Classification and visualisation of data from seismic and health monitoring of structures

Oliver R De Lautour, Piotr Omenzetter

Research output: Chapter in Book/Report/Conference proceedingChapter (peer-reviewed)peer-review


With the advent of modern sensing technology, monitoring of structures under dynamic excitation becomes a viable option for quick assessment of their condition and health. However, major challenges still exist in interpreting the monitoring data. The application of time series analysis methods to Structural Health Monitoring (SHM) is a relatively new but promising approach. Time series methods are inherently suited to SHM where data is sampled regularly over a long period of time, which is typical for monitoring systems. This study focuses on the classification of damage based on analysis of time series model coefficients. Autoregressive (AR) models were used to fit the acceleration time histories of a 3-storey laboratory bookshelf structure and data collected from the ASCE Phase II SHM Benchmark Structure in both healthy and damaged states. Preliminary inspection of the multivariate AR coefficients to check the presence of clusters corresponding to different damage states was achieved using two-dimensional projections obtained from either Principal Component Analysis (PCA) or Sammon mapping. Two classification techniques, Nearest Neighbour Classification (NNC) and Learning Vector Quantisation (LVQ) were used to classify damage into states based on analysis of the AR coefficients reduced in dimensionality via PCA. The results showed that NNC performed well, however, small improvements could be made using LVQ.
Original languageEnglish
Title of host publicationProceedings of the New Zealand Society for Earthquake Engineering Annual Conference
PublisherNew Zealand Society for Earthquake Engineering Inc.
Number of pages12
Publication statusPublished - Apr 2008


Dive into the research topics of 'Classification and visualisation of data from seismic and health monitoring of structures'. Together they form a unique fingerprint.

Cite this