DE10244539A1 - Global-adaptive correction of injection quantities/air mass measurement errors in combustion engine involves polynomial-based recursive learning technique with current engine operating information - Google Patents

Abstract

The method involves determining correction values for a characteristic field at at least three operating points and defining a correction plane from operating points and correction values. At least one control signal correction value is derived from operating parameter input values during engine operation and is learned using a polynomial-based recursive learning technique using current engine operating point information. The method involves placing values for at least one control signal for the internal combustion engine in a characteristic field depending on at least two operating parameters, whereby correction values for the characteristic field are determined at at least three operating points and a correction plane is defined from operating points and correction values. Each point of the correction plane is a possible correction value. At least one control signal correction value is derived from operating parameter input values during engine operation and is learned using a polynomial-based recursive learning technique using current engine operating point information. AN Independent claim is also included for the following: (a) a control unit for controlling an internal combustion engine.

Description

BACKGROUND THE INVENTION

The invention relates to a method and a control unit for controlling an internal combustion engine in particular a motor vehicle according to the preambles the respective independent Expectations.

Today's automotive engines are becoming increasingly popular higher Requirements for exhaust gas and consumption parameters posed. In order to meet these requirements, the pollutant emissions are these Internal combustion engines in a manner known per se by means of exhaust gas recirculation, i.e. by supplying exhaust gas to the intake air. Naturally point the sensors and actuators used there, such as an air mass meter or a fuel metering manufacturing or work tolerances which means that the best possible emission values are not always achieved to let. In order to optimally reduce the mentioned intervention options of using pollutants, it is actually required amount of fuel injected into the engine or actually that Air mass supplied to engine preferably to be precisely determined or any necessary correction of an operating parameter of the internal combustion engine, for example the amount of fuel injected.

In the correction mentioned known Values of one of at least two operating parameters of the internal combustion engine dependent Control signal of the internal combustion engine in the form of a map stored. In addition Correction values at several operating points of the internal combustion engine determined to correct the map. Based on the deviation between a desired one Value and an actual The value of one of the at least two operating parameters is given in each case Correction values are determined by at least three operating points and the assigned correction values at least one correction level is defined and each point of the correction plane serves as a possible correction quantity.

A corresponding method for the adaptive correction of characteristic diagrams of an injection system of a motor vehicle diesel engine is shown, for example, in DE 195 28 696 A1 out. The method described there uses an exhaust gas recirculation controller with a map in which - depending on the operating parameters of the diesel engine - a control signal for controlling an actuator is stored. The actuator influences the amount of air sucked in via an exhaust gas recirculation flap. The exhaust gas recirculation controller processes the output signal of a speed sensor and a lambda sensor and / or an air mass meter. If an error or an inaccuracy occurs in the fuel metering, the amount of fuel actually injected deviates from the amount of fuel desired by a driver, for example by means of an accelerator pedal sensor. If the actual quantity is less than a target value, the internal combustion engine does not deliver the desired torque. If the amount is too large, inadmissible exhaust gas emissions may occur. The quantity of fuel actually injected is therefore determined by means of a quantity calculation and compared at a comparison point with the target value. A correction quantity is then determined depending on the comparison result. With this correction quantity, the setpoint for the injected fuel quantity is corrected accordingly in an addition point.

A similar method for corrective control of an internal combustion engine is also known from the DE-OS 41 05 740 known. In the method described there for additive and multiplicative correction of a characteristic diagram, a first deviation signal represents an additive and a second deviation signal represents a multiplicative error. These additive and multiplicative errors are taken into account by additive and multiplicative correction factors in the entire map area. Such a "global correction" only provides sufficiently precise correction values if the map is divided into areas in which the additive errors dominate and in areas in which the multiplicative errors dominate.

It is also known to be local To optimize corrections to the above-mentioned maps, that to every map base a correction value is learned. Although the accuracy of the corrections across from the above methods is improved, it is disadvantageous that certain points in an operating cycle of the internal combustion engine already could be learned while at neighboring points for this has not been an opportunity since certain operating points are only relative rarely be approached. For example, it takes a relatively long time until after a refueling process the changed fuel properties taken into account in the entire map are.

The procedures above are therefore disadvantageous insofar as not in all operating points the internal combustion engine a global adaptive correction, for example the Amount of fuel injected and / or by a sensor recorded air mass enables becomes. Accordingly, any errors that occur in these operating parameters of the internal combustion engine insufficiently corrected.

SUMMARY THE INVENTION

The invention is therefore an object Reasons to further develop a method mentioned at the outset and a control unit in such a way that the disadvantages of the prior art are avoided and in particular the most accurate determination possible and thus the precise correction of the fuel quantity and / or air mass actually supplied to an internal combustion engine is made possible.

