Research On Fault Diagnosis Of Electric Valve Based On IPSO-SVM And DBN

As a general mechanical product,valve equipment is widely used in thermal systems to realize system process conversion or perform security protection of system.A modern nuclear power plant has a large amount of valves,the transmission of medium is mainly controlled by valves in nuclear power plants,and valves are indispensable to the safe operation of nuclear power plants.When the valves perform the corresponding function,frequent opening and closing can easily lead to the deterioration of the valve quality,which leads to accidents.Valve failures account for about 70 percent of nuclear engineering related failures,among them,the fault valve is mainly a cut-off type valve,in which the gate valve fault accounts for more than 35 percent.The safety and reliability of valve operation is closely related to the safety of nuclear power plant.Therefore,A test-bed is builded to collect the signals of the electric gate valve under normal and fault conditions.The key technologies of signal processing,fault diagnosis and fault degree evaluation of electric gate valve are studied.Signal processing and fault diagnosis methods for electric valves are proposed and a set of electric valve fault diagnosis system is developed.The main research work is as follows:1)Feature extraction of the vibration signal of the electric valve drive motor was carried out by Wavelet Packet Decomposition(WPD).The energy ratio of wavelet packet reconstruction node was taken as the extracted feature parameter;the fault detection method based on wavelet packet energy entropy was studied.The energy entropy was used as the detection index to judge whether a fault occurred.2)The parameters of Support Vector Machine(SVM)was optimized by using Improved Particle Swarm Optimization(IPSO),solving the problem that optimal parameters of SVM are difficult to determine and PSO algorithm is easy to fall into local optimum,and enhance the generalization ability of the SVM model.The IPSO-SVM fault diagnosis model was trained and established to identify the different operating states of the electric valve by the vibration signal characteristic samples of the normal and fault state of the electric valve.3)Acoustic emission sensor was used to detect valve internal leakage fault;Deep Belief Network(DBN)was used to realize approximate assessment of valve leakage level.4)A fault diagnosis system for electric gate valve was developed by using C#4.0 programming language,which integrates state monitoring,fault diagnosis and fault degree evaluation,signal feature extraction,model training and other functions.The function of each module was tested by experimental data to verify the validity of the system function.The results showed that the research method could correctly identify the operating state of the electric valve,and could make a more accurate evaluation of the leakage degree of the electric valve.System test results verified the effectiveness of each module function of the electric valve fault diagnosis system.The research content in this paper lays a foundation for further research and engineering application of intelligent fault diagnosis of electric valves.