Damage detection coupling convolutional neural networks and numerical simulations

Citation
30th Eg ICE International Conference on Intelligent Computing in Engineering 2023
Contribution to Conference
30th Eg ICE International Conference on Intelligent Computing in Engineering 2023
Scopus ID
2-s2.0-85207185898
Damage detection in structural health monitoring (SHM) typically relies on global structural dynamic parameters, which may be hardly sensitive to the onset of structural damage. As a result, data analysis in SHM has been increasingly drawing from the field of machine learning (ML) for detecting subtle patterns in structural response data, indicative of structural damage. However, ML for damage detection requires structural response data corresponding to damage, which is hardly available. This paper proposes a damage detection approach coupling convolutional neural networks with numerical simulations. Specifically, the capabilities of convolutional neural networks are utilized for classifying structural response data into different damage scenarios. Furthermore, structural response data for training the convolutional neural networks is generated through numerical simulations. The proposed approach is validated through simulations of a steel pylon, showcasing that the proposed approach is capable of correctly classifying images of structural response data corresponding to different damage scenarios.