Visual Interpretation of Topology Optimization Results Based on Deep Learning

In this study, an approach was proposed to interpret the shapes of electric motors obtained by topology optimization (TO). A convolutional neural network (CNN) model was trained on motor images and used to predict the degree of influence of local structures of motors on their characteristics. This prediction was obtained as heatmaps for visualizing the importance of each local structure. The proposed method was applied to a permanent magnet motor model. The proposed method facilitated successful visualization of the degree of influence on average torque, torque ripple, and mechanical stress. This was validated by comparing the obtained predictions with finite element analysis results. In addition, the proposed method allowed us to improve the motor characteristics through visualization. Therefore, this approach can improve the mechanical stress resistance while maintaining the torque characteristics.