An Estimation of Damage Index of RC Column Using Classifier with Deep Learning of Shear Failure Characteristics

Abstract

Using photographs of earthquake damage that contain many characteristics of shear failure of RC partial walls, fine tuning of a pre-trained CNN model was performed to generate classifiers that classify the damage index into four levels (II and below, III, IV, and V) based on photographs of earthquake damage to shear-failing RC columns. We checked the learning conditions that affect the estimation accuracy in generating the classifiers. One of the generated classifiers was able to estimate the damage index to shear columns with an 85.3% accuracy rate. By using the damage index estimated by the classifier, the remaining seismic capacity of two damaged buildings was generally evaluated appropriately.