Method and apparatus for efficient extraction of information from signals

• US 8,930,166 B1
• Filed: 12/06/2011
• Issued: 01/06/2015
• Est. Priority Date: 12/06/2011
• Status: Active Grant

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Das, Sreerupa, Maity, Ashis K.
• Primary Examiner(s): TSAI, CAROL S W
• Application Number:
  US13/311,912
• Time in Patent Office: 1,127 Days
• Field of Search: 702/184, 702/185, 705/36.R
• US Class Current: 702/184
• CPC Class Codes: G01R 31/2837 Characterising or performance testing, e.g. of frequency response transient response G01R27/28
• CPC Class Codes: G06F 18/2136 based on sparsity criteria, e.g. with an overcomplete basis
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06F 2218/00 Aspects of pattern recognition specially adapted for signal processing
• CPC Class Codes: G06F 30/367 Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods

Abstract

A method and device for extracting information from data representing a signal is disclosed. A set of data comprising a plurality of measurements of the signal generated at a first sampling rate is received from a sensor. A subset of the plurality of measurements is selected. A plurality of feature variables, each of which corresponds to a particular feature in a set of features that may be present in the signal are determined by deriving an underdetermined system of equations based on a selected basis function, the subset of the plurality of measurements, and the plurality of feature variables and corresponding features. The underdetermined system of equations is solved to determine a value for each feature variable using a non-linear optimization technique to minimize an L₁ norm of the set of features. Feature information is stored in a storage medium.

Claims

1. A method for extracting information from data representing a signal, comprising:

receiving, by a processing device, a set of data from a sensor, the set of data comprising a plurality of measurements of the signal generated at a first sampling rate;

randomly selecting, by the processing device, a subset of the plurality of measurements, the subset having fewer members than the set of data;

determining a value for each of a plurality of amplitude variables, each amplitude variable corresponding to a particular frequency component in a set of frequency components, the set of frequency components having a greater number of members than the subset of the plurality of measurements, by;

deriving an underdetermined system of equations based on;

a selected basis function;

the subset of the plurality of measurements; and

the plurality of amplitude variables and corresponding frequency components;

solving the underdetermined system of equations to determine the value for each amplitude variable using a non-linear optimization technique to minimize an L₁ norm of the set of frequency components, wherein the value of each amplitude variable identifies an amplitude of the corresponding frequency component in the signal; and

storing frequency component information including the value of at least some of the amplitude variables and corresponding frequency components in a storage medium.

2. The method of claim 1, wherein deriving the underdetermined system of equations is further based on the first sampling rate and the number of members in the set of data.

3. The method of claim 1, wherein the first sampling rate comprises a sampling rate in accordance with the Nyquist-Shannon sampling theory based on a selected highest frequency component of the signal, and wherein the number of members in the subset is about 1/10 of the number of members in the set of data.

4. The method of claim 1, wherein the plurality of measurements are in a time domain, and the set of frequency components are in a frequency domain.

5. The method of claim 1, wherein the signal is generated by a machine, and further comprising:

accessing, by a prognostics module, the frequency component information; and

based on the frequency component information, issuing an indication that the machine requires a maintenance task.

6. The method of claim 1, wherein the signal is generated by a machine, and further comprising:

performing the method for determining the value for each of the plurality of amplitude variables a plurality of times over a duration of time;

accessing the frequency component information for each of the plurality of times; and

based on the frequency component information for each of the plurality of times, issuing an indication that the machine requires a maintenance task.

7. The method of claim 1, wherein the selected basis function comprises one of a discrete Fourier transform function, a discrete cosine transform, a Hilbert transform function, and a Cepstrum transform function.

8. The method of claim 1, wherein the signal represents a vibration associated with a machine, and further comprising:

accessing, by a prognostics module, the frequency component information; and

based on the frequency component information, determining that the signal includes a frequency component that indicates an abnormal condition of the machine.

9. The method of claim 8, wherein the vibration is associated with one of a shaft, a gear and a bearing of the machine.

10. A processing device for extracting information from data representing a signal, comprising:

a communication interface configured to receive a set of data, the set of data comprising a plurality of measurements of the signal generated at a first sampling rate;

a processor communicatively coupled to the communication interface, and configured to;

randomly select a subset of the plurality of measurements, the subset having fewer members than the set of data;

determine a value for each of a plurality amplitude variables, each amplitude variable corresponding to a particular frequency component in a set of frequency components, the set of frequency components having a greater number of members than the subset of the plurality of measurements, by;

deriving an underdetermined system of equations based on;

a selected basis function;

the subset of the plurality of measurements; and

the plurality of amplitude variables and corresponding frequency components;

solving the underdetermined system of equations to determine the value for each amplitude variable using a non-linear optimization technique to minimize an L₁ norm of the set of frequency components; and

store frequency component information including the value of at least some of the amplitude variables and corresponding frequency components in a storage medium.

11. The processing device of claim 10, wherein deriving the underdetermined system of equations is further based on the first sampling rate and the number of members in the set of data.

12. The processing device of claim 10, wherein the first sampling rate comprises a sampling rate in accordance with the Nyquist-Shannon sampling theory based on a selected highest frequency component of the signal, and wherein the number of members in the subset is about 1/10 of the number of members in the set of data.

13. The processing device of claim 10, wherein the signal is generated by a machine, and wherein the processor is further configured to:

access the frequency component information; and

based on the frequency component information, issue an indication that the machine requires a maintenance task.

14. The processing device of claim 10, wherein the signal represents a vibration associated with a machine, and wherein the processor is further configured to:

access the frequency component information; and

based on the frequency component information, determine that the signal includes a frequency component that indicates an abnormal condition of the machine.

15. A method for extracting information from a signal, comprising:

randomly selecting, by a processing device, a subset of a plurality of measurements of the signal generated by a sensor at a first sampling rate;

determining, by the processing device, a value for each of a plurality of amplitude variables, each amplitude variable in the plurality of amplitude variables associated with a respective frequency component in a set of frequency components, the set of frequency components having a greater number of members than the subset, by;

deriving an underdetermined system of equations based on a selected basis function, the subset of the plurality of measurements, and the plurality of amplitude variables;

solving the underdetermined system of equations to determine the value of each of the amplitude variables using a non-linear optimization technique to minimize an L₁ norm of the sot of frequency components; and

based on the amplitude variables and corresponding frequency components, determining that the signal includes a frequency component that indicates an abnormal condition of the machine.

16. The method of claim 15, wherein the signal represents a vibration associated with one of a shaft, a gear and a bearing of the machine.

17. The method of claim 15, wherein the first sampling rate comprises a sampling rate in accordance with the Nyquist-Shannon sampling theory based on a selected highest frequency component of the signal, and wherein the number of members in the subset is about 1/10 of the number of members in the set of data.

18. The method of claim 15, wherein the set of measurements are in a time domain, and the set of frequency components are in a frequency domain.