Evaluation of machine learning models in predicting mechanical properties of additive-manufactured parts

Different machine learning (ML) models have been applied in predicting mechanical properties of additive-manufactured parts. However, the existing research primarily focuses on individual mechanical properties and overlooks interrelations among multiple properties and their combined responses to additive manufacturing (AM) process parameters. This research investigates ML models for predicting ultimate tensile strength (UTS), compressive strength (CS), and Young's modulus (YM) of fused filament fabricated parts. Gaussian process regression (GPR), support vector machines (SVMs), neural networks, and linear regression models are evaluated. GPR achieved the highest performance for UTS and CS with R 2 values of 0.95 and mean absolute errors (MAE) of 0.78 and 2.4 MPa, respectively. For YM, SVM performed best with an R 2 of 0.90 and an MAE of 23 MPa. Shapley Additive Explanations analysis reveals that printing temperature and infill density are the most influential parameters for UTS and CS, respectively, while YM is primarily affected by infill density and wall thickness. These findings highlight the critical role of AM process parameters, such as wall thickness, layer height, infill density, print speed, and print temperature, in determining the part performance, and demonstrate the potential of ML models to effectively predict mechanical properties of 3D-printed components.