ScienceDirect ScienceDirect

9th IFAC Symposium on Fault Detection, Supervision and 9th IFAC Symposium on Fault Detection, Supervision and Safety of Symposium Technical Processes 9th IFAC IFAC on Fault Fault Detection, Detection, Supervision Supervision and and Safety of Symposium Technical Processes 9th on Available online at www.sciencedirect.com 9th IFAC Symposium on Fault Detection, Supervision September 2-4, 2015. Arts et Métiers ParisTech, Paris, and France Safety of Technical Processes September 2-4, 2015. Arts et Métiers ParisTech, Paris, France Safety of Technical Processes Safety of Technical Processes September 2-4, 2015. Arts et Métiers ParisTech, Paris, France September September 2-4, 2-4, 2015. 2015. Arts Arts et et Métiers Métiers ParisTech, ParisTech, Paris, Paris, France France

ScienceDirect

IFAC-PapersOnLine 48-21 (2015) 1369–1374

Active Active Fault Fault Tolerant Tolerant Control Control in in aa Wind Wind Farm Farm with with Decreased Decreased Power Power Generation Generation Active Fault Tolerant Control in a Wind Farm with Decreased Power Active Fault Tolerant Control in a Wind Farm with Decreased Power Generation Due to Blade Erosion/Debris Build-Up Active Fault TolerantDue Control in a Wind Farm withBuild-Up Decreased Power Generation Generation to Blade Erosion/Debris Due to Blade Erosion/Debris Build-Up Due to Blade Erosion/Debris Build-Up Due to Blade Erosion/Debris Build-Up Hamed Badihi*, Youmin Zhang**, and Henry Hong***

