Load torque estimation in a three-phase induction motor through a data-driven sensor fusion approach

Abstract This paper introduces a new data-driven approach for load torque estimation in a three-phase induction motor. The method utilizes measurements of single-phase stator current and air-gap flux density, acquired via a transverse Hall effect sensor positioned within the stator slot. A modified CNN-Inception model first processes magnetic flux and single-phase current as individual image representations. These inputs are then integrated into a multi-channel RGB format–capturing flux, current, and their interaction patterns–to facilitate spatiotemporal feature learning. Experimental validation carried out with a 7.5 kW induction motor demonstrates the fusion model’s superiority, achieving mean absolute error (MAE) of 0.43 and 0.54 N m (1.5 and 2.2% mean absolute percentual error (MAPE)), outperforming flux-only of MAE 0.96 to 1.25 N m, and current-only of MAE 0.65 to 0.82 N m grayscale approaches for the motor rated power above 30%, and below the model also achieves good performance in low-load regimes handling torques as low as 0.59 N m, representing 1.5% of rated load with best case results of 0.37 N m of MAE and 2.6% of MAPE, which highlight the robustness of the approach even when operating near no-load conditions.