• Nie Znaleziono Wyników

The development of maintenance strategies of offshore wind farm

N/A
N/A
Protected

Academic year: 2021

Share "The development of maintenance strategies of offshore wind farm"

Copied!
72
0
0

Pełen tekst

(1)

The

development

of

maintenance strategies

of offshore wind farm

Tim Jonker

Technische Univ ersi tei t Del ft

(2)
(3)

The development of maintenance

strategies of offshore wind farm

by

Tim Jonker

Literature assignment ME54010

(4)
(5)

Preface

This report contains a literature study as part of the Master Mechanical Engineering with the track Transport Engineering and Logistics. Within this track the report topic is related to theme three: Real-time Coordination for Operational Logistics. In this report maintenance strategies for offshore wind farms are addressed. This is important because maintenance is a major expense of offshore wind energy. More specifically, it is investigated if through smart sensing technologies incipient failures could be detected before happening which allow for more advanced maintenance scheduling.

Tim Jonker Delft, September 2017

(6)
(7)

Contents

1 Introduction 1

1.1 Offshore wind turbines . . . 2

2 Maintenance strategies 5 2.1 Costs. . . 5 2.2 Maintenance strategies . . . 6 2.2.1 Corrective maintenance. . . 6 2.2.2 Preventive maintenance. . . 7 2.2.3 Predictive maintenance . . . 7

2.2.4 Other maintenance strategies . . . 9

3 Failure and reliability 11 4 Wind turbine components and common failures 15 4.1 Wind turbine components . . . 15

4.1.1 Rotor blades. . . 15 4.1.2 Hub. . . 16 4.1.3 Pitch system. . . 16 4.1.4 Yaw system . . . 17 4.1.5 Nacelle . . . 17 4.1.6 Tower. . . 17 4.1.7 Main shaft. . . 17 4.1.8 Gearbox . . . 18 4.1.9 Generator . . . 18 4.1.10Mechanical brakes. . . 18 4.1.11Electrical system . . . 18 4.1.12Control system . . . 19 4.1.13Sensors . . . 19 4.1.14Overview of failures . . . 19 5 Conditional monitoring 21 5.1 Conditional monitoring based on SCADA information. . . 21

5.1.1 Signal trending . . . 23

5.1.2 Power signal analysis . . . 26

5.2 Conditional monitoring based on vibrations. . . 27

5.2.1 Gearbox . . . 28

5.2.2 Bearings . . . 29

5.2.3 Generator . . . 29

5.2.4 Other components . . . 29

5.3 Other Condition Monitoring techniques . . . 30

5.3.1 Oil analysis . . . 30

5.3.2 Temperature modelling . . . 30

5.3.3 Electrical effects . . . 31

5.3.4 Visual inspection. . . 31

5.4 Structural health monitoring. . . 31

5.4.1 The four stages of SHM . . . 31

5.4.2 SHM techniques . . . 33

5.4.3 SHM in support structures. . . 35

5.5 Cost analysis . . . 36

(8)

vi Contents 6 Maintenance models 39 6.1 Maintenance architecture. . . 39 6.2 Fleet . . . 40 6.3 Maintenance logistics . . . 40 6.3.1 Strategic decisions. . . 41 6.3.2 Tactical decisions. . . 41 6.3.3 Operational issues. . . 41 6.3.4 Improvements. . . 42 6.4 Simulation models. . . 42 6.4.1 NOWIcob. . . 42 6.4.2 Dalgic’s Model. . . 44 6.5 Scheduling models. . . 46 6.5.1 SIMAP . . . 46 6.5.2 Scheduling tool. . . 47 7 Conclusion 49 A Available CM sensors 53

B Correlations of SCADA parameters for CM purposes 55

C Flowchart for SHM 57

D SHM techniques 59

(9)

1

Introduction

Because of finite fossil fuel reserves and stronger regulations regarding green house gas emissions, wind energy poses a suitable alternative. On top of that, the global energy demand increases rapidly, requiring a greater energy production [1]. Over the years wind turbines have increased significantly in size and power output (see Figure 1.1). The downside of these larger and greater number of wind turbines is that there is no space on land to place them. Furthermore, residents complain about landscape pollution and noise. Therefore, more and more wind turbines are placed offshore. Another advantage of offshore wind turbines is that on average, there is slightly more wind due to different heating rates of land and water. This means that more power can be generated offshore, compared to onshore, with an equal sized wind turbine [2]. Unfortunately, also the shores become more populated with wind farms. Therefore, wind farms are also placed further offshore each year, this can be seen in Figure1.2.

However, offshore wind turbines are harder to place and access when maintenance needs to be carried out. Maintenance is one of the main contributors to the high price of offshore wind turbines [3]. Therefore, much researches is devoted to explore better maintenance strategies. One promising strategy is predictive maintenance which through monitoring of components predicts failures and allows scheduling of preventive maintenance operations. Therefore, the aim of this literature research is to investigate the state of the art monitoring techniques, different maintenance strategies and investigate the option to integrate monitoring with a scheduling agent.

(10)

2 1.Introduction

Figure 1.1: In the last few years wind turbine size has increased significantly, resulting in a greater power output [2].

Figure 1.2: In the last few years wind turbines are placed further offshore [4].

1.1.

Offshore wind turbines

Most large scale offshore wind farms are located in Northern Europe, more specifically the North sea and the Baltic Sea. One of the reasons for this could be that land is scarce for the countries surrounding these waters. But, these countries are still compelled to meet strong regulations regarding greenhouse gas emissions. By 2020, 20% of the energy in the EU should rely on renewable energy [5]. In the absence of height differences for hydro power and poor weather conditions for solar energy, these countries are driven to offshore wind farms. On the other hand, also Canada, the United States and countries in eastern Asia like China start to develop wind farms [6].

(11)

1.1.Offshore wind turbines 3

there is the foundation; in shallow waters (<30 meters) a monopile structure is used, in transitional waters (between 20 and 80 meters) a stronger bottom founded structure is used and beyond that, floating wind turbines become financially more attractive [7].

Besides foundation however, the accessibility is also greatly influenced by the weather conditions and inherently the depth. Close to the shore, wind turbines are served by shore-based work boats. Further offshore, helicopters come to assist these work boats. However, in deep waters the travel time all the way back to shore becomes too long and therefore an offshore base or mothership is used [8].

Although a single land based wind turbine is completely common technology, offshore wind farms are a different story. One of the reasons is the large Operation and Maintenance (O&M) cost. Cur-rently, the wind turbine itself only accounts for one-third of the costs for the offshore project, the huge

O&M costs take up 30% [9]. This is also why wind turbines increase so rapidly in size, turbines are not critical for the total costs [6,10]. Offshore wind turbines are more expensive compared to onshore wind turbines; higher installation costs, more expensive foundation, more frequent maintenance and more expensive maintenance. The maintenance of offshore wind turbines is more frequent because of high humidity, ice and salt water exposure, which promotes quicker corrosion. Both these factors also explain why it is more expensive and dangerous to conduct maintenance offshore. On the other hand, offshore wind turbines deliver more power per turbine [7].

With the literature assignment different questions are answered:

• Which different maintenance strategies are available for offshore wind farms?

• What should be done to implement predictive maintenance?

• What is the current state of the art regarding monitoring techniques?

• Could a model be developed that contains predictive maintenance?

Answers and insights to these question will be provided during the report in different chapters. This report is structured in the following way: first different maintenance strategies will be discussed in Chapter2, in Chapter3a failure and reliability will be defined, Chapter4contains data of wind turbine components and their failure modes, Chapter 5 describes the state-of-the-art conditional monitoring techniques, finally in Chapter 6 different models are shown to simulate and schedule maintenance operations.

