Predicting High-Cycle Fatigue Life in Laser-Cut Structural Joints Using Stochastic Emulators and Active Learning

Overview & Synopsis

Laser cutting (LC) is widely adopted in modern manufacturing for its precision, speed, and ability to produce complex geometries with tight tolerances, particularly in the automotive, structural, and aerospace industries, where components must perform reliably under cyclic loading. However, LC changes the near-surface condition of materials by inducing microstructural transformations, modifying the surface topography, and generating tensile residual stresses, which can accelerate crack initiation and can markedly reduce the high-cycle fatigue life. Confronting these difficulties necessitates the development of prediction models that are data-efficient and capable of quantifying uncertainty. In this study, we propose a hybrid framework that couples state-of-the-art surrogate models and stochastic emulators with high-cycle fatigue experiments. Stochastic emulators can accurately and efficiently predict the observed data scatter, while an active learning approach helps selecting high-value tests via an epistemic uncertainty-driven criterion. The trained surrogate accurately predicted the fatigue life with quantified uncertainty and provided a scalable methodology for probabilistic design.