Diagnostic algorithm parameter optimization

• US 8,768,668 B2
• Filed: 01/09/2012
• Issued: 07/01/2014
• Est. Priority Date: 01/09/2012
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Patankar, Ravindra, Iyer, Shreyas, Felke, Tim
• Primary Examiner(s): Thangavelu, Kandasamy
• Application Number:
  US13/346,333
• Publication Number:
  US 20130179129A1
• Time in Patent Office: 904 Days
• Field of Search: 703/2, 703/6, 703/13, 714/26, 714/25, 702/184, 702/58, 706/2, 701/31.6, 701/213, 700/287, 342/357.58, 705/14.49
• US Class Current: 703/6
• CPC Class Codes: G05B 23/0243 model based detection method, e.g. first-principles knowledge model
• CPC Class Codes: G06F 11/07 Responding to the occurrence of a fault, e.g. fault tolerance
• CPC Class Codes: G06F 11/0739 in a data processing system embedded in automotive or aircraft systems
• CPC Class Codes: G06F 11/0751 Error or fault detection not based on redundancy power supply failures G06F1/30; network fault management H04L41/06
• CPC Class Codes: G06F 11/0766 Error or fault reporting or storing
• CPC Class Codes: G06F 11/079 Root cause analysis, i.e. error or fault diagnosis in a hardware test environment G06F11/22; in a software test environment G06F11/36
• CPC Class Codes: G06F 11/22 Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
• CPC Class Codes: G06F 11/261 by simulating additional hardware, e.g. fault simulation
• CPC Class Codes: G06F 11/321 Display for diagnostics, e.g. diagnostic result display, self-test user interface
• CPC Class Codes: G06F 11/366 using diagnostics G06F11/0703 takes precedence
• CPC Class Codes: G06F 17/10 Complex mathematical operations function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544
• CPC Class Codes: G06F 2207/7271 Fault verification, e.g. comparing two values which should be the same, unless a computational fault occurred
• CPC Class Codes: G06F 30/33 Design verification, e.g. functional simulation or model checking

Abstract

A system and method are provided for optimizing parameters of a plurality of selected diagnostic and/or prognostic algorithms in a tunable diagnostic algorithm library. A plurality of sensed data sets having an actual diagnostic label associated therewith is supplied to each of the diagnostic algorithms. A value for each parameter of each of the algorithms that are to be optimized is supplied. A computed diagnostic label is generated for each of the sensed data sets using each of the selected algorithms, a fault model, and the values for each parameter, each of the computed diagnostic labels and each of the actual diagnostic labels are supplied to a generic objective function, to thereby calculate an objective function value, and the value of one or more of the parameters is varied using an optimization routine that repeats certain of these steps until the objective function value is minimized.

Claims

1. A method for optimizing parameters of a plurality of selected diagnostic algorithms in a tunable diagnostic algorithm library, the method comprising the steps of:

a) supplying a plurality of sensed data sets to each of the selected algorithms implemented in a processor, each of the sensed data sets having an actual diagnostic label associated therewith;

b) supplying, to the processor, a value for each parameter of each of the selected algorithms that is to be optimized; and

in the processor;

c) generating a computed diagnostic label for each of the sensed data sets using each of the selected algorithms, a fault model, and the value for each parameter;

d) supplying each of the computed diagnostic labels and each of the actual diagnostic labels to a generic objective function, to thereby calculate an objective function value; and

e) varying the value of one or more of the parameters using an optimization routine that repeats steps c)-e) until the objective function value is minimized.

2. The method of claim 1, further comprising:

determining, in the processor and based on the fault model, all possible computed diagnostic labels that the selected algorithms will generate;

defining rows and columns of a confusion matrix having a plurality of entries and that includes a separate row for each of the possible computed diagnostic labels and a separate column for each actual diagnostic label; and

assigning a weight value to each entry in the confusion matrix to thereby generate a weighting matrix (W).

3. The method of claim 2, wherein each computed diagnostic label comprises a probability vector corresponding to each of the possible computed diagnostic labels.

