EP2786004B1 - Method and apparatus for the continous estimation of air-fuel ratio in an engine's cylinder - Google Patents

Description

The present invention relates to a method and a device for continuously estimating the cylinder richness of an engine. It concerns in particular, but not exclusively, a method and a device for continuously estimating the cylinder richness of a diesel engine, by means of a measurement of richness achieved by a sensor such as a probe.

It is known that the values relating to the richness of an engine can be measured by means of a probe, such as a lambda or NOx probe, which makes it possible to obtain precise values in stabilized speeds.

Nevertheless, it turns out that the implementation distance of the probe relative to the cylinder, the elements of the exhaust line arranged upstream of the probe, or the operating principles of the probe cause a delay and a filtering of this probe. which may vary depending on the stabilized operating points. As a result, the values relating to the richness thus measured are not perfectly representative of the real wealth specific to a cylinder during dynamic operation.

It is also known that the information relating to the richness of an engine can also be estimated by performing a calculation using the parameters of the admission air flow rate (measured or estimated) and the injected fuel flow rate (estimated). The result obtained has the advantage of having been calculated taking into account the dynamics associated with said parameters, as well as the slowness of establishing the air loop relative to that of the fuel circuit.

However, the determination of the wealth of an engine by proceeding in this way does not dispense with dispersions and drift of engine components, such as injectors, the fuel pump, the flow meter measuring the intake air. A dynamic but potentially derivative signal is thus obtained.

The estimation of the richness of an engine can also be achieved by mapping this estimate according to parameters characterizing the operating point which can be typically the speed and torque.

However, a wealth engine thus mapped presents neither the good dynamic representative of the real wealth of the cylinder (due in particular to a too low consideration of the slow establishment of the air loop compared to that of the circuit of fuel), nor sufficiently robust with respect to the dispersions generated by the engine components.

We know the document FR2834314 describing a method for estimating the richness of a cylinder of an internal combustion engine using a Kalman filter. The patent is also known EP 0 643 211 which relates to a device for estimating the air / fuel ratio of a mixture supplied to an internal combustion engine, by means of a probe which may be of type 02. The implementation of this invention results in the modeling of the behavior of the probe and the use of a Kalman filter.

This device makes it possible to estimate the cylinder-by-cylinder richness, by detecting in the exhaust puffs the relative impact of the different combustions on the overall wealth, the measurement of which is carried out by means of a probe disposed at the level of the exhaust manifold. . The estimation of the richness per cylinder is weighted by previously identified coefficients in a table.

• sa mise en oeuvre s'effectue uniquement sur des points de fonctionnement stabilisés ; elle ne peut s'effectuer en transitoire ;
• il ne permet pas de recaler la richesse globale en fonction des dispersions et dérives éventuelles ;
• le modèle de sonde utilisé ne fonctionne que sur un point de fonctionnement stabilisé et n'est pas adaptable d'un moteur à un autre en raison des dispersions ;
• l'utilisation des sondes NOx positionnées très en aval dans une ligne d'échappement et ayant un signal de richesse plus filtré que celui d'une sonde 02 placée plus en amont ne semble pas adaptée en cas d'utilisation de ce dispositif ; en effet, les pics de richesse occasionnés par les bouffées d'échappement et les consignes de « wobbling » ne sont pas visualisables avec ce type de sonde.

However, it turns out that this device has the following drawbacks:

• it is implemented only on stabilized operating points; it can not be transient;
• it does not make it possible to recalibrate the global richness according to possible dispersions and drifts;
• the probe model used only operates on a stabilized operating point and is not adaptable from one motor to another due to dispersions;
• the use of NOx probes positioned very downstream in an exhaust line and having a more filtered richness signal than that of a probe 02 placed further upstream does not seem suitable when using this device; Indeed, the peaks of wealth caused by the exhaust puffs and the instructions of "wobbling" are not visualizable with this type of probe.

Finally, we know the patent application WO 07 041 092 which discloses a system and a method for controlling a diesel engine by means in particular of sensors shaped to detect at least one specific component of the exhaust gas, as well as the patent application EP 1 413 728 which relates to a control unit and a control method of a NOx probe placed in the exhaust gas duct of an internal combustion engine.

However, the devices that are the subject of these patent applications do not make it possible to dispense with probes of type 02, and to allow both an estimation of the cylinder richness of an engine on stabilized and possibly transient points.

