EP0221386B1 - Method and device for adapting the mixture control in an internal-combustion engine - Google Patents

Description

State of the art

The invention is based on a method according to the type of the main claim and a device according to the type of the claim. The invention is a further development of the applicant's earlier application EP-A-191 923, in which the adaptive feedforward control influences both structurally determined areas of a basic map and, via a global factor, multiply each control value obtained from the map in the sense of a shift of all map support points .

In EP-A-152 604, which also goes back to the applicant, it is proposed that stored in a map and depending on To change the operating parameters of the internal combustion engine selected values in accordance with a learning process in such a way that not only a single predetermined map value, but also the respective map values located in its environment are additionally modified as a function of the change in the value concerned. In order to be able to introduce a learning method in the area of the precontrol, an actual value specification about the actual operating state of the internal combustion engine is required, the control factor or the manipulated variable of a lambda controller usually being evaluated as the actual value. This manipulated variable of the lambda controller therefore influences this pilot control range in the sense of adaptive learning of the pilot control range (for example injection time characteristic map with evaluation of a throttle valve signal and the engine speed) and at the same time serves as a feedback actual value in the current control of the mixture control on the basis of those possibly changed by adaptive learning Pre-control values from the map area.

In detail, the procedure can be such that the correction factor issued by the lambda controller is averaged, subjected to suitable boundary conditions and then incorporated into a structural map (structure adaptation) overlaying the basic map as well as into a global factor (global adaptation). This incorporation then takes place when leaving a feed area (adaptation area) defined around each map support point.

In this context, it is generally known (DE-OS 28 47 021; GB-PS 20 34 930 B) to design mixture metering systems in such a way that the metering or metering of the fuel takes place, for example, via so-called learning control systems. Such a learning control system contains, for example, values for the injection that are available when the machine is operating in a permanently active read / write memory.
The characteristic diagrams result in a quickly reacting pilot control, for example for the injection quantity or generally for fuel metering or also for other operating parameters on the internal combustion engine to be adapted as quickly as possible to changing operating conditions, such as ignition timing, exhaust gas recirculation rate and the like. Like. In order to get control systems to be learned, the individual map values can be corrected depending on the operating parameters and written into the respective memory.

EP-A-0 145 992 discloses a method for fuel metering which, in addition to a pilot control and a lambda control of the fuel / air ratio, has an adaptation option. A lambda control factor influences an injection time predetermined from operating parameters so that the actual fuel / air ratio approaches the stoichiometric value. Longer-term deviations of this control factor from a target value are recorded by arithmetically averaging the values of this control factor when the probe signal changes from the rich to the lean signal and when the corresponding change from the lean to the fat signal occurs. A difference between this mean value and a target value leads to the formation of a correction factor which is multiplied in the calculation of the fuel metering signal and which is ideally dimensioned such that it eliminates the longer-term deviation of the control factor. Values of this adaptive correction are stored in a memory for a large number of operating states, so that a suitable correction factor is available as quickly as possible when changing between two operating states. However, the specified method still allows improvements in the actual adaptation, ie in eliminating the longer-term deviations of the average control factor from an ideal value.

The following statements are based on the knowledge and the disclosure content of the earlier application EP-A-191 923

The invention is therefore based on the object of adaptive learning methods in the event of changing external conditions Improve fuel supply in internal combustion engines and ensure that the corrective action is distributed in the desired manner to the structural and global adaptation and includes a reference to the actual operating behavior of the internal combustion engine (frequency of the manipulated variable vibrations) also in the course of the adaptive learning process.

Advantages of the invention

The invention solves this problem with the characterizing features of main claims 1 and 7 and has the advantage that the learning ability is divided by different quantization of the factors for structural adaptation and global adaptation in such a way that external, slow-running operating change conditions (assuming a stable structure in the engine area) For example, changing air pressure) are essentially compensated for by the global factor and do not affect the structural area. On the other hand, the invention enables a division by means of a different weighting between global and structural adaptation such that where a structural correction is effectively required, this is also carried out by the structural factor in the respective map area and to a lesser extent is added to the global factor.

Finally, it is advantageous that, according to one embodiment, the averaging of the manipulated variable of the lambda controller, the formation of the structural and global factor as well as other work processes in the grid of jumps in the manipulated variable of the lambda controller, that is to say related to the number of probe passes, is switched off, so that the adaptation is dynamically adapted to the operating variable λ and does not run asynchronously in any time grid. The factors can therefore be optimally determined.

Furthermore, this relationship to the manipulated variable jump grid of the lambda controller results in the design of the adaptation as self-locking, since if the lambda controller runs against one of its actuating stops, i.e. no more grid jumps occur, it cannot be adapted further, and in any case First of all, the manipulated variable is recognized as implausible.

Advantageous further developments and improvements of the invention are possible through the measures listed in the subclaims.

drawing

Embodiments of the invention are shown in the drawing and are explained in more detail in the following description. 1 schematically shows a block diagram of the active learning area with evaluation of the manipulated variable curve of the lambda controller and division into structure and global adaptation, FIG. 2 shows the area of an adaptation area with the entry and exit time of the operating parameters, and FIG. 3 in the form a diagram the time sequence of the lambda controller manipulated variable.

