METHOD FOR MONITORING A ROTATING MACHINE

Abstract

A method (100) for monitoring a rotating machine comprising one or more rotating elements, each rotating element being an elementary source of vibration in the rotating machine, the monitoring method is implemented by a monitoring device comprising electronic circuitry adapted to implement the monitoring method.The monitoring procedure comprises at least the following steps: (Step 1) acquire kinematic data, velocity data, and a vibration signal of the rotating machine in steady state; (Step 2) from the acquired kinematic data, construct a dictionary representing the elementary sources of the rotating machine; (Step 3) model the operation of the rotating machine based on the acquired velocity data, the acquired vibration signal, and the dictionary; (Step 4) use the model to detect a malfunction of the rotating machine using a source separation method. Figure to be published with the abstract: Fig. 1.

Description

Title of the invention: METHOD FOR MONITORING A ROTATING MACHINE technical field

[0001] The invention relates to the monitoring of vibration signals from rotating machines. STATE OF PRIOR ART

[0002] Rotating machines are mechanical devices that use rotational motion to transmit power or support rotational motion between other machines. These machines have a wide range of applications, playing a vital role in sectors such as power transmission, transportation, energy production, and manufacturing. They are indispensable components in various industries, including power plants, trains, and automobiles. Gears and bearings, for example, are rotating machines responsible for transmitting power between two axes and are found in motors, transmissions, and manufacturing equipment. Electric motors, turbines, and compressors are also examples of rotating machines that convert energy into mechanical energy.

[0003] Rotating machines generate vibrations due to the forces and moments acting upon them. These vibrations result from static and dynamic imbalance, occurring when the center of mass is not aligned with the center of rotation and rotates around it. Defects such as cracks or manufacturing errors can also cause vibrations. These vibrations can be measured using various sensors, such as vibration sensors (accelerometers, rotational speed sensors, displacement sensors). Other sensors, such as acoustic sensors (microphones) or electromagnetic sensors (current or piezoelectric sensors), can also be used.

[0004] Vibration modeling of rotating machinery is crucial for several reasons. It allows for the design of more reliable and durable machines, the optimization of their performance, the securing of their operations, the monitoring of their condition to identify potential defects, the planning of maintenance operations, and the reduction of costs. For example, in the aeronautical sector, vibration modeling is essential for monitoring engines, turbines, compressors, gears, and bearings, thus contributing to the early identification of potential defects and the prevention of breakdowns. Indeed, by modeling vibrations, it is possible to understand the vibration signature of the machine, and therefore characterize the The contribution of each rotating component of the machine allows not only for the detection but also the localization of anomalies. In the context of industrial production, vibration modeling is used to monitor machine tools and other equipment, thereby identifying potential defects and optimizing overall performance.

[0005] Known vibration models are generally based on predetermined linear or nonlinear mixing models designed to explain the spectral components of the signals captured by sensors. These vibration models offer a way to understand the operation of rotating machines and to monitor faults by separating the contributions of the system's elementary signals. Currently, the most widespread method for determining these models is empirical, involving a spectral analysis of the rotating machine in steady-state operation under constant conditions. An automated approach for identifying these modulation patterns in rotating machine signals has not yet been proposed, and no model exists that is suitable for the complex architectures of these machines.

[0006] In this context, it is necessary to provide a method for monitoring a rotating machine which allows for the automatic modeling of the vibrations of a rotating machine in order to monitor the occurrence of defects. Description of the invention

[0007] To this end, according to a first aspect, a method for monitoring a rotating machine comprising one or more rotating elements is proposed, each rotating element being an elementary source of vibration in the rotating machine. The monitoring method is implemented by a monitoring device comprising electronic circuitry adapted to implement the monitoring method. The monitoring method comprises at least the following steps: - (Step 1) acquire kinematic data, velocity data and a vibration signal of the rotating machine in steady state; - (Step 2) from the acquired kinematic data, construct a dictionary representing elementary sources of the rotating machine; - (Step 3) model the operation of the rotating machine based on the acquired speed data, the acquired vibration signal and the dictionary; - (Step 4) use the model to detect a malfunction of the rotating machine by a source separation method.

