On-line alignment of a process analytical model with actual process operation

• US 8,880,203 B2
• Filed: 05/21/2010
• Issued: 11/04/2014
• Est. Priority Date: 05/21/2010
• Status: Active Grant

• Current Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Original Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Inventors: Wojsznis, Wilhelm K., Blevins, Terrence Lynn
• Primary Examiner(s): Padmanabhan, Kavita
• Assistant Examiner(s): Everett, Christopher E
• Application Number:
  US12/784,689
• Publication Number:
  US 20110288660A1
• Time in Patent Office: 1,628 Days
• Field of Search: None
• US Class Current: 700/30
• CPC Class Codes: G05B 19/41875 characterised by quality surveillance of production
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

A batch modeling and analysis system uses a simple and computationally inexpensive technique to align data collected from an on-going, currently running or on-line batch process with a batch model formed for the batch process so as to enable the reliable determination of the current operational state of the on-line batch process with respect to the batch model. This data alignment technique enables further statistical processing techniques, such as projection to latent sources (PLS) and principle component analysis (PCA) techniques, to be applied to the on-line batch data to perform analyses on the quality of the currently running batch. These analyses, in turn, provide useful information to a user, such as a batch operator, that enables the user to determine the quality of the batch at the present time, based on the batch model, and the likelihood that the desired batch output quality metrics will be reached at the end of the batch run.

Claims

1. A method of aligning an on-line process run to a process model having a model trajectory for each of a multiplicity of process variables, the method comprising:

receiving, at a computer device, value data from the on-line process run indicative of a current value for each of the multiplicity of process variables at a particular sample time;

storing the received value data from the on-line process run in a computer readable memory;

for each of the multiplicity of process variables, using the computer device to determine a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the process variable and a model trajectory value for that process variable within the process model at the corresponding one of the plurality of model scan positions;

for each of the plurality of model scan positions, using a computer device to sum the ones of the plurality of distances determined for each of the multiplicity of the process variables that correspond to the model scan position to produce, for each of plurality of model scan positions, a summed distance;

using a computer device to determine a minimum summed distance from the summed distances determined for the plurality of model scan positions; and

selecting one of the plurality of model scan positions associated with the determined minimum summed distance as the model scan position to which the on-line process run is currently aligned.

2. The method of aligning an on-line process run to a process model of claim 1, wherein using the computer device to determine a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the process variable and a model trajectory value for that process variable within the process model at the corresponding one of the plurality of model scan positions includes using a plurality of model scan positions comprising a most recently determined model scan position and one or more model scan positions forward in time than the most recently determined model scan position.

3. The method of aligning an on-line process run to a process model of claim 1, wherein using the computer device to determine a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the process variable and a model trajectory value for that process variable within the process model at the corresponding one of the plurality of model scan positions includes determining a Euclidian distance between the current value of the process variable and the model trajectory value for the process variable at the corresponding one of the plurality of model scan positions.

4. The method of aligning an on-line process run to a process model of claim 3, further including squaring the Euclidian distances.

5. The method of aligning an on-line process run to a process model of claim 3, wherein using the computer device to determine a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the process variable and a model trajectory value for that process variable within the process model at the corresponding one of the plurality of model scan positions, further includes determining a slope difference distance defining a difference between the slope of the current value of the process variable and the slope of the model trajectory for that process variable at one of the plurality of model scan positions and includes combining the Euclidian distance and the slope difference distance to produce a combined distance as the distance between the current value of the process variable of the on-line process run and the model trajectory.

6. The method of aligning an on-line process run to a process model of claim 5, further including weighting the Euclidian distance and the slope difference distance differently when combining the Euclidian distance and the slope difference distance to produce the combined distance between the current value of the process variable of the on-line process run and the model trajectory.

7. The method of aligning an on-line process run to a process model of claim 3, wherein using a computer device to sum the ones of the plurality of distances determined for each of the multiplicity of the process variables that correspond to the model scan position further includes weighting the ones of the plurality of distances determined for each of the multiplicity of process variables that correspond to the model scan position unequally when computing the sum.

8. The method of aligning an on-line process run to a process model of claim 3, further including assigning a penalty factor to one or more of the summed distances based on the model scan position with which the one or more of the summed distances—

is associated.

