Data-Driven Inter Turn Short Circuit Fault Detection of a Segmented SRM Based on Multi-Path Convolutional Neural Network and fCWT

This paper aims to propose a fault detection model for inter-turn short circuit faults in a Segmented Switched Reluctance Motor. To this end, a method is developed in the input of which the raw current signal of the motor is processed by the fast Continuous Wavelet Transform (fCWT) to generate the input matrix for the two-dimensional convolution layer. Compared with the conventional wavelet transforms, this method has proved to be considerably faster. The resulting matrix is input into a novel multi-path Convolutional Neural Network (CNN). This model uses a multi-path block which prevents the unintended elimination of crucial features for fault detection by using the feature map construction of the multi-path of layers. To evaluate this method, a six-phase SSRM is simulated using FEM simulation under healthy conditions and different levels of ITSC fault. Then, the current is acquired in a dataset and used for training and testing the model.