FR3079977A1 - METHOD AND DEVICE FOR MONITORING A ROTATING MACHINE - Google Patents

Abstract

The invention relates to a method for monitoring a rotating machine (1) comprising at least one rotating part capable of generating vibrations, each part having at least one characteristic frequency of fault, a vibratory signal representative of these vibrations having been acquired. by a sensor (21), a digital vibratory signal x (t) having been obtained, the method comprising a step (E1) of determination, for each characteristic frequency ad, of a selective spectral kurtosis of the digital signal which is defined by the part of the spectral kurtosis linked to said characteristic frequency.   

Description

GENERAL TECHNICAL AREA

The invention relates to the field of detection of mechanical faults in rotating machines by analysis of the vibro-acoustic signals generated by them and advantageously finds application in the detection of such faults in aircraft and helicopter engines for monitoring. such systems.

STATE OF THE ART

The detection of mechanical faults in rotating machines by analysis of vibro-acoustic signals is very effective since the vibrations emitted by rotating machines contain a lot of information on the internal forces in the system. These internal forces are often linked to certain fault mechanisms and exhibit distinctive vibration symptoms known as mechanical signatures.

Generally speaking, an emerging mechanical defect manifests itself in the form of recurrent short-term shocks whose spacing is almost constant. These spacings can be perfectly equal as in the case of a gear defect, or slightly different as in the case of a bearing defect. In both cases, the average spacing between the shocks designates the characteristic period of the fault or, in an equivalent manner, the inverse of the characteristic frequency. These shocks excite the structure of the mechanical system which responds according to its modes of resonance.

From a signal point of view, the time signature of a mechanical fault is therefore an almost periodic train of impulse responses. This signature is characterized by two pieces of information:

• the characteristic frequency (rate of recurrence of shocks);

• the properties of the transfer function.

Figure 1 illustrates the typical temporal vibro-acoustic signature generated by 4Tin mechanical defect.

Knowledge of the spectral properties of the impulse response can significantly improve the diagnosis of the mechanical system. Indeed, the information related to the repetition rates of the shocks is amplified in the resonance band and, consequently, an adequate filtering can make it possible to recover the fault signal while improving the signal-to-noise ratio.

Historically, an impact hammer test followed by a measurement of the impulse response is used to estimate the impulse response (for example see document US 7426426 B2). However, such a test is not practical in the majority of applications since the position of the excitation sources is often difficult to access.

Another technique is to compare the spectrum of a signal with a defect to that of a healthy signal (without defect). Again, this technique has the disadvantage that it requires the presence of a reference signal which is restrictive in several applications.

More recently, to get rid of a reference signal, spectral kurtosis (KS) has been proposed. As such, reference may be made to the following documents:

• J. Antoni, R.B. Randall, The spectral kurtosis: application to the vibratory surveillance and diagnostics of rotating machines, Mechanical Systems and Signal Processing (2006) 308-331.

• J. Antoni, The spectral kurtosis: a useful tool for characterizing non-stationary signais, Mechanical Systems and Signal Processing, Volume 20, Issue 2, 2006, Pages 282-307, ISSN 0888-3270.

KS has since been widely used in the field of fault detection based on the analysis of vibro-acoustic signals. KS is kurtosis calculated in frequency bands of a signal. To do this, it requires a measure of impulsivity (shock) in different frequency bands.

However, using the KS requires the noise to be stationary and Gaussian.

However, this assumption of stationarity is rarely satisfied. In particular, the vibratory signal contains a periodic deterministic part which questions the effectiveness of the KS.

Thus, in order to avoid the drawbacks linked to this deterministic part, in the aforementioned documents, the vibratory signal requires pre-processing to ^ eliminate the deterministic part.

However, this preprocessing requires on the one hand a significant computing time (for example an autoregressive filtering requires the calculation of a hundred coefficients) and, on the other hand, often distorts the signal (for example the filtering and whitening methods ).

