Method for monitoring rotating machine in order to detect faults in aircraft bearings

Through incremental transformation and spectrum standardization technology, the vibration signals in the aircraft engine are analyzed to form first-order and second-order spectrograms, solving the problem of bearing fault detection in high-speed non-steady state environments, and achieving high accuracy and robustness fault detection.

CN119096128APending Publication Date: 2024-12-06SAFRAN SA +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202380035635.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-04-22
Filing Date
2023-04-20
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art is difficult to effectively monitor bearing failures in aircraft engines, especially in high-speed and non-steady operating states, which are affected by high noise and complex vibration signal environments, resulting in inaccurate and unstable fault detection.

Method used

Using incremental transformation and spectrum standardization technology, the vibration signals are subjected to smoothly cyclic analysis of the first and second order cyclic patterns to form first and second order spectrograms, which are used to detect bearing fault characteristics and improve signal-to-noise ratio through spectrum autocorrelation and standardization processing. , remove noise interference.

Benefits of technology

It realizes reliable detection of aircraft bearing failures in high-speed and non-steady environments, improves the display and detection accuracy of fault characteristics, extends the service life of the machine and avoids faults.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure BDA0005095973870000111
    Figure BDA0005095973870000111
  • Figure BDA0005095973870000112
    Figure BDA0005095973870000112
  • Figure BDA0005095973870000121
    Figure BDA0005095973870000121
Patent Text Reader

Abstract

One aspect of the invention relates to a method for monitoring a rotating machine in order to detect a fault in a bearing, the method comprising the steps of: o acquiring a vibration signal measured by a vibration sensor from the rotating machine; o, determining a first-order spectrogram by performing first-order cyclostationary analysis on the vibration signal by using incremental transformation and spectrum standardization; o, determining a second-order spectrogram by performing second-order cyclostationary analysis on the vibration signal by using average cyclic coherence, incremental transformation and spectrum standardization; and o detecting a vibration characteristic of a fault in the bearing based on the first order spectrogram and the second order spectrogram.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The technical field of the invention is that of monitoring rotating machines, more specifically aircraft engine bearings by analyzing vibration signals and engine power transmission.

[0002] The invention relates to a method for monitoring a rotating machine in order to detect faults in a bearing. Background Art

[0003] The monitoring of aircraft bearings is of vital importance for the transport industry, and in particular for aviation. Bearings are, in fact, one of the most stressed and critical mechanical components of various equipment (compressors, turbojets, reducers, etc.). Therefore, the probability of damage to these mechanical components is high, which can lead to damage or even complete failure of the aircraft engine. In addition to the economic cost, such failures also endanger the safety of the equipment and users.

[0004] Predictive maintenance based on continuous monitoring of the health of the bearings will make it possible to detect and identify a possible fault in one of the bearings, thus preventing damage and / or failure of the bearing. The replacement of the bearing will therefore depend on the results of an analysis carried out using a system capable of such monitoring.

[0005] There are various methods for monitoring rotating machines, in particular by measuring the vibrations caused by the operation of the machines and the drive elements. However, the methods known from the prior art in the field are difficult to apply to aircraft machines, in particular aircraft and helicopter engines. In fact, monitoring the operation of aircraft engines is made difficult, firstly because of the accessibility and overall size of the shaft line and the rotating engine in a critical environment, which prevents the installation of a monitoring system that should use a small number of sensors and / or place the sensors at a certain distance from the bearings to be monitored, and secondly because of the fact that the engine speeds of these rotating machines are very high, which greatly changes the statistical properties of the vibration signal compared to the low speed regime.

[0006] In particular, bearings in aircraft need to rotate at very high speeds, on the same order as the resonant frequency of the system. This mode of operation significantly changes the statistical properties of the vibration sources and makes the vibration signals measured in situ very different from those obtained during the test phase (e.g. on a commercial test bench). For example, at high speeds, the superposition of transient waves generated by shocks reduces the impulsiveness of the faulty component, which is incompatible with most of the methods of the prior art, which are mainly based on the impulsive nature of the signal.

[0007] Another difficulty specific to aircraft operation is the presence of transient states, i.e. significant changes in speed, with very strong acceleration or deceleration phases, which are often sudden. These are also called non-steady-state states. In addition to these highly variable speeds, there are also variations in the radial and / or axial loads of the different engine components. However, the methods of the prior art do not allow such transient operating modes to be taken into account. However, not taking these non-steady-state states into account can be detrimental. On the one hand, it cannot be guaranteed that the recording in steady-state mode is long enough to allow analysis; on the other hand, the vibration characteristics of some bearing faults may be more apparent in some modes than in others, and therefore scanning several modes can increase the probability of detecting weak signals.

[0008] Furthermore, there are monitoring limitations directly related to the complexity of the engine. Here, the accessories of the turbine engine make it possible to record a very large number of vibration signatures in the signal, each of which is associated with one of the elements of the engine, such as gears, bearings, fans, blades, shafts, compressors, etc. These elements generate significant vibrations even in normal operation, which mask the vibration component generated by the faulty bearing and increase the noise. Therefore, traditional vibration analysis methods are not reliable, because for this type of application, source separation is not easy to achieve, especially if the signal is non-stationary.

[0009] In addition to the rotating elements of the turbine engine, other sources of vibration, known as asynchronous sources, make it more difficult to identify a faulty bearing because they add highly non-stationary components to the signal.

[0010] Part of the asynchronous sources is directly related to the design of the engine. In fact, a turbine engine generally comprises two rotating shafts that are not rigidly attached, each with a different rotation speed, with asynchronous operating modes during the steady-state phase of the engine speed and during the acceleration and deceleration phases. These are called the rotation speeds N1 and N2 of the shafts; this is also true for twin-engine aircraft. These asynchronous operating modes also generate additional harmonic contents, which do not facilitate the distinction of multiple frequencies in the signal spectrum associated with the operation of the different rotating elements of the engine. In fact, some harmonic components overlap or mask each other, resulting in a poor estimation of the operating state of the engine. In addition, the variability of the engine speed makes it difficult to detect a faulty bearing, since the harmonic signature associated with this bearing will also be highly variable. In addition, the variability of the engine speed also affects the variability of the statistical properties of the vibration signal.

[0011] In addition to the rotating elements of the engine, other asynchronous sources also add noise to the vibration signal through the presence of highly non-steady-state components. In fact, when an aircraft engine is operating, there are many asynchronous noise sources, whether solid, airborne, electromagnetic or thermal noise. The main sources of noise are the combustion of fuel to move the aircraft, the aerodynamic flow of fluids in the engine and around the device (such as flotation and turbulence), and the resonance of different elements in the engine and aircraft due to the vibrations generated by the engine. These multiple, mainly broadband noise sources mean that the signal-to-noise ratio level in the acquired vibration signal is low, and it is difficult to distinguish between the vibration signature of a faulty bearing and the noise therein.

[0012] Due to the unique and complex statistical properties of the vibration signals of rotating machines, the different noise sources mentioned above are not easily removed or separated using prior art methods. For example, aerodynamic disturbances are known to have broadband and random spectral characteristics, while fan / blade characteristics are periodic, while shafts and gears are sinusoidal components whose phase depends on the position of the shaft to which they are mounted.

[0013] Monitoring methods based on advanced analysis of time signals and spectrograms are known from the prior art (for example, FR2952177B1, FR3076348B1 and EP2496921A1). However, these methods do not solve the problems posed by aviation applications. That is, these methods are not robust or only marginally robust to non-steady-state operating conditions of turbine engines; they also do not allow taking into account the specific statistical structure of the vibration signal caused by high rotation speeds; and they are difficult to apply to highly noisy signals from asynchronous sources, resulting in a low level of signal-to-noise ratio.

[0014] Therefore, there is a need for a monitoring device for a rotating machine that is robust to high-speed, non-steady-state operating conditions of the machine to detect bearing faults in complex, high-noise environments. Summary of the invention

[0015] The present invention provides a method for solving the problems discussed previously by enabling monitoring of a rotating machine operating at high speeds by means of measuring vibration signals, thereby distinguishing the vibration signature of a faulty bearing from the vibration signatures of other elements and from the noise generated by the operation of the machine, while taking into account transient states of the engine.

[0016] A first aspect of the invention relates to a method for monitoring a rotating machine to detect faults in a bearing, the method comprising the following steps:

[0017] - acquiring a vibration signal from a rotating machine measured by a vibration sensor;

[0018] - Determine the first-order spectrogram by performing a first-order cyclostationary analysis of the vibration signal using delta transformation and spectrum normalization;

[0019] - Determine the second-order spectrogram by performing a second-order cyclostationary analysis of the vibration signal using average cyclic coherence, delta transformation and spectral normalization;

[0020] -Detect the vibration characteristics of bearing faults based on the 1st and 2nd order spectrograms.