(12)
(13)

2

Maintenance strategies

In this Chapter first the definition of maintenance is given afterwards the maintenance costs are put into perspective. The Chapter finishes with a summary of different available maintenance strategies.

All components will eventually fail and therefore maintenance is a necessary part of operation. The engineering definition of maintenance is:

”Actions necessary for retaining or restoring a piece of equipment, machine, or system to the spec-ified operable condition to achieve its maximum useful life”[11].

These actions can include lubrication, welding, fastening etc. But also, replacement of certain components to preserve functionality of the overall system (wind turbine). Maintenance operations can prolong the life of a system but failure can never be completely eliminated [12]. There are different strategies concerning maintenance, all of them have certain advantages and disadvantages.

2.1.

Costs

Before different maintenance strategies can be compared, a comparing criterion is required. As is almost everything in the world the aim of a wind farm is to make money. Therefore, high revenue and low costs are beneficial. An accepted method to compare costs of different energy harvesting methods is the Levelized Cost Of Energy(LCOE) shown in Equation2.1. As can be seen in Figure2.1the cost of offshore wind energy is rather high compared to fossil energy sources but also compared to nuclear energy and onshore wind energy [13].

𝐿𝐶𝑂𝐸 = 𝑇𝑜𝑡𝑎𝑙 𝑐𝑜𝑠𝑡𝑠 𝑜𝑣𝑒𝑟 𝑙𝑖𝑓𝑒𝑡𝑖𝑚𝑒

𝐸𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑝𝑟𝑜𝑑𝑢𝑐𝑒𝑑 𝑜𝑣𝑒𝑟 𝑙𝑖𝑓𝑒𝑡𝑖𝑚𝑒 (2.1)

(14)

6 2.Maintenance strategies

Figure 2.1: TheLCOEof main different energy sources in 2013 [13]

The feed-inn tariffs for generated energy are country dependent and are still subsidized. Germany pays a subsidized price of 190€/MWh [14], this means that margin is small. Oil and gas industry is mainly driven by maintenance of production as margins are large, wind industry is more driven by the price per kW (LCOE) because margins are small. Therefore, there is more scheduling and planning involved to keep the price of maintenance low compared to the oil and gas industry. The consequence is that it may be beneficial to conduct maintenance later or earlier than absolutely necessary. Main-tenance costs are constructed of: equipment costs, spare part costs, transport costs, crew cost and lost production costs. The time of lost production is most greatly influenced by the weather conditions followed by the unavoidable repair time. This is because during bad weather you need to wait before maintenance can be conducted [6]. The time available, based upon the weather, for maintenance is called the weather window. What makes the weather window so crucial, is that for certain opera-tions you require proper weather condiopera-tions for several consecutive days. Most standard maintenance operations cannot be conducted when waves exceed 1.5m or wind speeds are greater than 12 m/s [8]. This is because crew cannot safely work under these circumstances. As can be seen in Table2.1, waiting time for a weather window is worst in winter, unfortunately wind turbines also require the most maintenance during winter.

Table 2.1: Weather window probability, , and average waiting time, structured by season [6] Season Pw Tw (day) Winter 0,3 60 Autumn 0,5 30 Spring 0,6 10 Summer 0,8 3

2.2.

Maintenance strategies

There exist several different maintenance strategies which are based upon different principles. The three main strategies followed by some lesser known maintenance strategies are elaborated below, each with their advantages and applicability to offshore wind farms.

2.2.1.

Corrective maintenance

The first maintenance strategy is corrective maintenance, this strategy is characterized by conducting maintenance only when a specific component has failed. Therefore, this method is considered emer-gency based, and is never scheduled [15]. The advantage is that there is a rare chance of unnecessary maintenance trips and the number of maintenance trips is kept to a minimum. Furthermore, the im-plementation of corrective maintenance is easy. The disadvantages are however: the longer down times, potential higher maintenance costs, the requirement for having spare parts always available, crew & transport should always be available and potential damage to secondary components. These

(15)

2.2.Maintenance strategies 7

disadvantages make this maintenance strategy very expensive [16,17]. The availability of spare parts is a major contributor to these high costs as large inventories are required or expensive fast delivery contracts with suppliers. The alternative does not seem any better and is associated with long down-times. Moreover, corrective maintenance is always combined with lost production time because the turbine does not function after breakdown. This means that the wind turbine does not generate any power during scheduling, waiting for weather windows and repairing. The waiting time could be very long, especially during winter, increasing production losses [17].

Concerning wind turbines, corrective maintenance is usually adapted once the component will not affect: revenue, customer dissatisfaction or safety and health impact [18]. In other words, if the component is redundant in the complete system or has only a limited impact on the performance, corrective maintenance is appropriate [12].

2.2.2.

Preventive maintenance

Preventive maintenance is a time based maintenance strategy, it is characterized by planning and scheduling. Maintenance operations are scheduled based on available weather windows and crew availability but more importantly time based. The decision making, about which wind turbines specifi-cally will be addressed on a certain day may be based upon time since last check, visible deterioration or degradation. Most of the time, the executed jobs are of a preventive nature [15]. The advantage of preventive maintenance is that only occasionally a component completely malfunctions. Therefore, downtime and maintenance costs are much lower compared to corrective maintenance and this tech-nique is considered low risk. Furthermore, downtimes are shorter compared to corrective maintenance and damage to secondary equipment can be avoided. Preventive maintenance is considered three times less expensive compared to corrective maintenance for regular power plants [17]. However, in case of malfunction, it is impossible to predict which component requires attention. This could result in not having spare parts or appropriate tools immediately available. Furthermore, because many trips are only of a preventive nature this maintenance strategy is expensive and labor intensive [19].

Preventive maintenance is usually applied to original components during the warranty period. This is for two reasons: suppliers often specify a maintenance interval for a valid warranty and replacement is free of charge because of warranty. Furthermore, preventive maintenance is utilized once failure rates are known and failure intervals can be predicted accurately. However, this is also the downside of preventive maintenance: determining the maintenance interval. An interval too short; increases oper-ational costs, wastes production time and results in unnecessary replacements of components in good condition. Whereas if the interval is too long, unexpected failures could occur between maintenance operations. Therefore, preventive maintenance alone, would only be sufficient once failure intervals are known precisely. Otherwise, resources are spend on unnecessary maintenance trips or delayed repairs [18]. Commonly, the maintenance interval is based upon experience of similar components [17]. For wind turbines, which operate in variable conditions, time is not a trusted indicator of failure. This is because during one year the turbine may have been exposed to much higher loads than another year. Therefore, sometimes preventive maintenance is scheduled after a certain amount of maximum operational load hours [17].

2.2.3.

Predictive maintenance

The final maintenance strategy is predictive maintenance, this strategy is condition based. The main-tenance strategy is often also referred to as condition based mainmain-tenance or intelligent mainmain-tenance. Predictive maintenance means that components are being remotely monitored and once irregularities in the data arise, an opportunity is sought to send out a maintenance crew. Ideally, this opportunity arises before complete malfunction. Therefore, this strategy is condition-driven preventive mainte-nance but could also be considered as an early stage corrective maintemainte-nance strategy [15]. Predictive maintenance originated in the aerospace industry and relies on the principle ”if it is not broken, do not fix it” [20]. Predictive maintenance is a philosophy that uses actual operating conditions to op-timize operations. Through sensor technology, the longest interval between preventive maintenance operations is sought while avoiding corrective maintenance in the most cost efficient way [17]. The advantages are: there are no unnecessary trips, the nature of the problem is known beforehand, no damage to other components and down times are short. Because the problem is assessed beforehand proper spare parts and appropriate tools can be brought to the wind turbine. Besides that, down times are kept to an absolute minimum because repair is scheduled before complete malfunction. If used

(16)

8 2.Maintenance strategies

correctly, predictive maintenance is the most cost efficient maintenance strategy for critical compo-nents [12,16,17,20,21].

