Computer neural network supervisory process control system and method

• US 5,142,612 A
• Filed: 08/03/1990
• Issued: 08/25/1992
• Est. Priority Date: 08/03/1990
• Status: Expired due to Term

• Current Assignee: Rockwell Automation Technologies Incorporated (Allen-Bradley Co., Inc.)
• Original Assignee: E. I. du Pont de Nemours and Company (Corteva, Inc.)
• Inventors: Skeirik, Richard D.
• Primary Examiner(s): MacDonald, Allen R.
• Application Number:
  US07/562,268
• Time in Patent Office: 753 Days
• Field of Search: 364/513, 364/148, 364/149, 364/150, 364/151, 364/164, 364/165, 364/807, 364/200, 364/474.15, 364/500, 364/501, 364/502, 364/503, 395/11, 395/906, 395/22, 395/23, 395/68, 395/27, 395/26
• US Class Current: 706/23
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

A neural network for adjusting a setpoint in process control replaces a human operator. The neural network operates in three modes: training, operation, and retraining. In operation, the neural network is trained using training input data along with input data. The input data is from the sensor(s) monitoring the process. The input data is used by the neural network to develop output data. The training input data are the setpoint adjustments made by a human operator. The output data is compared with the training input data to produce error data, which is used to adjust the weights of the neural network so as to train it. After training has been completed, the neural network enters the operation mode. In this mode, the present invention uses the input data to predict output data used to adjust the setpoint supplied to the regulatory controller. Thus, the operator is effectively replaced. The present invention in the retraining mode utilizes new training input data to retrain the neural network by adjusting the weight(s).

Claims

1. A computer neural network process control method adapted for predicting output data provided to a controller used to control a process for producing a product having at least one product property, said method allowing a control aim to be set without a human operator, the computer neural network process control method comprising the steps of:

(1) operating the process with one or more sensors connected to sense process conditions and produce process condition measurements, and changing a controllable process state with an actuator,

(2) controlling said actuator with a process controller in accordance with a process condition measurement from one or more of said sensors and in accordance with a setpoint;

(3) configuring the neural network by specifying at least one input, at least one output, at least one training input, and at least one specified interval;

(4) training the neural network to produce a trained neural network comprising the substeps of;

(a) retrieving a first raw training input data;

(b) retrieving a second raw training input data;

(c) computing a corresponding first training input data based on said first raw training input data and said second raw training input data, said first training input data indicative of the action of a human operator of the process;

(d) retrieving a first input data,

(e) predicting a first output data using said first input data,

(f) computing a first error data in accordance with said first training input data and said first output data, and

(g) training the neural network to produce said trained neural network in accordance with said first error data;

(5) at said at least one specified interval, retrieving a second input data and predicting, with said trained neural network, second output data using said second input data; and

(6) retrieving said second output data for changing a setpoint of the controller for controlling the process.

2. The computer neural network process control method of claim 1, wherein step (3) comprises configuring the neural network by having at least two layers and/or up to full connectivity and/or at least one weight.

3. The computer neural network process control method of claim 1, wherein step (3) comprises configuring the neural network to have feedback and/or feedforward connectivity.

4. The computer neural network process control method of claim 1, wherein said step (4)(e) predicts said first output data using said first input data and a transfer function of said configured neural network of step (1).

5. The computer neural network process control method of claim 1, wherein said step (4)(f) comprises computing said first error data by subtracting said first training input data from said first output data.

6. The computer neural network process control method of claim 1, wherein said step (4)(g) comprises training said neural network by using said first error data to adjust a first weight data.

7. The computer neural network process control method of claim 6, wherein said training utilizes a back propagation method.

8. The computer neural network process control method of claim 1, wherein said step (5) comprises a step of using said at least one specified interval selected in accordance with a time constant for control of the process.

9. The computer neural network process control method of claim 1, wherein step (2) further comprises a step of computing, using the controller, controller output data in accordance with said process condition measurement and said setpoint.

10. The computer neural network process control method of claim 1, wherein step (6) of retrieving further comprises a step of changing said setpoint directly in accordance with said second output data.

11. The computer neural network process control method of claim 1, further comprising a step of using said setpoint for said first raw training input data used in step (4) (a) and/or said second raw training input data used in step (4) (b).

12. The computer neural network process control method of claim 11, wherein step (4) further comprises a step (a) (i) of using a human operator to make an adjustment to said controller setpoint, wherein said first training input data is indicative of said adjustment.

