In-situ controller autotuning by Bayesian optimization for closed-loop feedback control of laser powder bed fusion process

Abstract

Laser powder bed fusion (LPBF) additive manufacturing (AM) traditionally relies on static parameter assignment in an open loop, which can lead to defects when faced with complex thermal histories and process variability. Closed-loop control offers a promising alternative that can enhance stability and mitigate defects. However, controller performance relies heavily on precise parameter tuning, a process that is typically manual and system-specific. This study employs Bayesian Optimization (BO) as an automated, sample-efficient method for tuning in-layer controllers in LPBF, leveraging the repetitive nature of the process for either online (in-process) or offline (pre-process) tuning. We experimentally apply BO to tune an in-layer PI controller to modulate laser power, assessing its performance on wedge geometries prone to overheating. The results show that BO significantly reduces overheating, outperforming uncontrolled settings. Notably, this study presents the first microstructural analysis of parts produced with in-layer controlled tuning, identifying lack-of-fusion porosities caused by the controller’s corrective adjustments. In summary, BO demonstrates strong potential for automated controller tuning in LPBF, with implications for broader applications in AM, suggesting a path towards more adaptive and robust control across diverse machines and materials.