Expert system using pattern recognition techniques

• US 5,060,279 A
• Filed: 09/27/1988
• Issued: 10/22/1991
• Est. Priority Date: 04/10/1986
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Company (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Cowan, Colin F. N., Marton, Virgil N., Crawford, Thomas M.
• Primary Examiner(s): Razavi, Michael
• Application Number:
  US07/265,859
• Time in Patent Office: 1,120 Days
• Field of Search: 382/14, 382/8, 382/1, 382/15, 382/57, 364/513, 364/580, 364/550, 364/551
• US Class Current: 706/20
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: Y10S 706/911 Nonmedical diagnostics

Abstract

An expert system having a processor which can receive input data and operate on it to produce data indicative of a parameter or parameters which are considered to define or represent the condition of equipment or a system to be diagnosed and which provides an adaptive pattern recognition facility which can be trained or instructed to recognize different combinations of input data as indicative of particular conditions of the equipment. The system is trained by coupling it initially to known good equipment into which definable faults are introduced. The adaptive pattern recognition facility has a series of sections with adjustable weight coefficients which are adjusted in accordance with a given algorithm as successive data combinations associated with the various faults are fed to the system. Once trained, the system can diagnose unknown equipment and provide outputs indicative of faults.

Claims

1. An expert system for fault diagnosis, comprising:

processing means for receiving input data relating to a device or system to be diagnosed and manipulating said input data to produce output data representative of at least one parameter indicating the condition of said device or system, said processing means having adaptive pattern recognition means for recognizing certain patterns of input data and associating said patterns with known conditions, means for storing pattern data representing said known conditions during an instruction mode, and means for comparing input data from said device or system with said stored pattern data to determine the presence of said known conditions in said input data during a diagnostic mode; and

means for displaying a result of said comparison as said output data representative of said at least one parameter.

2. A system as in claim 1, further comprising a keyboard for inputting said input data.

3. A system as in claim 1, wherein said adaptive pattern recognition means comprises an adaptive combiner which assigns adjustable weights to data input during said instruction mode, said weights being stored by said pattern data storing means as said pattern data.

4. A system as in claim 3, further comprising means for adjusting said weights in accordance with a recursive least squares error technique to produce updated pattern data better fitting said input data, said pattern data storing means storing said updated pattern data as said pattern data.

5. A system as in claim 4, wherein said comparing means comprises data input during said diagnostic mode and said stored pattern data with expected data corresponding to data expected to arise for a device or system with the known conditions represented by said stored pattern data.

6. A system as in claim 1, wherein said displaying means includes a visual display unit which is responsive to said comparison results to provide said output data as a visible indication of a type and degree of fault of said device or system.

7. A system as in claim 6, wherein said output data indicates the action required to correct any diagnosed fault of said device or system.

8. An expert system for fault diagnosis, comprising:

means for receiving input data relating to a device or system to be diagnosed;

adaptive pattern recognition means for recognizing certain patterns of input data from said receiving means and associating said patterns with known conditions;

means for storing pattern data representing said known conditions during an instruction mode;

means for comparing diagnostic data from said device or system input during a diagnostic mode with said stored pattern data to determine the presence of said known conditions in said diagnostic data; and

means for displaying a result of said comparison as output data representative of at least one parameter indicating the condition of said device or system.

9. A method of diagnosing faults in a system, comprising the steps of:

(a) connecting a device or system with a predetermined number of known faults to an expert system;

(b) receiving low-fault data from said device or system with said predetermined number of known faults, adjusting first weights of an adaptive pattern recognition means until an output of said adaptive pattern recognition means closely approximates said low-fault data, and storing said first weights;

(c) introducing a known fault to said device or system;

(d) receiving fault data from said device or system with said introducing known fault and said predetermined number of known faults, adjusting second weights of said adaptive pattern recognition means until an output of said adaptive pattern recognition means closely approximates said low-fault data, and storing said second weights;

(e) connecting a device or system to be diagnosed to said expert system;

(f) applying input data from said device or system to be diagnosed to said adaptive pattern recognition means and adjusting third weights of said adaptive pattern recognition means until an output of said adaptive pattern recognition means closely approximates said low-fault data;

(g) comparing said third weights to said first and second weights to determine the presence of said known fault in said input data; and

(h) outputting the results of said comparison as output data representative of at least one parameter indicating the condition of said device or system.

10. A method as in claim 9, wherein said steps of adjusting the weights of said adaptive pattern recognition means includes the step of applying a recursive least squares error method to said output of said adaptive pattern recognition means.

11. A method as in claim 9, comprising the further steps of repeating said steps (c) and (d) for different degrees of said known fault and for different known faults to generate a series of said second weights representative of respective degrees of said known fault and said different known faults.