[0021] By "rotating machine" is meant an engine machine whose engine transforms the energy supplied to the engine into a rotational movement (e.g. through an axis). In the context of the present invention, this refers in particular to an aircraft, such as an airplane or a helicopter, but it may also refer to a wind turbine engine, a bearing vehicle engine, etc.

[0022] By "vibration sensor" is meant a sensor suitable for measuring vibrations to which a structure is subjected. For example, it may be an accelerometer-type sensor based on the piezoelectric effect, a laser vibrometer, a capacitive or eddy current displacement sensor, etc.

[0023] By "cyclostationary signal", it is meant a signal whose statistical properties vary periodically with time, as opposed to a stationary (steady-state) signal whose statistical properties do not vary with time. This is particularly true for vibrational features of rotating elements of an axis, whose harmonic content depends on the operating speed of the engine. Specifically, the statistical properties of such features will not be affected by the rotation period, a period being one or more time periods of shaft rotation, while the rotation period will have variability that varies with the non-steady state of the engine speed. Thus, "cyclostationary analysis" refers to analyzing signals whose statistical properties vary periodically with time.

[0024] According to the invention, faulty bearings in rotating machines can be detected by analyzing a high noise vibration signal acquired on the machine. The analysis is performed by studying the deterministic properties (also called periodic properties) and the cyclostationary properties of the harmonic characteristics of the fault. Therefore, the method is particularly suitable for high-speed applications such as helicopter, aircraft or drone engines, for example for monitoring accessory gearboxes (AGB), equipment brackets, transmission systems from engines to aircraft, etc. The method can also be applied to other types of rotating machines, such as car, truck or train axles, or to equipment such as gearboxes, alternators, pumps, aircraft landing gear wheels, etc.

[0025] The method is easy to implement within the scope of aeronautical applications, since it requires at least one vibration sensor, whose dimensions are compatible with the overall size and compactness issues of aeronautical systems. This means that only a small amount of equipment needs to be installed to make the method according to the invention operational, reducing the cost of monitoring. Furthermore, since vibration sensors are more robust to critical environments and more sensitive to high frequencies than speed or position sensors, the method is suitable for aeronautical applications.

[0026] Using the delta transform to form the 1st and 2nd order spectrograms allows the sources that interfere with the vibration signature of the faulty bearing to be removed. For example, the delta transform can be based on knowledge of the kinematics of the engine shaft to systematically remove these frequency components.

[0027] By "spectral normalization" is meant the operation of making the spectrum have zero mean and unit standard deviation. With the aid of spectral normalization, the 1st order spectrogram can be statistically reduced, since spectral normalization reduces and makes the spectral properties of the spectrogram consistent. In addition, spectral normalization implicitly whitens the signal spectrogram, thereby improving the signal-to-noise ratio and compensating for harmonic content associated with vibration modes and resonances of engine components.

[0028] These various tools used in the present invention thus make it possible to improve the robustness of the detection of the vibration signature of a faulty bearing with respect to the noise contained in the vibration signal. Thus, the detection of bearing faults can be performed at an early stage (when the first symptoms appear), thereby avoiding the failure of the turbine engine.

[0029] The method according to the invention thus provides a means of assisting a specialist or operator, for example, in establishing a reliable diagnosis of the health of an aircraft bearing by means of the emergence of harmonic signatures indicative of a bearing fault.

[0030] In addition to the features discussed in the previous paragraph, the method according to the first aspect of the invention may have one or more of the following additional features, which may be considered alone or in any technically possible combination.

[0031] In one embodiment, the step of detecting a vibration signature of a faulty bearing comprises identifying the faulty bearing from said vibration signature.

[0032] By means of this embodiment, it is possible to facilitate an expert or operator to diagnose a faulty bearing and the health of the rotating machine.

[0033] In one embodiment, the monitoring method according to the previous embodiment further comprises a maintenance step for the identified faulty bearing.

[0034] By means of this embodiment, a maintenance operation may be triggered to repair or replace the identified faulty bearing in order to extend the useful life of the rotating machine and prevent failure of the rotating machine.

[0035] In one embodiment, vibration signals are acquired at a plurality of different operating stages of the rotating machine.

[0036] The interest of analyzing the vibration signal according to the different operating phases of the rotating machine is to guarantee the diversity of the acquired vibration signatures and to ensure that if the rotating machine includes a faulty bearing, its vibration signature will actually be included in the signal. In fact, this signature is likely to appear only at a specific engine speed. Thus, the monitoring covers a large part of the aircraft's operating margins and increases the probability of detecting a bearing fault in the rich frequency content of the signal.

[0037] In one embodiment, the 1st order spectrogram and the 2nd order spectrogram are determined based on a Fourier transform applied to the vibration signal over a plurality of consecutive time windows having a duration between 0.1 seconds and 10 seconds.

[0038] By using a small time window to analyze the vibration signal, the statistical properties of the vibration signal (which vary with different engine speeds) can be considered to be in a steady state within the duration of the time window. Therefore, the analysis of the vibration signal ensures that the variability of the rotational operating speed of the machine is taken into account. This embodiment also facilitates detecting the vibration signature of a bearing fault by tracking the deviation of the characteristic frequency of the bearing.

[0039] In one embodiment, the acquiring step further comprises acquiring a rotating machine speed signal measured by a speed sensor.

[0040] The measured speed signal provides a speed reference at any measurement time. Here, the speed signal can be used to resample the vibration signal according to the operating cycle of the machine. Therefore, the speed signal can be used to reset the vibration signal to the engine speed. In other words, the vibration signal is used to identify the operating cycle of the machine in the vibration signal.

[0041] In one embodiment, the method further comprises the following steps in the step of determining the first-order spectrum diagram:

[0042] -For each time window in multiple time windows:

[0043] oResample the vibration signal from the velocity signal to obtain the angle signal;

[0044] oDetermine the 1st order spectrum by applying Fourier transform to the angle signal;

[0045] o determining a corrected 1st order spectrum by applying a delta transform to the 1st order spectrum;

[0046] o determining a normalized 1st-order spectrum by applying spectral normalization to the corrected 1st-order spectrum;

[0047] o Determine the flattened 1st order spectrum by applying spectral autocorrelation to the normalized 1st order spectrum;

[0048] - Concatenating the flattened 1st order spectra determined for each time window to form a 1st order spectrogram.

[0049] In one embodiment, the method further comprises the following steps in the step of determining the first-order spectrum diagram:

[0050] -For each time window in multiple time windows:

[0051] o determining a corrected flattened 1st order spectrum by applying a delta transform to the flattened 1st order spectrum;

[0052] o determining a reduced 1st order spectrum by applying spectral normalization to the corrected flattened 1st order spectrum;

[0053] The concatenation step is to concatenate the reduced 1st order spectra determined for each time window to form a 1st order spectrogram.

[0054] In one embodiment, the method further comprises the following steps in the step of determining the second-order spectrum diagram:

[0055] - for each time window in the plurality of time windows;

[0056] oResample the vibration signal from the velocity signal to obtain the angle signal;

[0057] oRemove the deterministic part of the angle signal to obtain the corrected vibration signal;

[0058] oDetermine the second-order spectrum based on the cyclic coherence of the corrected vibration signal;

[0059] oAverage the 2nd order spectrum;

[0060] o Determine the flattened 2nd order spectrum by applying spectral autocorrelation to the averaged 2nd order spectrum;

[0061] - Concatenating the flattened 2nd order spectra determined for each time window to form a 2nd order spectrogram.

[0062] In one embodiment, the method further comprises, in the step of determining the second order spectrogram, performing the following steps:

[0063] -For each time window in multiple time windows:

[0064] o determining a corrected flattened 2nd order spectrum by applying a delta transform to the flattened 2nd order spectrum;

[0065] o determining a reduced 2nd order spectrum by applying spectral normalization to the corrected flattened 2nd order spectrum;

[0066] The concatenation step is to concatenate the reduced 2nd order spectra determined for each time window to form a 2nd order spectrogram.

[0067] By applying autocorrelation, the vibration signature of a faulty bearing can be made visible in the 1st and 2nd order spectrograms, thereby improving the detection of the vibration signature in question.

[0068] A second aspect of the invention relates to a system for monitoring a rotating machine to detect faults in a bearing, the system comprising:

[0069] - Get modules, including:

[0070] oVibration sensor;

[0071] oMemory; and

[0072] oProcessor;

[0073] - Processing module 30, comprising:

[0074] oMemory; and

[0075] oProcessor;

[0076] - Connection means between the acquisition module and the processing module.

[0077] A third aspect of the invention relates to a computer program product comprising instructions which, when the program is executed on a computer, cause the computer to implement the steps of the method according to the first aspect of the invention.

[0078] A final aspect of the present invention relates to a computer-readable recording medium comprising instructions which, when executed by a computer, cause the computer to implement the steps of the method according to the first aspect of the present invention.

[0079] The invention and its various applications will be better understood after reading the following description and viewing the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The drawings are illustrated in a schematic manner and are not intended to limit the scope of the present invention in any way.

[0081] - Figure 1 is a block diagram illustrating the sequence of steps in the method according to the present invention.

[0082] - Figure 2 is a graphical representation of the sliding time window analysis of the vibration and velocity signals.

[0083] - Figure 3 It is the frequency spectrum of the vibration signal obtained on the rotating machine and the theoretical frequency spectrum of the kinematics of the rotating machine.

[0084] - Figure 4 A comparison between the Fourier transformed spectrum and the delta transformed spectrum of the same vibration signal is shown.

[0085] - Figure 5 The effect of applying spectral normalization to the spectrum is illustrated.

[0086] - Figure 6 The effect of applying spectral autocorrelation to the spectrum is illustrated.

[0087] - Figure 7 is Figure 1 Comparison of the spectra processed sequentially during the steps of the method described.

[0088] - Figure 8 is based on Figure 1 The step of the method is to compare the frequency spectra of the vibration signal before and after processing.

[0089] - Fig. 9 The periodic coherence of the random part of the vibration signal is shown.

[0090] - Fig.10 Indicates the average in the time dimension Fig. 9 The spectrum of the periodic coherence.

[0091] - Fig.11 is through Figure 1 An example of a 1st order spectrogram determined by the method described in .

[0092] - Fig.12 is through Figure 1 An example of a 2nd order spectrogram determined by the method described in .

[0093] - Fig.13 is the spectrum obtained by averaging the 1st-order spectrogram in the time dimension.

[0094] - Fig.14 is the spectrum obtained by averaging the 2nd order spectrogram in the time dimension.

[0095] - Fig.15 is a diagram for implementing Figure 1 A diagram of a system for the method described in FIG. DETAILED DESCRIPTION

[0096] Unless otherwise stated, identical elements appearing in different figures have a single reference label.

[0097] The present invention relates to a method for monitoring faulty bearings of rotating machines (such as bearings of aircraft engine shafts) by vibration analysis. The method requires as input a vibration signal acquired on a rotating machine during a non-steady-state operation phase lasting several seconds. As a supplement, the method can use the measured values ​​of the instantaneous rotation speed or instantaneous position of the shaft rigidly attached to the shaft of the monitored bearing and the kinematics of the shaft near the bearing as input. Then two spectral analyses of the vibration signal are performed in parallel. Preferably, these processes are applied to a limited sliding time window (usually one second) of the vibration signal. The first process includes a series of operations applied to the 1st order spectrum of the vibration signal to delete unwanted harmonics therefrom, improve the detection level of the vibration characteristics of the faulty bearing and perform statistical reduction. The second process involves deleting the purely random part of the vibration signal after the tonal part, and applying a series of operations to the 2nd order cyclostationary statistics to reveal any hidden periodic patterns. At the output, the method provides two spectra, and preferably two spectrograms, carrying the 1st order and 2nd order cyclostationary information of the vibration signal. The vibration characteristics of the bearing fault can then be detected and ultimately used to identify the bearing including the fault.

[0098] The provided method is therefore a support tool for an expert or operator to diagnose the health of a bearing in a rotating machine and, if necessary, helps to identify a faulty bearing for maintenance.

[0099] By the "health" of a bearing, or more generally a machine part, it is meant that the bearing complies with a certain set of requirements, such as those arising from specifications related to its design, manufacture and use. When the health of a bearing deteriorates for any reason, it is called a faulty bearing.

[0100] By "faulty bearing" is meant a bearing whose mechanical or geometrical properties have been damaged, causing anomalies when the shaft to which it is assembled is rotated, usually due to imbalance. Damage may be caused by discontinuities in the properties of the bearing material, due to hazards occurring during the manufacturing of the parts, or due to fatigue of the bearing during use or handling. For example, the material may be weakened during the manufacturing process and its use, generating high local stresses in the weakened area, or causing failure after being subjected to an impact. In addition, damage may also be caused by imbalance or geometric or mechanical deformation of the bearing, which is the case, for example, when the bearing (such as the bearings in turbine engines) is subjected to very high stresses. Therefore, the term "fault" covers all forms of damage that a bearing may experience: material failure, inclusions, cracks, lubricant failure or contamination, changes in material properties, deformation, misalignment, imbalance, overload, leakage current, etc. Each type of fault can be detected in the vibration signal because it produces a specific vibration signature.

[0101] By "vibration signature" is meant a set of harmonics whose frequency distribution is specific to a vibration source, such as the characteristics of a bearing rotated by the shaft to which it is assembled. Furthermore, the harmonic content associated with the bearing, inter alia, the height, amplitude, and statistical properties of the frequencies, are specific to the bearing in question. Furthermore, there may be relationships or non-linear relationships that explain the frequency distribution of the vibration signature. Furthermore, the vibration signature depends on the load applied to the bearing; specifically, the harmonic signature of the bearing may vary with the speed at which it rotates about the axis of the shaft. As a direct result, the vibration signature of the bearing does not appear systematically in the signal when the bearing is rotated, but only at certain engine speeds. The vibration signature of the bearing is specifically included in the spectrum or spectrogram of the vibration signal, although its amplitude is low compared to the vibration signatures of other elements in the machine.

[0102] By "spectrogram" is meant a representation of the frequency content of one or more signals over time. Additionally, the terms frequency content, harmonic content, harmonic components, and harmonics all refer to the set of frequencies (or harmonics) contained in the spectrum of a signal obtained by applying a Fourier transform from the time domain to the frequency domain.

[0103] A first aspect of the invention relates to a method for monitoring a rotating machine to detect bearing faults.

[0104] A rotary machine is, for example, an aircraft engine, comprising one or more axes which enable the energy consumed by the engine to be transformed into a rotational movement. The axis comprises several elements for displacing the aircraft, in particular bearings.

[0105] The example provided here as a preferred embodiment relates to the bearings of the accessory gearbox (AGB) of the CFM International CFM56 series turbine engine. The AGB comprises in particular two rotating shafts N1 and N2, each with a different operating speed. The rotating shaft N2 comprises at least one faulty bearing, the vibration characteristics of which are detected by means of the method 100 according to the invention. The rotating shaft may also comprise other non-faulty bearings.

[0106] Figure 1 A schematic representation of the main steps of a method 100 according to the invention is shown. The method 100 according to the invention comprises four main steps and one optional step.

[0107] The first step is a step 110 of acquiring a vibration signal by means of a vibration sensor, such as a piezoelectric accelerometer. The vibration signal is referred to as x(t) hereinafter. For example, the sensor is placed on the axis of the rotating machine or placed close to the axis. Preferably, the sensor is positioned on a fixed component of the rotating machine to be monitored. Furthermore, the sensor may be at a distance from the bearing to be monitored, such as a distance greater than 5 cm. The signal comprises vibrations generated during operation of the rotating machine.

[0108] The sampling frequency of the signal is high enough so that at least a portion of the vibration signature of the faulty bearing is within the acquired frequency range.

[0109] Preferably, the vibration signal is acquired over a total duration of a few seconds, for example over a duration of more than 10 seconds. Furthermore, the vibration signal is preferably acquired during an operating phase of the rotating machine such that its engine speed is non-steady, for example during a transient speed phase.

[0110] The vibration signal can also be acquired so as to include several different operating phases of the rotating machine. These can be transient phases, the duration of which can be short, for example less than one second, or long, for example more than one second. These transient phases can in particular be acceleration or deceleration phases. Preferably, the vibration signal includes a set of operating phases of the rotating machine that covers most of the operating margins of the rotating machine, i.e. the rotation speeds, accelerations and decelerations that the rotating machine may experience. This diversity of the acquired operating states makes it possible to ensure the diversity of the acquired vibration signatures and to ensure that if the rotating machine includes a faulty bearing, its vibration signature will actually be included in the signal.

[0111] The acquisition step 110 may also include acquiring a speed signal or a position signal. The speed or position signal measures the rotational speed or angular position of a reference shaft rigidly connected to the monitored bearing. For example, a speed or position sensor may be assembled to a shaft including the bearing, or measure the speed or position of the shaft including the bearing. For such a signal, the sensor is, for example, a proximity sensor, an encoder, or a tachometer. In the following, the term "speed signal" will be used to refer to a speed signal or a position signal. In fact, any signal can be used according to the method 100 of the present invention, and its properties will not change the implementation of the method 100. Preferably, the duration of the speed signal is the same as the duration of the vibration signal. Alternatively, the duration of the speed signal may be longer than the duration of the vibration signal. In the preferred embodiment, the speed signal is acquired simultaneously with the vibration signal by a tachometer on the rotating shaft N2, and the duration is the same. The speed signal is recorded as N2(t).

[0112] Acquisition may be performed on a test bench, where the rotating machine is isolated, or in situ when the machine is assembled on an aircraft.