Description of the embodiments

The various forms and variants of the present invention complement the basic ideas explained in detail in the main application in several respects, namely the type and division of the adaptation to a structural map which overlaps the basic map, and to the area of the global factor and the relationship between the factor formation Determination of the manipulated variable, the counting of settling delays u. Like. On the jump behavior of the lambda probe (manipulated variable jumps of the lambda controller), so that the adaptation is dynamically adapted to the operating parameter λ.

In order to understand the present invention, it is necessary to know the subject matter of EP-A-191 923, which, as already mentioned at the beginning, is assumed here.

Furthermore, it is pointed out that the means shown in the drawings, indicating the invention and its various aspects with the aid of discrete switching stages or blocks, do not restrict the invention, but in particular serve to illustrate basic functional effects and to indicate special functional sequences in a possible form of implementation. It is understood that the individual blocks, components, blocks, functions and. Like. Can be constructed in analog, digital or hybrid technology or, in whole or in part, corresponding areas of program-controlled digital systems or programs, for example, implemented by microprocessors, microcomputers, digital logic circuits and the like. Like. Since the help of program experts is available to the experts at any time, who can implement the corresponding functional sequences, commands and effects in a suitable program language, this step is no longer considered to be in need of explanation and the additional representation of flow diagrams when used by microprocessors, for example the understanding of the present invention can be dispensed with. The description of the invention given below is therefore only a preferred exemplary embodiment with regard to the overall functional and time sequence, the mode of operation achieved by the respective blocks discussed and the respective interaction of the sub-functions represented by the individual components, the references to the circuit blocks for reasons of better understanding.

First, to round off and for a better understanding briefly on some basic relationships in a z. In some cases, explanations of the parent application were repeated. The invention is based on the fact that for the formation of the injection time (or another indication of the amount of fuel to be supplied to an internal combustion engine when used in carburetors or the like) in normal operation (starting conditions, emergency operation, overrun fuel cutoff and the like are usually not included in this) Injection time map is used as a basis, which may be dependent on the speed throttle valve position and, for example, via a predefined one Number of speed and throttle valve signal points is spanned. For example, 15 speed and 15 throttle valve signal support points can be provided in numerical values. This basic injection map can then be designed, for example, for a special vehicle of a particular vehicle type. For example, to adapt to other vehicles with deviations such as in the engine area, throttle valve supports and. The same is superimposed on this basic map a structural map, which, expressed in numerical values, can have, for example, 8 speed and 8 throttle valve support points. These represent a subset of the 15x15 grid points of the injection time map.

A so-called global factor also serves to adapt boundary conditions that have a multiplicative effect on the mixture formation as a whole (here, for example, ambient pressure differences due to height, temperature, aging of aggregates or the like come into question).

The respective adaptation area for the learning process, here first the structural adaptation in the map, is shown in FIG. 2 and results from the agreement that, as already mentioned above, the 8x8 support points of the structural adaptation map of a subset from the 15x15 support points of the reason Map, which can also be referred to as the injection time map, when the fuel is supplied to the internal combustion engine via electromagnetically actuated injection valves. Since every second support point of the 15x15 map is a support point of the 8x8 structural map a basic and FSA support point (FSA = factor for structural adaptation) is surrounded by 8 basic support points within the map, which is due to the different network structure (single-crossed lines for basic support point and double-crossed lines for combined basic and FSA support point) ) can also be seen from the illustration in FIG. 2. These 8 basic support points span the adaptation surface 10. When the operating characteristic throttle valve position or angle DK and speed N defining the characteristic diagram enter the adaptation surface at the time t = T1 after X ignitions, the limited manipulated variable issued by the lambda controller 11 (see FIG. 1) is limited and without taking into account the respective one "History" divided into global and structural parts. A learning cycle is defined as the period between the time of entry t = T1 (entry point) and the time of exit t = T2 (exit point).

The learning process therefore proceeds as follows. The output of the lambda probe is fed to the lambda controller 11 at its input; At the output of the lambda controller 11 there is a normalized lambda manipulated variable Xr which, after passing through a limiting block 12 for limiting the actuating stroke with the limit values Xr'max and Xr'min, reaches the low-pass filter 13 as a limited, standardized manipulated variable Xr ', which is the limited manipulated variable Xr 'of the lambda controller is subjected to averaging. A low-pass output value Ya then results at the output of the low-pass filter, which can include a predefined (also recursive) low-pass formula, the output value of the low-pass filter Ya at the time point t = T1 being set equal to the input value becomes → Ya (t = T1) = Xr '(t = T1). If then more than X plus a predetermined number, for example approximately 32 ignition pulses, have passed since the entry into the adaptation area 10 according to FIG. 2, then the factor FSA to be learned in the structural area is transferred to the characteristic diagram when it leaves the adaptation area (time t = T2) , whereby this factor at time T2 is composed of the factor at time t <T1 and the new value of the low pass Ya at time T2, thus resulting in the following formula:

$\text{FSA (T2) = FSA (t <T1) + Ya (T2).}$

The term FSA (t <T1) is the factor stored for the structural area FSA during the last learning process.

