Artificial neural network and fuzzy logic based boiler tube leak detection systems

• US 6,192,352 B1
• Filed: 02/20/1998
• Issued: 02/20/2001
• Est. Priority Date: 02/20/1998
• Status: Expired due to Fees

• Current Assignee: Tennessee Valley Authority (Government of the United States of America), Tennessee Technological University
• Original Assignee: Tennessee Valley Authority (Government of the United States of America), Tennessee Technological University
• Inventors: Alouani, Ali Tahar, Chang, Peter Shih-Yung
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): DAVIS, GEORGE B
• Application Number:
  US09/026,822
• Time in Patent Office: 1,096 Days
• Field of Search: 706/8, 706/2, 706/902, 706/907
• US Class Current: 706/8
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G06N 3/043 based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
• CPC Class Codes: Y10S 706/902 Application using ai with detail of the ai system
• CPC Class Codes: Y10S 706/907 Power plant

Abstract

Power industry boiler tube failures are a major cause of utility forced outages in the United States, with approximately 41,000 tube failures occurring every year at a cost of $5 billion a year. Accordingly, early tube leak detection and isolation is highly desirable. Early detection allows scheduling of a repair rather than suffering a forced outage, and significantly increases the chance of preventing damage to adjacent tubes. The instant detection scheme starts with identification of boiler tube leak process variables which are divided into universal sensitive variables, local leak sensitive variables, group leak sensitive variables, and subgroup leak sensitive variables, and which may be automatically be obtained using a data driven approach and a leak sensitivity function. One embodiment uses artificial neural networks (ANN) to learn the map between appropriate leak sensitive variables and the leak behavior. The second design philosophy integrates ANNs with approximate reasoning using fuzzy logic and fuzzy sets. In the second design, ANNs are used for learning, while approximate reasoning and inference engines are used for decision making. Advantages include use of already monitored process variables, no additional hardware and/or maintenance requirements, systematic processing does not require an expert system and/or a skilled operator, and the systems are portable and can be easily tailored for use on a variety of different boilers.

Claims

1. A process for determining the occurrence and location of a boiler tube leak event in industrial/utility boilers, said process comprising:

(a) determining for a boiler, a set of tube universal leak sensitive variables ULSVs;

(b) calibrating the relationship between teaching data consisting of a plurality of known output patterns for each of said ULSVs in said set determined in step (a), and learning sample data consisting of a plurality of sample patterns obtained from actual boiler tube leak events, or simulated boiler tube leak events, or both, by training a universal leak detection system (ULDS) Artificial Neural Network (ANN) by supplying said learning data thereto and comparing said learning data with the corresponding teaching data patterns to thereby achieve a predetermined degree of convergence towards maximum pattern learning recognition;

(c) determining for said boiler, a set of tube local leak sensitive variables LLSVs;

(d) for each of said LLSVs determined in step (c), calibrating the relationship between teaching data consisting of a plurality of known output patterns for each individual variable of said LLSVs and learning sample data consisting of a plurality of sample patterns obtained from actual boiler tube leak events, or simulated boiler tube leak events, or both, by training, for each such LLSV, a corresponding a local leak detection system (LLDS) ANN forming a part of a group of LLDS ANNs by supplying said learning data thereto and comparing same with the corresponding teaching data patterns to thereby achieve a predetermined degree of convergence towards maximum pattern learning recognition;

(e) thereafter monitoring said industrial/utility boilers for the occurrence of a leak event by observing changes in values in a given boiler for each of said ULSVs and supplying said observed changes in values to said ULDS ANN for calculating a possibility that a leak event is occurring in the given boiler;

(f) comparing the possibility calculated in step (e) to a predetermined confidence threshold;

(g) if the possibility compared in step (f) is greater than said confidence threshold, concluding that a leak event is occurring and thereafter observing changes in values from said boiler for each of said LLSVs and supplying said observed changes in values to each of said corresponding local leak detection system (LLDS) ANNs for simultaneously calculating a possibility that the leak event is occurring at the location corresponding to one of a subsystem of the boiler; and

(h) comparing each possibility calculated by each LLDS ANN in step (g), one with the other, and concluding from said comparisons the location, in said boiler, of said leak event.

2. The process of claim 1, wherein detecting the occurrence and location of said leak event is effected during at least one development stage thereof, wherein an acoustical noise attributable thereto is not significantly greater in the immediate vicinity thereof than is a background acoustical noise attributable to operation of said boiler.

