scada data interpretation for condition-based monitoring of wind turbines

8/13/2019 SCADA Data Interpretation for Condition-Based Monitoring of Wind Turbines

1/26


2/26

Kesheng Wang

Jan Ola Strandhagen

Dawei TuEditors

Proceedings of

IWAMA 2013

The Third International Workshop of

Advanced Manufacturing and Automation

27 November 2013, Trondheim, Norway


3/26

Preface

IWAMA International Workshop of Advanced Manufacturing and Automation began in

2010 as a joint seminar between SFI Norman (SINTEF and NTNU) and Shanghai Key Laboratoryof Manufacturing Automation and Robotics (Shanghai University). In 2013 IWAMA expanded toinclude academic and industrial experts in the fields of advanced manufacturing and automationworldwide when the University of Manchester and Tongji University joined as co-organizers. AsIWAMA becomes an annual event, we expect more universities and industry sponsors will join theinternational workshop as co-organizers.

Manufacturing and automation have assumed paramount importance and are vital factors forthe economy of a nation and the quality of life. The field of manufacturing and automation isadvancing at a rapid pace and new technologies are also emerging in the field. The challengesfaced by todays engineers are forcing them to keep on top of the emerging trends through

continuous research and development.IWAMA aims at providing a common platform for academics, researchers, practicingprofessionals and experts from industries to interact, discuss trends and advances, and share ideasand perspectives in the areas of manufacturing and automation.

IWAMA 2013 takes place in Trondheim, Norway, 27 November 2013, organized byNorwegian University of Science and Technology, SFI Norman/SINTEF, Shanghai University,University of Manchester and Tongji University. The program is designed to improvemanufacturing and automation technologies for the next generation through discussion of the mostrecent advances and future perspectives, and to engage the worldwide community in a collectiveeffort to solve problems in manufacturing and automation. Participants from different countries,such as Norway, China, UK, Germany, Italy and India attend the workshop.

Manufacturing research includes a focus on the transformation of present factories, towardsreusable, flexible, modular, intelligent, digital, virtual, affordable, easy-to-adapt, easy-to-operate,easy-to-maintain and highly reliable smart factories. Therefore, IWAMA 2013 has mainlycovered 3 sessions in manufacturing engineering:

1. Intelligent Manufacturing Technology2. Intelligent Logistics and Supply Chains3. Intelligent Diagnosis and Prognosis of Wind Turbines

The proceedings of IWAMA2013 provide up-to-date, comprehensive and worldwide state-of-

the-art knowledge of the manufacturing and automation science and engineering. All paperssubmitted to the workshop have been subjected to strict peer-review by at least 2 expert referees.Finally, 28 papers have been selected to be included in the proceedings after a revision process.We hope that the proceedings will not only give the readers a broad overview of the latestadvances, and a summary of the event, but also provide researchers with a valuable reference inthis field.

On behalf of the organization committee and the international scientific committee of IWAMA2013, I would like to take this opportunity to express my appreciation for all the kind support,from the contributors of high-quality keynotes and papers, and all the participants. My thanks areextended to all the workshop organizers and paper reviewers, NTNU and SFI Norman for thefinancial support, and co-sponsors for their generous contribution. Thanks are also given to Quan

Yu, Yi Wang, Cecilia Haskins, Tonje Berg Hamnes, Line Holien, Jorurnn Auth, and Lilan Liu for

I


4/26

their hard editorial work of the proceedings and arrangement of the workshop. Thank also goes toLasse Postmyr from Tapir Academic Press for producing the proceedings.

Kesheng Wang, Ph.D. Professor of NTNU

Chairman of IWAMA 2013

Trondheim, Norway, 25thOctober, 2013

II


5/26


6/26

Organized and Sponsored by

NTNU(Norwegian University of Science and Technology);

SHU(Shanghai University);

SINTEF(The Foundation for Scientific and Industrial Research)

SFI Norman(Center for Research-based Innovation Norwegian Manufacture Future)UM (The University of Manchester)

TU (Tongji University)

Honorary Chairs:Prof. Minglun Fang

General Chairs:Prof. Kesheng Wang, Prof. Jan Ola Strandhagen and Prof. Dawei Tu

Local Organizing Committee:Prof. Jan Ola Strandhagen, Prof. Kesheng Wang, Prof. Heidi Dreyer, Lars Tore Gellein, Quan Yu,Tonje Berg Hamnes, Associate Prof. Cecilia Haskins, Jorunn Auth

International Program Committee:

Prof. Jan Ola Strandhagen, NorwayProf. Kesheng Wang, NorwayAssociate Prof. Per Schjlberg, NorwayProf. Heidi Dreyer, Norway

Associate Prof. Erlend Alfnes, NorwayLars Tore Gellein, NorwayGeir Vevle, NorwayAssociate Prof. Hirpa L. Gelgele, NorwayAssociate Prof. Wei Deng, NorwayLecturer Yi Wang, UKProf. Wladimir Bodrow, Germany

Prof. Janis Grundspenkis, LatviaProf. Hans Henrik Hvolby, DenmarkProf. Dawei Tu, ChinaProf. Minglun Fang, China

Prof. Tao Yu, ChinaProf. Binheng Lu, ChinaProf. Ming Chen, ChinaProf. Yun Chen, ChinaProf. Xinguo Ming, ChinaProf. Yingjie Yu, ChinaProf. Yayu Huang, China

SecretariatTonje Berg Hamnes (Norway) and Prof. Lilan Liu (China)

III


7/26

About Editors

Prof. Kesheng Wang holds a Ph.D. in production engineering from the Norwegian

University of Science and Technology (NTNU), Norway. In 1993 he was appointedProfessor at the Department of Production and Quality Engineering, NTNU. He is adirector of the Knowledge Discovery Laboratory (KDL) at IPK, NTNU at present. He isalso an active researcher and serves as a technical adviser in SINTEF. He was electedmember of the Norwegian Academy of Technological Sciences in 2006. He has published17 books, 10 book chapters and over 214 technical peer-reviewed papers in international

journals and conferences. Prof. Wangs current areas of interest are intelligentmanufacturing systems, applied computational intelligence, data mining and knowledgediscovery, swarm intelligence, condition-based maintenance and structured light systemsfor 3D measurements and RFID.

Prof. Jan Ola Strandhagenis a research director of the research center SFI NORMAN atSINTEF. He is also Professor at Department of Production and Quality Engineering, the

Norwegian University of Science and Technology (NTNU). He holds a PhD inProduction Engineering from NTNU (1994). His research has focused on productionmanagement and control, logistics, manufacturing economics and strategies. He hasmanaged and performed R&D projects in close collaboration with a wide variety of

Norwegian companies and participated as researcher and project manager in severalEuropean projects.

