The research of surface roughness prediction with machine learning according to process parameters in laser powder bed fusion

Abstract

Additively manufactured metal components are preferred in the industries that need to be reached high quality and durability. Although there are many advantages of freedom to design complex shapes and improved mechanical properties, low corrosion resistance and fatigue life still remain as challenges. One of the important factors of the challenges is surface roughness. A rough surface would wear and corrode more quickly than a smoother one. Especially, a down skin in an overhang structure has the highest surface roughness among the entire parts. To avoid rough surfaces, not only surface post processing but proper selection of process parameters is important. In this research, to decrease the down skin surface roughness of components fabricated by laser powder bed fusion (L-PBF), prediction of down skin surface roughness is conducted with machine learning algorithms in the step of selecting process parameters. The inputs of the data driven approach contain laser power (LP), scanning speed (SS), layer thickness (LT), hatching distance (HD) which are the main four factors in the L-PBF process, and overhang angles of specimens. Finally, through the data driven models, results show learning models are capable of predicting the down skin surface roughness of metal components fabricated.