A damage-tolerant approach for fatigue in cold-formed profiles combining numerical modelling and machine learning

A comprehensive damage-tolerant design framework for fatigue in cold-formed thin-walled mild steel profiles is presented, combining high-fidelity finite element analysis for stress intensity factor determination with two lightweight surrogate models for engineering workflows: a closed-form analytical expression based on linear-elastic fracture mechanics and a physics-informed neural network. Using the material’s fracture toughness and Paris’ law, design curves allow both static safety assessment and residual fatigue life prediction for profiles with pre-existing fatigue cracks. Residual stresses are also included, resulting in a mean stress effect that negatively affected fatigue performance. The proposed methodology is validated against full-scale fatigue test data. • Novel damage-tolerant design framework for thin-walled and cold-formed steel profiles. • Stress intensity factor determination via physics-informed machine learning model. • Determination of critical loads and residual fatigue life of cracked profiles. • Development of damage-tolerant fatigue design curves including residual stress effect. • Framework validated against full-scale fatigue tests.