Research on a CNN-CBAM-Based Method for Seal Fault Diagnosis of Flow Control Valves

The seal of a flow control valve is a critical component in hydraulic systems. Seal failure may cause leakage, reduced energy efficiency, and unplanned shutdowns, posing severe threats to safe system operation. Traditional seal fault diagnosis methods often suffer from strong hysteresis and excessive reliance on expert experience; hence an efficient, real-time online seal monitoring technique is urgently needed. This paper proposes a seal fault diagnosis method for flow control valves based on a Convolutional Neural Network with a Convolutional Block Attention Module (CNN-CBAM). First, a reciprocating seal test rig was established, and seal models with multiple fault modes—including abrasion, aging, scuffing, and poor lubrication—were constructed. Friction force–time series data were collected. Then, time–frequency analysis methods such as the S-transform, Short-Time Fourier Transform (STFT), and Continuous Wavelet Transform (CWT) were used to convert the signals into time–frequency maps, which served as input features to the CNN-CBAM network. On this basis, we analyzed how the three time–frequency methods affect the performance of the CNN-CBAM hybrid architecture. Experimental results show that the S-transform + CNN-CBAM model achieved a test accuracy of 99.33% on 3000 samples and better preserved the frequency-adaptive characteristics of the signal. The proposed method significantly improves diagnostic accuracy and robustness, demonstrating that the CNN-CBAM hybrid architecture can effectively identify reciprocating seal faults.