August 2011

The basic design criteria for wind generator installations—that they are placed in large open areas, mounted on tall towers, and not influenced by other nearby structures—also places them at very high risk of being affected by lightning issues. This risk becomes even greater when installations are in high flash density regions such as shown on the map in Figure 1.

Lighting may cause the catastrophic damage shown in Figure 2, where a direct strike to a blade occurs, but more commonly occurring effects and probably less noticeable ones may be those caused by a nearby strike to ground, or “cloud to cloud” discharge. Statistics on this subject show that of all lightning damage to wind generators only about 10 percent involve blades, while approximately 85 percent affect the power, control, and communication systems.

These events may cause severe equipment damage, but they will also cause less-obvious damage including premature component failure and interference of critical communication and control signals, all of which results in a disruption of service and loss of generation (revenue).

Ground Potential Rise
Wind generators that are installed in high-risk lightning environments should have lightning protection systems incorporated in their mechanical design, which would include air terminals on the nacelle, receptors in the blades, adequate down conductors, and a proper grounding system. This system would protect the wind generator from the physical damage of a direct strike by providing a safe and effective path to earth for the discharge current. It would not eliminate the effect the lightning would have on the electrical system, however.

Assuming that (external) lightning protection is installed and the generator is not mechanically damaged by a direct strike, the electrical system remains vulnerable to high voltage transients that may appear on the power, control, and communications systems. These transients, also referred to as surges, will involve these systems by being electromagnetically induced, entering the system through grid connections or by differences in the ground reference that are connected by utility circuits; specifically a substantial Ground Potential Rise (GPR) of local earth with regard to distant earth. The latter is caused by a local strike or strike to the structure, and the resulting local GPR as the lightning current dissipates as it travels through the earth’s impedance and at the same time being connected to a distant ground reference through the utility grid. Figure 3

Regardless of the cause of these voltage transients on the system, they will have a very damaging effect as they appear at the terminals of key pieces of equipment, since they will be of voltages magnitudes many times greater than what the equipment was designed for. While the transients are short lasting, from some tens of microseconds to less than a millisecond, they have high energy content during that short time and can produce insulation breakdowns with discharge currents of tens of thousand of amperes, occurring in multiple locations.

The most commonly used representation of the wave shapes for describing the characteristics of a lightning and surge current discharge is the 8/20 microsecond and 10/350 microsecond curve, where the later is associated with direct lighting strikes. One can easily see that even if the magnitudes are equal, the greater duration of the 10/350 µs wave has approximately 25 times the energy than the 8/20 µs.

Surge Protection Devices
The strategic placement of Surge Protective Devices (SPD) mitigate these issues as they function as either shorting or voltage clamping devices minimizing that transient voltage that would have otherwise appeared at the equipment terminals. These devices are applied between the energized conductor and ground and must be connected close to the terminals of the equipment they are intended to protect, and using the shortest and straightest routed conductors of sufficient cross-sectional area. Not implementing this design criterion will result in a higher than necessary voltage drop in the SPD circuit during the discharge that will expose the “protected equipment” to higher transient voltages than necessary.

The selection criteria is based upon the device’s ability to change states quickly enough for the brief time the transient is present, to be able to discharge the magnitude of the transient current that is associated with the transient voltage and without failing, and to do so minimizing the voltage drop across the SPD circuit so as to protect the equipment it is connected to, and to not interfere with the normal function of that circuit.

The most commonly used component for SPD application is the Metal Oxide Varistor (MOV) that functions as a voltage-clamping device. Other components are also used for SPD applications such as Silicon Avalanche Diode, controlled Spark Gaps, and Gas Discharge Tubes, with the latter two being switching devices which appear as short circuits or crowbars during their operation. Each technology has its own characteristics, making it more or less suitable for a specific application.

Figure 4 shows a table listing the basic characteristics of each of the primary components. These components are used to make a completed SDP that will include an insulated enclosure, terminals for connecting conductors, safety disconnection mechanism, status indications, and other features. Figure 5

Combinations of these devices can also be paired and coordinated so as to provide more optimal characteristics than they would offer individually. Figure 6 shows an example of two coordinated SPDs; a Spark Gap- and MOV-based technology that together are able to both discharge high energy lightning currents and clamp the residual voltage to an acceptable protection level.

