METHOD AND DEVICE FOR DETECTING A FAULT IN A MECHANICAL SYSTEM WITH AT LEAST ONE ROTATING ELEMENT

Description

[0001] La présente The invention lies in the field of systems for monitoring the operation of mechanical systems, in particular mechanical systems comprising at least one rotating member.

[0002] The present invention relates to a method for detecting a fault on a mechanical system comprising at least one rotating member as well as a device for detecting a fault configured to implement such a method.

[0003] Such a mechanical system comprises at least one rotating member, for example an input shaft and / or an output shaft. For the sake of simplification, a mechanical system comprising at least one rotating member is hereinafter referred to as a “mechanical system”.

[0004] For example, a mechanical system may include one or more bearings to guide the rotation of one or more rotating parts. A bearing includes, for example, a rolling bearing equipped with one or more rows of rolling elements such as balls, rollers or others.

[0005] Such a mechanical system may also be provided with at least one toothed wheel, a pinion, a toothed crown in order to provide a reduction or an increase in the rotation speed between two rotating members of the mechanical system, in particular between an input shaft and an output shaft.

[0006] Such a mechanical system may, for example, be equipped with an epicyclic gear train to provide a high reduction ratio of the rotation speed between two rotating parts of the mechanical system.

[0007] Such a mechanical system may, for example, be a gearbox or a power transmission box of a vehicle, in particular an aircraft.

[0008] A breakdown or malfunction of such a mechanical system may occur, for example, following the appearance of a defect on a bearing, in particular a rolling bearing, and / or on a toothed wheel, a pinion, a crown gear. Such a defect may, for example, take the form of a crack, flaking, a fissure, or even a break, on a toothed wheel, a pinion, a crown gear or a bearing. Such a defect may also take the form of a seizing of a rolling element.

[0009] Some surveillance systems designated by the acronym HUMS for "Health and Usage Monitoring System" in English, aim to monitor one or more mechanical systems through different sensors by tracking the evolution of a set of monitoring indicators. These monitoring indicators are calculated from the measurements of one or more sensors in order to characterize the state and operation of each mechanical system. For example, a monitoring indicator can be defined by a signal provided by one sensor or by combining the signals of several sensors. Several monitoring indicators can also exploit the measurements from a single sensor through various characteristics of the signal provided by this sensor, such as its time or frequency spectrum.

[0010] The evolution of each monitoring indicator can be compared to a detection threshold in order to detect or anticipate a possible fault or breakdown of the monitored mechanical system. The value of each detection threshold can be obtained through experience, through a statistical analysis of a history of measurements from several similar mechanical systems or through individual learning on a given mechanical system.

[0011] A monitoring indicator may take the form of a vibration indicator evaluated using a sensor comprising at least one accelerometer, a tachometer or a strain gauge for example. Such a monitoring indicator may in this case be equal to the maximum amplitude of a time-dependent vibration signal provided by an accelerometer for example.

[0012] For a vibration signal varying in the time domain, an indicator can be determined from statistical functions such as the quadratic mean, the crest factor, the asymmetry coefficient or the flattening coefficient of a distribution for example.

[0013] For a vibration signal expressed in the frequency domain, an indicator can be determined from the mean frequency or the standard deviation frequency.

[0014] Other indicators can be constructed from decompositions of the signal in time and frequency such as the wavelet transform, the empirical mode decomposition or the short-term Fourier transform for example.

[0015] Most of these indicators are calculated from pre-processing performed on a raw signal measured by a sensor in order to eliminate or significantly reduce noise and / or fluctuations in the speed of the monitored components. Among these pre-processing, such as angular resampling, calculations of the synchronous mean and synchronous variance can also be used, with the aim of attenuating certain components of the raw signal such as for example a random component and / or noise.

[0016] The publications “A Review of Vibration Based Technique for Helicopter Transmission Diagnostics” by Samuel PD (Journal of sound and vibration, 2005), “A Review on Vibration-Based Condition Monitoring of Rotating Machinery” by Monica Tiboni (Department of Mechanical and Industrial Engineering, University of Brescia, 2022) and “Cepstrum Analysis and Gearbox Fault Diagnosis” by RB Randall (Bruel & Kjaer) describe methods for analyzing vibration signals measured on rotating machines equipped with gears to establish diagnostics of these machines and determine the possible presence of faults.

[0017] Furthermore, document CN 102426102 describes a method intended to detect the appearance of a crack on a transmission shaft carrying a gear by analyzing a vibration signal measured near the gear and a signal comprising information relating to the rotation speed of the transmission shaft.

[0018] Document FR 3014502 describes a method and a device for detecting mechanical defects in a variable-speed rotating machine using vibration signatures generated by different speed ranges from measurement signals from vibration sensors fitted to non-rotating components of the machine during several operating cycles. These vibration signatures are then compared with a reference vibration signature established for each vibration sensor in a fault-free mode of the rotating machine.

[0019] Developing a sensitive and robust fault detection system is essential to prevent catastrophic failures of power transmission systems, especially for aircraft. Any improvement in detecting the occurrence of faults in a power transmission system significantly improves safety.

[0020] The prior art reports the monitoring of frequencies linked to a fault, frequencies known a priori, but does not take into account the linear combinations of these frequencies or the overall evolution of their trends. In general, the trend of the energy of the frequency channels of the fault is followed but the evolution of the energy of the associated harmonics and the number of harmonics having significant energy is not taken into account.

[0021] The present invention therefore aims to propose an alternative method and an alternative device aimed at improving the detection of the appearance or presence of a fault on a mechanical system comprising at least one rotating member.

[0022] The present invention relates, for example, to a method for detecting a fault on a mechanical system, the mechanical system comprising at least one member rotating in rotation around an axis AX of rotation as well as a vibration sensor emitting a temporal vibration signal, an angular sensor emitting a temporal angular signal varying as a function of an angular position of the member rotating around the axis AX of rotation and a computer.

[0023] The vibration sensor(s) thus make it possible to measure characteristics, such as vibrations for example, of the mechanical system as a whole or of one or more of its components and to emit a temporal vibration signal. x(t), namely in the time domain, carrying information relating to these characteristics, to the computer for example. The angular sensor makes it possible to measure an angular position of the organ rotating around the axis AX of rotation, relative to a reference of the mechanical system, for example a frame or a casing of the mechanical system, and to emit a time angular signal θ ( t ), namely in the time domain, carrying information relating to this angular position.

