Evaluating Anomaly For One Class Classifiers In Machine Condition Monitoring

• US 20070143038A1
• Filed: 11/27/2006
• Published: 06/21/2007
• Est. Priority Date: 12/07/2005
• Status: Active Grant

• Current Assignee: Siemens AG
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Neubauer, Claus, Yuan, Chao
• Granted Patent: US 7,567,878 B2
• Time in Patent Office: Days
• US Class Current: 702/35
• CPC Class Codes: G05B 19/4065 Monitoring tool breakage, life or condition
• CPC Class Codes: G05B 2219/31357 Observer based fault detection, use model
• CPC Class Codes: G05B 2219/34477 Fault prediction, analyzing signal trends
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

A method for monitoring machine conditions provides additional information using a one-class classifier in which an evaluation function is learned. In the method, a distance is determined from an anomaly measurement x to at boundary of a region R₁ containing all acceptable measurements. The distance is used as a measure of the extent of the anomaly. The distance is found by searching along a line from the anomaly to a closest acceptable measurement within the region R₁.

Claims

1. A method for evaluating an anomaly measurement x′

in a machine condition monitoring system wherein measurements x_i are evaluated in a one-class classifier having a decision region R₁ for the class C₁ such that an evaluation function ƒ

(x) is greater than or equal to a threshold T for a measurement x within the region, and less the T outside the region, the method comprising the steps of;

training the one-class classifier to establish the decision region R₁ from a set of training samples {x₁, x₂, . . . , x_N};

receiving the anomaly measurement x′

;

determining that the anomaly measurement is outside the region R₁;

determining a distance from the measurement x′

to a boundary of the region R₁; and

evaluating the anomaly measurement x′

based on the distance.

2. The method of claim 1, wherein the distance is a Euclidean distance.

3. The method of claim 1, wherein the step of determining a distance from the measurement x′

to the region R₁ is performed iteratively.

4. The method of claim 1, wherein the step of determining a distance from the measurement x′

to the region R₁ further comprises the steps of;

finding a closest sample q, in the region R₁, to the measurement x′

;

defining an average point m between the x′ and

q;

if a difference between ƒ

(m) and T is greater than a permissible error, then if ƒ

(m)<

T, assign m to x′ and

return to the defining step; and

if ƒ

(m)>

T, assign m to q and return to the defining step; and

if the difference is less than the permissible error, assigning a distance from the measurement x′

to in as the distance from the measurement x′

to a boundary of the region R₁.

5. The method of claim 4, wherein the step of finding a closest sample q, in the region R₁, to the measurement x′

further comprises using Euclidean distances between x′ and

the samples.

6. The method of claim 4, wherein the step of finding, a closest sample q, in the region R₁, to the measurement x′

further comprises the step of;

reducing a number of samples in the region R₁ by applying a k-mean algorithm.

7. The method of claim 4, wherein the step of evaluating the anomaly measurement x′

based on the distance further comprises evaluating a degree of the anomaly based on the distance.

8. The method of claim 4, wherein the step of defining an average point m between the measurement x′

and the closest sample q further comprises calculating

(x′

+q)/2.

9. The method of claim 1, further comprising the step of:

obtaining the evaluation function ƒ

(x) using an algorithm selected from the group consisting of a nearest neighbor algorithm, a probability density function and an inner product for a support vector representation machine.

10. A computer-usable medium having computer readable instructions stored thereon for execution by a processor to perform a method evaluating an anomaly measurement x′

in a machine condition monitoring system wherein measurements x_i are evaluated in a one-class classifier having a decision region R₁ for the class C₁ such that an evaluation function ƒ

(x) is greater than or equal to a threshold T for a measurement x within the region, and less the T outside the region, the method comprising the steps of;

training the one-class classifier to establish the decision region R₁ from a set of training samples {x₁, x₂,. . . , x_N};

receiving the anomaly measurement x′

;

determining that the anomaly measurement is outside the region R₁;

determining a distance from the measurement x′

to a boundary of the region R₁; and

evaluating the anomaly measurement x′

based on the distance.

11. The computer-usable medium of claim 10, wherein the distance is a Euclidean distance.

12. The computer-usable medium of claim 10, wherein the step of determining a distance from the measurement x′

to the region R₁ is performed iteratively.

13. The computer-usable medium of claim 10, wherein the step of determining a distance from the measurement x′

to the region R₁ further comprises the steps of;

finding a closest sample q, in the region R₁, to the measurement x′

;

defining an average point m between the x′ and

q;

if a difference between ƒ

(m) and T is greater than a permissible error, then if ƒ

(m)<

T, assign m to x′ and

return to the defining step; and

if ƒ

(m)>

T, assign m to q and return to the defining step; and

if the difference is less than the permissible error, assigning a distance from the measurement x′

to m as the distance from the measurement x′

to a boundary of the region R₁.

14. The computer-usable medium of claim 13, wherein the step of finding a closest sample q, in the region R₁, to the measurement x′

further comprises using Euclidean distances between x′ and

the samples.

15. The computer-usable medium of claim 13, wherein the step of finding a closest sample q, in the region R₁, to the measurement x′

further comprises the step of;

reducing a number of samples in the region R₁ by applying a k-mean algorithm.

16. The computer-usable medium of claim 13, wherein the step of evaluating the anomaly measurement x′

based on the distance further comprises evaluating a degree of the anomaly based on the distance.

17. The computer-usable medium of claim 13, wherein the step of defining an average point m between the measurement x′

and the closest sample q further comprises calculating

(x′

+q)/2.

18. The computer-usable medium of claim 10, further comprising the step of:

obtaining the evaluation function ƒ

(x) using an algorithm selected from the group consisting of a nearest neighbor algorithm, a probability density function and an inner product for a support vector representation machine.