Quality Assisted Multiobjective and Multidisciplinary Genetic Algorithms


Real-world engineering design optimization problems often involve multiple incommensurable and conflicting design objectives. They are usually constrained, and have a mix of continuous and discrete design variables. Two important goals for such problems are the generation of Pareto optimum solutions and the quality assessment (or "goodness") of such solutions. This dissertation presents an approach for generation of Pareto solutions and metrics for quality assessment of such solutions for multiobjective and multidisciplinary design optimization problems. In the approach for the generation of Pareto solutions, a new constraint handling technique is developed that works with a Multi-Objective Genetic Algorithm (MOGA) to solve constrained multiobjective optimization problems. This constraint handling technique uses a primary-secondary fitness assignment scheme to account for both individuals' (or designs') performance and their "matching" when the individuals are selected and paired as parents. To assess the goodness of solutions obtained from such a technique, several new quality metrics are introduced. These metrics include hyperarea difference, inferiority index, overall Pareto spread, kth objective Pareto spread, accuracy of the observed Pareto frontier, number of distinct choices and cluster. These metrics could be used to: (i) compare different Pareto solution sets obtained from different MOGAs, and/or (ii) compare different MOGAs against one another. Finally, a two-level MOGA is developed to handle multiobjective and multidisciplinary design optimization problems. A two-level quality assisted optimization scheme is also developed for "optimization for quality". This "optimization for quality" scheme can be used to optimize a set of quality metrics for the Pareto solutions in addition to optimizing the design objectives. Comparison studies are presented to demonstrate the proposed constraint handling technique with a MOGA, the two-level quality assisted MOGA, and the two-level "optimization for quality" scheme.