Numerical controller and machine learning device

A numerical controller configured to predict an abnormality which will occur in a machine to be controlled, the numerical controller comprising a processor configured to:

• acquire information, relating to operation of the machine, as time-series data,
• predict occurrence of the abnormality in the machine based on the time-series data,
• output prediction results on the occurrence of the abnormality in the machine, store results of machine learning of trends in the operation of the machine on occasions of occurrence of abnormalities in a database,
• acquire the time-series data as input data,
• predict the occurrence of the abnormality in the machine based on the input data and the results of the machine learning that are stored in the database, wherein, in the database,
• feature vectors representing the trends in the operation of the machine on the occasions of the occurrence of the abnormalities in the machine and abnormality information relating to the abnormalities are stored in association with each other, and
• results of clustering of the feature vectors into a plurality of clusters are stored, calculate approximate expressions for the time-series data included in the input data, generate a feature vector based on the calculated approximate expressions,
• determine which cluster among the plurality of clusters stored in the database is an approximation to the generated feature vector,
• produce ranking information on abnormalities associated with feature vectors belonging to the cluster determined as the approximation to the feature vector, and
• output the produced ranking information as prediction results on a duration of the occurrence of the abnormality in the machine, wherein
• the feature vector comprises at least machine operation data and the time-series data,
• the input data comprises at least two of motor pressure data, motor position data, or motor speed data,
• the processor is configured to refer to the database and, based on types, start time, and ending time of abnormalities associated with the feature vectors in the cluster that is the most approximate to a current feature vector, calculate scores of the types of the abnormalities based on the following expression:
• Rank of abnormality type in cluster=number of occurrences of abnormality type in cluster×(ending time of abnormality−start time of abnormality)×importance,
• where the importance is a value from 0 to 1, and is set in accordance with urgency on an occasion of occurrence of an abnormality, corresponding to the type of the abnormality, and
• the processor is configured to produce the ranking information in a descending order of the scores and output the ranking information.

A machine learning device configured to conduct machine learning of trends in operation of a machine on occasions of occurrence of abnormalities in the machine, based on time-series data representing information relating to the operation of the machine and acquired by a numerical controller which controls the machine, the machine learning device comprising a processor configured to:

• acquire the time-series data as input data,
• acquire abnormality information relating to the abnormalities which occur in the machine,
• conduct the machine learning of the trends in the operation of the machine on the occasions of the occurrence of the abnormalities, based on the input data and the abnormality information,
• store results of the machine learning in a database;
• calculate approximate expressions for the time-series data included in the input data,
• generate a feature vector based on the calculated approximate expressions,
• store the generated feature vector, associated with the abnormality information, in the database,
• carry out clustering of the feature vector, and
• store results of the clustering in the database, wherein
• the feature vector comprises at least machine operation data and the time-series data,
• the input data comprises motor pressure data, motor position data, and motor speed data,
• when the clustering of the feature vector is carried out, the processor is configured to
• compare distances from center values of a plurality of clusters to a current feature vector,
• select the cluster whose center value is the most approximate to the current feature vector, and
• in response to the distances from the center values of all the plurality of clusters being equal to or greater than a predetermined threshold, determine that there are no clusters approximate to the current feature vector,
• the processor is configured to refer to the database and, based on types, start time, and ending time of abnormalities associated with feature vectors in the cluster that is the most approximate to the current feature vector, calculate scores of the types of the abnormalities based on the following expression:
• Rank of abnormality type in cluster=number of occurrences of abnormality type in cluster×(ending time of abnormality−start time of abnormality)×importance,
• where the importance is a value from 0 to 1, and is set in accordance with urgency on an occasion of occurrence of an abnormality, corresponding to the type of the abnormality, and
• the processor is configured to produce ranking information in a descending order of the scores and output the ranking information.

A machine learning device configured to predict an abnormality which will occur in a machine controlled by a numerical controller, based on time-series data acquired by the numerical controller and representing information relating to operation of the machine, the machine learning device comprising a processor configured to:

• store results of machine learning of trends in the operation of the machine on occasions of occurrence of abnormalities in a database,
• acquire the time-series data as input data,
• predict occurrence of the abnormality in the machine based on the input data and the results of the machine learning that are stored in the database, wherein, in the database,
• feature vectors representing the trends in the operation of the machine on the occasions of the occurrence of the abnormalities in the machine and abnormality information relating to the abnormalities are stored in association with each other, and
• results of clustering of the feature vectors into a plurality of clusters are stored, calculate approximate expressions for the time-series data included in the input data,
• generate a feature vector based on the calculated approximate expressions,
• determine which cluster among the plurality of clusters stored in the database is an approximation to the generated feature vector,
• produce ranking information on abnormalities associated with feature vectors belonging to the cluster determined as the approximation to the feature vector, and
• output the produced ranking information as prediction results on a duration of the occurrence of the abnormality in the machine, wherein
• the feature vector comprises at least machine operation data and the time-series data, and
• the input data comprises at least one of motor pressure data, motor position data, or motor speed data,
• the processor is configured to refer to the database and, based on types, start time, and ending time of abnormalities associated with the feature vectors in the cluster that is the most approximate to a current feature vector, calculate scores of the types of the abnormalities based on the following expression:
• Rank of abnormality type in cluster=number of occurrences of abnormality type in cluster×(ending time of abnormality−start time of abnormality)×importance,
• where the importance is a value from 0 to 1, and is set in accordance with urgency on an occasion of occurrence of an abnormality, corresponding to the type of the abnormality, and
• the processor is configured to produce the ranking information in a descending order of the scores and output the ranking information.