AC Surge Protection
Lightning flashes from cloud to the lightning protection system, the wind generator structure, or a nearby ground will cause a local GPR with regard to distant ground references. Conductors spanning these distances will expose equipment to these significant voltages. The effects of these ground potential rises are primarily seen where a grid tied system is connected to the utility at the service entrance, since it is this intertie that connects local ground to a distant referenced ground. Surge protection should be placed at this location primarily to protect the service entrance side from damaging transients. The transients seen at this location are of the higher magnitude and duration (10/350 µsec versus 8/20 µsec), and therefore must be managed by surge protection with appropriately high discharge current ratings. Spark gaps in coordination with MOV SPDs are the ideal solution. The Spark Gap technology has the ability to discharge the high “lightning currents” providing an equipotential bonding function during the occurrence of the lightning transient, and coordinated MOV has the ability to clamp the residual voltage to an acceptable level for the equipment it is to protect.

In addition to the use of these coordinated Spark Gap/MOV SPDs at the service entrance, properly rated MOVs should be at critical locations within the generator system to mitigate the effects of induced lightning transients and internally generated switching transients.

Non-Power System Surge Protection
Just as power system equipment and components are susceptible to the affects of lightning, so is the equipment found on measurement and control, instrumentation, SCADA, and communication systems associated with these installations.

The basic concept of SPD protection is the same as on power circuits, but the equipment is usually less tolerable to overvoltage impulses and more susceptible to erroneous signals and to being adversely affected by the addition of series or parallel impedances to the circuits. Greater considerations must therefore be given to the characteristics of each SPD that is added to such circuits. Whether these components are communicating through twisted pairs, CAT 6 Ethernet, or coaxial RF, SPDs are available for these systems as well. As an example, when selecting SPDs for these circuits consideration must be given to their ability to discharge the transient currents without failure, provide an adequate voltage protection level, and have other unique characteristics so as not to interfere with the system’s function including series impedance, line to line and ground capacitance, and frequency bandwidth, etc. 

Conclusion
In spite of the high lightning risk that wind power system installations are exposed to, they can be protected by the application of Surge Protection Devices and Lightning Protection Systems. Regarding SPDs, one must give thoughtful and careful consideration to the placement of the SPD on the system and what they are intended to protect, the connection from the system to the SPDs, and the proper bonding of all grounded members and the ground system. Also check for the adequate discharge rating of the SPD, the voltage protection level the SPD provides, the suitability of the SPD for the system it is applied to, and consideration to failure mode of SPDs and status indication. Also be mindful of local and remote status indication and easily replaceable modules, and the suitability of SPDs on systems so as not to affect normal system function, specifically on non-power systems.  

Why should I care if my torque wrench or multimeter is calibrated? They’re built to be rugged, and they’re checked when they roll off the production line, so how far off could they be? Besides, I have a schedule to maintain, and I don’t have time for this nuisance. It can’t be that big of a deal.

If any or all of these statements represent your mindset, then it is apparent that instrument drift is not apparent to you. I have worked in the field of metrology—the science of measurement—for more than 28 years. While my calibration and engineering experience has been evenly divided among military weapons systems, manned shuttle launches at the Kennedy Space Center, and pharmaceutical and medical device manufacturers, a consistent theme rings true of the wide variety of instrumentation and gages with which I have dealt: you simply cannot predict which instrument/gage is going to exceed its allowable tolerance limits and have a negative impact on the products or processes where it has been used over its most recent calibration cycle. So, regardless of the industry in which the instrument is being used, if you are dealing with an acceptance tolerance for a given process, you cannot afford to have a situation where you thought the measurement was good and found out later that the instrument gave you a false indication. That can cause rework, scrapped product, or in the worst cases, product recall, injury, or fatality. We all hope that the processes in which instruments are being used to make qualifying decisions have enough “slop” in the acceptance tolerances to prevent these types of costly events from occurring, but history shows this is not always the case. And we shouldn’t base safety and cost on a “hope” anyway, should we?

