ABSTRACT
The long-term behavior of concrete bridges is highly sensitive to creep as it may reduce their serviceability and durability. Inaccurate creep prediction may cause cracks and excessive long-term deflections, permanent non-aesthetic deflections, or difficulties with closure. Moreover, creep may lead to prestress losses, which cause stress redistribution and a decrease of structural safety. For that, concrete creep is treated more precisely in bridge standards than other time-dependent deformations. Creep coefficient is a well-understood parameter used in the Eurocode 2 (EC2) model for creep prediction. This paper aims to calibrate the EC2 model to obtain better prediction of creep. The updated EC2 model is obtained by implementing a correction coefficient into the creep coefficient formula. This correction coefficient is calculated using classification and regression tree (CART), one of the most popular machine learning algorithms for decision tree. CART is built and carried out using the statistical software R. In this study, the Northwestern University database is used, and eleven variables (water-cement ratio, aggregate-cement ratio, cement content, type of cement, compressive strength, age at loading, temperature, relative humidity, volume-surface ratio, load over compressive strength at loading and time) are considered to compute the correction coefficient. A comparison between the EC2 model and the updated model is made and the results indicate that the updated model significantly improves the creep coefficient prediction. Thus, the adoption of the updated model will improve the long-term serviceability and durability of concrete bridges.
