Apparatus and method for batch property estimation

• US 7,242,989 B2
• Filed: 05/30/2003
• Issued: 07/10/2007
• Est. Priority Date: 05/30/2003
• Status: Active Grant

• Current Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Original Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Inventors: Blevins, Terrence L., Mehta, Ashish
• Primary Examiner(s): PATEL, RAMESH B
• Application Number:
  US10/449,437
• Publication Number:
  US 20040243380A1
• Time in Patent Office: 1,502 Days
• Field of Search: 700/29, 700/30, 700/31, 700/44, 700/45, 700/46, 700/47, 700/48, 700/52, 700/265, 700/266, 700/268, 700269-272, 700/32, 706/12, 706 15- 17, 706/21, 706/25, 422/3, 422/105, 422/109, 422/110, 422/129, 703/2
• US Class Current: 700/47
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 17/02 electric

Abstract

A method and apparatus that generates an estimate of a property of a batch process uses a non-parametric model to generate a plurality of rate of reaction estimates associated with the batch process. Each rate of reaction estimate may correspond, for example, to a particular time during the batch process. The plurality of rate of reaction estimates are then integrated to generate an estimate of a property of the batch at the particular time.

Claims

1. A method for estimating a property of a batch process, the method comprising:

determining rate of reaction estimates associated with the batch process using a non-parametric model; and

determining an estimate of a property of the batch process at a particular time based on the rate of reaction estimates.

2. A method as defined in claim 1, wherein determining the estimate of the property comprises integrating the rate of reaction estimates.

3. A method as defined in claim 2, wherein integrating the rate of reaction estimates comprises accumulating the rate of reaction estimates.

4. A method as defined in claim 1, wherein the non-parametric model comprises a neural network.

5. A method as defined in claim 1, further comprising determining an estimate of a difference between the property of the batch process at the particular time and a property of a model batch process at a time during the model batch process that corresponds to the particular time.

6. A method as defined in claim 5, wherein determining the estimate of the difference between the property of the batch process and the property of the model batch process comprises:

determining differences between the rate of reaction estimates of the batch process and rate of reaction estimates of the model batch process; and

integrating the differences between the rate of reaction estimates of the batch process and the rate of reaction of the model batch process.

7. A method as defined in claim 5, wherein determining the estimate of the difference between the property of the batch process and the property of the model batch process comprises subtracting the estimate of the property of the batch process at the particular time from a property of the model batch process at the time during the model batch process that corresponds to the particular time.

8. A method as defined in claim 7, wherein the property of the model batch process at the time during the model batch process that corresponds to the particular time is an estimated property.

9. A method as defined in claim 7, wherein the property of the model batch process at the time during the model batch process that corresponds to the particular time is a measured property.

10. A method as defined in claim 5, further comprising determining a predicted property at a batch process ending time based on the estimate of the difference between the property of the batch process at the particular time and the property of the model batch process at the time during the model batch process that corresponds to the particular time.

11. A method as defined in claim 10, wherein determining the predicted property at the batch process ending time comprises adding a desired property at the batch process ending time with the estimate of the difference between the property of the batch process at the particular time and the property of the model batch process at the time during the model batch process that corresponds to the particular time.

12. A method as defined in claim 1, further comprising determining an estimate of a deviation between an actual time at which the property of the batch process reaches a desired value and an expected time at which the batch process reaches the desired value.

13. A method as defined in claim 12, wherein determining the estimate of the deviation comprises determining an estimate of a difference between the property of the batch process at the particular time and a property of a model batch process at a time during the model batch process that corresponds to the particular time.

14. A method as defined in claim 13, wherein determining the estimate of the deviation further comprises dividing the estimate of the difference between the property of the batch process and the property of a model batch process by an estimate of a rate of reaction of the model batch process.

15. A method as defined in claim 14, wherein the estimate of the rate of reaction of the model batch process is an estimate of a rate of reaction of the model batch process proximate to the end of the model batch process.

16. A method as defined in claim 14, wherein the estimate of the rate of reaction of the model batch process is an estimate of a rate of reaction of the model batch process at a time during the model batch process that corresponds to the particular time.

17. A method as defined in claim 12, further comprising determining an estimate of time remaining until the property of the batch process reaches a desired value based on the estimate of the deviation.

18. A method as defined in claim 17, wherein determining the estimate of time remaining comprises subtracting the estimate of the deviation and an elapsed time from the expected time.

19. A method as defined in claim 1, further comprising multiplying the rate of reaction estimates with a gain adjustment.

20. A method as defined in claim 1, further comprising adding an offset adjustment to the rate of reaction estimates.

21. A method as defined in claim 20, further comprising multiplying the rate of reaction estimates with a gain adjustment.

22. A method for training a non-parametric model for estimating a rate of reaction of a batch process, the method comprising:

initializing the non-parametric model;

generating a plurality of rate of reaction estimates with the non-parametric model using a training record corresponding to a batch process;

generating an estimate of an ending property of the batch process based on the plurality of rate of reaction estimates;

comparing the estimate of the ending property with a measured ending property of the batch process to generate an error; and

adjusting the non-parametric model based on the error.

23. A method as defined in claim 22, wherein the non-parametric model comprises a neural network having a plurality of weights;

wherein initializing the non-parametric model comprises initialzing the plurality of weights; and

wherein adjusting the non-parametric model comprises adjusting the plurality of weights.

24. A method as defined in claim 22, wherein generating the estimate of the ending property of the batch process comprises integrating the plurality of rate of reaction estimates.

25. A method as defined in claim 22, further comprising determining a gain adjustment value based on a plurality of estimates of the ending property corresponding to a plurality of batch processes and a plurality of measurements of the ending property corresponding to the plurality of batch processes, wherein the plurality of estimates of the ending property are generated by the non-parametric model.