[0024] Un tel système mechanical system may comprise at least one of the following elements: at least one rotating member such as an input shaft and / or an output shaft, a rotating guide bearing and for example a bearing provided with rolling elements, a toothed wheel, a pinion, a toothed crown. Such a mechanical system may for example be a gearbox or a power transmission box of a vehicle.

[0025] The method according to the invention comprises the following steps: a processing of temporal vibration signals x m ( t ) emitted by the vibration sensor comprising the following sub-steps: ∘ emission of M temporal vibration signals x m ( t ) by the vibration sensor, M being greater than or equal to 2, m being a positive integer varying from 1 to M, « t » being the operating time of the mechanical system, o first transformation, with the calculator, of the M temporal vibration signals x m ( t ) in M ​​angular vibration signals x m ( θ ) as a function of the temporal angular signal θ m ( t ), " θ » being an angle of the organ rotating around the axis AX of rotation, o second transformation, with the calculator, of the M angular vibration signals xm (θ) into M frequency vibration signals x m ( f θ ) by a frequency operator ϑ, where “ f θ » is a frequency channel, o first determination of trend frequency channels f t depending on the M frequency vibration signals x m ( f θ ) , a trend frequency channel f t being a frequency channel f θ for which the M frequency vibration signals x m ( f θ ) meet a selection criterion relating to a growth trend, analyze, with the calculator, trend frequency channels f t comprising the following sub-steps: ∘ comparaison of each trend frequency channel f t with linear combinations of fault frequencies f d relating to the fault sought, the linear combinations respectively comprising sums of several of the fault frequencies f d multiplied by coefficients n i integers, ∘ second determination of fault frequency channels fx based on the comparison, a fault frequency channel f x being a trend frequency channel f t whose differences with linear combinations are lower than a comparison threshold, calculation, with the calculator, of a presence indicator I optimal of the fault on the mechanical system as a function of the number of fault frequency channels f x and the number of linear combinations of fault frequencies f d , triggering an alert for the presence of a fault based on a comparison between the presence indicator I optimal of the fault and a threshold for the presence of a fault.

[0026] A vibration sensor, within the scope of the invention, may comprise, for example, an accelerometer, a tachometer, an encoder-type sensor measuring an angular position of a shaft, a strain gauge, or others. Such a vibration sensor makes it possible to measure temporal vibration signals, comprising, for example, acceleration signals or speed signals. Such a vibration sensor may be positioned at different positions in the mechanical system and with various orientations. Such a vibration sensor may preferably be positioned on or near a rotating member, for example an input or output shaft, or a specific element to be monitored, such as a bearing, a toothed wheel, a pinion, or a toothed crown, for example. The vibration sensor may thus emit a temporal angular signal carrying information relating to the vibrations captured over time.

[0027] "Position" refers to the location where the vibration sensor is arranged on the mechanical system. "Orientation" refers to the angles of one or more preferred measurement directions of the vibration sensor.

[0028] The angular sensor can be positioned close to the rotating organ whose angular position it is to measure. The angular sensor can thus emit a temporal angular signal carrying information relating to the variation of this angular position over time.

[0029] The time-domain vibration signal and the time-domain angular signal may be signals formed by raw measurements emitted respectively by the vibration sensor and the angular sensor or by measurements obtained by more or less complex signal processing carried out by the computer or by a computer integrated into the corresponding sensor from such raw measurements, for example via standard filtering or sampling, or the application of transformations.

[0030] The temporal vibration signal and the temporal angular signal can be generated simultaneously, or even synchronously.

[0031] Thus, the vibration sensor emits M temporal vibration signals x m ( t ) during a predetermined duration of operation of the mechanical system. The temporal vibration signals x m ( t) are generally emitted at roughly equal regular intervals and each contains information relating to the vibrations experienced by part or all of the mechanical system during a measurement period T f predefined duration of the mechanical system. This measurement duration T f can be equal to several seconds, or even several minutes and cover several operating cycles of the mechanical system, for example ten rotation cycles of the organ rotating around the axis AX of rotation, the angle θ of the rotating organ being in this case equal to 3600°. However, temporal vibration signals x m ( t ) can also be emitted with different intervals between them. For example, a time-varying vibration signal x m ( t ) can be issued systematically after starting the mechanical system, possibly once a stabilized rotation speed has been reached.

[0032] At the same time, the angular sensor emits, for the predetermined duration, M temporal angular signals θ ( t ) which can be associated and synchronized with temporal vibrational signals x m ( t ).

[0033] Each temporal vibrational signal x m ( t ) and each temporal angular signal θ m ( t ) are emitted for a measurement duration T f , for example equal to a few tens of seconds or a few minutes. The temporal vibration signals x m ( t ) and the temporal angular signals θ m ( t ) can be stored in a computer memory or in a memory connected to the computer.

[0034] Then, the calculator applies two successive transformations to the M temporal vibration signals x m ( t) in order to transform them initially, in the usual way, into M angular vibration signals x m ( θ ) then to transform, using a frequency operator ϑ, the M angular vibration signals x m ( θ ) in M ​​frequency vibration signals x m ( f θ ) . Each frequency vibration signal x m ( f θ ) therefore comes from a temporal vibrational signal x m ( t ) and therefore of a vibration measurement carried out on the mechanical system. The M frequency vibration signals x m ( f θ ) are therefore spectral signals discretized according to the angular position θ of the rotating organ and vary according to a frequency channel f θ which is linked to this angular position θ . Frequency channels f θ are therefore relative to the angular positions of the rotating organ and not relative to time, like a frequency classically used. We designate by "frequency channels f θ » the frequencies of these discretized spectral signals.

[0035] Angular vibration signals x m ( θ ) and frequency vibration signals x m ( f θ ) can be stored in the computer memory or in the memory connected to the computer.

[0036] One or more trend frequency channels f t indicating a growth trend of M frequency vibration signals x m ( f θ ) are then determined according to a selection criterion. Thus, the M frequency vibration signals x m ( f θ ) associated with a trend frequency channel f t indicate a growth trend and satisfy this selection criterion relating to this growth trend. The expression "the M frequency vibration signals x m ( f θ ) indicate a growth trend" means that, for the same frequency channel f θ , the amplitudes of the M frequency vibration signals x m ( f θ ) are increasing from the first frequency vibration signal x 1 ( f θ ) towards the last frequency vibration signal x m ( f θ ) according to a curve whose characteristic can be this growth trend.