The disadvantage of predictive maintenance, is the need for sensor technology and decision making which leads to high investment costs and complex systems. The decision to schedule a maintenance operation could either be manual or automated. Eventually, an automated decision making algorithm would lead to more intelligence and therefore lowerO&Mcosts. This is because there are many factors that influence this decision, impossible to comprehend by the human brain. Some of the factors are: data trends, threshold values, prioritizing individual wind turbines, weather windows, spare part -cost and -availability, crew availability and routing. Furthermore, predictive maintenance is susceptible to false alarms. [21]. Smart sensing could help operators to notice incipient failures. Globally, a smart sensing system will acquire data directly from the wind turbine and process this information. Based upon the processed information, operators or programs decide whether maintenance is required. The advantage of automatic processing is that it is less labor intensive and requires lesser skills of the de-cision maker. Smart sensing will allow technicians to monitor more wind turbines simultaneously. The more automated the system becomes the more wind turbines can be supervised by a single employee. It is important to notice that smart sensing demands an online communication between the monitoring station and the wind turbine. With the developments of better wireless technologies this becomes easier. However, wireless technology should be combined with energy harvesting and energy efficient sensors. [22].

Different maintenance strategies are visualized in Figure2.2. Note that the degradation pattern and the scheduled maintenance interval are only for illustrative purposes. Most of the time, the degradation process is not linear and especially not known beforehand. Furthermore, there is no such thing as the single best maintenance strategy. Different maintenance strategies should be applied to different components to be cost effective. This principle is visualized in Figure 2.3[22]. However, in general predictive maintenance is considered the cheapest option, investment costs aside [21].

(17)

2.2.Maintenance strategies 9

Figure 2.3: Costs associated with different maintenance strategies [22]

2.2.4.

Other maintenance strategies

Besides the three common maintenance strategies above there are several lesser known maintenance strategies. Not necessarily these maintenance strategies are worse. However, they receive less atten-tion and only a few researches see them as viable alternatives.

Random maintenance

Firstly, random maintenance is an opportunity based strategy and therefore maintenance is only con-ducted once the opportunity arises. The decision to maintain a component may or may not depend on the condition of the component [15]. The advantage of this maintenance strategy is that it is able to cope with weather windows because these could be seen as opportunities. The disadvantage however, is that you are unable to tell beforehand if a component does in fact require maintenance. Therefore, unnecessary trips to wind turbines are made for inspection only. The main difference with preventive maintenance is that there is no schedule involved, specific wind turbines are addressed based upon sampling. Random maintenance is not necessarily better or worse compared to preventive mainte-nance, this is dependent on the failure curve. This curve is shown and explained later in this report

3.1. One of the believers in the power of random maintenance is Toshio Nakagawa who has written a dedicated book [24].

Opportunistic maintenance

The second type of alternative maintenance is described in [25]. Opportunistic maintenance is a variety on preventive- or corrective maintenance but is slightly different. The core of the strategy is that upon failure maintenance is carried out on that component, so far corrective. But for all other components in the wind turbine based on the age a certain kind of preventive maintenance is performed. [25] proposes multiple age groups, if a component is located in an old age group preventive replacement is performed. If the component is still young, preventive measures are executed e.g. greasing. The essence is that by preventive measures the age of a component could by reduced with a couple of years which would allow the component to never get older. According to the writers, opportunistic maintenance is a viable alternative in case of a facility with many failures or low available maintenance time as the result of weather for instance. Furthermore, the investment costs are lower because no extra sensor technology is required. During the research a mathematical optimum of six different age

(18)

10 2.Maintenance strategies

groups was found for a 50 turbine offshore wind farm. However, this number is greatly dependent on wind farm specifics [25].

Total Productive Maintenance

Total Productive Maintenance(TPM) is a maintenance strategy that originated in Japan and is centered around increasing the effectiveness of equipment. It was designed in the 1950s and finds its application mostly in production plants where operators work with the machines. The five pillars of the maintenance strategy are [17]:

1. Improve equipment effectiveness find which of the six big losses are the cause to low effec-tiveness. The six big losses are:

(a) Equipment breakdown

(b) Setup and adjustment slowdown (c) Idling and short-term stoppages (d) Reduced capacity

(e) Quality-related losses (f) Startup/restart losses

2. Involve operators in daily maintenance which means that machine operators are involved in the maintenance plan.

3. Improve maintenance efficiency and effectiveness concerns adequate detection, schedul-ing, spare part inventories, maintenance crew availability, etc.

4. Educate and train personnel perhaps the most important pillar. Everyone who is even remotely involved with the machine should be taught how to operate it, handle it with care and signal defects. This helps to prolong the lifetime of the machines.

5. Designing and managing equipment for maintenance prevention all machines should be designed in a way that reduces the chances of breakdowns. And in case of breakdown compo-nents should be easily replaced. Furthermore, preventive maintenance tasks such as lubrication should be easy accessible.

These pillars suggest thatTPMis more of a philosophy to improve maintenance and not directly appli-cable as a maintenance strategy. However, this does not mean thatTPMshould be forgotten, it offers useful insights when implementing other maintenance strategies [18].

Reliability Centered Maintenance

Another maintenance strategy which has been applied to wind turbines isReliability Centered Mainte-nance(RCM).RCMrelies on statistical data to determine when components will fail. Which could be considered an advanced form of preventive maintenance which is solely time based. Furthermore, by having insight in the system failure modes the critically of components could be assessed together with the effect on system performance [17]. The effects of a component are assessed on: safety for per-sonnel, environmental impact, production availability and material loss. This has led to fail-safe design where several functions within a system could be performed by multiple components. Which caused a major change in the safety-reliability perception. AlthoughRCMdid not find wide application in the wind farm industry, its studies on effect of failure are still valuable for determining which components should be monitored closely in case of predictive maintenance [9].

(19)

3

Failure and reliability

Before we dive into different failures and detection methods it is of paramount importance to under-stand the concepts of failure and reliability. These aspects will be addressed in this chapter, by giving an overview of the definitions and relations involved.

Failure is: ”the cessation of normal operation” [26]. In other words, failure is either present or not, it is a binary state. This implies that failure alone tells almost nothing. Engineers and wind farm managers require more information to say something useful regarding the performance of the wind turbine. Therefore, more definitions are used to describe the nature, frequency and effect of these failures. First of all reliability: ”The probability of a device performing its purpose adequately for the period of time intended under the operating conditions encountered”and availability: ”the probability of finding a system in the operating state at some time into the future”[27]. Reliability is usually given before operation and availability is quantitatively determined by data gathered during operation.

To quantify availability you need several other parameters:

• Mean Time To Failure(MTTF) is the time it takes after maintenance until a new breakdown of the system.

• Mean Time To Repair(MTTR) is the time required to conduct the maintenance operation, including waiting times.

• Mean Time Before Failure(MTBF) is the sum of theMTTFandMTTR.

• The failure rate𝜆 is the inverse ofMTBF.

• Repair rate𝜇 is the inverse ofMTTR

Availability is then calculated as shown in Equation3.1.

𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 1 −𝜆

𝜇 (3.1)

The failure rate is strongly dependent on the operating life of the wind turbine. This relation is shown in Figure3.1, which is known as the bathtub curve. The bathtub curve is usually approximated with a Power Law Process (PLP) shown in Equation 3.2. Where 𝛽 is used to identify the stage of degradation shown in Figure3.1[27,28]. 𝜃 is used as a shape parameter and has the units of time, it is always greater or equal then 0 for positive time.

𝜆(𝑡) = 𝛽 𝜃(

𝑡

𝜃) (3.2)

(20)

12 3.Failure and reliability

Figure 3.1: Failure rate in relation to service life. Curve is typically known as the bathtub curve [27]

The curve in Figure3.1is marked by three different phases: the first phase contains mostly instal-lation failures, the second phase contains the useful life of a product and during the final phase the product suffers from wear, consequently failure rate increases. The curve shown in Figure3.1is best applicable to electromechanical systems. Some components are more likely to fail in their early life; like control systems, whereas others; like wiring are more likely to fail after a long period. Therefore generally the overall system is captured with the bathtub curve [27].

The repair rate is very much dependent on the weather conditions as was explained before in section

2.1. But if waiting times are ignored, the time for actual repair work remains. The pure repair time is similar for on- and offshore- wind turbines. Therefore, the more publicly available data for onshore wind turbines is used to estimate the repair rate. The pure repair time is distributed as shown in Figure

3.2. From this figure, it is clear that a large portion of the failures take only a short time to repair. Therefore, the repair rate is mostly based on these short repairs. However, in offshore wind farms the travel time to the wind turbine is much longer. Consequently, regular short downtime operations can become long downtime operations. Which means that the repair rate will decrease drastically.

Figure 3.2: Downtime distribution of onshore wind turbines [27]

Short downtime operations are called minor repairs, which apply to repairs taking less then a day. Major repairs are those which require multiple days. However major repairs are far less frequent, they could take as long as 100 days. The division boundary of one day applies to onshore wind farms and is chosen at one day because only one crew team is required to resolve the failure. For offshore wind turbines, there is relatively less time available for repairs because of travelling times and accessibility.

(21)

13

This means that the amount of maintenance which could be performed on a single day, with a single crew team decreases. However, minor repairs also contain hard resets of the system which could be done remotely [27]. Logically, these repair times will not increase once turbines are placed offshore.

In onshore wind turbines 75% of the failures are as a result of minor failures, but these account for only 5% of the total downtime. Therefore, onshore wind turbine manufacturers main focus is on the major failures [27]. Whereas offshore, the downtime for minor failures will increase relatively more compared to those of major failures. Therefore, it could be beneficial to focus more on eliminating the minor failures first before addressing the bigger ones. On the other hand, you could also argue that because major failures require longer repairs also a longer weather window is required which makes the probability of finding one smaller. This matter requires more research which is out of scope of this report.

All in all, the different concepts highlighted above add up to an overall availability of the average wind turbine. The overall availability of offshore wind turbines compared to wind turbines onshore is compared in Figure3.3. It can be seen that availability is greatly dependent on which wind farm and which year is being considered [4]. On wind farm Middelgrunden offshore wind turbine availability is even greater in some years compared to the average onshore wind turbine availability. But generally, offshore wind turbines do not match the onshore availability.

Figure 3.3: Availability of different offshore wind turbines on different wind farms compared to onshore wind turbines [4].

Article Caroll toevoegen en Egmond aan Zee? Article Caroll toevoegen en Egmond aan Zee?

(22)
(23)

4

Wind turbine components and

common failures

Now that failure has been defined for wind turbines, the different components in a wind turbine are investigated. Once these components have been identified, an investigation into their failure rates is conducted.

4.1.

Wind turbine components

In this section an overview of the different wind turbine components and their common failures is given. Generally these components are listed as part of a separate sub systems. Which can be repaired or replaced as an entire module. The sub systems are mainly categorized by the function they fulfill within the entire system. An overview of most components that will be highlighted below is given in Figure

4.1.

Figure 4.1: Main sub systems in a common wind turbine [29]

4.1.1.

Rotor blades

The rotor blades are one of the key systems for a wind turbine. Once wind blows against the rotor blades, they start turning because of their design. The blades are equipped with a spherical and a concave side. During spinning, an under pressure forms on the spherical side with respect to the concave side. Resulting in a pulling force which allows the turbine to spin. Behind the rotor blade a

(24)

16 4.Wind turbine components and common failures

turbulence forms which is undesired to travel through. Therefore rotation speed is limited, to make sure that one blade never finds itself in the wake of another [30].

The blades are made form lightweight but stiff material, commonly a glass- or carbon- fiber re-inforced plastic. Some wind turbine’s rotor blades are equipped with extra features to prevent from lightning strikes or prevent ice forming [23]. Generally three-bladed wind turbines are being manufac-tured. This is because three blades optimally benefit from the amount of wind. More rotor blades would only lead to higher production costs and less speed to avoid turbulence. Two bladed wind turbines do exist but these are more unstable, create more noise and more distracting for human eyes [30].

Common flaws in the rotor are asymmetries, fatigue, reduced stiffness, propagation of cracks, increased surface roughness and deformation of the blades. Asymmetry is either the result of mass imbalance or errors of the pitch angle.

Fatigue failures are the result of material aging, long term repetitive stresses which cause failure at stress levels below yield strength. Fatigue stresses could result in delamination of the blades with reduced stiffness and the formation of cracks as result .

Increased surface roughness can be the result of long term exposure to pollution, icing or the existence of blowholes.

Unbalanced loading of the blade for a significant amount of time could result in deformation of the blade [29].

Furthermore, also other typical composite related failure modes could be identified on wind turbine blades, such as fiber breakage, matrix cracking and fiber splitting [31].

4.1.2.

Hub

The hub is the part of the wind turbine where the blades are connected to the stationary nacelle. The hub itself rotates and is connected to the main drive shaft which drives the generator. Generally the hub is made of steel or cast iron. This is because large forces are to be supported by this component [23].

Because of high repetitive irregular loads the hub is vulnerable to fatigue failures which can cause cracks.

4.1.3.

Pitch system

The pitch system connects the rotor blades to the hub. The pitch system allows blades to turn along their own long axis. This is used to regulate turning speed and therefore output. The turning speed can be adjusted by pitch angle because the pulling force can be controlled once the blade rotates. A common misunderstanding is that more wind leads to more output power. This is wrong, the generator only works to a specified maximum power output. Once wind speeds exceed the maximum power threshold, the pitch system is used to regulate the turning speed of the blades. In other words, the pitch system is used to control the power output. A common power curve for a wind turbine is shown in Figure4.2. The incoming wind power𝑃 is divided over generated power𝑃 and power losses 𝑃 (see Equation4.2). Where the pitch angle is used to control the power losses. The incoming wind power is related to the air density𝜌, swept area of the turbine 𝐴 and the wind speed 𝑣 this dependency is shown in Equation4.1.

𝑃 = 1

2𝜌𝐴𝑣 (4.1)

𝑃 = 𝑃 + 𝑃 (4.2)

The pitch angle is commonly adjusted by means of hydraulics to control power losses. The control of the pitch angle is usually fully autonomous based upon various sensor information [23].

(25)

4.1.Wind turbine components 17

Figure 4.2: Power curve of V112-3MW wind turbine [32]

Pitch system failures are either the result of bearing failure or hydraulic failure. Bearing failures can be caused by cracks and fatigue. Or due to breakage of the inner-/ outer- raceway, balls or cage. Hydraulic failures include oil leakage and sliding valve blockage [29].

