Hydraulic valve core fault diagnosis applied with a data-driven approach

The hydraulic valve is the critical element of the hydraulic system applied in advanced manufacturing. Tough working conditions cause the core stuck fault as the significant issue, which reduces the operating efficiency and even results in system breakdowns. Therefore, this research provides a data-driven approach to diagnose the core stuck fault of the hydraulic valve to address these challenges. Initially, the wavelet packet denoising (WPD) is employed to denoise the measured signals. The pre-processed signals are then subjected to the variational mode decomposition (VMD), with parameters optimized by the sparrow search algorithm (SSA), to extract the fault-related features. Finally, the hybrid model combining the convolutional neural network (CNN) and the long short-term memory (LSTM) is applied to diagnose the stuck fault. The research findings suggest that the proposed approach can effectively save the data processing time (45.13% on average) and achieve the excellent diagnostic performance of the core stuck fault (99.70% in combined sensor groups). These encouraging results can lead to the adequate confidence for further investigation of hydraulic valve studies and present applicable to diagnose other hydraulic machine faults.