Development of Multidisciplinary Design Optimization Algorithms for Ship Design Under Uncertainty.

Abstract

This dissertation presents the development and application of two new multidisciplinary design optimization (MDO) algorithms. The first MDO algorithm is a multi-level algorithm which computes target values for the design variables and uses the target values for driving the system level process. The second MDO algorithm is inspired by the principles of set-based design.

The multi-level MDO algorithm was implemented with techniques for optimization under uncertainty. The multi-level MDO algorithm under uncertainty was applied to a conceptual ship design problem and it was demonstrated that variability in the objective function was reduced and the probability of violating constraints at the optimum was reduced, while improving the discipline objective functions. The multilevel MDO algorithm was also implemented using surrogate models in place of the discipline analyses for a ship design problem. The optimization with surrogate models achieved a computational time savings with accurate predictions when compared to the true solvers.

Set-based design is a design approach where sets of feasible values for the design variables from different disciplines are determined and shared, with the goal of locating and working with the areas of feasible overlap. An MDO algorithm was developed with the core concept of describing the design using sets to incorporate features of set-based design and achieve greater flexibility than with a single-point optimization. The new algorithm was applied to a ship design problem and ship design application demonstrated the value of utilizing set-based design as a space-reducing technique before approaching the problem with a point-based optimization. Furthermore, incorporating flexibility in the constraints allowed the optimization to handle a problem with very strict constraints in a rational manner and minimize the necessary constraint violation.