Fault Diagnosis Network for Rotating Machinery Based on Multiscale Feature Fusion

Deep learning has been widely used in rotating machinery troubleshooting due to its excellent feature-capturing capability. This research presents a novel framework for machine fault diagnosis using deep learning techniques. We introduce a multimodal feature fusion network (MFFN) that leverages the robust feature learning capabilities of convolutional neural networks (CNNs) in the context of picture analysis. MFFN demonstrates the capability to concurrently handle multimodal fault data, resulting in resilient performance. Specifically, the wavelet transform converts the acquired sensor signals to time-frequency distribution (TFD). Next, we apply a deep CNN to simultaneously learn the discriminative representation from the TFD feature and the original signal. In order to assess the efficacy of the developed deep model, multiple experiments were conducted on several datasets for model analysis. The experimental findings indicate that the proposed method exhibits superior performance compared to conventional fault diagnosis techniques. MFFN can automatically find and pick out useful features that improve the accuracy of fault diagnosis. This is different from traditional methods that rely on experienced professionals to extract fine features. The MFFN we have presented demonstrates enhanced accuracy and stability as compared to a single signal input, effectively addressing the problem of overfitting to some extent.