Experimental validation of gas leak detection in screw thread connections of galvanized pipe based on acoustic emission and neural network

Galvanized steel pipes with screw thread connections are widely used in the household part of urban gas distribution system. For such pipes, leakages usually occur at the connections as opposed to the pipe bodies. A leak detection method has been proposed for galvanized steel pipes on the basis of acoustic emission (AE) and neural network. From the viewpoint of engineering application, this work conducts a thorough experimental investigation on the efficiency, accuracy, and applicability conditions of the method. An experimental platform consisting of one trunk pipe and three branch pipes with different diameters is set up to simulate the various leak scenarios. After feature extraction and analysis of the AE signals, a classifier on the basis of back propagation neural network is built for gas leak identification. The applicability of the classifier is investigated quantitatively considering different pipe diameters, number of connections, and leak rate. It validates that the proposed leak detection model can achieve an average accuracy above 95% of the leak with flow above 0.052 L/s on the trunk pipe, on the premise that the signals pass through no more than four screw thread connections. Then, the leak with flow above 0.1 L/s on branch pipes can be detected under the same sensor arrangement.