[0008] According to a particular arrangement, the vibration signal acquisition step comprises acquiring the vibration signal using an acquisition system comprising at least one of the following: a sensor, a conditioner, an analog filter anti-aliasing, a sample hold, and an analog-to-digital converter.

[0009] According to a particular arrangement, the acquisition step includes acquiring a digital vibration signal s(t) obtained over a predefined duration T = N.ôh with λ = 1 / Fs the sampling period and N is the digital length of the signal.

[0010] According to a particular provision, the speed data acquisition step includes acquiring a speed profile of the rotating machine and verifying that the standard deviation of the speed profile is less than or equal to 5% of an average value of the standard deviation of the speed profile.

[0011] According to a particular arrangement, the acquired vibration signal is processed to mitigate an effect of a transfer function of the rotating machine.

[0012] According to a particular arrangement, angular resampling is applied to the vibration signal as a function of the velocity data acquired to obtain an angular signal.

[0013] According to a particular arrangement, the angular signal is normalized by its energy according to ~ s(0) with# an angular variable and S (0) the angular signal. slS

[0014] According to a particular arrangement, the dictionary construction step includes determining elementary vibration sources of the rotating machine, each determined elementary vibration source being associated with a fundamental frequency determined from the acquired kinematic data.

[0015] According to a particular arrangement, the dictionary construction step includes decomposing each elementary source into a Fourier series according to s£( 0) = (^Lk)tin() + Bikcos (2,t kofiY with Umax is an overestimated number of harmonics, and B^ are the Fourier coefficients to be estimated with the phases

[0016] According to a particular arrangement, the dictionary is determined according to K, where K is the total number of elementary sources and with Df,:, k) =[ sini 2jT kofi), cos (l / rkofi). , sin(27T Hlnax ofi), cos{2îi H,!Utxofi) ]T for all 6

[0017] According to a particular arrangement, the modeling step includes determining a mixing function F on the elementary sources ^(#) with an additive noise £ defined as follows: s = F(515..., sK) + £.

[0018] According to a particular arrangement, the mixing function F is approximated with a mathematical function that minimizes the average error E 1^(^, sK) - s |A2 with s' the sources estimated based on 10.

[0019] According to a particular provision, the method includes a step 3.5 for calibrating the model of step 3, the model calibration step 3.5 comprising determining instantaneous frequencies of each source according to J i ' J tree

[0020] According to another aspect, a rotating machine monitoring device is proposed, characterized in that it comprises electronic circuitry for implementing a rotating machine monitoring method which includes at least the following steps: - (Step 1) acquire kinematic data, velocity data and a vibration signal of the rotating machine in steady state; - (Step 2) from the acquired kinematic data, construct a dictionary representing elementary sources of the rotating machine; - (Step 3) model the operation of the rotating machine based on the acquired speed data, the acquired vibration signal and the dictionary; - (Step 4) use the model to detect a malfunction of the rotating machine by a source separation method.

[0021] According to another aspect, a computer program product is proposed comprising program code instructions for executing the process according to the invention.

[0022] According to another aspect, a non-transient storage medium is proposed on which is stored a computer program comprising program code instructions to execute the process according to the invention, when said instructions are read from said non-transient storage medium and executed by a processor. Brief description of the drawings

[0023] The features of the invention mentioned above, as well as others, will become clearer upon reading the following description of at least one exemplary embodiment, said description being made in relation to the accompanying drawings, among which:

[0024] [Fig.1] schematically illustrates the process of a monitoring procedure;

[0025] [Fig.2] schematically illustrates an architecture of an EQL network;

[0026] [Fig.3] schematically illustrates the course of a training procedure;

[0027] [Fig.4] schematically illustrates a computer system adapted to implement the process.

[0028] DETAILED DESCRIPTION OF IMPROVEMENTS

[0029] Monitoring method

[0030] With reference to [Fig. 1], according to a first aspect, a method 100 is proposed for monitoring a rotating machine comprising one or more rotating elements, each rotating element being an elementary source of vibration in the rotating machine. The monitoring method is implemented by a monitoring device comprising electronic circuitry adapted to implement the monitoring method.