Hamed Badihi*, Youmin Zhang**, and Henry Hong*** Zhang**, and Henry Hong*** Hamed Badihi*, Youmin Zhang**, Hamed Badihi*, Youmin and Henry Hamed Badihi*, Youmin Concordia Henry Hong*** Hong*** Department of Mechanical and Industrial Engineering, University, Zhang**, and Department of Mechanical and Industrial Engineering, Concordia University, Montreal, Montreal, Quebec, Quebec, H3G H3G 1M8, 1M8, Canada Canada  Department of Mechanical and Industrial Engineering, Concordia University, Montreal, Quebec, H3G 1M8, Canada Department of Mechanical and Industrial Engineering, Concordia University, Montreal, Quebec, H3G 1M8, Canada ** Department of Information and Control Xi'an University of Technology, Xi'an, Shaanxi, 710048, China Department of Mechanical and Industrial Engineering, Concordia University, Montreal, Quebec, H3G 1M8, Canada ** Department of Information and Control Engineering, Xi'an University of Technology, Xi'an, Shaanxi, 710048, China ** Department of Information and Control Engineering, Xi'an University of Technology, Xi'an, Shaanxi, 710048, China ** Department of Information and Control Engineering, Xi'an University of Technology, Xi'an, Shaanxi, 710048, (e-mail: * [email protected], ** [email protected], *** [email protected]) ** Department Information and Control Engineering, Xi'an University of Technology, Xi'an, Shaanxi, 710048, China China (e-mail: of * [email protected], ** [email protected], *** [email protected]) (e-mail: * [email protected], ** [email protected], *** [email protected]) (e-mail: * [email protected], ** [email protected], *** [email protected]) (e-mail: * [email protected], ** [email protected], *** [email protected]) Abstract: Given the the importance importance of of reliability reliability issue in in wind wind farms, farms, the the current current paper presents presents the the design design Abstract: Given issue paper and development of a novel active fault-tolerant control scheme for an offshore wind farm against Abstract: Given the importance of reliability issue in wind farms, the current paper presents the design Abstract: Given the the importance of reliability reliability issue control in wind wind scheme farms, the the current paper presents presents the against design and development of importance a novel active fault-tolerant forcurrent an offshore wind farm Abstract: Given of in farms, paper the design decreased power generation caused byfault-tolerant turbine issue bladecontrol erosion scheme and debris build-up on the thewind blades over time. and development of aa novel active for an offshore farm against and development of novel active fault-tolerant control scheme for an offshore wind farm against decreased power generation caused by turbine blade erosion and debris build-up on blades over time. and development of a novel active fault-tolerant control scheme for an offshore wind farm against The proposed scheme employs a model-based fault detection and diagnosis approach to provide accurate decreased power generation caused by blade erosion and build-up on the time. decreased power generation caused by turbine turbine fault blade erosion debris build-up on blades over time. The proposed scheme employs a model-based detection anddebris diagnosis approach toblades provideover accurate decreased power generation caused blade erosion and and debris build-up on the the blades over time. and timely diagnosis information to by be turbine used in in fault an appropriate appropriate automatic signal correction algorithm. The The proposed scheme employs aa model-based detection and diagnosis approach to provide accurate The proposed scheme employs model-based fault detection and diagnosis approach to provide accurate and timely diagnosis information to be used an automatic signal correction algorithm. The The proposed scheme employs a model-based fault detection and diagnosis approach to provide accurate effectiveness of the proposed scheme is evaluated by simulations on an advanced offshore wind farm and timely diagnosis information to be used in an appropriate automatic signal correction algorithm. The and timely information to be is used in automatic correction algorithm. The effectiveness of the proposed scheme evaluated by simulations on ansignal advanced offshore wind farm and timely diagnosis diagnosis information to used in an an appropriate appropriate automatic signal algorithm. The benchmark model in proposed the presence of be wind turbulences, measurement noises and correction load variations. variations. effectiveness of the scheme is evaluated by simulations on an advanced offshore wind farm effectiveness of the proposed scheme is evaluated by simulations on an advanced offshore wind farm benchmark model in the presence of wind turbulences, measurement noises and load effectiveness of the proposed scheme is evaluated by simulations on an advanced offshore wind farm benchmark model in the presence of wind turbulences, measurement noises and load variations. Keywords: Fault Diagnosis, Fault-Tolerant Control, Control) Wind Turbine, Farm benchmark model in of turbulences, measurement noises and load variations. © 2015, IFAC (International Federation of Automatic HostingWind by Elsevier rights reserved. benchmark model in the the presence presence of wind wind turbulences, measurement noises and Ltd. load All variations. Keywords: Fault Diagnosis, Fault-Tolerant Control, Wind Turbine, Wind Farm Keywords: Fault Diagnosis, Fault-Tolerant Control, Wind Turbine, Wind Farm Keywords: Fault Fault Diagnosis, Diagnosis, Fault-Tolerant Fault-Tolerant Control, Control, Wind Turbine, Turbine, Wind Wind Farm Farm  Wind Keywords:  employs a model-based  1. employs a model-based FDD FDD approach approach to to provide provide accurate accurate 1. INTRODUCTION INTRODUCTION  and timely diagnosis information to be used in an appropriate employs a model-based FDD approach to provide accurate 1. INTRODUCTION employs a model-based FDD approach to provide accurate and timely diagnosis information to be used in an appropriate Wind remarkable potential 1. INTRODUCTION a signal model-based FDD approach to provide accurate INTRODUCTION Wind energy energy has has 1. remarkable potential for for fulfilling fulfilling the the employs automatic correction (ASC) algorithm. However, due and timely diagnosis information to be used in an appropriate and timely diagnosis information to be used in an appropriate automatic signal correction (ASC) algorithm. However, due increasing world energy demand in a sustainable and clean Wind energy has remarkable potential for fulfilling the and timely diagnosis information to be used in an appropriate Wind energy has remarkable potential for fulfilling the increasing world energy demandpotential in a sustainable and clean to the high degree of nonlinearity and uncertainty present in automatic signal correction (ASC) algorithm. However, due Wind energy has remarkable for fulfilling the automatic signal correction (ASC) algorithm. However, due to the high degree of nonlinearity and uncertainty present in way. In order to reduce the average cost of wind energy, large increasing world energythedemand demand incost sustainable and clean clean automatic signalmodelling correctionof(ASC) algorithm. However, due increasing world energy aaa sustainable and way. In order to reduce averagein of wind energy, large to wind turbines, wind energy systems is rather the high degree of nonlinearity and uncertainty present in increasing world energy demand in sustainable and clean to the high degree of nonlinearity and uncertainty present in wind turbines, modelling of wind energy systems is rather wind turbines are often installed in clusters called wind farms, way. In order to average cost of wind large to the high degree of nonlinearity andmodels uncertainty present in way. In order to reduce the average cost of wind energy, large wind turbines arereduce often the installed in clusters calledenergy, wind farms, challenging. Moreover, the available are too detailed wind turbines, modelling of wind energy systems is rather way. In order to reduce the average cost of wind energy, large wind turbines, modelling of wind energy systems is rather challenging. Moreover, the available models are too detailed particularly in offshore locations. As more and more offshore wind turbines are often installed in clusters called wind farms, wind turbines, for modelling of wind energy design systemsand is rather wind turbines are often installed in clusters called wind farms, particularly in offshore locations. As more and more offshore and uncertain direct use in control fault challenging. Moreover, the available models are detailed wind turbines are often installed in clusters called wind farms, challenging. Moreover, theuse available modelsdesign are too tooand detailed and uncertain for direct in control fault wind farms are developed, further from shore, both the particularly offshore locations. As more more offshore Moreover, the available models are too detailed particularly in offshore locations. As more and more offshore wind farmsin are developed, further fromand shore, both the challenging. diagnosis purposes. Considering such facts, this paper and uncertain for direct use in control design and fault particularly in offshore locations. As more and more offshore and uncertain for direct use in control design and fault diagnosis purposes. Considering such facts, this paper factors of complexity and limited accessibility together with wind farms are developed, further from shore, both the and uncertain for direct use in control design and fault wind farms are developed, further from shore, both the factors of complexity and limited accessibility together with proposes the application of data-driven models based on diagnosis purposes. Considering such facts, this paper wind farms are developed, further from shore, both the diagnosis purposes. Considering such facts, this paper proposes the application of data-driven models based on harsh climate conditions come into play and result in higher factors of complexity and limited accessibility together with diagnosis purposes. Considering such facts, this paper factors of complexity and limited accessibility together with harsh climate conditions come into play and result in higher fuzzy modelling and identification (FMI) technique for FDD proposes the application of data-driven models based on factors of complexity and limited accessibility together with proposes the application of data-driven models based on fuzzy modelling and identification (FMI) technique for FDD failure rates and maintenance challenges. This motivates the harsh climate conditions come into play and result in higher proposesSuch the application ofbedata-driven models based on harsh climate conditions come into play and result in higher failure rates and maintenance challenges. This motivates the design. models can more accurate than process fuzzy modelling and identification (FMI) technique for FDD harsh climate conditions come into play fault and result in higher fuzzy modelling and identification (FMI) technique for FDD design. Such models can be more accurate than process design and development of advanced detection and failure rates and maintenance challenges. This motivates the fuzzy modelling and identification (FMI) technique for FDD failure rates rates and maintenance challenges.fault This detection motivates and the design. design and and development of advanced models since they are on information (i.e. the Such models can be more accurate than process failure maintenance challenges. This motivates the design. Such models can be more accurate than process models since are based based on objective objective information (i.e. the diagnosis (FDD) as as fault-tolerant control (FTC) design and development fault detection and design. while Such they models can models be moremay accurate than process design and development of advanced fault detection and diagnosis (FDD) as well well of asadvanced fault-tolerant control (FTC) data), the process often suffer from models since they are based on objective information (i.e. the design and development of advanced fault detection and models since they are based on objective information (i.e. data), while the process models may often suffer from schemes in wind farms to improve their reliability and diagnosis well fault-tolerant control (FTC) models since they are based on the objective information (i.e. the the diagnosis (FDD) as well as fault-tolerant control (FTC) schemes in(FDD) wind as farms toas improve their reliability and data), incompleteness in representing modelled process. while the process models may often suffer from diagnosis (FDD) as well as fault-tolerant