Towards Digital Twin in Additive Manufacturing: A Multi-fidelity Approach for Enhancing LPBF Process Modeling with In-situ Data

Process modeling for additive manufacturing is essential to enable fully digital optimization workflows and reduce dependence on experimental studies to identify design flaws such as overheating, excessive thermal stress, and deformation. However, accurate track-level predictions in Laser Powder Bed Fusion (LPBF) demand significant computational resources and detailed material parameters, which are difficult to measure across the broad temperature ranges involved in the process. While post hoc calibration has been widely used to enhance model accuracy by aligning predictions with measurable outcomes, it still relies on manufacturing physical parts for validation. This study bridges the gap between physics-based simulation models and real-world processes by developing a multi-fidelity approach that incorporates in-situ data to enhance model accuracy. By employing machine learning to integrate simulations with real-time observations of temperature evolution, the multi-fidelity model improves predictive capabilities while reducing the experimental effort for first-time right production. This represents the first step towards the development of an AM digital twin for dynamically adjusting process parameters and controlling the printing process.