Field calibration of time-dependent behavior in segmental bridges using self-learning simulation

Abstract The construction of segmental bridges often shows unexpected deflections, mainly due to the complex time-dependent behavior of concrete. The control of excessive deflections is important for the successful completion of their construction. It is difficult to accurately predict these deflections purely by conventional numerical simulations, and often the preferred method is field calibration. The quality of the calibration is dependent on the experience of the engineers involved. Therefore, it is difficult to systematically quantify and benefit from the gained experience, and to accumulate it. We employ a novel method called Self-learning Simulation (SelfSim), which integrates numerical simulations and field observations. SelfSim “learns” the time-dependent behavior of concrete in the field from displacement measurements. The gained knowledge is retrieved using extracted stresses, strains, rates of stresses and strains, and a neural network-based constitutive model. SelfSim is able to provide predictions about the remaining or future construction with the gained knowledge. Deflections can be controlled systematically by using the information provided by SelfSim. As a demonstration, the proposed method is applied to the field calibration of the Pipiral Bridge, a concrete segmental bridge built by the balanced cantilever method.