The present invention therefore more particularly aims to solve these problems by proposing a method for continuously estimating the cylinder richness of an engine, with great accuracy and good dynamics, the implementation of the method can be carried out advantageously on the basis of stabilized operating points but also in transient.

To this end, the invention relates to a method for continuously estimating the cylinder richness of an internal combustion engine according to claim 1.

In this way and advantageously, the device according to the invention makes it possible to carry out a continuous estimation of the cylinder richness of an engine by using a calculated model of richness, and by recalibrating the result of this calculation with respect to the dispersions and possible drifts in order to obtain a continuous correction of the calculated value of wealth.

In this way, the correction of the calculated wealth Φ_Calc can be advantageously completed in order to refine the final value of the richness of the cylinder.

Advantageously, the combined use of these two drift identifiers makes it possible to take into account, without any limit, the nonuniformity of the wealth drift over the entire motor field, as well as its engine motor dispersion.

The invention also relates to a device for continuously estimating the cylinder richness of an internal combustion engine, characterized in that it comprises a drift identifier and a Kalman filter shaped to implement the one or more processes according to the invention.

One embodiment of the invention will be described below, by way of non-limiting example, with reference to the accompanying drawings in which:

The figure 1 is a schematic representation of the main elements used in the implementation of an alternative embodiment of the method according to the invention, with a highlighting of the relations existing between them.

The figure 2 is a schematic representation, in the form of curves, of the various richness and wealth drift signals produced during the implementation of the method according to the invention.

In this example, as shown on the figure 1 , the simple model of wealth Φ_Calc is calculated by a calculator 1 by means of the fuel charge to inject qlnj in a cylinder and the measurement of the air intake rate Qair in the same cylinder.

The fuel quantity to be injected qlnj is equivalent to the estimated value of the injected fuel flow, the air intake flow rate Qair being modeled or measured using a flowmeter.

The constitutive signal of the wealth model corresponds to the quotient of the injected fuel flow qlnj (or fuel_injected) on the intake air flow Qair or (Qair_cylindre), multiplied by the stoichiometric ratio y. The signal thus obtained has the particularity of being at the right dynamic, but potentially derived. $$\boxed{{\mathrm{\Phi }}_{\mathit{cylindre\_estimée}}=\gamma \cdot \frac{Q_{\mathit{carburant\_injecté}}}{Q_{\mathit{air\_cylindre}}}}$$

• de prendre en compte de manière satisfaisante les dynamiques associées aux deux grandeurs qlnj et Qair ; et
• d'obtenir une représentation correcte du phénomène constitué par la lenteur d'établissement de la boucle d'air par rapport à celle du circuit de carburant.

Advantageously, the constitutive signal of the wealth model allows:

• to take into account satisfactorily the dynamics associated with the two quantities qlnj and Qair; and
• to obtain a correct representation of the phenomenon constituted by the slowness of establishment of the air loop with respect to that of the fuel circuit.

Moreover, using a richness probe 2 which may be of the NOx type placed downstream in the exhaust line, preferably downstream of the catalyst SCR (Selective Catalytic Reduction), a measured signal of richness is defined. . This signal makes it possible to produce in transitory information on the wealth that is delayed and attenuated; on the other hand, in stabilized, the information carried by the signal is considered to be precise.

The richness probe 2 may be of the NOx type but it may also, for example, be of the lambda type.

The constitutive signal of the richness model and the measured signal of richness are carried as input of a drift identifier 3 which identifies and quantifies the possible dispersions and drifts of wealth on stabilized operating points.

The drift identifier 3 thus allows the realization of a possible learning drift resulting in the establishment of a mapping of these wealth drifts in the engine field.

In this way, the drift identifier 3 recalculates the calculated model of richness with respect to any dispersions and drifts using the measured signal of richness which has the advantage of being precise in stabilized regimes.

More precisely, the value of the calculated richness carried by the signal constituting the wealth model is corrected continuously by said drift learning carried out by the drift identifier 3, these drifts being preferentially recorded in a map and then read without discontinuity with interpolation.

However, the value of richness obtained at the output of the drift identifier 3 may be only incompletely recalibrated due to example of incomplete learning, or approximations inherent to the storage method.

Advantageously, in order to solve this problem, the correction of the calculated wealth Φ_Calc can be advantageously completed in order to refine the final value of said richness of the cylinder, by means in particular of a Kalman filter 4 which makes it possible to obtain a complementary dynamic identification of the wealth drift.

avec

• x̂ _k : l'estimation a priori du vecteur d'état à l'instant k ;
• x̂ _k : l'estimation a posteriori du vecteur d'état à l'instant k ;
• u _k-1 : le vecteur de variables déterministes ; en l'espèce, il s'agit de la richesse calculée dans la chambre de combustion.

The general equation of the Kalman filter is as follows, the equations to be considered are more particularly the equation (1) of the table «Update Time» and the equation (2) of the table «Update Measure ".

with

• x _k : the prior estimate of the state vector at time k;
• x _k : the posterior estimation of the state vector at time k;
• u _{k- 1} : the vector of deterministic variables; in this case, it is the calculated wealth in the combustion chamber.

• le signal de richesse modèle 02 de la sonde ; et
• le signal mesuré de richesse 02, sensor par la sonde 2.

Two signals are applied to the input of the Kalman filter:

• the model 02 richness signal of the probe; and
• the measured signal of richness 02, sensor by the probe 2.

Each of these signals is given relative confidence by introducing process noises w and measurement v.

The first filter processing step consists in applying to the model richness signal 02 the alteration that it would undergo in the event that it would be picked up by the probe 2 of the NOx type.

• un retard pur dû au transport des gaz d'échappement du cylindre vers la sonde 2, auquel s'ajoute le retard pur de la sonde 2 elle-même ; et
• un filtre (du 1^er ou 2^eme ordre par exemple).

This alteration consists of:

• a pure delay due to the transport of the cylinder exhaust gases to the probe 2, to which is added the pure delay of the probe 2 itself; and
• a filter ^(1st or ^2nd order, for example).

This alteration is only virtual because the probe 2 analyzes not this model signal 02, but the real signal of richness which differs from the previous one by the drift of wealth θ _{O 2} to identify.

ou $O_2^{\mathit{ff}}$

(et retardée, mais en amont de l'équation d'état). On crée ainsi le vecteur d'état 'a priori' de la structure du Kalman (se reporter à l'équation du vecteur d'état ci-dessous).The Kalman filter therefore creates a signal 'filtered computed wealth' $O_2^f$

or ${\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">O</figure-callout>}}_2^{\mathit{ff}}$

(and delayed, but upstream of the state equation). This creates the 'a priori' state vector of the Kalman structure (see the equation of the state vector below).

The second step of processing the Kalman filter, which is performed a posteriori, consists in comparing this virtual signal with the signal actually picked up and analyzed by the wealth probe 2. The difference between these two signals defines the instantaneous wealth drift θ _{O 2} .

However, the behavior (generally characterized by the delay and attenuation) of the probe 2 can not be perfectly known. he It is therefore necessary to limit the trust in this drift identification so as not to make it too random in the event of errors (even transient).

This limitation of confidence is obtained by means of the Kalman gain, which forces a convergence time before considering the fully identified drift. The finally estimated wealth signal is thus the model signal 02 recalibrated by the measurement of the probe 2 with a time of integration, linked to the necessarily limited confidence in the model of the behavior of the wealth probe 2.

avec :

• θ _{O 2} : la dérive identifiée, soit l'écart entre la richesse mesurée et le signal de richesse calculée passée par le modèle de comportement de la sonde ;
• O ₂ : la richesse calculée (modèle) ;
• $O_2^f$
  
  : la richesse calculée filtrée (ordre 1) afin de tenir compte du comportement de la sonde ;
• ${\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">O</figure-callout>}}_2^{\mathit{ff}}$
  
  : la richesse calculée filtrée (ordre 2) pour tenir compte du comportement de la sonde ;

The equation of state, which presents in this case four dimensions (assuming that the model of behavior used was in 2nd order), is as follows: $$\boxed{\begin{array}{l}{\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^{\mathrm{<figure-callout\; id="2"\; label="f\; O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">f\; </figure-callout>}}& \\ {O_{}}^{}{{}_2}^{\mathit{ff}}\end{array}\right]}_k=\left[\begin{array}{cccc}1+\mathrm{at}& -\mathrm{at}& 0& 0\\ 1& 0& 0& 0\\ 0& b& 1-b& 0\\ 0& 0& \mathrm{vs}& 1-\mathrm{vs}\end{array}\right]\times {\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^{\mathrm{<figure-callout\; id="2"\; label="f\; O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">f\; </figure-callout>}}& \\ {O_{}}^{}{{}_2}^{\mathit{ff}}\end{array}\right]}_{k-1}+\left[\begin{array}{c}\mathrm{at}\\ 1\\ 0\\ 0\end{array}\right]\cdot {\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">O</figure-callout>}}_{2,{\mathit{model}}_{k-1}}+w_{k-1}\\ {\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">O</figure-callout>}}_{2,{\mathit{sensor}}_k}=\left[0001\right]\cdot {\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^{\mathrm{<figure-callout\; id="2"\; label="f\; O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">f\; </figure-callout>}}& \\ {O_{}}^{}{{}_2}^{\mathit{ff}}\end{array}\right]}_k+v_k\end{array}}$$

with:

• θ _{O 2} : the drift identified, ie the difference between the measured richness and the calculated richness signal passed by the model of behavior of the probe;
• O ₂ : calculated wealth (model);
• $O_2^f$
  
  : calculated computed wealth (order 1) to take into account the behavior of the probe;
• ${\mathrm{<figure-callout\; id="2"\; label="O"\; filenames="imgf0001.png"\; state="\{\{state\}\}">O</figure-callout>}}_2^{\mathit{ff}}$
  
  : calculated computed wealth (order 2) to take into account the behavior of the probe;

• O _{2,mod el k-1} : la richesse calculée (modèle) à l'instant k-1 ;
• O _{2 ,sensork} , : la richesse mesurée par la sonde à l'instant k ;
• W _k-1 : le bruit du signal modèle 'richesse calculée' ;
• v_k : le bruit du signal mesuré de richesse par la sonde.

The vector composed of these four variables constitutes the state vector.

• O _{2, mod el k -1} : calculated wealth (model) at time k-1;
• O _{2 , sensor k} , : the richness measured by the probe at time k;
• W _{k -1} : the signal noise model 'calculated wealth';
• v _k : the noise of the measured signal of richness by the probe.

According to an alternative embodiment of the invention, the establishment of this state vector could be achieved by using a model of probe behavior with greater variability, or by not showing the said model, the latter being then defined upstream.

In this example of proposed implementation, the Kalman gain resulting from these calibration compromises is unique and pre-calculated, for the sake of simplicity of implementation.

However, according to an alternative embodiment of the invention, at each computation step, measurement noise and process noise, as well as the resulting optimal Kalman gain (infinite gain Kinf) can be updated.

According to another variant embodiment of the invention, the Kalman gain can be mapped by distinct zones of the motor field.

Advantageously, stabilized point drift learning makes it possible to accelerate the work of the Kalman filter, to make it more robust, and even to ensure that the result provided is accurate.

The combination of the Kalman filter and the stabilized learning technique ensures a high degree of adaptability. Indeed, the wealth drift, which depends first and foremost on component drifts such as the flowmeter and the injectors, is not expected to be uniform in the engine field, but variable from one point to another in transient. A Kalman filter alone, which by definition takes a significant amount of time to integrate a drift and converge there, would see its dynamic of identification and delivery. up to date conflict with the dynamics of the operating point and its associated wealth drift. The filter would thus keep - at least partially - a specific drift to the previous point whereas this one would already have radically changed on the following point. Failing to immediately integrate the new drift, that retained and applied would possibly be shifted, and could even distort the correction to aggravate the estimate rather than improve it.

The stabilized learning is freed from the model of behavior of the probe 2, insofar as it no longer intervenes on a stationary point (the two input signals being constant, there is no longer any attenuation and the cylinder-probe delay is inconsistent). A mapping of the wealth drift can thus be established (based for example on the torque-regime points), the latter obviously being dependent on the stabilized points actually encountered on the road. By carrying out an interpolated reading in this mapping, the skilled person is then able to pre-position the wealth drift of the current point, and to limit the remaining work of the Kalman filter, while minimizing the possible impact of drift history preservation.

• ▪utiliser la dérive lue dans la cartographie évolutive pour corriger directement le signal d'entrée de richesse modèle O₂ du filtre de Kalman 4 ;
• ▪ utiliser la dérive lue pour l'état 'a priori' du filtre de Kalman 4, permettant un pré-positionnement permanent plus ou moins abouti (selon le niveau de remplissage de la cartographie), qui est alors corrigé 'a posteriori' via le signal mesuré de richesse O_{2, sensor} par la sonde 2 et le gain de Kalman.

There are more ways to combine this mapping and the Kalman filter, these consist of:

• ▪use the drift read in the evolutionary cartography to directly correct the model O ₂ richness input signal of the Kalman 4 filter;
• ▪ use the drift read for the 'a priori' state of the Kalman 4 filter, allowing a permanent pre-positioning more or less successful (depending on the filling level of the map), which is then corrected 'a posteriori' via the measured signal of richness O _{2, sensor} by probe 2 and Kalman gain.

• le signal de richesse finale estimée (pointillés gras) est calé sur la dynamique du modèle de richesse modélisée (trait continu fin), mais recalé vis-à-vis de sa dérive au moins partiellement, sur des points transitoires comme sur des points stabilisés (dérive observée par le Kalman (flèches verticales), entre le signal sonde (mesure de richesse, trait continu en gras) et le signal virtuel en noir pointillé fin).

As shown in FIG. 3, the resulting identification of the wealth drift is carried out as follows:

• the estimated final richness signal (bold dots) is calibrated on the dynamics of the model of wealth modeled (fine continuous line), but recaled with respect to its drift at least partially, on transient points as on stabilized points ( drift observed by the Kalman (vertical arrows), between the probe signal (measurement of richness, continuous line in bold) and the virtual signal in fine dotted black).

• de procéder à une identification adaptative des dérives, et non d'une calibration fixe réalisée sur table valable uniquement sur un point de fonctionnement et/ou un seul moteur ;
• de procéder à une estimation de la richesse cylindre en bénéficiant à la fois des avantages d'une mesure obtenue au moyen d'une sonde, constitués par l'affranchissement des dérives et dispersions, mais également des avantages procurés par la simplicité du calcul à effectuer qui permettent un recalage dynamique de la richesse ;
• de lier le modèle d'émissions de particules au signal final obtenu, en s'assurant d'une grande robustesse en transitoire comme en stabilisé ; cela permet donc d'améliorer la performance et la robustesse de l'estimation des émissions de particules et celle du chargement associé du filtre à particules ;
• de pouvoir adapter la stratégie de recalage dynamique de la richesse à toutes sortes de ligne d'échappement ;
• de se dispenser en fonctionnement Diesel, de l'utilisation d'une sonde Lambda spécifique ; la sonde NOx dont la présence est règlementairement nécessaire en aval de la ligne d'échappement permet d'obtenir l'information richesse désirée, ce qui permet d'effectuer des économies.

Advantageously, the combination of the use of the drift identifier 3 and the Kalman filter 4 makes it possible in particular:

• to carry out an adaptive identification of the drifts, and not a fixed table calibration valid only on an operating point and / or a single engine;
• to make an estimation of the richness cylinder benefiting from the advantages of a measurement obtained by means of a probe, constituted by the franking of the drifts and dispersions, but also the advantages provided by the simplicity of the computation to carry out which allow a dynamic registration of wealth;
• to link the model of emissions of particles to the final signal obtained, making sure of a great robustness in transient as in stabilized; this therefore makes it possible to improve the performance and the robustness of the estimation of the emissions of particles and that of the associated loading of the particulate filter;
• to be able to adapt the strategy of dynamic registration of wealth to all sorts of escape routes;
• to dispense with diesel operation, the use of a specific Lambda probe; NOx probe whose presence is required by regulation downstream of the exhaust line provides the desired wealth information, which allows for savings.

Claims (7)

1. A method enabling continuous estimation of air-fuel ratio in an internal combustion engine's cylinder, said engine comprising a plurality of cylinders, at least one richness measurement sensor (2), a plurality of injectors connected to the cylinders, a flow meter, said method including the steps consisting of:

   • measuring a richness signal by means of the richness sensor (2).

   • calculating a simple richness model (Φ_Calc) corresponding to the quotient of the flow of injected fuel (qlnj) on the flow of intake air (Qair), multiplied by the stoichiometric ratio (γ),

   characterized in that said method further includes the steps consisting of:

   • identifying and quantifying the possible dispersions and deviations of richness on stabilised points of operation by means of at least one deviation identifier (3) from the signal constituting the richness model (Φ_Calc) and from the measured richness signal carried at the input of this deviation identifier (3), a mapping of the richness deviations in the engine field being realized by the deviation identifier (3), by teach-in of these possible deviations;

