A novel multi-scale convolution model based on multi-dilation rates and multi-attention mechanism for mechanical fault diagnosis

Most recent studies on deep learning models have focused on extracting richer features from the original signal by using multi-size convolutional kernels to improve the accuracy of mechanical fault identification. Meanwhile, the problem in which the feature resolution obtained by the convolution with larger convolution kernels is lower is disregarded. Such practice is harmful in fault classification. Moreover, the model may be disturbed by redundant information in the feature maps during training. To resolve the aforementioned problem, we propose a novel multiscale convolution model based on multi-dilation rates and multi-attention mechanism (MDRMA-MSCM) for mechanical fault diagnosis. In particular, MDRMA-MSCM adopts dilated convolution with a fixed-size convolutional kernel but different dilation rates to extract richer features, while, the quality of the feature maps is also ensured. Then, it uses the multi-attention mechanism to enhance feature maps in the temporal and channel dimensions. The output is the final representation for fault diagnosis. Comprehensive experiments on two benchmarks, namely, the bearing and gearbox datasets, demonstrate the effectiveness of the proposed method. In particular, MDRMA-MSCM achieves 99.71% and 99.99% average classification accuracy on the two datasets, respectively. Such results surpass those of other representative fault diagnosis methods in terms of diagnostic accuracy. Moreover, we tested the generalization capability of the model under variable working conditions and different noise datasets. The results indicated that the performance of our model is superior to those of the other models, demonstrating that our model exhibits better generalization capability. (C)& nbsp;2021 Elsevier Inc. All rights reserved.

Automated summary

This study proposes a novel multiscale convolution model based on multi-dilation rates and multi-attention mechanism (MDRMA-MSCM) for mechanical fault diagnosis. Comprehensive experiments demonstrate that the proposed method achieves higher classification accuracy than other methods on bearing and gearbox datasets, and exhibits superior generalization capability under variable working conditions and different noise datasets.