In addition, it has been found that even after application of these pretreatments, the KS is by construction sensitive to non-stationary noises which exist abundantly in complex systems operating under severe conditions. These non-stationary noises are often generated by other interfering sources of mechanical, electrical or / and electromagnetic nature.

PRESENTATION OF THE INVENTION

An object of the invention is to have a processing of vibrational signals which is more robust than spectral kurtosis.

To this end, the invention proposes a method for monitoring a rotating machine comprising at least one rotating part capable of generating vibrations, each part having at least one characteristic frequency of fault, a vibratory signal representative of these vibrations having been acquired by a sensor, a vibratory digital signal x (t) having been obtained, the method comprising a step of determining, for each characteristic frequency a _d , of a selective spectral kurtosis of the digital signal which is defined by the part of the spectral kurtosis related to said characteristic frequency.

The invention is advantageously supplemented by the following characteristics, taken alone or in any of their technically possible combinations:

the method comprises a step of obtaining a Wiener filter estimated from the selective spectral kurtosis thus determined, and a step of filtering the digital signal by means of the filter thus obtained so as to bring out a frequency signature generated by a defect ;

the process comprises a step of determining a spectrum of the square envelope defined by the discrete Fourier transform of the absolute value of the Hilbert transform of the filtered signal;

the determination of the selective spectral kurtosis comprises the following sub-steps: determination (E11) of a short-term Fourier transform of the digital signal defined by = DFT ^ _màf (h (ne _t Îât) x (n <5 _t )} 0 <m <S with At / <5 _t = N _h (l - p) where p denotes the overlap rate of the sliding window and S = T / At is the numerical length of the short-term Fourier transform with respect to to the variable iAt and T is the duration of acquisition of the signal; normalization (E12), for each frequency mAf, of the short-term Fourier transform obtained in the following manner Χ _Ν (ΐΔί; τηΔί) = ^{x (t} At ,my/·)

CT _x (mAf)

S_ - £

CT _x (male) = - ^ _{η = 0} | Ύ (ίΔί; male) | ² is the standard deviation of the short-term Fourier Transform from the time variable at the frequency mAf; determination (E13) for each mAf frequency, of the discrete Fourier transform with respect to a time axis

F _2X (a.Aa; mAp = | DFT ^ _aA J | X _w (ÎAt; mAp | ² }, OÙ Δα = is the frequency resolution compared to the secondary frequency aAa; determination (E14) of the selective spectral kurtosis, for the frequency ad, by means of the following functional: Κ55 ^ (τηΔβ = Za = c ./ <lUxCaAa; mAf) \ ² , with ad = kAa, c = 1,2,3 ... and such that c <S / 2k - 1.

- The method comprises a step (E0) of measurement and acquisition of a digital vibratory signal x (t), the acquisition being implemented by an accelerometer or else a microphone;

the vibratory signal is a vibro-acoustic signal, the acquisition being implemented by a microphone;

- The method comprises a step of angular resampling of the vibratory digital signal.

The invention also relates to a monitoring system comprising a processing unit configured to implement a method according to the invention and may further comprise a unit for measuring and acquiring a digital vibratory signal x (t), said acquisition measurement unit comprising a sensor preferably an accelerometer or a microphone.

The invention also relates to a computer program product comprising program code instructions for executing the steps of the method according to the invention, when this method is executed by at least one processor.

Thus, the invention allows the detection of faults in a rotating machine by envelope analysis based on the KSS of a vibratory signal.

In particular, the invention is based on the implementation of a selective spectral kurtosis having properties similar to spectral kurtosis while being at the same time:

• robust against the deterministic periodic contribution (therefore pretreatment is not required), and • robust against non-stationary noise (generated by mechanical or electromagnetic sources).

The selective spectral kurtosis proposed in the invention allows the measurement of periodic impulsivity and therefore has the capacity to promote repetitive impulses instead of simple impulses (as is the case with spectral kurtosis).

