Intelligent Materials Property Prediction for Petroleum Components: PPS–Carbon Fiber Composites Explainable AI

Abstract

The mechanical properties of polymer composites, such as Polyphenylene Sulfide (PPS) reinforced with carbon fibers, are critical for engineering applications demanding high strength, lightweight, and durability. Accurate predictions of these properties, including flexural modulus and impact strength, are challenging due to the complex nature of composites and their reliance on filler content and processing conditions. This study leverages machine learning (ML) to address these challenges by systematically applying and evaluating various regression models, including Support Vector Machines (SVM), Random Forests, Gradient Boosting, Decision Trees, and Ridge Regression. The dataset was rigorously preprocessed and augmented to 1000 synthetic data points, enhancing robustness and model performance. Among the tested models, SVM with optimized hyperparameters demonstrated superior predictive power, achieving an MSE of 0.313 and an R² of 0.979 for flexural modulus, and an MSE of 0.0034 and R² of 0.716 for impact strength. Explainable AI techniques, such as SHAP and LIME, provided transparency into feature contributions, revealing that filler content and processing temperature are key predictors. The study’s innovative use of data augmentation, model selection, and explainability marks a significant advancement in composite property prediction, offering a reliable, scalable approach to material design and optimization. This framework presents a pathway for more efficient, accurate predictions, ultimately enhancing the design and application of polymer composites in various engineering domains.