Statistical Beam-Shaping Melt Pool Modeling

Abstract

Laser powder bed fusion is one of the most popular metal additive manufacturing methods. It uses a high-intensity laser to selectively melt fine layers of metal powder and creates parts with complex geometries. Conventional laser powder bed fusion uses a Gaussian profile laser as it is the most mature laser type. However, due to the high energy concentration, Gaussian beams may create a melt pool with high volatility and a strong keyhole, which leads to problems such as high residual stress and porosities. Recently, beam-shaping lasers have been investigated for attempting a more uniform melt pool and controllable thermal gradient. However, varying beam shapes can lead to complex melt pool thermal distributions, making thermal gradients challenging to describe. In this work, novel statistical models are applied to represent thermal gradient distribution in complex scenarios. By leveraging machine learning techniques, the statistical thermal gradient model parameters can be predicted with process parameters like laser power, scanning speed, and beam shape as inputs. The machine learning model that can achieve this prediction is trained by finite element analysis results. The analysis demonstrates that the machine learning model can predict the parameters of the twin-gaussian thermal gradient model parameter with significantly high R-squared values (R2 > 0.9). Such information can be used to estimate the dominant microstructure with higher accuracy.