Surrogate-assisted PSO for identifying temperature-dependent parameters of aviation BLDC motors: methodology and validation

Precise identification of temperature-dependent dynamic parameters in Brushless DC (BLDC) motors proves essential for maintaining reliability and performance in aviation actuation systems subjected to thermal variations. This paper presents and validates a methodology for identifying key temperature-sensitive parameters (winding resistanceR, motor constantCe, and inductanceL) of an aviation BLDC motor. This methodology integrates three key components: (1) a physics-based Simulink model, (2) a neural network surrogate model trained on simulation data, and (3) particle swarm optimization. To rigorously evaluate the identification techniques, simulated data mimicking step-stress testing conditions with known underlying linear parameter-temperature relationships (ground truth) were employed. Two distinct PSO-based identification strategies were compared: Method 1 identifies parameters at each temperature point individually followed by fitting, while Method 2 directly optimizes the coefficients of the parameter-temperature relationship. The results demonstrate the high accuracy of the NN surrogate model (test set MAE = 0.0054). Method 2 achieved negligible error in identifying ground truth coefficients (KR,KCe,KL), demonstrating superior accuracy compared to Method 1's 2% error margin and non-monotonic parameter handling limitations. This study validates the direct coefficient identification approach (Method 2) combined with surrogate modeling as an accurate and robust methodology, establishing its potential for characterizing BLDC motor thermal dependencies.