Device and method for setting the pitch of blades in a wind turbine

Description

The invention relates to a device for controlling the pitch of blades of a wind turbine with the features of claim 1 ; and a method for controlling the pitch of blades of a wind turbine with the features of claim 6.

In propeller designs for wind turbines, rotor speed and power output can be controlled by pitching the rotor blades about their longitudinal axis (blade pitch control). Moreover, rotor blade pitching is the most effective protection against overspeed and extreme wind speeds, especially in large wind turbines. The blade bearings and the pitch control actuators are usually comprised in the hub of the wind turbine.

Therefore, efficient devices and methods for setting the pitch angle of at least one blade is required.

A device for setting the pitch angle comprises a measurement unit for measuring bending moments in and / or on the blades of the wind turbine during operation. Furthermore, it comprises a computing device for computing a pitch angle dataset with pitch angle offset data and a simulation device is executable for computing the difference between the measured bending moments and calculated bending moments over a predetermined time period, the pitch angle data set being determined in dependence of the simulation results. A pitch drive device is used to set corrected pitch angles of the blades in dependence of the computed pitch angle data set. With this it is e.g. possible to achieve a "self-balancing" structure if the blades receive an unbalanced aerodynamic load.

In one embodiment the computing device is an observer for a pitch control device providing set-points for pitch angles, so that the pitch angle dataset with pitch angle offsets and the set points for pitch angles are combined to the corrected pitch angles

(QC , Qc2, Θο3). The connection with the control device allows model based control schemes.

In a further embodiment the simulation device comprises a machine learning device for generating a functional relationship between the measured bending moments and the pitch angles, in particular allowing the training of a simulation model during the operation of the wind turbine and / or the implementation of a trained simulation model. With the machine learning device little or none a priori knowledge has to be implemented.

There is considerable data gathering at a wind tower so that in an embodiment the computing device obtains data regarding the wind speed, wind direction and / or status information about the wind turbine as input to compute the pitch angle data set. In particular with the machine learning device the additional data can be used to achieve better learning results.

The computing device determines in one embodiment the maximum of the difference between the measured bending moment and the computed bending moment. This is an efficient criterion for setting the pitch blade angle.

The issue is also addressed by the method of claim 6.

In this method bending moments in and / or on the blades are measured during operation of the wind turbine. Subsequently a pitch angle dataset is computed comprising a pitch angle offset data with a simulation device for computing the difference between the measured bending moments and calculated bending moments over a predetermined time period on a computing device, the pitch angle data being determined in dependence of the simulation results.

This leads to a setting of the pitch angles of the blades in dependence of the pitch angle data set by a pitch drive device.

Also an embodiment of the method can comprise a machine learning device for generating a functional relationship between the measured bending moments and the pitch angles, in particular allowing the training of a simulation model during the operation of the wind turbine and / or the implementation of a trained simulation model.

Given the time constants of wind turbines the simulation device estimates for each blade the average bending moment for a time period from 30 minutes to 6 hours, preferably for 30 minutes to 2 hours as calculated bending moments. Over this time periods short-term fluctuations are averaged out.

In a further embodiment the method calculates for each blade the relative deviation from the average bending moment as

Mbi

-∑? Mbi

n 1 1

and the maximum value of the relative deviation from the average bending moment.

In an alternative embodiment the simulation device calculates for one blade (i=1) the relative deviation from the average bending moment of the two neighboring blades:

AMri =

Mb2 + Mb3

It is also possible, that the relative deviation by absolute differences from the average bending moments.

One other efficient way in setting the pitch angles comprises that the corrected pitch angles for the blades are determined in dependence of a least one threshold value for the maximum value max(A ri) of the relative deviation from the average bending moment.

In one embodiment the method also comprises the computing device as an observer for a pitch control device providing set-points for pitch angles, so that the pitch angle dataset with pitch angle offsets and the set points for pitch angles are combined to the corrected pitch angles. Additional information for the computing device can comprise data regarding the wind speed, wind direction and / or status information about the wind turbine as input to compute the pitch angle data set. One possible criterion for setting the pitch angle in dependence of the bending moment is the maximum of the difference between the measured bending moment and the computed bending moment.

Embodiments of the device and method are shown in the figures, where

Fig. 1 schematically shows a wind turbine;

Fig. 2 schematically shows a first embodiment of a device for setting the pitch of a blade;

Fig. 3 schematically shows a second embodiment of a device for setting the pitch of a blade;

Fig. 4 shows a flow sheet for an embodiment of a method for setting the pitch angle of the blade.

Fig. 1 shows a wind turbine 100 on a wind tower 103. In this particular instance and in the further embodiments the wind turbine 100 comprises three blades 1 , 2, 3 which are connected with the wind turbine 100 through a hub 105.

A nacelle 104 of the wind turbine 100 comprises a number of units like a gearbox 101 for converting the rotational speed coming from the rotating blades 1 , 2, 3 to a rotational speed which is useful for generating electricity with a generator 102.

The nacelle 104 also comprises a computing device 20 which is used in connection with the setting the pitch angles Θι , Θ2, Θ3 of the blades 1 , 2, 3.

Generally the blade pitch angles Θι , Θ2, Θ3 are given against a predetermined null-marking at the hub. During assembly a marking on the blades is aligned with this null-marking on the hub. Those pitch angles Θι , Θ2, Θ3 are then used as base pitch angle values QM , 0 2, 0 3 for the control system. Under operation the changes of the pitch angles due to the control system ©control, i, 0controi,2, 0controi,3 are added to those base values forming the pitch angles Θι , Θ2, Θ3 of the blades 1 , 2, 3.

Θ, = ©controij + Θ bi with i=1 ,2,3 for the three blades

The computing device 20 can be part of a control system as will be described later. The location of the computing device 20 and the other units in the nacelle 104 shown in Fig. 1 only are exemplary. In particular, it is possible to locate the computing unit 20 elsewhere, e.g. in the wind tower 103.

The pitch angles Θι , Θ2, Θ3 of the blades 1 , 2, 3 can be individually adjusted by a pitch drive device 30 (shown schematically in Fig. 1) with actuators operating an all three bases of the blades 1 , 2, 3. By pitching the blades 1 , 2, 3 around the blade axes the blade pitch angles Θι , Θ2, Θ3 and consequently the angle of attack and the aerodynamic

forces are changed. Pitching influences the power output, and is therefore suitable for power limitation. The pitch angles Θι, 02, Θ3 at the hub 105 of the blades 1 , 2, 3 are suitably manipulated variables in a control system as will be shown below. The objective function is balancing of all bending moments M i, M 2, M 3. This can be e.g. the minimization of the deviations between the three bending moments. This implies that all three blades have experienced the same aerodynamic load over a certain period of time.

In the following it will become apparent that an embodiment comprises as self-learning pitch-balancing. This comprises the learning phase, the application phase and the control phase.

The learning is performed e.g. during the first hours of the wind turbine 100 or after major revision works.

During the subsequent application phase the learned corrections are used.

The control phase can be performed periodically or continuously.

Small imbalances in the pitch-alignment of the blades 1 , 2, 3, have a considerable effect not only on the power output but also on the mechanical loads, in particular on the gearbox 101. Therefore, the aerodynamic balancing of all three blades, in particular by setting the pitch angles individually for each blade 1 , 2, 3, is an important issue.

In Fig. 2 a first embodiment for a device (encased in a dashed line) and a method for setting the pitch angles Θι, 02, Θ3 is shown, i.e. a set of corrected pitch angles Qc 0c2, 0c3 is produced to adjust an existing setting for the pitch angles Θι, 02, Θ3. With this setting a balancing of the blades 1 , 2, 3 can be achieved.

The measured variables are bending moments ΜΜ , M 2, M 3 measured by a measurement unit 10 in the blades 1 , 2, 3 in generally known way.

In essence, the embodiment shown in Fig. 2 shows the use of a correlation between the bending moments M 1 , M 2, M 3 and the corrected pitch angles 0Ci , 0c2, 0c3.

The computing device 20 comprises a simulation device 22 which in one embodiment a priori knows the relationship between bending moments M 1 , M 2, M 3 and the best pitch angles Θι, 02, 03 for the performance at particular operating points. This knowledge can e.g. be obtained by training a machine learning device (e.g. neural network) to find optimal (or suitably defined) values for the pitch angles Θι , Θ2, Θ3 for different operating conditions subject to bending moment data M i , M 2, M 3.

In addition or alternatively, the learning can be performed online during the operation of the wind turbine 100, as mentioned above.

The computing device 20 using the simulation device 21 computes the difference between measured bending moments Mbi , M 2, M 3 and calculated bending moments Mbd , MbC2, M C3. In one embodiment, the measured bending moments M 1 , M 2, M 3 and / or calculated bending moments Mbci , M C2, M C3 can be averaged over a certain period of time (e.g. 1 h). Other measurement data can comprise wind speed data.

It can be assumed that over the period of time all three blades have experienced the same wind with all the local and / or temporal inhomogeneity.

Based on the difference between measured bending moments M 1 , M 2, M 3 and calculated bending moments Mbci , M C2, M C3, pitch offset angles Θ0ι , Θ02, Θ03 are computed (dataset 21) which are then used by the pitch drive device 30 to correct the pitch angles into corrected pitch angles 0Ci , 0c2, 0c3. It should be noted that it is not always necessary to correct all pitch angles at the same time. It might suffice to apply only one corrected pitch angle Θι , 02, 03.

A different control scheme is shown in Fig. 3 in which the computing device 20 with the simulation device 21 is used as an observer to a control device 40.

Both the computing device 20 and the control device 40 take the measured data for bending moments M 1 , M 2, M 3 as input. The computing device 20 is essentially operating as in the previous embodiment in determining a dataset 21 with the pitch angle offset data 0Oi , 0o2, 0os-

The control device 40 is determining set-points for the pitch angles 0spi , 0SP2, 0SP3- The pitch offset angles 0Oi , 0o2, 0o3 determined by the computing device 20 are combined (here added) and transmitted to the pitch drive device 30 using combined angles as corrected pitch angles Qc 0c2, 0c3 on the blades.

An optional feature in this embodiment is data 50 (e.g. wind speed, status of wind turbine etc.) transferred from the control device 40 to the observer 20, i.e. the computing device.

In the embodiments of Fig. 2 and 3, corrected pitch angles 0ci , Θο2, Θο3 are used to generate a set of self-balancing blades 1 , 2, 3, i.e. , the pitch angles Θι , Θ2, Θ3 are automatically adjusted so that the blades 1 , 2, 3 are balanced. This reduces the loads on the wind turbine 100 and reduces costs. Again, it should be noted that it is not always necessary to correct all pitch angles at the same time. It might suffice to apply only one corrected pitch angle.

In Fig. 4, a flowchart for an embodiment for a method for setting the pitch angles is shown.

In method step 401 initial values for the three pitch offset angles Θ0ι , Θ02, Θ03 are set.

In step 402 the measured bending moment data is filtered. Based on that filtered data, the average of the bending moments M i , M 2, M 3 (calculated) is estimated in step 403 by the computing device 20 based on the simulation device 22. The average is computed about a time period between 30 minutes and 2 hours, preferably over 1 hour.

Based on the time averaged values of the bending moments in step 404 relative deviations of the bending moments (e.g. expressed in percent) for each blade 1 , 2, 3 are computed by

<-""n —

-∑? Mbi

Here, the number of blades i is n=3. In the next step 405 the maximum estimated bending moment MM is determined, i.e. the blade n=1 , 2, 3 with the largest estimated bending moment: MM = abs (AMrk). That particular blade k is identified and the sign, i.e. the direction of the largest bending moment, is determined.

In the following steps 406 to 413 the correction to one of the blades k is determined in dependence of a least one threshold value for the maximum value max(A ri) (i.e. MM) of the relative deviation from the average bending moment.

In three steps 406, 408, 410 it is determined if MM is less than 0, 1 %, more than 2% or more than 1 %. If MM is less than 0, 1 %, no correction is required and the method stops (step 407).

If MM is more than 2%, a correction of 0, 1 ° is applied (step 409) and a new value for the corrected pitch offset is calculated (step 413).

If MM is more than 1 %, a correction of 0,05° is applied (step 41 1) and a new value for the corrected pitch offset is calculated (step 413).

Otherwise a correction of 0,02° is applied (step 412) and a new value for the corrected pitch offset is calculated (step 413).

The process can then be repeated after step 401.

One cycle of the steps 402 to 413 shown in Fig. 4 can be considered as one learning phase. The objective function is here, that the deviations between the bending moments between the blades becomes minimized. This is achieved in the process steps 404 and 405.

If the maximal relative deviation (step 404) is below a predetermined threshold, the learning phase can be terminated (not shown in Fig. 4).

In any case the maximal absolute value of a deviation from the common average value is determined in Step 405. For this particular blade k an evaluation regarding absolute value as well as direction of the maximal deviation can be performed.

The direction determines the blade pitch angle correction. If in this blade e.g. an overly large bending moment is determined, the blade has an overly high lift. Hence, the pitch needs to be set less aggressively. Therefore, the pitch angle is corrected dependent from the magnitude as 0.02°, 0.05° or 0, 1 °.

The correction is only applied for one blade for the next application phase or it is added to an already existing correction angle. The pitch angles of the other two blades remain unchanged. Alternatively the blade with the positive largest and the blade with the negative largest deviation with be corrected in one step.

At the end of the application phase the bending moments are assessed again, if their objective values are reached or if a correction of the blade with the largest deviation is necessary.

The difference (absolute and relative) of the bending moments of the blades strongly depends on the windspeed. The embodiment shown here is in particular applicable for the wind turbine 100 under partial load (e.g. 0.6 - 0.9 * vrated).

List of reference numbers

1 First blade

2 Second blade

3 Third blade

10 Measurement unit for bending moments

20 Computing device

21 Dataset with pitch angle offset data

22 Simulation device

Pitch drive device

Control device

Data

100 Wind turbine

101 Gearbox

102 Generator

103 Tower

104 Nacelle

105 Hub

Θι , Θ2, Θ3 Pitch angle of blade (measured at hub) 0O1 , 0O2, 0O3 Pitch offset angle

0Ci , 0C2, 0C3 Corrected pitch angles

0sp1 , 0sp2, 0sp3 Set-points for pitch angles

0b1 , 0b2, 0b3 base pitch angle values

MM , Mb2, Mb3 Bending Moment (measured)

Mbd , Mbc2, Mbc3 Bending Moment (calculated)
Patent claims

1. Device for setting the pitch angles (Θι , Θ2, Θ3) at a hub (105) of blades (1 , 2, 3) of a wind turbine (100), comprising

a measurement unit (10) for measuring bending moments (M i , M 2, M 3) in and / or on the blades (1 , 2, 3) during the operation of the wind turbine (100),

a computing device (20) for computing a pitch angle dataset (21) with pitch angle offset (Θ01 , Θ02, Θ03) data, a simulation device (22) is executable for computing the difference between the measured bending moments (Mbi , M 2, M 3) and calculated bending moments (Mbci , M C2, M C3) over a predetermined time period on the computing device (20), the pitch angle data set (21) being determined in dependence of the simulation results and

a pitch drive device (30) setting the corrected pitch angles (Qc Qc2, 0c3) of the blades (1 , 2, 3) in dependence of the pitch angle data set (21).

