Method for detecting a defect in a vibration sensor, associated device and computer program

Abstract

A method detects a defect in a vibration sensor including the whitening of the vibratory signal delivered by the vibration sensor, and the calculation of an indicator of asymmetry of the whitened vibratory signal. The whitening may be a cepstral whitening implemented simply by dividing the Fourier transform of the signal by its modulus. The asymmetry indicator is a counter of aberrant points in the whitened vibratory signal.

Claims

1. A method for detecting a defect in a vibration sensor, comprising: performing a spectral whitening of a vibratory signal delivered by the vibration sensor; and calculating an asymmetry indicator in the spectrally whitened vibratory signal, wherein calculating the asymmetry indicator comprises calculating a number of positive aberrant points and a number of negative aberrant points in the spectrally whitened vibratory signal, a point in the spectrally whitened vibratory signal being a positive or negative aberrant point when the absolute value thereof is higher than an aberration threshold, and calculating a difference between the number of positive aberrant points and the number of negative aberrant points, and wherein the method further comprises a step in which a defect in the vibration sensor is detected when the asymmetry indicator exceeds an alarm threshold.

2. The method according to claim 1, wherein the spectral whitening of the vibratory signal is a cepstral whitening.

3. The method according to claim 2, wherein the cepstral whitening of the vibratory signal comprises a Fourier transform of the vibratory signal, a calculation of a modulus of said transform, a division of said transform by its modulus and an inverse Fourier transform of a result of said division.

4. The method according to claim 1, wherein the asymmetry indicator is expressed in accordance with $\frac{.Math.N^{+}-N^{-}.Math.}{N},$ with N.sup.+ the number of positive aberrant points, N.sup.− the number of negative aberrant points and N the total number of aberrant points N=N.sup.++N.sup.−.

5. The method according to claim 1, wherein the vibration sensor is installed in a machine, and wherein the vibratory signal is exploited by a health monitoring system when no defect in the vibration sensor is detected.

6. A health monitoring system, comprising: a vibration sensor installed in a machine and a data processing unit coupled to the vibration sensor, the data processing unit being configured to implement the method according to claim 1.

7. The health monitoring system according to claim 6, wherein the vibration sensor is an accelerometer.

8. A non-transitory computer readable medium which stores thereon a computer program that when being executed by a computer, executes the method according to claim 1.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Other aspects, aims, advantages and features of the invention will emerge more clearly from a reading of the following detailed description of preferred embodiments thereof, given by way of non-limitative example, and made with reference to the accompanying drawings, on which:

(2) FIG. 1 is a diagram of a system for monitoring the state of health of a machine by vibratory analysis implementing the invention;

(3) FIG. 2 is a flow diagram illustrating the method for detecting a defect in a vibration sensor according to the invention;

(4) FIG. 3 illustrates a possible implementation of the spectral whitening carried out in the context of the invention;

(5) FIG. 4 illustrates a possible example of computation of an asymmetry indicator.

DETAILED DISCLOSURE OF PARTICULAR EMBODIMENTS

(6) With reference to FIG. 1, the invention can be implemented in a system 1 for monitoring the state of health of a machine by vibratory analysis. The machine may be a rotary machine, a combustion or explosion engine, or a test bench. A favoured application is that of the monitoring of aeroplane or helicopter engines.

(7) The system 1 comprises a unit 2 for acquiring the vibratory signal, a unit 3 for calculating state of health indicators and a restitution unit 4 for supplying an indication on the state of the machine and of the vibration sensor.

(8) The acquisition unit 2 comprises a vibration sensor 21 fixed to the machine being monitored by means of adhesive, for example an accelerometer, and an acquisition chain 22 for the vibratory signal delivered by the vibration sensor 21. The role of the acquisition chain is to convert the analogue signal delivered by the vibration sensor 21 into a digital signal. The chain typically comprises the following elements: a conditioner (amplification, galvanic isolation and excitation of the passive sensors, etc.), an analogue filter (for limiting the bandwidth of the sensor and thus preventing deterioration of the signal-to-noise ratio and spectral aliasing), a sample and hold unit (which takes a sample of the signal and keeps it constant during the conversion phase) and an analogue-to-digital converter. The raw vibratory signal thus digitised is transferred to the calculation unit 3.

(9) The calculation unit 3 is distinguished from a unit conventionally used in health monitoring in that it comprises a module 31 for detecting a defect in the vibration sensor configured to implement the method described hereinafter. The output of this sensor defect detection module is assessed by a first test module 32.

(10) If no sensor defect is detected (output N of the first test module 32), the vibratory signal is the subject of digital processing in a processing module 33 and indicators are calculated from the signals processed in an indicator calculation module 34. The indicators calculated are assessed by a second test module 35, where they are compared with thresholds in order to take a decision whether or not to trigger an alarm (output N and O of the second test module 35), this decision being supplied by the restitution unit 4, typically a display able to present information relating to a sound state of the machine being monitored (block 41) or to a defective state via the raising of an alarm (block 42).

(11) If a sensor defect is detected (output 0 of the first test module 32), the vibratory signal is not processed and the restitution unit 4 presents information relating to the presence of a sensor defect (block 43).

(12) The method for detecting a potential defect in the vibration sensor implemented in the detection module 31 is described hereinafter.

(13) The invention is based on the finding that the method disclosed in the aforementioned article associates any asymmetry in the vibratory signal with the presence of a defect in the vibration sensor. However, the applicant observes that the asymmetry of a vibratory signal is not solely related to a defect in the vibration sensor, but may also result from the non-linearity of the mechanical system and the distribution of the phases at the harmonics (sinusoids) because of the periodic vibrations generated for example by the gears. In a case of defect-free functioning of the vibration sensor, the vibratory signal therefore exhibits an asymmetry that risks being identified as excessive and therefore generating a false alarm.

(14) In order to make it possible to distinguish the various sources of asymmetry, and with reference to FIG. 2, the method comprises a first step E1 of spectral whitening of the vibratory signal X(n). This step makes it possible in fact to reject the entire deterministic component of the vibratory signal and, consequently, the asymmetry that it causes. Under these circumstances, the asymmetry in the whitened vibratory signal X.sub.b(n) can be solely assimilated to the presence of a defect in the vibration sensor.

(15) The deterministic part of the vibratory signal is periodic in stationary regime and therefore has a discrete frequency spectrum. The random part of the vibratory signal for its part has a continuous frequency spectrum. The spectral whitening of a signal consists of modifying the spectrum thereof so as to bring it close to the spectrum of a white noise. The whitened vibratory signal thus has a constant spectral power density where the peaks of the spectral envelope of the vibratory had been made uniform and the contribution of the deterministic part has consequently been erased.

(16) The separation of the deterministic part of the random part of the vibratory signal can in particular be achieved by means of a cepstral whitening that keeps the sources present in the phase of the vibratory signal and eliminates all the sources that are manifested on the amplitude. This technique offers a remarkable ability for blind elimination of the deterministic content of a vibratory signal while whitening the spectrum thereof. This technique consists of zeroing all the complex cepstrum of the signal while keeping only the component relating to zero quefrency. The cepstrum is thus zeroed and next transformed in the time domain, after recombination with the phase of the original sign. A description of this cepstral whitening technique can be found in the article by P. Borghesani et al entitled “Application of cepstrum pre-whitening for the diagnosis of bearing faults under variable speed conditions”, Mech. Syst. Sig. Process. 36 (2013) 370-384.

(17) Advantageously, cepstral whitening can be carried out without calculating the cepstrum by simply using the Fourier transform of the vibratory signal. In such a case, and with reference to FIG. 3, the whitening of the vibratory signal X(n) comprises a Fourier transform FT of the vibratory signal, the calculation MOD of the modulus of said transform, the division DIV of said transform by its modulus and an inverse Fourier transform IFT of the result of said division. With DFT the discrete Fourier transform and DFT.sup.−1 the inverse transform, the spectral whitening operation is written as

(18) $X_b\left(n\right)={\mathrm{DFT}}^{-1}\left\{\frac{\mathrm{DFT}\left\{X\left(n\right)\right\}}{.Math.\mathrm{DFT}\left\{X\left(n\right)\right\}.Math.}\right\}.$

(19) With reference to FIG. 2, the method comprises a second step E2 of calculating an asymmetry indicator of the whitened vibratory signal. This calculation makes it possible to evaluate the non-linearity of the transfer function of the vibration sensor or, seen differently, the presence of a defect, such as a fixing defect, in the vibration sensor.

