04. April 2019

RISC Software: Machine Learning Increases Safety of Rail Travel

The basic research project “iTPP 4.0” developed findings that make intelligent points for rail transport possible. The findings allow reliable prediction at any moment of potential wear and tear or defects in the points. To achieve this, machine learning algorithms derive decisions on planned maintenance from railway infrastructure sensors in a self-learning process.

Points: critical components in railway infrastructure

Points are essential elements of railway infrastructure. The availability of the entire network depends to a large extent on reliable points. The Austrian rail network has over 10,000 of them. Points have an average working life of around 25 years, during which maintenance work must be carried out several times to ensure trouble-free operation. If, following a switching command from the signal box, a set of points does not reach the required safe position in time because, for instance, pressure in the points setting mechanism is too low, unplanned servicing of these points becomes necessary. In the event of a failure or unscheduled standstills, trains can pass the points only at low speeds or not at all. These “low speed sections” can cause delays. Network availability is compromised.

Intelligent points

Railway points are already fitted with a wide variety of sensors measuring force, position and pressure. To date, evaluation of the sensor data has made an approximately 30-per-cent reduction in the number of failures possible. In order to predict even earlier when a set of points is likely to cause a service interruption, sensors have now been used that are not typically found in railway infrastructure. These include ultrasonic sensors, optical sensors for recording acoustic signals, and vibration sensors. Additionally, external influences such as the environment, climate and weather have been taken into consideration. The large amount of data gained in this way is collected in a central location and processed for maintenance prediction with the aid of mathematical algorithms and machine learning approaches. This enables early prediction of potential wear and tear or defects and consequently prevention of service interruptions.

Machine learning

The challenge is to recognize relevant patterns in the data sources in order to determine the extent of wear and tear in a set of points. To this end, training data from existing points is used and the malfunctions detected by the existing points diagnosis systems are analysed and categorized. This data served as the basis for the development of intelligent Points.

Rail transport: a safe and sustainable future

Freight traffic and passenger traffic are constantly increasing, and with them CO2 emissions caused by mobility. Moving this volume of traffic onto the railways is therefore essential if society is to achieve low CO2 emissions. Rail transportation must therefore become even more eco-friendly, safe and reliable. This also entails timely maintenance to avoid service interruptions due to material wear and tear or fatigue caused by cracks and grooves.

voestalpine Railway Systems: a global player in the railway business

voestalpine Railway Systems, a division of the voestalpine concern, is a global market and technology leader in the railway infrastructure sector. To strengthen the company’s competitive position and maintain its role as a railway points technology leader, the challenge must be met of developing intelligent system solutions for railway transportation that enable a considerable reduction in life cycle costs while at the same time increasing availability.

RISC Software GmbH: specialist in machine learning and big data

With its core competences of symbolic computation, mathematics and informatics, RISC Software GmbH is a specialist in the intelligent linking of big data and machine learning approaches. This project was able to call on a large amount of expertise from research and development projects in a wide range of fields of production, logistics and big data.

Project details

The project “Intelligent Turnout Performance Prognosis 4.0 (iTPP 4.0)” was funded as part of the FFG (Research Promotion Agency) project with the FFG number 855345 and by the strategic business and research programme “Innovatives Oberösterreich 2020”, the business strategy of the province of Styria for 2020, and the Styrian research strategy.

  • Project name: Intelligent Turnout Performance Prognosis 4.0 (iTPP 4.0)
  • Project budget: EUR 538,151
  • Duration: 01/09/2016 – 31/12/2018
  • Funding: FFG – Research Promotion Agency in cooperation with the province of Upper Austria and the province of Styria
  • Partners: RISC Software GmbH and voestalpine SIGNALING Zeltweg GmbH

(C) RISC Software GmbH