Efficient Calibration of LPTN Models for Digital Twins of ePowertrain Motors

Thermal phenomena hold vital significance in automotive applications involving permanent-magnet synchronous motors (PMSMs). The accumulation of heat in vulnerable regions, such as the magnets or end-windings, has the potential to impair device performance and inflict severe damage to components under extreme circumstances. Obtaining precise knowledge of the device temperatures is vital for implementing preemptive measures. However, conducting direct measurements to monitor thermal effects during production is often unfeasible. As an alternative, the development of real-time digital twins, capable of accurately representing the motor’s thermal effects using a limited set of sensor data, emerges as a viable solution. In this study, we present an approach for the development of these digital twins that relies on the utilization of a Lumped-Parameter Thermal Network (LPTN) representation. The LPTN delivers a prediction of the heat flows and temperatures within the PMSM. Then, data from sensors mounted on the device can then be used as input for a sensor data fusion algorithm, to obtain an accurate estimation of the temperatures at critical locations in the motor. With this approach, an accurate calibration of the LPTN parameters is required to ensure the reliability of the results. We present a calibration protocol based on individual tuning of thermal resistances and inertias in the LPTN through a gradient optimization method. This optimization is supported by a sensitivity analysis based on the analytical differentiation of the circuit dynamics. This approach has been proven to deliver fast and accurate results.