METHOD FOR DETECTING A BLADE MISALIGNMENT OF A ROTOR BLADE OF A ROTOR OF A WIND TURBINE

Abstract

Provided is a method for detecting at least one blade misalignment of a rotor blade of a rotor of a wind turbine having multiple rotor blades adjustable in their blade angle. The blade misalignment describes a blade angle deviation of a detected blade angle of the rotor blade from a reference blade angle. The wind turbine includes a gondola having the rotor and an azimuth adjustment device in order to adjust the gondola in an azimuth alignment having an azimuth angle, and to adjust the azimuth alignment. The azimuth angle is tracked using the azimuth adjustment device to a predeterminable azimuth setpoint angle, and the blade misalignment is detected as a function of an azimuth movement of the gondola. Provided herein is detection of aerodynamic imbalances with reduced costs.

Claims

1. A method for detecting at least one blade misalignment of a rotor blade of a rotor of a wind turbine, wherein: the wind turbine has multiple rotor blades having blade angles, respectively, that are adjustable, the blade misalignment represents a blade angle deviation of a detected blade angle of the rotor blade from a reference blade angle, and the wind turbine includes a gondola having the rotor and an azimuth adjustment device for adjusting the gondola in an azimuth alignment having an azimuth angle, and wherein the method comprises: to adjust the azimuth alignment, tracking, by the azimuth adjustment device, the azimuth angle to a predeterminable azimuth setpoint angle; and detecting the blade misalignment as a function of an azimuth movement of the gondola.

2. The method according to claim 1, wherein detecting the blade misalignment as a function of the azimuth movement of the gondola includes: detecting an adjustment activity of the azimuth adjustment device.

3. The method according to claim 2, wherein the adjustment activity is an adjustment activity of at least one azimuth motor of the azimuth adjustment device.

4. The method according to claim 1, comprising: counteracting, by the azimuth adjustment device, a change of the azimuth alignment of the gondola due to wind action by a position regulation of the azimuth angle.

5. The method according to claim 4, wherein: the azimuth adjustment device does not have a holding brake, or the tracking of the azimuth angle is performed continuously.

6. The method according to claim 1, wherein a state observer is used to detect the blade misalignment.

7. The method according to claim 1, wherein a state observer for detecting the blade misalignment uses as state variables at least: a gondola azimuth torque, and a blade misalignment.

8. The method according to claim 7, wherein the state observer uses as state variables: the gondola azimuth torque, an azimuth torque offset, and one blade misalignment of each of two rotor blades.

9. The method according to claim 7, wherein the gondola azimuth torque points in a gondola rotational direction and includes: one azimuth torque component, directed in the gondola rotational direction, of an aerodynamic blade torque (M.sub.aero,A, M.sub.aero,B, M.sub.aero,C) of each rotor blade, and an azimuth torque offset, so that a sum of the azimuth torque components of all rotor blades and the azimuth torque offset forms the gondola azimuth torque.

10. The method according to claim 1, wherein a Kalman filter is used as a state observer for detecting the blade misalignment.

11. The method according to claim 1, wherein an adjustment activity of the azimuth adjustment device is produced by evaluating an azimuth drive torque that represents a drive torque using which the gondola is adjusted in the azimuth alignment.

12. The method according to claim 11, wherein: the azimuth drive torque is determined as a function of one motor current or multiple motor currents or one motor torque or multiple motor torques of at least one azimuth motor, or the azimuth drive torque is determined as a product of the motor current, a sum of the multiple motor currents, the motor torque, or a sum of the multiple motor torques and a proportionality factor.

13. The method according to claim 11, wherein: the azimuth drive torque is used as an output variable of a state observer, the state observer outputs an estimated azimuth drive torque, or an azimuth drive variable representative of the estimated azimuth drive torque, and a difference between a detected azimuth drive torque and the estimated azimuth drive torque or a difference of variables representative thereof is returned as an observation error for adapting observer states in the state observer.

