Abstract

Laser-based powder bed fusion (LPBF) technology is one of the most applied additive manufacturing pro-cesses owing to, among others, its capacity of producing parts with mechanical properties superior to conventionally processed counterparts. Whereas to obtain full-dense components, the proper selection of processing parameters is mandatory and well explored, there is a gap in comprehending the influence of processing parameters on the resulting surface hardness and wear resistance. In this work, the effect of laser power, scanning speed, layer thickness, hatch distance, and material density on these properties is evaluated for materials commercially employed in LPBF. A machine learning-aided interpretable model is developed, featuring gradient boosting techniques (gradient boosting regressor (GBR), extreme gradient boosting regressor (XGBR), and AdaBoost) trained and evaluated by 5-fold cross-validation for the pre-diction of microhardness analyzed for literature data specific to selective laser melting of a variety of alloys and metal-based composites. Gaussian process regression is used to evaluate the wear rate, employing the testing parameters to learn the wear behavior, and interpreted in the context of an analytical model. Feature importance analysis has been carried out to understand the complex interactions during the pin-on-disc test. The trained models achieved high predictive performance (R2> 0.96) for wear rate prediction, con-sistent with mechanistic understanding, posing machine learning as a powerful tool for LPBF process design with minimum experimental effort in calibration. (c) 2023 CIRP.