4. The method of claim 3, further comprising generating a normalized confusion matrix (S) based on each of the actual diagnostic labels and using each of the probability vectors.

5. The method of claim 4, wherein the step of generating a normalized confusion matrix comprises calculating a normalized probability vector of each computed diagnostic label for each of the actual diagnostic labels.

6. The method of claim 4, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the objective function value is calculated using;

7. The method of claim 4, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the objective function value is calculated using;

8. A method for optimizing parameters of one or more diagnostic algorithms that are configured to supply one or more computed fault diagnostic labels, comprising the steps of:

a) supplying a plurality of sensed data sets to each of the one or more diagnostic algorithms implemented in a processor, each of the sensed data sets having an actual diagnostic label associated therewith;

b) supplying, to the processor, a value for each parameter of each of the one or more diagnostic algorithms that is to be optimized;

c) determining, in the processor and based on a fault model, all possible computed diagnostic labels that each of the one or more diagnostic algorithms will generate;

d) generating, in the processor, a confusion matrix having a plurality of entries and that includes a separate row for each of the possible computed diagnostic labels and a separate column for each actual diagnostic label;

e) assigning, in the processor, a relative weight value to each entry in the confusion matrix to thereby generate a weighting matrix (W); and

in the processor;

f) generating a computed diagnostic label for each of the sensed data sets using each of the one or more diagnostic algorithms and the value for each parameter, each computed diagnostic label comprising a probability vector;

g) supplying each of the computed diagnostic labels, each of the actual diagnostic labels, and the weighting matrix to a generic objective function, to thereby calculate an objective function value; and

h) varying the value of one or more of the parameters using an optimization routine that repeats steps f)-h) until the objective function value is minimized.

9. The method of claim 8, further comprising:

generating a normalized confusion matrix (S) based on each of the actual diagnostic labels and using each of the probability vectors.

10. The method of claim 9, wherein the step of generating a normalized confusion matrix comprises calculating a normalized score for each of the actual diagnostic labels.

11. The method of claim 10, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the objective function value is calculated using;

12. The method of claim 10, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the objective function value is calculated using;

13. A system for optimizing parameters of a plurality of diagnostic algorithms in a tunable diagnostic algorithm library, the system comprising:

a processor configured to implement each of the diagnostic algorithms and a fault model, and to receive (i) a plurality of sensed data sets, each having an actual diagnostic label associated therewith, and (ii) a value for each parameter of each of the diagnostic algorithms that is to be optimized, the processor further configured to;

a) generate a computed diagnostic label for each of the sensed data sets using each of the diagnostic algorithms and the value for each parameter;

b) supply each of the computed diagnostic labels and each of the actual diagnostic labels to a generic objective function, to thereby calculate an objective function value; and

c) selectively vary the value of one or more of the parameters using an optimization routine that repeats steps a)-c) until the objective function value is minimized.

14. The system of claim 13, wherein:

the processor is further configured to receive a weighting matrix (W); and

the weighting matrix is generated by;

determining, in the processor and based on the fault model, all possible computed diagnostic labels that each of the diagnostic algorithms will generate,

generating a confusion matrix having a plurality of entries and that includes a separate row for each of the possible computed diagnostic labels and a separate column for each actual diagnostic label, and

assigning a weight value to each entry in the confusion matrix.

15. The system of claim 14, wherein each computed diagnostic label comprises a probability vector corresponding to each possible computed diagnostic label.

16. The system of claim 15, wherein the processor is further configured to generate a normalized confusion matrix (S) based on each of the actual diagnostic labels and using each of the probability vectors.

17. The system of claim 16, wherein the processor is further configured to calculate a normalized score for each of the actual diagnostic labels.

18. The system of claim 16, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the processor is configured to calculate the objective function value using;

19. The system of claim 16, wherein:

the weighting matrix (W) and the normalized confusion matrix (S) are each p×

q matrices; and

the processor is configured to calculate the objective function value using;