A METHOD FOR COMPUTER-IMPLEMENTED MONITORING OF A COMPONENT OF A WIND TURBINE

Abstract

Provided is a method for computer-implemented monitoring of a component of a wind turbine, having access to a trained machine learning model which has been trained for one or more components of the same type of wind turbines. The trained machine learning model is configured to provide an output referring to a predetermined fault occurring at a component of a wind turbine by processing vibration signals in a predetermined domain which are measured in the vicinity of the component during the operation of the wind turbine. Vibration signals are mapped to corresponding vibration signals valid for the component based on one or more given kinematic parameters of the component and one or more given kinematic parameters of another component. The machine learning model is applied to the vibration signals valid for the component, resulting in an output referring to the predetermined fault occurring at the another component.

Claims

1. A method for computer-implemented monitoring of a component of a wind turbine, where the wind turbine is a first wind turbine and the component is a first component and where the method has access to a trained machine learning model which has been trained for one or more second components of a same type of one or more second wind turbines, the one or more second wind turbines being wind turbines of another type than the first wind turbine, where the trained machine learning model is configured to provide an output referring to a predetermined fault occurring at a second component of a second wind turbine by processing vibration signals in a predetermined domain which are measured in a vicinity of the second component during an operation of the second wind turbine, the method comprising: i) providing vibration signals in the predetermined domain measured in a vicinity of the first component during an operation of the first wind turbine; ii) mapping the vibration signals to corresponding vibration signals valid for the second component based on one or more given kinematic parameters of the first component and one or more given kinematic parameters of the second component; and iii) applying the machine learning model to the vibration signals valid for the second component, resulting in an output referring to the predetermined fault occurring at the first component.

2. The method according to claim 1, wherein the predetermined domain is the frequency domain or the cepstrum domain.

3. The method according to claim 1, wherein the one or more given kinematic parameters of the first component are described by a same function type as the one or more given kinematic parameters of the second component but with different function parameters.

4. The method according to claim 1, wherein the one or more given kinematic parameters of the first component are one or more specific values within the predetermined domain contained within the vibration signals measured in the vicinity of the first component in case that the predetermined fault occurs, and wherein the one or more given kinematic parameters of the second component are one or more specific values within the predetermined domain contained within the vibration signals measured in the vicinity of the second component in case that the predetermined fault occurs.

5. The method according to claim 1, wherein the component being monitored is a part or the drivetrain of the wind turbine.

6. The method according to claim 1, wherein the predetermined fault refers to a damage of a gearwheel or a damage of a bearing race or a damage of balls or rollers in a ball or roller bearing.

7. The method according to claim 1, wherein the machine learning model is based on one or more neural networks or on Principal Component Analysis.

8. An apparatus for monitoring of a component of a wind turbine, where the wind turbine is a first wind turbine and the component is a first component and where the method has access to a trained machine learning model which has been trained for one or more second components of a same type of one or more second wind turbines, the one or more second wind turbines being wind turbines of another type than the first wind turbine, where the trained machine learning model is configured to provide an output referring to a predetermined fault occurring at a second component of a second wind turbine by processing vibration signals in a predetermined domain which are measured in a vicinity of the second component during an operation of the second wind turbine, the apparatus comprising: a means for providing vibration signals in the predetermined domain measured in a vicinity of the first component during an operation of the first wind turbine; means for mapping the vibration signals to corresponding vibration signals valid for the second component based on one or more given kinematic parameters of the first component and one or more given kinematic parameters of the second component; and a means for applying the machine learning model to the vibration signals valid for the second component, resulting in an output referring to the predetermined fault occurring at the first component.

9. The apparatus according to claim 8, wherein the apparatus is configured to perform a method for monitoring a component of a wind turbine.

10. 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.

11. 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

[0022] Some of the embodiments will be described in detail, with reference to the following figures, wherein like designations denote like members, wherein:

[0023] FIG. 1 shows an example of a wind turbine being monitored by an embodiment of the invention;

[0024] FIG. 2 shows a wind turbine based on which the machine learning model used for monitoring the turbine of FIG. 1 has been trained;

[0025] FIG. 3 is a flowchart showing the steps of a method according to an embodiment the invention; and

[0026] FIG. 4 is a diagram illustrating the components used for performing the method shown in FIG. 3.

DETAILED DESCRIPTION

[0027] The method as described in the following is used for monitoring a component in a wind turbine. An example of such a wind turbine is shown in FIG. 1. The wind turbine is designated with reference numeral 1 and comprises a tower 2 on top of which a nacelle 3 is arranged. The nacelle is connected to a rotor having a hub 4 at which three rotating blades 5 are attached. A drivetrain for the wind turbine is accommodated within the nacelle 3. In the embodiment of FIG. 1, this drivetrain comprises a main bearing 6 for supporting the hub of the rotor as well as a gearbox 7 and a generator 8 for converting the mechanical power generated by the rotation of the rotor into electric power. Within the nacelle 3, there are provided three vibration sensors 9, 10 and 11. The vibration sensor 9 is located adjacent to the bearing 6, the vibration sensor 10 is located adjacent to the gearbox 7 and the vibration sensor 11 is located adjacent to the generator 8. Embodiments of the invention are not restricted to the use of just three vibration sensors. I.e., in other embodiments, a higher or lower number of vibration sensors may be used.

[0028] In the embodiment described herein, one of the components 6 to 8 of the drivetrain is monitored by processing the vibration signals detected by the sensor adjacent to the component. However, the method described in the following may be used for each of the components 6 to 8 so that all components within the nacelle are monitored. In another embodiment, the signals of several vibration sensors detecting vibrations of a respective components may be used for monitoring the respective component. Without loss of generality, embodiments of the invention are described in the following for monitoring the gearbox 7 based on the vibration signals of sensor 10.

[0029] As a prerequisite of the method described herein, there already exists a pre-trained machine learning model. This machine learning model has been trained by training data from another turbine than the turbine 1. This turbine is shown in FIG. 2 and designated with reference numeral 1′. The turbine has the same components as the turbine in FIG. 1. Nevertheless, the turbine 1′ is a turbine of another type than turbine 1 which is indicated by using apostrophes for the elements of the turbine in FIG. 2. The turbine 1′ may be a turbine having the same construction as the turbine 1 but being from another manufacturer. In a modified embodiment, the machine learning model has been trained based on training data from several turbines 1′ of the same type.

[0030] For the turbine 1′, the above-mentioned machine learning model has been trained based on vibration signals of the sensor 10′ where it was known whether the vibration signals refer to an operation of the wind turbine 1′ where a specific fault and particularly a specific damage was present within the gearbox 7′. The training was based on a high number of training data sets including a considerable amount of vibrations signals referring to an operation accompanied by the specific fault within the gearbox 7′. Depending on the circumstances, the machine learning model can be trained for different fault types. E.g., a fault type may refer to a damage of a gearwheel or a damage of an inner race or outer race of a bearing or a damage of the balls or rollers of a bearing. The machine learning model was trained for one of such particular fault types.

[0031] In a preferred embodiment, the machine learning model refers to a neural network. However, other machine learning models, e.g. Principal Component Analysis, may be used. The machine learning model may be trained by any suitable learning method. In the embodiment described herein, the learning method is based on supervised learning where the training data include the information whether a fault is present in the gearbox for the respective training data sets. Nevertheless, the machine learning model may also be used in combination with a training based on unsupervised learning.

[0032] In order to use the trained machine learning method with respect to the turbine 1′, vibration signals of the vibration sensor 10′ are input into the model during the operation of the wind turbine 1′ resulting in an output indicative of the specific fault. Depending on the circumstances, the output may be such that it indicates if the fault is present or not. However, the output may also be a probability with respect to the presence of the specific fault.

