Digital twin-driven optimization of laser powder bed fusion processes: a focus on lack-of-fusion defects

PurposeThe purpose of this research is to enhance the Laser Powder Bed Fusion (LPBF) additive manufacturing technique by addressing its susceptibility to defects, specifically lack of fusion. The primary goal is to optimize the LPBF process using a digital twin (DT) approach, integrating physics-based modeling and machine learning to predict the lack of fusion.Design/methodology/approachThis research uses finite element modeling to simulate the physics of LPBF for an AISI 316L stainless steel alloy. Various process parameters are systematically varied to generate a comprehensive data set that captures the relationship between factors such as power and scan speed and the quality of fusion. A novel DT architecture is proposed, combining a classification model (recurrent neural network) with reinforcement learning. This DT model leverages real-time sensor data to predict the lack of fusion and adjusts process parameters through the reinforcement learning system, ensuring the system remains within a controllable zone.FindingsThis study's findings reveal that the proposed DT approach successfully predicts and mitigates the lack of fusion in the LPBF process. By using a combination of physics-based modeling and machine learning, the research establishes an efficient framework for optimizing fusion in metal LPBF processes. The DT's ability to adapt and control parameters in real time, guided by machine learning predictions, provides a promising solution to the challenges associated with lack of fusion, potentially overcoming the traditional and costly trial-and-error experimental approach.Originality/valueOriginality lies in the development of a novel DT architecture that integrates physics-based modeling with machine learning techniques, specifically a recurrent neural network and reinforcement learning.