Method for adaptive threshold computation for time and frequency based anomalous feature identification in fault log data

• US 6,625,589 B1
• Filed: 08/04/2000
• Issued: 09/23/2003
• Est. Priority Date: 10/28/1999
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Varma, Anil, Roddy, Nicholas Edward
• Primary Examiner(s): Davis, George B.
• Application Number:
  US09/633,094
• Time in Patent Office: 1,145 Days
• Field of Search: 706/21, 706/45, 700/79
• US Class Current: 706/45
• CPC Class Codes: G05B 23/0218 characterised by the fault detection method dealing with either existing or incipient faults
• CPC Class Codes: G05B 23/0289 Reconfiguration to prevent failure, e.g. usually as a reaction to incipient failure detection
• CPC Class Codes: G06Q 10/06395 Quality analysis or management

Abstract

An algorithm for improving the probability of identifying the repair that will correct a fault aboard a machine, such as a locomotive. The invention utilizes historical fault log and repair data and further calculates the number of times a particular fault occurs in a given number of days and also the number of times a particular fault occurs on each day. Averages are calculated for these results and when the number of fault occurrences exceed some or more of those averages, then these faults are deemed statistically significant for subsequent processing.

Claims

1. A method for processing repair data and fault data comprising a plurality of faults from one or more machines, said method for facilitating analysis of a malfunctioning machine comprising:

(a) selecting from the fault data those faults having significant statistical relevance further comprising;

(a)(1) selecting at least one time window;

(a)(2) determining a threshold;

(a)(3) determining the number of occurrences of a fault during the time window;

(a)(4) determining the relationship between the number of occurrences of the fault and the threshold; and

(a)(5) in response to step (a4), determining the faults having significant statistical relevance;

(b) selecting a repair from the repair data;

(c) generating a case using the selected repair from step (b) and the selected faults from step (a);

(d) generating for each of the cases at least one repair and distinct fault cluster combination; and

(e) assigning a weight to each of the repair and distinct fault cluster combinations whereby the weight facilitates prediction of at least one repair for the malfunctioning machine.

2. The method of claim 1 wherein the at least one repair and distinct fault cluster combination comprises a plurality of combinations of the repair and the plurality of distinct statistically significant faults, wherein the number of combinations is equal to the number of unique combinations derivable from the set of the plurality of distinct statistically significant faults.

3. The method of claim 1 wherein the step of assigning weights comprises determining, for each repair and distinct fault cluster combination, the number of times the combination occurs in cases comprising related repairs, and the number of times the combination occurs in the plurality of cases, and dividing the number of times the combination occurs in cases comprising related repairs by the number of times the combination occurs in the plurality of cases.

4. The method of claim 1 wherein the statistically significant faults include those exceeding an average number of occurrences within the at least one time window.

5. The method of claim 1 wherein the statistically significant faults includes those exceeding a daily occurrence average within the at least one time window.

6. The method of claim 1 further comprising:

generating a new case from repair data and fault data, the new case comprising a repair and a plurality of distinct statistically significant faults;

generating, for the new case, a plurality of fault clusters for the plurality of distinct faults; and

redetermining a weight for each of the plurality of repair and fault clusters combinations to include the new case.

7. A system for processing repair data and fault data comprising a plurality of faults from one or more machines, said system for facilitating analysis of a malfunctioning machine comprising:

means for selecting from the fault data those faults having significant statistical relevance, further comprising;

means for selecting at least one time window;

means for determining a threshold;

means for determining the number of occurrences of a fault during the time window;

means for determining the relationship between the number of occurrences of the fault and the threshold;

in response to the means for determining the relationship, means for determining that the fault has significant statistical relevance;

means for generating a plurality of cases from the repair data and the selected fault data, wherein each case comprises a repair and a plurality of distinct statistically significant faults;

means for generating for each of the cases at least one repair and distinct fault cluster combination; and

means for assigning a weight to each of the repair and distinct fault cluster combinations whereby the weight facilitates prediction of at least one repair for the malfunctioning machine.

8. The system of claim 7 wherein the at least one repair and distinct fault cluster combination comprises a plurality of combinations of the repair and the plurality of distinct statistically significant faults, wherein the number of combinations is equal to the number of unique combinations derivable from the set of the plurality of distinct statistically significant faults.

9. The system of claim 7 wherein the means for assigning weights comprises means for determining, for each repair and distinct fault cluster combination, a number of times the combination occurs in cases comprising related repairs, and a number of times the combination occurs in the plurality of cases, and means for dividing the number of times the combination occurs in cases comprising related repairs by the number of times the combination occurs in the plurality of cases.

10. The system of claim 7 wherein the statistically significant faults include those having an above average number of occurrences within the at least one time window.

11. The system of claim 7 wherein the statistically significant faults includes those occurring at least once on more than the average number of days within the at least one time window.

12. A system for processing repair data and fault data comprising a plurality of faults from one or more machines, said system for facilitating analysis of a malfunctioning machine comprising:

a fault selector for selecting at least one time window, for determining a threshold, for determining the number of occurrences of the selected fault during the time window, for determining whether the number of fault occurrences exceeds the threshold, and for selecting faults for which the number of occurrences exceeds the threshold;

a case creator for generating a plurality of cases from the repair data and the selected faults, wherein each case comprises a repair and the selected faults having relevance to the repair;

a cluster creator for generating for each of the cases at least one repair and distinct fault cluster combination; and

a weight calculator for assigning a weight to each of the repair and distinct fault cluster combinations whereby the weight facilitates prediction of at least one repair for the malfunctioning machine.

13. The system of claim 12 wherein the at least one repair and distinct fault cluster combination comprises a plurality of combinations of the repair and the selected faults wherein the number of such combinations is equal to the number of unique combinations derivable from the set of the selected faults.

14. The system of claim 12 wherein the selected faults relevant to the repair include only the faults occurring over a period of time prior to the repair.

15. The system of claim 12 wherein the weight calculator comprises:

a counter for determining for each repair and distinct fault cluster combination, a number of times the combination occurs in cases comprising related repairs and a number of times the combination occurs in the plurality of cases; and

a divider for dividing the number of times the combination occurs in cases comprising related repairs by the number of times the combination occurs in the plurality of cases, wherein the result is the weight.

16. The system of claim 12 wherein the selected faults include those having an above average number of occurrences within the at least one time window.

17. The system of claim 12 wherein the selected faults includes those occurring on an above average number of days within the at least one time window.

18. An article of manufacture comprising:

a computer program product comprising computer usable medium having computer readable program code means embodied therein for causing the processing of repair data comprising a plurality of repairs and fault data comprising a plurality of faults from one or more machines, to facilitate analysis of a malfunctioning machine, computer readable program code in said article of manufacture comprising;

computer readable program code for causing a computer to select faults from among the fault data by selecting at least one time window, determining a threshold, determining the number of occurrences of a fault during the time window, determining the relationship between the number of occurrences of the fault and the threshold, and selecting faults in response to the relationship between the number of occurrences of the fault and the threshold;

computer readable program code for causing a computer to generate a plurality of cases from the repair data and the selected fault data, wherein each case comprises a repair and the selected faults;

computer readable program code for causing a computer to generate for each of the plurality of cases a plurality of clusters, wherein each cluster includes a repair and one element from the set of all unique combinations derivable from the selected faults; and

computer readable program code for causing a computer to assign a weight to each of the clusters, whereby the weight facilitates prediction of at least one repair for the malfunctioning machine.