Prof. Dawei Tureceived his bachelors degree, masters degree and PhD from ZhejiangUniversity in 1987, 1989 and 1993 respectively. Since 1993 he joined ShanghaiUniversity, and became a professor in 2000. Now he is the Executive Dean of the Schoolof Mechatronics Engineering and Automation, Shanghai University. His researchinterests include optoelectronic detecting, precision mechanics and instrumentation, 3Dvision and vision-based robot servo-control.

IV


8/26

Table of Contents

Preface............................................................................................................................... I

Keynotes

Knowledge Dynamic in Software Applications for Process ControllingWladimir Bodrow ......................................................................................................................... 1

Toward Intelligent Manufacturing: a Label Characters Marking and Recognition

Method for Steel Products with Machine VisionQijie Zhao, Dawei Tu and Peng Cao ............................................................................................ 9

The Development of Intelligent and Integrated RFID (II-RFID) System for

Manufacturing IndustriesKesheng Wang ........................................................................................................................... 25

Responsible and Dynamic LogisticsJan Ola Strandhagen ................................................................................................................... 49

Session IMT: Intelligent Manufacturing Technology

Development and Deployment of ACT System - RFID SolutionsJan Erik Evanger and Fredrik W. Goborg .................................................................................. 53

Assessment of Geometric Tolerance Information as a Carrier of Design Intent to

Manufacturing and InspectionHirpa G. Lemu ........................................................................................................................... 59

Phase to Depth Conversion Algorithm by Using Epipolar Restraint in Object-

adaptive Fringe Projection TechniqueJunzheng Peng, Yingjie Yu, Wenjing Zhou and Mingyi Chen .................................................. 71

Computer Vision Supported by 3D Geometric ModelingSven Fjeldaas ............................................................................................................................. 81

Studies on the Intelligent Manufacturing Technologies of the Fan Blade

Wenhua Zhu, Baorui Li, Bin Bai, Hu Han and Minglun Fang ................................................... 93Virtual Simulation of Production Line for Ergonomics EvaluationMing Chen and Jinfei Liu ......................................................................................................... 113

A Point Cloud Registration Strategy Combining Particle Swarm Optimisation

andIterative Closest Point MethodQuan Yu and Kesheng Wang ................................................................................................... 123

Graph Representation of N-Dimensional SpaceTomasz Kosicki ........................................................................................................................ 135

Study of the Role of Virtual Engineering Technologies in DesignHirpa G. Lemu ......................................................................................................................... 147

V


9/26

Session ILSC: Intelligent Logistics and Supply Chains

Lean Thinking in Product Development: Looking into Front-load Effects Using aSystem Dynamics ApproachAlemu Moges Belay and Torgeir Welo.................................................................................... 157

Ontology Based Interoperability Solutions within Textile Supply Chain: Review

andAnalysisYi Wang, Rishad Rayyaan and Richard Kennon ..................................................................... 171

A Framework for Real Time Production Planning and ControlEmrah Arica and Daryl J. Powell ............................................................................................. 185

Sources of Complexities in New Product and Process Development ProjectsGiedre Pigagaite, Pedro P. Silva, Bassam A. Hussein ............................................................. 197

Investigating the Fit of Planning Environments and Planning Methods, the Case

of anAutomotive Part ManufacturerPhilipp Spenhoff, Marco Semini, Erlend Alfnes and Jan Ola Strandhagen ............................. 211

Investigating the Applicability of Lean Principles in a Small, One-of-a-kindProduction CompanyLayli Dahir and Daryl Powell .................................................................................................. 221

Introducing Portfolio Maps as Basis for Standardization and Knowledge-basedDevelopmentSren Ulonska and Torgeir Welo ............................................................................................. 235

Incentive Regulation of Banks on Third Party Logistics Enterprises in Principal-

agent-based Inventory FinancingXuehua Sun and Xuejian Chu .................................................................................................. 249

Automation in the ETO Production Situation: The Case of a Norwegian Supplier

of Ship EquipmentBrge Sjbakk, Maria Kollberg Thomassen and Erlend Alfnes .............................................. 259

Competing in the Shell Eco-marathon. Challenges and Lessons LearnedBassam A. Hussein and Fariborz Ali Haidrloo ........................................................................ 271

Tackling the Storage Problem through Genetic Algorithms

Lapo Chirici and Kesheng Wang ............................................................................................. 281

Session IDPWT: Intelligent Diagnosis and Prognosis of Wind Turbines

(WINDSENSE Project)

Evaluation of Thermal Imaging System and Thermal Radiation Detector for

Real-time Condition Monitoring of High Power Frequency ConvertersAnders Eriksen, Dominik Osinski and Dag Roar Hjelme ........................................................ 295

SCADA Data Interpretation for Condition-based Monitoring of Wind TurbinesKesheng Wang, Vishal S. Sharma and Zhenyou Zhang ........................................................... 307

VI


10/26

Wind Turbine Fault Detection Based on SCADA Data Analysis Using ANNZhenyou Zhang and Kesheng Wang ........................................................................................ 323

Models for Lifetime Estimation - An Overview with Focus on Applications to

WindTurbinesThomas M. Welte and Kesheng Wang ..................................................................................... 337

VII


11/26

List of Authors

Alfnes, E. 211, 259

Arica, E. 49, 185

Bai, B. 93Belay, A. M. 157

Bodrow, W. 1

Cao, P. 9

Chen, M. 113

Chen, M. 71

Chirici, L. 281

Chu, X. 249

Dahir, L. 221

Eriksen, A. 295

Evanger, J. E. 53Fang, M. 93

Fjeldaas, S. 81

Goborg, F. W. 53

Haidrloo, F. A. 271

Han, H. 93

Hjelme, D. R. 295

Hussein, B. A. 197, 271

Kennon, R. 171

Kosicki, T. 135

Lemu, H. G. 59, 147Li, B 93

Liu, J. 113

Osinski, D. 295

Peng, J. 71

Pigagaite, D. 197Powell, D. J. 185, 221

Rayyaan, R. 171

Semini, M. 211

Sharma, V. S. 307

Silva, P. P. 197Sjbakk, B. 259

Spenhoff, P. 211

Strandhagen, J. O. 49, 211

Sun, X. 249

Thomassen, M. K. 259Tu, D. 9

Ulonska, S. 235

Wang, K. 25, 123, 281, 307, 323, 337

Wang, Y. 171

Welo, T. 157, 235

Welte, T. M. 337

Yu, Q. 123

Yu, Y. 71

Zhang, Z. 307, 323

Zhao, Q. 9Zhou, W. 71

Zhu, W. 93

VIII


12/26

SCADA Data Interpretation for Condition-based Monitoring of

Wind Turbines

Kesheng Wang

1

, Vishal S. Sharma

1, 2

and Zhenyou Zhang

1

1Department of Production and Quality Engineering, Norwegian University andScience and Technology, Norway2Dept. of Industrial and Production Engineering, National Institute of Technology,India

Abstract

Most of the wind turbines have of some kind of CBM devices/systems, which provide the

information about the device to the central data base i.e. SCADA database. Thesedevices/systems make use some kind of data processing techniques/methods in order tolocate faults. Finally all the information keeps on recoding in the SCADA data base. Sothe endeavour in this article is, to review the techniques/methods/algorithms developed,to carry out diagnosis and prognosis of the faults, based upon SCADA data obtained fromwind turbines. The focus here is also on the Computational Intelligence methodsdeveloped.

