A Simulation study on the Impact of Wake Deflection in Offshore Wind Farms

(1)

University of Groningen, Faculty of Economics and Business

Technology & Operations Management Master’s Thesis

A Simulation study on the Impact of Wake

Deflection in Offshore Wind Farms

Wouter Reuvers

Student number: 2686872

First Supervisor: Dr. E. Ursavas

Second Assessor: Dr. I. Bakir

Third Supervisor: Dr. M. Yildirim

January, 2020

(2)

1

Abstract

(3)

2

1. Introduction

The global share of wind energy in the total energy mix is increasing at a rapid pace. With a 37% growth of global installed capacity between 2015 and 2018, wind energy is able to cover close to 6% of global electricity demand (Wind Energy International Platform – Library, 2019). Offshore wind energy as a renewable energy source, offers tremendous opportunities in scalability and consistency. Because of this increased investment in offshore wind energy, ongoing research is conducted to optimize wind farm energy production and at the same time achieve cost reduction. This paper analyzes the effect of using optimized, wake deflection driven, wind farm control on power output, the turbine degradation probabilities, and thereby the operations and maintenance strategy of an offshore wind farm. The analysis is based on a simulation study testing the effectiveness of the wake deflection control method. Our research is relevant in the academic field of optimizing wind farm operations and maintenance, as we demonstrate the potential of wake deflection.

Offshore wind energy generation requires substantial high-capital expenditures and operational expenditures that are adding up to a higher a Levelized Cost of Energy (LCOE) than its onshore counterparts (Ashuri, Zaaijer, Martins, van Bussel, & van Kuik, 2014). Operations and Maintenance (O&M) costs for offshore wind farms are accountable for nearly 30% of total turbine life cycle costs (Fischer, Besnard, & Bertling, 2012). This greatly pressures the need for O&M cost reduction strategies. Integrated O&M planning for the entire offshore wind farm provides significant efficiencies between O&M costs and turbine downtime (Yildirim, Gebraeel, & Sun, 2017). As such, offshore wind turbines are grouped in designated wind farm areas. However, the clustering of wind turbines generates a new problem, called the wake effect. The wake effect, in short, is incoming airflow disturbance caused by upstream wind turbines (Boersma, Doekmeijer, Gebraad, Fleming, Annoni, Scholbrock, & van Wingerden, 2017). The disturbed wind both negatively affects wind generation and turbine fatigue on downstream turbines. This is relevant because many wind turbines therefore are not able to produce what they potentially could.

(5)

4 downstream turbines are enabled to overcompensate the loss of power generation from the upstream turbine (Gebraad, Thomas, Ning, Fleming, & Dykes, 2017). Thus, given a certain wind profile, wake deflection is expected to facilitate higher power generation. However, little research is done on the influence of wake deflection in more real-life conditions and wind farm setting. We test the effect of wake deflection on power output by changing the input parameters of wind speed, wind direction, and wind turbine spacing. Therefore, this research aims to gain insights into the effectiveness of wake deflection under multiple wind farm conditions.

In order to analyze the impact of wake deflection on power output, this study uses computer simulations to replicate wind farm scenarios. The simulations are performed using the open-source FLOw Redirectional and Induction in Steady-state (FLORIS) model. In FLORIS, the wake deflection effectiveness can be tested and validated for power optimization (Bay, King, Fleming, Mudafort, & Martínez-Tossas, 2019). The goal of the simulations is to calculate wind profile and wind turbine control impact power output. Subsequently, this effect on power generation is used as a basis for further analyses of wind turbine degradation.

Next to influencing power output, wake deflection has been found to influence wind turbine degradation. Namely, due to misaligned free-stream wind velocity on the rotor, the lower thrust force causes less turbine loading and thus makes it less prone to degradation (Fleming, Gebraad, Lee, & van Wingerden, 2015). However, research also found that the yaw misalignment, on the other hand, exerts aerodynamical forces on the turbine itself, affecting turbine fatigue (Hansen, 2015). Furthermore, evidence is found of the reduction of damage equivalent loading (DEL) on wind turbines in controlled wind farm simulations (Gebraad et al., 2016). DEL is a quantification of structural load on wind turbine fatigue. Although the authors provide causal outcomes on DEL signals, the wind farm used in the simulation is not a proper reflection of real-life wind farm conditions. We extend the simulation outcomes on DEL of Gebraad et al., (2016) with a more comprehensive simulation of wind farm conditions. In the simulations, the change in power output is used to predict DEL on wind turbine components based on sampled correlation data. To obtain valid results, we will group the wind turbines in the wind farm based on their relative position in the wind farm. Specifically, this study will analyze the relationships between four DEL signals and the power output of the different wind turbine classifications. These outcomes vary as we will provide arguments on why wind turbine classification has different implications for DEL prediction.

(6)

5 improvement possibilities with changing turbine lifetime expectations (Yildirim et al., 2017). A reduction in DEL influences the failure probability threshold and can thus suggest a decrease in maintenance intervals. For this reason, the potential lifetime increase in wind turbines components enables improvements to existing O&M strategies. We will use wind turbine degradation data to calculate the impact on the failure probabilities of the affected wind turbine components (Le & Andrews, 2016). In order to do so, the Mean Time To Failures (MTTF) are estimated based on sampled Weibull degradation distributions. Finally, by setting a realistic wind farm O&M scenario, the MTTF’s are translated to failure probabilities for the wind turbine. We will demonstrate that as a result of this, a wind farm that employs wake deflection control and thereby decreases failure probabilities, ensures a more profitable and effective O&M strategy.

To summarize, our first contribution to the scientific literature is by providing a detailed analysis of the influence of wake deflection on wind farm power output by simulating multiple wind farm conditions. This understanding of wake deflection extends the current literature by using historic weather data together with real-life wind turbine spacing in the simulation study. We show that wake deflection offers tremendous potential in significantly increasing total wind farm power output. This means that wind farm operators worldwide should further investigate the feasibility of using optimized wake deflection control in their daily operations. Moreover, this study contributes to the literature on wind turbine fatigue by estimating the influence of wake deflection on wind turbine degradation. Finally, the implications of wake deflection influence on O&M can contribute to extending existing academic O&M models.

In the following chapter, we will provide an overview of and critically reflect on the existing literature. Specifically, antecedent studies concerning wind energy, wake effect, wake deflection, wind turbine degradation, and maintenance are discussed. Next, the methodology used in the study is outlined. First, the conceptual model used in this research is explained. Secondly, the research objectives for this study are highlighted. After this, a description of the models and software used is provided, together with the simulation approach. Consequently, the outcomes of the simulations are analyzed and discussed. We will conclude the paper by discussing the results and limitations of this study.

