Enhanced Torque Estimation Based on a Cognitive Training Model for Robust PMSM in EV Applications

Abstract

Permanent Magnet Synchronous Motors (PMSM) which are used in commercial applications, requires precise torque calculation, which is necessary for the intended control. Conventional Model Predictive Control (MPC) performance is hampered by model parameter mismatches and high computational demands, precise torque control often necessitates the knowledge of rotor speed and position, which are traditionally obtained using mechanical sensors. The paper proposes Feedforward Neural Network model to estimate the parameter for desired switching of inverter for accurate position of rotor in optimized time. However, this model uses the d-q axis currents, voltages, rotor angle as inputs, and electromagnetic torque as the output. The model is developed with the help of Python programming based on Hyperband algorithm for hyperparameter tuning. Hyperband algorithm, efficiently optimizes hyperparameters by adaptive resource allocation, early stopping, reducing training time and improving accuracy. This integration allows the neural network(NN) to dynamically optimize its architecture, ensuring precise torque estimation. This approach addresses computational challenges and enhances the system's efficiency and responsiveness to real-time parameter variations and disturbances, leading to more robust and high-performing motor control applications.