control (FTC) data), process may suffer incompleteness in representing the modelled process. availability. schemes in wind farms to improve their reliability and data), while while the the process models models may often often suffer from from schemes in wind farms to improve their reliability and availability. in representing the modelled process. schemes in wind farms to improve their reliability and incompleteness incompleteness in of representing the AFTC modelled process. The the scheme is availability. incompleteness in representing the modelled process. availability. The effectiveness effectiveness of the proposed proposed AFTC scheme is evaluated evaluated In general, the FDD and FTC methods can be applied at both availability. In general, the FDD and FTC methods can be applied at both The by simulation on an advanced offshore wind farm benchmark effectiveness of the proposed AFTC scheme is evaluated The effectiveness of the proposed AFTC scheme is by simulation on an advanced offshore wind farm benchmark individual wind turbine and entire wind farm levels. Recently, In general, wind the FDD FDD andand FTC methods can belevels. applied at both both The effectiveness of the proposed AFTC schememeasurement is evaluated evaluated In general, the and FTC methods be applied at individual turbine entire windcan farm Recently, model in the presence of wind turbulences, by simulation on an advanced offshore wind farm benchmark In general, the FDD and FTC methods can be applied at both by simulation an advanced offshore wind farm benchmark model in the on presence of wind turbulences, measurement research works have been focused more on the applications individual wind turbine and entire wind farm levels. Recently, by simulation on an advanced offshore wind farm benchmark individual wind turbine and entire wind farm levels. Recently, research works have been focused more on the applications noises load variations. model in the wind turbulences, measurement individual wind turbine and entire wind farm levels. Recently, model and in presence of noises and loadpresence variations.of of such in FDD methods at turbine research works have been focused on the applications in the the presence of wind wind turbulences, turbulences, measurement measurement research works have been focused more on the applications of such methods, methods, in particular, particular, FDDmore methods at wind wind turbine model noises and load variations. research works have been focused more on the applications noisesremainder and load load variations. variations. level (for example, Odgaard et al., 2009b, Tabatabaeipour et The of the paper is organized as follows: of such methods, in particular, FDD methods at wind turbine noises and of such methods, in particular, FDD methods at wind turbine level (for example, Odgaard et al., 2009b, Tabatabaeipour et The remainder of the paper is organized as follows: In In of such methods, in particular, FDDetmethods at wind turbine al., 2012, Sloth et al., 2010, Badihi al., 2014b, Badihi et al., Section 2, the used wind farm benchmark model is briefly level (for example, Odgaard et al., 2009b, Tabatabaeipour et The remainder of the paper is organized as follows: In level (for example, Odgaard et al., 2009b, Tabatabaeipour et The remainder of the paper is organized as follows: In al., 2012, Sloth et al., 2010, Badihi et al., 2014b, Badihi et al., Section 2, the used wind farm benchmark model is briefly level (forMost example, Odgaard et al., 2009b, the Tabatabaeipour et The remainder of the paper isis organized as follows: In 2014a). of these works try to address FDD and FTC described. The considered fault described and analysed in al., 2012, Sloth et al., 2010, Badihi et al., 2014b, Badihi et al., Section 2, the used wind farm benchmark model is briefly al., 2012, Sloth et al., 2010, Badihi et al., 2014b, Badihi et al., Section 2, the used wind farm benchmark model is briefly 2014a). Most of these works try to address the FDD and FTC described. The considered fault is described and analysed in al., 2012, Sloth et al., 2010, Badihi et al., 2014b, Badihi et al., Section 2, the used wind farm benchmark model is briefly problems in two wind turbine benchmark models presented in Section 3. A short description about the used FMI technique 2014a). Most of these works try to address the FDD and FTC described. The considered fault is described and analysed in 2014a). Most of these works try to address the FDD and FTC described. The considered fault is described and analysed in problems in two wind turbine benchmark models presented in Section 3. A short description about the used FMI technique 2014a). Most of 2009a) these works try to address the FDD 2013). and FTC described. The the considered fault is described and analysed in (Odgaard et al., and (Odgaard and Johnson, A for modelling system is presented in Section 4. The problems in two wind turbine benchmark models presented in Section 3. A short description about the used FMI technique problems in two wind turbine benchmark models presented in Section 3. A A short short description about the used used FMI technique technique (Odgaard et two al., 2009a) and (Odgaard andmodels Johnson, 2013). in A for modelling the description system is about presented in Section 4. The problems in wind turbine benchmark presented Section 3. the FMI recent review of the literature in (Badihi et al., 2013) proposed AFTC scheme is presented in Section 5. Section (Odgaard et al., of 2009a) (Odgaard and Johnson, 2013). A for modelling system is presented in Section 4. The6 (Odgaard et 2009a) and (Odgaard and 2013). A for the system is in 4. recent review the and literature in (Badihi et al., 2013) proposed AFTCthe scheme is presented in Section 5. Section 6 (Odgaard et al., al.,references 2009a) and (Odgaard and Johnson, Johnson, 2013). A presents for modelling modelling the system is presented presented in Section Section 4. The The provides more on FDD and FTC for wind turbines. the simulation results with some comments and recent review of the literature in (Badihi et al., 2013) proposed AFTC scheme is presented Section 5. Section recent review of the literature in (Badihi et al., 2013) proposed AFTC scheme is presented in Section 5. Section 6 provides more references on FDD and FTC for wind turbines. presents the simulation results within some comments and6 recent review of the literature in (Badihi et al., 2013) proposed AFTC scheme is presented in Section 5. Section 6 With respect to the FDD and FTC at wind farm level, only aa presents discussions. Finally, conclusions are drawn in Section 7. provides more references on FDD and FTC for wind turbines. the simulation results with some comments and provides more references on FDD and FTC for wind turbines. presents the simulation results with some comments and With respect to the FDD and FTC at wind farm level, only discussions. Finally, conclusions are drawn in Section 7. provides more references on FDD and FTC for wind turbines. presents the simulation results with some comments and few research works on condition monitoring and fault With respect to the FDD FTC at wind farm level, Finally, conclusions are drawn in Section 7. With respect to the FDD and FTC farm level, only discussions. Finally, conclusions are drawn 7. few onand condition monitoring and only faultaaa discussions. 2. OVERVIEW OF WIND FARM MODEL With research respect to works thefarms FDD and FTC at at wind wind farm level, only discussions. Finally, conclusions areBENCHMARK drawn in in Section Section 7. 2. OVERVIEW OF WIND FARM BENCHMARK MODEL detection in wind are reported. Three examples can be few research works on condition monitoring and fault few research works on condition monitoring and fault detection in wind farms are reported. Three examples can be 2. OVERVIEW OF WIND FARM BENCHMARK MODEL few research works on condition monitoring and fault This paper considers an advanced wind farm simulation 2. OVERVIEW OF WIND FARM BENCHMARK MODEL found in Verma, 2011, and Verma, 2012, detection in windand farms are reported. Three examples be 2. OVERVIEW OF WIND FARM BENCHMARK MODEL paper considers an advanced wind farm simulation detection in farms are Three examples can be found in (Kusiak (Kusiak and Verma, 2011, Kusiak Kusiak and Verma,can 2012, detection inal.,wind wind farms are reported. reported. Three examples can be This benchmark model developed in the EU-FP7 project, This paper considers an advanced wind farm simulation Simani et 2014). However, some faults have to be dealt found in (Kusiak and Verma, 2011, Kusiak and Verma, 2012, This paper considers an advanced wind farm simulation benchmark model developed in the EU-FP7 project, found in and Verma, 2011, Kusiak Verma, Simani al., 2014). However, some faultsand have to be 2012, dealt This paper(Soltani considers an 2009). advanced wind farm simulation found at inet(Kusiak (Kusiak and farm Verma, 2011, Kusiak and Verma, 2012, AEOLUS et al., The model allows control benchmark model developed in the EU-FP7 with the wind control level only. Given the Simani al., However, some faults have to model EU-FP7 project, AEOLUS (Soltani etdeveloped al., 2009). in Thethe model allows project, control Simaniatet et the al., 2014). 2014). However, somelevel faultsonly. have Given to be be dealt dealt with wind farm control the benchmark benchmark model developed in the EU-FP7 project, Simani et al., 2014). However, some faults have to be dealt designers to develop and investigate farm level control AEOLUS (Soltani et al., 2009). The model allows importance of this issue, the current paper presents the design with at the wind farm control level only. Given the AEOLUS (Soltani et al., 2009). The model allows designers to develop and investigate farm level control with at the wind farm control level only. Given the importance thecontrol current paper presents the design AEOLUS under (Soltani et al.,operating 2009). The model for allows control with at theof this windissue, farm level only. Given the designers solutions various conditions an to develop investigate farm control and of novel active fault-tolerant control importance of this issue, current paper presents the design designers to develop and investigate farm level control solutions under various and operating conditions forlevel an optional optional importance of this issue, the current paper presents the design and development development of aa the novel active fault-tolerant control designers to develop and investigate farm level control importance of this issue, the current paper presents the design quantity layout of turbines installed a an wind farm. solutions under various operating conditions for optional (AFTC) scheme anaoffshore wind against and development novel active fault-tolerant control solutions under various operating conditions for an optional quantity and and layout of wind wind turbines installed in in wind farm. and of novel active fault-tolerant control (AFTC) scheme for forof wind farm farm against decreased decreased solutions under various operating conditions foraand an optional and development development ofanaaoffshore novel active fault-tolerant control Fig. 1 shows the default wind farm layout, Fig. quantity and layout of wind turbines installed in a wind farm.2 power generation caused by turbine blade erosion and debris (AFTC) scheme for an offshore wind farm against decreased quantity and layout of wind turbines installed in a wind Fig. 1 shows the default wind farm layout, and Fig. (AFTC) scheme for an offshore wind farm against decreased power generation caused by turbine blade erosion and debris quantity and layout of wind turbines installed in a wind farm. farm.2 (AFTC) scheme for an offshore wind The farm proposed against decreased Fig. 1 shows the default wind farm layout, and Fig. 2 build-up on the blades over time. scheme power generation caused by turbine blade erosion and debris Fig. 1 shows the default wind farm layout, and Fig. 2 power generation caused by turbine blade erosion and debris build-up on the blades over time. The proposed scheme Fig. 1 shows the default wind farm layout, and Fig. 2 power generation caused by turbine blade erosion and debris build-up on the blades over time. The proposed scheme build-up on the blades over time. The proposed scheme build-up on the blades over time. The proposed scheme 2405-8963 © 2015, IFAC (International Federation of Automatic Control) Copyright 2015 IFAC 1369Hosting by Elsevier Ltd. All rights reserved. Copyright 2015 responsibility IFAC 1369Control. Peer review© under of International Federation of Automatic Copyright © 2015 IFAC 1369 Copyright © 2015 IFAC 1369 10.1016/j.ifacol.2015.09.716 Copyright © 2015 IFAC 1369

