Objective The pivotal role valves play in ensuring the stable operation of natural gas pipeline systems necessitates real-time monitoring and defect identification to eliminate early leakage in natural gas valves. To this end, it is imperative to develop an effective method for detecting and identifying micro-leakage defects.

Methods An acoustic emission detection experimental setup was initially established targeting valve micro-leakage, with three pressure settings of 0.2 MPa, 0.4 MPa, and 0.6 MPa. This self-developed setup was utilized to capture the acoustic emission signals across five valve conditions (healthy state, external leakage due to valve stem corrosion, internal leakage due to untightness in valve closure, internal leakage due to valve core scratch, and external leakage due to flange gasket damage) through the acoustic emission detection method. The collected signals were then used to create time-frequency domain feature matrices. These intricate and redundant time-frequency domain feature matrices were transformed into two-dimensional and threedimensional feature factor matrices utilizing Principal Component Analysis (PCA). The optimal number of clusters for each feature factor matrix was determined through subsequent calculations. The two-dimensional and three-dimensional feature factor matrices were further analyzed using K-means and K-medoids clustering methods.

Results For the identification of micro-leakage defects of natural gas valves under low pressure conditions (0.2 MPa), the clustering effect of 3D PCA reached 97.8%, surpassing that of 2D PCA. As the pressure was gradually raised to 0.6 MPa, the clustering effect of 2D PCA continued to improve, eventually surpassing that of 3D PCA. In the evaluation of clustering methods, it was observed that the K-medoids method exhibited greater stability and higher operational efficiency compared to the K-means method.

Conclusion The classified processing method for acoustic emission signals, based on acoustic emission technology, feature extraction, PCA, and clustering algorithms, can enhance the accuracy and stability of valve micro-leakage detection. Moreover, it enables the classified identification of defects in natural gas pipeline valves. This approach not only offers monitoring means but also establishes safeguard measures crucial for ensuring the secure operation of natural gas pipeline systems.