Wind Turbine Gearbox Study Raises Reliability Hopes

Manufacturers continue to modify and redesign existing turbines and it is difficult to validate the modifications fast enough to prevent multiple units with unsatisfactory ‘solutions’ being deployed. Under current procedures, many years may be needed to develop confidence in a solution. By that time, the industry may have moved to larger turbines or different drivetrain arrangements. Moreover, the industry may never understand the fundamental failure mechanisms of the original problem.

It is useful to think of this product development cycle as consisting of several discrete steps that include activities outside those normally considered part of the design process. The Gearbox Reliability Collaborative (GRC) project was established to identify shortcomings and recommend improvements in this process.

The GRC Project

According to NREL, the GRC project has five key goals:

• Establish a collaborative of wind turbine manufacturers, gearbox designers, bearing experts, universities, consultants, national laboratories and others to jointly investigate issues related to wind turbine gearbox reliability and to share results and findings;
• Design and conduct field and dynamometer tests using two redesigned and heavily instrumented wind turbine gearboxes to build an understanding of how selected loads and events translate into bearing and gear response;
• Evaluate and validate current turbine, gearbox, gear and bearing analytical tools and models, and develop new tools/models as required;
• Establish a database of gearbox failures;
• Investigate condition monitoring methods to improve reliability.

Preparation of gearboxes

A gearbox representative of the common gearbox in service in 2006 was selected, featuring a three-point suspension drivetrain configuration with supports at the main bearing and gearbox mounting trunnions — a typical configuration for MW-class turbines.

A 750 kW rating was picked as large enough to represent common turbines currently in use, but small enough to be inexpensive to procure, modify and test in the NREL 2.5 MW dynamometer. The preparation included modifications to eliminate design shortcomings, update to current design practices and accommodate instrumentation.

Two gearboxes from the same manufacturer were removed from the field after about 40,000 hours of operation with sufficient damage to require a rebuild. These were carefully disassembled and inspected for damage. The two gearboxes were remanufactured and assembled, and instrumentation was installed during the gearbox assembly. The selected gearbox has a three-point suspension drivetrain.

Condition monitoring

Even now – as when the GRC got underway in 2007 — wind plant owner/operators are primarily practising reactive or time interval-based maintenance. A paradigm shift to condition-based maintenance (CBM), enabled by condition monitoring (CM) techniques, can help owner/operators cut operations and maintenance (O&M) cost.

The GRC dynamometer and field tests provide a great opportunity to investigate the strengths and limitations of different CM techniques. Primarily, the CM system was implemented by working with several commercial equipment suppliers. An integrated approach was taken because no single technique can provide the comprehensive and reliable solutions needed. Four CM techniques were initially applied: acoustic emission (AE) (specifically, stress wave); vibration; off-line (or kidney loop) real-time lubricant CM; and off-line oil sample analysis. As the GRC tests progressed, in-line (or main loop) real-time lubricant CM and electric signature-based techniques were added.

Field testing

Two field test campaigns have been conducted in the GRC project, NREL says. Both were conducted at Xcel Energy’s Ponnequin wind farm, located just south of the Wyoming-Colorado border, which has 44 machines of 750 kW and 660 kW.

The GRC test turbine is a three bladed, upwind, stall controlled turbine with a rated power of 750 kW. The generator has two sets of poles, which allow it to operate at two speeds. The turbine rotor operates at 22.4 rpm (1810 rpm on the HSS) and 14.9 rpm (1208 rpm on the HSS). The turbine has pitchable tip brakes and a high-speed shaft brake. For a normal shutdown, the tip brakes deploy first.

A measurement campaign verified whether predictions of maximum main shaft torque loads were accurate. The main shaft on turbine 29 at Xcel Energy’s Ponnequin wind farm was instrumented with a full bridge arrangement of strain gauges to measure torque. Data obtained from this test indicated that maximum torque value was 665 kNm, about twice the rated (350 kNm) torque.

