London, UK [Renewable Energy World Magazine] Wind power installations present asset managers and personnel concerned with wind turbine operation and maintenance with a unique set of problems. Conventional combustion or hydro plants exploit a predictable resource in the controlled circumstances of a turbine hall. Wind power is generated from a highly variable resource using a much larger number of turbines installed in remote locations with access issues. If asset managers were sports agents managing a portfolio of athletes, conventional combustion and hydro plant would be long distance runners, with the stresses and health issues associated with long-term continuous stable effort, while wind turbines would tennis pros, with a completely different and arguably more arduous set of stresses and health issues arising from stopping and starting, turning left and right, and responding rapidly to a situation that is constantly changing. It is clear the same approach cannot be adopted when maintaining the health, productivity and longevity of these fundamentally different kinds of asset.
The typically large number of turbines in a wind farm, their dispersed and remote locations and the presence of key turbine systems at the top of a tall tower present access issues which mean the consequences of component failure can be severe in terms of downtime and lost revenue.
Emphasis has therefore historically been placed on maintaining high levels of asset availability. However the efficiency of the turbines when available – the power performance – must also be inspected and scrutinised. This is not just because of the revenue-affecting issues that may be manifested as anomalous under-performance, but also because the signatures of incipient component failure can often be detected in poor performance and inspections and other interventions scheduled to avert potentially catastrophic downtime.
Scada DATA: The Key to Performance Monitoring
The performance of a wind farm can be affected by a number of factors. Shortfalls in performance or ‘performance deficits’ can arise from environmental effects, such as icing or turbulence, which can degrade the aerodynamic performance of the turbine. The cause can also be intrinsic to the turbine. The control system must intervene to pre-empt a potentially catastrophic overspeed situation by feathering the turbine blades and, if necessary, applying the mechanical brake. It may also be tuned to vary the pitch to avoid noisy resonant conditions. Control systems designed as a compromise between several conflicting constraints often have unforeseen and undesirable characteristics, possibly resulting in a significant performance degradation in a particular environment.
A wind farm’s existing SCADA (Supervisory Control and Data Acquisition) data stream is a valuable resource, which can be exploited by the operator to observe, and hence optimise, the performance of the wind farm. This is the basis of what we term ‘performance monitoring’.
Performance monitoring can complement condition monitoring. It can also be used independently of condition monitoring. Because it uses existing data, there is no installation cost, and hence no financial risk to the wind farm operator. Condition monitoring directly assesses the status of turbine systems and sub-systems using sensors such as accelerometers, strain gauges, and oil particle counters. The signals from these instruments are monitoring continuously. Performance monitoring is a fundamentally statistical endeavour, which uses the SCADA data that are already routinely recorded to investigate the relationships that describe how one parameter – such as output power – varies in response to another – such as input wind speed. The observed relationships are built up over time as the wind varies and the turbine responds to this variation in a manner characteristic of its condition. As performance monitoring uses data already acquired, no additional instruments need to be installed which would otherwise put warranties at hazard, and incur downtime and capital expenditure.
Unlocking the Secrets of the Power Curve
The ultimate success of a wind farm is determined by how much power it produces. The guideline for its power output is the warranted power curve, which shows the production expected for a given wind speed at a hypothetical site. The power curve is based on an ideal site, but degraded so that the turbine manufacturer has a reasonable chance of meeting it most of the time: after all, the manufacturer is not likely to warrant performance it is unlikely to meet.
Most operational wind farm sites are not ideal. Even offshore, where local terrain is not an issue, the power output of downwind turbines will be affected by wake effects from upwind turbines. Some turbines on a given site may struggle to meet their warranted performance, others may comfortably exceed the warranty. Maximising the performance of each turbine is not an easy task.
Factors which affect performance can be grouped into two categories. The first is conditions, which can include: wind shear, wind veer, turbulence, wake effects and icing. The second category is turbine operation, which can include: blade condition, turbine suitability, control algorithm, wind farm layout, maintenance and downtime, and error and alarm states.
With all these effects interacting, it should come as no surprise that operational power curves can depart radically from the warranted curve. A number of companies have developed software tools to aid in the rapid characterisation and quantification of these departures: this is the domain of performance monitoring, and is well outside the scope of condition monitoring. Some of the features observed in operational power curves are shown in Figures 1, 2, 3 and 4. Figure 1 is typical of many operational power curves seen on actual wind farms. Each point on the graph represents the measured power output of the turbine, at a given wind speed, over a 10-minute period. Large numbers of points, at wind speeds above cut-in, show zero power output. It is important to note that the data used to generate this graph have been filtered to show only points where the SCADA’s own record of the turbine’s status indicates that the turbine is available and generating for the full 10 minutes.
A power curve like that of Figure 1 can serve as the starting point for a highly focused diagnostic exercise, leading to a rapid identification of the cause of a power deficit. As part of a proactive O&M strategy, this can contribute to a dramatic improvement in energy yield.
Figure 2 shows a power curve where the measured power output, at a given measured wind speed, is unfeasibly high. In this case, it pointed to defective anemometry. Figure 3 shows the power curve of a turbine that is clearly performing significantly below the warranty. In this case, the power curve suggests possible degradation in the aerodynamics of the blades due to icing or damage. Figure 4, meanwhile, shows the power curve of a turbine that is performing dramatically below the warranty. In this case, there is clearly a defect in the control system.
Monitoring Can Cut Losses
Monitoring the performance of a wind farm and its constituent turbines is about optimizing the use of assets. Analysis of the power curve can show when a turbine is underperforming. In an extreme case, where several turbines on a farm exhibited a power curve like that shown in Figure 4, losses were of the order of £100,000 (US$150,000) a month as they coincided with a period of unseasonably high wind speeds. And, they went unnoticed for several months. Routine performance monitoring can prevent such losses.