9. The method of aligning an on-line process run to a process model of claim 8, wherein assigning the penalty factor includes assigning a penalty factor that increases as the model scan position increases in time from a most recently determined model scan position.

10. The method of aligning an on-line process run to a process model of claim 1, wherein using the computer device to determine a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the process variable and a model trajectory value for that process variable within the process model at the corresponding one of the plurality of model scan positions includes determining a slope difference distance as a distance between a slope of the current value of the process variable and a slope of the model trajectory for the process variable at each of the plurality of model scan positions.

11. A method of performing analysis of an on-line process run of a batch process, comprising:

generating, on a computer device, a statistical model having a model trajectory for each of a multiplicity of process variables prior to the on-line process run, the statistical model having a number of model scan positions;

storing the statistical model in a computer readable memory;

receiving, at a computer device, data from the on-line process run indicative of a current value for each of the multiplicity of process variables at a most recent time of measurement;

using the computer device to analyze the current value for each of the multiplicity of process variables to determine which of the number of model scan positions to which the on-line process run is aligned, wherein using the computer device to analyze the current value for each of the multiplicity of process variables includes;

for each of the multiplicity of process variables, determining a distance between the current value of the process variable of the on-line process run and a model trajectory for that process variable within the process model at each of a plurality of model scan positions;

for each of the plurality of model scan positions, summing the distances determined for the multiplicity of the process variables that correspond to the respective model scan position to produce a summed distance;

determining a minimum summed distance from the summed distances determined for the plurality of model scan positions; and

selecting the model scan position associated with the determined minimum summed distance as the model scan position to which the on-line process run is currently aligned; and

performing an analysis on the on-line data based on the determined alignment.

12. The method of performing analysis of an on-line process run of claim 11, wherein generating the statistical model includes collecting data for each of the multiplicity of process variables for a plurality of runs of the process, aligning the data for each of the plurality of runs of the process by skewing the data of each of the plurality of runs of the process to a normalized time frame, and developing each of the model trajectories of the statistical model as a statistical measure of the data in the normalized time frame.

13. The method of performing analysis of an on-line process run of claim 12, wherein the model trajectories include a measure of the mean of a process variable at each of a set of normalized times.

14. The method of performing analysis of an on-line process run of claim 12, wherein the model trajectories include a statistical measure of a variance of a process variable at each of a set of normalized times.

15. The method of performing analysis of an on-line process run of claim 11, wherein performing an analysis on the on-line data based on the determined alignment includes comparing the on-line data of one of the process variables to the model trajectory for that process variable.

16. The method of performing analysis of an on-line process run of claim 15, further including determining if the on-line data for a process variable falls outside of a range associated with the model trajectory for that process variable and indicating a problem if the on-line process data for the process variable falls outside of the range associated with the model trajectory for that process variable.

17. The method of performing analysis of an on-line process run of claim 11, wherein performing an analysis on the on-line data based on the determined alignment includes performing a principle component analysis procedure on the on-line data.

18. The method of performing analysis of an on-line process run of claim 11, wherein performing an analysis on the on-line data based on the determined alignment includes performing a projection to latent structure analysis on the on-line data to determine a predicted quality of the on-line process run.

19. The method of performing analysis of an on-line process run of claim 11, wherein performing an analysis on the on-line data based on the determined alignment includes determining a predicted quality of the on-line process run based on the statistical model.

20. The method of performing analysis of an on-line process run of claim 11, wherein determining a distance between the current value of the process variable of the on-line process run and a model trajectory for that process variable at each of the number of model scan positions includes using a number of model scan positions comprising the most recently determined model scan position and a preset number of one or more model scan positions forward in time than the most recently determined model scan position.

21. The method of performing analysis of an on-line process run of claim 20, wherein determining a distance includes determining a Euclidian distance between the current value of the on-line process variable when at the most recently determined model scan position and the model trajectory for the process variable at one of the number of model scan positions.

22. The method of performing analysis of an on-line process run of claim 21, wherein determining a distance further includes determining a slope difference distance defining a difference between a slope of the on-line process variable when at the most recently determined model scan position and the slope of the model trajectory for that process variable at one of the number of model scan positions and includes combining the Euclidian distance and the slope difference distance to produce the distance between the process variable of the on-line batch process run and the model trajectory.

23. The method of performing analysis of an on-line process run of claim 22, further including weighting the Euclidian distance and the slope differently when combining the Euclidian distance and the slope difference distance to produce the distance between the process variable of the on-line process run and the model trajectory.

