Experimental results of load-torque estimation in an induction motor using heteroscedastic product-of-experts (PoE) fusion of flux and current signals

Experimental results of load-torque estimation in an induction motor using heteroscedastic product-of-experts (PoE) fusion of flux and current signals Description This dataset accompanies the manuscript “Heteroscedastic product-of-experts fusion of flux and current for load-torque estimation in induction motors” (IEEE Transactions on Instrumentation and Measurement, manuscript TIM-26-09587). The title above is my working title; please replace it with the final title of the paper. The work estimates the load torque of a 7.5 kW three-phase induction motor, fed by the 60 Hz grid under healthy conditions, from a flux-probe signal and a stator-current signal. Each sensor is modeled as a local Gaussian expert, and the two are combined by precision-weighted fusion. The fused mean and fused standard deviation form a two-channel input to a compact convolutional neural network (CNN). The reference torque comes from a load cell and is averaged over windows of 0.2 s (2000 samples at 10 kHz). The record contains the results of all experiments reported in the manuscript and in the response to the reviewers: predictions, metrics, and configuration of every run (13 input configurations, seeds 42, 43 and 44; ten-model cross-validation ensembles evaluated on 7 held-out captures); per-epoch training logs for the runs that were logged; training summaries, the statistics of the local fusion weights by torque range, and the fundamental-frequency characterization of each capture. All files are flat CSV files; the README describes every file and column.