Machine learning and materials informatics approaches for evaluating the interfacial properties of fiber-reinforced composites

Fiber pullout tests have been frequently performed to determine the interfacial properties of fiber-reinforced composites. However, traditional experimental approaches and numerical investigations are restrained by being both labor-intensive and time-consuming. Hence, an accurate and effectual prediction of the interfacial properties is of paramount importance for composite design and tailoring. This work for the first time presents machine learning-assisted models to determine the interfacial properties based on previous micro-bond tests. Through a comparison between the pullout test results and prediction results, the effectiveness of the proposed model in the prediction of the interfacial shear strength and the maximum force is verified. The relationship between influencing attributes and interfacial properties can be reliably captured. It can be referred from the mean impact value analysis of the proposed models that the interfacial properties are significantly dependent on the fiber's diameters. This work reveals that gradient boosting regressor (GBR) and artificial neural networks (ANN) exhibit adequate generalization and interpretation abilities. Besides, both ANN and GBR, with small datasets, have tremendous potential for a wide array of applications in predicting the shear resistance properties in fiber-reinforced composites.

Automated summary

The study introduces machine learning-assisted models to predict the interfacial properties of fiber-reinforced composites for the first time, and the effectiveness of the model is verified through comparison between experimental and prediction results. It is found that the interfacial properties are significantly dependent on the fiber's diameter.