[0037] Each trend frequency channel f t is therefore a frequency channel f θ whose amplitudes x m ( f t ) frequency vibration signals x m ( f θ ) relating to this trend frequency channel f t increase during the operation of the mechanical system. Each trend frequency channel f t is therefore a frequency channel f θ which characterizes an increase in vibrations during the operation of the mechanical system. This increase in vibrations can thus potentially characterize the appearance, or even the presence, of a fault in the mechanical system. The trend frequency channels f t can be stored in the computer memory or in the memory connected to the computer.

[0038] The calculator then analyzes these trend frequency channels f t by comparing each trend frequency channel f t with linear combinations of fault frequencies f d relating to the specific fault being sought, and determines fault frequency channels fx following this comparison. The fault frequency channels fx can be stored in the computer's memory or in the memory connected to the computer.

[0039] Fault frequencies f d relating to the defect sought are known and defined in advance by feedback, by tests, or even by simulations. The defect frequencies f d may have been previously stored in the computer memory or in the memory connected to the computer. Indeed, a fault in a mechanical system often generates amplitude and frequency modulations on the vibrations of this mechanical system and gives rise to a set of excited frequency channels whose frequency can be expressed as a linear combination of the fault frequency(ies) f d .

[0040] Each fault frequency f d has for example been identified on a mechanical system presenting the desired defect and for which the amplitudes of the measured vibration signals are significant. The defect frequency(ies) f d associated with a specific fault being sought can be contained in a fault model. In this way, several fault models can exist, allowing the respective characterization of different types of faults likely to appear on the mechanical system.

[0041] Additionally, a fault model can be parametric. Fault frequencies f d associated with the fault considered can then be variable depending on operating parameters of the mechanical system, such as the operating phase of the mechanical system, and / or the torque of a rotating part, or even the temperature inside the mechanical system for example. The fault frequencies f d can also be variable depending on parameters external to the mechanical system, such as temperature for example.

[0042] Each linear combination comprises a sum of several of these fault frequencies f d multiplied by coefficients n i integers. The use of these linear combinations therefore makes it possible to compare the trend frequency channels f t with the fault frequencies f d known as well as with some of their harmonic frequencies, according to the coefficients n i used.

[0043] A fault frequency channel f x is thus determined and equal to a trend frequency channel f t whose differences with the linear combinations are lower than a comparison threshold, or even zero. In this way, the trend frequency channels are selected f t which are linear or “almost” linear combinations of the fault frequencies f d of the defect model relating to the defect sought.

[0044] One or more fault frequency channels f x can thus be determined equal respectively to one or more fault frequency channels f t when these coincide, within the comparison threshold, at a fault frequency f d , at one of its harmonic frequencies or at linear combinations of these fault frequencies f d . The comparison threshold may have been previously stored in memory.

[0045] Determination of one or more fault frequency channels f x thus characterizes a risk of the presence of a defect in the mechanical system.

[0046] To confirm this risk, the calculator calculates a presence indicator I optimal characterizing this risk of presence of a fault on the mechanical system as a function of the number of fault frequency channels fx and the number of linear combinations of fault frequencies f d . This presence indicator I optimal can then be compared to a threshold for the presence of a defect to confirm or deny this risk. The threshold for the presence of a defect may have been previously stored in the memory.

[0047] Thus, an alert of the presence of a fault is triggered based on a comparison between the presence indicator I optimal of the fault and the threshold for the presence of a fault. This alert of the presence of a fault can be triggered by an alert generator. This alert of the presence of a fault can be visual, audible and / or haptic for example.

[0048] The method according to the invention thus makes it possible to reliably determine the detection of the appearance of a fault on the mechanical system or its presence as soon as frequency channels show a significant growth trend, by identifying these frequency channels, then comparing them with linear combinations of fault frequencies. f d known and associated with the defect being sought. As a result, the detection of this defect can be anticipated, allowing it to be dealt with as early as possible, thus avoiding the risks of an aircraft being interrupted or cancelled, an emergency landing, or even an accident.

[0049] The method according to the invention may comprise one or more of the following characteristics, taken alone or in combination.

[0050] According to one example, the comparison threshold may be predetermined and may for example be equal to a frequency resolution associated with the discretized frequency vibration signals.

[0051] According to another example compatible with the previous ones, the predetermined duration during which the vibration sensor emits the M temporal vibration signals x m ( t ) can be slippery.

[0052] In this way, when the vibration sensor emits a (M+1)th temporal vibration signal x M +1 ( t ), the process is then applied to M temporal vibration signals x m ( t ) , m being between 2 and M+1. The (M+1)th temporal vibrational signal x M +1 ( t ) is for example stored in the memory and the first temporal vibration signal x 1 ( t ) may eventually be deleted from memory.

[0053] The predetermined duration may be, for example, several hours or even several days of operation depending on the mechanical system concerned. The interval between two temporal vibration signals may be, for example, several tens of minutes or several hours of operation of this mechanical system.

[0054] According to another example compatible with the previous ones, a number of the coefficients n i integers can be limited, and consequently the values ​​of the coefficients n i integers can be bounded.

[0055] In fact, the higher the value of the coefficients n i is high, the greater the risk that frequency channels meeting the criterion of fault frequency channels f x are confused with other physical phenomena of the system increases and, in fact, the risk of detecting false faults also increases.

[0056] A limited number of coefficients n i also allows to limit the number of linear combinations and, consequently, the calculation time of the default frequency channels fx.

[0057] According to another example compatible with the previous ones, the presence indicator I optimal of the fault may increase when the number of fault frequency channels f x increases and the fault presence alert can be triggered if the presence indicator I optimal is greater than the threshold for the presence of a defect.

[0058] For example, the presence indicator I optimal of the fault can be equal to a ratio of a number of the fault frequency channels f x by a total number of linear combinations of the fault frequencies f d and the fault presence alert can be triggered if the presence indicator I optimal is greater than the threshold for the presence of a defect.

[0059] The presence indicator I optimal is thus normalized and between 0 and 1.

[0060] According to another example compatible with the previous ones, the first determination of the trend frequency channels f t may include the following sub-steps: generation, with the calculator, of at least one series of amplitudes X fθ associated with a frequency channel f θ and comprising M amplitude values ​​belonging respectively to the M frequency vibration signals and relating to the frequency channel f θ , the number of amplitude series X fθ being equal to the number of frequency channels f θ , calculation, with the calculator, of a trend indicator I trend ( f θ ) for each frequency channel f θ depending on the series of amplitudes associated with the frequency channel f θ , and first determination, with the calculator, of trend frequency channels f t , each trend frequency channel f t being a frequency channel f θ meeting the selection criterion, this selection criterion then being relative to the trend indicator I trend ( f θ ) .

[0061] Each series of amplitudes X fθ thus contains as many amplitude values ​​as temporal vibration signals were emitted over the predetermined duration. Each series of amplitudes X fθ thus characterizes the “energy levels” associated with the frequency channel f θ to which this series of amplitudes is associated X fθ . Two series of amplitudes X fθ distinct are associated respectively with two frequency channels f θ distinct.

[0062] The first determination then makes it possible to identify, among the frequency channels f θ , trend frequency channels f t for which these energy levels are increasing significantly over the predetermined duration. Indeed, a trend indicator I trend ( f θ ) reflecting the growth trend of trend frequency channels f t , meets and satisfies the selection criterion, and can reflect the appearance of the defect considered.

[0063] Indeed, for almost all faults, the energy levels of the frequency channels associated with a fault increase slowly for a long time at the onset of the fault and then gradually increase more quickly until a rupture. The purpose of the trend indicator is therefore to identify the frequency channels that are in the slow growth phase. This allows them to be identified well before the rupture occurs.

[0064] Each trend frequency channel f t is a frequency channel f θ meeting the selection criterion, this selection criterion then being relative to the trend indicator I trend ( f θ ).

[0065] According to a first selection criterion, each trend frequency channel f t is a frequency channel f θ for which said trend indicator I trend ( f θ ) is greater than the trend threshold. Each trend frequency channel f t is, according to this first selection criterion, a frequency channel f θ whose amplitudes x m ( f T ) frequency vibration signals x m ( f θ ) relating to this trend frequency channel f t increase during the operation of the mechanical system more quickly than the amplitudes of a predetermined curve associated with the trend threshold.

[0066] The first determination then makes it possible to identify, among the frequency channels f θ , trend frequency channels f t for which these energy levels are increasing significantly over the predetermined duration, which may reflect the appearance of the fault considered if a growth trend in these energy levels is greater than the trend threshold.

[0067] According to a second selection criterion, the trend frequency channels f t are the P frequency channels f θ corresponding to the P trend indicators I trend ( f θ ) the largest, P being an integer greater than one. P is for example equal to 5. In this case, the trend frequency channels f t are for example the frequency channels f θ whose M temporal vibrational signals x m ( t ) are characterized by the strongest growth.

[0068] Each trend frequency channel f t is, according to this second selection criterion, a frequency channel f θ whose amplitudes x m ( f t ) frequency vibration signals x m ( f θ ) relating to this trend frequency channel f t grow most rapidly during the operation of the mechanical system. The first determination then makes it possible to identify, among the frequency channels f θ , trend frequency channels f t , for which these energy levels increase most rapidly over the predetermined duration, and likely to reflect the appearance of the fault considered.

[0069] Regardless of the selection criterion used, each trend frequency channel f t is therefore a frequency channel f θ which characterizes an increase in vibrations during the operation of the mechanical system. This increase in vibrations can thus potentially characterize the appearance, or even the presence, of a fault in the mechanical system. The trend frequency channels f t can be stored in the computer memory or in the memory connected to the computer.

[0070] The first and second selection criteria can be used independently or in a complementary manner, i.e. simultaneously.

[0071] Further, each amplitude value of said at least one series of amplitudes X fθ can be associated with a moment t m of a start of emission of the temporal vibratory signal x m ( t ) from which the amplitude value is derived.

[0072] Alternatively, the moment t m can also correspond to the start time of a measurement of the vibrations experienced by the mechanical system, the temporal vibration signal x m ( t ) carrying information relating to these vibrations. The start of emission of the temporal vibratory signal x m ( t ) and the start of a measurement of the vibrations experienced by the mechanical system can be significantly confused and simultaneous.

[0073] Furthermore, the calculation of a trend indicator I trend ( f θ ) may include the following sub-steps: third transformation of said at least one series of amplitudes X fθ by a transformation function Φ in at least one series of trends according to the relationship g fθ = φ ( X fθ ), construction of at least one trend line from said at least one trend series by linear regression, according to the relationship h fθ ( r ) = A fθ × r + B fθ , Or " r » is a variable, « h fθ ( r ) » is a trend line, « A fθ » is a slope coefficient of the trend line, « B fθ » is an ordinate at the origin of the trend line, estimation of a noise coefficient σ r for each trend line h fθ ( r ) , according to the relationship σ r f θ = 1 M ∑ m = M g f θ − A f θ × r + B fθ )) 2< | 1 / 2< , where « Σ m = M » is the sum function with m varying from 1 to M, « | | » is the absolute value function, calculation of the trend indicator I trend as a function of the noise coefficient relative to the trend line for each frequency channel f θ , according to the relationship I trend f θ = A f θ α + β × σ r f θ , Or " α » and « β » are predetermined parameters.

[0074] The trend indicator I trend is thus calculated as a ratio between the slope of the linear regression and the noise level.

[0075] The parameters αand β allow us to weight the sensitivity to noise, namely the influence of this noise.

[0076] The parameters α and β can be specific to each defect sought.

[0077] The transformation function Φ can for example be the unit function or the logarithm function or another mathematical function. The aim of this third transformation is to obtain a globally linear evolution of the amplitude values ​​after application of the transformation function Φ during the slow growth phase of these amplitude values ​​in the event of the appearance of a fault. The choice of the function Φ allows in particular to obtain good sensitivity either at the beginning or at the end of propagation of the fault. The construction of a trend line by linear regression is then more precise and more reliable to obtain a slope of evolution and an estimate of the noise coefficient σ r , and then to build a trend indicator I trend robust and simple.

[0078] Alternatively, transformations leading to obtaining a characteristic evolution curve linked to the nature of evolution of the amplitude values ​​associated with the frequency channels of the desired defect can be used. This characteristic evolution curve can be approached using curve fitting methods by a reduced parametric expression, specific to this characteristic evolution curve. Thus, the parameters of the characteristic evolution curve and the estimation of the noise coefficient would also make it possible to construct a trend indicator. I trend robust to identify trending frequency channels f t .

[0079] The method according to the invention may also include a step of filtering the vibration signals in order to filter the vibration signals, in particular when these vibration signals are noisy. This filtering step may be applied to both temporal vibration signals x m ( t ), angular vibration signals x m ( θ ) or frequency vibration signals x m ( f θ ) , than on the amplitude series X fθ . This filtering step allows the vibration signals and / or series to be smoothed in the usual way by applying, for example, median filtering or other existing filters.

[0080] The present invention also relates to a computer program comprising instructions which, when the program is executed, cause it to implement the method described above. The computer program may, for example, be executed by a computer. The instructions are, for example, stored in a memory of the computer or connected to the computer.

[0081] The present invention also relates to a monitoring device for monitoring a mechanical system and configured for implementing the method described above and thus for detecting the presence of a fault on a mechanical system comprising at least one member rotating in rotation around an axis AX of rotation. The monitoring device comprises at least one vibration sensor emitting a temporal vibration signal, an angular sensor emitting a temporal angular signal varying as a function of an angular position of the member rotating around the axis AX of rotation and a computer.

[0082] The present invention also relates to a mechanical system comprising at least one member rotating in rotation around an axis AX of rotation and a monitoring device as previously described for monitoring the mechanical system and for detecting the presence of a fault on the mechanical system.

[0083] This mechanical system can, for example, be a power transmission box of a vehicle, and of an aircraft in particular.

[0084] The present invention also relates to a power transmission box comprising such a mechanical system.

[0085] The present invention may finally relate to an aircraft comprising such a power transmission box.

[0086] The invention and its advantages will appear in more detail in the context of the description which follows with examples given for illustrative purposes with reference to the appended figures which represent: there figure 1 , a schematic side view of an aircraft, and the figure 2 , a diagram illustrating a method of detecting a fault in a mechanical system.

[0087] Elements present in several distinct figures are assigned a single reference.

[0088] There figure 1 represents a vehicle 1, and a rotary wing aircraft of the rotorcraft type in particular. This vehicle 1 comprises a mechanical system 10 provided with one or more rotating members 15 rotating around an axis AX of rotation. A rotating member 15 may for example comprise an output shaft or an input shaft.

[0089] Such a mechanical system 10 may comprise one or more rotational guide bearings, for example for the rotational guidance of at least one rotating member. A bearing comprises, for example, a rolling bearing provided with rolling elements.

[0090] Such a mechanical system 10 may also comprise, for example, at least one toothed wheel, a pinion, a toothed crown, fixed or mobile.

[0091] This mechanical system 10 may be, for example, a gearbox or a power transmission box of the rotary-wing aircraft 1. This mechanical system 10 may be connected, for example, to one or more engines 2, via one or more input shafts respectively, and may drive a rotor in rotation, such as, for example, a main rotor 3 via an output shaft, or possibly even an auxiliary rotor 4 as shown in the figure 1 .

[0092] Alternatively, such a mechanical system 10 may be arranged in a gearbox or power transmission box of a vehicle 1 or any other mechanical equipment.

[0093] Whatever its arrangement, the mechanical system 10 also comprises one or more vibration sensors 20, an angular sensor 25 measuring an angular position of a rotating member 15 around an axis AX of rotation and a computer 5. The mechanical system 10 may comprise several angular sensors 25 arranged respectively on several distinct rotating members 15.

[0094] Each vibration sensor 20 can measure and / or emit a temporal vibration signal relating to a vibration behavior of the mechanical system 10 as a whole, or to a particular vibration behavior of a rotating member 15, a bearing or even a gear for example. The temporal vibration signal carries information relating to the vibrations of the mechanical system 10 or one of its components. The vibration sensor(s) 20 may comprise for example an accelerometer, a tachometer, an encoder type sensor and / or a strain gauge.

[0095] The angular sensor 25 measures an angular position of a rotating member 15 around an axis AX of rotation. The angular position of the rotating member 15 around an axis AX of rotation can be defined relative to a reference frame of the mechanical system 10, for example a casing of the mechanical system 10. The time signal can carry information relating to the angular position of a rotating member 15 around an axis AX of rotation.

[0096] The angular sensor 25 may comprise an angular position sensor directly measuring a time signal indicating the variation of the angular position of the rotating member 15 as a function of time.

[0097] Alternatively, the angular sensor 25 may comprise an angular speed sensor or an angular acceleration sensor measuring a time signal relating respectively to an angular speed or to an angular acceleration which must undergo a simple integration or a double integration, in order to generate a time signal providing the variation of the angular position of the rotating member 15 as a function of time. This simple or double integration may be executed by the computer 5. This simple or double integration may also be executed by a computer integrated into the angular sensor 25.

[0098] The computer 5 may comprise at least one processor and at least one memory, at least one integrated circuit, at least one programmable system or at least one logic circuit, these examples not limiting the scope given to the expression “computer”. The computer 5 may also be connected to a memory by a wired connection or a wireless connection.

[0099] The memory may, for example, store instructions or algorithms relating to the performance of a fault detection method on the mechanical system 10 and one or more thresholds corresponding to this method. The memory may also store a computer program intended to be executed by the computer 5 in order to implement the fault detection method.

[0100] The vibration sensor(s) 20 as well as the angular sensor 25 and the computer 5 may be part of a monitoring device 9 of the mechanical system 10 intended to monitor the mechanical system in order to detect and identify a risk of the appearance or presence of a fault likely to cause a breakdown or malfunction of the mechanical system 10.

[0101] The computer 5 can be dedicated to the monitoring device 9 or be shared with other devices of the aircraft 1.

[0102] The fault detection method according to the invention comprises four main steps as shown in the diagram of the figure 2 . A first main processing stage 100 processes the temporal vibration signals x m ( t ) in order to determine trend frequency channels f t likely to correspond to a fault present on the mechanical system 10. A second main analysis step 200 carries out an analysis of these trend frequency channels f t to identify among them fault frequency channels f x , corresponding to the defect sought.

[0103] Then, a third main calculation step 300 allows the calculation of a presence indicator I optimal of the fault sought on the mechanical system 10 and a fourth main triggering step 400 triggers an alert of the presence of a fault based on this presence indicator I optimal of the fault and a threshold for the presence of a fault.

[0104] In this way, the detection of a fault in the mechanical system 10 can be reliably anticipated from the first signs of its appearance, the presence of this fault being able to be detected through the vibrations of this mechanical system 10.

[0105] The first main processing step 100 has several sub-steps.

[0106] Thus, during an emission step 120, M temporal vibration signals x m ( t ) are emitted by the vibration sensor 20, for example over a predetermined duration of operation of the mechanical system 10. This predetermined duration may be sliding. M is an integer greater than or equal to 2, and m is an integer varying from 1 to M identifying the number of the temporal vibration signal x m ( t ) and therefore the number of the vibration measurement of the mechanical system 10. " t» is the operating time of the mechanical system 10. The operating time t is considered to be stopped when the mechanical system 10 is not in operation.

