Integrated computational framework for predicting surface profile of laser powder bed fusion stainless steel SUS420 parts after laser polishing

The surface roughness of laser powder bed fusion (L-PBF) parts is one of the most important criteria governing their quality and fitness for use. The surface quality of L-PBF components is traditionally improved through mechanical polishing. However, in recent years, laser polishing (LP), in which laser radiation is used to melt a thin layer of metal on the component surface, has attracted increasing attention. Accordingly, a simulation model is considered the potential approach to explore insights into the process and to optimize the LP processing conditions. Previous studies which focus on building simulation for the LP process neglected the effect of the initial surface profile on the absorption of laser energy as laser irradiates the surface. In an effort to address this issue and to create a foundation for optimizing the processing conditions in LP, in this work, a new computational framework is proposed for predicting the surface profile of L-PBF parts following the LP process. For taking into account the effect of initial L-PBF processed roughness on the laser absorption, the proposed approach commences by using a mathematical model to reconstruct the measured surface profile of the printed part. Ray-tracing simulations are then performed to investigate the interaction between the laser radiation and the surface profile and to evaluate the absorption of the metal surface. Finally, heat transfer simulations, fast Fourier transform analyses, and a low-pass filter algorithm are employed to simulate the final surface profile of the L-PBF component following laser polishing. It is found that the predicted values of mean roughness (R_a) and root mean square (R_q) differ from experiment by no more than 7.7% and 9.8%, respectively. Additionally, the simulated correlation length (cl) which is the parameter illustrating the variation of height in the lateral direction is also compared with one attained from experiment. It is observed that the maximum error between simulated correlation length and one attained from the experiment is no more than 13.4%.