4.1.4.

Yaw system

The yaw system is used to make sure the nacelle is placed under an angle to optimally benefit from the incoming wind. The yaw system is located inside the nacelle and forms the connection between tower and nacelle. This means that the tower is always stationary and that the nacelle has one degree of freedom. The yaw system is, similarly to the pitch system, mostly hydraulically driven. Furthermore, it contains a large ball bearing to make rotations possible [29].

Failures found in yaw systems are similar to those found in pitch systems.

4.1.5.

Nacelle

The nacelle is the housing of the components in the wind turbine. The nacelle is used to support all the bearings and to shield the inner components from the elements. Therefore, it is important for the nacelle to be completely water resistant because sea water is to be kept out at all cost.

4.1.6.

Tower

The tower is no more then a a hollow monopile. Towers become taller and taller because at higher altitudes there are higher wind speeds. Which is beneficial because wind speed scales to a third power (Equation4.1). The tower has no degrees of freedom and is merely used to give height to the turbine. The tower forms the connection between the nacelle and the deep sea foundation.

The tower most commonly fails on structural damages like cracks and corrosion. Typically, tower failures only present themselves under extraordinary circumstances like lightning strikes, fire and severe storms. Furthermore, fatigue failure is also present in the tower, as a result of repetitive loading of wind and waves [31]. However, tower failure could also result from improper installation or manufacturing flaws [29].

4.1.7.

Main shaft

The main shaft is connected to the hub and gearbox. The rotational speed of this main shaft is equal to that of the hub, meaning slow. The main shaft is exposed to high bending moments because the

(26)

18 4.Wind turbine components and common failures

weight of the rotor blades is to be supported by this shaft. Therefore, the main shaft is supported by very strong radial bearings [23].

The main shaft commonly fails as a result of: crack, misalignment, corrosion and coupling failure. Failures in the main shaft can be problematic for all other components, because everything is connected to this shaft. The danger is mainly attributed to the fact that once failed, the main shaft excites vibrations in certain frequencies which can lead to resonance which accelerates degradation [29].

4.1.8.

Gearbox

The gearbox is used to increase the rotational speed of the main shaft to rotational speeds useful for the generator. Commonly, it requires three gearbox stages to go from 30 rpm to 1500 rpm which are common rotational speeds for wind turbine shafts. This means a gearbox ratio of 1:50. Normally, planetary gear sets are used in the gearbox. This is because they are compact and provide in line connection of the different shafts [23].

Gearbox failures are the cause of the longest downtime in wind turbines. The failures are usually located at the gears and bearings. Failures are caused by misalignment, installation errors, material defects, overload, surface wear and fatigue. Commonly, the flaws originate at the bearings where debris can cause abrasion of other components. But also, poor lubrication can result in tooth abrasion of the planetary gear wheels [29].

4.1.9.

Generator

The generator is used to transform the energy in the rotating shaft to electrical energy. This is done by means of induction.The problem with induction is that it delivers power at the input frequency. However, it is desirable to deliver a constant frequency to the power grid. Therefore, a Double Fed Induction Generator (DFID) is most commonly used. This type of generator allows different input frequencies and converts it to a steady 50/60 Hz. DFIDis therefore extremely useful in wind turbine industry because wind speed is never constant [33].

Generators can either fail as a result of electrical or mechanical failures. Electrical failures commonly consist of winding faults, short circuits and inter-turn faults. Electrical imbalance is an electrical prob-lem leading to failure of mechanical components because of increased temperature and vibrations. Mechanical failures include a broken rotor bar, bearing failure, displacements, mass imbalance and vibrations [29].

4.1.10.

Mechanical brakes

The mechanical brakes are installed to prevent the wind turbine from rotating too fast. Typically, mechanical brakes are used at wind speeds above 25 m/s (wind power 10 on the Beaufort scale) [32]. Otherwise, at these large wind speeds, the loads on the system become too large and mechanical failure prompts. Furthermore, there are also brakes installed at the yaw system. These allow the yaw system to shut down once wind direction is constant. Besides that, brakes are also used to shut down the system during maintenance operations, or once other critical components have failed. Normally, a disc brake with hydraulic powered calipers is being used [29].

A common failure in the mechanical brakes is excessive wear of either the calipers or brake disc. This flaw goes hand in hand with overheating of the entire brake system. Malfunction of the brakes can lead to catastrophic failure of the entire wind turbine. For instance, once the brake is not working when wind speeds rise above 25 m/s, the rotors start rotating too fast and all components in the wind turbine are subjected to excessive loads. Therefore, it is of paramount importance to keep the brakes functional at all time [29].

4.1.11.

Electrical system

The electrical system is the connection between the power grid and the generator. To maintain proper voltage and phase the energy needs to be controlled dependent on the amount of active and reactive power [23].

Overall the most common failure in wind turbines is the failure of the power electronic converter. These failures are caused by three factors: temperature, vibration and humidity. Of these three, temperature is the dominant one. The distribution of failures of different electric components is shown in Figure4.3.

(27)

4.1.Wind turbine components 19

Figure 4.3: Distribution of failures among different electrical components [29]

Capacitor failure includes: excessive leakage, short circuits, dielectric breakdown, migrating electric material, separated leads and increased dissipation factor.

PCBs common fail modes consist of broken buried metal lines, defect of vias, corrosion or crack traces, board delamination, misalignment of components, electrical leaks and cold solder joints.

IGBT modules most common reported failures are packing related and originate from thermome-chanical fatigue stress of the packing materials [29].

4.1.12.

Control system

The control system is a vital component regarding power output. It is located in either the nacelle or tower. Based upon various sensor information the control system makes sure that yaw and pitch angles are optimal for generating power. The main sensor information is power output, wind speed and wind direction. The control system should however not be confused with the system to regulate maintenance operations. The control system being described here, is solely used to optimize power performance and ensure safe operation. In larger wind farms, control systems of different wind turbines can be connected with each other to ensure an overall better performance [23].

Control systems can fail hard- or software related. Hardware failures are malfunction of: sensors, actuators, communication lines or the control board. Software failures consist of buffer overflow, insufficient memory, resource leaks and race condition. Most software failures can be resolved by resetting the system [29].

4.1.13.

Sensors

New modern wind turbines can be equipped with as much as 2000 sensors. All of these sensors provide information regarding operating conditions, performance ratings and healthiness of components. A few examples of different sensor measurements are: vibrations, temperature, wind direction, wind speed and turning speed [23].

Sensors are prone to failures regarding data processing and communication. The nature of these failures can be either hard- or software related. Sensor failures can be critical for wind turbine perfor-mance because the control system could stop functioning [29].

4.1.14.

Overview of failures

In the previous subsections the main components and their common failures in a wind turbine have been identified. In Figure4.4the different failure rates and their respective downtime combined with the annual downtime is shown. From this figure we learn that electrical system failures are the biggest

(28)

20 4.Wind turbine components and common failures

cause to the annual downtime.

Figure 4.4: Distribution of failures among different wind turbine subsystems from a WMEP study [27]

Reliawind grafiek toevoe-gen?

Reliawind grafiek toevoe-gen?