(20) The asymmetry indicator is a counter of aberrant points in the whitened vibratory signal. More precisely, the aberrant-point counter quantifies the asymmetry from the difference between the positive and negative aberrant points, these aberrant points being defined with respect to a symmetry threshold referred to as the aberration threshold. The aberration threshold is designated α, namely a positive integer smaller than the maximum value of the absolute value of the signal X.sub.b(n): 0<α<max(|X.sub.b(n)|). An aberration threshold α is for example chosen equal to twice the standard deviation of the L samples considered of the whitened vibratory signal. The positive and negative aberrant points, designed N.sup.+ and N.sup.− respectively, are defined as (cf. FIG. 4):

(21) $\hspace{1em}\{\begin{array}{c}{N}^{+}=\underset{n=1}{\overset{L}{.Math.}}{I}_{X_b\left(n\right)>a}\\ N^{-}=\underset{n=1}{\overset{L}{.Math.}}{I}_{X_b\left(n\right)<-a}\end{array}$

(22) where I.sub.αrg designates the indicator function (it returns 1 when the condition αrg is true). Let us designate N the total number of positive and negative aberrant points (N=N.sup.++N.sup.−), the calculation of the indicator counting aberrant points comprises the calculation of the difference between the number of positive aberrant points and the number of negative aberrant points and can be defined as follows:

(23) $I_X=\frac{.Math.N^{+}-N^{-}.Math.}{N}.$

(24) This indicator I.sub.x lies between 0 and 1 and is not influenced by the direction of the asymmetry. It returns 0 for a perfect symmetry and approaches 1 when the asymmetry is great.

(25) This indicator I.sub.x proves to be more effective than the indicator based on the optimised skewness described in the aforementioned article. This is because, unlike the indicator I.sub.x, optimised skewness is neither standardised nor bounded, and is therefore difficult to quantify. Furthermore, the asymmetry is characterised by directional pulses. However, these pulses do not affect the empirical probability distribution of the signal and are not detected by the skewness. The indicator I.sub.x, for its part, is capable of detecting this particular type of asymmetry. Being more sensitive to asymmetry than skewness, the indicator of the invention makes it possible to improve the ability to detect defects in the vibration sensor. It is furthermore simple to calculate.

(26) Once the asymmetry indicator is calculated, the method comprises a third step E3 of evaluating the asymmetry on the basis of a hypothesis test on the indicator I.sub.x. This is because, even if the whitened vibratory signal is symmetrical, the indicator does not necessarily return an exactly zero value. This is due to the estimation errors (bias and variance of the indicator) resulting from the finite length of the signal. A threshold, referred to as the alarm threshold μ, is adjusted, which defines the confidence level that it is wished to grant to the indicator. This threshold μ can be calculated from the empirical values of the indicator I.sub.x calculated on cases of sound operation of the sensor, or alternatively by statistical mathematical calculations. If the value of the indicator I.sub.x is below the alarm threshold μ, the signal is considered to be symmetrical, otherwise it is considered to be asymmetric. Thus a defect in the vibration sensor is detected when the asymmetry indicator I.sub.x exceeds the alarm threshold μ. In such a case (O in FIG. 2), an alert message can be displayed during a step E4 by the restitution unit 4. In the contrary case (N in FIG. 2), the vibration sensor is considered to be sound and the monitoring system functions normally during a step E5 in order to calculate indicators of abnormalities that will be compared with thresholds in order to decide on the presence or not of a defect in the machine.

(27) It will be understood that the method according to the invention is quick, easy to implement, and does not require signals from other sensors. It also allows robust and precise detection of defects by means of the spectral whitening allowing the separation of sources of asymmetry related to the sound or defective functioning of the vibration sensor. Furthermore, no history is required for detecting the defect, the vibratory signal being evaluated independently of the measurements made before or afterwards. Finally, the indicator counting aberrant points is more sensitive to asymmetry than skewness and therefore improves the detectability of the sensor defects.

(28) The invention is not limited to the method as described above, but also extends to a data processing unit comprising means configured to implement this method, as well as to a system for monitoring the condition of a machine by vibratory analysis that comprises a vibration sensor installed in the machine and such a data processing unit coupled to the vibration sensor. The invention also relates to a computer program product comprising code instructions for implementing the method when said program is executed on a computer.