Initial Situation

SBB uses HFO systems (hot box and stuck brake detection) to measure the bearing and wheel temperatures of passing trains on open tracks. Stuck brakes or overheated wheel bearings damage rolling stock and can lead to axle or wheel failure, resulting in derailment. If temperatures exceed defined limits, an alarm is triggered. On track sections where a train cannot be stopped at short notice, such as in the Gotthard Base Tunnel, this alarm may arrive too late for optimal operational intervention. Earlier detection makes it possible to remove at-risk trains from service before they reach critical sections, thereby avoiding track blockages.

In drei aufeinanderfolgenden Schwarz-Weiß-Bildern sprühen Funken unter einem fahrenden Zug hervor.

Project Overview

Client
SBB

Area
Data & High Performance

Solution
Early detection of stuck brakes from measurement data.

System Environment
Train Control Center (TCC), HFO systems (hot box and stuck brake detection), SBB historical measurement database

Type
Study

The SCS Contribution

To achieve this, SCS combined historical HFO measurement data from several checkpoints and developed a model that evaluates temperature development on an axle-by-axle basis to derive the risk of an impending stuck brake alarm.

Early Detection Before the Alarm

Critical temperature developments are identified in the measurement series before the defined alarm limits are exceeded.

Trained with Real Operational Data

The HFO measurements and evaluations stored by SBB over many years form the data basis for training and validating the model.

Designed for Minimal False Alarms

Since each axle is evaluated individually, the model had to achieve an exceptionally low false alarm rate without overlooking actual alarm cases.

Developed Together with SBB Experts

SBB’s domain knowledge was specifically incorporated into the feature selection, parameters, and evaluation of the model.

Result

The study demonstrated that imminent stuck brakes can be detected based on consecutive HFO measurements even before a limit is exceeded. This allows affected trains to be removed from service before critical track sections, preventing disruption to subsequent operations.

1:1,000,000

Targeted maximum false alarm rate per analyzed axle.

1 Checkpoint Ahead

Prediction of a possible alarm at the next HFO checkpoint.

Services Provided

Algorithms, ML & AI

Data & High Performance

The bald man in the white shirt, smiling at the camera, represents the friendly SCS team.

Your Contact Person

Pascal Iselin

Algorithms, Analytics & AI
+41 43 456 16 00

For questions regarding machine learning, data analysis, and the early detection of critical system states.

Further Case Studies

  • Ein Güterzug rast auf den Gleisen vorbei und transportiert dabei Frachtcontainer.

    SBB

    Detection of Wheel Defects with Smart Algorithms

    Detecting wheel damage from checkpoint measurement data.

  • A white and red high-speed train glides environmentally friendly through the green landscape under a blue sky.

    SBB

    Trustworthy AI for the Measuring Wheelset

    Explainable AI for Force Measurement on Wheel and Rail.