Abstract
Additive manufacturing (AM) part quality relies on many factors, including part geometry that impacts both the manufacturability and resulting dimensional accuracy of the part. To improve the dimensional accuracy of AM parts, data-driven approaches can be utilized to explore the effect of different process parameters on both simple and complex geometries. However, to provide general design guidelines, it is necessary to develop models and tools that accurately predict geometry-driven distortion across a broad range of geometries, while also being user-interpretable. Identifying and analyzing common part features that contribute to geometrical deviations and using them to design better parts could improve AM part quality. In this paper, a Gaussian process regression surrogate model was trained using 21 geometric features (selected from a set of 92 shape descriptors) from 324 different axisymmetric parts to predict maximum part distortion and identify the features that impact part distortion the most. Validated high-fidelity finite element analysis simulations were used to determine the maximum distortion corresponding to each part. Our results show the surrogate model approach can accurately predict part distortion, with a predictive error of approximately 0.07 mm for the testing set. The findings of this study can have implications for the exploration of new part designs by adjusting these identified features or incorporating them as design rules in AM product designs.