This task is solved by the characteristics of the independent Expectations. Advantageous refinements or developments are the subject of respective subclaims.

The method according to the invention and the control unit see to the solution this task in particular, temporarily during operation of the internal combustion engine or at certain intervals from operating parameters of Representing internal combustion engine Input signals a correction factor for at least one control signal to calculate the internal combustion engine and this correction factor with a recursive learning with the help of information about the respective current operating point of the internal combustion engine in one Save polynomial. From this polynomial, a momentarily valid in each case Correction factor with high precision be determined.

The correction factor is preferred dependent on of the currently injected fuel mass, the speed and if necessary, further operating parameters of the Internal combustion engine determined. Using the recursive learning process the correction factor is learned very quickly. Through further learning the accuracy of the correction factor can be increased even further. Thus, also for Operating points of the internal combustion engine in which no learning has yet taken place a correction factor can also be output. The method according to the invention requires only relatively short computing times and a minimal one Memory requirement for the temporary storage of the processed sizes.

In a preferred embodiment of the method according to the invention the exact amount of fuel injected from the Internal combustion engine supplied Air mass and the oxygen content in the exhaust gas (air value LAMBDA) determined. there the correction factor is a difference or a quotient between a target injection quantity and actually injected fuel amount formed.

According to an alternative embodiment actually fed to the internal combustion engine Air mass calculated from the fuel mass and the air value LAMBDA. The correction factor is calculated from the measured air mass and the calculated air mass.

The invention also allows for different Operating points of the internal combustion engine determined correction factors to save and from these stored correction factors, in particular in dynamic operation of the internal combustion engine, a global adaptive provide the determined correction factor.

The proposed recursive learning method has the further advantage that in all operating points the Internal combustion engine is learned, not only on the above Reference points. Thereby there can be adverse effects due to local effects at the support points, minimize local quantity errors, for example.

It is also advantageous that you can go through the use of the state vector mentioned receives a curved correction surface which adapts better to the real error than in the prior art known methods.

The invention is described below under Using the drawing, based on preferred embodiments explained in more detail from which further features and advantages of the invention result.

Show in the drawing

1 schematically a recursive learning method integrated into a control circuit of an engine control unit of an internal combustion engine according to the invention for the adaptive control of operating parameters of the internal combustion engine;

2 a schematic detailed view of the in the 1 shown recursive learning process;

3 a typical correction level for adaptive correction of the in 1 Operating parameters shown according to a first embodiment of the invention;

4 a linearly curved correction surface according to a second embodiment of the invention; and

5 a square curved correction surface according to a third embodiment of the invention.

DESCRIPTION THE PREFERRED EMBODIMENTS

The 1 shows schematically a flow control loop of an engine control unit of an internal combustion engine of a motor vehicle. The quantity control loop is used for metering fuel and air supplied by the internal combustion engine and is operated according to the invention by means of a recursive learning method. The actually injected fuel mass ME is calculated from the air mass ML supplied to the internal combustion engine, in the present case detected by means of an air mass meter (not shown) and from the air mass ratio LAMBDA measured by means of a lambda probe (not shown). With this quantity calculation, the actually injected fuel mass ME is derived from the formula
ME_ist = ML / (14.4 * LAMBDA).

As air quantity signal ML can either the amount of air detected with an air mass sensor or one out various operating parameters such as For example, the temperature and the pressure of the intake air volume calculated Size used become.

From the difference between the actual value the fuel mass ME_ist and one from the engine control unit provided target fuel mass ME_soll results in the fuel mass error DELTA_ME. The fuel mass error DELTA_ME, the speed N of the internal combustion engine and the target fuel mass ME_soll now go into said and recursive learning methods described in more detail below on.

It is already anticipated at this point that the advantage of the recursive learning process described below in particular lies in the fact that there is learning in all working points, not only because of the discretely available values of the operating parameters of the Support points defined by the internal combustion engine. This allows adverse effects due to local effects at the support points, For example, minimize local quantity errors in fuel metering.

As a recursive learning process in all Exemplary embodiments described below consistently the recursive least-squares parameter estimation method used. This method is described in the literature, for example in "Digital Control Systems", Rolf Isermann, published by Springer-Verlag, 1987, chap. 242, on that full reference is made in the present context, in detail described.

The variables used in this method and their relationship with the operating parameters of the internal combustion engine are in the 2 illustrated and concern:

1. the fuel mass error already mentioned DELTA_ME, namely at the respective sampling time of the quantity control loop;

2. a state vector PSI; his Elements are the speed N, the target fuel mass ME_soll and a constant one Factor (again at the sampling time of the quantity control loop);

3. a covariance matrix P, which the inverse product of a state matrix PSI with that in formula 24.2.7 of the PSI-T matrix defined in the textbook above, this state matrix PSI from the vectors mentioned in 2 PSI for all sampling times is formed;

4. an information vector THETA, whose elements are factors that make up an estimate of the fuel mass error can be formed;

5. a correction vector GAMMA, which Correction values for calculating a new value of THETA from a contains old value of THETA;

6. a forgetting factor OMEGA (cited above with "LAMBDA"), which a factor to the stronger Weighting of new sampling points compared to old sampling points.

