Predictive maintenance and prognostic health management of wind turbine parks

Use Case Description

The increasing complexity of industrial assets and their operating environments, including the need for a more real time and dynamic health assessment poses new challenges for accurate prognostics and health management, i.e., the research field that links studies of failure mechanisms to system lifecycle management. Traditional methods, which are primarily knowledge-based, originating from among others the mechanical and electrical engineering domain, no longer suffice since the health evolution and possible failures are often tightly related to the environmental conditions and operation histories of a particular individual system. The increasing availability of sensor data monitoring different aspects of the system behavior has stimulated research into data-driven PHM techniques. 

The goal of this use case is to study how data-driven and knowledge-based models can reinforce one another by combining them in a hybrid methodology to perform more accurate prognostics and management of ageing industrial machinery.

Results

Sirris' activities focused on the development of methodologies for prognostic and health management. More specifically, we developed data-driven methods for detection and forecast of anomalies (and failures) [1] and underperformance in wind turbines [2] and to support the replacement strategy of worn cutting tools used in the milling industry [3].  Furthermore, we explored hybridization methods to combine the advantages of data-driven and knowledge-based methods. As a result we developed a method for icing-risk forecast [4].  We also developed a method to extract operating modes of a wind turbine along with  visual analytics to facilitate the monitoring of those turbines [5][6][7]. We also developed a method for performance profiling of rotating machinery assets under varying operating conditions [8]. Finally, we investigated the integration of transfer learning and active learning strategies for human supervision of the learning process to cope with the lack of labelled data (e.g., for anomaly detection). 

We implemented a demonstrator for integrating the building blocks developed by the different research partners involved in the use case. The demonstrator showcases the results to industrial stakeholders, easing the comparative validation of the resulting techniques.

Journal/Conference publications

[1] Murgia, A., Verbeke, R., Tsiporkova, E., Terzi, L., & Astolfi, D. (2023). Discussion on the suitability of SCADA-Based Condition monitoring for wind turbine Fault diagnosis through Temperature Data Analysis. Energies, 16(2), 620.

[2] Murgia, A., Cabral, H., Tsiporkova, E., Astolfi, D., & Terzi, L. (2023, May). Data-driven characterization of performance trends in ageing wind turbines. In Journal of Physics: Conference Series (Vol. 2507, No. 1, p. 012019). IOP Publishing

[3] Verbeke, R., De Pauw, L., Fingerhut, F., Jacobs, T., Goudemé, T. & Tsiporkova, E. (2023). Automatic tool wear inspection by cascading sensor and image data In 2023 ECML/PKDD 2023 Post-Workshops and Tutorials Proceedings

[4] Murtas, G., Cabral, H., & Tsiporkova, E. (2023, June). Data-driven estimation of blade icing risk in wind turbines. In 2023 IEEE International Conference on Prognostics and Health Management (ICPHM) (pp. 320-327). IEEE.

[5] M. Dhont, E. Tsiporkova & V. Boeva. Layered integration approach for multi-view analysis of temporal data. Lecture Notes in Computer Science, 2020

[6] M. Dhont, E. Tsiporkova & V. Boeva. Performance Profiling of Operating Modes via Multi-View Analysis Using Non-Negative Matrix Factorisation. Springer-Verlag, Recent Advancements in Multi-View Data Analytics, 2021.

[7] M. Dhont,E. Tsiporkova & V. Boeva. Advanced Discretisation and Visualisation Methods for Performance Profiling of Wind Turbines. Energies, special issue: Wind Turbine Monitoring through Operation Data Analysis, 2021

[8] Fingerhut, F., Klein, S., Verbeke, M., Rajendran, S. & Tsiporkova, E. (2023, June). Multi-view contextual performance profiling in rotating machinery. In 2023 IEEE International Conference on Prognostics and Health Management (ICPHM) (pp. 343-350). IEEE. This paper received the the best industry paper award

To make possible the successful development of the use case,  different research groups have collaborated together over the years. We would like to thank the multiple PhD students who committed the most to the use case:

