Transfer learning device and transfer learning method

1 . A transfer learning device, comprising:

processing circuitry performing:

to acquire a plurality of pieces of machining data including data related to time-series machining conditions for causing a machining machine to perform machining for a plurality of steps and data related to a machining state executed by the machining machine for the plurality of steps on a basis of the machining conditions;

to calculate an importance with respect to each piece of machining data acquired on a basis of the plurality of pieces of the machining data acquired;

to extract a plurality of pieces of transfer learning data for performing transfer learning of a first machine learning model that receives, as an input, the data related to the machining conditions in time series for the plurality of steps generated in advance from among the plurality of pieces of the machining data acquired on a basis of the importance calculated and outputs data related to the machining state after execution for the plurality of steps; and

to perform the transfer learning of the first machine learning model by using the plurality of pieces of the transfer learning data extracted, and generate a second machine learning model that receives, as an input, data related to the machining conditions in time series for the plurality of steps and outputs data related to the machining state after execution for the plurality of steps.

2 . The transfer learning device according to claim 1 , the processing circuitry further performing to acquire a plurality of pieces of preliminary learning data including data used at the time of generating the first machine learning model and related to the machining conditions in time series for the plurality of steps and data related to the machining state after execution for the plurality of steps, wherein

the processing circuitry calculates the importance of each piece of machining data acquired on a basis of the plurality of pieces of the machining data acquired and the plurality of pieces of the preliminary learning data acquired.

3 . The transfer learning device according to claim 2 , wherein the processing circuitry includes:

to compress the plurality of pieces of the machining data and the plurality of pieces of the preliminary learning data; and

to calculate the importance for each piece of machining data acquired on a basis of the plurality of pieces of the machining data and the plurality of pieces of the preliminary learning data after being compressed.

4 . The transfer learning device according to claim 1 , the processing circuitry further performing to assign priority to each piece of transfer learning data extracted and select a plurality of pieces of selected transfer learning data from among the plurality of pieces of the transfer learning data on a basis of the priority, wherein

the processing circuitry performs the transfer learning of the first machine learning model using the plurality of pieces of the selected transfer learning data selected.

5 . The transfer learning device according to claim 4 , the processing circuitry further performing:

to evaluate the second machine learning model generated; and

to update the priority assigned to each piece of selected transfer learning data on a basis of an evaluation result of the second machine learning model performed, wherein

the processing circuitry samples the selected transfer learning data on a basis of the priority updated, and

the processing circuitry relearns the second machine learning model on a basis of the selected transfer learning data after the sampling.

6 . The transfer learning device according to claim 1 , wherein

the processing circuitry extracts feature amounts of the plurality of pieces of the transfer learning data, and

the processing circuitry performs the transfer learning of the first machine learning model on a basis of the feature amount extracted.

7 . A transfer learning method, comprising:

acquiring a plurality of pieces of machining data including data related to time-series machining conditions for causing a machining machine to perform machining for a plurality of steps and data related to a machining state executed by the machining machine for the plurality of steps on a basis of the machining conditions;

calculating an importance with respect to each piece of machining data acquired on a basis of the plurality of pieces of the machining data acquired;

extracting a plurality of pieces of transfer learning data for performing transfer learning of a first machine learning model that receives, as an input, the data related to the machining conditions in time series for the plurality of steps generated in advance from among the plurality of pieces of the machining data acquired on a basis of the importance calculated and outputs data related to the machining state after execution for the plurality of steps; and

performing the transfer learning of the first machine learning model by using the plurality of pieces of the transfer learning data extracted, and generating a second machine learning model that receives, as an input, data related to the machining conditions in time series for the plurality of steps and outputs data related to the machining state after execution for the plurality of steps.