Reliability and sensitivity analyses of fatigue-loaded industrial systems are promising for quantifying the impact of input parameter uncertainties on predicted fatigue life crack initiation. However, this requires first resorting to nonlinear finite element simulations, the results of which are then processed by multiaxial fatigue analyses. The computational cost of nonlinear finite element models generally makes sensitivity and reliability assessment very challenging in an industrial context, as it would require thousands of iterations. This paper aims to evaluate the probability of failure,p_f, of a welded substructure of a military vehicle subjected to random loading as the primary objective and then its by-products such as reliability sensitivities.

 In order to overcome the computational burden, an adaptation of Active Kriging Monte Carlo Simulation (AK-MCS) to sequential models is proposed and referred to as Active Kriging for Sequential Models (AK-SM). AK-SM introduces a novel enrichment strategy within the AK-MCS framework, leveraging the sequential nature of the numerical chain together with the lower computational cost of the fatigue solver. An imputation criterion based on functional decomposition and variance of the Kriging prediction is used to selectively avoid expensive finite element evaluations and prioritize affordable fatigue post-processor computations.

The results are first demonstrated on a simplified but representative finite element scenario to move toward the industrial case. AK-SM achieves significant reductions in computational cost, ranging from 70% to 90% compared to AK-MCS, while maintaining high accuracy in failure probability estimation. Subsequently, AK-SM is used to solve the industrial problem. Special attention is given to the probabilistic modelling of loading, material and geometrical properties. Reliability sensitivities are also evaluated as a by-product of the approach.