Condition Recognition And Quantitative Analysis Of Internal Leaks Through Valves Based On Acoustic Emission Method

Internal leaks through valves are common problems in petrochemical, thermal power, nuclear power and other industries. The overall reaction system fault diagnosis and safety early warning work are under heavy burden for these problems which may cause the medium pollution and the change of process parameters. Due to the lack of on-line inspection technology, the blind maintenance mode to repair and replace the defective valves, not only brings hidden trouble to the safe operation of the enterprise, but creates a large amount of resource waste. Therefore, there are important engineering value and academic significance to carry out the valve leak detection technology research.From sensitivity analysis of various factors related to the flow, sound field simulation of internal leaks through valves, different failure types of valve leakage simulation models are established with fluent software. Diverse factors such as pressure, temperature, medium, valve size are simulated. Based on the above analysis, the best testing point position, correlation value between impact factors and the leak rate, sound power are determined, which used to guide the experiment design and analysis of the acoustic emission change rules.Experiments study on the effect of pressure, leakage clearance, valve size, type of failure mode on acoustic emission and leak rates are carried out by valve test equipment. Changing laws of amplitude, ASL, RMS value at different factors are analysed. Finally, time-frequency domain features of acoustic emission signal acquired from different types of failure mode are studied.Acoustic emission signals are decomposed into a series of frequency bands with ‘db8’ wavelets based on wavelet packet method. Wavelet packet energy percentage, comentropy and AR value are applied to extract signal characteristic. Two-class and three-class classifications of SVM are used to the identification of failure modes, and the prediction accuracy is 90% and 80% respectively.SVR method is proposed to calculate the valve leak rate. Result shows that this data driven method has a wide range of application. Dimension of data characteristic is reduced by locality preserving projection method. When nearest Neighbor valve k=3 and grid search, PSO algorithm is selected, most of the prediction error is less than 20%. At last, in the view of enterprise valve system management, management information system and matching valve safety protection programs is designed.