Projection methods within model updating

Abstract

Based on certain mathematical subspace and projection techniques, two new nonlinear methods for the parametric identification of linear elastomecham'cal systems in the frequency domain are presented. In the form of natural generalizations the methods overcome serious problems of the classical Input Residual Method (IRM) and the nonlinear Output Residual Method (ORM). For the IRM the Projective Input Residual Method overcomes the necessity of measurements being complete without the need of reducing the mathematical model. For the ORM the Regularized Output Residual Method provides a smoothing of the classical ORM graph and thus a drastic elimination of local minima. For both methods the necessary experimental designs are realizable in terms of the number of measured components and the excitation configuration, i.e., number and structure of excitation vectors needed. It is shown that the methods are well-founded from a mathematical point of view and especially from a functional analysis one. Examples demonstrate the advantages of the methods presented.

Keywords:

• System identification
• non-linear parameter estimation
• projection methods
• model updating

Additional information

Notes on contributors

M. Oeljeklaus

*