Towards system identification using single channel measurement in wireless sensing network

Abstract

In the last few years, structural health monitoring has gained signifcant popularity to perform real-time condition assessment of civil structures. With the aid of mobile sensing network, movable wireless sensors have made a paradigm shift in cost- effective and faster deployment of sensors in large-scale structures. A wide range of system identifcation methods has been developed by different researchers to accurately identify modal parameters from the measured vibration data. However, most of these techniques are suitable only when all key locations of the structures are instrumented. In the case of mobile sensing network where a sensor is autonomously moved from one location to another, only a few sensors are available at any time. In this research, a newer time-frequency method, namely the empirical mode decomposition (EMD), is explored and improved to conduct system identification using single channel measurement. The original EMD method results in significant mode-mixing in the modal responses when utilizing closely-spaced modes and data with measurement noise. In this thesis, time-varying filtering based empirical mode decomposition (TVF-EMD) is proposed to undertake ambient modal identification. The proposed method is fully adaptive and suitable for automation since it uses only one channel of data at a time. An energy-based thresholding criterion is proposed to identify dominant frequency components of the vibration data. Once the key signal components are identified, a cluster diagram is integrated with TVF-EMD to identify modal responses that are utilized for modal identification. Such modification shows improved performance of the TVF-EMD in identifying modal parameters using single channel measurement under a wide-range of challenging situations including closely-spaced modes and measurement noise. The proposed method is verified using a suite of numerical, experimental and full-scale studies using wireless sensors in a decentralized manner. The proposed methodology shows significant potential towards its application in modern mobile and robotic sensors.