Vous êtes ici : GIPSA-lab > Formation > Thèses soutenues

Automatic Modal-Variation Tracking via a Filter-Free Random Decrement Technique Application to ambient vibration recordings on high-rise buildings


Directeur de thèse :     Nadine MARTIN

École doctorale : Electronique, electrotechnique, automatique, traitement du signal (EEATS)

Spécialité : Signal, image, parole, télécoms

Structure de rattachement : CNRS

Établissement d'origine : INP-PHELMA

Financement(s) : contrat à durée déterminée


Date d'entrée en thèse : 01/09/2011

Date de soutenance : 28/05/2015


Composition du jury :
M. Jérôme ANTONI professor, INSA Lyon (Rapporteur)
M. Laurent MEVEL Chargé de recherche, INRIA Rennes, (Rapporteur)
M. Philippe GUEGUEN Directeur de recherche IFSTTAR, ISTerre Grenoble (Président, Examinateur)
M. Julien HUILLERY Maître de conférence, Ecole Centrale Lyon (Examinateur)
Mme. Nadine MARTIN Directeur de recherche CNRS, GIPSA-Lab Grenoble (Directrice de thèse)


Résumé : This thesis proposes a novel approach to automatically monitor the variations of the frequencies and the damping ratios of actual high-rise buildings subjected to real-world ambient vibrations. The approach aims at dealing simultaneously with the following challenges: multicomponent signals recorded over the aforementioned buildings and having closely-spaced frequency modes with low, exponential and damped amplitudes and high additive noises. The approach relies on the application of the Random Decrement Technique directly over the multi-component signal under study which leads to the extraction of a Multi-mode Random Decrement Signature equivalent to the system impulse response. To characterize such a signature, we propose a signal model based on the physical structure of the building from where the modal parameters can be estimated. For the purpose of non-biased modal estimate, we propose to use an iterative method based on a Maximum-Likelihood Estimation. In order to initi alize t he parameters of the latter, a first step is designed which can be considered as an independent estimator of the modal parameters. The originality of this step lies in its ability to automatically define the number of modes of the estimated signature through the use of the statistical properties of a Welch spectrum. The modal parameters estimated by the spectral-based initialization step are finally refined by the Maximum-Likelihood Estimation step. The latter reduces the bias in the estimation and yields more reliable and robust results. All these steps are defined in order to be able to automatically monitor the health of a building via a long-term real-time tracking of the modal variations over time without the need to any user intervention . In addition, the proposed approach has paid very special attention to the automatic estimation of the most problematic modal parameter, i.e., the damping ratio. Such features making two of the original features as compared to existing techni q ues. The adaptability and functionality of AMBA is validated over six actual buildings excited by real-world ambient vibrations. From the obtained results, AMBA proved high efficiency in automatically estimating the frequencies and moreover the damping ratios in case of closelyspaced frequency modes and very low signal-to-noise ratio level. AMBA as well demonstrated a good performance for tracking the modal variations over time. Key words: automatic modal analysis, random decrement technique, modal variation, building monitoring, time-tracking.

GIPSA-lab, 11 rue des Mathématiques, Grenoble Campus BP46, F-38402 SAINT MARTIN D'HERES CEDEX - 33 (0)4 76 82 71 31