DE19547496A1 - System for determining exact air induction of IC engine - Google Patents

Abstract

A system for the exact determination of the amount of inducted air entering the cylinders of an IC engine is used as the basis for metering the correct volume of fuel for the promotion of efficient combustion. A monitoring regime with a learning facility ensures comprehensive control under both static and dynamic operating conditions. The principles of the method can be applied to take into account the effect of both linear and non-linear influences in technological situations other than the IC engine and for these applications use is made of neural networks in support of the monitoring function.

Description

The invention relates to methods for controlling internal combustion engines, in for example, a real-time estimate of the air mass in the cylinders of a ver internal combustion engine is used as the basis for metering the fuel. Real time Estimation here means that an immeasurable quantity can only be determined using ver current or simultaneous measured values of other sizes is determined so that the Estimated value is obtained simultaneously with the real value.

In practice, the air mass sucked into the combustion chambers is acceptable not measurable, but must be known in order to correctly meter the fuel to enable a desired by adjusting the air ratio λ accordingly Operating behavior of the internal combustion engine is achieved.

By real-time estimation of the air mass in the cylinders of a combustion engine tors, which includes both static and dynamic operating states for example, get a reliable default value for fuel injection.

With the proposed method, the air mass in the Combustion chambers can be determined to be both static and dynamic Operation of the internal combustion engine to ensure the desired mixture ratio.

The gasoline engine is chosen here as an example. The same applies analogously to the die selmotor or other combustion devices.

In the case of a gasoline engine, a measurable air mass flow _enters the intake manifold at the throttle point, from which the cylinders suck _{out the} air mass flow _out (load), which cannot be measured directly, during the individual intake cycles of the cylinders. As a result of the memory behavior of the finite Saugrohrvolumens the measurable air mass can flow at the throttle point in the suction pipe _in only stationary equal to the sucked air mass flow to be _out. However, the stationary operation of a gasoline engine in motor vehicles is the exception.

The invention enables the dynamic and stationary estimation of the air mass flow sucked _out into the combustion chambers, which cannot be measured directly. By integrating this air mass flow over the time of the intake strokes of the individual cylinders, the air mass located in the individual cylinders can be determined so that fuel can be metered in the desired manner to ensure safe operation. The air mass flow (load estimate) is estimated in real time, i.e. while the engine is running, and is independent of specimen scatter and aging influences.

Variables: the air mass flow at the throttle in the intake manifold _in and the thermodynamic states are compressed to estimate p and temperature T performed in the intake manifold, and requires possibly n is the engine speed, or given if the throttle position. Only the intake pipe volume and possibly the total displacement D must be known as geometry sizes. The methods presented for estimating the air masses in the cylinders can be transferred to other tasks, such as in throttle valve control for estimating the non-linear influences in a throttle valve moved by a servo drive or for estimating the fuel quantities entering the cylinder. This method can also be used to estimate non-linear influences such as friction in an electromechanical valve train in internal combustion engines or, for example, in an automatic transmission.

Assessment of the state of the art (example: air mass estimation in the cylinders)

In engine control devices of the prior art, either the mass flow method or the speed-sealing method is used for gasoline engines to estimate the intake air mass flow. In the Massenflußmethode is the (possibly not line ar filtered) air mass flow rate used in the draft tube _in as an estimate for the actual load, but only the stationary suction pipe due to the permits memory behavior ei ne correct estimate; this is therefore not suitable for dynamic operating states. When using an off-line, possibly non-linear filter to correct the storage behavior, the specimen scatter and / or parameter drift makes a dynamically correct estimate difficult or prevented. As a result of the storage behavior, this method shows strong deviations in the estimated air mass flow from the air mass actually entering the cylinders in dynamic operation.

The speed sealing method, on the other hand, uses the dependence of the load on the Suction behavior of a motor describing volumetric efficiency, an effect degree, which depends on the speed and the thermodynamic conditions in the intake manifold depends. This dependency is usually determined off-line, and in the control unit filed as a map, resulting in the same problems with regard to specimen litter and parameter drift result as with the mass flow method. With this method no dynamic error is generated by the storage behavior of the intake manifold, however, the quality depends on the exact knowledge of the volumetric efficiency.

Common to both methods is that the load cannot be correctly estimated either only in dynamic or in both stationary and dynamic operation. For the best possible conversion of the pollutants NO _x , HC and CO in a three-way catalytic converter, however, the stoichiometric mixture ratio is necessary over the entire operating range of a gasoline engine, so that a method for correctly estimating the air mass in the cylinders (load estimation) is required.

Representation of the procedure according to the application and devices for its implementation

The following two methods of load estimation represent a unique combination of the two known methods mentioned above, which are now both stationary as well as in dynamic operation provide a correct estimate and the observer Use the principle.

To correctly estimate the air mass in the cylinders in dynamic operation to take into account the storage behavior of the finite volume of the intake manifold.  

This is done as follows: The mass balance equation is used to estimate the intake air mass flow exploited

_in - _out = (1),

this means that the difference between the inflowing and outflowing mass flows of a Vo lumens represents the time derivative of the mass in this volume. Be Air mass flows are sought here, so that via the equation of state of the ideal Gas

pV = mRT (2)

this air mass together with the temperature T the pressure inside the intake manifold determined in this.

