Long title: Knowledge-based optimization of fleet-asset management

Assignment to our (IMW) research priorities:

  • Knowledge-based Maintenance
  • Data analysis
  • CRISP-DM
  • Predictive Modeling & Machine Learning

 

Duration: 01.10.2023 – 30.09.2027

Abstract:

The maintenance management of subway fleets currently relies solely on the analysis of failure factors and the expertise of experts. Consequently, the possibilities for continuous optimization of fleet assets are now exhausted. Balancing safety, reliability, and costs poses a challenge that hinders optimal progress. Within the scope of this project with Wiener Linien, sensor data and failure data from a specific system of a rail vehicle, the V-train, are to be utilized to provide meaningful information on maintenance metrics. Incorporating multimodal data facilitates data- and knowledge-driven maintenance, enabling life-cycle-dependent and dynamic optimization decisions for fleet deployment. Thus, at any point in the life cycle, the optimal and maximum possible safety, reliability, and cost metrics are intended to be simulated.

 

Results:

Exploration and development of a methodology for mapping and simulating the dependencies among vehicle data, faults, maintenance data, reliability, availability, maintainability, and supportability (RAMS), and life cycle costs (LCC) within a life-cycle-dependent, dynamic, and knowledge-based model. This model should be capable of simulating optimizations or changes in fleet deployment or fleet maintenance and their effects.

 

Partners:

 

Supervised theses within the scope of the project

  • Structured Asset-Data for knowledge based maintenance of Rail Vehicles (Bachelor Thesis)

    This thesis develops a standardized target structure for asset and maintenance data of rail vehicles as a foundation for a Knowledge-Based Maintenance (KBM) system. The focus lies on data standardization, alignment with EU Regulation 2019/779, and the modeling of asset information using a knowledge graph and entity-relationship model to support the WIN project methodology.

  • Turning Raw Data into an Actionable Dataset for Metro Vehicle Maintenance (Bachelor Thesis)

    This thesis develops a software-based framework for transforming heterogeneous metro vehicle sensor streams and maintenance logs into structured, analysis-ready datasets for predictive maintenance. The approach combines memory-efficient HDF5 storage, expert-in-the-loop sensor relevance analysis, and an interactive GUI to ensure scalable, transparent, and reproducible data preparation in the WIN project.

  • Analysis of Maintenance Programs for Metro Vehicles (Bachelor Thesis)

    This thesis identifies maintenance-relevant components of metro vehicles and analyzes current maintenance strategies across literature and industry using a systematic review and comparative analysis. The results provide a structured component and strategy overview that supports data-driven and predictive maintenance modeling within the WIN project.

  • Pattern Recognition for Predictive Maintenance in Metro Trains (Diploma Thesis) 

    This thesis evaluates supervised and unsupervised machine learning methods for anomaly detection in real-world metro fleet data, using Random Forest and LSTM autoencoder models on sensor and asset-history features. The work demonstrates the strong impact of temporal window design and contextual asset data on predictive maintenance performance within the WIN project environment.

 

Contact details:

Dipl.-Ing. Andreas Steiner

Email: andreas.steiner@tuwien.ac.at