Please use this identifier to cite or link to this item:
Publication type: Conference paper
Type of review: Peer review (publication)
Title: Transfer learning approaches for wind turbine fault detection using deep learning
Authors: Zgraggen, Jannik
Ulmer, Markus
Jarlskog, Eskil
Pizza, Gianmarco
Goren Huber, Lilach
et. al: No
DOI: 10.36001/phme.2021.v6i1.2835
Proceedings: Proceedings of the European Conference of the PHM Society 2021
Volume(Issue): 6
Issue: 1
Page(s): 12
Conference details: 6th European Conference of the Prognostics and Health Management Society, online, 28 June - 2 July 2021
Issue Date: 29-Jun-2021
Publisher / Ed. Institution: PHM Society
ISBN: 978-1-936263-34-9
Language: English
Subjects: Predictive maintenance; Deep learning; Transfer learning; Wind turbine; Renewable energy; Anomaly detection
Subject (DDC): 006: Special computer methods
620: Engineering
Abstract: Implementing machine learning and deep learning algorithms for wind turbine (WT) fault detection (FD) based on 10-minute SCADA data has become a relevant opportunity to reduce the operation and maintenance costs of wind farms. The development of practically implementable algorithms requires addressing the issue of their scalabililty to large wind farms. Two of the main challenges here are reducing the training times and enabling training with scarce or limited data. Both of these challenges can be addressed with the help of transfer learning (TL) methods, in which a base model is trained on a source WT and the learned knowledge is transferred to a target WT. In this paper we suggest three TL frameworks designed to transfer a semi-supervised FD task between turbines. As a base model we use a Convolutional Neural Network (CNN) which has been proven to perform well on the single turbine FD task. We test the three TL frameworks for transfer between WTs from the same farm and from different farms. We conclude that for the purpose of scaling up training for large farms, a simple TL based on linear regression transformation of the target predictions is an attractive high performance solution. For the challenging task of cross-farm TL based on scarce target data we show that a TL framework using combined linear regression and error-correction CNN outperforms the other methods. We demonstrate a scheme that enables the evaluation of different TL frameworks for FD without the need for labeled faults.
Further description: Best Paper Award
Fulltext version: Published version
License (according to publishing contract): CC BY 3.0: Attribution 3.0 Unported
Departement: School of Engineering
Organisational Unit: Institute of Data Analysis and Process Design (IDP)
Published as part of the ZHAW project: Machine Learning Based Fault Detection for Wind Turbines
Appears in collections:Publikationen School of Engineering

Files in This Item:
File Description SizeFormat 
2021_Zgraggen-etal_Transfer-learning-approach-wind-turbine-fault-detection.pdf2.53 MBAdobe PDFThumbnail

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