Method for isolating sources of drifts in output properties for machines and processes

• US 7,174,283 B2
• Filed: 07/08/2002
• Issued: 02/06/2007
• Est. Priority Date: 07/16/2001
• Status: Expired due to Fees

• Current Assignee: Pronetix Limited
• Original Assignee: Pronetix Limited
• Inventors: Berkooz, Oded, Evenor, Moshe, Roach, Robert Landon
• Primary Examiner(s): PALADINI, ALBERT WILLIAM
• Application Number:
  US10/483,879
• Publication Number:
  US 20040176116A1
• Time in Patent Office: 1,674 Days
• Field of Search: 703/2, 703/6, 700 28- 31, 700/38, 700/45, 700/51, 702/84
• US Class Current: 703/2
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis

Abstract

A method for determining a root cause for a drift in a process or a machine, the function of which is determined by a model describing a relationship between setup and/or input variables and output property or properties, comprising the steps of: (a) Assigning desired output properties for a product of said machine or process, and allowed ranges for said properties; (b) Assigning original values to said variables for obtaining by said machine or process said assigned output properties within said assigned ranges; (c) Measuring the various output properties of the machine or process; (d) When measuring an output property of the machine or process that drifts from its assigned value, substituting the one or more measured output properties in their corresponding model equations, and calculating for each variable in each equation its presumed value, assuming it is the sole cause of the drift; and (e) Analyzing all presumed calculated variable values and the measured values of the output properties of the machine or process to determine the variable that is most probably responsible for the drift.

Claims

1. A method for determining a root cause for a drift in at least one of a machine and a process, the function of which is determined by a model describing a relationship between at least one of setup and input variables and at least one output property, comprising:

a. Assigning desired output properties for a product of at least one of said machine and said process, and allowed ranges for said properties;

b. Assigning original values to said variables for obtaining by at least one of said machine and said process said assigned output properties within said assigned ranges;

c. Measuring the various output properties of at least one of the machine and the process;

d. When measuring an output property of at least one of the machine and the process that drifts from its assigned value, substituting the measured output properties in their corresponding model equations, and calculating for each variable in each equation its presumed value, assuming it is the sole cause of the drift; and

e. Analyzing all presumed calculated variable values and the measured values of the output properties of at least one of the machine and the process to determine the variable that is most probably responsible for the drift.

2. A method according to claim 1, wherein the model is a model type that predicts process outputs given the values substituted for the variable, with sufficient accuracy.

3. A method according to claim 1 wherein the variable which is determined as the one responsible for the drift is the variable for which all its presumed calculated values are found to be the same for all output properties.

4. A method according to claim 1 wherein the variable which is determined as the one responsible for the drift is the variable for which all its presumed calculated values are found to be the same for all output properties, taking into account variations due to statistical distributions of the output properties and measurement errors.

5. A method according to claim 1, wherein:

a. The measuring step comprises measuring of a series of output measurements during different production times;

b. When a drift is realized in a measured output property, assigning a first early output measurement as a starting measurement, and a later output measurement as an ending measurement;

c. The substituting step comprises, for each variable, defining its said original assigned value as the starting input value, and substituting the ending measurement in the model, and calculating for each variable its presumed ending input value, assuming it is the sole cause of the drift;

d. In the analyzing step, for a plurality of points in the range between the starting input value and the ending input value of each variable, calculating from the model corresponding output values, assuming the other variables having values as originally assigned for obtaining the assigned output property; and

e. Determining the variable whose output values calculated in step (d) are most similar to the output measured values, as the one most probably responsible for the drift.

6. A method according to claim 1, wherein the determining step comprises:

a. For each variable, drawing a curve between its calculated output values, therefore obtaining an output curve corresponding to each variable; and

b. Determining the variable whose output curve is most similar to the output measured values, as the one most probably responsible for the drift.

7. A method according to claim 5, taking into account statistical distributions of the output properties and measurement variation errors.

8. A method according to claim 1, wherein the analyzing step comprises:

a. Finding a new set of variables being each within a limited range from its original value, that when the new set of variables, supplemented one at a time with the calculated amount of drift in each variable, is substituted in the model, results in sufficiently remote one from the other output property values;

b. Calculating from the model separately for each presumed variable drift, by evaluating the model while substituting the new set and each of the presumed drifts in turn; and

c. Performing an experiment in which the said found new set of variables are substituted in at least one of the machine and the process, and obtaining an experimental output; and

d. Assigning the variable for which the calculated output is closest to the said experimental output as the variable responsible for the drift.

9. A method according to claim 8, wherein said new set of variables is found by using the model.

10. A method according to claim 8, wherein said new set of variables is found by applying directional derivatives to the model separately with respect to each variable.

11. A method according to claim 8, wherein said new set of variables is selected such that it provides outputs which differ by more than the statistical variation of at least one of the machine and the process, while staying as close as possible to the original values as assigned, to obtain sufficiently remote one from the other output property values.

12. A method according to claim 8, taking into account statistical distributions of the output properties and measurement variation errors.

13. A method according to claim 1, wherein the input variables of at least one of the machine and the process represent at least one of inputs and internal mechanisms.

14. A method according to claim 6, taking into account statistical distributions of the output properties and measurement variation errors.