Fast and Accurate Design of BLDC Motors Using Bayesian Neural Networks

Abstract

Brushless direct current (BLDC) motors are gaining popularity over traditional direct current (DC) motors due to their higher efficiency, compact size, and precise control capabilities. This study proposes a fast and accurate approach to BLDC motor design using a Bayesian neural network (BNN). The BNN, a specialized form of the multi-layer perceptron (MLP), offers strong resistance to overfitting and performs effectively with noisy or limited datasets, making it well-suited for complex motor design problems. In the proposed method, the BNN is applied within an inverse modeling framework to map desired motor performance parameters to the corresponding design variables. A dataset for an outer-rotor BLDC motor—containing both design parameters and the resulting output torque—is generated through finite element analysis (FEA). Finally, a demonstration of BLDC motor design using the BNN validates the effectiveness of the proposed approach.