Wind turbines are dotted across the countryside and coastlines. Photo: Jason Tranter

If you asked the average vibration analyst what type of situations they least like to deal with, their checklist might contain:

1. Variable speed and load from one test to the next

2. Variable speed and load during the actual test

3. Difficult and limited machine accessibility

4. Complex gearboxes – planetary gearboxes being the worst

5. Very low speed shafts

Well, you have just described a wind turbine. The wind conditions are constantly changing, so each vibration measurement taken could potentially be at a different speed and load condition. And what is worse is that the speed can vary as the blades rotate. Even the nacelle (the house at the top of the tower) will rotate as the wind direction changes. And one more small challenge is that the whole structure can vibrate and resonate due to the construction of the tower and nacelle.

Therefore, routine monitoring by vibration analysts visiting the wind turbines on a routine basis is almost out of the question. That’s not to say that it is not done – it is simply very, very challenging to acquire data that can be compared to previous readings in order to detect changes in the patterns.

Although we will discuss the technical issues, you cannot skip the challenge associated with accessing the wind turbines. At best they are on land not too far from civilization. At worst they could be out at sea. And for safety reasons then you have to stop the turbine, climb up the tower, set-up your measurement equipment, climb back down, start the turbine, record the data, stop it again, climb back up to retrieve the instrumentation, then climb back down again – and re-start the turbine. It is a long way up to the nacelle – you had better be fit!

Variable speed and load

One of the key requirements for successful vibration analysis is to be able to compare the current readings to either a previously collected set of readings, or to a set of alarm limits. We want to see how the vibration patterns have changed. In a standard power station, the majority of the machines will run at the same speed and load from one test to the next. Comparisons with older data are easy, and alarm limits can be generated based on experience with the machine, or based on statistical analysis of the history of data. But it is not that easy with a wind turbine.

As the wind speed varies, the load on the blades, shaft, bearings, gears and generator will change. The speed of the machine will also change. The result is that the peaks in the spectrum will not line-up with peaks in previous spectra, and the amplitudes of peaks are no longer comparable. Not only does the load affect the amplitude of the peaks in the spectrum, natural frequencies will either cause the measured vibration amplitudes to be higher or lower than when the machine was running under a different speed or load.

It is certainly possible to “order normalize” the spectrum, so that the speed-related peaks in the spectrum will be aligned, but that does not address the changes in amplitude.

The solution is to define one or more bands of operation where spectra (and time waveforms) collected within that band can be deemed “comparable”. The “band of operation” may be specified by the RPM of the input shaft, or the power generated by the turbine, or perhaps another parameter. You will then need to wait until the required conditions are met before the vibration measurements are acquired. Alarm limits can also be defined for that “band of operation”.

Variable speed during the measurement

When the analyzer (or monitoring system) acquires the “time record” that is used to compute the spectrum (via the FFT calculation), it is assumed that the machine being monitored operates at a constant speed during that test.

For example, if you acquire a 1600 line spectrum with an Fmax of 1000 Hz, the analyzer will acquire 1.6 seconds of vibration data in order to compute the FFT (for just one average). A 1500 RPM generator will rotate 40 times during the test, but the 15 RPM input shaft will rotate just 40 percent of one rotation. In order to capture 10 rotations, we need an Fmax of 40 Hz (with resolution set to 1600 lines), and the measurement will take 40 seconds!

If the speed of the wind turbine varies during the test, the peaks in the spectrum can blur – the peaks will be wider than they should be, and the amplitude of each peak will be reduced. And this blurring effect may not be consistent from one test to the next. (Note: The blurring effect will be more noticeable at higher frequencies, that is, the harmonics will blur more than the fundamental.)

Therefore, depending upon the nature of the turbine, and the wind conditions, this effect can either be tolerated, or the “order tracking” technique must be employed. Either the once-per-rev tachometer signal must be fed into the analyzer (with an internal “tracking ratio synthesizer”) such that the analyzer varies its sample rate in proportion to the RPM, or a shaft-encoder must be used to generate a “pulse train” that contains, for example, 360 pulses per rotation of the shaft which is used to control the analyzer’s sample rate.

Note that we could also use time synchronous averaging, where the collection of the time records are synchronized by the tachometer and averaged together and then the FFT is calculated. This method however, requires lots of averages (i.e. lots of time), it does not deal with speed variation that occurs during the acquisition of the time record, and all non-synchronous vibration is averaged out – including key data related to bearing faults. It can, however, be used effectively to detect tooth damage in the gearbox.

Gearbox measurements

There is one more challenge when monitoring gearboxes, planetary gearboxes in particular. In an ideal world the accelerometer would be placed close to the bearing and/or gear of interest. However, not only do these gearboxes have a large number of bearings and gears, it is difficult to get an accelerometer close to certain bearings, such as planet bearings.

When analyzing spectra, either conventional spectra or demodulated spectra (or PeakVue, Shock Pulse, etc.), it is necessary to resolve three issues:

Condition monitoring measurement of a wind turbine.