2. Theoretical Background

(7)

6

2.1. Wind energy

The capacity of offshore wind energy is skyrocketing as global installed capacity is planned to be doubled by 2023 (IEA, 2019). However, O&M costs for offshore wind farms can be multiple times higher than wind farms on land (Levitt, Kempton, Smith, Musial, & Firestone, 2011). This raises the question of what counterbalances these disadvantages to explain the high investments in offshore wind energy. Firstly, many countries have little to no land available for large scale onshore wind farm projects. Even when there is space, onshore wind farms receive a lot of societal resistance due to noise emission and visual impact (Angelakoglou, Botsaris, & Gaidajis, 2014). Although people encourage wind energy, the Not In My BackYard principle takes the upper hand (Krohn & Damborg, 1999). This means, for example, that people are positive towards placing wind turbines, as long as they don’t place it in their neighborhood. Next to more space and fewer protests, the wind on the sea is much more suitable for wind energy generation. One reason for this is that the average wind speeds on the sea are significantly higher than wind speeds on land (Esteban, Diez, López, & Negro, 2011). This allows for a higher effectivity of power generation per installed turbine. Secondly, offshore wind is more stable and less fluctuant than onshore wind. This leads to fewer harmful turbulence effects. The absence of these allows that the offshore turbine can be mounted closer to the surface than its onshore counterpart (Esteban et al., 2011). This makes the turbine more compact and stable.

2.2. Wind turbines

(8)

(9)

8 Figure 2: Simplified power curve of a 3 Mega Watt (MW) HAWT. Source: (Uit het Broek, Veldman,

Fazi, & Greijdanus, 2019)

The power output of a HAWT depends on the thrust force that can be extracted from the wind. The physics behind it lies in momentum theory and conservation of energy theory. For a more detailed scientific and technical overview of wind energy extraction, the reader is referred to Njiri & Söffker, 2016, and Porté-agel, Bastankhah, & Shamsoddin, 2019.

2.3. Wake effect

Offshore wind turbines are often grouped together in wind farms for economic reasons. However, this comes at a cost. The wind turbines in the rear are affected by the deprivation and destabilization of the wind by the upstream wind turbines. This turmoil is referred to as the wake effect. This section explains what the wake effect is and how it affects wind farms.

(10)

9 The wake effect affects incoming airflow of downstream turbines when the turbines are close enough placed to one another and when the given downstream turbine(s) is aligned in wind direction with the upstream turbine. A helicopter photo of the wake effect is provided in Figure 3. Historically seen, wind turbines are often placed around 7 RD downwind from each other in the most frequent wind direction (Wu & Porté-Agel, 2011). According to a study of the Energy Research Centre Netherlands (ECN), the optimal commercial value of wind farm power density [MW/km2], is between 7 and 8 RD for very large wind turbines (Bulder, Bedon, & Bot, 2018).

Because of this commercially driven placement, the wind turbines will affect each other, while the wind has not restored to its original state within the 7 RD. A combination of depleted kinetic energy and increased turbulence will constrain downstream turbines of maximum power extraction. The power loss due to the wake effect is reportedly around 12% for offshore wind farms (Barthelmie, Frandsen, Hansen, Schepers, Rados, Schlez, & Neckelmann, 2009). This emphasizes the need for reducing this significant power loss due to wind turbine wakes.

(11)

10

2.4. Wake effect modeling, simulation, and control

Since wind turbines are placed within the wake range of each other due to spatial and the priorly discussed economic considerations, it is essential to understand the influence of wake effect on wind power generation. Developing and testing theoretical wake effect models is an ongoing research subject because it enables better power prediction and therefore benefits planning and budgeting.

In this study, simulations will be used to calculate wake deflection influence on specific wind farm situations. For this, the FLORIS model has been used as a simulation tool. The FLORIS model was created by a collaboration of the National Renewable Energy Laboratory (NREL) in the US and the Technical University Delft in the Netherlands (Delft, Data, & Control, 2018). The FLORIS model is one of the newest wake modeling tools available and has the advantages that it combines multiple validated wake models into definitive results (Annoni, Bay, Taylor, Pao, Fleming, & Johnson, 2018; Bay, King, Fleming, Mudafort, & Martínez-Tossas, 2019). Also, the FLORIS model has kept relatively low fidelity and is therefore low in computational costs. Because the FLORIS simulation package is based on this combination of wake effect models, the differences and reliability of these are elaborated on in this section.

A simple but effective model was provided by Jensen, (1983). It captures the linear influence of the thrust coefficient on the far-wake area as a cone shape. The model is limited because it does not include turbulence as a factor. Yet, due to its simplicity, low computational costs and effective results the (updated) model is still used by commercial operators (Rathmann, Hansen, Hansen, Mortensen, & Murcia Leon, 2018). The Jensen model has been replaced by the Multi-Zone model which is able to model overlapping wakes in yawed positions (Gebraad & Van Wingerden, 2014). Yet, still this model fails to incorporate the turbulence intensity.

A breakthrough in wake effect modeling was presented by the Gaussian model (Bastankhah & Porté-Agel, 2014). It describes the velocity deficit based on the Navier-Stokes equation, which describes the motion of incompressible fluids, and added turbulence on the wind turbine’s wake. This is of great importance as it is able to conserve momentum, and therefore better replicates wake effect physics (Annoni, Fleming, et al., 2018).

(12)

11

2.5. Wake deflection

Wake deflection is a control method for upstream wind turbines, that makes sure that the downstream wind turbines are less affected by the wake effect caused by the upstream wind turbines. This is done by misaligning the upstream turbine’s yaw into a calculated optimal degree relative to the incoming wind. The optimal degree is calculated by a wind farm optimization solver that maximizes total power output. Figure 4 shows how the wake is deflected by misaligning the first turbine. The figure shows that by misaligning the upstream turbine, its wake is deflected partially from the downwind turbine. This control mechanism is a promising tool in reducing the wake effect impact on downstream wind turbines (Fleming et al., 2014). However, maximizing total power for the entire farm does mean that upstream individual wind turbine power generation must be decreased. The wind farm yaw angle control therefore shifts from, so-called, greedy to optimized control. In the next paragraphs, the relevant wake deflection models used in FLORIS are discussed.

When the efficiency gains by downstream turbines exceed the loss in efficiency of the upstream turbine, wake deflection is considered to be beneficial. Several studies validated the effectiveness of wake deflection both theoretically and empirically (Bastankhah & Porté-Agel, 2016; Fleming et al., 2017, 2019). Simulation studies showed that the annual energy production can be increased by up to 4% (Gebraad et al., 2017).

Figure 4: An illustration of wake deflection (Zalkind & Pao, 2016).

(13)

12 study, a few other control techniques and models exist (Boersma et al., 2017). Since FLORIS uses the most promising methods, only these will be included in the current research.

2.6. Wind turbine degradation

Recent research suggests that wake deflection control can also have a positive effect on preventing wind turbine degradation, which could decrease downtime and costs for wind farms. This section focuses on wind turbine degradation and how it is affected by wind deflection control.

Wind turbines at sea often operate in harsh conditions as wind and waves have a free flow. In addition, the turbine is a rotating device with dozens of sensitive parts and materials, all with a delicate fatigue. This combination makes wind turbine components prone to early failure. Fatigue is a complex result of the turbine’s material being under pressure of mechanical loading. Fatigue basically equals the lifetime of the turbine until it breaks down or is expected to breakdown under certain degradation thresholds (del Campo, Ragni, Micallef, Diez, & Simão Ferreira, 2015). In order to quantify fatigue the measure of DEL was introduced, it relates load to degradation of the turbine.

This paper focusses on aerodynamical loading caused by incoming wind flow. Here interesting dynamics come into play when we introduce the wake effect. Generally, load on wind turbines increases with thrust force until a certain threshold. Thus, when less wind flow enters the turbine it experiences less load. However, it has been suggested that when wind speed is low, the wake effect can even relatively increase the load on the turbine due to turbulence (Lee, Churchfield, Moriarty, Jonkman, & Michalakes, 2012). To make the situation even more complex, purposely misaligning the yaw of the wind turbine causes temporal increased load due to aerodynamical periodical forces on a wind turbine (Hansen, 2015). To conclude, there are three forces that determine the change in DEL caused by yaw misalignment in wake deflection control.

(14)

13 critical wind turbine parts decreases significantly with the wake deflection control strategy (Gebraad et al., 2016).

The effects on power generation and turbine degradation are highly relevant in optimizing wind farm control. The degradation results have in turn a significant impact on optimizing O&M strategies and planning. In the next section, the implications for the O&M strategy are discussed under the assumption that a wind farm adopts the wake deflection control strategy to optimize power production.

2.7. Operations & Maintenance

This paragraph aims to provide a context for offshore wind farm O&M. Offshore wind farms are significantly more costly compared to onshore wind farms, and therefore cost reduction is an even more important strategy towards profitability. Firstly, the installation costs are higher due to far more difficult construction factors (Fischer, Besnard, & Bertling, 2012). Specialized machinery and technology are needed to install a wind turbine in harsh conditions. After installation, the farm needs to be connected to the grid, which is more expensive due to the length, technology, and again difficult construction conditions. Apart from the fact that performing maintenance to an offshore wind turbine is more expensive, the need for maintenance is also greater. The higher wind speeds and turbulence propagation lead to faster degradation of wind turbine parts. Besides, maintenance is done via ship or helicopter, and the costs of these operations are tremendously high. Also, broken turbines need to wait longer for repair due to slow maintenance reaction times and the grouping of maintenance crew deployment, as there are high set-up costs for a maintenance expedition (Yildirim et al., 2017). Le & Andrews (2016) provide a wind turbine degradation model that is used to simulate degradation and associated maintenance. The authors base the failure probabilities of specific wind turbine components on estimated Weibull degradation distributions. In turn, the aforementioned study translates the degradation states in a wind turbine lifetime to corresponding maintenance actions. With this information, they were able to estimate a detailed cost model for the offshore wind farm. This research will use the existing degradation and maintenance model to extend current O&M literature (Le & Andrews, 2016).

3. Methodology

(15)

14 determine choices and expectations for O&M strategies and costs. This methodology section starts with discussing the conceptual model based on the literature. We continue by stating the research objectives formed based on the conceptual model. After, we provide the simulation set up to create a relevant context for the tests formulated.

3.1. Conceptual model

In this section, the conceptual model used to structure this study is explained. In Figure 5, the conceptual model of this research is provided to clarify the methodology of this research. The conceptual model suggests positive or negative relationships between the variables based on existing literature.

Figure 5: The conceptual model of research relationships.

We will test the intermediary proposed relationships between the concepts presented in Figure 5. In addition to testing the positive or negative influence, the intensity of the inferred causation is also tested. This paper presents four research objectives arising from the conceptual model. In the following sections, we will provide our research objectives and the underlying reasoning for these research objectives. Moreover, we will discuss the method used for testing our research objectives.

3.1.1 Power output wind farm

As described before, evidence of the possibility of power optimization through yaw deflection control already exists. Yet, most of the simulation studies use the SOWFA software in obtaining the results, which is accurate but too computationally expensive to use in active control settings. This research aims to validate the preceding evidence in this matter by making use of FLORIS. Besides, this study will perform extensive simulations to gain specific insights concerning power optimization.

(16)

15 This research objective will be tested with FLORIS in different wind farm scenarios. Furthermore, real wind data is used to test the impact of wind speed and wind direction on the 3x3 wind farm. The weather data is described in 3.2.2. Wake deflection is considered ‘effective’ when it enables an increase in the total power production of a wind farm.

3.1.2. Damage equivalent load output wind farm

Preliminary studies predict an overall decrease in DEL by using wake deflection through simulations in SOWFA/FAST (Fleming et al., 2017; Gebraad et al., 2016). This study extends the current literature by investigating the impact of wake deflection on DEL for the wind turbines in a more realistic simulation setting.

Research objective 2: analyze the impact of wind farm optimized controlled wake deflection strategy on the average DEL for wind turbines.

The DEL is approximated for four load signals:

1. Blade out-of-plane bending moments. These moments affect the degradation of the rotor, specifically the hub and the blades.

2. Drivetrain low-speed shaft (LSS) torsion. This DEL affects the degradation of the drivetrain shaft.

3. Yaw bearing moments. These moments affect the degradation of the yaw, the bearing and the gears in particular.

4. Tower bending moments. The DEL accompanied by these moments affect the tower of the structure.

Correlations for every DEL signal are calculated by averaging results from Gebraad et al. (2016). Their findings can be found in the Appendix. The correlations are generalized to model a relationship between power change and the change in DEL. Next to that, the influence of the wind turbine position in the wind farm is used to differentiate between the results. These linear correlations are linked with FLORIS power output for the 3x3 wind farm example to gain commercially interesting insights.

3.1.3. Turbine degradation and failure probability

For wind turbine degradation it is assumed that the varying mechanical stress due to the variability of wind speed measured in DEL, is fully responsible for the wind turbine component degradation. Therefore, a change in DEL will directly influence the lifetime and failure distribution of the specific wind turbine component.

(17)

16 The wind turbine component failure distribution data will be sampled from Le & Andrews (2016) to calculate Mean Time To Failure (MTTF). The authors provide a comprehensive wind turbine degradation and maintenance model. The expected MTTF is based on preliminary estimated Weibull degradation distribution parameters. With the results on DEL changes, a new MTTF can be calculated, as the DEL proportionally influences the MTTF. Finally, real degradation data from rotating machinery is used to predict failure probabilities based on the expected lifetime (Yildirim et al., 2017). To estimate the failure probabilities at a given age of the wind turbine, a 20-year scenario for the example 3x3 wind farm setting is constructed.

3.1.4. Cost analysis example wind farm

A reduction in failure probabilities is hypothesized to reduce the frequency of maintenance need and thereby reducing maintenance and downtime costs. Next to that, the increased power output will lead to higher wind farm efficiency. The combined extra revenue and cost-saving naturally account for an increase in the profit model.

Research objective 4: to calculate the effect of the decrease in failure probabilities due to wake deflection strategy on expected maintenance costs. Together with the calculation of the impact on overall power output, we will analyze the change in profitability for the wind farm.

The expected failure probabilities for the wind turbine components in the given time period will be multiplied by the specific maintenance costs. The estimated relative gains between the failure probabilities are quantified in the business model. With it, implications of decreased failure probabilities are discussed as a contribution to O&M strategies. Although the change in profit for the scenario can be estimated, the exact result on LCOE requires much more data and is therefore requires further research.

3.2. Simulation set up

(18)

17

3.2.1. Base simulation

In the simulation study of Gebraad et al. (2016), both the SOWFA and the FLORIS model are used to predict power output in their simulation set up. The authors show that the FLORIS model predicts the wake effect and its impact on power output accurately enough to substitute the SOWFA model in daily wind farm control. As this paper builds on the obtained DEL results in the aforementioned study, the relation between the power optimization and its effect on degradation needs to be estimated. In order to do so, the same simulation set up as Gebraad et al. (2016) is used as a basis for further analysis.

