EP2195519A1

EP2195519A1 - Engine state parameter estimation comprising the measurement of the internal pressure of a cylinder

Info

Publication number: EP2195519A1
Application number: EP08837795A
Authority: EP
Inventors: Sébastien CASTRIC; Vincent Talon
Original assignee: Renault SAS
Current assignee: Renault SAS
Priority date: 2007-10-12
Filing date: 2008-08-19
Publication date: 2010-06-16
Anticipated expiration: 2028-08-19
Also published as: FR2922262B1; EP2195519B1; FR2922262A1; WO2009047412A1

Abstract

The invention relates to a system for estimating the state parameters of an internal combustion engine (2), including: at least one cylinder (3), a moving piston (4) driven by means of a crankshaft (6), means (20, 21) for measuring temporal variations in the angle of the crankshaft and the internal pressure of the cylinder, at least one physical model (231) for calculating a plurality of intermediate temporal variables from said crankshaft angle and cylinder internal pressure measurements and from a measurement of at least one state parameter of the engine, a means (24) for creating tables of discretised temporal variables from the intermediate temporal variables, and a learning model (28) for estimating at least one state parameter of the engine (2) from said tables of discretised temporal variables.

Description

Estimation of engine status parameters by measuring the internal pressure of a cylinder

The invention relates to the estimation of state parameters of a rotary internal combustion engine comprising a plurality of cylinders.

Anti-pollution standards and declining fuel consumption are becoming increasingly important issues for car manufacturers. It is therefore necessary to control the consumption of vehicles, therefore carbon dioxide emissions, while rejecting as little as possible of polluting gases such as oxides of nitrogen NO _x , carbon monoxide CO, unburned fuels HC and particulates, especially for diesel engines. This passes necessarily by a perfect control of the combustion. It is therefore necessary to know the magnitudes of various parameters of the state of the engine, which are for example:

filling η _v (amount of fresh gas admitted into the cylinders) - the driving torque

- the engine speed

- phasing of fuel injections φ _ιnj

- the mass of fuel introduced for each injection M _inj

the temperature at the outlet of the cylinder T _avt - the pressure at the outlet of the cylinder P _avt

- NO _x oxide emissions

- CO carbon monoxide emissions

- emissions of unburnt HC fuels

- Particle emissions Thanks to the estimation of these quantities it is possible to control the motor in closed loop, that is to say to optimize at every moment the control of the engine thanks to the knowledge of its state. As is known, many sensors are generally embedded in vehicles for the measurement of a plurality of magnitudes and state parameters, which causes a significant cost in the manufacture of vehicles. For example, reference may be made to the European patent application published under the number EP 1 367 248, in which an estimation of the emission of nitrogen oxides NO _x is carried out on the basis of a mathematical model called "virtual sensor". And a plurality of engine status parameter measurements. But the implementation of this model requires the measurement of many state parameters.

Reference may also be made for example to the US patent application published under the number US 2004/0073381, in which is described a means for estimating the emission of NO _x nitrogen oxides from the correlation between a plurality of measurements of engine control parameters and a table of static magnitudes based on an average of the nitrogen oxide temperatures. But this means also requires the measurement of many state parameters and has the disadvantage of using static tables which are expensive in design time. The aim of the invention is therefore to provide a system for controlling the state parameters of an engine that makes it possible to respond to the needs mentioned above and, in particular, to propose a system for estimating these state parameters which enables to remove some sensors such as that of the pressure at the cylinder outlet or upstream of the turbine of a turbocharger P _avt or the temperature upstream of the turbine T _avt .

These estimators also make it possible to know the pollutant emissions at each moment, in order to deduce the rate of emission of nitrogen oxides NO _x . It is therefore advantageous to remove the nitrogen oxide sensor. This control of emissions also makes it possible to envisage the reduction of the volumes of the systems of postprocessing of these emissions.

Another object of the invention is to provide a system for estimating the state parameters of an engine which makes it possible to to avoid the preliminary design of many engine tuning maps and thus significantly reduce the development time.

In one embodiment, the system estimates at least one state parameter of an internal combustion engine comprising: at least one cylinder and a movable piston driven through a crankshaft; means for measuring the time variation of the crankshaft angle and the internal pressure of said cylinder; at least one physical model for calculating a plurality of intermediate time variables from said measurements of the crankshaft angle and cylinder internal pressure and from a measurement of at least one engine state parameter; means for creating time variable tables discretized from said intermediate time variables; and a learning model for estimating at least one engine state parameter from said discrete time variable tables.

The learning model can be, for example, of the neural network type, the statistical type or the type of kriging. Advantageously, the measurement of the internal pressure of the cylinder can be carried out by means of a pressure sensor. It should be noted that each cylinder may be equipped with such a pressure sensor or, more simply, only one of the engine cylinders.

Advantageously, the system comprises means for initializing the learning model by carrying out preliminary tests.

In another aspect, a method for estimating at least one state parameter of an internal combustion engine, comprising at least one cylinder and a movable piston driven through a crankshaft, comprises the following steps : A first step of measuring the temporal variation of the crankshaft angle and the internal cylinder pressure; a second step of calculating, through at least one physical model, a plurality of intermediate time variables from said crankshaft angle and pressure measurements; internal cylinder and from a measurement of at least one engine state parameter; a third step of discretizing said intermediate temporal variables, intended for creating tables of discrete temporal variables; and a fourth step of estimating, through a learning model, at least one engine state parameter from said discrete time variable tables.

Other objects, features and advantages of the invention will appear on reading the following description, given solely by way of nonlimiting example, and with reference to the appended drawing, in which, the figure illustrates a system according to the invention. invention for estimating certain state parameters of an engine.

The figure schematically shows a system for estimating the state parameters of an internal combustion engine 2. The internal combustion engine 2 comprises a cylinder 3 in which a piston 4 moves by means of an internal combustion engine 2. a connecting rod 5 connecting the piston 4 to the crankshaft 6. A combustion chamber 7 is delimited by said cylinder 3, said piston 4 and a cylinder head 8. The cylinder head 8 is provided with at least two valves 9 and 10 which make it possible to connect the combustion chamber 7 with respectively the intake manifold 9a, for air optionally mixed with a part of the exhaust gas, and the exhaust gas manifold 10a. The engine 2 also comprises a fuel injector 11 arranged to inject fuel into the combustion chamber 7. The estimation system comprises two measurement sensors 20 and 21 as well as an electronic calculation unit 22 comprising three modules : a calculation module 23, a discretization module 24 and an estimation module 25.

The sensor 20 makes it possible to measure at any moment the angle θ of the crankshaft 6, the sensor 21 makes it possible to measure the internal pressure P _cy ι of the cylinder 3 which corresponds to the pressure inside the combustion chamber 7. These sensors 20 and 21 each emit a temporal measurement signal, transmitted respectively by the connections 20a and 21a, in the direction of the electronic calculation unit 22.

The calculation module 23 comprises several physical models 231 to 237 which make it possible to calculate a certain number of intermediate time variables from the input time signals θ,

P _cy ι and from the measurements of certain state parameters of the motor 2, brought by the connection 22a to the input of the calculation module 23.

These intermediate temporal variables thus calculated are transmitted by leads 26 to the input of the discretization module 24. The intermediate time variables may be, for example, cylinder temperature T _cy u heat release Q, the mass fraction of gas burned X _b , the mass of liquid fuel M _car bu _q and vaporized M _car b _vap , the mass of fresh gas M _gf and flue gas M _g b, the rate of burned gas X _g b, or the polytropic coefficient k .

The state parameters of the motor 2 brought by the connection 22a are, for example, parameters such as the engine speed, the fuel injection _timing φ _ιnj or the fuel mass introduced for each injection M _ιnj . These are variables distinct from the calculated intermediate temporal variables.

The intermediate temporal variables are discretized in the module 24 to generate tables of discrete temporal variables. This discretization of the signals takes place at precise moments for certain measurements of angles θ of the crankshaft 6.

The estimation module 25 receives these tables of temporal variables discretized by the connections 27 in order to estimate the desired state parameters, such as, for example, the filling η _v or the temperature at the outlet of the cylinder T _avt . The internal pressure P _cy ι of the cylinder 3 thus makes it possible to construct intermediate temporal variables in order to deduce from it certain parameters of the state of the engine 2. This construction of the temporal variables is carried out by means of models 231 to 237 which are based only on temporal variables, excluding any space variable.

As previously described, the models 231 to 237 receive as inputs the variables P _cy u θ and certain state parameters brought by the connection 22a. It is also possible that a physical model can use as input a plurality of intermediate temporal variables, brought by the connections (30), which are the result of a calculation made by another model, thus increasing the number of computation combinations. intermediate variables.

All these intermediate temporal variables, which depend on the measurement of the variable P _cy ι and the angle θ of the associated crankshaft 6, are calculated only when the valves 9a and 10a are closed, that is to say one when no gas transfer is effected in or from the combustion chamber 7.

In the figure, there is shown by way of example, some physical models 231 to 237 of the calculation module 23. But this module

23 may include a greater number of models, the example given is not limiting. The calculations made by the different physical models are described below:

Physical model 231: calculation of the cylinder temperature in the combustion chamber 7Vw.