3. The process of claim 1, wherein the determination of said sets of tube ULSVs and tube LLSVs and the training of said ANNs in steps (a)-(d) thereof is effected at a time substantially different from the time during which steps (e)-(h) are effected.

4. The process of claim 3, wherein the time during which steps (e)-(h) are effected, is at least 24 hours subsequent to the time during which said steps (a)-(d) are effected.

5. The process of claim 1, wherein the comparisons made in step (h) thereof are effected with an inference engine.

6. The process of claim 1, wherein step (g) thereof there are provided eleven LLDS ANNs, each of which corresponds to a location in the boiler wherein it has been predetermined that a leak event is likely to occur, said locations comprising:

7. The process of claim 6, wherein for each of said eleven locations and associated LLSV, the arrangement of the inputs of the corresponding LLSVs and location numbers (L#) comprise:

8. The process of claim 7, wherein twenty LLSVs are utilized and comprise:

9. A process for identifying tube leak sensitive variables (LSV) requisite for later determination of the occurrence, the location, or both, of a boiler tube leak event in an industrial/utility boiler, said process comprising:

(a) collecting changes in values associated with a monitoring of said variables during operation of said boiler;

(b) calculating a sensitivity function of each particular variable for which changes in values are collected in step (a);

(c) comparing each said sensitivity function calculated in step (b), with a predetermined sensitivity threshold, eliminating those variables whose sensitivity function is less than said threshold and collecting those variables whose sensitivity function is greater than said threshold;

(d) calculating a possibility that one of any of the variables collected in step (c), contains information redundant with information contained in any other of said collected variables;

(e) eliminating a given variable, if the possibility that such variable contains redundant information;

(f) collecting those variables determined in step (e), to contain information not redundant with information contained in any other variable;

(g) comparing a change in values for each of the variables collected in step (f) with standard principles of thermodynamics and mechanics and eliminating those variables whose changes in values do not correlate; and

(h) collecting for later determination of the occurrence, the location, or both, of said boiler tube leak event, those variables which correlate with said standard principles in step (g), supra, as tube LSVs.

10. The process of claim 9, wherein calculating the sensitivity of each particular variable in step (b) thereof quantifies for each, a sensitivity function S(v_i) wherein:

11. The process of claim 9, wherein calculating the sensitivity of each particular variable in step (b) thereof quantifies for each, a sensitivity function S(v_i) wherein:

12. The process of claim 9, wherein the LSV identified is a universal leak sensitive variable (ULSV).

13. The process of claim 12, wherein about three separate ULSVs are identified.

14. The process of claim 9, wherein the LSV identified is a local leak sensitive variable (LLSV).

15. The process of claim 14, wherein about twenty separate LLSVs are identified.

16. The process of claim 9, wherein the LSV identified is a group leak sensitive variable (GLSV).

17. The process of claim 16, wherein about eleven separate GLSVs are identified.

18. The process of claim 9, wherein the LSV identified is a subgroup leak sensitive variable (SGLSV).

19. The process of claim 18, wherein about eleven separate SGLSVs are identified.

20. The process of claim 9, wherein the LSV identified is selected from the group consisting of ULSV, LLSV, and mixtures thereof.

21. A process for determining the occurrence and location of a boiler tube leak event in an industrial/utility boiler, said process comprising:

(a) determining for said boiler, a set of tube group leak sensitive variables GLSVs;

(b) arranging said set of GLSVs into individual groups;

(c) for each of said set of GLSVs arranged in step (b), calibrating the relationship between teaching data consisting of a plurality of known output patterns of said obtained from actual boiler tube leak events, simulated boiler tube leak events, or both, by training for each such group of GLSVs, a corresponding Artificial Neural Network ANN by supplying the training data thereto and comparing same with the corresponding teaching data patterns to thereby achieve a predetermined degree of convergence towards maximum pattern recognition;

(d) thereafter monitoring said industrial/utility boiler for the occurrence of a leak event by observing changes in values in said boiler for each of said set of GLSVs and supplying said observed changes in values to each of said corresponding ANNs for calculating a possibility that a leak event is occurring;

(e) comparing the possibility calculated in step (d), to a predetermined confidence threshold;

