open access

https://hdl.handle.net/2115/100430

Structural Damage Identification Using One-dimensional Convolutional Neural Network

英語

dynamic analysis

2026-07-01

Proceedings of the 10th International Symposium on Reliability Engineering and Risk Management (ISRERM 2026)

Aging transportation infrastructure poses increasing challenges, including higher maintenance costs and uncertainty regarding structural reliability and safety, making early damage detection and risk reduction essential. Structural Health Monitoring (SHM) plays a key role in meeting this need by leveraging sensor data to assess structural conditions and identify potential issues before they escalate. Modern SHM systems increasingly integrate machine learning to analyse complex vibration patterns and extract features that traditional methods may overlook. Convolutional Neural Networks (CNNs), with their ability to automatically learn spatial and temporal features directly from raw data, offer a powerful approach for localizing and quantifying structural damage without requiring preprocessing. This paper presents a CNN‑based method for identifying both the location and severity of damage in structural components, evaluated on two distinct datasets to demonstrate its versatility. The first dataset consists of simulated time‑domain responses from a bridge structure, while the second contains experimental modal data from a wooden cantilever beam. In both cases, damage is introduced through localized stiffness reductions. Results from the two datasets demonstrate strong predictive accuracy and consistent improvements over conventional artificial neural network models. Future work will explore the integration of Aimed Multilevel Sampling (AMS) to reduce computational effort and further improve prediction robustness.

会議発表論文

Version of Record