Graph based knowledge models for capitalizing, predicting and learning: A proof of concept applied to the dam systems

Abstract

The capitalization and the analysis of historical information is nowadays a prerequisite for any effective risk management and assessment in a wide range of domains. Despite the development of mathematical models, procedures, support decision systems and databases, some engineering disciplines, such as civil engineering, remain resistant to the use of new digital technology due to the gap between the expectations of the engineers and the support that the tools may really provide. It is essential to propose a tool able to process both cross disciplinary and interdisciplinary knowledge flux and feedback from experience in a common and convenient unifying framework. The aim is to assist and to support engineering work and to make the task of knowledge modelling easier. The domain of dam systems is no exception to the rule. Dam failures are still commonplace. These failures stem from a lack of understanding about the complex relationships between three different factors: random hazards, the limit states of dam structures along with human activities and decisions. No generic and holistic approach is currently available that permits the processing of both knowledge and data, performs inferences and is easily usable for all types of users. This paper proposes the basic principles of a convenient design methodology for capitalizing, learning and predicting based on the formalism of conceptual graphs. The aim is to provide an easily usable tool able to (1) capitalise heterogeneous knowledge and store a database about dams, (2) issue alerts on current projects, (3) draw lessons from past dam failures and (4) tackle key issues in forensic civil engineering.