(29)

5

Conditional monitoring

In previous chapters different maintenance strategies have been discussed. It was concluded that predictive maintenance is the most cost efficient method for critical components. A prerequisite for predictive maintenance is online continuous Condition Monitoring (CM). Therefore, this chapter will highlight the state of the art technologies regarding conditional monitoring. The focus will be on those components causing the longest down times, which were established in Figure4.4. TypicallyCM

consists of three stages [34]:

1. Data acquisition, collecting raw data from the installed sensors;

2. Data processing and diagnosis, involves signal processing techniques to convert the raw data to readable data. Afterwards a diagnosis determines if and where the damage is located or developing;

3. Prognosis of remaining useful life, an important step towards predictive maintenance. If the re-maining life can be accurately estimated an accurate maintenance plan could be made. However, this proves to be very hard and is currently mainly based on qualitative expert judgment. A key aspect of everyCMtechnique is the sampling rate, this should be carefully chosen to achieve desired performance. Low sampling frequencies prevent the bandwidth from becoming over populated and reduce the required storage size. But with low sampling frequencies, a lot of data is lost and only superficial conclusions can be drawn. High sampling frequencies should be handled with caution, as they easily flood storage systems and congest the bandwidth. On the other hand, a more detailed damage profile could be made once more data is available [35].

One way to overcome this dilemma, is using a variable sampling rate. Low sampling rates could be used until an anomaly is discovered. Which should be the signal to increase the sampling rate to localize and estimate the severity of the damage [35]. [35] distinguishes two different definitions for this: ordinary

CMand diagnosing, this distinction was not found elsewhere.

In this chapter different approaches toCM are highlighted. First different SCADA-based CM ap-proaches are described and supported by an example. Followed by the most widely used monitoring technique which relies on vibrations. Afterwards, lesser known CM techniques are reported. Next, structural health monitoring is discussed which is more concerned with failures to the structural part of the wind turbine. Finally, the chapter concludes with a study into the cost benefits ofCMapplied to a gearbox.

5.1.

Conditional monitoring based on SCADA information

Every wind turbine is equipped with Supervisory Control And Data Acquisition(SCADA) systems. The original purpose of this system was to monitor the production and confirm operation. Later, it was discovered that the gathered data could also be used to alert supervisors of incipient malfunctions.

SCADAsystems provide 10 minute averaged measurements regarding [36]:

• Active power output (net one-directional energy transfer [37]);

(30)

22 5.Conditional monitoring

• Reactive power (energy returning to the source [37]);

• Power factor (ratio of real power to apparent power [37]);

• Phase currents;

• Average wind speed and standard deviation, measured with an anemometer;

• Gearbox bearing temperature;

• Gearbox lubrication oil temperature;

• Generator winding temperature;

• Nacelle temperature (1h average)

These measurements could be extended with vibration transducers which give an earlier warning to failure compared to temperature, shown in Figure5.1[35]. Furthermore, [38] reports thatSCADAdata also includes: Generator bearing temperature and shaft speeds of the turbine and generator. [34] adds that currently not only 10 min average results are given but also minimums, maximums and standard deviations. Furthermore, [34] says that newSCADAsystems also include information on the voltages and currents of the pitch motors and tower vibration data.

Figure 5.1: Development of a mechanical failure [22]

Some researchers ([22,35]) think the way forward withCMof wind turbines is in additional better

CMsensors whereas others ([38]) believe that better use of SCADA data is the path to follow. The reason why [38] believes that additionalCMis not the way forward is because of:

• CMsystems are very expensive with costs over £10k;

• CMreport false positives (false alarm) and sometimes true negatives (no alarm);

• CMsystems focus on mechanical failures whereas electric and hydraulic components are more prone to failures;

• Direct-drive and hybrid-drive wind turbines start to take ground with respect to conventional gear driven wind turbines. These new types of wind turbines use less mechanical components and therefore more components cannot be monitored with currentCMtechniques.

On the other hand, there are also problems with theSCADAdata currently available [38].

• SCADA only reports data every 10 minutes. Therefore, conventional interpretation techniques like spectral analysis cannot be used;

(31)

5.1.Conditional monitoring based on SCADA information 23

• Because operational conditions (wind -speed and -direction) vary over time,SCADAdata covers a wide range. Therefore, it is hard to diagnose whether change in data is due to failure or operational circumstances. This results in faults being detected only once it is already too late. A scatter-plot of rawSCADA data is given in Figure 5.2. This figure illustrates the difficulty of reading the data.

• Current analyzing methods rely on comparing data with those of neighbouring wind turbines. Or they use advanced analyzing techniques like neural networks which require a lot of training data.

• The advance detection period is short. This is problematic if maintenance scheduling relies on theSCADAanalysis. A distribution of the advance detection period is shown in Figure5.3. The distribution shows that most failures are detected within six months prior to failure. This is too short if maintenance operations are preferably scheduled during summer [34].

Figure 5.2: SCADAdata of a 750 kW wind turbine, temperature refers to the generator bearing temperature [38].

Figure 5.3: Distribution of the advance detection period for faults detected by analysis based onSCADAinformation [34]

5.1.1.

Signal trending

One way to analyze SCADA data is through signal trending, which is a relatively easy method. It is centered around the comparison of incoming SCADAdata with reference data. This reference data could come from other neighbouring wind turbines or the historical data of the same wind turbine [34]. [38] has developed a method to do signal trending which also indicates the severity of the damage. The required steps are shown below.

(32)

24 5.Conditional monitoring

Pre-Processing

The problem with the data shown in Figure5.2is that the data is noisy even though it already displays 10 minute averages. This noise makes it difficult to find correlations between different parameters. Fortunately, there are methods which help to reveal these correlations. First of all, a pre-processing of the data is required. [38] uses 11 steps to do this.

1. Define theSCADAdata of interest.

2. Remove collected data while turbine was in standby.

3. Recover the upper and lower limit of the wind speed𝑉 and𝑉 .

4. Divide the range[𝑉 , 𝑉 ] in 𝑁 smaller ranges. N can be found by using Equation5.1.

𝑁 ≈ 𝑉 − 𝑉

0.5 (5.1)

5. Initialize𝐾 = 1

6. Identify the indices of wind speed data located in the range[𝑉 + (𝐾 − 1)𝑉 , 𝑉 + 𝐾𝑉 ] where𝑉 is calculated with Equation 5.2, this should be close to0.5

𝑉 = 𝑉 − 𝑉

𝑁 (5.2)

7. Further divide subrange𝐾 into 𝑚 different bins. [38] suggests using 5 different bins.

8. Now estimate the expected wind speed ( ̄𝑉 ) value in each bin with Equation5.3.

̄𝑉 = ∑ (𝑝 𝑥 𝑣 ) (5.3)

Where 𝑝 = is the probability that the wind speed is in bin 𝑗. 𝑛 represents the number of wind speed data in bin𝑗, 𝑛 is the number of wind speed data in subrange 𝐾 and 𝑣 is the wind speed corresponding to bin𝑗.

In other words, the average wind speed of each subrange is determined based on a 𝑚-bin histogram method. This is because this method is capable of filtering outliers.

9. Repeat this procedure for the generator power data and other relevant data gathered bySCADA.

10. Increase𝐾 by one, 𝐾 = 𝐾 + 1.

11. Iterate steps 6-10 until𝐾 > 𝑁.

Moreover, it should be pointed out that different references than wind speed might also be used. However, wind speed proves to work best for uncovering hidden correlations [38]. The effect of the pre-processing procedure is shown in Figure 5.4. From this figure it becomes clear that it is hard to see the correlations in the raw data but with a simple pre-processing algorithm faulty bearings can be identified.

(33)

5.1.Conditional monitoring based on SCADA information 25

Figure 5.4: SCADAdata of a 1500 kW wind turbine, temperature refers to the generator bearing temperature. Red dots are gathered before maintenance and black dots after maintenance [38].