The cylinder temperature can be calculated thanks to the ideal gas law:

Or :

P _Cy i is the internal pressure of the cylinder - V _cy ι is the volume of the cylinder

M _t is the total mass enclosed in the cylinder r is the constant of the perfect gases, r = 287 J / kg.K The total mass M _t can be determined by mapping according to the engine speed and the pressure of the intake manifold 9a.

The volume of the cylinder V _cy ι is determined by measuring the angle θ of the crankshaft 6. An analytical law makes it possible to determine V _cy ι as a function of θ:

K _yl [θ] = V _1n + S _pι ^• R _n + L _bι - R _n ^• cos (0) - 4L _b ? - Rj - ^ή n {θ) ²

Or :

V _1n is the dead volume S _pι is the surface of the piston R _vι is the radius of the crankshaft Lbi is the connecting rod length The variable T _cy ι thus calculated is transmitted directly to the discretization module 24.

Model 232: Calculation of heat release Q. Heat release Q represents heat exchanges between the gas and the outside, during chemical reactions that take place during the combustion phase of the fuel. That is, it represents the sum of the heat released by combustion minus the heat lost at the walls. Q is calculated as follows:

dQ c dV

. . γ dP

.. γ ^c yyl IY ⁱ p dθ γ - l P cyl yl aυ γ - \ cy v ^{c ≠} d "θ γ - - \ c 1 ^{c ≠ d}

Where: θ is the angle of the crankshaft γ is the ratio of the specific heats C _P IC _V where C _p and C _v are the mass heats, respectively at constant pressure and volume

In general, we take γ equal to 1, 4. The Q variable thus calculated is transmitted as input to the physical models 233, 234 and 235, as well as to the discretization module 24.

Model 233: calculation of the mass fraction of burnt gases Xt ₁ . The mass fraction of burnt gases X _b evolves during combustion. An image of X _b can be obtained by the release of heat Q. In fact the heat released is proportional to the mass of fuel burned. So the integral of heat Q is directly related to the mass of fuel already burnt. This integral is normalized between 0 and 1. It represents then the evolution of the combustion. It is called X _b .

X _b - Norm dQ dθ

Where ω is the angular velocity of the motor in radians per second.

The variable Xb thus calculated is transmitted directly to the discretization module 24.

Model 234: calculation of the mass of liquid fuel M _r .arh_n _q and vaporized M _rnr h van. Amongst the state parameters of the motor transmitted by the connection 22a, it is possible to use the phasings of each fuel injection φ _in] , the mass of fuel introduced for each injection M _nj , as well as the duration of each injection T _nj . Thanks to these parameters, it is possible to reconstruct the injection rate in the cylinder 3. This makes it possible to calculate:

^ - ^ - carb hq ~. ~. -T ^ - = Qm _ιnj - Qm _vap vap _ dM carb

Q m = _vap - Qm _comb

^~ dt Or :

_Ιnj Qm is the average fuel flow rate injected qm _vap is the vaporized fuel flow rate Qm means _com b is the fuel flow combustion means also:

Where PCI is the lower heating value (about 43500 kJ / lcg for diesel).

The average flow rate of combustion being directly proportional to the heat release Q, it can be calculated from the previous physical model 232.

At any moment, the mass of liquid and vaporized fuel present in the combustion chamber 7 can thus be known.

The variables M _because bjiq and M _car b_vap thus calculated are transmitted directly to the discretization module 24.

Model 235: calculation of the mass of fresh gas M ^ and of burnt gas Mg _J1 . The flue gases have two origins: one part (called EGR) is the partially recycled exhaust gases from the exhaust manifold 10a to the intake manifold 9a, another part (called GBR) are the residual gases of the preceding cycle which have not been drained. At the initial instant when the inlet valve 9 is closed, is known initial mass of fresh gas M _gf _ _t, mixture of oxygen and nitrogen, and the initial mass of burnt gas M _g ,, b_ mixture of carbon dioxide, water and nitrogen. The evolution of the fresh gases towards the flue gases is dependent on the heat generation Q. In fact the combustion always takes place locally with the richness 1, that is to say that when one burns 1 gram of fuel one burns on average 14.7 grams of fresh gas. We have seen that:

Is

Qm ^ = PCO - Qm _conb

Where: - PCO is the burning power (about 14.7)

Qwig _f .gb is the average flow of air Which allows to calculate:

dM SI

^~ dT = -0 * ^» « gf - gb

dM _g dt - - Qm g ₁₁ fT - gb

The variables M _{g /} and M _g b thus calculated are transmitted as inputs to the physical model 236, as well as to the discretization module 24.

Model 236: Calculation of the burnt gas ratio X ^, diluent used to reduce NO ₂₁ emissions.

