Neural network based analysis system for vibration analysis and condition monitoring

• US 6,301,572 B1
• Filed: 12/02/1998
• Issued: 10/09/2001
• Est. Priority Date: 12/02/1998
• Status: Expired due to Term

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Harrison, Gregory A.
• Primary Examiner(s): Powell, Mark
• Assistant Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/204,334
• Time in Patent Office: 1,042 Days
• Field of Search: 706/52, 702/54, 702/185
• US Class Current: 706/52
• CPC Class Codes: G01H 1/00 Measuring characteristics of vibrations in solids by using direct conduction to the detector G01H9/00, G01H11/00 take precedence
• CPC Class Codes: G01H 1/003 of rotating machines G01H1/10 takes precedence

Abstract

A system and a method for tracking long term performance of a vibrating body such as a gas turbine, includes a vibration sensor who time domain outputs are transformed to the frequency domain, using a fast Fourier transform processing. Frequency domain outputs are provided as inputs to a fuzzy adaptive resonance theory neural network. Outputs from the network can be coupled to an expert system for analysis, to display devices for presentation to an operator or are available for other control and information purposes.

Claims

1. A vibration analysis system for monitoring an operating device comprising:

at least one vibration sensor;

transform circuitry, coupled to the sensors, for transforming sensor outputs to frequency spectrum information;

a fuzzy adaptive resonance-type neural network for receipt and processing of the information to generate selected output information.

2. A system as in claim 1 wherein the transform circuitry comprises a Fast Fourier Transform.

3. A system as in claim 1 which includes circuitry for converting frequency and amplitude information to serial form for presentation to the network.

4. A system as in claim 3 which includes an expert system for automatically analyzing the output information.

5. A system as in claim 1 wherein the transform circuitry comprises a programmed processor.

6. A system as in claim 1 wherein the neural network comprises a programmed processor and a plurality of network implementing instructions.

7. A system as in claim 6 wherein the transform circuitry is coupled to serializing circuitry which presents serial information to the network implementing instructions.

8. A system as in claim 7 which includes normalizing circuitry coupled between the transform circuitry and serializing circuitry.

9. A system as in claim 1 wherein parameters of the network have been adjusted in response to presenting data from at least one properly functioning source of vibrations.

10. A system as in claim 1 which includes windowing circuitry coupled between the sensor and the transform circuitry.

11. A system as in claim 1 which includes a filter and wherein sensor outputs are filtered before being coupled to the transform circuitry.

12. A system as in claim 11 which includes a digitizer coupled between the filter and the transform circuitry.

13. A system as in claim 1 which includes normalization circuitry, coupled to the transform circuitry, for providing a normalized frequency domain representation to the network.

14. A system as in claim 13 wherein the normalizing circuitry includes baseline noise removal circuitry.

15. A system as in claim 1 wherein the network includes a serial input port.

16. A system as in claim 15 wherein characteristics of a normal operating device have been incorporated into the network.

17. A system as in claim 16 wherein the memory of the network includes vibration-related information from a normally operating device.

18. A system as in claim 17 which includes circuitry for detecting differences between the stored information and signals from the sensor indicative of a current operating condition.

19. A system as in claim 17 which includes circuitry for detecting operational trends, in the input spectrum, which indicate at least one of device wear and device damage.

20. A method of analyzing vibrations generated by a device comprising:

training a resonant-type neural network to recognize a normal class of vibrations from the device using frequency domain inputs from comparable, normally functioning devices;

detecting vibrations from the device to be monitored;

converting the detected vibrations to a frequency domain representation using a variance reduced transform process;

presenting the frequency domain information serially to the trained network;

processing the information in the network; and

determining if the detected vibrations are outside of the normal operating mode of the device.

21. A method as in claim 20 which includes filtering the detected vibrations.

22. A method as in claim 20 which includes detecting the operating speed of the device.

23. A method as in claim 22 which includes differentiating between periodic, speed related vibrations and non-speed related vibrations.

24. A method as in claim 20 which includes retraining only portions of the network in response to having carried out a maintenance procedure on the device.

25. A method as in claim 20 which includes coupling outputs from the network to an analysis system.

26. A software based vibration analysis system comprising:

executable instructions for transforming an input time domain signal to a frequency domain representation;

instructions for analyzing and normalizing the frequency domain representation; and

instructions for implementing a resonant-type neural network including instructions for implementing network short term and longer term memory wherein nominal, expected characteristics of a selected type of device being analyzed can be stored.

27. A system as in claim 26 which includes instructions for detecting speed signals from the device and for distinguishing between substantially periodic, speed related and non-speed related signals.

28. A system as in claim 27 which includes instructions for removing base line noise from the frequency domain representation, and normalizing the frequency domain represented.

29. A system as in claim 28 which includes instructions for analyzing differences, detected by the network, between an expected spectral pattern and a spectral pattern for the input signal.

30. An apparatus comprising:

a movable vessel;

a turbine for moving the vessel;

at least one vibration sensor coupled to the turbine;

at least one speed sensor coupled to the turbine;

a processor for comparing digital frequency domain representations of the signals from at least the vibration sensor to prestored expected, frequency domain representations of turbine vibrations wherein the processor implements a software based, resonant-type neural network to carry out the comparing process.

31. An apparatus as in claim 30 which includes processing circuitry, coupled between the speed sensor and the processor and including executable instructions for distinguishing between speed related vibrations and non-speed related vibrations.

32. An apparatus as in claim 30 wherein the vessel is selected from a class which includes a marine platform, a land platform and an airborne platform.

33. An apparatus as in claim 32 which includes an output device and executable instructions for presenting to an operator indications of operational differences between the frequency domain representation of current signals from the vibration sensor and the prestored, expected frequency domain representations.

34. An apparatus as in claim 33 which include instructions for providing maintenance indicators to the operator.

35. A system as in claim 32 wherein the processor includes a digital signal processor.

36. A system as in claim 32 wherein the processor includes a programmable computer for executing at least some of the instructions.

37. A system as in claim 32 wherein the system transmits detected vibration information to an external collecting point, for incorporation into a general vibration database.