Machine learning device and method for learning a correction value in a motor current control, correction value calculation device comprising a machine learning device and motor drive device

Abstract

Machine learning device (1) that learns correction values ​​associated with a current control based on a current command and a current feedback value in a motor drive device for a three-phase AC motor (104), wherein the machine learning device comprises: a state observation unit (11) that observes a state variable comprising data relating to an error between a rotor position command for the three-phase AC motor (104) and an actual rotor position of the three-phase AC motor (104), data relating to the temperature in the motor drive device, data relating to the temperature of the three-phase AC motor (104) and / or data relating to the voltage of each part of the motor drive device at the time of drive control of the three-phase AC motor (104) by the motor drive device;and a learning unit (12, 16) which learns a current feedback offset correction value used to correct an offset amount contained in the current feedback value, an inter-phase current feedback phase inequality correction value used to correct an inequality between phases in the current feedback value, and a current command correction value used to correct a current command for an insensitivity range to compensate for a lower current amount due to a switching insensitivity range by which switching elements of upper and lower branches in the same phase of an inverter for motor power supply (102) are not switched on simultaneously, according to a training data set defined by the state variable.

Images