kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
A Fault Detection Framework Using Recurrent Neural Networks for Condition Monitoring of Wind Turbines
KTH, School of Electrical Engineering and Computer Science (EECS), Electrical Engineering.ORCID iD: 0000-0002-6428-2241
2021 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

The global energy system is experiencing a transition to a sustainable system with ambitious targets for increased use of renewable energy. One key trend for this transition has been the large introduction of wind power and integration into the electricity grid. In order to succeed in this transition, there is a need to develop efficient tools to support the handling of the assets. Asset management is a coordinated activity for the organization to get value from an asset. As the main part of asset management, maintenance includes all the technical and corresponding administrative actions to keep or restore the asset to the desired state in which it can perform its required functions. Traditional maintenance is usually based on scheduled monitoring and physical inspections. However, with new access to data and information about condition-based maintenance shows to be an efficient solution for asset management. This thesis explores data-driven solutions for electrical equipment to generate alerts towards potential operation risks, which targets digital, efficient, and cost-effective asset management. Specifically, the thesis investigates wind turbines.

This thesis proposes a fault detection framework for cost-effective preventive maintenance of wind turbines by using condition monitoring systems. The thesis utilizes the data from supervisory control and data acquisition systems as the main input. For log events, each event is mapped to corresponding components based on the Reliawind taxonomy. For operation data, recurrent neural networks are applied to model normal behaviors, which can learn the long-time temporal dependencies between various time series. Based on the estimation results, a two-stage threshold method is proposed as post-processing to determine operation conditions. The method evaluates the shift values deviating from the estimated behaviors and their duration time to attenuate minor fluctuations. A two-level condition monitoring system is constructed to apply the proposed fault detection framework, which targets to detect possible faults of components and conduct performance analysis of turbines. The fault detection framework is tested with the experience data from onshore wind farms. The results demonstrate that the framework can detect operational risks and reduce false alarms.

Abstract [sv]

Energisystem i världen över genomgår en omställning till ett hållbart system med ambitiösa mål för ökad användning av förnybara energikällor. En central trend i denna omställning har varit en markant ökning av elektricitet från vindturbiner och integrering i elnätet. För att lyckas med denna omställning behöver effektiva verktyg utvecklas för förvaltning av tillgångar och underhållsstyrning. Asset Management är en samordnad aktivitet för en organisation för att få värde från en tillgång där underhåll utgör en stor del. Traditionellt underhåll baseras vanligtvis på schemalagd övervakning och fysiska inspektioner. Men med ny åtkomst av data och information om tillgångar tillståndsbaserat underhåll visar sig vara en effektiv lösning för kapitalförvaltning. Denna avhandling undersöker datadrivna lösningar för elektrisk utrustning att generera varnar för potentiella driftsrisker, som är inriktade på digital, effektiv och kostnadseffektiv underhållsstyrning. Specifikt studerar avhandlingen vindkraftverk.

Denna avhandling föreslår ett ramverk för feldetektering för kostnadseffektivt förebyggande underhåll av vindkraftverk med tillståndsövervakning. Avhandlingen använder data från övervakningskontroll och datainsamlingssystem. För logghändelser mappas varje händelse till motsvarande komponenter baserat på Reliawind taxonomin. För driftdata används återkommande neurala nätverk för att modellera normalt beteende, vilka lära sig de tidsmässiga beroendeförhållanden mellan olika tidsserier. Baserat på uppskattade resultat föreslås en tvåstegs tröskelmetod som efterbehandling för att bestämma driftsförhållandena. Metoden utvärderar skiftvärden som avviker från uppskattat beteende och dess varaktighet för att dämpa mindre fluktuationer. Ett tillståndsövervakningssystem med två nivåer är konstruerat för att tillämpa det föreslagna feldetekteringsramverket, som är inriktat på att upptäcka eventuella fel hos komponenter och genomföra prestandaanalys av turbiner. Ramverket för feldetektering testas med erfarenhetsdata från vindkraftparker på land. Resultaten visar att ramverket kan upptäcka operativarisker och minska falska larm.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2021.
Series
TRITA-EECS-AVL ; 2021:4
Keywords [en]
asset management, condition monitoring, fault detection, preventive maintenance, recurrent neural networks, two-stage threshold, wind power
Keywords [sv]
kapitalförvaltning, underhållsstyrning, tillståndsövervakning, feldetektering, förebyggande underhåll, återkommande neurala nätverk, tvåstegs tröskel, vindkraft
National Category
Other Electrical Engineering, Electronic Engineering, Information Engineering
Research subject
Electrical Engineering
Identifiers
URN: urn:nbn:se:kth:diva-288175ISBN: 978-91-7873-741-3 (print)OAI: oai:DiVA.org:kth-288175DiVA, id: diva2:1513609
Public defence
2021-01-28, Kollegiesalen, Brinellvägen 8, Stockholm, 13:00 (English)
Opponent
Supervisors
Note