This burnt gas content X _g t is the proportion of flue gas present at the closure of the intake valve 9 with respect to the total mass M _t enclosed in the cylinder 3.

y ^M Φ ^M Φ

M _t M _gf + M _g

The variable X _g b thus calculated is transmitted directly to the discretization module 24. Model 237: calculation of the polytropic coefficient in compression phase k.

In the compression phase, between the closing of the inlet valve and the beginning of the combustion, the polytropic coefficient k is calculated. This coefficient k makes it possible in particular to model the adiabatic transformations (no exchange of heat and matter with the outside) thanks to the following formula:

p, y ^k _ A rcyl ^V cyl ^{~ Λ}

Where A is a constant

The value of k is identified throughout the compression phase thanks to the formula:

Where: θ, is the angle of the current crankshaft - Aθ is the calculation interval This calculation interval Aθ may correspond to at least one sampling step of the internal pressure signal of the cylinder P _cy ι as a function of the angle θ of the crankshaft 6. In general, the interval is taken from the order of 10 sampling steps of said signal. The variable k thus calculated is transmitted directly to the discretization module 24.

After having determined the intermediate temporal variables described above, they are discretized via the module 24 in order to establish discretized variable tables. Said tables obtained are the inputs of the estimation module 25. This module 25 comprises learning models 28 which can be of the neural network type, as illustrated in the figure, or statistics or of the kriging type.

It is interesting to use kriging models which are interpolation models using stochastic methods that allow a calculation of probabilities applied to statistical data processing.

There are three types of kriging: simple kriging, ordinary kriging and universal kriging. The difference between these types of estimation lies in knowing the statistics of the variable to be interpolated:

- simple kriging: the stationary variable has a known mean;

- ordinary kriging: the stationary variable has an unknown mean; - universal kriging: the variable is non-stationary.

Preferably, an ordinary kriging will be used. In general, kriging is based on the correlation between the variables that one wishes to estimate and the discretized variables that are the inputs of the model. In the ordinary kriging method, the estimation of a variable can be written in the following form:

Where: y is the estimate of the variable in a non discretized point xo

X ₁ represent the discretizations of the variables

- y (xι) are the values associated with the discretizations X ₁ - X ₁ are the kriging coefficients

The X ₁ are the tables of discretized variables obtained by the module 24, the y (x _t ) are the values of the variables which one wishes to estimate, like for example the filling η _v . The principle of kriging is to determine the X ₁ coefficients, which are dependent on the X ₁ , by studying the degree of similarity between the y (X ₁ ) from the covariance between the points x, as a function of the distance between these points. . The weights X ₁ associated with each of the values y (xι) are chosen so as to obtain a prediction y of minimum variance.

To identify the model, it requires prior testing. The estimation of new points is then based on these tests. The learning models 28 are thus previously identified on tests carried out on the engine test bench or on the vehicle.

Claims

System for estimating at least one state parameter of an internal combustion engine (2) comprising at least one cylinder (3), a movable piston (4) driven via a crankshaft ( 6), means (20, 21) for measuring the time variation of the crankshaft angle and the internal pressure of said cylinder, characterized in that it comprises: at least one physical model (231) for calculating a plurality of variables intermediate time intervals from said measurements of crankshaft angle and cylinder internal pressure and from a measurement of at least one engine condition parameter; means (24) for creating time variable tables discretized from said intermediate time variables; and a learning model (28) for estimating at least one engine state parameter (2) from said discrete time variable tables.

2. System according to claim 1, comprising a learning model (28) of the neural network type.

3. System according to one of claims 1 to 2, comprising a learning model (28) of the statistical type.

4. System according to one of claims 1 to 3, comprising a krigeage learning model (28).

5. System according to one of claims 1 to 4, comprising means for initializing the learning model (28) by performing prior tests.

Method for estimating at least one state parameter of an internal combustion engine (2) comprising at least one cylinder (3), a movable piston (4) driven via a crankshaft ( 6), comprising a first step of measuring the temporal variation of the crankshaft angle and the internal cylinder pressure, characterized in that it comprises: a second calculation step, via at least one physical model (231), a plurality of intermediate temporal variables from said measurements of the crankshaft angle and the internal pressure of said cylinder and from a measurement of at least one engine state parameter; a third step of discretizing said intermediate temporal variables, intended for creating tables of discrete temporal variables; and a fourth step of estimating, via a learning model (28), at least one engine state parameter (2) from said discrete time variable tables.

The method of claim 6, wherein the estimating step comprises using a neural network learning model (28).

8. Method according to one of claims 6 to 7, wherein, the estimation step comprises the use of a learning model (28) of the statistical type.

9. Method according to one of claims 6 to 8, wherein, the estimation step comprises the use of a krigeage learning model (28).

10. Method according to one of claims 6 to 9, wherein, the estimation step comprises a step of initialization of the learning model (28) by performing prior tests.