A Machine Learning Framework for the Reconstruction of Composite Fatigue and Fracture Properties: A Synthetic Data...

Area of Science:

• Materials Science
• Computational Mechanics
• Machine Learning

Background:

• Natural fiber-reinforced composites offer sustainable alternatives to conventional materials.
• Predicting mechanical properties like fatigue life and fracture toughness is crucial for their application.
• Existing analytical models often require extensive experimental validation.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) framework for reconstructing fatigue life and fracture toughness in natural fiber composites.
• To assess the predictive accuracy of six regression algorithms using a synthetic dataset.
• To establish a reproducible ML evaluation pipeline for composite material studies.

Main Methods:

• A synthetic dataset of 600 samples was generated using established Basquin fatigue and Rule of Mixtures fracture equations, with calibrated stochastic noise.
• Eight natural fiber types and five matrix systems were included, simulating diverse composite compositions.
• Six regression algorithms (Random Forest, Gradient Boosting, SVM, Neural Network, Ridge, Lasso) were evaluated using a 70-15-15 split, 5-fold cross-validation, and grid search optimization.

Main Results:

• Gradient Boosting achieved R² = 0.93 for fatigue life prediction, closely matching the noise-ceiling of 0.96.
• Stacking Ensemble achieved R² = 0.87 for fracture toughness prediction, reaching 89% of the noise-ceiling (0.98).
• Feature importance analysis highlighted composite indicators, stress amplitude, and fiber length as key predictors.

Conclusions:

• The developed ML framework effectively reconstructs fatigue life and fracture toughness, acting as accurate automated surrogates for governing equations.
• The study demonstrates the potential of ML in predicting composite properties, reducing the need for extensive empirical testing.
• The proposed pipeline serves as a methodological template for future machine learning applications in composite materials research.