반지도학습 딥러닝 모델을 이용한 진동 기반 구조물 손상 탐지 기술
Vibration-Based Structural Damage Detection Using Semi-Supervised Deep Learning Model
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 연구는 LK-Block Encoder를 반지도학습하여 구조물의 진동 응답을 통한 손상 탐지 프레임워크를 제안하고 이를 검증하기 위해전단 빌딩을 이용한 실험을 진행하였다. 1-D CNN 기반 은닉층 블록을 4개 직렬로 연결한 LK-Block Encoder를 반지도학습한 후 특징 공간 상에서 정상 상태의 진동 데이터와 손상 상태의 진동 데이터를 분류하는 임계치를 설정하여 진동 데이터로부터 손상을 탐지할 수 있는 프레임워크를 제시하였다. 이를 검증하기 위해 전단 빌딩의 기둥 두께를 다르게 하여 손상 시나리오를 가정한 후 제안한 프레임워크를 통해 손상 탐지를진행하였다. 그 결과, 큰 손상 상태에 대해서 97.03%의 정확도를 보였다. 정상 상태의 데이터 셋을 4개로 증가시켜 모델을 학습한 결과 작은 손상 상태에 대해서도 99.66% 정확도로 손상을 탐지하는 것을 보였다. 또한 탐지 대상과 다른 종류의 손상으로 모델을 학습시키는 경우에도 정상 상태 데이터를 증가시키면 손상 탐지 정확도가 높아지는 것을 확인하였다. 이를 통해 고유진동수 분석으로 탐지가 어려운 작은 손상에 대해서 제안한 프레임워크가 높은 성능으로 손상을 탐지하는 것을 보였다.
Abstract
This study proposes a framework for damage detection in structures through vibration responses using semi-supervised learning of an LK-Block Encoder, and validated through experiments with a shear building. The framework utilizes an LK-Encoder, which is composed of four serially connected 1-D CNN-based hidden layer blocks, and applies semi-supervised learning to set a threshold in the feature space for distinguishing between normal and damaged vibration data, thereby enabling damage detection from vibration data. To validate this approach, the study assumed damage scenarios by varying the thickness of shear building columns, and damage detection was conducted using the proposed framework. Results showed an accuracy of 97.03% for large damage condition. When the dataset of the normal condition was increased fourfold, small damage was detected with an accuracy of 99.66%. Additionally, it was confirmed that even when the model was trained with different types of damage from the target detection, increasing the amount of normal data improved damage detection accuracy. These results indicate that the proposed framework effectively detects even minor damage, which is difficult to identify through natural frequency analysis, with high accuracy.
참고문헌 (11)
- Huth, O., "Damage identification using modal data : Experiences on a prestressed concrete bridge", Journal of Structural Engineering, 131(12), pp.1898~1910 (2005)
- Shadan, F., "A frequency response-based structural damage identification using model updating method", Structural Control and Health Monitoring, 23(2), pp.286~302 (2016)
- Azim, M. R., "Damage detection of steel girder railway bridges utilizing operational vibration response", Structural Control and Health Monitoring, 26(11), pp.e2447~ (2019)
- Ghorbani, E., "Hybrid output-only structural system identification using random decrement and Kalman filter", Mechanical Systems and Signal Processing, 144, pp.106977~ (2020)
- Entezami, A., "Damage Detectionin Largely Unobserved Structures under Varying Environmental Conditions : An AutoRegressive Spectrum and Multi-Level Machine Learning Methodology", Sensors, 22(4), pp.1400~ (2022)
- Meixedo, A., "Online unsupervised detection of structural changes using train–induced dynamic responses", Mechanical Systems and Signal Processing, 165, pp.108268~ (2022)
- Salawu, O. S., "Detection of structural damage through changes in frequency : a review", Engineering structures, 19(9), pp.718~723 (1997)
- Mousavi, Z., "Developing deep neural network for damage detection of beam-like structures using dynamic response based on FE model and real healthy state", Applied Acoustics, 168, pp.107402~ (2020)
- Seventekidis, P., "Simulation Error Influence on Damage Identification Classifiers Trained by Numerical Data", Society for Experimental Mechanics Annual Conference and Exposition, pp.11~25 (2023)
- Rastin, Z., "Unsupervised structural damage detection technique based on a deep convolutional autoencoder", Shock and Vibration, 2021(1), pp.6658575~ (2021)
- Kim, B., "Deep Learning-Based Assessment of Civil Structure: Inspection using Instance Segmentation and Monitoring using Semi-Supervised Learning", Ph.D, pp.2023