Surveillance system and method having an adaptive sequential probability fault detection test

• US 6,892,163 B1
• Filed: 03/08/2002
• Issued: 05/10/2005
• Est. Priority Date: 03/08/2002
• Status: Expired due to Term

• Current Assignee: Intellectual Assets LLC
• Original Assignee: Intellectual Assets LLC
• Inventors: Bickford, Randall L., Herzog, James P.
• Primary Examiner(s): Barlow, John
• Assistant Examiner(s): Taylor, Victor J.
• Application Number:
  US10/095,835
• Time in Patent Office: 1,159 Days
• Field of Search: 702/183, 702/72, 702/189, 702/185, 702/181, 700/30, 700/29, 714/26, 706/15
• US Class Current: 702/181
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G06F 18/24155 Bayesian classification

Abstract

System and method providing surveillance of an asset such as a process and/or apparatus by providing training and surveillance procedures that numerically fit a probability density function to an observed residual error signal distribution that is correlative to normal asset operation and then utilizes the fitted probability density function in a dynamic statistical hypothesis test for providing improved asset surveillance.

Claims

1. A method for performing surveillance of an asset, the steps including:

acquiring residual data correlative to expected asset operation;

fitting an equation to said acquired residual data and storing said fitted equation in a memory means;

collecting a current set of observed signal data values from the asset;

using said fitted equation in a sequential hypotheses test to determine if said current set of observed signal data indicates unexpected operation of the asset that is indicative of a fault condition;

outputting a signal correlative to a detected fault condition for providing asset surveillance.

2. The method of claim 1 wherein the step of acquiring residual data correlative to expected asset operation includes acquiring training data values correlative to expected asset operation and obtaining deviations between said training data values and values obtained by using a parameter estimation model wherein said parameter estimation model values are a function of said training data values.

3. The method of claim 2 wherein the step of fitting an equation to said acquired residual data and storing said fitted equation in a memory means includes the step of fitting a general probability density function (PDF) to said acquired residual data such that said general PDF is a function of statistical moments of said residual data.

4. The method of claim 3 wherein the step of fitting said general PDF includes the step of fitting a function of a first statistical moment of said residual data, a function of a second statistical moment of said residual data, and a function of at least one higher order statistical moment of said residual data.

5. The method of claim 4 wherein said function of a first statistical moment of said residual data is a sample mean.

6. The method of claim 4 wherein said function of a second statistical moment of said residual data is a sample variance.

7. The method of claim 4 wherein said function of at least one higher order statistical moment of said residual data is a sample skewness.

8. The method of claim 4 wherein said function of at least one higher order statistical moment of said residual data is a sample kurtosis.

9. The method of claim 4 wherein said function of at least one higher order statistical moment of said residual data is a fifth or higher order statistical moment.

10. The method of claim 3 wherein the step of fitting an equation to said acquired residual data and storing said fitted equation in a memory means includes the step of fitting a general probability density function (PDF) to said acquired residual data by fitting a standard Gaussian PDF to said acquired residual data and by then adding successive higher order terms to said standard Gaussian PDF until an adequate agreement between the so derived general PDF and the actual distribution of said acquired residual data is achieved.

11. The method of claim 3 wherein the step of using said fitted equation in a sequential hypotheses test to determine if said current set of observed signal data indicates unexpected operation of the asset that is indicative of a fault condition further includes the step of performing at least one of a group of four sequential hypothesis tests comprised of a positive mean test, a negative mean test, a nominal variance test, and a inverse variance test all as a function of said general probability density (PDF) function.

12. The method of claim 11 wherein the step of performing at least one of a group of four sequential hypothesis tests comprised of a positive mean test, a negative mean test, a nominal variance test, and a inverse variance test all as a function of said general probability density function (PDF) includes the step of processing said at least one of a group of four sequential hypothesis tests and comparing an outcome of each processed test to upper and lower threshold limits for determining unexpected operation of the asset that is indicative of a fault condition.

13. A surveillance system for monitoring an asset, said system comprising in combination:

a data acquisition means for acquiring a current set of signals engendered from said asset correlative to asset status;

a digitizing means for digitizing said current set of signals for defining a current set of digitized signals;

a parameter estimation means for producing a set of estimated signal values as a function of said current set of digitized signals;

a fault detection means for detecting the presence of a fault as a function of said set of estimated signal values and said current set of digitized signals;

said fault detection means including a stored fault detection model comprised of an empirically derived probability density function utilized in a sequential probability test of at least one result of a mathematical operation between said set of estimated signal values and said current set of digitized signals;

a communication means for communicating detected faults to a remote location for providing asset surveillance.

