ED Sciences Physiques et de l'Ingénieur
Upgrading Robustness and Resilience Assessment of Urban Infrastructures Using Artificial Intelligence-Based Metamodeling for Smart Urban Design
by Minh Tuan BUI (I2M - Institut de Mécanique et d'Ingénierie de Bordeaux)
The defense will take place at 14h00 - Bâtiement A9.a, Amphithéâtre 3 Université de Bordeaux, Bâtiment A9.a, 351 Cours de la libération, 33405, Talence, France
in front of the jury composed of
- Sidi Mohammed ELACHACHI - Professeur des universités - Université de Bordeaux - Directeur de these
- Humberto YANEZ-GODOY - Maître de conférences - Université de Bordeaux - CoDirecteur de these
- Sylvie YOTTE - Professeure des universités - Université de Limoges - Rapporteur
- Franziska SCHMIDT - Ingénieure de recherche - Université Gustave Eiffel - Rapporteur
- Frédéric DUPRAT - Professeur des universités - INSA Toulouse - Examinateur
- Olivier PILLER - Directeur de recherche - INRAE - Examinateur
- Rasool MEHDIZADEH - Maître de conférences - École des Mines de Nancy - Examinateur
Urban water networks are essential to the functioning of society, with major challenges related to the preservation of public health, the economy, and the environment. These networks are fundamentally composed of buried pipes. Accelerated urbanization driven by population growth, combined with extreme climate phenomena, increases the risk of failure and threatens the long-term sustainability of these networks. Proactive maintenance is therefore necessary for their extension and/or replacement. The approaches developed in this thesis, based on a geomechanical perspective, propose decision-support tools for planning the full range of these operations. Among these approaches, traditional geomechanical modeling simulations of buried pipes rely on three-dimensional finite element calculations. These computations enable accurate representation of their complexity but become computationally prohibitive in the presence of uncertainties. Moreover, from a normative standpoint, the spatial variability of the soil is frequently neglected, and the soil is assumed to be homogeneous over very large sections, leading to inaccurate predictions. In this context, this thesis proposed a metamodeling methodology based on artificial intelligence to assess the geomechanical robustness and geomechanical resilience of buried pipes under uncertain conditions, achieving significant computational gains. The developed methodology comprised two main phases: (1) the creation of databases used to build three-dimensional finite element geomechanical models, incorporating global sensitivity analysis and modeling of spatial variability, particularly of the soil surrounding the pipes; (2) the construction of a metamodel based on an optimal experimental design, the evaluation of machine learning models using six different algorithms, and the development of a predictive model employing the Monte Carlo method. The obtained results demonstrate that the Young's modulus of the natural soil on which the pipe is laid is the dominant parameter influencing geomechanical robustness. The Gradient Boosting model proves to be the most effective with limited data and enables a reduction in simulation time of up to 99.86%. Geomechanical robustness and geomechanical resilience are evaluated using two indicators: the maximum displacement and the maximum stress of the pipe. Displacement highlights the dominant influence of soil type and correlation length, whereas stress exhibits reduced sensitivity to large correlation lengths. It is shown that an optimal diameter-to-thickness ratio of the pipe maximizes its geomechanical robustness and geomechanical resilience. Recommendations regarding different soil types in planning have been given in cases considering both geomechanical robustness and geomechanical resilience. A grid presenting different levels of geomechanical indicators is proposed to guide risk-based decision-making in specific sections of the network. By integrating artificial intelligence-based metamodeling and accounting for the spatial variability of the soil, this methodology ensures efficient, reliable, and computationally inexpensive management of urban water networks.