Keywords

SCADA data, CI, CBM, WT

AbbreviationsBT: Boosting Regression TreeOEM: Original equipment manufacturerCART: Standard classification & Regression treeFMECA: Failure Analysis, Failure Mode Effect and Criticality AnalysisCBM: Condition Based MonitoringREP: Representative TreeCBMS: Condition Based Monitoring SystemDE: Differential evolutionkNN: k Nearest Neighbor NetworkTPM: Total Productive MaintenanceANN: Artificial Neural NetworkWIP: Work in progress

NNE: Neural Network EnsembleES: Evolution StrategyRF: Random ForestRCM: Reliability Centred MaintenanceSCADA: Supervisory Control & Data AcquisitionPSO: Particle Swarm optimization

SVM: Support Vector Machine

307

International Workshop of Advanced Manufacturing and Automation 2013 (IWAMA 2013)


13/26

CMMS: Computerised Maintenance and Management SystemsWT: Wind TurbinePCA: Principal Component analysisMLP : Multi-Layer PerceptronOEE: Overall Equipment Effectiveness

CI: Computational IntelligenceFFT: Fast-Fourier transformBT: Bragging TreeTSA: Time synchronous averagingFIS: Fuzzy inference systemANFIS: Adaptive neuro fuzzy inference system

1 Introduction

Renewable energy sources are playing an important role in the global energy mix, as ameans of reducing the impact of energy production on climate change. Wind energy

made its first actual step though simpler wind devices date back thousands of years agowith the vertical axis windmills found at the Persian-Afghan borders around 200 BC andthe horizontal-axis windmills of the Netherlands and the Mediterranean following muchlater (1300-1875 AD). Further evolution and perfection of these systems was performedin the USA during the 19thcentury mainly for pumping water between 1850 and 1970.The first large wind machine to generate electricity of 12 kW was installed in Cleveland,Ohio, in 1888. Toward the end of World War I, use of 25 kW machines throughoutDenmark was widespread. Further development of wind turbines in the USA was inspired

by the aerospace developments, while subsequent efforts in Denmark, France, Germany,and the UK (between 1935 and 1970) showed that large-scale WTs could work.European developments continued after World War II and lot of development have beenmade till date(Kaldellis & Zafirakis, 2011). During the last thirty years, securities ofenergy supply and environmental issues have reheated the interest for wind energyapplications. Also the challenge of wind energy applications is the target of 1000 GW ofwind power by 2030. Further the trend in future is towards producing big and offshoreturbines. The future forecast is indicated that the demand for wind power will increasemany folds in the years to come. The rated capacity and size both will increase in time tocome.

2 Brief Background of WTsA wind turbine is a machine for converting the kinetic energy in wind into mechanicalenergy. Wind turbines are mainly classified into two general types: horizontal axis andvertical axis. A horizontal axis machine has its blades rotating on an axis parallel to theground. A vertical axis machine has its blades rotating on an axis perpendicular to theground. There are a number of available designs for both and each type has certainadvantages and disadvantages. However, compared with the horizontal axis type, veryfew vertical axis machines are available commercially. Further horizontal wind turbinesare also of two types upwind and downwind WTs. In case of upward WT the rotor is inthe front of the unit. In order to keep it oriented into the wind, a yaw mechanism is

needed. Also the extended nacelle is required to position the rotor far enough away from

308



14/26

the tower, in order to avoid any problems with a blade strike. In a downwind turbine therotor is on the back side of the turbine. The nacelle is designed to seek the wind, thusthere is no need to have a separate yaw mechanism. The flexible blade could be used butthis advantage may also be a disadvantage, as the flexing may cause fatigue to the blades.Tower shadow is problem with downwind machines because the rotor blade is behind the

tower. This may cause turbulence and lead to increased fatigue on the unit. Keeping inview the pros and cons of both the types, mostly upward WTs are used. The various

parts of horizontal wind turbine (Figure 1) and their purpose are indicated: 1: blade (twoor three blades, wind blowing over the blades causes the blades to "lift" and rotate); 2:rotor ( blades and the hub together); 3: pitch (blades are turned out of the wind to keep therotor from turning in winds that are too high or too low to produce electricity); 4: brake(disc brake which can be applied mechanically, electrically, or hydraulically); 5: low-speed shaft (turns at about 30 to 60 rpm); 6: gear box(connect the low-speed shaft to thehigh-speed shaft in order to increase speed from 30 or 60 rpm to about 1200 to 1500 rpmfor producing electricity); 7: generator (induction generator that produces 60-cycle AC

electricity). 8: controller ( starts or stops the WT at wind speeds of about 8 to 16 mph and65 mph respectively; 9: anemometer (measures the wind speed and transmits wind speeddata to the controller); 10: wind vane (measures wind direction and communicates withthe yaw drive to orient the turbine); 11: nacelle (the rotor attaches to the nacelle, whichsits atop the tower and includes the gear box, low and high-speed shafts, generator,controller, and brake, a cover protects the components inside the nacelle); 12: high-speedshaft (drives the generator); 13: yaw drive (upwind turbines face into the wind; the yawdrive is used to keep the rotor facing into the wind as the wind direction changes;downwind turbines don't require a yaw drive); 14: yaw motor (powers the yaw drive); 15:tower (made tubular steel or steel lattice, wind speed increases).

(a) (b)

Fig. 1(a) Parts of Wind Turbine http://www.wwindea.org/technology/ch01/estructura-en.htm, (b): CBMS on WT (Kusiak et al, 2010)

3 Fault Diagnosis and Prognosis

A fault is a physical defect, imperfection or flaw that occurs within the system. This may

cause a failure: the non-performance of some action that is due or expected. Fault

309



15/26

detection is determination of faults present in a system and time of detection. Faultisolation is determination of kind, location and time of detection of a fault; follows faultdetection. Fault identification is determination of size and time-variant behavior of afault; follows fault isolation and Fault diagnosis is determination of kind, size, locationand time of a fault; follows fault detection, and includes fault isolation and identification

[Donders, 2002]. Prognosis is an approach that combines information on each machinescurrent condition with historical data from machines of the same class, models of the

physics of failure of components and short term projected usage to predict the futureprobability of failure of that individual machine. That is, prognosis gives a probabilisticforecast that is specific to each machine, allowing a strategy that balances the risk ofrunning a machine with damage indications against the lost revenue while waiting formaintenance [Hyers et al, 2006]. Data Mining is a process of extraction of usefulinformation and patterns from huge data. It is also called as knowledge discovery process,knowledge mining from data, knowledge extraction or data /pattern analysis [Ramageri,2010]. Various maintenance practices are have been described and classified, empirically,

into the following approaches [Alsyouf, 2004]: reactive Approach; preventive approach;predictive approach, diagnostic (expert systems) approach ; autonomous approach ; leanapproach ; proactive approach. The ultimate goal of Fault Diagnosis and Prognosis is todecide the appropriate maintenance strategy.