SAFEPROCESS 2015 1370 September 2-4, 2015. Paris, France

Hamed Badihi et al. / IFAC-PapersOnLine 48-21 (2015) 1369–1374

illustrates the overall structure of the model under consideration.

T8

T10

T9

Wind

D3 D2 T6

T5

T7

D3

D2 T2

T1

T3 D1

D1

T4 D1

With respect to the outputs, the component of wind turbines in Fig. 2 generates a set of outputs including a set of measurements 𝑴𝑴 required for use by the wind farm controller along with a set of coefficients of thrust 𝑪𝑪𝑻𝑻 for turbines, necessary to calculate the wake effects (i.e., low speed turbulent air flows behind turbine) by wind field component.

Fig. 1. Wind farm layout (D1=600m, D2=500m, D3=300m). 𝑪𝑪𝑻𝑻

Wind Turbines

𝑽𝑽𝒏𝒏𝒏𝒏𝒏𝒏 , 𝑽𝑽𝒓𝒓𝒓𝒓𝒓𝒓

Wind Field

𝑴𝑴 𝑷𝑷𝒅𝒅,𝒒𝒒

turbine is represented using a simple model of an offshore 5 MW baseline turbine proposed by the U.S. National Renewable Energy Laboratory (NREL) (see Jonkman et al. (2009)). As it is shown in Fig. 3, the baseline control system in each individual wind turbine basically acts upon the power demand 𝑃𝑃𝑑𝑑,𝑞𝑞 in (1) specified by the wind farm controller. To this end, the turbine’s baseline control system employs a blade-pitch controller as well as a torque controller to compute appropriate blade-pitch reference 𝛽𝛽𝑟𝑟,𝑞𝑞 and generator torque reference 𝜏𝜏𝑟𝑟,𝑞𝑞 , respectively. A complete description of the wind turbine benchmark model can be found in (Jonkman et al., 2009).

Wind Turbine # q

𝑃𝑃𝑎𝑎

Wind Farm Controller

𝑃𝑃𝑑𝑑

Local Wind Profile Baseline 𝛽𝛽𝑟𝑟.𝑞𝑞

Network Operator

𝑃𝑃𝑑𝑑,𝑞𝑞

Fig. 2. Illustration of overall model structure (This figure is based on Soltani et al. (2009)). As it is shown in Fig. 2, this benchmark model is composed of four major components: A) Network Operator The network operator determines the total active power demand 𝑃𝑃𝑑𝑑 required for safe and reliable connection of wind farm to the electrical grid. The baseline model for network operator can function in different modes such as: absolute, delta, and frequency regulation modes. Basically, in the frequency regulation mode used in this paper, the measured grid frequency is used as a feedback signal to set up active power control in real-time and maintain the necessary balance between power generation and load, which in turn regulates the grid frequency to its reference value despite a changing grid load. B) Wind Farm Controller As it is shown in Fig. 2, the wind farm controller plays an interface role that ensures appropriate distribution of operator total demanded power 𝑃𝑃𝑑𝑑 among wind turbines in the farm while providing an estimate of total available power 𝑃𝑃𝑎𝑎 in the wind farm to the operator (e.g., in the case of delta mode operator). The baseline wind farm controller in (1) operates using a proportional distribution algorithm that sends a set of power demands 𝑃𝑃𝑑𝑑,𝑞𝑞 (𝑘𝑘) at the time step 𝑘𝑘 (i.e., 𝑷𝑷𝒅𝒅,𝒒𝒒 in Fig. 2) to each of 𝑁𝑁 individual turbines based on a simple estimate of their current available powers 𝑃𝑃𝑎𝑎,𝑞𝑞 (𝑘𝑘) and the total available 𝑃𝑃𝑎𝑎 (𝑘𝑘) and total demanded 𝑃𝑃𝑑𝑑 (𝑘𝑘) powers in the farm. 𝑃𝑃𝑎𝑎,𝑞𝑞 (𝑘𝑘) (1) , 𝑞𝑞 = 1, … , 𝑁𝑁 𝑃𝑃𝑑𝑑,𝑞𝑞 (𝑘𝑘) = 𝑃𝑃𝑑𝑑 (𝑘𝑘) 𝑃𝑃𝑎𝑎 (𝑘𝑘) C) Wind Turbines This component simulates the dynamics of the wind turbines installed in the farm based on the measured nacelle wind speed 𝑉𝑉𝑛𝑛𝑛𝑛𝑛𝑛 , effective wind speed 𝑉𝑉𝑟𝑟𝑟𝑟𝑟𝑟 , and power demands 𝑃𝑃𝑑𝑑,𝑞𝑞 at each individual turbine. Each

Control System

𝑃𝑃𝑔𝑔,𝑞𝑞

𝜏𝜏𝑟𝑟,𝑞𝑞

Fig. 3. The 𝑞𝑞th wind turbine in the farm (𝑞𝑞 = 1, … , 𝑁𝑁). Note that in addition to the generated power 𝑃𝑃𝑔𝑔,𝑞𝑞 , the turbine model provides many other measured variables. D) Wind Field The interactions between the wind turbines installed in a wind farm can be represented through the wind field model. This model simulates the wind speed throughout the farm based on an ambient field model together with a wake model which describes wakes meandering behind turbines and their effect on the ambient wind field. 3. BLADE EROSION/DEBRIS BUILD-UP FAULT Basically, decreased power generation in a wind farm may be due to several different malfunctions. However, blade erosion along with debris build-up on the blades due to dirt, ice, etc., constitute the most probable fault which results in a lower power generation because of changes in the aerodynamics of the wind turbine, and thereby lowering the maximally obtained power. In more detail, the aerodynamic torque 𝜏𝜏𝑎𝑎𝑎𝑎𝑎𝑎 applied to the rotor by the wind is defined in (2), in which 𝜔𝜔𝑟𝑟𝑟𝑟𝑟𝑟 is the rotor angular speed, 𝜌𝜌 is the air density, 𝐴𝐴 is the swept area of the turbine rotor, and 𝑉𝑉𝑤𝑤 is the wind speed (Pao and Johnson, April, 2011). 1 (2) 𝜌𝜌 𝐴𝐴 𝑉𝑉𝑤𝑤3 (𝑡𝑡) 𝐶𝐶𝑝𝑝 (𝛽𝛽(𝑡𝑡), 𝜆𝜆(𝑡𝑡)) 2𝜔𝜔𝑟𝑟𝑟𝑟𝑟𝑟 (𝑡𝑡) In (2), the power coefficient 𝐶𝐶𝑝𝑝 (𝛽𝛽, 𝜆𝜆) is a three-dimensional surface as a function of the tip-speed ratio 𝜆𝜆 and the blade pitch angle 𝛽𝛽 where the latter two terms determine the operating condition of a variable speed wind turbine. For variable-speed wind turbines, the turbine is ideally operated at the peak of the 𝐶𝐶𝑝𝑝 surface in order to capture as much power as possible. However, over time, the blade erosion and debris build-up fault shifts the turbine’s 𝐶𝐶𝑝𝑝 surface downward,