PRESENTATION OF THE FIGURES

Other characteristics, objects and advantages of the invention will emerge from the description which follows, which is purely illustrative and not limiting, and which should be read with reference to the appended drawings in which, in addition to FIG. 1 already discussed:

- Figure 2 illustrates steps of a monitoring method according to the invention;

- Figure 3 illustrates sub-steps of a step of the monitoring method according to the invention;

- Figure 4 illustrates a spectral kurtosis and a selective spectral kurtosis implemented in the invention;

- Figure 5 illustrates spectra of a square envelope for a vibration signal with a defect (top curve) and for two vibration signals with no faults (the two curves below).

DETAILED DESCRIPTION OF THE INVENTION

FIG. 2 illustrates an exemplary implementation of a monitoring system and method according to the invention.

A rotating machine 1 is monitored to detect possible mechanical faults. Such a rotating machine is, for example, a helicopter, airplane or even car engine.

A rotary machine is therefore generally a part or a set of several rotating parts. In the case of several parts, the latter are housed in a casing which envelops the whole.

A measurement unit 2 makes it possible to measure and acquire a vibratory signal generated by the rotation of parts of the rotary machine.

Such an acquisition unit comprises a sensor 21 configured to measure the vibration signal. The sensor 21 makes it possible to obtain an analog signal representative of the vibration signal and is connected to an acquisition chain 22 configured to supply a digital vibration signal x (t). A sensor 21 is for example an accelerometer.

Alternatively, the measuring unit 2 makes it possible to measure a vibro-acoustic signal. In this case, the sensor 21 is a microphone. In what follows we only speak of a vibratory signal which is by no means limiting and also includes the notion of vibro-acoustic signal.

The acquisition chain 22 includes, as such, a conditioner (not shown), an analog anti-aliasing filter (not shown), a blocking sampler (not shown) and an analog to digital converter (not shown). The elements of the acquisition chain 22 are well known to those skilled in the art and will not be described in more detail here.

The digital vibratory signal x (t) obtained is then brought to a processing unit 3 configured to implement a monitoring method with the aim of monitoring the rotating machine 1 and detecting any mechanical faults. In particular, as will be detailed in the following, a monitoring signal is obtained making it possible to highlight possible faults.

The processing unit 3 is in particular configured to implement various steps which will be described below.

The processing unit 3 makes it possible to communicate the monitoring signal to a display unit 4 for display so that it can be viewed by an operator in particular. The display unit can also display various indicators generated from the monitoring signal.

The monitoring process is now described.

A vibratory digital signal x (t) = x (n5 _t ) is acquired (step EO) via the measurement and acquisition unit 2. This digital signal is the digital version of the analog signal measured and acquired by the sensor 21 and the acquisition chain 22.

The digital signal x (t) has a predefined duration T = L.ô _t , with 5 _t = 1 / F _S is the sampling period and L the digital length of the signal, during which the rotating machine 1 operates under a regime stationary.

For example, an acquisition time of two (2) seconds with a sampling frequency of 50 kHz is used for an aircraft engine.

It is considered that the rotating machine 1 operates under a stationary regime when it operates at an almost constant speed and load.

This steady state condition is assumed to be satisfied if the standard deviation of the speed profile v (t) does not exceed 5% of its average value:

- 2, v (nô _t y <-j— 2, v (n5 _t ).

n = 0 n = 0

It is considered that the kinematics of the rotary machine 1 to be monitored is known. In particular, the characteristic frequencies of the different rotating parts are known.

These characteristic frequencies are, in known manner, the frequencies of occurrence of the fault in each rotating part.

The advantage of knowing these characteristic frequencies is that if a rotating part has a defect, it will then be particularly visible at this frequency.

Thus, for each characteristic frequency a _d of the rotating machine, a selective spectral kurtosis is determined (step E1). Selective spectral kurtosis (denoted KSS) is defined by the part of spectral kurtosis linked to the characteristic frequency.