3.1 Fault Diagnosis and Prognosis Systems on WTCondition monitoring can detect faults early and can prevent major breakdowns. This isassociated with significantly decrease in maintenance costs. Furthermore it allows foroptimisation of maintenance schedules, thereby reducing downtime and enhancing

equipment availability, safety and reliability [Entezami et al, 2012]. Garca et al (2012)provided comprehensive review of the CBM of wind turbines, describing the differentmaintenance strategies. In their study they have focused their attention on the sensorstechnology (acquiring and analysing data for fault diagnosis). Hameed et. al (2010) andHameed et al (2009) have also provided significant vital information in the area of CM.On a WT major faults are generated because of the main gearbox, generator, main

bearings, rotor blades and their possibility of failures in terms of percentage are 32, 23, 11and < 10% respectively as stated by the insurer German Lloyd. The CBMS is a tool that

provides information on the status of a component and can also predict an expectedfailure/fault. The possible applications of CBMS on WT are shown in the Figure 1(b).

The fault diagnoses sensors used and techniques employed for CBMS on various parts ofthe WT are summarized in Table 1. [Lu, 2009, Hameed et al, 2009, Hameed et al, 2010,Hyers et al, 2006, Fischer et al, 2012, Liu et al, 2012]. Digital filtering, modelling, signaland spectrum analysis are the main parts of the data processing in CBMS. Data analysisin wind turbines could be based on many of the techniques, few important aresummarised in the Table 2. The next step is to predict the remaining life of thecomponents. So as to adopt a suitable maintenance strategy for WTs [Nadakatti et al,2008].

310



16/26

Table 1Summary of CBMS on WT

Part of WT Fault Technique Sensor/ Monitoring quantity

Gearbox Gear toothdamages, Bearingfaults

Vibration monitoring & spectrumanalysis, supplemented with

AE sensing detects pitting, cracking

Transducers, velocity sensors,accelerometers, spectral emittedenergy sensors, AE sensors

Oil analysis Temperature,contamination,moisture

Generator Stator, bearing ,rotor inside

Current signature analysis Current measurement

RotorBlades

Creep & corrosion,fatigue, imbalance,roughness

Shearography, Radiography, AEsensing

AE sensors, strain gauges, Fiber bragggrating

Tower andblades

Ultrasonic testingtechniques

time-frequency techniques andwavelet transforms

Ultrasonic sensors, Fiber bragg grating

PitchmechanismYaw systemPowerElectronics/Electricalsystem

Voltage andcurrentanalysis, electricalresistance

Spectrum analysis, eddy current,thermography

Current and voltage measuringequipments

Table 2Summary of sensory signals and signal processing methods [Garca et al, 2012]

Methods Description Applications

Statistical methods Root mean square, peak amplitude Widely used thediagnosis of failuresMaximum value, minimum value, mean, peak to peak, standard

deviation, shape factor, crest factor, impulse factor, definiteintegral, energy ratio and kurtosis.

Trend analysis Particular algorithms power output patterns from WT generators Monitoring pitchmechanisms, poweroutput generators

Filtering methods Least median squares, filtering with and without a classical

statistical method (based on standard deviation)

Vibration filtering

Time-domain analysis Variations in current signals and trends, vibration analysis, oilanalysis, AE analysis

Imbalances betweenthe rotor and the statorphases, faults in therotor windings of thegenerator.

Cepstrum analysis Inverse Fourier transform of the logarithmic power spectrum Gear vibrations

Time synchronousaveraging (TSA)

Waveform signal Cracked gear toothIdentifies source oferror

Fast-Fouriertransform (FFT)

Conversion of a digital signal from the time domain into thefrequency domain, constant speed wind energy converters

Bearings

Amplitude

demodulation

Low-amplitude and low-frequency, periodic signals that might

be masked by other higher energy vibrations

Bearings

Order analysis Out of phase signals can be separated and analysed, suitable forvariable speed wind energy converters.

Overall rotor conditionincluding surfaceroughness, massimbalance andaerodynamicasymmetry.

Wavelet transforms Time-frequency technique ,non-stationary signals Gears, bearings,mechanical systems

Hidden Markovmodels

Classification of patterns in trend analysis Bearings

CI Techniques Computational Intelligent algorithms (Lei, Lin, He, & Kong,2012),(Amjady & Hedayatshodeh, 2012),(Uraikul, Chan, &Tontiwachwuthikul, 2007)

All Systems

311



17/26

Some existing CBMS are: Gram & Juhl TCM System; DMT WindSafe System;Siemens Monitoring & Safety System; PRFTECHNIK [Wiggelinkhuizen et al, 2008]

4 SCADA Data Based CBM for WT

Supervisory Control and Data Acquisition (SCADA) is an application that collects datafrom a system and send them to a central computer for monitoring and control usages.Current CBM systems essentially provide the necessary sensor and data capturecapability required for monitoring. This monitoring system consists of wireless sensorsthat communicate with an embedded microprocessor mounted on the devices. Its purposeis to allow diagnostic and fault detection algorithms to be deployed down at the sensor /hardware level. This will decrease the volume of data that must be transmitted and storedvia the more traditional centralised SCADA system approach. The SCADA data resideson a server / PC in the form of a database, and the raw data resides on the hard-drive ofthe microprocessor / sensor system mounted on the turbine (Refer Figure 2). A typical

SCADA data is 10 minute averaged data [Zaher and McArthur, 2007]. Thus the collectedand stored SCADA data must then be examined in order to deduce the overall health ofthe turbine as well as its internal components [Uraikul et al, 2007]. An operational windfarm typically generates vast quantities of data. The SCADA data contain informationabout every aspect of a wind farm, from power output and wind speed to any errorsregistered within the system[Kim et al, 2011]. SCADA data may be effectively used totune a wind farm, providing early warning of possible failures and optimising poweroutput across many turbines in all conditions. Typical parameters recorded by SCADA ona WT could be broadly categorized into following types which could be used in faultdiagnosis and prognosis activity: Wind parameters, such as wind speed and wind

deviations; Performance parameters, such as power output, rotor speed, and blade pitchangle; Vibration parameters, such as tower acceleration and drive train acceleration;Temperature parameters, such as bearing temperature and gearbox temperature[Vermaand Kusiak, 2012].

