A MACHINE LEARNING-BASED ACCELERATED PYROLYSIS CHARACTERIZATION AND OPTIMIZATION OF HIGH-TEMPERATURE COMPOSITES

Abstract

Manufacturing polymer-based composites for high-temperature applications involves a series of complex steps during which the material undergoes several transformations. These typically include a lay-up step, a curing process, a high-temperature pyrolytic process to convert the resin phase into amorphous carbon, followed by several resin backfill steps, and finally, graphitization to achieve the desired crystalline structure of carbon atoms. The parameters used during the pyrolysis process significantly affect the degradation reactions, the final yield, laminate permeability, and hence the final properties of the material. Typically, extensive testing is required to characterize pyrolysis kinetics and identify optimal processing conditions. This paper addresses these challenges through a novel probabilistic machine-learning (ML)-based approach for the accelerated characterization and optimization of the pyrolysis process, utilizing theory-based transformations of limited experimental data affected by noise and errors. Gaussian Process Regression (GPR), a Bayesian probabilistic approach to regression, is used to determine optimal test parameters to accurately characterize pyrolysis kinetics and achieve the desired yield while satisfying specific constraints. This approach can be used to improve the processing efficiency of high-temperature composites and increase their performance with minimal experimental effort.