   • identifying and quantifying the possible transient dispersions and richness deviations by means of at least one supplementary deviation identifier consisting of a Kalman filter (4) from the measured richness signal and from the model richness signal of the sensor,

   • correcting continuously the calculated richness (Φ_Calc) by using the deviation read in the mapping for directly correcting the model richness input signal of the Kalman filter (4) or by using the deviation read in the mapping for the a priori status of the Kalman filter (4).
2. The method according to Claim 1,
   characterized in that the richness sensor (2) is of the NOx type.
3. The method according to Claim 1 or Claim 2,
   characterized in that the first processing step of the Kalman filter (4) consists in applying to the first model richness input signal (O₂ ) the virtual alteration which it would undergo if it were picked up by the sensor (2), this alteration consisting in:

   • a dead time due to the transport of the exhaust gases of the cylinder towards the sensor (2), to which is added the dead time of the sensor (2) itself;
   and

   • a filter;

   a "filtered calculated richness" signal (O₂ ^f or O₂ ^ff ) and a vector of "a priori" status are created according to the following equations, assuming that the model of behaviour used is of the 2^nd order: $$\boxed{\begin{array}{l}{\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^f\\ {O_2}^{\mathit{ff}}\end{array}\right]}_k=\left[\begin{array}{cccc}1+a& -a& 0& 0\\ 1& 0& 0& 0\\ 0& b& 1-b& 0\\ 0& 0& c& 1-c\end{array}\right]\times {\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^f\\ {O_2}^{\mathit{ff}}\end{array}\right]}_{k-1}+\left[\begin{array}{c}a\\ 1\\ 0\\ 0\end{array}\right]\cdot O_{2,{\mathit{model}}_{k-1}}+w_{k-1}\\ O_{2,{\mathit{sensor}}_k}=\left[0001\right]\cdot {\left[\begin{array}{c}{\theta }_{O_2}\\ O_2\\ {O_2}^f\\ {O_2}^{\mathit{ff}}\end{array}\right]}_k+v_k\end{array}}$$

   

   with:

   θ_o2 : the identified deviation, namely the gap between the measured richness and the calculated richness signal passed by the behaviour model of the sensor;

   O₂ : the calculated richness (model);

   O₂ ^f : the filtered calculated richness (order 1) so as to take into account the behaviour of the sensor;

   O₂ ^ff : the filtered calculated richness (order 2) so as to take into account the behaviour of the sensor;
   the vector composed of these four variables constitutes the status vector.

   O_{2,model k-1} : the calculated richness (model) at the moment k-1;

   O_{2,sensor k} : the richness measured by the sensor at the moment k;

   w_k-1 : the noise of the "calculated richness" model signal;

   v_k : the noise of the signal measured for richness by the sensor;

   each of the input signals (O₂) and signal measured for richness (O _2,sensor) by the sensor (2) comes to assign a relative trust by introducing process noises (w) and measurement noises (v).
4. The method according to Claim 3,
   characterized in that the second processing step of the Kalman filter (4), which is realized a posteriori, consists in comparing the virtual signal with the signal actually picked up and analysed by the richness sensor (2); the gap between these two signals defines the actual richness deviation (θ_o2 ).
5. The method according to one of Claims 3 and 4,
   characterized in that a limitation of the trust in the identification of the deviation is carried out by means of the Kalman gain (4), which forces a convergence time before considering the fully identified deviation; the finally estimated richness signal is therefore the model signal (O₂ ) reset by the measurement of the sensor (2) averaging an integration time, linked to the necessarily limited trust in the behaviour model of the richness sensor (2).
6. The method according to one of Claims 3 to 5,
   characterized in that the combination of the mapping obtained and of the Kalman filter (4) is carried out by:

   • using the deviation read in the evolving mapping to correct directly the model richness input signal (O₂) of the Kalman filter (4);

   • using the deviation read for the "a priori" status of the Kalman filter (4), permitting a permanent pre-positioning more or less accomplished, according to the filling level of the mapping, which is then corrected "a posteriori" via the measured richness signal (O_{2, sensor} ) by the sensor (2) and the Kalman gain.
7. An apparatus enabling continuous estimation of air-fuel ratio in an internal combustion engine's cylinder, said engine comprising a plurality of cylinders, at least one richness measurement sensor, a plurality of injectors connected to the cylinders, a flow meter,
   characterized in that it includes a deviation identifier (3) and a Kalman filter (4) adapted to implement the method or methods according to any one of the preceding claims.

Images