Robust PPLS Modeling And Its Application In Process Monitoring

In recent years,safety accidents have often occurred in industrial production.In order to reduce property loss and casualties,it is necessary to carry out a reliable and accurate process monitoring for the industrial process to ensure safe and smooth operation of production and product quality.In addition,plenty of data are stored,which results in the use and development of the multivariate statistical monitoring methods.Traditional process monitoring methods are usually assumed that the sample data is subject to a normal distribution and that the sample rate is consistent.However,the actual variable data distribution is complex and the sampling period is diverse,which brings difficulties to the process monitoring.In this dissertation,in view of the practical problems such as outliers,multirate and dynamic characteristics in real industry,based on PPLS model,a robust model is builded and applied to process monitoring.The main research is as follows:1.In order to solve the problem of outliers in industrial process,the shortcomings of PPLS model are analyzed.This section improves the robustness of the PPLS model based on the assumption that the raw data satisfy T distribution rather than Gaussian distribution.By adjusting the freedom degree of T distribution,the proposed robust probabilistic partial least squares(RPPLS)model can overcome the shortcomings of PPLS model.Furthermore,on the basis of RPPLS model,two monitoring indicators GT2 and GSPE are proposed to monitor the process state and the model changes,respectively.Comparing the monitoring performance in the TE process based on PPLS and RPPLS shows that RPPLS is more effective than PPLS in terms of the fault accuracy and the missing alarm rate.2.Considering the multirate problems in the industrial proces,this section presents a semi-supervised robust probabilistic partial least squares(Semi-supervised RPPLS)method which can handle the data that the sample sizes of input variables and output variables are unequal.The model should be developed based on complete data samples.But the dataset are divided into two parts,the first part that contains samples of both the process variables and corresponding quality variables is denoted as the labeled dataset;the other part that only consists of the process variables samples,which is treated as the unlabeled dataset.The unlabeled dataset are employed together with the small amount of labeled dataset to develop a valid statistical model.Furthermore,on the basis of semi-supervised RPPLS model,three monitoring indices GT2,GSPE_x and GSPE_y are proposed to evaluate the process state and the model changes,respectively.The proposed method is proven to be more effective by comparing with the down-sampling RPPLS method in the monitoring of TE process.3.Considering the dynamic characteristics in the actual conditions,a dynamic robust PPLS modeling method is proposed based on the extension process data matrix with state space mode.In the proposed method,not only the correlation between variables is considered,but also the correlation of variables in time dimension is fully taken into account.Therefore,the developed model can extract the useful information in the process.Besides,the GT2 and GSPE indicators are proposed based on the dynamic robust PPLS which can monitor dynamic process accurately and timely through the integration of dynamic and static process information.The application in the TE process demonstrates that the dynamic robust PPLS can monitor the occurrence of process faults more accurately and effectively.