Welcome to the new version of CaltechAUTHORS. Login is currently restricted to library staff. If you notice any issues, please email coda@library.caltech.edu
Published September 2004 | public
Report Open

Real-time Loss Estimation for Instrumented Buildings

Abstract

Motivation. A growing number of buildings have been instrumented to measure and record earthquake motions and to transmit these records to seismic-network data centers to be archived and disseminated for research purposes. At the same time, sensors are growing smaller, less expensive to install, and capable of sensing and transmitting other environmental parameters in addition to acceleration. Finally, recently developed performance-based earthquake engineering methodologies employ structural-response information to estimate probabilistic repair costs, repair durations, and other metrics of seismic performance. The opportunity presents itself therefore to combine these developments into the capability to estimate automatically in near-real-time the probabilistic seismic performance of an instrumented building, shortly after the cessation of strong motion. We refer to this opportunity as (near-) real-time loss estimation (RTLE). Methodology. This report presents a methodology for RTLE for instrumented buildings. Seismic performance is to be measured in terms of probabilistic repair cost, precise location of likely physical damage, operability, and life-safety. The methodology uses the instrument recordings and a Bayesian state-estimation algorithm called a particle filter to estimate the probabilistic structural response of the system, in terms of member forces and deformations. The structural response estimate is then used as input to component fragility functions to estimate the probabilistic damage state of structural and nonstructural components. The probabilistic damage state can be used to direct structural engineers to likely locations of physical damage, even if they are concealed behind architectural finishes. The damage state is used with construction cost-estimation principles to estimate probabilistic repair cost. It is also used as input to a quantified, fuzzy-set version of the FEMA-356 performance-level descriptions to estimate probabilistic safety and operability levels. CUREE demonstration building. The procedure for estimating damage locations, repair costs, and post-earthquake safety and operability is illustrated in parallel demonstrations by CUREE and Kajima research teams. The CUREE demonstration is performed using a real 1960s-era, 7-story, nonductile reinforced-concrete moment-frame building located in Van Nuys, California. The building is instrumented with 16 channels at five levels: ground level, floors 2, 3, 6, and the roof. We used the records obtained after the 1994 Northridge earthquake to hindcast performance in that earthquake. The building is analyzed in its condition prior to the 1994 Northridge Earthquake. It is found that, while hindcasting of the overall system performance level was excellent, prediction of detailed damage locations was poor, implying that either actual conditions differed substantially from those shown on the structural drawings, or inappropriate fragility functions were employed, or both. We also found that Bayesian updating of the structural model using observed structural response above the base of the building adds little information to the performance prediction. The reason is probably that Real-Time Loss Estimation for Instrumented Buildings ii structural uncertainties have only secondary effect on performance uncertainty, compared with the uncertainty in assembly damageability as quantified by their fragility functions. The implication is that real-time loss estimation is not sensitive to structural uncertainties (saving costly multiple simulations of structural response), and that real-time loss estimation does not benefit significantly from installing measuring instruments other than those at the base of the building. Kajima demonstration building. The Kajima demonstration is performed using a real 1960s-era office building in Kobe, Japan. The building, a 7-story reinforced-concrete shearwall building, was not instrumented in the 1995 Kobe earthquake, so instrument recordings are simulated. The building is analyzed in its condition prior to the earthquake. It is found that, while hindcasting of the overall repair cost was excellent, prediction of detailed damage locations was poor, again implying either that as-built conditions differ substantially from those shown on structural drawings, or that inappropriate fragility functions were used, or both. We find that the parameters of the detailed particle filter needed significant tuning, which would be impractical in actual application. Work is needed to prescribe values of these parameters in general. Opportunities for implementation and further research. Because much of the cost of applying this RTLE algorithm results from the cost of instrumentation and the effort of setting up a structural model, the readiest application would be to instrumented buildings whose structural models are already available, and to apply the methodology to important facilities. It would be useful to study under what conditions RTLE would be economically justified. Two other interesting possibilities for further study are (1) to update performance using readily observable damage; and (2) to quantify the value of information for expensive inspections, e.g., if one inspects a connection with a modeled 50% failure probability and finds that the connect is undamaged, is it necessary to examine one with 10% failure probability?

Additional Information

A report on research supported by the CUREE-Kajima Joint Research Program Phase V under the supervision of K.A. Porter and J.L. Beck

Files

EERL2004-08.pdf
Files (2.7 MB)
Name Size Download all
md5:a7fc70c48a7e404dda6b85b35efdfae3
2.7 MB Preview Download

Additional details

Created:
August 19, 2023
Modified:
January 13, 2024