Comparative Study of Machine Learning Approaches for Predicting Creep Behavior of Polyurethane Elastomer

Area of Science:

• Materials Science
• Polymer Science
• Computational Materials Science

Background:

• Long-term mechanical properties, particularly creep, are critical for viscoelastic polymers like polyurethane elastomers.
• Understanding creep behavior is essential for predicting material performance and ensuring structural integrity over time.
• Traditional methods for creep analysis can be time-consuming and complex.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the creep properties of polyurethane elastomers.
• To investigate the influence of creep time, temperature, stress, and material hardness on creep behavior.
• To compare the predictive performance of different machine learning algorithms for creep analysis.

Main Methods:

• Utilized multilayer perceptron network, random forest, and support vector machine regression models.
• Employed genetic algorithm and k-fold cross-validation for hyper-parameter optimization.
• Trained and tested models using experimental data on polyurethane elastomer creep.

Main Results:

• All three machine learning models demonstrated excellent fitting ability on the training dataset.
• Models exhibited varying prediction capabilities on the testing set, highlighting sensitivity to different factors.
• Correlation coefficients between predicted and experimental strains exceeded 0.913 (often >0.998) on the testing set.

Conclusions:

• Machine learning offers a powerful approach for predicting the long-term creep properties of viscoelastic polymers.
• The developed models provide accurate estimations of creep behavior under various conditions.
• This research facilitates more efficient and reliable material performance assessment for polyurethane elastomers.