Full Program »

A Gaussian process model based on the wavelet transform for earthquake damage detection

This paper presents the development of a Gaussian Process model for the Continuous Wavelet Transform of acceleration response measurements. The fundamental assumption is that while the structure is not damaged, the wavelet coefficients at any point in time are a transformed realization of a Gaussian Process and the occurrence of damage would cause the wavelet coefficients to deviate from the original undamaged process. The proposed statistical model accounts for the effect of the input motion to the structural response, which eliminates the need for prior signal normalization. The model has been applied to the experimental data acquired from a series of shake table tests conducted at the University of Nevada, Reno. The results from the application are presented and it is demonstrated that the proposed model captures differences in the distribution of model parameters between records where the structure is undamaged and records where structural damage has occurred. Finally, probabilistic metrics for the presence of damage are developed based on Multivariate Gaussian Theory and the presented results indicate the capability of the proposed model for damage detection and classification.

Author(s):

Konstantinos Balafas    
Stanford University
United States

Ram Rajagopal    
Stanford University
United States

Anne Kiremidjian    
Stanford University
United States