Kernel-based method for detecting boiler tube leaks

• US 8,275,577 B2
• Filed: 09/18/2007
• Issued: 09/25/2012
• Est. Priority Date: 09/19/2006
• Status: Active Grant

• Current Assignee: SmartSignal Corporation (Emerson Electric Company)
• Original Assignee: SmartSignal Corporation (Emerson Electric Company)
• Inventors: Herzog, James P.
• Primary Examiner(s): Charioui, Mohamed
• Application Number:
  US11/856,897
• Publication Number:
  US 20080071501A1
• Time in Patent Office: 1,834 Days
• Field of Search: 702/59, 702/65, 702/67, 702/68, 702/104, 702/119, 702/179, 702/180, 702/182, 702/183, 702/188, 702/189
• US Class Current: 702/183
• CPC Class Codes: F22B 37/10 Water tubes; Accessories therefor working of metal tubes B21D; pipes in general F16L; repairing leaks in water tubes F16L55/16; F28F11/00; baffles, screens, or deflectors formed of water tubes F23M9/10; cleaning internal or external surfaces of water tubes F28G
• CPC Class Codes: F28F 2265/16 for preventing leakage
• 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

Abstract

A method and apparatus are provided for diagnosing faults in a monitored system that is monitored by sensors. An empirical model is generated for a targeted component of the monitored system. The empirical model is trained with an historical data source that contains example observations of the sensors. Substantially real-time estimates are generated based on instrumented data corresponding to the targeted component. The substantially real-time estimates are compared and differenced with instrumented readings from the sensors to provide residual values. The residual values are analyzed to detect the faults and determine a location of the faults in the monitored system.

Claims

1. A method of diagnosing faults in a monitored system, the monitored system being monitored by sensors, comprising:

constructing an empirical model for a targeted component of the monitored system, wherein the empirical model is trained with a historical data source that contains example observations of the sensors;

generating substantially real-time estimates based on instrumented data corresponding to the targeted component;

comparing and differencing the substantially real-time estimates with instrumented readings from the sensors in the form of input observations to provide residual values;

analyzing the residual values to detect the faults and determine a location of the faults in the monitored system; and

adapting, by a processor, the empirical model with input observations indicating normal operation of the monitored system only after a predetermined time period from acquiring each such input observation.

2. The method of claim 1, wherein the targeted component consists of steam generating equipment which contains a first set of sensors to monitor hot side conditions of steam generating equipment and a second set of the sensors to monitor steam/water conditions of the steam generating equipment.

3. The method of claim 2, further comprising determining the location of a tube leak by graphically displaying, on a visual interface, a representation of one or more components of the steam generating equipment and the locations of tubes within one or more of the components, and indicating residual values at the representation of physical locations of sensors that correspond to the residual values, the sensors measuring the temperature near tubes at different locations within the one or more components.

4. The method of claim 3 wherein the sensors comprise tube-bundle thermocouples.

5. The method of claim 2, wherein the steam generating equipment is a boiler of a fossil-fuel power plant.

6. The method of claim 2, wherein the steam generating equipment is a steam generator of a nuclear power plant.

7. A method of claim 1, wherein the empirical model generates estimated sensor values according to a nonparametric kernel-based method.

8. A method of claim 7, wherein the empirical model generates estimated sensor values according to a similarity-based modeling method.

9. A method of claim 7, wherein the empirical model generates estimated sensor values according to a kernel regression modeling method.

10. The method of claim 1, wherein adapting corriprises having the empirical model implement an adaptation algorithm to learn new normal variation patterns in operation of the monitored system.

11. The method of claim 10, wherein the adaptation algorithm utilizes at least one of:

manual (user-driven), trailing, out-of-range, in-range, and control-variable driven adaptation algorithms.

12. The method of claim 1, wherein the empirical model is updated to form a subset of the example observations and in real-time with each new input observation localized within a learned reference library to those example observations that are relevant to the input observation, according to predetermined relevance criteria.

13. The method of claim 1, wherein the generating substantially real-time estimates based on data corresponding to the targeted component comprises generating at least one inferred real-time estimate.

14. The method of claim 1, wherein the predetermined time period is set to at least the amount of time the system being monitored can still operate with a known fault.

15. The method of claim 3 wherein the representation is a grid with columns and rows, and wherein squares formed by the grid represent a bundle of tubes.

