Neural Network Approach for E-Motor Development

In this study, the development of electric motor design optimization methods and algorithms for electric vehicles, which have become widespread as a result of energy policies, is discussed. The rapidly increasing need for micro transportation within the scope of small cities has increased the interest in short-range transportation vehicles such as electric bicycles and electric scooters. Therefore, an electric scooter model is considered and the desired motor requirements are determined by analyzing its dynamic model. Then, IPM topologies are investigated and the appropriate topology is decided. IPM design parameters are dealt with in the ANSYS RMXprt environment, and all design combinations by selecting the appropriate test matrix in Taguchi's experiment design method are modeled in ANSYS RMXprt and logged in the appropriate file format together with the obtained results. The motor design models of all experiments are saved as the. png format in the aspect format to be determined. Then, the labeled pictures with the obtained results in the experimental design are trained in MATLAB on a neural network model with appropriate input and output. Thereafter, the trained neural network derives the appropriate motor geometry in terms of the design requirements. The derived motor geometry is converted into a 2D technical drawing format with the help of a package program (Img2CAD) and uploaded to the ANSYS Maxwell environment. To assess the motor performance are performed in ANSYS Maxwell. The proposed methodology shows that the results of parameter estimation and geometry generation in solution space with the trained neural network give sufficient performance.