QC 20210104

Available from: 2021-01-04 Created: 2020-12-30 Last updated: 2022-06-25Bibliographically approved
List of papers
1. An Anomaly Detection Approach Using Wavelet Transform and Artificial Neural Networks for Condition Monitoring of Wind Turbines' Gearboxes
Open this publication in new window or tab >>An Anomaly Detection Approach Using Wavelet Transform and Artificial Neural Networks for Condition Monitoring of Wind Turbines' Gearboxes
2018 (English)In: 2018 POWER SYSTEMS COMPUTATION CONFERENCE (PSCC), Institute of Electrical and Electronics Engineers (IEEE) , 2018Conference paper, Published paper (Refereed)
Abstract [en]

This paper presents an anomaly detection approach using artificial neural networks and the wavelet transform for the condition monitoring of wind turbines. The method aims to attain early anomaly detection and to prevent possible false alarms under healthy operations. In the approach, nonlinear autoregressive neural networks are used to estimate the temperature signals of the gearbox. The Mahalanobis distances are then calculated to measure the deviations between the current states and healthy operations. Next, the wavelet transform is applied to remove noisy signals in the distance values. Finally, the operation information is considered together with the refined distance values to detect potential anomalies. The proposed approach has been tested with the real data of three 2 MW wind turbines in Sweden. The results show that the approach can detect possible anomalies before failure events occur and avoid reporting alarms under healthy operations.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2018
Keywords
condition monitoring system, neural networks, the wavelet transform, and wind power
National Category
Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
urn:nbn:se:kth:diva-238158 (URN)10.23919/PSCC.2018.8442916 (DOI)000447282400160 ()2-s2.0-85053108763 (Scopus ID)
Conference
2018 POWER SYSTEMS COMPUTATION CONFERENCE (PSCC)
Note

QC 20181107

Available from: 2018-11-07 Created: 2018-11-07 Last updated: 2026-06-10Bibliographically approved
2. An anomaly detection approach based on machine learning and scada data for condition monitoring of wind turbines
Open this publication in new window or tab >>An anomaly detection approach based on machine learning and scada data for condition monitoring of wind turbines
2018 (English)In: 2018 International Conference on Probabilistic Methods Applied to Power Systems, PMAPS 2018 - Proceedings, Institute of Electrical and Electronics Engineers Inc. , 2018Conference paper, Published paper (Refereed)
Abstract [en]

This paper presents an anomaly detection approach using machine learning to achieve condition monitoring for wind turbines. The approach applies the information in supervisory control and data acquisition systems as data input. First, machine learning is used to estimate the temperature signals of the gearbox component. Then the approach analyzes the deviations between the estimated values and the measurements of the signals. Finally, the information of alarm logs is integrated with the previous analysis to determine the operation states of wind turbines. The proposed approach has been tested with the data experience of a 2MW wind turbine in Sweden. The result demonstrates that the approach can detect possible anomalies before the failure occurrence. It also certifies that the approach can remind operators of the possible changes inside wind turbines even when the alarm logs do not report any alarms.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers Inc., 2018
Keywords
Condition monitoring, electricity generation, machine learning, preventive maintenance, and supervisory control and data acquisition systems
National Category
Other Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
urn:nbn:se:kth:diva-238018 (URN)10.1109/PMAPS.2018.8440525 (DOI)000451295600131 ()2-s2.0-85053114161 (Scopus ID)9781538635964 (ISBN)
Conference
2018 International Conference on Probabilistic Methods Applied to Power Systems, PMAPS 2018, 24 June 2018 through 28 June 2018
Note

Conference code: 138773; Export Date: 30 October 2018; Conference Paper; Funding details: CSC, China Scholarship Council; Funding text: This work is financed by Chinese Scholarship Council.

