A study on torque modelling of switched reluctance motors

In this paper we develop and verify the suitability of two torque models of the switched reluctance motor (SRM). The first torque model is constructed analytically in terms of the well known flux saturation characteristics of the SRM. The torque modeling problem renders to an optimization process: minimizing the discrepancy between the model estimated torque and measured torque by means of tuning 18 coefficients in the torque model. Both statistic search - Genetic Algorithm (GA), and deterministic search - Levenberg-Marquardt (LM) gradient expansion method, are employed to search the optimal solution. Through comparative study, we show that the combination of the two: GA searches the neighborhood of the global minimum and LM refines, gives the best results. The second torque model is constructed using artificial neural network (ANN), which provides a model-free black-box approach. While the simulation results show the effectiveness of both models, the experimental results indicate that the analytic model using domain knowledge outperforms the ANN model.