Comparative Analysis of Machine Learning Models for Predicting the Mechanical Behavior of Bio-Based Cellular...

Area of Science:

• Materials Science
• Computational Mechanics

Background:

• Growing demand for sustainable materials drives interest in biocomposites.
• Machine learning (ML) offers potential for understanding biocomposite mechanical behavior efficiently.

Purpose of the Study:

• Investigate the mechanical behavior of biocomposite sandwich structures under compression.
• Evaluate the effectiveness of ML algorithms in predicting load-bearing capacities based on geometric variations.

Main Methods:

• Experimental mechanical tests were conducted on biocomposite sandwich structures under quasi-static out-of-plane compression.
• Three ML models were evaluated: Generalized Regression Neural Networks (GRNN), Extreme Learning Machine (ELM), and Support Vector Regression (SVR).
• Model performance was assessed using R-squared (R²), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE).

Main Results:

• The GRNN model exhibited the highest predictive accuracy, with excellent performance on both training and testing datasets (e.g., R² of 0.9999 for training, 0.9993 for testing).
• ELM showed moderate performance, while SVR demonstrated the lowest accuracy, indicating limited effectiveness.
• GRNN successfully captured the nonlinear load-displacement behavior, including critical peaks and fluctuations, with strong generalization capabilities.

Conclusions:

• Advanced ML models, particularly GRNN, can accurately predict the mechanical behavior of biocomposites.
• This predictive capability enables more efficient and cost-effective development and optimization of sustainable materials.
• The study highlights the increasing utility of ML in materials science for understanding complex mechanical responses.