The notion of selectivity is linked to the fact that the KSS is, in other words, a selection of a part of the KS usually calculated.

The KSS advantageously makes it possible to characterize the transients linked to a target characteristic frequency. The objective of determining the KSS is to estimate an optimal Wiener filter from the latter in order to minimize the noise and best bring out the digital vibrational signal, the signature generated by a fault.

Alternatively, an angular resampling step (step E2) can be implemented on the acquired digital signal. The advantage of such a step is to synchronize the cyclic content to the motor shaft. The angular resampling requires the presence of a tachometric signal, acquired jointly with x (t), or a method of estimation of the instantaneous frequency from the vibratory signal making it possible to transform the signal in the angular domain (x (0) where Θ is the angular variable).

FIG. 3 illustrates the main steps allowing the determination of the KSS for a characteristic frequency a _d , noted mathematically KSS “^ (f).

First, the short-term Fourier transform (TFCT) is applied (step E12) using the sliding window / r (n <5 _t ) of length / ν _Λ to the digital vibrational signal x (n <5 _t ).

The TFCT is defined as follows

Ύ (ίΔί; mû /) = DFT ^ _{m & f} {h (nô _t - iAt) x (nô _t )} 0 <m <S with At / 5 _t = N _h (l - p) where p denotes the overlap rate of the sliding window and S = T / At is the digital length of the TFCT compared to the variable ίΔί (with T designating the duration of acquisition of the signal).

The TFCT is a matrix of dimension N _h by S designating the frequency and time axes respectively.

Then, for each mAf frequency, the TCFT is normalized as follows (step E13):

^{W 7} v _x (m & D

WHERE, o _x (mAf) = | 2 ^ ₌ θ | Ύ (ίΔί; ητΔ / ·) | ² is the standard deviation of the TFCT from the time variable at the frequency mAf.

The size of the window N _h should preferably be a power of 2 and is such that Δ / is of the order of a few hundred Hertz.

The discrete Fourier transform of X _N (lAt; mAf) is then calculated (step E14) for each frequency mAf, relative to the time axis and is defined as follows:

r _2X (a. Δα; ιώΔ /) = | ϋΡΤ ^ ₃ . _Δα {| Χ _Ν (ΐΔί; mAf) \ ² }, where Δα = is the frequency resolution with respect to the axis of the cyclic frequencies aAa. The quantity obtained is a matrix having the same dimension as the TFCT but comprising two frequency variables of Hertz unit.

Finally, let / c, the integer such that, the target fundamental frequency is written in the form: a _d = kAa, the KSS is then obtained (step E15) and is defined as follows

KSS “£ (mAf) = | Γ _2Χ (3Δα; τηΔ / ·) | ² , a = ck with c = 1,2,3 ... and such that c <S / 2k - 1.

In addition, if the target fundamental frequency does not coincide with the frequency resolution Δα, an interpolation (for example cubic) can be implemented to perform the sum.

Illustrated in Figure 4 is a KS (left) and a KSS (right) of a rotating machine for which a gear unit may have chipping in one of its wheels. The KSS is obtained for a characteristic frequency linked to the reducer.

Note that the KS has negative values. These values are due to the fact that the KS is affected by the deterministic part. On the other hand, the KSS allows to better isolate the signature of the target fault (a little before 0.1 Hz) while for the KS there are more oscillations.

In addition, in the case where the angular resampling is carried out, the KSS can be applied to the angular signal χ (θ) in a manner perfectly similar to that described for x (t).

Advantageously, the KSS obtained is used to estimate an optimal Wiener filter.

In particular, the Wiener filter associated with the train of interest of frequency a _d is estimated (step E3) from the KSS as follows

H _ad (màf) = JxSSf (mAf).

The signal x (t) is then filtered (step E4) by this filter in order to favor the impulsivity associated with the frequency of the suspect part from the spectral point of view of the rotating machine being monitored. The filtered signal, noted y (t), is then given by: y (ours) = IDFT) ^ _n5t {H _ad (jôn Χ (/ δ /)}, where X (jôf) is the Fourier transform of x ( t) that is to say,

X (j8f) = OF ^ _jS fx (ours)}, and ôf = F _s / L is the frequency resolution.

In addition, an intermediate interpolation step may be required to be able to calculate the coefficients of H _Ud at frequencies jSf from those calculated at frequencies mhf.

Finally, to locate the characteristic frequency a _d and its harmonics, the spectrum of the square envelope is calculated (step E5).

Such a spectrum makes it possible to construct a decision as to the existence or not of the defect in question. It is defined as:

SESfjôf) = | ΌΡΎ ^ _δί {\ ΗΜβη {γ (ηδί)} \ ² } \.

We illustrated in Figure 5 a comparison of the spectrum of the square envelope applied on the system whose KS and KSS are presented in Figure 4, with those applied on two other healthy systems (middle curve and bottom curve) . The curve at the top indicating the system with the fault, shows the presence of lines located in the frequency of rotation of the suspect wheel equal to 0.0011 (which is equal to the frequency characteristic of the fault) and its harmonics.

From the spectrum, a vibratory signature of a defect is obtained. A fault detection step (step E6) is then implemented. For example, an alarm on the existence of a fault can be triggered once the value of the square envelope spectrum exceeds 3 times its standard deviation.

In addition, this spectrum can be displayed (step E7) on the display unit 4.

As already mentioned, the KSS is the part of the KS linked to a characteristic frequency.

Indeed, the Applicant has broken down the KS so as to identify the components of KS which are linked to the fault frequency sought a _d In what follows, the developments carried out by the Applicant are presented so as to demonstrate the substitution of the KS such as known by the KSS.

Let KS be defined at each mAf frequency by the following function:

where, M _4X (mAf) and M _2X (jnAf) are the moments of order 4 and 2 respectively. Two possible estimators of M _4X (mAf) and M _2X (mAf) are written as a function of the TFCT of% (t) as follows:

SL · i = o

SL · t = o

Is

Ψ _2Χ (3Δα; ζηΔ /) = | DFT / _At ^ _aA „{| X (iAt; mA / ') | ² }

On the one hand, for a = 0,

S-l

Ψ _2χ (0; ητΔ /) = - ^ | Α (ίΔί; mAf) \ ² = M _2X (mAf) = a _x (mAf) ² i = o

On the other hand, using Parseval's equality,

Sl Sl | Α (ίΔί; md /) | ⁴ = S ^ | Ψ2χ (3Δα; mAf) \ ² = S.M4X (mAf) i = 0 a = 0

Thus, the KS can be written as a function of Ψ:

= + | _{V (0: mW} a = 0

Sl | r _x (aâa; mAf) \ ² , a = l because

Ψ _2Χ (3Δα; mAf) 1 (\ X (iAt; mAf) \ ² ) 1

Ψ _2Χ (Ο; mW = S ^DFT “- (j = 5 mAf ^} = r _2X (aAa; m Af)

Let k be the integer such that the target characteristic frequency is written in the form a _d = kAa, therefore the sum in KS, can be broken down into two parts: a first sum on the target frequency a _d and its harmonics, and a second sum on the rest of the frequency components:

K _4X (m Af) = —1 + lr _2X (aAa; mAf) l ² +] Γ _2Χ (αΑα; m Af)] ² .

a = c.k a * c.k

It is recalled that the KSS was proposed as follows

KSS “£ (mAf) = lr _2X (aAa; mAf) l ² .

a = c.k

In KS, the sum is made over the entire frequency axis. Thus, all the frequency components interfere including all the transients coming or not from different existing faults, the non-stationary random pulses or even the deterministic components which calls into question its efficiency while the KSS only takes into account the linked frequency components to the part of the system to be monitored which allows it to be more robust to interference.