The base scenario is a six wind turbine farm consisting of two rows of three NREL 5-MW wind turbines, which the standard wind turbine used in simulation studies (Jonkman, Butterfield, Musial, & Scott, 2009). The 5 MW wind turbines have an RD of 126 meters. The simulation uses a fixed mean wind velocity of 8 meters per second (m/s) and a 6% turbulence intensity. Wind shear, wind veer, and air density are kept constant. The wind turbines are placed 5 RD spacing in the downwind direction and 3 rotor diameters in the crosswind direction. The wind turbines are allowed a 40-degree positive offset relative to the wind direction. Thus, a negative offset is not allowed as previous studies show that a negative offset greatly increases blade loads (Fleming et al., 2014).

The base simulation consists of two scenario’s that are compared in total power output. In the first scenario, the farm is operated in the conventional way of aligning the turbines with the wind direction. This setting is described as ‘greedy’. The second scenario is the ‘optimized’ one, in which total power output is maximized by calculating the individual yaw configurations. The model therefore acts like a solver to maximize total power by changing the yaw angle decision variables. In the latter scenario, the wake deflection strategy can be observed, as upwind turbines sacrifice individual power output for the benefit of the downstream turbine’s output. When the wind direction is aligned with the downwind plant configuration the total power gain was 13% (Gebraad et al., 2016). Admittedly, the base scenario is not very economical in terms of layout. This is because it assumes a constant 270 degrees wind direction and with 270 degrees aligned turbines the downwind turbines will suffer from the full wake effect. However, this scenario does facilitate chances of wake deflection effectiveness.

(19)

18

3.2.2. Wind farm simulation setting and data

To extend the theoretic results based on Gebraad et al. (2016) this paper enhances the simulation to a more commercially realistic one. We create a larger wind farm with symmetrical 3 times 3 wind turbines. The argument to increase the amount of wind turbines is to better demonstrate the effect of the wakes on the different layers of the wind farm. Mainly because of changing wind direction there will always be a ‘middle’ row. Next to that, the spacing between the wind turbines is set to the commercial distances of 7 RD downwind and 5 RD crosswind, for the reference wind direction of 270 degrees.

In addition to a more realistic farm layout, the wind conditions are also simulated to mimic real-life conditions. For the simulation, real-life weather data is used to reproduce the input wind speed and wind direction. The data is retrieved from data.knmi.nl, the Dutch weather monitoring organization. Specifically, the data of the planned offshore wind farm ‘Borsele 1’ from 1st January 2004 until the 31st of December 2004 is used as input data (KNMI DataCentrum, 2019). The Borsele 1 weather sensors kept track of hourly weather data in the Dutch North Sea to support wind farm construction decisions. The time period used is randomly chosen from the dataset and is regarded as a good reflection of offshore wind. The wind speed and wind direction at 80 meters height and 100 meters height is averaged to estimate wind at the 5 MW turbine hub height of 90 meters above sea level. A descriptive visual overview of the data used can be found in Appendix 2.

3.2.3. Turbine position classification

The turbines in the first row in the simulation setting are the stereotypical front turbines. To clarify, these are the first turbines that are reached by the wind from a wind direction perpendicular to the square wind farm layout. The front turbines are not affected by any wake effect. When the plant operates with greedy control these turbines generate the most power because they receive undisturbed wind. Meanwhile, the front turbines have the most impact on the rest of the wind plant because their wake effect is potentially the strongest and may reach most other wind turbines. In corner positions in the wind farm at specific wind directions, the front turbine may not affect any other turbine at all in the greedy position.

The turbines between the first and the last row are the middle turbines. These turbines are affected by some turbine’s wake effect. At the same time, the middle turbine also affects other middle turbines or back turbines. With greedy wind farm control, the middle turbine just behind a front turbine will experience the most wake effect and therefore be the least efficient in the wind farm.

(20)

19 greedy position, as it is not necessary to deflect wakes for downstream turbines. Therefore, in optimized control, the rear turbines will experience the highest power gains of the wind plant.

4. Results & Analysis

In the results and analysis section, the findings of this paper's research objectives are presented. The report is structured to follow the research objective chronology. First, the antecedent theory of wake deflection impact on power output is tested in FLORIS. Secondly, the simulations are extended to obtain specific insights on the influence of the simulation input variables. Subsequently, the effectiveness of wake deflection, in terms of increasing power output, is assessed. After that, multiple yearly simulations are performed to replicate real-life wind farm output. We continue by analyzing the impact of wake deflection on DEL for the wind turbines. In addition, the relationship between power change and DEL change is provided. Thereafter, this research analyzes the impact of the DEL change on wind turbine degradation and failure probabilities. Finally, the consequences of the alteration in failure probabilities and its impact on maintenance are calculated in a hypothetical wind farm setting.

4.1. Simulations of wake deflection in FLORIS

In this section, the preceding simulation scenario from Gebraad et al. (2016) is replicated to test FLORIS model validity. The simulation set up is expanded and multiple simulations are created to test the consequences of different wind farm layouts on wake deflection effectiveness.

(21)

20

Table 1: Simulation results for various wind farm configuration switching to optimal control.

Simulation: A B C D E

Wind plant 3x2 3x2 3x3 3x3 8x10

Spacing 5D/3D 5D/3D 7D/5D 20D/5D 7D/5D

Yaw angle turbine 1 25.85 26.73 24.87 2.84E-06 25.72

Yaw angle turbine 2 39.8 32.74 24.74 1.81E-21 21

Yaw angle turbine 3 0.35 0.04 0 0 18.06

Yaw angle turbine 6 0.45 0.07 0 0 18.11

Greedy MWh 6.68 5.26 9.19 12.91 75.35

Optimized MWh 7.55 5.91 9.94 12.91 78.74

Total power gain 13.0% 12.4% 8.2% 2.3E-06% 4.5%

The power output simulation is performed in FLORIS using a similar configuration as the base scenario (simulation A). With a total power gain of 12.4%, the replicated simulation B shows comparable results with simulation A, and therefore the new simulation output can substitute the antecedent results. These new results indicate that the correlations discovered by Gebraad et al. (2016) can be generalized.

In simulation C the wind farm is expanded to 9 turbines in a 3x3 configuration. In addition, the spacing between the wind turbines is set to realistic commercial distances. It can be observed in Table 1 that this further spacing reduces the effectiveness of wake deflection compared to simulation B. This paper will use the 3x3 configuration as the working model for further analysis as it better represents a (section of) real-life wind farm and the dynamics that come in play due to the wake effect.

The RD for the wind farm was stretched until wake deflection was marginally effective in simulation D. It was tested that up until 20RD downwind distance, wake deflection was still slightly effective for power generation as can be seen in Table 1, with a very little power gain achieved. From 21RD onwards, the wake effect’s influence is not worth deflecting it. Apparently, at this point the gain in power generation for the downwind turbines does no longer offset the loss of power generation in the front turbines.

(22)

21 80 wind turbines, there are smaller relative gains compared to smaller rectangular wind farms since the proportion of rear turbines in the farm is lower. Visualization of simulation D and E can be found in the Appendix.

