Hybrid-modeling for PTFE polymerization reaction with deep learning-based reaction rate model

Abstract The tetrafluoroethylene (TFE) polymerization process is an essential industrial process to produce polytetrafluoroethylene (PTFE), which is extensively utilized in aerospace and medical domains. A precise mechanism model is a prerequisite for comprehensively understanding this process. However, significant uncertainties in the kinetic model parameters may hinder attaining an optimal reaction rate. This study proposes a hybrid polymerization reaction model that integrates process mechanism modeling and data-driven modeling to address this challenge. In the hybrid modeling approach, the mechanism model for the polymerization reaction is developed based on reaction kinetic and thermodynamic assumptions. Additionally, a long short-term memory (LSTM) neural network is employed to predict the reaction rate for chain initiation by leveraging temporal relationships derived from archived measurements. The proposed methodology is implemented using a PTFE polymer reactor system, and experimental comparisons affirm its superior performance and effectiveness.