Artificial Neural Networks for Electric Machine Efficiency and Flux Map Computation

Abstract

This work discusses the utilization of artificial neural networks for interior permanent magnet motor efficiency and flux map estimation. Efficiency maps for electric vehicle motors, which provide the efficiency at different operating points require a large number of transient FEA computations, often with fine time steps and meshes so that losses are calculated accurately. Flux maps provide the variation of motor torque, d and q-axes flux at different operating points and also require a large number of computations. In some cases, 3D simulations may also be required, especially for magnet loss and voltage computations. A large number of such calculations is time consuming and computationally intensive. This work discusses the use of artificial neural networks for efficiency and flux map estimations for a 380 kW motor designed for heavy duty vehicles. The neural network predictions are compared with FEA calculations. The efficiency maps are also interpolated using conventional techniques such as nearest neighbor and spline interpolation. The errors in all cases are compared.