At each sampling time, the Sizes DELTA_ME and PSI newly formed. Initially, PSI and P become new ones Correction vector and then from PSI, GAMMA, and the forgetting factor OMEGA a new matrix P formed.

From the available values from PSI, GAMMA, THETA and DELTA_ME will then become a new information vector THETA educated. THETA contains then one depending on the speed first factor THETA1, a second one dependent on the fuel mass Factor THETA2 and a constant factor THETA3 for calculation the correction value required at each operating point of the internal combustion engine for the Fuel mass. This relationship defines a correction level, the first two factors being the slope of the plane towards the speed and the amount of fuel and the third factor one represent constant offset.

The exact relationships are in the 3 illustrated, in which a first embodiment of the method according to the invention is shown. The mathematical relationship therefore applies to the correction level mentioned
ME_korr = N * THETA1 + ME_soll * THETA2 + THETA3

The recursive learning process will in the present embodiment only activated when the change dynamics the fuel mass, namely the time derivative DELTA_ME_soll / DELTA_t of the setpoint ME_soll, one does not exceed a certain value. Dynamic negative influences on the learning process or adaptive correction, e.g. very short-term changes (peak values) of the Internal combustion engine supplied Fuel mass, minimized.

From the information vector THETA and the state vector PSI becomes the correction value of the fuel mass ME_korr calculated according to the relationship ME_korr = THETA * PSI. This correction value then becomes additional functions made available in the engine control unit, for example the functions Fuel metering, start of injection and / or exhaust gas recirculation. Thereby it is ensured that the learning process also has a correction value for the Fuel mass outputs when the learning process is not activated is, e.g. in the case of an inactive lambda sensor.

The covariance matrix P and the information vector THETA will be in a non-volatile Memory, for example an EEPROM. This will stand at the start the internal combustion engine immediately the sizes necessary for the calculation of ME_korr Available.

With reference to the 4 and 5 Two further exemplary embodiments of the invention will now be described, which differ in particular by modifications of the underlying recursive learning method.

The one in the 4 In the embodiment shown, the state vector PSI is expanded by further elements, in the present case by the product of the speed of the internal combustion engine and the fuel mass. The correction value for the fuel mass results from the following equation:
ME_korr = N * THETA1 + ME_soll * THETA2 + THETA3 + N * ME_soll * THETA4

The one in the 5 In the exemplary embodiment shown, the state vector PSI is expanded by the speed in the square and the temperature of the internal combustion engine. With this expansion, the correction plane becomes a curved surface or a correction space. The advantage of this curved surface is that a better adaptation of the correction value of the fuel mass to the real fuel mass error of the internal combustion engine can be achieved. This is because the curved correction surface obtained by using a state vector PSI with more than three elements can be better adapted to the real error.

The correction value for the fuel mass results from the equation:
ME_korr = N * THETA1 + ME_soll * THETA2 + THETA3 + N * N * THETA4

In other versions, the described in detail in Chapter 24.6 of the above textbook Modifications of recursive parameter estimation algorithms are used for example the discrete filtering method in covariance form (DSFC). Here, the covariance matrix P is the product of the matrices S and the S_T replaced. Alternatively, U-D factorization (DUDC} where P is the product of the matrices TT, D and U_T is replaced. By means of these modifications, achieve better numerical properties in digital calculation.

In an additional variant, instead of a correction value for the fuel mass, a correction value for the air mass ML_korr is calculated. In this case, the air mass ML_act actually flowing into the combustion chamber of the internal combustion engine is calculated from the value of the fuel mass ME provided by the control unit of the internal combustion engine and the oxygen content in the exhaust gas LAMBDA measured by means of a lambda probe, according to the formula:
ML_ist = ME * 14.4 * LAMBDA

The difference between this calculated air mass and the air mass measured by means of an air mass meter the air mass error DELTA_ML. This error now goes in the same way the recursive learning process.

Another variant becomes the correction factor for the fuel mass not as a difference, but as a quotient ME_soll and ME_is formed. The same applies in this variant for the air mass correction.

The recursive learning method described above can generalize to all conceivable quantity or quantity replacement signals of the internal combustion engine applied, e.g. on the needle stroke / spray duration, the fuel pressure, the torque, and / or the air and exhaust gas mass flows.

