Project update

IoT sensing, edge AI and data orchestration for safer railways

Thinger.io presents the latest milestones of airCAT, the aerOS-funded initiative that fuses IoT sensing, edge AI and data orchestration to deliver real-time railway vibration monitoring and safeguard railways.

By capturing high-speed vibration data and turning it into real-time predictive-maintenance insights, airCAT is redefining track safety.

The goal: boost railway safety and infrastructure reliability by uniting IoT sensing, edge computing and machine learning for early anomaly detection.

The field campaign

Recording vibrations on a high-speed train

A real-world acquisition campaign aboard an IRYO train on the Madrid–Barcelona line.

During the recent IoT Solutions World Congress in Barcelona, our team ran a real-world acquisition campaign aboard an IRYO high-speed train on the Madrid–Barcelona line. Our fully integrated sensor suite captured 300K+ datapoints: 3-axis accelerometer, gyroscope and magnetometer readings, environmental metrics (temperature, humidity, luminosity, pressure) and high-precision GPS coordinates.

The campaign at a glance

Train
IRYO high-speed train
Line
Madrid – Barcelona
Datapoints captured
300K+
Motion
3-axis accelerometerGyroscopeMagnetometer
Environment
TemperatureHumidityLuminosityPressure
Position
High-precision GPS coordinates

Everything the on-board node recorded on the Madrid–Barcelona journey.

This rich dataset underpins AI models designed to flag vibration patterns that may signal track defects or structural issues.

RevPi · system log
May 19 18:31:58 RevPi20679 node[1501]: --- Datos de sensores ---
May 19 18:31:58 RevPi20679 node[1501]: Frecuencia de muestreo: 40.24 Hz
May 19 18:31:58 RevPi20679 node[1501]: Tamaño del archivo de log: 2.04 MB
May 19 18:31:58 RevPi20679 node[1501]: GPS: Lat=40.586055, Lon=-3.996745, Alt=866.60m
May 19 18:31:58 RevPi20679 node[1501]: Acelerómetro: [-157.00, 1.00, 840.00]
May 19 18:31:58 RevPi20679 node[1501]: Giroscopio: [0.00, 0.00, 0.00]
May 19 18:31:58 RevPi20679 node[1501]: Magnetómetro: [-1139.05, -502.81, 377.82]
May 19 18:31:58 RevPi20679 node[1501]: Heading: 244.66°
May 19 18:31:58 RevPi20679 node[1501]: Luz: 0.00 lx
May 19 18:31:58 RevPi20679 node[1501]: Temperatura: 31.64°C
May 19 18:31:58 RevPi20679 node[1501]: Humedad: 18.99%
May 19 18:31:58 RevPi20679 node[1501]: Presión: 91797.11 hPa

Console log from the RevPi during the field test

An industrial cabinet with terminal blocks, circuit breakers, a DC/DC converter, the orange Kunbus Revolution Pi modules and a Teltonika router
Kunbus RevPi edge device
The airCAT edge node, a wrapped enclosure with a dome sensor on top, placed in the luggage rack of a train carriage
IRYO train interior with the airCAT edge node
The edge node that travelled on the train: the Kunbus RevPi in its cabinet, and the enclosure in place inside the IRYO carriage.

Edge AI

Early prototypes in action

A lightweight onboard model turns the raw readings into a first map of vibration levels along the line.

Using the collected data, we prototyped a lightweight onboard ML model that computes a composite vibration-severity score and filters outliers, producing an initial geospatial heat-map of vibration levels. These results validate our end-to-end data pipeline and set a clear baseline for more advanced edge-native models with minimal latency.

A map of north-eastern Spain with the Madrid–Barcelona railway line coloured by vibration severity, mostly green with amber and red segments
The journey from Madrid to Barcelona: the vibration-severity score of the onboard model, plotted along the line.

aerOS domains

Orchestration in progress

We are deploying the aerOS architecture, structured in three key domains.

airCAT-OBD (On-Board Domain)

Edge data capture, local model execution and in-train visualizations.

airCAT-MLMTD (Model Trainer Domain)

Cloud-based AI training and optimisation.

airCAT-ENTRYPOINT (Management Domain)

Unified system oversight and orchestration.

Our Kunbus RevPi industrial edge device, running a 64-bit ARM OS, already hosts the Thinger.io stack and will shortly run aerOS containers (Service Fabric, Data Fabric, Local Orchestrator) for edge-to-cloud lifecycle management.

At the edge

Thinger.io on the train, in a Docker container

Deployed in a local Docker container, Thinger.io gives the on-board node its dashboards, its logging and its link to the aerOS Data Fabric.

Real-time dashboards

GPS and multi-sensor telemetry, live on the train.

Persistent CSV logging

Every reading stored on the node for offline AI training.

aerOS Data Fabric

Direct integration for a streamlined data flow.
A Thinger.io dashboard titled ClimaStick Real-Time with accelerometer, temperature and humidity charts and three colour toggles
The ClimaStick real-time dashboard on the Thinger.io platform, with the accelerometer and temperature and humidity series of the railway vibration monitoring node.

Dissemination

Working with UC3M and the aerOS team

The models are refined with a university research group and the results shared with the AIoT and railway communities.

We collaborate closely with the Applied Artificial Intelligence Group (GIAA) at Universidad Carlos III de Madrid to refine our AI models and validate them against extended datasets covering normal and anomalous segments. Alongside the aerOS team, we are preparing scientific papers and presentations to share architecture insights with the broader AIoT and railway communities.

What's next

The roadmap

From the first field campaign to full edge-cloud orchestration.

Our upcoming roadmap includes:

  • Full deployment of all aerOS domains across edge and cloud
  • Advanced AI training with enriched datasets
  • OTA model-update trials and complete edge-cloud orchestration
  • Publication of findings in leading journals and conferences

The roadmap as published in May 2025, after the first field campaign.

airCAT continues to showcase the transformative potential of IoT + AI + edge computing, powered by aerOS orchestration and the flexibility of Thinger.io. Stay tuned for fresh field results and deployment breakthroughs as we help shape the future of predictive railway maintenance.

Tell us what you think

We welcome your feedback to keep improving together. Fresh field results and deployment news will follow here.

Join the Discord communityTalk to us about your project