Dodging Gotchas
In the energy industry there have been recent cases where improperly calibrated instruments have kicked off a series of events, leading to billions of dollars of property damage as well as the deaths of employees and contractors. The attitude of “we haven’t had a problem yet, so why worry about whether or not my calibrations are being done correctly?” is a very risky attitude to take, both in terms of safety and cost to your company. Figure 1

The cost to have your calibrations done correctly—to ensure traceable measurement is preserved from the upper echelons of NIST through reference level calibration laboratories, to the standards that your calibration supplier uses to determine whether your equipment is good or bad, to the way you use those instruments in your processes—is often a fraction of the cost of a product recall, a lawsuit related to an accident, or to a faulty product or service. It’s what is commonly referred to as the “cost of quality.” Any cost must also have a benefit, otherwise why spend the money? The benefit of calibrations done correctly is a reliable system of instrumentation that ensures decisions about the process or product are not incorrectly based on erroneous information, regardless of the instrument you pick up and use in the process.

Unfortunately, and you’re not going to like this, but not all calibrations are performed correctly. At this point I must claim bias because I work for a commercial calibration provider and, of course, we want your business. We’ve thought through some not-so-evident scenarios and have designed our processes to catch and correct some problems so that you don’t have to concern yourself with them causing problems in your processes. Allow me to remove myself from this business bias in order to point out the flaws that can become “gotchas” in your attempt to preserve measurement assurance in your processes. Most companies that provide B2B calibration services are ISO-17025 accredited. At a minimum, this ISO standard covers some very important points that are critical to ensuring you get the calibration that you need. It also misses some important points, many of which most people assume are not a problem.

Sharing Expectations
Let’s start with your expectation for the service you get when you pay for a calibration. Do you know what information you must convey to the calibration provider? Do you assume the calibration provider is the expert and no discussion is needed to convey your expectations? While they are the experts in calibration, they may miss some of these critically important points if you are not clear with each other. It’s up to you to understand where these pitfalls can occur and how to avoid them. This concept follows the “buyer beware” approach to conducting business and making purchasing decisions.

Stop for a minute and think about your personal buying decisions. When you go to the drive-through at a fast food place, how mad are you when you get all the way home just to find out they gave you the wrong food, or forgot to put something in the bag when they handed it to you? Your expectation is likely that the person on the speaker should be capable of taking down your order and passing that through to the people cooking your food, and on to the people putting it together for you at the pickup window. How can this fairly simple process go so wrong?

Now, consider how much more complex a process it can be when placing your order for the calibration of an instrument. Or how about placing your order for the calibration of 50-100 instruments, or 1,000 instruments? To my knowledge, the largest quote my company has processed was for just over 10,000 instruments and, considering the varying calibration cycles on each one, the total quote was for about 18,000 calibrations per year. So what are the odds of getting that order 100-percent correct, and what are your expectations for each of these calibrations? The good news is that we have designed our systems to handle this variety and volume. The reality is that we still get some things wrong. But that is what disclaimers, as well as warranties, are for. Believe it or not, many prospective clients simply can’t give us the information we need to make a decision about whether or not we have the capability to perform the calibration service, let alone the cost of the service.

Accuracy Counts
It all starts with properly identifying your instruments with a valid manufacturer name and model number. In many cases other information is important, too, such as instrument range, accuracy, description, cal cycle, and next due date, etc. This information helps your calibration provider to identify the instrument, which points to the accuracy or tolerances against which the instrument will be calibrated. Without this information it’s a best guess, and hence the disclaimer. But for those instruments which can be clearly identified up front a warranty can come into play if, for some reason, the calibration report does not provide the tolerances that you expected to see. If you regularly order calibration without data (certificate only), how can you possibly verify that what you expected is what you received? That would be like ordering a meal at the drive-through and not wanting the food, just taking the receipt. The substantial part of the order was the food, and the receipt simply shows that you paid for certain items. It’s pretty much the same with calibration: the substantial part of the order is the calibration report that shows the data on your instrument, not the calibration certificate that simply identifies your instrument. Dealing with calibration this way makes no sense whatsoever. In fact, ordering a calibration with the data and the lab’s measurement uncertainties is all-encompassing in preserving measurement traceability and may soon be a requirement for all accredited laboratories (i.e., it may not be a choice for them to provide the lesser levels of service, because it breaks traceability).

Speaking of expected tolerances, do you know what tolerances the instrument should be tested against? Does your calibration provider know how to determine these? Likely you chose the particular instrument because the OEM published information telling you how it will perform (accuracy specifications) over a given timeframe (recommended cal cycle). and this met your needs (product/process acceptance tolerances). So it stands to reason that your calibration provider should also check the instrument against these same OEM instrument tolerances. This is where the ISO standard is lacking. It states simply that the calibration provider must understand their client’s needs and expectations. It can’t possibly cover all of the tolerances and test points for the huge variety of instruments that require calibration. But if that conversation between the calibration provider and their client isn’t explicit, the result could be that you are getting something that missed the mark on your expectations. How can that happen, you ask? Let me count the ways.

