This study presents a damage localization method for building structures using dynamic displacement responses based on a convolutional neural network (CNN). The proposed method is based on the interrelation of dynamic displacement response measured from a building in a healthy state. Based on the interrelation constructed by CNN in advance, damaged stories are localized by investigating the discrepancy of dynamic responses between healthy and damaged states. Hence, this allows identification of the location of damage without the use of structural response labeled with damage information in the CNN training stage. Specifically, to construct the CNN presenting the interrelation of structural response of the building under the healthy state, dynamic displacement response is utilized in the both input and output of CNN. Then, when the building is suspected to be damaged, the displacement response measured from the building is used as an input data in the previously trained CNN. Based on the discrepancy between the output obtained by inputting damaged state response into CNN and the response in a healthy state, the location of damage in the building is identified. To express this discrepancy, indicators for damage localization are newly defined in this study, which can be calculated by healthy and damaged state responses with the trained CNN. Through the investigation of the distribution of those indicators extracted from multiple stories of the structure, the location of damage in building structures is identified. We validated the proposed method for identifying damage locations through a numerical study and an experimental study.
Bibliographical noteFunding Information:
This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science, ICT & Future Planning, MSIP) (No. 2018R1A5A1025137).
All Science Journal Classification (ASJC) codes
- Civil and Structural Engineering
- Building and Construction
- Mechanics of Materials