A Knowledge-driven Physics-Informed Neural Network model; Pyrolysis and Ablation of Polymers

Aref Ghaderi, Ramin Akbari, Yang Chen, Roozbeh Dargazany

Published 2022-09-23Version 1

In aerospace applications, multiple safety regulations were introduced to address associated with pyrolysis. Predictive modeling of pyrolysis is a challenging task since multiple thermo-chemo-mechanical laws need to be concurrently solved at each time step. So far, classical modeling approaches were mostly focused on defining the basic chemical processes (pyrolysis and ignite) at micro-scale by decoupling them from thermal solution at the micro-scale and then validating them using meso-scale experimental results. The advent of Machine Learning (ML) and AI in recent years has provided an opportunity to construct quick surrogate ML models to replace high fidelity multi-physics models, which have a high computational cost and may not be applicable for high nonlinear equations. This serves as the motivation for the introduction of innovative Physics informed neural networks (PINNs) to simulate multiple stiff, and semi-stiff ODEs that govern Pyrolysis and Ablation. Our Engine is particularly developed to calculate the char formation and degree of burning in the course of pyrolysis of crosslinked polymeric systems. A multi-task learning approach is hired to assure the best fitting to the training data. The proposed Hybrid-PINN (HPINN) solver was bench-marked against finite element high fidelity solutions on different examples. We developed PINN architectures using collocation training to forecast temperature distributions and the degree of burning in the course of pyrolysis in multiple one- and two-dimensional examples. By decoupling thermal and mechanical equations, we can predict the loss of performance in the system by predicting the char formation pattern and localized degree of burning at each continuum.