Decision criterion

Once data has been pre-processed, a model of the current performance is made. This is later compared to healthy historic data to identify faulty equipment. As was previously mentioned in 4.1.3, the pitch system is used to regulate the rotor speed by rotating the blades along their long axis, extracting more or less energy from incoming wind. To a certain point, the pitch system will try to extract maximum energy. But above that the pitch system reduces the rotor speed. This turning causes high non-linearity in the gathered data and therefore [38] suggests that only the data of maximum extraction should be used to avoid dealing with this high non-linearity.

Different potential correlations between different SCADA data are shown in AppendixB. All data points available before the pitch system becomes active are called{𝑥 , 𝑦 } where 𝑖 = (1, 2...𝑛), further-more𝑥 and 𝑦 represent two parameters from the correlation listB. With a𝑘 order polynomial,𝑦 can be approximated from𝑥 , shown in Equation5.4. ̂𝑦 represents the approximate of 𝑦

⎡ ⎢ ⎢ ⎣ ̂𝑦 ̂𝑦 ⋮ ̂𝑦 ⎤ ⎥ ⎥ ⎦ = ⎡ ⎢ ⎢ ⎣ 1 𝑥 𝑥 … 𝑥 1 𝑥 𝑥 … 𝑥 ⋮ ⋮ ⋮ ⋱ ⋮ 1 𝑥 𝑥 … 𝑥 ⎤ ⎥ ⎥ ⎦ ⎡ ⎢ ⎢ ⎣ 𝑎 𝑎 ⋮ 𝑎 ⎤ ⎥ ⎥ ⎦ (5.4)

Or more conveniently written down as:

y= Xa (5.5)

The𝑎 can be approximated with a least square fitting, this gives 𝑎 as shown in Equation5.6[39].

a= (X X) X y (5.6)

Matrix𝑎 now defines the model of the system, this model could be compared with the historical model 𝑏. A comparing method is shown in Equation5.7. Where𝑥 and 𝑥 represent the minimum and maximum value of the data [38].

𝑐 =

∫ |∑(𝑎 − 𝑏 )𝑥 |𝑑𝑥

𝑥 − 𝑥 (5.7)

Once𝑐 ≈ 0 the turbine is healthy, meaning no significant differences with historical data. Whenever 𝑐 > 0, the current wind turbine model is inconsistent with historical data and could indicate incipient failure. The complicated comparing criterion (Equation5.7) is used because it depends on the variation of 𝑦 against 𝑥 on the entire interval [𝑥 , 𝑥 ]. Because basically, it is an integral of the partial differential| ̂|. This is more reliable then the instantaneous difference between 𝑎 and 𝑏 [38].

The method proposed by [38] described above, is also empirically tested and shows very promis-ing results. These include detection of generator windpromis-ing faults, gear teeth faults, blade failure and

(34)

26 5.Conditional monitoring

generator bearing failure. The start and end of the process is shown in Figure5.5[38]. The effect is clear, over a four month interval the𝑐- value increases, indicating incipient failure. This allows for a maintenance operation somewhere in these months. From the raw data alone, it would be impossible to detect deterioration. Furthermore, the 𝑐- value allows for a quantification of the severity of the situation. Unfortunately, the severity of the real situation is not given in the article. Besides that, it should be noticed that there is a huge difference in c- value depending on which correlation is evalu-ated. Therefore, care should be taken with developing a decision criterion for which𝑐 a maintenance operation is required. If multiple correlations, as shown in Figure 5.5, exist, it could be beneficial to integrate or combine them to get a better estimate of the fault.

Figure 5.5:SCADAdata for detection of blade failure, (top) rawSCADAdata (bottom) processedSCADAdata [38]

5.1.2.

Power signal analysis

Another method which usesSCADA data to detect imbalances is Power signal analysis. The working principle, is that by investigating the power output, failures could be detected. [22], usesSCADAdata to detect rotor imbalance by means of power signal analysis. It is based on the difference between real output current and expected output current of the turbine. A healthy turbine would produce a pure single sinusoidal output current,𝑖 (𝑡) shown in Equation5.8.

𝑖 (𝑡) = 𝑎𝑐𝑜𝑠(𝜔 𝑡) (5.8)

Where𝜔 represents the angular shaft speed of a healthy turbine and 𝑎 the amplitude of instantaneous current for a healthy wind turbine. An instability in the rotor will change the shaft rotation speed to𝜔 see Equation5.9.

𝜔 (𝑡) = 𝜔 + 𝑐𝑐𝑜𝑠(𝜔 𝑡) (5.9)

Where𝜔 represents the angular shaft speed produced by the fault. Which results in a phase shift in current obtained by Equation5.10.

(35)

5.2.Conditional monitoring based on vibrations 27

Here𝛾 is a constant equal to Which produces a faulty current (𝑖 ) as in Equation5.11.

𝑖 (𝑡) = 𝑎𝑐𝑜𝑠[𝜔 𝑡𝛾𝑠𝑖𝑛(𝜔 𝑡)]

= 𝑎𝑐𝑜𝑠(𝜔 𝑡)𝑐𝑜𝑠(𝛾𝑠𝑖𝑛(𝜔 𝑡)) − 𝑎𝑠𝑖𝑛(𝜔 𝑡)𝑠𝑖𝑛(𝛾𝑠𝑖𝑛(𝜔 𝑡)) (5.11)

If we assume𝑐 ≪ 𝜔 which leads to 𝛾 ≪ 1 we could simplify Equation5.11to Equation5.12because 𝑐𝑜𝑠(𝛾𝑠𝑖𝑛(𝜔 𝑡)) ≈ 1 and 𝑠𝑖𝑛(𝛾𝑠𝑖𝑛(𝜔 𝑡)) = 𝛾𝑠𝑖𝑛(𝜔 𝑡) 𝑖 (𝑡) = 𝑎𝑐𝑜𝑠(𝜔 𝑡) − 𝑎𝛾𝑠𝑖𝑛(𝜔 𝑡)𝑠𝑖𝑛(𝜔 𝑡) = 𝑎𝑐𝑜𝑠(𝜔 𝑡) −𝑎𝛾 2 𝑐𝑜𝑠((𝜔 − 𝜔 )𝑡) + 𝑎𝛾 2 𝑐𝑜𝑠((𝜔 + 𝜔 )𝑡) (5.12)

This means that by frequency demodulation of the current output, rotor imbalances can be detected [22].

Other mechanical failures could lead to amplitude modulation of the stator current. This effect could be observed because the amplitude becomes time variant. [22] does not further specify which type of mechanical failures could be detected by amplitude modulation.

However not described by [22], it could be that the current data obtained by SCADAshould also be pre-processed. Initially, the pre-processing algorithm given by [38] presented in subsection5.1.1

could be used.

Other SCADA based analysis

Besides the strategies mentioned before, there exist many other techniques that rely onSCADA data. Usually, these techniques employ some sort of model of healthy wind turbine behaviour. Based on the model and operational conditions these models predict the behaviour of the wind turbine. Once different behaviour is shown by the incoming data it indicates failure. This strategy can rely on different techniques, one of them is Artificial Neural Networks which is briefly described in [34].

5.2.

Conditional monitoring based on vibrations

In this section conditional monitoring based on vibration information is being discussed. This is differ-ent from previous section because that involvedCMbased onSCADAinformation. Vibration analysis is ”the most popular technology employed in wind turbines, especially for rotating equipment” [40].

