The design of products and processes has gradually shifted from a purely physical process towards a process that heavily relies on computer simulations (virtual prototyping). To optimize this virtual design process in terms of speed and final product quality, statistical methods and mathematical optimization techniques are widely used nowadays. The main contributions of this thesis are in two areas. The first is the area of gradient estimation for optimization problems for which a single simulation can be carried out quickly. For these optimization problems classical optimization methods can be used. The quality of the gradient estimations is important for the success of these methods. We propose and analyze different (new) gradient estimation schemes. The second is the area of optimization methods for optimization problems with time consuming simulation runs. We propose a new method based on local approximation functions and a new geometry and trust region concept. We discuss simulation-based optimization methods for integer valued problems, and conclude with an application in the design of heat sinks.
|Qualification||Doctor of Philosophy|
|Award date||28 Apr 2006|
|Place of Publication||Tilburg|
|Publication status||Published - 2006|