Quality Prediction And Process Monitoring Based On Statistical Models For Multiphase Batch Processes

As an important production way,batch processes have been widely applied in various fields of modern industry.For batch processes,two critical problems to be solved are safe and reliable process run and stable and consistent product quality.Process monitoring and quality prediction can be employed to solve these two problems.Particularly,process monitoring and quality prediction based on multivariate statistical analysis have gained increasing attention due to only using the normal process data for modeling and have been under wide investigation.In batch processes,some batch processes include multiple phases within a batch caused by operational of phenomenological regimes,which are called multiphase batch process.Due to exiting the multiple phases,it is more challenging to statistical modeling,online quality prediction and process monitoring in multiphase batch processes.At the same time,it also provides huge research space.This dissertation proposes several efficient methods for online quality prediction and process monitoring,which are summarized as follows:(1).Final product quality depends on the common cumulative effects of different phases.However,traditional prediction models based on the final product quality do not consider these critical cumulative effects so that they hardly reflect the internal correlations between process variables and quality variables accurately.As a result,the prediction accuracy of model is affected.An online quality prediction method based on the phase cumulative quality(PCQ)model is proposed to address this problem.This method isolates the local cumulative effects of different phases on quality by constructing the PCQ model.Then,the final product quality prediction is achieved by summing all the predicted value of PCQ,which the common cumulative effects of different phases on quality are considered.(2).The transitions existing in multiphase batch processes have important effects on product quality.Therefore,it is critical for quality prediction to model in transitions.To improve the performance of quality prediction,an online quality prediction method considering the transition is proposed.First,an improved repeatability factor is defined to divide the batch process into different steady phases and transitions.Furthermore,an online quality prediction method for transitions is proposed based on the switch point of the transition and the just-in-time cumulative model.The input of the just-in-time model is optimized by introducing the switch point and a new similarity factor.Furthermore,the just-in-time cumulative models for transitions are built,considering the just-in-time cumulative effects of transitions on quality.Therefore,the prediction accuracy of the transition is improved.(3).Final product quality not only depends on the cumulative effects of process variables but also the cumulative effects of quality itself.However,these cumulative effects are not considered in the traditional methods.For this problem,an online quality prediction method based on the double cumulative model is proposed.The double cumulative model is constructed by building the internal and external cumulative model,which considers the cumulative effect of process and quality self enough.Also,this method avoids the further data estimation and considers the cumulative effects in the first phase.At the same time,for quality prediction based on the just-in-time model need building more models,two determinate models are constructed to improve the efficiency and performance of quality prediction.(4).To improve the monitoring performance for multiphase batch processes,an online monitoring method based on two steps feature vector selection-based kernel variable correlation analysis(TSFVS-KVCA)is proposed in this paper.This method resolves the computational complexity and the instability of high-dimensional kernel due to the extraction of nonlinear information in multiphase batch processes.Also,the common nonlinear information between two neighboring phases and the specific nonlinear information in a certain phase are considered based on the basis vectors extracted by the TSFVS-KVCA method,which not only improve the performance of process monitoring in the steady phase but also provide a favorable platform for modeling in the transition.Furthermore,an online monitoring method for transitions is proposed based on the identification of the dominant phase.Apply above methods to multiphase batch process simulations and the results of simulation demonstrate effectiveness and reliability of the proposed method.Finally,the conclusions are drawn and further researches on multiphase batch processes are illustrated.