Remaining life prediction for individual components from sparse data

• US 8,768,657 B2
• Filed: 01/12/2007
• Issued: 07/01/2014
• Est. Priority Date: 01/12/2006
• Status: Active Grant

• Current Assignee: Jentek Sensors, Inc.
• Original Assignee: Jentek Sensors, Inc.
• Inventors: Goldfine, Neil J., Zilberstein, Vladimir A., Weiss, Volker, Sheiretov, Yanko K.
• Primary Examiner(s): Rivas, Omar Fernandez
• Assistant Examiner(s): CALLE, ANGEL J.
• Application Number:
  US11/653,067
• Publication Number:
  US 20070239407A1
• Time in Patent Office: 2,727 Days
• Field of Search: 703/2, 707/101, 707/102, 706/21, 702/183
• US Class Current: 703/2
• CPC Class Codes: F05B 2270/109 to prolong engine life
• CPC Class Codes: F05B 2270/404 active, predictive, or anticipative
• CPC Class Codes: G01B 7/34 for measuring roughness or irregularity of surfaces
• CPC Class Codes: G01N 27/82 for investigating the presence of flaws
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Predicting the remaining life of individual aircraft, fleets of aircraft, aircraft components and subpopulations of these components. This is accomplished through the use of precomputed databases of response that are generated from a model for the nonlinear system behavior prior to the time that decisions need to be made concerning the disposition of the system. The database is calibrated with a few data points, to account for unmodeled system variables, and then used with an input variable to predict future system behavior. These methods also permit identification of the root causes for observed system behavior. The use of the response databases also permits rapid estimations of uncertainty estimates for the system behavior, such as remaining life estimates, particularly, when subsets of an input variable distribution are passed through the database and scaled appropriately to construct the output distribution. A specific example is the prediction of remaining life for an aircraft component where the model calculates damage evolution, input variables are a crack size and the number of cycles, and the predicted parameters are the actual stress on the component and the remaining life.

Claims

1. A method for rapid decision making for a nonlinear system, the method comprising:

providing a database of system responses, the database relating a plurality of input variables to at least one output variable predicting a future state of a component, at least one of the plurality of input variables being a current component damage level, each output variable having a nonlinear dependence on the plurality of input variables;

calibrating the database'"'"'s relations between the plurality of input variables to the at least one output variable predicting the future state with first empirical information about the nonlinear system;

obtaining a sensor response from a sensor;

estimating a current component damage level probability distribution from the sensor response and second empirical data representing a dependence of the sensor response on the damage level;

inputting the current component damage level probability distribution to a multivariate inverse method;

estimating a future state probability distribution using the database and the multivariate inverse method; and

using the future state probability distribution to make a decision.

2. The method as claimed in claim 1 wherein the database is precomputed, using a model for the nonlinear system'"'"'s behavior, prior to using the database in estimating the future state probability distribution.

3. The method as claimed in claim 1 wherein the inputting comprises inputting the plurality of input variables, including the current component damage level probability distribution, to the multivariate inverse method, and the method further includes dividing the plurality of the measured input variables into portions, determining a weighting value for each portion, and further calculating the future state probability distribution by passing the measured input variables and weighted value of each portion through the database of system responses.

4. The method as claimed in claim 3 wherein calculating the future state probability distribution involves passing a fraction of the plurality of input variables into the database.

5. The method as claimed in claim 4 further comprising using a Latin hypercube methodology to determine the future state probability distribution, by passing a substantially reduced number of inputs through the database to construct the future state probability distribution.

6. The method as claimed in claim 1 wherein the sensor response is obtained using an eddy current sensor.

7. The method as claimed in claim 1 wherein current component damage level probability distribution is crack size.

8. The method as claimed in claim 1 wherein at least one input variable is number of cycles.

9. The method as claimed in claim 1 wherein the future state probability distribution is a distribution of crack sizes at a prescribed number of cycles.

10. The method as claimed in claim 1 wherein the future state probability distribution is a number of cycles required to reach a specific crack size.

11. The method as claimed in claim 1 wherein estimating the current component damage level probability distribution from the sensor response comprises subtracting a baseline response from the sensor response.

18. The method as claimed in claim 1 wherein the sensor response is an image of effective material properties.

12. A method for rapid decision making for a nonlinear system, the method comprising:

generating a database of system responses from a damage evolution model, the database relating a plurality of input variables to at least one output variable predicting a future state of a component, at least one of the plurality of input variables being a current component damage level, each output variable having a nonlinear dependence on the plurality of input variables;

storing the database for future use;

calibrating the database'"'"'s relations between the plurality of input variables to the at least one output variable predicting the future state with first empirical information about the nonlinear system;

obtaining a sensor response from a sensor;

estimating a current component damage level probability distribution from the sensor response and second empirical data representing a dependence of the sensor response on the damage level;

inputting the current component damage level probability distribution to a multivariate inverse method;

estimating a future state probability distribution using the database and the multivariate inverse method; and

using the future state probability distribution to make a decision.

13. The method as claimed in claim 12 wherein the first empirical information is sensor based.

14. The method as claimed in claim 13 wherein the sensor is an eddy current sensor.

15. The method as claimed in claim 12 wherein a shape of the future state probability distribution is assumed.

16. The method as claimed in claim 12 wherein the future state probability distribution is determined by measurement.

17. The method as claimed in claim 12 wherein the component is a shot peened metal.

19. The method as claimed in claim 12 wherein the component damage level represents a size of a crack in the component.

20. The method as claimed in claim 12 wherein the plurality of input variables comprises a number of cycles the component has been in use.

21. The method as claimed in claim 12 wherein the future state probability distribution is remaining life.

22. The method as claimed in claim 12 wherein the future state probability distribution is a distribution of cycles in which a prescribed crack size is reached and further including determining a decision regarding fitness for service as a function of the distribution of cycles.

26. The method of claim 12, wherein the second empirical data defines a probability of detection for the sensor.

27. The method of claim 12, wherein calibrating the database comprises performing one or more coupon tests to estimate parameters of the damage evolution model used to generate the database.

23. A method for rapid decision making for a nonlinear system, the method comprising:

providing a database of system responses generated from a model, the database relating a plurality of input variables to at least one output variable predicting a future state of a component, at least one of the plurality of input variables being a measure of temporal usage, at least one of the plurality of input variables being a current component damage level, each output variable having a nonlinear dependence on the plurality of input variables;

calibrating the database with first empirical information about the nonlinear system;

obtaining a sensor response from a sensor;

estimating a current component damage level probability distribution from the sensor response and second empirical information representing a dependence of the sensor response on the current component damage level;

receiving the temporal usage;

inputting the temporal usage and the current component damage level probability distribution to a multivariate inverse method;

estimating a future state probability distribution using the database and the multivariate inverse method; and

using the future state probability distribution to make a decision.

24. The method of claim 23, wherein the temporal usage is a number of cycles a component has been used.

25. The method of claim 23, wherein the temporal usage is a number of hours a component has been used in flight.