Neural network control of spot welding

• US 6,018,729 A
• Filed: 09/17/1997
• Issued: 01/25/2000
• Est. Priority Date: 09/17/1997
• Status: Expired due to Fees

• Current Assignee: Lockheed Martin Energy Research Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Energy Research Corporation (Martin Marietta Corporation), General Motors Corporation (Motors Liquidation Co. GUC Trust)
• Inventors: Schroth, James G., Zacharia, Thomas, Allen, John D. Jr.
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): STARKS, WILBERT L
• Application Number:
  US08/932,439
• Time in Patent Office: 860 Days
• Field of Search: 706/21, 706/20, 706/23, 219/117, 219/110, 219/130, 219/121
• US Class Current: 706/21
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A spot welder comprises a neural network for processing, in real time, current and voltage energizing a weld in progress. The neural network generates a predicted time of optimal weld strength and/or nugget size for the weld in progress. A controller terminates the weld in progress at the predicted time. A method for controlling a spot welder comprises the steps of: sensing in real time current and voltage energizing a spot weld in progress; predicting a time of optimal weld strength and/or nugget size with a neural network responsive to the sensed current and voltage; and, terminating the weld in progress at the predicted time. A sensor for electromotive forces (EMF) induced by the spot welder can generate a signal for canceling out a large fraction of EMF components in at least one or both of the current and voltage signals. EMF components are substantially precluded in the current signal if the current sensor uses a buried shunt. Termination of the weld in progress at the predicted time is prevented when the predicted time precedes a predetermined minimum weld duration. The weld in progress is terminated at a predetermined maximum weld duration when the predicted time is after the predetermined maximum weld duration.

Claims

1. A spot welder, comprising:

at least two electrodes having tips engageable with a stack-up of material;

a power supply for energizing said electrodes to spot weld said stack-up when said electrodes are engaged with said stack-up;

a current sensor coupled to said power supply and generating a first signal representative of current supplied to said electrodes;

a voltage sensor coupled to said power supply and generating a second signal representative of voltage across said electrodes and said stack-up;

a neural network for processing, in real time, data representative of said first and second signals for a weld in progress according to a learning rule determined by data from previous welds, said neural network generating a predicted time of optimal weld strength for said weld in progress; and

,

a controller responsive to said neural network for terminating said weld in progress responsive to said predicted optimal time.

2. The spot welder of claim 1, further comprising a third sensor for detecting electromotive forces (EMF) induced by said spot welder and generating a third signal for substantially canceling out EMF components in at least one of said first and second signals.

3. The spot welder of claim 2, wherein:

said current sensor comprises a buried shunt, substantially precluding said EMF components in said first signal; and

,

said third signal cancels out said EMF components in said second signal.

4. The spot welder of claim 3, further wherein said host receives said set-up instructions from a production control system.

5. The spot welder of claim 1, wherein said learning rule is updated by said data representative of said first and second signals from successive welds.

6. The spot welder of claim 1 further comprising a robot arm for positioning said electrodes responsive to said controller.

7. The spot welder of claim 1, wherein said controller further comprises:

an analog to digital converter (ADC) for said current and voltage signals;

a digital signal processor (DSP) for filtering and amplifying digital signals from said ADC for use by said neural network;

a neural network controller for generating a stop current signal at said predicted time of optimal weld strength;

a welder controller responsive to said stop current signal for generating a digital stop current command; and

,

a digital to analog converter (DAC) responsive to said digital stop current command for supplying an analog stop current command to said power supply.

8. The spot welder of claim 7, wherein the controller further comprises a host for providing set-up instructions to said ADC, said DSP, said neural network controller, said welder controller, and said DAC.

9. The spot welder of claim 1 wherein said controller is programmed with a minimum weld time and a maximum weld time, said minimum weld time representing a minimum length of time before said controller terminates said weld in progress, and said maximum weld time representing a maximum length of time before said controller terminates said weld, said controller being responsive to said predicted time of optimal weld strength only between said minimum and maximum weld times.

