Description

The future PhD student will be registered at ENSAM (École nationale supérieure d'arts et métiers ).

 

Research Work

 

Scientific Context

 

Water leakage in urban underground pipelines can not only bring economic loss due to the wastage, but also can lead to structural damages of roads and buildings, which brings severe problems in the city infrastructure management. Addressing these issues requires efficient detection to minimize the impact on public safety. There are several methods for detecting water leaks, such as pressure/flow sensors or fiber optics, but they require sensors to be installed in advance. To complete the existing techniques, in this study, we will explore the application of ground penetrating radar (GPR) to detect underground water leakage, because GPR is non-destructive and high-efficient way.

 

 

 

Thesis Objective

 

This thesis aims to detect water leakages without the need for excavation based on radar methods. The physical basis of radar depends on the dielectric contrast between dry and wet soil around the pipes. This contrast appears as an anomaly in the radargram, providing an indication of the potential leak location.

 

Based on the previous research works, Leak-scan proposes to use 3D GPR data to optimize the detection process powered by AI techniques. Its objectives contain (1) inverse the time-domain data into spatial domain based on full wave inversion (FWI); (2) combine the simulated data with experimental data to build hybrid dataset, (3) use learning-based methods to detect the GPR images with and without leakage.  

 

Concretely, this doctoral project will aim to design a comprehensive GPR detection solution. Indeed, to implement this technique with high precision and obtain a real-time view of the soil condition, certain scientific hurdles remain to be overcome. The challenge lies in identifying water leaks with the consideration of the nature of soils, the multi-layered structure in the urban trench construction, the volume of leaked water, and the pipe dimensions, etc. Meanwhile, the AI ​​modeling of wave propagation in the ground will be applied in the interpretation of data. This process is supported by a federated learning infrastructure that combines data from multiple sources to build a global perspective. The entire framework leverages an edge-cloud infrastructure to ensure real-time responsiveness.

 

The data used in this thesis will be a hybrid one: the combination of the collected experimental data from the Project Radir and the simulated data from gprMax.

 

 

 

Thesis Work Planning

 

The provisional work plan is detailed as follows :

 

— Enrollment in Doctoral School 432 (ED SMI 432 - SCIENCES DES MÉTIERS DE L'INGÉNIEUR), state of the art review, and definition of a research methodology.

 

— Modeling the behavior of electromagnetic waves in unsaturated soil, paper submission.

 

— Development of a reliable and efficient PINNs-FWI model for the detection and characterization of water leaks. Paper submission

 

— Proposal of innovative methods for optimizing AI models and distributed computing, and paper submission

 

— Evaluation of field strategies for the experiments obtained by Project Radir, article submission

 

— Dissertation manuscript written, presentation of results, defense.

 

 

 

Expected Scientific/Technical Output

 

● The research results are expected to be published in top-tier international conferences and journals.

 

● The thesis will lead to the development of a complete GPR solution for pipeline leak detection