1370

𝜏𝜏𝑎𝑎𝑎𝑎𝑎𝑎 (𝑡𝑡) =

SAFEPROCESS 2015 September 2-4, 2015. Paris, France


resulting in lower energy capture through not only a decrease in the peak value of the 𝐶𝐶𝑝𝑝 surface but also a change in the location of the peak of the 𝐶𝐶𝑝𝑝 surface. The sensitivity of energy loss to the changes at the peak of the 𝐶𝐶𝑝𝑝 surface due to this fault is considered and quantified in (Fingersh and Carlin, 1998), which concludes that the fault can lead to a substantial off optimal operation and power loss. Given the importance of the blade erosion and debris build-up fault in wind turbines, it is necessary to detect, diagnose, and accommodate such a fault in a timely and effective manner. However, it is difficult to handle this fault at a wind turbine control level, mainly because of the fact that a lower generated power may be due to either debris build-up on the blades or simply that the true wind speed is lower than the measured/estimated wind speed. Conversely, at a wind farm level, it is possible to compare the performance and operation features of the different wind turbines in a given wind farm. The described benchmark model in Section 2, in its original form, does not include any fault. However, it is possible to fairly model a realistic scenario for the blade erosion and debris build-up fault and incorporate it in the benchmark model. In the following lines, the modeling of the considered fault in this paper is described. In general, the generator power 𝑃𝑃𝑔𝑔 can be expressed as: 𝑃𝑃𝑔𝑔 (𝑡𝑡) = 𝜂𝜂𝑔𝑔 𝜂𝜂𝑚𝑚 𝑃𝑃𝑎𝑎𝑎𝑎𝑎𝑎 (𝑡𝑡) (3) in which 𝜂𝜂𝑔𝑔 is the generator efficiency, 𝜂𝜂𝑚𝑚 is the efficiency of transmission system, and 𝑃𝑃𝑎𝑎𝑎𝑎𝑎𝑎 is the aerodynamic rotor power given by (Pao and Johnson, April, 2011): 𝑃𝑃𝑎𝑎𝑎𝑎𝑎𝑎 (𝑡𝑡) = 𝜏𝜏𝑎𝑎𝑎𝑎𝑎𝑎 (𝑡𝑡) 𝜔𝜔𝑟𝑟𝑟𝑟𝑟𝑟 (𝑡𝑡)

𝑦𝑦(𝑘𝑘 + 1) = 𝑓𝑓(𝚽𝚽(𝑘𝑘)) + 𝑒𝑒

𝚽𝚽(𝑘𝑘) = [𝑦𝑦(𝑘𝑘), … , 𝑦𝑦(𝑘𝑘 − 𝑚𝑚 + 1), 𝑢𝑢(𝑘𝑘), … , 𝑢𝑢(𝑘𝑘 − 𝑛𝑛 + 1)]

1371

(6)

in which 𝚽𝚽(𝑘𝑘) is an information data vector including the past model inputs 𝑢𝑢 and outputs 𝑦𝑦, 𝑘𝑘 is the discrete-time-step, {𝑚𝑚, 𝑛𝑛} ∈ ℤ denote the orders of model that are defined by the user, and 𝑒𝑒 denotes the modelling error. A T-S type fuzzy model can approximate the unknown function 𝑓𝑓(. ) in terms of q rules as follows (Babuska, 1998): 𝐑𝐑𝐑𝐑𝐑𝐑𝐑𝐑𝒊𝒊 : 𝐈𝐈𝐈𝐈 𝚽𝚽(𝑘𝑘) is 𝑨𝑨𝑖𝑖 𝐭𝐭𝐭𝐭𝐭𝐭𝐭𝐭 𝑦𝑦̂𝑖𝑖 (𝑘𝑘 + 1) = 𝐹𝐹𝑖𝑖 (𝚽𝚽(𝑘𝑘)) where, 𝑨𝑨𝒊𝒊 are the antecedent fuzzy sets of the ith rule ( 𝑖𝑖 = 1,2, … , 𝑞𝑞 ). 𝐹𝐹𝑖𝑖 (. ) can have a simple but practically efficient linear form as follows (Babuska, 1998): (7) 𝐹𝐹𝑖𝑖 (. ): 𝑦𝑦̂𝑖𝑖 (𝑘𝑘 + 1) = 𝒂𝒂𝒊𝒊 𝚽𝚽 + 𝑏𝑏𝑖𝑖 where 𝒂𝒂𝒊𝒊 and 𝑏𝑏𝑖𝑖 denote the parameter vector and scalar offset of the ith rule, respectively. The dynamics of a nonlinear system can be represented using the fuzzy fusion (8) over all model outputs (Babuska, 1998): 𝑞𝑞 𝑞𝑞 (8) 𝑦𝑦̂ = (∑𝑖𝑖=1 𝜇𝜇𝑖𝑖 (𝚽𝚽)𝑦𝑦̂𝑖𝑖 )/(∑𝑖𝑖=1 𝜇𝜇𝑖𝑖 (𝚽𝚽)) where 𝑦𝑦̂ is the aggregated output of the model, and 𝜇𝜇𝑖𝑖 represents the degree of fulfillment of ith rule.

It is worth mentioning that, in order to identify T-S models including the antecedent fuzzy sets, consequent parameter vectors and scalar offsets, the well-established Gustafson– Kessel (GK) clustering algorithm (Gustafson and Kessel, 1978) is used in this paper.

(4)

5. AN AFTC SCHEME

Now, substituting (4) into (3) and using (2) yields: (𝜂𝜂𝑔𝑔 . 𝜂𝜂𝑚𝑚 ) (5) 𝜌𝜌 𝐴𝐴 𝑉𝑉𝑤𝑤3 (𝑡𝑡) 𝐶𝐶𝑝𝑝 (𝛽𝛽(𝑡𝑡), 𝜆𝜆(𝑡𝑡)) 2 The equation (5) shows that the generated power 𝑃𝑃𝑔𝑔 is a direct function of the power coefficient 𝐶𝐶𝑝𝑝 that itself can be changed due to the blade erosion and debris build-up fault. With respect to this fact, the fault can be simply modeled by scaling the generated power in a wind turbine. In this paper, a realistic scaling factor of 0.97 (3% power loss) is used. Therefore, the benchmark model presented in Section 2 is modified with a fault scenario representing occurrence of 3% power loss in a wind turbine in the farm shown in Fig. 1. 𝑃𝑃𝑔𝑔 (𝑡𝑡) =

