Exploiting Graph Neural Networks for Prescriptive maintenance of CERN’s technical infrastructure


Lorenzo Giusti

Lorenzo Giusti

Lorenzo Giusti is currently a senior research scientist at CERN working on geometric and topological deep learning for particle physics in the cryogenics group. Lorenzo holds a PhD in Data Science at La Sapienza, University of Rome, specialized in topological neural networks. His research includes a period of visiting at the University of Cambridge and as a research scientist intern at NASA’s Jet Propulsion Laboratory, where he led a project on Martian terrain modeling using spacecraft imagery and Neural Radiance Fields.


CERN, the European Organization for Nuclear Research, is the largest centre for scientific research in particle physics, and it is known for its complex system of systems comprising advanced particle accelerators and detectors. To fulfill the physics program and deliver the required luminosity for the experiments, advanced tools are required to operate, maintain, and guide device consolidation. It is, therefore, critical to monitor the activities objectively and guide the implementation of strategies to improve performance, optimize costs and highlight key areas needing prioritization. Moreover, the availability of reliable, cost-effective, and energy-efficient sensors entails growing data that captures the underlying phenomena happening within the CERN’s technical infrastructure. To identify failures early and perform prescriptive maintenance of such a complex system of systems, this project aims to reveal hidden dependencies and approach the maintenance operations within the technical infrastructure of the largest particle accelerator complex of the world using graph neural networks.