Research On Fault Diagnosis Of One-way Valve Of High-pressure Diaphragm Pump Based On Deep Belief Network

Check valve is one of the most important components in reciprocating high-pressure diaphragm pumps,and its working state directly affects the safety of high-pressure diaphragm pump operation and the production efficiency of the enterprise.During the production process,the reciprocating high-pressure diaphragm pump has a harsh working environment and variable working conditions,the conveyed mineral slurry usually have the characteristics of high pressure,high temperature,high corrosion and high concentration,its own mechanical structure is complex,so each moving part affects each other.Therefore,the collected check valve vibration signals are non-stationary,non-linear,complex in composition,and accompanied by strong noise,which makes feature extraction and fault diagnosis difficult.Aiming at this problem,a check valve fault diagnosis method based on deep belief network is proposed,combining with the advantages of time-frequency distribution images in characterizing signal feature information,and the advantages of deep learning being able to mine the internal connections of data and establish complex mapping relationship between feature and state.The main research contents are as follows.(1)In order to solve the problem that the vibration signal of the check valve has complex non-stationary nonlinear components and is accompanied by strong noise,and the characteristic representation of single domain,such as time domain or frequency domain,cannot represent the vibration signal well and will be affected by more irrelevant information,the time-frequency distribution image of vibration signal is used to characterize the state characteristics of the check valve.Obtain the time-frequency distribution images of four time-frequency analysis methods of Short-time Fourier Transform(STFT),Wigner-Ville Distribution(WVD),S-transform(ST),and Generalized S-transform of the check valve vibration signal.Through analysis and comparison,the Generalized S-transform time-frequency distribution image has the advantages of less noise interference and better time-frequency aggregation,and the Generalized S-transform time-frequency image under various operating states has obvious differences in energy density distribution,which can more accurately complete the fault detection of the check valve.(2)The research on the fault mechanism of the reciprocating high-pressure diaphragm pump check valve is not completely mature,and there is no definite index or time function or frequency function to characterize the state information of the check valve.Therefore,based on the time-frequency distribution image,which can distinguish the operating state of check valve,a pattern recognition method is introduced,and a check valve fault diagnosis method based on Generalized S-transform and Deep Belief Network(DBN)is proposed.First,the Generalized S-transform is used to obtain the time-frequency distribution image of the vibration signal in each state.Then,2-Dimension Non-negative Matrix Factorization(2DNMF)is used to calculate the row-based matrix,column-based matrix,and feature matrix of the time-frequency image,at the same time,complete the reduction of the time-frequency image to reduce the amount of calculation.Finally,the feature matrix is converted into a feature vector and input to the DBN fault classification model to complete the fault diagnosis of the check valve.The feasibility of this method is proven by experimental analysis and comparison.(3)Before using DBN for classification and recognition,the feature matrix needs to be rearranged into vectors,which will lose the features on the two-dimensional plane of the time-frequency distribution image,and DBN have many parameters and a large amount of calculation.In view of this problem,combining with the advantages of Convolutional Neural Network(CNN),a check valve fault diagnosis method based on Generalized S-transform and Convolutional Deep Belief Network(CDBN)is proposed.Threshold segmentation method is used to segment the compressed Generalized S-transform time-frequency distribution image,highlight the higher energy density distribution of the time-frequency image,and then directly input the feature matrix into the CDBN fault classification model for fault classification.Through experimental analysis and comparison,the results show that the CDBN fault classification model converges quickly and the recognition accuracy is high,indicating that this method can well complete the fault diagnosis of check valves.The thesis takes check valve as the research object.Aiming at the problem that the feature information in the check valve vibration signal is not easy to extract,a fault diagnosis method combining time-frequency distribution image and deep learning is proposed to realize the feature extraction and fault diagnosis of check valve.