predict the performance of a design. Engineers attempt to modify design parameters to find the best design that satisfies design requirements. Although many automated design optimization techniques have been developed, no single technique has emerged to efficiently search the typically high-dimensional and highly nonlinear parameter spaces for the best design.;An optimization approach based on the interdigitation of three optimization technologies is presented. The approach, called Inter-GEN, is an interdigitation of genetic algorithms, expert systems, and numerical optimization. Inter-GEN uses a control strategy to couple and switch among optimization technologies to get the greatest optimization improvement in as few runs of the simulation code as possible. There are two search phases in the Inter-GEN approach: knowledge-directed and knowledge-supplementation. Knowledge-directed search uses expert system technology tightly coupled with numerical optimization techniques to exploit the engineer's knowledge of the physics of the parameter space. The expert system specifics regions of the parameter space within which numerical optimization can focus. When the knowledge-directed search phase no longer provides improvement, a knowledge-supplementation search phase is entered. The knowledge-supplementation search phase supplements the engineer's incomplete knowledge of the parameter space by coupling genetic algorithms and numerical optimization to explore and exploit the parameter space efficiently. The genetic algorithm identifies various "hills" within the parameter space and provides a non-gradient-based method of avoiding constraint boundaries and local optima. Numerical optimization enhances the genetic algorithm's ability to exploit unimodal regions of the parameter space (i.e., it focuses on getting the best "local" solution).;Inter-GEN has been implemented and tested using a single control strategy on a diverse set of eleven test cases: six engineering problems which have proven difficult to numerical optimization techniques, four 3-dimensional problems whose surface is completely specified, and the preliminary design of a GE 10-stage aircraft engine turbine. The test results show Inter-GEN to be a more efficient and more robust optimization approach than either numerical optimization or genetic algorithms used in isolation.