14. The method according to claim 1, wherein a model description underlies a state observer for detecting the blade misalignment in which for each rotor blade an aerodynamic blade torque M.sub.aero is described by: $M_{a e r o} = \frac{1}{3} * \frac{1}{2} * π * ρ_{a i r} * c_{s} (λ, α + Y) * l_{bl}^{3} * {(v_{W i n d} + ω_{Y a w} * l_{bl} * \sin (φ))}^{2}$ with: ρ.sub.air=air density $λ = \frac{ω_{R o t} * l_{bl}}{v_{W i n d}} = speed ratio (ω_{R o t} = rotor velocity or rotor speed in rad / s)$ l.sub.bl=rotor blade length α=blade angle γ=blade misalignment of the rotor blade c.sub.s(λ, α)=aerodynamic properties of the rotor blade ω.sub.Yaw*l.sub.bl*sin(φ)=apparent wind velocity due to an azimuth movement at the rotor blade; with φ: rotor angle, wherein each blade torque M.sub.aero represents a torque acting on the rotor due to the rotor blade in a rotational direction transverse to an axis of rotation of the rotor, wherein each of the blade torques is transformed using an angle transformation as a function of a rotor angle of the rotor related to the respective blade into an azimuth torque component directed in the gondola rotational direction, a gondola azimuth torque is calculated as a sum of the azimuth torque components and an azimuth torque offset, a gondola acceleration torque is calculated as a function of a change of an azimuth speed and a mass moment of inertia of the gondola, an azimuth drive torque is determined as a function of a difference between the gondola azimuth torque and the gondola acceleration torque, and the azimuth drive torque, or a variable representative thereof, forms an output variable of the model of the state observer.

15. The method according to claim 14, wherein two of the blade misalignments, the azimuth torque offset, and the azimuth torque form system states of the model, and a sum of blade misalignments of all rotor blades is zero and the blade misalignment of one of the rotor blades is calculated from the blade misalignments of remaining rotor blades.

16. The method according to claim 1, wherein a model description underlies a state observer, which is described by the following equation system and in which three rotor blades A, B, and C are used: $M_{aero, A} = f (v_{w}, n_{R}, α, γ_{A}); M_{aero, B} = f (v_{w}, n_{R}, α, γ_{B});$ $M_{aero, C} = f (v_{w}, n_{R}, α, γ_{C});$ $M_{yaw, A} = f (M_{aero, A, φ}); M_{yaw, B} = f (M_{aero, B, φ}); M_{y a w, C} = f (M_{aero, C, φ})$ $γ_{c} = - γ_{A} - γ_{B}$ $M_{yaw, Nac} = M_{yaw, A} + M_{yaw, B} + M_{y a w, C} + M_{y a w, o f f}$ $M_{acc} = i_{ge} .Math. i \frac{{dn}_{DRV}}{dt} .Math. J_{Nac}$ $M_{D R V} = \frac{1}{i_{g}} .Math. (M_{yaw, Nac} - M_{a c c})$ with M.sub.aero,A, M.sub.aero,B and M.sub.aero,C: aerodynamic blade torque M.sub.aero of the rotor blade A, B, or C, respectively; v.sub.w: wind velocity; n.sub.R: rotor speed; α: blade angle; φ: rotor angle M.sub.yaw,A, M.sub.yaw,B, and M.sub.yaw,C: azimuth torque component of the rotor blade A, B, or C, respectively; γ.sub.A, γ.sub.B and γ.sub.c: blade misalignment of the rotor blade A, B, or C, respectively; M.sub.yaw,off: azimuth torque offset; M.sub.yaw,Nac: gondola azimuth torque; M.sub.acc: gondola acceleration torque; $M_{acc} = i_{ge} .Math. i \frac{{dn}_{DRV}}{dt} .Math. J_{Nac};$ i.sub.ge: transmission ratio between azimuth motor and gondola; n.sub.DRV: speed of azimuth motor; J.sub.Nac: mass moment of inertia of gondola M.sub.DRV: motor torque of azimuth motor.

17. A method for correcting at least one blade misalignment of a rotor blade of a rotor of a wind turbine having multiple rotor blades that have adjustable blade angles, wherein the blade misalignment represents a blade angle deviation of a detected blade angle of the rotor blade from a reference blade angle, and the wind turbine includes a gondola having the rotor and an azimuth adjustment device for adjusting the gondola in an azimuth alignment having an azimuth angle, and wherein the method comprises: to adjust the azimuth alignment, tracking, by the azimuth adjustment device, the azimuth angle to a predeterminable azimuth setpoint angle; detecting the blade misalignment as a function of an azimuth movement of the gondola; and for each rotor blade, determining a correction angle as a function of the respective detected blade misalignment; and correcting the blade angle using the correction angle, or comparing a detected blade misalignment to a predeterminable deviation limiting value and generating an error or warning message if an absolute value of the detected blade misalignment is greater than the deviation limiting value.

18. The method according to claim 17, wherein the correction angle corresponds to the detected blade misalignment, or the correction angle is tracked to the detected blade misalignment with a delay function having a time constant of at least one hour.

19. A wind turbine having a rotor having multiple rotor blades that have adjustable blade angles, wherein the wind turbine includes a gondola having the rotor and an azimuth adjustment device in order to adjust the gondola in an azimuth alignment having an azimuth angle, to adjust the azimuth alignment, the azimuth angle is tracked by means of the azimuth adjustment device to a predeterminable azimuth setpoint angle, and the wind turbine includes a controller configured to: detect a blade misalignment as a function of an azimuth movement of the gondola, wherein the blade misalignment represents a blade angle deviation of a detected blade angle of the rotor blade from a reference blade angle.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0122] The invention is explained in more detail by way of example hereinafter with reference to the appended figures.

[0123] FIG. 1 shows a wind turbine in a perspective illustration.

[0124] FIG. 2 schematically shows a rotor of a wind turbine in a front view.

[0125] FIG. 3 schematically shows a gondola having a rotor in a top view.

[0126] FIG. 4 schematically shows a model of a proposed state observer in a block diagram.

DETAILED DESCRIPTION

[0127] FIG. 1 shows a wind turbine 100 having a tower 102 and a gondola 104. A rotor 106 having three rotor blades 108 and a spinner 110 is arranged on the gondola 104. The rotor 106 is set into a rotational movement by the wind in operation and thus drives a generator in the gondola 104.

[0128] The wind turbine 100 includes an electrical generator 101, which is indicated in the gondola 104. Electrical power can be generated by means of the generator 101. A feed unit 105, which can be designed in particular as an inverter, is provided to feed electrical power into an electrical supply grid at a grid connection point PCC. A turbine controller 103 is provided to control the wind turbine 100 and also the feed unit 105.

[0129] FIG. 2 shows a rotor 200 having 3 rotor blades 202 in a top view from the front of a spinner 204. The rotor 200 is thus shown in FIG. 2 from the viewpoint of the wind. In operation, the wind results in a rotation of the rotor 200 in rotor rotational direction 206. The rotor 200 and thus each rotor blade 202 therefore has a rotor angle 208, which is also designated by the Greek letter φ.

[0130] The rotor angle 208 or φ can be related to a rotor reference angle 212, which is shown by way of example in FIG. 2 as a dashed vertical line. Accordingly, if this vertical were the reference variable and for a rotor blade which stands vertically upward (12 o'clock position), the rotor angle would thus have the value zero.

[0131] Only one rotor angle is shown by way of example for a rotor blade in FIG. 2. A rotor angle results accordingly for the other two rotor blades, which is greater or less by 120° or 240°.

[0132] A blade angle direction 210 is also shown by way of example in FIG. 2, however only by way of example for a rotor blade in which the relevant rotor blade can be adjusted in its rotor blade angle α.

[0133] FIG. 3 schematically shows a gondola 300 in a top view. It includes a rotor 306 having a first, second, and third rotor blade 301, 302, and 303. The second rotor blade 302 is located here in a 12 o'clock position, thus stands vertically upward. The first and third rotor blades 301, 303 are in an 8 o'clock or 4 o'clock position, respectively.

[0134] The gondola 300 is rotatable around a vertical tower axis 308 in the gondola rotational direction 310, thus adjustable in its azimuth alignment. A gondola azimuth torque 312 thus also points in this gondola rotational direction 310.

[0135] As an illustration, a wind 314 acting on the gondola 300, and thus the rotor 306, is shown or indicated in FIG. 3. The wind 314 acts on each of the three rotor blades 301 to 303 and thus results in each case in an aerodynamic blade torque 331 to 333 at the respective rotor blade 301 to 303. Each of these aerodynamic blade torques 331 to 333 results in each case in an azimuth torque component directed in the gondola rotational direction 310. This is dependent on the rotor angle φ, which bears the reference sign 208 in FIG. 2.

[0136] How strongly and in which direction the respective aerodynamic blade torque 331 to 333 contributes to the gondola azimuth torque 312 is thus dependent on this rotor angle φ. According to the definition of the rotor angle φ or 208 of FIG. 2, the respective azimuth torque component is calculated from the respective aerodynamic blade torque 331 to 333, multiplied by sin(φ). The azimuth torque component of the first rotor blade 301 is thus directed counter to the gondola azimuth torque 312, whereas the azimuth torque component of the third rotor blade 303 points in the direction of the gondola azimuth torque 312. With uniform wind, the azimuth torque components of the first and third rotor blade 301, 303 can cancel out in the position shown, and thus the azimuth torque components of all three rotor blades can cancel out, since the second rotor blade 302 has an azimuth torque component having the value zero in the vertical position.

[0137] However, if the individual rotor blades experience different forces from the wind and thus different aerodynamic blade torques, because their blade angles α are different, they do not cancel out, and result in a perceptible component of the gondola azimuth torque 312. This component changes due to rotation of the rotor, in particular cyclically and with changing sign.

[0138] The gondola azimuth torque 312 additionally has a further component due to an azimuth torque offset, which is not shown in FIG. 3, however.

[0139] For illustration, a blade profile is indicated for the second rotor blade 302 in FIG. 3, and a blade angle α is also shown therein for illustration. It is assumed here that an alignment of the rotor blade, namely of the second rotor blade 302 here, with its blade chord 316 in the direction of a rotor plane 318 corresponds to the blade angle zero. The blade angle α shown here by way of example thus has a blade angle α to this rotor plane 318, which in the example shown or the snapshot shown has approximately the value of 30°. In a part-load operation, the rotor blade angle α typically has a value of approximately 0° to 5°.

[0140] Such a blade angle α is predetermined identically for all rotor blades and is designated as the collective blade angle. An individual component can optionally be overlaid. Instead of the collective blade angle, an individual blade angle can also be used in each case for each rotor blade if a single blade adjustment is used. An individual component can then also be overlaid.

[0141] The blade misalignment to be detected is thus the deviation of a detected blade angle from the actual blade angle α, or from another reference blade angle, which is not shown in FIG. 3. The blade misalignment is to be detected individually for each rotor blade but can be detected in a common process.

[0142] To detect the blade misalignment of each rotor blade, the use of a state observer is proposed, which is based on a model that is shown in FIG. 4. FIG. 4 thus shows a model 400, in which a rotor angle φ, a wind velocity V.sub.W, a rotor speed n.sub.R, and a collective blade angle α are entered as input variables. It is assumed here that individual blade control is not provided, so that the same setpoint value for the blade angle is provided for each rotor blade. This is set for each rotor blade, wherein then the set rotor blade angle deviates in each case from the setpoint value by the blade misalignment. A motor speed n.sub.DRV of an azimuth motor is also entered as an input variable.

[0143] Furthermore, a blade misalignment γ.sub.A of a first rotor blade A, a second blade misalignment γ.sub.B of a second rotor blade B, and a third blade misalignment γ.sub.C of a third rotor blade C are taken into consideration. The actual rotor blade angle of the respective rotor blade then deviates by this blade misalignment from the setpoint value, thus the collective blade angle α. The first two blade misalignments γ.sub.A, γ.sub.B are shown as input variables, but are not input as input variables in the model 400. They are to be determined by the state observer using this model 400 and each form a system state of the observer.

[0144] The third blade misalignment γ.sub.C is calculated from these two first blade misalignments γ.sub.A, γ.sub.B. It corresponds to the negative value of the sum of both first two blade misalignments γ.sub.A and γ.sub.B. The subtraction block 402 is provided for this purpose. If the assumption that the sum of all three blade misalignments is zero is not made, the third blade misalignment γ.sub.C can also be used as an additional input variable.

[0145] The model 400 calculates for each of the three rotor blades A, B, and C in an aerodynamic block 411 to 413 an aerodynamic blade torque M.sub.aero,A, M.sub.aero,B, and M.sub.aero,C, which are only shown as M.sub.A, M.sub.B, and M.sub.C in FIG. 4 for better clarity.

[0146] Each of these aerodynamic blade torques M.sub.A, M.sub.B, and M.sub.C is transformed in a corresponding first, second, or third transformation block 421 to 423 with further consideration of its rotor angle into an azimuth torque component M.sub.yaw A, M.sub.yaw,B, and M.sub.yaw,C which are shown as M.sub.y,A, M.sub.y,B, and M.sub.y,C in FIG. 4 for better clarity.