4. FUZZY MODELLING The proposed AFTC scheme in this paper employs a modelbased FDD system that relies on the nominal dynamic models of the system. This section describes a data-driven modelling approach based on FMI technique used for modelling of the system. The dynamic model can be represented by a Takagi-Sugeno (T-S) type dynamic fuzzy model identified from input/output system’s measurements obtained from the simulation of the system under normal (fault-free) operation. In the following lines, the used fuzzy modelling and identification method is described briefly. Without loss of generality, the following form defines a single-input single-output (SISO) system:

This section presents an AFTC scheme designed to improve the reliability and availability of wind farms against the fault discussed in Section 3. As already mentioned, the considered fault results in a lower power generation which negatively affects the total active power generated by all wind turbines in a farm as a whole. Generally speaking, when such a fault occurs in one or more turbines in a wind farm, the wind farm controller still has to follow the operator total demanded power 𝑃𝑃𝑑𝑑 using the proportional distribution algorithm described in (1), no matter which turbine(s) is/are faulty. Whereas, to compensate the power loss caused by the faulty turbine(s), it is necessary to avoid overloading the remaining healthy turbines but only correcting the reference power signal(s) to the faulty turbine(s) and thereby accommodating the fault effects. In fact, overloading the healthy wind turbines may lead to high structural loading and fatigue. Here, it is assumed that the fault may occur in any turbine installed in a farm. But, no more than one faulty turbine can occur at a time. The AFTC scheme proposed in this paper is shown in Fig. 4. As it is observed in this figure, the AFTC scheme basically relies on an integrated FDD and ASC process that covers the entire farm. Basically, the main idea behind the proposed scheme is to monitor the consistency of generated powers from turbines in a farm through dividing them into groups of three, and then to conduct the FDD in each group in real-time. Finally, the FDD information is used for ASC and accommodation of the fault in any faulty turbine. The FDD system in Fig. 4 employs a model-based fault

1371



detection and estimation algorithm based on FMI technique to provide the most up-to-date information about true status of the wind farm system. The ASC mechanism means that the nominal controllers at both wind turbine and wind farm levels are kept unchanged; only the output of the torque controller in any faulty turbine is corrected according to the real-time fault information from the FDD system. Here, the supervision process shown in Fig. 4 is not described explicitly, because it is very simple in the considered case. In fact, according to the provided information from the FDD system, the supervisor only identifies the faulty turbine in the farm together with its relevant estimated power loss (fault magnitude) 𝑃𝑃̂𝑞𝑞 due to the fault. The FDD system and ASC mechanism are further described in the following two sections, respectively.

in which 𝑁𝑁 is the number of wind turbines in the farm and the ceiling function 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 returns the smallest integer that is greater than or equal to 𝑁𝑁/3. For example, in case of a wind farm with 20 turbines (𝑁𝑁=20), 𝑅𝑅 is seven. The FDD system shown in Fig. 4 is basically composed of 𝑅𝑅 modules each used for monitoring the consistency of generated powers in a group of three turbines in the farm with 𝑁𝑁 turbines. Fig. 5 shows the general structure of the FDD system including its modules and inputs/outputs. Module # 𝟏𝟏 𝑷𝑷𝒓𝒓

𝑴𝑴

𝑪𝑪𝑻𝑻

Wind Turbines

𝑽𝑽𝒏𝒏𝒏𝒏𝒏𝒏 , 𝑽𝑽𝒓𝒓𝒓𝒓𝒓𝒓

FDD System

Fig. 5. FDD system including 𝑅𝑅 modules each for FDD process in a group of three wind turbines. In Fig. 5, 𝑴𝑴(𝑘𝑘) is a vector of performance data including measured variables and control commands/references in wind turbines. However, the FDD system only needs the vectors of reference powers and generated powers defined in (10) and (11), respectively.

𝑴𝑴 𝑰𝑰 𝑷𝑷𝒅𝒅,𝒒𝒒

FDD System

𝑃𝑃𝑎𝑎

Wind Farm Controller

𝑃𝑃𝑑𝑑

Network Operator

Fig. 4. Schematic of the proposed AFTC scheme based on an integrated FDD system and ASC mechanism. Based on FDD information 𝑰𝑰(𝑘𝑘), the supervisor applies appropriate signal ̂ 𝒒𝒒 (𝑘𝑘) . correction/modification using power loss estimates 𝑷𝑷 Note that the farm includes 𝑁𝑁 turbines (𝑞𝑞 = 1, … , 𝑁𝑁). 5.1 FDD System With respect to the facts discussed in Section 3, the considered fault is difficult to be detected and identified at a wind turbine level. Therefore, as it is observed in Fig. 4, the proposed AFTC scheme here addresses the considered fault at a wind farm level through comparing the powers generated by different wind turbines in a given wind farm. However, monitoring the consistency of generated powers from different turbines in a farm is not a straightforward process even in the case of adjacent turbines where wind condition may be almost similar. In fact, the reference power signals are not necessarily similar for the turbines in a farm. Therefore, in order to monitor the consistency of generated powers from any two arbitrary wind turbines, it is required to consider not only the generated power responses from the turbines but also the relevant turbines’ reference power signals. An efficient approach is to divide turbines installed in a wind farm into groups of similar size and then monitor the consistency of generated powers from turbines in each group. To this end, the current paper suggests that all turbines in a wind farm be divided into groups of three regardless of their locations in the farm (i.e., the turbines in a group are not necessarily adjacent). In fact, the smallest number of groups required for covering a wind farm is computed by: 𝑅𝑅 = 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑁𝑁/3)

𝑰𝑰

Module # 𝑹𝑹 Wind Field

̂ 𝒒𝒒 𝑷𝑷

Supervisor

Module # 𝒑𝒑 𝑷𝑷𝒈𝒈

(9)

𝑷𝑷𝒓𝒓 (𝑘𝑘) = [𝑃𝑃𝑟𝑟,𝑞𝑞 (𝑘𝑘)] , 𝑞𝑞 = 1, … , 𝑁𝑁 𝑞𝑞 = 1, … , 𝑁𝑁 𝑷𝑷𝒈𝒈 (𝑘𝑘) = [𝑃𝑃𝑔𝑔,𝑞𝑞 (𝑘𝑘)] , The reference powers 𝑃𝑃𝑟𝑟,𝑞𝑞 in (10) is computed as: 𝑃𝑃𝑟𝑟,𝑞𝑞 (𝑘𝑘) = 𝜏𝜏𝑟𝑟,𝑞𝑞 (𝑘𝑘) ∙ 𝜔𝜔𝑔𝑔,𝑞𝑞 (𝑘𝑘)

(10) (11)

(12) in which 𝜏𝜏𝑟𝑟,𝑞𝑞 and 𝜔𝜔𝑔𝑔,𝑞𝑞 are the generator torque reference and generator angular speed for 𝑞𝑞 th wind turbine in the farm, respectively. The FDD information vector 𝑰𝑰(𝑘𝑘) includes all results from 𝑅𝑅 modules as follows: (13) 𝑰𝑰(𝑘𝑘) = [𝐼𝐼1 (𝑘𝑘), 𝐼𝐼2 (𝑘𝑘), … , 𝐼𝐼𝑅𝑅 (𝑘𝑘)] As already mentioned, each module in the FDD system in Fig. 5 monitors the consistency of generated powers in a group of three turbines based on a model-based FDD approach. All modules are essentially the same but each is used for a particular group of three turbines. Fig. 6 presents the functionality of an arbitrary module that is called Module #𝑃𝑃 in Fig. 5. This module monitors the generated power by three different turbines called Ti, Tj, and Tk that are included in an individual group. As it is shown in Fig. 6, the module employs two dynamic models that are essentially the same but with different inputs/outputs for estimating the nominal relative performance between turbine Ti and turbines Tj and Tk. The dynamic models are designed and developed using the FMI technique already described in Section 4. In more detail, as it is observed in Fig. 6, each dynamic model uses the relative reference power ∆𝑃𝑃𝑟𝑟 as the input and estimates the nominal relative generated power ∆𝑃𝑃𝑔𝑔 that needs to be compared with the actual relative generated power ∆𝑃𝑃𝑔𝑔 using a threshold test on the instantaneous values of the produced residuals 𝑟𝑟 . In this way, as already mentioned, the consistency of generated powers from any two arbitrary wind

