Machine learning apparatus, control device, laser machine, and machine learning method

1. A machine learning apparatus configured to learn a shift amount from being coaxial to being non-coaxial when cutting a workpiece by a laser machine configured to emit a laser beam and an assist gas coaxially and non-coaxially, the machine learning apparatus comprising:

a state-observation section configured to observe machining condition data included in a machining program given to the laser machine, and measurement data of a dimension of dross generated at a cutting spot of the workpiece when the machining program is executed, as a state variable representing a current state of an environment in which the workpiece is cut; and

a learning section configured to learn the shift amount in association with cutting quality of the workpiece, using the state variable.

2. The machine learning apparatus of claim 1 , wherein the machining condition data includes at least one of a material of the workpiece, a thickness of the workpiece, a machining speed at which the workpiece is cut, a nozzle diameter of the laser machine, a supply pressure of the assist gas, a focus position of the laser beam, and an output characteristic value of the laser beam.

3. The machine learning apparatus of claim 1 , wherein the measurement data includes individual dimensions of the dross on both sides of the cutting spot of the workpiece, or a dimension difference between the dross on both sides of the cutting spot.

4. The machine learning apparatus of claim 1 , wherein the state-observation section further observes, as the state variable, measurement data of a kerf-width on a rear surface of the workpiece.

5. The machine learning apparatus of claim 1 , wherein the learning section includes:

a reward calculator configured to obtain a reward associated with the dimension of the dross; and

a function-update section configured to update a function representing a value of the shift amount, using the reward.

6. The machine learning apparatus of claim 5 , wherein the reward calculator obtains the reward that differs in response to a dimension difference between the dross on both sides of the cutting spot of the workpiece.

7. The machine learning apparatus of claim 6 , wherein the reward calculator obtains the reward that differs in response to a difference in the machining condition data, in addition to the dimension difference.

8. The machine learning apparatus of claim 1 , further comprising a decision-making section configured to output a command value of the shift amount to be commanded to the laser machine, based on a learning result by the learning section,

wherein the state-observation section observes the state variable wherein the dimension of the dross formed when the machining program is executed in accordance with the command value is observed as the measurement data for a next learning cycle.

9. A control device configured to control a laser machine configured to emit a laser beam and an assist gas coaxially and non-coaxially, the control device comprising:

the machine learning apparatus of claim 1 ; and

a state-data acquisition section configured to acquire the machining condition data and the measurement data.

10. A laser machine comprising:

a machining head configured to emit a laser beam and an assist gas coaxially and non-coaxially;

a measuring section configured to measure a dimension of dross generated at a cutting spot of a workpiece when a machining program is executed; and

the control device of claim 9 .

11. A machine learning method of learning a shift amount from being coaxial to being non-coaxial when cutting a workpiece by a laser machine configured to emit a laser beam and an assist gas coaxially and non-coaxially, the method comprising:

observing, by a processor, machining condition data included in a machining program given to the laser machine, and measurement data of a dimension of dross generated at a cutting spot of the workpiece when the machining program is executed, as a state variable representing a current state of an environment in which the workpiece is cut; and

learning, by the processor, the shift amount in association with cutting quality of the workpiece, using the state variable.