Internal valve leakage in a natural gas pipeline not only brings huge economic losses to the petroleum enterprises, but also causes immeasurable environmental pollution. Therefore, the diagnosis of internal valve leakage and the prediction of leakage rates are the basis to ensure the safe operation of natural gas pipeline. In this paper, based on acoustic emission detection system, the internal valve leakage signals were collected, which were analyzed and processed to diagnose the internal valve leakage and predict the leakage rates. Due to the complex work environment and serious noise interference, the collected acoustic emission signals contain a large amount of environmental noise. Therefore, singular spectrum analysis was proposed to reduce the environmental noise in acoustic emission signals. Radial basis function neural network was used to predict the leakage rates. Experimental results demonstrate that pure internal leakage source signals can be obtained via singular spectrum analysis. The prediction accuracy of leakage rates based on the characteristic parameters of pure AE signals is better than the accuracy without signals denoising. Therefore, singular spectrum analysis is an effective denoising method for acoustic emission signals, which can improve the prediction accuracy of internal valve leakage rate.
