An online motor rotor temperature estimation based on BP neural network with experimental data-driven

The rotor temperature monitoring is crucial for reliability and performance of permanent magnet synchronous motor. Overheating is the main reason for motor demagnetization, which may lead to power output interruption. However, direct measurement of rotor temperature is difficult to be widely applied in electric drive products due to expensive equipment and complicated integration. This paper introduces an innovative approach for estimating rotor temperature of permanent magnet synchronous motor based on BP neural network with big data-driven. The architecture of BP neural network is constructed by adopting Bayesian optimization theory, meanwhile related control parameters of rotor temperature estimation are picked out by Pearson correlation analysis. And then, BP neural network is trained forward and backward by using sufficient experimental data to obtain optimized weights and biases. Finally, wireless temperature measurement equipment is integrated into electric drive system on whole vehicle. Experimental results show that the maximal error is within 6.2°C under different operation environment and road conditions. The proposed method in this paper can meet requirements of rotor temperature estimation precision without extra cost and be extend to other motor application fields.