Predicting the roughness of overhanging surfaces in laser powder bed fusion via in-situ thermal imaging

The production of overhanging surfaces in Laser Powder-Bed Fusion (LPBF) has long been a challenging task due to poor heat dissipation and lack of support of loose powder, resulting in surface defects and increased roughness due to dross formation and sintering. Surface quality is a critical aspect of AM mechanical components that undergo fatigue loading, as a rough surface can act as a preferential crack initiation site and lead to premature failure. Predicting the quality of the as-built surfaces could be used to identify critical areas that require rework or post-processing, or to find regions that require optimization of the process parameters to improve the final quality. The orientation of the surface itself (i.e., the degree of inclination of the surface) could be used to predict the final surface quality and will be employed as benchmarking reference throughout the work (referred to as "geometry-based" model). This study demonstrates the effectiveness of using data mining on high-speed thermal video images to create a real-time predictive model based on "in-situ" data for estimating surface roughness (Sa) of overhanging surfaces printed at different inclinations. The results showed that the model based on "in-situ" data has a prediction accuracy that is more than 2 times higher than the one obtained with a model that is purely based on geometric data, i.e., a model that relies only on the inclination angle of the surface during the print. The proposed method is tested on different materials (AISI 316L stainless steel and AlSi10Mg) and process conditions (continuous and pulsed laser, low and high power) to show the flexibility and extended applicability of the proposed solution. The newly developed method opens new possibilities for in-situ quality control and process optimization of surface quality in Laser Powder Bed Fusion (LPBF).