[0031] The monitoring method comprises at least the following steps: - (Step 1) acquire kinematic data, velocity data and a vibration signal of the rotating machine in steady state; - (Step 2) from the acquired kinematic data, construct a dictionary representing elementary sources of the rotating machine; - (Step 3) model the operation of the rotating machine based on the acquired speed data, the acquired vibration signal and the dictionary; - (Step 4) use the model to detect a malfunction of the rotating machine by a source separation method.

[0032] Step 1 - Acquisition

[0033] According to a particular arrangement, step 1 of the vibration signal acquisition comprises acquiring the vibration signal using an acquisition system including at least one of the following: a sensor, a signal conditioner, an analog anti-aliasing filter, a sample-and-hold circuit, and an analog-to-digital converter. Preferably, the acquisition system includes all of the following elements: a sensor, a signal conditioner, an analog anti-aliasing filter, a sample-and-hold circuit, and an analog-to-digital converter.

[0034] At the output of the acquisition system, a digital vibration signal is obtained over a predefined duration T = N.Δh, where Δt - 1 / Fs is the sampling period and N is the digital length of the signal. During this period, the rotating machine preferentially operates in steady state, meaning that the rotating machine has a nearly constant speed and load. This steady-state operating condition is considered to be met if the standard deviation of the speed profile does not exceed 5% of its average value.

[0035] <O.O5E[X01-

[0036] With^[.] the empirical mean.

[0037] According to a particular arrangement, the acquired vibration signal is then processed to attenuate the effect of a transfer function of the sensor that acquired the vibration signal. One way of doing this is to whiten the signal to remove the effect of the function Transfer of the rotating machine. This step allows us to obtain a model of the rotating machine that is not affected by the sensor used and that will be adapted for future acquisitions and new sensors.

[0038] According to a particular arrangement, angular resampling is applied to the vibration signal as a function of the velocity data acquired to obtain an angular signal.

[0039] According to a particular arrangement, the speed data can be acquired from a tachometer signal or can be estimated with an algorithm from the vibration signal, acquired simultaneously with s(t), thus allowing an approximation of the speed and the transformation of the vibration signal in the angular domain (s(0) where G is the angular variable. The length of this signal corresponds to the total number of cycles of the rotating machine.

[0040] The angular signal is then normalized by its energy according to ~ ü. Xe) with# an angular variable and s(6) the angular signal.

[0041] Notion of order

[0042] For the remainder of this document, it is necessary to clarify the concept of rotation order in a gear system. A rotating machine can contain one or more rotating parts. By exploiting the kinematics of the rotating machine, it is possible to determine the rotation orders of each rotating element. These orders are frequencies normalized with respect to a reference rotating element.

[0043] By way of example, for a single rotating element, there is only one first-order component. In contrast, a gear set can contain several rotating elements (wheels) of different diameters that mesh together to transmit rotational motion according to a reduction ratio. In this case, the vibration signal is represented by multiple modulations between the meshing forces generated during contact between the teeth and the modulating signals caused by the wheels.

[0044] If we consider a simple example where two gears 1 and 2 are in contact, the order of the meshing force °e is related to the rotation orders of the gears and as follows: oe = Zlol = Z2o2. Where Zf and are the number of teeth of the two gears 1 and 2.

[0045] For a planetary gear system, there are two or more planets carried by a planet carrier meshed on one side with a central sun gear and on the other side with a ring gear. The gear configuration (locked wheel or moving wheel) depends on its operation. In the case of a reduction gear, the input shaft is connected to the sun gear, and the output shaft is connected to the planet carrier.

[0046] If we consider a case in which the crown is blocked, and that Z2, Z3 are the number of teeth of the crown, the solar gear and the satellites respectively and °s is the order of the solar gear, then the orders of the different rotating elements are as follows: - Order of the solar gear: °s - Order of the satellite carrier: „ _ z: .. Ops - z2+z, os - Considered engagement frequency: „ „ 7 _ ziz? °e - \Ups - "s

[0047] Step 2 - Dictionary construction

[0048] The dictionary construction step includes determining elementary vibration sources of the rotating machine, each determined elementary vibration source being associated with a fundamental frequency determined from the acquired kinematic data.

[0049] According to a particular arrangement, the dictionary is constructed by determining the characteristic commands of the rotating machine, then these characteristic commands are used to define the different elements of the dictionary.

[0050] Advantageously, the elementary vibration sources, denoted by , of the rotating machine are considered to be periodic, with their fundamental frequencies determined from the kinematic model of the rotating machine. Each elementary source is then represented using a Fourier series decomposition as follows: %(0) = EfjA^cos()sin(2tt k ) + Bikcos (2tt k ) ' with ^max an overestimated number of harmonics, and B^ the Fourier coefficients to be estimated with the phases

[0051] The dictionary is determined according to jy — with K the number of sources total elementary and with D^:, k) =[ sin( 2rt kofi), cos (27rkofi). , .wn(2zr ofi), cos(2tt H^ofi)]1 for ALL 6.

[0052] Then the sources M0) are each a linear combination of Dp

[0053] Step 3 - Modeling

[0054] The modeling step includes determining a mixing function F on the elementary sources with additive noise £ defined as follows: s = ..., sK) + £

[0055] Thus, in other words, instead of looking for a black box F, modeling step 3 involves finding a mathematical expression for F that can then be used for monitoring the rotating machine in question. A mathematical expression here means a symbolic expression in the sense that It is expressed in terms of mathematical operations such as addition and / or multiplication between different sources.

[0056] One of the main advantages of using a mathematical expression is the ability to provide symbolic expressions that are often more easily interpreted than purely statistical, black-box models. This allows for an understanding of the operation of the rotating machine and the various complex interactions between its different rotating components. Furthermore, it allows for the automatic discovery of the form of the relationship between the sources, e.g., whether linear or non-linear. Finally, the mathematical expression is capable of exploring a wider range of functional forms, which can be particularly useful when the relationship between the sources is inherently complex.

[0057] According to a particular arrangement, the mixing function F is approximated with a mathematical function that minimizes the average error E -, sK ) -s |A2 with % the sources estimated based on Dj.

[0058] Finding the mathematical function F can then be viewed as a symbolic regression problem. According to the embodiment presented here, finding the mathematical function F is carried out using a neural network. The function V' is represented by a neural network architecture as defined below.

[0059] Example of implementation with a neural network

[0060] According to the example presented here, for function V7, it is proposed to use an Equation Learner (EQL) neural network architecture constructed from linear layers with unusual activation functions, unary or binary mathematical operators, for example: the "identity", "multiplication", and "addition" operators. This architecture is primarily dedicated to linear regression based on neural networks. If we consider an architecture with two linear layers with "multiplication" and "identity", the set of activation functions (see Fig.2) represent the primitive functions of the final expression. The loss function of the neural network contains two terms, the first term is represented by E |y?(Sj,sK) -s |A2 and the second term is a regularization which penalizes the number of non-zero parameters X = {W : weight matrices, b : bias matrices], defined as follows: . J (X) = IhjD) - S |A 2 + AR (x)

[0061] The term parsimony is essential in — —L, J / )Lç Ia2 + AR (x) For find an interpretable and compact expression for the function V that represents the multi-modulation model. The goal of this term is to make, as much as possible, nuis network settings. That is, to disable the settings as much as possible.

