| description abstract | This paper presents an improvement to the eigensystem realization algorithm (ERA) with natural excitation technique (NExT), which is called the ERA-NExT-AVG method. The method uses a coded average of row vectors in each Markov parameter for evaluating modal properties of a structure. The modification is important because, for the existing stochastic system identification methods, the state-space model, obtained from output sensor data, is usually overparameterized resulting in large systems. Solving such a problem can be computationally very intensive especially in the applications when using the computational capabilities of embedded sensor networks. As a way to improve the efficiency of the ERA-NExT method, the proposed method focuses on the number of components in a single Markov parameter, which can theoretically be minimized down to the number of structural modes. Applying the coded average column vectors as Markov parameters to the ERA, the computational cost of the algorithm is significantly reduced, whereas the accuracy of the estimates is maintained or improved. Numerical simulations are performed for a shear frame model subjected to Gaussian white noise ground excitation. The efficiency of the proposed method is evaluated by comparing the accuracy and computational cost of the estimated modal parameters using the proposed method, with several other stochastic modal identification methods including the ERA-observer Kalman filter identification, ERA-NExT, and autoregressive models. The performance of the method is then evaluated by applying it to ambient vibration data from the Golden Gate bridge, collected using a dense wireless sensor network, and its vertical and torsional modes are successfully and accurately identified. | |
