Designing Large Segmented Flexible Conical Plain Bearings as Wind Turbine Main Bearings

Main bearings are failure-prone components in wind turbines, significantly increasing the LCOE of wind energy. Therefore, segmented plain bearings are being discussed as replacement for the currently used rolling bearings. Segmented plain bearings allow an up-tower replacement of faulty segments without the need to dismantle the WT drivetrain, as is required for rolling bearing replacements. One such bearing is the segmented, flexible, conical “FlexPad” plain bearing as alternative to existing double row tapered roller bearings. Designing the FlexPad bearing manually is very complex and thus time-consuming. Hence, during the research project a new holistic design method for large and highly loaded, conical plain bearings was developed. To prove the suitability of the method, two bearing prototypes were built and tested. The prototype, designed with the new method, outperformed the first, manually designed bearing, in terms of the design objectives. Most importantly, the friction losses were reduced by 32.1%. The project results show that the new method is able to efficiently design safely operating FlexPad bearing variants for wind turbines.


Introduction
Wind turbines (WT) are a key technology towards a carbon neutral energy production worldwide.The EU aims to be climate neutral till 2050 and Germany set itself a target of a fully carbon neutral electricity production till 2035 [1][2][3].To increase the competitiveness of wind energy, the levelized cost of electricity (LCOE) needs to be lowered further.Currently about 30% of the LCOE is caused by service and maintenance activities [4].The main rolling bearings are critical components in terms of maintenance costs due to their failure probability of 15-30% within their 20-year design lifetime [5].Main bearing replacements are expensive procedures as an external crane is needed which costs about 200.000 € per day for offshore WTs [6].Hence, a discussed alternative for rolling bearing as main bearings in the wind industry are segmented plain bearings.Segmented plain bearings allow an up-tower sub-component wise replacement of faulty segments without the need to dismantle the WT drivetrain, as it is needed for rolling bearing replacements.
The use of segmented plain bearings for WT is still in the prototype stage and subject of ongoing research.The large and highly loaded structures of WTs deform under load.Plain bearings are sensitive towards these deformations as their clearance is typically only about 1‰ of the bearing's diameter.Exceeding the plain bearings capacity to endure the deformation will lead to edge loading and mixed friction, resulting in excessive wear and thus premature bearing failure.The FlexPad bearing, shown in Figure 1, was designed with these challenges in mind and fulfils the requirement of an easy to maintain WT main bearing.Its flexible support structure for the segments ensures they stay parallel to the shaft

Method
The FlexPad bearing is defined by ten key design parameters (see Figure 1).Five parameters describe the bearing global parameters: outer and inner diameters (Do and Di), span (S), inclination (α) and number of segments (Nsegm).The segments are supported by a flexible arm, giving each segment the ability to adapt to the shaft's deflection.Each arm is defined by five parameters: arm thickness (sarm), dimension and position of the groove (tgroove, xgroove, bgroove), and the normalized arm length (larm,norm).Each of the ten parameters affects the performance of the bearing individually.Furthermore, there are cross-influences between the parameters affecting performance.Thus, finding an optimal design manually is very difficult.Therefore, an optimization method is needed to design this bearing efficiently.
The modelling and simulation of the FlexPad bearing for one parameter set during the WEA-GLiTS project required about eight hours, including a lot of error-prone manual user interaction.Improvement to the simulation model and automatization of the entire process, from parameter definition, modelling and finally running the simulation reduced this time to about one hour, without the need of user interaction.With this improved modelling approach a new design method was developed, first described by Rolink et al. [11].Further finetuning of this method leads to the approach depicted in Figure 2 and consists of five steps: 1. Definition of design parameter space Each parameter can vary in a defined range.These ranges need to be set by preliminary studies (e.g.spot sampling) or can be defined by experienced engineers.

Sampling
Within the defined parameter space, discrete and unique parameter sets are defined.Sampling can be carried out utilizing a sampling algorithm, e.g.Latin Hypercube Sampling.

Simulation
Each of the sampled parameter set is simulated.For the FlexPad, this is done by using an in-house developed toolchain.The new modelling approach automatically links the necessary steps to perform the elasto-hydrodynamic-multi-body simulations.

Surrogate model
The simulation results are post-processed and used to derive surrogate models for every objective stated in table of the next step.For the FlexPad bearing, a linear regression function with first and second order terms was used.This approach is further described by Rolink et al. [10].

Optimization
In the last step, surrogate models are used to find an optimized bearing design which bests fulfills the design objectives.Regarding the FlexPad prototype, the following design objectives were optimized:

Objective Target Explanation
Friction power Minimize Frictional losses of bearing need to be minimized for an efficient bearing

Pressure area ratio Maximize
Area on the segment with a hydrodynamic pressure over 0.1 MPa.Maximized for good sliding area utilization Maximum hydrodynamic pressure Maximize to limit Maximize to limit of sliding material with safety margin for a high-power density of bearing A bearing designed and optimized with the design method (V2) is depicted at the bottom of Figure 2. All objectives, except for hydrodynamic pressure, could be optimized in the direction of the target by using the new design method (cf.Table 1).The hydrodynamic pressure was reduced by 4%, despite the target being maximizing to a limit.This objective is of secondary importance and ensures that the sliding material strength is not exceeded.As the limit has not been exceeded, this objective can also be stated achieved.
The V1 design was developed without utilizing the new design method and is a scaled down version of the FlexPad bearing prototype developed in the WEA-GLiTS project [8,12].The outer diameter of V1 was reduced from 926 mm to 503 mm to decrease the prototype manufacturing costs.Both V1 and V2 were built and extensively tested at the Chair for Wind Power Drives (CWD).Within the FlexPad project the V1 bearing was used to validate the simulation models implemented in the aforementioned toolchain and as a baseline for optimizing the FlexPad's design.The second prototype (V2) was designed using the design method presented in this chapter.It was used to validate the design method itself and demonstrate that the computed optimization results can also be measured on the test bench.

Test bench setup
Both prototypes, V1 and V2, were tested on the same test bench at the CWD.The experiments were used to validate the improved modelling approach and the new design method for the FlexPad bearing.Also, the general qualification to design a safely operating plain bearing for WT with the design approach should be proven.Figure 3 (left) shows the setup of the V1 prototype on the test bench.The following section describes the measurement system setup used on the test bench and afterwards the applied load cases during testing.

Test load cases
Both prototypes were tested on the test bench shown in Figure 3 left, which allowed load application in the following directions at variable shaft speeds: • Thrust and vertical forces • Bending moments around the horizontal and vertical axes All load cases shown in the following are derived from multi-body simulations of an 850 kW turbine.Each conical FlexPad test specimen underwent a similar test program.The prototypes were tested under constant load cases, dynamic load cases, idling, and start-stop operation to demonstrate the bearings' overall feasibility as a WT main bearing.The bearing design was performed with the rated operation load case shown in Table 2.These loads were applied for 12 hours to determine the steady-state temperature of the test bearing.The start-stop load case is a critical load case for plain main bearings due to solid-state friction between the shaft and bearing segments during WT start-up, resulting in inevitable wear.Each bearing prototype underwent approximately 10,000 start-stop cycles of 30 s each, as shown in Figure 7.During the first five seconds, of the steady forces are applied without rotation.In the following ten seconds, the shaft speed and thrust force (Fx) increase.This operating point is held for five seconds.The loads then are ramped down to zero and finally, the shaft rotation is ramped down to stand still.After another five seconds the cycle restarts.Each bearing variant successfully completed 10,000 start-stop cycles without any damage or significant wear beyond the initial run-in wear.

Figure 7. Load and shaft speed curves of exemplary start-stop load cycle
Idling is another critical load case for plain main bearings.Forces and torques act on the bearing even at low speeds (see Table 2).As the shaft speed is only 0.6 min -1 , the bearing does not operate fully in the hydrodynamic regime and runs continuously in mixed friction.Each prototype variant was operated for approximately 100 h under this load conditions.These 100 h only represent a fraction of the expected operation in idling of a WT in the field, but the test bench time was limited and the start-stop operation is the more critical operation condition.

Results and Discussion
The project's design method relies on simulation data, making a validated simulation model crucial for its reliability.The validation of the simulation model is shown below exemplarily for prototype V1.

Validation of the simulation model
The validation was based on temperature measurements and thermo-elasto-hydrodynamic (TEHD) simulations conducted by the project partner IST with the simulation software FIRST.The friction losses in the TEHD model are converted into a temperature distribution in the lubrication gap, as shown in Figure 8 bottom left.To validate the simulation model, the measured surface temperatures of the pads at the test bench were compared to the simulated ones (Figure 8).The measured values, shown in blue, were gathered after ten hours of operation with static 100% production loads (cf.Table 2).The simulated values of the oil temperature at the sensor position, shown in red, resulted from simulating the same load case.Overall, the simulation model and the measurement show very good agreement, with a maximum deviation of less than 5%.The validation of the lubrication condition was based on the measurement of the lubrication gap height.Figure 9 compares the results of the simulation and measurement for the static production load case in Table 2.The measured lubrication gap height corresponds qualitatively to the simulated one, although the absolute height of the lubrication gap does not match.This discrepancy is due to the complicated application and calibration of the sensors in the installed state.The simulation model is considered validated due to the good quantitative agreement of the temperatures and the qualitative agreement of the lubrication gap height.