QC 20190115

Available from: 2019-01-15 Created: 2019-01-15 Last updated: 2026-06-10Bibliographically approved
3. Wind Turbine Health Assessment Framework Based on Power Analysis Using Machine Learning Method
Open this publication in new window or tab >>Wind Turbine Health Assessment Framework Based on Power Analysis Using Machine Learning Method
2019 (English)In: Proceedings of 2019 IEEE PES Innovative Smart Grid Technologies Europe, ISGT-Europe 2019, Bucharest, Romania: Institute of Electrical and Electronics Engineers (IEEE) , 2019Conference paper, Published paper (Refereed)
Abstract [en]

This paper proposes a performance assessment framework to estimate operation status of wind turbines. The overall objective is to propose a method for health assessment to support preventive maintenance strategies for wind turbines. The framework uses the data in the supervisory control and data acquisition systems as input. The framework consists of three main stages: power curve prediction, sliding window method analysis and performance assessment. At the first stage, k-means and density-based clustering are applied to eliminate noisy measurements. Then both parametric and non-parametric methods are applied to estimate the ideal power curve, which is used as a reference value to assess the actual one. At the second stage, the sliding window method is used to calculate the deviation between actual power data and ideal values, which indicates the real time performance of wind turbines. At the third stage, different performance zones are defined to assess health conditions. The proposed approach has been applied with the experience data of six onshore wind turbines from a single wind farm. The results indicate that the introduced framework can monitor the operation conditions and evaluate the performance of wind turbines.

Place, publisher, year, edition, pages
Bucharest, Romania: Institute of Electrical and Electronics Engineers (IEEE), 2019
Keywords
Health assessment, machine learning, maintenance, power curve, wind power
National Category
Energy Engineering
Identifiers
urn:nbn:se:kth:diva-268237 (URN)10.1109/ISGTEurope.2019.8905495 (DOI)000550100400058 ()2-s2.0-85075882283 (Scopus ID)
Conference
2019 IEEE PES Innovative Smart Grid Technologies Europe, ISGT-Europe 2019, Bucharest, Romania, September 29 - October 2, 2019
Note

QC 20200415

Available from: 2020-04-15 Created: 2020-04-15 Last updated: 2022-06-26Bibliographically approved
4. A fault detection framework using RNNs for condition monitoring of wind turbines
Open this publication in new window or tab >>A fault detection framework using RNNs for condition monitoring of wind turbines
2021 (English)In: Wind Energy, ISSN 1095-4244, E-ISSN 1099-1824Article in journal (Refereed) Published
Abstract [en]

This paper proposes a fault detection framework for the condition monitoring of wind turbines. The framework models and analyzes the data in supervisory control and data acquisition systems. For log information, each event is mapped to an assembly based on the IEA Reliawind taxonomy. For operation data, recurrent neural networks are applied to model normal behaviors, which can learn the long-time temporal dependencies between various time series. Based on the estimation results, a two-stage threshold method is proposed to determine the current operation status. The method evaluates the shift values deviating from the estimated behaviors and their duration time to attenuate the effect of minor fluctuations. The generated results from the framework can help to understand when the turbine deviates from normal operations. The framework is validated with the data from an onshore wind park. The numerical results show that the framework can detect operational risks and reduce false alarms.

Place, publisher, year, edition, pages
John Wiley & Sons, 2021
Keywords
condition monitoring, fault detection, recurrent neural networks, two-stage threshold, wind power
National Category
Electrical Engineering, Electronic Engineering, Information Engineering
Research subject
Electrical Engineering
Identifiers
urn:nbn:se:kth:diva-288169 (URN)10.1002/we.2628 (DOI)000621394100001 ()2-s2.0-85101516519 (Scopus ID)
Note

QC 20210308

Available from: 2020-12-30 Created: 2020-12-30 Last updated: 2026-01-30Bibliographically approved

Open Access in DiVA

fulltext(2668 kB)457 downloads
File information
File name FULLTEXT02.pdfFile size 2668 kBChecksum SHA-512
dbcf545c9c837ce22d3f8b51d7944bff0f7f5f1de4774ed2fef871c6be84e3084a7b66b5f687265f2ded0f6d14d1b5205544f6cd2803e7506c391baa10754250
Type fulltextMimetype application/pdf
errata(45 kB)112 downloads
File information
File name FULLTEXT03.pdfFile size 45 kBChecksum SHA-512
9d15740ad5e3b88d1ca7f9d9b833cdb0be617625bff4ee0a3c0be54b7088da501bad826aaf7dce86417a7036ee47610b5e31fb32199316aa110921a6a35d584d
Type fulltextMimetype application/pdf

Other links

zoom link for online defense

Authority records

Cui, Yue

Search in DiVA

By author/editor
Cui, Yue
By organisation
Electrical Engineering
Other Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 577 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 1436 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf