Advanced Fiber Beam Finite Element Model for Neural Network Training in Vibration-Based Bridge Monitoring

Recent advancements in civil infrastructure monitoring have witnessed the increasingly high-performance sensor technologies and data-driven algorithms, opening up new possibilities for assessing structural conditions. In recent years, there has been a growing interest in leveraging the potential of Artificial Intelligence for civil infrastructure monitoring. One promising approach is the use of computational models to train and test data-driven algorithms aiming to tackle damage detection problems. To enhance the effectiveness of such procedures based on simulated data, this study proposes a high-performance beam finite element model for training a neural network model able to predict the dynamic response of the structure and for generating various damage scenarios. Compared to 2D and 3D finite element models, the advanced fiber beam model offers superior computational efficiency while accurately capturing the nonlinear behavior of structural elements. Specifically, a force-based beam finite element based on a damage-plasticity model is implemented to describe damage and degradation of materials in reinforced concrete girders. Through the simulation of the dynamic structural response under withe noise excitation, a neural network model representing the structure in the undamaged conditions is obtained. The prediction error of such network model is investigated as a suitable measure for the definition of a damage indicator able to detect the presence of damage (concrete cracks and reinforcement yielding). The integration of an advanced fiber beam model, accurate constitutive law and neural network models shows promising potential in the monitoring of existing bridges.