A Convolutional-Neural-Network-Based Damage Detection Method and Its Application to a Shake Table Test of an 18-Story Steel Frame Building Structure

Date: 4/24/2019

Time: 03:15 PM

Room: Elliott Bay

Rapid structural damage detection of buildings based on measurement is indispensable information for the decision making of reinforcement and retrofitting of structures after earthquakes. In this case, not the damage existing of the whole building, but the damage of individual members of the structure should be recognized. For steel frame building structures, beam-end fractures and buckling of columns cause the degradation of story stiffness, which will lead to the collapse or tilt of the whole structure. It is difficult to recognize this kind of local damage using conventional building damage detection methods such as the identification of response characteristics and the wave propagation methods. However, these damages can be recognized directly from the ripples of the waveforms which are generated by the shock due to the fractures. Recently, with the development of technology of image classification and objects detection, the Convolutional Neural Networks (CNNs) have been proved as one of the effective methods for feature extraction from 2D image data. In order to train an effective damage recognition machine, a large amount of training data is necessary. Because the seismic response data accompanied with structural damage is very rare, in this study, a classifier for detecting beam end fracture of steel frame structures from acceleration waveforms using CNN model was built. And a robust classifier for damaged data and undamaged data was developed. The CNN model was trained using a great number of acceleration waveforms which were generated by numerical simulations. Finally, the proposed method was applied to the shake-table-test data of a specimen of an 18-story high-rise steel frame building to verify the feasibility to recognize the beam end fractures from acceleration waveforms directly.

Presenting Author: Luyao Wang

Authors