Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Computational Science, Engineering & Technology Series
ISSN 1759-3158
Edited by: Y. Tsompanakis, P. Iványi and B.H.V. Topping
Chapter 3

Neural Network-Based Techniques for Damage Identification of Bridges: A Review of Recent Advances

S. Arangio

Department of Structural and Geotechnical Engineering, Sapienza University of Rome, Italy

Full Bibliographic Reference for this chapter
S. Arangio, "Neural Network-Based Techniques for Damage Identification of Bridges: A Review of Recent Advances", in Y. Tsompanakis, P. Iványi and B.H.V. Topping, (Editors), "Civil and Structural Engineering Computational Methods", Saxe-Coburg Publications, Stirlingshire, UK, Chapter 3, pp 37-60, 2013. doi:10.4203/csets.32.3
Keywords: structural health monitoring, bridges, damage identification, neural networks, soft computing approach, Bayesian inference.

A modern approach for the design of bridges should be able to assess the structural health and performance during their entire life-cycle. From this perspective, effective structural monitoring assumes a primary role for detecting the circumstances that may eventually lead to damage and unsafe operations in a timely manner, so that costly replacements can be avoided or delayed by effective preventive maintenance.

The extraction of useful information from the large amount of measurement data, that are often time series of the accelerations, is still considered a challenge and suitable techniques are needed for assessing the health of the bridge from such data. Various methods for structural assessment and identification of damage, based on the use of neural networks, have been proposed in the past few decades and shown to be effective. In the first part of this work, a review of some case studies, focusing in more detail on those methods proposed in the last five years, is carried out. This analysis shows that most of the applications are still based on the use of traditional neural network models, where the internal architecture is assigned on the basis of rules of thumb or regularization procedures. In the second part of the work, the neural network model is re-examined from the Bayesian probability logic viewpoint. Using this approach, it is shown that the conventional learning approach can be derived as a particular approximation of the Bayesian framework, and it is schematically recalled how Bayesian inference can be applied to neural networks in order to optimize the architecture of the models and determine the relative importance of different inputs. In the last part, these Bayesian neural networks are applied for processing the data coming from the long term monitoring of the Tianjin Yonghe Bridge. The dataset was provided by the Asian-Pacific Network of Centers for Research in Smart Structure Technology (ANCRiSST) that made the monitoring data available to the researchers as a benchmark problem.

purchase the full-text of this chapter (price £20)

go to the previous chapter
go to the next chapter
return to the table of contents
return to the book description
purchase this book (price £75 +P&P)