Prediction method and analysis of cogging torque based on AdaBoost-BRNN

Owing to the complex and changeable problems of motor performance optimization and the highly nonlinear relationship between the structural parameters of different parts, the traditional finite element analysis method is slow, and the analytical method is complex and has many limitations. Thus, it is difficult to determine different variables for different indicators to achieve a quick response. Therefore, taking cogging torque as the goal of motor performance analysis, a cogging torque prediction and analysis method based on the AdaBoost integrated Bayesian Regularization neural network (BRNN) learning algorithm is proposed. Artificial NN, a new data processing technology, can effectively analyze the complex relationship between variables and is widely employed in engineering technology. At present, A 4-pole 24-slot surface-mount permanent magnet synchronous motor simulation model is designed in Ansys Maxwell to get the test data. According to the characteristics of cogging torque, a cogging torque analysis model with pole embrace, stator slot width, and pole arc offset as inputs and cogging torque as output, is established. Data are collected by segmenting the level factors, and the Bayesian regularization method is used for model training to improve the NN for generalization ability. Finally, a strong predictor model is obtained using AdaBoost integrated learning. The test set obtained by dividing the experimental data is substituted into the predictor, and the results are compared with the actual experimental results. In addition, the robustness of the model is judged according to the influence of its parameters on the prediction results for the experiments. The experimental results show that the BRNN has good generalization ability, and the error of the strong predictor based on AdaBoost is smaller than the average error of the weak predictors at each level and has good fitting ability compared with the actual experimental data. The results confirm the applicability of the above method to the motor performance optimization process for cogging torque prediction analysis.