Residual stress computation in direct metal deposition using integrated artificial neural networks and finite element analysis

Abstract

Direct Metal Deposition (DMD) as a part of additive manufacturing processes, utilizes a laser heat source to deposit metallic material to construct components layer by layer, allowing for controlled manufacturing. DMD processes inherently induce residual stresses and distortions in the build part. While finite element (FE) analysis is one of the approaches to predict the residual stress, the high computational time required for the FE analysis makes it undesirable for the early stages of product development. This study extends a previously developed approach to more complex structures to evaluate the accuracy and computational efficiency of predicting residual stresses. For the present study, six different metallic structures with various complex shapes made from AISI 304 L were considered. Comparing the residual stress fields predicted by the integrated ANN-FE with thermomechanical FE analysis, the results demonstrated consistent distributions for all stress components across structures. Although localized regions exhibited higher errors, these areas coincided with low-stress zones, posing no critical concerns. Despite higher errors in some areas, most structures displayed prediction errors below 15%. Moreover, the analysis achieved a substantial computational efficiency enhancement of approximately 6 times for all structures analyzed.