13. The computer neural network process control method of claim 1, wherein said retrieving steps (4) (a), (4) (b) and/or (4) (d) further comprises the step of retrieving from an historical database.

14. The computer neural network process control method of claim 13, wherein step (4) (b) comprises the step of retrieving said second raw training input data from said historical database with an associated first timestamp.

15. The computer neural network process control method of claim 14, wherein step (4) (d) retrieves said first input data from said historical database using said associated first timestamp.

16. The computer neural network process control method of claim 14, further comprising a step of using a clock for producing said associated first timestamp of step (4) (b).

17. The computer neural network process control method of claim 14, wherein step (4) (d) further comprises a step of retrieving said first input data with an associated second timestamp selected in accordance with said associated first timestamp.

18. The computer neural network process control method of claim 1, further comprising the steps of:

(7) retrieving a second raw training input data with a corresponding second input data; and

(8) retraining said trained neural network in accordance with a second training input data computed using said second raw training input data, a second error data computed using said second training input data and a second input data predicted using said corresponding second input data.

19. The computer neural network process control method of claim 18, wherein said step (8) further comprises a step of stopping the use of step (6) for changing said setpoint when said second error data exceeds a metric.

20. The computer neural network process control method of claim 18, wherein step (7) of retrieving further comprises the step of retrieving past output data.

21. The computer neural network process control method of claim 1, wherein step (4) (d) comprises a step of using an aim value for a product property value as said first input data.

22. The computer neural network process control method of claim 20, wherein step (3) of configuring further comprises a step of specifying said at least one specified interval to expire whenever a change in said aim value occurs.

23. The computer neural network process control method of claim 1, wherein step (4) (d) comprises a step of using an aim value for a process condition as said first input data.

24. The computer neural network process control method of claim 20, wherein step (3) of configuring further comprises a step of specifying said at least one specified interval to expire whenever a change in said aim value occurs.

25. The computer neural network process control method of claim 1, wherein said first input data of step (4) (d) comprises the step of retrieving an associated second timestamp.

26. The computer neural network process control method of claim 25, further comprising a step of using a lab to provide said first input data and to provide said associated second timestamp indicating the time at which a lab sample was taken from the process and was used to produce said first input data.

27. The computer neural network process control method of claim 1, further comprising the steps:

(5) saving said first output data;

(6) retrieving a second raw training input data;

(7) computing a second training input data using said second raw training input data;

(8) retraining the neural network when said second training input data is different from said saved first output data.

28. The computer neural network process control method of claim 1, further comprising a step of using said setpoint for said first input data used in step (4) (d) and/or said second input data used in step (5).

29. A computer neural network process control system adapted for predicting output data provided to control a setpoint of a controller used to control a process for producing a product having at least one product property, the computer neural network process control system comprising:

(a) a neural network, responsive to a first input data indicative of sensor or aim or lab data, comprising,

(1) training means, responsive to a first training input data indicative of the action of a human operator of the process, for training said neural network in accordance with a training set having said first input data and said first training input data to produce trained neural network weights, and

(2) predicting means using said trained neural network weights for predicting the output data in accordance with a second input data indicative of sensor or aim or lab data; and

(b) said controller, responsive to said predicting means, comprising,

(1) sending means for adjusting said setpoint in accordance with the output data, and

(2) setpoint adjustment means for accepting a change to said setpoint made by said human operator.

30. The computer neural network process control system of claim 29, further comprising storage means for storing said first training input data as a past operator action value, and

wherein said training means is connected to retrieve said past operator action value as said first input data.

31. The computer neural network process control system of claim 29, wherein said neural network further comprises (3) computing means for computing error data in accordance with said first training input data and said first input data,, said error data used by said training means to train said neural network.

32. The computer neural network process control system of claim 31, wherein said predicting means is connected to retrieve said weights for predicting the output data, and wherein said training means is connected to modify said weights in accordance with said error data of said computing means.

33. The computer neural network process control system of claim 29, wherein said controller further comprises

(3) input means for receiving controller input data indicative of sensor or aim or lab data;

and

(4) computing means for computing controller output data in accordance with controller input data and said setpoint, and

(5) sending means for sending said controller output data to an actuator for changing a controllable process state.

34. The computer neural network process control system of claim 33, further comprising:

(c) one or more sensors for sensing process conditions and for producing process condition measurements, and

wherein said controller is connected to use one or more of said process condition measurements as said controller input data.

