Batch Process Fault Detection And Diagnosis Based On Slow Feature Analysis

In recnet years,with the urging market requirement for various product types and high product quality,the manufacturing of higher-value-added products that are mainly produced through batch processes have become increasingly important in many industrial processes.Efficient fault detection and diagnosis play key roles in ensuring the plant safety and improving the product quality of batch processes.Massive process data collected by the ubiquitous sensors in batch processes have facilitated the development of data-driven based batch process fault detection and diagnosis.Recently,a new dynamic data analytic technology called slow feature analysis(SFA)has emerged,which is a biologically inspired data-driven based feature analysis method.SFA has the ability of extracting slowly varying features from process data to characterize the dominant varying trends of industrial processes.Considering that bathc process is inherent dynamic time-varying instead of having steady operation point and varies from one stage to another stage due to the change of process operation conditions,and each stage has slowly varying tends.This dissertation uses SFA to extract the underlying driving forces of each stage that result in the batch process dynamic time-varying.Furhtermore,aiming at the complicated characteristics of batch process data,we studies SFA based batch process fault detection and diagnosos method.The main research work and results are as follows.(1)To deal with the high nonlinearity and inherently time-varying dynamics of batch process data,a multiway global preserving kernel slow feature analysis(GKSFA)based batch process fault detection method is proposed.Firstly,kernel SFA(KSFA)is extended to detect batch process fault and multiway KSFA is developed.To overcome the limitation of multiway KSFA that only preserves the local structure information of process data,global preserving structure analysis is further integrated into multiway KSFA and multiway global preserving kernel slow feature analysis(GKSFA)is proposed.Multiway GKSFA not only explores the local dynamic data relationships of batch process,but also considers the mining of global data structure information.Lastly,a rule based on the cumulative slowness contribution is designed to determine the number of the retained slow features in multiway GKSFA model.The simulation studies on the fault detection in a numerical nonlinear system and the benchmark fed-batch penicillin fermentation process demonstrate that the proposed method can improve the batch process fault detection performance effectively.(2)Taking into consideration of the problem that KSFA lacks the ability to utilize the class label information of continuous process data,a discriminant KSFA(DKSFA)based continuous process fault detection and diagnosis method is proposed.Firstly,by incorporating the discriminatibe analysis technology into KSFA,DKSFA is developed to extract the discrimant information between normal operation data and fault pattern data.Then,support vector data description(SVDD)is applied to describe the distribution region of normal operation data and one monitoring index is constructed for fault detection.After a fault is detection,the direction of fault snapshot data is calculated by applying pairwise DKSFA on fault snapshot data and normal operation data.Lastly,the fault pattern of fault snapshot data is identified by measuring the similarity between its own fault direction and the directions of historical fault data.The simulation studies on the fault detection and fault identification in the continuous stirred tank reactor(CSTR)system illustrate that the proposed method achieves satisfactory fault detection performance and can recognize the pattern of fault snapshot data accurately.(3)In order to make full use of the class label information of batch process data and considering the problem of nonlinear fault variable identification,a discriminant GKSFA(DGKSFA)based batch process fault detection and fault variable identification method is proposed.When DKSFA is extended to detect batch process fault,DKSFA only preserves the local neighborhood relationship of normal operation data and omits the global data structure exploration.To overcome this problem,by closely integrating global structure analysis and DKSFA,DGKSFA is developed to extract discriminant feature of batch process as well as preserve global and local geometrical structure information of normal operation data.For the purpose of fault detection,a monitoring index is constructed based on the distance between the optimal kernel feature vectors of test data and reference data.To tackle the challenging issue of nonlinear fault variable identification,a new nonlinear contribution plot method is further proposed to identify the fault variable after a fault is detected,which is derived from the idea of process variable pseudo-sample trajectory projection of fault snapshot data in DGKSFA based nonlinear biplot.The simulation studies on the fault detection and fault variable identification in a numerical nonlinear dynamic system and the benchmark fed-batch penicillin fermentation process demonstrate that the proposed method can effectively detect fault and distinguish fault variables from normal variables.(4)Considering the multiphase characteristics of batch process and the problem of fault variable identification,a batch process fault detection and fault variable identification method baded on global preserving statistics SFA(GSSFA)is proposed.Firstly,statistics SFA(SSFA)is developed to extract slowly varying information at the same sampling time among different batches based on various statistics of the original variables.Then,based on SSFA model,the phase recognition factor(PRF)is defined to automatically achieve steady phases and transitions division.After phase division,GSSFA is further proposed not only to explore the inherent time-varying dynamic information of batch process but also to consider the mining of global data structure information.Furthermore,a novel process monitoring strategy based on the GSSFA model is developed to monitor batch processes steady phases and transitions.In order to identify fault variable,an improved reconstruction based contribution(IRBC)plot based on GSSFA model is proposed to locate fault variable.The RBC of various statistics of original process variable to the monitoring indices is calculated in the proposed RBC method.Based on the calculated RBC of process variables' statistics,a new contribution of process variable is built to identify fault variable.The simulation studies on the fault detection and fault variable identification in a numerical multiphase system and the benchmark fed-batch penicillin fermentation process demonstrate the effectiveness and superiority of the proposed method.