10. An improved controller for a spot welder having current and voltage sensors for generating respective signals representative of current and voltage levels during a spot weld in progress, the improvement comprising:

a neural network for processing, in real time, data representative of said respective signals for said spot weld in progress according to a learning rule determined by data from previous welds, said neural network generating a predicted time of optimal weld strength for said weld in progress; and

,

a controller responsive to said neural network for terminating said weld in progress responsive to said predicted time of optimal strength.

11. The improved controller of claim 10, further comprising a further sensor for electromotive forces (EMF) induced by said spot welder and generating a further signal for substantially canceling out EMF components in at least one of said respective signals.

12. The improved controller of claim 11, wherein:

said current sensor comprises a buried shunt, substantially precluding said EMF components in said first signal; and

,

said further signal cancels out said EMF components in said signal representative of voltage.

13. The improved spot welder controller of claim 11, wherein said controller is programmed with a minimum weld time and a maximum weld time, said minimum weld time representing a minimum length of time before said controller terminates said weld in progress, and said maximum weld time representing a maximum length of time before said controller terminates said weld, said neural network controller being responsive to said predicted time of optimal weld strength only between said minimum and maximum weld times.

14. The improved spot welder controller of claim 10, wherein said controller is programmed with a minimum weld time and a maximum weld time, said minimum weld time representing a minimum length of time before said controller terminates said weld in progress, and said maximum weld time representing a maximum length of time before said controller terminates said weld, said neural network controller being responsive to said predicted time of optimal weld strength only between said minimum and maximum weld times.

15. The improved spot welder controller of claim 10, wherein said learning rule is updated by said data representative of said first and second signals from successive welds.

16. A method for controlling a spot welder, comprising the steps of:

sensing in real time current and voltage levels energizing a spot weld in progress;

predicting a time of optimal weld strength with a neural network responsive to said sensed current and voltage levels according to a learning rule determined by data from previous welds; and

,

terminating said weld in progress responsive to said predicted time of optimal weld strength.

17. The method of claim 16, further comprising the steps of:

sensing in real time electromotive forces (EMF) induced by said spot welder; and

,

substantially canceling out EMF components in at least one of said sensed current and voltage levels in accordance with said sensed EMF prior to processing of said current and voltage levels by said neural network.

18. The method of claim 17, further comprising the steps of:

sensing said current with a buried shunt, substantially precluding said EMF components in said first signal; and

,

substantially canceling out said EMF components in said second signal.

19. The method of claim 17, further comprising the steps of:

preventing termination of said weld in progress responsive to said predicted time when said predicted time is prior to a predetermined minimum weld duration; and

,

terminating said weld in progress after a predetermined maximum weld duration when said predicted time is after said predetermined maximum weld duration.

20. The method of claim 16, further comprising the steps of:

preventing termination of said weld in progress responsive to said predicted time when said predicted time is prior to a predetermined minimum weld duration; and

,

terminating said weld in progress after a predetermined maximum weld duration when said predicted time is after said predetermined maximum weld duration.

21. A spot welder, comprising:

at least two electrodes having tips engageable with a stack-up of material;

a power supply for energizing said electrodes to spot weld said stack-up when said electrodes are engaged with said stack-up;

a current sensor coupled to said power supply and generating a first signal representative of current supplied to said electrodes;

a voltage sensor coupled to said power supply and generating a second signal representative of voltage across said electrodes and said stack-up;

a neural network for processing, in real time, data representative of said first and second signals for a weld in progress according to a learning rule determined by data from previous welds, said neural network generating a predicted time of optimal nugget size for said weld in progress; and

,

a controller responsive to said neural network for terminating said weld in progress responsive to said predicted time of optimal nugget size.

22. A method for controlling a spot welder, comprising the steps of:

sensing in real time current and voltage levels energizing a spot weld in progress;

predicting a time of optimal weld strength with a neural network responsive to said sensed current and voltage levels according to a learning rule determined by data from previous welds; and

,

terminating said weld in progress responsive to said predicted time of optimal nugget size.