Systems and methods for estimating future risk of failure of a wind turbine component using machine learning
A method for estimating future risk of failure of a component of an industrial asset. The method includes receiving a plurality of different types of data associated with the industrial asset or a fleet of industrial assets. The plurality of different types of data includes, at least, reliability data (such as time-to-event data). The method also includes generating a failure prediction model for the component based on the reliability data and available time-series measurements. Further, the method includes applying the failure prediction model to the different types of data based on the types of data available in the received data. The applied failure prediction model includes one of a default model, a conditional survival model, or a joint conditional survival model. Thus, the method includes estimating, via the failure prediction model, the future risk of failure of the industrial asset and implementing a control action as needed.
1. A computer-implemented method for estimating future risk of failure of a component of an industrial asset, the method comprising:
receiving, via a controller, a plurality of different types of data associated with the industrial asset or a fleet of industrial assets, the plurality of different types of data comprising, at least, reliability data, the reliability data comprising, at least, time-to-event data from the industrial asset or the fleet of industrial assets;
determining a number of failures or a number of suspensions present in the time-to-event data;
comparing the number of failures or the number of suspensions to a threshold;
generating, via the controller, a failure prediction model for the component based on the reliability data and available time-series measurements;
applying, via the controller, the failure prediction model to the plurality of different types of data associated with the industrial asset or the fleet of industrial assets based on types of data available in the received plurality of different types of data, the applied failure prediction model comprising one of a default model, a conditional survival model, or a joint conditional survival model;
using the default model as the failure prediction model when the number of failures or the number of suspensions is below the threshold;
estimating, via the failure prediction model, the future risk of failure of the industrial asset; and
implementing, via the controller, a control action based on the future risk of failure, wherein implementing the control action comprises at least one of shutting down the industrial asset or shutting down the component of the industrial asset.
2. The computer-implemented method of claim 1 , wherein the time-to-event data comprises the number of failures, the number of suspensions, and at least one of a number of historical failures, a number of historical suspensions, time/cycles to failure, time/cycles to wear, or combinations thereof.
3. The computer-implemented method of claim 2 , further comprising:
performing a quality check on at least one of the reliability data or the time-series measurements to evaluate for errors; and
correcting the errors found during the quality check.
4. The computer-implemented method of claim 1 , wherein the default model comprises at least one of one or more physics-based models or one or more WeiBayes-type models.
5. The computer-implemented method of claim 1 , wherein generating the failure prediction model for the component based on the reliability data and the available time-series measurements further comprises:
when the number of historical failures or the number of suspensions is above the threshold, determining whether the time-series measurements associated with the industrial asset are available, and if so,
building the failure prediction model using the reliability data and the time-series measurements associated with the industrial asset.
6. The computer-implemented method of claim 5 , wherein generating the failure prediction model for the component based on the reliability data and the available time-series measurements further comprises:
when the number of historical failures or the number of suspensions is above the threshold and the time-series measurements associated with the industrial asset is available, generating a plurality of models with different association structures;
selecting one of the plurality of models based on an accuracy thereof as a preferred model, the preferred model being the failure prediction model, which is the joint conditional survival model.
7. The computer-implemented method of claim 6 , wherein the joint conditional survival model comprises a time-series model that characterizes a time evolution of the time-series measurements, a survival model that uses the reliability data, and an association structure comprising parameters to join the time-series model and the survival model together based on the reliability data.
8. The computer-implemented method of claim 7 , wherein the reliability data further comprises static data, the static data comprising at least one of a start date of the industrial asset, a current status of the industrial asset, mitigation actions implemented to extend a life of the component of industrial asset, genealogy information relating to the industrial asset, inspection data associated with wear of the industrial asset, a current status of the industrial asset comprising at least one of a failure status, an operational status, a wear status, or combinations thereof.
9. The computer-implemented method of claim 8 , wherein generating the failure prediction model for the component based on the reliability data and the available time-series measurements further comprises:
when the number of historical failures or the number of suspensions is above the threshold, determining whether the time-series measurements associated with the industrial asset is available, and if not,
building the conditional survival model using the reliability data, the conditional survival model being the failure prediction model.
10. The computer-implemented method of claim 9 , wherein estimating the future risk of failure of the industrial asset further comprises:
estimating the future risk of failure of the industrial asset using the conditional survival model.
11. The computer-implemented method of claim 6 , wherein generating the failure prediction model for the component based on the reliability data and the available time-series measurements further comprises:
when the number of historical failures or the number of suspensions is above the threshold and the time-series measurements associated with the industrial asset is available, determining which of the time-series measurements associated with the industrial asset are relevant to a failure mode for each of the industrial assets in the fleet by correlating the time-series measurements with the time-to-event data; and
generating the joint conditional survival model based on the failure mode for each of the industrial assets in the fleet.
12. The computer-implemented method of claim 11 , wherein estimating the future risk of failure of the industrial asset further comprises:
estimating the future risk of failure of the industrial asset using the joint conditional survival model.
13. The computer-implemented method of claim 1 , wherein the time-series measurements comprises data from one or more sensors of the industrial asset.
14. The computer-implemented method of claim 1 , further comprising training the failure prediction model with additional failure data after generating the failure prediction model for the component.
15. The computer-implemented method of claim 1 , further comprising developing one or more competing risk models for evaluating the future risk of failure of the industrial asset.
16. The computer-implemented method of claim 1 , wherein implementing the control action further comprises at least one of scheduling a preventative maintenance action, scheduling a repair action, performing asset management actions, generating an alarm, or performing a sensitivity analysis.
17. The computer-implemented method of claim 1 , wherein the industrial asset comprises a wind turbine.
18. A system for estimating future risk of failure of a component of a wind turbine, the system comprising:
a controller comprising at least one processor, the at least one processor configured to perform a plurality of operations, the plurality of operations comprising:
receiving a plurality of different types of data associated with the wind turbine or a fleet of wind turbines, the plurality of different types of data comprising, at least, reliability data, the reliability data comprising, at least, time-to-event data from the wind turbine or the fleet of wind turbines;
determining a number of failures or a number of suspensions present in the time-to-event data;
comparing the number of failures or the number of suspensions to a threshold;
generating a failure prediction model for the component based on the reliability data and available time-series measurements;
applying the failure prediction model to the plurality of different types of data associated with the wind turbine or the fleet of wind turbines based on types of data available in the received plurality of different types of data, the applied failure prediction model comprising one of a conditional survival model or a joint conditional survival model;
using the default model as the failure prediction model when the number of failures or the number of suspensions is below the threshold;
estimating, via the failure prediction model, the future risk of failure of the wind turbine; and
implementing a control action based on the future risk of failure, wherein implementing the control action comprises at least one of shutting down the industrial asset or shutting down the component of the industrial asset.