Evaluation of fatigue properties in flame-resistant magnesium alloy welded joints by crystal plasticity analysis and data science approach

Abstract

The effect of weld geometry on fatigue properties of flame-resistant magnesium alloy welded joints was evaluated using computational and data science methods. First, a workflow for fatigue life prediction of flame-resistant magnesium alloy welded joints was proposed. The workflow consists of thermo-elastic-plastic analysis simulating welding, macroscopic stress field analysis to identify stress concentration zones, crack initiation analysis using crystal plasticity analysis and Tanaka-Mura model, and crack propagation analysis using finite element method and Paris’ law. The fatigue life calculations were repeatedly performed by using the proposed workflow for different shapes of weld toe, excess weld metal and undercut, and the results were compiled into a database. A surrogate model was developed from the database by machine learning to predict the fatigue indicator parameter (FIP), which is an index of crack initiation life obtained by the Tanaka-Mura model. The effect of weld geometry on fatigue life was evaluated by exploring the surrogate model using a Markov chain Monte Carlo method. The results obtained are consistent with the previous findings that toe radius and undercut depth have a strong influence on fatigue.