PhDs

• Ahmed Hassan (UGent) : Dynamic Bayesian networks for condition monitoring of mechatronic systems

• Alejandro Morales Hernández (UHasselt): Essays on Machine Learning: Advances in Forecasting and Optimization

• Arash Heidari (UGent): Data-Efficient Multi-Objective Optimization

• Dandan Peng (LMSD - KU Leuven): Condition Monitoring of Rotating Machinery Based on AI Methodologies

• Fabian Fingerhut (Sirris): Project Adaptive and continual learning for context drift detection and mitigation in industrial applications

• Lorenzo Perini (DTAI - KU Leuven): Flexible and weak supervision techniques for learning

• Luca Stradiotti (DTAI - KU Leuven): Flexible and Limited Supervision for Anomaly Detection

• Michiel Dhont (Sirris) : Multi-source Trend Analytics for Intelligent Transportation Systems

• Nezmin Kayedpour (UGent) : Data-Driven Adaptive Operational Strategies for Wind Energy Conversion Systems Incorporating Frequency Containment Reserve

• Zhenxiang Cao (STADIUS - KU Leuven): Artificial intelligence segmentation time series

Outreach

External events and workshops

• On April 28th, 2022, Sirris organized a seminar on "AI for Prognostic and Health Management of your industrial assets". The workshop presents to companies the achieved results and it is meant to exchange insights on the current state of practice in industry. By discussing the intermediate results, the industrial stakeholders gain insights into the possibilities of data-driven and hybrid prognostic and health management approaches on their own data, which will enable them to better assess the usefulness for their context.

• On April 29th, 2021, Sirris organized the 'Meet the Experts' workshop, in which a panel of 4 representative companies that are active in the domain of Prognostic Health Management of Industrial Assets (3E, Yazzoom, I-care and Toqua) gave more insights in the current industrial state-of-the-art and several research partners presented their current realizations in the context of the use case. This allowed us to validate the current realizations and gather valuable feedback for the next steps.

• On June 22nd, 2021, Sirris organized a joint webinar with Flanders Make on 'How can AI enhance Smart Maintenance and Prognostic Health Management for Industrial Assets?'. In this workshop, the recent advances in the domain of Prognostic Health Management were discussed, as well as its implications to enhance the maintenance operations for industrial assets, resulting from the insights gained in the context of the two respective use cases from the Flanders AI Research Program. Several research partners that are active in these use cases presented their current results to a broad audience from industry and academia.

• Finally, Sirris presented the Flanders AI Research Program and relevant research results from the use case during the following outreach and intake activities that it organized:

  • EluciDATA Tech Talk on December 8th, 2020 and December 9th, 2021.

  • Industry seminar on ‘Fleet-based analytics for data-driven operation and maintenance optimization’ (April 29th, 2021)

  • Mastercourse session on ‘How to leverage AI to benchmark your industrial assets’ in June 15th, 2021 and March 8th, 2022

  • AI innovations driving the future of construction (May 3rd, 2022).

  • AI sessie op Advanced Engineering Expo (May 11-12th, 2022).

  • Flanders AI Forum (June 7th, 2022). Within the Forum organized with the support of AI Impulse program, Sirris presented a session on: State of the Art AI for Contextual anomaly detection

  • Asset Performance Conference (2023). Within the conference, Sirris presented the work: Handling lack of labeled data and trust in deep learning models: A drivetrain use case

Uptake

Follow-up projects

• Hybrid models for Flexible UtiLity scaLE STorage expansion (FULLEST).

• DIgital twin in Support of susTaInabLe and resiLient energy systems (DISTILL)

Basic Use Case Info

Period

2019-2023 

Contact

Alessandro.Murgia@sirris.be

Research Groups

Sirris (EluciDATA Lab)

KU Leuven DTAI/STADIUS

Ghent University  IDLab/EEDT/GAIM/KERMIT/LMSD

Hasselt University QM

What's next?  

AI for prognostic and health management of assets in the field