Joint approximate diagonalization technique for modal identification of the Donghai Bridge

Abstract The second-order blind identification (SOBI) and its variants have been extensively explored for output-only modal identification of civil structures under varied excitations. At the core of these methods is the matrix joint approximate diagonalization (JAD) technique, while their efficiency and accuracy are largely determined by how the target-matrices for JAD are constructed from multi-channel structural responses. This study first formulates the JAD framework for structural identification, where different techniques in formulating the target-matrices are summarized and mathematical tools to conduct JAD are also presented. Then two novel ways stemming from conventional identification methods are presented as alternatives to construct the target-matrices for ambient identification, to maintain a low-order formulation and even avoiding the formation of covariance matrix. Subsequently, in view of the large number of candidate target-matrices which are analytically usable, a guiding principle is proposed for selecting reliable target-matrices, where the closeness of the eigenvectors of the target-matrices are compared beforehand, therefore eliminating of distorted target-matrices and also improving the efficiency of the subsequent JAD. The proposed techniques are applied to modal identification of the Donghai Bridge from monitoring data and the proposed JAD-based methods are compared in this context. The results suggest the effectiveness of the proposed techniques and also provide a performance evaluation of these methods.