Data-driven prediction of hole geometry deviation and circularity in SLS-fabricated PA12 parts after chemical vapor smoothing using machine learning

Selective Laser Sintering (SLS) of polyamide 12 (PA12) is widely used to manufacture functional components with complex geometries; however, achieving dimensional accuracy in hole features remains challenging because of anisotropic shrinkage, non-uniform cooling, and post-processing effects. This study presents an experimental and data-driven investigation of hole geometry in SLS-fabricated PA12 components after chemical vapor smoothing (CVS). PA12 samples with different inner and outer hole diameters were fabricated using five exposure cycles and three build orientations (X, Y, and Z). A fixed CVS treatment was applied to all samples under standard processing conditions. Hole diameter deviation and circularity were measured and used to develop supervised machine learning models, including polynomial regression, support vector regression, random forest, gradient boosting, XGBoost, and LightGBM. The models were comparatively evaluated to predict hole diameter deviation and circularity using exposure cycle, build orientation, and nominal hole dimensions as input variables. The results showed that exposure cycle and build orientation influenced hole geometry, with Z -oriented samples exhibiting larger dimensional deviations. Although CVS improved edge smoothness and continuity, slight dimensional variations were observed after treatment, depending on feature size, build orientation, and prior thermal exposure. Ensemble-based models provided the highest prediction accuracy, with the optimized XGBoost model achieving R 2 values of 0.909 for circularity (RMSE = 0.011, MAE = 0.009) and 0.911 for diameter deviation (RMSE = 0.047, MAE = 0.034). The findings demonstrate the potential of data-driven models for feature-level prediction of hole geometry under the investigated processing conditions.