A computational approach based on high-performance beam finite element for predictive response in monitoring existing bridges

In recent years, the structural safety of existing bridges has become an increasingly relevant topic due to the age and extent of the Italian infrastructure assets. Several technologies, such as the application of machine learning techniques, have been developed to automate inspections and monitoring processes of existing bridges. One promising approach is the use of simulated data from numerical models to train data-driven algorithms for detecting structural damage. To improve the effectiveness of the algorithm training, it is necessary to create an extensive dataset including various damage scenarios. This procedure entails performing numerous nonlinear analyses, thereby highlighting the importance of adopting an efficient numerical model to reduce the computational effort. This work proposes a high-performance computational approach to predict the nonlinear response of reinforced concrete and prestressed concrete bridges. Specifically, this work adopts an advanced fiber beam element based on a damage-plasticity model, which offers superior computational efficiency, compared to 2D and 3D finite element models. The proposed damage-plastic model introduces two different damage variables for tensile and compressive behaviour to consider the re-closure of tensile cracks when moving from tension to compression states. To accurately assess the frequency variation due to the cracking of structural components, this research proposes a modification of the damage-plastic model which accounts for the partial closure of cracks. Both constitutive models are imple-mented in OpenSees software framework. Computational aspects and solution algorithms are extensively detailed in this thesis. Several applications are presented in this work to demonstrate the effectiveness of the proposed computational approach in simulating the nonlinear static and dynamic responses of concrete bridge structures. The advanced fiber beam element is validated by comparing numerical results with experimental measurements from tests conducted on reinforced concrete and prestressed concrete beams. Additionally, an application at the structural level of the proposed numerical method is discussed simulating a full-scale test of an existing prestressed reinforced concrete bridge. The application of the model within the new promising developments in Structural Health Monitoring (SHM) is explored. Especially, this research proposes an approach for training Artificial Neural Networks (ANNs) to detect structural damage using simulated data derived from numerical results. An unsupervised method has been employed to train a neural network. The prediction error of such network model is investigated as a suitable measure for the definition of a damage indicator. Finally, regarding the new advancements in vision-based techniques, this thesis also explores the integration of the proposed fiber beam element into the process of creating synthetic environment, that is virtual dataset generated to train algorithms of visual recognition systems. In conclusion, the integration of the advanced fiber beam model with an accurate constitutive law and machine-learning techniques shows promising potential for future innovations in the monitoring of existing bridges.