Data-Driven ANN-Based Torque Estimation of High Power Induction Motor Drives for Railways

Modern railway traction drives require precise traction torque control for regulatory compliance and reliable operation. Inaccurate core loss modeling, flux observer errors, inverter non-linearities and measurement uncertainties are common sources of torque estimation errors. Torque estimation error can be reduced by improving motor models and parameter estimation, inverter modeling, variable measurements, etc. However, this is not trivial, and some level of uncertainty will always remain. In this paper, an Artificial Neural Networks (ANN)-based technique for torque estimation error compensation in railway traction drives is proposed. Data used for ANN training are collected during the commissioning process of the traction drive. The proposed method will be implemented in the 400 kW traction drive of a high-speed train. Torque estimation at high speeds was found to be the most critical. Using the proposed method, the estimation error was reduced from 12% to 2% in this case.