Change detection using directional statistics

Patent metadata

Field	Value
Publication number	US-10378997-B2
Application number	US-201715490282-A
Country	US
Kind code	B2
Filing date	Apr 18, 2017
Priority date	May 6, 2016
Publication date	Aug 13, 2019
Grant date	Aug 13, 2019

Abstract

A method includes capturing multivariate time-series data comprising two or more data sets from a system captured over a past time period and a present time period, applying at least two sliding time windows to the multivariate time-series data in determining respective data matrices, computing an orthonormal matrix for each of the data matrices, wherein the orthonormal matrix is a signature of fluctuation patterns of a respective data matrix, computing a difference between at least two of the data sets in the past and the present time periods through the orthonormal matrices, and detecting a fault in at least one of the systems by comparing the difference to a threshold.

First claim

What is claimed is:

1. A method comprising:
capturing multivariate time-series data comprising two or more data sets from a system captured over a past time period and a present time period;
applying at least two sliding time windows to the multivariate time-series data in determining respective data matrices;
computing an orthonormal matrix for each of the data matrices, wherein the orthonormal matrix is a signature of fluctuation patterns of a respective data matrix;
computing a difference between at least two of the data sets in the past and the present time periods through the orthonormal matrices;
detecting a fault in at least one of the systems by comparing the difference to a threshold and removing the system from service upon detected the fault.

2. The method of claim 1 , wherein the multivariate time-series data is captured by a plurality of sensors.

3. The method of claim 2 , further comprising measure, by the plurality of sensors, one or more of acceleration, speed, strain, temperature, pressure, and voltage.

4. The method of claim 2 , further comprising measure, by the plurality of sensors, one or more of throttle position, gear number, and power status.

5. The method of claim 1 , wherein a first sliding time window of the two sliding time windows is a test window covering a predefined time duration up to the present time period, and wherein a second sliding time window of the two sliding time windows is a training window covering a portion of the past time period.

6. The method of claim 1 , wherein computing the difference comprising determining a change score corresponding to a distance between the orthonormal matrices.

7. The method of claim 1 , wherein the steps of capturing multivariate time-series data, applying the at least two sliding time windows to the multivariate time-series data in determining respective data matrices, computing the orthonormal matrix for each of the data matrices, and computing the difference between at least two of the data sets are iteratively repeated until the detection of the fault.

8. The method of claim 1 , further comprising performing the steps of capturing multivariate time-series data, applying the at least two sliding time windows to the multivariate time-series data in determining respective data matrices, computing the orthonormal matrix for each of the data matrices, and computing the difference between at least two of the data sets for at least one sub-system of the system.

9. A non-transitory computer readable medium comprising computer executable instructions which when executed by a computer cause the computer to perform a method comprising:
capturing multivariate time-series data comprising two or more data sets from a system captured over a past time period and a present time period;
applying at least two sliding time windows to the multivariate time-series data in determining respective data matrices;
computing an orthonormal matrix for each of the data matrices, wherein the orthonormal matrix is a signature of fluctuation patterns of a respective data matrix;
computing a difference between at least two of the data sets in the past and the present time periods through the orthonormal matrices;
detecting a fault in at least one of the systems by comparing the difference to a threshold and removing the system from service upon detected the fault.

10. The computer readable medium of claim 9 , wherein the multivariate time-series data is captured by a plurality of sensors.

11. The computer readable medium of claim 10 , further comprising measure, by the plurality of sensors, one or more of acceleration, speed, strain, temperature, pressure, and voltage.

12. The computer readable medium of claim 10 , further comprising measure, by the plurality of sensors, one or more of throttle position, gear number, and power status.

13. The computer readable medium of claim 9 , wherein a first sliding time window of the two sliding time windows is a test window covering a predefined time duration up to the present time period, and wherein a second sliding time window of the two sliding time windows is a training window covering a portion of the past time period.

14. The computer readable medium of claim 9 , wherein computing the difference comprising determining a change score corresponding to a distance between the orthonormal matrices.

15. The computer readable medium of claim 9 , wherein the steps of capturing multivariate time-series data, applying the at least two sliding time windows to the multivariate time-series data in determining respective data matrices, computing the orthonormal matrix for each of the data matrices, and computing the difference between at least two of the data sets are iteratively repeated until the detection of the fault.

16. The computer readable medium of claim 9 , further comprising performing the steps of capturing multivariate time-series data, applying the at least two sliding time windows to the multivariate time-series data in determining respective data matrices, computing the orthonormal matrix for each of the data matrices, and computing the difference between at least two of the data sets for at least one sub-system of the system.

Assignees

• IBM

Inventors

• Ide Tsuyoshi
• Kalagnanam Jayant R
• Phan Dung T

Classifications

• G01M13/00 Primary

  Testing of machine parts · CPC title

• G05B23/024 Primary

  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 title