16. A monitoring apparatus for diagnosing faults in a system monitored by sensors, comprising:

a reference data store containing instrumented data corresponding to a targeted component of the system;

a processor to

construct an empirical model for a targeted component of the monitored system, wherein the empirical model is trained with a historical data source that contains example observations of the sensors;

generate substantially real-time estimates based on the instrumented data corresponding to the targeted component;

compare and difference the substantially real-time estimates with instrumented readings from the sensors in the form of input observations to provide residual values;

analyze the residual values to detect the faults and determine a location of the faults in the monitored system; and

adapt the empirical model with input observations indicating normal operation of the system only after a predetermined time period from acquiring each such input observation.

17. The monitoring apparatus of claim 16, wherein the processor constructs an empirical model to generate estimated sensor values according to a nonparametric kernel-based method.

18. The monitoring apparatus of claim 17, wherein the processor constructs an empirical model to generate estimated sensor values according to a similarity-based modeling method.

19. The monitoring apparatus of claim 17, wherein the processor constructs an empirical model to generate estimated sensor values according to a kernel regression modeling method.

20. The monitoring apparatus of claim 16, further comprising a visual interface to graphically display a representation of the physical location of components of the targeted component and indicate residual values at the physical locations of sensors that correspond to the residual values and on the representation.

21. The monitoring apparatus of claim 20 wherein the sensors comprise tube-bundle thermocouples.

22. The monitoring apparatus of claim 20 wherein the representation is a grid with columns and rows, and wherein squares formed by the grid represent a bundle of tubes.

23. The monitoring apparatus of claim 16, wherein the processor adapts by constructing an empirical model that implements an adaptation algorithm to learn new normal variation patterns in operation of the system.

24. The monitoring apparatus of claim 23, wherein the adaptation algorithm utilizes at least one of:

manual (user-driven), trailing, out-of-range, in-range, and control-variable driven adaptation algorithms.

25. The monitoring apparatus of claim 16, wherein the processor constricts an empirical model that is updated to form a subset of the example observations and in real-time with each new input observation localized within a learned reference library to those example observations that are relevant to the input observation, according to predetermined relevance criteria.

26. The apparatus of claim 16, wherein the predetermined time period is set to at least the amount of time the system being monitored can still operate with a known fault.

27. A method for characterizing tube leak faults in a monitored system, comprising:

collecting historical sensor data for a targeted component of the monitored system;

producing residual signals that are correlative of given tube leak fault types in the targeted component;

analyzing the residual signals to generate residual signal signatures according to at least one predetermined analysis algorithm;

collecting the residual signal signatures in a database of time-varying measurements;

analyzing, by a processor, the database of residual signal signatures according to at least one predetermined classification algorithm to characterize salient features of the residual signal signatures relating to the given tube leak fault types, and

determining the location of a tube leak at a particular tube by providing, on a visual interface, a graphical display of a representation of the physical location of tubes within the targeted component and indicating residual values at a representation of the physical location of sensors corresponding to the residual values and on the display, the sensors being disposed and arranged to measure the temperature at a plurality of locations near different tubes located within the targeted component.

28. The method of claim 27, wherein the at least one predetermined analysis algorithm comprises at least one of residual threshold alerting, window ratio rule, sequential probability ratio test, and run-of-signs algorithms.

29. The method of claim 27, wherein the at least one predetermined classification algorithm comprises at least one of K-means, LVQ neural network, and SBM classification algorithms.

30. The method of claim 27 wherein the sensors comprise tube-bundle thermocouples.

31. The method of claim 27 wherein the representation is a grid with columns and rows, and wherein squares formed by the grid represent a bundle of tubes.

32. A method of monitoring a system, comprising:

collecting historical sensor data for a component of the monitored system;

generating estimate values by using the historical sensor data;

producing residual values that indicate a difference between the estimate values and current input values;

analyzing the residual values to determine if a particular type of fault exists; and

determining the location of the part on the system that caused the fault by providing a graphical display, on a visual interface, of a representation of the physical location of the parts of the system being monitored, and a representation of residual values at a moment in time and on the graphical display, each residual value being represented at the physical location of a sensor on the display and corresponding to the residual value.

33. The method of claim 32 wherein the system is a steam generating system, and wherein the parts are tubes within a system component, and wherein the sensors are disposed and arranged to measure the temperature at a plurality of locations near different tubes located within the system component.

34. The method of claim 33 wherein the representation is a grid with columns and rows, and each square formed by the grid represents a bundle of tubes.

35. The method of claim 32 wherein the location with a sensor changes in color to indicate the amount of the residual value at that location.

36. The method of claim 32 wherein the residual values are generated by using similarity based modeling wherein the estimate values are generated by using a calculation that uses both the historical sensor data and the current input values.