Surveillance of industrial processes with correlated parameters

• US 5,586,066 A
• Filed: 06/08/1994
• Issued: 12/17/1996
• Est. Priority Date: 06/08/1994
• Status: Expired due to Term

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Wigeland, Roald A., Kubic, William L., Gross, Kenny C., White, Andrew M.
• Primary Examiner(s): Ramirez, Ellis B.
• Assistant Examiner(s): Wachsman, Hal D.
• Application Number:
  US08/255,586
• Time in Patent Office: 923 Days
• Field of Search: 395/914, 395/915, 376/215-217, 364/576, 364/550, 364/551.02, 364/554, 364/492, 364/185, 364/184, 340/506, 340/507, 340/521, 340/522, 340/511
• US Class Current: 702/181
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G05B 23/027 Alarm generation, e.g. communication protocol; Forms of alarm
• CPC Class Codes: Y10S 706/914 Process plant
• CPC Class Codes: Y10S 706/915 Power plant

Abstract

A system and method for surveillance of an industrial process. The system and method includes a plurality of sensors monitoring industrial process parameters, devices to convert the sensed data to computer compatible information and a computer which executes computer software directed to analyzing the sensor data to discern statistically reliable alarm conditions. The computer software is executed to remove serial correlation information and then calculate Mahalanobis distribution data to carry out a probability ratio test to determine alarm conditions.

Claims

1. A system for surveillance of an industrial process to assess at least one of an operational state of the industrial process and the operational state of sensors monitoring the process, comprising:

at least one sensor means for sensing raw data information from the industrial process corresponding to the operational state of the industrial process;

means for converting the raw data information to computer data;

computer means for receiving the computer data and generating sensor data analyzable by the computer means;

computer software means executable by the computer means for carrying out manipulation of the sensor data, said computer software means including, (a) means for removing serial correlation information between any of the means for sensing raw data information and (b) means for calculating Mahalanobis distribution data and determining a statistical deviation from previously accumulated training data for the industrial process, wherein the training data is characteristic of raw data information with the means for sensing operating normally and the industrial process properly functioning; and

means for applying a probability ratio test to the Mahalanobis distribution data to determine alarm conditions for said industrial process and said sensor means.

2. The system as defined in claim 1 wherein said computer software means includes means for assuring the auto- and cross-correlation structure data are unchanged relative to an earlier time.

3. The system as defined in claim 1 wherein said means for sensing comprises a plurality of individual sensors.

4. The system as defined in claim 1 wherein said means for removing serial correlation comprises at least one of (1) means for Fourier filtering the raw data information, (2) means for performing autoregressive filtering of the raw data information or (3) means for passing the raw data information without filtering.

5. The system as defined in claim 4 wherein said means for Fourier filtering comprises a fast Fourier transformation software package.

6. The system as defined in claim 4 wherein low frequency components are filtered by the means for Fourier filtering and the high frequency components are removed by the means for performing autoregressive filtering.

7. The system as defined in claim 1 wherein said means for applying a probability ratio test further comprises means for performing a sequential probability ratio test.

8. The system as defined in claim 1 wherein said means for calculating Mahalanobis distribution data includes means for calculating Mahalanobis distributions for a plurality of combinations in which each one of said means for sensing is masked stepwise.

9. The system as defined in claim 8 wherein the Mahalanobis distribution data are recalculated at later times by the means for calculating Mahalanobis data, thereby ensuring no significant time drift of the raw data information output from the means for sensing.

10. The system as defined in claim 1 wherein said means for removing serial correlation comprises means for superpositioning and manipulating the raw data information and derivatives thereof characteristic of one of Fourier filtered data and autoregressively filtered data.

11. The system as defined in claim 10 wherein the means for removing serial correlation by autoregression includes means for skipping selected ones of the raw data information.

12. The system as defined in claim 10 wherein the means for removing serial correlation by autoregression includes recalculation means for comparing newly calculated filtered data with previously calculated filtered data.

13. The system as defined in claim 1 wherein said means for calculating Mahalanobis distribution data includes means for transforming principal components.

14. The system as defined in claim 1 wherein the means for applying a probability ratio test includes means for specifying an offset for the Mahalanobis distribution data characteristic of abnormal operation of the industrial process.

15. The system as defined in claim 1 wherein the probability ratio test comprises forming a ratio:

##EQU9## where if P_n >

B implies no abnormality and if P_n <

A implies abnormality where A and B are assigned values for the particular said industrial process.

16. The system as defined in claim 15 wherein said system includes means for identifying a faulty one of said means for sensing, said means for identifying comprising means for masking sequentially each of said means for sensing and comparing the Mahalanobis distribution data, thereby obtaining P_n <

A when the faulty one of said means for sensing is unmasked.

17. The system as defined in claim 16 wherein said means for identifying further includes means for masking sequentially more than one of said means for sensing, thereby enabling identification of more than one faulty one of said means for sensing.

18. A method for performing surveillance of an industrial process using industrial sensors to assess at least one of an operational state of the industrial process and viability of the sensors, comprising the steps of:

sensing raw data information from the industrial process using a plurality of sensors;

converting the raw data information to computer data and outputting the computer data to a computer;

generating analyzable sensor data for the computer;

executing computer software using the computer to remove serial correlation information present for each sensor alone between the raw data information accumulated by each of the sensors at different times;

accumulating training data for the industrial process and the sensors characteristic of raw data information known to be representative of a properly functioning sensor and industrial process;

further executing the computer software to calculate a statistical deviation from the training data by determining Mahalanobis distribution data for both the raw data information and the training data; and

applying a probability ratio test to determine alarm conditions for abnormal states of the industrial process and the sensors.

19. The method as defined in claim 18 wherein the steps of removing serial correlation information comprises performing at least one of the step of Fourier filtering and autoregressively filtering the raw data information.

20. The method as defined in claim 18 further including the steps of performing a sequential probability ratio test after applying a probability ratio test of the Mahalanobis distribution data.

21. The method as defined in claim 18 wherein the step of determining the Mahalanobis distribution data includes sequentially calculating distribution data for multiple situations in which each one of said sensors is masked sequentially.

22. The method as defined in claim 18 wherein the Mahalanobis distribution data is recalculated periodically to ensure continuity with previous calculations.

23. The method as defined in claim 18 wherein the industrial process is selected from the group consisting of a nuclear power plant process, a fossil power plant process, an operating automobile, an operating aircraft, an operating ship, a water and waste treatment process, a pumping station process, an operating petroleum fluid pipeline, a chemical plant process and a biomedical process.

24. The method as defined in claim 18 wherein said step of executing computer software includes segmenting the raw data information into linear segments thereby characterizing nonlinear variable behavior as a plurality of segments of linear behavior.

25. The method as defined in claim 18 wherein the step of executing computer software includes the step of time averaging over slowly changing portions of the raw data information.

26. The method as defined in claim 18 wherein the step of executing computer software includes skipping analysis of selected portions of the raw data information over slowly changing portions of the raw data information.

27. The method as defined in claim 18 further including the step of generating an alarm for at least one of an automated response by a device operating the industrial process or a manual response by an operator.