Online Temperature Estimation of Permanent Magnet Synchronous Machines (PMSM) Using Non-linear Autoregressive Neural Networks with Exogenous Input (NARX)

PMSMs are widely used in high-performance industry applications. This popularity is due to their high torque-to-inertia ratio, high efficiency, low maintenance, fast dynamic response, among others features. However, the construction of such machines includes some components that are highly sensitive to the temperature, hence, requiring control strategies that mitigate failures and loss management, taking the machine temperatures into account. Sensor-based temperature measurements of such parts are difficult to be implemented, and are not always well-accurate. Therefore, this paper proposes an approach based on artificial neural network model to estimate the temperature at the most critical points of a PMSM, namely, the permanent magnet, stator teeth, windings, and stator yoke. In this study, the variables, ambient and coolant temperatures, motor speed, and the stator voltages and currents in the direct and quadrature axes are taken as inputs to a Non-linear Autoregressive Neural Networks with Exogenous Input (NARX). To develop and test the proposed temperature estimator, a 140-h multivariate database from a torque-controlled 52 kW PMSM was used. The obtained results have shown that the proposed method successfully estimates the temperature at the selected points.