Kriging methods for constrained multi-objective electromagnetic design optimization

Design problems in electrical engineering are typically solved using a computationally expensive numerical method, such as the finite element method. As the designs of most interest are those which are optimal in some way, any algorithm used to search for such designs must perform well in a low number of iterations, if they are to be identified within a reasonable time. Kriging is a method of making predictions (based on a set of observations) which has been used as a surrogate model to reduce computational cost in optimization searches. After reviewing the state-of-the-art in kriging-assisted single and multi-objective opti- mization, this thesis proposes several novel algorithms for efficiently solving constrained (or unconstrained) multi-objective optimization problems. Using a combination of these novel algorithms and existing methods, a practical optimization tool is integrated into commercial electromagnetic design software (Opera). The tool is demonstrated on four different optimal design problems, which cover the four possibilities of un- constrained and constrained single and multi-objective optimization. Some areas of potential future work are given in the final chapter.