Publications / 2017 Proceedings of the 34rd ISARC, Taipei, Taiwan

Developing a Pattern Model of Damage Types on Bridge Elements Using Big Data Analytics

Soram Lim, Sehwan Chung and Seokho Chi
Pages 849-855 (2017 Proceedings of the 34rd ISARC, Taipei, Taiwan, ISBN 978-80-263-1371-7, ISSN 2413-5844)
Abstract:

Bridge plays an important role for transportation networks but its damage can threat public safety and hinder economic activity. Timely bridge management is needed not only to keep operating services but also to ensure traffic safety. Nowadays economically developed countries such as the U.S., Canada, Japan, and Korea have difficulties to maintaining over 30–year-old bridge structures, the aged structures. While the number of the aged bridges have increased, limited budget, time, and professional inspectors have downgraded a quality of periodic inspection. Therefore, the demands to understand bridge damage patterns explaining damage types, locations, or time frames have been emphasized. The primary purpose of this study is to develop a model using big data analyses to produce patterns that shows relationships between bridge-related variables and damage types on different bridge elements. The scope of this research covers the total of 6,773 bridges in Korea and uses Bridge Management System (BMS) data managed by the Korea Institute of Civil Engineering and Building Technology (KICT). The BMS data include fundamental bridge specifications, structural information, environmental information, and inspection and maintenance history. Some external data were also collected: contractor names and their performance ranking, temperature, and weather conditions. After preprocessing, key predictors (i.e., independent variables) selected by correlation analysis and random forest algorithms and targets (i.e., dependent variables) were both used to be clustered to represent critical damages Pattern analysis algorithms (i.e., mining association rules, clustering, rule-based classification) were then applied to discover a series of damage patterns. For the validation, cross-validation and matching with new inspection data were performed. The research findings have potential to support bridge inspectors with possible damage information through bridge element by element and decision makers for strategic preventive bridge maintenance planning.

Keywords: Bridge Management, Infrastructure Maintenance, Damage Pattern, Big Data analytics, O&M