On-line training neural network system for process control

• US 5,212,765 A
• Filed: 08/03/1990
• Issued: 05/18/1993
• 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:
  US07/563,092
• Time in Patent Office: 1,019 Days
• Field of Search: 364/513, 364/148, 364/149, 364/150, 364/151, 364/164, 364/165, 364/803, 364/200, 364/474.15, 364/500, 364/501, 364/502, 364/503, 395/22, 395/23, 395/11, 395/68, 395/906
• US Class Current: 417/44.2
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

An on-line training neural network for process control system and method trains by retrieving training sets from the stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network. When multiple presentations are needed to effectively train, a buffer of training sets is filled and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed. An historical database of timestamped data can be used to construct training sets when training input data has a time delay from sample time to availability for the neural network. The timestamps of the training input data are used to select the appropriate timestamp at which input data is retrieved for use in the training set. Using the historical database, the neural network can be trained retrospectively by searching the historical database and constructing training sets based on past data.

Claims

1. A computer neural network process control system adapted for predicting output data provided to 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:

(1) a neural network, connected to retrieve training input data from a data source, and connected to retrieve input data from a data source in accordance with time specifications, and connected to store output data to a data source, comprising;

(a) specification storing means for storing specifications for said training input data, said input data, and said output data;

(b) weight storing means for storing weights for said neural network;

(c) training means for adjusting said weights; and

(d) predicting means for predicting said output data in accordance with said input data and said weights; and

(2) monitoring means, responsive to said training input data, for monitoring, which comprises;

(a) comparing means for detecting new training input data;

(b) computing means for determining said time specifications for said input data; and

(c) triggering means, responsive to said comparing means, for triggering in accordance with said new training input data, said triggering means connected to initiate training by said training means of said neural network.

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

(1) at least one module, comprising;

(i) at least one neural network module, connected to retrieve training input data and connected to retrieve input data in accordance with time specifications, and connected to store said output data, comprising;

(a) specification storing means for storing specifications for said input data, said training input data, and said output data;

(b) weight storing means for storing weights for said neural network module;

(c) training means for adjusting said weights; and

(d) predicting means for predicting output data in accordance with said input data and said weights; and

(2) modular timing and sequencing means, responsive to module data specifications, and connected to retrieve data in accordance with said data specifications, comprising;

(i) neural network module timing means, comprising;

(a) comparing means for detecting new training input data;

(b) computing means for determining said time specifications for said input data; and

(c) triggering means for initiating training, said triggering means connected to initiate training by said training means of said neural network.

3. The modular computer neural network process control system of claim 2, wherein said neural network module further comprises neural network specifications.

4. The modular computer neural network process control system of claim 3, further comprising:

(3) configuration means for entering said neural network specifications, comprising;

(a) a template presenting a partial neural network specification; and

(b) input means for accepting user input via said template to complete said partial neural network specification.

5. The modular computer neural network process control system of claim 2, wherein said modular timing and sequencing means further comprises:

(ii) sequencing means for controlling said execution of said at least one module in accordance with module sequence specifications.

6. The modular computer neural network process control system of claim 2, wherein said modular timing and sequencing means further comprises:

(ii) timer means for controlling the execution of said at least one module, in accordance with time specifications.

7. The modular computer neural network process control system of claim 2, wherein said at least one module further comprises,

(ii) a feedback module, and wherein said modular timing and sequencing means further comprises

(iii) feedback module timing means for controlling the execution of said feedback module.

8. The modular computer neural network process control system of claim 7, wherein said at least one module further comprises:

(3) a sensor connected to sense a condition in the process;

(4) an actuator connected to change a controllable state of the process; and

(5) a controller connected to use data from said sensor as its controller input data for computing controller output data in accordance with said controller input data and in accordance with a parameter, and connected to send said controller output data to said actuator; and

wherein at least one of said at least one modules is connected to change said parameter of said controller.

9. The modular computer neural network process control system of claim 2, wherein said at least one module further comprises,

(ii) an expert system module, and wherein said modular timing and sequencing means further comprises

(iii) expert system module timing means for controlling the execution of said expert system module.

10. The modular computer neural network process control system of claim 2, wherein said at least one module further comprises

(ii) a feedforward module, and wherein said modular timing and sequencing means further comprises

(iii) feedforward module timing means for controlling the execution of said feedforward module.

11. The modular computer neural network process control system of claim 2, wherein said at least one module further comprises

(ii) a statistical test module, and wherein said modular timing and sequencing means further comprises

(iii) statistical test module timing means for controlling the execution of said statistical test module.

12. The modular computer neural network process control system of claim 2, wherein said at least one module further comprises

(ii) an event module, and wherein said modular timing and sequencing means further comprises

(iii) event module timing means for controlling the execution of said feedforward module.

13. The modular computer neural network process control system of claim 2, further comprising:

(3) an historical database, connected to provide said training input data and/or said input data, and/or connected to store a history of said output data with associated timestamps.

14. A computer neural network process control system adapted for predicting output data provided to 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:

(1) stack means for storing at least two training sets;

(2) a neural network, connected to retrieve said training sets from said stack and connected to store output data, comprising;

(a) specification storing means for storing specifications for input data, training input data, and said output data;

(b) weight storing means for storing weights for said neural network;

(c) training means, for adjusting said weights in accordance with said training sets; and

(d) predicting means for predicting said output data in accordance with said input data and said weights;

(3) constructing means, responsive to said training input data, for constructing said neural network, comprising;

(a) comparing means for detecting new training input data;

(b) computing means for determining time specifications for said input data;

(c) retrieval means for retrieving said new training input data, and for retrieving said input data in accordance with said time specifications;

(d) bumping means for removing a training set from said stack means, and for storing said retrieved new training input data and said input data as a training set in said stack means; and

(e) triggering means for initiating training, said triggering means connected to initiate training by said training means of said neural network.