PhD "Collaborative and Adaptive AIs for Predictive Inspection of Network Infrastructure" M/F

about the role

Your role will be to conduct a PhD thesis on: "Collaborative and Adaptive AIs for Predictive Inspection of Network Infrastructures" 

• Global Context and Problem Statement

 The monitoring and maintenance of network infrastructures currently face several major challenges: high intervention costs, lengthy inspection times, subjectivity in evaluations, and insufficient inspection frequency. In this context, automating the inspection process represents a key strategic challenge for operators like Orange, as well as for other players such as EDF and Enedis. Resource optimization, reduction of operational costs, and improvement in inspection accuracy are at the heart of this issue.

• Scientific Objective – Expected Results and Challenges to Address

The objective of this PhD thesis is to design a network infrastructure inspection system based on collaborative AIs, leveraging adaptive embedded multi-vector (drones, vehicles) and multimodal (IMU, visible, GPS, Lidar) fusion. Key scientific challenges to address include:

1.     Preparing an annotated dataset,

2.      Upstream geometric fusion: developing methods to efficiently associate data from heterogeneous sensors while considering the context (GPS position, weather, vector speed, etc.),

3.     Adaptive multimodal fusion: designing dynamic fusion techniques capable of adapting to changing environmental conditions,

4.     Creating models that combine the segmentation of defective infrastructures with the assessment of their severity level and the prioritization of interventions.

The proposed approaches include the use of recurrent neural networks for temporal fusion, the development of multi-sensor auto-calibration algorithms, and the design of neural networks for geometric transformation estimation. Expected outcomes include validated demonstrators in real-world environments and patent filings