Worst case tolerance design of magnetic devices by evolutionary algorithms

This paper presents an evolutionary approach to worst case tolerance design of magnetic devices. The algorithm improves on the classical nominal design, accounting for parameter variations and tolerances, so that the system performance does not exceed upper and lower specifications imposed in advance by the designer. The evolutionary algorithm adjusts central values and tolerances of parameters so that a user-defined objective function is optimized and the design constraints are met. The feasibility of the candidate sets of design parameters and tolerances is estimated by means of vertex analysis. The algorithm ensures that the optimized design set of parameters and tolerances is 100% feasible in a large number of practical cases. This paper gives an example of the design of a magnetic device and discusses the salient features and limitations of the approach.