1. Computing the speed of each shaft, and the gear mesh frequencies, can be quite a challenge with planetary gearboxes.

2. Computing the bearing frequencies will be very complicated due to the large number of bearings and different shaft speeds. Both jobs are made even more difficult if the manufacturer is not willing to provide the details of the bearings used and gear ratios and tooth counts.

3. The amplitude of the vibration measured when a planet bearing begins to fail, for example, will be lower than the vibration from a bearing in contact with the gearbox case due to the transmission path involved.

It should be noted that time waveform analysis, oil analysis and wear particle analysis should always be employed on gearboxes. Using spectra alone does not provide the earliest warning or the most complete picture.

Almost all of the vendors of portable data collectors and analyzers now manufacture on-line monitoring systems designed specifically for the wind turbine application. There are an awful lot of wind turbines, and each one requires its own monitoring system. These vendors all recognize the challenge and the opportunity.

Systems are designed to monitor the speed of the turbines, and other process parameters, so that they can correctly determine when the turbine is operating in the pre-defined “band of operation”.

In fact, many of these systems can define multiple “bands of operation”. Each band will have its own set of alarm limits, and all readings are tagged with their band of operation so that graphical comparisons can be performed.

It is important to have multiple bands for two reasons:

1.Unless the weather and load conditions are reasonably constant, the turbine will not be operating in any one band for a large proportion of time. By defining multiple bands, the system will monitor and check the turbine far more frequently.

2. The bearings, gearbox, and generator will react differently under different speed and load conditions. It is therefore very helpful to monitor the machine-train during the majority of operating conditions. For example, a problem with the support structure may only be detected when the turbine is operating at highest load, and skidding in bearings may only occur when they are lightly loaded.

The challenge with all on-line monitoring systems

All on-line monitoring systems face a number of challenges that can limit their effectiveness, but these challenges are compounded when applied to wind turbines. I have already discussed the issue related to varying speed and load, but let’s take a look at some of the other challenges.

One of the most critical decisions is selecting the number of sensors that should be installed on the gearbox, generator and main bearings, and selecting their location. Every sensor costs money, and it requires another channel in the monitoring system. And when you multiply these additional costs with the number of wind turbines, you can see that it is a very sensitive issue.

As with all vibration monitoring applications, it is essential that the monitoring system can at least acquire enough data to warn when the vibration levels are increasing – even if there is not enough data to actually diagnose the problem remotely. But when monitoring large planetary gearboxes, the spectral data can be very complex as discussed previously.

Knowing the failure modes of the turbine can help immeasurably. If you know which gears and bearings are most likely to fail, then you can position the accelerometers accordingly.

The “central monitoring service” is the group of people who will respond to the alarms, analyze the data and make final recommendations. It is essential that this group has access to the required data and has the experience to make recommendations. Obviously a communication link must be established with the wind turbine monitoring systems.

Centralized or de-centralized

The monitoring system must not only acquire data when the turbine is operating within pre-defined bands, it must compare the data to alarm limits and take the appropriate action. There are at least two approaches: perform all of these operations within the system that is installed in the nacelle and communicate directly with a central monitoring service, or install a more sophisticated system centrally within the wind park and use it to communicate with both the wind turbine monitoring systems, and with the central monitoring service. Many wind farms have a wired or wireless network, and the monitoring system may be allowed to tap into that network.

Windfarms with CMS

The effectiveness of alarm checking software

Many vibration analysts running ‘normal’ vibration monitoring programmes do not have an effective set of alarm limits set up for their machines which would allow them to run an exception report that provides useful, actionable information. Instead, healthy machines falsely trip the alarms, and machines with problems go undetected. The solution has been to manually analyze each and every measurement. This is not possible when performing on-line monitoring.

It is therefore very important that the alarm limits are set up carefully, and they need to be refined frequently. Too many on-line monitoring systems generate “thousands of alarm exceptions” – as a result, faith in the system is lost. There are methods that can be used to set up effective alarm limits, such as statistical alarm generation, but that will need to be covered in a separate article.

It should be noted that there is an ISO standard under development, ISO/FDIS 10816-21, that provides alarm limits for wind turbines with gearboxes, and ISO/CD 16079-1, that provides general guidelines for condition monitoring and diagnostics for wind turbines.

New wind turbines are being installed at an amazing pace, and while some of the earlier reliability problems have been resolved, there is no doubt that reliability will be an on-going issue. Condition monitoring technologies such as vibration monitoring, wear particle analysis, and performance monitoring will play a very important role in the viability of wind farm operation. As long as monitoring system vendors and wind turbine manufacturers continue to improve their designs, focus on reliability, and share information, renewable energy from wind power will continue to grow as a source of affordable and clean energy around the world. 

Jason Tranter

Managing Director,

Mobius Institute, Australia,

jason@mobiusinstitute.com