US Patent No. 10,148,211

MACHINE LEARNING APPARATUS AND METHOD FOR LEARNING CORRECTION VALUE IN MOTOR CURRENT CONTROL, CORRECTION VALUE COMPUTATION APPARATUS INCLUDING MACHINE LEARNING APPARATUS AND MOTOR DRIVING APPARATUS


Patent No. 10,148,211
Issue Date December 04, 2018
Title Machine Learning Apparatus And Method For Learning Correction Value In Motor Current Control, Correction Value Computation Apparatus Including Machine Learning Apparatus And Motor Driving Apparatus
Inventorship Sou Saito, Yamanashi (JP)
Taku Sasaki, Yamanashi (JP)
Tsutomu Shikagawa, Yamanashi (JP)
Assignee FANUC CORPORATION, Yamanashi (JP)

Claim of US Patent No. 10,148,211

1. A motor driving apparatus with current control based on a current command and a current feedback value in a three-phase AC motor, the motor driving apparatus comprising:a state observation unit that observes a state variable including at least one of data regarding an error between a rotor position command to the three-phase AC motor and a rotor actual position of the three-phase AC motor, data regarding temperature in the motor driving apparatus, data regarding temperature of the three-phase AC motor, and data regarding a voltage of the motor driving apparatus;
a learning unit that learns a current feedback offset correction value used for correcting an offset amount included in the current feedback value, an inter-current-feedback-phase unbalance correction value used for correcting an unbalance between phases in the current feedback value, and a current command correction value for a dead zone used for correcting a current command in order to compensate a decreased amount of current due to a switching dead zone by which switching elements of upper and lower arms in the same phase of an inverter for motor power supply provided in the motor driving apparatus are not simultaneously turned on, in accordance with a training data set defined by the state variable;
a decision-making unit which decides the current command correction value for a dead zone, on the basis of a result of learning by the learning unit in accordance with the training data set in response to an input of the state variable;
a current command correction unit for a dead zone which corrects the current command for the three-phase AC motor by using the current command correction value for a dead zone, and generates a corrected current command; and
a current control unit which controls driving of the three-phase AC motor based on the corrected current command,
wherein the voltage includes at least one of an input AC voltage input into the motor driving apparatus, DC link voltage between a rectifier provided in the motor driving apparatus and rectifying the input AC voltage and the inverter for motor power supply, and control voltage used to drive a control apparatus provided in the motor driving apparatus,
wherein the learning unit comprises:
a reward computation unit that computes a reward on the basis of the error between the rotor position command to the three-phase AC motor and the rotor actual position of the three-phase AC motor; and
a function update unit that updates, on the basis of the state variable and the reward, functions for computing the current feedback offset correction value, the inter-current-feedback-phase unbalance correction value, and the current command correction value for a dead zone, and
wherein the reward computation unit increases the reward when the error between the rotor position command to the three-phase AC motor and the rotor actual position of the three-phase AC motor is smaller than an error observed by the state observation unit before the present error, and the reward computation unit decreases the reward when the error is larger than the error observed before.