Natural gas pipeline leakage detection based on new wavelet basis transform and singular value decomposition in 2D-CNN

Pipeline leakage fault diagnosis faces challenges, such as background noise interference, insufficient feature extraction and limited sample data in detection tasks. Therefore, the two-dimensional convolutional neural network (2D-CNN)-based natural gas pipeline leakage fault diagnosis method is proposed by innovatively employing wavelet based transform and singular value decomposition (SVD) techniques. First, the acoustic signals collected by acoustic sensors are processed by SVD to eliminate noise interference. Subsequently, custom Morlet wavelet basis functions are designed to transform the data into 2D time-frequency images. Finally, the 2D-CNN model is trained and tested for pipeline leakage detection. To evaluate the detection effectiveness of the proposed method, two case studies are implemented with GPLA_12 and ML_12 collected from public and personal experimental platforms. The results demonstrate that the proposed method can achieve fault recognition accuracy rates of 97.55% and 90.83%, respectively. Compared with other competitive methods, superior performance is further highlighted for application prospects.