[0033] The idea of embodiments of the invention is to enable a monitoring of the turbine 1 based on the trained machine learning model although the training of the model was performed for another wind turbine 1′. This is achieved by an appropriate mapping taking into account kinematic parameters with respect to both turbines 1 and 1′. This will be described in the following with respect to FIG. 3 and FIG. 4.

[0034] FIG. 3 shows a flowchart illustrating an embodiment of the invention for monitoring the gearbox 7 of the turbine 1. In step S1, vibration signals VS originating from the sensor 10 are recorded. The vibration signals need not refer to the direct sensor outputs but may be pre-processed signals. In the embodiment described herein, the vibration signals are the sensor data after having applied a Fourier transform, i.e. the vibration signals are signals within the frequency domain.

[0035] In a step S2, the vibration signals VS valid for the sensor 10 of the turbine 1 are converted by a mapping into vibration signals VS' which are vibration signals which would have been occurred in the turbine 1′ in case that the operation states of both gearboxes 7 and 7′ were the same. I.e., in case that the vibration signals VS refer to a fault within the gearbox 7, this fault would also be present in the gearbox 7′ if vibration signals VS' were detected. In order to perform the mapping, the knowledge of kinematic parameters KP1 with respect to wind turbine 1 and of kinematic parameters KP2 with respect to turbine 1′ is used. The kinematic parameters KP1 refer to the component 7 and the specific fault being monitored. Analogously, the kinematic parameters KP2 refer to the component 7′ and the specific fault being monitored.

[0036] In the following, an example for a mapping from VS to VS' is described. As the kinematic parameters for both turbines 1 and 1′, so-called damage frequencies occurring at the specific fault are used. Those damage frequencies are known beforehand and can be expressed by the following formula:

f=n.Math.f.sub.c+z.Math.f.sub.m  (1)

[0037] The frequencies f.sub.c and f.sub.m are completely described by the kinematics of the respective components 7 and 7′ and the rotation speed of the shaft. In the above formula, f.sub.c represents the gearmesh frequency in case that a damage of a gearwheel is the specific fault. Analogously, f.sub.c may refer to the frequency of an inner race or an outer race in case of a specific fault referring to cracks in these races or it may refer to a ball or roller spin frequency in case of a specific fault referring to the balls or rollers within a ball/roller bearing. The frequency f.sub.c is known but is different for the components 7 and 7′. Furthermore, f.sub.m in the above formula represents a known sideband frequency present when the specific fault occurs. Analogously to the frequency f.sub.c, the frequency f.sub.m is known but is different for the components 7 and 7′. Moreover, n, z are integers.

[0038] In the following, the frequencies occurring in the gearbox 7 of wind turbine 1 are designated by the index A, whereas the frequencies occurring in the gearbox 7′ of wind turbine 1′ are designated as B. In other words, index A refers to the component 7 and index B refers to the component 7′.

[0039] In the embodiment described herein, a linear mapping T(⋅) is used for converting the vibration signals VS into the vibrations signals VS′. I.e., the mapping is described by the function T(x)=ax+b. However, other mappings than linear mappings also provide good results in case that a corresponding mapping can be found.

[0040] Using the above formula (1), the kinematic parameters for the gearboxes 7 and 7′ are as follows:

f.sup.A=n.Math.f.sup.A.sub.c+z.Math.f.sup.A.sub.m

f.sup.B=n.Math.f.sup.B.sub.c+z.Math.f.sup.B.sub.m

[0041] Based on the above equations, parameters a and b shall be determined such that f.sup.B=T(f.sup.A). Thus, the following applies:

f.sup.B=T(f.sup.A)⇔n.Math.f.sup.B.sub.c+z.Math.f.sup.B.sub.m=a.Math.(n.Math.f.sup.A.sub.c+z.Math.f.sup.A.sub.m)+b⇔0=n.Math.(a.Math.f.sup.A.sub.c−f.sup.B.sub.c)+z.Math.(a.Math.f.sup.Am−f.sup.Bm)+b