1372

SAFEPROCESS 2015 September 2-4, 2015. Paris, France


turbines is monitored by considering not only the generated power responses from the turbines but also the relevant turbines’ reference power signals. The power consistency itself is determined by threshold test (14) on the residuals 𝑟𝑟: (14) (𝑟𝑟𝑚𝑚 − 𝛿𝛿𝑟𝑟𝑠𝑠𝑠𝑠𝑠𝑠 ) ≤ 𝑟𝑟(𝑘𝑘) ≤ (𝑟𝑟𝑚𝑚 + 𝛿𝛿𝑟𝑟𝑠𝑠𝑠𝑠𝑠𝑠 ) where 𝑟𝑟𝑚𝑚 and 𝑟𝑟𝑠𝑠𝑠𝑠𝑠𝑠 are respectively the mean and the standard deviation values of residuals 𝑟𝑟 during fault-free operation. The tuning parameter 𝛿𝛿 should be properly chosen to minimize false detection and missed detection rates. Obviously, the generated powers from any two arbitrary wind turbines are consistent as long as (14) is satisfied. ∆𝑃𝑃𝑟𝑟,𝑖𝑖𝑖𝑖 = 𝑃𝑃𝑟𝑟,𝑖𝑖 − 𝑃𝑃𝑟𝑟,𝑗𝑗

Dynamic Model

∆𝑃𝑃𝑔𝑔,𝑖𝑖𝑖𝑖

𝑟𝑟𝑖𝑖𝑖𝑖 ∑

− +

∆𝑃𝑃𝑔𝑔,𝑖𝑖𝑖𝑖 = 𝑃𝑃𝑔𝑔,𝑖𝑖 − 𝑃𝑃𝑔𝑔,𝑗𝑗

Dynamic Model

∆𝑃𝑃𝑔𝑔,𝑖𝑖𝑖𝑖

Decision Making

∆𝑃𝑃𝑟𝑟,𝑖𝑖𝑖𝑖 = 𝑃𝑃𝑟𝑟,𝑖𝑖 − 𝑃𝑃𝑟𝑟,𝑘𝑘

Threshold Test 𝑟𝑟𝑖𝑖𝑖𝑖 ∑

−

𝑰𝑰𝒑𝒑

+

∆𝑃𝑃𝑔𝑔,𝑖𝑖𝑖𝑖 = 𝑃𝑃𝑔𝑔,𝑖𝑖 − 𝑃𝑃𝑔𝑔,𝑘𝑘

Threshold Test

and full-load regions, the estimated fault magnitude is used to correct the nominal reference torque control signal computed by the torque controller. This signal correction is defined below for the 𝑞𝑞th wind turbine: 𝑃𝑃̂𝑞𝑞 (𝑘𝑘) (15) 𝜏𝜏𝑟𝑟,𝑞𝑞,𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (𝑘𝑘) = 𝜏𝜏𝑟𝑟,𝑞𝑞 (𝑘𝑘) + 𝜔𝜔𝑔𝑔,𝑞𝑞 (𝑘𝑘) In (15), 𝜏𝜏𝑟𝑟,𝑞𝑞,𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 is the corrected generator torque reference and 𝑃𝑃̂𝑞𝑞 is the fault magnitude both in the 𝑞𝑞th wind turbine in the farm. As already mentioned, the fault magnitude is estimated by the FDD system based on the logic presented in Table 1. With respect to this table and considering (15), in fault-free operation (i.e., the third row in Table 1), there is no need to modify/compensate the nominal reference torque control signal 𝜏𝜏𝑟𝑟,𝑞𝑞 in (15). Then, the fault magnitude is assigned to be exactly zero in order to maintain the nominal performance of the system. Conversely, in the presence fault, the residuals show sensitivity and move out of the threshold range in (14). Then, the fault magnitude will be estimated using the absolute of relevant generated residuals. Accordingly, the obtained estimates of fault magnitude will be used directly in (15) for signal correction and fault accommodation. 6. SIMULATION RESULTS

Fig. 6. Details of an example module (Module #𝑃𝑃 in Fig. 5). As it is shown in Fig. 6, the subsequent process of decision making will be conducted online on the residuals and the results of threshold tests to detect the possible faulty turbine and then identify/estimate the power loss (fault magnitude) 𝑃𝑃̂ in it. In total, there are four possibilities with details presented in Table 1. The FDD decision made in this table including information about the faulty turbine and relevant fault magnitude will be used in the signal correction mechanism that is described in the next section.

No.

As already mentioned, the dynamic models used in the model-based FDD system are basically the same and developed using the SISO fuzzy modelling and identification technique described in Section 4. To train and evaluate the model, a set of 80,000 measured data for each of input and output were used. The data were obtained with a sampling rate of 80 Hz from the simulation of wind turbine under faultfree operation. It is worth mentioning that each set of the data was split into equal halves: one half for training and the other half for validation. The modelling accuracy of the identified fuzzy model can be demonstrated in terms of the Variance Accounted For (VAF) index (Babuska, 1998). The developed model has VAF of 98% that indicates satisfactory accuracy.

FDD Decision 𝑰𝑰𝒑𝒑

Consistent

Inconsistent

Faulty Turbine

1

Ti and Tj

Ti and Tk

Tk

Fault Magnitude 𝑃𝑃̂ |𝑟𝑟𝑖𝑖𝑖𝑖 (𝑘𝑘)|

2

Ti and Tk

Ti and Tj

Tj

|𝑟𝑟𝑖𝑖𝑖𝑖 (𝑘𝑘)|

3

Ti and Tj Ti and Tk ̶

̶

0

4 ̶

Ti and Tj Ti and Tk

Ti

|𝑟𝑟𝑖𝑖𝑖𝑖 (𝑘𝑘)| or |𝑟𝑟𝑖𝑖𝑖𝑖 (𝑘𝑘)|

This section presents the evaluation of the AFTC scheme via simulation tests performed in MATLAB/Simulink using the nonlinear benchmark model presented in Section 2. The fault scenario considered for evaluation of the AFTC scheme is modelled by the occurrence of 3% power loss in wind turbine T4 in the wind farm (see Fig. 1) within time period of [650700] sec. Simulations are conducted for a realistic wind field with mean speed of 12 m/s, a turbulence intensity of 10%, and over 1000 seconds of run time. 6.1 Dynamic Model Accuracy

Table 1. Decision making logic in Fig. 6. The FDD decision is sent in the form of 𝑰𝑰𝒑𝒑 for 𝒑𝒑th module Turbine Power Consistency

