Machine learning device, numerical control device and machine learning method for learning threshold value of detecting abnormal load

Patent metadata

Field	Value
Publication number	US-2017357243-A1
Application number	US-201715609336-A
Country	US
Kind code	A1
Filing date	May 31, 2017
Priority date	Jun 9, 2016
Publication date	Dec 14, 2017
Grant date	—

Abstract

A machine learning device for learning a threshold value of detecting an abnormal load in a machine tool, includes a state observation unit, and a learning unit. The state observation unit observes a state variable obtained based on at least one of information about a tool, main spindle revolution rate, and amount of coolant of the machine tool, material of a workpiece, and moving direction, cutting speed, and cut depth of the tool, and the learning unit learns the threshold value of detecting an abnormal load based on training data created from an output of the state observation unit and data related to detection of an abnormal load in the machine tool and on teacher data.

First claim

What is claimed is:

1 . A machine learning device for learning a threshold value of detecting an abnormal load in a machine tool, comprising:
a state observation unit that observes a state variable obtained based on at least one of information about a tool, main spindle revolution rate, and amount of coolant of the machine tool, material of a workpiece, and moving direction, cutting speed, and cut depth of the tool; and a learning unit that learns the threshold value of detecting an abnormal load based on training data created from an output of the state observation unit and data related to detection of an abnormal load in the machine tool and on teacher data.

2 . The machine learning device according to claim 1 , wherein the learning unit comprises:
an error calculation unit that calculates an error between the training data and the teacher data; and a learning model updating unit that updates a learning model for defining an error of a condition correlated to a detection of an abnormal load in the machine tool based on an output of the state observation unit and an output of the error calculation unit.

3 . The machine learning device according to claim 1 , wherein
the training data is data related to a predicted load current value, which is a value of a prediction of a current through a load, and the teacher data is data related to a measured load current value, which is a value of an actual measurement of a current through the load.

4 . The machine learning device according to claim 1 , wherein the information about a tool comprises information on type, material, and diameter of the tool.

5 . The machine learning device according to claim 1 , wherein the machine learning device is on a cloud server.

6 . The machine learning device according to claim 1 , wherein the machine learning device is on a fog server.

7 . The machine learning device according to claim 1 , wherein the machine learning device is connectable to at least one other machine learning device, and mutually exchange or share an outcome of machine learning with the at least one other machine learning device.

8 . The machine learning device according to claim 1 , wherein the machine learning device comprises a neural network.

9 . A numerical control device that comprises the machine learning device according to claim 1 and controls the machine tool, wherein the numerical control device detects an abnormal load in the machine tool based on the threshold value of detecting an abnormal load learned by the machine learning device.

10 . The numerical control device according to claim 9 , wherein the numerical control device judges that an abnormal load has been detected in the machine tool when a measured load current value, which is a value of an actual measurement of a current through a load, is greater than the sum of the threshold value of detecting an abnormal load and a predefined offset amount.

11 . The numerical control device according to claim 10 , wherein
the numerical control device operates in phases comprising a learning phase for learning the threshold value of detecting an abnormal load and an application phase for controlling the machine tool to actually machine the workpiece, learning for the threshold value of detecting an abnormal load is performed in the learning phase, based on respective load current values for machining conditions, by performing prearranged exercise machining, and an abnormal load is detected in the machine tool in the application phase by comparing the learned threshold value of detecting an abnormal load with the measured load current value, which is a value of an actual measurement of a current through the load.

12 . The numerical control device according to claim 11 , wherein the learning for the threshold value of detecting an abnormal load is performed also in the application phase.

13 . A machine learning method for learning a threshold value of detecting an abnormal load in a machine tool, comprising:
observing a state variable obtained based on at least one of information about a tool, main spindle revolution rate, and amount of coolant of the machine tool, material of a workpiece, and moving direction, cutting speed, and cut depth of the tool; and learning the threshold value of detecting an abnormal load, based on training data created from the state variable and data related to detection of an abnormal load in the machine tool and on teacher data.

14 . The machine learning method according to claim 13 , wherein learning the threshold value of detecting an abnormal load comprises:
calculating an error between the training data and the teacher data; and updating a learning model for defining an error of a condition correlated to a detection of an abnormal load in the machine tool based on the state variable and the calculated error between the training data and the teacher data.

15 . The machine learning method according to claim 13 , wherein
the training data is data related to a predicted load current value, which is a value of a prediction of a current through a load, and the teacher data is data related to a measured load current value, which is a value of an actual measurement of a current through the load.

Assignees

• Fanuc Corp

Inventors

• Takayama Kanta
• Sato Kazuo
• Maeda Hideaki

Classifications

• G05B2219/49307

  Learn, learn operational zone, feed, speed to avoid tool breakage · CPC title

• G06N3/08

  Learning methods · CPC title

• G05B19/4065 Primary

  Monitoring tool breakage, life or condition · CPC title

• G05B19/40938

  Tool management · CPC title

• G05B2219/39352

  Feedback error learning, ffw ann compensates torque, feedback from pd to ann · CPC title