There are many success stories about using SCADA data for CBM. Initially Monostoriet al (2009) discussed a framework for optimizing wind energy systems design, operationand maintenance. They emphasized, as wind turbines have several built-in sensorsmeasuring various physical characteristics during the operation and SCADA systemsserve with huge amount of data. There after Wiggelinkhuizen et al (2008) presented theframework of the EU funded Condition Monitoring for Offshore Wind Farms project.

They studied a small wind farm of five turbines having been instrumented with severalcondition monitoring systems and also with the traditional measurement systems formeasuring mechanical loads and power performance. Data from vibration and traditionalmeasurements, together with data collected by the turbines system control and dataacquisition (SCADA) systems, have been analyzed to assess. They could determinefailures, detect early stage of failure and asses the components health. They developed,applied, and tested several data analysis methods and measurement configurationssuccessfully and conclude that for all types of measurements SCADA data, time series,vibration monitoring could be used for CBM. Zaher and McArthur, 2007 also proposedan idea to use the combination of anomaly detection and data-trending techniques

312



18/26

encapsulated in a multi-agent framework for the development of a fault detection systemfor WTs.

Fig. 2The SCADA data based CBM system

Sensor 1 Micoprocessor

mounted on device

Sensor 2 MicoprocessorCentralised SCADA

System

SCADA DATA

RAW DATA

SCADA master

10 mins average data

Fault diagnosis

Fault prognosis

Maintenance

DATA MINING HMI

Corroborate

PAST/PRESENT DATA

PRESENT DATA

Operator Workstation

WANRTU

remote location

and/or

PLC

StrategyAlgorithmsand/or

modelsSensor 1 Micoprocessor

mounted on device

Sensor 2 Micoprocessor

ADD ON DATA(AOD)

pre-processing

SCADADATA

AODCM

Add on data

condition monitoring

AODCM

DATA

SCADA data

AOD dataHigh frequency data

ADD ON DATA(AOD)

313



19/26

Table 4: Summary of papers using SCADA data for CBMS

Faults/outputvariables

Parametersection/algorithms/Classification

Modellingalgorithms/method

Input/output data type Major findings Reference

Power

prediction

PCA MLP,REP

tree , M5Ptree , BT,k-NN

Inputs: wind, speed,

wind direction, outsidetemperature; Outputs:turbine control

k-NN modelling

best amongstudied

[Kusiak, et

al, 2009]

Future statesof the powerdrive

Time series modelfused with thedynamicequations

Data miningalgorithmswith theevolutionarystrategy

Input:wind speed,blade pitch angle,generator torque;Outputs:rotor speedand turbine power

Optimal control toimprove the poweroutput withoutlarge rotor speedchanges.

[Kusiak et al,2009]

Vibrations of awind turbine,performanceevaluation

NN, SVM,CART BT,RF

Input :Torque value,blade pitch angle rate,wind deviation, bladepitch angle, windspeed; Outputs:drivetrain acceleration andtower acceleration

NN algorithm bestamong studied

[ Kusiak andZhang,2010]

Power output CART,BT,SVM,

NN

Radial-basis-

function MLP

Input:current wind

speed, generatorspeed, wind direction,wind directiondifference; Output:blade pitch/ yaw angle

Control approach

generatedoptimized settingsof the blade pitchand yaw angle

[ Kusiak, et

al, 2010]

Wind turbinefailure

PSO, DE,ES

Wind speed differencetests

Detected normaland abnormalstates, PSO bestamong the studied

[Ye et al,2010]

Parametersoptimizing theenergyconversion

Extracts theprocess and thetime seriesmodels fromhistorical SCADAand wind data

NNE,geneticwrapperapproach

Input:Wind conditions,electricity demand;Output:blade, pitchangle, generatortorque.

Intelligent controlsystem smoothedthe power output,generator torque,and rotor speed

[ Kusiak etal, 2010]

WTperformancemonitoring &prognostics

WTperformanceanalyticpower curveanalysis

Power residual Performanceanalytic powercurve analysisseparates out pre-failure data fromother normaloperating data

[Uluyol et al,2011]

Predictturbine faults

NN,NNE,BT,SVM

Wind, performanceVibration andtemperatureparameters

Power curvedetermines thehealth of WT,status/ fault,category

[ Kusiak andLi, 2011]

Identificationand predictionof status

patterns inWT

PCA BT, RF,ripper,k-NN

Power, wind speed,rotor speed, andgenerator speed

A data-miningapproach applied,Random forest

algorithm bestamong studied

[Kusiak andVerma,2011]

vibrations ofWT

modified k-meansclustering

NN,NNE,BT,SVM,randomforest,CART,kNN,NN

SCADA vibration datasampled at 10 sintervals

NNE suitable drivetrain accelerationand the NN modelis most suitable forpredicting toweracceleration.

[Zhang andKusiak,2012]

Carbonbrushes ofgenerator

Chi-squarestatistic, boostingtree algorithm,wrapper(geneticsearch)

Multilayeredperceptron,boostingtree,k-NNSupportvectormachine

Wind , performance,vibration andtemperatureparameters

Drive trainacceleration,generator speed,torque andaerodynamicasymmetry.Boosting tree

algorithm bestamong the studied

[Verma andKusiak,2012]

314



20/26

5 Data-driven Methods for Analysis of SCADA Data from WTs

Approaches to analyse conditions of WT can be categorized into model-based and datadriven ones. Model-based approaches attempts to incorporate physical models of systeminto estimation of the condition of WT. However, uncertainty due to the assumptions and

simplifications of adopted model may pose significant limitations on this approach.Especially for SCADA data analysis we should use data-driven approaches. Data-driventechniques utilize monitored operational data related to system health. They canbe beneficial when understanding of first principles of system operation is notstraightforward or when the system is so complex that developing an accuratemodel is prohibitively expensive. Data-driven approaches can often be deployedquickly and cheaply, and still provide wide coverage of system behavior. Anadded value of data-driven techniques is their ability to transform highdimensional noisy data into lower dimensional information useful for decision-

making [Dragomir et al., 2007].Pre-processing of SCADA is a must of extraction of useful information and patternsfrom huge data. The various data-driven methods being used for analysis of SCADA datafrom WTs can be divided into two categories: (1) statistical techniques (regressionmethods, ARMA models, etc.) and (2) Computational Intelligence (CI) techniques(Artificial Neural Networks, Fuzzy systems, etc.).We will focus on ComputationalIntelligence techniques in this paper.

5.1 Artificial Neural NetworksANNs can be used for a wide range of applications. They are inspired by the mechanismof the brain and can be classified by different categories as depending upon the learningmechanism or how they are trained (supervised \ unsupervised). ANNs accept input

parameters, process them, and produce output parameters according to a nonlineartransfer function. Some of the key features for NN are their high processing speeds whichare due to their massive parallelism, their proven ability to be trained, to produceinstantaneous and correct responses from noisy or partially incomplete data, and theirability to generalize information over a wide range. These features make them a goodchoice for applying to WTs data analysis. Hofemann et al explored the possibilities toapply ANN not only within a wind park but on turbines located at different sites.

Following the idea to develop a tool to forecast the particular loads of any wind turbine inthe field without the need to install additional measuring systems, a model was developedthat needed only signals contained in the SCADA data as input signals. They trained thenetwork to predict the blade root moments of two wind turbines located at different sites.A supervised ANN was chosen, the load data was the target data of the network and wasused to train the network as well as to verify the made predictions. Parameters as windspeed, wind direction, power output, pitch angle and yaw moment were used as inputdata.

5.2 Fuzzy Systems

315



21/26

An opportunity for increased transparency and openness of data-driven model is offeredby fuzzy logic methods, which are increasingly proposed in modern diagnosticand prognostic technologies. Based on the principles of Zadehs fuzzy set theory,fuzzy logic provides a formal mathematical framework for dealing with thevagueness of everyday reasoning [Zadeh, 1965]. As opposed to binary reasoningbased on ordinary set theory, within the fuzzy logic framework measurementuncertainty and estimation imprecision can be properly accommodated.

Fuzzy systems are very useful in two general contexts like in situations involvinghighly complex systems whose behaviors are not well understood and in situations wherean approximate, but fast, solution is desired. A further advantage of fuzzy systems is thatexisting expert knowledge can be implemented to improve the approximation by tuning,removing or adding of membership functions and rules.

5.3 Adaptive Neuro-fuzzy Inference SystemHybrid CI model are able to integrate Artificial Neural Network models and Fuzzy Logicsystem models to make a better performance. Fuzzy neural networks have shown to bevery advantageous in dealing with real-world problems. These neuro-fuzzy systemscombine the benefits of these two powerful paradigms into a single capsule. This givesthe ability to accommodate both data and existing expert knowledge about the problemunder consideration. In ANFIS the advantages of NN are combined with FIS. Thereby theFIS is used to set up a set of rules whose membership functions parameters are tuned in atraining phase. Recently Schlechtingen et al (2013), proposed a system for wind turbinecondition monitoring using ANFIS. For this purpose: (1) ANFIS normal behavior modelsfor common SCADA data were developed in order to detect abnormal behavior of thecaptured signals and indicate component malfunctions or faults using the prediction error.They used 33 different standard SCADA signals and described, for which 45 normal

behavior models were developed. The performance of these models was evaluated interms of the prediction error standard deviations to show the applicability of ANFISmodels for monitoring wind turbine SCADA signals. The computational time needed formodel training was compared to NN models showing the strength of ANFIS in trainingspeed. For automation of fault diagnosis FIS were used to analyze the prediction errorsfor fault patterns. The outputs were both the condition of the component and a possibleroot cause for the anomaly. The output was generated by the aid of rules that capture the

existing expert knowledge linking observed prediction error patterns to specific faults.The work was based on continuously measured wind turbine SCADA data from 18turbines of the 2 MW class covering a period of 30 months. The system proposed in thismethod shows a novelty approach with regard to the usage of ANFIS models in thiscontext and the application of the proposed procedure to a wide range of SCADA signals.

5.4 Data Mining ApproachesA considerable work has been carried out in using SCADA data on WT diagnosis and

prognosis by Kusiak and his group. The work carried out by various researchers has beenpresented in Table 4. Kusiak et al (2009) presented various models for monitoring windfarm power output. An evolutionary computation algorithm was used to build a nonlinear

316



22/26

parametric model to monitor the wind farm performance. The k-NN model producedgood performance for the wind farm operating in normal conditions. In their later studythey [Kusiak et al., 2009] proposed the concept of anticipatory control applied to windturbines. A modified evolutionary strategy algorithm was used to solve a nonlinearconstrained optimization problem. The proposed approach has been tested on the data

collected from a 1.5 MW wind turbine. In a subsequent study Kusiak et al (2010),proposed a framework for optimization of the power produced by wind turbines. Theyproposed control approach generated optimized settings of the blade pitch and yaw angle.They used integration of data mining and evolutionary computation in their approach. Ina later study on SCADA data Kusiak and Zhang (2010) found that vibrations of a windturbine have a negative impact on its performance. Data-mining algorithms were used to

build models with turbine parameters of interest as inputs, and the vibrations of drivetrain and tower as outputs. The performance of each model was thoroughly evaluated

based on metrics widely used in the wind industry. The neural network algorithmoutperforms other classifiers and is considered to be the most promising approach to

study wind turbine vibrations. Ye et al (2010), proposed that with design of wind speeddifference tests to detect both hard failures (cause complete shutdown of the turbine e.g.gear failure) and soft failures (degrade the turbine performance but do not necessarilystop the turbine e.g. anemometer faults). They used PSO based approach to learn fromhistorical data to decide the location and size of the boundary i.e. abnormal state from thenormal state. Further Kusiak et al (2010) presented an intelligent wind turbine controlsystem based on models integrating the following three approaches: data mining, model

predictive control, and evolutionary computation. The results produced by the intelligentcontrol system were found to be better than those of the current wind turbine controlsystem. For turbulent wind, the intelligent control system smoothed the power output,generator torque, and rotor speed without compromising the electricity demand. Uluyol etal (2011) proved that the wind turbine power curve analytic is useful for assessing windturbine performance and generating robust indicators for component diagnostics and

prognostics. This approach makes use of SCADA information and provides easyconfiguration based on process control approaches for condition-based monitoring. Theyused operational regime-based condition indicators that prevented false alarms andincreased the possibility of fault isolation. This approach could also detect slow

performance degradation caused by component wear as well as degradation due to animpending failure. Kusiak and Li (2011) used fault data provided by SCADA system.

