Data Compression of Structural Seismic Responses via Principled Independent Component Analysis

Abstract

This paper proposes a novel lossy data compression scheme for structural seismic responses based on principled (truncated) independent component analysis (PICA). It is first shown that independent component analysis (ICA) is able to transform a multivariate data set into a sparse representation space where is optimal for coding and compression, such that both the intradependencies and interdependencies (i.e., redundant information) between the multichannel data are removed for efficient data compression. Two examples are presented to demonstrate the compression performance of PICA, using the real-measured structural seismic responses from the 1994 Northridge earthquake, of the Fire Command Control (FCC) building and the USC hospital building, respectively. It is compared with the popular wavelet transform coding technique, which is only able to handle single-channel data separately. Results show that PICA achieves dramatically higher compression ratio (CR) than the wavelet method while retaining excellent reconstruction accuracy. It is also shown that PICA slightly outperforms the (principled) principal component analysis (PCA) method—which used to be considered optimal multivariate data compression scheme—with respect to both CR and reconstruction accuracy. Equipped with the FastICA algorithm that enjoys a cubic convergence rate, PICA has potential for rapid and reliable data transfer, communication (e.g., multihop wireless sensor network), storage, and retrieval in online or post-disaster (e.g., earthquake) monitoring and assessment applications of civil infrastructures.