Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/88472
DC FieldValueLanguage
dc.contributorDepartment of Civil and Environmental Engineering-
dc.creatorZhang, Qiuhu-
dc.identifier.urihttps://theses.lib.polyu.edu.hk/handle/200/10813-
dc.language.isoEnglish-
dc.titleSparse bayesian learning approach for damage detection in a population of nominally identical structures-
dc.typeThesis-
dcterms.abstractThis thesis is dedicated to the development of a general damage detection framework for nominally identical structures (NISs) rather than only a particular single structure. The developed damage detection framework is formulated in an unsupervised learning scheme, which only makes use of response measurements from undamaged structures. It consists of two phases: the baseline and inspection phases. In the baseline phase, historical response measurements from multiple nominally identically undamaged structures are utilized to establish a data-driven baseline model for representing healthy population features of all NISs. Three types of sparse Bayesian modelling approaches are proposed to deal with multiple sources of uncertainty in the measured responses, including measurement noise (intra-structure uncertainty) and structural variability in the materials and/or manufacturing processes (inter-structure uncertainty). The first modelling approach is introduced simply by pooling the inter-structure and intra-structure uncertainties such that standard sparse Bayesian learning (SSBL) can be implemented to model the population features of NISs. In the second modelling approach, an extension to SSBL termed heteroscedastic sparse Bayesian learning (HSBL) is proposed to address heteroscedastic training data, resulting from the pooling of multiple sources of uncertainty. In the third modelling approach, another extension to SSBL termed panel sparse Bayesian learning (PSBL) is proposed, in which different sources of uncertainty can be modelled separately. Their performance is assessed in terms of three model quality indices, including the root mean square residual (RMSR), the mean standardized log loss (MSLL) and the sparsity ratio K. In the inspection phase, Bayesian residuals between new response measurements and population features predicted by the baseline model are examined for the identification of damage in NISs. Three categories of probabilistic diagnostic logics including frequentist null hypothesis significance testing (NHST), Bayesian point null hypothesis testing (PNHT), and the novel Bayesian NHST are compared in the capacities of the detection of damage, the quantification of damage extent, and the warning of diagnostic risk. The impact on structural damage diagnostics, of the three types of sparse Bayesian modelling approaches for constructing the baseline model in the baseline phase is investigated. The optimal baseline modelling approach and the optimal damage diagnostic logic are found. A case study of online condition assessment for railway wheels is conducted throughout this thesis to validate the feasibility and effectiveness of the proposed methods.-
dcterms.accessRightsopen access-
dcterms.educationLevelPh.D.-
dcterms.extentxxi, 269 pages : color illustrations-
dcterms.issued2020-
dcterms.LCSHStructural dynamics-
dcterms.LCSHStructural health monitoring-
dcterms.LCSHHong Kong Polytechnic University -- Dissertations-
Appears in Collections:Thesis
Show simple item record

Page views

46
Last Week
0
Last month
Citations as of Mar 24, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.