Try Troubleshooting
Incorrect Calibration Procedure: Some calibration service providers rely on military cal procedures to provide the step-by-step methodology for their technicians. The pitfall here is that military cal procedures have sometimes been written for their specific use where functions/ranges of an instrument are omitted, different test points are checked that may not provide a full calibration, and/or accuracy specifications have been changed for a specific purpose. If you wanted a calibration against the OEM tolerances and your instrument was calibrated using a military procedure, you may not be getting what you expected—and your calibration provider may be none the wiser that they are not delivering to your expectation. This may not be an intentional act on your calibration provider’s part, but it certainly does not preserve your measurement traceability since you could be failing good product or passing bad product due to this miscommunication and/or lack of understanding.

Incorrect Tolerances (OEM): An OEM’s organization is comprised of human beings, who are known to make mistakes. How well an organization performs to keep things in control depends on how the company is managed, which changes over time. These changes in the OEM’s organization can cause inadvertent errors to occur. For example, the marketing department is eager to get their instrument out ahead of their competitors and may publish a marketing piece that has preliminary specs. Months later, the engineering department may make slight changes to the accuracy specifications and print these in the operating manual. If this misalignment between the marketing specs and the engineering specs is not caught, then a conflict has been created as to how the instrument will perform. This happens more often than you are probably aware.

So which spec were you looking at when you bought the instrument, and which spec is your calibration provider using for the tolerances they apply? Are they diligent in catching these OEM errors and clarifying them on your behalf? It’s not necessarily anyone’s fault, although the root cause lies with the OEM, it’s just a reality of the manufacturing and selling process. Becoming aware of these pitfalls can help you understand where your processes can get riddled with unexpected errors, and perhaps help you to prevent them.

Incorrect Tolerances (Accuracy Translation Error): Instrument specifications can sometimes be tricky to interpret. How would you interpret the following accuracy statement, provided by the OEM for the frequency accuracy of an electronic counter? It states that “Accuracy: ±1 count ± timebase error” (see Table 1).

Two pages of specs later, the timebase error is defined as shown in Table 2. Do you know whether your instrument uses a standard crystal or has Opt. 010 installed (oven oscillator)? After all, this changes the tolerances that will be applied to your instrument. This can be an area where you and your calibration provider interpret the specs differently, causing problems with your quality system. It can happen on complex specs, like this one, and it can happen with presumably simpler specs as well.

Conclusion
While there are many places where pitfalls can happen in the calibration buying process, I think you can get the picture from these examples of the importance of taking a cautious approach to any purchasing decision. If you’re getting cheap calibrations without data reports, remember that you’ll usually get what you pay for. If your calibrations are not preserving the integrity of your measurements (which impact your decisions about a product or process), then you’re throwing away your money because you’re not getting the quality you expect and your product is at some unknown level of risk. Wouldn’t it be better to make sure your product/service maintains the level of quality you intend, even if doing it the right way costs a little more? Whether cutting corners is an intentional effort or an oversight on your calibration provider’s behalf, it still costs you in the end.

The purpose of the traceable measurement chain is to make sure that your decisions about your product/process quality are valid. You can’t do that if you don’t know where you are with your instrument’s performance. Reduce your risks by digging in and asking more questions in any of your purchasing decisions. Take an active role on your company’s behalf to ensure they are getting the value for the money. Among these buying decisions is calibration service, because calibration not only has a direct impact on your product, it helps support the quality and reliability of the wind industry itself.

Roller bearings are essential parts in most devices and machines whose primary function is to transmit power and motion. They support the rotating parts within the stationary ones, and thus providing basic functionality in the machine. From the point of view of classical machine design, roller bearings are just another machine element that needs to be chosen according to the operating conditions, yet they offer a potential for stand-alone study and investigation.

