MLP-based Torque and Speed Prediction in BLDC and SRM

Motor performance prediction is necessary in order to increase efficiency, reliability and effectiveness of electric drive systems. Since of their efficiency and low maintenance requirements, brushless DC (BLDC) motors and switched reluctance motors (SRMs) are used in the electric vehicle industry, robotics, automation, aerospace engineering and renewable energy systems. Existing motor modelling and simulation approaches require accurate parameters and are computationally demanding; thus, they cannot be used for monitoring and control purposes in real time. In this study, we present the intelligent multi-layer perceptron (MLP) regressor for estimating the torque and speed of BLDC and SRM motors under different operating conditions. Hyper parameters optimization increases prediction accuracy and generalization ability of the proposed MLP approach. From the experiments performed, it is clear that the MLP Regressor performs test MSE values of 0.0020 and 0.0050 in BLDC torque and speed respectively while performing 0.0028 and 0.0061 MSE values in the SRM torque and speed respectively. In comparison to the Linear Regression approach, the proposed MLP decreases the torque MSE value of the BLDC by 60.0% and the SRM by 54.8%, while decreasing the speed MSE value of the BLDC by 47.4% and the SRM by 47.0%. The training time of the MLP Regressor is 15.4s and 18.3s for the BLDC and SRM, respectively. From the obtained results, it is clear that the proposed MLP framework gives more accurate predictions than the Linear Regression and Physical Modeling frameworks especially for the nonlinear nature of the SRM motors.