Abnormality diagnostic device, abnormality diagnostic method, and program

An abnormality diagnostic method comprising:

• diagnosing, by a computer, a type of abnormality that occurs in an abnormality diagnostic target, which is a device or a system, on the basis of differences between abnormality simulation results for each type of abnormalities obtained by simulating a plurality of types of abnormalities in the abnormality diagnostic target and a plurality of observation results obtained by observing the abnormality diagnostic target using a plurality of detectors,
• wherein
• the diagnosing of the type of abnormality comprises:
• reconstructing the plurality of time-series observation results to multidimensional data in which a combination of time-series observation results of two or more of the plurality of detectors are collected for each observation time among the plurality of time-series observation results, and setting a plurality of combinations of two analysis results among a plurality of analysis results obtained through principal component analysis on the reconstructed multidimensional data;
• deriving a plurality of first trajectories on a phase plane in which a first axis represents a principal component related to one time-series observation result, a second axis represents a principal component analysis result related to the other time-series observation result, and corresponding elements are set as coordinates for each of the plurality of combinations of the two analysis results;
• and
• diagnosing a type or a cause of the abnormality on the basis of a difference between a shape of each of the plurality of first trajectories and a shape of each of a plurality of second trajectories on a phase plane in which a first axis represents a principal component of an abnormality simulation result corresponding to the first axis related to the first trajectory, a second axis represents a principal component of an abnormality simulation result corresponding to the second axis of an observation result related to the first trajectory, and corresponding elements are set as coordinates on the basis of a combination of abnormality simulation results matching a combination of observation results related to each of the plurality of first trajectories among a plurality of abnormality simulation results, and
• wherein
• the abnormality diagnostic method further comprises:
• converting the shape of each of the plurality of second trajectories and the shape of each of the plurality of first trajectories for each type of abnormality according to complex autoregressive models and calculating a plurality of dissimilarity distances between the complex autoregressive models used for the conversion;
• and
• identifying the shape of the second trajectory having a smallest dissimilarity distance from the shape of the first trajectory among the calculated plurality of dissimilarity distances and diagnosing a type of abnormality related to the identified second trajectory as the type of abnormality in the abnormality diagnostic target.

A non-transitory computer-readable storage medium storing a program for causing a computer to:

• diagnose a type of abnormality that occurs in an abnormality diagnostic target, which is a device or a system, on the basis of differences between abnormality simulation results for each type of abnormalities obtained by simulating a plurality of types of abnormalities in the abnormality diagnostic target and a plurality of observation results obtained by observing the abnormality diagnostic target using a plurality of detectors,
• wherein
• the diagnosing of the type of abnormality comprises:
• reconstructing the plurality of time-series observation results to multidimensional data in which a combination of time-series observation results of two or more of the plurality of detectors are collected for each observation time among the plurality of time-series observation results, and setting a plurality of combinations of two analysis results among a plurality of analysis results obtained through principal component analysis on the reconstructed multidimensional data;
• deriving a plurality of first trajectories on a phase plane in which a first axis represents a principal component related to one time-series observation result, a second axis represents a principal component analysis result related to the other time-series observation result, and corresponding elements are set as coordinates for each of the plurality of combinations of the two analysis results;
• and
• diagnosing a type or a cause of the abnormality on the basis of a difference between a shape of each of the plurality of first trajectories and a shape of each of a plurality of second trajectories on a phase plane in which a first axis represents a principal component of an abnormality simulation result corresponding to the first axis related to the first trajectory, a second axis represents a principal component of an abnormality simulation result corresponding to the second axis of an observation result related to the first trajectory, and corresponding elements are set as coordinates on the basis of a combination of abnormality simulation results matching a combination of observation results related to each of the plurality of first trajectories among a plurality of abnormality simulation results, and
• wherein
• the program further causes the computer:
• to convert the shape of each of the plurality of second trajectories and the shape of each of the plurality of first trajectories for each type of abnormality according to complex autoregressive models and to calculate a plurality of dissimilarity distances between the complex autoregressive models used for the conversion;
• and
• to identify the shape of the second trajectory having a smallest dissimilarity distance from the shape of the first trajectory among the calculated plurality of dissimilarity distances and to diagnose a type of abnormality related to the identified second trajectory as the type of abnormality in the abnormality diagnostic target.