Empirical design of experiments using neural network models

• US 7,451,122 B2
• Filed: 03/29/2006
• Issued: 11/11/2008
• Est. Priority Date: 03/29/2006
• Status: Expired due to Fees

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Mylaraswamy, Dinkar, Menon, Sunil K., Dietrich, Paul F., Olson, Lewis P.
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/394,317
• Publication Number:
  US 20070239633A1
• Time in Patent Office: 958 Days
• Field of Search: None
• US Class Current: 706/15
• CPC Class Codes: G06N 3/02 Neural networks

Abstract

Methods and apparatus are provided pertaining to a design of experiments. The method comprises generating a data set from historical data; identifying and removing any fault data points in the data set so as to create a revised data set; supplying the data points from the revised data set into a nonlinear neural network model; and deriving a simulator model characterizing a relationship between the input variables and the output variables. The apparatus comprises means for generating a data set from historical data; means for identifying and removing any fault data points in the data set so as to create a revised data set; means for supplying the data points from the revised data set into a nonlinear neural network model; and means for deriving a simulator model characterizing a relationship between the input variables and the output variables.

Claims

1. A method for a design of experiments for modeling the effects of two or more input variables on one or more output variables, the method comprising the steps of:

(a) generating a data set comprising data points from historical data for the input variables and the output variables, each data point comprising corresponding values for one or more of the input variables and one or more of the output variables from the historical data;

(b) identifying fault data points in the historical data, a fault data point being a data point from the data set from the historical data in which an output variable value is determined to be caused by factors other than the input variables;

(c) removing the identified fault data points from the data set, thereby generating a revised data set with no fault data points, a no fault data point being a data point from the data set from the historical data that is not determined to be a fault data point;

(d) supplying the no fault data points from the revised data set into a nonlinear neural network model; and

(e) deriving a simulator model characterizing a relationship between the input variables and the output variables using the nonlinear neural network model with the supplied data.

2. The method of claim 1, wherein the output variables comprise one or more performance indicators for a turbine engine, and the input variables comprise factors affecting performance of the turbine engine.

3. The method of claim 1, further comprising the steps of:

(f) selecting a plurality of statistical measures to characterize the no fault data points with respect to each of the input variables in the revised data set;

(g) determining statistical measure values for each statistical measure for each of the input variables, based on the no fault data points in the revised data set;

(h) supplying the statistical measure values into the derived simulator model; and

(i) determining calculated values from the derived simulator model of one or more of the output variables corresponding to the respective statistical measure values of the input variables.

4. The method of claim 3, wherein the statistical measures are selected from the group consisting of the minimum, lower quartile, median, mean, mode, upper quartile, and maximum values from the revised data set for each variable.

5. The method of claim 3, wherein the statistical measures comprise the minimum, median and maximum values from the data set for each input variable.

6. The method of claim 3, wherein the determined statistical measure values and calculated values are used to test the derived simulator model.

7. The method of claim 3, further comprising the steps of:

(j) generating an original algorithm for modeling the output variables as a function of one or more of the input variables;

(k) selecting dominant input variables, each of the dominant input variables being the input variables having dominant effects on the output variables; and

(l) generating an enhanced algorithm for modeling the output variables as a function of one or more of the dominant input variables.

8. The method of claim 7, wherein the dominant input variables are selected based on statistical analysis of variance (ANOVA) calculations.

9. The method of claim 7, further comprising the steps of:

selecting additional input variables, for inclusion in the enhanced algorithm, each of the additional input variables being an input variable that was not in the original algorithm but was selected as a dominant input variable in step (k); and

removing input variables, for exclusion from the enhanced algorithm, that were in the original algorithm but were not selected as dominant inputs variables in step (k).

10. A method for generating an enhanced algorithm for representing the effects of two or more input variables on an output variable, comprising:

(a) generating an original algorithm for modeling the output variables as a function of one or more of the input variables;

(b) generating a data set comprising data points from historical data for the input variables and the output variables, each data point comprising corresponding values for one or more of the input variables and one or more of the output variables;

(c) identifying fault data points in the historical data, a fault data point being a data point in which an output variable value is determined to be caused by factors other than the input variables;

(d) removing the identified fault data points from the data set, thereby generating a revised data set with data points;

(e) supplying the data points from the revised data set into a nonlinear neural network model;

(f) deriving a simulator model characterizing a relationship between the input variables and the output variables using the nonlinear neural network model with the supplied data;

(g) selecting a plurality of statistical measures to characterize the data points with respect to each of the input variables in the revised data set;

(h) determining statistical measure values for each statistical measure for each of the input variables, based on the data points in the revised data set;

(i) supplying the statistical measure values into the derived simulator model;

(j) determining calculated values from the derived simulator model of one or more of the output variables corresponding to the respective statistical measure values of the input variables;

(k) selecting dominant input variables, each of the dominant input variables being the input variables having dominant effects on the output variables; and

(l) generating an enhanced algorithm for modeling the output variables as a function of one or more of the dominant input variables.