[0113] The method 100 according to the invention then comprises a step 120 of determining a 1st order spectrogram and a step 130 of determining a 2nd order spectrogram. Since these two steps are independent of each other, ie one step does not require results from the other step to be implemented, they can be performed simultaneously.

[0114] The step 120 of determining the 1st order spectrogram and the step 130 of determining the 2nd order spectrogram are preferably 1st order and 2nd order cyclostationary analysis steps of the vibration signal, respectively, so as to obtain 1st order and 2nd order spectrograms, respectively.

[0115] Preferably, the acquired vibration signal is analyzed over a plurality of consecutive reduced time windows. This is also referred to as analysis by sliding the time window or by convolution of the vibration signal with a window function. In order to obtain the 1st and 2nd order spectrograms, the aim is to determine the 1st and 2nd order spectra from the vibration signal for each time window, respectively. The time interval of the time windows (also referred to as window size) is between 0.1 seconds and 10 seconds; preferably, the interval is between 0.5 seconds and 2.5 seconds. In the following, x i (t) and Refers to the vibration signal and speed signal contained in the i-th time window respectively. Overlapping can be applied to consecutive time windows. In this case, the time interval between each time window is less than the size of the window. For example, the time interval between two time windows can be 0.25 seconds less than the size of the time window.

[0116] In the illustrated embodiment, the size of the time windows is 1 second and there is no overlap. The time interval between two time windows is 1 second.

[0117] Figure 2 The vibration signal ( Figure 2 a) and speed signal ( Figure 2 b) The division of the signal is illustrated here to represent the analysis by sliding time windows. The vibration signal is then decomposed into a set of vibration signals labeled x0 to x7. The speed signal is decomposed into a set of speed signals of equal duration, labeled to In this time representation, the richness and complexity of the vibration signal content is observed, and therefore there is a need to analyze this signal in the spectral domain to facilitate its processing and analysis.

[0118] The step 120 of determining the 1st order spectrogram comprises five consecutive main sub-steps numbered 121 to 125 and two optional sub-steps numbered 126 and 127. These steps 121 to 127 are repeated for each time segment obtained by convolving the vibration signal with the window function. The step 120 of determining the 1st order spectrogram also comprises a final step 128.

[0119] Sub-step 121 is to analyze the vibration signal x i (t) A step of performing angle resampling. Preferably, for example, by interpolation, based on the velocity signal of axis N2 Resampling is performed. Resampling may also depend on the type of speed sensor used and its characteristics. Resampling is used to obtain the angle signal.

[0120] The resampling can be obtained using the angle of axis N2, which itself is given by Then mark the angle signal as

[0121] In sub-step 122, the angle signal The Fourier transform (FT) is used to determine the first-order spectrum, which is labeled X i (α), where α is the frequency.

[0122] Here, the 1st order spectrum can be obtained by Obtain, among which Specifies the TF operator. The spectrum can be conventionally obtained using a Fast Fourier Transform (FFT).

[0123] exist Figure 3 An illustration of the spectrum of the vibration signal is provided in . The spectrum shows all the frequencies contained in the vibration signal. Frequencies that may be related to the rotation of the AGB shaft, i.e. fundamental frequencies and harmonic frequencies with an order greater than 1, are indicated by the solid circles at the top of the peaks. Not all potential frequencies are necessarily related to the theoretical dynamics of the system, therefore, simply removing all of these components may result in a distortion of the spectrum by removing components related to the faulty bearing.

[0124] Sub-step 123 is here a step of applying a delta transformation (TΔ) to obtain a corrected 1st order spectrum. Preferably, based on the kinematic knowledge related to the axis N2 (e.g. based on the axis frequency set ) to apply TΔ.

[0125] TΔ is applied here to remove frequency content and vibration signatures generated by mechanical sources that are not related to the bearings of the rotating shaft N2, in particular those emitted by shafts and gears of the power transmission close to the bearing(s) of the monitored shaft. For example, these frequencies may be known in advance by numerical dimensional analysis with dedicated software (such as Abaqus, COMSOL or Catia). In this case, sub-step 123 comprises obtaining the set of shaft frequencies Sub-sub-step 123a of .

[0126] In this embodiment, 11 frequencies are known and associated with the rotation of the 11 axes of the CFM56 AGB. These 11 frequencies form the axis frequency set referenced to N2 Tables 1 and 2 summarize the dynamics of the AGB, where the axis frequency set There are 11 frequencies in the table. It is indicated that these frequencies are obtained when the rotation speed of shaft N2 is 10000 rpm. In this table, the high-pressure shaft is shaft N2. The designation F1 tooth (Hz) corresponds to the shaft rotation speed in Hertz, and the designation F1 tooth (N2) corresponds to the shaft rotation speed related to the high-pressure shaft rotation speed.

[0127] [Table 1]

[0128]

[0129] [Table 2]

[0130]

[0131]

[0132] Possible imbalances and / or unbalances of the different shafts (specifically the gears and impellers assembled thereon) may generate additional unwanted harmonic content in the 1st order spectrum. Then it is necessary to consider the harmonic content in the 1st order spectrum in the frequency band [0,α max ], where α max A predefined maximum order is specified. Preferably, the predefined maximum order is higher than the highest harmonic associated with the faulty bearing, which can be theoretically determined based on the bearing geometry. Alternatively, the predefined maximum order can be at least greater than the pth harmonic of the faulty bearing, where p is an integer greater than or equal to 2, and preferably greater than or equal to 5, when the Nyquist criterion allows. In this case, sub-step 123 includes obtaining the total set Sub-sub-step 123b, which then obtains the axis frequency set Sub-sub-step 123a of . Total set can be defined such that

[0133] Furthermore, the set of potential harmonics can be restricted to a subset that includes statistically significant harmonics. If at frequency α, with the total set The associated Z spectrum (i.e. obtained by Z transform) exceeds a given statistical threshold ζ, then the total set associated with frequency α The harmonics Y(α) of are considered significant. In this case, sub-step 123 includes obtaining a set of significant harmonics Sub-substep 123c, followed by obtaining the total set Sub-sub-step 123b of: Can be To define, where Z Y (α) is the same as the total set The statistical threshold ζ is, for example, greater than or equal to 2. Preferably, the statistical threshold is greater than or equal to 4. In fact, the larger the threshold, the more it can ensure that only the harmonics of the bearing are considered and less noise will be considered. For example, the statistical threshold ζ can be equal to 6 to meet the six-sigma principle.

[0134] By "Z-transform" is meant an operation for normalizing a spectrum, where the spectrum is denoted by Y, and the total set The normalization operation consists in centering the spectrum Y by subtracting the trend μ from the spectrum Y and then replacing the centered spectrum Y by c Divide by the variance σ of spectrum Y to make the center spectrum Y c Standardization makes in A spectrum normalization operator, also called TZ operator, is specified. The trend is, for example, the mean or value of the spectrum Y. Preferably, the trend is a moving median estimated by a median filtering operation of the spectrum Y.

[0135] Furthermore, in order to ensure a minimum deformation of the 1st order spectrum, sub-step 123 may include a sub-sub-step 123d of estimating correction values. Thus, the correction values ​​are used to correct the significant harmonic set The correction value can be determined based on the statistics of the harmonics around the harmonic of interest. For example, the correction value can be based on the trend μ of the adjacent samples for each significant harmonic in the 1st order spectrum. X(α) is determined by the median or mean. For example, some samples on the left side of the significant harmonic sample and some samples on the right side can be regarded as neighbors of the significant harmonic. The number of samples on the right side and the number of samples on the left side can be equal. These numbers of samples are, for example, greater than or equal to 5. Preferably, 25 samples on the left side and 25 samples on the right side are used to determine the correction value. Then, the correction value will be different and adapted for each significant harmonic in the 1st order spectrum.

[0136] Therefore, we can define the significant harmonic subsets based on and the correction value of each significant harmonic is determined to be marked as And the corrected 1st order spectrum obtained by TΔ is such that:

[0137]

[0138] in Specify X i (α) relative to the significant harmonic subset Therefore, the delta transform is a method to remove the significant harmonics from the 1st order spectrum. Advantageously, correcting the harmonics of the 1st order spectrum corresponding to the significant harmonics makes it possible to preserve the continuity and regularity of the 1st order spectrum while removing frequency components associated with the elements and shafts of the rotating machine other than the bearings of the shaft N2.

[0139] Figure 4 The 1st order spectrum of the vibration signal obtained by TΔ is illustrated in Figure 1. The 1st order spectrum is plotted with a dashed line and the spectrum obtained by a simple TF is plotted with a solid line. The crosses mark the harmonics of the vibration signature associated with a fault in the outer ring of the bearing of shaft N2. The circles in turn indicate the harmonics associated with the known vibration signature of the AGB shaft without bearing faults. It can be noted that the harmonics related to the kinematics of the accessory gearbox shaft are removed without any visible deformation of the spectrum, while the harmonics related to bearing faults (which are not integer multiples of these axes) are retained.

