A twin-mesh approach for random field analysis in high-dimensional dynamic models

Non-deterministic methods such as the random field approach suffer from the curse of dimensionality from computational burden of industrially sized models. The computational complexity increases at high dimensions due to increased time complexity of eigenvalue computations, leading to the Karhunen-Loeve series expansion becoming intractable. This makes the corresponding propagation routines to become inviable. A novel methodology is proposed for efficient propagation of a random field by tackling this problem from the discretization perspective. The method uses a twin-model that efficiently discretizes a random field on a coarse mesh grid using a KL expansion, which is then propagated on a high-dimensional grid of the Finite Element model. A two-dimensional model of moderate-dimensionality with 10000 elements is used to illustrate the numerical efficiency of this approach through a convergence study focusing on the resolution of the twin-model when applied in a dynamic analysis. The method is also well suited for higher dimensions.