If the pressure change due to the temperature change is neglected, what when considering a gasoline engine intake manifold due to the slow gradients is permissible (the partial derivation can be neglected), so the amendment tion of the intake manifold pressure is a measure of the difference in the air mass flows, such as by differentiating equation (2) over time

will be shown. The pressure change due to the temperature change should not be neglected a correction term must also be taken into account, which results from the differentiation rules result in:

This physical relationship can be broken down into two fundamentally different approaches apply to the estimation, which are shown below. Are for verification each simulation results that have been validated.

Load estimation through differentiation of the intake manifold pressure

The first method is based on the differentiation of the intake manifold pressure. From the relationship (2), differentiation according to time gives an estimate for the air mass flow _out into the cylinders, neglecting the slow and relatively small temperature change

This approach is shown in Fig. 2 in the signal flow diagram. If the time derivative of the intake manifold temperature is not to be neglected, then proceed according to equation (4); the signal flow diagram according to FIG. 3 then results.

To estimate the air mass flow into the cylinders with this approach, it is necessary to entering the intake manifold volume mass air flow _into the intake manifold pressure p and measure the temperature T in the intake manifold. In Fig. 4, the observer result obtained online by means of this differentiation approach is shown, dashed the air mass flow at the intake manifold inlet, drawn through the air mass flow in the cylinder and dotted the observed air mass flow in the cylinder. In this scaling, the dotted line is almost congruent with the solid line, that is, the air mass flow into the cylinders is very well estimated, assuming that the pressure signal is noise-free.

For technical implementation, there is a calculation rule in an engine control unit to implement, which thus the measured variables intake manifold pressure and air mass flow processed in the intake manifold. The estimate of the air mass flow into the cylinders is obtained as follows:

• - The intake manifold pressure p is differentiated
• - the weighted with the ratio of the quotient of differentiated intake manifold pressure and intake manifold will flow from the inflow into the intake manifold air mass _in subtracts
• - This difference represents the estimated value for the air mass flow into the cylinders _out

This procedure is shown in the signal flow chart Fig. 2. If the time derivation of the intake manifold temperature T should not be neglected, a correction term corr is to be added to the estimated value that has just been obtained

calculated. This procedure is shown in the signal flow chart Fig. 3.

The estimated value for the air mass flow must be integrated over the time of the intake stroke of each individual cylinder in order to obtain the air mass in the respective cylinders. If the change in temperature is also to be taken into account, the technical implementation must be carried out in accordance with the signal flow diagram in FIG . Otherwise the above comments apply.

The first method uses the time derivative of the intake manifold pressure as the measured variable p, this is problematic due to the measurement noise. Therefore, the following is a second method described, which is a correct estimate of the air mass flow allowed into the cylinders without differentiation.

For this purpose, an observer is designed that enables the intake manifold pressure p to be exceeded to conclude a parameter adaptation to the air mass flow.

Estimate the air mass flow into the cylinders "Volumetric Efficiency" Observer Approach

The differentiation of the intake manifold pressure can be avoided by choosing a special observer approach. Again, the mass balance equation (1) is used. As an example of a possible observer approach, the air mass sucked out of the suction pipe is to be understood as a disturbance variable which interferes with the pressure build-up inside the suction pipe. Based on this approach, a disturbance observer can be designed if the dependence of the air mass flow drawn in by the engine on the intake manifold pressure and the speed is taken into account at the same time. This dependency is referred to as "volumetric efficiency" and describes the suction behavior of the engine in the dependencies mentioned as a dimensionless efficiency between zero and one. Influences such as acoustic resonances, choked-flow effect on the intake valves and valve overlaps etc. are summarized in this volumetric efficiency. The volumetric efficiency is the main cause of the hump-shaped torque-speed curve in internal combustion engines. When considering engines with variable valve timing, the volumetric efficiency also depends on the phase angle of the timing. Using the volumetric efficiency η _vol , the air mass flow drawn in by the engine can be determined using the speed-density method as follows (half speed due to the four-stroke process assumed here):

With this knowledge, the signal flow plan for the intake manifold can be drawn together with the suction cylinder as in Fig. 5. Here too, the time derivation of the intake manifold temperature is neglected; if it is to be taken into account, the equation (4) must be followed.

In relation (7) the quantities n, p and T can be measured, the volumes D (stroke space) and V are known and constant, as is the gas constant R. The volumetric Efficiency is unknown.

The aim must therefore be to design an adaptive disturbance variable observer who can learns volumetric efficiency correctly and can therefore correctly estimate the load.

Simple observer approach

In FIG. 6, a potential observer approach is a nonlinear disturbance observer shown as an example in which real-time operating the current η _vol learns and thus a correct estimate of the _out supplies the sucked air mass flow. For this purpose an adaptation integrator is used, the output of which simulates the respective η _vol , the adaptation takes place via the observer error. This is the "simple" observer approach because, in contrast to the observer approach described below, the use of artificial intelligence is still dispensed with.

In this approach, the stability can be demonstrated using linear methods if the measured variables speed n and temperature in the intake manifold T are assumed to be constant. In FIG. 7, the result is shown observer, the estimated air mass flow is in this scaling almost congruent with the actual air mass flow into the cylinders.

For technical implementation, an observer calculation rule, for example according to the signal flow chart Fig. 6, to implement in an engine control unit, which thus processes the sizes intake manifold pressure and temperature, air mass flow into the intake manifold and engine speed.