[0140] The mechanical spectrum can be considered as a random statistical sequence with a frequency-dependent mean and variance. Based on this definition, the harmonics associated with the vibration signature (here, the bearing) can be considered as outliers. These outliers are also unidirectional, i.e., they follow the positive direction on the frequency axis. Several factors may be responsible for the frequency variability of these statistics, but there are two main ones:

[0141] - the dynamics of a mechanical system (here a rotating machine) leads to time correlations which take the form of resonant modes and deformations in the frequency domain;

[0142] - The presence of random noise that is strongly correlated to the operation of the mechanical system, which is exacerbated in certain frequency bands, generating a higher noise variance.

[0143] In all cases, this statistical frequency variability must be compensated for in order to obtain a standard or common spectrum.

[0144] To this end, sub-step 124 is a spectral normalization step of the corrected 1st order spectrum based on robust statistics to obtain a normalized 1st order spectrum. This compensates for spectral trends (also known as spectral mean) and noise variances while being robust to harmonics associated with outliers. Specifically, these outliers form the vibration signature of the bearing(s) of the rotating machine whose health status is diagnosed.

[0145] The interest of this normalization is therefore to flatten the 1st order spectrum in order to compensate for the influence of the average transfer function of the rotating machine. More precisely, this normalization makes it possible to reduce the spectral energy associated with the resonant modes of the rotating machine, colored noise and other undesirable noise sources in order to analyze the vibration behavior of the N2 shaft bearing.

[0146] It can be noted that in the case of the vibration signal spectrum, the outliers appear strictly on the positive side of the statistical distribution, i.e. on the right side of the frequency axis, which makes it possible to simplify the modeling of the statistical properties of the signal spectrum (in this case the 1st order spectrum) in order to normalize it.

[0147] Therefore, the spectrum normalization sub-step 124 may include a sub-sub-step 124a for estimating the trend of the first-order spectrum Preferably, the trend is estimated based on the determination of a moving median. For example, such a determination of a moving median may be a median filtering operation of the 1st order spectrum, which comprises calculating the median of the 1st order spectrum within a time window, such that Where Medfilt(*) is the median filter operator.

[0148] Therefore, the spectrum normalization sub-step 124 may include a sub-sub-step 124b for normalizing the spectrum according to the trend μ of the first-order spectrum. ΔX (α) is estimated to determine the centered first-order spectrum X Δc (α). For example, the centered order 1 spectrum is defined such that

[0149] Furthermore, the spectrum normalization sub-step 124 may also include a sub-sub-step 124c of determining the dispersion of the frequency sequence. For example, the dispersion is calculated based on the mean median deviation of the frequency sequence of the 1st order spectrum, and its interest is to provide a robust estimate of the dispersion of the sequence with respect to outliers. Thus, the mean median deviation (denoted as EMM ΔX (α)) can be defined as:

[0150]

[0151] Advantageously, according to this equation, the mean-median deviation is related to the variance of the frequency series (denoted by σ ΔX (α)) is linearly related, so that Where k is a predetermined real number that depends on the distribution of the frequency sequence.

[0152] In practice, this distribution is unknown and needs to be estimated empirically. Therefore, the spectrum normalization sub-step 124 may include a sub-sub-step 124d for empirically estimating the standard deviation associated with the variance of the frequency series of the 1st order spectrum. This empirical estimation can be achieved using the fact that outliers appear strictly on the positive side of the frequency axis. Therefore, a robust standard deviation can be estimated from the right side of the empirical probability distribution function. In other words, this is equivalent to defining the coefficient k to be equal to the mean. Therefore, the sub-sub-step 124d of empirically estimating the standard deviation can be a step of determining the variance or standard deviation of the frequency series, which depends linearly on the mean median deviation, with the coefficient k i , so that in Specify the indicator function, when 1 when , 0 otherwise.

[0153] Finally, the spectrum normalization sub-step 124 comprises a sub-sub-step 124e for determining a normalized 1st order spectrum, specifically called the Z spectrum and denoted as The standardized first-order spectrum can be further determined based on the trend and variance of the first-order spectrum, so that in Specifies the spectrum normalization operator, also known as the TZ operator. Thus, for any α, the interest in using TZ is to ensure that the normalized 1st-order spectrum is defined by zero mean and a robust unit standard deviation.

[0154] exist Figure 5 An illustration of the effect of spectral normalization is provided in . It can be seen that the signal dynamics after normalization are flattened, with zero mean and unit standard deviation. It can also be noted that peaks emerge in the spectrum that were barely visible before normalization.

[0155] Sub-step 125 is to apply spectral autocorrelation to obtain a flattened first-order spectrum (denoted as ) step. The application of this autocorrelation makes it possible to improve the detection of the characteristics of the bearings of the shaft N2 and to reduce the energy of the harmonics generated by the kinematics of the rotating machine.

[0156] The spectral features generated by the bearing fault are present in the 1st order spectrum in the form of periodic patterns relative to axis N2. Specifically, these patterns are peaks located at one or more frequencies of the fault and on the sidebands of the modulation frequency. The modulation frequency can be the rotational frequency of the shaft of the bearing in question or the rotational frequency of its train (the Fundamental Train Frequency or FTF). Therefore, these vibration features exhibit regularities related to periodicity, which can be exploited by spectral autocorrelation to improve the presence of the features of the (multiple) bearings of axis N2, while reducing the harmonics generated by other axes that do not exhibit such periodic patterns relative to axis N2.

[0157] For example, according to the relationship Apply spectral autocorrelation, where Specifies the spectral autocorrelation operator.

[0158] In other words, applying spectral autocorrelation can correct for spectral leakage around the peak of the bearing signature caused by the sliding nature of the bearing elements.

[0159] exist Figure 5 An illustration of the effect of the spectral autocorrelation operation is provided in . It can be observed that the emergence of the vibration signature of the fault in the outer ring of the shaft N2 bearing clearly emerges from the rest of the spectrum, which has lost some of its dynamic range. As a result, the readability of the spectrum is significantly improved.

[0160] At the end of the sub-step 125 of applying spectral autocorrelation, the features of the bearings on shaft N2, especially the features of the faulty bearings, are significantly amplified compared to the unwanted features. These bearing features can then be used to diagnose the health of shaft N2 and, if applicable, the bearing fault.

[0161] However, applying spectral autocorrelation may amplify some harmonic series related to the dynamics of the shaft due to modulation or interference between the vibration features of the shaft; this is the case, for example, when there is non-negligible spatial coherence between the vibration features. Moreover, the spectrum obtained by spectral autocorrelation preserves artificial statistical trends related to the bias induced by the spectral autocorrelation operator.

[0162] In order to correct this undesired bias and amplification, the method 100 according to the invention can optionally be used to implement sub-step 126, which is a step of applying a second TΔ to flatten the 1st order spectrum. The interest of this step is to obtain a corrected flattened 1st order spectrum, which is indicated as Preferably, TΔ is applied based on the kinematic knowledge related to the axis N2 that has been determined in the sub-step 123 of applying TΔ. Specifically, the significant harmonic subset previously determined in the sub-step 123 of applying TΔ may be and the correction value of each significant harmonic to implement the second TΔ, so that Note that Δ{*} specifies the same TΔ operator as in sub-step 123 .

[0163] The interest of this step 126 of applying a second TΔ is to remove all or some of the unwanted harmonic series related to the kinematics of the axis, which may have appeared by applying the spectral autocorrelation transform in the previous step.

[0164] The method 100 according to the present invention may further implement a sub-step 127, which is to apply a second spectrum normalization to the corrected flattened 1st-order spectrum in order to determine a reduced 1st-order spectrum (denoted as ) step. This sub-step is particularly interesting for the flattened 1st order spectrum after correction The flattening is performed in order to compensate for the bias introduced by applying the spectral autocorrelation. The second application of spectral normalization also makes it possible to make the 1st order spectrum statistically general, i.e. its statistical properties are comparable to those of other normalized spectra, e.g. spectra obtained for the same engine under other measurement conditions (such as a different measurement bench).

[0165] Preferably, spectral normalization is performed by applying a second TZ, as described in a sub-sub-step of spectral normalization sub-step 124 .

[0166] In this case, sub-step 127 consists in estimating the trend μ of the corrected flattened 1st-order spectrum R1 Sub-sub-step 127a of (α) is preferably estimated by determining a moving median. Determining the moving median may be a median filtering operation of the corrected flattened 1st order spectrum, which comprises calculating the median of the corrected flattened 1st order spectrum within a time window, such that Where Medfilt(*) is the median filter operator.

