On the reliability of a class of system identification techniques insights from bootstrap theory.pdf

Structural Safety 24 (2002) 261–280
ate/strusafe
On the reliability of a class of system identification
techniques: insights from bootstrap theory
T. Kijewski, A. Kareem*
NatHaz Modeling Laboratory, Department of Civil Engineering and Geological Sciences, University of Notre Dame,
156 Fitzpatrick Hall, Notre Dame, IN 46556, USA
Abstract
The estimation of structural damping from systems with unknown input can be particularly challenging
but is vital to furthering the understanding of energy dissipation in structural motions. The random
decrement technique and spectral analysis monly invoked system identification schemes for this
problem. Although there is well-established theory regarding the bias and variance errors of power spectra
and random decrement signatures, the theoretical error formulae give only approximate indications of
errors inherent to the estimated dynamic properties. This study addresses this through a bootstrap
approach to assess the quality of system identification by providing surrogate estimates of damping and
natural frequency to generate useful statistics and confidence intervals. # 2002 Elsevier Science Ltd. All
rights reserved.
Keywords: Bootstrapping; System identification; Monte Carlo; Damping; Random decrement technique; Spectral
analysis
1. Introduction
With the advancement of modern structures to new heights, the issues of serviceability and
fort e to the forefront in design, making damping, now more than ever, a
critical design parameter. Despite its significant role, damping continues to be an enigma in
design [1,2]. As damping does not relate to physical properties of the structure in a direct way like
mass and stiffness, it cannot be estimated with much certainty in design. While the examination of
existing structures has resulted in international databases of actual damping values, there is con-
siderable scatter in the data, partly attributed to the amplitude dependence of damping, but more
so due to er