Improved pattern search algorithm using an objective function effect based move schedule for three-dimensional component layout

Layout of components inside mechanical and electro-mechanical products is a critical aspect of engineering product development. The automated 3D component layout framework developed at the Computational Design Lab at Carnegie Mellon was an important step towards the automation of product layout. While Pattern Search algorithms have been successfully used in the 3D component layout framework, and are the current state of the art for automated layout, many questions related to tuning its parameters and performance remain unanswered.; The focus of this dissertation is to identify these questions and answer them using domain knowledge and information about the 3D component layout search space. The outcome of such an exercise is a better understanding and improved performance of the Pattern Search algorithm for 3D component layout. One important question identified is the scheduling of patterns in the search. To answer this question the dissertation presents a new method that schedules patterns in decreasing order of their effect on the objective function resulting in an improved Pattern Search algorithm. The new schedule improves performance of Pattern Search algorithms on 3D component layout problems and on minimization of general 2D multimodal surfaces. The dissertation also explores various schedules for decreasing step sizes of patterns in Pattern Search and demonstrates the robustness of the geometric schedule.; Two new applications of the 3D component layout framework developed in this dissertation further advance the use of Pattern Search algorithms in engineering optimization problems.