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] in the neural network. An example is to use Where L is the number of layers used in the network and 0 < q < 1. We can also use smooth relaxations of standards such as • j. Iwl2 w> a ^0.5<» = / 4 . . . U / h4, 3^, 3a \ 2 V ■ W + 4a + 8 ) w < a which approximates IJ0-''. An architectural improvement is also suggested to systematically group and access the coefficients of the estimated sources, particularly where the number of overestimated harmonics is high. The network is divided into two distinct components: the first dedicated to source estimation, followed by the second to determining the mathematical expression representing the data. The number of layers in the EQL network is a hyperparameter that depends on the complexity of the vibration model. Therefore, when the number of layers is overestimated, it is advisable to use the residual connection technique, which involves concatenating the output of each symbolic layer (hl, h2) to the next, as illustrated in [Fig. 3]. This forces the network to obtain more compact expressions, which also avoids problems of gradient explosion or vanishing, since the gradient will propagate more easily through the network. The network is trained using a multi-phase procedure. The first phase involves training without regularization to find a reasonably good initialization point for the network. Next, the objective function is regularized. Then, thresholding is applied to the neural network parameter, eliminating low coefficients from the final expression derived from the architecture. The network is then retrained to refine the system parameter estimation in the third phase. The network's hyperparameters are: F, a (regularization parameter), 2, 1e (the number of layers used in the architecture, which depends on the complexity of the gear model), and the number of activation functions in the symbolic layers. Step 3.5 - Model calibration The process then includes a step 3.5 for recalibrating the model from step 3. Step 3.5 for recalibrating the model involves determining instantaneous frequencies of each source according to f. = ()il'faritre

[0071] In other words, step 3.5 of model calibration takes as input a vibration signal that may or may not be the same as the vibration signal used to estimate the rotating machine model. For example, while we need to use a steady-state signal for model determination, the vibration signal used for calculating indicators can be acquired in steady-state or variable-state conditions. We also need a tachometer that allows us to measure the instantaneous frequency of the input shaft f. Knowing the characteristic orders of the different sources, the instantaneous frequencies of each source are determined as follows:

[0072]

[0073] If the vibration signal used for calibration is the same as that used for modeling, the calibrated model is the same as the modeled model if the vibration signal is corrected with respect to its transfer function. Otherwise, a calibration method for the model determined during modeling is applied, which amounts to estimating the constants involved in the model.

[0074] When the phase is stationary, a simple approach will be, for example, to record the amplitudes of the peaks of the spectrum corresponding to each source or combination of sources.

[0075] In the case of a non-stationary phase, the recalibration is done by sliding window assuming that the operating conditions evolve slowly over time, which allows us to assume that the signal is stationary over the short analysis window.

[0076] Step 4 - Using the model

[0077] The method then involves using the model to detect a malfunction of the rotating machine. This is achieved by monitoring the evolution of the estimates of the elementary sources related to the rotating elements of the system. According to a particular arrangement, step 4 can be carried out by applying estimation methods based on the already determined data model, such as the method described in US patent 0232880, which allows the contributions of the different elements to be separated, or another method for separating and differentiating the elementary sources constituting the vibration signal of the system. The amplification of the estimates of the elementary sources over time provides information on the health of the corresponding elements.

[0078] Step 5 - Repair

[0079] According to a particular provision, the process may then include repairing the detected malfunction using a third-party device.

[0080] Computer program product

[0081] According to another aspect, a computer program product is proposed comprising program code instructions for executing the monitoring process.

[0082] Storage medium

[0083] According to another aspect, a non-transient storage medium is proposed on which is stored a computer program comprising program code instructions to execute the monitoring process 100, when said instructions are read from said non-transient storage medium and executed by a processor.

[0084] Monitoring device

[0085] According to another aspect, a monitoring device is proposed comprising electronic circuitry (computer system 200) adapted to implement a process 100.

[0086] As schematically shown in [Fig.4], the computer system 200 may include, connected by a communication bus 210: a processor 201; a random access memory 202; a read-only memory 203, for example of type ROM (“Read Only Memory”) or EEPROM (“Electrically-Erasable Programmable Read Only Memory”); a storage unit 204, such as a hard disk drive (HDD) or a storage media reader, such as an SD card reader (“Secure Digital”); and an input / output interface manager 205.

[0087] The processor 201 is capable of executing instructions loaded into RAM 202 from ROM 203, external memory, a storage medium (such as an SD card), or a communication network. When the computer system 200 is powered on, the processor 201 is capable of reading instructions from RAM 202 and executing them. These instructions form a computer program enabling the processor 201 to implement process 100.

