Loughborough University
Leicestershire, UK
LE11 3TU
+44 (0)1509 263171
Loughborough University

Loughborough University Institutional Repository

Please use this identifier to cite or link to this item: https://dspace.lboro.ac.uk/2134/24272

Title: Condition monitoring of wind turbine drive trains by normal behaviour modelling of temperatures
Authors: Tautz-Weinert, Jannis
Watson, Simon J.
Issue Date: 2017
Publisher: © Dirk Abel, Christian Brecher, Rik W. De Doncker, Kay Hameyer, Georg Jacobs, Antonello Monti, Wolfgang Schröder
Citation: TAUTZ-WEINERT, J. and WATSON, S.J., 2017. Condition monitoring of wind turbine drive trains by normal behaviour modelling of temperatures. IN: Werkmeister, A.T. (ed). Conference for Wind Power Drives (CWD 2017), Conference Proceedings, Aachen, Germany, 7th-8th March 2017. Aachen: Dirk Abel, Christian Brecher, Rik W. De Doncker, Kay Hameyer, Georg Jacobs, Antonello Monti, Wolfgang Schröder [pub], pp.359-372
Abstract: Condition monitoring and early failure detection are needed to reduce operational costs of wind turbines, particularly for offshore farms where accessibility is restricted. Failure detection technologies should be simple and reliable in order to contribute to the overall aim of cost reduction. Operational data from the Supervisory Control And Data Acquisition (SCADA) system are a potential source of information for condition monitoring and have the advantage of being recorded at each turbine without the costs of additional sensors. Detection of drivetrain failures using these ten-minute data has been successfully demonstrated in the last five years. This paper summarises and evaluates different ways of so-called normal behaviour modelling of temperature using SCADA data, i.e. the prediction of a measured temperature under the assumption that the system is behaving normally. After training, the residual of modelled and measured temperature acts as an indicator for possible wear and failures. Multiple approaches are discussed: linear modelling, artificial neural networks in auto-regressive, feedforward and layer recurrent configurations, adaptive neuro-fuzzy inference systems and state estimation techniques. A case study with real data reveals differences of approaches, sensitivity to training data and settings of algorithms. Early failure detection of a gearbox failure is demonstrated, although challenges in achieving reliable monitoring without many false alarms become apparent.
Description: This conference paper appears here with the permission of the conference organisers and publishers.
Sponsor: This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 642108 (Advanced Wind Energy Systems Operation and Maintenance Expertise, http://awesome-h2020.eu/).
Version: Accepted for publication
URI: https://dspace.lboro.ac.uk/2134/24272
Publisher Link: https://www.cwd.rwth-aachen.de/1/conference/
ISBN: 9783743134560
Appears in Collections:Conference Papers and Presentations (Mechanical, Electrical and Manufacturing Engineering)

Files associated with this item:

File Description SizeFormat
2016-12-01-CWD2017-JTautzWeinert-paper-revision-submission-nocomments.pdfAccepted version481.09 kBAdobe PDFView/Open

 

SFX Query

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.