.png){kind=icon}
.png){kind=icon}
Return
Anomaly Detection on Bridges Using Deep Learning With Partial Training
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3447571.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Bridges are exposed daily to environmental and operational factors that may cause weariness, fatigue, and damage. Continuous structural health monitoring (SHM) has been crucial to ensuring public safety, preventing accidents, and avert costly damages. In this regard, advances in Machine Learning and Big Data technologies have enabled automated, real-time structural monitoring. However, challenges persist, notably the scarcity of labeled data, rendering supervised learning impractical. Additionally, state-of-the-art methods demand extensive training data to generalize and achieve satisfactory performance, which can be limited in real-world scenarios. This paper presents a novel three-step method supported by advanced Machine Learning and signal processing techniques aimed at detecting anomalous signals. This method is trained solely on structural acceleration signals, eliminating the need for labeled data. Among the contributions of this work, it can be mentioned that a remarkable accuracy in the detection of structural damage was demonstrated quantitatively. (F1 Score of 93%), while requiring significantly less training data volume than alternative methods (less than 25% of the total) and opening up different lines of research.
Keywords:
Data models
Monitoring
Bridges
Anomaly detection
Training data
Machine learning
Deep learning
Unsupervised learning
Structural engineering
Safety
Defect detection
Public security
Structural health monitoring
bridges
damage detection
anomaly detection
machine learning
deep learning
unsupervised learning
Journal
IF:
3.6
Papers:
9.8W
Citations:
29.4W