A Electric Vehicle Reducer Bearing Fault Diagnosis Method Based on Space-Time Aware Convolution and Transformer Structure

The reliability and efficiency of industrial machinery are critically dependent on the early detection of faults in critical components such as reducer bearings. This paper presents a novel approach to fault diagnosis method in reducer bearings that leverages the power of space-time aware convolution and the Transformer structure. The proposed method integrates the temporal dynamics and spatial features of the bearing vibrations to enhance the fault detection process. Our approach begins with the acquisition of vibration signals from the electric vehicle reducer bearings. These signals are then processed using a space-time aware convolutional neural network (ST-CNN) that is designed to capture both the spatial patterns and the temporal evolution of the signals. The ST-CNN is followed by a Transformer-based model that further refines the fault detection by attending to the most relevant features in the sequence of processed signals. We have conducted extensive experiments on a dataset comprising various fault types and severities. The results demonstrate that our method outperforms traditional fault detection techniques, including those based solely on CNNs or LSTMs. Our model achieved an accuracy of 92.4% on the self-built data set, far exceeding other comparison models.