Monitoring of a machine with a rotating machine component

• US 10,330,567 B2
• Filed: 04/14/2016
• Issued: 06/25/2019
• Est. Priority Date: 04/15/2015
• Status: Active Grant

• Current Assignee: Siemens AG
• Original Assignee: Siemens AG
• Inventors: Ochsenfeld, Henning
• Primary Examiner(s): Seven, Evren
• Application Number:
  US15/098,664
• Publication Number:
  US 20160305844A1
• Time in Patent Office: 1,167 Days
• Field of Search: None
• CPC Class Codes: G01H 1/003 of rotating machines G01H1/10 takes precedence
• CPC Class Codes: G01H 1/006 of the rotor of turbo machines
• CPC Class Codes: G01M 13/028 Acoustic or vibration analysis
• CPC Class Codes: G01M 7/00 Vibration-testing of structures; Shock-testing of structures G01M9/00 takes precedence

Abstract

In a method for the automated monitoring of a machine with a rotating machine component, at successive measurement times parameter values (P_t) of operating parameters of the machine and in each case over a predefined measurement duration, vibrations of the machine are recorded. A vibration spectrum (Y_t(f)) of the vibrations recorded during the measurement duration is determined for each measurement duration. A mean power density (Φ_t⁽ⁱ⁾) of the vibration spectrum (Y_t(f)) is in each case determined from the vibration spectrum (Y_t(f)) for a multiplicity of frequency ranges. During a learning phase a model function ({circumflex over (Φ)}⁽¹⁾(P) to {circumflex over (Φ)}^(N)(P)) for an averaged power density is determined for each frequency range as a function of the operating parameters. During a monitoring phase for each measurement time and each frequency range an expected mean power density ({circumflex over (Φ)}⁽ⁱ⁾(P_t)) is determined from the model function ({circumflex over (Φ)}⁽¹⁾(P) to {circumflex over (Φ)}^(N)(P)) established in the learning phase for the frequency range, and compared with the mean power density (Φ_t⁽ⁱ⁾) established from the recorded vibrations.

Claims

1. A method for monitoring of a machine with a rotating machine component, in an automated manner, and carrying out maintenance measures for the rotating machine component, the method comprising:

at successive measurement times recording value (P_t) of at least one operating parameter, and recording vibrations of the machine over a predefined measurement duration;

for each measurement duration determining a vibration spectrum (Y_t(f)) of vibration amplitudes and vibration frequencies (f) of the vibrations recorded during the measurement duration;

determining a mean power density (Φ

_t⁽ⁱ⁾ of the vibration spectrum (Y_t(f)) for a plurality of frequency ranges of the recorded vibration frequencies;

for each measurement time and each frequency range, forming a deviation measure for a deviation of the mean power density (Φ

_t⁽ⁱ⁾) from an expected mean power density ({circumflex over (Φ

)}⁽ⁱ⁾(P_t)) selected for a respective frequency and the at least one parameter value (P_t) recorded at the respective measurement time, and carrying out maintenance measures for the machine with the rotating machine component.

2. The method of claim 1, wherein the expected mean power density ({circumflex over (Φ

)}⁽ⁱ⁾(P_t)) is established during a learning phase comprising;

at successive measurement times recording value (P_t) of at least one operating parameter, and recording vibrations of the machine over a predefined measurement duration;

for each measurement duration determining a vibration spectrum (Y_t(f)) of vibration amplitudes and vibration frequencies (f) of the vibrations recorded during the measurement duration;

determining a mean power density (Φ

_t⁽ⁱ⁾) of the vibration spectrum (Y_t(f)) for a plurality of frequency ranges of the recorded vibration frequencies;

for each frequency range determining a model function ({circumflex over (Φ

)}⁽¹⁾(P) to {circumflex over (Φ

)}^(N)(P)) forming the mean power densities (Φ

_t⁽ⁱ⁾)) corresponding with respective measurement times and at least one parameter value (P_t), thereby establishing a plurality of expected power density ({circumflex over (Φ

)}⁽ⁱ⁾(P_t)) values as a function of respective frequency and the at least one parameter value (P_t).

3. The method of claim 1, wherein the machine is a transmission.

4. The method of claim 1, wherein the at least one of the operating parameters is selected from a group consisting of torque, rotational speed, and temperature of the machine or a portion thereof.

5. The method of claim 1, wherein the width of the frequency ranges grows monotonously with increasing vibration frequency (f).

6. The method of claim 1, wherein the model function ({circumflex over (Φ

)}⁽¹⁾(P) to {circumflex over (Φ

)}^(N)(P)) is determined for each frequency range as a linear function of the operating parameters, with a linear multiple regression model being trained for estimation of the model function ({circumflex over (Φ

)}⁽¹⁾(P) to {circumflex over (Φ

)}^(N)(P)).

7. The method of claim 1, further comprising forming, as a deviation measure for each measurement time and each frequency range, a power density difference (z_t⁽ⁱ⁾) normalized to a standard normal distribution between the mean power density (Φ

_t⁽ⁱ⁾) established from the recorded vibrations and the expected mean power density ({circumflex over (Φ

)}⁽ⁱ⁾(P_t)).

8. The method of claim 7, wherein the normalized power density difference (z_t⁽ⁱ⁾) for a measurement time and a frequency range is formed according to

9. The method of claim 8, wherein the smoothed expected value (μ

_t⁽ⁱ⁾) for a measurement time is formed depending on the power density difference (δ

_t⁽ⁱ⁾) for the measurement time and the smoothed expected value (μ

_t⁽ⁱ⁾) for the measurement time preceding the measurement time according to

μ

_t⁽ⁱ⁾=λ

δ

_t⁽ⁱ⁾+(1−

λ

)μ

_t-1⁽ⁱ⁾,

wherein λ

is a smoothing parameter.

10. The method of claim 8, wherein the smoothed standard deviation (σ

_t⁽ⁱ⁾) for a measurement time is formed depending on the power density difference (δ

_t⁽ⁱ⁾) and the smoothed expected value (μ

_t⁽ⁱ⁾) for this measurement time and the smoothed standard deviation (σ

_t⁽ⁱ⁾) for the measurement time preceding the measurement time according to

(σ

_t⁽ⁱ⁾)=λ

(δ

_t⁽ⁱ⁾−

μ

_t⁽ⁱ⁾)+(1−

λ

)(σ

_t-1⁽ⁱ⁾)²,

wherein λ

is a smoothing parameter.

11. The method of claim 10, wherein the smoothing parameter is formed according to

12. The method of claim 1, further comprising providing a tolerance interval (T) for the deviation measure;

and for each measurement time and for each frequency range determining whether the deviation measure lies within the predefined tolerance interval (T).