Integration of artificial neural network with finite element analysis for residual stress prediction of direct metal deposition process. (June 2021)

• Record Type: Journal Article
• Authors: Hajializadeh, F.; Ince, A.
• Is Part Of: Materials today communications. Volume 27(2021)
• Publication Date: 2021-06
• Subjects: Additive manufacturing -- Direct metal deposition -- Residual stress -- Machine learning -- Neural network -- Finite element analysis; Materials science -- Periodicals; 620.11
• DOI: 10.1016/j.mtcomm.2021.102197
• Languages: English
• ISSNs: 2352-4928

Abstract

Graphical abstract: Highlights: An integrated modeling framework of Artificial Neural Network (ANN) and finite element (FE) is proposed. ANN-FE is used to predict residual stress distributions of Direct Metal Deposition (DMD) process. Three different DMD geometric parts made of AISI 304 L are used for the model validation. ANN-FE is capable of accurate and efficient prediction of residual stress distributions of DMD parts. Abstract: Direct Metal Deposition (DMD) process is considered to be an efficient and reliable manufacturing method for the production of complex parts in many design applications. The thermo-mechanical nature of the DMD process induces a significant amount of residual stresses and distortions on fabricated parts. Evaluation of residual stress distributions requires considerable amount of modeling and experimental works. Therefore, there is a need for an accurate, and feasible assessment method(s) for engineers to estimate residual stresses based on chosen process parameters and geometrical features of DMD built parts. A novel artificial neural network-based modelling approach integrated with finite element analysis is proposed to address shortcomings of conventional thermo-mechanical finite element-based models and improve the computational efficiency of predicting residual stresses of AISI 304 L parts built on the basis of the DMD process. Predicted results showed that the novel approach is capable of accurate and efficient prediction of residual stress distributions of three different geometric structures e.g. a plane wall shape, L-shape wall, and rectangular box structures. Furthermore, the computational time of predicting the residual stresses for the wall, L-shape wall, and rectangular box structures is significantly improved with respect to the classical finite element thermo-mechanical analysis.