Method of controlling a manufacturing process using multivariate analysis

• US 5,442,562 A
• Filed: 12/10/1993
• Issued: 08/15/1995
• Est. Priority Date: 12/10/1993
• Status: Expired due to Fees

• Current Assignee: Eastman Kodak Company
• Original Assignee: Eastman Kodak Company
• Inventors: Swanson, Ronald E., Hopkins, Robert W., Scheible, John J., Miller, Paige
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): BROWN, THOMAS
• Application Number:
  US08/165,155
• Time in Patent Office: 613 Days
• Field of Search: 364/148-152, 364/157-158, 364/160-163, 364/468, 364/401-403
• US Class Current: 700/108
• CPC Class Codes: G05B 21/02 electric

Abstract

In a manufacturing process using multivariate analysis for statistical process control, data representing a plurality of process and/or product variables is collected during operation of the process. A surrogate variable, representing the overall state of the process is evaluated. The surrogate variable is a function of a plurality of intermediate variables, which in turn are a function of the process and/or product variables. When the value of the surrogate variable is outside of a predetermined limit, the contributions to the surrogate variable by the intermediate variables are calculated. The intermediate variable with the largest contribution is identified and the contributions to the identified intermediate variable by the process variables is calculated., The process variables with the largest contribution to the identified intermediate variable are identified, and used to diagnose and correct the problem with the process.

Claims

1. A computer implemented method of controlling a manufacturing process, comprising the steps of:

a.) during operation of the process, measuring the process with sensors of data representative of a plurality of process/product variables;

b.) in a computer, connected to receive data from the sensors using multivariate analysis, said computer generating a surrogate variable having a value representative of whether or not the process is in control, said surrogate variable being a function of a plurality of intermediate variables, each of said intermediate variables being a weighted function of the process/product variables;

c.) when the value of the surrogate variable is outside a predetermined limit, said computer determining which of the intermediate variables primarily contributed to the value of the surrogate variable;

d.) for the intermediate variable determined in the previous step, identifying which of the process variables primarily contributed to the value of the determined intermediate variable; and

e.) said computer modifying the process to change the process variable identified in the previous step to bring the value of the surrogate variable within the predetermined limit.

2. The method claimed in claim 1, wherein said multivariate analysis is Principle Component Analysis (PCA) or Partial Least Squares Analysis (PLS), and said surrogate variable is the Quantity T_i², defined as ##EQU4## where:

D is the number of intermediate variables;

t_id² is the square of the d^th intermediate variable for the i^th observation for PCA or the square of the d^th intermediate variable for the X block for the i^th observation for PLS;

σ

_d² is the square of the standard deviation of the d^th intermediate variable across all observations for PCA or the square of the standard deviation of the d^th intermediate variable for the X block for PLS.

3. The method claimed in claim 2, wherein said step of determining which intermediate variable primarily contributed to the value of the surrogate variable, comprises calculating the contribution to T² by the intermediate variables, and selecting the intermediate variable with the largest contribution, wherein the contributions to T² are the individual terms ##EQU5##

4. The method claimed in claim 3, wherein said step of identifying which process variables primarily contributed to the determined intermediate variable comprises calculating the contribution of each process variable to the determined intermediate variable, wherein the intermediate variable is calculated according to the following equation, ##EQU6## and wherein the contributions to the intermediate variable are x_ij p_jd, where x_ij is the value of the j^th process variable for observation i and p_jd is the loading for the d^th intermediate variable and j^th process variable.

5. A computer implemented method of controlling a manufacturing process, comprising the steps of:

a.) during operation of the process, measuring the process with sensors of data representative of a plurality of process/product variables;

b.) in a computer, connected to receive data from the sensors using multivariate analysis, said computer generating a surrogate variable having a value representative of the overall state of the process, said surrogate variable being a function of a plurality of intermediate variables, each of said intermediate variables being a weighted function of the process/product variables;

c.) when the value of the value of the surrogate variable is outside a predetermined limit, said computer determining which of the intermediate variables primarily contributed to the value of the surrogate variable;

d.) for the intermediate variable determined in the previous step, identifying which of the process variables primarily contributed to the value of the determined intermediate variable;

e.) comparing the pattern of process variables determined in the previous step to a known pattern of process variables representing a particular process failure;

f.) when a match of said patterns of process variables is made, stopping the process; and

g.) said computer modifying the process to change the process variables identified in the previous step to correct said particular process failure.

6. The method claimed in claim 5, wherein said multivariate analysis is Principle Component Analysis (PCA) or Partial Least Squares Fit (PLS), and said surrogate variable is the Quantity T_i², defined as ##EQU7## where:

D is the number of intermediate variables;

t_id² is the square of the d^th intermediate variable for the i^th observation for PCA or the square of the d^th intermediate variable for the X block for the i^th observation for PLS;

σ

_d² is the square of the standard deviation of the d^th intermediate variable across all observations for PCA or the square of the standard deviation of the d^th intermediate variable for the X block for PLS.