Figure 6: Wind turbines, numbered, and wake effect in the greedy scenario for simulation C.

Figure 7: Wind farm and wake effect in the optimized scenario for simulation C.

(23)

22

Table 2: Individual turbine details simulation C

Simulation C Greedy Yaw Angle Opt. Yaw Angle Greedy Power MW Power Portion greedy Optimized Power MW Power Portion opt. Power Change Turbine 1 0 24.87 1.69 18.4% 1.41 14.2% -16.7% Turbine 2 0 24.75 0.62 6.7% 0.84 8.5% 37.1% Turbine 3 0 0 0.76 8.3% 1.06 10.7% 40.4% Turbine 4 0 24.87 1.69 18.4% 1.41 14.2% -16.7% Turbine 5 0 24.75 0.62 6.7% 0.84 8.5% 37.1% Turbine 6 0 0 0.76 8.3% 1.06 10.7% 40.4% Turbine 7 0 24.87 1.69 18.4% 1.41 14.2% -16.7% Turbine 8 0 24.75 0.62 6.7% 0.84 8.5% 37.1% Turbine 9 0 0 0.76 8.3% 1.06 10.7% 40.4% Total - - 9.19 100.0% 9.94 100.0% 8.2%

(24)

23

4.1.1. Wake deflection effectiveness

Previous studies have shown that wake deflection is most effective at lower wind speeds as the wakes have the most relative influence on incoming wind for downstream turbines (Boersma et al., 2017). In this section the 3x3 wind farm is simulated for increasing wind speed in steps of 0.5 m/s. In the first simulation, the wind direction is kept constant on 270 degrees. Figure 9 is a plot obtained by running the mentioned simulation with increasing wind speed intervals.

Figure 9: Wake deflection effectiveness at different wind speeds.

As illustrated in Figure 9, wake deflection starts being relevant from 5.5 m/s. From there it starts with the peak effectiveness of 14.4% higher power output for the wind farm. The wake deflection effectiveness decreases until 14.5 m/s. From 3 m/s to 5 m/s wind speed the wind turbine does produce power, but weaker wakes at that stadium make wake deflection ineffective. A reasonable explanation is that at lower wind speeds the amount of kinetic energy to be absorbed from the wind is low enough to make it recover faster after an upstream turbine depletes the wind. This means that the upstream turbines are better off taking full advantage of the incoming wind, as downstream turbines will not experience a significant decrease in wake incurred from upstream wake deflection. For wind speed levels above 14.5 m/s wake deflection becomes ineffective at a wind direction of 270 degrees. An explanation for this result is that at these wind speeds the rear turbines almost hit their 5 MW maximum power output in greedy configuration already. Therefore, there is simply no potential to increase power for the total wind farm in an optimized setting.

The next simulation set up is inverted from the previous one since the wind speed is kept constant at 8 m/s and the wind direction is simulated with steps of 1 degree. It is assumed that the wind turbines

(25)

24 can rotate 360 degrees. Figure 10 shows a histogram displaying the average relative power gain obtained per grouped wind directions.

Figure 10: Average power gain for wind direction bins

As illustrated in Figure 10, there are large differences for wake deflection effectiveness in the wind rose. Wake deflection is most effective just a little over the quarterly marks. With a peak at 2 and 182 degrees with a 20.7% power gain. The main explanation for the fluctuation is the constraint that negative yaws are not allowed, explained in section 3.2.1. Besides, at certain wind directions the wakes created do not influence downstream turbines to justify wake deflection due to farm layout. The series is repeated at the 180 degrees mark due to the mirrored configuration of wind turbines.

It is found that the highest effectiveness for wake deflection is reached at a wind speed of 5.78 m/s at a direction of 182.2 degrees. For this scenario, the mean power gain is 28.7%. Details of the simulation results provided in this section can be found in Appendix 4.

4.1.2. Simulation analysis 3x3 wind farm

To investigate the dynamic influence of wind speed and wind direction on the wake deflection effectiveness, this research conducts multiple yearly simulations for the 3x3 wind farm. This section provides an analysis of the power output for the 3x3 wind farm with a year of simulated weather data. Each simulation adds uncertainty in wind speed and wind direction to reproduce real-life conditions. For all simulations the other relevant input variables are kept constant. The results are presented in Table 3. The table provides details concerning the simulation run and the associated turbine results obtained in FLORIS. This section continues by providing insights gained from the various simulations.

0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 0-29 30-59 60-89 90-119 120-149 150-179 180-209 210-239 240-269 270-299 300-329 330-360 % P o w er G ain

(26)

25 Subsequently, the thresholds for wake deflection effectiveness are identified. Next to that, this section provides clarifying wind rose diagrams to visualize simulation inputs.

Table 3: Simulation results of the 3x3 wind farm with a year of weather data.

Simulation: F G H I

Average wind speed in m/s 8.0 8.0 9.0 9.0

Average1_{wind direction in degrees} ₂₇₀ _253.2 ₂₇₀ _253.2

Total power produced in GW greedy 80.68 120.9 149.57 180.83 Total power produced in GW optimized 87.31 122.67 154.1 182.04 The average power gain for turbine 1 -16.7% 1.2% -8.7% 1.0% The average power gain for turbine 2 37.1% 3.3% 18.9% 2.1% The average power gain for turbine 3 40.4% 2.4% 23.0% 1.4% The average power gain for turbine 4 -16.7% 2.4% -8.7% 1.7% The average power gain for turbine 5 37.1% 5.1% 18.9% 2.9% The average power gain for turbine 6 40.4% 4.8% 23.0% 2.6% The average power gain for turbine 7 -16.7% 1.3% -8.7% 0.9% The average power gain for turbine 8 37.1% 4.1% 18.9% 2.3% The average power gain for turbine 9 40.4% 3.0% 23.0% 1.6% The average power gain for the wind farm 8.2% 1.8% 4.4% 1.1%

Simulation F is an extended simulation of the hourly results of simulation C for the 3x3 wind farm before. The relative results for the simulated year remain stable, as the input variables stay the same and the wind farm performs constantly. Note that the power production is relatively low due to the high wake effect in this simulation set up. On the other hand, this allows for effective wake deflection as the average power gain for the wind farm is 8.2%.

Simulation G runs with a constant wind speed of 8 m/s, but this time the hourly wind direction is sampled from the weather data set. It can be observed that the average power gain is lower than that of simulation F, and also more equally distributed. There are two reasons for this outcome. First, the wind direction changes 360 degrees in this simulation and therefore the turbine classification is leveled. This can be seen in the left wind rose of Figure 11. The wind roses show the relative wind

1_{Taking the mean of average wind direction would yield false results. Therefore the average is calculated by a}

(27)

26 direction and wind speed for a specific simulation. Secondly, as was depicted in Figure 10, due to negative yaw alignment not being allowed, for certain wind directions wake deflection is not effective.

Figure 11: The wind roses generated by simulations G and H respectively. The frequency of the wind speed, the colored bars, are displayed in the wind roses.

