Multi-fidelity surrogate modeling of fatigue cracks integrating the effects of parameter randomness

Penetrating fatigue cracks critically compromise the structural integrity and operational reliability of mechanical components. Predicting three-dimensional fatigue crack propagation involves complex multi-parameter interactions and computationally intensive simulations, posing significant challenges for efficient life assessment. To bridge this gap, this work developed a probabilistic multi-fidelity surrogate model that integrates experimental characterization and finite element analysis. The framework employs a D-vine copula model to quantify uncertainties and dependencies in key parameters, while an adaptive stochastic configuration network ensemble surrogate emulates crack-growth responses with high computational efficiency. Validated through ASTM-compliant fatigue tests on 25CrNiMo alloy specimens. Results demonstrate that the proposed methodology enables efficient, high-accuracy fatigue life evaluation for engineering structures operating under uncertainty, providing a robust tool for reliability-driven design and maintenance.