An investigation of the ensemble machine learning techniques for predicting mechanical properties of printed parts in additive manufacturing

Abstract

This study investigates the ensemble machine learning models to predict the mechanical properties of the 3D-printed Polylactic Acid (PLA) specimens. We studied the effects of five process parameters, including the build orientation, infill angle, layer thickness, printing speed, and nozzle temperature, on the printed parts tensile strength and surface roughness. Machine learning models are developed using the experimental data collected from the printed 27 specimens. Gradient Boosting Regression, Extreme Gradient Boosting Regression, Adaptive Boosting Regression, Random Forest Regression, and Extremely Randomized Tree Regression models were developed during the machine learning modeling stage to predict the surface roughness and tensile strength of the printed parts. This research demonstrates the effectiveness of Extremely Randomized Tree Regression model in providing accurate tensile strength predictions with root mean square error (RMSE) of 1.03, mean absolute error (MAE) of 0.82, and mean absolute percentage error (MAPE) of 2.20%. Similarly, Random Forest Regression model shows better accuracy in predicting surface roughness having RMSE of 0.408, MAE of 0.31, and MAPE of 9.28%. Moreover, the comparative study confirms that ensemble machine learning techniques are more useful than the traditional support vector and k-nearest neighbor machine learning models for predicting the surface roughness and tensile strength of the printed parts. The results highlight a novel approach of using ensemble machine learning models in identifying complex correlations in the dataset, establishing the foundation for improved product design and mechanical property optimization through adjustment of the process parameters combination.