Method and device for monitoring the wear of a cylinder

Abstract

Method and device for monitoring the wear of a cylinder. Method for evaluating the wear of a cylinder in an aircraft turboshaft engine comprising: A step of obtaining, for at least one flight of an aircraft comprising said cylinder, a value for each input variable of a wear quantification law, and A step of evaluating the amount of wear of said cylinder induced by said at least one flight by providing said values ​​as input to said quantification law, said quantification law being previously determined in particular by means of training a classification model and selecting parameters from among the parameters used to train the classification model. Figure for the abstract: Fig. 6.

Description

Title of the invention: Method and device for monitoring the wear of a cylinder. Technical field

[0001] The general field of application of the invention is that of aeronautics.

[0002] The invention relates more particularly to monitoring and prediction methods, also known as "prognostics and health monitoring methods," for the maintenance of a turboshaft engine, especially the cylinders of a turboshaft engine. The invention also relates to devices for implementing such methods. Previous technique

[0003] A turboshaft engine typically includes various actuator systems such as VSV (Variable Stator Vane) systems for actuation of variable stators in the engine's compressor, VBV (Variable Bleed Valve) systems for purging air from the compressor by actuation of valves, and TBV (Transient Bleed Valve) systems for regulating the air charge in the high-pressure compressor. These systems include at least one actuator whose position influences the operation of the system it comprises. In particular, within a VSV system, the position of a hydraulic actuator allows the stator vanes to be rotated to a chosen angle via mechanical transmission components.Changing the angle allows the compressor's compression ratio to be varied, thus optimizing the efficiency of the high-pressure section of the turbocharger in steady state, but also increasing its operating margins during transient phases.

[0004] Generally, a cylinder consists of a body and a piston that moves relative to the body according to a position defined by a control system. In one stroke, a piston can make several dozen or even hundreds of movements, typically ranging from a few millimeters to several centimeters. Thus, a piston can travel a cumulative distance of up to more than one kilometer per year.

[0005] These regular relative movements can lead to progressive wear of the cylinder.

[0006] Thus, it is necessary to regularly check the condition of the cylinders in order to ensure the proper functioning of the turbomotor.

[0007] Cylinder wear is generally monitored by performing manual inspections following a maintenance procedure provided by the manufacturers. These inspections allow for an assessment of the cylinder condition but also require a Turbocharger immobilization, see dismantling of the cylinders. These inspections are relatively long and costly.

[0008] French patent FR 3 117 457 describes a method for detecting wear on a cylinder by analyzing the effective response time of the cylinder to move in response to a command.

[0009] It is desirable to be able to limit unplanned aircraft downtime. In particular, there is a need for improved monitoring of cylinder wear.

[0010] The invention aims to meet all or part of these needs. Description of the invention

[0011] To this end, the invention proposes a method for determining a wear quantification law for a cylinder in an aircraft turboshaft engine comprising: obtaining, for each flight, a plurality of aircraft flights: - a parameter value for each parameter in a set of parameters; - of a quantity of wear induced by said theft on a cylinder; at least one training of a classification model from the parameter values ​​of at least a part of the parameters of the parameter set, such that said model once trained is capable of determining a cylinder wear state from values ​​for said at least a part of the parameters of the parameter set; the selection of at least one parameter from said at least a part of parameters according to the influence of each parameter on the performance of determining the trained model; the determination of a law for quantifying the wear of a cylinder from the quantities of wear induced and the values ​​of said at least one selected parameter, each selected parameter being an input variable for said law.

[0012] Preferably, the unselected parameters are not input variables for the quantization law.

[0013] The performance of model determination can be determined by any known criterion, in particular by determining a confusion matrix, the recall criterion, the sensitivity criterion, the precision criterion, or the specificity criterion. This list is not exhaustive and other criteria may be used.

[0014] The influence of the parameter on the determination performance corresponds to the impact that the presence, or absence, of said parameter has on the performance. In other words, does the presence, or absence, of the parameter improve the determination performance of the model or, on the contrary, degrade it?

[0015] Correspondingly, the invention also relates to a device for determining a wear quantification law for a cylinder in an aircraft turboshaft engine comprising:

[0016] - a data acquisition module configured to obtain for each flight of a plurality of aircraft flights a parameter value for each parameter of a set of parameters and a quantity of wear induced by said flight on a cylinder, - optionally, a labeling module configured to assign to each parameter value a cylinder wear state; - a training module configured to perform at least one training of a classification model from the parameter values ​​of at least some of the parameters in the parameter set, such that said model once trained is able to determine a cylinder wear state from values ​​for said at least some of the parameters in the parameter set; - a selection module configured to select at least one parameter from said at least a set of parameters based on the influence of each parameter on the determination performance of the trained model; and - a law determination module configured to determine a wear quantification law for a cylinder from the induced wear quantities and the values ​​for at least one selected parameter, each selected parameter being an input variable for said law.

