Diagnostic system with learning capabilities

• US 6,442,542 B1
• Filed: 10/08/1999
• Issued: 08/27/2002
• Est. Priority Date: 10/08/1999
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Ramachandran, Ramesh, Bonissone, Piero Patrone, Johnson, John Andrew, Steen, Phillip Edward, Shah, Rasiklal Punjalal, Ramani, Vipin Kewal, Chen, Yu-To
• Primary Examiner(s): Alam, Hosain T.
• Application Number:
  US09/415,408
• Time in Patent Office: 1,054 Days
• Field of Search: 707/1, 707/5, 707/6, 707/104, 707/3, 707/4, 702/35, 706/14, 706/16, 706/25, 706/47, 714/25-26, 714/37
• US Class Current: 714/25
• CPC Class Codes: G01R 31/2846 using hard- or software simulation or using knowledge-based systems, e.g. expert systems, artificial intelligence or interactive algorithms
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: Y10S 707/915 Image
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing

Abstract

A diagnostic system is provided for identifying faults in a machine (e.g., CT scanner, MRI system, x-ray apparatus) by analyzing a data file generated thereby. The diagnostic system includes a trained database containing a plurality of trained data, each trained data associated with one of plurality of known fault types. Each trained data is represented by a trained set of feature values and corresponding weight values. Once a data file is generated by the machine, a current set of feature values are extracted from the data file by performing various analyses (e.g., time domain analysis, frequency domain analysis, wavelet analysis). The current set of feature values extracted is analyzed by a fault detector which produces a candidate set of faults based on the trained set of feature values and corresponding weight values for each of the fault types. The candidate set of faults produced by the fault detector is presented to a user along with a recommend repair procedure. In cases where no fault is identified or in response to a misdiagnosis produced by the diagnostic system, the user may interactively input a faulty condition associated with the machine being diagnosed (e.g., based on his/her experience). The diagnostic system further includes a learning subsystem which automatically updates the plurality of trained data based on the faulty condition input by the user.

Claims

1. A system for diagnosing a machine by analyzing a data file generated by the machine, comprising:

a trained database which contains a plurality of trained data associated with a plurality of fault types;

a feature extractor which extracts a plurality of feature values from the data file;

a fault detector which receives said plurality of feature values extracted and produces a candidate set of faults based on said plurality of trained data;

a user interface which presents said candidate set of faults produced by said fault detector to a user and allows said user to interactively input a faulty condition associated with the machine; and

a learning subsystem which updates said plurality of trained data based on said faulty condition input by said user.

2. The system of claim 1, wherein each of said plurality of trained data is associated with a respective one of said fault types represented by a trained set of feature values and corresponding weight values, said corresponding weight values indicate a relationship between said respective one of said fault types and said corresponding trained set of feature values.

3. The system of claim 2, wherein said fault detector is configured to:

access said trained set of feature values for each respective fault type;

calculate a distance vector for each of said fault types, representing how closely the extracted feature values match with said trained set of feature values based on said trained set of feature values and said corresponding weight values; and

identify said candidate set of faults based on said distance vector calculated for each fault type.

4. The system of claim 2, wherein said fault detector is configured to:

access said trained set of feature values for each respective fault type;

compute a plurality of regions for each of said fault types, each respective region represented by corresponding maximum and minimum values;

determine, for each of said fault types, which of said plurality of regions, the extracted feature values fall into based on said corresponding maximum and minimum values computed; and

based thereon, identify said candidate set of faults.

5. The system of claim 2, wherein said learning subsystem is configured to update said trained set of feature values corresponding to one of said fault types that corresponds to said faulty condition input by said user, based on said feature values extracted.

6. The system of claim 2, wherein said learning subsystem is configured to update said weight values corresponding to one of said faulty types, that corresponds to said faulty condition input by said user, based on a total number of trained cases for all fault types and a total number of trained cases for said one of said faulty types.

7. The system of claim 1, further comprising a parser which receives the data file generated by said machine and removes extraneous data from the data file before features are extracted from said data file.

8. The system of claim 1, further comprising a gross filter which categorizes the data file as normal or faulty data, and wherein features are extracted only from data files categorized as faulty data.

9. The system of claim 1, wherein said machine being diagnosed is an imaging system.

10. A method for diagnosing a machine by analyzing a data file generated by the machine, comprising:

receiving the data file generated by the machine;

extracting a plurality of feature values from the data file received;

accessing a plurality of trained data associated with a plurality of known fault types;

producing a candidate set of faults based on said plurality of feature values extracted and said plurality of trained data accessed;

presenting said candidate set of faults produced to a user;

allowing said user to interactively input a faulty condition associated with the data file; and

updating said plurality of trained data based on said faulty condition input by said user.

11. The method of claim 10, wherein each of said plurality of trained data includes a trained set of feature values and corresponding weight values for each of said plurality of known faulty types.

12. The method of claim 11, wherein said step of producing a candidate set of faults further comprises the steps of:

accessing said trained set of feature values and corresponding weight values for each respective fault type;

calculating a distance vector for each of said fault types, representing how closely the extracted feature values match with said trained set of feature values based on said trained set of feature values and said corresponding weight values; and

identifying said candidate set of faults based on said distance vector calculated for each fault type.

13. The method of claim 11, wherein said step of producing a candidate set of faults further comprises the steps of:

accessing said trained set of feature values for each respective fault type;

computing a plurality of regions for each of said fault types, each respective region represented by corresponding maximum and minimum values;

determining, for each of said fault types, which of said plurality of regions, the extracted feature values fall into based on said corresponding maximum and minimum values computed; and

based thereon, identifying said candidate set of faults.

14. The method of claim 11, wherein said step of updating said plurality of trained data further comprises the step of updating said trained set of feature values corresponding to one of said fault types, that corresponds to said faulty condition input by said user, based on said feature values extracted.

15. The method of claim 11, wherein said step of updating said plurality of trained data further comprises the step of updating said weight values corresponding to one of said faulty types, that corresponds to said faulty condition input by said user, based on a total number of trained cases for all fault types and a total number of trained cases for said one of said faulty types.

16. The method of claim 10, further comprising the step of removing extraneous data from the data file before features are extracted from said data file.

17. The method of claim 10, further comprising the step of categorizing the data file as normal or faulty data, and wherein features are extracted only from data file categorized as faulty data.

18. The method of claim 10, wherein said machine being diagnosed is an imaging system.