Validation of the design method
Both tested prototypes successfully completed the test bench program consisting of static and dynamic loads of rated operation for a WT.In addition, each bearing withstood 10,000 start-stop cycles and 100 h of idling.These two load cases are critical for plain bearings due to the continuous mixed friction operation.During the tests no bearing failure occurred.A visual inspection of all sliding segments after testing showed no extensive damage and an intact sliding surface.Therefore, the general qualification of the new design method to design a safely operating plain bearing for WT is proven.
To validate the effectiveness of the design method the measured frictional torque of the two bearing variants is compared.The friction torque measurement during an eleven-hour production load case test is shown in Figure 10, left.The decrease in frictional torque at the beginning is a common occurrence in plain bearings and is referred to as the running-in process.To obtain a comparative value, the measured frictional torque on the test bench was averaged over six minutes after 10.5 h (Figure 10, red line).The table on the right in Figure 10 compares this value with the simulation result for both tested bearing variants.The deviation between the measurement and simulation is approximately 43 Nm for both variants.It is expected that the deviation in measurements is a result of the friction in the rolling bearing of the test bench located between the drive and load unit (cf. Figure 3 left).Due the design of the bench, the torque measurement includes the losses in the rolling bearing.As the same load case was used for both V1 and V2, the frictional torque of the rolling bearing in the load unit should remain constant.Since the deviation in simulated and measured frictional torque also remains constant, it can be concluded that this deviation is a result of friction in the rolling bearing.Disregarding this offset it can be seen, that the friction torque for V2 was reduced by 32.1% with the design method compared to V1 (based on the simulated values).This validates the effectiveness of the design method to optimize for the defined objectives.

Conclusion
WTs could use segmented plain bearings as main bearings in the future.Currently, this concept is still subject to ongoing research.The large and highly loaded structures of a WT deform under load.Plain bearings are sensitive towards these deformation as their clearance is only about 1‰ of the bearing's diameter.Exceeding the plain bearings capacity to endure the deformation will lead to excessive wear and thus premature bearing failure.The FlexPad bearing was designed with these challenges in mind and to be a replacement for tapered roller bearings.However, designing the FlexPad bearing manually is very complex and thus time-consuming.To handle the complexity a new design method was developed and validated during the FlexPad project.The method uses a detailed elasto-hydrodynamic-multibody simulation model.Different designs of the FlexPad have been simulated and translated into a surrogate model.The surrogate model allows for the utilization of a mathematical optimization algorithm to design a new FlexPad bearing.
Two prototypes were built to validate the design method.The V1 design is a scaled down version of the FlexPad bearing prototype developed manually in a previous project.The second bearing (V2) was designed and optimized with the design method presented herein.The optimization objectives were to minimize friction power, maximize pressure area ratio, and maximize hydrodynamic pressure (to a limit).Both bearing designs were built and extensively tested at the CWD under realistic load conditions for WT.Both successfully completed the test bench program consisting of static and dynamic loads of rated operation for a WT.In addition, each bearing withstood 10,000 start-stop cycles and 100 h of idling.These two load cases are critical for plain bearings due to the continuous mixed friction operation.Simulations showed that the bearing designed using the new method had 32.1% less friction power during rated operation.The reduction of friction between the two designs was validated on the test bench.The project results show that the new method is able to design a FlexPad bearing variant with the optimization targets achieved.
In summary, the new method enables to design safely operating FlexPad bearings while meeting specific optimization targets.The method's current limitation to the FlexPad bearing can be expanded in future research by adapting it to other segmented plain bearing designs for WT.

Figure 1 .
Figure 1.The FlexPad bearing and its parameters.

Figure 2 .
Figure 2. Steps of the design method

Figure 3 .
Figure 3. FlexPad bearing (V1) on the test bench (left) | FlexPad in operation with the applied measurement system in close-up (centre) | Comparison of the two prototypes tested (right)3.1.Temperature and lubrication gap height measurementTo measure the temperature distribution, two PT100 temperature sensors were integrated into each pad, as shown in Figure4.The sensors' measuring tip is located 0.5 mm below the pad sliding surface, allowing for precise measurement of the oil temperature in the lubrication gap.This enables precise determination of the steady-state temperature of the bearing, which is crucial in assessing the temperature safety of the plain bearing.Figure4right shows a detailed section of the temperature measurement system.The temperature sensors are identifiable by the white cables.

Figure 4 .Figure 5 .
Figure 4. Position of the temperature sensors under the sliding surface (left) and their application holes in the cutaway view (right) Each cone of the FlexPad bearing was equipped with two eddy current sensors.These sensors allow to measure the distance between the shaft and the sliding surface, and therefore the lubrication gap height.The measurement is unaffected by the oil in the lubrication gap.The sensors were positioned flush with the cone running surfaces, as shown in Figure 5.

Figure 6
Figure 6 displays one of the dynamic load cases, with clear force and moment oscillations in the four-second detail section on the right.

Figure 6 .
Figure 6.Exemplary dynamic load case and shaft speed with detailed 4 s section (right)

Figure 8 .
Figure 8. Validation of the TEHD simulation model via test bench measurements of the pad temperature distribution on the gearbox side of prototype V1

Figure 9 .
Figure 9.Comparison of lubrication gap height between measurement and simulation

10 Figure 10 .
Figure 10.Validation of the design methodology by comparing the frictional torque between test bench tests and simulations of V1 and V2 for the same load case

Table 1 .
Comparison of the reference FlexPad design with the optimized design.

Table 2 .
Exemplary production and idling load case with constant values