Coupling time evolution model with empirical regression model to estimate mechanical wear

• US 8,600,917 B1
• Filed: 04/18/2011
• Issued: 12/03/2013
• Est. Priority Date: 04/18/2011
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Schimert, James, Wineland, Arthur Ray
• Primary Examiner(s): Starks, Wilbert L
• Application Number:
  US13/088,683
• Time in Patent Office: 960 Days
• Field of Search: 706/14, 706/45
• US Class Current: 706/14
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

Mechanical systems wear or change over time. Data collected over a system'"'"'s life can be input to statistical learning models to predict this wear/change. Previous work by the inventors trained a flexible empirical regression model at a fixed point of wear, and then applied it independently at time points over the life of an engine to predict wear. The embodiment disclosed herein relates those wear predictions over time using a time evolution model. The time evolution model is sequentially updated with new data, and effectively tunes the empirical model for each engine. The combined model predicts wear with dramatically reduced variability. The benefit of reduced variability is that engine wear is more evident, and it is possible to detect operational anomalies more quickly. In addition to tracking wear, the model is also used as the basis for a Bayesian approach to monitor for sudden changes and reject outliers, and adapt the model after these events.

Claims

1. A method for monitoring wear in a mechanical system, comprising:

(a) repeatedly measuring a parameter over time during operation of the mechanical system;

(b) calculating residuals at time points over the life of the mechanical system using an empirical model that models values of said parameter as a function of values of other parameters, said residuals representing the respective differences between each measurement of said parameter and each corresponding parameter value predicted by the empirical model;

(c) determining whether the measurements evolve as expected under a time evolution model that relates predictions of residuals over time; and

(d) flagging an event in response to the measurements of said parameter deviating over time from the behavior predicted by the time evolution model by more than a threshold value,

wherein in accordance with said time evolution model, wear at time t equals the wear at a previous time (t−

1) plus a local growth rate at time (t−

1).

2. The method as recited in claim 1, further comprising isolating and rejecting outliers and adapting the time evolution model to reflect the onset of a structural change in response to the measurements deviating over time from the behavior predicted by the time evolution model by more than said threshold value.

3. The method as recited in claim 1, wherein said adapting comprises increasing the uncertainty in a prior distribution as measured by a prior covariance matrix.

4. The method as recited in claim 1, further comprising updating the time evolution model in response to the measurements deviating over time from the behavior predicted by the time evolution model by less than said threshold value.

5. The method as recited in claim 4, wherein the local growth rate is a random walk.

6. The method as recited in claim 1, wherein said time evolution model is a second-order polynomial dynamic linear model.

7. The method as recited in claim 1, wherein step (c) comprises comparing the respective predictive abilities of standard and alternative time evolution models, wherein said alternative model is similar in form to said standard model, but allows for more extreme wear observations.

8. The method as recited in claim 1, further comprising isolating a fault in said mechanical system when said event is flagged.

9. The method as recited in claim 1, wherein said mechanical system is a gas turbine engine and said parameter is engine exhaust gas temperature.

10. A method for monitoring wear in a mechanical system, comprising:

(a) repeatedly measuring a parameter over time during operation of the mechanical system;

(b) calculating the value of a monitoring statistic for each of said measurements;

(c) calculating the value of a cumulative monitoring statistic that is a product of sequential values of said monitoring statistic;

(d) determining whether the value of said monitoring statistic is less than or not less than a threshold value;

(e) determining whether the value of said cumulative monitoring statistic is less than or not less than said threshold value; and

(f) flagging an event in response to the values of said monitoring statistic and said cumulative monitoring statistic being less than said threshold value,

wherein said mechanical system is a gas turbine engine and said parameter is engine exhaust gas temperature.

11. The method as recited in claim 10, wherein said event is an abrupt structural change in said mechanical system.

12. The method as recited in claim 10, further comprising isolating and rejecting outliers included in said measurements and increasing the uncertainty in a prior distribution as measured by a prior covariance matrix to reflect the onset of a structural change in response to the values of said monitoring statistic and said cumulative monitoring statistic being less than said threshold value.

13. The method as recited in claim 10, further comprising updating a time evolution model in response to the values of said monitoring statistic and said cumulative monitoring statistic being not less than said threshold value.

14. The method as recited in claim 10, further comprising isolating a fault in said mechanical system when said event is flagged.

15. A system for monitoring the health of a mechanical system, comprising a computer system programmed to perform the following operations:

(a) receiving values representing the results of measurements of a parameter over time during operation of the mechanical system;

(b) predicting residuals at time points over the life of the mechanical system using an empirical model that models values of said parameter as a function of values of other parameters, said residuals representing the respective differences between each measurement of said parameter and each corresponding parameter value predicted by the empirical model;

(c) determining whether the measurements evolve as expected under a time evolution model that relates predictions of residuals over time; and

(d) flagging an event in response to the measurements of said parameter deviating over time from the behavior predicted by the time evolution model by more than a threshold value,

wherein said mechanical system is a gas turbine engine and said parameter is engine exhaust gas temperature.

16. The system as recited in claim 15, wherein said computer system is further programmed to perform the following operations:

isolating and rejecting outliers and adapting the time evolution model to reflect the onset of a structural change in response to the measurements deviating over time from the behavior predicted by the time evolution model by more than said threshold value.

17. The system as recited in claim 16, wherein said adapting comprises increasing the uncertainty in a prior distribution as measured by a prior covariance matrix.

18. The system as recited in claim 15, wherein said computer system is further programmed to update the time evolution model in response to the measurements deviating over time from the behavior predicted by the time evolution model by less than said threshold value.

19. The system as recited in claim 15, wherein step (c) comprises comparing the respective predictive abilities of standard and alternative time evolution models, wherein said alternative model is similar in form to said standard model, but allows for more extreme changes in the mechanical system or component.

20. A system for monitoring the health of a mechanical system, comprising a computer system programmed to perform the following operations:

(a) receiving values representing the results of measurements of a parameter over time during operation of the mechanical system;

(b) calculating the value of a monitoring statistic for each of said measurements;

(c) calculating the value of a cumulative monitoring statistic that is a product of a plurality of values of said monitoring statistic;

(d) determining whether the value of said monitoring statistic is less than or not less than a threshold value;

(e) determining whether the value of said cumulative monitoring statistic is less than or not less than said threshold value; and

(f) flagging an event in response to the values of said monitoring statistic and said cumulative monitoring statistic being less than said threshold value,

wherein said mechanical system is a gas turbine engine and said parameter is engine exhaust gas temperature.