Damage Progression in Shaking Table Tests of Steel Braced Frames and Disaster Damage Classification Based on Deep Learning of Point Clouds

Deep learning is applied to point clouds data acquired from shaking table tests of steel braced frames. The continued use is possible in cases where story drift angle is 1/200 or less and ductility factor is 2 or less. In deep learning for point clouds, classification accuracy by PointNet is 84%. PointNet++’s classification accuracy is 93%. PointNet++ performs hierarchical feature learning of local regions, making it possible to improve the accuracy rate through learning.