A possible calculation rule can determine the estimated value for the air mass flow _out as follows:

• - The reality "intake manifold" is represented in the observer as a parallel model
• - The real integration of the air mass flow diff weighted by the factor reference is simulated by a delay element of the first order, the Output simulates the course of the intake manifold pressure
• - By the additional connection of the intake manifold pressure weighted by the factor λ to the input of this delay element, the delay element reproduces the real integrating behavior when the estimated and real intake manifold pressure match, this is illustrated in FIG. 8
• - The error between the curves of the real and estimated intake manifold pressure arises from the mismatched air mass flow _differences , that is to say from the incorrect estimate for the air mass flow flowing into the cylinders _-out , since the input value _in is the same for reality and observer
• - The air mass flow _out flowing into the cylinders depends on the relationship 7 of speed and density via the so-called volumetric efficiency η _vol , whose operating point-dependent influence can be simulated in the observer by a multiplication with an integrator output
• - The input of this integrator is a learning step size ν, with the Be care error between real and estimated intake manifold pressure that the resulting adaptation makes the observer error zero (active pulmonary stability of the observer)
• - If the observer error between the real and estimated intake manifold pressure is zero, the delay element acts as an integrator with the air mass flow difference as the only input by applying its intake manifold pressure weighted by the factor λ, see Fig. 8
• - so the observer structure is equal to reality, which means that at over matching route structure and matching route exits (suction pipe pressure and estimated value) also the route inputs (difference of air mass flows) are the same
• - Since the difference between the air mass flows of the observer is formed from the measured _in and the estimated _out -, the estimated air mass flow _{out is} therefore equal to the real air mass flow into the cylinders _out .

The estimated value for the air mass flow must over the time of the intake stroke Every single cylinder can be integrated to the air mass in the respective cylinder to maintain.  

A disadvantage of this observer approach is the forgetting of the knowledge that has been learned, since the parameter η _{vol of} the old working point, which is reproduced by the integrator output, is forgotten again when operating in another working point. The result of this is that an error will always arise due to learning by the observer in the case of working point shifts.

This can be avoided if the results of the learning processes in the map get saved.

Another approach that also records the learned results in the map can be achieved through the processes of artificial intelligence; this is the third Procedure - with the associated equipment - shown in the following chapter.

Observer approach with a neural network

The aim of this observer approach is to estimate the air mass flow in the zy linder by learning the volumetric efficiency of the engine using the method of artificial intelligence. This forms the basis for the prediction of the air mass flow mes.

For this purpose, a neural network can be set up, for example, as an input dimension has all sizes on which the volumetric efficiency depends (suction pipe pressure, speed, possibly valve timing. . . ). Instead of a neural network but also any adaptive approach that allows a static not to be used to represent linearity over a one- or multi-dimensional entrance space.

A General Regression Neural Neural Network is used here as an example. This neural network, hereinafter referred to as GRNN, is used in the Observer approach described above the influence of real volumetric Ef to learn efficiency and make it available as knowledge. For better understanding the mode of operation of this special neural network should be explained very briefly den: The (multidimensional) entrance space is in receptive fields of many neurons divided up. These neurons become dependent on the distance of the input status of the GRNN variously more or less activated and then carry through it respective weight according to their activation to the initial value of the GRNN this input state at. The neurons represent integrators, which in their effects mainly work within their receptive field. This means, that when using a GRNN the adaptations described in the above approach integrator is replaced by many piece-wise integrators. In principle can however, any other adaptive approach can be used that allows a static Represent non-linearity over a one- or multi-dimensional entrance space.

The extended signal flow plan in the Laplace area by the GRNN is shown in FIG. 9. The basic function of this observer approach using GRNN can be realized, for example, by the method described in Az: 195 31 962.4 and the associated device. In the present case, if the dynamics of the intake manifold pressure are not negligible, the strictly positive real condition, which was also mentioned in Az: 195 31 962.4, can be circumvented by a special structure of the neural network. This special structure of a neural network is characterized in that the network has internal dynamics, that is to say it has several states in addition to the weight integrators, in order to ensure a correct correlation between the error signal and the network input generating the error signal.

The observer result of this approach using artificial intelligence is shown in FIG. 10, the learning begins at time t = 0 without prior knowledge. This figure shows the observer's settling process, a normal process for all observations. This oscillation arises from the learning process in constantly new operating points in the speed-intake manifold pressure phase level, through which the volumetric efficiency is learned. After the learning process, the observer result will be congruent with the air mass flow to be estimated.

For technical implementation, an observer calculation rule, for example according to Si gnalflußplan Fig. 9, to implement in an engine control unit, which thus processes the sizes intake manifold pressure and temperature, air mass flow into the intake manifold and engine speed. A possible calculation rule can use a neural network with a special structure (here general regression neural network) to determine the estimated value for the air mass flow _out as follows:

• - The reality "intake manifold" is represented in the observer as a parallel model
• - The real integration of the air mass flow diff weighted by the factor reference is simulated by a delay element of the first order, the Output simulates the course of the intake manifold pressure
• - By the additional connection of the intake manifold pressure weighted by the factor λ to the input of this delay element, it is achieved that the delay element simulates the real integrating behavior when the estimated and real intake manifold pressure match, this is illustrated in FIG. 11
• - The error between the curves of the real and estimated intake manifold pressure arises from the mismatched air mass flow _differences , that is to say from the incorrect estimate for the air mass flow flowing into the cylinders _-out , since the input value _in is the same for reality and observer
• - The air mass flow _out flowing into the cylinders depends on the relationship 7 of speed and density via the so-called volumetric efficiency η _vol , whose non-linear dependence on the variables speed and intake manifold pressure in the observer by multiplying with the output of a neural network these two inputs can be replicated
• - The weights of the neurons of this neural network are with a Learning step size ν due to the observer error corresponding to the activation of the individual neurons at the various working points so that the the resulting adaptation makes the observer error zero (control tech stability of the observer)
• - If the observer error between the real and estimated intake manifold pressure is zero, the delay element acts as an integrator with the air mass flow difference as the only input by applying its intake manifold pressure weighted by the factor λ, see Fig. 8
• - so the observer structure is equal to reality, which means that at over matching route structure and matching route exits (suction pipe pressure and estimated value) also the route inputs (difference of air mass flows) are the same
• - Since the difference between the air mass flows of the observer is formed from the measured _in and the estimated _out -, the estimated air mass flow _{out is} therefore equal to the real air mass flow into the cylinders _out .

The estimated value for the air mass flow must over the time of the intake stroke Every single cylinder can be integrated to the air mass in the respective cylinder to maintain. If valve timing can be influenced (VANOS, other valve train than via a camshaft) this knowledge must also be used, for example as additional input dimension in the neural network.

Air mass flow prediction

The approach presented by a learnable observer provides the basis for predicting the air mass flow into the cylinders, which means that it is possible to deduce the air mass flow solely from the intake manifold states of pressure and temperature together with the speed (possibly phase angle of the intake valves). This is shown in FIG. 12, which estimates the load solely on the basis of the knowledge previously learned, without knowing the air mass flow flowing into the intake manifold. In this procedure, the signal flow plan according to FIG. 8 can be simplified, for example, by measuring the pressure p, temperature T and speed n, the nonlinear characteristic fields learned by the GRNN, and permanently implemented as characteristic fields. As a result, the cost of implementation can be reduced considerably.

This fact can be used to measure the air mass in the cylinders to predict by moving from the throttle position angle to the flowing into the intake manifold Air mass depending on the thermodynamic conditions in the near future is closed so that the intake manifold condition is predicted, from which then on the Load can be closed in the near future.

Air mass flow control

The approaches presented by observers provide the basis for regulating the air mass flow into the cylinders. After learning, for example, the non-linear pump behavior of the cylinders η _vol , this information is available in order to specify the setpoint value of the intake manifold pressure through which a desired air mass flow into the cylinders can be achieved at a predetermined speed. In this way, the air mass in the combustion chambers can be specifically adjusted via an intake manifold pressure control.

Further application examples

As already noted in Chapter 1, the methods and Facilities also for other tasks with the internal combustion engine or on their technical facilities are transferred.

As described in the previous chapters, the observers did the An apply a load estimate to the intake manifold of an internal combustion engine. The result The load estimate can, for example, be interpreted as a setpoint or as a disturbance variable will. Depending on the interpretation, the information can be passed on in the engine control be working. For example, if the load estimate is a disturbance variable in a process is interpreted, then according to the prior art the disturbance variable by ei ne feedforward control can be suppressed. In the same way, after State of the art undesirable influences are interpreted as disturbance variables.

This perspective allows the application of the method according to the application to egg A large number of technical facilities with non-linear repercussions within of the process under consideration should be identified or compensated for.

If, in addition to the procedures described above, the throttle valve control is treated, then it is known that the control of the throttle valves position is difficult to achieve due to non-linear repercussions. The reason for the difficulty are nonlinear effects like the nonlinear spring constant that Friction effects and the moments on the throttle valve due to the air flow into it Intake manifold. If the influence of these effects could be estimated and the result de signal of these effects is used for feedforward control, then remains an easy to control route. The goal must therefore be to see the effects in the to estimate the effective sum.

Based on this idea, suitable observers can be designed to estimate the impact. A structure with artificial intelligence, such as, for example, with General Regression Neural Network in FIG. 8, is suitable as a universal observer - that is, observers for linear, non-linear and multidimensional relationships . In FIG. 8, the volumetric efficiency η _{vol was} estimated using the GRNN approach . In the same way, it is possible to estimate the undesirable influences and subsequent feedforward control in the throttle valve. In another application - the fuel path - the fuel entering the cylinders can be estimated in the same way when the fuel is injected into the intake manifold.

The fields of application are also said to be from the automotive technology sector to an electromechanical valve train and an automatic gearbox be called.