A second campaign of field tests was conducted using the fully instrumented GRC Gearbox 1 installed in turbine 12 at Ponnequin. During testing operations the wind turbine faulted several times due to high-speed bearing temperatures exceeding 90ºC. There were also two incidents of substantial oil loss. An inspection revealed that the high-speed stage gear teeth showed signs of significant overheating. As of June 2011, the failure database had collected 37 incidents. Bearing failures featured in 36 and gear failures in 22.

Gearbox run-in

The identified damage modes in wind turbine gearboxes include the early onset of gear and bearing surface damage, such as scuffing and micropitting. The risk of this is difficult to accurately assess but the influence factors are well known — both are strongly affected by surface roughness of the contacting parts. Fortunately, it is well accepted that risk can be mitigated by controlled initial wear-in or run-in. A controlled run-in has been a requirement in the wind turbine gear industry since 2002, when the ANSI/AGMA/AWEA 6006 wind turbine gear standard was published. This is normally performed in factory serial acceptance testing in timed load stages of 30-60 minutes. GRC performed run-in of its two 750 kW gearboxes in the NREL 2.5 MW dynamometer test facility using a modified version of the standard procedure currently performed at most WTG gearbox manufacturing facilities.

Particle counting — or on-line ISO cleanliness measurement — was used as the gatekeeper for moving to the next load stage through identifying when the rate of particle generation from wear stops increasing and falls off. The GRC used on-line oil monitoring equipment to help determine whether the initial run-in plan of 30 minutes of operation at 25%, 50%, 75% and 100% of rated torque was sufficient or excessive.

Bringing down the cost of wind energy will require a significant rise in long-term gearbox reliability (Source: Photographic Services, Shell International Ltd)

Test data from both gearboxes indicate that particle counts did in fact fall off at the 1+ hour range, depending on load and temperature. The industry has recognised that cleanliness is a better measure than time spent at a specific load level as it verifies that run-in has occurred. Future work should include the use of alternative particle counting devices, verified ISO cleanliness, reconfiguration of the lubrication system piping to increase the response time of the particle counters, and detailed investigation of bearing and gear contact surfaces to verify that run-in conditioning is well correlated to particle counts.

Some debate remains on how beneficial oil conditioning might be. GRC tests have demonstrated that lubrication oil filtration, moisture prevention by breathers, and heat control are useful to keep turbine gearbox oil dry and clean.

The gearbox experts in the collaborative programme also deemed that the torque histogram in the original gearbox specification was too low. A field test measured a maximum torque of 665 kNm during a transition from low- to high-speed generator. The highest reverse torque of -287 kNm was measured during a transition from high- to low-speed generator. The torque spikes occurred despite the presence of a softstarter. The magnitude of the torque spikes and the resulting damage during generator shifts and starts could be greatly reduced with proper controller tuning.

Effect of non-torque loads

Non-torque loads such as shaft thrust and bending have historically been left out of the dynamometer test validation process. However, in wind turbines, especially those with three-point suspensions, many experts suspect this assumption is not valid and could be a major contributor to gearbox problems. The GRC systematically applied non-torque loads in dynamometer tests to assess their effects. Results indicate that these loads affect tooth contact patterns in the low-speed stage in a manner that could shorten gearbox life. Preliminary analysis has included assessment of low-speed shaft bending, gearbox displacement and rotation, and tooth mesh patterns between the planets and the ring gear. Bending in the main shaft was found to affect the planet-ring gear mesh pattern and planet carrier misalignment. The gear load distribution shift was large enough, in some cases, to significantly increase edge loading, known to contribute to high contact stresses and shorter gear life.

Another non-torque load is the axial force of thrust of the turbine rotor on the drivetrain. In typical configurations, this load should be transmitted solely through the main bearing to the drivetrain main frame support structure, from which the gearbox should be uncoupled. But the main bearing permits small axial displacements of the main shaft during occasional thrust reversals that have been observed in field testing.

