Model Updating Using Neural Networks

Abstract

Accurate models are necessary in critical

applications. Key parameters in dynamic

systems often change during their life cycle due

to repair and replacement of parts or en-

vironmental changes. This dissertation presents

a new approach to update system models,

accounting for these changes. The approach

uses frequency domain data and a neural net-

work to produce estimates of the parameters

being updated, yielding a model representative

of the measured data. Current iterative methods

developed to solve the model updating problem

rely on min- imization techniques to nd the set

of model parameters that yield the best match

between experimental and analytical responses.

Since the minimization procedure requires a fair

amount of computation time, it makes the

existing techniques infeasible for use as part of

an adaptive control scheme correcting the

model parameters as the system changes. They

also require either mode shape expansion or

model reduction before they can be applied,

introducing errors in the procedure.

Furthermore, none of the existing techniques

has been applied to nonlinear systems. The

neural network estimates the parameters being

updated quickly and accurately without the

need to measure all degrees of freedom of the

system. This avoids the use of mode shape

expansion or model reduction techniques, and

allows for its implementation as part of an

adaptive control scheme. The proposed

technique is also capable of updating weakly

nonlinear systems. Numerical simulations and

experimental results show that the proposed

method has good accuracy and generalization

properties, and it is therefore, a suitable

alternative for the solution of the model

updating problem of this class of systems.