[0017] In general, the invention thus proposes to model the wear caused by a flight on a cylinder by means of parameters selected by training a classification model.

[0018] Such a selection makes it possible to obtain a compact quantification law, comprising a limited but relevant number of input variables.

[0019] Preferably, the method further comprises the configuration of a device for estimating the wear of a cylinder from the quantification law, the estimation device being intended to allow the evaluation of a quantity of wear induced by at least one flight as a function of values ​​for the input variables of said law provided to the estimation device.

[0020] The invention therefore also relates to a wear estimation device configured by implementing a determination method according to the invention. The estimation device thus configured is intended to allow the evaluation of the amount of wear on a cylinder induced by at least one flight as a function of values ​​for the input variables of the quantification law determined by implementing said determination method.

[0021] The quantity of wear induced can be a depth or volume of wear of cylinder induced by at least one flight.

[0022] The said parameter values ​​obtained may be related to contextual, operational and / or structural parameters.

[0023] By "contextual parameters" are meant parameters defining a flight context. A contextual parameter may characterize a route or environmental conditions. For example, a contextual parameter may be: a flight path, a flight duration, an altitude, a temperature, particularly at arrival and / or departure airports, a pollution level, particularly at arrival and / or departure airports, an outside temperature, particularly at arrival and / or departure airports, a distance traveled by the aircraft. The values ​​of these parameters for a given flight can be obtained via the Flight Radar database.

[0024] By "operational parameters" are meant parameters relating to the operation of the aircraft, in particular the turboshaft engine. An operational parameter may include pressure, temperature (for example, fuel temperature or ambient turboshaft engine temperature), engine speed, current, power, and flow rate. The values ​​of these parameters for a given flight may be obtained by means of onboard sensors.

[0025] By "structural parameters" we mean parameters relating to the design of the cylinder, the turboshaft engine and / or the aircraft. The values ​​of these parameters can be obtained from control laws and / or physical models, and from manufacturer documentation.

[0026] Preferably, said parameter values ​​include at least contextual parameter values, and possibly operational ones.

[0027] The invention can thus exploit classically acquired and easily accessible data, making its deployment particularly efficient.

[0028] At least one classification model training is performed from the values ​​of at least a part of the parameters of the parameter set so as to obtain a trained model configured to provide as output a wear state from parameter value(s) supplied as input for at least a part of the parameters.

[0029] At the end of at least one training session, at least one less trained model is obtained.

[0030] In particular embodiments of the invention, several training sessions of a classification model are performed using the values ​​of at least a portion of the parameters in the parameter set, said at least a portion of parameters preferably being different for each of said training sessions.

[0031] A first part of a parameter is different from a second part of a parameter when at least one parameter is present in the first part and absent in the second or when at least one parameter is absent from the first part and present in the second.

[0032] Two different parts may have a different number of parameters.

[0033] The classification model can be a supervised or unsupervised learning model.

[0034] In particular embodiments of the invention, the model is a supervised learning model.

[0035] Any classification model may be used. In particular, the model may be a decision tree forest model, a gradient boosting model such as XGBoost (Extreme Gradient Boosting), a support vector model (SVM), a principal component analysis (PCA) model, a singular value decomposition (SVD) model, or a hierarchical clustering model. These examples are neither limiting nor exhaustive.

[0036] In particular embodiments, the classification model is a supervised learning model trained from parameter values ​​that have been previously labeled.

[0037] In particular, a wear status can be assigned to each flight and therefore to each parameter value relating to that flight. The wear status can be considered a category, or label, for the parameter values ​​of each flight. For example, the wear status is chosen from: "worn cylinder" and "unworn cylinder".

[0038] The state of wear can be indicative of wear experienced by the cylinder compared to a previous state, for example a nominal state. Alternatively, the state of wear can be defined independently of a previous state.

[0039] The wear condition determined for a flight or a parameter value of said flight can be indicative of wear induced by said flight on the cylinder for which the amount of wear is obtained. In other words, for each flight for which a wear condition is obtained, said wear condition can indicate whether said flight caused wear to the cylinder.

[0040] The state of wear can be determined from the amount of wear induced. For example, if the amount of wear is less than a threshold, then the state of wear is defined as "not worn".

[0041] The wear status can be manually associated with each parameter value or with each flight, for example from data entered manually by maintenance operators during maintenance operations.

[0042] Alternatively or additionally, the wear state can be determined automatically. In particular, once the classification model has been trained, it can be advantageously used to enrich a database storing parameter values ​​associated with their wear state, for example, to subsequently improve the classification model or to train a new model.

