Development of a Supervised Learning Model to Predict Permeability of Porous Carbon Composites

Abstract

Predicting the permeability of porous thermal protection system (TPS) materials is essential for understanding their performance during high-speed entry. High-fidelity formulation of Klinkenberg permeability for TPS materials is intractable because unique parameters are needed at each temperature, for various gaseous species, and at every stage of decomposition of resin in the porous material. A supervised learning model based on support vector machine is developed to predict the permeability of TPS materials and is found to be a robust technique to capture the complex relationship between temperature, average pressure, porosity, and permeability of the material. The ability of different gaseous species to permeate through the material is captured through the supervised learning model by constructing an input variable called species identifier, which relates the molecular weight and viscosity of the gaseous species. The model is also extended to capture the permeability of the full composite, which includes both the fibers and the resin. It is demonstrated that the new model captures the nonlinear relationship between permeability and degree of char during the decomposition of the resin in the porous material. The supervised learning model of the physical simulations offers a robust approach for predicting permeability of porous materials in both continuum and noncontinuum flow regimes.