14. The system of claim 13 wherein said parameter estimation means for producing said set of estimated signal values is of a type individually selected from a group comprised of a multivariate state estimation technique method, a neural network method, a mathematical process model method, a autoregressive moving average method, and a Kalman filter method.

15. The system of claim 14 wherein said empirically derived probability density function utilized in said sequential probability test is obtained by acquiring training data values correlative to expected asset operation and obtaining residual data by computing differences between said training data values and values obtained by using said parameter estimation model on said training data values and fitting a general probability density function (PDF) to said residual data.

16. The system of claim 15 wherein said fitted general PDF is a function of statistical moments of said residual data.

17. The system of claim 16 wherein said general PDF is a function of a first statistical moment of said residual data, of a second statistical moment of said residual data, and at least one higher order statistical moment of said residual data.

18. The system of claim 17 wherein said first statistical moment of said residual data is a sample mean.

19. The system of claim 17 wherein said second statistical moment of said residual data is a sample variance.

20. The system of claim 17 wherein said at least one higher order statistical moment of said residual data is a sample skewness.

21. The system of claim 17 wherein said at least one higher order statistical moment of said residual data is a sample kurtosis.

22. The system of claim 17 wherein said at least one higher order statistical moment of said residual data is a fifth or higher statistical moment.

23. The system of claim 15 wherein said general probability density function (PDF) of said acquired residual data is comprised of a standard Gaussian PDF and at least one higher order term to said standard Gaussian PDF.

24. A method for training a plurality of process models wherein each process model represents a set of parameter estimation models and a set of fault detection models of an asset to be monitored by a surveillance system, the steps including:

acquiring at least one set of signal values associated with a current status of said asset;

forming a training data set comprised of at least one said set of signal values;

providing a memory means for storing and retrieving said training data set;

processing said training data set comprised of at least one said set of signal values associated with said current status of said asset for training at least one parameter estimation model;

estimating at least one set of process parameters by using at least one said parameter estimation model for processing said training data set;

empirically fitting a general probability density function to at least one said set of signals values in said training data set as a function of at least one said estimated set of process parameters and said training data set for obtaining at least one empirically derived fault detection model; and

storing within said memory means said at least one said parameter estimation model and at least one said empirically derived fault detection model for subsequent use in surveillance of said asset.

25. A method for performing surveillance of an asset, the steps including:

numerically fitting a probability density function to a distribution correlative to normal operation of the asset;

acquiring a current set of observed signal data values from the asset;

calculating a current set of estimated signal data values from a stored model for said asset;

transforming said calculated current set of estimated signal data values with said current set of observed signal data values for defining residual data values;

utilizing said fitted probability density function in a statistical hypothesis test performed on said residual data values for detecting a fault condition; and

outputting a signal correlative to each detected fault condition for providing asset surveillance.

26. The method of claim 25 ether including a step of forming said stored model.

27. The method of claim 26 wherein the step of forming said stored model includes a step of preparing said stored model using historical operating data from the asset.

28. The method of claim 27 wherein the step of preparing said stored model using historical operating data from the asset further includes a step of creating a training data set having discrete observations correlative to the expected normal operation of the asset wherein each single observation is comprised of a vector of data values from a digitized set of signals correlative to each signal parameter to be included in said stored model.

29. The method of claim 28 wherein said step of forming said stored model includes a step of creating a parameter estimation model by training said stored model using said training data set.

30. The method of claim 25 wherein said step of numerically fitting a probability density function to a distribution correlative to normal operation of the asset includes a step of preparing said fitted probability density function using historical operating data from the asset.

31. The method of claim 30 wherein the step of preparing said fitted probability density function using historical operating data from the asset further includes a step of creating a training residual data set having discrete residual data values correlative to the expected normal operation of the asset wherein each residual data value is comprised of a deviation between a historical operating data signal value and a estimated signal value produced by said stored model for said asset.

32. The method of claim 31 wherein said step of forming said fitted probability density function includes a step of numerically determining an equation correlative to said training residual data set having discrete residual data values correlative to the expected normal operation of the asset.