Adaptive multiple forgetting factor recursive least square (AMFF-RLS) for real-time structural identification

Publication Type:
Conference Proceeding
From Materials to Structures: Advancement Through Innovation - Proceedings of the 22nd Australasian Conference on the Mechanics of Structures and Materials, ACMSM 2012, 2013, pp. 879 - 884
Issue Date:
Filename Description Size
2012000187OK.pdf1.21 MB
Adobe PDF
Full metadata record
System identification refers to any systematic way of deriving or improving models of systems through the use of experimental and field testing input-output data. In the field of civil engineering, identification of the state of the structure during the dynamic loads, such as earthquake, to predict the current state of the structure and detect any damage or hazard,when it occurs, has posed a great challenge to the research community. Therefore, online and real-time structural parameters identification has recently drawn more attractions, although few research works have been reported especially for cases where measurement data are contaminated by highlevel noise. The Recursive Least Square with single forgetting factor has been widely used in estimation and tracking of time-varying parameters in the fields of electrical and mechanical engineering. However, when there are multiple parameters that each (or some) varies with a different rate, this method cannot perform well. On the other hand, a priori information on the changing rate of the parameters might not be available, and the forgetting factors must be updated adaptively. This paper presents a new adaptive tracking technique, based on the Recursive Least Square (RLS) approach with Adaptive Multiple Forgetting Factors (AMFF). The proposed method considers an adaptive rule for each of the forgetting factors assigned to each of the parameters and thus, enables simultaneous estimation of the time-varying stiffness and damping of the storeys of the structure. Numerical examples show that results of this RLS-based approach are accurate and robust, even when the observed data are contaminated with different types and significantlevels of noise. © 2013 Taylor & Francis Group.
Please use this identifier to cite or link to this item: