Model reduction for fatigue life estimation of a welded joint driven by machine learning

Abstract

In structural mechanics, the design of component assemblies is fundamental for ensuring structural integrity and durability. Fatigue, a common failure mode, particularly challenges the validation of welded joints' fatigue resistance. Various analytical and numerical methods estimate fatigue life but often involve costly processes requiring extensive parameter adjustments and software integration. This study applies machine learning (ML) for metamodeling of a complete finite element analysis and fatigue analysis workflow for estimating the fatigue life of welded joints, considering geometric characteristics, loading conditions, and weld classes. The comparison of the effect of learning database configuration for several regression estimators has led to a reduced model that provides design rules. Our findings demonstrate the significant potential of ML to streamline complex frameworks and accurately estimate the fatigue life of welded joints, advancing AI's application in mechanical engineering.

Keywords

• Metamodeling
• Machine learning
• Fatigue life
• Welded joints
• FEA structural mechanics