Convolutional Neural Network-Based Online Stator Inter-Turn Faults Detection for Line-Connected Induction Motors

Stator inter-turn faults (SITFs) constitute a significant portion of induction motor failures. In this paper, a novel two-dimensional (2D) convolutional neural network (CNN)-based SITFs detection technique is proposed for line-connected induction motors, where the fundamental frequency phasor magnitude (FPM) and the 3rd harmonic components of the stator currents are used as signals. The proposed technique extracts FPMs from the measured three-phase stator currents of an induction motor through the digital Fourier filtering and then reconfigures them into three-dimensional (3D) images using the current to image transformation (CIT) mechanism to form image datasets. The same preprocessing procedure is applied to the 3rd harmonic components. The proposed approach is validated using experimental data measured in the lab for a 2.2 kW induction motor under various healthy and SITFs conditions, along with 12 motor loadings and an unbalanced voltage supply. It shows robust SITFs detection performance under load variations and power supply asymmetry. FPMs and SITFs work equally well for the SITFs detection and fault severity assessment; while the faulty phase can be effectively identified by FPMs.