Simulation H runs the other way around, with the constant wind direction of 270 degrees but now the hourly wind speed is sampled from the weather data set. This distinction from simulation G can be observed in Figure 11 as the difference in the wind roses. In the year used for simulation, there were 775 hours of wind speeds below 3 m/s, which is the cut-in speed threshold for this type of wind turbine. Moreover, at the constant wind direction of 270 degrees, at wind speeds below 5.42 m/s wake deflection was not considered effective. This threshold signals the tipping of the tradeoff between maximization of own power production and reducing the wake for the downstream turbines. Apparently, below 5.42 m/s the marginal gain for upstream turbines to behave greedy is higher than gain for the downstream turbines would get in case the wake effect would be reduced. There are a total of 1485 hours for which the wind speed is between 3 and 5.42 m/s. It can be observed that there also is an upper bound, as for wind speeds higher than 14.16 m/s wake deflection has become ineffective in increasing total power output. There are 1411 hours of wind speeds higher than 14.16 m/s. All in all, it can be computed that wake deflection was initiated for 5113 hours of the year, which translates into 58.2% of the time.

(28)

27 for which wake deflection was enabled. The distribution of the wind speed and wind direction is visualized in Figure 12. Hypothetically, when only assessing the hours of active wake deflection, the average power gain can be computed as 1.1% divided by 36%, which is 3.1%. Therefore, this section can be concluded with the insight that wake deflection is expected to be effective in a limited number of hours within a year. Wake deflection is effective in increasing total wind farm power output only when both wind speed and wind direction allow for positive yaw misalignment. Yet, when both conditions are met, wake deflection proves to improve power output significantly.

Figure 12: wind rose generated by simulation I.

4.2. Damage equivalent loading

(29)

28 specific correlations between power change per turbine position and the DEL of every specific load signal.

This paper focusses on four important load signals that influence the degree of degradation for essential wind turbine components. The load signals described in 3.2.2 and are the blade out-of-plane bending moments, drivetrain low-speed shaft (LSS) torsion, yaw bearing moments and tower bending moments.

The four load signals identified are crucial for monitoring the wind turbine degradation. However, there are various more fatigue-related aspects involved with wind turbines, but there is insufficient data for these loadings in order to provide a valid analysis.

Table 4: Change in Damage Equivalent Load from switching wind plant control from greedy to optimized based in simulation A.

Turbines ∆ Power ∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. ∆ DEL

Front Turbine Avg. -12.7% -13.35% -5.25% -15.78% -2.65% -9.26%

Middle Turbine Avg. 3.2% -14.20% -31.10% -9.78% -17.05% -18.03% Rear Turbine Avg. 85.3% -7.35% -4.95% -1.05% -28.73% -10.52%

Average all 25.3% -11.63% -13.77% -8.87% -16.14% -12.60%

DEL prediction is an unadvanced research topic and therefore this research paper will use statistics based on simulation data (Gebraad et al., 2016). Their results are summarized in Table 4 and their original data can be found in the Appendix. The load signals were estimated in the FAST simulation module by estimating signal newton meter and vibrations.

(30)

29 turbine misaligns its yaw with the wind direction, the wind will unevenly cause load on the turbine, since the wind hits the turbine asynchronously. Also, with decreased wake effect, the turbulence accompanied by it also decreases, which results in a more stable wind inflow for the downwind turbines.

Table 5: Change in DEL from switching wind plant control from greedy to optimized related to the change in power output of the individual turbine.

The statistics in Table 5 represent averages for every signal based on the turbine’s position and its change in power output. It can be observed that for every x% change in power output the average DEL of the particular component in a wind turbine changes with the corresponding relative value. All correlations given in Table 5 are visualized in the graphs of Figure 13. For example, when a front turbine experiences a 10% decrease in power output due to misaligning its yaw, the average DEL on the blade out-of-plane bending moments (signal 1) decreases with 11.8%. Most correlations are positive, as can be seen in Table 5, meaning that an increase in power yields an increase in DEL. However, for the rear turbines and DEL 4 for the middle turbines, there is a negative relationship between power change and DEL change. This suggests that even though the turbine increases power output, its DEL is decreasing.

The power output on itself does not explain the difference in DEL, there are various other factors such as turbulence and vertical wind shear that have a major influence on the fatigue of the wind turbine (Kragh & Hansen, 2014). As can be observed in Table 5 with the finding that for rear turbines a 1% increase in power translates in an average decrease in DEL of 0.26%. Therefore, power change from greedy to optimized control is used as a predictor of DEL change instead of plain power output.

Turbines ∆ Power ∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. ∆ DEL

Front Turbine Avg. 1% 1,18 0,35 1,39 0,13 0,76

Middle Turbine Avg. 1% 0,34 0,39 0,36 -0,10 0,25

Rear Turbine Avg. 1% -0,24 -0,21 -0,09 -0,49 -0,26

(31)

30 70,00 80,00 90,00 100,00 110,00 120,00 130,00 80 85 90 95 100 105 110 115 120 DE L ∆ In d exe d

Turbine Power ∆ Indexed

Front Turbines

∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. DEL

90,00 95,00 100,00 105,00 110,00 80 85 90 95 100 105 110 115 120 DE L ∆ In d exe d

Middle Turbines

∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. DEL

85,00 90,00 95,00 100,00 105,00 110,00 115,00 80 85 90 95 100 105 110 115 120 DE L ∆ In d exe d

Rear Turbines

(32)

31 Figure 13: DEL relative change in relation to the power output of the front turbine.

The graphs of Figure 13 visualize the correlations between the change in turbine power and the effect on change in DEL. The graphs show nicely that the correlations are linear when using index figures. We now know that correlations between power change due to wake deflection and the associated DEL signals make it possible to calculate DEL output for different simulation settings.

4.3. DEL linked to simulation results

The wind farm simulations in FLORIS predicted relative power gain for switching from greedy to optimized control as was seen in sections 4.1. and 4.2. In the last section, the relationship between relative power gain and DEL was established. Here, the simulation results on power output are extended with the associated DEL change investigated. Based on the turbine classification, the DEL change relative to the greedy load can be determined for each signal for the 3x3 wind farm example. At first, the results of simulation F, the simulation run with constant wind speed and wind direction, is extended with the expected structural load change. After that, in the simulations where the wind turbines in the 3x3 wind farm experience a changing wind direction, their position classifications are calculated according to the proportional wind directions faced. With this data available, it is possible to link DEL change to power change for simulation I, the simulation with both changing wind speed and wind direction. Subsequently, this section ends with discussing the computed relative DEL changes for the real-life simulation.

In simulation F the wind farm is aligned with the constant wind direction, and therefore the wind turbines per row deliver the same output. For this reason, the turbines are grouped together in Table 6.

Table 6: Results of switching to an optimal control yaw misalignment strategy for simulation F.

Simulation F ∆ Power ∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. DEL Front turbines 1/4/7 -16.7% -19.7% -5.8% -23.1% -2.2% -12.7% Middle turbines 2/5/8 37.1% 12.6% 14.4% 13.3% -3.6% 9.2% Rear turbines 3/6/9 40.4% -9.7% -8.6% -3.8% -19.9% -10.5%

Average turbine 15.6% -9.4% -1.7% -7.7% -8.5% -6.6%

(33)

32 turbines increases as well. Yet, the loading only increased by a third compared to the power increase. As expected from the negative correlations found in section 3.1., although the hindmost turbines yield the highest power increase in the farm, the DEL for all signals decreases. Apparently also the tower bending moments, DEL 4, in the middle turbines deliver similar negative results.

