A new data-driven framework for prediction of molten pool evolution and lack of fusion defects in multi-track multi-layer laser powder bed fusion processes

Abstract

This study proposes a new data-driven approach to predict the area fraction of lack of fusion and morphology of 3D-printed parts produced using the laser powder bed fusion process. Different processing parameters were considered to develop this methodology, such as laser power, laser speed, hatch spacing, and layer thickness. Melt pool geometrical parameters were numerically predicted using a mathematical model optimized using the Nelder-Mead optimization method and a machine learning model. The Gaussian process was later employed to further modify the layerwise deposition. The accuracy of the proposed methodology was then extended to predictions of the area fraction of lack of fusion and the morphological aspects of the melt pools in 3D-printed parts by comparing them with the experimental observations. The results revealed that this data-driven approach, which is independent of the thermophysical properties of the material, is capable of reasonably predicting the area fraction of lack-of-fusion formed during material deposition using the laser powder bed fusion process.