Claims (9)

1. Method for monitoring a rotary machine (1) comprising at least one rotating part capable of generating vibrations, each part having at least one characteristic frequency of fault, a vibratory signal representative of these vibrations having been acquired by a sensor (21), a vibratory digital signal x (t) having been obtained, the method comprising a step (E1) of determining, for each characteristic frequency a _d , of a selective spectral kurtosis of the digital signal which is defined by the part of the spectral kurtosis linked to said characteristic frequency, the determination (ET) of the selective spectral kurtosis including squstelapes.suiyantes .: ~ determination (Έ11) of a short-term Fourier transform of the digital signal defined by X (i &t; ηιΔ /) = - ίΔέ) χ (ηό7)} 0 < m <S with.àt / - A _{/ t} (1 p) ... p.ù ... p .... çfesi.gn.e..je..taux. ... of .. overlap ... the..Slidingwindow.andS - T / t ... eslJ.aJongueur ... ^^^^^ te £ nie..eaL [a.Be.orLà .. ^ - normalization (Έ12), for each frequency m & f. of the transform of Short-term Fourier obtained as follows = îLèîlZFle. _o q .................................................. .................................................. ........................................ '......... . a _x (yn & f) ~ j Σ màfW is the standard deviation of Sa Short Fourier Transform ΙθΠΏ.θ.β.3 £. [φ.β0.Ο ... Ι. ^ - determination (E13) for each frequency m & f, of ia discrete Fourier transform with respect to a time axis z - ¹ · - <, Z r _{2 ¥} ra.Aœ: m & f) = - 0ΡΤ / _Δί ... _{> άΔα} { | Χ _Α > (. ΊΔί; màfA *}, b where ~ is the frequency resolution with respect to the secondary frequency aAa: -determina ^ means of the following functional: KSSf / (rn & .f) ~ | Γ _2ί (3Δα; ΐηΔ /) Ι ² , a ~ ck Ayecn ^ - k & ayc ~ 1,2,3 .... etteLguec <S / 2k 1, 2. Method according to claim 1, comprising a step (E3) of obtaining a Wiener filter estimated from the selective spectral kurtosis thus determined, and a step of filtering (E4) of the digital signal by means of the filter thus obtained so as to bring out a frequency signature generated by a defect. 3. Method according to one of the preceding claims comprising a step (E5) of determining a spectrum of the square envelope defined by the discrete Fourier transform of the absolute value of the Hilbert transform of the filtered signal. 4. Method according to one of the preceding claims, in which the determination (E1) of the selective spectral kurtosis comprises the following substeps: - determination (E11) of a short-term Fourier transform of the digital signal defined by X (iAt; mAf) = ------- θ- < m <S with At / S ^ - N _h (l — p) where p denotes the overlap rate of the sliding window and S - T / At is the numerical length of the short-term Fourier transform with respect to the variable iAt and T is the signal acquisition time; - normalization (E12), for each mAf frequency, of the transform Short-term Fourier obtained as follows XJjApmAf) - _o qo _x (mAf) - | mAff ² is the standard deviation of the short-term Fourier Transform compared to the time variable at the frequency mAf) - determination (E13) for each frequency mAf, of the discrete Fourier transform with respect to a time axis i r _a (aA "; mA /) - j ΒΡΤ ^ _> ^ {| Χ _# (ΐΔί; ^ Δ / ψ], where Aa - is the frequency resolution with respect to the secondary frequency αΔ"; - determination (E14) of the selective spectral kurtosis, for the frequency a _a , by means of the following functional: KSS% £ (mAf) = | r _z ^ (aA «; tnA / ') | ^a , ac.k With a _d - kAa, c - 1,2,3 ... and such that c <S / 2k — L 64. Method according to one of the preceding claims, comprising a step (EO) of measurement and acquisition of a digital vibratory signal x (t), the acquisition being implemented by an accelerometer. 65. Method according to one of claims 1 to 34, wherein the vibratory signal is a vibro-acoustic signal, the acquisition having been carried out by a microphone. 