[0088] All or part of the process 100 can thus be implemented in software form by executing a set of instructions by a programmable machine, for example a DSP (Digital Signal Processor) or a microcontroller, or be implemented in hardware form by a dedicated machine or component, for example an FPGA (Field Programmable Gate Array) or ASIC (Application-Specific Integrated Circuit). Generally, the computer system 200 includes electronic circuitry adapted and configured to implement, in software and / or hardware form, the process in relation to the computer system 200 in question.

Claims

Demands

1. A method (100) for monitoring a rotating machine comprising one or more rotating elements, each rotating element being an elementary source of vibration in the rotating machine, the monitoring method being implemented by a monitoring device comprising electronic circuitry adapted to implement the monitoring method, the monitoring method being characterized in that it comprises at least the following steps: - (Step 1) acquiring kinematic data, speed data and a vibration signal of the rotating machine in steady state; - (Step 2) from the acquired kinematic data, constructing a dictionary representing elementary sources of the rotating machine; - (Step 3) modeling an operation of the rotating machine as a function of the acquired speed data, the acquired vibration signal and the dictionary;- (Step 4) use the model to detect a malfunction of the rotating machine by a source separation method.;

2. A method according to any one of the preceding claims, wherein the vibration signal acquisition step comprises acquiring the vibration signal using an acquisition system comprising at least one of: a sensor, a conditioner, an analog anti-aliasing filter, a sample-and-hold circuit, and an analog-to-digital converter.

3. Method according to claim 2, wherein the acquisition step comprises acquiring a digital vibration signal S (J) obtained over a predefined duration T = N.dh with 5t = 1 / Fs the sampling period and V is the digital length of the signal.

4. A method according to claim 3, wherein the speed data acquisition step comprises acquiring a speed profile of the rotating machine and verifying that the standard deviation of the speed profile is less than or equal to 5% of an average value of the standard deviation of the speed profile.

5. A method according to claim 4 wherein the acquired vibration signal is processed to attenuate an effect of a transfer function of the rotating machine.

6. A method according to claim 5, wherein angular resampling is applied to the vibration signal as a function of the velocity data acquired to obtain an angular signal.

7. Method according to claim 6, wherein the angular signal is normalized by its energy according to ~ with# an angular variable sW = ^fi and s ( # ) the angular signal.

8. A method according to any one of the preceding claims, wherein the dictionary construction step comprises determining elementary vibration sources of the rotating machine, each determined elementary vibration source being associated with a fundamental frequency determined from the acquired kinematic data.

9. A method according to claim 8, wherein the dictionary construction step comprises decomposing each elementary source into a Fourier series according to -L^A^os^.^sin^k^B) + Bik cos (Inko-fi), with Umax an overestimated harmonic number, and and B^ the Fourier coefficients to be estimated with the phases

10. A method according to claim 9 wherein the dictionary is determined according to jy — ^a] with 'c total number of elementary sources and with d):. k) = [ sin(2n kofi), cosifinkofi), , sin(2tr Hlnux ofi), uisi la U„xofi) ]' for all 0

11. A method according to any one of the preceding claims, wherein the modeling step comprises determining a mixing function F on the elementary sources ^(#) with additive noise £ defined as follows: S = F(Sp sK) + £

12. A method according to claims 10 and 11 in combination, wherein the mixing function F is approximated with a mathematical function V that minimizes the mean error E^(Sj, sK) - s |A2 , with s < the sources estimated based on Dh

13. A method according to any one of the preceding claims, comprising a step 3.5 of recalibrating the model from step 3, the step 3.5 Model calibration including determining instantaneous frequencies of each source according to f=°*fti 1 tree

14. A rotating machine monitoring device characterized in that it comprises electronic circuitry (200) for implementing a method of monitoring a rotating machine which includes at least the following steps: - (Step 1) acquiring kinematic data, speed data and a vibration signal of the rotating machine in steady state; - (Step 2) from the acquired kinematic data, constructing a dictionary representing elementary sources of the rotating machine; - (Step 3) modeling an operation of the rotating machine as a function of the acquired speed data, the acquired vibration signal and the dictionary; - (Step 4) using the model to detect a malfunction of the rotating machine by a source separation method.

15. Product computer program comprising program code instructions to perform the process (100) according to any one of claims 1 to 13.

Images