A BMS-oriented framework using artificial neural networks for fast multi-hazard risk assessment of bridges

The aim of this work is to develop a neural-based framework able to set an effective ANN to quickly estimate the level of various risks to which bridges are prone. Moreover, the tool needs to satisfy the necessities previously highlighted, in order to represent a significant step forward. Specifically, it needs to be able to: I) handle different risk typologies (and related components); II) using only documentary data; III) be adaptable enough to become a part of a larger framework or a BMS, providing information useful at different levels of management, planning, and decision-making, beyond just the inspection level. To achieve this result, different steps and goals need to be previous achieved, starting from those dedicated to gathering the necessary knowledge and highlighting areas of improvement to those related to promoting innovative solutions to the problem. Specifically, the necessary goals addressed in this work are as follows: I) Firstly, to review the state of the art regarding the multi-risk assessment for bridges, outlining practices, trends and limitations of existing procedures; II) Secondly, to investigate the current practice and existing BMS, highlighting analogies and differences among different systems; III) Thirdly, to study the most recent applications of ANN algorithms in predicting bridge conditions, also searching for implementations in existing BMSs, underscoring capabilities and limitations, and making comparisons between existing methodologies; IV) Finally, based on the knowledge built on the previous steps, to propose a novel framework to build efficient ANN for rapid multi-risk evaluations of bridges. Then, assess the effectiveness of the framework on a sufficiently large case study. The main result of this work will be an instrument useful for Management Bodies in supporting their management activities related to existing bridge. This thesis is organized into six main chapters, each addressing specific aspects related to the development and application of ANN of for BMS. Chapter 2 provides a comprehensive literature review on multi-risk hazard assessment, focusing on the methodologies and approaches used to assess risks associated with bridges under various hazards at a territorial level. Chapter 3 presents the evolution and the structure of the main existing BMS, exploring in detail their frameworks and modules, as well as their specificities. Chapter 4 presents the theoretical foundations of ANN, detailing the analytical aspects. Chapter 5 extends the discussion to the specific application of ANN in bridge management systems, highlighting their potential to enhance risk assessment and decision-making processes. Moreover, considering the results from previous Chapters, a novel BMS-oriented framework using ANN for fast risk assessment of existing bridges is proposed and discussed in detail. Chapter 6 proposes a case study where the developed framework is applied to real-world data of Italian bridges, demonstrating the effectiveness and flexibility of the proposed methodology. The thesis concludes with a summary of findings and suggestions for future developments in the field, emphasizing the significance of ANN in supporting bridge management practices and addressing emerging challenges in infrastructure resilience and safety.