[0167] Then, sub-step 127 includes sub-sub-step 127b for estimating the trend of the corrected flattened 1st-order spectrum. To determine the centered corrected flattened first-order spectrum μ R1 (α). For example, the centered corrected flattened 1st order spectrum is defined such that

[0168] Furthermore, the spectrum normalization sub-step 127 may also include a sub-sub-step 127c for determining the dispersion of the frequency sequence. For example, the dispersion is calculated based on the mean median deviation of the frequency sequence of the corrected flattened 1st order spectrum, which is interesting in that it is a robust estimate against outliers. Therefore, the mean median deviation (denoted as ) can be defined as:

[0169]

[0170] Advantageously, according to this equation, the mean-median deviation is related to the variance of the frequency series (denoted by σ ΔR (α)) is linearly related, so that where l is a predetermined real number that depends on the distribution of the frequency sequence.

[0171] As with the 1st order spectrum in sub-step 124, the distribution associated with the corrected flattened 1st order spectrum is unknown and has to be estimated. Therefore, the spectrum normalization sub-step 127 may include a sub-sub-step 127d of empirically estimating the standard deviation associated with the variance of the frequency series of the corrected flattened 1st order spectrum. This empirical estimation may be achieved, as in sub-step 124, using the fact that outliers appear strictly on the positive side of the frequency axis. Therefore, a robust standard deviation may be estimated from the right side of the empirical probability distribution function. In other words, this is equivalent to defining the coefficient l to be equal to the mean. Therefore, the sub-sub-step 127d of empirically estimating the standard deviation may be a step of determining the variance or standard deviation of the frequency series, which depends linearly on the mean median deviation, with a coefficient of l. i , so that in Specify the indicator function, when 1 when , 0 otherwise.

[0172] The spectrum normalization sub-step 127 includes determining a reduced first-order spectrum (denoted as ). In addition, the reduced 1st-order spectrum can also be determined based on the trend and variance of the corrected flattened 1st-order spectrum, such that: in Specifies the spectral normalization operator, also known as the TZ operator. For any α, the interest in using TZ is also to ensure that the reduced 1st-order spectrum is defined by zero mean and robust unit standard deviation.

[0173] exist Figure 7 A graphical representation of the spectrum obtained at different sub-steps of step 120 of determining the 1st order spectrum is provided in FIG. Figure 7 .a is the first-order spectrum obtained by TΔ; Figure 7 .b is the flattened first-order spectrum after spectral autocorrelation; Figure 7 .c is the reduced 1st order spectrum obtained at the end of sub-step 127. Figure 7 In the spectrum of , the black circles indicate the harmonic positions of the vibration signature of the faulty bearing. These different representations show the importance of the method according to the invention for improving the visualization of the vibration signature of the faulty bearing in the 1st order spectrum.

[0174] Finally, sub-step 128 is the step of concatenating the reduced 1st order spectra to form a 1st order spectrogram. The concatenation is achieved by juxtaposing the reduced 1st order spectra in time order in the vibration signal. The 1st order spectrogram is denoted as Z1(i,α), such that i=0,1,…T-1, where T is the number of time windows.

[0175] The 1st order spectrogram provides complete information about the cyclostationary statistical content of the 1st order bearing at all states scanned during the acquisition.

[0176] Advantageously, considering several operating states of a rotating machine can significantly improve the visibility and detectability of faults.

[0177] The step 130 of determining the 2nd order spectrogram comprises five consecutive main sub-steps numbered 131 to 135 and two optional sub-steps numbered 136 and 137. These steps 131 to 137 are repeated for each time period obtained by convolving the vibration signal with the window function. The step 130 of determining the 2nd order spectrogram also comprises a final step 138.

[0178] Sub-step 131 is to analyze the vibration signal x i (t) A step of resampling the angle. Preferably, based on the velocity signal of axis N2 Resampling is performed. Resampling is used to obtain the angle signal.

[0179] The resampling can be achieved using the angle of axis N2, which itself is given by Then, the angle signal is marked as

[0180] Sub-step 132 is to delete the angle signal ). The interest of this step is to remove the angular periodicity of the signal, regardless of the origin of this deterministic part. In other words, this step makes it possible to estimate the random part of the signal, specifically called the residual signal and denoted r, by removing its deterministic part. i (t).

[0181] The means for obtaining the residual signal may be tools known in the literature, such as a design based on a Wiener filter.

[0182] Preferably, the tool used to estimate the residual signal is the so-called "frequency-domain adaptive noise cancellation" or FD-SANC method (Antoni, RBRandall, Randall, Unsupervised noise cancellation for vibratorysignals: part II-a novel frequency-domain algorithm, Mechanical Systems and Signal Processing, Vol. 18, No. 1, 2004, pp. 103-11). The interest of the FD-SANC method is that it is efficient, fast and completely independent of the source of the deterministic part of the signal. The FD-SANC method is based on the design of an optimal Wiener filter, denoted h(t), which maximizes the signal-to-noise ratio of the harmonics in a least squares sense. The residual signal r is then i (t) is written as signal x i A simple convolution between (t) and the Wiener filter h(t) makes r i (t) = ∫h(s)x i (ts)d.

[0183] Figure 8 The residual signal obtained by FD-SANC (dashed line) is compared with the angle signal (solid line). It can be observed that the FD-SANC algorithm estimates and removes most of the harmonics related to the deterministic signal present in the spectrum.

[0184] Sub-step 133 is the step of determining the cyclic coherence of the residual signal. Cyclic coherence is a powerful tool for detecting cyclostationary components in a signal, even in the case of low signal-to-noise ratios.

[0185] The cyclic coherence is determined based on an estimate of the cyclic correlation of the signal (here the residual signal). Therefore, the sub-step 133 of determining the cyclic coherence may include a sub-sub-step 133a of determining the cyclic correlation of the residual signal. Preferably, the cyclic correlation of the residual signal (denoted as ) is estimated from the double TF of the autocorrelation function on the residual signal, so that where E{*} is the mathematical expectation operator, and τ is the time offset. As a reminder, is the TF operator, and α and f are two frequencies.

[0186] Then, the sub-step 133 of determining the cyclic coherence may include a sub-step 133b of estimating the cyclic coherence of the residual signal based on the cyclic correlation of the residual signal. Preferably, the cyclic coherence (denoted as ) is obtained by spectrum normalization, e.g.

[0187] Fig. 9 The cyclic coherence determined from the residual signal of the CFM56 AGB is shown. The bispectral distribution is a function of the N2 order (horizontal axis) and the absolute spectral frequency in Hertz (vertical axis). Note that there are a large number of vertical spectral lines in the signal, which are pure cyclostationary 2nd order components.

[0188] Sub-step 134 is a step of averaging the cyclic coherence. Of interest, the averaged cyclic coherence is a spectral representation that summarizes the 2nd order cyclostationary cyclic content of the residual signal. Preferably, the averaged cyclic coherence (denoted as ) is calculated from the integral of the squared magnitude of the cyclic coherence. For example, the cyclic coherence can be averaged so that where F s is the sampling frequency of the vibration signal.

[0189] Fig.10 The cyclic coherence of the average of the residual signal is illustrated, where the distribution of the integral of the square of the amplitude, i.e. the distribution of a random signal, is represented relative to the spectral frequency axis. Indications H0 to H5 identify the harmonics present in the signature of the faulty bearing, in this case the outer ring of shaft N2, making it possible to detect the same type of fault on another shaft.

[0190] Sub-step 135 is a step of applying spectral autocorrelation to obtain a flattened 2nd order spectrum. As with sub-step 125, the interest in applying spectral autocorrelation is to improve the detection of bearing features of shaft N2 and to reduce the energy of harmonics generated by the kinematics of the rotating machine.

[0191] Preferably, a flattened 2nd order spectrum (denoted as ) is implemented in the same manner as sub-step 125. Here, it can be based on the relationship to apply spectral autocorrelation, where Again, specify the spectral autocorrelation operator.

[0192] It should be noted that, unlike the step 120 of determining the 1st order spectrum, the step 130 of determining the 2nd order spectrum does not necessarily include a normalization operation. In fact, the 2nd order spectrum is already flat due to its cyclic coherence properties. This is because the cyclic coherence implicitly includes a spectral whitening operation. Therefore, the step 130 of determining the 2nd order spectrum may include an intermediate step, before or after the sub-step 135 of applying the spectral autocorrelation, but this will not improve the visualization of the (one or more) characteristics of the (one or more) bearings.

[0193] Sub-step 136 is an optional step of applying TΔ from the flattened 2nd order spectrum to obtain a corrected flattened 2nd order spectrum. For the same reasons discussed with sub-step 126, the interest of this step is to correct the undesired bias and amplification introduced by the application of the autocorrelation operator. In fact, it is necessary to remind that the application of spectral autocorrelation may amplify some harmonic series related to the dynamics of the shaft due to modulation or interference between the vibration characteristics of the shaft; and the spectrum obtained by spectral autocorrelation retains artificial statistical trends related to the bias caused by the autocorrelation operator.

