FR3096031A1 - Model registration by segment or plane in a turbomachine - Google Patents

Abstract

The present invention relates to a method for resetting a model of turbomachine operating parameter (mod_PARAM), the model being defined as a segment law indicating the value of said parameter as a function of a variable (mod_PARAM (Var)), or being defined as a law per plane indicating the value of said parameter as a function of two variables (mod_PARAM (Var1, Var2),, the resetting method comprising the following steps: - obtaining a value of the parameter (Val_PARAM) (step E1) , - calculation of an error (ε) by comparison of said value of the parameter (Val_PARAM) with the corresponding value of the model (Val_mod_PARAM), said value of the model (Val_mod_PARAM) belonging to one of the segments or planes of the model (mod_PARAM) (step E31), - application of a corrector (112) by minimizing said error (ε) to determine a correction (corr) (step E32), - registration of the segment of the model (mod_PARAM) or of the plane of the model to the correction aid (corr) (step E33). the abstract: Fig. 7a

Description

Model registration by segment or plane in a turbomachine

The present invention relates to the updating of predictive models in the context of a turbomachine.

As a preliminary remark, several definitions are given. As illustrated in FIG. 1 , we place ourselves within the framework of a turbomachine 1 comprising two successive compressors 2, 3 (low-pressure compressor 2 and high-pressure compressor 3) followed by a combustion chamber 4. These definitions are applicable for the entire asked.

Ps3 is the static pressure measured or calculated in the plane upstream of the combustion chamber).

Xn12R is the speed of low pressure compressor 2, reduced to the temperature of said compressor T12 (to overcome temperature variations), expressed in revolutions per minute.

PCN12R (or N1 in the case of a direct drive ) is the speed of low pressure compressor 2, reduced on T12 (to overcome temperature variations), expressed as a percentage of the maximum low pressure speed.

Xn25R is the speed of high pressure compressor 3, reduced on the T25 (to compensate for temperature variations), expressed in revolutions per minute.

PCN25R (or N2) is the speed of high pressure compressor 3, reduced to the temperature of said compressor T25 (to overcome temperature variations), expressed as a percentage of the maximum high pressure speed.

PT2 is the total external pressure (provided by the aircraft).

P25 is the modeled static pressure in the high pressure compressor.

A model is a mathematical law describing the evolution of a physical quantity (parameter) as a function of one or more physical variables.

In operation, it happens that turbomachines undergo false detections of pumping (stall of the blades of one of the two compressors) in the cruising phase. These events have a strong operational impact (engine endoscopy).

In these two cases, it was observed at the time when the events took place a gap failure between the two Ps3 channels, that is to say between the two channels for acquiring the static pressure upstream of the combustion chamber.

The impact of false surge detections has a significant operational impact in that the aircraft is grounded until the engine has been bored out to check for damage.

The Ps3 acquisition line sometimes consists of a pipe which takes the pressure upstream of the combustion chamber 4 and two pressure sensors located directly in the aircraft computer (FADEC, for full authority digital engine control ) .

The measurement of Ps3 is carried out using two independent sensors. In order to consolidate the information from the two sensors, a selection logic between the two sensors has been implemented. It is assumed here that the sensors take valid measurements (no electrical failure and measurement contained in a range of physically plausible measurements), and that the two sensors take measurements away from each other. This configuration causes a gap failure, but it is currently impossible to vote for one or the other of the measurements, not knowing which is the closest to the actual Ps3 value.

To overcome this problem, a model of Ps3 based on thermodynamic laws is calculated. This model theoretically makes it possible to remove the doubt by providing a third quantity (analytical redundancy), independent of the measurements of Ps3, which will make it possible to vote for one or the other of the readings via the selection logic. Figure 2 illustrates this principle, with the two acquisition channels V10, V20, the mod_Ps3 model and the flip-flop which occurs when the V10 channel again becomes closer to the mod_Ps3 model than the V20 channel which had diverged from the V10 channel before . The rocker causes the computer to observe a significant pressure variation ∆Ps3

Nevertheless, we observe in practice values of the model quite far from the real value of Ps3. This can lead to erroneous lane arbitration. The Applicant realized, after studies, that the false detection of pumping was due to a sudden change in the selection of Ps3: as the two measurements of Ps3 were apart, the selected channel went from the strongest measurement to Ps3 to the weakest measurement in one calculation step since the model was initially closer to the most important Ps3 and then approached the weakest Ps3. It is this false jump ∆Ps3 of at least 15% relative value that can trigger a false surge detection when the pressure has not actually dropped.

There is therefore a need to guard against this type of event, in particular by improving the management of arbitration, in particular concerning the pressure Ps3, but for any other parameter.

More generally, there is a need to better process thermodynamic models, so that they better reflect reality, whether for Ps3 or other parameters.

In addition, various improvements or uses of the thermodynamic model could be made to improve the speed, efficiency and relevance of thermodynamic models.

An object of the invention is to propose solutions to the problems mentioned.

To this end, a method is proposed for adjusting a model of the operating parameter of a turbomachine or an aircraft,
the model being defined as a law by segment indicating the value of said parameter as a function of a variable, or being defined as a law by plane indicating the value of said parameter as a function of two operating variables,
said law being affine on each segment or being affine on each plane, the parameter model being stored in a memory.

The operating parameters and the variables relate for example to a temperature or a pressure, or even to a compressor speed (typically the speeds Xn12 and Xn25 of the low pressure body and of the high pressure body. More generally, they can be any parameter of operation for which there is a measurement and a model allowing analytical redundancy

The registration process includes the following steps:
- obtaining a value of the parameter,
- calculation of an error by comparing said value of the parameter with the corresponding value of the model, said value of the model belonging to one of the segments or planes of the model,
- application of a corrector by minimizing said error to determine a correction,
- recalibration of the segment of the model or of the plane of the model using the correction, to reposition said segment or plane and thus obtain a recalibrated model of the physical parameter.

In one embodiment, the step of obtaining the value of the parameter is done by:
- a direct measurement of said parameter using a sensor, or
- a measurement of a third parameter on which said parameter depends, or
- a simulation.

In one embodiment, the corrector is a PID corrector or an integral corrector.

In one embodiment, when the model is a law by segment, the resetting is done by freezing a point of the segment and by moving another point of the segment using the correction, the two points preferably being ends of the segment .

In one embodiment, when the model is a law by segment, the recalibration is done by not keeping any point of the segment fixed, for example by moving the two ends of the segment using the correction.

In one embodiment, the displacement of the ends of the segment is done according to their respective distance with said corresponding value of the model.

In one embodiment, the distribution of the correction to be applied at one end of the segment is equal to the ratio of the distance of the corresponding value of the model at the other end of the segment, over the length of the segment.

In one embodiment, the step of recalibrating the segment of the model comprises a linear interpolation between two recalibrated points.

In one embodiment, when the model is a law by plane, the plane has the shape of a rectangle which is cut into triangles, and the registration is done by freezing one or two vertices of the triangle and by moving the last two or the last vertex of the triangle using the correction.

In one embodiment, when the model is a law by plane, the plane is cut into triangles, and the registration is done by moving the three vertices of the triangle.

In one embodiment, the displacement of each vertex of the triangle is done according to the area of the sub-triangle defined by the two other vertices and said corresponding value of the model.

In one embodiment, the distribution of the correction to be applied to a vertex of the triangle is equal to the ratio of the area of the sub-triangle defined by the other vertices and said corresponding value of the model, to the area of the triangle.

In one embodiment, the triangle registration step comprises a linear interpolation from the registered points.

In one embodiment, the parameter is the pressure Ps3 or the pressure Ps3 divided by the pressure P25 and in which:
- the variable is, when the model is a law by segment, the PCN25R regime and
- the variables are, when the model is a distribution per plane, the PCN25R and the PCN12R, or the PCN25R and the PT2.

In one embodiment, the model to be readjusted is chosen as a function of a variable, the memory stores a plurality of models expressed as a function of the aircraft air bleed, the variable possibly being the level of aircraft air bleed in the compressors.

In one embodiment, the gains of the corrector are different for different segments or planes of the model.

A method for analyzing the aging of a turbomachine is also proposed, the method consisting in implementing the following steps:
- F1: Implementation of a registration process as described previously,
- F2: Save the adjusted model in a non-volatile memory,
steps F1 and F2 being repeated at least twice, and preferably more,
- F3: Compare the different readjusted models to deduce an evolution of the state of the turbomachine.

Other characteristics, objects and advantages of the invention will emerge from the description which follows, which is purely illustrative and not limiting, and which must be read in conjunction with the appended drawings in which:

Figure 1 schematically illustrates a turbomachine.

FIG. 2 illustrates a method of arbitration between two acquisition channels using a thermodynamic model.

FIG. 3 graphically illustrates a process for recalibrating the pressure Ps3.

FIG. 4 illustrates by a block diagram a process for recalibrating a parameter model, such as the pressure Ps3.

Figure 5 illustrates a corrector.

FIGS. 6a and 6b illustrate a method for registering a 2D model by segment.

FIG. 7a illustrates, for a segment, a method of registering a 2D model into a segment by weighting.

FIG. 7b illustrates, for several segments, a method of registering a 2D model into a segment by weighting.

Figure 8 illustrates a 3D model per plane.

FIG. 9 illustrates by a block diagram a process for recalibrating a 3D parameter model, such as the pressure Ps3, as a function of the pressures PCN12R and PCN25R.

FIG. 10a illustrates, for a plane, a method of registering a 3D model into a segment by weighting.

FIG. 10b illustrates the choice of a triangle from among the rectangle forming a plane of the 3D model.

FIG. 10c illustrates the choice of the weighting for a triangle among the rectangle forming a plane of the 3D model.

FIG. 11 illustrates by a block diagram a model selection as a function of a variable, prior to the registration of the model.

Figure 12 illustrates a turbomachine aging analysis method.

The context and definitions given in the introduction are repeated here.

Firstly, a method for adjusting the static pressure model upstream of the combustion chamber will be described. This pressure will be called pressure Ps3 and this model will be called “model of Ps3” and referenced mod_Ps3. This is a thermodynamic model.

The final goal of the Ps3 model is in particular to make it possible to arbitrate between two redundant acquisition channels V10, V20, whose function is to measure the pressure Ps3. Each V10, V20 acquisition channel includes a 10, 20 sensor. The 10, 20 sensor is standard and will not be described here.

A process for arbitration between the two acquisition channels V10, V20 will be described below.

A calculation unit 100 is provided, which comprises a processor 110 and a memory 120. The calculation unit 100 can be a FADEC (“ full authority digital engine control ”) or else be a distinct component, positioned as close as possible to the tracks acquisition V10, V20 for more responsiveness.

The memory 120 stores a mod_Ps3 model, which makes it possible to obtain the value of the pressure PS3 as a function of at least one variable Var, which is the PCN25R speed (high pressure compressor speed): the mod_Ps3 model is then written under the form mod_Ps3(PCN25R). In practice, the mod_Ps3 model involves several sub-models, such as in particular the Ps3 model on the high-pressure compressor pressure P25 (this model is called mod_Ps3/P25) and the mod_Ps3/P25 model is itself expressed as a function of the speed of the PCN25R high-pressure compressor reduced to its temperature T25. This model is then written in the form mod_Ps3/P25(PCN25R/T25).

Then simply multiply the value of Ps3/P25 by P25 to obtain the value of Ps3.

Rather than directly readjusting the mod_Ps3 model, it is thus preferable to readjust the mod_Ps3/P25 model. The denomination of "Ps3 model", in the form mod_Ps3, includes models that do not directly express the pressure Ps3 but allow it to be obtained subsequently, such as the model mod_Ps3/P25.

In a first step E1, one of the two acquisition channels V10, V20, using its sensor 10, 20, measures a value Val_Ps3 of the pressure Ps3 on the turbomachine (for a real value of the physical quantity which is used as a variable, i.e. PCN25R). At this stage, it is assumed that the two acquisition channels V10, V20 are healthy and that the two sensors 10, 20 give a correct measurement. In other words, there is no failure of sensors 10, 20 or deviation beyond a predetermined threshold between the two measurements.

This measurement of a value Val_Ps3 of the pressure Ps3 is then sent to the calculation unit 100.

A data conversion or processing step E2 can be implemented: for example, Val_Ps3 is a static pressure value Ps3, whereas the model mod_Ps3/P25 uses the pressure Ps3 reduced on the P25: it is therefore necessary to divide the value of the static pressure by P25 to obtain the value Val_Ps3/P25.

Then, in a step E3, the calculation unit 100 readjusts the model of Ps3 stored in its memory 120 using said measurement of the value of the pressure Ps3. By readjusting, we mean that there is at least one point of the mod_Ps3 model (in practice a plurality, or even an infinity, if the model is continuous) whose ordinate has been moved (therefore at constant abscissa). Note Rmod_PS3/P25 the readjusted model. Thereafter, we will simplify the writing by keeping mod_PS3/P25 which designates a model before and after resetting.

In this case, there is at least one point P of the mod_Ps3(Var) curve whose value Val_mod_Ps3(Var) has changed before and after the recalibration, for a given value of the variable. In the preferred embodiment, one works with mod_Ps3/P25(PCN25R) and Val_mod_Ps3/P25(PCN25R).

Finally, a step E4 of storing the readjusted Ps3 model in the memory 120 is defined. In another embodiment, it does not delete it.

Preferably the steps E1, E2 and E3 are repeated at regular intervals, of type at each calculation step. The calculation step corresponds to about -0.015s. During a calculation step, the two steps E1 and E3 can be implemented or else a step E1 and in parallel step E3 using data from step E1 of the previous step are implemented.

As the mod_Ps3 model is updated at regular intervals, the arbitration can be done more quickly and therefore more correctly, avoiding the ∆Ps3 jumps linked to the untimely change of lane V10, V20.

The resetting is advantageously carried out using a corrector 112 which is integrated in a loop of the control chain. This corrector will be described in detail below.

A method of arbitration between two acquisition channels V10, V20 is also defined, the arbitration method comprising a step A1 for implementing a registration method as described above and a step A2 for choosing the acquisition channel V10, V20, during which the processor chooses a channel V10, V20 from among the two channels V10, V20. The choice is made according to the acquisition channel V10, V20 which is closest to the readjusted model. Step A2 is classic and will not be described here.

Secondly, a specific process for resetting a mod_PARAM model of a turbomachine or aircraft parameter (for example temperature, pressure, absolute or relative) will be described, with reference to the general representation of FIG. . We will speak of “parameter of interest”. The model is again a thermodynamic model. The model describes the evolution of the parameter as a function of one or more variables Var which are also in reality turbomachine or aircraft parameters (for example temperature, pressure, absolute or relative). It is stored in the memory 120 of the calculation unit 100.

This method is fully applicable to the method for adjusting the pressure Ps3 described above. We will also use the pressure Ps3 as an example of parameter PARAM and the pressure PCN25R as variable Var but the method can be applied to any physical parameter PARAM of an aircraft and any variable Var (for example the pressure PT2): for example mod_Ps3 /P25(PCN25R), mod_Ps3/P25(PCN25R, PCN12R), mod_Ps3/P25(PCN25R, PT2), mod_T25(PCN12R, PT2), mod_Xn25(PCN12R, PT2) where Mach is the speed of the aircraft, mod_T3(T25 ), etc.

A model is defined here as a law per segment (in a so-called 2D configuration) or per plane (in a so-called 3D configuration) indicating the value of said parameter of interest as a function respectively of a variable Var (2D) or of two variables Var1 , Var2 (3D). The law is linear respectively on each segment (or in other words, affine by piece: i.e. its equation is in the generic form z=ax+c) or on each plane (equation in the generic form z= ax+by+c).

The advantage of a model defined as a law by segment (2D) or by plane (3D) is the application of the principles of linear control. For example, the model mod_Ps3/P25(Xn25r) or mod_Ps3/P25(PCN25R) is nonlinear in its entirety.

We place ourselves in the same framework as before, with the two acquisition channels V10, V20.

In a step E1, a value Val_PARAM of the parameter of interest PARAM is obtained. This obtaining can be done within the framework of step E1 described previously, by a measurement of a sensor 10, 20 of one or more acquisition channels V10, V20, with in particular the acquisition of a third parameter and said interest parameter is deduced therefrom. Alternatively or additionally, the parameter of interest PARAM can be obtained using a simulation.

The following steps and sub-steps are implemented by the processor 110 and the memory 120 of the calculation unit 100.

A data conversion step E2 can be implemented when the measured parameter does not correspond to the model parameter: for example, as explained previously, Val_Ps3 is a static pressure value Ps3, whereas the model mod_Ps3/P25 uses the pressure Ps3 reduced on the P25. In the case of a third-party parameter, said calculation unit 100 calculates a value of the parameter of interest Val_PARAM from the value of the third-party parameter.

Then, the registration step E3 is implemented. This resetting step E3 comprises several sub-steps.

In a sub-step E31, the processor 110 retrieves the value Val_mod_PARAM from the model mod_PARAM which corresponds to the value of the parameter of interest Val_PARAM obtained in step E1.

The value of the model Val_mod_PARAM is therefore on one of the segments or planes of the model mod_PARAM. This correspondence can be done via the value of the variable Var of the model mod_PARAM: one takes the value of the model Val_mod_PARAM whose abscissa corresponds to that of the value Val_PARAM of the parameter of interest. To do this, it may actually be necessary to perform two measurements: one on the PARAM parameter and one on the Var variable, to obtain a pair of data.

In the case of the pressure Ps3, it is thus possible to have a measurement of the PCN25R at the same time as the measurement of the Ps3.

With the two values Val_mod_PARAM and Val_PARAM, the sub-step E31 comprises the calculation of an error ε, typically by subtraction: ε=Val_mod_PARAM-Val_PARAM. This error ε is illustrated in figure 5 .

In a sub-step E32, this error ε is processed by a corrector 122, whose role is to minimize said error ε. The corrector 122 makes it possible to calculate a correction corr which is a deviation to be applied to the coordinates of the points of the corrected law, obtained via the PID corrector, from the error (difference between the measurement and the model) and which must be made to the mod_PARAM model. Due to the segmentation (segment or plane) of the m_PARAM model, the corrector is only implemented on the segment or the plane considered during the implementation of step E3.

A particular corrector will be described below.

Finally, in a sub-step E33, the correction corr is used to readjust the segment or the plane of the model mod_PARAM. This step consists in recalculating a segment or a plane, from the previous mod_PARAM model and from the correction corr calculated in the sub-step E32. In particular, the resetting consists in moving a minimum number of points of the mod_PARAM model in a sub-step E331 and in interpolating the rest of the model between these points in a sub-step E332: two points for the model by segments and three points for the model by plane.

Several embodiments of the registration will be described below.

It is further noted, for example in FIG. 3, that the readjustment of a segment will also influence the adjacent segments in the case where the end of the readjusted segment is moved. A step of interpolation of the adjacent segments can also be implemented.

The corrector chosen is a PID corrector (proportional integral derivative), illustrated in figure 5, where Gp, Gd and Gi are respectively the gain of the proportional corrector, of the derivative corrector and of the integral corrector, S being the variable in the frequency domain (variable of Laplace).
The integral corrector (the I of the PID) makes it possible to introduce a certain inertia into the looped system, which makes it possible to avoid hyper-sensitivity to disturbances and crazy points, compared to an all or nothing corrector. The integral corrector also makes it possible to control the speed of resetting, and to avoid an instantaneous drift of the model m(param) towards the mean between the two channels V10, V20 in the event of drift of one of the sensors 10, 20.
A proportional corrector (the P of the PID) and a derivative corrector (the D of the PID) are implemented to more finely adjust the corrector 122 if necessary but are not used (the empirical approach has shown that their contribution is marginal compared to that of the integrator which naturally transcribes much better the behavior desired for the registration). We can thus have Gp=Gd=0.
The adjustment of the corrector is made in such a way that the mod_PARAM model is readjusted quickly enough to take account of the reconfigurations of the turbomachine (for example a change in the levels of air bleeds from the high pressure compressor).

Model by segment (2D)

We place ourselves here on the segment of the model mod_PARAM which is concerned by the measurement Val_PARAM carried out in step E1. This segment has two extremal points, on the left and on the right, denoted A and B.

Point-to-point registration

The first solution, illustrated in FIGS. 6a and 6b , consists in accounting for the correction by modifying the coordinates of a single point of the segment, for example one of the extremal points A or B, while the other is fixed.

In this case, the output of the corrector 122 directly impacts point B (respectively point A), and point A (respectively point B) remains fixed. This solution, however, forces at least one of the points of the mod_PARAM model to be fixed to serve as a reference, from which the other segments of the mod_PARAM model will be impacted. Thus, during the resetting step E2 and more precisely during the sub-step E231, only one of the two extreme points is moved. Then, the interpolation step E232 is implemented.

This solution is the simplest and fastest to calculate.

Weighted registration of the two points of the segment

The second solution, illustrated in FIGS. 7a and 7b , consists in distributing the correction in a weighted manner to allow the selected segment to be readjusted in a more representative and more efficient manner. In an advantageous embodiment, the weighting is performed as a function of the distance between value Val_PARAM, here Val_Ps3/P25, and points A and B of the segment.

Figures 7a and 7b illustrate the readjustment over an interval and a calculation step:
- step E1: the measured value Val_PARAM is obtained by one or two acquisition channels V10, V20; in the example, it is Val_Ps3 ,
- step E2 (image (a) of FIG. 7b ): the measured value Val_PARAM is converted to be homogeneous with the mod_PARAM model; by simplification, one keeps the same reference Val_PARAM,
- step E31 (image (b) of FIG. 7b ): ε is measured which is the difference between the measured value Val_PARAM and value of the model Val_mod_PARAM; in the example with pressure Ps3: Val_PARAM=Val_PS3/P25, i.e. the pressure Ps3 measured divided by the model pressure P25 and Val_mod_PARAM = Val_mod_Ps3/P25, the pressure Ps3 of the model readjusted (by previous iterations ) that we divide by P25 model,
- step E32 (image (b) of FIG. 7b ): the error ε is minimized via the corrector 122, by integrating it, to calculate a correction corr,
- step E331 ( FIG. 7a ): the distance is then measured (or before step E31) from the point Val_mod_PARAM, here Val_mod(Ps3/P25), to the point A, which constitutes the lower limit of the interval of the variable Var (here PCN25R) and which depends on the linearization of the model chosen, with respect to the distance between points A and B. Finally, the correction is distributed on the ordinate of points A (to give A') and B (to give B'),
- step E332 (image (c) of FIG. 7b ): a new segment is interpolated between the two readjusted points A' and B'.

The principle of operation is to distribute the correction corr of the corrector 122 of an interval on the ordinates of the points A and B according to the same principle as previously: in one embodiment, X% of the correction is distributed on the ordinate of the point B, with X the ratio between the distance from point Val_mod_PARAM to point A over the distance from point A to point B. 100-X% of the correction is distributed on the ordinate of point A (30% and 70% on the Figure 7a ).

Once the two points A' and B' have been replaced, all that is required in step E232 is to interpolate the model between these two points. The law being defined by segment, the linear interpolation (or affine) is simple.

Alternatively, any other (distinct) points of the segment can be moved by the correction: it suffices to choose two points and the linear (or affine) interpolation makes it possible to complete the rest of the segment considered.

This method thus allows an effective and fast recalibration to obtain a model recalibrated mod_PARAM. Nevertheless, this mod_PARAM model only depending on one variable Var (PCN25R in the case of mod_Ps3), it may be insufficient for certain flight situations, in particular when the parameter of interest PARAM depends on several variables Var1, Var2.

Model by plan (3D)

In this respect, to take several variables into account, the mod_PARAM model can be a function of two variables (mod_PARAM(Var1, Var2)) and be expressed in the form of a law defined by planes, the law being linear on each plane as shown in Figure 8 .

FIG. 9 illustrates the implementation of a registration method in the case of a model by plane.

For example, in the case of pressure Ps3, when activating a bleed air level, the model mod_Ps3/P25(PCN25R) (i.e. the model Ps3 reduced on P25 according to of PCN25R) is modified because part of the air compressed by the high pressure compressor is sent to the aircraft air system). The corrector 122 of the 2D model per segment possibly makes it possible to adapt to this reconfiguration if the gains of the corrector 122 are adjusted so that the registration of the model is rapid, but this may pose other difficulties.

Still in the example of the pressure Ps3, to overcome the problem of air samples, a model of Ps3/P25 which no longer depends not only on PCN25R, but also on PCN12R, is then implemented: we then define mod_Ps3/ P25(PCN25R, PCN12R). When activating the samples, the law linking PCN25R and PCN12R is modified, which allows us to account for the reconfiguration of the system. The readjustment of this law therefore requires a new “3D” corrector.

Point-to-point registration

The first solution, not illustrated, consists in accounting for the correction by fixing the coordinates of a single point of the rectangle, for example one of the vertices A, B, C or D of the rectangle and by modifying the coordinates of two points of the rectangle , for example two of the vertices A, B, C or D. Alternatively, one can fix two points fixing the coordinates of two points of the rectangle, for example two of the vertices A, B, C or D of the rectangle and by modifying the coordinates of a point of the rectangle, for example two of the vertices A, B, C or D.

The points concerned are moved during the sub-step E331 then the interpolation step E332 over the entire rectangle is implemented. As we work on three points each time, we are assured of the existence of the interpolated rectangle.

Weighted registration

To allow a weighted readjustment, for which no point is fixed, the model mod_PARAM is linearized by cutting the rectangle ABCD into triangles ABC, ABD, typically two complementary triangles ( figure 8 ). Indeed, three points A, B, C are always coplanar, before and after readjustment, which ensures the existence of the interpolation of the readjusted triangle at the interpolation sub-step E332, once the sub-step E331 adjustment of the three points carried out. The three new points resulting from the correction can thus be used to describe the Cartesian equation of a plane, thus allowing us to linearly interpolate the mod_PARAM model.

Indeed, if a correction weighted on three points of the surface were applied on four points, for example the four vertices ABCD of the rectangle, there would be a deformation of the rectangle if the four points of the rectangle were no longer coplanar (impossible to interpolate the coordinates of the parameter PARAM thanks to the Cartesian equation of a plane).

In the sub-step E331, it is first a question of selecting the triangle to readjust according to the value of Val_PARAM (called point X) obtained by the steps E1 and E2. For this, a difference in slope between the segment AC which divides the rectangle in two and the segment AX ( figure 10b ). You can use any vertex B, C or D.

Indeed, with the four points A, B, C, D forming a rectangle and the point X corresponding to the measured point Val_PARAM, it is necessary to determine if X belongs to the triangle ABC or to the triangle ACD (remember that these triangles were chosen arbitrarily compared to ABD and DBC).

For this, a comparison of the values of the rates of variation ΔAC, ΔAX of the straight lines (AC) and (AX) is carried out during the sub-step E331. Indeed if ΔAX > ΔAC then ACD is selected and if ΔAX ≤ ΔAC then ABC is selected. Then it is a matter of distributing the correction.

Unlike the 2D model with segments, the distances between point X and the points of triangle ABC do not reflect the distribution of the correction to be applied. The distribution is therefore made in proportion to the areas of the triangles XAB, XAC and XBC ( figure 10c, where x is the area of XBC, y the area of AXC and z the area of XAB).

We define the ratios corr_A, corr_B, corr_C by corr_a=x/(x+y+z), corr_b=y/(x+y+z), and corr_z=z/(x+y+z).

The ratio corr_A is applied to the readjustment of point A, corr_B to that of point B and corr_C to that of point D.

Finally, the interpolation sub-step E332 is implemented from the three points readjusted by a simple Cartesian plane equation, to interpolate the entire triangle.

Matrix segment (2D) model

It has been said that the 2D segment model has limitations, especially when another variable could have a strong influence on the mod_PARAM model.

Illustrated in FIG. 11 , another solution for taking account of another variable consists in storing in the memory 120 a matrix M of 2D mod_PARAM model. Instead of having a model in the form mod_PARAM(Var1, Var 2), we have a model in the form mod_PARAM_Var2(Var1), where mod_PARAM_Var2 designates an applicable model for a given value (or a set of given values) of the variable Var2.

Figure 11 illustrates mod_Ps3_PCN12R(PCN25R). Here, PCN12R does not necessarily symbolize an exact value of the variable but a level, which can be an interval or be discrete.

In the case of the pressure Ps3 where the parameter PARAM is Ps3/P25 and where the variable Var1 is PCN25R, the memory 120 can store a plurality of model mod_Ps3 according to the samples, that is to say PCN12R.

In this embodiment, there are a limited number of templates stored. Therefore, PCN12R values can be expressed by a number of aircraft air bleed levels.

Consequently, before the step E31 described above, the model mod_PARAM_Var2 is chosen in a step E30, according to the value of the variable Var2, then the model mod_PARAM_Var2 is readjusted as a 2D model during the steps E31, E32 and E33. Parallel to step E1, there is a step of measurement or acquisition of the variable Var2 which determines the choice of the model mod_PARAM_Var2

Adjustment of the dynamics of the correctors

The adjustment of the dynamics of the 2D corrector is carried out by taking into account two contradictory needs:
- the dynamic must be slow enough so that the known cases of drift of one of the acquisition channels V10, V20 do not cause the model to drift by following the average of the channels V10, V20 (so that we can vote for one of the two paths at the moment the gap failure rises),
- the dynamics must be fast enough for the rev ranges concerned to be readjusted all the same (in particular the revs covered up to the take-off revs during take-off).

As we have one corrector 122 per 2D model segment or per 3D model plane, it is possible to adjust the correctors (essentially the integrating corrector by the way) independently of each other:
- a fast dynamic will then be applied to the speed ranges covered quickly during a classic mission. This makes it possible to respond to the constraint of resetting these rev ranges in a very short time,
- a slow dynamic will be applied to the speed ranges on which the recalibration time is not a major constraint (examples: ground idle, cruise, climb). In the case of the pressure Ps3, this makes it possible to best guard against the risks of recalibration on the average of the channels Ps3 in the event of drift of one of the two over these speed ranges.

Thirdly, a turbomachine aging analysis method will be described, as illustrated in FIG . The example is taken again with the pressure Ps3 and the previous recalibration models, but the principle is applicable in the same way to any recalibration process making it possible to generate a recalibrated model Rmod_PARAM.

At each readjustment, step E3 is implemented and a “readigned” model mod_PARAM (mod_Ps3, mod_Ps3/P25, etc.) is generated. When the purpose of this readjustment is to allow a more effective arbitration, the readjusted model mod_Ps3/P25 comes to replace the model mod_Ps3/P25 previously which becomes de facto obsolete. In this regard, an overwrite can be performed in memory 120.

However, as each mod_Ps3/P25 model differs from the previous model (on a few segments or a few planes, at least), it is possible to observe, step by step, the overall evolution of the mod_Ps3/P25 model by comparing all (or a certain number) of failed models.

Thus, the various resetting methods described previously are advantageously implemented in a method for measuring turbomachine aging.

The turbomachine analysis method thus comprises a step F1 of implementing a readjustment method comprising the steps E1, E2, E3, E4 and a step F2 of storing the readjusted mod_PARAM model in a memory, which can be the memory 120. Unlike step E4, which may involve deleting the previous model, step F2 involves a definitive save (that is to say a non-transitory save) of the mod_PARAM model.

Steps F1 and F2 are repeated at least twice and preferably a large number of times.

In particular, the behavior of a compressor can be degraded in different ways depending on its environment (cold, sand, etc.) or unexpected events (ingestion of a bird causing pumping or slight damage to the blades). The retiming allows the model to “age” with its engine. It must therefore be able to readjust on one or two missions, but not be sensitive to variations in Ps3 over a few seconds.

As it is a question of analyzing the turbomachine, that is to say of seeing its evolution over time, it is preferable for the memory 120 to store corrected mod_PARAM models generated at time intervals greater than one day, or even month or quarter or semester.

Once all these data have been acquired, a comparison step F3 is implemented by the processor 110 to compare the different mod_PARAM readjusted models. This comparison makes it possible to deduce the state of the turbomachine.

In the case of pressure Ps3 for example, even with PCN25R, a "young" HP compressor will have a higher Ps3 than an "old" HP compressor. The degradation of the compression rate therefore results in the lowering of the Ps3 to a given PCN25R. The comparison of the models therefore makes it possible to deduce a change in the state of the engine.

Step F3 can be performed by the calculation unit 100 directly, so that the state of the turbine engine or of the aircraft is known as soon as an operator requires it. Alternatively, this step F3 is done in the design office, after data recovery. In the same way, step F2 can be performed using the memory 120 of the calculation unit, but the recalibrated models Rmod_PARAM can also be transmitted to a memory external to the aircraft or to the turbomachine, in particular in a design office, to then implement the F3 state.

For example, an analysis of the aging of the high-pressure compressor can be established thanks to the evolution of the mod_Ps3/P25(PCNR25R) model. As compressor efficiency decreases over time, monitoring mod_Ps3/P25(PCNR25R) models provides continuous information reflecting the current compressor.

Claims (17)

Method for adjusting an operating parameter model (mod_PARAM) of a turbomachine (1) or an aircraft, ,
the model being defined as a law by segment indicating the value of said parameter as a function of a variable (mod_PARAM(Var)), or being defined as a law by plane indicating the value of said parameter as a function of two variables (mod_PARAM(Var1, Var2), ,
said law being affine on each segment or being affine on each plane, the parameter model being stored in a memory (120),
the registration method comprising the following steps:
- obtaining a value of the parameter (Val_PARAM) (step E1),
- calculation of an error (ε) by comparing said value of the parameter (Val_PARAM) with the corresponding value of the model (Val_mod_PARAM), said value of the model (Val_mod_PARAM) belonging to one of the segments or planes of the model (mod_PARAM) ( step E31),
- application of a corrector (112) by minimizing said error (ε) to determine a correction (corr) (step E32),
- recalibration of the segment of the model (mod_PARAM) or of the plane of the model using the correction (corr), to reposition said segment or plane and thus obtain a recalibrated model of the physical parameter (step E33). Method according to claim 1, in which the step of obtaining the value of the parameter (Val_PARAM) is done by:
- a direct measurement of said parameter (PARAM) using a sensor (10, 20), or
- a measurement of a third-party parameter on which said parameter depends (PARAM), or
- a simulation. Method according to claim 1 or 2, in which the corrector (112) is a PID corrector or an integral corrector. Method according to any one of Claims 1 to 3, in which, when the model (mod_PARAM(Var)) is a law by segment, the readjustment is done by freezing a point of the segment (A, B) and by moving another point of the segment (A, B) using the correction (corr), the two points (A, B) preferably being ends of the segment. Method according to any one of Claims 1 to 4, in which, when the model is a law by segment (mod_PARAM(Var)), the readjustment is done by not keeping any point of the segment fixed, for example by moving the two extremities (A, B) of the segment using the correction (corr). Method according to claim 5, in which the displacement of the ends of the segment (A, B) is done according to their respective distance with the said corresponding value of the model (Val_mod_PARAM). Method according to claim 5 or 6, in which the distribution of the correction (corr) to be applied at one end of the segment (A, B) is equal to the ratio of the distance of the corresponding value of the model (Val_mod_PARAM) at the other end of the segment (B, A), along the length of the segment (AB). Method according to any one of Claims 4 to 7, in which the step of registering the segment of the model comprises a linear interpolation between two registered points. Method according to any one of Claims 1 to 4, in which, when the model (mod_PARAM(Var1, Var2)) is a law by plan, the plan has the shape of a rectangle (ABCD) which is cut into triangles ( ABC, ABD), and the resetting is done by freezing one or two vertices of the triangle and by moving the last two or the last vertex of the triangle using the correction (corr). Method according to any one of Claims 1 to 4, in which, when the model (mod_PARAM(Var1, Var2)) is a law by plane, the plane is cut into triangles, and the registration is done by moving the three vertices of the triangle (A,B,C). Method according to Claim 10, in which the displacement of each vertex (A, B, C) of the triangle is done according to the area of the sub-triangle (XBC, XAC, XAB) defined by the two other vertices (BC, AC, AB) and said corresponding value of the model (Val_mod_PARAM). Method according to claim 11, in which the distribution of the correction (corr) to be applied to a vertex of the triangle is equal to the ratio of the area of the sub-triangle defined by the other vertices and the said corresponding value of the model, on the area of the triangle (ABC). A method according to any of claims 9 to 12, wherein the step of registering the triangle comprises linear interpolation from the registered points (ABC). A method according to any one of claims 1 to 13, wherein the parameter is pressure Ps3 or pressure Ps3 divided by pressure P25 (Ps3/P25) and wherein
- the variable (Var) is, when the model is a law by segment (mod_Ps3/P25(Var), the PCN25R regime and
- the variables (Var1, Var2) are, when the model is a law by plane (mod_Ps3 (Var1, Var2)), the PCN25R and the PCN12R, or the PCN25R and the PT2. Method according to any one of Claims 1 to 14, in which the model (mod_PARAM) to be readjusted is chosen as a function of a variable, the memory (120) stores a plurality of models (mod_PARAM) expressed as a function of the sampling of aircraft air, the variable possibly being the level of aircraft air bleed into the compressors. Method according to any one of Claims 1 to 15, in which the gains of the corrector are different for different segments or planes of the model (mod_PARAM). Method for analyzing the aging of a turbomachine (1), the method consisting in implementing the following steps:
- F1: Implementation of a registration method according to any one of claims 1 to 16,
- F2: Save the adjusted model (mod_PARAM) in a non-volatile memory (120),
steps F1 and F2 being repeated at least twice, and preferably more,
- F3: Compare the different readjusted models (mod_PARAM) to deduce an evolution of the state of the turbomachine.