Ablation and mechanical properties of carbon/high silica/phenolic composites: Using experimental analysis and five machine learning models for correlating and generalizing to diverse compositional ratios

Abstract

This paper investigates the thermal and mechanical properties of carbon/high silica/phenolic composites with varying reinforcement ratios. Five hybrid samples were fabricated: 100% carbon, 75% carbon/25% silica, 50% carbon/50% silica, 25% carbon/75% silica, and 100% silica. A three-point bending test evaluated their strength, while an ablation test at 3000°C for 1 minute measured backside temperature, linear ablation rate, and mass ablation rate. Results indicated that the 100% carbon sample had the highest bending strength, while the 50% carbon/50% silica sample achieved the lowest linear and mass ablation rates, demonstrating an effective balance between fire retardancy and insulation, resulting in minimal backside temperature during ablation. Additionally, five machine learning models (Linear Regression, Decision Trees, Random Forests, Gradient Boosting Machines, and Neural Networks) were utilized to predict mass ablation rate, linear ablation rate, and strength. Decision Trees and Gradient Boosting Machines exhibited the highest prediction accuracy, while Linear Regression struggled with non-linear data, resulting in lower accuracy for ablation rate predictions. Notably, these models were also able to generalize to other percentages, showcasing their robustness and versatility in optimizing material compositions beyond the tested scenarios. This study highlights the potential of machine learning in predicting the properties of advanced composites, contributing to the development of high-temperature resistant materials.