Machine learning models for the structural design of thrust active magnetic bearing and its optimization

Abstract This paper characterizes the design and structural optimization of a thrust active magnetic bearing (TAMB) that utilizes a thrust active magnet to control the rotor’s position and provide the necessary force. The magnet is designed using magnetostatic finite element analysis (FEA) in Ansys Workbench, with mesh convergence studied. Parameters that heavily influence the performance are then identified. The datasets consisting of influencing parameters and the output are generated using FEA. Two ANN models: Levenberg-Marquardt and Bayesian Regularization are trained with the help of generated datasets. Finally, TAMB is optimized by reducing the magnet volume for an axial force using PSO.