[0107] Preferably, the M temporal vibrational signals x m ( t ) correspond to vibration measurements carried out at operating speeds of the mechanical system 10 that are substantially similar and close. Operating speeds are considered to be close, for example, if the rotation speed differences of the rotating member 15 are less than 10% and if the torque differences of the rotating member 15 are less than 30%.

[0108] Each temporal vibrational signal x m ( t ) corresponds to a measurement of the vibrations of the mechanical system 10 carried out during a measurement period T f , preferably identical for all temporal vibration signals x m ( t ) .

[0109] Simultaneously, M temporal angular signals θ ( t ) are emitted by the angular sensor 25, " θ » being an angle of the rotating member 15 around the axis AX of rotation.

[0110] The M temporal vibrational signals x m ( t ) and the M temporal angular signals θ m ( t ) are transmitted, in the form of electrical or optical, analog or digital signals, by a wired or wireless link, to the computer 5.

[0111] Alternatively, the vibration sensor(s) 20 and the angular sensor 25 may continuously transmit a temporal vibration signal. x(t) and a temporal angular signal θ ( t ) to the computer 5 which records at regular intervals the M temporal vibration signals x m ( t ) and the M temporal angular signals θ ( t ) in memory.

[0112] Then, during a first transformation step 140, the computer 5 transforms, in a known manner, the M temporal vibration signals x m ( t ) in M ​​angular vibration signals x m ( θ ) as a function of the temporal angular signal θ ( t ) .

[0113] This transformation of the temporal vibrational signal x m ( t ) from the time domain to the angular domain advantageously makes it possible to associate the variations of the angular vibration signal with the angular positions of the rotating member 15, and therefore with positions of the elements of the mechanical system 10. This first transformation step 140 therefore allows resampling of the time vibration signal x m ( t) in the angular domain in order to attenuate in particular the effect of fluctuations in the rotation speed of the rotating member 15 on the vibration measurements. Indeed, the rotation speed of the rotating member 15 is not systematically constant, in particular during transient phases, for example during the start-up of the mechanical system 10 or changes in speed and in the event of high stresses on the mechanical system 10.

[0114] Then, during a second transformation step 150, the computer 5 transforms the M angular vibration signals x m ( θ ) in M ​​frequency vibration signals x m ( f θ ) by a frequency operator ϑ. This second transformation can be formulated according to the relation x m ( f θ ) = ϑ( x m ( θ )), Or " f θ » is a frequency channel. Frequency channels f θ are derived from the second transformation using the frequency operator ϑ from the M angular vibration signals x m ( θ ) function of the angle θ of the rotating member 15 around the axis AX of rotation.

[0115] This second transformation from the angular domain to the frequency domain advantageously makes it possible to highlight the energy levels of the frequency channels f θ vibrations of the mechanical system 10.

[0116] The frequency operator ϑ can be different depending on the defect being sought. The frequency operator ϑ can for example be a Fourier transform, an envelope spectrum or another transformation.

[0117] Finally, during a first determination 160, the calculator 5 determines trend frequency channels f t depending on the M frequency vibration signals x m ( f θ ) , a trend frequency channel f t being a frequency channel f θ for which the M frequency vibration signals x m ( f θ ) satisfy a selection criterion relating to a growth trend. A trend frequency channel f t corresponds for example to a frequency channel f θ for which the energy levels of the vibrations of the mechanical system 10 are increasing and therefore likely to correspond to the appearance of a defect present on the mechanical system 10.

[0118] The first determination of 160 trend frequency channels f t may include several sub-steps.

[0119] During a generation sub-step 170, the calculator 5 generates at least one series of amplitudes X fθ associated with a frequency channel f θ and comprising M amplitude values ​​belonging respectively to the M frequency vibration signals and relating to this frequency channel f θ . The number of amplitude series X fθ is equal to the number of frequency channels f θ . The amplitudes of each amplitude series X fθ characterize the vibration energy for each of the frequency channels f θ .

[0120] Each amplitude value of the amplitude series X fθ can be associated with the number m relating to each vibration measurement.

[0121] Alternatively, each amplitude value of the amplitude series X fθ can be associated with a moment t m of a start of emission of the temporal vibratory signal x m ( t ) from which the amplitude value is derived. This association is particularly interesting when the vibration measurements are not carried out at regular intervals, in order to take into account the irregular intervals between two vibration measurements.

[0122] Additionally, a sub-step of filtering the amplitude values ​​of the amplitude series X fθ can be applied. Alternatively, such a filtering sub-step can also be applied to time series of vibration signals x m ( t ), to the series of angular vibration signals x m ( θ ) .

[0123] Then, during a calculation sub-step 180, the calculator 5 calculates a trend indicator I trend ( f θ ) for each frequency channel f θ depending on the series of amplitudes associated with the frequency channel f θ .

[0124] Then, during a first determination sub-step 190, the calculator 5 determines trend frequency channels f t by checking, for each frequency channel f θ , if the trend indicator I trend ( f θ ) meets and satisfies the selection criterion.

[0125] For example, a first selection criterion may include a trend threshold. Each trend frequency channel f t is then a frequency channel f θ for which the trend indicator I trend ( f θ ) is greater than the trend threshold. A trend frequency channel f t is therefore the frequency channel f θ for which the series of amplitudes is globally increasing when m varies from 1 to M, and therefore presenting an increasing trend greater than a predefined trend and corresponding to the trend threshold.

[0126] According to a second selection criterion, the trend frequency channels f t are the P frequency channels f θ whose trend indicators I trend ( f θ ) are the largest, P being an integer greater than one. The trend frequency channels f t are the frequency channels f θ for which the amplitude series are globally increasing when m varies from 1 to M and which include the M temporal vibration signals x m ( t ) characterized by the strongest growths in their respective energy levels.

[0127] Only one selection criterion from the first and second selection criteria may be used.

[0128] Alternatively, the first and second selection criteria can be used simultaneously and in a complementary manner. For example, when the number of frequency channels f θ , whose trend indicator I trend ( f θ ) is greater than the trend threshold, is less than P, then only these frequency channels f θ , whose trend indicator I trend ( f θ ) is greater than the trend threshold, are trend frequency channels f t . Conversely, when the number of frequency channels f θ , whose trend indicator I trend ( f θ ) is greater than the trend threshold, is greater than P, then only the P frequency channels f θ , whose trend indicators I treng ( f θ ) are the largest, are trend frequency channels f t .

[0129] In addition, the 180 calculation of the trend indicator I trend ( f θ ) can be broken down into several sub-steps.

