Predictive Maintenance · Real-time Analytics
Talgo needed a fleet monitoring system that could ingest train telemetry in real time, handle communication failures, process about 5GB of signal data per train per day, present operational information to engineers in less than one second, and scale to additional train series without bespoke redevelopment. The company also wanted to move maintenance from fixed preventive schedules toward predictive and condition-based interventions.
Reported impact
Talgo used TSMART to shift maintenance toward predictive and condition-based operations, improving fleet reliability, availability, and maintainability while reducing unplanned failures and saving inspection effort. The platform now supports proactive intervention before trains reach maintenance bases and can be extended to new train series or packaged as a service for third parties.
Atos and Talgo built TSMART, a fleet monitoring and predictive-maintenance platform for high-speed trains, using Google Cloud capabilities for data storage, real-time data processing, advanced geo analytics, and machine learning. Telemetry from trains is streamed into the platform in real time, where TSMART handles more than 30,000 signals per second per train from 100 trains around the world and processes roughly 5GB of data per train per day. Processed signals are exposed to maintenance engineers through engineer cockpit visualization screens with end-to-end latency of about one second. On top of the streaming layer, Talgo and Atos added condition-based and predictive maintenance capabilities delivered through an MLOps architecture so the team can monitor system health, predict equipment lifespan, increase inspection frequency, and detect failures earlier. Additional modules cover event analysis, gauge monitoring, configuration monitoring, graphing, notifications, and administration.
Train telemetry streams into TSMART in real time, where Google Cloud storage and processing services handle signal ingestion, analytics, and machine-learning workloads. Predictive-maintenance models delivered through the MLOps architecture evaluate equipment health and expected lifespan, then engineer cockpit screens present alarms, events, geolocation, and signal values with about one second of end-to-end latency for maintenance decisions.
"200 hours of maintenance inspection time projected to be saved per train per year"
"Currently, the system manages streaming and analysis of more than 30,000 signals per second per train from 100 trains around the world"
"The data and analytics platform behind TSMART was developed using Google Cloud capabilities for data storage, real-time data processing, advanced geo analytics and machine learning."
"The data is ingested in real time and presented to users in friendly visualizations, maintaining end-to-end latency of around one second from signal arrival to display"
"The predictive maintenance is not only based on AI models and rules. It was also delivered from an MLOps architecture"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
The 2025 People and Well-Being table totals 3,594 employees.
Related
NVIDIA Sample Aug 25, 2026 Siemens Siemens needed to help industrial teams overcome fragmented engineering tools, labor shortages, and slow design and maintenance workflows so they could design, simulate, and operate complex products and factories faster… 10%–15% reductions in capital expenditure NVIDIA CUDA · NVIDIA DGX B200 · NVIDIA DGX H200 
In-house Sample Aug 13, 2026 Lockheed Martin Static flight-hour tables and usage metrics ignored real-time consumption, causing chronic under-resourcing, limited stock effectiveness, and a reactive sustainment supply chain that could not keep pace with modern… predict component failure up to 72 hours before it occurs LogC2 dashboard · deep-learning models · digital data fabric NVIDIA Sample Jul 21, 2026 Wistron Corporation Wistron needed to scale production of NVIDIA AI infrastructure while improving quality, energy efficiency, and worker training across global plants, but traditional CFD studies, sparse defect data, and reactive factory… 10% energy reduction Apple Vision Pro · Cadence Reality Digital Twin Platform · NVIDIA Cosmos 
