Deep residual network framework for structural health monitoring
MetadataShow full item record
Funding and Sponsorship
Convolutional neural networks have been widely employed for structural health monitoring and damage identification. The convolutional neural network is currently considered as the state-of-the-art method for structural damage identification due to its capabilities of efficient and robust feature learning in a hierarchical manner. It is a tendency to develop a convolutional neural network with a deeper architecture to gain a better performance. However, when the depth of the network increases to a certain level, the performance will degrade due to the gradient vanishing issue. Residual neural networks can avoid the problem of vanishing gradients by utilizing skip connections, which allows the information flowing to the next layer through identity mappings. In this article, a deep residual network framework is proposed for structural health monitoring of civil engineering structures. This framework is composed of purely residual blocks which operate as feature extractors and a fully connected layer as a regressor. It learns the damage-related features from the vibration characteristics such as mode shapes and maps them into the damage index labels, for example, stiffness reductions of structures. To evaluate the efficacy and robustness of the proposed framework, an intensive evaluation is conducted with both numerical and experimental studies. The comparison between the proposed approach and the state-of-the-art models, including a sparse autoencoder neural network, a shallow convolutional neural network and a convolutional neural network with the same structure but without skip connections, is conducted. In the numerical studies, a 7-storey steel frame is investigated. Four scenarios with considering measurement noise and finite element modelling errors in the data sets are studied. The proposed framework consistently outperforms the state-of-the-art models in all the scenarios, especially for the most challenging scenario, which includes both measurement noise and uncertainties. Experimental studies on a prestressed concrete bridge in the laboratory are conducted. The proposed framework demonstrates consistent damage prediction results on this beam with the state-of-the-art models.
Showing items related by title, author, creator and subject.
Development and application of a deep learning–based sparse autoencoder framework for structural damage identificationPathirage, C.; Li, Jun; Li, L.; Hao, Hong; Liu, Wan-Quan; Wang, R. (2019)© The Author(s) 2018. This article proposes a deep sparse autoencoder framework for structural damage identification. This framework can be employed to obtain the optimal solutions for some pattern recognition problems ...
Wang, S.; Li, Jun; Luo, H.; Zhu, H. (2019)Most current studies with vibration-based Structure Health Monitoring (SHM) techniques has been focused on aboveground civil infrastructure. However, a few studies have been conducted for underground structures. The ...
Pathirage, C.; Li, Jun; Li, L.; Hao, Hong; Liu, Wan-Quan; Ni, P. (2018)© 2018 Elsevier Ltd Artificial neural networks are computational approaches based on machine learning to learn and make predictions based on data, and have been applied successfully in diverse applications including ...