[0043] From at least one trained model, at least one parameter is selected from the set of parameters according to the influence that at least one parameter has on the performance of determining a state of wear by said at least one trained model.

[0044] Many selection methods, well known to those skilled in the art, can be implemented.

[0045] In particular, the selection may be based on: - a "Wrapper" type method, in particular a forward selection method, a backward selection method, an exhaustive feature selection method, or a recursive feature selection method, - an integrated selection method also called an "embedded-based method" in English; - a principal component analysis, commonly called PCA (Principal Component Analysis); - an independent component analysis, commonly called ICA (Independent Component Analysis); - a t-SNE (t-distributed stochastic neighbor embedding) algorithm.

[0046] This list is neither exhaustive nor limiting.

[0047] The wear quantification law can be determined in such a way as to allow the evaluation of a wear depth, in other words, the depth of material that has been removed, for example, from the cylinder body. Thus, the quantification law can allow the evaluation of the depth of material removal at the bore guiding the piston due to the operation of the cylinder.

[0048] In alternative embodiments, the wear quantification law is determined so as to allow evaluation of a wear volume.

[0049] The determination of the wear quantification law may involve the determination of a linear wear function having as input variable at least one first selected parameter and possibly a wear amplification function.

[0050] The amplification function can vary according to one or more distinct selected second parameters and preferably not correlated to at least one selected first parameter.

[0051] The wear quantification law can take the form: P = f(yi, ..., yk) = ayi + g(y2, ..., yk), with a a linear wear coefficient, g the amplification function and (yb ..., yk) the selected parameters, of which there are k+1, with k>0, yi being said to be at least a first selected parameter.

[0052] The function g can depend on one or more selected parameters (y2, .. .yk). Alternatively, the function g is a constant value.

[0053] Preferably, when the function g is a constant value, the quantization law has as its input variable only the at least one first selected parameter.

[0054] The wear amplification function can be determined from a statistical analysis of one or more second parameters selected according to the wear state of the cylinder.

[0055] The amplification function can be determined by implementing the following steps: for each selected parameter other than the aforementioned at least one first parameter: - Statistical analysis of the values ​​taken for said parameter as a function of the quantities or states of wear in order to determine a correlation between the values ​​of said parameter and the quantity or state of wear, - For each parameter value or range of values ​​for said selected parameter, determination of an amplification coefficient from the statistical analysis.

[0056] The statistical analysis for a selected parameter may include generating a histogram for each wear state or group of wear quantities, each bar of the histogram representing a predefined range of parameter values. For each range of parameter values, an amplification coefficient may be determined by comparing the bars of the histograms generated for said selected parameter corresponding to said range of parameter values.

[0057] The linear wear coefficient can be a scalar coefficient.

[0058] In particular embodiments of the invention, the selection of at least one parameter further includes the selection of at least one flight time, possibly selected independently of the trained model.

[0059] The at least one first selected parameter may include at least one flight duration.

[0060] In particular embodiments, a quantification law is determined for a specific type of cylinder, the resulting wear quantities then being acquired for cylinders of that particular cylinder type. For example, the quantification law can model the wear of VSV cylinders, each resulting wear quantity being a wear quantity of a VSV cylinder.

[0061] The determination method can be used to determine a specific quantification law modeling the wear of VSV cylinders, and / or a specific quantification law modeling the wear of VBV cylinders and / or a specific modeling law modeling the wear of TBV cylinders.

[0062] Similarly, the quantification law may be specific to a particular type of turboshaft engine and possibly a particular type of cylinder.

[0063] Thus, the quantification law can be specific to a type of turbomotor and / or a type of cylinder.

[0064] Equivalently, the quantification law can be determined for a particular type of aircraft, and possibly a particular type of turboshaft engine and / or a particular type of cylinder.

[0065] In particular embodiments of the invention, the quantification law can be determined for any type of cylinder, for any type of turboshaft engine and for any type of aircraft.

[0066] The quantification law determined by means of a determination method according to the invention can advantageously be used to evaluate a quantity of wear of a cylinder in an aircraft turboshaft engine induced by at least one flight, in particular, using an estimation device according to the invention.

[0067] Thus, the invention also relates to a method for evaluating the wear of a cylinder in an aircraft turboshaft engine comprising: - a step of obtaining, for at least one flight of an aircraft comprising said actuator, a value for each input variable of a wear quantification law, and - an evaluation step, in particular by means of an estimation device, of a quantity of wear of said cylinder induced by said at least one flight by providing said values ​​as input to said quantification law, said quantification law being previously determined by the implementation of a determination method according to the invention, said quantification law preferably being specific to the type of said cylinder and / or to the type of said turboshaft engine comprising said cylinder and / or to the type of said aircraft comprising said cylinder.

[0068] Correspondingly, the invention relates to a device for evaluating the wear of a cylinder according to the invention comprising: - a data acquisition module configured to obtain, for at least one flight of an aircraft equipped with said actuator, a value for each input variable of a wear quantification law; and

[0069] - an estimation device according to the invention configured to evaluate a quantity of wear of said cylinder induced by said at least one flight taking into account said values ​​obtained.

[0070] The invention thus makes it possible to quantify the wear of a cylinder without the need for manual control of the cylinder and therefore without immobilizing the aircraft.

[0071] Unlike the prior art, the invention allows for relatively precise monitoring of the wear status of the cylinders and can advantageously allow for optimization of the maintenance operations. In particular, this helps to reduce the number of unnecessary downtimes but also to reduce the number of unforeseen downtimes.

[0072] The evaluation method may advantageously include the determination of a cumulative wear quantity. By "cumulative wear quantity" is meant a total wear quantity experienced by the cylinder with reference to a previous state of the latter, for example a nominal state.

[0073] In particular, the evaluation method can be implemented for one flight or a plurality of flights. When the parameter values ​​intended to be provided as input variables for the wear quantification law relate to a plurality of flights, the evaluation of the quantity of wear can be carried out by evaluating the quantities of wear induced for each of the flights considered separately and then adding the said quantities of wear induced evaluated for each of the said flights.

[0074] The wear quantification law used may be a quantification law specific to the type of cylinder of said cylinder to be evaluated.

[0075] The wear quantification law used may be a quantification law specific to the type of turboshaft engine and / or aircraft comprising said cylinder to be evaluated.

[0076] Preferably, the wear quantification law used is a quantification law specific to the type of cylinder of said cylinder to be evaluated and to the type of turbomotor comprising said cylinder to be evaluated.

[0077] The evaluation method preferably includes a step of checking the wear of the cylinder by comparing said quantity of induced wear and / or said quantity of cumulative wear with at least one threshold value.

[0078] Several threshold values ​​can be predetermined indicating, for example, whether a maintenance operation should be carried out or not, in particular whether the cylinder needs to be repaired, changed, or requires a visual inspection.

[0079] The evaluation method may include generating an alarm when the amount of induced and / or cumulative wear exceeds said threshold value.

[0080] In particular embodiments of the invention, the input variable values ​​provided as input to the quantification law are estimated or predictive. In particular, the evaluation can predict the amount of wear induced by at least one future flight, possibly several. Alternatively, input variable values ​​can be estimated to fill in missing values ​​for past flights.

[0081] The method then advantageously allows estimating a probable amount of wear of a cylinder and / or anticipating a future amount of wear and thus programming possible maintenance operations accordingly.

[0082] All or part of the steps of the evaluation process can be implemented by computer, preferably all steps of the process are implemented by computer.

[0083] Equivalently, all or parts of the steps of the determination process can be implemented by computer, preferably all the steps of the process are implemented by computer.

[0084] Thus, the invention also relates to a computer program comprising code instructions which, when implemented, allow the execution of the steps of an evaluation process according to the invention, this program being capable of being implemented in a device for evaluating the wear of a cylinder of a turboshaft engine of an aircraft configured to implement the evaluation process according to the invention.

[0085] The invention also relates to a computer program comprising code instructions which, when implemented, allow the execution of the steps for determining a law for quantifying cylinder wear in an aircraft turboshaft engine, this program being capable of being implemented in a device for determining the wear of a cylinder of an aircraft turboshaft engine configured to implement the determination method according to the invention.

[0086] In particular, this program preferably includes code instructions which, when implemented, allow the configuration of an estimation device according to the invention.

[0087] The invention further relates to a recording medium readable by means of a computer comprising a computer program according to the invention.

[0088] The aforementioned features and advantages, as well as others, will become apparent from the detailed description that follows. This detailed description refers to the attached drawings. Brief description of the drawings

[0089] The attached drawings are schematic and are intended primarily to illustrate the principles of the exposition.

[0090] On these drawings, from one figure to another, identical elements (or parts of elements) are identified by the same reference signs.

