Federated Learning Framework to support AI-Driven Prescriptive Maintenance in Large-Scale Cryogenic Infrastructures at CERN

This work presents a Federated Learning (FL) framework for anomaly detection and prescriptive maintenance tested on CERN’s helium compression systems. To keep its magnets in a superconducting state, the Large Hadron Collider (LHC) relies on an extensive cryogenic system, where predictive maintenance classical approaches are already central in ensuring operational efficiency and reducing infrastructure downtime. Leveraging FL, our solution enables AI-powered predictive maintenance via decentralized model training, using acquisitions from 20 custom motor-compressor systems located in 4 compressor stations around the 27 km LHC accelerator. This approach enables model performance improvement while optimizing energy consumption and maintaining consistent anomaly detection performance. The machine learning model used to detect deviations from nominal behaviour is based on deep autoencoders, an architecture that has proven to be particularly effective for reconstruction tasks. FL is enabled thanks to CERN’s FL platform, CAFEIN®, which has already demonstrated its effectiveness in medical applications. Experimental results confirm the efficacy of our framework, trained in federated learning, in detecting anomalies and estimating the RUL (Remaining Useful Life) of the compressors, with performance aligned with prior findings using centralized setup. This approach exemplifies how collaborative AI-driven methodologies can foster innovation, sustainability, and efficiency in industrial settings, accelerating collaboration between research centres, and at the same time complying with data privacy limitations.