The optimization of 3D printed PLA/EPO/Lignin Biocomposite’s tensile strength using three different supervised learning applications

Abstract

Developing a sustainable and biodegradable biocomposite for 3D printing necessitates iterative experimentation to achieve the desired composition and properties. Combining polylactic acid (PLA), epoxidized palm oil (EPO), and lignin offers a promising formulation for a 3D biocomposite filament with diverse potential applications. To optimize its performance specifically for 3D printing applications, the use of machine learning techniques can expedite the process of property optimization. Employing three distinct machine learning models, namely random forest (RF), support vector regression (SVR), and artificial neural network (ANN), facilitates a comparative analysis to determine the most effective approach for predicting the tensile strength values across different biocomposite compositions. Through model training and evaluation, the comparison reveals that both RF and SVR demonstrate superior accuracy compared to ANN. Notably, RF exhibits exceptional consistency, boasting an average R2 score of 0.9777 and an average mean squared error of 1.5475. SVR follows closely with an average R2 score of 0.9777 and an average mean squared error of 7.7751, while ANN lags behind with an average R2 score of 0.6551 and an average mean squared error of 117.5218. Assessing the performance of these machine learning models underscores their potential applicability in enhancing the production of biocomposite filaments for 3D printing, thereby facilitating the refinement of biocomposite properties.