Updating of Finite Element Models Based on a Double Condensation Procedure Using Frequency Response Functions Data

Author(s):  
Rémi Berriet ◽  
René Fillod ◽  
Noureddine Bouhaddi

Abstract In order to take into account information from test data, not only at the resonances, but also in the other parts of the measured frequency spectrum, it is of interest to use directly measured Frequency Response Functions (FRF) instead of modal data. We also avoid by this way an experimental modal analysis. In return we have to introduce damping terms into the analytical model, we have to weight the FRF data in a systematic manner and to compute simultaneously a large amount of data. The presented procedure analyses overall these three aspects: definition of modal damping parameters, definition of weighted FRF data and condensation of the problem. This last notion is particularly pointed out. The condensation is performed in two steps : a static condensation of the model on the degrees of freedom corresponding to the location of the sensors, and a simultaneous condensation of experimental and analytical FRF data by a common transformation matrix. The first applications are performed on a simulated test case with large stiffness, mass and modal damping perturbations introduced in the initial model as well as strong noise pollution of measured responses and applied forces.

1998 ◽  
Vol 120 (2) ◽  
pp. 509-516 ◽  
Author(s):  
J. A. Morgan ◽  
C. Pierre ◽  
G. M. Hulbert

This paper demonstrates how to calculate Craig-Bampton component mode synthesis matrices from measured frequency response functions. The procedure is based on a modified residual flexibility method, from which the Craig-Bampton CMS matrices are recovered, as presented in the companion paper, Part I (Morgan et al., 1998). A system of two coupled beams is analyzed using the experimentally-based method. The individual beams’ CMS matrices are calculated from measured frequency response functions. Then, the two beams are analytically coupled together using the test-derived matrices. Good agreement is obtained between the coupled system and the measured results.


2018 ◽  
Vol 23 (No 3, September 2018) ◽  
pp. 314-320
Author(s):  
Eun-Taik Lee ◽  
Hee-Chang Eun

Measurements by sensors provide inaccurate information, including external noises. This study considers a method to reduce the influence of the external noise, and it presents a method to detect local damage transforming the measured frequency response functions (FRFs) to reduce the influence of the external noise. This study is conducted by collecting the FRFs in the first resonance frequency range from the responses in the frequency domain, taking the mean values at two adjacent nodes, and transforming the results to the proper orthogonal decomposition (POD). A damage detection method is provided. The curvature of the proper orthogonal mode (POM) corresponding to the first proper orthogonal value (POV) is utilized as the damage index to indicate the damage region. A numerical experiment and a floor test of truss bridge illustrate the validity of the proposed method for damage detection.


1998 ◽  
Vol 120 (2) ◽  
pp. 503-508 ◽  
Author(s):  
J. A. Morgan ◽  
C. Pierre ◽  
G. M. Hulbert

This paper presents a new method to calculate the so-called Craig-Bampton component mode synthesis (CMS) matrices from measured frequency response functions. The procedure is based on a modified residual flexibility method, from which the Craig-Bampton CMS matrices are recovered. Experimental implementation of the method requires estimating the modal parameters corresponding to the measured free boundary modes and the Maclaurin series expansion coefficients corresponding to the omitted modes. Theoretical developments are presented in the present paper, Part 1. The performance of the new method is then demonstrated in Part 2 (Morgan et al., 1998) by comparison of experiment and analysis for a simple two-beam system.


Author(s):  
Jeffrey A. Butterworth ◽  
Lucy Y. Pao ◽  
Daniel Y. Abramovitch

Fitting discrete-time models to frequency-response functions without addressing existing transport delay can yield higher-order models including additional non-physical nonminimum-phase (NMP) zeros beyond those that may appear as a result of sampling. These NMP zeros can be attributed to a discrete-time representation of a Pade´ approximation to account for the transport delay [1, 2]. Here, we explore this idea in greater detail and this discussion motivates the main contribution of this paper, the presentation of a procedure for fitting a discrete-time model to experimentally measured frequency response data. The appearance of NMP zeros in a system model can complicate controller design and limits the desired closed-loop performance. This discrete-time model-fitting procedure presents a technique that will help yield a model that reflects the measured frequency-response functions accurately, while minimizing the presence of non-physical NMP zeros. The key benefit being that, with respect to previous model fits, it may be possible to eliminate all NMP zeros in the discrete-time model. In the case of model-inverse-based control design, this will allow the stable inversion of the model without the use of approximation methods to account for NMP zeros.


Sign in / Sign up

Export Citation Format

Share Document