15 Mars – Thesis defense - Quynh Anh Hoang
14 h Visio
Using conceptual graphs to build failure knowledge model for forensic in civil engineering.
The construction industry is very complex and involves many risks. Although they provide essential benefits, infrastructures such as dams, bridges, etc., may, in case of failure, lead to severe consequences in terms of loss of human life and material damage. Numerous studies have looked at dam failures, which have led to changes in regulations to increase the safety of these structures. However, despite this progress, dam failures, sometimes significant, continue to occur throughout the world.
Risk management approaches provide tools to minimize the probability and/or consequences of failures. The traditional approach to risk management is based on identifying risk events, then assessing their probability and potential impact before deciding on risk mitigation measures. However, these approaches take little account of feedback and lessons learned from past failures when they could provide valuable lessons for advancing practice.
Forensic engineering is a discipline that aims at investigating failures in order to learn lessons and improve practices in the design, management, and operation of structures. Forensic investigation is the study of structures, materials, components, or infrastructure that have failed in order to determine the causes of those failures. Cross-analysis of different failures enables the identification of common factors at the origin of the failures. The results of these analyses can be used to improve risk awareness and induce better practices in the design and construction of similar structures. This can help avoid the repetition of these failures and contribute to the improvement of the safety of these structures.
In order to take advantage of failure information obtained from forensic investigations, numerous failure databases have been created over the past decade. However, the main weakness of these databases is the difficulty of their direct exploitation and the lack of intrinsic consistency of their data. Indeed, while they are useful for referencing cases and sometimes offering statistical data, they do not allow cross-analysis or automatically infer knowledge from past cases due to the lack of a unified vocabulary and inference engine. A database for this purpose should be organized in such a way that it is not only easy to find the information but also to use it for further study and analysis.
To address these issues, we have proposed to use Conceptual Graphs (CGs) to build a knowledge model of failures. The structure of the model we built makes it applicable to any type of failure and structure, but we have more specifically developed the model for dam failures. CGs are a formalism of representation of knowledge and reasoning in the form of a graph. They provide a unified vocabulary that forms vocabulary support allowing to share and reuse a representation of a phenomenon or a situation in the studied domain. CGs provide visual reasoning mechanisms that can be used to describe failure processes by simulating interactions between system components, but also to search for information and create new knowledge from existing knowledge.
In order to illustrate how the model works and its ability to deal with a variety of failures, we applied it to five case studies (two concrete dam failures and three embankment dam system failures). These examples are used to demonstrate the interest of the model to find similar failure cases, to propose possible causes of failure in the context of forensic engineering, and as a support to risk analysis.