[0147] To consider their respective rotor angle, the transformation blocks 421 to 423 receive the general rotor angle φ as an input variable and convert this accordingly into the individually applicable rotor angles, in that the rotor angle φ is increased or decreased by 120° or 240° if necessary.

[0148] Via the summation block 404, the three azimuth torque components M.sub.y,A, M.sub.y,B, and M.sub.y,C are added up to form the gondola azimuth torque M.sub.y, wherein an azimuth torque offset M.sub.y,O is also added. The gondola azimuth torque M.sub.y stands as a simplification for the formula symbol M.sub.yaw,Nac, and the azimuth torque offset M.sub.y,O stands as a simplification for the formula symbol M.sub.yaw,Offset.

[0149] The azimuth torque offset M.sub.y,O is shown as an input variable, but is not input into the model, rather forms a system state of the model and thus of the state observer, which is to be ascertained by the state observer. The gondola azimuth torque M.sub.y is also a system state in this model and thus the state observer.

[0150] The model 400 also considers the motor speed n.sub.DRV as an input variable, which is converted via a transmission factor i.sub.g in the first amplification block 406 into a gondola speed n.sub.G. The transmission factor i.sub.g designates the transmission ratio of an adjustment transmission between azimuth motor and gondola. The transmission ratio i.sub.g can also be designated by the formula symbol i.sub.ge or i.sub.gear.

[0151] In the derivation block 408, the gondola speed n.sub.G is derived according to time, so that a gondola acceleration a.sub.G results. For the sake of simplicity, a factor of 2π was omitted in the illustration. In the second amplification block 410, this gondola acceleration a.sub.G is calculated using the mass moment of inertia J.sub.G to form the gondola acceleration torque M.sub.A.

[0152] The mass moment of inertia J.sub.G of the gondola can also be designated by the formula symbol J.sub.Nac. The gondola acceleration torque M.sub.A can also be designated by the formula symbol M.sub.acc.

[0153] In the difference block 414, the difference of the gondola azimuth torque M.sub.y and the gondola acceleration torque M.sub.A is formed, so that an azimuth drive torque M.sub.AG results. This can be multiplied in the third amplification block 416 by the inverse transmission ratio i.sub.g, so that the motor torque M.sub.m results, for which the formula symbol M.sub.Motor can also be used.

[0154] The azimuth drive torque M.sub.AG and the motor torque M.sub.m form the output variable of the model and thus the output variable of the state observer. This estimated motor torque is compared to a detected motor torque, and returned via a corresponding feedback into the model in order to equalize the system states, thus the first blade incorrect angle γ.sub.A, the second blade incorrect angle γ.sub.B, the azimuth torque offset M.sub.y,O, and the gondola azimuth torque M.sub.Y.

[0155] The following was thus recognized and the following solutions were proposed.

[0156] Automatic detection of an aerodynamic imbalance based on a Kalman filter and a position-regulated azimuth drive is particularly proposed.

[0157] In wind turbines having azimuth holding brake, an aerodynamic imbalance results in a tower torsion, which is difficult to detect metrologically. Using the proposed use of position-regulated azimuth adjustment devices, such a holding brake is omitted in operation, since the drive is position-regulated and thus holds against the torque. Aerodynamic imbalances in the azimuth movement, which forms a regulation error, and in the drive torque of the motors, which forms a manipulated variable for the azimuth adjustment device, thus become visible.

[0158] A solution is proposed here in which Kalman filters, which are regulation observers in the broader meaning, observe the online estimation of parameters and the observation of regulation states. One advantage of such a solution is the lean structure and simple parameterization.

[0159] With a position-regulated azimuth adjustment device, the azimuth drive does not have a holding brake as in previous azimuth systems, because it is position-regulated. This means the motors are regulated so that the regulating error (difference of setpoint position and actual position) is as small as possible in that the motor torque is used as a manipulated variable. This means that the azimuth drive yields somewhat in the event of a torque, for example from the wind, and then actuates the setpoint position again.

[0160] The moment of the motor is proportional to the current and can thus be ascertained. In addition, each motor has a speed sensor and the gondola movement is thus also very well known due to the large transmission ratio of >1:1000.

[0161] The aerodynamic imbalance results in a gondola azimuth torque which is dependent on the blade position, thus the rotor angle of the blade misalignment and the wind velocity. This torque is counteracted by the motors of the azimuth drive and the inertia of the gondola.

[0162] The aerodynamic imbalance is estimated continuously and online based on the azimuth movements of the gondola with the aid of a Kalman filter. A detection of the azimuth drive torque is an important requirement. Alternatively, the gondola azimuth torque could also be detected, which can be ascertained at the position-regulated azimuth drive from the azimuth speed or its change and the torque of the azimuth motors.

[0163] The Kalman filter has a total of 4 states: [0164] gondola azimuth torque [0165] blade misalignment blade A [0166] blade misalignment blade B [0167] azimuth torque offset.

[0168] In this case, the gondola azimuth torque is the sum of the offset and the aerodynamic torques which results from the pitch misalignment of blade A, B, and C, which can also be designated synonymously as the blade misalignment. The aerodynamic torques are calculated from the pitch angle, which can also be designated synonymously as the blade angle, the pitch misalignment, the blade properties, and the wind velocity. The offset comprises all other torques which act on the azimuth drive, for example due to an oblique incident flow.

[0169] The azimuth motor torque is used as a “measured value” in the meaning of the Kalman filter, which is compared to the sum of the gondola azimuth torque and the acceleration torque (moment of inertia multiplied by azimuth acceleration). The Kalman filter now progressively optimizes the internal states, thus the states of the model of the Kalman filter, until the “measured value” from the Kalman filter corresponds in the best possible manner with the real measured value.

[0170] As a result, after a short time the pitch misalignments in blade A, B, and C are obtained, wherein it is assumed that the sum of all pitch misalignments corresponds to 0°.

[0171] Subsequently, the pitch misalignment can be corrected automatically in operation of the wind turbine and the imbalance can thus be avoided.

[0172] Particularly the following can be achieved or is to be achieved as much as possible using the proposed solution:

[0173] 1. Optimizing the turbine operation: Pitch misalignments result in a yield reduction of the wind turbine and increased sound emissions depending on the degree of the misalignment.

[0174] 2. Maintaining the certification conditions: The certification of the wind turbine is presently based on the assumption that the pitch angles deviate at most by +/−0.3° from one another.

[0175] 3. Load reduction/possible cost saving: Due to the automated recognition and elimination of the aerodynamic imbalance, the assumed pitch error, thus the pitch misalignment of +/−0.3° in simulations, in particular blade load simulations, can be reduced, by which a load reduction is achieved. A lesser aerodynamic imbalance reduces the load on the azimuth drive, due to which potentially fewer or weaker motors can be used.

[0176] The following alternative approaches can be avoided or improved:

[0177] Visual methods: The pitch angle is checked visually.fwdarw.Disadvantage: no turbine operation; complex measurement technology required; manual procedure.

[0178] Monitoring the tower torsion: If the azimuth torque is not known, the azimuth torque can also be ascertained from the tower torsion. However, this requires complex measurement technology and is thus quite costly.

[0179] Automatically recognizing the aerodynamic imbalance in operation of the wind turbine without additional hardware is sought.

[0180] This subsequently enables the correction of the pitch misalignment by adding an offset to the respective pitch angle.

[0181] The following advantages result:

[0182] Cost savings: no additional costs in the case of an azimuth adjustment device which uses a position regulation.

[0183] Minimizing risks: In all wind turbines, the aerodynamic imbalance is monitored and corrected and not only in conspicuous turbines.

[0184] Simplified and shortened startup: A determination of the aerodynamic imbalance by an external service provider can be omitted.

[0185] Simple software care and parameterization: Many parameters describe the physics of the wind turbine and are known or only have to be specified in their order of magnitude.

[0186] No dependencies for the regulation of the wind turbine and no cross-sensitivity to windshear, turbulence, or mass imbalance.

[0187] After approximately 10 minutes of operation of the wind turbine, the pitch misalignments of all blades are known.

[0188] The following is to be added to the Kalman filter used.

[0189] The Kalman filter operates so that it adapts system states of its model in dependence on a comparison of the output variable of the model and a corresponding measured variable of the system. The adaptation takes place so that a difference between the output variable of the model and the corresponding measured variable of the system is minimized. It is assumed that the measured variable and also the system states to be observed can have measurement noise.

[0190] A time-discrete description is used in the system description and therefore also in the model. Variables are observed therein at the current discrete point in time k and at the prior point in time k−1, thus at the step k and at the prior step k−1.

[0191] The states x.sub.k for the step k are calculated therein as a function of the states of the prior step k−1 and as a function of input variables u.sub.k for the current step k. The following generalized relationship reflects this:

x.sub.k=ƒ(x.sub.k-1,u.sub.k)

[0192] For the blade misalignments and the azimuth torque offset, it can be assumed that they are constant over time. The following then applies for them x.sub.k=x.sub.k-1.

[0193] The states are represented by the vector x.sub.k, in which all observed states are combined. The input variables are represented by the vector u.sub.k. In simple systems, only one state and/or only one input variable could also be present, which is not the case here, however.

[0194] Furthermore, it is presumed that the states have noise, having the covariance Q.

[0195] Measured values z.sub.k, thus output variables of the model, are calculated directly as a function of system states x.sub.k and input variables u.sub.k, which the following generalized relationship reflects:

z.sub.k=h(x.sub.k,u.sub.k)

[0196] The measured values are represented by a vector z.sub.k. In the present case, or according to a proposed aspect, however, only one measured value or only one output variable is present.

[0197] It is assumed that the measured variables have a noise R.

[0198] In the prior equations, ƒ(x.sub.k-1, u.sub.k) and h(x.sub.k, u.sub.k) can be designated as transfer functions, namely as the state transfer function ƒ(x.sub.k-1, u.sub.k) and as the output transfer function h(x.sub.k, u.sub.k). Their generalized representation indicates that the transfer functions can be nonlinear.

[0199] The mode of operation of the Kalman filter can be divided into two parts.

[0200] A prediction can be viewed as the first part. A linearization of the transfer functions is carried out:

[00005] ${{F_{k} = \frac{\partial f (x, u)}{\partial x} .Math.}_{x_{k - 1}, u_{k}} and H_{k} = \frac{\partial h (x, u)}{\partial x} .Math.}_{x_{k}, u_{k}}$

[0201] The linearization is thus carried out by partial derivation of the functions according to the respective states, specifically at the points x.sub.k-1 or x.sub.k and u.sub.k. A linearized state transfer matrix F.sub.k and an output transfer matrix H.sub.k result.

[0202] The states x.sub.k are predicted by the above-described relationship or calculated for this purpose, thus using the generalized relationship:

x.sub.k=ƒ(x.sub.k-1,u.sub.k)

[0203] In addition, a covariance of the observed states is predicted as the covariance matrix P.sub.k of the estimated values. The covariance matrix P.sub.k of the estimated values can also be designated synonymously as the covariance matrix P.sub.k of the observed states. For the prediction, it can be calculated for this purpose from a prior covariance matrix P.sub.k-1, the linearized state transfer matrix F.sub.k, and the covariance matrix Q.sub.k of the process noise, which thus identifies a process noise that is superimposed on the current system states by input variables. The covariance matrix P.sub.k of the estimated variables is calculated according to the equation:

P.sub.k=F.sub.k*P.sub.k-1*F.sub.k.sup.T+Q.sub.k

[0204] The correction, namely of the states, can be viewed as the second part.

[0205] For this purpose, initially a residual covariance matrix S.sub.k is calculated, namely from the current output transfer matrix H.sub.K, the covariance matrix P.sub.K of the estimated values, and a measured variable covariance matrix R.sub.k:

S.sub.k=H.sub.kP.sub.kH.sub.k.sup.T+R.sub.k

[0206] With the aid of the covariance matrix P.sub.k of the estimated values of the output transfer matrix H.sub.K and residual covariance matrix S.sub.k, the return amplification matrix K.sub.k can be calculated, which can also be designated as optimal amplification:

K.sub.k=P.sub.kH.sub.k.sup.TS.sub.k.sup.−1

[0207] Based thereon, the states x.sub.k are adapted in the observer, which can also be designated as updating these states x.sub.k. For this purpose, a difference is calculated from the measured output variable z.sub.k and its calculation from the estimated states x.sub.k of the observer and the input variables u.sub.k of the observer, according to the output transfer function h(x.sub.k, u.sub.k), using the following equation:

x.sub.k=x.sub.k+K.sub.k(z.sub.k−h(x.sub.k,u.sub.k))

[0208] A difference between measured and observed output variable is thus multiplied by the optimum amplification and added to the current states.