[0130] During a third transformation sub-step 182, the calculator 5 transforms each series of amplitudes X fθ into a trend series using a transformation function Φ according to the relationship g fθ = Φ ( X fθ ) . The transformation function Φ aims to linearize as much as possible the amplitude values ​​of each series of amplitudes X fθ .which exhibit a growth trend consistent with the default model. The transformation function Φ may be different depending on the defect sought. The transformation function Φ can for example be the unit function if the defect has a substantially “linear” growth or the logarithm function if the defect has a substantially “exponential” growth.

[0131] During a construction sub-step 184, the calculator 5 constructs a trend line for each trend series by linear regression, according to the relationship h fθ ( r ) = A fθ × r + B fθ , with " r ", a variable, « h fθ ( r ) », a trend line, « A fθ ", a slope coefficient of the trend line, « B fθ ", an ordinate at the origin of the trend line.

[0132] During an estimation sub-step 186, the calculator 5 estimates a noise coefficient σ r for each trend line h fθ (r), according to the relation σ r f θ = 1 M ∑ m = 1 M g f θ − A f θ × r + B f θ 2 1 / 2 , with " Σ m = 1 M », the sum function with m varying from 1 to M, « | | », the absolute value function.

[0133] During a calculation sub-step 188, the calculator 5 calculates the trend indicator I trend depending on the slope coefficient A fθ from the trend line h fθ ( r ) and the noise coefficient relative to the trend line for each frequency channel f θ , according to the relationship I trend f θ = A f θ α + β × σ r f θ .

[0134] The parameters α and β are predetermined parameters and allow the sensitivity to noise to be weighted. α allows in particular the establishment of an upper limit of the trend indicator I trend in order to limit its divergence when the noise σ r ( f θ ) tends towards 0, while β allows to adjust its sensitivity to noise.

[0135] The parameters α and β can be specific to each defect sought, but common to all frequency channels.

[0136] The main step of analyzing 200 trend frequency channels f t includes the following sub-steps.

[0137] During a comparison step 210, the calculator 5 compares each trend frequency channel f t with linear combinations of fault frequencies f d relating to the fault sought. The fault frequencies f d are previously known and contained in a fault model associated with the fault sought. Indeed, most mechanical faults generate on a mechanical system 10 modulations of amplitudes and frequencies of the vibrations of this mechanical system 10. Consequently, for a given fault present on the mechanical system 10, specific frequencies, called “fault frequencies f d » can be excited during the operation of the mechanical system 10, namely present energy levels which follow a growth “law” consistent with the desired defect. In addition, these defect frequencies f d vary little depending on the operating parameters of this mechanical system 10, these parameters having an influence mainly on the amplitude of the vibration signal for these fault frequencies f d .

[0138] Linear combinations of fault frequencies f d respectively comprise sums of several, or even all, of the fault frequencies f d multiplied by coefficients n i integers. The values ​​of the coefficients n i are bounded and the number of coefficients n i integers is limited. The coefficients n i are for example between -4 and 4. Consequently, the number of these linear combinations of fault frequencies f d is also limited.

[0139] These linear combinations of fault frequencies f d are such that: n 1 F 1 − ⋯ n i F i ⋯ − n k F k N 1 min < n 1 < N 1 max ⋮ N k min < n k < N k max n 1 ⋯ n i ⋯ n k ∈ ℤ with k, the number of fault frequencies f d , F 1 , ... , F i , ... , F k , THE k fault frequencies f d with 1 ≤ i ≤ k, ni , a coefficient of the linear combination associated with the frequency F i N i min , a lower limit of ni , N i max , an upper limit of ni .

[0140] Then, during a second determination step 220, the calculator 5 determines fault frequency channels f x , based on the comparison of trend frequency channels f t with linear combinations of fault frequencies f d . A fault frequency channel f x is thus determined equal to a trend frequency channel f t whose differences with linear combinations are lower than a comparison threshold. The comparison threshold is low and for example equal to a frequency resolution associated with discretized frequency vibration signals.

[0141] This second determination amounts to searching among the trend frequency channels f t those allowing to minimize the differences of each trend frequency channel f t with these linear combinations of fault frequencies f d . Following these comparisons, the trend frequency channels are selected. f t for which the result of these differences is a zero or near-zero frequency, either lower than the comparison threshold, as being the default frequency channels f x .

[0142] This second main analysis step 200 therefore makes it possible to identify fault frequency channels. f x , for which their energy levels have an increasing trend and which correspond to frequencies characteristic of the defect sought.

[0143] During the main calculation step 300, the presence indicator I optimal of the fault on the mechanical system 10 is calculated.

[0144] For example, the presence indicator I optimal can be set to increase as the number of fault frequency channels f x increases. The fault presence alert is then triggered if the presence indicator I optimal is greater than the threshold for the presence of a defect.

[0145] For example, the presence indicator I optimal of the fault can be equal to a ratio of a number of the fault frequency channels f x by a total number of linear combinations of the fault frequencies f d . The fault presence alert is then triggered if the presence indicator I optimal is greater than the threshold for the presence of a defect.

[0146] So, the presence indicator I optimal is standardized, namely between 0 and 1. In this way, a threshold for the presence of a single fault can be applicable to the detection of several different types of fault likely to appear on the mechanical system 10.

[0147] In addition, the presence indicator I optimal is zero when none of the fault frequencies f d of the fault model does not express an increasing trend and has therefore not been identified among the trend frequency channels f t . Conversely, the presence indicator I optimal is equal to 1 when all fault frequencies f d of the default model express an increasing trend.

[0148] Furthermore, the greater the number of fault frequency channels f x is important, the higher the presence indicator I optimal is important. Thus, the more the presence indicator I optimal is important, the greater the risk of the presence of a defect, and the more confident one can be about the presence of the defect.

[0149] The method according to the invention therefore makes it possible to determine a presence indicator I optimal which is reliable, efficient and robust for detecting a fault on the mechanical system 10, from the first signs of the appearance of this fault.

[0150] Naturally, the present invention is subject to numerous variations in its implementation. Although several embodiments have been described, it is understood that it is not conceivable to exhaustively identify all possible embodiments. It is of course possible to replace a means described by an equivalent means without departing from the claims.

