A Comparative Study of Machine Learning Techniques for Predicting Mechanical Properties of 3D-Printed Hybrid TPU/PLA Hybrid Composites

Abstract

Thermoplastic Polyurethane (TPU) and Polylactic Acid (PLA) are mostly used thermoplastics due to the special mechanical properties in Fused Deposition Modelling (FDM). PLA is brittle and hard, unlike TPU, which is extremely flexible, tough, and impact-resistant. Although additive manufacturing, materials development, optimization of processes, and the application of machine learning have made great progress, the majority of the current literature deals with single-material systems or homogeneous composites. Even though mixed composites and multi-material methods were explored, no study was done on structured layered hybrid composites with particular reference to the use of TPU alternating with PLA as a layer. Furthermore, ML models have been widely used in prediction, but a large-scale comparative study of layered hybrid TPU/PLA composites has not been conducted. This research focuses on the PLA/TPU alternate layered hybrid composite in an equal (1:1) ratio to acquire a balanced strength and ductile factor to be used in advanced applications in additive manufacturing. The mechanical properties of FDM-fabricated specimens, like Young’s Modulus (E or EM) and Tensile Strength (TS or TES), are found at various process parameters. The parameters considered for the Taguchi L27 experimental design are the Thickness of the Layer (LT), the Printing Speed (PS), the Temperature of the Nozzle (TPU/PLA NT), the number of Shell/Wall Layers (SL), and the Bed Temperature (BT) as parameters. The experiment was repeated three times, which gave evidence of the accuracy and repeatability of the experiment and obtained 27 experimental results. The experimental results indicate that the layer thickness and nozzle temperature are the most significant parameters that influence the mechanical performance. This is due to the fact that the thinnest layer has the most interlayer bonding and the highest tensile strength of 40.795 MPa and Young’s modulus of 1.188 GPa. With faster printing speed, the mechanical properties are reduced due to a reduction in interlayer adhesion. The tensile strength is most sensitive to the layer thickness. The nonlinear correlations between the process parameters and the mechanical properties are very complicated and are modelled using Machine Learning (ML) techniques such as Linear Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Adaptive Boosting (AdaBoost) and eXtreme Gradient Boosting (XGBoost). XGBoost has the highest predictive capacity with an R2 of above 0.99 for tensile strength and Young’s modulus, among others.