(f) concluding, if the possibility compared in step (e), is greater than said confidence threshold, that a leak event is occurring and thereafter observing changes in values from said boiler for each GLSV in each said set and supplying said observed changes in values to each of said corresponding ANNs for simultaneously calculating a possibility that the leak event is occurring at the location corresponding to one of said GLSVs in one of said set;

(g) comparing each possibility calculated by each ANN in step (f), one with the other, and concluding from said comparison the location in said boiler of the group in which said leak event is occurring;

(h) determining for said boiler a set of tube subgroup leak sensitive variables SGLSV;

(i) arranging said SGLSVs into individual subgroups to form subgroups of SGLSVs;

(j) for each of said subgroups of SGLSVs arranged in step (i), calibrating the relationship between teaching data consisting of a plurality of known output patterns of said subgroup and learning sample data consisting of a plurality of sample patterns obtained from actual boiler tube leak events, simulated boiler tube leak events, or both, by training for each such SGLSV, a corresponding Artificial Neural Network ANN by supplying said learning data thereto and comparing same with the corresponding teaching data patterns to thereby achieve a predetermined degree of convergence towards maximum pattern recognition;

(k) observing changes in values in said boiler for each SGLSV associated with the subgroup identified in step (g), and supplying said observed changes in values to said SGLSV corresponding ANN trained in step (j), for calculating a possibility that the leak event is occurring at the location corresponding to one of said SGLSVs; and

(l) comparing each possibility calculated by the SGLSV ANN in step (k), one with the other, and concluding from said comparison the location in said boiler of said leak event.

22. The process of claim 21, wherein the occurrence and location of said leak event is effected during at least one development stage thereof wherein an acoustical noise attributable thereto is not significantly greater in an immediate vicinity thereof than is a background acoustical noise attributable to operation of said boiler.

23. The process of claim 22, wherein step (f) thereof are provided four group leak detection system (GLDS) ANNs, each of which said GLDS ANNs determines the likelihood of a leak event in each group in the boiler wherein a leak event is likely to occur, said locations comprising an economizer, a waterwall, a superheater, and a reheater.

24. The process of claim 23, wherein for each of said GLDS ANNs, the arrangement of corresponding GLSVs and associated variables comprise:

25. The process of claim 24, wherein eleven GLSVs are identified and comprise:

26. The process of claim 21, wherein the determination of said set of tube GLSVs and the training of said ANNs in steps (a)-(c) and steps (h) and (j) thereof is effected at a time substantially different from the time during which steps (d)-(g) and steps (k) and (l) are effected.

27. The process of claim 26, wherein said steps (d)-(g) and (k) and (l) are effected at least 24 hours subsequent to the time wherein steps (a)-(c) and steps (h) and j) are effected.

28. The process of claim 27, wherein the occurrence and location of said tube leak event is effected during at least one development stage thereof, wherein an acoustical noise attributable thereto is not significantly greater in an immediate vicinity thereof than is a background acoustical noise attributable to operation of said boiler.

29. The process of claim 27, wherein the conclusions made in steps (h) and (k) thereof are effected with an inference engine.

30. A system for identifying tube leak sensitive variables LSVs requisite for later determination of the occurrence, the location, or both, of a boiler tube leak event in an industrial/utility boiler, said system comprising:

(a) first information collection means for storing changes in values associated with a monitoring of said LSVs during operation of said boiler;

(b) first calculating means for determining a sensitivity function of each particular LSV for which changes in values are assembled in said first collection means;

(c) first comparing means for matching the sensitivity function calculated in said first calculating means with a predetermined sensitivity function threshold and identifying those variables whose sensitivity function is greater than said threshold function;

(d) second information collection means for assembly of those LSVs identified in said first comparing means as having a sensitivity function greater than said threshold function;

(e) second calculating means for determining the possibility that any one of the LSVs assembled in said second information collection means is redundant with information contained in any other of said LSVs assembled therein;

(f) third information collection means for assembly of those LSVs determined in said second calculating means to contain information not redundant with information contained in any other of such LSVs;

(g) second comparing means for matching observed changes in values for each of the LSVs collected in said third information collection means with standard principles of thermodynamics and mechanics; and

(h) fourth information collection means for assembly of the resulting tube LSVs identified in said second comparing means as correlating with said standard principles.

