Temperature Estimation of a PMSM using a Feed-Forward Neural Network

Temperature monitoring of electrical machines is crucial to exploit the full potential of the machine while protecting it against thermal overload. Accordingly, the development of suitable monitoring methods has been an important research focus in recent decades. Especially in the automotive industry, precise methods for temperature monitoring without additional hardware are important since high thermal utilization is necessary to reach highest power densities. Artificial intelligence techniques such as neural networks have shown promise in recent years for estimating temperatures in electrical machines. In this work, a feed-forward neural network is implemented to estimate the temperature of a permanent magnet synchronous machine based on the operating point and past temperature values. The performance of the thermal monitoring method using feed-forward neural networks is evaluated by comparison with measurements.