[0091] [Fig. 1] Fig. 1 represents a functional architecture of a device for determining a wear quantification law for a cylinder according to an example of the invention,

[0092] [Fig. 2] Fig. 2 represents the main steps of a determination process of a quantification law according to an example of the invention that can be implemented by a device as illustrated in [Fig. 1],

[0093] [Fig. 3] Fig. 3 represents a functional architecture of a device for evaluating the wear of a cylinder according to an example of the invention,

[0094] [Fig. 4] Fig. 4 represents the main steps of an evaluation process the wear of a cylinder according to an example of the invention which can be implemented by a device as illustrated in [Fig.3],

[0095] [Fig. 5] Figure 5 represents an example of the implementation of the evaluation process according to the invention,

[0096] [Fig. 6] Figure 6 represents an example of implementation of the process determination according to the invention,

[0097] [Fig.7] Figure [Fig.7] illustrates an example of the hardware architecture of a device determination and / or an evaluation device according to the invention, and

[0098] [Fig.8] Figure 8 summarizes the results of implementing a process evaluation applied to a VSV cylinder according to the invention based on a quantification law determined according to a determination method according to the invention. Description of embodiments

[0099] Figure 1 shows a device 1 for determining a law quantifying the wear of a cylinder disposed in an aircraft turboshaft engine according to the invention.

[0100] Device 1 is configured to implement a method for determining 100 a law quantifying the wear of a cylinder in an aircraft turboshaft engine, the implementation steps of which are illustrated in [Fig.2].

[0101] The determination device 1 comprises a data acquisition module M102 configured to implement a step 102 of obtaining, for each flight of a plurality of aircraft flights, a parameter value for each parameter of a parameter set and a quantity of cylinder wear induced by said flight, a training module M104 configured to implement a step 104 of training at least one classification model from the parameter values ​​of at least a part of the parameters, a selection module M106 configured to implement a step 106 of selecting at least one parameter from the part of the parameters used to train the classification model, the selection being carried out from the trained classification model,and an M108 determination module configured to implement a step 108 of determining a wear quantification law from the values ​​for the selected parameters and the induced wear quantities.

[0102] Device 1 may further include a labeling module M103 configured to implement a parameter value labeling step associating each flight or parameter value with a cylinder wear state.

[0103] The device 1 may further include a configuration module Ml 10 configured to implement a configuration step 110 of a wear estimation device 4 for a cylinder.

[0104] In particular embodiments, the determination device 1 has the hardware architecture of a computer 3, as shown in [Fig. 7]. It should be noted that certain elements of this architecture may be confused with corresponding elements of the turboshaft engine.

[0105] More specifically, the determination device 1 may include a PC processor, a read-only memory ROM, a random-access memory RAM and communication means.

[0106] The read-only memory of the determination device 1 constitutes a recording medium readable by the processor and on which is recorded a computer program according to the invention, comprising instructions for the execution of the steps of the determination process 100 according to the invention detailed below and in particular illustrated in [Fig.2].

[0107] This computer program equivalently defines functional modules (software) of the determination device 1, such as in particular the data acquisition module M102, the training module M104, the selection module M106, the determination module M108 of a wear quantification law, optionally the labeling module M103 and / or the configuration module M110. The functions of these modules are described in more detail in the following description with reference to the steps of the determination process 100.

[0108] The determination method according to the invention comprises a step 102 of obtaining for each flight of a plurality of flights VI, ... VM, with M an integer strictly greater than 1, of parameter values ​​VAL and of a quantity of cylinder wear QTE, M being an integer strictly greater than 1, preferably greater than or equal to 500, or even greater than or equal to 1000, or even greater than or equal to 10000, or even greater than or equal to 100000.

[0109] A parameter value VAL is obtained for each parameter of a set of parameters Yl, ..., YN, with N an integer strictly greater than 1, preferably greater than 50.

[0110] The values ​​may come from one or more sources 10, in particular they may be stored in one or more databases DB1, DB2, DB3, as is notably represented in the example of [Fig.6].

[0111] The determination method can be implemented for a particular type of cylinder and / or for a particular type of turboshaft engine so as to determine a law for quantifying the wear of a cylinder specific to the particular type of cylinder and / or the particular type of turboshaft engine respectively.

[0112] In particular embodiments of the invention, the determination method aims to determine a wear law for a single type of cylinder, for example a VSV cylinder, optionally in a particular type of turboshaft engine. The induced wear quantities are the wear quantities induced by the respective flights on a cylinder of the particular type, where applicable in a turboshaft engine of the particular type.

[0113] Parameters can be contextual, operational, and / or structural parameters. In particular, parameter values ​​can include parameter values ​​from the public Flight Radar database.

[0114] The parameter set may include one or more, or even all, of the following parameters: - atmospheric pressure, notably measured at a departure and / or arrival airport, and / or in a turboshaft engine, for example at the level of a cylinder, - a temperature, notably measured at a departure and / or arrival airport, and / or in a turboshaft engine, for example at the level of a cylinder, - a humidity level, notably measured at a departure and / or arrival airport, - a level of pollution, in particular measured at a departure and / or arrival airport, the level of pollution being determined for example from a rate of dust particles, a concentration of sea salt, sulfate, carbon dioxide, and / or carbon monoxide. - a flight time, in particular the flight time may include all or part of the following durations: a duration for the taxiing phase, preceding takeoff and / or following landing, also called the "taxi out" phase and the "taxi in" phase, respectively, a duration for the takeoff phase, a duration for the climb phase, a duration for the cruise phase, a duration for the descent phase and a duration for the landing phase. - a distance traveled by a piston in a hydraulic cylinder and / or by an aircraft, - a speed, in particular a speed for a cylinder piston and / or for an aircraft, - an altitude, - an engine speed, - a level of vibration in the turbocharger, for example at the level of a cylinder, - a position command for a cylinder, - a pressure, for example in the turboshaft engine, particularly in the cylinder chamber

[0115] The determination method may include a labeling step during which a wear state LAB is associated with each of the parameter values. Preferably, the wear state is determined for a flight and associated with each parameter value of said flight.

[0116] In particular embodiments, the state of wear is the quantity of wear.

[0117] The state of wear can be determined from data obtained from maintenance operations. The maintenance data can be stored in a BD4 database.

[0118] From the VAL parameter values ​​obtained in step 102, and possibly the LAB wear states, at least one training of a MOD classification model is carried out during a step 104.

[0119] The values ​​of all or part of the parameters in the parameter set can be used to train the classification model.

[0120] The classification model can be a supervised or unsupervised model. The model is intended to determine the wear state of a cylinder from values ​​for the parameters used to drive it.

[0121] At least one training of the classification model is carried out in order to perform step 108 of selecting at least one parameter.

[0122] Many selection methods from one or more trainings of a classification model may be suitable for carrying out the invention.

[0123] By way of non-limitation, specific examples of implementation of the training and selection steps are described below.

[0124] The classification model can be a random forest model trained on the parameter values ​​VAL of the parameter set and the wear states LAB, the wear states serving as labels for the classification. Once trained, the model is advantageously capable of determining the wear state of a cylinder from parameter values ​​provided as input to the model.

[0125] The selection of at least one parameter may involve the application of a "feature importance" algorithm 14, as illustrated in [Fig. 6], which calculates an index based on the mean decrease in impurity (Mean Decrease Impurity Importance) for each parameter. Thus, for each parameter, the "feature importance" algorithm determines an importance index corresponding to the sum of the decreases in impurity for all nodes of the trained model where the parameter is used, weighted by the proportion of parameter values ​​passing through the node, averaged over all trees in the forest. The selection can then be made based on the importance index determined for each parameter.

[0126] In alternative embodiments, several trainings of the classification model can be performed so as to allow comparison of the model's performance when trained with a parameter and without that same parameter. The comparison can be performed for all parameters in the parameter set, making it possible to assess the importance or insignificance of the parameter in the performance of the trained models.

[0127] In alternative embodiments, the classification model is a PCA (Principal Component Analysis) algorithm.

[0128] As illustrated in [Fig.6], the classification model, once trained, can allow an enrichment 12 of the labeled parameter values.

[0129] Once at least one parameter has been selected, the determination process includes a step 108 of determining a law P for quantifying the wear of a cylinder, preferably the determination law taking as an input variable only said at least one selected parameter.

[0130] For example, about ten parameters can be selected from a set of about sixty parameters. This order of magnitude is given by way of illustration and is not limiting. Indeed, in other embodiments, two parameters can be selected from a set of about one hundred parameters. In still other embodiments, more than 90% of the parameters in a set of parameters can be selected.

[0131] Preferably, less than 90% of the parameters in the parameter set are selected, preferably even less than 80%, preferably even less than 50%, preferably even less than 20%.

[0132] The determination of a wear quantification law of a cylinder is carried out from the values ​​of parameters VAL RED for the selected parameter(s) {Yi}, i being in the interval [1 ; N[, each selected parameter being an input variable for the quantification law.

[0133] Preferably, the wear quantification law P=f(Yi) includes a first linear function fi, taking as input variable at least one first selected parameter Yk, and optionally a second function f2 which can depend on one or more second selected parameters, different from at least one first selected parameter, the quantification law being able to be written: P=fi(Yk)+f2(Yj), with k in the interval [1 ; K], K greater than or equal to 1, and j in the interval ]K ; N[.

[0134] The second function may include an amplification coefficient whose value varies according to the value of one or more of the second selected parameters Yj.

[0135] The second function can be piecewise continuous.

[0136] The amplification coefficient can be determined by statistical analysis of the parameter values ​​of all or part of the selected second parameters.

[0137] By way of illustration, it is assumed that a single second parameter Yj is selected, taking values ​​in the interval I. The determination of the amplification coefficient may involve: - For each LAB wear state, the generation of a histogram whose bars are successive and disjoint subintervals {h, I2, ...} covering the interval I, and which represents the distribution of the number of flights VI, ... VN in each sub-interval; - Comparison of histograms according to wear states so as to quantify, for each sub-interval, the correlation between the values ​​of said parameter Yj for said sub-interval and the wear states LAB; - For each sub-interval, the determination of an amplification coefficient as a function of the quantification of the correlation.

[0138] Once determined, the law quantifying the wear of a cylinder can advantageously be applied to estimate or predict the wear of a cylinder in an aircraft turboshaft engine.

[0139] Preferably, the determination method includes for this purpose a configuration step 110 of a device for estimating the wear of a cylinder from the determined quantification law.

[0140] Said estimation device 4 can optionally be configured to estimate wear of different types of cylinders according to several quantification laws determined according to the determination method according to the invention and being specific to said types of cylinders.

[0141] This is also applicable to quantification laws that would be specific to types of turboshaft engines and / or types of aircraft.

[0142] Any quantization law can be stored in a memory 20.

[0143] Figure 3 illustrates a device 2 for evaluating the wear of a cylinder in an aircraft turboshaft engine.

[0144] Device 2 is configured to implement a method for evaluating the wear of a cylinder in an aircraft turboshaft engine, the implementation steps of which are illustrated in [Fig.4].

[0145] The device 2 for evaluating the wear of a cylinder includes a data acquisition module M202 configured to implement a step 202 for obtaining a value for each input variable of the quantification law for at least one flight including said cylinder, and an estimation device M204 configured to implement a step 204 for evaluating a quantity of wear of said cylinder induced by said at least one flight by providing said values ​​as input to said modeling law.

[0146] In particular embodiments, the evaluation device 2 has the hardware architecture of a computer, as shown in [Fig.7].

[0147] More specifically, the evaluation device 2 may include a PC processor, a read-only memory ROM, a random-access memory RAM and communication means.

[0148] The read-only memory of device 2 constitutes a storage medium readable by the processor and on which a computer program conforming to to the invention, comprising instructions for carrying out the steps of the evaluation process 200 according to the invention detailed below and in particular illustrated in [Fig. 4].

[0149] This computer program equivalently defines functional modules (software) of the evaluation device 2, such as in particular the data acquisition module M202, and the estimation device M204. The functions of these modules and devices are described in more detail in the following description with reference to the steps of the evaluation process 200.

[0150] Preferably, device 2 is not disposed in an aircraft.

[0151] Similar to step 102 of the determination process, the values ​​for the input variables can come from one or more sources 10, in particular they can be stored in one or more databases DB1, DB2, DB3.

[0152] The evaluation method 200 may include recording the quantity of wear induced in a memory, different or similar to the memory 20 in which the determination law is recorded.

[0153] Figure 5 illustrates an example of an embodiment of the evaluation method 200 according to the invention in which a value for each input variable of the quantification law is obtained for several flights Va, Vb, Vc.

[0154] For each flight Va, Vb, Vc, a quantity of wear Pa, Pb, Pc respectively is evaluated by providing the values ​​relating to said flight as input to said quantification law, then a quantity of wear induced by the whole of said flights Va, Vb, Vc is determined by accumulating the quantities of wear Pa, Pb, Pc evaluated for each of the flights considered separately.

[0155] The evaluation method 200 may further include a step 206 of comparing the amount of wear induced by at least one flight with a threshold value.

[0156] An alarm may be triggered if the result of the comparison indicates an excessive amount of wear.

[0157] The comparison step 206 may include several comparisons; for example, one comparison may estimate the need to replace the cylinder and another the need to perform a visual inspection during a maintenance operation. Information determined from this comparison may be transmitted to an operator.

[0158] The evaluation device may further include a comparison module (not shown) configured to implement comparison step 206.

[0159] The evaluation device may also include a transmission module (not shown) configured to transmit the information to an operator.

[0160] Obtaining the values ​​in the evaluation method according to the invention may involve estimating the value for one or more input variables of the law of quantification. In particular, such an estimation can be made to assess the wear of a cylinder induced by at least one future flight.

[0161] Such a prediction can make it possible to anticipate a failure of the cylinder and to better plan maintenance operations. In particular, by detecting future wear likely to lead to a failure of the cylinder and therefore of the turboshaft engine incorporating it, before such wear occurs, the evaluation method and device according to the invention can allow for more efficient and judicious aircraft maintenance scheduling. Examples

[0162] Figure 8 illustrates an EXP curve for evaluating the wear depth of a cylinder over time, assessed using a wear quantification law for a cylinder determined by implementing a determination method according to the invention. The REF curve represents a reference curve.