31. The system of claim 30, wherein said sensitivity function determined in said first calculating means for each particular LSV is quantified as S(v_i) wherein:

32. The system of claim 30, wherein said sensitivity function determined in said first calculating means for each particular LSV is quantified as S(v_i) wherein:

33. A system for determining the occurrence and location of a boiler tube leak industrial/utility boiler, said system comprising:

(a) tube universal leak detection system (ULDS) means for determining a likelihood of an occurrence of a tube leak event, said ULDS means operatively associated with inputs of observed changes in said industrial/utility boiler of universal leak sensitive variables (ULSV), and comprising a first Artificial Neural Network (ANN) trained on a desired convergence between ULSV teaching data, and ULSV learning data;

(b) a plurality of local leak detection system (LLDS) means, each of which is operatively associated with inputs of one of a plurality of sets of observed changes in said industrial/utility boiler of local leak sensitive variables (LLSV) each of said LLDS means comprising a corresponding second ANN trained on the desired convergence between LLSV teaching data and LLSV learning data; and

(c) inference engine means for receiving an output from each of said plurality of sets of LLDS means and for determining the location in the boiler of said leak event.

34. The system of claim 33, wherein said ULSV teaching data consists of a plurality of known output patterns of a plurality of ULSVs and said ULSV learning data consists of a plurality of sample patterns obtained from actual boiler tube leak events, simulated boiler tube leak events, or both, and wherein said LLSV teaching data consists of a plurality of known output patterns of individual LLSVs and said LLSV learning data consists of a plurality of sample patterns obtained from actual boiler tube leak events, simulated boiler tube leak events, or both.

35. The system of claim 34, wherein there are provided at least one first ANN and about eleven of said corresponding second ANN.

36. The system of claim 35, wherein there are provided about three inputs of said ULSVs to said first ANN.

37. The system of claim 36, wherein three inputs of said ULSVs comprise:

38. The system of claim 35, wherein there are provided about twenty inputs of said LLSVs to eleven corresponding second ANNs.

39. The system of claim 38, wherein said about twenty inputs of said LLSVs comprise:

40. The system of claim 39, wherein said at least first ANN and eleven of said corresponding second ANNs each are comprised of one input layer provided with about three neurons, a first hidden layer provided with about forty neurons, a second hidden layer provided with about twenty-four neurons, and an output layout provided with at least one neuron.

41. A system for determining the occurrence and location of a boiler tube leak industrial/utility boiler, said system comprising:

(a) a first plurality of first tube group leak detection system (GLDS) means for determining the likelihood of the occurrence of a tube leak event, each of said GLDS means operatively associated with inputs of observed changes in said industrial/utility boiler of group leak sensitive variables (GLSV) and comprising a plurality of corresponding first Artificial Neural Networks (ANN) trained on a desired convergence between GLSV teaching data, and GLSV learning data, and operatively associated with inputs of observed changes in said industrial/utility boiler of said GLSVs;

(b) first inference engine means for receiving an output from each of said plurality;

of GLDS means and for determining the group in which the boiler leak event is occurring;

(c) a plurality of tube subgroup leak detection system (SGLDS) means for determining a likelihood in the group identified by said first inference engine means of the location of said leak event, each of which is operatively;

associated with inputs of observed changes in said industrial/utility boiler of subgroup leak sensitive variables (SGLSV) each of said SGLDS means comprising a corresponding second ANN trained on the desired convergence between SGLSV teaching data, and SGLSV learning data; and

(d) second inference engine means for receiving an output from each of said plurality of SGLDS means and for determining the location in the boiler of said leak event.

42. The system of claim 41, wherein said GLSV teaching data consists of a plurality of known output patterns of a plurality of GLSVs and said GLSV learning data consists of a plurality of sample patterns obtained from actual boiler tube leak events, simulated boiler tube leak events, or both, and wherein said SGLSV teaching data consists of a plurality of known output patterns of individual SGLSVs and said SGLSV learning data consists of a plurality of sample patterns obtained from actual boiler tube leak events, simulated boiler tube leak events, or both.

43. The system of claim 42, wherein there are provided about four corresponding first ANNs and about six corresponding second ANNs.

44. The system of claim 43, wherein there are provided to each of said corresponding first ANNs about four inputs of said GLSVs.

45. The system of claim 44, wherein an arrangement of four inputs to said plurality of corresponding first ANNs associated with each of four groups of GLSVs is:

46. The system of claim 45, wherein eleven GLSVs are identified and comprise:

47. The system of claim 43, wherein there are provided to each of said corresponding second ANNs about six inputs of said SGLSVs.