Modelling and design optimization of switched reluctance motors based on machine learning methods: Review

Accurate modelling of Switched Reluctance Motors (SRM) is a complex and time consuming task that involves a lot of tradeoffs and assumptions. Numerical and analytical modelling techniques have been extensively used in literature. Due to recent advancements in artificial intelligence and machine learning (ML) methods, there is need to explore nonlinear modelling of SRMs using machine learning surrogate models. This paper presents a comprehensive review of machine learning techniques of optimizing Switched Reluctance Motor (SRM) models. Several machine learning models have been developed to optimize SRM parameters such as stator pole arc, rotor pole arc, stack length, turns per pole, and phase current. The SRM parameter values have been obtained from Finite Element Analysis data of an 8/6, 600 V machine. The design parameters are optimized based on weighted combination objective function. The objective is to minimize torque ripples while maximizing efficiency. The ML techniques considered in this work include the Bayesian algorithm, Feed Forward Neural Network (FNN), Back Propagation Neural Network (BPNN), Artificial Neural Network (ANN), Radial Basis Function Neural Network (RBFNN), Sequential Least Squares Programming (SLSP) and Gaussian Process Regressor (GPR). The BPNN is trained with various techniques such as Levenberg Marquardt (LM), Differential Evolution (DE), Gradient Descent Algorithm (GDA), Quasi-Newton algorithm, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Best parameters, percentage torque ripple reduction and percentage efficiency improvement are then used to evaluate the performance of the machine learning optimization models.