[0163] The curves illustrate the evolution of a cumulative wear depth in pm (ordinate) as a function of a number of engine cycles of the turbomotor comprising said cylinder (abscissa).

[0164] The change in slope in the reference curve REF can be explained by a change in material undergoing wear once a certain depth of wear is reached.

[0165] The quantification law in this example includes the following input variables: - duration of the flight in cruise Af., - flight duration during the ascent phase of Ata, - flight time during the "taxi in" phase, AU, in other words the duration of the aircraft's movement after landing, once the engines have been switched off, - sea salt concentration at departure and arrival airports, (Sd, Sa), - specific humidity at departure and arrival airports (Hd, Ha), - temperatures at departure and arrival airports (Td, Ta), and takes the form: P = a*Atc + g(Ata, Atti, Sd, Sa, Hd, Ha, Td, Ta).

[0166] In this example, the linear wear function, f = a*Atc, has as its input variable the duration of the flight in cruise, and the wear amplification function g(Ata, Atti, Sd, Sa, Hd , Ha, Td, Ta) has as its input variables the duration of flight in the climb phase, the duration of flight in the "taxi in" phase, the sea salt concentration at the departure and arrival airports, the specific humidity levels at the departure and arrival airports, the temperatures at the departure and arrival airports.

[0167] Note that the change in slope observed on the curve is explained by a change in materials of the cylinder undergoing wear.

[0168] Although the present invention has been described with reference to specific embodiments, it is evident that modifications and changes can be made to these examples without departing from the general scope of the invention as defined by the claims. In particular, individual features of the various embodiments illustrated / mentioned can be combined in additional embodiments. Therefore, the description and drawings should be considered in an illustrative rather than a restrictive sense.

[0169] The characteristics described with reference to a process are transposable, alone or in combination, to a device, and conversely, all the characteristics described with reference to a device are transposable, alone or in combination, to a process.

Claims

Demands

1. Method for determining (100) a law for quantifying the wear of a cylinder in an aircraft turboshaft engine comprising: obtaining, for each flight of a plurality of aircraft flights: - a parameter value (VAL) for each parameter of a set of parameters (Yl,... .YN); - of a quantity of wear (QTE) induced by said flight on a cylinder; at least one training of a classification model (MOD) from the parameter values ​​(VAL) of at least a part of the parameters in the parameter set, such that said model once trained is capable of determining a cylinder wear state from values ​​for said at least a part of the parameters in the parameter set; the selection of at least one parameter (Yi) of said at least a part of the parameters according to the influence of each parameter in the determination performance of the trained model; the determination of a quantification law (P) of the wear of a cylinder from the quantities of wear induced (QTE) and the values ​​(VAL) of said at least one selected parameter (Yi), each selected parameter being an input variable for said law.

2. Method of determination according to claim 1, wherein the classification model (MOD) is a supervised learning model trained from parameter values ​​that have been previously labeled (LAB).

3. Method of determination according to any one of claims 1 or 2 wherein the quantification law is specific to a type of turbomachine and / or a type of cylinder.

4. A method of determination according to any one of claims 1 to 3 in which the quantification law includes the determination of a linear wear function having as input variable at least one first selected parameter and optionally of a wear amplification function.

5. A method of determination according to claim 4, wherein the amplification function is determined from a statistical analysis of at least one second selected parameter, distinct from at least one first selected parameter.

6. A method for determining according to any one of claims 1 to 5, comprising the configuration of a device for estimating (4) the wear of a cylinder from the quantification law, the estimating device being intended to evaluate a quantity of wear induced by at least one flight as a function of values ​​for the input variables of said law.

7. Method for evaluating (200) the wear of a cylinder in an aircraft turboshaft engine comprising: - A step of obtaining (202), for at least one flight of an aircraft comprising said cylinder, a value for each input variable of a wear quantification law, and - A step of evaluating (204) a quantity of wear of said cylinder induced by said at least one flight by providing said values ​​as input to said quantification law, said quantification law (P) being previously determined by implementing a determination method (100) according to any one of claims 1 to 6.

8. Evaluation method according to claim 7 further comprising a step of checking (206) the wear of the cylinder by comparing said quantity of induced wear and / or a quantity of cumulative wear, determined beforehand, with at least a threshold value.

9. A computer program comprising code instructions which, when implemented, enable the execution of steps in a process according to any one of claims 1 to 8.

10. Device for evaluating the wear of a cylinder of a turboshaft engine of an aircraft comprising: - a data acquisition module (M204) configured to obtain for at least one flight of an aircraft comprising said cylinder a value for each input variable of a wear quantification law; and - an estimation device (4) configured according to claim 6 and intended to evaluate a quantity of wear induced by said at least one flight as a function of the values ​​of the input variables supplied to it.

Images