Abstract
After a significant earthquake, the updated status of the structural elements is usually determined based on a qualitative visual inspection. Although visual inspection provides a prompt assessment of the damaged elements, the output of this subjective method is influenced by the experience and decision of a trained inspector, which may vary from case to case. In this study, an innovative machine learning-based procedure is developed to automate damage state identification of non-ductile reinforced concrete moment frames (RCMFs) utilizing visual indices of crack patterns of the concrete surface. An extensive database including 264 surface crack patterns is constructed corresponding to 61 non-ductile beam-column joint specimens tested under the quasi-static cyclic loads at different drift ratios. Two visual damage indices, including the cumulative length of cracking and the areal density of crushing, are extracted using image processing filtering on the collected images. Supplementary information of the specimen, such as aspect ratio and concrete compressive force, is also considered as input variables. Various machine learning-based predictive models are developed for estimating the maximum drift ratio of a damaged non-ductile beam-column joint based on the available information of the specimen. Two scenarios are considered based on the accessibility of the compressive strength of the concrete as the input data, and a predictive model is obtained for each scenario. The peak experienced drift ratio predicted by the proposed methodology is finally used as an engineering demand parameter attributed to the corresponding fragility curves to determine the damage state of non-ductile RCMFs.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have