Mixed variable optimization methods for complex engineering system design

During the last decade, MDO (Multidisciplinary Design Optimization) has emerged as a new engineering discipline. MDO can be described as a methodology for solving large complex engineering system design problems. A complex system is made up of several distinct disciplines or subsystems. These disciplines interact with each other during the design process. The GSE (Global Sensitivity Equation) based CSSO (Concurrent Subspace Optimization) method, one of the most popular MDO methods, has shown promise of solving complex design problems. However most research to date has concentrated primarily on continuous design variables only.; The purpose of this work is to develop methods appropriate for the realistic mixed variable problems that are encountered in MDO problems. For coupled engineering systems, the mixed variable subsystems can be essentially divided into two categories depending on the existence or absence of meaningful sensitivities. One of these categories corresponds to all subsystems for which the derivatives of the system responses with respect to discrete design variables exist. The other corresponds to those subsystems for which the derivatives do not exist or have no physical meaning. For the former type problems, a modified SLP (Sequential Linear Programming) is developed and a suitable mixed-integer LP solver can then be used for this subspace optimization. The modified CSSO proposed in this dissertation has been demonstrated to successfully solve several complex system design problems. For the latter case, the mixed system is further decomposed into pure discrete and pure continuous subsystems. A different modified CSSO method based on the decomposition scheme is used to solve a composite plate design problem. This method is proved to be useful but could be stuck at local minimum. A diversification scheme is necessary to prevent this from happening. An attempt of using Artificial Neural Network (ANN) in place of the discrete subsystem analysis was made. Finally, a DOE (Design of Experiments) method using Taguchi Orthogonal Arrays was developed and was proved effective to solve structural optimization problems with material selection.