In Situ Monitoring Network for Deposition Morphology and Residual Stress Reconstruction

In laser metal deposition (LMD), complex thermo-mechanical coupling and irregular layer morphology significantly affect residual stress distribution. However, most simulations rely on idealized geometries, limiting prediction accuracy. This study proposes a data-driven framework integrating in situ vision-based morphology reconstruction with thermo-mechanical simulation for high nitrogen steel (HNS). An improved DeepLabv3+ network is developed to extract deposition layer contours under strong illumination and spatter interference, achieving a mean intersection over union (mIoU) of 97.32% and an overall accuracy of 99.42%. The reconstructed morphology is incorporated into a finite element model to enable dynamic heat source tracking and realistic geometric representation. The proposed method demonstrates high morphology reconstruction accuracy, with all measurement errors controlled within 0.91%. The simulated temperature field agrees well with experimental measurements. Furthermore, the predicted residual stress distribution is consistent with X-ray diffraction (XRD) results under different laser power conditions. The results indicate that local surface morphology significantly influences stress concentration, with protrusion regions exhibiting stress peaks up to 989 MPa, markedly higher than those in concave regions. This study improves the accuracy of residual stress prediction in LMD by incorporating real morphology data and provides insight into the relationship between morphological features and stress evolution in additively manufactured HNS components.