Structural health monitoring using wireless sensor networks and Bayesian probabilistic methods

Abstract

Structural health monitoring (SHM) has emerged as an active, interdisciplinary research field over the past two decades due to the need to better manage and maintain complex structural systems to ensure their safety, serviceability and sustainability. Structural health monitoring employs sensing technologies and data processing methods to perform condition assessment and damage detection of structural systems, such as buildings, bridges, aircrafts, ships, etc. In this thesis, structural health monitoring using wireless sensor networks and Bayesian probabilistic methods are presented. In the hardware aspects, the emerging wireless sensor networks (WSN) for SHM have attracted a lot of attention from both academic and industrial communities. Wireless sensor networks have the potential to improve SHM dramatically with their onboard computation and wireless communication capabilities. However, some issues need to be addressed before wireless sensors can be utilized in SHM. Accurate synchronized sensing among wireless sensors is a key issue enabling the implementation of such smart systems for SHM based on vibration measurements. However, perfect synchronized sensing is unlikely to be achieved in WSN. The effect of non-synchronous sensing when using wireless sensors on structural modal identification is investigated and a methodology for correcting such errors is proposed in the first part of this thesis. To process the large amount of structural response data collected by the wireless monitoring system, Bayesian probabilistic techniques for system identification and model updating are explored for SHM purpose. Bayesian probabilistic approaches are very powerful because they explicitly treat uncertainties entering the mathematical models of the system and the excitations. Under a Bayesian statistical framework, not only the optimal (most probable) values of the model parameters are obtained, but also the uncertainties associated with the updated parameters of interest can be quantified. For structural modal parameter identification, a Bayesian spectral decomposition (BSD) method is proposed. This method is an output only modal identification method based on measured ambient or operational response data, which makes use of the statistical properties of eigenvalues and associated eigenvectors of the output spectral density matrix to obtain not only the optimal values of the identified modal parameters but also their associated uncertainties. Calculation of the uncertainties of the identified modal parameters is very important if one plans to proceed with the updating of a nominal finite element model based on modal estimates. This method identifies each mode separately. For each mode, this method identifies modal spectral parameters (modal frequencies and damping ratios) and mode shapes separately. By using this variable separation approach, the computational effort is greatly reduced. An energy-efficient distributed computing strategy is proposed for implementation of this method in wireless sensor networks that provide such distributed computing environment. For structural model updating, two Bayesian probabilistic methods are proposed. One is based on modal data and the other is based on modal flexibility data. In the modal data based approach, an efficient Bayesian probabilistic method in conjunction with an auxiliary deterministic approach for structural model updating with incomplete modal data is presented. The initial estimates obtained in the preliminary deterministic updating stage can facilitate the optimization in the Bayesian updating stage, which makes the algorithm more efficient and robust. In the modal flexibility based approach, measured modal flexibility data is used for model updating. It is shown that the modal flexibility matrix can be easily and accurately estimated from a few of lower frequency modes of vibration of the structure. By incorporating modal flexibility data in the Bayesian updating formulation, the model parameters are updated and their associated uncertainties are quantified. Quantification of the uncertainties of the updated model parameters is very important if one plans to proceed with damage detection by comparing the updated model parameters of the undamaged state and those of the possible damage state of the structure. Illustrative examples using both numerical data and experimental data are presented to show the application and the effectiveness of the proposed methods.

Date of Award	2013
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology