METHOD FOR COMPUTER-IMPLEMENTED MONITORING OF A WIND TURBINE

Abstract

A method for monitoring of a wind turbine including: a) acquiring respective values of one or more operation variables of the wind turbine at the corresponding operation time point; b) determining the value of a tower clearance resulting from the acquired values of the operation variables with the aid of a trained data driven model, where the value of an output variable which is the tower clearance or a variable correlated with the tower clearance is predicted by feeding the acquired value of each operation variable as a digital input to the trained data driven model which outputs the predicted value of the output variable as a digital output, the tower clearance being shortest the distance between the tower and the tip of a specific rotor blade from the rotor blades when the specific rotor blade is in its lowermost position in which it points downward in the vertical direction.

Claims

1. A method for computer-implemented monitoring of a wind turbine comprising a tower which extends in a vertical direction and a nacelle which is arranged on top of the tower, where an electric generator is disposed within the nacelle having several rotor blades is connected to the nacelle at a rotor hub, the rotor being configured to be rotated by wind around a rotor axis where a mechanical energy of a rotation of the rotor is converted into electric energy by the electric generator, the method comprising the following steps which are carried out for a corresponding operation time point of a number of operation time points of the wind turbine: a) acquiring respective values of one or more operation variables of the wind turbine at the corresponding operation time point based on sensor data recorded for the wind turbine; and b) determining a value of a tower clearance resulting from the acquired value or values of the one or more operation variables with an aid of a trained data driven model, where the value of an output variable which is the tower clearance or a variable correlated with the tower clearance is predicted by feeding the acquired value of each operation variable as a digital input to the trained data driven model which outputs the predicted value of the output variable as a digital output, wherein the trained data driven model is trained by training data sets, each comprising a value of the output variable for values of the one or more operation variables, the tower clearance being a shortest distance between the tower and a tip of a specific rotor blade from the several rotor blades when a specific rotor blade is in a lowermost position in which the specific rotor blade points downward in a vertical direction.

2. The method according to claim 1, wherein the output variable is a blade tip deflection of the specific rotor blade to the tower at the corresponding operation time point, where the tower clearance is derived in step b) from the blade tip deflection.

3. The method according to claim 1, wherein the one or more operation variables include one or more bending moments of the specific rotor blade of the rotor.

4. The method according to claim 3, wherein the one or more bending moments include a flapwise bending moment of the specific rotor blade at a root attached to the rotor hub or at a predetermined distance from the root and/or an edgewise bending moment of the specific rotor blade at the root attached to the rotor hub or at a predetermined distance from the root.

5. The method according to claim 3, wherein the one or more operation variables additionally include one or more of the following variables: an azimuth angle describing an angular position of the specific rotor blade around the rotor axis; a rotational speed; an electric power produced by the wind turbine; a pitch angle of the specific rotor blade describing an angle in which the specific rotor blade is attached to the rotor hub; a wind speed at the rotor hub along a rotor axis and/or the wind speed along at least one axis perpendicular to the rotor axis; a nacelle acceleration along the rotor axis and/or along an axis perpendicular to the rotor axis.

6. The method according to claim 1, wherein one or more predetermined actions are performed, in case that the value of the tower clearance determined in step b) falls below a predetermined threshold.

7. The method according to claim 6, wherein the one or more predetermined actions comprise one or more actions resulting in a greater tower clearance, the one or more actions including a change of a pitch angle of the rotor blades and/or a reduction of a rotational speed of the rotor.

8. The method according to claim 6, wherein the one or more predetermined actions comprise a generation of an alarm signal perceivable by surveillance staff and/or a stop of a rotation of the rotor.

9. The method according to claim 1, wherein the trained data driven model is a neural network.

10. The method according to claim 9, wherein the neural network is a feed-forward neural network or a convolutional neural network or a recurrent neural network.

11. The method according to claim 1, wherein the trained data driven model is a classification and regression tree classifier.

12. An apparatus for computer-implemented monitoring of a wind turbine comprising a tower which extends in a vertical direction and a nacelle which is arranged on top of the tower, where an electric generator is disposed within the nacelle and a rotor having several rotor blades is connected to the nacelle at a rotor hub, the rotor being configured to be rotated by wind around a rotor axis, where a mechanical energy of a rotation of the rotor is converted into electric energy by the electric generator, where the apparatus is configured to perform a method comprising the following steps which are carried out at a corresponding operation time point of a number of operation time points of the wind turbine: a) acquiring respective values of one or more operation variables of the wind turbine at the corresponding operation time point based on sensor data recorded for the wind turbine; and b) determining a value of a tower clearance resulting from the acquired value or values of the one or more operation variables with an aid of a trained data driven model, wherein the value of an output variable which is a tower clearance or a variable correlated with the tower clearance is predicted by feeding the acquired value of each operation variable as a digital input to the trained data driven model which outputs the predicted value of the output variable as a digital output, wherein the trained data driven model is trained by training data sets, each comprising a value of the output variable for values of the one or more operation variables, the tower clearance being a shortest distance between the tower and a tip of a specific rotor blade from the several rotor blades when the specific rotor blade is in a lowermost position in which the specific rotor blade points downward in a vertical direction.

