Abstract
Efficient vibroacoustic response prediction on complex structures, such as spacecraft, represents a challenging task, even for the computers and numerical techniques of today. This is particularly evident in the mid-frequency range, where structures begin exhibiting chaotic behaviour, rendering element-based techniques inefficient or unreliable. In this article, an efficient random formulation for reduced finite element method (FEM) models is proposed, such that Monte Carlo simulations can be carried out robustly within practically acceptable timeframes. The introduced novel non-parametric stochastic FEM is inherently compatible with various existing component mode synthesis techniques. It is particularly well adapted to use with popular modal reduction approaches, such as the Craig-Bampton method. The mathematical framework for the method is outlined, enabling the deterministic reduced matrices to be robustly perturbed at the subsystem level. Properties, such as matrix positive-(semi)definiteness, mean system eigenvalues, and representation accuracy are preserved. This new stochastic FEM is validated against a full parametric Monte-Carlo simulation and test data of a real spacecraft structure, establishing its reliability and computational efficiency. In the proposed coupled FEM-BEM approach, the acoustic domain is modelled with hierarchical matrix accelerated collocation BEM. This alleviates the memory requirements for the large, dense BEM matrices, and the need for spatial discretisation of acoustic FEM. The full implementation is outlined for a simple geometry discretised with high a density mesh, showing consistent convergence of the employed iterative solver.