Machine learning-driven prediction of manufacturing parameters and analysis of mechanical properties of PC/ABS specimens produced by the fused deposition modeling additive manufacturing method

In this thesis, the tensile strength, flexural strength, and impact energy of PC/ABS specimens manufactured by material extrusion additive manufacturing were investigated, together with the sensitivity of these mechanical properties to key process parameters such as printing speed, infill density, and raster angle. Specimens produced under parameter combinations determined via a Taguchi-based experimental design were subjected to tensile, flexural, and impact tests, and the resulting data were evaluated not only through direct comparisons but also by analysis of variance. The results revealed that infill density has a dominant effect on tensile and flexural strength, while the interaction between printing speed and infill density plays a decisive role for impact energy. In the second stage of the study, multi-output regression models based on machine learning, including Random Forest, k-Nearest Neighbors, and Multilayer Perceptron, were employed to simultaneously predict the mechanical properties, and model performance was tested after hyperparameter tuning with Optuna. The findings indicate that non-linear model structures provide high accuracy particularly in predicting tensile and flexural strength and can be used as decision-support tools for optimizing process parameters; thus, a holistic framework is presented that integrates experimental and data-driven approaches to characterize the mechanical behavior of PC/ABS in the considered manufacturing process.