1373

5.2 ASC Mechanism

6.2 AFTC Scheme Performance

The purpose of this section is to provide an explanation of the ASC mechanism designed for the AFTC scheme shown in Fig. 4. Here, the original wind turbine controllers and wind farm controller all are kept unchanged, but the final control output in the faulty turbine is modified. As already mentioned, the FDD system not only detects the faulty turbine in the wind farm, but also estimates the fault magnitude. Therefore, the estimated fault magnitude can act upon a reference control signal in the faulty turbine to accommodate the fault effects. Since the torque controller is active in both partial-

As already mentioned in Section 5, the AFTC scheme monitors the consistency of generated powers from turbines in a farm through dividing them into groups of three, and then conducts the FDD in each group in real-time. In reference to the wind farm shown in Fig. 1 which includes 10 turbines, at least four groups of three turbines are required to cover the entire farm. Table 2 presents three example group configurations obtained by dividing the turbines into four groups of three while covering the entire wind farm. In this

1373



paper, the group configuration in the first row of the table is chosen for implementation of the AFTC scheme. Table 2. Three example group configurations for the wind farm shown in Fig. 1 No. 1 2 3

Group #1 T3,T4,T7 T4,T7,T10 T4,T7,T10

Group #2 T6,T9,T10 T7,T9,T10 T3,T6,T9

Group #3 T2,T5,T8 T3,T6,T8 T2,T3,T9

Group #4 T1,T2,T5 T1,T2,T5 T1,T5,T8

Fault Indicator [-]

Estimated Fault Magnitude [MW]

Based on the obtained simulation results, the AFTC scheme can effectively detect, identify and accommodate the considered fault in the faulty turbine (i.e., turbine T4). Fig. 7 shows the estimated fault magnitude as well as the fault indicator signal produced by the FDD system. 0.2 0.15 0.1 0.05 0 0

200

400 600 Time [Sec]

800

1000

800

1000

(a)

1 0.5 0 0

200

400 600 Time [Sec]

(b) Fig. 7. FDD results. (a) Estimated fault magnitude, and (b) Fault indicator signal. The generator power response is a well-suited performance index for investigating the overall performance and, in particular, the fault-tolerance property of the AFTC scheme. Fig. 8 shows the generator power response for the faulty turbine while the fault-free response with baseline controllers is used as a frame of reference for comparison. As it is observed in this figure, the autonomous structure of AFTC scheme which is basically due to its ASC mechanism does not affect the nominal performance of the baseline controllers under fault-free conditions. This feature can be favourable in terms of easier acceptance and Validation & Verification (V&V) by wind turbine industry. Fault-Free Operation with Baseline Controllers Faulty Operation with Baseline Controllers Faulty Operation with AFTC Scheme

Generator Power [MW]

5

5

Generator Power [MW]

4

3

4.5

4 3

4.5 4

2

4

1

3.5 0

2 -1 0

3.5

660 660

200

680 680

400 600 Time [Sec]

700 700

720

720 800

1000

1 Fig. 8. Generator power response during fault-free and faulty conditions for wind turbine T4 in Fig. 1. 0 7. CONCLUSIONS -1 This paper addresses the design and400development of a novel 0 100 200 300 500 600 700 Time [Sec] active fault-tolerant control (AFTC) scheme for an offshore wind farm against decreased power generation caused by turbine blade erosion and debris build-up fault. The proposed

scheme employs a model-based fault detection and diagnosis approach to provide accurate and timely diagnosis information to be used in an appropriate automatic signal correction algorithm. The effectiveness of the proposed scheme is evaluated by simulation on an advanced offshore wind farm benchmark model. Simulation results clearly indicate the effectiveness of the AFTC scheme. Extending the proposed approach in this paper to address more demanding fault situations including more than one faulty turbine at a time remains as one of future works. REFERENCES Babuska, R. (1998) Fuzzy Modeling for Control. Kluwer Academic Publishers, Springer. Badihi, H., Zhang, Y. M. & Hong, H. (2013) A Review on Application of Monitoring, Diagnosis, and Fault-Tolerant Control to Wind Turbines. Proc. of International Conference on Control and Fault-Tolerant Systems (SysTol'13). Nice, France. Badihi, H., Zhang, Y. M. & Hong, H. (2014a) An Active FaultTolerant Control Approach to Wind Turbine Torque Load Control against Actuator Faults. Proc. of AIAA SciTech 2014. Maryland, USA. Badihi, H., Zhang, Y. M. & Hong, H. (2014b) Fuzzy GainScheduled Active Fault-Tolerant Control of a Wind Turbine. Journal of the Franklin Institute, 351 (7), 3677–3706. Fingersh, L. & Carlin, P. (1998) Results from the NREL VariableSpeed Test Bed. Proc. of 17th ASME Wind Energy Symp. USA. Gustafson, D. E. & Kessel, W. C. (1978) Fuzzy Clustering With a Fuzzy Covariance Matrix. IEEE Conference on Decision and Control. San Diego, CA, USA. Jonkman, J., Butterfield, S., Musial, W. & Scott, G. (2009) Definition of a 5 MW Reference Wind Turbine for Offshore System Development. IN NREL/TP-500-38060 (Ed. Colorado, USA, , National Renewable Energy Laboratory. Kusiak, A. & Verma, A. (2011) A Data-Driven Approach for Monitoring Blade Pitch Faults in Wind Turbines. IEEE Transactions on Sustainable Energy, 2 (1), 87–96. Kusiak, A. & Verma, A. (2012) A Data-Mining Approach to Monitoring Wind Turbines. IEEE Transactions on Sustainable Energy, 3 (1), 150-157. Odgaard, P., Stoustrup, J. & Kinnaert, M. (2009a) Fault Tolerant Control of Wind Turbines - a Benchmark Model. the 7th IFAC Symposium on Fault Detection, Supervision and Safety of Technical Processes. Spain. Odgaard, P. F. & Johnson, K. E. (2013) Wind Turbine Fault Diagnosis and Fault Tolerant Control - an Enhanced Benchmark Challenge. Proc. of American Control Conference (ACC). Washington, DC, USA. Odgaard, P. F., Stoustrup, J., Nielsen, R. & Damgaard, C. (2009b) Observer Based Detection of Sensor Faults in Wind Turbines. European Wind Energy Conference. Pao, L. Y. & Johnson, K. E. (April, 2011) Control of Wind Turbines; Approaches, Challenges, and Rescent Developments. IEEE Control Systems Magazine. Simani, S., Farsoni, S. & Castaldi, P. (2014) Residual Generator Fuzzy Identification for Wind Farm Fault Diagnosis. Proc. of the 19th IFAC World Congress. Cape Town, South Africa. Sloth, C., Esbensen, T. & Stoustrup, J. (2010) Active and Passive Fault-Tolerant LPV Control of Wind Turbines. American Control Conference (ACC), 2010. Soltani, M., Knudsen, T. & Bak, T. (2009) Modeling and Simulation of Offshore Wind Farms for Farm Level Control. European 800 Offshore 900 1000 Conference and Exhibition. Stockholm, Sweden. Wind Tabatabaeipour, S. M., Odgaard, P. F., Bak, T. & Stoustrup, J. (2012) Fault Detection of Wind Turbines with Uncertain Parameters: A Set-Membership Approach. Energies, 5 (7), 2424-2448.

1374