Field testing has also shown significant loading differences between the upwind and downwind rows of planet bearings in the first stage of the gearbox. The effect of the overhung weight of the rotor causes motions and elastic deflections in the planet carrier and housing that contribute significantly to shifting of the upwind/ downwind planet bearing load share. Dynamic upwind/downwind variations occur most dramatically at low to moderate loads (50% or lower). There is also evidence that bearing and planet pin fits contribute to carrier dynamic deflection. In addition, the load share between the planets varies once per revolution and these time-varying loads have been shown to reduce gearbox life. These results suggest the complete carrier, gearbox support structure, and gear and bearing elements should be modelled together to capture the system effects.

The load distribution across the annulus gear in the planetary stage is highly important because it is a central parameter for calculated life predictions in the design process. Normally, this load distribution is measured with strain gauges placed within the roots of the ring gear teeth inside the gearbox. There are a number of disadvantages to this approach; chief among them are limited space in the roots, inaccessibility for maintenance, exposure to harsh conditions within the ‘box, and cost. Therefore, the GRC annulus gear was instrumented externally along one root to determine if the load distribution could be measured more easily from the outside.

It was observed that the external strain could be modelled well; however, the externally measured load distribution is flattened out when compared to the internal strain. The external strain gauges can indicate the centroid of the load distribution accurately, thereby verifying whether the microgeometry design of the gear is sufficient to prevent edge-loading. However, due to the flattened strain profile, the gauges cannot give an accurate measure of KhB, a key design parameter in gear fatigue life rating that indicates how much edge loading is occurring. Because of the ease of installation and time and cost savings, external gauging could help gearbox designers validate some design assumptions, the report’s authors conclude.

Influence of assembly error

Modelling, testing and failure analysis uncovered effects of assembly error including a difference in the deformation of the planet carrier under equivalent operating conditions with an increased misalignment of the planet pin that significantly cuts bearing life prediction below the 20-year design life, NREL says.

Gearbox 1 failure analysis revealed many assembly issues, including the lack of tapped holes or other provisions for handling the hollow shaft. This made assembly and disassembly difficult and increased the risk of assembly damage. The blind assembly of gearbox bearing sets led to cocked rollers or spacer interference, which caused damage at the roller spacings in multiple locations. Loose interference fits also led to damage. The interface between the outer ring of a bearing and the shoulder on the bearing cap had severe fretting corrosion, most likely caused by the loose fit of the outer ring of the bearing. The hollow shaft seal was severely scuffed and the "O"-ring vaporised due to excessive endplay in some bearings on the low-speed shaft. This endplay was caused by the spinning of the inner ring of another bearing, another effect of incorrect assembly fits, the report says.

Gearbox 1 failure analysis

After multiple oil losses in the field and condition monitoring testing in the dynamometer, Gearbox 1 was sent for disassembly and inspection, which revealed that the primary failure mode of the highspeed (HS) gearset was severe scuffing.

The root cause of failure was most likely lubricant starvation. The wear pattern indicates the HS gear mesh was misaligned causing higher load at the rotor end of the teeth. The sun spline, meanwhile, suffered from severe fretting corrosion. The root causes of the failure were probably lubricant starvation and poor load distribution (only about 50% of the teeth carried load). Although the HS gearset, sun spline, and a bearing set showed evidence of overheating due to lubricant starvation, there was no other evidence that the gearbox ran out of oil. It may have leaked oil, but did not run completely dry.

Subsequently, the oil transfer ring for the planet carrier was found cocked and jammed in the carrier. Hand pressure showed that it was prone to jamming, which was most likely the cause of failures downstream in the lubricant path. Most gears had some teeth with mild to severe fretting corrosion; its root cause was parking of the wind turbine. All teeth of the intermediate gearset had a spot of fretting corrosion and scuffing; the root cause was probably trapping of debris between a pair of teeth. The hunting gear ratio caused the damage to imprint on all teeth of the intermediate gearset.

The annulus gear had a 25 mm patch of scuffing, probably due to trapping of debris. The non-hunting gear ratio with a common factor of three caused the damage to imprint on every third tooth. Rust was also found on the carrier bore for the rotor shaft and on the outer diameter fit for the shrink ring, because no rust preventative was applied when the gearbox was removed from the rotor shaft.