Estimation of Heat Sources Through Power Measurements in Thermal Models of PMSMs

Electric powertrains are commonly used in the automotive industry nowadays. Thermal phenomena are critical for their performance, as the temperatures of some components must remain below certain thresholds during operation. This is the case of Permanent-Magnet Synchronous Motors (PMSM): magnet temperatures cannot exceed a limit value to avoid demagnetization and motor failure. It is not possible, however, to directly monitor magnet temperatures by means of sensors. A feasible alternative to obtain this information consists in the use of virtual sensors, which fuse a numerical model of the motor with readings from a reduced set of physical sensors, for instance via the use of Kalman filters. The required models can be formulated using Lumped-Parameter Thermal Networks (LPTN), able to describe the thermal behaviour of the motor in real time, at the cost of simplifying the representation of the thermal dynamics of the system. This simplification translates into parameter and geometry uncertainties, which can be alleviated by means of offline calibration. LPTNs are also sensitive to the heat sources in the system and these, in turn, are highly dependent on the operation conditions. Although numerical models exist to describe heat source behaviour, the accurate prediction of heat sources is challenging, particularly at the motor magnets. Improper modelling of thermal losses can result in unrealistic predictions of the motor temperatures. In this work, we present a method to estimate the internal temperatures of PMSMs using additional measurements of the motor power consumption. This information can be related to the thermal losses of the motor, helping Kalman filters to improve estimation accuracy. The proposed approach was tested with a benchmark example and in a standard WLTP bench test with an automotive-grade PMSM. Results confirmed the ability of the method to correct the inaccuracies that stem from heat source uncertainties.