Deutsch
Share
close
Share
close
  • Sprache / language
  • deen
Research Project

Anticipate risks.
Plan maintenance more effectively.

Research Project

Anticipate risks.
Plan maintenance more effectively.

Research Project

Predictive Maintenance in Rail Transport

 

As part of the research project “Diagnosis and Prognosis for Maintenance Management in Rail Transport,” IABG , on behalf of the German Center for Rail Transport Research (DZSF), investigated how different data sources and models can be combined to obtain additional diagnostic and predictive information for maintenance. Demonstrators for infrastructure and rail vehicles illustrate the potential for more proactive and needs-based maintenance.

 

From Condition Data to Predictive Decisions

A single data source is often insufficient for reliable maintenance decisions. Only by systematically linking different data and models is it possible to assess the current condition and predict its future development.

The research project therefore combined physical models, artificial intelligence methods, and image, geographic, route, and operational data. Two demonstrators show how this can generate additional diagnostic and predictive information for specific maintenance tasks.

 

The added value for the rail industry:

  • Identify risks earlier and assess developments proactively
  • Plan and prioritizemaintenance measures more effectively
  • Allocate resources based on actual needs rather than solely according to fixed intervals
  • Support safety and availability through better decision-making criteria
"Green Along the Tracks" Findings: Potential Hazards Posed by Individual Trees

Vegetation Management

Vegetation along railroad tracks must be regularly inspected and maintained to identify risks to railroad operations at an early stage. DB Guideline 882 defines, among other things, requirements for clearing and stabilisation zones. However, on-site inspections by qualified personnel are time-consuming and costly.

The demonstrator we developed maps out the entire process – from inventory and condition assessment through growth forecasting to the needs-based planning of inspections and measures. Multispectral and LiDAR data from trains, drones, or aircraft, as well as satellite data, can be used for this purpose. AI-supported methods analyse the data and make it possible to identify individual trees, detect deadwood, and determine height and distance from the track.

In conjunction with growth models, it is possible to estimate when a critical condition might be reached. This allows inspection intervals and maintenance measures to be better aligned with the actual risk and reduces unnecessary on-site inspections.

In addition, a study was conducted to determine how sight lines that must be kept clear at technically unsecured railroad crossings can be automatically identified and assessed for critical vegetation.

User Interface of the Demonstrator (Web Application)

Lifetime Wear and Tear on Vehicle Structures

Structural fatigue develops over a long period of time and is initially neither visible nor immediately recognisable as a loss of function. For maintenance tailored to actual needs, it is therefore crucial to know how heavily a vehicle structure has actually been stressed and how further operations will affect its remaining service life.

The vehicle demonstrator links track characteristics – such as speed, track alignment, switches, and stops – to the resulting local structural stresses. A trained machine-learning model predicts these stresses for a selected route; a structural integrity model then derives the accumulated damage from these predictions.

This allows routes already traveled to be evaluated in terms of their service life consumption to date and enables the estimation of additional stresses for planned routes. Maintenance could thus be better aligned with a vehicle’s actual usage profile. Further validation is required for application to real-world fleets.

What the Project Reveals About Predictive Maintenance

The two demonstrators address different tasks but follow the same principle: Only by linking complementary data sources and models can information be generated that goes beyond mere status monitoring.

This allows maintenance decisions to be more closely aligned with actual usage, the predicted condition, and specific risks. At the same time, the demonstrators highlight the prerequisites for operational deployment: up-to-date and appropriate data, application-specific validation, and integration into existing maintenance processes.

From Research to Practical Application

The demonstrators developed are proofs of concept and not products ready for immediate use. However, they demonstrate how data, physical models, and machine learning methods can be integrated into end-to-end diagnostic and predictive chains.

For specific applications, these approaches can be adapted to the respective infrastructure, vehicles, available data, and operational processes. In the future, such analysis chains could also serve as building blocks for a digital twin.

Customer Testimonial

Project: “Diagnosis and Prognosis for Maintenance Management in Rail Transport”

The project was characterised by highly competent, reliable and open collaboration throughout. Particularly noteworthy was the willingness to consider alternative proposals and different solution approaches, and to work together in finding pragmatic compromises. The project outcomes were documented to a high standard, coordination was transparent and goal-oriented, and the reliable way of working enabled efficient project delivery with minimal management oversight.

Overall, the project was delivered in a highly professional, efficient and collaborative manner, making it a genuine pleasure to work together.

Axel SimrothGerman Centre for Rail Traffic Research at the Eisenbahn-Bundesamt (DZSF)

The Project at a Glance

Research Project:
Diagnosis and Prognosis for Maintenance Management in Rail Transport
Client:
German Center for Rail Transport Research at the Federal Railway Authority (DZSF)
Project Duration:
2023/2026
Methods:
Model combination, data fusion, machine learning, geospatial data analysis, and structural integrity assessment

Contact