[0194] Preferably, the sub-step 136 of applying TΔ is implemented in a similar manner as the sub-step 126. Specifically, the application of TΔ may be based on the subset of significant harmonics determined in the TΔ application sub-step 123. and the correction value of each significant harmonic. The corrected flattened second-order spectrum can be obtained and is marked as Make Note that Δ{*} specifies the same TΔ operator as in substeps 123 and 126 .

[0195] Sub-step 137 is an optional step of applying spectral normalization to the corrected flattened 2nd order spectrum to obtain a reduced 2nd order spectrum. As with sub-step 127, the interest of this step is to flatten the 2nd order spectrum to compensate for the bias introduced by applying the spectral autocorrelation. Such spectral normalization can also make it possible to make the 2nd order spectrum statistically general, i.e. its statistical properties are comparable to those of other normalized spectra, for example, spectra obtained for the same engine under other measurement conditions (such as a different measurement bench).

[0196] Preferably, the spectral normalization is performed by applying TZ, as described in step 120 of determining the 1st order spectrum.

[0197] In this case, sub-step 137 consists in estimating the trend μ of the corrected flattened 2nd order spectrum R2 Sub-sub-step 137a of (α) is preferably estimated by determining a moving median. Determining the moving median may be a median filtering operation of the corrected flattened 2nd order spectrum, which comprises calculating the median of the corrected flattened 2nd order spectrum within a time window, such that Where Medfilt(*) is the median filter operator.

[0198] Then, sub-step 137 includes sub-sub-step 137b for estimating the trend value of the corrected flattened second-order spectrum. To determine the centered, corrected, flattened second-order spectrum For example, the centered corrected flattened 2nd order spectrum is defined such that

[0199] In addition, the spectrum normalization sub-step 137 may also include a sub-sub-step 137c for determining the dispersion of the frequency sequence. For example, the dispersion is calculated based on the mean median deviation of the frequency sequence of the corrected flattened 2nd order spectrum, which is interesting in that it is a robust estimate against outliers. Therefore, the mean median deviation (denoted as Can be defined as:

[0200]

[0201] Advantageously, according to this equation, the mean-median deviation is related to the variance of the frequency series (denoted as ) is linearly related, so that where l is a predetermined real number that depends on the distribution of the frequency sequence.

[0202] As with the 1st order spectrum in sub-steps 124 and 127, the distribution associated with the corrected flattened 2nd order spectrum is unknown and has to be estimated. Therefore, the spectrum normalization sub-step 137 may include a sub-sub-step 137d of empirically estimating the standard deviation associated with the variance of the frequency series of the corrected flattened 2nd order spectrum. This empirical estimation can be achieved, as in sub-step 124, by exploiting the fact that outliers appear strictly on the positive side of the frequency axis. Therefore, a robust standard deviation can be estimated from the right side of the empirical probability distribution function. In other words, this is equivalent to defining the coefficient g to be equal to the mean. Therefore, the sub-sub-step 137d of empirically estimating the standard deviation can be a step of determining the variance or standard deviation of the frequency series, which depends linearly on the mean median deviation, with the coefficient g i , so that in Specify the indicator function, when 1 when , 0 otherwise.

[0203] The spectrum normalization sub-step 137 includes determining a reduced 2nd order spectrum (denoted as ). In addition, the reduced second-order spectrum can also be determined based on the trend and variance of the corrected flattened second-order spectrum, such that: in Specifies the spectral normalization operator, also known as the TZ operator. For any α, the interest in using TZ is also to ensure that the reduced 1st-order spectrum is defined by zero mean and robust unit standard deviation.

[0204] Finally, sub-step 138 is the step of concatenating the reduced 2nd order spectra to form a 2nd order spectrogram. The concatenation is performed by juxtaposing the reduced 2nd order spectra in time order in the vibration signal. The 1st order spectrogram is denoted as Z2(i,α), where i=0,1,…T-1, where T is the number of time windows.

[0205] The 2nd order spectrogram provides complete information about the cyclostationary statistical content of the 2nd order bearing at all states scanned during the acquisition.

[0206] Advantageously, the fact that several operating states of the rotating machine are taken into account significantly improves the manifestation and detectability of faults.

[0207] The fourth step of the method 100 according to the invention is a step 140 of detecting the vibration signature of the bearing fault. The detection is performed in particular based on the 1st order spectrogram and the 2nd order spectrogram.

[0208] Fig.11 and Fig.12 The 1st and 2nd order spectrograms determined from the AGB vibration signal are shown, in this case acquired over 115 seconds. The vibration signature of the faulty bearing is clearly visible in these spectrograms and can be clearly identified by the white vertical lines. The faulty bearing can then be identified based on the detected signature, allowing a reliable assessment of the bearing's health.

[0209] Detecting and identifying faulty bearings enables early diagnosis of the occurrence of bearing faults in order to immediately trigger maintenance and prevent the risk of failure of the rotating machine. Thus, the method provided makes it possible to extend the service life of the machine.

[0210] For illustration, consider the general case of a bearing fault signature, which includes a fault frequency α f and its frequency α m Multiple harmonics of modulation:

[0211] - In the case of a bearing element fault signature, for example, the frequency α f is the characteristic frequency of the bearing element (BSF-Ball Spin Frequency), and α m is the train frequency (FTF).

[0212] -For outer race faults, the characteristics are usually realized by harmonics without sidebands, and α f is BPFO (outer ring ball passing frequency).

[0213] -For inner race fault, frequency α f is the characteristic frequency of the inner race fault, and α m is the shaft rotation frequency.

[0214] D(α f ,α m ; N f ,N m )={N f α f +n m α m / n f =1…N f ,n m = ±1,…±N m} indicates

[0215] The frequency set that defines the bearing fault characteristics. This set is the fault frequency α f and the potential modulation frequency α m Therefore, this feature is determined by the harmonic number N set in advance by the operator. f and sideband pair N m Parameterized.

[0216] A metric associated with a given type of fault (associated with the set D) can be defined, for example as the mean of the harmonics:

[0217]

[0218] The value of the indicator is then compared with a threshold, for example, threshold = 6 (i.e., the occurrence rate of the feature in the spectrum is 6 times the standard deviation). If I(Z,D)>threshold, there is a fault, otherwise there is no fault.

[0219] The detection of faulty bearings can then be performed by an automatic detection algorithm implementing a previous threshold comparison. Alternatively, abnormal spectral signature recognition software based on reference signatures can be used.

[0220] The detection can also be performed using a machine learning algorithm; this can be a supervised or unsupervised learning algorithm; moreover, the algorithm can be based on a database of predetermined thresholds or reference features of faulty bearings and / or non-faulty bearings. The database can be constructed based on theoretical or empirical knowledge of the dynamics of faulty bearings or non-faulty bearings. The detection can also include issuing an alarm, which contains identification information to identify the faulty bearing, for example in the form of an identification number associated with the bearing. For example, an alarm is issued to an operator or an expert. This alternative embodiment makes it easier for an expert or operator to diagnose the health of the faulty bearing and its impact on the health of the rotating machine. It also makes it possible to analyze a large number of vibration signals and / or features simultaneously, in particular improving the early detection of faults. Finally, this alternative makes it possible to perform detection based on quantitative criteria, for example in relation to a reference system or a learning tool. The interest here is in using only criteria that are objective and not affected by operator analysis bias.

[0221] Alternatively, an operator or expert can perform the detection based on their knowledge of vibration analysis and specific requirements, for example in terms of the safety of rotating machines. Advantageously, the 1st and 2nd order spectrograms make it easier to read the vibration signature of the bearing. In fact, the 1st and 2nd order spectrograms here show the harmonic distribution in the form of a vibration signature, the degree of importance of which is assessed by the number of standard deviations in relation to the background noise. Thus, an operator or expert can easily analyze and interpret these spectrograms by observing them in order to detect faulty bearings and their nature.

[0222] Since each vibration signature of a faulty bearing is different depending on the nature of one or more faults in the bearing, the method 100 advantageously enables the nature of the fault in the bearing to be detected. The detection of the bearing fault then enables the health of the bearing to be assessed. In this case, the identification information may also include information related to the nature of the bearing fault, such as summarizing the nature of the fault in the form of keywords.

[0223] The detection step 140 may also include the step of averaging the 1st and 2nd order spectrograms. For example, the averaged 1st order spectrogram (denoted as ), so that i = 0, 1, ... T-1, and determine the average second-order spectrum (denoted by Make Alternatively, the detection of a faulty bearing may be performed based on averaged 1st and 2nd order spectra.

[0224] Fig.13 and Fig.14 The averaged 1st and 2nd order spectra for detecting faulty bearings of a CFM56 AGB are shown. The 1st order spectra ( Fig.13 ) on indicators H0 to H3 and 2nd order spectrum ( Fig.14 ) identifies the specific harmonics of the faulty bearing on shaft N2. This representation also allows the characteristics of the faulty bearing to appear in the 1st and 2nd order spectrum diagrams and facilitates the detection and identification of bearing faults.

