Neural Network Based Optimization of an IPMSM Within a BLDC Drive

This paper introduces a neural network based optimization framework for Interior Permanent Magnet Synchronous Machines (IPMSMs) in squarewave drive applications (i.e., Brushless Direct Current or BLDC drives) for enhanced machine performance. The focus was on reducing torque ripple while enhancing or maintaining the torque-to-current ratio. A Finite Element Method model and a Design of Experiments are employed to generate training data for neural networks. The networks enabled a multi-objective optimization, producing two designs that significantly reduce torque ripple and simultaneously increase average torque. A significant decrease in simulation time alongside an innovative approach to designing IPMSMs for squarewave drives is presented.