This paper presents a damage identification and performance assessment study of a four-story masonry-infilled reinforced concrete building in Sankhu, Nepal, using ambient vibration and point cloud data measurements. The building was severely damaged during the 2015 Gorkha earthquake. A set of accelerometers was used to record the ambient response of the building in order to extract its modal parameters, and a series of lidar scans were collected to estimate the surface defects of certain structural components. An initial model of the structure is created using a recently proposed strut model for masonry infills and a novel modeling approach for infilled RC frames. Dimensions are extracted from lidar-derived point cloud data in the absence of as-built drawings. The FE model updating is first performed through a deterministic formulation where optimal model parameters are estimated through a least squares optimization, and then through a Bayesian inference formulation where the joint posterior probability distribution of the updating parameters are estimated based on the prior knowledge of updating parameters and likelihood of measured data. The error functions for both formulations are defined as the difference between identified and model-predicted modal parameters. Two cases of model updating are performed using different parameterizations and different prior information about the damage. In the first case, updating parameters include walls and columns along the four stories of the building and exclude structural components observed to be severely damaged. The prior knowledge about structural component stiffness values is based on the expected material properties. In the second case of model updating, updating parameters include walls and columns of only the first story, and the prior stiffness values are estimated from the point-cloud measurements. The prior values are then updated using the vibration measurements. The damage identification results are in good agreement with visual observations and point cloud damage quantifications. The most probable model parameters in the Bayesian approach are also found to be in good agreement with the optimal results obtained in the deterministic formulation. Finally, it is shown that the probabilistic natural frequency predictions provide more realistic confidence bounds when both modeling errors and parameter uncertainties are accounted for in the prediction process.
Read full abstract