Machine learning-based pre-print build adhesion prediction in FDM using STL geometric features

Build adhesion failure is one of the major causes of defects and print loss in fused deposition modelling (FDM). This study proposes an automated framework to extract layer-wise geometric features from STL models and predict build adhesion requirements before printing begins. A Python-based graphical user interface was developed using Tkinter and STL libraries to obtain manufacturing-relevant parameters, including volume, dimensions, number of layers, first-layer area, last-layer area, and centre of mass. A dataset containing 450 STL models was generated and used to train multiple machine learning classifiers. Comparative evaluation showed that Decision Tree, Random Forest, and Gradient Boosting classifiers achieved the highest predictive performance on the evaluated dataset, while other models also demonstrated competitive classification capability. Feature importance and correlation analyses were conducted to identify key predictors influencing adhesion behaviour. The proposed approach enables pre-print decision support, reduces trial-and-error parameter tuning, and improves print success rates, particularly for low-cost 3D printers.