Systems and methods for learning of normal sensor signatures, condition monitoring and diagnosis

Patent metadata

Field	Value
Publication number	US-9443201-B2
Application number	US-201113703156-A
Country	US
Kind code	B2
Filing date	May 12, 2011
Priority date	Jun 9, 2010
Publication date	Sep 13, 2016
Grant date	Sep 13, 2016

Abstract

Systems and methods to monitor a signal from an apparatus are disclosed. A feature extracted from the signal is automatically defined. Signals are received over a period of time wherein the apparatus is in a normal operational mode. Features are classified in a learning mode and are applied to create a reference model that defines a within-normal operational mode. In a testing mode a signal generated by the apparatus is received, a feature is extracted and classified. Instantaneous data generated in operational mode by the apparatus is classified by the system as abnormal if it does not lie within boundaries of the reference model or contains information/structure in an orthogonal subspace. A learned reference model is augmented by a user or automatically. In one illustrative example the apparatus is a power generation equipment and the signal is an acoustic signal.

First claim

The invention claimed is:

1. A method to determine a status of an apparatus, the method comprising:
capturing by one or more sensors a first signal generated by the apparatus to be processed by a processor in a learning mode;
the processor automatically defining a feature by using the first signal, wherein the feature includes a representation of a characteristic of the first signal in a transform domain;
the processor learning a reference model in a learning mode that defines a normal operational mode of the apparatus in a first feature space;
capturing a second signal generated by the apparatus by the one or more sensors in an operational mode to extract an operational feature; and
comparing the operational feature to the reference model to determine whether the apparatus is operating in the normal operational mode and when a decision criteria is met, then analyzing the captured second signal in a second feature space that is different from the first feature space to determine whether there are additional features that define the normal operational mode.

2. The method of claim 1 , wherein the decision criteria is whether a predetermined threshold amount of energy is accounted for in the second feature space.

3. The method of claim 1 , wherein the decision criteria is whether the apparatus is operating in the normal operational mode.

4. The method of claim 1 , further comprising adding the additional features to the reference model when it is determined that the additional features define the normal operational mode.

5. The method of claim 1 , wherein the analysis of the second signal in the second feature space is performed with principal component analysis, independent component analysis or mutual interdependence analysis.

6. The method of claim 1 , wherein the step of automatically determining a feature is performed with principal component analysis, independent component analysis or mutual interdependence analysis.

7. The method of claim 1 , wherein the representation of the characteristic is one of the group consisting of a maximum amplitude, an average amplitude, an energy content, independent components, principal components, a crest factor, a deviation from an average, a kurtosis, and a skew all determined over a period of time in the transform domain.

8. The method of claim 1 , wherein the second feature space is orthogonal to the first feature space.

9. The method of claim 1 , wherein the first and second signals are acoustic signals and the apparatus is a power apparatus.

10. The method of claim 1 , further comprising: classifying an operational feature extracted from a signal generated by a second apparatus.

11. The method of claim 10 , further comprising: relaxing a boundary of the reference model automatically.

12. The method of claim 1 wherein defining the feature comprises selecting a first sub-set of content of the captured signal representing normal operation as the feature and not selecting a second sub-set of content of the captured signal not representing normal operation as indicated by a lack of structure, and wherein analyzing comprises analyzing the second sub-set of the content of the captured second signal for the structure where the structure indicates out-of-normal operation.

13. A method to determine a status of an apparatus where a reference model has been learned in a first feature space that defines a normal mode of operation, the method comprising:
capturing a signal generated by the apparatus by one or more sensors in an operational mode to extract an operational feature; and
comparing the operational feature to the reference model to determine whether the apparatus is operating in the normal mode of operational and when a decision criteria is met, then analyzing the captured signal in a second feature space that is different from the first feature space to determine whether there are additional features that define the normal operational mode.

14. The method of claim 13 , wherein the decision criteria is whether a predetermined threshold amount of energy is accounted for in the second feature space.

15. The method of claim 13 , further comprising adding the additional features to the reference model when it is determined that the additional features define the normal operational mode.

16. The method of claim 13 , wherein the analysis of the signal in the second feature space is performed with principal component analysis, independent component analysis or mutual interdependence analysis.

17. A system comprising:
apparatus that operates in a normal mode of operation;
one or more sensors that capture signals while the apparatus is in an operating mode;
memory that stores a reference model that defines a normal mode of operation in a first feature space; and
a processor connected to the one or more sensors and to the memory that extracts an operational feature from capture signals and determines accesses the reference model to determine whether the apparatus is operating in the normal mode of operational and when a decision criteria is met, then analyzes the captured signals in a second feature space that is orthogonal to the first feature space to determine whether there are additional features that define the normal mode of operation.

18. The system of claim 17 , wherein the decision criteria is whether a predetermined threshold amount of energy is accounted for in the second feature space.

19. The system of claim 17 , wherein the processor adds the additional features to the reference model when it is determined that the additional features define the normal mode of operational.

20. The system of claim 17 , wherein the processor performs the analysis of the signal in the second feature space with principal component analysis, independent component analysis or mutual interdependence analysis.

21. The system of claim 17 , wherein features in the first feature space and the second feature space are selected from the group consisting of a maximum amplitude, an average amplitude, an energy content, an energy content in a selected frequency band, independent components, principal components, a crest factor, a deviation from an average, a kurtosis, and a skew all determined over a period of time in the transform domain.

Assignees

• Claussen Heiko
• Rosca Justinian
• Brummel Hans-Gerd
• Thompson Edward David
• Siemens Ag

Inventors

• Claussen Heiko
• Rosca Justinian
• Brummel Hans-Gerd
• Thompson Edward David

Classifications

• G06F18/2433

  Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection · CPC title

• G06N99/005 Primary

  Physics · mapped topic

• G06K9/6284

  Physics · mapped topic

• G06N20/00 Primary

  Machine learning · CPC title