Multi-fault diagnosis and fault degree identification in hydraulic systems based on fully convolutional networks and deep feature fusion

Normal and stable operations of hydraulic systems are of great importance to the safety and efficiency of industrial production processes. Accurate and prompt diagnosis of fault types and degrees can ensure hydraulic systems return to normal in the early stage of faults and thus can prevent serious accidents. However, the structure of hydraulic systems is complex, and some faults may occur simultaneously. In addition, many hydraulic systems have multi-rate data collected from different sensors. Such problems cause great challenges to fault diagnosis in hydraulic systems. Motivated by the above issues, this paper proposes a deep learning method to diagnose faults and identify fault degrees in hydraulic systems using fully convolutional networks (FCNs) and deep feature fusion. The main contributions are twofold: (1) A new fault diagnosis framework is designed to identify both the fault types and degrees in the presence of multiple faults; and (2) deep feature extractors composed of multiple superimposed convolutional blocks are designed to extract deep features from multi-rate time series, and such features are then fused via flattening and concatenating and fed into the fault diagnosis model. Case studies based on a hydraulic system test bed are provided to demonstrate the effectiveness and superiority of the proposed fault diagnosis method.