Figure index

1 Schematic representation of the intake manifold of a gasoline engine,
2 Representation of the first approach described for estimating the air mass flow into the cylinders in the signal flow diagram. The time derivative of the intake manifold pressure p is required.
3 representation of the first described approach for estimating the air mass flow into the cylinders, taking into account the time derivative of the intake manifold temperature,
4 Result with the differentiation approach described above, section, dashed the air mass flow at the intake manifold inlet, the air mass flow drawn into the cylinders and dotted the observed air mass flow into the cylinders,
5 Alleviate the signal flow diagram of the air path in the frequency range with the intake manifold and Zy
6 Signal flow diagram of a possible non-linear observer approach for load estimation. The unknown parameter η _vol , depending on intake manifold pressure and speed, is simulated in all working points of the engine by a single integrator. The parameters λ and ν in the circle of observers can be used to set the dynamics of the observer
7 Result with the simple observer approach described above, detail, dashed the air mass flow at the intake manifold inlet, the air mass flow drawn into the cylinders and the observed air mass flow dotted into the cylinders. Here, too, the lines drawn and dotted are almost congruent, so a correct estimate is made. The advantage of this approach compared to the previous approach is that no measurement signal has to be differentiated,
8 clarification of the observer principle,
9 Signal flow diagram of a possible non-linear observer approach with GRNN to simulate the dependence of the volumetric efficiency on the intake manifold pressure and the speed. In the case of adjustable valve timing, this dependency must be taken into account by an additional input dimension of the GRNN,
10 Result with the learnable observer approach described above without any prior knowledge, detail, dashed the air mass flow at the intake manifold inlet, the air mass flow drawn into the cylinder and dotted the observed air mass flow into the cylinder. The oscillation of the observer at each individual operating point in the speed-intake manifold pressure phase plane is clearly recognizable,
11 clarification of the observer principle,
12 Result with the above observer approach, evaluation solely on the basis of already learned knowledge, no online adaptation, detail, dashed line the air mass flow at the intake manifold inlet, the air mass flow drawn into the cylinders and the observed air mass flow dotted into the cylinders,
13 The neural network comprises internal states (dynamics H₂ (s)) for the correlation between the treasure value and the network input.

Claims (84)

1. A method for controlling internal combustion engines, characterized in that the air mass flow sucked in by the cylinders is calculated by a process observer described in detail in the following claims in such a way that a desired mixing ratio is achieved with the fuel to be supplied. 2. Process for the control of internal combustion engines, characterized in that the time change of the intake manifold pressure is used to estimate the load. 3. Process for the control of internal combustion engines, characterized in that the mass balance equation (1) is used in the process observer. 4. The method according to claims 1 and 2 or 3, characterized in that the process observer differentiates the intake manifold pressure (equation 3, Fig. 2 or equation 4, Fig. 3), for example in analog signal processing. 5. The method according to claims 1 and 2 or 3, characterized in that the Process observer processes the change in intake manifold pressure p over time, for example in digital signal processing. 6. The method according to claims 1 and 2 or 3, characterized in that the Process observer the temporal change in the size intake manifold air temperature T processed. 7. The method according to claims 1 and 2 or 3 and 4 and 5, characterized records that the process observer the changes in the intake manifold pressure quantities p and intake manifold air temperature T processed as input variables. 8. The method according to claims 1-5 and 7, characterized in that the process observer works in principle as follows:

• - The intake manifold pressure p is differentiated
• - the weighted with the ratio of the quotient of differentiated intake manifold pressure and intake manifold temperature is subtracted from the inflow into the intake manifold air mass flow _in
• - This difference represents the estimated value for the air mass flow into the cylinders _out
• - This estimate for the air mass flow into the cylinders _out is integrated via the intake stroke of the individual cylinders and thus results in the air mass in the individual cylinders

Changes in signal processing using mathematical transformations Part of the claim. 9. The method according to claims 1-7, characterized in that the process observer works in principle as follows:

• - The intake manifold pressure p is differentiated
• - the weighted with the ratio of the quotient of differentiated intake manifold pressure and intake manifold temperature is subtracted from the inflow into the intake manifold air mass flow _in
• - A correction term formed from the measured variables T and p is added
• - this sum represents the estimated value for the air mass flow into the cylinders _out
• - This estimate for the air mass flow into the cylinders _out is integrated via the intake stroke of the individual cylinders and thus results in the air mass in the individual cylinders

Changes in signal processing using mathematical transformations Part of the claim. 10. The method according to claims 1-5 and 6 and 8 or 7 and 9, thereby characterized in that the process observer values of different sizes too saves or caches various times in the sense of a look-up table to reduce the effort of signal processing (e.g. sizes and T). 11. The method according to claims 1-10, characterized in that character variables such as V or D are time-varying. 12. The method according to claims 1-11, characterized in that, for example, se instead of measuring the air mass flow _in other sizes such as the throttle valve angle can be used as an observer input to save effort. 13. The method according to claims 1-10, characterized in that the process watcher the sizes of the air mass flow into the intake manifold _in, intake manifold pressure p and temperature T as measured or estimated sizes proces tet in the intake manifold. 14. A method for controlling internal combustion engines, characterized in that, based on the speed-sealing method and the mass flow method, a process observer, which is described in detail in the following claims, is designed to calculate an estimate for _out that with to be supplied fuel a desired mixing ratio is reached. 15. The method according to claim 14, characterized in that the process observer contains the mass balance equation (1). 16. The method according to claims 14 and 15, characterized in that the Process observers did not differentiate the intake manifold pressure and the well-known theory the observer applies.   17. The method according to claims 14-16, characterized in that the process observer is implemented adaptively, for example, the error between real Reduce intake manifold pressure p and the observed estimate. 18. The method according to claims 14-17, characterized in that the observer is set for example as follows:

