Methods and systems for fault diagnosis in observation rich systems

• US 8,359,110 B2
• Filed: 03/23/2009
• Issued: 01/22/2013
• Est. Priority Date: 03/23/2009
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Palo Alto Research Center, Inc. (Xerox Holdings Corp.)
• Inventors: Kuhn, Lukas D., de Kleer, Johan
• Primary Examiner(s): Gami, Tejal
• Application Number:
  US12/409,157
• Publication Number:
  US 20100241251A1
• Time in Patent Office: 1,401 Days
• Field of Search: 700/31, 700/29, 700/47, 700/79, 700/21, 706/52
• US Class Current: 700/31
• CPC Class Codes: G05B 23/0245 based on a qualitative model, e.g. rule based; if-then decisions
• CPC Class Codes: G05B 23/0262 Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
• CPC Class Codes: Y02P 90/80 Management or planning

Abstract

Diagnostic systems and methods are presented for determining the current condition of a production plant and the resources thereof, in which successively more complex diagnostic abstractions are used to determine the plant condition, with a more complex abstraction being selected when the most recently selected diagnostic abstraction is logically inconsistent with the current fault status indications.

Claims

1. A control system for controlling operation of a production system with a plant that achieves one or more production goals, the control system comprising:

at least one processor;

an electronic memory operatively coupled with the at least one processor;

a planner implemented using the at least one processor, and operative to provide plans for execution using one or more plant resources in the plant;

a model of the plant; and

a diagnosis system comprising;

a plurality of diagnostic abstractions stored in the electronic memory and individually representing one or more fault assumptions regarding resources of the plant with a complexity of the fault assumption or assumptions of each diagnostic abstraction being different, the plurality of diagnostic abstractions includes at least one abstraction representing a persistent fault assumption and at least one abstraction representing an intermittent fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively;

a belief model comprising at least one fault status indication for at least one resource of the plant, a list of good or exonerated resources of the plant, a list of bad or suspected resources of the plant, and a list of unknown resources of the plant; and

a diagnoser including;

an abstraction diagnosis component implemented using the at least one processor, and operative to initially select a first one of the plurality of diagnostic abstractions having the least complex fault assumption or assumptions and to subsequently select another one of the diagnostic abstractions having more complex fault assumptions when a most recently selected diagnostic abstraction is logically inconsistent with the current fault status indications in the belief model, a domain diagnosis component implemented using the at least one processor, and operative to determine a current plant condition and to update the belief model based at least partially on the currently selected diagnostic abstraction, a previously executed plan, at least one corresponding observation from the plant, and the plant model; and

said planner generating at least one plan to schedule jobs to improve productivity based on said belief model to avoid using one or more resources known, or believed with high probability, to be faulty.

2. The control system of claim 1, wherein the plurality of diagnostic abstractions includes at least one abstraction representing a single fault assumption and at least one abstraction representing a multiple fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

3. The control system of claim 2, wherein the plurality of diagnostic abstractions includes at least one abstraction representing an interaction fault assumption and at least one abstraction representing a non-interaction fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

4. The control system of claim 1, wherein the plurality of diagnostic abstractions includes at least one abstraction representing an interaction fault assumption and at least one abstraction representing a non-interaction fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

5. The control system of claim 4, wherein the domain diagnosis component is operative to identify at least one interaction fault involving interaction of two or more resources of the plant.

6. The control system of claim 1, wherein at least one fault status indication in the belief model indicates a fault probability for at least one resource of the plant.

7. A method of determining a current condition of resources of a production plant, the method comprising:

storing in electronic memory, a plurality of diagnostic abstractions individually representing one or more fault assumptions regarding resources of the plant with a complexity of the fault assumption or assumptions of each diagnostic abstraction being different, the plurality of diagnostic abstractions includes at least one abstraction representing a persistent fault assumption and at least one abstraction representing an intermittent fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively;

selecting a first one of the plurality of diagnostic abstractions having the least complex fault assumption or assumptions regarding resources of the plant;

determining a current plant condition based at least partially on the currently selected diagnostic abstraction, a previously executed plan, at least one corresponding observation from the plant, and a plant model;

selectively selecting another one of the diagnostic abstractions having more complex fault assumptions when a most recently selected diagnostic abstraction is logically inconsistent with the current fault status indications;

maintaining a belief model in the electronic memory, the belief model comprising at least one fault status indication for at least one resource of the plant, a list of good or exonerated resources of the plant, a list of bad or suspected resources of the plant, and a list of unknown resources of the plant;

updating the belief model based at least partially on the currently selected diagnostic abstraction, a previously executed plan, at least one corresponding observation from the plant, and the plant model; and

generating at least one plan to schedule jobs to improve productivity based on said belief model to avoid using one or more resources known, or believed with high probability, to be faulty.

8. The method of claim 7, wherein the plurality of diagnostic abstractions includes at least one abstraction representing a single fault assumption and at least one abstraction representing a multiple fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

9. The method of claim 7, wherein the plurality of diagnostic abstractions includes at least one abstraction representing a persistent fault assumption and at least one abstraction representing an intermittent fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

10. The method of claim 7, wherein the plurality of diagnostic abstractions includes at least one abstraction representing an interaction fault assumption and at least one abstraction representing a non-interaction fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively.

11. The method of claim 10, wherein determining a current plant condition comprises identifying at least one interaction fault involving interaction of two or more resources of the plant.

12. A non-transitory computer readable medium having computer executable instructions for performing the steps of:

storing in electronic memory, a plurality of diagnostic abstractions individually representing one or more fault assumptions regarding resources of the plant with a complexity of the fault assumption or assumptions of each diagnostic abstraction being different, the plurality of diagnostic abstractions includes at least one abstraction representing a persistent fault assumption and at least one abstraction representing an intermittent fault assumption regarding resources of the plant with successively more complex fault assumptions, respectively;

selecting a first one of the plurality of diagnostic abstractions having the least complex fault assumption or assumptions regarding resources of a production plant;

determining a current plant condition based at least partially on the currently selected diagnostic abstraction, a previously executed plan, at least one corresponding observation from the plant, and a plant model;

selectively selecting another one of the diagnostic abstractions having more complex fault assumptions when a most recently selected diagnostic abstraction is logically inconsistent with the current fault status indications;

maintaining a belief model in the electronic memory, the belief model comprising at least one fault status indication for at least one resource of the plant, a list of good or exonerated resources of the plant, a list of bad or suspected resources of the plant, and a list of unknown resources of the plant;

updating the belief model based at least partially on the currently selected diagnostic abstraction, a previously executed plan, at least one corresponding observation from the plant, and the plant model; and

generating at least one plan to schedule jobs to improve productivity based on said belief model to avoid using one or more resources known, or believed with high probability, to be faulty.