Research On Quality Prediction For Batch Processes

In recent years,with the increasingly high quality requirements of industrial products in modern society,the manufacturing of higher-value-added products that are mainly produced through batch processes have become more and more important in many industries.However,the product quality indicators of batch processes are usually obtained only by offline measuring after the end of the process.Multi-stage and nonlinearity are typical features of batch industrial processes.A complete batch operation cycle can be divided into multiple periods,characteristics of variable information in each period of the internal process will have different effects on the final product quality.Therefore,for the multi-stage batch processes,not only should we pay attention to the operation of the entire process,but also in-depth analysis of the process variable information and its correlation with the final product quality characteristics in each period of the process.In this thesis,the partial least squares(PLS)and its improved algorithm are used to extract the correlation between the process variable and the final product quality.On this basis,a new time division method is proposed,and then a multi-stage quality prediction method is developed:(1)According to the relationship between process variables and the quality of the final product with time change characteristic,a time division method based on variable sliding window width is proposed,thus the batch process in accordance with the relevant relationship between process variables and quality variables with different periods is divided into a number of detailed stable periods and transitional periods.(2)After the time division of the batch process is completed,different statistical models are established for each time period according to the different correlation characteristics presented in the stable periods and the transitional periods.(3)In the process of modeling,since the process behavior of each period has different influence on the final product quality,it is necessary to define a "relevance index" related to the final product quality as weight for each sub-stage.In this paper,we have proposed a weighting method based on the correlation between process variables and quality variables.(4)The prediction models of each transitional period and the stable period combining with each weight are weighted summed to establish a comprehensive quality prediction model,and a real-time online quality prediction method is proposed.(5)Using the injection molding process as the experimental background to verify the experiment and the experimental results show that the proposed method is feasible and effective.