Uncertainty propagation in reinforcement bond performance of 3D-printed concrete via generative-augmented ensemble learning
Baixi Chen, Xianlin Wang, Haijun Zhou, Sheng Jiang, Yiwei Weng
Abstract
The uncertainty in reinforcement bond performance poses a significant challenge to the reliable application of 3D printed concrete structures in practice. To address this issue, this study employs an ensemble learning method to develop a data-driven model, trained on a dataset augmented using a deep generative adversarial network. By integrating the data-driven model with Monte Carlo simulation, uncertainty propagation analysis is conducted, with concrete compressive strength identified as the primary source of uncertainty. With the coefficient of variation (COV) of concrete compressive strength ranging from 0.05 to 0.25, the influence of key parameters on bond strength uncertainty is systematically investigated. The results indicate that reinforcement type is the most critical factor: steel wire reinforcement amplifies uncertainty by more than 50% during propagation, whereas steel bar and steel nail reinforcements reduce uncertainty by over 60%. In addition, a concrete cover thickness greater than ten times the reinforcement diameter and a reinforcement bond index exceeding 0.075 are shown to effectively mitigate bond strength uncertainty. While larger aggregates and thinner print layers can reduce uncertainty levels, their use is not recommended due to potential compromises in the overall performance of 3D printed concrete. These findings provide practical guidance for designing reliable 3D printed concrete systems and contribute to the advancement of 3D concrete printing technologies in a safe and sustainable manner.