System and method for abnormal event detection in the operation of continuous industrial processes

• US 20060058898A1
• Filed: 08/26/2005
• Published: 03/16/2006
• Est. Priority Date: 09/10/2004
• Status: Active Grant

• Current Assignee: ExxonMobil Research & Engineering Company (Exxon Mobil Corporation)
• Original Assignee: ExxonMobil Research & Engineering Company (Exxon Mobil Corporation)
• Inventors: Wang, Robert K., Kendi, Thomas A., Woo, Stephen S., Emigholz, Kenneth F., McLain, Richard B., Dash, Sourabh K.
• Granted Patent: US 7,349,746 B2
• Time in Patent Office: Days
• US Class Current: 700/29
• CPC Class Codes: C10G 11/187 Controlling or regulating
• CPC Class Codes: G05B 23/0208 characterized by the configuration of the monitoring system
• CPC Class Codes: G05B 23/0243 model based detection method, e.g. first-principles knowledge model
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis

Abstract

Thousands of process and equipment measurements are gathered by the modern digital process control systems that are deployed in refineries and chemical plants. Several years of these data are historized in databases for analysis and reporting. These databases can be mined for the data patterns that occur during normal operation and those patterns used to determine when the process is behaving abnormally. These normal operating patterns are represented by sets of models. These models include simple engineering equations, which express known relationships that should be true during normal operations and multivariate statistical models based on a variation of principle component analysis. Equipment and process problems can be detected by comparing the data gathered on a minute by minute basis to predictions from these models of normal operation. The deviation between the expected pattern in the process operating data and the actual data pattern are interpreted by fuzzy Petri nets to determine the normality of the process operations. This is then used to help the operator localize and diagnose the root cause of the problem.

Claims

1. A system for abnormal event detection in an industrial process comprising a set of models to detect abnormal operations in said process wherein said process has been divided into equipment groups and process operating modes with process measurements as imputs to said models.

2. The system of claim 1 wherein said equipment groups and process operating modes are described by a principle component analysis model.

3. The system of claim 1 wherein said equipment groups include all major material and energy interactions in the same group.

4. The system of claim 1 wherein said equipment groups include quick recycles in the same group.

5. The system of claim 2 wherein the equipment groups have minimal interaction among them.

6. The system of claim 5 wherein said equipment groups include feed forward and feed back paths between upstream and downstream equipment groups.

7. The system of claim 1 wherein the operating modes are specific time periods wherein process behavior is significantly different.

8. The system of claim 1 wherein inputs to the system are measurements that measure the performance and/or physical state of the system of the process.

9. The system of claim 8 wherein the inputs determine the physical state of the system.

10. The system of claim 9 wherein said inputs to the system include temperatures, and/or pressures, and/or flows.

11. The system of claim 8 wherein said input measurements have a signal to noise ratio above a predetermined value.

12. The system of claim 1 wherein models outputs are combined using fuzzy Petri nets.

13. The system of claim 9 wherein the signal to noise ratio is above 4.

14. The system of claim 8 wherein said input measurements satisfy a cross correlation test.

15. The system of claim 2 wherein said inputs to said principle component analysis model are scaled.

16. The system of claim 10 further comprising an operator interface.

17. The system of claim 15 wherein said operator interface includes Pareto charts.

18. The system of claim 15 wherein said operator interface includes an indication of the model that gives the most abnormal indication.

19. The system of claim 15 wherein said operator interface includes a continuous indication of normality of major process areas.

20. The system of claim 15 wherein the fuzzy net includes a normality trend that indicates the greatest likelihood of abnormality.

21. The system of claim 2 further comprising engineering models.

22. The system of claim 2 further comprising inferential measurements.

23. The system of claim 19 wherein said engineering models are redundancy checks based on known engineering relationships.

24. The system of claim 21 wherein said redundancy checks are 2×

2 checks.

25. The system of claim 21 wherein said redundancy checks are multiple redundancy checks.

26. The system of claim 1 wherein said set of models has been restructured based on process operations.

27. The system of claim 10 wherein said output from said fuzzy Petri nets is a normality trend.

28. The system of claim 1 wherein said industrial process is a chemical process.

29. The system of claim 1 wherein said industrial process is a petrochemical process.

30. The system of claim 1 wherein said industrial process is a refining process.

31. A method to developing an abnormal event detector for a industrial process comprising

a) dividing the process into equipment groups and/or operating modes,

b) determining input variables and their operating range for said equipment and/or operating modes,

c) determining historical data for said input variables.

d) determining a historical data training set having no abnormal operation,

e) creating a set of models, one for each of said equipment groups and/or operating modes using said historical data training set.

32. The method of claim 31 further comprising creating engineering models based on known engineering relationships and combining with the set of models for said equipment groups and/or operating mode.

33. The method of claim 31 wherein said models are recalibrated with a new historical data training set.

34. The method of claim 31 wherein said set of models are principle component analysis models.

35. The method of claim 34 further comprising the step of combining output from the models using fuzzy Petri nets.

36. The method of claim 31 further comprising the step of displaying output from the set of models as a normality trend.

37. The method of claim 31 further comprising the step of using fuzzy Petri nets to interpret output from said set of models as normal or abnormal.

38. The method of claim 31 further comprising the step of using fuzzy Petri nets to combine model outputs into a single continuous summary of normality and abnormality.

39. The method of claim 34 further comprising the step of supplementing the principle components analysis models with known engineering relationships.

40. The method of claim 39 wherein said known engineering relationships are 2×

2 redundancy checks.

41. The method of claim 39 wherein said known engineering relationships are multidimensional redundancy checks.

42. The method of claim 39 further comprising the step of using fuzzy Petri nets to combine the set of principle components analysis models with known engineering relationships.

43. The method of claim 31 wherein said industrial process is a chemical process.

44. The method of claim 31 wherein said industrial process is a petrochemical process.

45. The method of claim 31 wherein said industrial process is a refining process.

46. A method for determining an abnormal event for an industrial process comprising the steps of determining if output from a set of models exceeds a predetermined value wherein said set of models include models corresponding to equipment groups and/or operating modes and/or engineering models.

47. The method of claim 46 wherein output from the set of models is combined using fuzzy Petri nets to provide a continuous summary indication of the normality of the process.

48. The method of claim 47 further comprising the step of creating Pareto chart for an equipment group, an operating mode, or engineering model.

49. The method of claim 46 wherein said industrial process is a chemical process.

50. The method of claim 46 wherein said industrial process is a petrochemical process.

51. The method of claim 46 wherein said industrial process is a refining process.