Since bearings are tightly integrated into the functionality of the machine in which they operate, any improvement in the bearing operating conditions has a direct impact on the machine itself, its life cycle cost, and also in the environment. Design techniques that reduce the friction torque or roller losses of the bearing lead to improved efficiency, and advanced manufacturing processes and materials provide bearings with increased service life while simultaneously reducing the environmental footprint associated with manufacturing and recycling [1]. Additionally, life cycle cost analysis indicates that the largest part of costs associated with the operation of a machine is ownership costs, in contrast to acquisition or disposal costs [2]. Unscheduled maintenance costs are an integral part of ownership costs, and an adequate service life of the building blocks of the running machine is easily justifiable. Standstill costs due to faulty or overloaded bearings should therefore be avoided.

In this article we are mainly interested in the service life of roller bearings, pertaining to advanced calculations based on the knowledge of their inner geometry as defined in international standards. Predicting service life within an application-dependent accuracy is a major factor in a successful machine design.

Service Life Calculation
The notion of bearing service life is statistical in its meaning, and its origins can be traced back to the Lundberg-Palmgreen theory [3]. It is statistically determined that in a population of identical bearings which experience the same operating conditions (temperature, loading, speed, and lubrication), the number of bearings in that population which survive after a given service duration depends on the extent of this duration. The service life in which 90 percent of the bearings survive is termed the basic rating life L₁₀ and is generally given by bearing manufacturers. The higher the rating life, the more reliable a single bearing in operation will be in fulfilling its service requirements.

Until recently, the standard way to calculate the rating life of a bearing was based on ISO 281 [4]. The basic idea behind this method is to calculate, based on the actual radial and axial load, an equivalent load depending on the bearing type (radial or axial). The rating life is then calculated as

where i=r for radial bearings and i=a for axial bearings, C_i  is the basic dynamic load rating, P_i is the dynamic equivalent load and n=3 for ball bearings and n=10/3 for roller bearings. The basic dynamic load rating is given by the manufacturer, but can also be calculated according to the formulas given in [4] when the bearing geometry is known.

This method of calculation cannot cope with general loading conditions, bearing ring misalignments, or clearance in the bearing. Therefore, there is an increasing trend in using information of the inner geometry of the bearing, according to [5]. Due to the fact that these methods are computationally intense, the engineering practice is the use of specialized computer software such as the KISSsoft calculation program [6].

The basic principle described in the ISO/TS 16281 [5] is the complete analysis of the internal loading on each roller element, taking into account the inner geometry of the bearing, tilting, and/or misalignment and internal clearance. The rating life, calculated in accordance to [5], is termed reference rating life L_10r. The contacts between the roller element and the raceways in ball bearings are treated as point contacts, whereas in roller bearings as line contacts (Figure 1).

There are several factors that influence the reference rating life of a bearing. For a given external loading condition, the Hertzian pressure on the roller element contact surface is greatly affected by the cross-sectional raceway groove radius of the rings (for balls bearings) and the roller profile (for roller bearings). In general, increased Hertzian pressure leads to reduced service life. Any factor which affects diametral clearance—be it a press fit of the bearing on the shaft and/or housing, increased operating temperature and thus thermal expansion, or local deformations of rings as according to housing deformations—has a direct effect on the internal loading distribution, and therefore on the service life. Lubrication conditions and the potentially different reliability are encapsulated in the ISO life modification factor a_ISO and the life modification factor for reliability a₁ [4]. The modified reference rating life is thus given as

Industrial Application
It is helpful to illustrate how the above notions are used in industrial applications. For this purpose we use a typical case from the wind energy sector, as seen in Figure 2. It is a 2MW wind turbine shaft, rotating at 1750 RPM. Two cylindrical bearings are used to bear the significant radial load, and a deep groove bearing for any thrust component. The coupling of the shaft to the turbine blades on the right and the driven spur gear (220 mm pitch diameter) on the left are clearly visible. In the ideal situation, in which the bearings sit perfectly on their rigid housing, the modified reference service life of the critical radial bearings is around 99’000 h for the left one, and 736’600 h for the right one. Figure 3

The high service life of the right bearing is to be attributed to perfect loading conditions, which are rarely true in real-life applications. Manufacturing errors and deformations due to loading or assembly can cause the housing structure to diverge from its perfect theoretical alignment. This has a direct influence on the outer ring of the bearing, and the subsequent development of increased contact pressure between the roller bodies and the raceways. The deformation of the housing could be modeled as a misalignment error of the bearing outer ring. A small error of 0.3° loads the roller elements eccentrically and high contact pressures are developed. As a consequence, the bearing’s modified reference service life is drastically reduced to around 8’280 h. Table 1