24. The method of performing analysis of an on-line process run of claim 20, further including weighting the distances determined for the multiplicity of process variables unequally when computing the sum.

25. The method of performing analysis of an on-line process run of claim 20, further including assigning a penalty factor to one or more of the sums based on the model scan position with which a sum is associated.

26. The method of performing analysis of an on-line process run of claim 25, wherein assigning the penalty factor includes assigning a penalty factor that increases as the model scan position increases in time from the most recently determined model scan position.

27. The method of performing analysis of an on-line process run of claim 11, wherein generating the statistical model having a model trajectory for each of a multiplicity of process variables includes statistically combining measured data from multiple model scan positions to develop a model trajectory value for a process variable at a particular model scan position.

28. The method of performing analysis of an on-line process run of claim 11, wherein analyzing the current value for each of the multiplicity of process variables to determine which of the number of model scan positions to which the on-line process run is aligned includes determining an indicator variable alignment and using the determined indicator variable alignment in conjunction with the difference between the process variable values of the on-line process run and the model trajectory values for those process variables at each of a number of model scan positions to determine which of the number of model scan positions to which the on-line process run is aligned.

29. A computer device for use in aligning an on-line batch process run to a batch process model having a model trajectory for each of a multiplicity of batch process variables, comprising:

a non-transitory computer readable memory having a computer implemented application stored thereon, the application including;

a first routine that receives, at a computer device running the first routine, value data from the on-line batch process run indicative of a current value for each of the multiplicity of batch process variables at a particular sample time;

a second routine that stores the received value data from the on-line batch process run in a computer readable memory;

a third routine that, for each of the multiplicity of batch process variables, determines a plurality of distances corresponding to a plurality of model scan positions, each of the plurality of distances indicating a distance between the current value of the batch process variable and a model trajectory value for that batch process variable within the batch process model at the corresponding one of the plurality of model scan positions;

a fourth routine that, for each of the plurality of model scan positions, sums the ones of the plurality of distances determined for each of the multiplicity of the batch process variables that correspond to the model scan position to produce, for each of plurality of model scan positions, a summed distance;

a fifth routine that determines a minimum summed distance from the summed distances determined for the plurality of model scan positions; and

a sixth routine that selects one of the plurality of model scan positions associated with the determined minimum summed distance as the model scan position to which the on-line batch process run is currently aligned.

30. The computer device of claim 29, wherein the third routine uses a plurality of model scan positions comprising a most recently determined model scan position and one or more model scan positions later in time than the most recently determined model scan position.

31. The computer device of claim 29, wherein the third routine determines a Euclidian distance between the current value of the batch process variable when at the most recently determined model scan position and a value of the model trajectory for the batch process variable at one of the plurality of model scan positions.

32. The computer device of claim 31, wherein the third routine squares the Euclidian distance to determine the distance for the current value of the process variable.

33. The computer device of claim 31, wherein the third routine determines an absolute value of the Euclidian distance to determine the distance for the current value of the process variable.

34. The computer device of claim 31, wherein the third routine further determines a slope difference distance defining a difference between a slope of the batch process variable when at the most recently determined model scan position and a slope of the model trajectory for that batch process variable at one of the plurality of model scan positions and includes combining the Euclidian distance and the slope to produce a combined distance as the distance between the process variable of the on-line process run and the model trajectory.

35. The computer device of claim 34, wherein the third routine further includes weighting the Euclidian distance and the slope difference distance differently when combining the Euclidian distance and the slope difference distance to produce the distance between the process variable of the on-line process and the model trajectory.

36. The computer device of claim 31, wherein the fourth routine weights the distances determined for different ones of the multiplicity of process variables for a particular model scan position unequally when computing the summed distances.

37. The computer device of claim 31, wherein the fourth routine assigns a penalty factor to one or more of the summed distances based on the model scan position with which a summed distance is associated.

38. The computer device of claim 37, wherein the fourth routine assigns the penalty factor as a variable that increases in magnitude as the model scan position increases in time away from the most recently determined model scan position.

39. The computer device of claim 31, wherein the third routine determines the plurality of distances as a difference between a slope of the on-line batch process variable when at the current value and a slope of the model trajectory for the batch process variable at one of the plurality of model scan positions.