Analysis of Earthquake Records from Structures: An Overview

The paper presents an overview of the methods that are available to analyze seismic records from structures. A typical analysis involves data processing, system identification, and damage detection. Data processing aims to minimize the ambient and instrument noise in the data, as well as possible low-frequency drifts, outliers, and other unwanted signals. System identification deals with determining the dynamic characteristics of a structure from its recorded response. There are a large number of methods available in the literature for system identification, varying from simple Fourier analysis to stochastic adaptive filtering. Unless data require otherwise, simple methods should be preferred for identification, because they are more robust and results are easier to interpret. Modal identification is the most widely used form of system identification. An alternative is the discrete-time filters, which provide a convenient model for identification of linear as well as nonlinear structures. Special techniques can be developed to identify a particular component of response, such as torsion, soil-structure interaction, and inter-story drift. Damage detection is a subject that is closely related to nonlinear system identification. Since a damaged structure almost always behaves in a nonlinear fashion, the problem of damage detection becomes equivalent to identification of the nonlinear behavior in the structure. The standard method for damage detection has been to observe the changes in the frequencies of the structure, However, unless it is a major damage, frequencies are not very sensitive to damage, particularly to localized damage. More reliable methods for damage detection can be developed by using time-frequency analyses and wave propagation techniques.