Generally, mechanical failures can be detected in an early stage based on vibration information (see Figure 5.1). ISO 10816-21 [41] and IEC 61400-25-6 [42] define the location points and orientation of the vibration sensors offering the ability to exactly locate faulty components. Figure 5.6 shows a gearbox and generator where vibratory sensors have been applied appropriately. ISO 13373 -1 [43] helps to process and present vibration data which is a crucial step towards CM. A guideline for im-plementation of vibration based CMcan be found in [44]. The frequencies of the vibrations play an important role into determining the severity of the fault. Different frequency ranges require different sensors. Low frequencies are detected with position transducers, middle frequencies with velocity sen-sors, high frequencies require accelerometers and spectral emitted energy sensors are used to detect the highest frequencies [45]. Commonly, vibration analysis works in the high frequency range and therefore uses accelerometers. Frequency analysis can be used because different components excite different distinctive frequencies, during normal and faulty operation. Commonly, a signal processing technique, like the Fast Fourier Transform (FFT), is required to convert the time-domain signal into a frequency domain signal [22,34].

(36)

28 5.Conditional monitoring

Figure 5.6: Locations of vibration sensors on the gearbox and generator forCMin SKF WindCon system [34]

Similar to the procedure of5.1.1[42] suggests that the vibration sensor information should be pre-processed by a bin method. The sole difference with [38], is that power is used as the reference instead of wind speed. After acquisition and pre-processing the data, the remaining data should be processed. This requires signal processing techniques, for different situations different methods should be used. An overview of different signal processing techniques, both in time and frequency domain, is given in [34].

Because vibration based condition monitoring is such an isolated field, expert monitoring centers are hired for monitoring the wind turbine components. These service centers send color coded peri-odic reports to the wind farm owner who is then responsible for the maintenance. This division has a negative influence on the progress of vibration based analysis because the wind farm owners do not deliver feedback to the service centers. Which means that their training dataset is never extended. Therefore, better communication between these two actors would improve failure detection and reduce false alarms [34].

Different components display different frequencies upon faulty operation. These frequencies will be highlighted in the following subsections.

5.2.1.

Gearbox

The gearbox will excite a certain characteristic frequency𝑓 induced by gear faults. These vibrations could be measured by installed vibration sensors on the gearbox. 𝑓 is dependent on the rotating frequency of each shaft𝑓 , and each gear meshing frequency𝑓 , . Where𝑖 and 𝑗 denote the number

of the shaft and gear. 𝑓 can be determined with Equation 5.13. 𝐼 and 𝐽 give the total number of shafts and gear pairs in the gearbox [29].

𝑓 = ∑ 𝑙 𝑓 , ± ∑ 𝑚 𝑓 , {𝑙 , 𝑚 = 0, 1, 2, … } (5.13)

Acoustic Emission (AE) is another phenomenon that expresses itself in frequencies similar to the characteristic frequency𝑓 . The working principle is explained in subsection5.4.2. AEsensors could be equipped on gearboxes as well to measure𝑓 [29].

Vibrations will result in amplitude and frequency modulations observed in the electrical signals of components connected to the gearbox. Most commonly, the current sensors of the generator. These characteristic frequencies are called𝑓 and are determined by Equation5.14.

(37)

5.2.Conditional monitoring based on vibrations 29

In Equation 5.14 𝑓 is the fundamental frequency and 𝑘 is used as a positive integer to represent possible harmonics.

However, the three techniques presented above provide a possibility to detect gearbox failures it is still a troublesome procedure. This is because normally the gearbox already displays several frequencies, so it is hard to distinguish new frequencies, and sometimes failure only alters the amplitude of existing frequencies which requires additional analysis [29].

Therefore, it could be useful to extend and combine vibration information with other measurements like: bearing temperature, lubrication temperature, lubrication viscosity and the presence of particles in the lubricant [29].

5.2.2.

Bearings

Bearings will display characteristic vibration frequencies once failure is incipient. These frequencies are distinctive for the different components of a ball bearing. Namely, the outer raceway 𝑓 , inner raceway𝑓 , balls 𝑓 and cage 𝑓 . Equations5.15to5.18report these frequencies as a function of: 𝑓 , the rotational frequency;𝑁, the number of balls; 𝐷 , the ball diameter; 𝐷 , the ball pitch diameter and 𝜃, the ball contact angle.

𝑓 = 0.5𝑁𝑓 (1 − 𝐷 𝑐𝑜𝑠𝜃 𝐷 ) (5.15) 𝑓 = 0.5𝑁𝑓 (1 +𝐷 𝑐𝑜𝑠𝜃 𝐷 ) (5.16) 𝑓 = 0.5𝐷 𝐷 𝑓 [1 − ( 𝐷 𝑐𝑜𝑠𝜃 𝐷 ) ] (5.17) 𝑓 = 0.5𝑓 (1 − 𝐷 𝑐𝑜𝑠𝜃 𝐷 ) (5.18)

Gear failures are easier to detect by means of vibrations then gearbox failures because they always add new frequencies. These can be identified by means of frequency analysis. However, to achieve optimal detection the measurements should be combined withAE, electrical signals and lubricant data [29].

5.2.3.

Generator

As was discussed in subsection4.1.9generators commonly fail on winding faults. Fortunately, winding faults can be detected because they alter the magnetic field present in an induction device. Once the magnetic field is changed, it is observed in the characteristic frequency 𝑓 of electrical signals. 𝑓 could be determined based on the number of pole pairs𝑝, the fundamental frequency 𝑓 and the slip 𝑠 according to Equation5.19[29].

𝑓 =

𝑘 ± ( )

𝑓 {𝑘 = 1, 3; 𝑛 = 1, 2, … , 2𝑝 − 1} (5.19) Slip is the relative difference between synchronous speed and actual speed of the rotor of the gener-ator [46]. However vibration techniques work well winding faults could also be detected with torque measurements, shaft displacement, winding temperature, gearbox vibration (not recommended) and electric machine vibration [29].

With the same parameters as in Equation5.19also potential breaking of the rotor bar can be detected with Equation5.20[29].

𝑓 = (1 ± 2𝑘𝑠)𝑓 {𝑘 = 1, 2, 3, … } (5.20)

5.2.4.

Other components

Many other components could also be diagnosed by using vibration frequency analysis andAE. However, because every component in the wind turbine is connected with another, and most components rotate themselves it is extremely difficult to distinguish which frequencies are the result of faulty equipment. Therefore, [29] already suggests combining different measurements with each other to achieve a

Cytaty

Powiązane dokumenty

According to the UNAIDS report from 2013, the number of people living with HIV infection across the world has been estimated at 78 million since the beginning of the epidemic, with

Credit money represents any circulating medium which has little real value relative to its monetary valuec. The Monetary Act of 1792 resulted in an early monetary system that relied

In this section we shall present some considerations concerning convergence of recurrence sequences, and their applications to solving equations in Banach

In this paper, we propose a solution for secure private data storage that protects confidentiality of user’s data, stored in cloud.. Solution uses order

During initialisation phase all &#34;CREATE&#34; queries are reconstructed and original data schema is mapped to encrypted one; all mapping parameters (meta information and

Impulse response analysis in infinite order cointegrated vector autoregressive processes, Journal of Econometrics 81: 127–157.

The space X of all countable ordinal numbers, endowed with the order topology, is sequentially compact and therefore countably compact4. This shows that Theorem 2 is false if R is

23 Tekst jedn. Maciej Zieliński, Wykładnia prawa.. Taka wskazówka sądu jest bardzo oczywista. Z kolei druga dana w cytowa- nym judykacie odsyła, przy ustalaniu znaczenia tego