Abstract
Polymer additive manufacturing (AM) has grown rapidly in the past decade, with material extrusion, vat photopolymerization, powder bed fusion and jetting now widely used for functional polymer parts. The mechanical performance of these parts depends strongly on process parameters such as layer height, build orientation, energy input and post-processing conditions, which motivate the development of predictive models for process-property relationships. Classical approaches based on Taguchi designs, ANOVA and response surface methodology have provided valuable insight, but the potential of modern machine learning (ML) techniques is not yet fully exploited. This review surveys recent work on ML-based prediction of mechanical properties of polymer AM parts using process parameters as inputs. Across the literature, well-tuned artificial neural networks, tree-based ensembles and support vector regression typically achieve prediction errors below about 5-10% for strength and modulus, showing that data-driven surrogates can substantially reduce experimental trial-and-error in process optimization. Ongoing challenges include small datasets, missing standardized error metrics, and limited coverage of non-quasi-static phenomena like fatigue, impact, and environmental degradation.