Data-Driven Prediction of Mechanical Properties in 3D-Printed Composites Using Hybrid Machine Learning Models

Authors

• Rakesh Mittapally Business Intelligence Architect/AI and ML Engineer

Abstract

The study investigated the advancement of 3D printing technology has revolutionized manufacturing across multiple industries, with composite materials playing a crucial role in enhancing material properties and expanding technological capabilities. This study investigates the intricate relationships between 3D printing process parameters and the mechanical properties of composite materials, employing advanced machine learning techniques to predict and optimize tensile strength. The research explores the impact of key printing parameters: printing speed, nozzle temperature, and filler material percentage on the tensile strength of 3D-printed composite materials. A comprehensive experimental dataset was collected, analyzing 30 different printing configurations to understand their effects on material performance. Three machine learning regression models were evaluated for predictive accuracy: AdaBoost Regression, Multilayer Perceptron (MLP) Regressor, and Gaussian Process Regressor. Each model was trained and tested to predict tensile strength based on input parameters. Correlation analysis revealed a strong positive relationship between filler material percentage and tensile strength, with a correlation coefficient of 0.94.The correlation heatmap and descriptive statistics highlighted complex interactions between printing parameters. Printing speed showed a moderately negative correlation with tensile strength (-0.54), while nozzle temperature demonstrated minimal direct influence. Performance metrics revealed significant challenges in model generalization.

The AdaBoost Regression model showed the most stable performance, with the Gaussian Process Regressor and MLP Regressor struggling to generalize beyond training data. This underscores the complexity of predicting composite material properties and the need for sophisticated modeling approaches. The study contributes to the understanding of 3D printing composite materials by demonstrating the potential of machine learning in predicting and optimizing material characteristics. The findings offer insights into process parameter optimization, highlighting the critical role of careful parameter selection in achieving desired mechanical properties. Future research should focus on improving model generalization, expanding the dataset, and exploring advanced machine learning techniques to enhance predictive accuracy in composite material 3D printing.