‘Monitor’ of an entire offshore wind farm

For a wind farm at sea, the maintenance and repair costs are responsible for almost a third of the kWh price. It is logical that wind farm managers wish to know as exactly as possible which turbines require maintenance and when. How can this be achieved? “By determining the exact load on a few of the turbines in a wind farm. This information is the basis for a computer model that predicts the load on the other turbines,” Tom Obdam of ECN Wind says confidently. This is something he has taken for granted for several months now, since demonstrating it with the Flight Leader project.

Every medal has its reverse side. The advantages of wind farms at sea (more unimpeded wind, no noise nuisance and no pollution of the horizon) are offset by the disadvantage of difficult access to the turbines, which means high maintenance and repair costs. Ideally, maintenance would be carried out just before a component is worn out. But you then need to know when that time has arrived, as the maintenance team at sea is extremely dependent on the weather. Postponed maintenance leads to emergency repairs and these are very undesirable.
At ECN Wind Energy, Tom Obdam carried out research into the central issue of the maintenance problem: can the measurement of the load on one or two twee turbines be considered representative for all the turbines in the wind farm? Obdam intends to use the answers to draw up a maintenance schedule that takes into account the wear of each individual wind turbine in the farm. This would be based on standard measurements from the control system of each turbine. In other words, Obdam is striving for accurate, tailor-made maintenance.

Neural networks
An important resource for this study is ECN Wind Energy’s test farm in the Wieringermeer. The turbines lined up here are fitted with all sorts of instruments for scientific measurements. Naturally, this includes the standard sensors (SCADA parameters) that are used for regulating the turbine. Obdam examined the relationship between the scientific load meters and the customary measuring instruments. “The objective was to see whether the SCADA signals can be used reliably for predicting the mechanical load on a wind turbine. The advantage of this is that you do not have to fit commercial wind turbines with expensive additional load meters.”
To carry out this task, the ECN researchers set up a neural network. This self-learning software technique proved very satisfactory for linking the mechanical load on the turbine to the control signals. It did this by sampling the SCADA instruments every ten minutes. Obdam then examined whether the network predicted the load on another ECN turbine with sufficient accuracy. “The result was extremely encouraging. But it did become obvious that we needed to make a very clear distinction between turbines standing in free-stream wind conditions and turbines standing in the wake of one or more of the other turbines.”
The project started at the beginning of 2008. The two ECN turbines in the wind farm were adapted to make them suitable for the investigation. The required software was written and the test phase went ahead. This showed that the data collected from a turbine standing in free stream cannot be used directly to predict the load for a turbine with a turbulent wind flow. “The great advantage of a neural network was then demonstrated,” Obdam explains. “You can ‘teach’ the network artificially what it is like working in turbulent conditions. If you do this and then get the network to predict the load again, the results become sufficiently accurate, although the quality of the prediction is not as good as for the turbines in free stream.”

Further refinement
A demo version of the Flight Leader software is now ready for use, but the work is not finished yet. Obdam and his colleagues will continue to refine the program by means of additional research using the turbines in ECN’s test farm, preferably in collaboration with a turbine manufacturer or wind farm owner. A practical test is also in the pipeline. After all, the program is intended for offshore wind farms but all the results have been obtained via onshore research. So how can we find out to what extent the results apply out at sea as well? “This is indeed a problem,” Obdam agreed. “Luckily, the oldest offshore wind farm in the Netherlands, NoordzeeWind’s Offshore Windpark Egmond aan Zee, has two turbines with load meters. We may use the data from these instruments for follow-up research. This is a great opportunity! Unfortunately, the turbines are both at the edge of the farm. The ideal situation would be one turbine at the edge and one turbine in the middle of the farm.”
The research was started to satisfy the need of managers of offshore wind farms to keep down their maintenance costs. But as a market-orientated researcher, Obdam does not rule out the possibility of interest from manufacturers of wind turbines. “Turbines that are located behind the first row do not get the full force of unimpeded wind. Their maximum mechanical load is always lower, therefore, but it fluctuates more than that of the turbines in full stream at the front. This lower but very variable load can tax the machine more heavily over time than a higher but uniform load.”
The expertise of the turbine manufacturers shows a gap with respect to this phenomenon, and they would like to see it filled. The Flight Leader research can yield important information about the variation of the load on turbines in wind farms. This may lead to new constraints, Obdam pointed out. “Who knows, the design requirements may be different in future for turbines in the middle of a wind farm and turbines at the edge.”

Contact
Tom Obdam
ECN Wind Energy
Tel.: 022 456 4395
E-mail: Tom Obdam 

Info
Click here to read or download Tom Obdam’s contribution about Flight Leader to the European Wind Energy Conference 2009.  

This ECN-Newsletter is free for publication, under the strict condition that the source is mentioned: www.ecn.nl/en/news