• - The reality "intake manifold" is represented in the observer as a parallel model
• - The real integration of the air mass flow difference weighted by the factor is formed by a delay element of the first order, the output of which simulates the course of the intake manifold pressure
• - By the additional connection of the intake manifold pressure weighted by the factor λ to the input of this delay element, it is achieved that the delay element simulates the real integrating behavior when the estimated and real intake manifold pressure match, this is illustrated in FIG. 8
• - the error between the profiles of the real and estimated intake manifold pressure arises conferences by the non-matching Luftmassenstromdif, that is _out by the incorrect estimate of the air flowing into the cylinder air mass flow, since the input _value for reality and observers equal
• - The air mass flow _out flowing into the cylinders depends on the relationship 7 of the speed and density via the so-called volumetric efficiency η _vol , the influence of which on the operating point can be simulated in the observer by multiplication with an integrator output
• - The input of this integrator is subjected to the learning error between the real and the estimated intake manifold pressure over a learning step size ν in such a way that the resulting adaptation makes the viewing error zero (control-technical stability of the viewer)
• - If the observer error between the real and the estimated intake manifold pressure is zero, the delay element acts by acting on its input with the intake manifold pressure weighted by the factor λ like an integrator with the air mass flow difference as the only input, see Fig. 8
• - This means that the observer structure is the same as reality, which means that if the route structure and the route match, outputs (intake manifold pressure and estimated value therefor) also the route inputs (difference in air mass flows) are the same
• - Since the difference between the air mass flows of the observer is formed from measured _in and estimated _-out , the estimated air mass flow _{out is} therefore equal to the real air mass flow into the cylinders _out
• - The estimated value for the air mass flow must be integrated over the time of the intake stroke of each individual cylinder in order to obtain the air mass in the respective cylinders
• - Possible mathematical transformations are part of the claim.

19. The method according to claims 14-18, characterized in that the volumetric efficiency η _{vol is} adapted at an operating point in order to obtain the desired result after the relationship 7. 20. The method according to claims 14-19, characterized in that the process observer uses prior knowledge, for example about η _vol roughly as a map, and then additionally adapts a multiplicative or additive-acting variable. 21. The method according to claim 20, characterized in that the Prozessobobach ter adapts several multiplicative or additive acting variables or uses more prior knowledge than only about η _vol . 22. The method according to claims 14 to 21, characterized in that the adaptation of the process observer and thus the approximation of the observer route structure to the reality according to FIG. 8 takes place on the basis of a method derived from the linear stability theory, for example by the counter polynomial of the observer circuit by means of the parameters λ and ν, behave stably according to the desired observer. 23. The method according to claims 14 to 22, characterized in that the parameters λ and ν, can be designed to vary in time, for example at each operating point to achieve the desired behavior of the observer group as a function of n (for example using a map). 24. The method according to claims 14-21, characterized in that the Prozessbe observer using known mathematical error models or the non-linear ones Stability theory is implemented adaptively. 25. The method according to claims 14-24, characterized in that the Prozessbe observer uses known prior knowledge (e.g. the sizes, or in Maps stored). 26. The method according to claims 14-25, characterized in that for the adap the process observer to the time-variant or time-invariant variables known regression methods are used. 27. The method according to claims 14-26, characterized in that the pro zeßbobachter processed both measured and estimated quantities. 28. The method according to claims 14 to 27, characterized in that the pro zeßobachter both measured and estimated sizes to different Cached times. 29. The method according to claims 14-28, characterized in that the learned knowledge is stored in certain working points so that it is not forgotten, and then used as prior knowledge additive or multiplicative (z. B. η _vol as a function of n and p ). 30. The method according to claims 14-29, characterized in that at Ver Use of devices to achieve variable intake manifold geometry, variable Control times or valve trains other than via camshafts the volumetric Efficiency still has an additional dependency than just on p and n. 31. A method for controlling internal combustion engines, characterized in that the speed sealing method with the mass flow method as described is used for air mass flow determination or control by the of mass flow of air drawn into the cylinders from one to the following say detailed process learners such be shown is calculated that with the fuel to be supplied a desired mixture ratio is reached. 32. The method according to claim 31, characterized in that the observer is set as follows for example:

• - The reality "intake manifold" is represented in the observer as a parallel model
• - The real integration of the air mass flow difference weighted by the factor is formed by a delay element of the first order, the output of which simulates the course of the intake manifold pressure
• - By the additional connection of the intake manifold pressure weighted by the factor λ to the input of this delay element, it is achieved that the delay element simulates the real integrating behavior when the estimated and real intake manifold pressure match, this is illustrated in FIG. 8
• - the error between the profiles of the real and estimated intake manifold pressure arises conferences by the non-matching Luftmassenstromdif, that is _out by the incorrect estimate of the air flowing into the cylinder air mass flow, since the input _value for reality and observers equal
• - The air mass flow _out flowing into the cylinders depends on the relationship 7 of speed and density via the so-called volumetric efficiency η _vol , whose influence on the operating point can be simulated in the observer by multiplying it by the output of an adaptation device
• - The input of this adaptation device is subjected to the learning error between the real and the estimated intake manifold pressure over a learning step size ν in such a way that the resulting adaptation makes the observer error zero (control-technical stability of the observer)
• - If the observer error between the real and the estimated intake manifold pressure is zero, the delay element acts by acting on its input with the intake manifold pressure weighted by the factor λ like an integrator with the air mass flow difference as the only input, see Fig. 8
• - This means that the observer structure is the same as reality, which means that if the route structure and the route match, outputs (intake manifold pressure and estimated value therefor) also the route inputs (difference in air mass flows) are the same
• - Since the difference between the air mass flows of the observer is formed from measured _in and estimated _-out , the estimated air mass flow _{out is} therefore equal to the real air mass flow into the cylinders _out
• - The estimated value for the air mass flow must be integrated over the time of the intake stroke of each individual cylinder in order to obtain the air mass in the respective cylinders
• - Possible mathematical transformations are part of the claim.

