This study aims to optimize the geometric accuracy of thin-walled components manufactured by fused filament fabrication using high-temperature polylactic acid, focusing on the influence of process parameters on critical geometrical features.
A full factorial experimental design (108 conditions) was adopted to investigate the effects of nozzle diameter, printing speed, temperature and build orientation on five geometrical attributes (eccentricity, angularity, spacing, straightness and parallelism) across three wall profiles. Dimensional data were extracted via image-based metrology and analyzed using statistical hypothesis testing and ensemble learning models (Random Forest and Gradient Boosting).
Eccentricity and spacing were accurately predicted by machine learning models (R² > 0.86), with nozzle size and print speed identified as the most influential parameters. Angularity showed moderate predictability but was statistically controlled under specific settings. Straightness and parallelism were better explained through hypothesis testing because of high-frequency deviations. Optimal parameter sets were identified for each geometrical feature.
This work integrates image-based dimensional metrology, rigorous statistical validation and interpretable machine learning to model multiple geometric targets simultaneously. The proposed hybrid framework advances predictive control in fused filament fabrication of thin-walled structures and supports process optimization for precision-demanding applications in aerospace, biomedical and structural prototyping.
