Device health estimation by combining contextual information with sensor data

• US 10,078,062 B2
• Filed: 12/15/2015
• Issued: 09/18/2018
• Est. Priority Date: 12/15/2015
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Palo Alto Research Center, Inc. (Xerox Holdings Corp.)
• Inventors: Eldardiry, Hoda M. A., Liao, Linxia, Honda, Tomonori, Saha, Bhaskar, Abreu, Rui
• Primary Examiner(s): Fan, Bo
• Application Number:
  US14/969,984
• Publication Number:
  US 20170167993A1
• Time in Patent Office: 1,008 Days
• Field of Search: 702132
• CPC Class Codes: G01N 25/72 Investigating presence of flaws
• CPC Class Codes: G05B 13/00 Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion G05B19/00 takes precedence
• 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

Abstract

A method and system for detecting fault in a machine. During operation, the system obtains control signals and corresponding sensor data that indicates a condition of the machine. The system determines consistent time intervals for each of the control signals. During a consistent time interval the standard deviation of a respective control signal is less than a respective predetermined threshold. The system aggregates the consistent time intervals to determine aggregate consistent intervals. The system then maps the aggregate consistent intervals to the sensor data to determine time interval segments for the sensor data. The system may generate features based on the sensor data. Each respective feature is generated from a time interval segment of the sensor data. The system trains a classifier using the features, and applies the classifier to additional sensor data indicating a condition of the machine over a period of time to detect a machine fault.

Claims

1. A computer-executable method for detecting fault in a machine, comprising:

obtaining a control signal associated with controlling the machine and sensor data that indicates a condition of the machine during a time period when the control signal controls the machine;

determining consistent time intervals for the control signal, wherein during a consistent time interval the standard a standard deviation of the control signal is less than a predetermined threshold;

mapping the consistent time intervals to the sensor data to determine a plurality of time interval segments for the sensor data;

generating a plurality of training features based on the sensor data, wherein each respective feature is generated in association with from a time interval segment;

providing the plurality of training features as input to a classifier to train the classifier to classify abnormal sensor data during a respective consistent time interval;

generating new features for the classifier with same conditions as in classifier training by determining time intervals of a primary control signal that have same values for the primary control signal as a value of the primary control signal when generating the training features; and

detecting a machine fault by providing new features associated with additional machine sensor data as input to the classifier to detect abnormal sensor data during a respective consistent time interval.

2. The method of claim 1, wherein the control signal is at least one of a spindle motor speed, spindle load, and actual spindle speed;

and wherein the sensor data is temperature data indicating a temperature associated with the machine.

3. The method of claim 1, wherein generating the plurality of training features includes computing at least one of an average, a standard deviation, a maximum fast Fourier transform (FFT) value, and a FFT frequency at maximum amplitude for the sensor data.

4. The method of claim 3, wherein the generated training features form a high-dimensional feature space, further comprising:

applying principal component analysis (PCA) to project the high-dimensional feature space into a low-dimensional space; and

applying linear discriminant analysis (LDA) to determine an optimal coordinate transformation that provides maximum separation between classes.

5. The method of claim 1, wherein determining consistent time intervals further comprises generating a temporal segment representation of the machine'"'"'s operation context.

6. The method of claim 1, further comprising:

removing one or more control signal intervals that are inconsistent from a plurality of control signals before determining aggregate consistent intervals based on the plurality of control signals.

15. The method of claim 1, further comprising:

aggregating consistent time intervals of a plurality of control signals to determine aggregate consistent intervals.

16. The method of claim 15, wherein aggregating the consistent time intervals comprises determining an intersection of sets of consistent time intervals over all control signals.

7. A non-transitory computer-readable storage medium storing instructions which when executed by a computer cause the computer to perform a method for detecting fault in a machine, the method comprising:

obtaining a control signal associated with controlling the machine and sensor data that indicates a condition of the machine during a time period when the control signal controls the machine;

determining consistent time intervals for the control signal, wherein during a consistent time interval a standard deviation of the control signal is less than a predetermined threshold;

mapping the consistent time intervals to the sensor data to determine a plurality of time interval segments for the sensor data;

generating a plurality of training features based on the sensor data, wherein each respective feature is generated in association with a time interval segment;

providing the plurality of training features as input to a classifier to train the classifier to classify abnormal sensor data during a respective consistent time interval;

generating new features for the classifier with same conditions as in classifier training by determining time intervals of a primary control signal that have same values for the primary control signal as a value of the primary control signal when generating the training features; and

detecting a machine fault by providing new features associated with additional machine sensor data as input to the classifier to detect abnormal sensor data during a respective consistent time interval.

8. The storage medium of claim 7, wherein the control signal is at least one of a spindle motor speed, spindle load, and actual spindle speed;

and the and wherein the sensor data is temperature data indicating a temperature associated with the machine.

9. The storage medium of claim 7, wherein generating the plurality of training features includes computing at least one of an average, a standard deviation, a maximum fast Fourier transform (FFT) value, and a FFT frequency at maximum amplitude for the sensor data.

10. The storage medium of claim 7, wherein determining consistent time intervals further comprises generating a temporal segment representation of the machine'"'"'s operation context.

11. The storage medium of claim 7, wherein the method further comprises:

removing one or more control signal intervals that are inconsistent from a plurality of control signals before determining aggregate consistent intervals based on the plurality of control signals.

17. The storage medium of claim 7, wherein the method further comprises:

aggregating consistent time intervals of a plurality of control signals to determine aggregate consistent intervals.

18. The storage medium of claim 17, wherein aggregating the consistent time intervals comprises determining an intersection of sets of consistent time intervals over all control signals.

12. A computing system comprising:

one or more processors;

a memory; and

a non-transitory computer-readable medium coupled to the one or more processors storing instructions stored that, when executed by the one or more processors, cause the computing system to perform a method comprising;

obtaining a control signal associated with controlling the machine and sensor data that indicates a condition of the machine during a time period when the control signal controls the machine;

determining consistent time intervals for the control signal, wherein during a consistent time interval a standard deviation of the control signal is less than a predetermined threshold;

mapping the consistent time intervals to the sensor data to determine a plurality of time interval segments for the sensor data;

generating a plurality of training features based on the sensor data, wherein each respective feature is generated in association with from a time interval segment;

providing the plurality of training features as input to a classifier to train the classifier to classify abnormal sensor data during a respective consistent time interval;

generating new features for the classifier with same conditions as in classifier training by determining time intervals of a primary control signal that have same values for the primary control signal as a value of the primary control signal when generating the training features; and

detecting a machine fault by providing new features associated with additional machine sensor data as input to the classifier to detect abnormal sensor data during a respective consistent time interval.

13. The computing system of claim 12, wherein the control signal is at least one of a spindle motor speed, spindle load, and actual spindle speed;

and wherein the sensor data is temperature data indicating a temperature associated with the machine.

14. The computing system of claim 12, wherein generating the plurality of training features includes computing at least one of an average, a standard deviation, a maximum fast Fourier transform (FFT) value, and a FFT frequency at maximum amplitude for the sensor data.

19. The computing system of claim 12, wherein the method further comprises:

aggregating consistent time intervals of a plurality of control signals to determine aggregate consistent intervals.

20. The computing system of claim 19, wherein aggregating the consistent time intervals comprises determining an intersection of sets of consistent time intervals over all control signals.