Training data screening device, robot system, and training data screening method

1 . A training data screening device for screening training data for constructing a learning model by machine learning from collected data collected by a data collector, the training data screening device comprising:

a data evaluation model constructed by machine learning on at least a part of the collected data, or by machine learning on data different from the collected data;

a data evaluator that evaluates the input collected data using the data evaluation model;

a memory for storing evaluated data, which is the collected data evaluated by the data evaluator; and

a training data screener that screens the training data for constructing the learning model from the evaluated data stored by the memory by an instruction of an operator to whom an evaluation result of the data evaluator is presented, or automatically screens the training data based on the evaluation result,

wherein

the collected data includes time series information of measured values based on detection values obtained by at least one sensor provided with a machine to be controlled,

the data evaluator evaluates the collected data for each partial time series information being the time series information corresponding to a part of time of the time series information of the detection values, using the data evaluation model, and

the data evaluation model is constructed to output an evaluation value corresponding to each of a plurality of reference operations, which are subdivisions of the operations performed by the operator, when the partial time series information is input.

2 . The training data screening device according to claim 1 , comprising:

an evaluation presenter that presents the evaluation result of the data evaluator to the operator; and

an instruction receiver that receives an instruction from the operator regarding whether or not to screen the evaluated data as the training data, wherein

the training data screener screens the training data for constructing the learning model based on the instruction input to the instruction receiver.

3 . The training data screening device according to claim 1 , wherein

if the best evaluation value among the evaluation values output by the data evaluation model with respect to each of the plurality of reference operations when the partial time series information is input is better than a threshold value, the data evaluator assigns correspondence information to the collected data, indicating that it is evaluated that the partial time series information corresponds to the reference operation with the best evaluation value, and stores the collected data with the correspondence information in the memory as the evaluated data.

4 . The training data screening device according to claim 3 , wherein

the evaluated data is presented for screening by the operator or used for automatic screening from the evaluated data, in a manner in which the correspondence information as the evaluation result is assigned.

5 . The training data screening device according to claim 4 , wherein

the evaluated data is presented for screening by the operator from the evaluated data so that each range of the partial time series information to which the correspondence information as the evaluation result is assigned can be distinguished.

6 . The training data screening device according to claim 5 , wherein

the evaluated data including, in addition to the range in which the best evaluation value among the evaluation values output by the data evaluation model with respect to each of the plurality of reference operations is better than the threshold value, the range in which the best evaluation value is not better than the threshold value is presented for screening by the operator from the evaluated data.

7 . The training data screening device according to claim 6 , wherein

the operator can specify the range in which the best evaluation value among the evaluation values output by the data evaluation model with respect to each of the plurality of reference operations is not better than the threshold value, and the operator can assign to the specified range, correspondence information indicating an operation that is not included in the plurality of reference operations.

8 . The training data screening device of claim 5 , wherein

the evaluated data is visually presented for screening by the operator from the evaluated data in the form of a graph representing the detection values by the sensor or information based on the detection values.

9 . The training data screening device of claim 5 , wherein

the training data screener can screen the training data from the evaluated data for each range of the partial time series information to which the correspondence information has been assigned as the evaluation result.

10 . A robot system comprising:

a learning model constructed by machine learning using the training data screened by the training data screening device according to claim 1 ; and

a robot that performs work based on the output of the learning model.

11 . A training data screening method for screening training data for constructing a learning model by machine learning from collected data collected by a data collector, the training data screening method comprising:

a data evaluation step for evaluating the input collected data using a data evaluation model constructed by machine learning on at least a part of the collected data, or by machine learning on data different from the collected data;

a storing step for storing evaluated data, which is the collected data evaluated in the data evaluation step; and

a training data screening step for screening the training data for constructing the learning model from the evaluated data stored in the storing step by an instruction of an operator to whom an evaluation result in the data evaluation step is presented, or automatically screening the training data based on the evaluation result,

wherein

the collected data includes time series information of measured values based on detection values obtained by at least one sensor provided with a machine to be controlled,

the data evaluation step evaluates the collected data for each partial time series information being the time series information corresponding to a part of time of the time series information of the detection values, using the data evaluation model, and

the data evaluation model is constructed to output an evaluation value corresponding to each of a plurality of reference operations, which are subdivisions of the operations performed by the operator, when the partial time series information is input.

12 . The training data screening method according to claim 11 , wherein

the data evaluation step includes assigning a label if an evaluation value is equal to or greater than a predetermined threshold value.