Multimodal fusion fault diagnosis method under noise interference

In practical industrial production environments, the collection of fault signals is often accompanied by significant background noise. The presence of substantial noise makes feature extraction from fault signals very challenging, thereby reducing fault diagnosis performance. To address this issue, this paper proposes a multimodal fusion fault diagnosis method based on a multiscale stacked denoising autoencoder and dual-branch feature fusion network (MSSDAE-DBFFN). First, the noisy vibration signals are denoised using the MSSDAE. Then, the denoised vibration signals are divided into two branches for feature extraction and fusion. In one branch, the vibration signals are converted into gramian angular summation field (GASF) images using the GASF, and feature extraction is performed with a multiscale convolutional network. In the other branch, the waveforms are subjected to feature extraction using a wavelet scattering network. Finally, the fused features are sent to a classifier to complete the fault diagnosis task. To demonstrate the effectiveness of the proposed method, it is compared with four different denoising methods and five different classification methods across two datasets. The experimental results show that MSSDAE-DBFFN outperforms the other methods in both denoising and classification across five different signal-to-noise ratios (SNR). At an SNR of −10 dB, the SNRs after denoising are 4.582 dB and 5.489 dB, respectively, while the accuracy rates are 89.33 % and 91.67 %, respectively.