Multilevel Monte Carlo simulations of composite structures with uncertain manufacturing defects

By adopting a Multilevel Monte Carlo (MLMC) framework, this paper shows that only handful of costly fine scale computations are needed to accurately estimate statistics the failure composite structure, as opposed many thousands typically in classical analyses. The introduces MLMC method and provides an extension called with selective refinement efficiently calculated structural probabilities. Simple-to-implement, self-adaptive algorithms given, results demonstrate huge computational gains for two novel, real world example problems composites performance analysis: (i) effects fibre waviness on compressive strength material (ii) uncertain buckling panel random ply orientations. For most challenging test case estimating 1?150 probability speed-up factor >1000 over Carlo. In absolute terms, time was reduced from 218 CPU days just 4.4 hours, making stochastic simulations would otherwise be unthinkable now possible.