26. A method as defined in claim 22, further comprising:

determining a plurality of errors based on a plurality of estimates of the ending property corresponding to a plurality of batch processes and a plurality of measurements of the ending property corresponding to the plurality of batch processes, wherein the plurality of estimates of the ending property are generated by the non-parametric model; and

determining an offset adjustment value based on the plurality of errors.

27. A batch property estimator, comprising:

a non-parametric model configured to receive data associated with a batch process and to generate a plurality of rate of reaction estimates related to the batch process; and

a first integrator, coupled the non-parametric model, to integrate the plurality of rate of reaction estimates to generate an estimate of a property of the batch process at a particular time.

28. A batch property estimator as defined in claim 27, further comprising a memory coupled to the first integrator to store to store an estimate of an ending property of the batch process.

29. A batch property estimator as defined in claim 27, further comprising a property difference estimate generator to generate an estimate of a difference of the property of the batch process at the particular time with a property of a model batch process at a time corresponding to the particular time.

30. A batch property estimator as defined in claim 29, wherein the property difference estimate generator comprises:

a model reaction rates store to generate rate of reaction estimates of a model batch process that correspond to the plurality of rate of reaction estimates;

a first adder coupled to the model reaction rate store and the non-parametric model to subtract the plurality of rate of reaction estimates from the rate of reaction estimates of the model batch process to generate a plurality of rate of reaction differences; and

a second integrator coupled to the adder to integrate the plurality of rate of reaction differences to generate an estimate of the difference of the property of the batch process at the particular time with the property of the model batch process at the time corresponding to the particular time.

31. A batch properly estimator as defined in claim 29, wherein the property difference estimate generator comprises:

a model batch property store to generate a property of a model batch process that corresponds to the estimate of the property of the batch process at the particular time; and

an adder coupled to the model batch property store and the first integrator to subtract the estimate of the property of the batch process at the particular time from the property of the model batch process.

32. A batch property estimator as defined in claim 29, further comprising a predicted ending property generator coupled to the property difference estimate generator to generate a prediction of the property of the batch process at an expected ending time based on the estimate of the difference of the property of the batch process at the particular time with the property of the model batch process at the time corresponding to the particular time.

33. A batch property estimator as defined in claim 32, wherein the predicted ending property generator comprises an adder to add the estimate of the difference of the property with a desired property of the batch process at the ending time.

34. A batch property estimator as defined in claim 29, further comprising a time deviation generator coupled to the property difference estimate generator to generate an estimate of a deviation between an actual time at which the property of the batch process reaches a desired value and an expected time at which the property of the batch process reaches the desired value based on the estimate of the difference of the property of the batch process at the particular time with the property of the model batch process at the time corresponding to the particular time.

35. A batch property estimator as defined in claim 34, wherein the time deviation generator comprises:

a model reaction rate store to generate a rate of reaction estimate of a model batch process; and

a divider coupled to the model reaction rate store and to the property difference estimate generator, to divide the model reaction rate store and to the property by the rate of reaction estimate of a model batch process to generate the estimate of the deviation.

36. A batch property estimator as defined in claim 35, wherein the rate of reaction estimate of the model batch process comprises an estimate of a rate of reaction of the model batch process at time corresponding to the particular time.

37. A batch property estimator as defined in claim 35, wherein the rate of reaction estimate of the model batch process comprises an estimate of a rate of reaction of the model batch process proximate to an end of the model batch process.

38. A batch property estimator as defined in claim 34, further comprising a time remaining estimate generator coupled to the time deviation generator to generate an estimate of time remaining until the property of the batch process reaches the desired value based on the estimate of the deviation.

39. A batch property estimator as defined in claim 38, wherein the time remaining estimate generator comprises an adder to subtract the estimate of the deviation and an elapsed time of the batch process from an expected ending time.

40. A batch property estimator as defined in claim 27, wherein the non-parametric model comprises a neural network.

41. A method for facilitating control of a batch process, the method comprising:

determining rate of reaction estimates associated with the batch process using a non-parametric model;

determining estimate data related to the batch process based on the rate of reaction estimates, the estimate data related to the batch process including at least one of an estimate of a property of the batch process at a particular time and an estimate of a difference between the property at the particular time and a property of a model batch process at a time during the model batch process that corresponds to the particular time; and

using the estimate data to facilitate control of the batch process.

42. A method as defined in claim 41, wherein using the estimate data comprises determining additional estimate data based on the estimate data.

43. A method as defined in claim 42, wherein the additional estimate data comprises a predicted property at an ending time.

44. A method as defined in claim 42, wherein the additional estimate data comprises an estimate of a deviation between an actual time at which the property reaches a desired value and an expected time at which the batch process reaches the desired value.

45. A method as defined in claim 42, wherein the additional estimate data comprises data associated with an estimate of a time at which the property reaches a desired value.

46. A method as defined in claim 42, wherein the additional estimate data comprises an estimate of time remaining until the property reaches a desired value.

47. A method as defined in claim 42, further comprising using the additional estimate data to facilitate control or the batch process.

48. A method as defined in claim 47, wherein using the additional estimate data comprises displaying at least some of the additional estimate data on a display device.

49. A method as defined in claim 47, wherein using the additional estimate data comprises providing at least some of the estimate data to a controller associated with control of the batch process.

50. A method as defined in claim 49, wherein using the estimate data further comprises using at least some of the estimate data to control the batch process via the controller.

51. A method as defined in claim 41, wherein using the estimate data comprises displaying at least some of the estimate data on a display device.

52. A method as defined in claim 41, wherein using the estimate data comprises providing at least some of the estimate data to a controller associated with control of the batch process.

53. A method as defined in claim 52, wherein using the estimate data further comprises using at least some of the estimate data to control the batch process via the controller.