Statistical Design of Thermal Protection System Using Physics-Informed Machine learning

TL;DR

This paper introduces a physics-informed neural network approach combined with Bayesian uncertainty quantification and Sequential Monte Carlo methods to efficiently estimate thermal protection material properties, significantly reducing computational costs.

Contribution

It develops a novel PINN-based framework integrated with Bayesian and SMC techniques for fast, accurate thermal property estimation in high-temperature environments.

Findings

PINN reduces computational time compared to traditional methods.

SMC enhances parallel computation and speed.

Combined approach achieves substantial speedup in property estimation.

Abstract

Estimating the material properties of thermal protection films is crucial for their effective design and application, particularly in high-temperature environments. This work presents a novel approach to determine the properties using uncertainty quantification simulations. We quantify uncertainty in the material properties for effective insulation by proposing a Bayesian distribution for them. Sampling from this distribution is performed using Monte Carlo simulations, which require repeatedly solving the predictive thermal model. To address the computational inefficiency of conventional numerical simulations, we develop a parametric Physics-Informed Neural Network (PINN) to solve the heat transfer problem. The proposed PINN significantly reduces computational time while maintaining accuracy, as verified against traditional numerical solutions. Additionally, we used the Sequential Monte Carlo (SMC) method to enable vectorized and parallel computations, further enhancing computational speedup. Our results demonstrate that integrating MCMC with PINN decreases computational time substantially compared to using standard numerical methods. Moreover, combining the SMC method with PINN yields multifold computational speedup, making this approach highly effective for the rapid and accurate estimation of material properties.