[0225] The method 100 according to the invention optionally comprises a step 150 of a maintenance operation. The maintenance operation may be implemented upon detection of one or more faulty bearings, the characteristics of which enabled detection in the previous step. For example, the maintenance operation is an operation of repairing or replacing said bearing.

[0226] Maintenance operations can be implemented, in particular, depending on the nature of the detected bearing fault and the assessed bearing health. More specifically, the criticality of the bearing fault can be assessed in order to assess the necessity of performing maintenance operations. This criticality can be assessed based on 1st and 2nd order spectral graphs and specific requirements in terms of safety of the rotating machine.

[0227] Alternatively, in the step 120 of determining the spectrum of the angle signal, the Welch procedure may be used to determine the 1st order spectrum. Thus, the interest is in reducing the variance of the random noise in the 1st order spectrum of the angle signal. In this case, determining the 1st order spectrum is equivalent to determining the power spectral density of the angle signal.

[0228] In case the spectrum normalization sub-step 124 has not been implemented, the second spectrum normalization sub-step 127 may include sub-sub-steps of sub-step 124 in order to implement TZ and determine a reduced 1st order spectrum.

[0229] Alternatively, the resampling steps 121 and 131 may be implemented in the step 115 for resampling the vibration signal, instead of being performed in the steps 120 and 130 of determining the 1st and 2nd order spectrograms, respectively.

[0230] Advantageously, since the method 100 allows for the analysis of signals within a short time window, the method 100 can be implemented to perform real-time monitoring of the bearings of a rotating machine. This enables an operator or expert to trigger maintenance operations instantly, i.e., as soon as the signature of a faulty bearing is detected.

[0231] Alternatively, the vibration signal may be an acoustic signal measured by an acoustic sensor such as a microphone or a piezoelectric sensor.The different steps of the method 100 may be applied to vibration signals and acoustic signals.

[0232] In an alternative embodiment, the 1st order spectrogram is composed of the first vibration signal and the 2nd order spectrogram is composed of the second vibration signal. The first and second vibration signals are preferably acquired at different locations on the rotating machine (e.g., on two different axes) using a vibration sensor (such as an accelerometer). The vibration signals can also be acquired on the same axis. Advantageously, this embodiment makes it possible to acquire more information in the spectrogram to improve the robustness of the detection of the vibration signature of a faulty bearing.

[0233] Thus, the first vibration signal is used in step 120 of determining the 1st order spectrogram, and the second vibration signal is used in step 130 of determining the 2nd order spectrogram. Preferably, the first and second vibration signals are acquired for the same duration and their acquisition is synchronized so as to be performed simultaneously.

[0234] The method 100 according to the invention may be implemented by a system 10 for monitoring a rotating machine to detect faults in a bearing, such as Fig.15 As shown in the system. The system 10 includes an acquisition module 20, a processing module 30 and an optional detection module 40.

[0235] The acquisition module 20 is used to implement the acquisition step 110 and includes a vibration sensor 21a, a memory 22 and a processor 23. This is, for example, an acquisition device suitable for acquiring a vibration signal. Preferably, the memory 22 includes instructions that, when executed, enable the processor 23 to perform the acquisition step 110 of the method 100. Alternatively, the vibration sensor 21a is an acoustic sensor 21b. Optionally, the acquisition module includes an angular velocity sensor 21c.

[0236] The processing module 30 is used to implement the steps of processing the acquired vibration signal, which are steps 120 and 130 of determining the first-order and second-order spectrograms, respectively. The processing module 30 includes a processor 31 and a memory 32. The memory 32 includes instructions, which, when executed, enable the processor 31 to perform the steps 120 of determining the first-order spectrogram and 130 of determining the second-order spectrogram of the method 100.

[0237] The system 10 may further include a communication device 50 to enable data communication between the acquisition module 20 and the processing module 30. For example, it is a wired connection of the USB or Ethernet type, or a wireless connection of the WiFi type.

[0238] The detection module 40 is used to implement the step 140 of detecting the vibration characteristics of the faulty bearing when the detection is automatically performed by an algorithm. The detection module includes a processor 41 and a memory 42. The memory 42 includes instructions, which, when executed, enable the processor 41 to perform the step 140 of detecting the vibration characteristics of the faulty bearing of the method 100.

[0239] The system 10 may also include a communication device 60 to enable data communication between the processing module 30 and the detection module 40. For example, this may be a wired connection of the USB or Ethernet type, or a wireless connection of the Wifi type.

Claims

1. A method (100) for monitoring a rotating machine to detect faults in a bearing, the method (100) comprising the following steps: - acquiring (110) a vibration signal from a rotating machine measured by a vibration sensor; - determining (120) a 1st order spectrogram by concatenating a plurality of reduced 1st order spectra, the reduced 1st order spectra being obtained by: o applying a delta transform to correct a plurality of 1st order spectra obtained by Fourier transforming the vibration signal, the delta transform removing interference sources in the 1st order spectra from a subset of significant harmonics and a correction value for each significant harmonic; o determining a plurality of reduced first-order spectra by applying spectral normalization to the corrected first-order spectra, the reduced first-order spectra having a zero mean and a unit standard deviation; - determining (130) a 2nd order spectrogram by cascading a plurality of reduced 2nd order spectra, the reduced 2nd order spectra being obtained by: o determining and averaging the cyclic coherence from the vibration signal to obtain multiple averaged 2nd order spectra, the cyclic coherence being determined by double Fourier transform; o applying a delta transform to a plurality of averaged 2nd order spectra and then applying spectral normalization to obtain a plurality of reduced 2nd order spectra, the delta transform being applied from a subset of significant harmonics and a correction value for each significant harmonic; - Detecting (140) a vibration signature of a bearing fault based on the reference signature and from the 1st order spectrogram and the 2nd order spectrogram.

2. The monitoring method (100) according to the preceding claim, wherein the detecting step (140) further comprises identifying a faulty bearing from the vibration signature.

3. The monitoring method (100) according to claim 3, further comprising a maintenance step (150) for the identified faulty bearing.

4. The monitoring method (100) according to one of the preceding claims, wherein the vibration signals are acquired at a plurality of different operating phases of the rotating machine.

5. The monitoring method (100) according to one of the preceding claims, wherein the 1st order spectrogram and the 2nd order spectrogram are determined based on a Fourier transform applied to the vibration signal over a plurality of consecutive time windows having a duration between 0.1 seconds and 10 seconds.

6. The monitoring method (100) according to any one of the preceding claims, wherein the acquiring step (110) further comprises acquiring a speed signal of the rotating machine measured by a speed sensor.

7. The monitoring method (100) according to claim 6, in the step (120) of determining the first-order spectrum diagram, further comprising the following steps: -For each time window in multiple time windows: o resampling (121) the vibration signal from the velocity signal to obtain an angle signal; o Determine the (122) 1st order spectrum by applying a Fourier transform to the angle signal; o determining a (123) corrected 1st order spectrum by applying a delta transform to the 1st order spectrum; o determining (124) a normalized 1st order spectrum by applying spectral normalization to the corrected 1st order spectrum; o determining (125) a flattened 1st order spectrum by applying spectral autocorrelation to the normalized 1st order spectrum; - concatenating (128) the flattened 1st order spectra determined for each time window to form a 1st order spectrogram.

8. The monitoring method (100) according to claim 7, in the step (120) of determining the first-order spectrum diagram, further comprising the following steps: -For each time window in multiple time windows: o determining (126) a corrected flattened 1st order spectrum by applying a delta transform to the flattened 1st order spectrum; o determining (127) a reduced 1st order spectrum by applying spectral normalization to the corrected flattened 1st order spectrum; The concatenation step (128) concatenates the reduced 1st order spectra determined for each time window to form a 1st order spectrogram.

9. The monitoring method (100) according to any one of claims 6 to 8, in the step (130) of determining the second-order spectrum diagram, further comprising the following steps: -For each time window in multiple time windows: o resampling (131) the vibration signal from the velocity signal to obtain an angle signal; o removing (132) the deterministic part of the angle signal to obtain a corrected vibration signal; o Determine the (133) 2nd order spectrum according to the cyclic coherence of the corrected vibration signal; o Averaging the 2nd order spectrum (134); o determining (135) a flattened 2nd order spectrum by applying spectral autocorrelation to the averaged 2nd order spectrum; - concatenating (138) the flattened 2nd order spectra determined for each time window to form a 2nd order spectrogram.

10. The monitoring method according to claim 9, in the step (130) of determining the second-order spectrum diagram, further comprising the following steps: -For each time window in multiple time windows: o determining (136) a corrected flattened 2nd order spectrum by applying a delta transform to the flattened 2nd order spectrum; o determining (137) a reduced 2nd order spectrum by applying spectral normalization to the corrected flattened 2nd order spectrum; The concatenation step (138) is to concatenate the reduced 2nd order spectra determined for each time window to form a 2nd order spectrogram.