Fault Diagnosis in an Asynchronous Motor Using Three-Dimensional Convolutional Neural Network

Modern industrial processes make extensive use of electric induction motors; timely fault diagnosis of such equipment is a critical phase that allows planning effective predictive maintenance which can significantly reduce downtime and optimize productivity. In this study, we present a novel technique for the diagnosis of both simple mechanical and electrical defects as well as their combination in an asynchronous motor based on current signal measurements using various operating states. The suggested technique implements the fault detection mechanism by mapping stator signals recorded from a three-phase induction motor in a steady state. These signals are used to indicate a healthy condition as well as failed states, which are mapped onto a 3D point cloud using a voxelization approach. The use of this data structure aids in the generation of rich features that keep intact the spatial relationship between the three stator currents, which is then utilized to train a 3D convolutional neural network with the produced voxels (3D CNN). The proposed method surpasses the classical deep learning diagnosis-based method by projecting the stator currents into 2D RGB images because our method prevents data collapsing. The suggested technique for real-time monitoring of the operational state of an induction motor is resilient and fast, according to experimental results acquired by utilizing data from a real test bed.