Efficient robust systems design through the use of hybrid optimization and distributed computing

The optimization of many realistic large-scale engineering systems can be computationally expensive. The evaluation of a single design configuration can take minutes or hours, and although computing power is steadily increasing, the complexity of the analysis codes continues to keep pace. In this dissertation, a new method to utilize parallel processing and hybrid optimization methods to allow for rapid solution to these complex problems in introduced. The hybrid algorithm switches back and forth between global and local optimization algorithms based on information about the topography of the local design space. Local design space information is gathered in real-time by performing regression and statistical analysis of the current population of a Genetic Algorithm. This information is then used to decide when it is beneficial to execute a local search algorithm. The efficiency of the hybrid optimization approach is further increased by execution in a distributed computing environment. The design space is intelligently partitioned and the hybrid optimizer is run in each of these subspaces. In this way multiple local optima can be identified simultaneously. In addition, uncertainty, which is common in engineering design problems, is handled through a Monte Carlo-based robust design procedure. To demonstrate the usefulness of this approach, results are presented from four case studies, including multimodal benchmarking problems and complex engineering design problems.