Controlling a manufacturing process with a multivariate model
First Claim
1. A computer-implemented method for controlling a batch-type manufacturing process with a finite duration, the method comprising:
- receiving dependent variable data and manipulated variable data associated with the batch-type manufacturing process, the dependent variable data including measured past and present values of a first set of process parameters observed by one or more sensors, the manipulated variable data including measured past and present values of a second set of process parameters measured from a plurality of process tools, wherein the first set of process parameters, representative of dependent variables, and the second set of process parameters, representative of manipulated variables, are X-type variables in the batch-type manufacturing process;
determining, using a multivariate model of the manufacturing process, one or more multivariate statistics based on at least the dependent variable data and the manipulated variable data, wherein each multivariate statistic, which comprises a Hotelling value, a residual standard deviation value, a principal component score or a partial least squares component score, measures a deviation of the batch-type manufacturing process from a multivariate space of normal process behavior;
determining future values of the manipulated variables by optimizing an objective function that comprises J=θ
Y(YSP−
Ypred)2+θ
MV(EMV)2+θ
DModX(EDModX)2+θ
T2(ET2)2+θ
t(Et)2, wherein (i) YSP represents at least one setpoint or target value for Y-type yield variables representative of yield or quality at the end of the finite duration of the batch-type manufacturing process, (ii) Ypred represents at least one predicted value for the yield variables, (iii) EMV represents an amount of deviation in the manipulated variable data from a desired trajectory subject to a penalty weight θ
MV, (iv) EDModX represents an amount of the residual standard deviation value subject to a penalty weight θ
DModX, (v) ET2 represents an amount of the Hotelling value subject to a penalty weight θ
T2, (vi) Et represents an amount of the principal component score or the partial least squares component score subject to a penalty weight θ
t, (vii) and θ
Y represents a penalty weight;
adjusting at least one of the second set of process parameters based on the future values of the manipulated variables.
9 Assignments
0 Petitions
Accused Products
Abstract
A method, controller, and system for controlling a manufacturing process (batch-type or continuous-type) with a multivariate model are described. Dependent variable data and manipulated variable data are received. Dependent variable data represents values of uncontrolled process parameters from a plurality of sensors. Manipulated variable data represents controlled or setpoint values of controllable process parameters of a plurality of process tools. A predicted operational value, multivariate statistic, or both are determined based on the received data, and operating parameters of the manufacturing process are determined based on the predicted score, multivariate statistic, or both.
133 Citations
30 Claims
-
1. A computer-implemented method for controlling a batch-type manufacturing process with a finite duration, the method comprising:
-
receiving dependent variable data and manipulated variable data associated with the batch-type manufacturing process, the dependent variable data including measured past and present values of a first set of process parameters observed by one or more sensors, the manipulated variable data including measured past and present values of a second set of process parameters measured from a plurality of process tools, wherein the first set of process parameters, representative of dependent variables, and the second set of process parameters, representative of manipulated variables, are X-type variables in the batch-type manufacturing process; determining, using a multivariate model of the manufacturing process, one or more multivariate statistics based on at least the dependent variable data and the manipulated variable data, wherein each multivariate statistic, which comprises a Hotelling value, a residual standard deviation value, a principal component score or a partial least squares component score, measures a deviation of the batch-type manufacturing process from a multivariate space of normal process behavior; determining future values of the manipulated variables by optimizing an objective function that comprises J=θ
Y(YSP−
Ypred)2+θ
MV(EMV)2+θ
DModX(EDModX)2+θ
T2(ET2)2+θ
t(Et)2, wherein (i) YSP represents at least one setpoint or target value for Y-type yield variables representative of yield or quality at the end of the finite duration of the batch-type manufacturing process, (ii) Ypred represents at least one predicted value for the yield variables, (iii) EMV represents an amount of deviation in the manipulated variable data from a desired trajectory subject to a penalty weight θ
MV, (iv) EDModX represents an amount of the residual standard deviation value subject to a penalty weight θ
DModX, (v) ET2 represents an amount of the Hotelling value subject to a penalty weight θ
T2, (vi) Et represents an amount of the principal component score or the partial least squares component score subject to a penalty weight θ
t, (vii) and θ
