Determination device and determination method

1. A determination device comprising:

a memory; and

a processor coupled to the memory and configured to:

obtain sensor data on motion of a device from a plurality of sensors,

extract, from the sensor data, time-series data related to an anomaly based on a threshold value used in detecting the anomaly with use of the sensor data,

determine an analogous relationship between or among the plurality of sensors from the time-series data related to the anomaly and convert the time-series data related to the anomaly into structural data having a graph structure in which the analogous relationship is represented by a line and the plurality of sensors are represented by dots, and

generate, from the structural data, data for each cause of the anomaly using kernel density estimation, and generate a classifier that identifies a cause of the anomaly from the data generated for each cause of the anomaly.

2. The determination device according to claim 1 , wherein the processor is configured to set a value of sensor data within a normal range defined by the threshold value among the sensor data at 0.

3. The determination device according to claim 1 , wherein the structural data is a Gram matrix.

4. The determination device according to claim 3 , wherein the processor is configured to conduct dimensional compression of the structural data.

5. The determination device according to claim 1 , wherein the processor is further configured to:

detect the anomaly with use of the threshold value,

identify a cause of the anomaly with use of the classifier, and

perform output based on a result of detection of the anomaly and an identification result of the cause of the anomaly.

6. A determination method implemented by a computer, comprising:

obtaining sensor data on motion of a device from a plurality of sensors;

extracting time-series data related to an anomaly from the sensor data based on a threshold value used in detecting the anomaly with use of the sensor data;

determining an analogous relationship between or among the plurality of sensors from the time-series data related to the anomaly and converting the time-series data related to the anomaly into structural data having a graph structure in which the analogous relationship is represented by a line and the plurality of sensors are represented by dots; and

generating, from the structural data, data for each cause of the anomaly using kernel density estimation, and generating a classifier that identifies a cause of the anomaly from the data generated for each cause of the anomaly.

7. The determination method according to claim 6 , wherein the extracting includes setting a value of sensor data within a normal range defined by the threshold value among the sensor data at 0.

8. The determination method according to claim 6 , wherein the structural data is a Gram matrix.

9. The determination method according to claim 8 , wherein the generating includes conducting dimensional compression of the structural data.

10. The determination method according to claim 6 , further comprising:

detecting the anomaly with use of the threshold value;

identifying a cause of the anomaly with use of the classifier; and

performing output based on a result of the detecting of the anomaly and an identification result of the cause of the anomaly.

11. A non-transitory computer-readable storage medium storing a determination program causing a computer to execute a process, the process comprising:

obtaining sensor data on motion of a device from a plurality of sensors;

extracting time-series data related to an anomaly from the sensor data based on a threshold value used in detecting the anomaly with use of the sensor data;

determining an analogous relationship between or among the plurality of sensors from the time-series data related to the anomaly and converting the time-series data related to the anomaly into structural data having a graph structure in which the analogous relationship is represented by a line and the plurality of sensors are represented by dots; and

generating, from the structural data, data for each cause of the anomaly using kernel density estimation and generating a classifier that identifies a cause of the anomaly from the data generated for each cause of the anomaly.

12. The non-transitory computer-readable storage medium according to claim 11 , wherein the extracting includes setting a value of sensor data within a normal range defined by the threshold value among the sensor data at 0.

13. The non-transitory computer-readable storage medium according to claim 11 , wherein the structural data is a Gram matrix.

14. The non-transitory computer-readable storage medium according to claim 13 , wherein the generating includes conducting dimensional compression of the structural data.

15. The non-transitory computer-readable storage medium according to claim 11 , the process further comprising:

detecting the anomaly with use of the threshold value;

identifying a cause of the anomaly with use of the classifier; and

performing output based on a result of the detecting of the anomaly and an identification result of the cause of the anomaly.