Research On Modeling And Correcting Methods In Continuous Twin-Screw Wet Granulation Process

The continuous twin-screw wet granulation process is an important part of the pharmaceutical production process.Establishing a forecasting model between granulation process operating variables and raw material attributes and quality indicators lays the foundation for process quality index control.However,changes in the properties of raw materials in the actual production process can cause changes in the working conditions or slow time-varying of the granulation process,which leads to a decrease in the accuracy of the prediction model.Therefore,it is of great theoretical and practical significance to study how to effectively ensure the performance of the prediction model to meet the production requirements under the condition that the characteristics of the continuous twin-screw wet granulation process are changed.This thesis first introduces the basic principles and process flow of continuous twin-screw wet granulation process.Based on previous research results,a mechanism model that reflects the correlation between operating variables,raw material properties and the average particle size of the final quality index is established.The validity of the mechanism model is verified by simulation experiments.The mechanism model of continuous twin-screw wet granulation was used to simulate the actual production process to produce the data needed for research.The PLS method was used to establish a model for predicting the average particle size.Then use the model-based principal component analysis(MBPCA)to establish a monitoring model to monitor the performance of the particle average particle size prediction model to determine whether the characteristics of the granulation process have changed.When changes in the granulation process characteristics are monitored,a method for identifying changes in the granulation process characteristics based on the combination of wavelet packet analysis and SVM is proposed,that is,using wavelet packet analysis to extract the statistical sequence characteristics of the monitoring model,and using the extracted features as input training SVM classifier to identify whether the granulation process is slow time-varying or operating conditions change.When identifying whether the granulation process is slow time-varying or the operating conditions are changed,this paper proposes a targeted model correction strategy.Specifically,when slow time-varying is identified in the granulation process,recursive PLS method is used to correct the model to eliminate the effect of slow time-varying.When the operating conditions of the granulation process are identified,a model correction strategy based on a combination of cases and just-in-time learning(JIT)is proposed.First,match similar cases from the model case library.If there are similar cases,switch the similar cases directly;if there are no similar cases,use the JIT-based method to re-establish the particle average particle size prediction model to correct the model and add it to the model case library to update the case library.Finally,the validity of this model correction strategy is verified by simulation research.