Prediction of fatigue life of metallic structures with welded joints using automatic learning systems

Welded metallic joints are prone to fatigue damage, which may lead to sudden and catastrophic structural failure. In this research, fatigue failures of metallic structures with welded joints are analyzed using an approach based on automatic learning technology. A database of physics-based parameters, including material properties, loading histories, and stresses around potential cracking sites, is constructed based on experimental results and numerical analyses. Various automatic learning tools are used to search for the mathematical formulas and data patterns embedded in the database. The obtained rules and formulas can be used to support design of welded metallic structures. This approach provides a new way to locate fatigue-prone areas, predict fatigue lives, and may lead to designs of more fatigue resistant structures. It complements the classical deterministic and statistical fatigue failure predictions.