13. The apparatus according to claim 12, wherein the apparatus is configured to perform a method of monitoring a wind turbine.

14. A computer program product, comprising a computer readable hardware storage device having computer readable program code stored therein, said program code executable by a processor of a computer system to implement the method according to claim 1 when the program code is executed on a computer.

15. A computer program with program code for carrying out the method according to claim 1 when the program code is executed on a computer.

Description

BRIEF DESCRIPTION

[0033] Some of the embodiments will be described in detail, with reference to the following figures, wherein like designations denote like members, wherein:

[0034] FIG. 1 is an embodiment of a wind turbine for which the method as described in the following is implemented;

[0035] FIG. 2 is schematic side views of a wind turbine illustrating the tower clearance occurring between the rotor and the tower of the turbine; and

[0036] FIG. 3 is a flow diagram illustrating the steps performed in an embodiment of the invention.

DETAILED DESCRIPTION

[0037] Embodiments of the invention as described in the following provides a method for computer-implemented monitoring of a wind turbine in order to detect small tower clearances between the rotor and the tower of the turbine. As a consequence, damages of the turbine due to a collision between a rotor blade and the tower are avoided.

[0038] FIG. 1 shows an embodiment of a wind turbine for which the method as described in the following is implemented. The wind turbine is designated with reference numeral 1 and comprises a tower 2 extending in the vertical direction. A nacelle 3 is located on top of this tower. An electric generator 4 is disposed within the nacelle. One side of the nacelle is connected to a rotor 5 having rotor blades 6 fixed to a rotor hub 7. The hub 7 is attached to the nacelle 3. The rotor 5 rotates around a horizontal rotor axis A which extends along the longitudinal direction of the nacelle. Three rotor blades 6 are attached to the rotor hub 7. The blades are arranged with an angular distance of 120° in the rotor plane which is perpendicular to the rotor axis A.

[0039] The position of the wind turbine 1 with respect to the earth's surface is described by a Cartesian coordinate system comprising an x-axis, a y-axis and a z-axis. The x-axis and the y-axis extend in a horizontal plane whereas the z-axis extends in a vertical direction perpendicular to the earth's surface. The vertical direction corresponds to the extension of the tower 2. The nacelle 3 can be rotated around a longitudinal axis of the tower, i.e. an axis parallel to the z-axis of the Cartesian coordinate system. In case of wind, the nacelle is positioned such that the wind substantially falls in the direction of the rotor axis A on the rotor 5, resulting in a rotation of the rotor.

[0040] Wind acting on the rotor blades 6 causes bending moments occurring at the respective rotor blades. This will result in a variation of the so-called tower clearance between the rotor 5 and the tower 2. In FIG. 1 and FIG. 2, this tower clearance is indicated by reference numerals TC. The tower clearance refers to the shortest distance between the tip of a rotor blade when passing its lowermost position and the surface of the tower.

[0041] FIG. 1 shows a scenario where no wind is acting on the rotor blades 6 of the rotor 5. Without wind, a relatively big tower clearance TC occurs between the rotor 5 and the tower 2. FIG. 2 shows the same wind turbine as FIG. 1 with the only difference that high wind is acting on the wind turbine. Evidently, this will result in a bending moment on the rotor blades 6 so that the rotor blades are bent towards the y-axis corresponding to the wind direction. This will reduce the tower clearance TC. In case of very strong winds, the tower clearance may become zero resulting in a collision between the rotor blades and the tower causing damages on the wind turbine.

[0042] In order to avoid such damages, the turbine 1 is monitored by the aid of a neural network in order to detect small tower clearances. The monitoring method is implemented in a controller (not shown) disposed within the wind turbine which can initiate countermeasures in case of small tower clearances. Contrary to methods as disclosed in the prior art, the tower clearance is not measured directly but by a prediction performed by a corresponding neural network.

[0043] FIG. 3 shows a flow diagram illustrating a monitoring method according to an embodiment of the invention which is used for the wind turbine as shown in FIG. 1 and FIG. 2. As already mentioned, this method uses a neural network designated as NN in FIG. 3. The neural network is trained by training data where each training data set refers to values of a plurality of sensors installed in the wind turbine 1 forming input data as well as to a value of the tower clearance of a specific rotor blade as an output occurring for the input data of the training data set.

[0044] In the method described herein, the training data are generated by numerical simulations which are performed by the well-known software BHawC which is an aeroelastic code intended for calculating wind turbine responses. This software simulates the wind turbine with all its components, including its sensors, for a range of environmental conditions. The software BHawC provides the tower clearance directly. Instead, the blade tip deflections provided by the software BHawC or by another aerolastic code may be used in order to derive the tower clearance therefrom.

[0045] In the embodiment described herein, the neural network NN is trained by training data comprising the tower clearance TC. Nevertheless, in an alternative embodiment, the neural network may also be learned by training data where the output refers to the above described blade tip deflections. In this case, the tower clearance is calculated afterwards from the output generated by the trained neural network NN.

