Guided differential dilated convolutional network for fault diagnosis of hydraulic manipulators

Abstract

Fault diagnosis for hydraulic manipulators plays a crucial role in ensuring operational safety but still faces challenges in fault localization. To address this issue, a guided differential dilated convolutional network (GDDCN) is proposed in this study. First, a novel interference masking mechanism is designed to provide dual guidance for feature extraction and classification. Then, a learnable differential kernel with the center parameter fixed at zero and side parameters opposite in sign is designed to adaptively extract gradient features. Afterward, a multiscale gated dilated convolution (MGDC) module is developed to capture global temporal features across multiple scales and achieve gated feature fusion. Finally, the fused features are fed into a fully connected classification module for fault classification. The results show that the GDDCN achieves diagnosis of the faulty joint with an average accuracy of 99.39% on a hydraulic manipulator platform. The superiority and effectiveness of the GDDCN are further confirmed through comparative and ablation studies.