Reducing computational effort in field optimisation problems

Design and optimisation of many practical electromechanical devices involve intensive field simulation studies and repetitive usage of time‐consuming software such as finite elements (FEs), finite differences of boundary elements. This is a costly, but unavoidable process and thus a lot of research is currently directed towards finding ways by which the number of necessary function calls could be reduced. New algorithms are being proposed based either on stochastic or deterministic techniques where a compromise is achieved between accuracy and speed of computation. Four different approaches appear to be particularly promising and are summarised in this review paper. The first uses a deterministic algorithm, known as minimal function calls approach, introduces online learning and dynamic weighting. The second technique introduced as ES/DE/MQ – as it combines evolution strategy, differential evolution and multiquadrics interpolation – offers all the advantages of a stochastic method, but with much reduced number of function calls. The third recent method uses neuro‐fuzzy modelling and leads to even further economy of computation, although with slightly reduced accuracy of computation. Finally, a combined FE/neural network approach offers a novel approach to optimisation if a conventional magnetic circuit model could also be used.