If such misalignment errors can’t be avoided, a potential corrective action could be to manufacture the roller elements with an optimal profiling for that case. For the sake of simplicity and illustration we present the results of non-symmetrically profiled rollers, in which the left half of the roller employs a standard logarithmic profiling while the right half employs a profile value which is double that of the logarithmic one. The new pressure distribution is shown in Figure 4, where it is clear that the pressure peaks have been eliminated. In this case, the bearing life has been increased to around 26’200 h. The increased manufacturing costs of custom roller profiling could be balanced by the savings in maintenance work, related to the replacement of the short-lived bearing, and in any case the cost-benefit analysis is highly application-dependent.

Conclusion
Modern wind power plants need to achieve long service life with minor maintenance expense. Operation and maintenance can be accurately calculable, and with reliable calculation software, the chance of loss of bearings can be drastically minimized beforehand. With the implementation of the calculation method ISO/TS 16281 considering the aspect of service life, KISSsoft provides close-to-reality prognoses about future behavior, required stability, and service life of a gearbox. 

References:
1) Roller Bearing Contribute to the Environment”, Japan Bearing Association, Global environment committee.
2) “A Life Cycle Cost Calculation and Management System for Machine Tools”, Rafael Enparantza, Oscar Azkarate, Jose Zendoia, Proceedings of the 13th CIRP International Conference on Life Cycle Engineering, 2006.
3) “Roller Bearing Analysis”, Tedric A. Harris, John Wiley & Sons, Inc. 2001.
4) ISO 281, Roller bearings – Dynamic load ratings and rating life, 2007.
5) ISO/TS 16281, Roller bearings – Methods for calculating the modified reference rating life for universally loaded bearings, 2008.
6) KISSsoft, Calculation programs for machine design, www.kisssoft.ag.

Components of wind turbines are affected by asymmetric loads, variable wind speeds, and severe weather conditions which cause wind turbines to change their states. A typical wind turbine undergoes various states during its daily operations. The wind turbine states follow a certain pattern, such as: 1) turbine OK to run-up idling; 2) turbine online to maintenance/repair mode; 3) turbine weather conditions to external stop, and; 4) turbine OK to fault mode, and so on. A state change from normal turbine operations to a fault mode adversely impacts the performance of wind turbines and its components. Monitoring these states can greatly augment the maintenance operations.

Wind turbine monitoring can be done in two ways: condition-based, and performance-based. Condition monitoring requires installation of additional equipment/sensors to continuously monitor relevant parameters in real time. Performance monitoring utilizes historical data for prediction of turbine performance. Since performance monitoring relies on the existing data, there is no additional cost to the wind farm operators. Analyzing the historical data through data-mining techniques is a promising approach to performance monitoring. The current SCADA systems already record wind turbine parameters. An improved SCADA system coupled with data-mining algorithms can be useful in identifying critical performance indicators of wind turbines. This article demonstrates application of data-mining techniques to prediction of fault modes of a wind turbine.

Turbine States Information
In addition to wind turbine parameters, a SCADA system records the states of wind turbines.  Consider 17 possible states in which a turbine can be found. The state “turbine in fault mode” (state No. 17 in Table 1) can usually be expressed in more than 400 ways. Components of a wind turbine such as wind turbine blades (e.g., blade angle asymmetry), turbine yaw (e.g., yaw runaway), wind turbine generator (e.g., generator brush worn), and wind turbine gearbox (e.g., gearbox over-temperature) can be affected. Using domain knowledge, the states listed in Table 1 can be broadly categorized into four main groups (Figure 1). At the top level, prediction of fault modes of wind turbine is crucial in identifying an actual fault.

Figure 2 displays the distribution of wind turbines states. The values are averaged for a wind farm over a year period. Almost 90 percent of the time the turbine is operating normally. However, faults make up 6.82 percent of the total time. Any effort to minimize the frequency of faults would enhance wind turbine availability.

Developing Prediction Models
In this section, data-mining models are developed for predicting turbine states. The data collect by the SCADA system is sampled at the frequency of 0.1 Hz. The faults that have occurred at 17 wind turbines in a three-month period are plotted in Figure 3. Based on the frequency of fault modes, three wind turbines are identified (labeled as A, B, and C in Figure 3). Data from turbine A is used for training and testing data-mining algorithms, whereas turbine B and C are used for performance evaluation.

