Supervised Machine Learning Models for Mechanical Properties Prediction in Additively Manufactured Composites

Abstract

This paper analyses mechanical property prediction through Machine Learning for continuous fiber-reinforced polymer matrix composites printed using the novel Material Extrusion Additive Manufacturing technique. The composite is formed by a nylon-based matrix and continuous fiber (carbon, Kevlar, or fiberglass). From the literature, the elastic modulus and tensile strength were taken along with printing parameters like fiber content, fiber fill type, matrix lattice, matrix fill density, matrix deposition angle, and fiber deposition angle. Such data were fed to several supervised learning algorithms: Ridge Regression, Bayesian Ridge Regression, Lasso Regression, K-Nearest Neighbor Regression, CatBoost Regression, Decision Tree Regression, Random Forest Regression, and Support Vector Regression. The Machine Learning analysis confirmed that fiber content is the most influential parameter in elasticity (E) and strength (σ). The results show that the K-Nearest Neighbors and CatBoost provided the closest predictions for E and σ compared to the other models, and the tree-based model presented the narrowest error distribution. The computational metrics point to a size versus prediction time tradeoff between these two best predictors, and adopting the prediction time as the most relevant criterion leads to the conclusion that the CatBoost model can be considered, when compared to the others tested, the most appropriate solution to work as a predictor in the task at hand.

Keywords

• additive manufacturing
• mechanical properties
• composites
• machine learning