Identification of structural systems from measured response

Abstract

This report presents an approach to the problem of parameter estimation of finite element models of complex
\nstructural systems using measured response. The method computes the element constitutive parameters of a finite
\nelement model of a structure by finding a constrained nonlinear minimum of the difference between the measured
\nresponse of the real structure and the computed response of a finite element model of the structure. Constraints are
\nused to bound the constitutive parameters. The optimization problem is solved by the method of Recursive Quadratic
\nProgramming. A simple and straightforward method of computing the gradient of the loss function on an
\nelement-by-element basis is presented.
\nThe method presented herein has been implemented as a general purpose parameter estimation program for
\nstructural systems. This program has all the flexibilities of a general purpose finite element package in that it can
\ntreat complex structures with different topologies, geometries, and element types. It is suited to static, modal dynamic,
\nor forced dynamic response, sparsely sampled in space, time, and state. We consider both an equation-error
\nestimator and an output-error estimator for the static and modal dynamic problems, and an equation-error estimator
\nfor the forced dynamic problem.
\nMonte Carlo simulation is used to study the behavior of the parameter estimation algorithms in the presence of
\nnoise in the measurements. Extensive simulations are carried out for both the static and forced dynamic cases. A
\ncase study, using real field measurements is presented for the modal dynamic case.