Characterize traction-separation relation and interfacial imperfections by data-driven machine learning models.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Interfacial mechanical properties are critical for composite material performance in applications like aerospace and robotics.
• Accurate measurement of microscale interfacial interactions, specifically traction-separation (T-S) relations, is experimentally challenging.
• Machine learning (ML) offers a promising data-driven approach to model complex interfacial behaviors.

Purpose of the Study:

• To develop precise data-driven models for characterizing interfacial mechanical properties in composite materials.
• To investigate traction-separation (T-S) relations and identify interface defect locations using ML.
• To provide a validated, accessible framework for predicting interfacial behavior.

Main Methods:

• Integration of machine learning (eXtreme Gradient Boosting - XGBoost), finite element analysis (FEA), and empirical experiments.
• Training ML models on macroscale force-displacement curves derived from FEA and mechanical tests.
• Utilizing multi-output regression for T-S relations and classification for defect identification.

Main Results:

• Highly accurate prediction of T-S relations with R² = 0.988.
• Successful identification of interface imperfection locations with 81% accuracy.
• Experimental validation using 3D printed double cantilever beam specimens.

Conclusions:

• The developed data-driven models precisely characterize interfacial mechanical properties and defects in composites.
• The approach offers a robust and efficient method for evaluating structural reliability and failure criteria.
• A provided code package enables broader application of these ML models for diverse material interfaces.