As was shown in Figure 12 with the wind rose of simulation I, the wind direction was not constant nor evenly distributed. Therefore, the wind turbines do not match with a position classification and thus it is not possible to link their power output change with DEL change. As we recall from Table 5, the correlation between power output and DEL heavily depends on the turbine position classification. Therefore, based on the wind direction data used as input for the simulation I, the turbines are reclassified by the frequency of their position within the wind farm. This is done by analyzing the wind direction data and computing the relative position for the wind direction bin average. Appendix 5 shows the details of this test. With these results, it is possible to compute the frequency of occurrence of these bins and therefore also determine how many hours of the year a certain wind turbine fits one of the three positions.

Table 7: Individual turbines and frequency of turbine classification with dynamic wind direction

Turbine Front Middle Rear

1 66.8% 0.0% 33.2% 2 42.5% 25.2% 32.3% 3 55.1% 0.0% 44.9% 4 48.0% 35.8% 16.2% 5 0.0% 100.0% 0.0% 6 16.2% 35.8% 48.0% 7 73.9% 0.0% 26.1% 8 32.3% 25.2% 42.5% 9 43.3% 0.0% 56.7% Average 42.0% 24.7% 33.3%

(34)

33

Table 8: Results of switching to an optimal control yaw misalignment strategy for simulation I.

Simulation I ∆ Power ∆ DEL 1 ∆ DEL 2 ∆ DEL 3 ∆ DEL 4 Avg. DEL

Turbine 1 0.8% 0.6% 0.1% 0.7% -0.1% 0.3% Turbine 2 2.0% 1.0% 0.4% 1.3% -0.3% 0.6% Turbine 3 1.4% 0.8% 0.1% 1.0% -0.2% 0.4% Turbine 4 1.6% 1.0% 0.4% 1.2% -0.1% 0.7% Turbine 5 3.1% 1.1% 1.2% 1.1% -0.3% 0.8% Turbine 6 3.0% 0.6% 0.3% 0.9% -0.7% 0.3% Turbine 7 0.9% 0.7% 0.2% 0.9% 0.0% 0.4% Turbine 8 2.6% 0.9% 0.3% 1.3% -0.5% 0.5% Turbine 9 1.9% 0.7% 0.1% 1.0% -0.4% 0.3% Average 1.1% 0.5% 0.2% 0.6% -0.2% 0.3%

Table 8 is a replication of Table 6 but now with the input data of simulation I. As shown in Table 8, the average DEL increases for signals 1, 2 and 3 for all turbines. This can be explained by the relatively higher average power gain. Although the turbines produce more MW, the DEL for signal 4 remains the same or decreases slightly. Overall, the yearly differences are small, however, when wake deflection would be effective more frequently throughout the year, the results become very relevant.

4.4. Wind turbine degradation

In this section, the impact of DEL change on wind turbine degradation and failure probabilities are analyzed. The section begins with providing degradation outcomes found in previous studies for the specific wind turbine components. Secondly, an important assumption linking DEL on component degradation is given. Next to that, the states of degradation are discussed together with their corresponding maintenance action.

(35)

34 Figure 14: Wind turbine component condition graph.

We assumed that the decrease of DEL for the four load signals equals the decrease in failure probabilities with the associated wind turbine components. Therefore, we expect that the corresponding components are fully affected by DEL change. The affected components are listed in Table 9. Next to that, the classification of the maintenance associated is provided. The classification has an influence on the cost, the vessel needed and the duration of the maintenance action. The details of the maintenance classification can be found in Appendix 6.

Table 9: Influenced turbine components by DEL 1-4.

Load Turbine

part Component

Condition

states Maintenance description Classification

1 Rotor Hub C/F Replacement Type 1

1 Rotor Hub D Minor corrosion repair Type 4

1 Rotor Blade C/F Replacement Type 1

1 Rotor Blade D Minor repair Type 5

2 Drivetrain Shaft F Shaft replacement Type 1

2 Drivetrain Shaft D/C Minor repair, alignment Type 4

3 Yaw Bearing F Complete replacement Type 1

3 Yaw Bearing C Gear tooth repair Type 3

3 Yaw Bearing D Corrective repair Type 4

4 Structure Tower C/F Replacement Type 1

4 Structure Tower D Corrosion repair Type 5

(36)

35 maintenance actions to restore it to as good as new condition, depending on the condition state of the component.

4.4.1. Failure probabilities in a wind farm scenario

In order to arrive at failure probabilities, we must set a time period for which the probability will be calculated. In order to come up with real-life results, a wind farm scenario is constructed. Secondly, this section extends the theory on wind turbine degradation by calculating the MTTF of the wind turbine components. Thereafter, we will create a link between the wind turbine component MTTF and the failure probabilities of these components.

It is assumed that the 3x3 wind farm operates in normal conditions for 10 years with greedy control. The wind farm operator would want to switch to optimized control for the coming 10 years based on the results of simulation F. Simulation F is used to be able to group the turbines in position classification. Therefore, the DEL figures in Table 6 are used. In order to budget and plan the operations and maintenance strategy for the next 10 years the failure probabilities of the wind turbine components at age 20 are calculated and compared.

The failure probabilities are based on the shape parameter β, and scale parameter η of the Weibull distribution function. With these parameters, the MTTF is calculated as 𝑀𝑇𝑇𝐹 = 𝛼𝛤(1 + 1/𝛽).

(37)

36

Table 10: Failure probabilities greedy and optimized control

Table 10 lists the investigated wind turbine component conditions with their Weibull degradation distributions and its corresponding MTTF. The optimized DEL is the wind farm turbine averaged DEL change per load signal from Table 6. In Table 10, it can be observed that a change in DEL alters the MTTF for the specific component conditions. To determine the failure distributions of the components failing in the next 10 years starting from age 10, the MTTF is used to calculate relative age at age 20 in comparison to its expected lifetime. This calculation is used to determine the position in the matrix, from where the failure probability needs to be retrieved. Table 10 shows that a change in DEL, through optimal control, has a significant impact on the failure probabilities of the wind turbine components compared to greedy control.

4.4.2. Impact on operations and maintenance costs

This section extends the results on failure probabilities by linking it to O&M costs and planning. Based on the findings that wake deflection causes differences in failure probabilities the saving on maintenance costs can be estimated. Next to that, the analysis of the financial impact of wake deflection is completed by estimating the additional revenues achieved with higher power production.

The relevant cost specifications used in this analysis are sampled from Le & Andrews (2016) in order to gain concrete insights on the implications for predicted expenditures. The costs were estimated in pounds and are converted to euro’s in this research paper with an exchange rate of 1:1.2.

