Machine learning device, numerical control system, setting device, numerical control device, and machine learning method

A setting device configured to

• select an action based on a threshold value that is set beforehand from among actions acquired from a machine learning device, and
• set the action that is selected into a machining program,
• wherein the machine learning device is configured to perform machine learning with respect to a numerical control device configured to cause a machine tool to operate based on the machining program, the machine learning device comprising:
• a memory that stores program;
• and
• a processor configured to execute the program and control the machine learning device to:
• acquire, as the numerical control device executes the machining program set with at least a cutting amount for one time and a cutting rate and causes the machine tool to perform cutting work, state information including the cutting amount for one time and the cutting rate;
• output action information including adjustment information for the cutting amount for one time and the cutting rate included in the state information;
• acquire determination information that is information regarding at least a magnitude of pressure applied to a tool during the cutting work, a shape of a waveform of the pressure applied to the tool, and a period of time taken for the cutting work, and, based on the determination information that has been acquired, to output a reward value used in reinforcement learning depending on a predetermined condition;
• and
• update a value function based on the reward value, the state information, and the action information,
• wherein the predetermined condition is either of a condition for prioritizing machining time and a condition for prioritizing lifetime of the tool,
• the processor outputs a first reward value under the condition for prioritizing machining time and outputs a second reward value under the condition for prioritizing the lifetime of the tool, and
• the processor updates a first value function based on the first reward value, the state information, and the action information under the condition for prioritizing machining time and updates a second value function based on the second reward value, the state information, and the action information under the condition for prioritizing the lifetime of the tool.

A numerical control system comprising:

• a machine learning device;
• a setting device;
• and
• a numerical control device configured to execute a machining program set by the setting device,
• wherein the machine learning device comprises:
• a memory that stores program;
• and
• a processor configured to execute the program and control the machine learning device to:
• acquire, as the numerical control device executes the machining program set with at least a cutting amount for one time and a cutting rate and causes the machine tool to perform cutting work, state information including the cutting amount for one time and the cutting rate;
• output action information including adjustment information for the cutting amount for one time and the cutting rate included in the state information;
• acquire determination information that is information regarding at least a magnitude of pressure applied to a tool during the cutting work, a shape of a waveform of the pressure applied to the tool, and a period of time taken for the cutting work, and, based on the determination information that has been acquired, to output a reward value used in reinforcement learning depending on a predetermined condition;
• and
• update a value function based on the reward value, the state information, and the action information,
• wherein the predetermined condition is either of a condition for prioritizing machining time and a condition for prioritizing lifetime of the tool,
• the processor outputs a first reward value under the condition for prioritizing machining time and outputs a second reward value under the condition for prioritizing the lifetime of the tool, and
• the processor updates a first value function based on the first reward value, the state information, and the action information under the condition for prioritizing machining time and updates a second value function based on the second reward value, the state information, and the action information under the condition for prioritizing the lifetime of the tool.

A numerical control device comprising:

• a machine learning device;
• and
• a setting device,
• a numerical control device being configured to execute a machining program set by the setting device,
• wherein the machine learning device comprises:
• a memory that stores program;
• and
• a processor configured to execute the program and control the machine learning device to:
• acquire, as the numerical control device executes the machining program set with at least a cutting amount for one time and a cutting rate and causes the machine tool to perform cutting work, state information including the cutting amount for one time and the cutting rate;
• output action information including adjustment information for the cutting amount for one time and the cutting rate included in the state information;
• acquire determination information that is information regarding at least a magnitude of pressure applied to a tool during the cutting work, a shape of a waveform of the pressure applied to the tool, and a period of time taken for the cutting work, and, based on the determination information that has been acquired, to output a reward value used in reinforcement learning depending on a predetermined condition;
• and
• update a value function based on the reward value, the state information, and the action information,
• wherein the predetermined condition is either of a condition for prioritizing machining time and a condition for prioritizing lifetime of the tool,
• the processor outputs a first reward value under the condition for prioritizing machining time and outputs a second reward value under the condition for prioritizing the lifetime of the tool, and
• the processor updates a first value function based on the first reward value, the state information, and the action information under the condition for prioritizing machining time and updates a second value function based on the second reward value, the state information, and the action information under the condition for prioritizing the lifetime of the tool.

A machine learning method for a machine learning device configured to perform machine learning with respect to a numerical control device configured to cause a machine tool to operate based on a machining program, the machine learning device comprising a memory that stores program and a processor configured to execute the program and perform the machine learning method comprising:

• acquiring, as the numerical control device executes the machining program set with at least a cutting amount for one time and a cutting rate and causes the machine tool to perform cutting work, state information including the cutting amount for one time and the cutting rate;
• outputting action information including adjustment information for the cutting amount for one time and the cutting rate included in the state information;
• acquiring determination information that is information regarding at least a magnitude of pressure applied to a tool during the cutting work, a shape of a waveform of the pressure applied to the tool, and a period of time taken for the cutting work and, based on the determination information that has been acquired, outputting a reward value used in reinforcement learning depending on a predetermined condition;
• and
• updating a value function based on the reward value, the state information, and the action information,
• wherein the predetermined condition is either of a condition for prioritizing machining time and a condition for prioritizing lifetime of the tool,
• the processor outputs a first reward value under the condition for prioritizing machining time and outputs a second reward value under the condition for prioritizing the lifetime of the tool, and
• the processor updates a first value function based on the first reward value, the state information, and the action information under the condition for prioritizing machining time and updates a second value function based on the second reward value, the state information, and the action information under the condition for prioritizing the lifetime of the tool.