Intelligent process control using predictive and pattern recognition techniques

• US 6,609,119 B1
• Filed: 06/17/1999
• Issued: 08/19/2003
• Est. Priority Date: 03/14/1997
• Status: Expired due to Term

• Current Assignee: Dubai Aluminium Company Limited (Emirate of Dubai (United Arab Emirates))
• Original Assignee: Dubai Aluminium Company Limited (Emirate of Dubai (United Arab Emirates))
• Inventors: Meghlaoui, Abdelhamid
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/335,062
• Time in Patent Office: 1,524 Days
• Field of Search: 204/400, 204/435, 205/336, 706/23
• US Class Current: 706/25
• CPC Class Codes: C25C 3/20 Automatic control or regulation of cells controlling or regulating in general G05
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: Y02P 70/50 Manufacturing or production processes characterised by the final manufactured product

Abstract

A neural control logic scheme based on prediction and pattern recognition techniques is used to control electrochemical processes such as aluminum electrolytic cells. The predictive capacity of feedforward neural networks is used to predict the future values of decision variables to be used by the cell'"'"'s control logic, enabling the control logic to apply anticipated actions to cells in different conditions, thus avoiding anode effects and improving cell stability. The pattern-recognition capacity of LVQ-type neural networks is used to provide a closed-loop control structure to the feeding of the cell as a function of cell resistance, alumina concentration and cell condition. The closed-loop control structure enables the cell to operate at a near-optimal regime regardless of the condition of the cell.

Claims

1. A method for closed loop control of an electrochemical cell having an electrolytic bath resistance, a rate of change over time of the resistance and at least one cell condition, each condition having a characteristic curve described by a non-linear regression function, comprising the steps of:

a. using a neural network with a predictive algorithm to predict the bath resistance and its rate of change over time;

b. using a neural network with a pattern-recognition algorithm to recognize the condition of the cell, wherein each condition is identified with a codebook of associated triggers;

c. deducing the alumina concentration in the cell on a real-time basis from the resistance and the nonlinear regression function associated with the characteristic curve; and

d. using the neural networks to operate a closed-loop feeding control mechanism for the cell whereby a controller controls the rate at which a reactant is fed into the electrolytic bath.

2. The method of claim 1 wherein the cell condition is identified by measuring the concentration of a reactant in the electrolytic bath and the resistance of the electrolytic bath.

3. The method of claim 2 wherein cell is an aluminum electrochemical cell and the reactant is alumina.

4. The method of claim 3, further including the steps of

a. applying a decision criteria to the control logic; and

b. adjusting the decision criteria to maintain an approximately optimal concentration of alumina in the cell.

5. The device of claim 4 wherein the electrochemical cell produces aluminum.

6. The device of claim 3 wherein the reactant is alumina.

7. The method of claim 1, wherein the feeding control logic controls the cell based upon the resistance of the electrolytic bath and the concentration of the reactant in the electrolytic bath.

8. The method of claim 1, wherein the set of decision criteria have the following rule:

a. at low alumina concentration, the decision criteria are tightened, or

b. at high alumina concentration, the decision criteria are relaxed.

14. The device of claim 8 wherein the cell condition is identified in real-time.

9. A device for controlling an electrochemical cell having an electrolytic bath and a cell condition comprising:

an algorithm for predictive control of the cell; and

an algorithm for pattern-recognition control of the cell; and

a neural network which utilizes the pattern-recognition algorithm to recognize and identify the cell condition;

a neural network which utilizes the predictive algorithm to predict the cell resistance and the rate of change of the resistance over time; and

a controller for controlling the rate at which reactant is fed into the cell, wherein the controller comprises a feeding control logic, and further wherein the feeding control logic utilizes at least pattern-recognition and predictive control methods.

10. The device of claim 9 wherein the condition of the cell is identified in real time.

11. The device of claim 9 wherein the reactant is alumina.

12. The device of claim 11 wherein the electrochemical cell produces aluminum.

13. The device of claim 9 wherein the electrochemical cell produces aluminum.

15. The device of claim 9 wherein the reactant is alumina.

16. A device for closed-loop control of the production of aluminum in electrochemical cells having a resistance and a cell condition comprising:

a neural network for predictive control of the resistance of the cell;

a neural network for pattern-recognition control of the condition of the cell;

a feeding control logic controlled by the neural networks; and

a controller for controlling the rate of addition of a reactant to the cell according to the feeding control logic;

wherein the resistance and the cell condition change over time and wherein the controller controls the cell to operate efficiently independent of the condition of the cell by using non-linear regression functions to deduce the concentration of reactant in the cell and using the predicted resistance and the condition of the cell to control the feeding of the cell in sufficient time to optimize the feeding by optimizing the reactant concentration in the cell.

17. The device of claim 16 wherein the controller controls the cell to minimize anode effects and sludge formation while controlling the cell to operate at near-optimal aluminum concentration.

18. A method for determining the optimal alumina concentration in a cell having a cell condition that changes over time during the electrochemical production of aluminum comprising the steps of:

training a neural network to utilize pattern-recognition and predictive control techniques to control a feeding control logic for the cell; and

using the feeding control logic to continuously maintain an optimal concentration of aluminum in the cell, independent of the cell condition.

19. A device for closed-loop control of an electrochemical cell having an electrolytic bath containing a reactant, and a cell condition, the device comprising:

a neural network for predictive control of the cell which predicts future values of resistance in the cell and the rate of change of the resistance over time;

a neural network for pattern-recognition control of the cell which recognizes the present condition of the cell;

output from both neural networks;

a codebook of triggers, cell conditions and reactant concentrations; and

a controller for controlling the rate at which reactant is fed into the cell, wherein the controller utilizes a feeding control logic, the codebook and the output.

20. The device of claim 19 wherein the reactant is alumina.

21. A controller for controlling the rate at which a reactant is fed into an electrochemical cell comprising the steps of:

identifying a set of typical conditions in the electrochemical cell;

measuring the resistance and alumina concentration for each of the conditions;

plotting the curves characteristic of each condition;

associating a nonlinear regression to each curve;

training a first neural network to predict future values of the resistance in the cell and the rate of change over time of the resistance;

training a second neural network to recognize the present condition of the cell;

establishing a codebook of triggers, cell concentrations and alumina concentrations to be used by a control logic; and

utilizing the first and second neural networks, the codebook and the control logic to control the electrochemical cell.

22. The device of claim 21 wherein the reactant is alumina.

23. A method for determining the optimal alumina concentration in an electrochemical cell having a cell condition during the electrochemical production of aluminum comprising the steps of:

training a first neural network to predict values of the cell resistance and a second neural network to recognize the present condition of the cell;

identifying typical conditions of an aluminum electrochemical cell and associating a characteristic curve to each cell condition;

associating a nonlinear regression function to each curve from which the resistance versus alumina concentration may be determined;

establishing a codebook of triggers, cell conditions and alumina concentration;

using the neural networks, the codebook and a control logic to perform closed loop control of the cell under low, medium and high alumina concentrations;

comparing the performance of the cell under the low, medium and high alumina concentrations; and

determining the optimal operational value of the alumina concentration in the cell.

24. A method for closed loop control of an electrochemical cell for producing aluminum, comprising the steps of:

using two levels of control, a first control level comprising predicting a cell resistance and its rate of change over time, a second control level comprising recognizing at least one cell condition, each condition having a characteristic curve described by a non-linear regression function;

estimating a real-time alumina concentration from the, non-linear relationship of resistance versus alumina concentration;

establishing a set of decision criteria based on the cell condition, the estimated alumina concentration and the predicted values of the cell resistance and its rate of change over time; and

feeding alumina into the electrolytic bath based on the set of decision criteria.

25. A device for controlling an electrochemical cell having an electrolytic bath resistance and a cell condition comprising:

means for predicting the cell resistance and its rate of change over time;

means for recognizing the cell condition; and

means for feeding a reactant into the electrolytic bath based on a set of decision criteria, wherein the set of decision criteria are based on the cell condition, an estimated real-time reactant concentration, and the predicted values of the cell resistance and its rate of change over time, and wherein the real-time reactant concentration is estimated from a non-linear relationship of the cell resistance versus reactant concentration.

26. A device for controlling an electrochemical cell having an electrolytic bath resistance and a cell condition comprising:

a first neural network that predicts the cell resistance and its rate of change over time;

a second neural network that recognizes and identifies the cell condition; and

a feed controller that sets the rate at which a reactant is fed into the cell, wherein the feed controller comprises a feed control logic, and wherein the feed control logic utilizes at least the first and second neural networks.