Speaker
Description
The Gran Telescopio Canarias (GTC) is the largest optical and infrared telescope in operation and one of the Spanish Unique Scientific and Technical Infrastructures (ICTS). Its size, the range of technological domains involved and the precision it requires make it a complex system, whose operation relies on specialised personnel and on sophisticated control and monitoring tools.
The telescope's operational data are collected continuously and stored in a data lake built by the GTC engineering operations group: telemetry from its subsystems and services, instrument status, and environmental monitoring.
The objective now is to build a complete and interpretable prognostics and health management capability for the telescope. Watching the system is not enough. From the analysis of its operational data must come either an operational indication for the team, or an automatic action on the system. The expected result is fewer failures, better maintenance planning, higheravailability, and a protected scientific return.
Prognostics and health management is a layered chain: detect a deviation from normal behaviour, assess what it means, anticipate how it will evolve, and end in a recommended action. The project builds that chain on three pillars: statistical characterisation of nominal behaviour, machine learning models for anomaly detection, and simulation of operational scenarios that are rare or absent in the historical record.
The first pillars already operate on real subsystems of the telescope, validated against historical episodes recorded in the archive. Every alarm explains itself: it names the physical measurements behind it, so the engineering team can decide what to do.
We also discuss what this costs. Learning from an archive is not a single ingestion: fifteen years of telemetry are reprocessed every time the analysis changes. Time-critical analysis stays on premises, for operational safety. The heavy analytical workload runs on the Spanish Supercomputing Network, hosted at CESGA. We close with the transfer of these methods to other research infrastructures.