Quality prediction for polypropylene extrusion based on neural networks

• Published in: IOP conference series. Materials Science and Engineering Vol. 1257; no. 1; pp. 12034 - 12039
• Main Authors: Tan, C H, Yusof, K M, Alwi, S R W
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.10.2022
• Subjects: Approximation; Artificial neural networks; Extrusion; Linear functions; Modelling; Neural networks; Nonlinear response; Polypropylene; Quality; Root-mean-square errors
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/1257/1/012034

Abstract In the polypropylene (PP) industry, melt index (MI) is considered the most important quality variable. Different grades of PP have their specific range of MI. Accurate prediction of MI is essential for efficient monitoring and off-grade reduction. Neural Networks (NN) modelling is proposed as the technique for MI estimation. It has powerful adaptive capabilities in response to nonlinear behaviours. By training the NN, it can discover the relationship between inputs and outputs and makes it capable of function approximation. The goal of this research is to develop NN model to predict the MI based on PP extrusion parameters. Different types of NN such as artificial neural networks (ANN), stacked neural networks (SNN) and deep neural networks were trained and compared to understand their efficiency in solving the problem. The simulation results show that deep neural networks can perform the highest accuracy prediction with the lowest root mean square error (RMSE), followed by SNN and ANN. All three modelling proved that NN can perform non-linear function approximation for polymer extrusion.