Parameters for States Prediction
A SCADA system records various parameters, which can be categorized into: 1) non-controllable parameters such as wind speed and wind deviations; 2) performance parameters such as power output and rotor speed; 3) vibration parameters such as tower acceleration and drive train acceleration, and; 4) temperature parameters such as gearbox temperature and generator temperature. It is important to mention that not all the parameters are causing wind turbines to change its states. Therefore, we need to identify a set of parameters that impact the wind turbine states. This is the place where algorithms build using data-mining techniques are useful. Data-mining algorithms use statistical measure such as information gain and correlation coefficient to identify relevant set of parameters. Table 2 shows 10 parameters relevant to wind turbines states (data from turbine A), identified by three different data-mining algorithms. The techniques used are wrapper with genetic search (WGS), wrapper with best first search (WBFS), and boosting tree algorithms (BTA). For detailed description of these techniques refer to [1-3].

Data-Mining Models
A combination of relevant input parameters (found in previous section) can be used to develop prediction models for state faults. The evaluation of data-mining algorithms is based on the prediction accuracy of turbine states namely turbine OK (TO), weather downtime (WD), maintenance downtime (MD), and fault mode (FM) (Figure 4). The diagonal elements (e.g. TPTO, TPWD) are correctly predicted turbine states, whereas non-diagonal elements are wrongly predicted (e.g., FPFM, MD: turbine in fault mode is predicted as turbine in weather downtime).

Data-mining algorithms namely random forest algorithm (RFA) [4] is used to build eight prediction models at various time stamps, with the maximum prediction length of 5 min. The selection of data-mining algorithm is based upon their performance on training data at time stamp t. The accuracy was found to be in the range 81-99 percent for all turbine states (Table 3).

Robustness of Data-Mining Models
In order to check the robustness of developed models, various unseen faults were tested. The aim is to check the response of the model when unobserved faults are encountered. Figure 5A (a-b) Figure 5B displays the distribution states for turbines B and C plotted over a three-month period. The number of faults varies across turbines, however, turbines are found to be operational most of the time.  The results shown in Table 4 and Table 5 illustrate the response of data-mining algorithms on turbine A and B. Except of the gearbox-related faults, other fault modes of wind turbines are correctly identified.

Conclusion
The parameters recorded by SCADA systems can be useful for monitoring purposes. Using prediction models derived by data-mining algorithms the states of wind turbines can be predicted ahead of the time, which can be helpful in maintenance planning. The models built using data-mining algorithms can be integrated with current SCADA system to enhance performance monitoring of wind turbines. Acknowledgement: The research reported in the paper has been supported by funding from the Iowa Energy Center, Grant 07-01.  

References:
1) Kohavi, R. and John, G.H., 1997, “Wrappers for feature subset selection,” Artificial Intelligence, 97(1-2), pp. 273-324.
2) Sbihi, A., 2007, “A best first search exact algorithm for the Multiple-choice Multidimensional Knapsack Problem,” Journal of Combinatorial Optimization, 13, pp. 337-351.
3) Kudo, T. and Matsumoto, Y., 2004, “A Boosting Algorithm for Classification of Semi-Structured Text,” EMNLP.
4) Liwa, A. and Wiener, M., 2002, “Classification and regression by random forest,” R News, 2(3), pp. 18-22.

When the fourth-largest wind farm operator in China increases its installed capacity by 50 percent within one year, it indicates the company knows its business. And one of the business aspects that all wind farm operators need to take care of is “turbine reliability,” or keeping the turbines turning as much as possible and keeping maintenance activities down to a minimum. Reduced or controlled maintenance is always a valuable contribution to bottom line profitability, but when you’ve got 2,000MW of installed capacity, the target by end 2010 for Guohua Energy Investment Company—based in Beijing, China—then you are talking about huge amounts of money.

