Temperature Estimation of PMSM Using a Difference-Estimating Feedforward Neural Network
Jun Lee, Jung-Ik Ha
Abstract
Keeping the temperature of a permanent magnet synchronous machine (PMSM) in a safe range is essential for maintaining machine performance. In this paper, a feedforward neural network (FNN) for the temperature estimation of a PMSM is proposed. An FNN obtains the past temperature and operating condition values and gives the temperatures information at the next moment. Similar to a thermal-circuit-based model, the proposed model estimates the temperatures of multiple parts of interest with a heat point of view. The proposed FNN estimates the temperature differences, rather than the temperatures, and can enhance the learning and estimation performance of the FNN. Various FNN structures are compared, and the model generation steps are presented. For the experiments performed, where the temperature of the PMSM changes from 20 to 130 °C, the best FNN-based model showed closed-loop estimation errors less than 4.5 °C for the side of the permanent magnet and four parts on the winding, even in 2 hours of closed-loop estimation.