Distribution Changes during Thermal Degradation of Poly(styrene peroxide) by Pairing Tree-Based Kinetic Monte Carlo and Artificial Intelligence Tools

Abstract

A tree-based kinetic Monte Carlo (kMC) model is presented that differentiates between 38 end-group pairs for isothermal degradation of poly(styrene peroxide) (PSP). The binary trees allow for fast and accurate calculation of reaction probabilities, with mass-weighted binary trees for the accurate sampling of peroxide bond fissions and hydrogen abstractions along chains. The kinetic parameters are tuned via artificial neural networks (ANNs) to successfully predict literature experimental data, among other lumped product yields. ANNs are also utilized for sensitivity analysis to unravel the effects of individual reactions on the time evolution of experimental responses and other simulation outputs, including the variations of the chain length distributions of the macrospecies. PSP degradation is characterized by three stages of degradation considering both instantaneous and time-averaged concentrations. The first stage features rapid unzipping and results in the fast production of major products benzaldehyde and formaldehyde, the second stage features the most significant level of hydrogen abstractions involving PSP and other macrospecies types, and the third stage exhibits the consumption of the remaining peroxide bonds toward oligomeric species in a wide time frame until the degradation process is finalized by the depletion of peroxide bonds. This proof-of-concept study based on unprecedentedly detailed analyses of the chemistry via kinetic Monte Carlo paves the way to further improve our understanding of chemical recycling of solid plastic waste.