Paper web breakage prediction using principal components analysis and classification and regression trees

• US 6,522,978 B1
• Filed: 08/13/2001
• Issued: 02/18/2003
• Est. Priority Date: 09/15/1999
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Bonissone, Piero Patrone, Chen, Yu-To
• Primary Examiner(s): Hoff, Marc S.
• Assistant Examiner(s): Raymond, Edward
• Application Number:
  US09/929,534
• Time in Patent Office: 554 Days
• Field of Search: 702/35, 702/36, 702/85, 702/115, 702/182, 702/183, 702/184, 162/263, 735/98, 736/06, 736/49, 737/67, 737/94, 738/658, 341/17, 344/45, 700/78, 700/79, 700/80, 700/83
• US Class Current: 702/35
• CPC Class Codes: G01N 33/346 Paper sheets

Abstract

A system and method for predicting web breaks in a paper machine. Principal components analysis (PCA) and classification and regression tree (CART) modeling are used to predict web break sensitivity from measurements taken from the paper mill. Also, the CART model is used to isolate the root cause of the predicted web break sensitivity.

Claims

1. A system for predicting a web break in a paper machine located about a paper mill, comprising:

a paper mill database containing a plurality of measurements obtained from the paper mill, each of the plurality of measurements relating to a predetermined paper machine variable;

a processor for processing each of the plurality of measurements into break sensitivity data; and

a break predictor responsive to the processor for predicting a web break sensitivity within the paper machine from the plurality of processed measurements.

2. The system according to claim 1, wherein the break predictor comprises an inductive reasoning tool.

3. The system according to claim 1, wherein the break predictor comprises a classification and regression tree model.

4. The system according to claim 3, wherein the classification and regression tree model is trained with historical web break data.

5. The system according to claim 1, wherein the break sensitivity data comprise time-based transformations of the plurality of measurements.

6. The system according to claim 1, wherein the break sensitivity data comprise principal components in determining web breakage from the plurality of measurements.

7. The system according to claim 5, wherein the break sensitivity data further comprise a principal component first derivative value, a principal component second derivative value and a principal component difference from steady-state value.

8. The system according to claim 1, further comprising a fault isolator responsive to the break predictor for determining the paper machine variables affecting the predicted web break sensitivity.

9. The system according to claim 8, wherein the fault isolator comprises a classification and regression tree model having a set of rules linking paper machine variables to the predicted web break sensitivity.

10. The system according to claim 8, wherein the fault isolator identifies the paper machine variables that are root causes for the predicted web break sensitivity.

11. The system according to claim 1, wherein the plurality of measurements contained in the paper mill database are generated from various processes occurring within the paper mill.

12. The system according to claim 1, wherein the paper mill database comprises a raw materials database, a preprocess database, a paper machine database, an operation shift database and a maintenance schedule database.

13. A system for predicting a web break in a paper machine located about a paper mill, comprising:

a paper mill database containing a plurality of measurements obtained from the paper mill, each of the plurality of measurements relating to a predetermined paper machine variable;

a processor for processing each of the plurality of measurements into break sensitivity data; and

a break predictor responsive to the processor for predicting a web break sensitivity within the paper machine from the plurality of processed measurements, wherein the break predictor comprises an inductive reasoning tool.

14. The system according to claim 13, wherein the break predictor comprises a classification and regression tree model.

15. The system according to claim 14, wherein the classification and regression tree model is trained with historical web break data.

16. The system according to claim 14, wherein the break sensitivity data comprise time-based transformations of the plurality of measurements.

17. The system according to claim 16, wherein the break sensitivity data comprise principal components in determining web breakage from the plurality of measurements.

18. The system according to claim 17, wherein the break sensitivity data further comprise a principal component first derivative value, a principal component second derivative value and a principal component difference from steady-state value.

19. The system according to claim 18, further comprising a fault isolator responsive to the break predictor for determining the paper machine variables affecting the predicted web break sensitivity.

20. The system according to claim 19, wherein the fault isolator comprises a classification and regression tree model having a set of rules linking paper machine variables to the predicted web break sensitivity.

21. The system according to claim 20, wherein the fault isolator identifies the paper machine variables that are root causes for the predicted web break sensitivity.

22. The system according to claim 13, wherein the plurality of measurements contained in the paper mill database are generated from various processes occurring within the paper mill.

23. The system according to claim 13, wherein the paper mill database comprises a raw materials database, a preprocess database, a paper machine database, an operation shift database and a maintenance schedule database.

