Quality-relevant Batch Process Fault Detection Method Based On Multi-way Canonical Variate Analysis

In modern society,the requirement of market is changing rapidly.The batch process has the characteristics of small batches?varieties and higher-value-added,and it has been widely used in the fields of biopharmaceuticals,polymer reactions and fine chemicals.Compared with the continuous process,the mechanism and the operation of batch process are more complicated,the quality of production is easier affected by various factors.In fault monitoring,it is of great significance to ensure safe and efficient operation for industrial systems to detect the fault and determine whether the process fault affects product quality or not.In order to judge whether the fault affects the quality of production or not,in this paper,MCVA(Multi-way Canonical Variate Analysis,MCVA)method is applied to study the quality-related fault diagnosis of batch processes.The main research work is as follows:In order to judge whether the process fault affects the quality of the production or not,a method based on Multi-way Multi-subspace CVA for batch process quality-related fault diagnosis is proposed.Firstly,a batch-variable expansion method for the three-dimensional data of the batch process is proposed.Furthermore,the CVA algorithm is used to analyze the correlation between the process data and quality data.The measurement data of process and the measurement data of quality are mapped into five subspaces,which are process-quality correlated subspace,quality-uncorrelated principle subspace,quality-uncorrelated residual subspace,process-uncorrelated principle subspace and process-uncorrelated residual subspace.Through the monitoring of the five subspaces of the batch process,the information changes of all the subspaces in the production process can be completely monitored.The proposed method was applied to penicillin fermentation process to verify its effectiveness.Considering the multi-stage of batch processes,a multi-stage MMCVA method based on quality-related of batch process fault detection is proposed.Considering that process datahave different influence on the quality data in the different period,the quality-related information was introduced to realize the division of the stage.Furthermore the batch-variable expansion method is applied to establish the multi-subspace monitoring model in all stages of the batch process.By analyzing and comparing the simulation results of penicillin fermentation process,it is verified that this method can mine multi-stage information effectively and monitor the process fault more effectively.In order to predict the trend of process deterioration in the early stages of the fault,a method of fault prediction based on quality-related feature trend analysis(QRFTA)is proposed.This method extracts the dynamic feature of the quality-related information by using a canonical variable analysis algorithm and then prediction model is established by using a support vector machine algorithm.This method is validated on the penicillin fermentation process.