The recursive learning method described above is preferably one in an electronic engine control unit Internal combustion engine of a motor vehicle in the form of a control program realizable.

Claims (17)

Method for controlling an internal combustion engine, in particular of a motor vehicle, in which values of at least one control signal of the internal combustion engine are stored in a map, depending on at least two operating parameters of the internal combustion engine, correction values for correcting the map being able to be determined at at least three operating points, the correction values being based on of the deviation between a desired value and an actual value of one of the at least two operating parameters, and wherein at least one correction level is defined by at least three operating points and the assigned structural values and each point of the correction level serves as a possible correction variable, characterized in that in operation the internal combustion engine at predeterminable time intervals from the input signals representing at least two operating parameters, at least one correction value for the at least one control signal of the Br Internal combustion engine is calculated and the at least one correction value is learned by means of a polynomial-based, recursive learning method, using information about the respective current operating point of the internal combustion engine. A method according to claim 1, characterized in that as the recursive learning method, the recursive method of the smallest Error squares is used. A method according to claim 2, characterized in that the reject at each sampling time The fuel mass (DELTA_ME) and a state vector PSI are newly formed, a new correction vector being formed first from PSI and a covariance matrix P and then a new covariance matrix F from a state matrix PSI, a correction vector GAMMA and a forgetting factor OMEGA, and the existing ones Values of PSI, GAMMA, THETA and DELTA_ME, a new information vector THETA is formed, which has a first factor THETA1 which is dependent on the speed, a second factor THETA2 which is dependent on the fuel mass and a constant factor THETA3 for calculating the correction value required for each operating point of the internal combustion engine contains the fuel mass. A method according to claim 3, characterized in that in the correction plane the factors THETA1 and THETA2 the slopes the plane in the direction of the speed and the amount of fuel of the internal combustion engine and the factor THETA3 represent a constant offset. Method according to one of claims 2 to 4, characterized in that that a correction factor for that Injection quantity according to the relationship ME_corr = N * THETA1 + ME_soll * THETA2 + THETA3 is calculated. A method according to claim 5, characterized in that from the information vector THETA and the state vector PSI a correction value of the fuel mass ME_korr according to the relationship ME_korr = THETA * PSI is calculated. Method according to one of claims 2 to 4, characterized in that that the state vector PSI is the product of the speed of the internal combustion engine and fuel mass is expanded, the correction value for the fuel mass according to the context ME_korr = N * THETA1 + ME_soll * THETA2 + THETA3 + N * ME_soll * THETA4 is calculated. Method according to one of claims 2 to 4, characterized in that that the state vector PSI squared around the speed and the temperature of the Internal combustion engine is expanded, taking from the correction level a curved one area or a correction room is formed. A method according to claim 8, characterized in that the correction value for the fuel mass according to the context ME_korr = N * THETA1 + ME_soll * THETA2 + THETA3 + N * N * THETA4 is calculated. A method according to claim 1, characterized in that as a recursive learning method the discrete filtering method in Covariance form (DSFC) is used, the covariance matrix P is replaced by the product of the matrices S and the S_T. A method according to claim 1, characterized in that as a recursive learning method, the method of U-D factorization (DUDC) is used, the covariance matrix P being the product of Matrices U, D and U_T is replaced. A method according to claim 1 or 2, characterized in that that a correction value for the air mass ML_korr is calculated, using the fuel mass ME and the oxygen content measured by means of a lambda probe in the exhaust gas (air value LAMBDA) according to the relationship ML_ist = ME * 14.4 * LAMBDA which is actually in the combustion chamber of the internal combustion engine flowing Air mass ML is being calculated. A method according to claim 12, characterized in that from the difference between the calculated air mass and the an air mass error measured by means of an air mass meter DELTA_ML is calculated, which goes into the recursive learning process. Method according to one of claims 3 to 13, characterized in that that the correction factor for the fuel mass or air mass as a quotient of ME_soll and ME_ is formed. Method according to one of the preceding claims, characterized characterized that the recursive learning process is only activated if the time derivative DELTA_ME_soll / DELTA_t of the setpoint ME_should not exceed a certain value. Method according to one of the preceding claims, characterized characterized that the covariance matrix P and the information vector THETA can be cached in a permanent memory. Control unit for controlling an internal combustion engine, in particular a motor vehicle, in which values of a control signal are stored in a memory in a map, depending on at least two operating parameters of the internal combustion engine, with means which determine correction values for correcting the map at at least three operating points, the correction values , can be determined on the basis of the deviation between a desired value and an actual value of one of the at least two operating parameters, and by means which define at least one correction level by means of at least three operating points and the assigned correction values and each point of the correction level as a possible correction Use tur size, characterized by means for calculating a correction factor for at least one control signal of the internal combustion engine by means of a recursive learning method according to one of the preceding claims.