Improving the conversion accuracy between bridge element conditions and NBI ratings using deep convolutional neural networks

Abstract

Interest in the analysis of bridge element conditions in the U.S. has increased lately. Since 2014, the Federal Highway Administration is publishing bridge element data to better predict the performance of bridges for improving the allocation of management resources. However, because bridge elements data are still limited, bridge engineers often rely on National Bridge Inventory (NBI) condition ratings to predict the performance of bridges, which have been assembled since the 1970s. Therefore, it is valuable to investigate the correlation that exists between NBI ratings and element conditions to improve our knowledge of the latter. The objective of this article is to perform the analysis of both bridge element condition data and NBI ratings to back-map NBI deterioration curves into element deterioration profiles using deep convolutional neural networks. The proposed approach better estimates NBI ratings from bridge element conditions by at least 24.8% when compared to other techniques. By using an error tolerance of ±1 on the NBI ratings, the proposed procedure can accurately predict more than 90.0% of the ratings, while element deterioration rates have a 60% probability of being predicted within the range of the empirical rates.

Keywords:

• Artificial neural network
• bridge asset management
• bridge element deterioration
• convolutional deep learning
• machine learning
• National Bridge Inventory

Acknowledgements

The authors are thankful to Mr. Harjit Bal and Mr. Vijay Sampat of the New Jersey Department of Transportation (NJDOT) for their fruitful suggestions and for providing one of the element databases utilised in this study. The support of Project Manager Eddy Germain and Ankur Patel for BRP is greatly acknowledged. The contents of this paper reflect views of the authors who are responsible for the facts and accuracy of the data presented herein. The contents of the paper do not necessarily reflect the official views or policies of the agencies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The authors also acknowledge the financial support of the New Jersey Department of Transportation through the Bridge Resource Program (BRP) Project # 2017-02 which made this research possible.