Probabilistic Postearthquake ASCE 41-17 Compliant Performance Level Identification for Shear-Dominated RC Shear Walls via Crack Image Analysis

Abstract

In this paper, a probabilistic methodology based on image analysis for identifying the postearthquake performance level of reinforced concrete shear walls is proposed. A databank of 270 surface crack maps of 87 rectangular reinforced concrete shear walls obtained from quasi-static experiments at different drift ratio values is employed to develop the methodology. The specimens included within the databank exhibit a diverse range of structural design parameters and geometric properties. The complexity of the surface crack patterns is extracted by succolarity, lacunarity, and generalized fractal dimensions of the images of specimen images. For a specific fractal geometry index, fragility functions are developed to determine the probability of exceedance from an ASCE 41-17-compliant seismic performance level. Four typical probability distributions are used to generate the fragility functions: lognormal, gamma Weibull, and beta. Two goodness of fit tests including the K–S test and the Lilliefors test are used to assess the fragility curves’ fitness. The results show that, among the fractal geometry indices, succolarity has the best goodness of fit results, and the Weibull distribution fits the most seismic performance levels. In this study, seismic performance level indices are developed to optimize the goodness of fit by combining generalized fractal dimensions, lacunarity, and succolarity. A damaged wall specimen not in the databank at available performance levels is selected to present the application of the methodology. The results are also compared with other deterministic approaches available in the literature. By using the proposed methodology, the seismic performance level can be projected without the requirement for the structural characteristics data that are usually unavailable for seismically damaged buildings in the aftermath of an earthquake by only relying on image-derived data.