Y represents a penalty weight;adjusting at least one of the second set of process parameters based on the future values of the manipulated variables. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A multivariate controller for a batch-type manufacturing process with a finite duration, the controller comprising:
-
a hardware control module in communication with a plurality of process tools and a plurality of sensors to monitor manipulated variable data from the process tools and dependent variable data from the sensors, the control module including a multivariate model for determining one or more multivariate statistics based on at least the manipulated variable data and the dependent variable data, each multivariate statistic comprising a Hotelling value, a residual standard deviation value, a principal component score or a partial least squares component score, and each multivariate statistic measuring a deviation of the batch-type manufacturing process from a multivariate space of normal process behavior, wherein the dependent variable data includes measured past and present values of a first set of process parameters observed by the plurality of sensors and the manipulated variable data includes measured past and present values of a second set of process parameters, wherein the first set of process parameters, representative of dependent variables, and the second set of process parameters, representative of manipulated variables, are X-type variables in the batch-type manufacturing process; and a hardware solver module to;
i) receive, from the multivariate model, the one or more multivariate statistics, and ii) optimize an objective function using the multivariate statistics determined from the multivariate model to generate future values of the manipulated variables for providing to the plurality of process tools, wherein the objective function comprises J=θ
Y(YSP−
Ypred)2+θ
MV(EMV)2+θ
DModX(EDModX)2+θ
T2(ET2)2+θ
t(Et)2, wherein (i) YSP represents at least one setpoint or target value for Y-type yield variables representative of yield or quality at the end of the finite duration of the batch-type manufacturing process, (ii) Ypred represents at least one predicted value for the yield variables, (iii) EMV represents an amount of deviation in the manipulated variable data from a desired trajectory subject to a penalty weight θ
MV, (iv) EDModX represents an amount of the residual standard deviation value subject to a penalty weight θ
DModX, (v) ET2 represents an amount of the Hotelling value subject to a penalty weight θ
T2, (vi) E1 represents an amount of the principal component score or the partial least squares component score subject to a penalty weight θ
t, (vii) and θ
Y represents a penalty weight. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28)
-
-
29. A system for controlling a batch-type manufacturing process with a finite duration, the system comprising:
-
a hardware data acquisition means for acquiring manipulated variable data associated with the manufacturing process, including measured past and present values of a set of process parameters measured from a plurality of process tools; and
acquiring dependent variable data associated with the manufacturing process, including measured past and present values of a second set of process parameters observed by a plurality of sensors, wherein the first set of process parameters, representative of dependent variables, and the second set of process parameters, representative of manipulated variables, are X-type variables in the batch-type manufacturing process;a hardware multivariate control means incorporating a multivariate statistical model for receiving at least the manipulated variable and dependent variable data and determining multivariate statistical information that measures a deviation of the batch-type manufacturing process from a multivariate space of normal process behavior, wherein the multivariate statistical information comprises one or more of a Hotelling value, a residual standard deviation value, a principal component score or a partial least squares component score; a hardware process control means for determining future values of the manipulated variables by optimizing an objection function using at least the multivariate statistical information determined by the multivariate control means, the objective function comprising J=θ
Y(YSP−
Ypred)2+θ
MV(EMV)2+θ
DModX(EDModX)2+θ
T2(ET2)2+θ
t(Et), wherein (i) YSP represents at least one setpoint or target value for Y-type yield variables representative of yield or quality for the end of the finite duration of the batch-type manufacturing process, (ii) Ypred represents at least one predicted value for the yield variables, (iii) EMV represents an amount of deviation in the manipulated variable data from a desired trajectory subject to a penalty weight θ
MV, (iv) EDModX represents an amount of the residual standard deviation value subject to a penalty weight θ
DModX, (v) ET2 represents an amount of the Hotelling value subject to a penalty weight θ
T2, (vi) Et represents an amount of the principal component score or the partial least squares component score subject to a penalty weight θ
t, (vii) and θ
Y represents a penalty weight,wherein the process control means is configured to adjust at least one of the second set of process parameters representative of manipulated variables based on the future values of the manipulated variables. - View Dependent Claims (30)
-
Specification