Detection of seismic damage in buildings using structural responses
Authors: P Omenzetter, O de Lautour, University of Auckland
Paper number: 349 (EQC 2006/U535)
Abstract
As a result of an earthquake and its aftershocks built infrastructure may sustain damage. One of the major challenges for quick and efficient recovery in the aftermath of a hazardous event is rapid estimation of the damage. If the state of buildings, bridges, dams and other structures could be quickly and reliably assessed healthy, undamaged structures could be immediately reopened for continuous, uninterrupted service, while damaged structures would be closed and prioritised for later detailed evaluation, repair, demolition or replacement. Doing so will minimise casualties and economic losses and will aid quick recovery of an affected area. Accurate estimation of seismic damage is, however, a time and resource consuming task. Traditionally, it can be achieved by visual inspection of infrastructure following an earthquake. However, given the usually large stocks of structures to inspect and limited number of qualified personnel damage assessment is a slow process. The fact that damage can often be inconspicuous adds to the difficulty.
An alternative to visual inspection can be using measurements of structural responses during strong motion events taken by sensors located in the structure. This approach becomes feasible with the development of continuous seismic monitoring arrays. In New Zealand, the EQC and FRST funded GeoNet monitoring project that is currently expanding its coverage to buildings and bridges, can be used for structural damage detection. However, raw data from seismic sensors are of limited value. The challenge is to analyse the measured structural responses so that structural damage can be detected and quantified. This research studies several techniques that enable such purposeful data analyses.
Damage detection by analysis of structural responses is based on the premise that it is possible to choose certain response signal features that are different for responses of healthy and damaged structures. Once the features are selected another analytical tool is required to actually tell the difference between the features corresponding to different structural states. In this research, we modelled structural accelerations using autoregressive time series models in order to find suitable damage sensitive features, and used pattern recognition techniques for feature classification. The approach was thoroughly investigated through several experimental studies and results of damage detection and quantification are promising.
Technical Abstract
The ability to estimate seismic induced damage to civil infrastructure is undoubtedly one of the most important challenges faced by structural engineers. In this research, structural damage was detected and assessed by analysing the structural response.
Autoregressive (AR) time series models were used to fit the acceleration time histories obtained when the structure was in both undamaged and damaged states. The AR coefficients were selected as damage sensitive features and statistical pattern recognition techniques were investigated for interpreting changes in the values of these features caused by damage. Initially, an offline damage detection method was developed in which Back-Propagation Artificial Neural Networks (BP ANNs) were used for both classification and quantification tasks where the damage states were recognised or percentage of remaining stiffness at a specific location was estimated, respectively. The method was applied to three experimental structures: a 3-storey bookshelf structure, the ASCE SHM Phase II Experimental Benchmark Structure and a RC column. In addition, for damage classification tasks, two supervised classification techniques of Nearest Neighbour Classification (NNC) and Learning Vector Quantisation (LVQ), and an unsupervised method of Self-Organising Maps were studied. Damage classification and/or quantification using BP ANNs, NNC and LVQ techniques was achieved with very good results confirming the usefulness of AR coefficients as damage sensitive features and the studied pattern recognition techniques as damage classifiers.
An online damage detection method was also developed based on recursive identification of the AR models using the forgetting factor and Kalman filter approaches and BP ANNs. An analytical linear 3-DOF model with time varying stiffness was investigated and the results showed that damage could be detected and quantified as it occurred. Damage detection in nonlinear systems was addressed with the investigation of an analytical 1-DOF elastoplastic oscillator and a 3-DOF Bouc-Wen hysteretic model. In both cases the on-set of nonlinearity was detected with good accuracy.
Order a research paper
Many of these research papers have PDF downloads available on the site.
If you'd like to access a paper that doesn't have a download, get in touch to ask for a copy.