Research On Process Monitoring For Slow-varying Batch Processes Based On Multivariate Statistical Analysis

Batch processes play an important role in the process industry,whose products have close relationship with people's everyday life.In order to timely handle abnormal process behaviors and guarantee safe and stable operations,it's necessary to monitor batch processes.Affected by various factors,such as catalyst passivation,equipment wear and sensor drift,slow-varying in practical batch processes is common.Batch-to-batch slow-varying behaviors are normal process behaviors,which do not affect the process safety or the quality of products.So the batch process monitoring systems should distinguish between slow-varying behaviors and faults.In addition,the slow-varying behaviors could lead to model mismatch problem,and cause misinformation and false alam of fault.Thus,batch-to-batch slow-varying behaviors bring new challenges to batch process modeling and on-line monitoring.Through elaborately analyzing slow-varying behavior characteristics,this thesis researches slow-varying batch process monitoring by multivariate statistical analysis:This thesis develops a slow-varying behavior characteristic estimation model by using multiway partial least squares(MPLS).The between-batch difference matrix is calculated by differencing the two bath process data with some batches interval,and this matrix extracts between-batch difference caused by slow-varying.The estimation model based on MPLS is built by analyzing the statistical rules combined in the data information of the calculateddifference matrix,and consequently slow-varying on-line estimation can be realized by themodel.Furthermore,the on-line monitoring data is compensated by the estimated values to eliminate the impact of slow-varying on monitoring data.In order to avoid slow-varying adversely affecting process monitoring,on the basis of data compensation,this thesis proposes a batch process monitoring method.Combined with the advantages of that multiway principal component analysis(MPCA),high-dimensional and highly coupled data can be processed.The monitoring model is built based on MPCA,and the process statistical rules and the control limit of statistical variables is extracted and calculated by the model respectively.During on-line monitoring,the data to be monitored is beforehand compensated by the slow-varying estimation calculated by slow-varying model.This monitoring method is significantly helpful for slow-varying process monitoring,and can guarantee the accuracy and robustness of the process monitoring model.The proposed method is used to monitor the penicillin fermentation process to verify its effectiveness.The simulating results shows the proposed monitoring method is able to distinguish slow-varying and process faults,and can accurately monitor batch processes.