Image-assisted seismic damage evaluation for reinforced concrete interior beam-column joints

Abstract

Seismic damage quantification for reinforced concrete (RC) interior beam-column joints is crucial for post-earthquake stability and safety assessment of existing RC buildings. Loading-induced cracks during earthquake excitation usually degrade structural strength, ductility, and stiffness, underscoring the importance of rational assessment for RC joints. In this study, an image-assisted seismic damage evaluation method for RC interior beam-column joints was suggested by incorporating image-assisted technology, feature selection techniques, and regression models. Three seismic damage evaluation indicators derived from drift ratio (DID), stiffness (DIS), and strength (DIF) were first introduced as the poor prognosis markers of RC damaged joints. Secondly, image-assisted feature extraction method was presented by pre-processing image to a consistent binary format, with textural and geometric attributes of surface crack patterns extracted as part of features for three evaluation tools. Next, hybrid-feature parameters by combing image features and structural parameters were jointly selected by combing MRMR algorithms and F-test. Lastly, seismic damage evaluations of RC joints were performed based on four machine learning models (e.g., Neural Network, Gaussian Process, Random Forest, and Gradient Boosting) and 194 images from 37 available studies. Results indicate the highest accuracy for seismic damage derived from drift ratio can reach a robust R² of 0.83. SHapley Additive exPlanations (SHAP) analysis of the proposed model further deepens the understanding of seismic damage recognition and provides insights into the intricate relationships between predictors and damage evaluation tools.