DC Motor Benchmark with Prediction Based on Mixture of Experts

Abstract

The Mixture of Experts (MoE)–based approach is applied to verify the possibility of using surrogate models for searching the optima of complex multicriteria problems with constraints. This approach can successfully solve problems when the design space is limited by a higher number of constraints and traditional methods of Design of Experiments (DoE) in conjunction with one surrogate model are not able to partition the design space acceptably enough for further prediction. The methodology is tested on a well-known DC motor benchmark, where the electromagnetic and temperature fields were solved analytically, in a simplified form.