[0046] In a preferred embodiment of the invention, the following sensor data are used as input data within the corresponding training data sets: [0047] the azimuth angle AA describing the angular position of the specific rotor blade around the rotor axis A; [0048] the rotational speed RS of the rotor 5; [0049] the electric power EP produced by the wind turbine 1; [0050] the pitch angle PI of the specific rotor blade describing the angle in which the specific rotor blade is attached to the rotor hub 7; [0051] the wind speed HWx in the direction of the x-axis, the wind speed HWy in the direction of the y-axis and the wind speed HWz in the direction of the z-axis, where said wind speeds are the speeds occurring at the rotor hub 7; [0052] the acceleration NAx of the nacelle 3 in the direction of the x-axis and the acceleration NAy of the nacelle 3 in the direction of the y-axis; [0053] the flapwise bending moment FW of the specific rotor blade at its root attached to the rotor hub 7 and the edgewise bending moment EM of the specific rotor blade at its root attached to the rotor hub 7.

[0054] As mentioned above, the output data of a respective training data set refers to the tower clearance TC of the specific rotor blade when its tip is in its lowermost position.

[0055] The neural network NN indicated in FIG. 3 is trained beforehand by using the above training data sets. In a preferred embodiment, a deep neural network is used. This neural network comprises more than one hidden layer between its input layer and its output layer. E.g., a feed-forward network or a recurrent neural network may be used in the method of embodiments of the invention. However, any other data driven model learned by machine learning may be implemented in the method of embodiments of the invention, e.g. a decision tree classifier trained by corresponding training data.

[0056] The trained neural network NN is implemented in the controller of the wind turbine. As indicated in FIG. 3, the trained neural network NN receives during the operation of the wind turbine input data IN at corresponding operation time points t. The input data are measured values of the above-mentioned quantities AA, RS, EP, PI, HWx, HWy, HWz, NAx, NAy, FM and EM. In step S1 indicated in FIG. 3, the trained neural network NN predicts from those values the output variable OV which is the tower clearance TC.

[0057] In a subsequent step S2, the predicted value of the tower clearance TC at the corresponding operation time point t is compared with a threshold TH. In case that the tower clearance falls below the threshold, there is the risk of collisions between the rotor blades and the tower. Hence, in case that the tower clearance is below the threshold TH, one or more predetermined actions AC are initiated by the controller of the wind turbine 1.

[0058] In a particularly preferred embodiment, the predetermined actions refer to a reduction of the rotational speed of the rotor 5 of the turbine 1 and/or a suitable change of the pitch angle of all rotor blades 6. Those measures have the consequence of a greater tower clearance TC reducing the risk that rotor blades 6 collide with the tower 2.

[0059] Additionally, or alternatively, an alarm signal may also be generated in case that the tower clearance TC falls below the threshold TH. This alarm signal is transmitted to a control station so that the staff in control station is informed about the risk of a collision between the rotor blades and the tower. Thereafter, the staff may initiate appropriate countermeasures in order to lower the risk of a collision between the rotor blades and the tower. In another embodiment, a predetermined action may also refer to an emergency stop, i.e. an immediate stop of the rotation of the rotor. This emergency stop will be initiated in case of very low tower clearances predicted by the trained neural network NN of FIG. 3.

[0060] In another embodiment of the invention, only a part of the input data IN indicated in FIG. 3 are used for training the neural network and performing predictions by the trained neural network. In this respect, the inventors found out that only using the azimuth angle AA, the flapwise bending moment FM and the edgewise bending moment EM as the input data will result in reasonable predictions with less computational effort in comparison to the use of all input data indicated in FIG. 3.

[0061] The method as described in the foregoing was tested by the inventors based on a neural network in the form of a feed-forward network having two hidden layers, each layer comprising fifty nodes. Inter alia, the neural network was learned with all the input data indicated in FIG. 3 and also with only the input data referring to the azimuth angle and the flapwise and edgewise bending moments. The inventors could show that quite accurate predictions of the tower clearance could be achieved. The coefficient of determination R.sup.2 which is a well-known quantity was calculated for predictions of the trained neural networks. Coefficients in the range between 0.97 and 0.99 could be achieved.

[0062] Embodiments of the invention as described in the foregoing have several advantages. Particularly, the tower clearance of a wind turbine can be monitored in an easy and efficient way without installing sensors at the tip of the rotor blades. This is achieved by the aid of a neural network predicting the tower clearance based on other sensor data of the wind turbine. To do so, the neural network is trained by training data. The training data can be simulated data, e.g. taken from the well-known software BHawC. Nevertheless, the training data may also be extracted from real measurements of a corresponding wind turbine.

[0063] Although the present invention has been disclosed in the form of preferred embodiments and variations thereon, it will be understood that numerous additional modifications and variations could be made thereto without departing from the scope of the invention.

[0064] For the sake of clarity, it is to be understood that the use of “a” or “an” throughout this application does not exclude a plurality, and “comprising” does not exclude other steps or elements.