See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222572791

An efficient computational procedure for random vibro-acoustic simulations

Article  in  Journal of Sound and Vibration · February 2008

DOI: 10.1016/j.jsv.2007.07.068

CITATIONS READS

6 53

2 authors:

Jean-Pierre Coyette Karl Meerbergen

Université Catholique de Louvain - UCLouvain KU Leuven
82 PUBLICATIONS   828 CITATIONS    114 PUBLICATIONS   2,350 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nonlinear eigenvalue problems, Parametric eigenvalue problems, linear algebra for machine learning View project

Actran development program View project

All content following this page was uploaded by Karl Meerbergen on 17 March 2014.

The user has requested enhancement of the downloaded file.

 An efficient computational procedure for random
vibro-acoustic simulations
Karl Meerbergen
January 5, 2009

Vibro-acoustic models are often subjected to random excitations. Examples are acous-
tic diffuse fields (e.g. in reverberant test chambers) and turbulent boundary layer exci-
tations (e.g. in aerodynamic noise studies).
The framework for modelling such distributed excitations is the mathematical concept
of (weakly) stationary random process. Such processes are usually characterized, in the
frequency domain, by power spectra and are practically defined by referring to a reference
power spectrum and a suitable spatial correlation function. The power spectral density
(PSD) is the Fourier transform of the auto-correlation function and is the characterization
of a random process in the frequency domain.
In a finite element context, the random excitation is usually defined on a part of the
boundary surface. With this random excitation a cross PSD matrix is associated. The
response PSD matrix can be very large. Note that it is not feasible storing this matrix,
since it is a dense matrix of very high dimensions. On the other hand, the excitation
cross PSD matrix can be stored explicitly, since its size is typically much lower. The goal
is to approximate the response PSD matrix by a low rank matrix in factored form, whose
storage requirements are low.
In earlier work [1], the pseudo load-case method was proposed that aims at producing
a low rank approximation of the response cross PSD matrix from a low rank partial
eigendecomposition of the excitation PSD matrix. However, for turbulent boundary
layers, such decomposition can be very expensive and does not always lead to a significant
reduction of the memory consumption.
We discuss a new method for handling random acoustic excitations in finite element
models. The idea is to directly approximate the cross PSD matrix of the response by a
low rank matrix. We illustrate the low rank approximation can be computed efficiently by
the implicitly restarted block Lanczos method. The novelty lies in applying the Lanczos
method to the cross PSD matrix of the response rather than the excitation. We give a
theoretical explanation for finite element models with modal damping.
We also show numerical examples that show a significant reduction in computation
time compared to the pseudo load-case method.
The scientific details are published in [2].

1
 References
[1] J.P Coyette, T. Knapen, G Lielens, K. Meerbergen, and P. Ploumhans. Simulation
of randomly excited acoustic insulation systems using finite element approaches. ICA
Conference, Kyoto, April 4-9, 2004.

[2] J.P. Coyette and K. Meerbergen. An efficient computational procedure for random
vibro-acoustic simulations. Journal of Sound and Vibration, 310(1–2):448–458, 2008.

View publication stats