76. Method according to one of the preceding claims, comprising a step (E2) of angular resampling of the vibratory digital signal. §7. Monitoring system comprising a processing unit configured to implement a method according to one of the preceding claims. 68. Monitoring system according to claim §7, further comprising a unit (1) for measuring and acquiring a digital vibratory signal x (t), said unit (1) for measuring acquisition comprising a sensor ( 21) preferably an accelerometer or a microphone. 469. A computer program product comprising program code instructions for executing the steps of the method according to one of the preceding claims, when this method is executed by at least one processor. 1. Method for monitoring a rotary machine (1) comprising at least one rotating part capable of generating vibrations, each part having at least one characteristic frequency of fault, a vibratory signal representative of these vibrations having been acquired by a sensor (21), a vibratory digital signal x (t) having been obtained, the method comprising a step (E1) of determining, for each characteristic frequency a _d , of a selective spectral kurtosis of the digital signal which is defined by the part of the spectral kurtosis linked to said characteristic frequency, the determination (E1) of the selective spectral kurtosis comprising the following substeps: - determination (E11) of a short-term Fourier transform of the digital signal defined by X (iAt; mAf) = DFT ^ g _àf (h (ne _t - ίΔί) χ (ηδι)} 0 <m <S with At / 8 _t = N _h (l - p) where p denotes the rate of overlap of the sliding window and S = T / At is the numerical length of the short-term Fourier transform with respect to the variable iAt and T is the duration of signal acquisition; - normalization (E12), for each frequency mAf, of the short-term Fourier transform obtained as follows X _N (iAt; mAf) = ^x ^ ' ^m ^ _O ù o _x (mAf) = - ^ _{n = 0} \ X (iAt; mAf) \ ² is the standard deviation of the short-term Fourier Transform from the time variable at the frequency mAf; - determination (E13) for each frequency mAf, of the discrete Fourier transform with respect to a time axis r _2X (a. Aa-, mAf) = | DFTf _At ^ _aA J | X _w (jAt; mAf)! ² }, where Aa = is the frequency resolution with respect to the secondary frequency aAa; - determination (E14) of the selective spectral kurtosis, for the frequency a _d , by means of the following functional: KSS “f (mAf) = | F _2X (aAa; mA /) | ² , a = ck With a _d = kAa, c = 1,2,3 ... and such that c <S / 2k - 1. 2. Method according to claim 1, comprising a step (E3) of obtaining a Wiener filter estimated from the selective spectral kurtosis thus determined, and a step of filtering (E4) of the digital signal by means of the filter thus obtained so as to bring out a frequency signature generated by a defect. 3. Method according to one of the preceding claims comprising a step (E5) of determining a spectrum of the square envelope defined by the discrete Fourier transform of the absolute value of the Hilbert transform of the filtered signal. 4. Method according to one of the preceding claims, comprising a step (EO) of measurement and acquisition of a digital vibratory signal x (t), the acquisition being implemented by an accelerometer. 5. Method according to one of claims 1 to 3, wherein the vibratory signal is a vibro-acoustic signal, the acquisition having been implemented by a microphone. 6. Method according to one of the preceding claims, comprising a step (E2) of angular resampling of the digital vibrational signal. 7. Monitoring system comprising a processing unit configured to implement a method according to one of the preceding claims. 8. Monitoring system according to claim 7, further comprising a unit (1) for measuring and acquiring a digital vibrating signal x (t), said unit (1) for measuring acquisition comprising a sensor (21 ) preferably an accelerometer or microphone. 9. A computer program product comprising program code instructions for executing the steps of the method according to one of the preceding claims, when this method is executed by at least one processor.