Machine learning-driven optimization for surface roughness prediction of vertical orientation measurements on 3D printed components

Surface roughness accuracy can be challenging in Additive Manufacturing (AM) because traditional point-based measurements often fail to capture the full range of surface characteristics. This study investigates how machine learning (ML) can improve roughness prediction by integrating image-based analysis with real-time adaptive control. A full factorial experimental design examines the effects of infill density, print speed, nozzle temperature, and layer height on the surface roughness of vertically oriented printed parts. A dataset of 81 experiments, including roughness image data, is used to test and validate the XGBoost model. The proposed model outperforms traditional regression methods, achieving an R${}^2$ of 97.06% and a Mean Squared Error (MSE) of 0.1383, compared to 95.72% and 0.224 of the conventional approach. These results demonstrate the potential of ML-driven optimization to significantly improve precision and consistency in AM, aerospace, healthcare, and automotive industries.