Data driven method and system for predicting operational states of mechanical systems

• US 8,165,826 B2
• Filed: 09/30/2008
• Issued: 04/24/2012
• Est. Priority Date: 09/30/2008
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Schimert, James, Basu, Sabyasachi, Wineland, Art, Farner, Scott
• Primary Examiner(s): Moffat, Jonathan C Teixeira
• Application Number:
  US12/241,643
• Publication Number:
  US 20100082267A1
• Time in Patent Office: 1,302 Days
• Field of Search: 702/34, 702/184
• US Class Current: 702/34
• CPC Class Codes: G05B 19/4065 Monitoring tool breakage, life or condition
• CPC Class Codes: G05B 2219/37252 Life of tool, service life, decay, wear estimation
• CPC Class Codes: G07C 3/00 Registering or indicating the condition or the working of machines or other apparatus, other than vehicles

Abstract

There is provided an automated data driven method for predicting one or more operational states, such as wear or degradation, of a mechanical system over time. The method has the steps of collecting data on the mechanical system from a data recording device, preprocessing the collected data, selecting a training data set that represents a base condition for statistical comparison, fitting a statistical model to the training data set to relate a predicted response to nuisance variables at the base condition, and using an output model to predict what an observed response would have been at the base condition and calculating the difference between the observed response and the predicted response to predict the one or more operational states of the mechanical system.

Claims

1. An automated data driven method for predicting one or more operational states of a mechanical system over time, the method comprising the steps of:

collecting data on the mechanical system from a data recording device, the collected data comprising technical parameters of the mechanical system, environmental data, and operational data;

preprocessing the collected data to determine when during operation of the mechanical system the data is collected, what segment of the mechanical system is to be predicted, and how to reduce the amount of data to produce snapshots of data representative of the health of the mechanical system;

selecting a training data set from the snapshots of the collected data that represents a base condition for statistical comparison, wherein the training data set is created when a mechanical system output is stable yet there is variability in nuisance variables to represent operational conditions of the mechanical system;

fitting a nonparametric statistical regression model to the training data set using a computer to relate a response to the nuisance variables at the base condition, wherein a relationship between the response and the nuisance variables is not specified by model parameters; and

,

using the computer to apply the fitted nonparametric statistical regression model to a single set of response and nuisance variables observed at one time during operation of the mechanical system to generate a predicted response representing what the response would have been at the base condition and calculating the difference between the response and the predicted response to predict the one or more operational states of the mechanical system.

2. The method of claim 1 wherein the collecting data step further comprises the step of determining one or more technical parameters of the mechanical system to be measured relevant to the predicting of the one or more operational states.

3. The method of claim 1 wherein the method further comprises the step of using the predicted one or more operational states for trend analysis.

4. The method of claim 1 wherein the nonparametric statistical regression model is a random forest technique.

5. The method of claim 1 wherein the mechanical system is selected from the group comprising engines, turbines, tires, and brakes.

6. The method of claim 1 wherein the mechanical system is in an apparatus selected from the group comprising aircraft, automobiles, trucks, watercraft, power generator units, and military vehicles.

7. The method of claim 1 wherein the operational state is selected from the group comprising wear, degradation, part failure, and anomalies.

8. The method of claim 1 wherein the method uses the nonparametric statistical regression model to model aircraft gas turbine engine exhaust gas temperature (EGT) as a function of environmental, flight and engine parameters and to generate the output model as a predictive model of engine wear.

9. The method of claim 1 wherein the nuisance variables are selected from the group comprising environmental influences, flight conditions, system faults, and mechanical parameters.

10. An automated data driven method for predicting wear of a mechanical system over time, the method comprising the steps of:

collecting data on the mechanical system from a data recording device, the collected data comprising technical parameters of the mechanical system, environmental data, and operational data;

selecting from the collected data one or more of the technical parameters of the mechanical system relevant to the predicting of wear of the mechanical system;

preprocessing the collected data to summarize operation of the mechanical system to determine when during operation of the mechanical system the data is collected, what segment of the mechanical system is to be predicted, and how to reduce the amount of data to produce snapshots of data representative of the health of the mechanical system;

selecting a training data set from the snapshots of the collected data that represents a base condition for statistical comparison, wherein the training data set is created when a mechanical system output is stable yet there is variability in nuisance variables to represent operational conditions of the mechanical system;

fitting a nonparametric statistical regression model to the training data set using a computer to relate a response to the nuisance variables at the base condition, wherein a relationship between the response and the nuisance variables is not specified by model parameters;

using the computer to apply the fitted nonparametric statistical regression model to a single set of response and nuisance variables observed at one time during operation of the mechanical system to generate a predicted response representing what the response would have been at the base condition and calculating the difference between the response and the predicted response to predict the one or more operational states of the mechanical system;

plotting the predicted wear; and

,

using the plotted wear for trend analysis.

11. The method of claim 10 wherein the nonparametric statistical regression model is a random forest technique.

12. The method of claim 10 wherein the mechanical system is selected from the group comprising engines, turbines, tires, and brakes.

13. The method of claim 10 wherein the mechanical system is in an apparatus selected from the group comprising aircraft, automobiles, trucks, watercraft, power generator units, and military vehicles.

14. The method of claim 10 wherein the method uses the nonparametric statistical regression model to model aircraft gas turbine engine exhaust gas temperature (EGT) as a function of environmental, flight and engine parameters and uses the output model to predict engine wear.

15. An automated data driven system for predicting one or more operational states of a mechanical system over time comprising:

a data collection component for collecting data on the mechanical system, the collected data comprising technical parameters of the mechanical system, environmental data, and operational data;

a preprocessing component for preprocessing the collected data in order to determine when during operation of the mechanical system the data is collected, what segment of the mechanical system is to be predicted, and how to reduce the amount of data to produce snapshots of data representative of the health of the mechanical system;

a training data set selection component for selecting a training data set from the snapshots of the collected data that represents a base condition for statistical comparison, wherein the training data set is created when a mechanical system output is stable yet there is variability in nuisance variables to represent operational conditions of the mechanical system;

a statistical nonparametric regression modeling component for fitting a statistical nonparametric regression model to the training data set to relate a response to the nuisance variables at the base condition, wherein a relationship between the response and the nuisance variables is not specified by model parameters;

a fitted statistical nonparametric regression model; and

,

a predicting component that applies the fitted nonparametric regression model to a single set of response and nuisance variables observed at one time during operation of the mechanical system to generate a predicted response representing what the response would have been at the base condition and calculating the difference between the response and the predicted response to predict the one or more operational states of the mechanical system.

16. The system of claim 15 wherein the statistical nonparametic regression modeling component is a random forest technique.

17. The system of claim 15 wherein the mechanical system is selected from the group comprising engines, turbines, tires, and brakes.

18. The system of claim 15 wherein the mechanical system is in an apparatus selected from the group comprising aircraft, automobiles, trucks, watercraft, power generator units, and military vehicles.