Robust Covariance Matrix Adaptation Evolution Strategy: Optimal Design of Magnetic Devices Considering Material Variation

Uncertainties caused by material variation can significantly impair the characteristics of devices. Therefore, it is important to design devices whose performance is not significantly damaged even when material variations occur. Robust optimization seeks for the optimal solutions that are robust to fluctuations due to uncertainties caused by material variation, geometrical variation due to assembly tolerances, and changes in physical properties over time in real-world problems. However, naive robust optimization requires iterative calculations to compute the expected values, which need a huge computational burden. This paper introduces a novel robust optimization method for magnetic devices using the covariance matrix adaptation evolution strategy (CMA-ES). In this method, called RCMA-ES (robust CMA-ES), the expected value of the objective function is evaluated using the local average of neighboring individuals without increasing the computation cost. For validation, RCM-ES and robust genetic algorithm (RGA), one of the robust optimization methods without increasing the computational load, was applied to the topology optimization of a magnetic shield and actuator, considering the uncertainty in the BH characteristics. RCMES was demonstrated to be particularly more effective for topology optimization with a large number of dimensions compared to RGA and provides robust optimal shapes that are insensitive to variations in BH characteristics.