Abstract
Predictive maintenance for electric drives increasingly leverages data-driven diagnostics to detect early-stage faults under varying operating conditions. Building on the analysis and simulation of induction motor behavior under fault conditions in industrial electric drive systems, this paper presents the design, training, and validation of a deep convolutional neural network (CNN) capable of automatically identifying multiple fault types. The CNN operates on a 2D time–feature representation of multivariate signals and is trained using Adam optimization with a 70/30 train–holdout split. Results show 92.31% accuracy on the holdout set, with a confusion matrix indicating minimal inter-class confusion. For deployment, the trained model together with the normalization parameters is integrated into a masked Simulink block to enable automated end-to-end inference on new simulation runs.