Machine learning device learning failure occurrence mechanism of laser device

A machine learning device that is connected with at least one laser device to be mutually communicable with the laser device via a network, the laser device comprising:

• at least one laser oscillator;
• a power source unit that supplies driving current to the laser oscillator;
• at least one output light sensor that detects light output of laser light emitted from the laser oscillator;
• and
• a control unit that outputs at least a current output command corresponding to a light output command to the power source unit and receives a detection signal from the output light sensor, wherein
• the machine learning device comprises:
• a state observation unit that observes a state variable of an inside and an outside of the laser device, the state variable including time-series data of light output which is detected by the output light sensor and the light output command, through the control unit of the laser unit;
• a determination result acquisition unit that acquires a determination result on correctness with respect to a quantitative failure occurrence mechanism, the quantitative failure occurrence mechanism being outputted by the machine learning device for each failure in the laser device;
• a learning unit that receives output from the state observation unit and output from the determination result acquisition unit and learns the quantitative failure occurrence mechanism corresponding to the each failure while associating the quantitative failure occurrence mechanism with the state variable of the laser device, the state variable being observed by the state observation unit, and the determination result on correctness with respect to the quantitative failure occurrence mechanism, the determination result being acquired by the determination result acquisition unit;
• and
• a decision making unit that refers to a learning result of the learning unit when an occurrence of the each failure is detected through at least collation between the time-series data of the light output and the light output command and decides the quantitative failure occurrence mechanism which is to be outputted from the machine learning device.