AI-driven Predictive Maintenance

TNT members involved in this project:
Lars Christian Gleichmann, M. Sc.
Quy Le Xuan, M. Sc.
Dr.-Ing. Marco Munderloh
Prof. Dr.-Ing. Jörn Ostermann

Maintenance is an essential and indispensable part of many industries, especially for those with high asset intensity, for example, automotive, airline, energy, and manufacturing. Predictive maintenance (PdM) is a modern maintenance strategy aiming not only to monitor and assess the current health state, but also to early detect possible faults, as well as to predict the remaining useful life of the systems of interest. The ability to predict when failures might occur will allow maintenance to be efficiently scheduled and performed in a proactive manner.

At TNT we follow the goal of developing AI solutions for industrial applications and showing the industry the potential of AI systems for further development. In this context, with a focus on AI-driven predictive maintenance as a representative application case, our research topics include (but are not limited to)

  • Remaining useful life (RUL) prediction,
  • Faults detection and diagnostics,
  • Anomaly detection.

The main challenges faced in practice when working with PdM are the limited amount of labelled data (run-to-failure, faulty, anomalous) available for training and the poor empirical performance due to the distribution/domain shift between training and test data.

In our research, we aim to develop reliable AI solutions that should show better generalization for cross-domain data with possible distribution shifts caused, for example, by the difference in operating conditions and the production variability. In addition, model uncertainty awareness and interpretability are further key aspects that we particularly take into account when designing and developing our solutions.

Various methods have been applied and developed within this project, including:

  • Uncertainty-aware deep neural networks
  • Self-supervised domain adaptation
  • Continual/Lifelong learning

  • Conference Contributions
    • Quy Le Xuan, Yeremia Gunawan Adhisantoso, Marco Munderloh, Jörn Ostermann
      Uncertainty-Aware Remaining Useful Life Prediction for Predictive Maintenance Using Deep Learning
      16th CIRP Conference on Intelligent Computation in Manufacturing Engineering, CIRP ICME, 2022
    • Yeremia Gunawan Adhisantoso, Quy Le Xuan, Christoph Kellerman, Marco Munderloh, Jörn Ostermann
      Introduction to Deep Degradation Metric in Smart Production Ecosystems
      16th CIRP Conference on Intelligent Computation in Manufacturing Engineering, CIRP ICME, 2022