Data‐driven rapid damage evaluation for life‐cycle seismic assessment of regional reinforced concrete bridges

• Published in: Earthquake engineering & structural dynamics Vol. 51; no. 11; pp. 2730 - 2751
• Main Authors: Xu, Ji‐Gang, Feng, De‐Cheng, Mangalathu, Sujith, Jeon, Jong‐Su
• Format: Journal Article
• Language: English
• Published: Bognor Regis Wiley Subscription Services, Inc 01.09.2022
• Subjects: Ageing; Aging; Aging (artificial); aging corrosion effects; Concrete bridges; Corrosion; Corrosion effects; Damage assessment; damage state; Earthquake damage; Earthquake prediction; Earthquake resistance; Earthquakes; Learning algorithms; Machine learning; Marking and tracking techniques; Mathematical models; Modelling; Numerical models; Performance assessment; Performance testing; RC bridges; Regional analysis; Reinforced concrete; Seismic activity; seismic assessment; Seismic response; tagging
• DOI: 10.1002/eqe.3699
• ISSN: 0098-8847

Rapid and accurate post‐earthquake damage evaluation of regional reinforced concrete (RC) bridges is a key issue for assessing the seismic resilience of cities and communities. Especially, RC bridges are susceptible to the aggressive environment, which can induce time‐dependent aging effects such as corrosion, and thus, it should be considered in the assessment. This paper presents an approach for regional seismic performance assessment of RC bridges in a life‐cycle context based on machine‐learning techniques. The life‐cycle seismic demand and capacity of the bridges are, firstly, obtained by the elaborated numerical model, which includes the deterioration induced by the aging corrosion effect. Then, the tagging‐based damage state (green, yellow, or red) is easily obtained by comparing the pairs of demand and capacity through machine learning. Four hundred and eighty bridge models are generated to develop the machine‐learning models and the performance of the machine learning models is evaluated. Results show that the extreme gradient boosting (XGBoost) model has the best performance, which has an accuracy of 81% in predicting the damage states. The proposed approach is demonstrated with a single bridge example and bridges in a sample region. It is shown that the machine learning model can accurately predict the post‐earthquake damage states of the single bridge, and it can also rapidly assess the damage states of the bridges in the sample region. Approximately 30% bridges in the sample region will experience damage states shift after 100 years, which highlights the importance of considering the aging effects on the post‐earthquake damage assessment of bridges.