Predictive control of rolling mills using neural network gauge estimation

• US 5,586,221 A
• Filed: 07/01/1994
• Issued: 12/17/1996
• Est. Priority Date: 07/01/1994
• Status: Expired due to Fees

• Current Assignee: Syracuse University
• Original Assignee: Syracuse University
• Inventors: Zagrobelny, Joseph, Isik, Can
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): KATBAB, ABDOLLAH
• Application Number:
  US08/270,143
• Time in Patent Office: 900 Days
• Field of Search: 395/22, 395/51, 395/906, 364/148, 364/162, 364/165, 364/183
• US Class Current: 706/23
• CPC Class Codes: B21B 37/16 Control of thickness, width, diameter or other transverse dimensions B21B37/58 takes precedence
• CPC Class Codes: G01B 7/06 for measuring thickness measuring during the manufacture of coatings C23C14/54
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A system for controlling the output of a rolling mill. An intelligent control system is part of a control loop between the mill and a PID controller. The control loop does not rely on the output of an exit gauge sensor in normal operation. The intelligent control system can be an artificial neural network or a parallel cascade network, and has an output node for generating an output signal that is predictive of the exit gauge at a future time. A comparator coupled to the artificial neural network output signal and to a reference signal derives an error signal which is fed to the PID controller for modulating the metal thickness.

Claims

1. A control system for controlling a complex industrial process of the type having a plurality of nonlinear, time-varying states that are mutually coupled in an uncertain manner, the industrial process having a process input, and a process output that is dependent on said time-varying states, the industrial process being responsive to a control signal for changing said process output, the system comprising:

an artificial neural network having an input layer comprising a plurality of input nodes, said input nodes being coupled to state signals that are representative of said time-varying states at a current time and said neural network having a hidden layer and an output node for generating an output signal, said neural network being trained in a training cycle wherein said each state signal is delayed by a predetermined time when presented to said neural network with said process output, thereby sychronizing said process output with past state signals so that said output signal is predictive of said process output at a future time;

comparator means coupled to said output signal and coupled to a reference signal for deriving an error signal that is representative of a difference therebetween; and

control means responsive to said error signal for controlling said industrial process.

2. The system in accordance with claim 1, wherein said control means comprises a proportional integral-derivative controller.

3. The system in accordance with claim 1, wherein said artificial neural network comprises 25 input nodes, 15 hidden nodes, and only one output node, said input nodes being partially interconnected with said hidden nodes.

4. The system in accordance with claim 1, wherein said control means comprises an expert system.

5. A system for controlling the output of a rolling mill of the type having a roller that produces a product of a desired thickness, the mill having an exit gauge for measuring a process output, and a plurality of time varying states, the system comprising:

an artificial neural network having an input layer comprising a plurality of input nodes, said input nodes being coupled to state signals that are representative of said time-varying states at a current time and said neural network having a hidden layer and an output node for generating an output signal, said neural network being trained in a training cycle wherein said each state signal is delayed by a predetermined time when presented to said neural network with said process output, thereby synchronizing said process output with past state signals so that said output signal is predictive of said exit gauge process output at a future time;

comparator means coupled to said output signal and coupled to a reference signal indicative of a desired exit gauge for deriving an error signal that is representative of a difference between said reference signal and said output signal; and

a controller responsive to said error signal for controlling the product thickness.

6. The system in accordance with claim 5, further comprising:

an exit gauge sensor having an output;

a selector for monitoring said output signal of said neural network;

in the event said neural network is operating within predetermined limits, said selector coupling said output signal of said neural network to said comparator means, and in the event said neural network is operating outside said limits said selector coupling said output of said exit gauge sensor to said comparator means.

7. The system in accordance with claim 5, wherein said artificial neural network comprises 25 input nodes, 15 hidden nodes, and only one output node, said input nodes being partially interconnected with said hidden nodes.

8. The system in accordance with claim 5, wherein said controller comprises a proportional integral-derivative controller.

9. A method for controlling the output of a rolling mill of the type having a roller that produces a product of a desired thickness, the mill having a plurality of time varying states and a process output, said method comprising the steps of:

coupling an intelligent control system to state signals that are representative of said time-varying states at a current time, said intelligent control system having an output node for generating an output signal, said intelligent control signal being trained in a training cycle wherein said each state signal is delayed by a predetermined time when presented to said control system with said process output, thereby synchronizing said process output with past state signals so that said output signal is predictive of said process output at a future time;

generating a reference signal indicative of a desired exit gauge;

deriving a first error signal that is representative of a difference between said reference signal and said output signal; and

controlling the product thickness in accordance with said first error signal.

10. The method in accordance with claim 9, wherein said step of modulating is performed with a proportional integral-derivative controller.

11. The method in accordance with claim 9, wherein said step of deriving is performed with a comparator.

12. The method in accordance with claim 9, wherein said step of coupling is performed by coupling an input layer of an artificial neural network to said signals, said neural network having a hidden layer.

13. The method in accordance with claim 12, further comprising the steps of:

measuring an exit gauge at a point beyond a roll gap of the rolling mill;

providing an exit gauge signal responsive to a result of said step of measuring;

determining whether said artificial neural network is operating within predetermined limits;

in the event said artificial neural network is operating outside predetermined limits, deriving a second error signal that is representative of a difference between said reference signal and said exit gauge signal and modulating the product thickness in accordance with said second error signal.

14. The method in accordance with claim 9, wherein said step of coupling is performed by coupling a parallel cascade network to said signals, said parallel cascade network having a plurality of input nodes determined during training thereof, each said input node comprising a dynamic linear element cascaded with a static nonlinear system, said dynamic linear element being derived from a cross correlation between a said time-varying state and the output of the rolling mill.

15. The method in accordance with claim 14, further comprising the steps of:

measuring an exit gauge at a point beyond a roll gap of the rolling mill;

providing an exit gauge signal responsive to a result of said step of measuring;

determining whether said parallel cascade network is operating within predetermined limits;

in the event said parallel cascade network is operating outside predetermined limits, deriving a second error signal that is representative of a difference between said reference signal and said exit gauge signal and modulating the product thickness in accordance with said second error signal.