2. Device according to claim 1 , wherein the computing device (20) is an observer for a pitch control device (40) providing set-points (0s i , Θδ 2, 0s 3) for pitch angles (Θι , Θ2, Θ3), so that the pitch angle dataset (21) with pitch angle offsets (Θ0ι , Θ02, Θ03) and the set points (0s i , Θδ 2, 0s 3) for pitch angles (Θι , Θ2, Θ3) are combined to the corrected pitch angles (Qc Qc2, 0c3).

3. Device according to claim 1 or 2, wherein the simulation device (22) comprises a machine learning device for generating a functional relationship between the measured bending moments (M 1 , M 2, M 3) and the pitch angles (Θι , Θ2, Θ3)., in particular allowing the training of a simulation model during the operation of the wind turbine (100) and / or the implementation of a trained simulation model.

4. Device according to at least one of the preceding claims, wherein the computing device (20) has data regarding the wind speed, wind direction and / or status information about the wind turbine (100) as input to compute the pitch angle data set (21).

5. Device according to at least one of the preceding claims, wherein the computing device (20) determines the maximum of the difference between the measured bending moment and the computed bending moment.

6. Method for setting the pitch angles (Θι , Θ2, Θ3) at a hub (105) of blades (1 , 2, 3) of a wind turbine (100), comprising

a) the measurement of bending moments (M i , M 2, M 3) in and / or on the blades (1 , 2, 3) during operation of the wind turbine (100),

b) the computing of a pitch angle dataset (21) comprising pitch angle offset (Θ0ι , Θ02, Θ03) data with a simulation device (22) for computing the difference between the measured bending moments (Mbi , M 2, M 3) and calculated bending moments (Mbci , M C2, M C3) over a predetermined time period on a computing device (20), the pitch angle data set (21) being determined in dependence of the simulation results and

c) setting the pitch angles (Θι , Θ2, Θ3) of the blades (1 , 2, 3) in dependence of the pitch angle data set (21) by a pitch drive device (30).

7. Device according to claim 6, wherein the simulation device (22) comprises a machine learning device for generating a functional relationship between the measured bending moments (M 1 , M 2, M 3) and the pitch angles (Θι , Θ2, Θ3), in particular allowing the training of a simulation model during the operation of the wind turbine (100) and / or the implementation of aa trained simulation model.

8. Method according to claim 6 or 7, wherein the simulation device (22) estimates for each blade (1 , 2, 3) the average bending moment for a time period from 30 minutes to 6 hours, preferably for 30 minutes to 2 hours as calculated bending moments (Mbci , M c2, Mbc3).

9. Method according to claim 8, wherein the simulation device (22) calculates for each blade (1 , 2, 3) the relative deviation from the average bending moment as

-∑? Mbi

and the maximum value max(A ri) of the relative deviation from the average bending moment.

10. Method according to claim 8, wherein the simulation device (22) calculates for one blade (i=1) the relative deviation from the average bending moment of the two neighboring blades:

AMri =

Mb2 + Mb3

1 1 . Method according to claim 9 or 10, wherein AMri is determined by absolute differences from the average bending moments.

12. Method according to at least one of the claim 9 to 1 1 , wherein the corrected pitch angles (0Ci, Qc2, 0c3) for the blades (1 , 2, 3) are determined in dependence of at least one threshold value for the maximum value max(A ri) of the relative deviation from the average bending moment.

13. Method according to at least one of the claims 6 to 12, wherein the computing device (20) is an observer for a pitch control device (40) providing set-points (0s i, 0S 2, Qspz) for pitch angles (Θι, Θ2, Θ3), so that the pitch angle dataset (21) with pitch angle offsets (0Oi, Θ02, Θ03) and the set points (0s i, 0S 2, 0S 3) for pitch angles (Θι, Θ2, Θ3) are combined to the corrected pitch angles (Qc 0c2, 0c3).

14. Method according to at least one of the claim 6 to 13, wherein the computing device (20) uses data regarding the wind speed, wind direction and / or status information about the wind turbine (100) as input to compute the pitch angle data set (21 ).

15. Method according to at least one of the claims 6 to 14, wherein computing device (20) determines the maximum of the difference between the measured bending moment and the computed bending moment.