Control system for electric arc furnace

• US 5,204,872 A
• Filed: 04/15/1991
• Issued: 04/20/1993
• Est. Priority Date: 04/15/1991
• Status: Expired due to Fees

• Current Assignee: Neural Applications Corporation
• Original Assignee: Milltech-HOH, Inc.
• Inventors: Staib, William E., Staib, Robert B.
• Primary Examiner(s): Reynolds, Bruce A.
• Assistant Examiner(s): HOANG, TU BA
• Application Number:
  US07/685,629
• Time in Patent Office: 736 Days
• Field of Search: 373/104, 373/102, 373/105, 373/106, 373/107, 373/108, 373/47, 373/49, 373/50, 364/513, 364/807, 307/201
• US Class Current: 373/104
• CPC Class Codes: H05B 7/148 Automatic control of power electrode feeding arrangements H05B7/109; automatic feeding of electrodes for spot or seam welding or cutting B23K9/12; disposition of electrodes in or on furnaces F27D11/10; control of position in general G05D3/00; regulating electric characteristics of arcs in general G05F1/02; regulating electric power in general G05F1/66
• CPC Class Codes: Y02P 10/25 Process efficiency
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

An improved arc furnace regulator employs neural circuits connected in a multi-layer network configuration with various weighted relationships between the successive layers which are automatically changed over time as a function of an error signal by means of the back-propagation method so that the regulator gradually improves its control algorithm as a result of accumulated experience. The network is implemented in software which can be developed and run on a PC with extra co-computing capability for greater execution speed. A second trainable neural network which emulates the arc furnace is used to develop the error signal, and is trained in mutually exclusive time periods with the training of the regular network.

Claims

1. In a control system for an electric arc furnace of a type having electrode means, drive means for changing a position of said electrode means, means for obtaining one or more furnace state measurements, means responsive to said furnace state measurements for establishing one or more state measurement set-points for optimizing operation of said furnace, and furnace regulator means responsive to said set-points and said furnace state measurements to use a selected control algorithm to dynamically calculate a change in electrode position for causing said furnace state measurements to more closely approach their respective set-points, and for causing said drive means to move said electrode means accordingly;

the improvement wherein;

said regulator means includes means for establishing variable coefficients governing said selected control algorithm;

means for presenting said furnace state measurements and set-points and said variable coefficients as inputs to said regulator means;

said regulator means being arranged to calculate a regulator output as a function of said inputs representing said furnace state measurements, said input representing said set-points and said variable coefficients, and to present said regulator output to said drive means to control the position of said electrode means;

means for comparing said set-points and said regulator output for calculating a regulator error signal;

and means responsive to said regulator error signal for practically continuously modifying said variable coefficients accordingly.

2. In a control system for an electric arc furnace of a type having electrode means, drive means for changing a position of said electrode means, means for obtaining one or more furnace state measurements, means responsive to said furnace state measurements for establishing one or more state measurement set-points for optimizing operation of said furnace, and furnace regulator means responsive to said set-points and said furnace state measurements to calculate a change in electrode position for causing said furnace state measurements to more closely approach their respective set-points, and for causing said drive means to move said electrode means accordingly;

the improvement wherein said regulator means comprises;

furnace regulator emulator neural network means having at least one layer and at least one input means, each said layer having at least one neural element, said furnace regulator emulator neural network means also having at least one coupling means for connecting each said neural element to at least one other said neural element or at least one said input means, each said coupling means having a regulator emulator changeable weight controlling a degree of coupling provided thereby;

means for presenting inputs to said furnace regulator emulator neural network means representing said furnace state measurements and said set-points;

said furnace regulator emulator neural network means being arranged to calculate a regulator output as a function of said inputs representing said furnace state measurements, said inputs representing said set-points and said regulator emulator changeable weight, and to present said regulator output to said drive means to control the position of said electrode means;

means for comparing said set-points and said inputs representing said furnace state measurements for calculating a regulator error signal;

and means responsive to said regulator error signal for practically continuously modifying said regulator emulator changeable weight accordingly.

3. In a control system as in claim 2, the further improvement comprising:

clock means for repeatedly obtaining said furnace state measurements and causing said control system to repeatedly recalculate said regulator output and said regulator emulator changeable weight as a function of several of said furnace state measurements obtained at different times.

4. In a control system as in claim 3, the further improvement wherein said means for calculating a regulator error signal comprises:

furnace emulator means for emulating the operation of a furnace;

means for presenting said furnace state measurements, said set-points, sand said regulator output;

said furnace emulator means being arranged to calculate a set of simulated furnace state values as a function of said inputs representing said furnace state measurements, said set-points and said regulator output;

and means for comparing said furnace state measurements and said set-points for calculating said regulator error signal.

5. In a control system as in claim 4, the further improvement wherein:

said furnace emulator means comprises neural network means adapted to emulate a furnace, said neural network means having at least one layer and at least one input means, each said layer having at least one neural element, said neural network means also having at least one coupling means for connecting each said neural element to at least one other said neural element or at least one said input means, each said coupling means having a furnace emulator changeable weight controlling a degree of coupling provided thereby;

means for comparing said furnace state measurements and said simulated furnace state values for calculating a furnace emulator error signal;

and means responsive to said furnace emulator error signal for practically continuously modifying said furnace emulator changeable weight accordingly.

6. In a control system as in claim 5, the improvement further comprising:

clock means for repeatedly obtaining said furnace state measurements and causing said control system to repeatedly recalculate said simulated furnace state values and said furnace emulator changeable weight as a function of several of said furnace state measurements obtained at different times.

7. In a control system as in claim 5, the improvement further comprising:

clock means for repeatedly obtaining said furnace state measurements and causing said control system to repeatedly recalculate said regulator output and said simulated furnace state values as a function of several of said furnace state measurements obtained at different times, thereby producing recalculated values of said regulator output and said simulated furnace state;

and means responsive to said clock means for modifying said regulator emulator changeable weight and said furnace emulator changeable weight in mutually exclusive time periods as functions of said recalculated values, and holding constant said changeable weight of one of said coupling means while modifying said changeable weight of the other of said coupling means.

8. A system for controlling a process, said system comprising:

process emulator means arranged to emulate said process;

a regulator including trainable regulator neural network means arranged to regulate said process;

means for establishing a desired state of said process and presenting said desired state to said regulator neural network means;

said regulator neural network means outputting a process control signal to control said process;

means for presenting said process control signal to said process emulator means;

said process emulator means being arranged to calculate a simulated state of said process as a function of said process control signal;

means for comparing said simulated state with said desired state to derive a process control error signal; and

means for training said regulator neural network means as a function of said process control error signal.

9. A process control system as in claim 8, wherein said process emulator means includes trainable process emulator neural network means;

and further comprising means for obtaining measurements of a present state of said process, means for presenting said present state measurements and said process control signal to said process emulator neural network means, means for comparing said present state measurements with said simulated state to derive a process emulator error signal; and

means for training said process emulator neural network means as a function of said process emulator error signal.

10. A process control system as in claim 8, further comprising:

clock means for repeatedly obtaining said present state measurements and causing said control system to repeatedly recalculate said process control error signal and said simulated state as a function of several of said present state measurements obtained at different times, thereby producing recalculated values of said process control error signal and said simulated state; and

means responsive to said clock means for training of said regulator neural network means and said process emulator neural network means to occur in mutually exclusive time periods and as functions of said recalculated values.

11. A process control system comprising:

at least two trainable neural networks, including a process emulator neural network arranged to emulate a process exhibiting a behavior and a regulator neural network arranged to regulate said process;

means for obtaining a measurement of a present state of said process and presenting said measurement of the present state of said process to said regulator neural network means;

means for establishing a desired state of said process;

said regulator neural network means being arranged to output a process control signal adapted to control said process;

means for presenting the measurement of the present state of said process and said process control signal to said process emulator neural network means, said process emulator neural network means being arranged to calculate a simulated state of said process as a function of said process control signal;

means for comparing said simulated state with said desired state to derive a process control error signal;

means for training said regulator neural network means as a function of said process control error signal;

means for providing a reference signal representing a state of said process and exemplary of the behavior of said process;

means for comparing said reference signal with said simulated state to derive a process emulator error signal;

and means for training said process emulator neural network means as a function of said process emulator error signal.

