역사 구조 건전성 평가를 위한 확률론적 손상 지수 개발

In this paper, damage indices are reported for unsupervised learning based pattern recognition. Since it is impossible to collect data through destructive testing under operation for establishment of training pattern, damage indices are developed for unsupervised learning. Threshold values are estimated using acceleration responses obtained from multiple points of a subway station. To verify the feasibility of the proposed damage indices, numerical analyses are performed using ABAQUS. It is assumed that damages occur at the points that stress is concentrated by considering earthquake. Dynamic train load excites the subway station.