They carried out fault prediction at three levels: (1) fault and no-fault prediction; (2) faultcategory (severity); and (3) the specific fault prediction. They used power curve todetermine the health of a wind turbine. The model extraction at level 1 and 3 was

performed using the Neural Network (NN), Neural Network Ensemble (NN Ensemble),the Boosting Tree Algorithm (BTA), and the Support Vector Machine (SVM). Whereasat level 2 they used Neural Network (NN), the Standard Classification and RegressionTree (CART), the Boosting Tree Algorithm (BTA), and the Support Vector Machine(SVM). Furthermore they could successfully predict faults 5-60 min before they occur ateach level. Kusiak and Verma [2011] used an association rule mining algorithm toidentify frequent status patterns of turbine components and systems to predicted using

historical wind turbine data. They explored five data-mining algorithms namely bagging,

317



23/26

ripper, rotation forest, random forest, and k-nearest neighbor (k-NN, k=10). Theyconcluded that the best predication results were obtained with random forest algorithm.Verma and Kusiak [2012], proposed a methodology to predict the generator brush wornfault in wind turbines. They used SCADA systems information from 27 wind turbines.They used algorithms (Chi-square statistic, Boosting tree algorithm, Wrapper (genetic

search) for parameter selection, by which the ten most important parameters wereidentified. They built the model using four well-known data-mining algorithms, namely,multilayered perceptron (MLP), boosting tree, k-NN (k10), and support vector machine(SVM). These algorithms were studied to evaluate the quality of the models for

predicting generator brush faults and it was found that the boosting tree algorithmprovided the best prediction results. Zhang and Kusiak [2012], proposed three monitoringmodels for detecting abnormal vibration of wind turbines in time domain based onSCADA data. The sampling interval of SCADA data was 10 s, it allows to detectabnormal statuses of wind turbines in the time domain. The data collected includedtraining, test, and error data. They used a modified k-means clustering algorithm to first

vibration monitoring model. The k-means algorithm grouped data into clusters byexamining their similarity. The clusters were then labelled as normal or abnormal statusesof wind turbine vibration based on the error reports of wind turbines. They furtherincorporated the concept of control charts to develop models for monitoring of turbinevibration. They also addressed the detecting of abnormal drive train and tower vibrationof a wind turbine. They compared seven different data-mining algorithms, namely,neural network ensemble (NNE) , neural network (NN) , boosting regression tree (BT) ,support vector machine (SVM) , random forest with regression (RF) , standardclassification and regression tree (CART) , and k nearest neighbor neural network (kNN). Four metrics, the mean absolute error (MAE), standard deviation of absolute error (SDof AE), mean square error (MSE), and the standard deviation of square error (SD of SE),are utilized to evaluate the performance of data-mining algorithms in model extraction.They found that the NNE model is recognized as the most suitable for determining drivetrain acceleration and the NN model is considered to be the most suitable algorithm fordeveloping the model for predicting tower acceleration. Yang et al [2013] in their work

proposed a technique which would interpret the SCADA data collected from windturbines. They developed an effective method for processing raw SCADA data, further

propose an alternative condition monitoring technique based on investigating thecorrelations among relevant SCADA data; and realised the quantitative assessment of the

health condition of a turbine under varying operational conditions. Both laboratory andsite verification tests have been conducted and were found satisfactory.

6 Discussions and Future Challenges

From the literature reviewed it has been demonstrated successfully by researchers that bykeeping track of wind speed and power output parameters, the overall health of theturbine can be supervised. Furthermore SCADA data could be usefully used for CBM ofWTs and locating faults. There has been success in using SCADA data for power

prediction, optimal control settings, performance evaluation, predict turbine faults(predicting drive train acceleration/tower acceleration/gear box failure) and also the

vibrations on a WT. Many AI techniques have been applied including NN, Fuzzy,

318



24/26

ANFIS, GAs etc [Yang et al,2008, Garcia et al, 2006]. Use of AI model in CBM of windturbines is justifiable as there are many variables involved and it not easy to establishaccurate mathematical model for such kind of complicated systems. It is also proposed touse data fusion techniques for monitoring the health of WTs [Dempsey and Sheng,2013]. Few challenges that need to overcome before SCADA data analysis becomes fully

successful are: SCADA data can differ from turbine to turbine and SCADA data changeswith the operational conditions. So because of this it becomes difficult to differentiate

between a real fault and fake fault which is quite a challenge.More over WT SCADA data are usually 10 min average data, so some of the

information is lost. Thus referring to the research of wind turbine condition monitoring, aframe work is proposed that takes SCADA data and also an add on high frequency datafrom the sensors (some) to diagnose and prognose with both conventional data andSCADA data. After comparison of the two results, the appropriate method can be chosenfor maintenance decision making.

A number of models have been proposed tried and tested[Wilkinson et al, 2013,

Schlechtingen & Ferreira Santos, 2011], however the accuracy of the model dependsupon the careful selection of variables and the quality of the data (free from noise). So

pre-processing of the data plays an important role in the accuracy of the models.Performance of AI based data mining algorithms and CBM algorithms are showing quite

promising results. Hence by using computational intelligence concepts more efficientmodels could be obtained thus enhancing the accuracy and robustness of the model. It is

proposed that Data mining (AI based) and evolutionary computations could be integratedfor building the models for prediction and monitoring.

7 Conclusions

The major outcomes from this review are listed below:

A frame work is proposed for diagnosis and prognosis of CBM of WTs using acombinational approach. These make use of conventional SCADA data and add onhigh frequency data from the sensors. In addition to this it also analyses historicalSCADA data.

Few challenges that need to overcome in SCADA data analysis are: firstly SCADAcan differ from turbine to turbine, secondly SCADA data changes with the operationalconditions. So pre-processing the raw data can equally contribute to the success of the

CBM algorithm itself. Some researchers have successfully demonstrated the use of AI algorithms for SCADA

data analysis. Seeing their performance it is believed that AI models could enhance theaccuracy and robustness of models.

Though the comparison results of various models are mentioned but no one clear cutperfect modelling technique could emerge. So this area remains open for furtherexplorations.

References

319



25/26

Alsyouf, I. (2004). Cost Effective Maintenance for Competitive Advantages. Thesis for the degreeof Doctor of Philosophy (Terotechnology) School of Industrial Engg., Vxj University,

Sweden. Vxj University Press, ISSN: 1404-4307.Amjady, N. and Hedayatshodeh, M. (2012). A New Power Transformer Fault Diagnosis System

and its Application for Wind Farms. Journal of Basic and Applied Scientific Research, Vol 2,

no. 5, pp 47584764.Dempsey, P. J. and Sheng, S. (2013). Investigation of data fusion applied to health monitoring ofwind turbine drivetrain components. Wind Energy, pp 479489.

Donders, S. (2002). Fault Detection and Identification for Wind Turbine Systems: a closed-loopanalysis.Masters Thesis,University of Twente Faculty of Applied Physics Systems and Control

Engineering.Dragomir, O.E., Gouriveau, R., Zerhouni, N., Dragomir, F. (2007). Framework for a Distributed

and Hybrid Prognostic System, In 4th IFAC Conference on Management and Control ofProduction and Logistics.

Entezami, M., Hillmansen, S., Weston, P. and Papaelias, M. P. (2012). Fault detection anddiagnosis within a wind turbine mechanical braking system using condition monitoring.

Renewable Energy, Vol. 47, pp. 175182.Fischer, K., Besnard, F. and Bertling, L. (2012). Reliability-Centered Maintenance for WindTurbines Based on Statistical Analysis and Practical Experience.IEEE Transactions on EnergyConversion, Vol. 27, no.1, pp.184195.