Cross-checking turbine performance against error and alarm codes can identify, for example, where the power output is being curtailed because of an overheating bearing. Used alongside condition monitoring, such evidence can strengthen the case for preventive maintenance. In the absence of condition monitoring, such evidence may be the first indication that something is wrong. A typical display produced by performance monitoring software showing the instance and duration of events and alarms is shown in Figure 5.
This allows the analyst to see which events and alarms coincide with periods of anomalous power performance, to observe any cascades of inter-related events when performing root-cause analysis, and attribute yield deficits and revenue variance to specific issues.
Routine performance monitoring over time builds up a body of evidence to characterise past performance and predict future performance.
Wind farm operators are familiar with the term ‘lost yield’ – it is a measure of the energy output of a turbine that is lost due to down-time. This is only one component of the aggregate yield deficit against the potential yield; another component is what might be described as ‘performance loss’ – the loss due to degraded performance.
Routine measurement of performance loss allows the operator to quantify the revenue loss associated with power curves such as those of Figure 1, Figure 3 and Figure 4. Negative performance loss can highlight instrumentation faults, such as that evident in Figure 2.
Associating performance loss with errors and alerts allows the operator to develop a maintenance strategy that minimises revenue loss, rather than downtime.
What are the benefits of performance monitoring? Consider a hypothetical wind farm with the following characteristics: 50 MW rated power, 30% capacity factor, a power price of £75/MWh (US$124.5/MWh), and revenues of approximately £10 million (US$17 million) per annum.
A 1% improvement in production would yield £100,000 ($150,000) more annual revenue; a 3% improvement would yield £300,000 ($450,000). This represents a substantial benefit for modest cost.
Software Makes the Process Possible
As shown above, performance monitoring is key to understanding and optimising turbine performance. However, the analysis of SCADA data requires time, skill and expertise, which can be too expensive to provide on a routine basis. This is the motivation behind the development of performance monitoring software, such as SgurrEnergy’s sgurrtrend. Performance monitoring software can automate the processing of data, and provide data visualisation through a number of views that support trained personnel in the rapid assessment of a wind farm’s performance. Such software can provide the customer with a periodical, perhaps weekly or monthly, review of a wind farm’s performance, as well as guidance on how to improve that performance.
Power curves such as those given above are very useful tools to analyse the behaviour of underperforming turbines. But the first step in rapid performance assessment is the rapid identification of which turbines are underperforming, and when. SgurrEnergy have been exploring a range of metrics to assist in this.
One promising approach is illustrated in Figure 7, left. Each point on this graph characterises the performance of each turbine for one week. Experience has shown that turbines whose behaviour can broadly be desribed as ‘normal’ produce points on this graph that lie on a straight diagonal line, known as the ‘main sequence’, seen here towards the bottom left of the graph. Clicking on any point in the performance metric graph will bring up the power curve corresponding to that turbine for that week. This is a powerful tool for rapidly identifying where significant performance losses occur.
Once a suspect turbine and time interval have been brought to the analyst’s attention by the performance metrics view, the power curve can be examined in detail, as shown in Figure 6.
The time interval can be edited; warranted or other benchmark power curves overlaid; different time intervals compared, either as scatter-plots or as interpolated curves, and by clicking on any point on the graph, the analyst can quickly view errors and alerts that were active at the time that point was generated. This is another powerful tool in identifying the causes of degraded performance.
An important guideline in the development of performance monitoring software has been the idea of data visualisation, of presenting data in a variety of views to make the analyst’s job easier. The performance metrics view is one starting point for analysis, and is well suited to identifying short-term departures from optimal performance. The yield deficit summary view is another starting point, which gives the analyst and the wind farm operator an overview of the wind farm’s performance over time.
Some care is required in interpreting these graphs — as multiple error codes may be active concurrently — and so some skill and experience are needed to establish a causal relationship between error codes and yield deficit.
Nevertheless, these graphs help a great deal in finding which error conditions are associated with large revenue losses. Other ways of presenting data are being explored.
Learning the Lessons and Looking to the Future
SgurrEnergy’s experience of reviewing the performance of wind farms suggests that improvements of 1%–3% in revenue are readily attainable. With suitable software tools, experienced analysts can provide a monthly review of a wind farm’s performance, and guidance on how to improve it.
Performance monitoring software can present data in a variety of ways to support analysts in:
- Rapidly identifying where performance shortfalls occur;
- Associating those shortfalls with errors and alerts; and,
- Quantifying those shortfalls to target the efficient use of O&M resources.
The exploration of novel ways of presenting data continues. Routine performance assessment has no initial capital cost, no down-time for installation, and hence no risk. In 2009, for the first time more MWs of capacity came out of warranty than were installed.
As the focus shifts from getting megawatts of capacity into the ground and more emphasis is placed on the efficiency of getting megawatt hours of energy out of the wind, we will see an increasing reliance on performance monitoring techniques. In addition, poor performance today may indicate poor availability tomorrow, and the signature of component failures that impact asset availability are often evident in the data analysed to assess power performance.
The methods developed to perform these assessments have now achieved a sufficient degree of sophistication that they can be performed rapidly, for example by using performance metrics to rapidly focus the analyst’s investigation and obtain the greatest value for money and cost effectiveness of their time for the client.
Software that extracts the critical information inherent in the vast quantities of performance data and presents it in an easily interpreted format leaves the analyst to concentrate on what they do best – bringing their experience and expertise to bear to best effect.