Learning device, learning method, and program therefor for shorten time in generating appropriate teacher data

A learning method executed in a learning device in a control system which comprises:

• a controller which outputs a command value obtained by compensating a target value based on a compensation amount;
• and
• a servo mechanism controlled to perform a predetermined process on an object to be processed, wherein a command value output by the controller is input to the servo mechanism, and the servo mechanism outputs a control variable as a response to the command value,
• wherein the learning device provides, to an adjuster including a learned neural network model which is an artificial intelligence (AI) model learned to output the predetermined compensation amount to the controller based on a specific parameter of the object to be processed, the learned neural network model, and
• wherein the learning method is executed to make the learning device perform:
• generating a candidate compensation amount for controlling the servo mechanism based on operation data including the target value, the command value and the control variable, performing learning with the generated candidate compensation amount and the specific parameter of the object to be processed as teacher data, and generating or updating a learned neural network model;
• and
• providing the learned neural network model that has been generated or updated to the adjuster,
• adopting the candidate compensation amount as the teacher data when the generated candidate compensation amount is not a value deviating by a predetermined threshold value or more from a candidate compensation amount included in the teacher data used by the learned neural network model for learning, and
• adopting the candidate compensation amount as the teacher data when the generated candidate compensation amount is in the range of a candidate compensation amount included in the teacher data used by the learned neural network model for learning.

A non-transient computer-readable recording medium, recording a program for operating a control system which comprises:

• a learning device;
• a controller which outputs a command value obtained by compensating a target value based on a compensation amount;
• and
• a servo mechanism controlled to perform a predetermined process on an object to be processed, wherein a command value output by the controller is input to the servo mechanism, and the servo mechanism outputs a control variable as a response to the command value,
• wherein the learning device provides, to an adjuster including a learned neural network model which is an artificial intelligence (AI) model learned to output the predetermined compensation amount to the controller based on a specific parameter of the object to be processed, the learned neural network model, and
• wherein the program operates the learning device to perform:
• generating a candidate compensation amount for controlling the servo mechanism based on operation data including the target value, the command value and the control variable, performing learning with the generated candidate compensation amount and the specific parameter of the object to be processed as teacher data, and generating or updating a learned neural network model;
• and
• providing the learned neural network model that has been generated or updated to the adjuster,
• adopting the candidate compensation amount as the teacher data when the generated candidate compensation amount is not a value deviating by a predetermined threshold value or more from a candidate compensation amount included in the teacher data used by the learned neural network model for learning, and
• adopting the candidate compensation amount as the teacher data when the generated candidate compensation amount is in the range of a candidate compensation amount included in the teacher data used by the learned neural network model for learning.