Research On Cavitation State Monitoring For Control Valve Based On Neural Network

As a control element in the hydraulic system,the control valve adjusts the flow rate and pressure of the oil flowing through the hydraulic system by changing the displacement of the spool.During the use of the control valve,the pressure and velocity in the flow field will change due to the complex flow channel structure,resulting in cavitation and vibration,which will affect its control accuracy.There are differences of the vibration characteristics at different surfaces of the valve body induced by the cavitation phenomenon.Therefore,it is of great significance to improve the safety performance of the control valve by recognizing the cavitation state.The difference of cavitation-induced vibration at different measuring points of the control valve and the sensitivity analysis of cavitation development state are carried out.Firstly,an acceleration signal acquisition test platform for cavitation vibration of control valve is built,and the acceleration signal of cavitation-induced vibration of control valve at multiple measuring points is collected synchronously.Secondly,vibration acceleration level and center of gravity frequency are proposed to analyze the position of measuring points.The vibration frequency spectrum of control valve is divided by 1/3 octave method,and the most sensitive frequency band and measuring point position to cavitation development are finally obtained.The results show that the vibration signals of the same surface measuring point at the control valve body are similar and the signals of different surface measuring points are quite different.The results shows that the vibration signals of the control valve are anisotropic.The development of the cavitation state mainly causes the vibration intensity of the frequency band above the center frequency of 10000 Hz to increase,which is suitable as the characteristic frequency band for monitoring the cavitation state of the control valve.Finally,the cavitation state coefficient is proposed to accurately distinguish noncavitation,cavitation inception,critical cavitation and severe cavitation.The research on cavitation state recognition method of control valve based on Gram angle field-convolutional neural network is carried out.First,the time series obtained by the sensors is cut using a non-overlap moving time window,and the extremely long time series obtained by the sensor is cut into a series of time segments.Secondly,the image data is obtained by using Grammiar angle field transformation to transform the time segments.Again,each sensor data is used as an independent channel,and the deep learning model is used to perform feature-level fusion of multi-point data.Then,the convolutional neural network Alex Net model is used to classify and identify the four types of cavitation states of the control valve.The T-SNE algorithm is used to visualize the input layer data and output layer data of the model.Finally,according to the robustness and anti-interference performance of the model,the recognition accuracy of the model under different noise levels and different openings is analyzed.The recognition accuracy of the proposed model for multi-opening data sets and the recognition accuracy under different levels of Gaussian white noise are studied.The online monitoring system of cavitation state of control valve is studied.Firstly,the online monitoring and cavitation state classification system of control valve is built by Matlab software.The online monitoring software is designed by App designer module in Matlab,including feature monitoring sub-module and cavitation state realtime classification sub-module.Then,when using the online monitoring system to classify the cavitation state of the control valve,the recognition accuracy is high.