Fault Diagnosis of Rolling Bearings using Multi-scale Convolution Neural Network with Hybrid Attention Mechanism

Abstract

With the continuous development of artificial intelligence technology, mechanical fault diagnosis methods based on deep learning (DL) have made great progress. Nevertheless, the operating conditions of mechanical equipment are subject to substantial random factors in real industrial scenarios and there are different levels of environmental noise. This fact undoubtedly puts forward higher requirements for the adaptability and robustness of the model. In this paper, a multi-scale convolution neural network with hybrid attention mechanism (MSCNN-HAM) is proposed to solve the above issues. First, to extract multiscale features and filter invalid information, the one-dimensional vibration signal is input into the multiscale feature learning module. Second, a hybrid attention module is introduced to obtain more effective features. Third, the deep feature is extracted by the module including a series of small convolution kernels. Finally, fault diagnosis is realized through a classifier. The designed method is tested on experiments with different levels of environmental noise, and the final result proved its effectiveness and superiority.