In-situ laser powder bed fusion: real-time assessment of residual stress through thermal gradient analysis

Metal additive manufacturing (AM) processes have garnered significant attention due to their ability to enhance design flexibility and manufacturability. However, the rapid heating and cooling inherent in these processes often lead to deviations in the as-built properties, diverging from the desired outcomes. Residual stress emerges as a critical factor contributing to these deviations, posing a risk of build failure and adversely affecting the functionality of fabricated parts. Despite advancements in in-situ monitoring sensors, which allow for a closer examination of process physics, including temperature gradients and melt pool geometries, the quantitative relationship between these observed physical phenomena and residual stress remains underdeveloped. The inability to measure and analyze real-time residual stress hinders effective control of as-built properties. To address these challenges, this study proposes a data-efficient computational machine learning model integrated with process-related physical phenomena. The main novelty of the proposed model is its use of as-built top surface thermal data to predict overall as-built part residual stress during the real-time fabrication process. In the powder bed fusion process, the majority of the as-built volume is buried in powder, making it difficult to measure directly, and the continuous addition of material on top results in changing residual stresses over time in the unseen volume. The developed model has an average RMSE value equals to 0.065 and correlation factor equals to 0.92. By establishing a quantitative link between observed in-situ phenomena and residual stress, the proposed framework facilitates a deeper understanding of the metal AM process.