Load Component Condition Type Weibull β Weibull η MTTF greedy DEL opt. MTTF opt. FP greedy FP opt. 1 Hub Degraded 4 1.2 15.38 14.5 91.6% 15.8 0.9163 0.8795 1 Hub Failure 1 1.2 17.3 16.3 91.6% 17.8 0.8823 0.8371 1 Blade Degraded 5 1.2 23.02 21.7 91.6% 23.6 0.0812 0.0470 1 Blade Failure 1 1.2 25.9 24.4 91.6% 26.6 0.0014 0.0005

2 Shaft Degraded 4 1.2 160 150.5 98.3% 153.2 6E-44 6E-45

2 Shaft Failure 1 1.5 180 162.5 98.3% 165.4 3E-51 2E-52

3 Bearing Degraded 4 1.2 29.12 27.4 92.3% 29.7 0.318 0.245 3 Bearing Critical 1 1.2 32.76 30.8 92.3% 33.4 0.008 0.004

3 Bearing Failure 1 1.2 36.4 34.2 92.3% 37.1 0.002 0.001

4 Tower Degraded 5 1.2 14.93 14.0 91.5% 15.3 0.979 0.968

(38)

37 Assuming 8 m/s mean wind speed the average power output for the 3x3 wind farm is 1.02 MWh with greedy wind turbine control and 1.1 MWh for the optimal yaw-controlled wind turbines. The average price per MWh is assumed to be 50 euro (Electricity market report, 2019). In the calculation of the opportunity costs caused by the downtime, the average data for the type of maintenance operations is used. The waiting time for a crew for the right weather is left out of the model.

The total amount of repair site visits in the simulation of Le & Andrews (2016) was 72.17. With a total expected number of repairs of 76.59, this implies that most crew visits are for a single repair only. In this study, therefore, it is assumed that for every repair the typical rate of the vessels is paid.

Table 11: Maintenance cost saving expectation (Exp) for the next 10 years per wind turbine.

Table 11 shows the expected maintenance need for specific wind turbine components for a single wind turbine. The maintenance costs are based on standard rates found in Appendix 6. The expected costs are the costs for a certain maintenance action, times the failure probability of this component’s condition. Since the tower is already past its MTTF, the expected maintenance costs are very high as the failure probability is close to 1. The total expected maintenance costs are reduced by €16,000 when compared to the greedy scenario for the next 10 years per wind turbine. However, these cost savings can rise exponentially when the scenario would be extended, and the components are closer to their expected lifetime. Besides, only a small portion of the wind turbine components are assessed. Component Maintenance Cost of

repair (€) Repair rate (€) Downtime cost (€) FP greedy Exp. Cost greedy (€) DEL opt. FP opt. Exp. Cost opt. (€)

Hub Minor repair 3,600 6,000 270 0.9163 9,044 91.6% 0.8795 8,680

Hub Replacement 52,800 84,000 22,719 0.8823 140,737 91.6% 0.8371 133,535 Blade Minor repair 4,800 8,400 270 0.0812 1,093 91.6% 0.0470 633 Blade Replacement 240,000 84,000 22,719 0.0014 494 91.6% 0.0005 163

Shaft Minor repair 6,000 6,000 270 0.0000 0 98.2% 0.0000 0

Shaft Shaft replacement 44,400 84,000 45,354 0.0000 0 98.2% 0.0000 0 Bearing Corrective repair 6,000 6,000 270 0.3184 3,906 92.2% 0.2448 3,003 Bearing Gear tooth repair 8,400 18,000 900 0.0080 218 92.2% 0.0036 100 Bearing Replacement 10,800 84,000 45,354 0.0018 253 92.2% 0.0006 89 Tower Corrosion repair 24,000 8,400 270 0.9793 31,993 91.7% 0.9679 31,623 Tower Replacement 316,800 84,000 22,719 0.0317 13,406 91.7% 0.0164 6,953

(39)

38 With optimized control and wake deflection there can be a yearly power increase of 6630 MWh compared to greedy control, as was the result of simulation F. For 10 years of this average production, this would mean an additional revenue of €3,315,000. Together with the €16,000 times 9 turbines is 144,000 of saved costs, the expected profitability of this wind farm rises with €3,459,000 for the 3x3 wind farm in its 10-year prognosis. When the wake deflection strategy would be tested in larger wind farms the profitability would add up significantly.

5. Discussion

This paper aimed to answer the research question: what is the influence of wake deflection on operations and maintenance in offshore wind farms? This was done by investigating intermediary steps between wind turbine operation and control and its impact on maintenance and profitability. This chapter first summarizes the findings of this study in the conclusion. Secondly, the contributions to the scientific research field are elaborated on. After that, some limitations of this study are provided. Finally, partially build on the limitations, suggestions for future research are discussed.

5.1. Conclusion

The relative global dependence on offshore wind energy rises on a yearly basis. With these large planned investments, operators are in search of constant improvement in wind turbine effectiveness and cost savings. The promising control method to battle wake effect disturbance, and therefore increase effectiveness and decrease expected maintenance costs, is analyzed in this research paper. Overall it can be concluded that wake deflection in combination with farm optimized control can provide significant benefits for the wind farm operator. For analyzing the specific dynamics leading to this observation this paper contributes to the scientific research body.

In summary, the findings on the effect of wake deflection on wind farm power output are that, given the assumption that the wind farm experiences wake effect nuisance, wake deflection enables a few percentage point gains in total power output. Therefore, research objective 1 can be concluded as: farm optimized wake deflection control increases the total power output of the wind farm. The increase in power due to this control method can rise to 28.7% for the 3x3 wind farm. However, it must be noted that the effectiveness of wake deflection differs hugely for the combination of wind farm inputs.

(40)

39 Where downwind turbines benefit from wake deflection, the front turbine always sacrifices a portion of its capacity. The second insight is that wake deflection remains effective for wind farms spaced much further downwind. Furthermore, when changing the wind speed input variable, it can be concluded that wake deflection is effective between a wind speed of 5.42 m/s and 14.16 m/s. With the wind speed constant and altering the wind direction variable it was observed that there is a major disparity in wake deflection effectiveness for different wind directions. This is mainly explained by farm layout and the constraint of non-negative yawing.

The 3x3 wind farm simulated with a year of real wind data lets us observe that the average power gain is still significant with a 1.1% total power gain. This means that in realistic conditions wake deflection boosts power production in a wind farm. The average increase in power is however lower than in controlled simulations. This is explained by the results that only 36% of the hour's wake deflection was effective with the right combination of wind speed and wind direction.

Next to power production, this paper investigated the correlation between power gain and DEL change for four load signals. It can be concluded that, through different influences of the turbine position on structural load change, the DEL experiences different correlations for every signal in combination with the turbine classification. As expected, the average DEL decreases overall in the 3x3 wind farm example. Hence, an answer on research objective 2 will yield that wind farm optimized controlled wake deflection strategy decreases the average DEL on wind turbines. When zooming in it is found that front turbines experience a relatively higher decrease in DEL compared to decreasing in power. The middle turbines experience an increase in DEL while relatively increasing more in power. Perhaps most interesting, the rear turbines experience a decrease in DEL while increasing in power.

The implications for DEL reduction are translated to wind turbine degradation. Research objective 3 can be concluded by; a farm optimal controlled wake deflection strategy decreases wind turbine degradation and therefore reduces component failure probability. The paper suggests that DEL reduction for specific signals reduces the degradation distribution of the corresponding wind turbine components. It can be concluded that the wake deflection strategy reduces the MTTF for all selected wind turbine components and therefore extends the turbine component lifetime.

A Simulation study on the Impact of Wake Deflection in Offshore Wind Farms

University of Groningen, Faculty of Economics and Business

Technology & Operations Management Master’s Thesis