There are moves in the wind energy industry to extend the operational life of wind turbines beyond their nominal design life, typically 20 years. There are many issues to be dealt with to examine the technical risks and commercial returns associated with life extension, one of which is the continuing performance, reliability and life of the many rolling bearings used in wind turbines.
Historically, rolling bearings have caused significant reliability issues, the most notable of which are the WEC/WSF failures that afflicted certain gearbox and main bearings. With the advent of sophisticated pitch systems to control dynamic loads on blade, main shaft and gearbox bearings (as well as structural loads on the tower and nacelle), pitch bearings are subjected to increased cumulative angles of oscillation compared to earlier designs.
There are international standards for the design of wind turbine machinery, for example, IEC 61400-4 for gearboxes. Presently there are no such standards for pitch and yaw systems though IEC 61400-1 section 9.8 gives some overall guidance. Within IEC 61400-4, design of rolling bearings is covered by ISO 281, “Rolling bearings – dynamic load ratings and rating life”. This proposes 90% failure probability for each bearing, that is, the L10 life. However, higher reliability may be agreed between supplier and customer, especially as series reliability becomes important for a gearbox containing many bearings.
In terms of steel quality, IEC 61400-4 uses DIN ISO 683-17 that in turn references ASTM E45. It also refers to failure modes other than surface and sub-surface initiated fatigue that are covered by ISO 281. Amongst these are adhesive wear, corrosion, excessive voltage, current leakage, overload, indentation from debris and indentation from handling. IEC 61400-4 also specifies appropriate levels of lubricant cleanliness and temperature.
IEC 61400-4 gives recommended values for maximum contact stress in the range 1300 to 1650MPa (depending upon bearing location), but calculated from the Miner’s sum dynamic equivalent bearing load (derived from the gearbox torque spectrum).
During site development, meteorological data would have been gathered for assessment of wind speed and direction. Data would have then been used to estimate bearing loads for wind turbine selection and design. This data could also be gathered throughout life after construction, including torque loadings. It is therefore possible to compare design with actual field data so that a judgment can be made on the rate of life consumption. This data can then be used to determine the margin of safety against design assumptions.
As stated above, the reliability level assumed in bearing life estimates can be agreed between supplier and customer, providing it is better than 90%. If higher reliability has been agreed and bearing life estimates meet the desired level (for example 20 years), this will be an important factor to consider in life extension. Effectively, bearings would have been “oversized” to reach that level of reliability. Life estimates of L5, L2 or even L1 will result in better series reliability for a gearbox containing many bearings.
Towards end of design life, judgments on life extension will draw upon the condition of the various bearings. Gearboxes generally have condition monitoring systems for vibration, oil temperature, oil cleanliness and oil water content. Vibration, through spectral analysis, can indicate bearing defects linked to their specific frequencies, though this becomes more challenging for slow speed bearings. It is recognized that a surface defect needs to be above about 10mm in extent for reliable detection. However, sub-surface initiated fatigue can be characterized by a network of cracks that release copious quantities of material over a short period of time. Thus spectral analysis can only be expected to reveal a “snapshot” of bearing condition with little confidence in future failure likelihood.
Boroscopy may be used to inspect the internal condition of gearbox bearings but this may be limited by access to all running surfaces so incipient spalling may be missed.
Deviations from design assumptions on lubricant condition may be incorporated into revised life estimates. For example, higher operating temperatures, poorer oil cleanliness and higher water content can be used to re-assess bearing life. It is also possible that life estimates may not have been performed at that time with earlier versions of current codes.
For main shaft, pitch and yaw bearings (that are generally grease-lubricated), condition assessment is more difficult. The Fe content is frequently measured to form a judgment on bearing condition, together with collection and identification of larger particles. It is often difficult to determine the levels of particulate contamination that can be considered normal. Furthermore, pitch bearings may appear to be oscillating normally, but the rate of deterioration can accelerate as components break up leading to seizure.
An indication of remaining life may be gained from consideration of maximum contact stresses and actual duty to which bearings have been subjected. Since wind turbines that are up for life extension would have been designed up to 20 years ago, analytical methods have improved enabling more accurate estimates of peak stresses. Stresses well in excess of the guidance given in IEC 61400-4 would cast doubt on continued operation beyond 20 years. Similarly, less sophisticated pitch control systems may be in play, either increasing the likelihood of fretting corrosion or increasing peak stresses. These are factors to be considered in terms of bearing life.
Operational and maintenance histories are key factors in remaining life assessment. Previous failures should be analysed to determine root causes from which lessons can be learned. Failure data can be analysed with Weibull statistics to determine mean life and Weibull slope. These give information on actual L10 from in-service failures and whether a “wear out” condition is approaching (ie increasing failure rate) compared with the original design assumptions. In the absence of failures, a Weibayes analysis can be performed.
Once design or actual L10 life estimates are established, for systems containing many bearings (like gearboxes), series reliability calculations can be performed to estimate the probability of system failure Fs from an equation of the form:
where Fn is the probability that an element n is in a failed state. Probability of failure for each element may be derived from the L10 life according to future years of operation. Once a cost for repair or replacement of a failed system has been established, it is possible to establish the likely future cost from the probability of failure. This therefore forms the basis of decision-making on the basis of future economic risks.
So the issues to be considered in life extension of wind turbine bearings can be summarized as follows:
- Review of design assumptions concerning reliability level (L10 or better), lubricant condition (temperature, cleanliness, water content) and comparison with actual operating condition
- Consideration of other failure modes not considered by ISO 281, for example, corrosion, hydrogen effects, skidding
- Review of the potential effects of steel cleanliness on bearing life
- Review of maximum contact stresses seen in service and operational duty for comparison with recommendations
- A comparison of design and actual meteorological data to determine rate of life consumption, together with design and actual torque data
- Review of bearing condition from monitoring data (vibration, oil condition, debris analysis), but taken with caution as “snapshots” may not give a reliable indication of impending failure
- Review of operational and maintenance histories, putting any failure data into Weibull or Weibayes analyses to determine actual L10 life and Weibull slope (in anticipation of wear-out)
- Perform an economic risk assessment based upon design or actual L10 life, series reliability for systems with many bearings, change of failure probability as a function of future years operation and resulting likely costs