Creep and shrinkage of concrete are complex phenomena that are difficult to model. Consequently, model prediction and real behavior in the field often show a significant difference. The conventional approach to improve the prediction begins with a choice of a constitutive model followed by calibration of the model parameters. However, due to the inherent error of constitutive models, the improvement is limited.

A new approach of self-learning simulation is adopted in the thesis to characterize the time-dependent behavior of concrete. The method does not require a predefined constitutive model. The material model becomes gradually more accurate during the simulation procedure using the measured displacements. A rate-dependent neural network material model is introduced and integrated with the self-learning simulation.

The proposed method is applied to both synthetic and experimental examples for verification and then applied to the field calibration of a concrete segmental bridge. Deflection measurements of the bridge during the construction are utilized to characterize the time-dependent behavior of the concrete in the field. The obtained material behavior is used to provide updated casting curves and estimation of the long-term behavior of the bridge.

The method integrates computational simulation and construction. It can be applied to characterize any material behavior, including the time-dependent behavior of concrete in segmental bridges. The study presents a scheme to improve the camber of segmental bridges using a self-learning simulation