Detecting occurrence of abnormality

Patent metadata

Field	Value
Publication number	US-9625354-B2
Application number	US-201214345415-A
Country	US
Kind code	B2
Filing date	Jul 27, 2012
Priority date	Sep 21, 2011
Publication date	Apr 18, 2017
Grant date	Apr 18, 2017

Abstract

A method, apparatus and computer program for detecting occurrence of an anomaly. The method can exclude arbitrariness and objectively judge whether a variation of a physical quantity to be detected is abnormal or not even when an external environment is fluctuating. The method includes acquiring multiple primary measurement values from a measurement target. Further, calculating and a reference value for each of the multiple primary measurement values by optimal learning. The method further includes calculating a relationship matrix which indicates mutual relationships between the multiple secondary measurement values. Further the method includes calculating an anomaly score for each of the secondary measurement value which indicates the degree of the measurement target being abnormal. The anomaly score is calculated by comparing the secondary measurement value with a predictive value which is calculated based on the relationship matrix and other secondary measurement values.

First claim

The invention claimed is:

1. A method for detecting an occurrence of an abnormality in a measurement target, the method comprising:
acquiring multiple primary measurement values from the measurement target;
calculating multiple reference values so that a reference value is calculated for each primary measurement value using optimal learning;
calculating multiple secondary measurement values with the multiple primary measurement values and the corresponding multiple reference values, wherein the multiple secondary measurement values are calculated by subtracting the corresponding multiple reference values from the multiple primary measurement values;
calculating a relationship matrix indicating a plurality of mutual relationships among the multiple secondary measurement values; and
calculating an anomaly score for each secondary measurement value, which indicates the degree the measurement target is abnormal by comparing a predictive value with the secondary measurement value, wherein the predictive value is calculated based on the relationship matrix and other multiple secondary measurement values.

2. The method according to claim 1 , wherein said reference value is calculated by using a linear mapping and maximizing a projection element of a normal sample in terms of a projection matrix.

3. The method according to claim 1 , wherein said relationship matrix is calculated by calculating a weighted adjacency matrix from a graph indicating a relationship among said multiple secondary measurement values and performing maximum aposteriori probability estimation of a normal distribution using a Laplace prior distribution.

4. The method according to claim 3 , wherein:
the absolute value of a weight of said weighted adjacency matrix increases as the strength of the relationship among said multiple secondary measurement values increases; and
the weight is zero when there is no relationship among the multiple secondary measurement values.

5. The method according to claim 3 , further comprising outputting said graph.

6. The method according claim 1 , wherein said predictive value is calculated for each secondary measurement value by using a logarithmic loss according to a conditional distribution of the other multiple secondary measurement values.

7. An apparatus that detects whether an abnormality has occurred in a measurement target, the apparatus comprising:
primary measurement value acquisition means for acquiring multiple primary measurement values from the measurement target;
reference value calculation means for calculating multiple reference values so that a reference value is calculated for each primary measurement value using optimal learning;
secondary measurement value calculation means for calculating multiple secondary measurement values with the multiple primary measurement values and the corresponding multiple reference values, wherein the multiple secondary measurement values are calculated by subtracting the corresponding multiple reference values from the multiple primary measurement values;
relationship matrix calculation means for calculating a relationship matrix indicating a plurality of mutual relationships among the multiple secondary measurement values; and
anomaly score calculation means for calculating an anomaly score for each secondary measurement value, which indicates the degree the measurement target is abnormal by comparing a predictive value with the secondary measurement value, wherein the predictive value is calculated based on the relationship matrix and the other multiple secondary measurement values.

8. The apparatus according to claim 7 , wherein said reference value calculation means calculates said reference value by using a linear mapping and maximizing a projection element of a normal sample in terms of a projection matrix.

9. The apparatus according to claim 7 , wherein said relationship matrix calculation means calculates said relationship matrix by calculating a weighted adjacency matrix from a graph which indicates a relationship among said multiple secondary measurement values and performing maximum aposteriori probability estimation of a normal distribution using a Laplace prior distribution.

10. The apparatus according to claim 9 , wherein said relationship matrix calculation means configures said weighted adjacency matrix so that the absolute value of a weight of said weighted adjacency matrix increases as the strength of the relationship among said multiple secondary measurement values increases, and the weight is zero when there is no relationship among the secondary measurement values.

11. The apparatus according to claim 9 , further comprising output means for outputting said graph.

12. The apparatus according to claim 7 , wherein said abnormality degree calculation means calculates said predictive value for each secondary measurement value by using a logarithmic loss according to a conditional distribution of the other multiple secondary measurement values.

13. A computer readable non-transitory article of manufacture tangibly embodying computer readable instruction which, when executed, cause a computer to carry out the steps of a method, the method comprising:
acquiring multiple primary measurable values from the measurement target;
calculating multiple reference values so that a reference value for each primary measurement value using optimal learning;
calculating multiple secondary measurement values with the multiple primary measurement values and corresponding multiple reference values, wherein the multiple secondary values are calculated by subtracting the corresponding multiple reference values from the multiple primary measurement values;
calculating a relationship matrix indicating a plurality of mutual relationships among the multiple secondary measurement values; and
calculating an anomaly score, for each secondary measurement value, which indicates the degree the measurement target is abnormal by comparing a predictive value with the secondary measurement value, wherein the predictive value is calculated based on the relationship matrix and other multiple secondary measurement values.

14. The computer according to claim 13 , further comprising: calculating said reference value by using a linear mapping and maximizing a projection element of a normal sample in terms of a projection matrix.

15. The computer according to claim 13 , wherein said relationship matrix is calculated by calculating a weighted adjacency matrix from a graph indicating a relationship among said multiple secondary measurement values and performing maximum aposteriori probability estimation of a normal distribution using a Laplace prior distribution.

16. The computer according to claim 15 , further comprising outputting said graph.

17. The computer according to claim 13 , wherein said predictive value is calculated for each secondary measurement value by using a logarithmic loss according to a conditional distribution of the other secondary measurement values.

Assignees

• Ide Tsuyoshi
• Morimura Tetsuro
• Globalfoundries Inc

Inventors

• Ide Tsuyoshi
• Morimura Tetsuro

Classifications

• G07C5/0808

  Diagnosing performance data (testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005) · CPC title

• G01M99/008 Primary

  by doing functionality tests · CPC title

• G06Q10/06

  Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling · CPC title

• G07C3/00

  Registering or indicating the condition or the working of machines or other apparatus, other than vehicles · CPC title

• G06Q10/20 Primary

  Administration of product repair or maintenance · CPC title