Apparatus using a neural network for power factor calculation

• US 5,450,315 A
• Filed: 09/26/1994
• Issued: 09/12/1995
• Est. Priority Date: 09/26/1994
• Status: Expired due to Term

• Current Assignee: Square D Company (Schneider Electric SE)
• Original Assignee: Square D Company (Schneider Electric SE)
• Inventors: Stefanski, John J.
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): Garland, Steven R.
• Application Number:
  US08/312,342
• Time in Patent Office: 351 Days
• Field of Search: 364/148, 364/477, 364/482, 364/483, 364/551.01, 364/551.02, 395/21-23, 395/24, 219/108-110, 219/114, 219/117.1, 219/130.01, 219/130.21
• US Class Current: 700/48
• CPC Class Codes: B23K 11/252 using digital means
• CPC Class Codes: B23K 31/006 relating to using of neural networks

Abstract

A weld controller utilizes a neural network to compute power factor of a secondary circuit of a welding transformer supply power to a workpiece through a pair of welding electrodes. Phase controlled switches supply power to the transformer and the neural network uses the phase angle at the time of energization of the switches and the length of time that the switches conduct to compute the power factor. The computed power factor is compared with previous computations of power factor online and any changes are interpreted as changes in resistance of the secondary circuit. This provides a measure of contact wear and the weld controller can compensate for these changes by increasing weld power. The neural network is adaptable for use with other types of control systems.

Claims

1. An apparatus for calculating power factor of a control system utilizing phase controlled switches supplying output power to an inductive and resistive load, said apparatus comprising:

A a microprocessor for generating firing pulses for every half cycle period of conduction for controlling operation of said phase controlled switches, said firing pulses having variable firing angle delay times;

B a first detector for determining a first time interval alpha representative of said firing angle delay time for energizing said phase controlled switches;

C a second detector for determining a second time interval gamma representative of a total conduction time of said phase controlled switches for each half cycle period of conduction;

D a neural network operating in said microprocessor, said neural network for calculating said power factor, said network comprising three layers fully connected including an input layer for receiving data representing alpha and gamma, a hidden layer having a single neuron and an output layer having a single neuron for computing said power factor, and for scaling and providing constant data relating to said power factor; and

E control means within said microprocessor for using said data relating to said power factor with a predetermined schedule for varying said firing angle delay times.

2. The apparatus of claim 1 wherein said neural network is trained off-line using backpropagation techniques to compute values for network weights and biases, said training using data from a preprocessed solution of equations relating power factor to alpha and gamma.

3. The apparatus of claim 2 wherein said equations relating power factor to alpha and gamma are

4. The apparatus of claim 1 wherein said control system is a resistance welder.

5. The apparatus of claim 4 wherein said resistance welder includes a stepper program for increasing weld power as contact tips deteriorate, and wherein said calculated power factor for use in determining actual weld power.

6. The apparatus of claim 4 wherein said resistance welder includes a means of compensating for contact tip wear and uses said calculated power factor for determining a change in resistance of said output contacts and associated interconnecting wiring by comparing consecutive power factor calculations to detect changes in said power factor, said changes to reflect changes in resistance requiring said compensation.

7. An apparatus for controlling the quality of a weld in a resistance weld controller having a pair of welding electrodes connected to a secondary winding of welding transformer and an AC source connected to a primary winding of said welding transformer, said AC source for supplying alternating polarity half cycles of welding current to said weld transformer using phase controlled switches, said apparatus comprising:

A means for measuring a first time interval representative of a firing angle delay time for energizing said phase controlled switches;

B means for measuring a second time interval representative of a total conduction time of said phase controlled switches for each half cycle period of conduction;

C a neural network for calculating a power factor, said network comprising three layers fully connected including an input layer for receiving data representing said first and second time intervals, a hidden layer having a single neuron and an output layer having a single neuron for computing said power factor, and for scaling and providing constant data relating to said power factor; and

D control means for using said data relating to said power factor with a predetermined weld schedule for varying said firing angle delay times.

8. The apparatus of claim 7 wherein said neural network is trained off-line using backpropagation techniques to compute values for network weights and biases, said training using data from a preprocessed solution of equations relating power factor to said first time interval and said second time interval.

9. The apparatus of claim 8 wherein said equations relating power factor to said first time interval and said second time interval are

10. The apparatus of claim 7 wherein said resistance welder includes a stepper program for increasing weld power as contact tips deteriorate, and wherein said calculated power factor for use in determining actual weld power.

11. The apparatus of claim 7 wherein said resistance welder includes a means of compensating for contact tip wear and uses said calculated power factor for determining a change in resistance of said output contacts and associated interconnecting wiring by comparing consecutive power factor calculations to detect changes in said power factor, said changes to reflect changes in resistance requiring said compensation.

12. An apparatus for use with a resistance welder for calculating power factor of a secondary circuit of a weld transformer supplying weld current to a pair of weld electrodes, said weld transformer receiving power through a phase controlled switch, said apparatus comprising:

A a neural network for learning in advance samples of time dependent signals relating to a first time interval representing a phase angle delay time for energizing said phase controlled switch and to a second time interval relating to length of time said phase controlled switch conducts and for processing said first and second time intervals to produce an output signal corresponding to said power factor when actual data representing said first and second time intervals are inputted to said neural network;

B a first means for inputting to said neural network data representing said first time interval during operation of said resistance welder;

C a second means for inputting to said neural network data representing said second time interval during operation of said resistance welder; and

D means for outputting from said neural network data representing said calculated power factor for use of and during operation of said resistance welder.

13. The apparatus of claim 12 wherein said neural network comprises a plurality of identical three layer, fully connected neural networks, each of said plurality of three layer networks including an input layer for receiving data representing said second time interval, a hidden layer and an output layer having a single neuron for computing said power factor, and for scaling and providing constant data relating to said power factor.

14. The apparatus of claim 13 wherein each of said plurality of three layer networks represents one set of weights and biases for a particular value of said first time interval.

15. The apparatus of claim 14 wherein said neural network is trained off-line using backpropagation techniques to compute values for said sets of weights and biases, said training using data from a preprocessed solution of equations relating power factor to said first time interval and said second time interval.

16. The apparatus of claim 15 wherein said equations relating power factor to said first time interval and said second time interval are