33. The method according to claims 31 and 32, characterized in that for the adaptation of the process observer to the time-variant variables, characteristic diagrams are used as an adaptation device in order to take their influences into account. This is done, for example, by using a discrete grid over the input space of p and n as a map for η _vol . 34. The method according to claims 31 and 32, characterized in that for the adaptation of the process observer to the time-variant variables, known regression methods are used as an adaptation device to reach the adaptation. This is done, for example, by using a mathematical function approach for η _vol by adjusting the function coefficients according to a regression approach. 35. The method according to claims 31-33, characterized in that the Pro zeßobobachter is able to learn by using artificial intelligence processes can be used as an adaptation device. 36. The method according to claims 31-34, characterized in that a new ronal network is used as an adaptation device. 37. The method according to claims 31-34, characterized in that fuzzy Logic is used as an adaptation device. 38. The method according to claims 31-34, characterized in that a com combination of fuzzy logic and neural processes (neuro-fuzzy) as Adapti onseinrichtung is used. 39. Method according to claims 31-35, characterized in that as a neuro a general regression neural network as adaptations device is used. 40. Process, characterized in that a different approach than in the process is used according to claims 34-38 for learning, for example wise by including special mathematical or hardware components in the Process observers are involved.   41. The method according to claim 31, characterized in that the Prozessbe observer using known mathematical error models, as in the literature described, becomes adaptive. 42. The method according to claims 31 and 35, characterized in that the process observer uses a neural network, is implemented in accordance with FIG. 8 and uses the method described in AZ: 195 31 962.4. 43. The method according to claims 31 and 35, characterized in that the Process observer a mathematical error model, as in AZ: 195 31 962.4 be wrote, used to achieve stable learning. 44. The method according to claims 31 and 35, characterized in that the Stability of learning is guaranteed by means of the Ljapunov theory. 45. The method according to claims 1, 14 and 31, characterized in that the process observer during of the total operating time of the internal combustion engine all variables as a time variant ant looks at and considers all variables in the learning process. 46. The method according to claims 1, 14 and 31, characterized in that the process observer only the Part of the variables taken into account in learning, which changes due to its time variance affects the process. 47. The method according to claims 1, 14 and 31, characterized in that the Process observers only limited the very slow time-variant variables retrained in time periods and at predefinable intervals and in characteristic diagrams stores this knowledge. 48. The method according to claims 1, 14 and 31, characterized in that the learned knowledge or the estimated value of the process observer for the fuel measurement is taken into account. 49. The method according to claims 14 and 31, characterized in that for Simplification of the implementation of the "learned knowledge" stored in maps is and that at predetermined time intervals by means of the described The maps are refreshed. 50. Process according to claims 1-49, characterized in that the volu metric efficiency of several sizes than just the intake manifold pressure and the Speed depends; this applies, for example, to the VANOS process or if a different valve train than the camshaft. 51. The method according to claims 1-49, characterized in that the volume The efficiency of the sizes intake manifold pressure, speed and valve opening times is determined (VANOS, different valve train than via a camshaft, Switching suction tube).   52. Process according to claims 1-50, characterized in that the method Ren is applied to any device in which an air mass flow due to the storage behavior of a finite volume with a reasonable opening wall cannot be measured. 53. The method according to claims 1-49, characterized in that instead of Load estimate of the air mass flow an estimate of the disturbance variables at the Dros selflap is done. 54. The method according to claims 1-49, characterized in that instead of Load estimate of the air mass flow an estimate of the disturbance variables in the elec tromagnetic valve train takes place. 55. The method according to claims 1-49, characterized in that instead of Load estimate of the air mass flow an estimate of the disturbance variables at a gearbox to be switched automatically. 56. The method according to claims 1-49, characterized in that instead of Load estimate of the air mass flow an estimate of the be in the cylinders sensitive amount of fuel takes place. 57. Method according to claims 1-49, characterized in that an estimate of a size that occurs in a linear or nonlinear fashion from one or depends on several process states. 58. The method according to claims 1-49, characterized in that instead of Estimation of multiple sizes happens in a linear or nonlinear fashion depend on one or more process states within a process. 59. The method according to claims 14 to 52, characterized in that the time derivative of the intake manifold temperature is not neglected. 60. Process according to claims 1-59, characterized in that to Kor relation of the observer error signal with the ons causing this error additional quantities of the neural network within the neural network States next to the weight integrators can be used. This is done to those without these additional internal states of the neural network for adap tion of the weights valid SPR condition between network input and observer annul errors. 61. The method according to claim 60, characterized in that to correlate the Be observer error signal with the input variables causing this error Neural network the additional states between excitation and weight device of the individual neurons. 62. The method according to claim 61, characterized in that the additional states between excitation and weighting of the individual neuroses are inserted in order to correlate the observer error signal with the input variables causing this error of a general regression neural network, as illustrated in FIG. 13. This signal flow diagram shows in the lower branch the transformation of the identification problem (f (.) Unknown) to be achieved by an observer approach to the chain arrangement of known dynamics H 1 (s), unknown static relationship f (.) (Also multidimensional) and known dynamics H 2 (s ). The upper branch shows the structure of a General Regression Neural Network with excitation W (.) And weights Θ and the states in between in the dynamic H₂ (s). See AZ: 195 31 362.4. 