Data And Model Dual-driven Based Fault Diagnosis Of Papermaking Wastewater Treatment Process

Sequencing Batch Reactor Activated Sludge Process(SBR),as the most widely used technology of wastewater treatment,plays an important role in the wastewater treatment process.The fault diagnosis system can detect the occurrence time and position of the SBR process in time,and improve the stability and continuity of the SBR process.It is of great significance to apply fault diagnosis to improve production efficiency and reduce equipment maintenance cost.This paper takes the SBR process of a paper mill in Guangzhou as the research object,and collects the historical data of the process as the basis of data driven fault diagnosis.The mechanism model is established according to the SBR process of the plant,which is the basis of model fault diagnosis.Firstly,based on the popular machine learning method,SVM is trained based on historical data to monitor the faults of SBR process.In view of the characteristics of unbalanced data in wastewater treatment process,OCSVM is established for fault monitoring.The fault data collected by paper mill and simulated fault samples including fixed bias,drift bias,total failure and precision degradation was used as test data.The results show that the method can effectively monitor the paper factory faults,as well as fixed bias and total failure.Secondly,based on PCA method,the MPCA multivariate statistical fault diagnosis model is established.MPCA can monitor the paper factory faults and simulated fault of fixed bias,and total failure well and the misdiagnosis rate is low,and the fault source can be located accurately.Aiming at the characteristics of multi operation stages of typical batch process in SBR process,a multi-way principal element analysis method sub MPCA is proposed to divide sub periods by correlation relation and cluster analysis.Clustering analysis is the better way to divide sub periods,and the optimal number of sub periods is determined.Compared with MPCA,subMPCA can monitor the simulated faults of fixed deviation and drift deviation,which can ensure the detection rate,and realize zero misdiagnosis.Thirdly,based on the accepted BSM1 model,according to the actual operation of SBR process in paper mill,the SBR process model is established to simulate the dissolved oxygen concentration and the liquid level of SBR pool.The absolute and relative errors between the simulation and the actual data are calculated to verify the precision of the model.Then,the extended Kalman filter is constructed to generate the residual error,and the fault indication signal is extracted from the residual error.The fault detection and location are realized by comparing residual error with the corresponding threshold.Finally,based on the previous research,OCSVM,MPCA and EKF models are integrated,namely OCSVM to realize fault monitoring,MPCA to realize fault location and EKF to reconstruct fault signal through filtering.A fault diagnosis framework integrating fault monitoring,fault location and fault signal reconstruction is built.And four kinds of simulated faults are used as test data to verify the performance of fault monitoring,the accuracy of fault location and the accuracy of signal reconstruction.