11. The method of claim 10, wherein the output variables comprise one or more performance indicators for a turbine engine, and the input variables comprise factors affecting performance of the turbine engine.

12. The method of claim 10, wherein the statistical measures are selected from the group consisting of the minimum, lower quartile, median, mean, mode, upper quartile, and maximum values from the revised data set for each variable.

13. The method of claim 10, wherein the statistical measures comprise the minimum, median and maximum values from the revised data set for each variable.

14. The method of claim 10, wherein the determined statistical measure values and calculated values are used to test the derived simulator model.

15. The method of claim 10, wherein the dominant input variables are selected based on statistical analysis of variance (ANOVA) calculations.

16. The method of claim 10, further comprising the steps of:

selecting additional input variables, for inclusion in the enhanced algorithm, each of the additional input variables being an input variable that was not in the original algorithm but was selected as a dominant input variable in step (k); and

removing input variables, for exclusion from the enhanced algorithm, that were in the original algorithm but were not selected as dominant inputs variables in step (k).

17. The method of claim 10, further comprising:

deriving a linear regression model characterizing the relationship between the input variables and the output variables utilizing the enhanced algorithm;

analyzing results of the derived linear regression model;

analyzing results of the derived simulator model using the enhanced algorithm; and

comparing the results of the derived linear regression model with the results of the derived simulator model.

18. A system for modeling the effects of two or more input variables on an output variable comprising:

(a) means for generating a data set comprising data points from historical data for the input variables and the output variables, each data point comprising corresponding values for one or more of the input variables and one or more of the output variables;

(b) means for identifying fault data points in the historical data, a fault data point being a data point in which an output variable value is determined to be caused by factors other than the input variables;

(c) means for removing the identified fault data points from the data set, thereby generating a revised data set with data points;

(d) means for supplying the data points from the revised data set into a nonlinear neural network model; and

(e) means for deriving a simulator model characterizing a relationship between the input variables and the output variables using the nonlinear neural network model with the supplied data.

19. The system of claim 18, wherein the output variables comprise one or more performance indicators for a turbine engine, and the input variables comprise factors affecting performance of the turbine engine.

20. The system of claim 18, wherein the system further comprises:

(f) means for selecting a plurality of statistical measures to characterize the data points with respect to each of the input variables in the revised data set;

(g) means for determining statistical measure values for each statistical measure for each of the input variables, based on the data points in the revised data set;

(h) means for supplying the statistical measure values into the derived simulator model; and

(i) means for determining calculated values from the derived simulator model of one or more of the output variables corresponding to the respective statistical measure values of the input variables.

21. The system of claim 18, further comprising:

means for implementing the derived simulator model.

22. A program product comprising:

(a) a program configured to at least facilitate;

(i) generating a data set comprising data points from historical data for the input variables and the output variables, each data point comprising corresponding values for one or more of the input variables and one or more of the output variables from the historical data;

(ii) identifying fault data points in the historical data, a fault data point being a data point from the data set from the historical data in which an output variable value is determined to be caused by factors other than the input variables;

(iii) removing the identified fault data points from the data set, thereby generating a revised data set with no fault data points, a no fault data point being a data point from the data set from the historical data that is not determined to be a fault data point;

(iv) supplying the no fault data points from the revised data set into a nonlinear neural network model; and

(v) deriving a simulator model characterizing a relationship between the input variables and the output variables using the nonlinear neural network model with the supplied data; and

(b) a computer-readable signal bearing media bearing the program.

23. An apparatus comprising:

(a) a processor;

(b) a memory coupled to the processor; and

(c) a program residing in memory and being executed by the processor, the program configured to characterize a relationship between one or more input variables and one or more output variables through at least the following steps;

(i) generating a data set comprising data points from historical data for the input variables and the output variables, each data point comprising corresponding values for one or more of the input variables and one or more of the output variables from the historical data;

(ii) identifying fault data points in the historical data, a fault data point being a data point from the data set from the historical data in which an output variable value is determined to be caused by factors other than the input variables;

(iii) removing the identified fault data points from the data set, thereby generating a revised data set with no fault data points, a no fault data point being a data point from the data set from the historical data that is not determined to be a fault data point;

(iv) supplying the no fault data points from the revised data set into a nonlinear neural network model; and

(v) deriving a simulator model characterizing a relationship between the input variables and the output variables using the nonlinear neural network model with the supplied data.

24. The apparatus of claim 23, wherein the processor is operable to implement the derived simulator model.