Method for monitoring multivariate processes

• US 5,949,678 A
• Filed: 10/16/1996
• Issued: 09/07/1999
• Est. Priority Date: 12/22/1993
• Status: Expired due to Term

• Current Assignee: Telefonaktiebolaget LM Ericsson
• Original Assignee: Telefonaktiebolaget LM Ericsson
• Inventors: Sundin, Karl Olov Lasse, Wold, Svante
• Primary Examiner(s): Grant, William
• Assistant Examiner(s): PATEL, RAMESH B
• Application Number:
  US08/666,346
• Time in Patent Office: 1,056 Days
• Field of Search: 364/188, 364/149, 364/158, 364/126, 364/151, 364/578, 364/513, 364/138, 364/468.01, 364/468.02, 364/468.03, 364/148, 364/150, 364/152, 364/191, 364/139
• US Class Current: 700/83
• 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: G05B 23/0289 Reconfiguration to prevent failure, e.g. usually as a reaction to incipient failure detection

Abstract

Variables associated with a multivariate process are monitored on a periodic basis. As a result of an observation of the variables, a residual standard deviation between the monitored variable values and a predictive model can be determined. The residual standard deviation is plotted as a series of connected points. Through the use of an adaptive model, the observations of the variables, and in turn, the residual standard deviation and the predictive model are adapted to changes in the multivariate process.

Claims

1. A method for monitoring a multivariate process comprising the steps of:

registering measured values of variables associated with said multivariate process as an observation at a predetermined time;

describing the variables as a multidimensional space, wherein each variable represents one dimension in the space;

representing each observation as a point in the multidimensional space, whereby a series of observations carried out at different times will be represented by a point cluster in space;

calculating for the point cluster at least one first and one second principal directions in space;

determining projections of the observations onto the first and second principal directions, whereby a model of the multivariate process is obtained;

updating during the course of the multivariate process the first and second principal directions of the point cluster in space, whereby the model of the multivariate process is dynamically adapted to the multivariate process in real time.

2. A method according to claim 1, wherein the multivariate process is illustrated as a picture on a screen, wherein the first and the second principal directions define a plane shown on the screen, wherein the projections of the observations are illustrated as points on the screen, and wherein the multivariate process is displayed in the form of a set of connected points which moves over the screen from one point to another point representing consecutive observations of the process over time.

3. A method according to claim 1, wherein the multivariate process is illustrated as a picture on a screen, wherein said first, said second and a third principal directions define a three-dimensional space which is shown on the screen, wherein the projections of the observations are illustrated as points on the screen, and wherein the multivariate process is displayed in the form of a set of connected points which moves over the screen from one point to another point representing consecutive observations of the process over time.

4. A method according to claim 2 or 3, wherein the multivariate process is illustrated by determining the residual standard deviation between a latest observed vector presentation of the process and an on-line updated model of the process used as a quantity indicating said residual standard deviation, and by a set of connected points on the screen changing color in dependence on said residual standard deviation.

5. A method according to claim 4, wherein said set of connected points is displayed on said screen using a first color when said residual standard deviation is within said normal range and using a second color when said residual standard deviation is outside of said normal range.

6. A method according to claim 2 or 3, further comprising the step of:

initiating an alarm when the set of connected points enters a predefined range which lies outside said normal range for the multivariate process.

7. A method according to claim 2 or 3, wherein a part of the multivariate process which is of current interest is displayed more brightly on said screen.

8. A method according to claim 2, wherein positions of the individual variables are projected onto said plane, and a midpoint of the point cluster in this plane is illustrated as points on the screen.

9. A method according to claim 8, wherein said multivariate process, indicated in the form of a set of connected points as well as a score plot, is simultaneously displayed in said plane defined by the first and second principal directions on separate or on the same screen image, whereby the movement of the set of connected points over the screen can be immediately related to multivariate process-driving variables,, whereby the direction of the movement of the set of connected points away from the midpoint of the point cluster indicates an influence on the multivariate process by one or more variables which in the score plot lie in the same direction of travel as the movement of the set of connected points which indicates the influence of variable/variables in drift in the process.

10. A method for monitoring a multivariate process, comprising the steps of:

registering measured values of variables associated with said multivariate process as an observation at a certain time;

describing the variables as a multidimensional space, wherein each variable represents one dimension in the space;

representing each observation as a point in the multidimensional space, whereby a series of observations carried out at different times are represented by a point cluster in space;

calculating for the point cluster at least one first and one second principal directions in space;

determining projections of the observations onto the first and second principal directions, whereby a model of the multivariate process is obtained;

updating during the course of the multivariate process the first and second principal directions of the point cluster in space, whereby the model of the process is dynamically adapted to the process in real time;

p1 calculating a residual standard deviation between a latest observed vector presentation of the multivariate process and an on-line updated model of the multivariate process; and

initiating an alarm when the residual standard deviation of the multivariate process to the model exceeds a predetermined limit.

11. A method according to claim 10, wherein the residual standard deviation is displayed as a function of the observations made over time.

12. An apparatus for monitoring a multivariate process comprising:

means for registering measured values of variables associated with said multivariate process as an observation at a predetermined time;

means for describing the variables as a multidimensional space, wherein each variable represents one dimension in the space;

means for representing each observation as a point in the multidimensional space, whereby a series of observations carried out at different times will be represented by a point cluster in space;

means for calculating for the point cluster at least one first and one second principal directions in space;

means for determining projections of the observations onto the first and second principal directions, whereby a model of the multivariate process is obtained; and

means for updating during the course of the multivariate process the first and second principal directions of the point cluster in space, whereby the model of the multivariate process is dynamically adapted to the multivariate process in real time.

13. An apparatus according to claim 12, wherein the multivariate process is illustrated as a picture on a screen, wherein the first and the second principal directions define a plane shown on the screen, wherein the projections of the observations are illustrated as points on the screen, and wherein the multivariate process is displayed in the form of a set of connected points which moves over the screen from one point to another point representing consecutive observations of the process over time.

14. An apparatus according to claim 13, wherein the multivariate process is illustrated by determining the residual standard deviation between a latest observed vector presentation of the process and an on-line updated model of the process used as a quantity indicating said residual standard deviation, and by a set of connected points on the screen changing color in dependence on said residual standard deviation.

15. An apparatus according to claim 14, wherein said set of connected points is displayed on said screen using a first color when said residual standard deviation is within said normal range and using a second color when said residual standard deviation is outside of said normal range.

16. An apparatus according to claim 12, wherein the multivariate process is illustrated as a picture on a screen, wherein said first, said second and a third principal directions define a three-dimensional space which is shown on the screen, wherein the projections of the observations are illustrated as points on the screen, and wherein the multivariate process is displayed in the form of a set of connected points which moves over the screen from one point to another point representing consecutive observations of the process over time.