Apparatus and methods for detecting system faults using hidden process drivers

• US 20060047482A1
• Filed: 08/24/2005
• Published: 03/02/2006
• Est. Priority Date: 08/25/2004
• Status: Active Grant

• Current Assignee: Siemens Corp. (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Neubauer, Claus, Yuan, Chao, Cataltepe, Zehra
• Granted Patent: US 7,216,061 B2
• Time in Patent Office: Days
• US Class Current: 702/185
• CPC Class Codes: G05B 23/0221 Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
• CPC Class Codes: G06F 11/0736 in functional embedded systems, i.e. in a data processing system designed as a combination of hardware and software dedicated to performing a certain function testing or monitoring of automated control systems G05B23/02
• CPC Class Codes: G06F 11/0751 Error or fault detection not based on redundancy power supply failures G06F1/30; network fault management H04L41/06

Abstract

An apparatus for detecting faults in a system monitored by a plurality of sensors is provided. The apparatus includes a hidden process driver unit that generates a hidden process driver based upon sensor values received from a group of correlated sensors selected from among the plurality of sensors. The apparatus also includes a sensor estimating unit that generates sensor estimates for each of the plurality of sensors based upon the hidden process driver and a known process driver provided by an uncorrelated sensor. The apparatus further includes a fault determining unit that indicates a fault when a residual based upon a difference between a sensor value supplied by one of the plurality of sensors and a corresponding one of the sensor estimates lies outside an acceptable range of residual values.

Claims

1. A computer-based method for detecting faults in a system, the method comprising:

grouping a plurality of sensors communicatively linked with the system into at least one group of correlated sensors based upon a measure of correlation between the plurality of sensors;

generating a hidden process driver for the at least one group of correlated sensors based upon a functional relationship between the hidden process driver and a set of sensor values produced by each of the correlated sensors;

generating a sensor estimate based upon the hidden process driver corresponding to the at least one group of correlated sensors and at least one known process driver; and

detecting a fault when a residual based on a difference between a sensor value produced by a particular one of the plurality of sensors and a sensor estimate corresponding to that particular sensor lies outside a predetermined acceptable range of residuals for the particular sensor.

2. The method of claim 1, further comprising determining a non-parametric form of the functional relationship based upon a set of training data produced by the plurality of sensors.

3. The method of claim 1, further comprising estimating at least one parameter of the functional relationship based on a set of training data produced by the plurality of sensors.

4. The method of claim 3, wherein estimating at least one parameter comprises estimating the at least one parameter based upon a least-squares regression on the set of training data.

5. The method of claim 3, wherein the functional relationship is defined by one of a linear function, a quadratic function, and an exponential function.

6. The method of claim 1, further comprising determining an inverse function for each of the plurality of correlated sensors based upon the functional relationship between the process driver and the set of sensor values.

7. The method of claim 6, further comprising determining a plurality of confidence weights, each corresponding to a distinct one of the plurality of correlated sensors, and computing a weighted average based upon the confidence weights.

8. A method of estimating values of sensors used in monitoring a system, the method comprising:

grouping a plurality of sensors communicatively linked with the system into at least one group of correlated sensors based upon a measure of correlation between the plurality of sensors;

generating a hidden process driver for the at least one group of correlated sensors based upon a plurality of inverse sensor functions and confidence weights, each of the plurality of inverse sensor functions and confidence weights uniquely corresponding to a different one of the group of correlated sensors;

generating a sensor estimate based upon the hidden process driver corresponding to the at least one group of correlated sensors and at least one known process driver.

9. The method of claim 8, further comprising iteratively determining the confidence weights by computing successive values for each confidence weight until the successive values converge.

10. The method of claim 8, wherein at least one of the inverse sensor functions is an inverse function of at least one of a linear function, a quadratic function, and an exponential function.

11. The method of claim 8, wherein at least one of the inverse sensor functions corresponding to one of the group of correlated sensors is an inverse function of a function having a non-parametric form, and further comprising determining the non-parametric form based upon a plurality of segmented linear functions determined using a set of training data produced by the one of the group of correlated sensors to which the at least one inverse sensor function corresponds.

12. An apparatus for detecting faults in a system monitored by a plurality of sensors, the apparatus comprising:

a hidden process driver unit that generates at least one hidden process driver based upon sensor values received from at least one group of correlated sensors selected from among the plurality of sensors;

a sensor estimating unit that generates at least one sensor estimate based upon the at least one hidden process driver and at least one known process driver provided by at least one uncorrelated sensor from among the plurality of sensors; and

a fault determining unit that indicates a fault when a residual based upon a difference between a sensor value supplied by one of the plurality of sensors and a corresponding sensor estimate lies outside a predetermined acceptable range of residual values.

13. The apparatus of claim 12, wherein the hidden process driver unit comprises a function determining module that determines a plurality of inverse sensor functions, each of the plurality of inverse sensor functions uniquely corresponding to a distinct one of the sensors belonging to the group of correlated sensors.

14. The apparatus of claim 13, wherein the function determining module is configured to determine the plurality of inverse sensor functions based upon at least one of a linear function, a quadratic function, and an exponential function.

15. The apparatus of claim 13, wherein the function determining module is configured to determine the plurality of inverse sensor functions based on at least one function having a non-parametric form.

16. The apparatus of claim 13, wherein the function determining module is configured to determine the plurality of inverse sensor functions by computing at least one parameter.

17. The apparatus of claim 13, wherein the function determining module is configured to determine the at least one parameter based upon at least one least-squares statistic

18. The apparatus of claim 12, wherein the hidden process driver comprises a confidence weight determining module that determines a plurality of confidence weights, each of the plurality of confidence weights corresponding to a distinct one of the sensors belonging to the group of correlated sensors.

19. The apparatus of claim 18, wherein the confidence weight determining module iteratively determines the confidence weights by computing successive values for each confidence weight until the successive values converge.