24. A method for predicting a web break in a paper machine located about a paper mill, comprising:

obtaining a plurality of measurements from the paper mill, each of the plurality of measurements relating to a predetermined paper machine variable;

processing each of the plurality of measurements into break sensitivity data; and

predicting a web break sensitivity within the paper machine from the plurality of processed measurements.

25. The method according to claim 24, wherein processing each of the plurality of measurements comprises time-based transformations of the plurality of measurements.

26. The method according to claim 24, wherein processing each of the plurality of measurements comprises generating principal components in determining web breakage from the plurality of measurements.

27. The method according to claim 25, wherein processing each of the plurality of measurements comprises generating a principal component first derivative value, a principal component second derivative value and a principal component difference from steady-state value.

28. The method according to claim 24, wherein predicting the web break sensitivity comprises processing the break sensitivity data using an inductive reasoning tool.

29. The method according to claim 24, wherein predicting the web break sensitivity comprises processing the break sensitivity data using a classification and regression tree model.

30. The method according to claim 29, further comprising training the classification and regression tree model with historical web break data to learn how to predict web break sensitivity.

31. The method according to claim 30, further comprising testing the trained classification and regression tree model with the historical break data to test how well the model predicts web break sensitivity.

32. The method according to claim 30, wherein the training comprises preprocessing the historical web break data.

33. The method according to claim 32, wherein the preprocessing comprises:

reducing the quantity of the historical web break data;

reducing the number of variables contained in the historical web break data;

transforming the values of the historical web break data;

extracting features that affect web break sensitivity from the historical web break data; and

generating the classification and regression tree model to predict a web break sensitivity from the extracted features.

34. The method according to claim 33, wherein reducing the quantity of historical web break data includes selecting data associated with a web break having a predetermined cause.

35. The method according to claim 34, wherein reducing the quantity of historical web break data includes selecting data within a predetermined time period of a web break.

36. The method according to claim 34, further comprising segmenting the historical web break data.

37. The method according to claim 36, wherein segmenting the data includes selecting data associated with a predetermined paper grade.

38. The method according to claim 36, wherein segmenting the historical web break data includes dividing the data into break positive data and break negative data.

39. The method according to claim 38, wherein dividing the data into break positive data includes segmenting the break positive data by time-series analysis.

40. The method according to claim 33, wherein reducing the number of variables includes processing the data utilizing a technique selected from the group consisting of knowledge engineering, visualization, classification and regression trees and logistic regression.

41. The method according to claim 33, wherein reducing the number of variables includes processing the data utilizing principal components analysis.

42. The method according to claim 33, wherein transforming the values of the historical web break data includes smoothing or filtering the data.

43. The method according to claim 24, further comprising the step of isolating the paper machine variables affecting the predicted web break sensitivity.

44. The method according to claim 24, wherein the obtaining of the plurality of measurements comprises receiving measurements generated from various processes occurring within the paper mill.

45. A method for predicting a web break in a paper machine located about a paper mill, comprising:

obtaining a plurality of measurements from the paper mill, each of the plurality of measurements relating to a predetermined paper machine variable;

processing each of the plurality of measurements into break sensitivity data comprising time-based transformations of the plurality of measurements; and

predicting a web break sensitivity within the paper machine from the plurality of processed measurements.

46. The method according to claim 45, wherein processing each of the plurality of measurements comprises generating principal components in determining web breakage from the plurality of measurements.

47. The method according to claim 46, wherein processing each of the plurality of measurements comprises generating a principal component first derivative value, a principal component second derivative value and a principal component difference from steady-state value.

48. The method according to claim 47, wherein predicting the web break sensitivity comprises processing the break sensitivity data using an inductive reasoning tool.

49. The method according to claim 48, wherein predicting the web break sensitivity comprises processing the break sensitivity data using a classification and regression tree model.

50. The method according to claim 49, further comprising training the classification and regression tree model with historical web break data to learn how to predict web break sensitivity.

51. The method according to claim 50, further comprising testing the trained classification and regression tree model with the historical break data to test how well the model predicts web break sensitivity.

52. The method according to claim 51, wherein the training comprises preprocessing the historical web break data.

53. The method according to claim 52, wherein the preprocessing comprises:

reducing the quantity of the historical web break data;

reducing the number of variables contained in the historical web break data;

transforming the values of the historical web break data;

extracting features that affect web break sensitivity from the historical web break data; and

generating the classification and regression tree model to predict a web break sensitivity from the extracted features.

54. The method according to claim 45, wherein the obtaining of the plurality of measurements comprises receiving measurements generated from various processes occurring within the paper mill.