Garcia, M. C., Sanz-Bobi, M. A. and Del Pico, J. (2006). SIMAP: Intelligent System for PredictiveMaintenance. Computers in Industry, Vol. 57, no. 6,pp. 552568.

Garca Mrquez, F. P., Tobias, A. M., Pinar Prez, J. M. and Papaelias, M. (2012). Conditionmonitoring of wind turbines: Techniques and methods. Renewable Energy, Vol. 46, pp. 169178.

Hameed, Z., Ahn, S. H. and Cho, Y. M. (2010). Practical aspects of a condition monitoring systemfor a wind turbine with emphasis on its design , system architecture , testing and installation.

Renewable Energy, Vol. 35, no. 5, pp. 879894.Hameed, Z., Hong, Y. S., Cho, Y. M., Ahn, S. H. and Song, C. K. (2009). Condition monitoring

and fault detection of wind turbines and related algorithms: A review. Renewable andSustainable Energy Reviews, Vol. 13, no. 1,pp. 139.

Hyers, R. W., Mcgowan, J. G., Sullivan, K. L., Manwell, J. F. and Syrett, B. C. (2006). Conditionmonitoring and prognosis of utility scale wind turbines. Energy Materials: Materials Scienceand Engineering for Energy Systems, Vol. 1, no. 3, pp. 187203.

Kaldellis, J. K. and Zafirakis, D. (2011). The wind energy (r)evolution: A short review of a longhistory.Renewable Energy, Vol. 36, no. 7,pp. 18871901.

Kim, K., Parthasarathy, G., Uluyol, O., Foslien, W., Shuangwen, S. and Fleming, P. (2011). Use ofSCADA Data for Failure Detection in Wind Turbines. Conference Paper NREL/CP-5000-

51653.Kusiak, A. and Li, W. (2011). The prediction and diagnosis of wind turbine faults. Renewable

Energy, Vol. 36, no. 1, pp. 1623.Kusiak, A., Li, W. and Song, Z. (2010). Dynamic control of wind turbines. Renewable

Energy,Vol. 35, no. 2, pp. 456463.Kusiak, A., Song, Z. and Zheng, H. (2009). Anticipatory Control of Wind Turbines With Data-

Driven Predictive Models.IEEE Transactions On Energy Conversion, Vol. 24, no. 3, pp. 766774.

Kusiak, A. and Verma, A. (2011). Prediction of Status Patterns of Wind Turbines: A Data-MiningApproach.Journal of Solar Energy Engineering, Vol. 133, no. 1, pp. 011008101100810.

Kusiak, A. and Zhang, Z. (2010). Analysis of Wind Turbine Vibrations Based on SCADA Data.Journal of Solar Energy Engineering,Vol. 132, no. 3, pp. 031008103100812.

320



26/26

Kusiak, A., Zheng, H. and Song, Z. (2009). Models for monitoring wind farm power. RenewableEnergy, Vol. 34, no. 3, pp. 583590.

Kusiak, A., Zheng, H. and Song, Z. (2010). Power optimization of wind turbines with data miningand evolutionary computation.Renewable Energy, Vol. 35, no. 3, pp. 695702.

Kusiak, B. A. and Verma, A. (2010). The Future of Wind Turbine diagnostics.

windsystemsmag.com,pp. 6671.Lei, Y., Lin, J., He, Z. and Kong, D. (2012). A method based on multi-sensor data fusion for faultdetection of planetary gearboxes. Sensors,Vol. 12, no. 2, pp 200517.

Liu, W. Y., Zhang, W. H., Han, J. G. and Wang, G. F. (2012). A new wind turbine fault diagnosismethod based on the local mean decomposition.Renewable Energy, Vol. 48,pp. 411415.

Lu, B. (2009). A Review of Recent Advances in Wind Turbine Condition Monitoring and FaultDiagnosis. Power Electronics and Machines in Wind Applications, IEEE,pp. 1 7.

Nadakatti, M., Ramachandra, A. and Kumar, A. N. S. (2008). Artificial intelligence-basedcondition monitoring for plant maintenance.Assembly Automation, Vol. 28, no. 2, pp. 143150.

Ramageri, B. M. (2010). Data mining techniques and applications. Indian Journal of ComputerScience and Engineering,Vol. 1, no. 4, pp. 301305.

Schlechtingen, M. and Ferreira Santos, I. (2011). Comparative analysis of neural network andregression based condition monitoring approaches for wind turbine fault detection. MechanicalSystems and Signal Processing,Vol. 25, no. 5, pp. 18491875.

Uluyol, O., Parthasarathy, G., Foslien, W. and Kim, K. (2011). Power Curve Analytic for WindTurbine Performance Monitoring and Prognostics. Annual Conference of the Prognostics and

Health Management Society,pp. 18.Uraikul, V., Chan, C. W., and Tontiwachwuthikul, P. (2007). Artificial intelligence for monitoring

and supervisory control of process systems. Engineering Applications of ArtificialIntelligence,Vol. 20, no.2,pp. 115131.

Verma, A. and Kusiak, A. (2012). Fault Monitoring of Wind Turbine Generator Brushes: A Data-Mining Approach. Journal of Solar Energy Engineering, Vol. 134, no. 2, pp. 0210011

0210019.Wiggelinkhuizen, E., Verbruggen, T., Braam, H., Rademakers, L., Xiang, J. and Watson, S.

(2008). Assessment of Condition Monitoring Techniques for Offshore Wind Farms. Journal ofSolar Energy Engineering, Vol. 130, no. 3, pp. 03100410310049.

Wilkinson, M., Darnell, B. and Harman, K. (2013). Annual Comparison of Methods for WindTurbine Condition Monitoring with SCADA Data.EWEA 2013 Vienna, 4-7 February 2013,pp.47.

Yang, S., Li, W. and Wang, C. (2008). The intelligent fault diagnosis of wind turbine gearboxbased on artificial neural network. International Conference on Condition Monitoring andDiagnosis, pp. 13271330.

Ye, X., Yan, Y. and Osadciw, L. A. (2010). Learning Decision Rules by Particle Swarm

Optimization ( PSO ) for Wind Turbine Fault Diagnosis. Annual Conference of the Prognosticsand Health Management Society.

Zadeh, L.(1965). Fuzzy sets,Inform. Control, Vol. 8, pp. 338353.Zaher, A. S. and McArthur, S. D. J. (2007). A Multi-Agent Fault Detection System for Wind

Turbine Defect Recognition and Diagnosis.IEEE Lausanne Power Tech, pp. 2227.Zhang, Z. and Kusiak, A. (2012). Monitoring Wind Turbine Vibration Based on SCADA Data.

Journal of Solar Energy Engineering, Vol. 134, no. 2, pp. 021004-1-021004-12.


scada data interpretation for condition-based monitoring of wind turbines

Documents