000138937 001__ 138937
000138937 005__ 20190509132227.0
000138937 0247_ $$2doi$$a10.5075/epfl-thesis-4481
000138937 02470 $$2urn$$aurn:nbn:ch:bel-epfl-thesis4481-4
000138937 02471 $$2nebis$$a5814139
000138937 037__ $$aTHESIS
000138937 041__ $$aeng
000138937 088__ $$a4481
000138937 245__ $$aModel-free data interpretation for continuous monitoring of complex structures
000138937 269__ $$a2009
000138937 260__ $$bEPFL$$c2009$$aLausanne
000138937 300__ $$a155
000138937 336__ $$aTheses
000138937 520__ $$aStructural Health Monitoring (SMH) has the potential to provide reliable, quantitative data on the real condition of structures. Early detection of damage or deterioration can enable preventive repair and reduce maintenance costs. Although obtaining data from monitoring may be easy and fast, the phase of data interpretation is usually complex and expensive in terms of time and resources. An approach to damage detection involves comparisons between predictions from behavioral models and measurements. However, civil engineering structures are difficult to model accurately and this challenge is compounded when structures are built in uncertain environments. Such situations encourage the enhancement of traditional approximate structural assessments through in-service measurements and interpretation of monitoring data. Practitioners encounter difficulties in selecting the data interpretation technique that is most appropriate for continuous monitoring. Only a few studies exist on statistical methods for interpretation of measurements from continuous monitoring. These studies have not included the complexities that arise in practical situations such as measurement noise, missing data and outliers. Moreover, structures are today equipped with hundreds of sensors that measure aspects related to the response and the environment. Therefore algorithms that are able to analyze enormous amounts of heterogeneous data are required for data interpretation. The most innovative contribution of the research work is the definition of new methodologies based on Moving Principal Component Analysis (MPCA) and Robust Regression Analysis (RRA) to detect changes in the behavior of civil engineering structures. The methodology analyzes changes in the patterns of measurement series and is thus able to indicate when and where the damage occurs. The algorithms are designed to learn characteristics of time series generated by sensors during a period called the initialization phase when the structure is assumed to behave normally. The information from the initialization phase subsequently helps in identifying behaviors that can be classified as anomalous. In this manner the proposed methodologies identify anomalous behavior without explicit (and costly) knowledge of structural characteristics such as geometry and behavior models. The methodologies have been tested on data from numerically simulated structures. Studies involved a range of measurement noise levels and a range of damage severities. A comparative study with other statistical methods demonstrates superior performance of the proposed methodologies. The MPCA methodology is extended to accommodate situations of missing data and outliers. Situations of missing complete measurements and missing a single measurement at random are managed with a minimum of extra data pre-processing. A simple and robust cleaning procedure algorithm based on the Inter Quartile Range analysis is presented to remove outliers from measurement series. A clustering algorithm to group measurement series ensures that MPCA is applicable for structures having a significant number of sensors. MPCA is applied to all clusters generated by the clustering algorithm. Finally, MPCA and RRA methodologies have been validated on full scale structures. The results show the ability to identify abrupt permanent changes due to construction stages through time-scale reversal. The results indicate that the proposed methodology can support structural assessments of civil engineering structures.
000138937 6531_ $$aStructural Health Monitoring (SHM)
000138937 6531_ $$amodel-free data interpretation
000138937 6531_ $$amoving principal component analysis
000138937 6531_ $$arobust regression analysis
000138937 6531_ $$aclustering
000138937 6531_ $$amissing data
000138937 6531_ $$aoutliers
000138937 6531_ $$asystème de surveillance d'ouvrage
000138937 6531_ $$ainterprétation des données sans modèle
000138937 6531_ $$aanalyse du Composant Principal Mobiles
000138937 6531_ $$aanalyse de Régression Robuste
000138937 6531_ $$aregroupement
000138937 6531_ $$adonnées manquantes
000138937 6531_ $$aaberrantes
000138937 700__ $$aPosenato, Daniele
000138937 720_2 $$aSmith, Ian$$edir.$$g106443$$0241981
000138937 8564_ $$uhttps://infoscience.epfl.ch/record/138937/files/EPFL_TH4481.pdf$$zTexte intégral / Full text$$s5222384$$yTexte intégral / Full text
000138937 909C0 $$xU10237$$0252031$$pIMAC
000138937 909CO $$pthesis-bn2018$$pDOI$$pENAC$$ooai:infoscience.tind.io:138937$$qDOI2$$qGLOBAL_SET$$pthesis
000138937 918__ $$dEDIC2005-2015$$aENAC
000138937 919__ $$aIMAC
000138937 920__ $$b2009
000138937 970__ $$a4481/THESES
000138937 973__ $$sPUBLISHED$$aEPFL
000138937 980__ $$aTHESIS