7. The method claimed in claim 6, wherein said step of determining which intermediate variable primarily contributed to the value of the surrogate variable, comprises calculating the contribution to T² by intermediate variable, and selecting the intermediate variable with the largest contribution, wherein the contribution to T² are the individual terms ##EQU8##

8. The method claimed in claim 7, wherein said step of identifying which process variables primarily contributed to the determined intermediate variable comprises calculating the contribution of each process variable to the determined intermediate variable and selecting those process variable whose contribution values are above a certain predetermined limit.

9. The method claimed in claim 8, wherein the intermediate variable is calculated according to the following equation, ##EQU9## and wherein the contributions to the intermediate variable are x_ij p_jd, where x_ij is the value of the j^th process variable for observation i and p_jd is the loading for the d^th intermediate variable and j^th process variable.

10. A computer implemented method of controlling a process, comprising the steps of:

a.) during operation of the process, measuring the process with sensors of data representative of a plurality of process/product variables;

b.) in a computer, connected to receive data from the sensors using multivariate analysis, generating a surrogate variable having a value representative of whether or not the process is in control, said surrogate variable being a function of a plurality of intermediate variables, each of said intermediate variables being a weighted function of the process/product variables;

c.) when the value of the surrogate variable is outside a predetermined limit, said computer determining which of the intermediate variables primarily contributed to the value of the surrogate variable;

d.) for the intermediate variable determined in the previous step, identifying which of the process variables primarily contributed to the value of the determined intermediate variable; and

e.) said computer modifying the process to change the process variable identified in the previous step to bring the value of the surrogate variable within the predetermined limit.

11. The method claimed in claim 10, wherein said multivariate analysis is Principle Component Analysis (PCA) or Partial Least Squares Analysis (PLS), and said surrogate variable is the Quantity T_i², defined as ##EQU10## where:

D is the number of intermediate variables;

t_id² is the square of the d^th intermediate variable for the i^th observation for PCA or the square of the d^th intermediate variable for the X block for the i^th observation for PLS;

σ

_d² is the square of the standard deviation of the d^th intermediate variable across all observations for PCA or the square of the standard deviation of the d^th intermediate variable for the X block for PLS.

12. The method claimed in claim 11, wherein said step of determining which intermediate variable primarily contributed to the value of the surrogate variable, comprises calculating the contribution to T² by the intermediate variables, and selecting the intermediate variable with the largest contribution, wherein the contributions to T² are the individual terms ##EQU11##

13. The method claimed in claim 12, wherein said step of identifying which process variables primarily contributed to the determined intermediate variable comprises calculating the contribution of each process variable to the determined intermediate variable, wherein the intermediate variable is calculated according to the following equation, ##EQU12## and wherein the contributions to the intermediate variable are x_ij p_jd, where x_ij is the value of the j^th process variable for observation i and p_jd is the loading for the d^th intermediate variable and j^th process variable.

14. A computer implemented method of controlling a process, comprising the steps of:

a.) during operation of the process, measuring the process with sensors of data representative of a plurality of process/product variables;

b.) in a computer, connected to receive data from the sensors using multivariate analysis, generating a surrogate variable having a value representative of the overall state of the process, said surrogate variable being a function of a plurality of intermediate variables, each of said intermediate variables being a weighted function of the process/product variables;

c.) when the value of the value of the surrogate variable is outside a predetermined limit, said computer determining which of the intermediate variables primarily contributed to the value of the surrogate variable;

d.) for the intermediate variable determined in the previous step, identifying which of the process variables primarily contributed to the value of the determined intermediate variable;

e.) comparing the pattern of process variables determined in the previous step to a known pattern of process variables representing a particular process failure;

f.) when a match of said patterns of process variables is made, stopping the process; and

g.) said computer modifying the process to change the process variables identified in the previous step to correct said particular process failure.

15. The method claimed in claim 14, wherein said multivariate analysis is Principle Component Analysis (PCA) or Partial Least Squares Fit (PLS), and said surrogate variable is the Quantity T_i², defined as ##EQU13## where:

D is the number of intermediate variables;

t_id² is the square of the d^th intermediate variable for the i^th observation for PCA or the square of the d^th intermediate variable for the X block for the i^th observation for PLS;

σ

_d² is the square of the standard deviation of the d^th intermediate variable across all observations for PCA or the square of the standard deviation of the d^th intermediate variable for the X block for PLS.

16. The method claimed in claim 15, wherein said step of determining which intermediate variable primarily contributed to the value of the surrogate variable, comprises calculating the contribution to T² by intermediate variable, and selecting the intermediate variable with the largest contribution, wherein the contribution to T² are the individual terms ##EQU14##

17. The method claimed in claim 16, wherein said step of identifying which process variables primarily contributed to the determined intermediate variable comprises calculating the contribution of each process variable to the determined intermediate variable and selecting those process variable whose contribution values are above a certain predetermined limit.

18. The method claimed in claim 17, wherein the intermediate variable is calculated according to the following equation, ##EQU15## and wherein the contributions to the intermediate variable are x_ij p_jd, where x_ij is the value of the j^th process variable for observation i and p_jd is the loading for the d^th intermediate variable and j^th process variable.