Prediction Method of Large-Diameter Ball Valve Internal Leakage Rate Based on CNN-GA-DBN
Jize Wan, Mingjiang Shi, Yanbing Liang, Liansheng Qin, Liyuan Deng
Abstract
Large-diameter ball valves are primarily applied in oil and gas long-distance pipelines for emergency shutdown. Due to the problems of friction between the transmission medium and the valve cavity, as well as corrosion and aging, the ball valves are prone to internal leakage, failing emergency shutdown. Therefore, it is essential to accurately detect whether the internal leak occurs in the valve. This work adopts the acoustic emission (AE) technology to detect the leaking ball valve and proposes a novel method for leakage rate prediction, which uses a convolutional neural network (CNN) combined with a deep belief network (DBN) for feature learning and a genetic algorithm (GA) to optimize DBN, building a prediction model based on CNN-GA-DBN for internal leakage rate. Combined with the built experimental platform, the internal leakage signals of ball valves under different conditions were acquired, and the experimental verification was operated. The results show that the optimal values of mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), and correlation coefficient (CORR) obtained using CNN-GA-DBN are 0.8668, 14.2879, 1.0681, and 0.9963, respectively. It indicates that the proposed method can provide powerful support for the internal leakage rate prediction of ball valves.