Quality-related Fault Monitoring For Multi-phase Batch Process Based On KPI-CCA

Batch process is an important measure for the refinement reform of manufacturing industry.In order to improve product quality,production efficiency and economic benefits,concentrated on the research hotspot of quality-related fault monitoring,this thesis considers with the multi-phase,dynamic characteristics and transmissibility of batch process,combined with three key parts: phase partition,fault monitoring and propagation path analysis.It provides a systematic framework for quality-related fault monitoring of batch process.The achivements are as follows:(1)Propose a phase partition method based on joint canonical variable matrixConsidering the influence of quality variables and dynamic characteristics of batch process on phase partition,a phase partition method based on joint canonical variable matrix was proposed.Using the cooperative idea of dynamic and static combination,firstly,the time slice of joint canonical variable matrix was constructed according to canonical correlation analysis(CCA).Secondly,the slow feature time slice was extracted by slow feature analysis,the production operation period was divided in two steps according to the clustered similarity.The problem of distinguishing strong dynamics from switching points was solved,and the differential phase recognition factor was introduced to realize the accurate positioning of switching points.On the one hand,it can improve the partition sensitivity,on the other hand,it can effectively improve the accuracy of quality-related fault monitoring model.(2)Propose a quality-related fault monitoring method based on multi-phase local nearest neighbor standardized canonical correlation analysisAiming at the problem of reducing monitoring performance after Z-score,a quality-related fault monitoring method based on multi-phase local neighborhood standardized canonical correlation analysis was proposed.Firstly,considering the error of phase partition,the local neighborhood standardization processing was carried out for the historical data samples in the sub-phase,so as to eliminate the impact of abnormal jump points of monitoring statistics caused by false partition;Then,the indirect monitoring of key performance indicator(KPI)was studied.The canonical correlation coefficient was extracted by CCA and decomposed by generalized singular value decomposition to obtain the projection space of KPI-related and unrelated process variables,realize quality-related and quality-unrelated fault monitoring,and solve the problem that quality variables was difficult to measure online.(3)Propose a fault propagation path analysis method based on mutual information interpretation structure modelIn terms of the nonlinearity and the overlapping of production modules of batch process,a fault propagation path analysis method based on mutual information interpretation structure model was proposed.In order to measure the nonlinear correlation between variables,this method combined mutual information and interpretation structure model,integrated process knowledge and historical data,builds information fusion reachability matrix,clearly showed the partition of process modules through variable relationship coordinate diagram,made up for the lack of excessive dependence of knowledge model on process flow diagram.The visualization of causal topology from global,local,horizontal and vertical perspectives accurately identified the propagation path of fault.(4)Field experiment and result analysis of Escherichia coli fermentation productionFinally,the actual data of E.coli fermentation process was used to verify the practicability of a series of algorithms including phase partition,fault monitoring and propagation path analysis.The experimental results fully verified that the systematic framework can comprehensively solve the quality-related fault monitoring problems of batch process.