Method for windmill farm monitoring

Patent metadata

Field	Value
Publication number	US-11339763-B2
Application number	US-201715828450-A
Country	US
Kind code	B2
Filing date	Dec 1, 2017
Priority date	Jun 3, 2015
Publication date	May 24, 2022
Grant date	May 24, 2022

Abstract

A method for monitoring turbines of a windmill farm includes: providing a global nominal dataset containing frame data of the turbines of the windmill farm and continuous reference monitoring data of the turbines for a first period in a fault free state, the reference monitoring data including at least two same monitoring variables for each turbine; building a nominal global model based on the global nominal dataset which describes the relationship in between the windmill turbines and clustering the turbines according thereto; assigning the data of the global nominal dataset to respective nominal local datasets according to the clustering; and building a nominal local model for the turbines of each cluster based on the respective assigned nominal local datasets, the nominal local model being built such that a nonconformity index is providable which indicates a degree of nonconformity between data projected on the local model and the model itself.

First claim

What is claimed is:

1. A method for monitoring turbines of a windmill farm, the method comprising:
providing a global nominal dataset containing frame data of the turbines of the windmill farm and continuous reference monitoring data of the turbines for a first period in a fault free state, the reference monitoring data comprising at least two same monitoring variables for each turbine;
building a nominal global model based on the global nominal dataset and relationships between the windmill turbines present in the windmill farm by statistical techniques during a fault-free time period, and clustering the turbines according thereto;
assigning the data of the global nominal dataset to respective nominal local datasets according to the clustering;
building a nominal local model for the turbines of each cluster based on the respective nominal local datasets and based on multivariate statistical algorithms or artificial intelligence techniques, wherein the nominal local model is built in that way, that a nonconformity index (NC) is providable which is indicating a degree of nonconformity between data and the model;
providing a test dataset with continuous test monitoring data of the turbines of the windmill farm for a further period, wherein those continuous test monitoring data are structured in the same way than the continuous reference monitoring data in the nominal global dataset and wherein the clustering of the nominal global dataset is also applied on the test dataset;
cluster wise projection of continuous test monitoring data of the test dataset on the respective assigned nominal local models of the turbines and deriving a nonconformity index (NC) for each respective turbine therefrom; and
indicating a first turbine as critical in case that the respective related nonconformity index exceeds a given limit.

2. The method for monitoring turbines of a windmill farm of claim 1 , wherein the nominal local model is based on Principal Component Analysis, Linear Discriminant Analysis, Support Vector Machines, or artificial intelligence techniques.

3. The method for monitoring turbines of a windmill farm of claim 1 , further comprising refining the nominal local model during one or more iterations, each iteration comprising:
identifying one or more turbines of the cluster of turbines as outliers; and
rebuilding the nominal local model without data collected from the one or more turbines of the cluster of turbines identified as outliers.

4. The method for monitoring turbines of a windmill farm of claim 3 , wherein data collected from the one or more turbines of the cluster of turbines identified as outliers is removed from the subset of the global nominal dataset during each iteration.

5. The method for monitoring turbines of a windmill farm of claim 1 , wherein the global nominal dataset comprises data for each turbine of the turbines of the windmill farm comprising electrical measurements, temperature measurements, motional measurements, or ambient condition measurements.

6. The method for monitoring turbines of a windmill farm of claim 1 , wherein the global nominal dataset comprises data about a type of each turbine of the turbines of the windmill farm or a spatial proximity of each turbine of the turbines of the windmill farm to each other turbine of the turbines of the windmill farm.

7. The method for monitoring turbines of a windmill farm of claim 1 , wherein each of the global nominal dataset and the test dataset comprises a three mode dataset comprising several process variables (index J) of several turbines (index I) along several time samples (index K).

8. The method for monitoring turbines of a windmill farm of claim 1 , wherein the global nominal dataset or the test dataset are collected at least predominantly by a SCADA system.

9. The method for monitoring turbines of a windmill farm of claim 1 , further comprising a computing device with a respective software program module running on the computing device configured to automatically perform the method.

10. The method for monitoring turbines of a windmill farm of claim 1 , wherein an automatic fault analysis is initiated upon identifying the first turbine as critical.

11. The method of claim 10 , wherein the automatic fault analysis comprises:
turbine level parsing;
time level parsing; and
variable level parsing.

12. The method for monitoring turbines of a windmill farm of claim 2 , wherein the artificial intelligence techniques comprise a neural network.

13. The method for monitoring turbines of a windmill farm of claim 5 ,
wherein the electrical measurements comprise generated electrical power measurements, voltage measurements, current measurements, or power factor measurements,
wherein the temperature measurements comprise nacelle temperature measurements or electrical generator temperature measurements,
wherein the motional measurements comprise blade speed measurements, or electrical generator speed measurements, or
wherein the ambient condition measurements comprise wind direction measurements, wind speed measurements, or ambient temperature measurements.

14. A method for monitoring turbines of a windmill farm, the method comprising:
collecting a global nominal dataset from a plurality of turbines of a windmill farm during a reference period that is determined to be fault free;
building a nominal global model based on the global nominal dataset and a plurality of relationships between the plurality of turbines present in the windmill farm by statistical techniques during a fault-free time period;
identifying a cluster of turbines from the plurality of turbines based on the nominal global model;
building a nominal local model for the cluster of turbines based on a subset of the global nominal dataset comprising data collected from the cluster of turbines during the reference period;
iteratively refining the nominal local model, each iteration comprising:
identifying an outlier turbine of the cluster of turbines as being an outlier based on results obtained from the nominal local model;
removing the outlier turbine from the cluster of turbines;
removing data collected from the outlier turbine from the subset of the global nominal dataset; and
rebuilding the nominal local model based on the subset of the global nominal dataset after removing data collected from the outlier turbine from the subset of the global nominal dataset;
collecting a test dataset from the cluster of turbines during an operational period of the windmill farm;
deriving a nonconformity index for each turbine of the cluster of turbines that measures conformity between the test dataset and the nominal local model; and
identifying a first turbine of the cluster of turbines as critical when the nonconformity index for the turbine of the cluster of turbines exceeds a given limit.

15. The method for monitoring turbines of a windmill farm of claim 14 , wherein each of the global nominal dataset and the test dataset comprises a three mode dataset comprising several process variables (index J) of several turbines (index I) along several time samples (index K).

16. The method for monitoring turbines of a windmill farm of claim 14 , wherein the global nominal dataset or the test dataset are collected at least predominantly by a SCADA system.

17. The method for monitoring turbines of a windmill farm of claim 14 , further comprising continuing to refine the local nominal model until no outlier turbines can be identified in the cluster of turbines.

18. A method for determining a cluster of turbines of a windmi

Assignees

• Hitachi Energy Switzerland Ag

Inventors

• Chioua Moncef
• Chen Ni Ya
• Yu Rongrong
• Zhou Yingya
• Chen Yao

Classifications

• F03D17/00 Primary

  Monitoring or testing of wind motors, e.g. diagnostics (testing during commissioning of wind motors F03D13/30) · CPC title

• F05B2240/96

  as part of a wind turbine farm · CPC title

• Y02B10/30

  Wind power · CPC title

• G06F11/30

  Monitoring · CPC title