63. The method according to claim 62, characterized in that the correlation of the loading observer error signal with the input variables causing this error Neural network additional states between excitation and weighting of the individual neurons are inserted in such a way that with states after the Output of the neural network the desired correlation is achieved. - DEVICES 64. Device for controlling internal combustion engines with one sensor each Air mass flow measurement, for intake manifold temperature measurement and for the intake manifold Pressure measurement, characterized in that a computer with a method according to claims 1 to 13, the fuel supply of an internal combustion engine controls such that a predeterminable mixing ratio of fuel / air is reached becomes. 65. Device for regulating internal combustion engines with sensors for Luftmas Sense current measurement, for intake manifold temperature measurement, for intake manifold pressure measurement and for measuring other quantities, characterized in that a computer with a method according to claims 1 to 13, the fuel supply of a Ver controls internal combustion engine such that a predeterminable mixing ratio force material / air is reached. 66. Device for controlling combustion devices with one sensor each for air mass flow measurement, intake manifold temperature measurement and suction Pipe pressure measurement, characterized in that a computer with a method according to claims 1 to 13, the fuel supply of a combustion device controls such that a predeterminable mixing ratio of fuel / air is reached becomes. 67. Device for controlling combustion devices with sensors for air mass flow measurement, for intake manifold temperature measurement, for intake manifold pressure measurement solution and for measuring other sizes, characterized in that a computer with a method according to claims 1 to 13, the fuel supply Combustion device controls such that a predeterminable mixing ratio nis fuel / air is reached. 68. Device for controlling internal combustion engines with one sensor each Air mass flow measurement, for intake manifold temperature measurement and for the intake manifold  Pressure measurement, characterized in that a computer with a method according to claims 14 to 30, the fuel supply of an internal combustion engine controls such that a predeterminable mixing ratio of fuel / air is reached becomes. 69. Device for regulating internal combustion engines with sensors for Luftmas Sense current measurement, for intake manifold temperature measurement, for intake manifold pressure measurement and for measuring other quantities, characterized in that a computer with a method according to claims 14 to 30, the fuel supply of a Ver controls internal combustion engine such that a predeterminable mixing ratio force material / air is reached. 70. Device for controlling combustion devices with one sensor each for air mass flow measurement, intake manifold temperature measurement and suction pipe pressure measurement, characterized in that a computer with a process ren according to claims 14 to 30, the fuel supply of a combustion Direction controls such that a predeterminable mixture ratio fuel / air is achieved. 71. Device for controlling combustion devices with sensors for air mass flow measurement, for intake manifold temperature measurement, for intake manifold pressure measurement solution and for measuring other sizes, characterized in that a computer with a method according to claims 14 to 30, the fuel supply Combustion device controls such that a predeterminable mixing ratio Fuel / air is reached. 72. Device for controlling internal combustion engines with one sensor each Air mass flow measurement, for intake manifold temperature measurement and for the intake manifold Pressure measurement, characterized in that a computer with a method according to claims 31 to 49, the fuel supply of an internal combustion engine controls such that a predeterminable mixing ratio of fuel / air is reached becomes. 73. Device for regulating internal combustion engines with sensors for air mass Sense current measurement, for intake manifold temperature measurement, for intake manifold pressure measurement and for measuring other quantities, characterized in that a computer with a method according to claims 31 to 49, the fuel supply of a Ver controls internal combustion engine such that a predeterminable mixing ratio force material / air is reached. 74. Device for controlling combustion devices with one sensor each for air mass flow measurement, intake manifold temperature measurement and suction pipe pressure measurement, characterized in that a computer with a process ren according to claims 31 to 49, the fuel supply of a combustion Direction controls such that a predeterminable mixture ratio fuel / air is achieved.   75. Device for controlling combustion devices with sensors for air mass flow measurement, for intake manifold temperature measurement, for intake manifold pressure measurement solution and for measuring other sizes, characterized in that a computer with a method according to claims 31 to 49, the fuel supply Combustion device controls such that a predeterminable mixing ratio Fuel / air is reached. 76. Device for regulating internal combustion engines with sensors for Luftmas Sense current measurement, for intake manifold temperature measurement, for intake manifold pressure measurement and for measuring other quantities, characterized in that a computer with a method according to claims 31 to 49 using methods according to claims 57 to 63, the fuel supply of an internal combustion engine controls that a predeterminable mixture ratio of fuel / air is reached. 77. Device for controlling combustion devices with one sensor each for air mass flow measurement, intake manifold temperature measurement and suction Pipe pressure measurement, characterized in that a computer with a method according to claims 31 to 49 using methods according to claims 57 to 63 controls the fuel supply of a combustion device in such a way that a predefinable fuel / air mixing ratio is achieved. 78. Device for controlling a process, characterized in that a poss cher process observer is used according to claims 57 to 63, a Estimating the disturbance or learning its dependence on process states. 79. Devices according to claims 64 to 78, using a digital computer realize the process according to claims 1 to 63. 80. Devices according to claims 64 to 78, with an analog computing realize the process according to claims 1 to 63. 81. Devices according to claims 64 to 78, with a hybrid rake Werk (analog and digital) implement the method according to claims 1 to 63.