That is why Zhou Weihua, deputy general manager of the Production Department, made a decision in 2008 to evaluate his maintenance strategy and the technology to be applied within that strategy. A critical decision was to look for the most reliable and experienced way to determine the health of his turbines in the fastest and most effective way. Figure 1

Investigating Online Monitoring
With a few different turbine designs and sizes out in the field, he needed the best possible way to get reliable and regular information about the general condition of selected turbines of these designs and types. At the same time he wanted to identify any typical negative trends that might occur in the various, but differing, field operating conditions that the turbines would be deployed to. Not being satisfied with his previous assessment process of only using temperature readings of gearbox oil and generator bearing housings, he wanted to investigate online monitoring, where vibration signals from critical components are collected 24 hours a day and analyzed to determine exactly if and where problems are developing, and how severe any problems were. This type of monitoring would also provide valuable knowledge that could allow critical maintenance to be planned and avoid unexpected and very costly breakdowns. He knew that this type of condition monitoring is widely used in the United States, Europe, and Australia, etc.

Gaining Firsthand Experience
His first step was to make a deep investigation of what technology was available, and the quality and knowledge of the suppliers. These investigations covered all the key suppliers of such technology, but contacts with other wind farm operators in China quickly brought him to consider SKF and its turbine monitoring system for further discussions. “Of course I knew of SKF,” says Zhou, “they are the number one bearing specialist in the world with expert knowledge in many industries over many years in many countries, also here in China. And I had heard of the SKF WindCon turbine monitoring system, but had never used it before.

“After direct contact with SKF I heard that they had recently been successful with remote online monitoring, where data from the monitored turbine in the wind farm were transmitted by Internet to their specialists for analysis,” he says. “In order to get firsthand experience of their remote online capability I visited Intelligence Centre Wind (ICW), their major wind turbine diagnostic center in Hamburg, Germany.” Figure 2

At the time of Zhou’s visit to SKF, ICW was receiving and processing data from more than 600 turbines from across Europe and also from the United States. “In Hamburg I saw examples of almost all the capabilities of an SKF WinCon system, because there were so many systems deployed in so many wind farms,” he says. “The level of detail that the system could give was clearly demonstrated, and the explanations by the people making the analysis clearly indicated that they knew all about wind turbines and the critical machinery. The Hamburg visit gave me the confidence to set up a field trial on some turbines in Guohua wind farms in 2009.”

Farm Field Trials
The field trial involved 11 SKF WindCon systems to be deployed across three wind farms; in Jiangsu province, Shadong province, and Inner Mongolia. Around the time SKF were installing their systems in the Guohua turbines, it had opened a Remote Condition Monitoring Centre in Shanghai. Having SKF experts so close was an added benefit to Zhou, and he was anxious to get the systems operating and see what the results would be.  The data was collected on a local server at each wind farm and transmitted to an SKF server in Shanghai for analysis. Any immediate emergency situation would be reported at once, but an “emergency situation” was not detected at start-up of the monitoring, and a report was issued once per month to the wind farm site managers and Zhou. The reports contained the detailed vibration spectra for the key components being monitored, together with a brief statement that summarized the analysis of the data in terms of the condition of the component. But there was also a “report grid” that indicated the components being monitored, and each one had one of three colors; green, yellow, or red to give immediate indication of the condition, with green being OK, yellow being a slight to major deviation from the “acceptable levels or trends,” and red being something requiring immediate attention. Figure 3

The first results were good, because Zhou could “see” for the first time what was really happening with his critical machinery. The SKF WindCon systems delivered the vibration spectra, similar to those he saw in Hamburg, and the SKF experts in Shanghai interpreted the data in terms of the condition of the bearings, potential misalignment, and gear damage, etc.

During the six months prior to January 2010 the SKF WindCons provided good information on the turbines, allowing Zhou to determine if and what maintenance would be required. In that period one of the installed SKF WindCons identified a severely damaged generator gearbox bearing at the Inner Mongolia wind farm.  Figure 4

Success and Further Deployment
“The SKF WindCons did their job,” Zhou says. “They kept me informed, and the early identification of the severity of the damaged bearing in Inner Mongolia was enough to allow us to plan for replacement at minimum cost and disturbance to the planned electricity supply from the turbine.”

Since the tests Guohua has installed another 58 SKF WindCons, and an additional 280 were ordered for new turbines that went into service in 2010. This is the start of a mass application of remote monitoring, and Zhou says that “we have some knowledge regarding vibration analysis within our own maintenance teams, but I intend to build up more knowledge within the Guohua organization and set up our own remote monitoring centre in Beijing for our turbines. A lot of this knowledge and assistance will come from SKF as they provide us with their remote diagnostic monitoring during the coming years.”