12. A process control system as in claim 11, further comprising:

clock means for repeatedly obtaining measurements and causing said control system to repeatedly recalculate said process control error signal and said simulated state as a function of several of said measurements obtained at different times, thereby producing recalculated values of said process control error signal and said simulated state;

said regulator neural network and emulator neural network being responsive to said clock means to train of said regulator neural network and emulator neural network to occur in mutually exclusive time periods and as functions of said recalculated values.

13. In a control system for a chaotic process employing at least one regulatable means, means for obtaining at least one present process state measurement, means for establishing at least one state measurement set-point for optimizing operation of said process, and regulator means responsive to said set-point and said present process state measurement to use a selected control algorithm to dynamically calculate a change in a condition of said regulatable means for causing said process state measurement to more closely approach said set-point, and for adjusting said regulatable means accordingly;

the improvement wherein;

said regulator means includes means for establishing variable coefficients governing said selected control algorithm;

means for presenting said present process state measurement, said coefficients, and said set-point inputs to said regulator means;

said regulator means is arranged to calculate a regulator output as a function of said outputs representing said present process state measurement, said set-point and said coefficients, and to present said regulator output to adjust said regulatable means;

means for comparing inputs representing said set-point and inputs representing said process state measurement for calculating a regulator error signal;

and means responsive to said regulator error signal for practically continuously modifying said coefficients accordingly.

14. In a control system as in claim 13, the further improvement comprising:

clock means for repeatedly obtaining said process state measurement and causing said control system to repeatedly recalculate said regulator output and said coefficients as a function of several of said process state measurements obtained at different times.

15. In a control system as in claim 13, the further improvement wherein said means for calculating a regulator error signal comprises:

means for emulating said process;

means for presenting inputs representing said present process state measurement, said set-point, and said regulator output to said process emulator;

said means for emulating said process being arranged to calculate a simulated process state value as a function of said inputs representing said present process state measurement and said regulator output;

and means for comparing said set-point and said simulated input representing said set-point and said simulated process state value for calculating said regulator error signal.

16. In a control system as in claim 15, the further improvement wherein:

said process emulator means comprises process emulator neural network means for realizing a neural network adapted to emulate a process, said process emulator neural network means having at least one layer and at least one input means, each said layer having at least one neural element, said process emulator neural network means also having at least one coupling means for connecting each said neural element to at least one other said neural element or at least one said input means, each said coupling means having a process emulator changeable weight controlling a degree of coupling provided thereby;

means for comparing said input representing said present process state measurement and said simulated process state value for calculating a process emulator error signal;

and means responsive to said process emulator error signal for practically continuously modifying said process emulator changeable weight accordingly.

17. In a control system as in claim 16, the improvement further comprising:

clock means for repeatedly obtaining said process state measurement and causing said control system to repeatedly recalculate said simulated process state value and said process emulator changeable weight as a function of several of said process state measurements obtained at different times.

18. In a control system as in claim 16, the improvement further comprising:

clock means for repeatedly obtaining said process state measurements and causing said control system to repeatedly recalculate said regulator output and said simulated process state values as a function of several of said process state measurements obtained at different times, thereby producing recalculated values of said regulator output and said simulated process state;

and means responsive to said clock means for modifying said regulator emulator changeable weight and said process emulator changeable weight in mutually exclusive time periods as functions of said recalculated values, and holding constant said changeable weight of one of said coupling means while modifying said changeable weight of the other of said coupling means.