35. The computer neural network process control system of claim 33, wherein said predicting means further comprises:

(a) output data storage means for storing an output data value;

wherein said predicting means is connected to store the output data in said output data storage means; and

wherein said training means further comprises,

(v) comparing means for retrieving the output data in said output data storage means, and connected to execute said weight adjusting means when said first training input data is different from the output data.

36. The computer neural network process control system of claim 29, wherein said controller comprises a hardware system, and said neural network comprises a separate hardware system.

37. The computer neural network process control system of claim 29, wherein said neural network comprises a software system adapted to run on a computer platform.

38. The computer neural network process control system of claim 29, wherein said training means is connected to retrieve the setpoint of the controller to compute said first training input data.

39. The computer neural network process control system of claim 29, wherein said training means further comprises:

(i) storage means for saving a past raw training input data value indicative of a past value of said setpoint;

(ii) retrieving means for retrieving a current raw training input value indicative of a current value of said setpoint;

(iii) computing means for computing a first training input data indicative of a change between said current raw training input value and said past raw training input data value and for saving said current raw training input value in said storage means; and

(iv) weight adjusting means, responsive to said first training input data, for training said neural network weights in accordance with a training set having said first input data and said first training input data.

40. The computer neural network process control system of claim 29, further comprising storage means for storing the output data as a control action value, and

wherein said training means is connected to retrieve said control action value as said first input data.

41. The computer neural network process control system of claim 29, further comprising lab means for receiving a sample from the process, which said lab means provides said first input data based on said sample.

42. The computer neural network process control system of claim 29, further comprising sensor means, responsive to said neural network means, for providing said first input data.

43. A computer neural network process control system adapted for predicting output data provided to control a setpoint of a controller used to control a process for producing a product having at least one product property, the computer neural network process control system comprising:

(a) a neural network, responsive to a first input data indicative of sensor or aim or lab data, comprising,

(1) training means, responsive to a first training input data indicative of the action of a human operator of the process, for training said neural network in accordance with a first training set having said first input data and said first training input data to produce trained neural network weights, and

(2) predicting means using said trained neural network weights for predicting the output data in accordance with a second input data indicative of sensor or aim or lab data;

(b) said controller, responsive to said predicting means, comprising,

(1) sending means for adjusting said setpoint in accordance with the output data; and

(2) setpoint adjustment means for accepting a change to said setpoint made by said human operator; and

(c) an historical database comprising,

(1) storing means for storing said first training input data with an associated first timestamp, and for storing said first input data indicated by said associated first timestamp, and

(2) retrieving means, responsive to said storing means, for retrieving said first training set comprising said first training input data and said first input data indicated by said associated first timestamp, and connected to provide said first training set to said training means.

44. The computer neural network process control system of claim 43, further comprising lab means for receiving a sample from the process, which said lab means provides said first input data based on said sample.

45. The computer neural network process control system of claim 44, further comprising clock means, associated with said lab means, for providing said associated first timestamp to said storing means.

46. The computer neural network process control system of claim 43, further comprising sensor means, responsive to said neural network means, for providing said first input data.

47. The computer neural network process control system of claim 46, further comprising clock means for providing an associated second timestamp of said first input data to said historical database, and

wherein said retrieving means selects said first input data indicated by said associated first timestamps using said associated second timestamp.

48. The computer neural network process control system of claim 43, further comprising output means for storing the output data in said historical database.

49. A computer neural network process control system adapted for predicting output data provided to control a setpoint of a controller used to control a process for producing a product having at least one product property, the computer neural network process control system comprising:

(a) a neural network, responsive to a first input data indicative of sensor or aim or lab data, comprising,

(1) predicting means using said trained neural network weights for predicting the output data in accordance with a second input data indicative of sensor or aim or lab data; and

(b) said controller, responsive to said predicting means, comprising,

(1) sending means for adjusting said setpoint in accordance with the output data, and

(2) setpoint adjustment means for accepting a change to said setpoint made by aid human operator.

50. A computer neural network process control system adapted for predicting output data provided to control a setpoint of a controller used to control a process for producing a product having at least one product property, the computer neural network process control system comprising:

(a) a neural network, responsive to a first input data indicative of sensor or aim or lab data, comprising,

(1) training means, responsive to a first training input data indicative of the action of a human operator of the process, for training said neural network in accordance with a training set having said first input data and said first training input data to produce trained neural network weights, and

(2) predicting means using said trained neural network weights for predicting the output data in accordance with a second input data indicative of sensor or aim or lab data; and

(b) a controller, responsive to said predicting means, comprising,

(1) setpoint adjustment means for accepting a change to said setpoint made by said human operator.