Time series methods are inherently suited to the analysis of regularly sampled Structural Health Monitoring (SHM) data and deserve to be better and more extensively explored. This study focuses on the use of statistical pattern recognition techniques to classify seismic damage based on analysis of the time series model coefficients. Autoregressive (AR) models were used to analyze time histories from a 3-storey laboratory bookshelf structure excited on a shake table and the ASCE Phase II Experimental SHM Benchmark Structure in both healthy and damaged states. The coefficients of these AR models were used as damage sensitive features. Three supervised pattern recognition techniques, Back-propagation Artificial Neural Networks, Nearest Neighbor and Learning Vector Quantization were used to classify damage into states, quantify its severity and determine location. In order to visualize the data and reduce its dimensionality it was compressed using Principal Component Analysis or Sammon mapping. The minimum numbers of sensors required for reliable damage detection were also addressed. The results show that seismic damage can be detected and quantified by the three pattern recognition techniques with a very good accuracy using compressed data and small number of sensors.
|Title of host publication||Proceedings of the 14th World Conference on Earthquake Engineering|
|Number of pages||8|
|Publication status||Published - Oct 2008|