Method and device for conducting a process in a controlled system with at least one precomputed process parameter.

• US 5,600,758 A
• Filed: 11/10/1994
• Issued: 02/04/1997
• Est. Priority Date: 11/11/1993
• Status: Expired due to Term

• Current Assignee: Siemens AG
• Original Assignee: Siemens AG
• Inventors: Martinetz, Thomas, Soergel, Guenter, Broese, Einar, Gramckow, Otto
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): KATBAB, ABDOLLAH
• Application Number:
  US08/337,660
• Time in Patent Office: 817 Days
• Field of Search: 395/21, 364/148, 364/157
• US Class Current: 706/23
• CPC Class Codes: B21B 37/58 Roll-force control; Roll-gap control B21B38/105 takes precedence
• CPC Class Codes: B21B 37/74 Temperature control, e.g. by cooling or heating the rolls or the product B21B37/32, B21B37/44 take precedence
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 17/02 electric

Abstract

Known methods for conducting a process in an automatically controlled system preset the system at the beginning of each process run based on at least one process parameter. The process parameter is precomputed with a model of the process which is supplied with input values. During the process the input values and the process parameter are measured and are used to adaptively improve the precomputed process parameter after the process run. The present invention simplifies and improves the precomputed value of the process parameter by supplying at least part of the input values to a neural network. The network response of the neural network forms a correction value for the approximate value delivered by the model for the process parameter to be precomputed. The network parameters of the neural network are modified after each process run to adapt the precomputed value to the actual process events.

Claims

1. A method for controlling a process with a computing means having a mathematical model of the process and with a neural network having variable network parameters, the method comprising steps of:

a) supplying input values to the mathematical model;

b) precomputing at least one selected process parameter with the mathematical model at the beginning of a process run, based on the input values supplied to the mathematical model;

c) presetting the at least one process parameter;

d) measuring the input values and the at least one process parameter during the process; and

e) adaptively improving the at least one process parameter after the process based on the measured at least one process parameter and based on the measured input values, the step of adaptively improving including sub-steps of;

I) supplying at least part of the measured input values to the mathematical model;

ii) supplying at least part of the measured input values to the neural network;

iii) forming a computed at least one process parameter with the mathematical model;

iv) forming a network response with the neural network;

v) linking the computed at least one process parameter with the network response to form a linked result;

vi) comparing the linked result with the measured process parameter to form a deviation; and

vii) adaptively modifying the variable network parameters of the neural network such that the deviation is reduced, the variable network parameters being adaptively modified to train the neural network on-line.

2. The method of claim 1 wherein the variable network parameters of the neural network are adaptively modified on-line, after each process run, using the measured input values and the measured process parameters.

3. The method of claim 1 further comprising a step of additionally adaptively modifying variable model parameters of the mathematical model after a plurality of process runs, and based on a deviation between the measured process parameters and a result computed by the mathematical model.

4. The method of claim 3 wherein the variable model parameters of the mathematical model are additionally adaptively modified off-line.

5. The method of claim 1 wherein the sub-step of linking is additive linking.

6. The method of claim 1 wherein the sub-step of linking is multiplicative linking.

7. The method of claim 1 wherein the sub-step of linking is dual linking.

8. The method of claim 1 wherein the process is a rolling process.

9. The method of claim 8 wherein the at least one selected process parameter precomputed and preset includes a rolling force in rolling stands of a rolling train.

11. The method of claim 1 wherein the at least part of the measuring input values supplied to the neural network include values whose influence on the at least one process parameter cannot be accurately described by the mathematical model.

10. The method of 8 wherein the at least one selected process parameter precomputed and preset includes a temperature variation of a rolled strip.

12. An arrangement for conducting a process in a controlled system, the arrangement comprising:

a) means for presetting the system based on at least one precomputed process parameter;

b) a computing means including a mathematical model of the process for precomputing the at least one process parameter based on input values;

c) means for measuring input values and process parameters during the process;

d) a neural network, the neural network

i) having variable network parameters,

ii) adaptively modifying the at least one process parameter precomputed by the computing means, the at least one process parameter being adaptively modified to train the neural network on-line,

iii) being supplied with at least part of the input values measured by the means for measuring, and

iv) providing a network response based on its variable network parameters and based on the at least part of the input values; and

e) means for linking the at least one process parameter precomputed by the computing means with the network response of the neural network the linked result being provided to the computing means for adapting the at least one process parameter.

13. A method for estimating at least one process parameter of a process, the at least one process parameter used for controlling the process, the method comprising the steps of:

a) supplying model input values to a computing means implementing an analytical process model of the process;

b) determining at least one estimated computed value of the at least one process parameter using the analytical process model, the analytical process model determining the at least one estimated computed value based on the model input values;

c) providing network input values to a neural network structure;

d) forming at least one network response value associated with the at least one process parameter using the neural network structure, the neural network structure forming the at least one network response value based on the model input values;

e) linking at least one estimated computed value to the at least one network response value for generating at least one estimated process parameter value associated with the at least one process parameter; and

(f) providing the estimated process parameter value to a controller for controlling the process.

14. The method of claim 13, wherein the at least one estimated process parameter value is generated by adding the at least one estimated computed value to the at least one network response value.

15. The method of claim 13, wherein the at least one estimated process parameter value is generated by multiplying the at least one estimated computed value by the at least one network response value.

16. The method of claim 13, wherein the at least one estimated process parameter value is generated by adding one to the at least one network response value for forming a dual linkage value and multiplying the dual linkage value by the at least one estimated computed value.

17. The method of claim 13, wherein the model input values and the network input values both include at least one shared element value associated with the at least one process parameter.

18. The method of claim 17, wherein the model input values and the network input values are identical.

19. The method of claim 13, further comprising the step of:

f) adaptively modifying the at least one process parameter for on-line training of the neural network.

20. The method of claim 13, wherein the at least one network response value associated with the at least one process parameter includes one on-line training value for on-line training the neural network structure.

21. The method of claim 13, further comprising the step of:

g) training the neural network structure based on a difference between the at least one estimated process parameter value and the at least one network response value.

22. The method of claim 13, wherein the process is preset with the at least one network response value, the process including a rolling train process.

23. The method of claim 22, wherein the rolling train process is controlled by a rolling train, the rolling train having a rolling grip, the rolling grip having a thickness profile which corresponds to the at least one network response value.

24. The method of claim 13, wherein the process is controlled by a controller, the at least one network response value being supplied to the controller, the process including a rolling train process.

25. The method of claim 13, wherein the process includes rolling processes.

26. The method of claim 13, wherein the least one estimated process parameter value, the at least one estimated computed value and the at least one network response value are all formed as future values for the at least one process parameter of the process.