Claims

1. Method for detecting a fault in a mechanical system (10), said mechanical system (10) comprising at least one rotating member (15) rotating about an axis (AX) of rotation as well as a vibration sensor (20) emitting a temporal vibration signal x(t), an angular sensor (25) emitting a temporal angular signal θ(t) varying as a function of an angular position of said rotating member (15) about said axis (AX) of rotation and a calculator (5), said method including: - processing (100) temporal vibration signals xm(t) emitted by said vibration sensor (20) including the following sub-steps: o said vibration sensor (20) transmitting M temporal vibration signals xm(t), Mbeing greater than or equal to 2, m being a positive integer varying from 1 to M, "t" being the operating time of said mechanical system (10), o first transformation (140), with said calculator (5), of said M temporal vibration signals xm(t) into M angular vibration signals χm(θ) as a function of said temporal angular signal θm(t), "θ" being an angle of said rotating member (15) about said axis (AX) of rotation, second transformation (150), with said calculator (5), of said M angular vibration signals xm(θ) into M frequency vibration signals xm(fθ) by a frequency operator ϑ, where "fθ" is a frequency channel, characterized in that said processing (100) of temporal vibration signals xm(t) includes the following sub-step: o first determination (160) of trend frequency channels ft as a function of said M frequency vibration signals χm(fθ), a trend frequency channel ft being a frequency channel fθ for which said M frequency vibration signals xm(fθ) satisfy a selection criterion relating to a growth trend, and that said method includes the following steps: - analysis (200), with said calculator (5), of said trend frequency channels ft including the following sub-steps: o comparing (210) each trend frequency channel ft with linear combinations of fault frequencies fd relating to said sought fault, said linear combinations respectively comprising sums of a plurality of said fault frequencies fd multiplied by integer coefficients Ni, o second determination (220) of fault frequency channels fx, as a function of said comparison, a fault frequency channel fx being a trend frequency channel ft whose differences with said linear combinations are less than a comparison threshold, - calculating (300), with said calculator (5), an indicator of the presence Ioptimal of said fault on the mechanical system (10) as a function of the number of said fault frequency channels fx and the number of said linear combinations of said fault frequencies fd, - triggering (400) an alert of the presence of a fault as a function of a comparison between said indicator of the presence Ioptimal of said fault and a threshold of the presence of a fault.

2. Method according to Claim 1, for which said first determination (160) of the trend frequency channels ft includes the following sub-steps: - generating (170), with said calculator (5), at least one series of amplitudes Xfθ associated with a frequency channel fθ and comprising M amplitude values belonging respectively to said M frequency vibration signals and relating to said frequency channel fθ, the number of said series of amplitudes Xfθ being equal to the number of frequency channels fθ, - calculating (180), with said calculator (5), a trend indicator Itrend(fθ) for each frequency channelfθ as a function of said series of amplitudes associated with said frequency channel fθ, and - first determination (190), with said calculator (5), of trend frequency channels ft, each trend frequency channel ft being a frequency channel fθ satisfying said selection criterion, said selection criterion relating to said trend indicator Itrend(fθ).

3. Method according to Claim 2, for which according to said selection criterion, each trend frequency channel ft is a frequency channel fθ for which said trend indicator Itrend(fθ) is greater than said trend threshold.

4. Method according to any of Claims 2 - 3, for which according to said selection criterion, said trend frequency channels ft are the P frequency channels fθ corresponding to said largest P trend indicators Itrend(fθ), P being an integer greater than one.

5. Method according to any of Claims 2 to 4, for which each amplitude value of said at least one series of amplitudes X{fθ is associated with a moment tm of a start of emission of said temporal vibration signal xm(t) from which said amplitude value is derived.

6. Method according to one of Claims 2 to 5, for which said calculation (180) of a trend indicator Itrend(fθ) includes the following sub-steps: - third transformation (182) of said at least one series of amplitudes Xfθ by a transformation function Φ into at least one series of trends according to the relationship gfθ =Φ(Xfθ), - constructing (184) at least one trendline from said at least one series of trends by linear regression, according to the relationship hfθ(r) = Afθ × r + Bfθ, where "r" is a variable, "hfθ(r)" is a trendline, "Afθ" is a guiding coefficient of said trendline, "Bfθ" is an ordinate at the origin of said trendline, - estimating (186) a noise coefficient σr for each trendline hfθ(r), according to the relationship σr(fθ) = 1 M ∑ m = 1 M g f θ − (Afθ×r + Bfθ))2|1 / 2, where " Σ m = 1 M " is the sum function with m varying from 1 to M, "| |" is the absolute value function, - calculating (188) said trend indicator Itrend according to said noise coefficient relating to said trendline for each frequency channel fθ, according to the relationship I trend f θ = A f θ α + β × σ r f θ , where "α" and "β" are predetermined parameters.

7. Method according to Claim 6, for which said transformation function Φ is the unit function or the logarithm function.

8. Method according to one of Claims 7 or 8, for which said parameters α and β are specific to each fault sought.

9. Method according to one of Claims 1 to 9, for which said indicator of presence Ioptimal of said fault increases when said number of said fault frequency channels fx increases and said fault presence alert is triggered if said indicator of presence Ioptimal is greater than said threshold of the presence of a fault.

10. Method according to one of Claims 1 to 9, for which a number of said integer coefficients Ni is limited, and values of said integer coefficients Ni are bounded.

11. Method according to Claim 10, for which said indicator of presence Ioptimal of said fault is equal to a ratio of a number of said fault frequency channels fx to a total number of linear combinations of said fault frequencies fd and said fault presence alert is triggered if said indicator of presence Ioptimal is greater than said threshold of the presence of a fault.

12. Method according to one of Claims 1 to 11, for which said comparison threshold is equal to a frequency resolution associated with said discretized frequency vibration signals.

13. Method according to one of Claims 1 to 12, for which said at least one vibration sensor (20) emits said M temporal vibration signals xm(t) during a predetermined sliding period of operation of said mechanical system (10).

14. Calculator program comprising instructions which, when said program is executed, lead to implementing the method according to one of Claims 1 to 13.

15. Monitoring device (9) configured to monitor a mechanical system (10), said mechanical system (10) comprising at least one rotating member (15) rotating about an axis (AX) of rotation, said monitoring device (9) comprising at least one vibration sensor (20) emitting a temporal vibration signal, an angular sensor (25) emitting a temporal angular signal varying as a function of an angular position of said rotating member (15) about said axis (AX) of rotation and a calculator (5), characterized in that said monitoring device (9) is configured to implement the method according to one of Claims 1 to 13.

16. Mechanical system (10) comprising at least one rotating member (15) rotating around an axis (AX) of rotation, characterized in that said mechanical system (10) has a monitoring device (9) according to Claim 15.

17. Power transmission box (6) comprising a mechanical system (10) according to Claim 16.