High-performance machine learning and data analysis for next-generation railway design

The MINERVE project aims to create a digital twin of the French railway infrastructure,which will then be used to assess the state of the railway and predict its evolution. To achieve such goal, the project requires a robust architecture capable of collecting, storing, and analyzing large amounts of data from various sources. During my thesis, we developed a theoretical big data architecture tailored to MINERVE's objectives, based on the Zaloni zone data lake architecture, designed to be implemented on the Paris-Saclay mesocenter, a high-performance computing facility.We then used this theoretical architecture to implement two use-cases that would support the development of MINERVE Digital Twins. The first use case is the development of a pipeline for the semantic segmentation of large railway point clouds, a needed work to enhance the automationof railway infrastructure monitoring and maintenance. The second use case focuses on the generation of high-frequency seismic waves from low-frequency ones, a necessary step in the making of an accurate Digital Twin, as the ability to detect earthquakes and other seismic events is crucial for the safety and reliability of railway operations.