Improved data compression techniques to analyze big data in structural health monitoring

Abstract

With growing number of complex and slender structures worldwide, long-term structural health monitoring (SHM) has been intensively pursued to retrofit and control these structures under extreme climatic events. Modern sensing technology including wireless sensors and high quality data acquisitions have improved the capability of SHM where a relatively enormous amount of data could be measured remotely and sent wirelessly for a longer period of time. Unlike wired vibration sensors, wireless sensors are inexpensive and easier to install with less labour-intensive process, thereby leading to a significant cost-saving to the infrastructure owner. However, the modern sensing technology and remote data acquisition has some several limitations due to their limited bandwidth, time synchronization and inadequate sampling issues. The large amount of data collected from the structural systems often causes missing data, network jam or packet loss while transmitting the big data. In this research, the theory of compressive sampling (CS) is implemented as a promising data compression technique that can recover undersampled vibration signals of dynamical systems, thereby reducing overall burden of analyzing big data in SHM. The l1-norm minimization (LNM) and discrete cosine transform (DCT) are exploited to perform data compression and enhance data recovery of the compressed big data. A novel time-frequency blind source separation is integrated with the data compression technique to evaluate the accuracy of the proposed method in modal identification. The results of the proposed data compression techniques are verified using a suite of numerical, experimental and full-scale studies. The results reveal that DCT could be considered as a powerful data compression tool even for the vibration data containing damage signatures, low energy modes and low signal-to-noise ratio.