[0042] In order to determine a and b, the following two equations (i) and (ii) are solved:

0=n.Math.(a.Math.f.sup.A.sub.c−f.sup.B.sub.c)+b  (i)

0=z.Math.(a.Math.f.sup.A.sub.m−f.sup.B.sub.m)  (ii)

[0043] Solving equation (ii) will provide the parameter a as follows:

a=f.sup.B.sub.m/f.sup.A.sub.m for z≠0  (ii)

[0044] Solving equation (i) will provide the parameter b as follows:

b=n.Math.(f.sup.B.sub.c−a.Math.f.sup.A.sub.c)=n.Math.(f.sup.B.sub.c−f.sup.A.sub.c.Math.f.sup.B.sub.m/f.sup.A.sub.m)=n.Math.(f.sup.B.sub.c.Math.f.sup.A.sub.m−f.sup.A.sub.c.Math.f.sup.B.sub.m)/f.sup.A.sub.m  (i)

[0045] The above mapping is dependent on n. The mapping was derived within the frequency domain. Nevertheless, the mapping may also be derived in the same way for the well-known cepstrum-domain.

[0046] Based on the mapping T, the vibration signals VS are converted in step S2 into the vibration signals VS' as shown in FIG. 3. In a next step S3, the vibration signals VS' are fed as input data to the trained machine learning model which is designated as ML in FIG. 3. As mentioned above, the machine learning model provides an output indicative of the specific fault where this fault is designated as FT in FIG. 3. Hence, when applying the machine learning model ML to the vibration signals VS′, the output OU referring to the fault FT is generated. Due to the mapping, this output will indicate whether there is a fault within the gearbox 7 of the wind turbine 1 although the machine learning model has been trained for a different gearbox 7′ in another wind turbine 1′.

[0047] Embodiments of the invention as described in the foregoing are based on the knowledge that vibration signals of rotating machines are characterized by variations in amplitudes and frequencies that can be explained by the kinematics of the machine components. These characteristics change when anomalies occur. Thus, a fault or damage type has a vibrational pattern that can be described by the component kinematics. Therefore, for each fault type, the vibration signals from one component can be mapped to another component using the kinematics of both components.

[0048] FIG. 4 shows an apparatus for performing the method described with respect to FIG. 3. The apparatus comprises a means M1 providing the vibration signals VS. In the embodiment described herein, means M1 comprises the vibration sensor 10 detecting the vibrations. Nevertheless, in a modified embodiment, the vibration sensor is not part of the apparatus. In this case, the vibration signals have been acquired and stored beforehand and means M1 just accesses the storage for retrieving the vibration signals. To do so, means M1 is part of a computer program implemented on a computer. This computer program also comprises means M2 and M3 which are described in the following.

[0049] Means M2 performs the mapping as shown in step S2 of FIG. 3. The vibration signals VS' resulting from this mapping are processed by means M3 which performs step S3 of FIG. 3. I.e., the machine learning model ML is implemented in means M3 and, by applying this model to the vibration signals VS′, the output OU indicative of the specific fault is generated. This output may be stored in a storage and/or may be provided on a user interface, e.g. a display, in order to inform the user about the occurrence of a fault.

[0050] The embodiment described above has several advantages. Particularly, a pre-trained machine learning model can be used for monitoring a wind turbine other than the wind turbine for which the machine learning model has been trained. This can be achieved by an appropriate mapping based on known kinematic parameters with respect to the monitored wind turbine and the wind turbine used for training the machine learning model. Due to the use of a pre-trained machine learning method, there is no need for developing fault detection algorithms for new wind turbines. Moreover, the method of embodiments of the invention artificially generates new training data and allows for further development and design of new machine learning algorithms.

[0051] 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.

[0052] 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.