Intelligent identification of internal leakage in natural gas pipeline control valves based on Mamba-ARN

Abstract

Natural gas pipeline control valves are widely used in industrial and civil fields. However, in actual operation, due to factors such as wear, tiny sealing defects may occur between the valve core and the valve seat, leading to natural gas leakage. The complex application environment makes the vibration signal often affected by noise and background factors, which reduces the signal quality and thus affects the accuracy of diagnosis. To address this problem, this study proposes a lightweight intelligent fault diagnosis method based on Mamba Attention Residual Network (Mamba-ARN). The methodology converts one-dimensional vibration signals into a time-frequency feature map using continuous wavelet transform (CWT), which is fed into the Mamba-ARN model for leak detection. To enhance the model's performance, it integrates a residual network structure, ECA and CBAM attention mechanisms, Swish activation functions, and CutMix data augmentation. The main contributions of this method include: (1) By combining Mamba, ECA, CBAM modules and Bottleneck structure, the model can adaptively enhance the feature expression ability, improve the accuracy of feature selection and extraction, reduce computational complexity, and enhance robustness and training stability; (2) The use of Swish activation function and CutMix data augmentation technology not only improves the computational efficiency and stability of the model, but also significantly improves the generalization ability of the model by increasing data diversity, and verifies its versatility and robustness under complex working conditions through transfer tasks under different noise conditions. Experimental results show that the accuracy of the model in the internal leakage recognition task reaches 99.8 %, which is significantly better than the other eight mainstream comparison methods (CNN: 96.9 %, CNN-LSTM: 83.52 %, Transformer: 93.9 %, ResNet-18: 97.9 %, CNN-Transformer: 95.2 %, Mamba-SE: 98.7 %, Mamba- Non-local: 97.8 %, Mamba-MixUp: 97.6 %), further verifying the generalization ability and stability of this method in complex environments.