Research On Application And Optimization Of Fault Diagnosis And Prediction Methods In Pneumatic Products

At present,electrical valve positioners are widely used in the pneumatic industry.If the positioner fails,oscillations will occur in the entire control loop,which will accelerate the loss of equipment and reduce production quality,ultimately resulting in a waste of resources.Fault diagnosis and fault prediction of the positioner can ensure the normal operation of the entire pneumatic system.The main research contents of this dissertation are as follows:(1)The fault of the positioner is analyzed,and the common fault phenomenon,the cause of the fault and the treatment method of the fault are explained.Then the positioner experiment platform is built to collect the state parameter data of the positioner.(2)Five typical mechanical failures of the positioner are selected,and the real failure state of the positioner is simulated by artificially creating faults for the positioner.Then the positioner in each fault state is subjected to a fault simulation test,and the parameter data of each state of the positioner is collected at the same time.The collected data is denoised by wavelet threshold,then each state parameter of the positioner is constructed with features,and the XGBoost algorithm is used for feature selection.Then the feature selection results are optimized and compared and analyzed.Finally,the processed data is normalized.(3)Based on the research background of big data,a fault diagnosis method based on BP neural network is proposed for electric valve positioner.The single hidden layer and double hidden layer Back Propagation(BP)neural networks are used to establish the fault diagnosis model of the positioner.Then some parameters of different neural networks are traversed and optimized within a certain range,and finally the best parameter points of each neural network are found,and then the data collected in the positioner fault simulation experiment is used to train and test the fault diagnosis model.The effects of the fault pattern recognition of the two BP neural networks are comprehensively compared and analyzed from the three aspects of network structure complexity,network training time and network test accuracy.Finally,it is concluded that the single hidden layer BP neural network is more suitable for the fault diagnosis of the positioner.(4)For the electric valve positioner,a state parameter prediction method based on Long-Short Term Memory(LSTM)neural network is proposed.LSTM neural network,Convolutional Neural Networks-Long-Short Term Memory(CNN-LSTM)neural network and Convolutional Long-Short Term Memory(Conv LSTM)neural network are used to establish the state parameter prediction model of the positioner.Univariate prediction and multivariate prediction methods are used to predict each state parameter of the positioner.By comparing the test accuracy of the network,it is concluded that the multivariate LSTM prediction model is more suitable for predicting the control current of the positioner,the multivariable Conv LSTM prediction model is more suitable for predicting the input pressure of the positioner,the univariate CNN-LSTM prediction model is more suitable for predicting the feedback current of the positioner,the multivariate CNN-LSTM prediction model is more suitable for predicting the output pressure of the positioner.