Development of soft-sensors for polymerization processes

• Main Author: Kng, Madeline Wan Rong
• Other Authors: Chen Tao
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Chemical engineering::Polymers and polymer manufacture
• Online Access: http://hdl.handle.net/10356/61991
• Institution: Nanyang Technological University

Polymerization processes are important industrial processes where the polymer product can be made into many household and laboratory products. One limitation to this process is that there is no online indication of an important polymer quality, melt flow rate (MFR), as at least 2 hours of laboratory analysis is required to determine the polymer quality. To enable real-time monitoring of MFR, various works have been done to conduct prediction of the polymer quality. One approach is modeling based on fundamental principles governing the process which will require good knowledge of the polymerization process and much effort and time due to the complexity of the process. The other approach is to develop an artificial intelligence model using data mining tools such as artificial neural network, which has an advantage over the mechanistic approach due to its excellent ability to model nonlinear relationships. In this project, a software sensor using artificial neural network built-in in MATLAB is developed to predict the MFR of an industrial polymerization plant. A suitable network function, newrb, is recommended in this report. Historical data is found to have an influence on the model and update of model is necessary when training input gets outdated. The normalization method using the mean and standard deviation of the process variables can be further improvised by other univariate and multivariate analysis in order to removing outliers, reduced and classified to gain better prediction results.