Extracting Structural Damage Features: Comparison Between PCA and ICA

Abstract

How to effectively extract structural features from structural damage signals is always a hot problem in structural engineering domain. In this paper, principal component analysis (PCA) and independent component analysis (ICA) are disscussed in detail for selecting the feature from the measured time series data. Considering of the structural engneering data with unknow covariance and different scales, a standardization PCA based samples is used. In order to speed up the calculation for components, second-order-statistics spatio-temporal decor- reltion algorithm is applied. Then, The components from PCA and different ICA algorithms are tested by the benchmark dataset from IASC-ASCE SHM group in British Columbia University. The results show that both PCA and ICA can effec- tivey reduce the influence from noise; different cumulate contribution rate in PCA plays different roles, and 99% is preferred . For two-damage level, both PCA and ICA are good; but for multi-damage level, ICA is better than PCA with 99% cu- mulate contribute rate. Therefore, ICA extracts structural features more accurately than PCA.