[0209] Finally, in preparation for the next step, an update of the current covariance matrix P.sub.k of the estimated values, thus the observed system states, is carried out according to the following equation:

P.sub.k=(I−K.sub.kH.sub.k)P.sub.k

[0210] The covariance matrix P.sub.k of the estimated values thus updated is then required in the next step in order to predict, thus to calculate, the new covariance matrix in the first part, namely the prediction, as was shown above. For this purpose, the covariance matrix P.sub.k of the estimated values of this step thus updated forms the prior covariance matrix P.sub.k-1 in the next step.

[0211] All of these calculations are repeated step-by-step, so that the now current step is then the prior step.

[0212] The Kalman filter is thus a recursive method, in which the return amplification matrix K.sub.k, which can also be designated as a correction matrix or correction vector, is newly determined or adapted in each step.

[0213] In summary, in each case for the step k: [0214] x.sub.k: observed states [0215] u.sub.k: input value [0216] z.sub.k: output variable [0217] F.sub.k: state transfer matrix [0218] H.sub.k: output transfer matrix [0219] K.sub.k: return amplification matrix [0220] P.sub.k: covariance matrix of the estimated values [0221] Q.sub.k: covariance matrix of the process noise [0222] R.sub.k: covariance matrix of the output variable [0223] S.sub.k: residual covariance matrix [0224] I: unit matrix (for each step)

[0225] The following initial values can be used for the exemplary case described in FIG. 4.

[0226] P.sub.k can be a 4×4 matrix, in which only the main diagonal can be occupied for P.sub.0. As initial values, thus values in the main diagonal, values can be selected according to a rough expectation, namely how good the initial values x.sub.0=0 are. For example, the values in the main diagonal can be selected as P.sub.0(1,1)=(10{circumflex over ( )}6){circumflex over ( )}2; P.sub.0(2,2)=(10{circumflex over ( )}6){circumflex over ( )}2; P.sub.0(3,3)=0.1{circumflex over ( )}2; P.sub.0(4,4)=0.1{circumflex over ( )}2

[0227] Precise initial values are not important, the order of magnitude is more fundamentally decisive than the correct value.

[0228] Q.sub.k and R.sub.k can be assumed to be constant for simplification. Q.sub.k characterizes how fast the states can change independently of the model equations. The main diagonal can be occupied here with: Q.sub.0(1,1)=0; Q.sub.0(2,2)=Q.sub.0(3,3)=(10{circumflex over ( )}(−7)){circumflex over ( )}2; Q.sub.0(4,4)=(10{circumflex over ( )}2){circumflex over ( )}2

[0229] R.sub.k describes the measurement noise and, for example, (10{circumflex over ( )}4){circumflex over ( )}2 can be assumed as the initial value.

[0230] According to one aspect, the following results, also with reference to FIG. 4:

x.sub.k=[M.sub.y,M.sub.y,o,γ.sub.A,γ.sub.B].sup.T

u.sub.k=[φ,v.sub.W,n.sub.R,α,n.sub.DRV].sup.T

z.sub.k=M.sub.m

[0231] The transfer functions result from the system description shown above, also from the relationships according to FIG. 4.

[0232] It is to be noted that if a Luenberger observer is used, no noise is taken into consideration, not for states, input variables, or output variables. It is therefore possible to return the observer deviation via a fixed correction vector. The recursive change of the return amplification matrix K.sub.k and the recursive calculations required for this purpose, which are necessary with the Kalman filter, can be omitted with the Luenberger observer. The fixed correction vector can be calculated beforehand.

[0233] System states can also be designated in simplified form and synonymously as states. They also include the states of the observer.

[0234] The various embodiments described above can be combined to provide further embodiments. These and other changes can be made to the embodiments in light of the above-detailed description. In general, in the following claims, the terms used should not be construed to limit the claims to the specific embodiments disclosed in the specification and the claims, but should be construed to include all possible embodiments along with the full scope of equivalents to which such claims are entitled. Accordingly, the claims are not limited by the disclosure.

METHOD FOR DETECTING A BLADE MISALIGNMENT OF A ROTOR BLADE OF A ROTOR OF A WIND TURBINE

Inventors

Cpc classification

Classification Explorer

F03D17/00

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Classification Explorer

F05B2270/328

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Classification Explorer

F03D7/0224

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Classification Explorer

F05D2260/83

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Classification Explorer

F05B2270/329

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Classification Explorer

G01B21/24

PHYSICS

International classification

Classification Explorer

G01B21/24

PHYSICS

Classification Explorer

F03D17/00

MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING

Abstract

Claims

Description