19. In a control system for an electric arc furnace of a type having electrode means, an alternating current power supply for said electrode means including a transformer having primary and secondary windings, drive means for changing the position of said electrode means, means for obtaining at least one present furnace state measurement, means responsive to said furnace state measurement for establishing at least one state measurement set-point for optimizing operation of said furnace, and furnace regulator means responsive to said set-point and said present furnace state measurement to use a selected control algorithm to dynamically calculate a change in electrode position for causing said furnace state measurement to more closely approach said set-point, and for causing said drive means to move said electrode means accordingly;

the improvement wherein;

said furnace state measurement means includes means for deriving a measurement of the instantaneous voltage at said primary winding of said power supply transformer;

and means for presenting inputs to said regulator means representing said set-point and said furnace state including at least said derived measurement of the instantaneous voltage at said primary winding;

said regulator means being arranged to calculate a regulator output as a function of said inputs representing said furnace state inputs, and said set-point, and to present said regulator output to said drive means to control the position of said electrode means.

20. A control system as in claim 19 wherein:

said regulator means includes means for establishing modifiable coefficients governing said control algorithm;

and further comprising means for comparing said set-point and said input representing said furnace state for calculating a regulator error signal, and means responsive to said regulator error signal for practically continuously modifying said coefficients accordingly.

21. A control system as in claim 19 wherein said transformer is a three-phase transformer defining a neutral voltage point, said power supply includes three phase conductors connected to said primary winding of said transformer, and said means for deriving a measurement of the instantaneous voltage at said primary winding of said power supply transformer measures the voltage between at least two of said phase conductors.

22. A control system as in claim 21 further comprising means for calculating said derived measurement of the instantaneous voltage at real-time speed, and for presenting a representation thereof as one of said inputs to said regulator means.

23. A control system as in claim 21 wherein said means for deriving a measurement of the instantaneous voltage at said primary winding of said power supply transformer presents at said regulator input representations of the voltages between each of said phase conductors and said neutral point.

24. A control system as in claim 19 wherein said transformer is of a type having three phases and defining a neutral voltage point, said power supply includes three phase conductors connected to said primary winding of said transformer, and said means for deriving a measurement of the instantaneous voltage at said primary winding of said transformer is arranged to determine the voltage between at least one of said phase conductors and said neutral voltage point.

25. A system for controlling a process, said system comprising:

process emulator means arranged to emulate said process;

a regulator including trainable regulator neural network means arranged to regulate said process;

means for establishing a desired state of said process and presenting said desired state to said regulator neural network means;

said regulator neural network means outputting a process control signal to control said process;

means for presenting said process control signal to said process emulator means;

said process emulator means being arranged to calculate a predicted future state of said process as a function of said process control signal;

means for comparing said predicted future state with said desired state to derive a process control error signal;

and means for training said regulator neural network means as a function of said process control error signal.

26. The process control system of claim 25 wherein said process emulator means calculates said predicted state of said process as of a selected future time.

27. The process control system of claim 25 further comprising means for executing process control actions in response to said process control signal from said regulator means;

said means for executing process control actions exhibiting a response time delay between receiving said process control signal and completing execution of said process control actions;

and wherein said process emulator means calculates said predicted state of said process as of a time which is ahead of a present time by an advance interval which is a function of said response time delay.

28. The process control system of claim 27 wherein said advance interval is substantially equal to said response time delay.

29. A process control system as in claim 25, wherein said process emulator means includes trainable process emulator neural network means;

and further comprising means for obtaining measurements of a present state of said process, means for presenting said present state measurements and said process control signal to said process emulator neural network means, means for retaining at least one stored predicted future state which was calculated at a previous time, means for comparing said present state measurements with said stored predicted future state to derive a process emulator error signal; and

means for training said process emulator neural network means as a function of said process emulator error signal.

30. A process control system as in claim 25, further comprising:

clock means for repeatedly obtaining said present state measurements and causing said control system to repeatedly recalculate said process control error signal and said predicted future state as a function of several of said present state measurements obtained at different times, thereby producing recalculated values of said process control error signal and said predicted future state; and

means responsive to said clock means for causing the training of said regulator neural network means and said process emulator neural network means to occur in mutually exclusive time periods and as functions of said recalculated values.