Data-driven investigation of elastoplastic and failure analysis of additively manufactured parts under bending conditions

• Comprehensive analysis of elastoplastic and failure behavior of Ti-6Al-4 V alloy components fabricated via additive manufacturing • Conducts bending tests on LPBF-fabricated Ti-6Al-4 V specimens, analyzing elastic properties and varying sample dimensions, demonstrating the methodology’s applicability across diverse scenarios. • Utilizes Deep Neural Networks to predict material behavior, validated through k-fold cross-validation, and optimized using Genetic Algorithms to minimize root-square normalized error between simulations and experimental data. This study presents an advanced investigation the elastoplastic and failure behavior under bending condition applied to a Ti-6Al-4 V alloy part fabricated by Laser Powder Bed Fusion (LPBF) techniques. The objective is to develop a robust framework to accurately predict the material behavior of Ti-6Al-4 V alloy under bending loads by integrating the Finite Element Method (FEM), Deep Neural Networks (DNN), and Genetic Algorithms (GA) as an intelligent data-driven system. The prediction model incorporates many input components, including elastic modulus, Poisson’s ratio, Swift hardening model parameters, and modified GTN fracture model coefficients. The DNNs are trained using data generated from FEM simulations. Model evaluation is performed through k-fold cross-validation, ensuring robust performance assessment. GA are employed to optimize the coefficients of the elastic, plastic, and fracture models, minimizing the root-square normalized error (RSNE) between simulation results and experimental data. If the required error threshold is not achieved in an iteration, the process continues automatically, incorporating new data until the desired accuracy is reached. The findings demonstrate a successful characterization of elastic, plastic, and fracture-related properties, highlighting the capability of the proposed methodology to accurately predict the material behavior of components manufactured by various techniques under different conditions. This approach highlights the potential for extending the methodology to other materials and manufacturing processes, enabling precise prediction of material behavior in diverse applications.