Wind turbine drivetrain wear detection using azimuth variation clustering

1. A method for monitoring a wind turbine azimuth drivetrain, the method comprising:

accumulating a plurality of azimuth variation characteristics data from at least one wind turbine over a period of time;

identifying a plurality of clusters of values within the plurality of azimuth variation characteristics data, where each cluster of values in the plurality of azimuth variation characteristics data is associated with a respective wind turbine;

associating each respective cluster of values in the plurality of clusters of values with a respective condition of a respective main drivetrain for the respective wind turbine associated with the respective cluster of values;

identifying at least one selected cluster of values that comprise measurements made in time proximate to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance;

based on identifying the at least one selected cluster, training a machine learning model with the at least one selected cluster labeled as corresponding to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance, where the machine learning model is configured to process azimuth variation characteristics to identify received azimuth variation characteristics data that correspond to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

receiving, subsequent to associating the respective cluster of values with the respective condition, a measured set of azimuth variation characteristics data for a subject azimuth drivetrain of a subject wind turbine;

processing, subsequent to and based on training the machine learning model, the measured set of azimuth variation characteristics data with the machine learning model to determine an identified cluster with values of azimuth variation characteristics data that corresponds to values in the measured set of azimuth variation characteristics data for the subject azimuth drivetrain, where the identified cluster has been determined to, based on training of the machine learning model, to comprise azimuth variation characteristics corresponding to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

determining that a respective condition associated with the identified cluster is a subject condition associated with a subject main drivetrain of the subject wind turbine; and

reporting the subject condition as a condition of the subject main drivetrain.

2. The method of claim 1 , wherein the respective condition of the main drivetrain comprises a condition of likely to have upcoming damage to the main drivetrain.

3. The method of claim 1 , further comprising

associating each cluster of values in the plurality of clusters of values with a respective condition of a respective azimuth drivetrain of the respective wind turbine associated with the respective cluster of values; and

determining that a respective condition associated with the identified cluster is a subject condition associated with the subject azimuth drivetrain,

where the reporting further comprises reporting the subject condition as a condition of the subject azimuth drivetrain.

4. The method of claim 1 , where identifying the plurality of clusters of values comprises processing the azimuth variation characteristics data with an unsupervised learning process to identify clusters of values within the azimuth variation characteristics data.

5. The method of claim 1 , wherein the training a machine learning model comprises:

the identifying a plurality of clusters of values within the plurality of azimuth variation characteristics data;

the associating each respective cluster of values with the respective condition of the respective main drivetrain; and

the identifying the subject condition based upon processing by the machine learning model.

6. The method of claim 1 , further comprising:

receiving, in association with each respective value within the plurality of azimuth variation characteristics data, a respective control mode indication within a plurality of control modes of the at least one wind turbine that was in effect during measurement of the each respective value within the plurality of azimuth variation characteristics data; and

associating, based on the respective control mode indication that was in effect during measurement of each respective value within the plurality of azimuth variation characteristics data, each respective value in the plurality of azimuth variation characteristics data with a respective bin of a plurality of bins, where each respective bin is associated with a respective control mode of the at least one wind turbine that is associated with that respective bin,

where the identifying the plurality of clusters of values identifies separate clusters of values within each bin of the plurality of bins,

where a measurement control mode within the plurality of control modes was in effect for the at least one wind turbine when the measured set of azimuth variation characteristics data was measured, and

where determining the identified cluster comprises identifying the identified cluster within the respective bin associated with the measurement control mode.

7. The method of claim 1 , further comprising:

receiving maintenance data indicating a plurality of maintenance actions performed on the at least one wind turbine over the period of time,

where associating each respective cluster of values in the plurality of clusters of values with a respective condition of a respective main drivetrain comprises:

determining at least one respective set of azimuth variation characteristics data within the plurality of azimuth variation characteristics data that is associated with at least one maintenance action within the plurality of maintenance actions;

determining a particular condition of the main drivetrain that is associated with the at least one maintenance action; and

associating the particular condition with a cluster of values comprising the at least one respective set of azimuth variation characteristics.

8. The method of claim 7 , where determining at least one respective set of azimuth variation characteristics data within the plurality of azimuth variation characteristics data that is associated with at least one maintenance action within the plurality of maintenance actions comprises performing association rule learning.

9. A wind turbine azimuth drivetrain monitor, comprising:

a model building engine that, when operating:

accumulate a plurality of azimuth variation characteristics data from at least one wind turbine over a period of time;

identifies a plurality of clusters of values within the plurality of azimuth variation characteristics data, where each cluster of values in the plurality of azimuth variation characteristics data is associated with a respective wind turbine; and

associates each respective cluster of values in the plurality of clusters of values with a respective condition of a respective main drivetrain for the respective wind turbine associated with the respective cluster of values;

identify at least one selected cluster of values that comprise measurements made in time proximate to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance; and

a model based monitoring engine that, when operating:

based on identifying the at least one selected cluster, train a machine learning model with the at least one selected cluster labeled as corresponding to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance, where the machine learning model is configured to process azimuth variation characteristics to identify received azimuth variation characteristics data that correspond to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

receives, subsequent to associating the respective condition, a measured set of azimuth variation characteristics data for a subject azimuth drivetrain of a subject wind turbine;

process, subsequent to and based on training the machine learning model, the measured set of azimuth variation characteristics data with the machine learning model to determine an identified cluster with values of azimuth variation characteristics data that corresponds to values in the measured set of azimuth variation characteristics data for a subject azimuth drivetrain, where the identified cluster has been determined to, based on training of the machine learning model, to comprise azimuth variation characteristics corresponding to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

determines that a respective condition associated with the identified cluster is a subject condition associated with a subject main drivetrain of the subject wind turbine; and

reports the subject condition as a condition of the subject main drivetrain.

10. The wind turbine azimuth drivetrain monitor of claim 9 , where the model building engine, when operating, identifies the plurality of clusters of values by at least identifying values of azimuth variation characteristics data that are within a threshold value of one another.

11. The wind turbine azimuth drivetrain monitor of claim 9 , where the model building engine, when operating, further:

receives, in association with each respective value within the plurality of azimuth variation characteristics data, a respective control mode indication within a plurality of control modes of the at least one wind turbine that was in effect during measurement of the each respective value within the plurality of azimuth variation characteristics data; and

associates, based on the respective control mode indication that was in effect during measurement of each respective value within the plurality of azimuth variation characteristics data, each respective value in the plurality of azimuth variation characteristics data with a respective bin of a plurality of bins, where each respective bin is associated with a respective control mode of the at least one wind turbine that is associated with that respective bin,

where the model building engine, when operating, identifies the plurality of clusters of values by at least identifying separate clusters of values within each bin of the plurality of bins,

where a measurement control mode within the plurality of control modes was in effect for the at least one wind turbine when the measured set of azimuth variation characteristics data was measured, and

where the model building engine, when operating, determines the identified cluster by at least identifying the identified cluster within the respective bin associated with the measurement control mode.

12. The wind turbine azimuth drivetrain monitor of claim 9 , where the model building engine, when operating, further:

receives maintenance data indicating a plurality of maintenance actions performed on the at least one wind turbine over the period of time,

where the model building engine, when operating, associates the respective condition of the main drivetrain with each cluster of values by at least:

determining at least one respective set of azimuth variation characteristics data within the plurality of azimuth variation characteristics data that is associated with at least one maintenance action within the plurality of maintenance actions;

determining, by at least performing association rule learning, a particular condition of the main drivetrain that is associated with the at least one maintenance action; and

associating the particular condition with a cluster of values comprising the at least one respective set of azimuth variation characteristics.

13. A method for monitoring a wind turbine azimuth drivetrain, the method comprising:

storing a definition of a plurality of clusters of values within a plurality of azimuth variation characteristics data,

where the plurality of azimuth variation characteristics data is accumulated from at least one wind turbine over a period of time, and

where each respective cluster of values in the plurality of clusters of values is associated with a respective condition of a respective main drivetrain for a respective wind turbine associated with the respective cluster of values;

identifying at least one selected cluster of values that comprise measurements made in time proximate to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance;

based on identifying the at least one selected cluster, training a machine learning model with the at least one selected cluster labeled as corresponding to one of a drive mechanism failure or an indication of a drive mechanism requiring maintenance, where the machine learning model is configured to process azimuth variation characteristics to identify received azimuth variation characteristics data that correspond to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

receiving a measured set of azimuth variation characteristics data for a subject azimuth drivetrain of a subject wind turbine;

processing, subsequent to and based on training the machine learning model, the measured set of azimuth variation characteristics data with the machine learning model to determine an identified cluster within the plurality of clusters of values, the identified cluster comprising values of azimuth variation characteristics data that correspond to values in the measured set of azimuth variation characteristics data for the subject azimuth drivetrain, where the identified cluster has been determined to, based on training of the machine learning model, to comprise azimuth variation characteristics corresponding to the one of a drive mechanism failure or the indication of a drive mechanism requiring maintenance;

determining that a respective condition associated with the identified cluster is a subject condition associated with a subject main drivetrain; and

reporting the subject condition as a condition of the subject main drivetrain of the subject wind turbine.

14. The method of claim 13 , wherein the respective condition of the main drivetrain comprises likely to have upcoming damage to the main drivetrain.

15. The method of claim 14 , further comprising:

dispatching, based upon determining the subject condition is in the second set of conditions, repair crews to repair the subject azimuth drivetrain.

16. The method of claim 14 , further comprising:

monitoring, based upon determining the subject condition is in the first set of conditions, subsequently measured sets of azimuth variation characteristics data for the subject azimuth drivetrain of the subject wind turbine to determine a progression of wear of the subject azimuth drivetrain; and

scheduling, based on the determined progression of wear of the subject azimuth drivetrain, maintenance of the subject azimuth drivetrain.

17. The method of claim 16 , further comprising:

selecting an identified time for maintenance of the subject azimuth drivetrain based on anticipated generation requirements for the subject wind turbine,

where scheduling maintenance of the subject azimuth drivetrain is based on the identified time for maintenance of the subject azimuth drivetrain.

18. The method of claim 17 , wherein the monitoring subsequently measured sets of azimuth variation characteristics data comprises determining a rate of wear of the subject azimuth drivetrain, and

wherein the identified time is selected based on the determined rate of wear.

19. The method of claim 17 , wherein the monitoring subsequently measured sets of azimuth variation characteristics data comprises determining a projected remaining time until damage of the subject azimuth drivetrain, and

wherein the identified time is selected to be before an expiration of the remaining time until damage of the subject azimuth drivetrain.

20. The method of claim 13 ,

where a respective condition of the azimuth drivetrain is associated with each cluster of values in the plurality of clusters of values,

where the plurality of clusters of values comprises a first set of clusters and a second set of clusters, where

a first set of conditions is associated with clusters in the first set of clusters where each condition in the first set of conditions is associated with azimuth drivetrain wear, and

a second set of conditions is associated with clusters in the second set of clusters where each condition in the second set of conditions is associated with azimuth drivetrain damage;

where reporting the subject condition comprises:

notifying, based upon determining the subject condition is in the first set of conditions, a worn condition of the subject azimuth drivetrain, and

notifying, based upon determining the subject condition is in the second set of conditions, a damaged condition of the subject azimuth drivetrain.