Motor control apparatus in which power regeneration are adjusted, controller for converter, machine learning apparatus and method thereof

Patent metadata

Field	Value
Publication number	US-10135255-B2
Application number	US-201615046966-A
Country	US
Kind code	B2
Filing date	Feb 18, 2016
Priority date	Jul 31, 2015
Publication date	Nov 20, 2018
Grant date	Nov 20, 2018

Abstract

A machine learning apparatus includes a state observation unit which observes a state variable including at least one of data associated with the value of a DC voltage in a DC link, data associated with the amount of power supply regeneration representing the amount of power returned from the DC link to an AC power supply by a power supply regeneration operation by a converter, and data associated with the occurrence or nonoccurrence of an overvoltage alarm indicating that the DC voltage of the DC link has exceeded a preset overvoltage alarm level, and a learning unit learns conditions associated with the power supply regeneration operation by the converter, in accordance with a training data set defined by the state variable.

First claim

What is claimed is:

1. A machine learning apparatus for a controller to drive a motor within a machine or a robot, the learning apparatus is configured to learn a condition associated with a power supply regeneration operation for converting DC power in a DC link used to generate AC power for motor driving into AC power, and returning the AC power to an AC power supply, in a converter which generates DC power in the DC link by conversion from AC power in the AC power supply, the machine learning apparatus comprising:
a processor configured to
observe a state variable comprising at least one of
(1) data associated with a value of a DC voltage in the DC link, wherein the DC voltage is obtained by a DC voltage measuring circuit connected to the DC link,
(2) data associated with an amount of power supply regeneration representing an amount of power returned from the DC link to the AC power supply by the power supply regeneration operation by the converter, wherein the amount of power supply regeneration is obtained by a power supply regeneration amount measuring circuit connected to a link between the AC power supply and the converter, and
(3) data associated with an occurrence or a nonoccurrence of an overvoltage alarm indicating that the DC voltage of the DC link has exceeded a preset overvoltage alarm level, wherein the occurrence or nonoccurrence of the overvoltage alarm is obtained from an overvoltage alarm receiving circuit connected to the DC voltage measuring circuit; and

learn a condition associated with the power supply regeneration operation by the converter, in accordance with a training data set defined by the state variable,

wherein the converter is controlled in accordance with the learned condition to convert the DC power in the DC link into the AC power to be returned to the AC power supply,
wherein the processor is further configured to:
compute a reward based on the state variable, and
update, based on the reward, a function for changing a power supply regeneration starting voltage representing a voltage defined in the DC link as a criterion for starting a power supply regeneration operation by the converter, and a power supply regeneration stopping voltage representing a voltage defined in the DC link as a criterion for stopping the power supply regeneration operation started by the converter.

2. The machine learning apparatus according to claim 1 , wherein the processor is configured to increase the reward when a current DC voltage of the DC link is higher than a previous DC voltage of the DC link.

3. The machine learning apparatus according to claim 1 , wherein the processor is configured to reduce the reward when a current DC voltage of the DC link is lower than a previous DC voltage of the DC link.

4. The machine learning apparatus according to claim 1 , wherein the processor is configured to increase the reward when a current amount of power supply regeneration is smaller than a previous amount of power supply regeneration.

5. The machine learning apparatus according to claim 1 , wherein the processor is configured to reduce the reward when a current amount of power supply regeneration is larger than a previous amount of power supply regeneration.

6. The machine learning apparatus according to claim 1 , wherein the processor is configured to increase the reward when the nonoccurrence of the overvoltage alarm is observed.

7. The machine learning apparatus according to claim 1 , wherein the processor is configured to reduce the reward when the occurrence of the overvoltage alarm is observed.

8. The machine learning apparatus according to claim 1 , the processor is configured to learn the condition in accordance with the training data set obtained for a plurality of converters.

9. A controller for a converter, the controller comprising the machine learning apparatus according to claim 1 , the controller configured to
decide a power supply regeneration starting voltage and a power supply regeneration stopping voltage in response to input of a current state variable of the state variable, based on a result of learning by the processor in accordance with the training data set; and
control the converter to perform power conversion between the AC power in the AC power supply and the DC power in the DC link,
wherein the controller is configured to
issue an instruction to the converter to start a power supply regeneration operation when the DC voltage of the DC link rises and becomes higher than the power supply regeneration starting voltage decided by the controller, and
after start of the power supply regeneration operation by the converter, issue an instruction to the converter to stop the power supply regeneration operation when the DC voltage of the DC link drops and becomes lower than the power supply regeneration stopping voltage decided by the controller.

10. The controller according to claim 9 , wherein the processor is configured to re-learn and update the condition in accordance with an additional training data set defined by the current state variable.

11. The controller according to claim 9 , further comprising:
the DC voltage measuring circuit configured to measure the DC voltage of the DC link;
the power supply regeneration amount measuring circuit configured to measure the amount of power supply regeneration; and
the overvoltage alarm receiving circuit configured to receive the data associated with the occurrence or nonoccurrence of the overvoltage alarm.

12. A motor control apparatus, comprising:
the controller for a converter according to claim 9 ; and
an inverter controller configured to control an inverter to perform power conversion between the DC power in the DC link and AC power on an AC motor side for driving a motor.

13. A machine learning method for a controller to drive a motor within a machine or a robot, the learning apparatus is configure to learn a condition associated with a power supply regeneration operation for converting DC power in a DC link used to generate AC power for motor driving into AC power, and returning the AC power to an AC power supply, in a converter which generates DC power in the DC link by conversion from AC power supplied from the AC power supply, the method comprising:
observing by a processor a state variable comprising at least one of
(1) data associated with a value of a DC voltage in the DC link, wherein the DC voltage is obtained by a DC voltage measuring circuit connected to the DC link,
(2) data associated with an amount of power supply regeneration representing an amount of power returned from the DC link to the AC power supply by the power supply regeneration operation by the converter, wherein the amount of power supply regeneration is obtained by a power supply regeneration amount measuring circuit connected to a link between the AC power supply and the converter, and
(3) data associated with an occurrence or a nonoccurrence of an overvoltage alarm indicating that the DC voltage of the DC link has exceeded a preset overvoltage alarm level, wherein the occurrence or nonoccurrence of the overvoltage alarm is obtained from an overvoltage alarm receiving circuit connected to the DC voltage measuring circuit; and

learning by the processor a condition associated with the power supply regeneration operation by the converter, in accordance with a training data set defined by the state variable,
wherein the converter is controlled in accordance with the learned condition to convert the DC power in the DC link into the AC power to be returned to the AC power supply,
wherein the method further comprises:
computing a reward based on the state variable; and
upda

Assignees

• Fanuc Corp

Inventors

• Kuroki Wataru

Classifications

• H02P23/12

  Observer control, e.g. using Luenberger observers or Kalman filters · CPC title

• H02J4/00 Primary

  Circuit arrangements for mains or distribution networks not specified as AC or DC; Circuit arrangements for mains or distribution networks combining AC and DC sections or sub-networks (arrangements using intermediate DC-AC-DC conversion H02J1/002; arrangements using high-voltage DC [HVDC] links H02J3/36) · CPC title

• H02P23/0004

  Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control · CPC title

• G05B13/028

  using expert systems only · CPC title

• H02M1/32

  Means for protecting converters other than automatic disconnection · CPC title