Real-time Temperature Estimation of Permanent Pagnets in a PMSM Using a Hybrid ANN and Lumped-Parameters Thermal Network

Since Permanent Magnet Synchronous Motors (PMSM) have high popularity in EVs accurate temperature estimation of the Permanent Magnets (PMs) as the most vulnerable components of these motors has a high degree of importance. Lumped Parameter Thermal Network (LPTN) is one of the conventional methods, used for temperature estimation. Despite some merits of this method like simplicity and low computational burden, finding some of the parameters like airgap thermal resistance and updating the parameters during the lifespan in different conditions are still open problems. In this paper, a novel PM temperature estimation method is proposed using both LPTN and Artificial Neural Network (ANN). In this method, equivalent thermal resistances of motor airgap in the LPTN are estimated and updated in realtime using an ANN considering the motor operating conditions. The ANN is trained just once and can be deployed for the rest of the PMSM lifespan. By updating the parameters of the LPTN in real-time, negligible error in the estimation of PM temperature is guaranteed. The effectiveness of the proposed method is validated using a 1.5 kW PMSM prototype. It is shown that compared with the conventional methods in which airgap the resistances of LPTN have been derived analytically, the estimation error is reduced significantly by the proposed method. The results show that the temperature estimation error of this method lies under 0.5 oK frequently with the MSE of 0.05. However, using the considered conventional method of LPTN the error lies under 5 oK and MSE is about 7.