Detection thresholds in structural health monitoring M.D. Trifunac n , M. Ebrahimian Department of Civil Engineering, University of Southern California, Los Angeles, CA 90089-2531, USA article info Article history: Received 19 April 2014 Accepted 28 July 2014 Keywords: Full scale testing of structures Identification of earthquake damage Structural health monitoring using wave propagation methods Detection signal to noise ratio abstract Most health-monitoring methods for analysis of full-scale structures detect changes in the soil-system response by comparison of the identified response parameters “before” and “after” large excitations. However, because the deformations of the foundation soil, which are almost always present, are nonlinear, and because permanent deformations and changes in the soil depend differently on the excitation histories of different sites, it is difficult at best to (1) make use of classical concepts, which are based on signal-to-noise ratio, and (2) the vibrational description of the response in terms of transfer functions. This situation then necessitates a selection of methods and metrics that detect changes in the system parameters that (1) can be evaluated in almost real-time conditions, and (2) are as insensitive as possible to the contributions to the response of the soil–structure interaction. This, in turn, requires case- by-case, specific, and complex methods of analysis, which can only occasionally be generalized to other buildings and which can also be quite different from one event to the next. In this paper, we present examples of what has been done thus far to avoid the associated complexities for a group of full-scale structures, two of which experienced damaging response during earthquake shaking. In the examples presented, we emphasize those features of the analyses that are related to detection thresholds and to the time- and case-dependent amplitudes of the detection “noise”. & 2014 Elsevier Ltd. All rights reserved. 1. Introduction One of the principal goals of dynamic structural health mon- itoring in general, and especially after strong earthquake shaking, is to identify the location(s) and the extent of structural damage, and to do so with confidence and as soon as possible. This is needed for identification of the degrees of building damage, for ensuring public safety, and for optimization of post-earthquake mitigation measures [2]. Related to this is the more general subject of structural-system identification, which aims to find the best representative dynamic model of a structure, which in turn can be used to reproduce the recorded vibration data. In this paper, it will be assumed that some form of instrumental recording is available in the structures that were subjected to strong earthquake shak- ing, and that the data are in a form suitable for analysis. We will also consider experimental and computational methods for pre- and post-earthquake analysis of structures that may have experi- enced strong and damaging earthquake shaking. Structural health-monitoring methods for full-scale structures can be grouped broadly into (1) vibrational, and (2) wave- propagation methods. Both aim to identify a change in structural properties. The vibrational methods can identify changes in system frequencies and in the associated mode shapes. The wave-propagation methods search for changes in the wave travel times through the structural members, and when the recording has sufficient spatial resolution, these times can be used to determine the location and relative severity of the damage [90,123,128]. The success of both methods depends upon (1) their ability to compare the post-damage response to some pre-damage reference level, (2) the quality of the dynamic model of the structure, and (3) the signal-to-noise ratio of the amplitude(s) of the relative changes in the governing parameter(s) of the response. This process is far more challenging and difficult than classical signal extraction and identification from noisy data because the “noise” in structural health monitoring includes large fluctuations of system parameters over time, that are associated with other, simultaneously occurring processes that cannot always be identified or separated from the analysis. For this reason, the term “detection thresholds” in the title of this paper refers to time- dependent physical processes, which must also be calibrated and whose changes must be followed in time. It cannot be thought of, in the classical sense, as representing some equivalent, constant “noise” amplitude in the system. After starting work on this paper, the authors discovered that the amplitudes and the complexity of the noise thresholds are far greater than expected and that the sources of this noise” can be different for each case study. The structural models adopted for analysis and the environmental factors that contribute to each Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/soildyn Soil Dynamics and Earthquake Engineering http://dx.doi.org/10.1016/j.soildyn.2014.07.014 0267-7261/& 2014 Elsevier Ltd. All rights reserved. n Corresponding author. Fax: þ1 213 744 1426. E-mail address: trifunac@usc.edu (M.D. Trifunac). Soil Dynamics and Earthquake Engineering 66 (2014) 319–338