An essential inventive measure in this context is that an adjustment of the factor for the structural area per learning cycle is only permissible by a predetermined percentage value, specifically only 3% in the preferred exemplary embodiment, ie the Δ value of the FSA is relative Coarse gradation only at 0.03 and therefore, as will be explained further below, significantly different from the Δ value of the factor for the global adaptation FGA. Attention is drawn overall to the illustration in FIG. 1, switches or function blocks which have not been explained so far being specified further below in connection with supplementary aspects in the present invention.

The learning process, i.e. the formation of the factor for the global adaptation FGA, takes place as in the structural adaptation within a learning cycle (time between T1 and T2 in FIG. 2) and in approximately the same way, with only the limited and filtered manipulated variable Ya des Lambda controller 11 can be filtered again with a corresponding filter formula.

wobei Zg(T2) der zum Zeitpunkt T2 am Ausgang des zweiten Filters anstehende Wert der Lambda-Regler-Stellgröße ist.The factor FGA to be learned is also formed when leaving the adaptation surface (t = T2)

$\text{FGA (T2) = FGA (T <T1) + Zg (T2),}$

where Zg (T2) is the value of the lambda controller manipulated variable pending at the time T2 at the output of the second filter.

Global and structural adaptation are prohibited in the operating states
Lambda control switched off,
with warm-up enrichment, start u. Operating conditions of the internal combustion engine that cannot be evaluated for an adaptation.

In contrast to the permitted adjustment of the factor of the structural adaptation FSA of, for example, 3% per learning cycle, a much finer gradation is permitted for the global factor, for example only 1/8 or 1/16 of the resolution of the structural factor FSA, like the two grading blocks 14a and 14b in Figure 1 also indicate. Through this different Grading is to be achieved that, provided the structure in the internal combustion engine area is stable (for example, throttle valve intake manifold does not change, internal combustion engine is stable), however, it changes in a multiplicative manner, that is to say in a globally changing manner, such as gradual air pressure changes and the like. Like., These are not communicated to the structural, but only the global factor. By making the grading for possible changes in the structural factor FSA much larger, for example with the specified 3%, than the changes in the global factor (grading e.g. 0.19%), slow changes can be made with only a slow rate of change (this remains then probably less than 3% per learning cycle) will always be compensated for by the global factor, i.e. it must be absorbed before the change threshold (3%) for the structural factor FSA is reached. A trip over a passport height can be assumed as an example, in which the air density changes slowly with increasing altitude and so adaptive compensation is effected via the global factor FGA.

Larger expected changes can occur, however, if a vehicle is parked, for example, in certain, even extreme, low-pressure weather conditions and restarted in a corresponding high-pressure weather situation.

To compensate for this effect, which also has a multiplicative effect, but which can then suddenly be greater than, for example, the specified 3% According to an advantageous measure of the present invention, only global adaptation via the global factor FGA is permitted after starting for a limited number of learning cycles. One can therefore also see in the illustration in FIG. 1 a switch 15 which is to be understood here as an example and which disconnects the connection to the low-pass filter and thus forms a structural ban. Only when the multiplicative deviation has been compensated is the adaptive compensation permitted if the structure changes. This results in a considerably improved stability of the learning map and avoids unnecessary "breathing", also with the advantage that support points positions in the map area (structure) that can be evaluated perfectly for any subsequent shutdown of the internal combustion engine are available for subsequent use.

Accordingly, the total injection time t _i output during normal operation is made up as follows:

ist die Grundeinspritzzeit aus dem 15α·15N-Kennfeld und t_S ist die Ventilverzugszeit.In this formula, FSA and FGA are the factors for structural adaptation (from 8α · 8N map with α = throttle valve angle and N = speed) and for global adaptation, π _i Fi are factors from other functions for forming t _i , t $\genfrac{}{}{0ex}{}{\text{'}}{\text{L}}$

is the basic injection time from the 15α · 15N map and t _S is the valve delay time.

In the structural map, the factors are not interpolated. In order to compensate for the jump in the transition from one interpolation point to the next, the manipulated variable of the lambda control becomes the opposite percentage changed by the same amount by which the factors stored at the two bases differ.

A further embodiment of the present invention is based on the knowledge that, if the learning process assumes convergence per learning step (learning cycle), the full deviation is adopted as a correction value in the learning factors. The deviation is divided in such a way, namely above the low-pass filter 13 in the parallel branch for structural and global adaptation, weighting elements 16a, 16b, that the greater part of the deviation is added to the structural factor FSA and the smaller part to the global factor FGA. The weaker weighting (k) of the global factor means that corrections for structural reasons, if (inevitably) they are also taken up by the global factor, do not radiate too strongly to the other map points (structure). The different distribution of the correction value caused by the weighting blocks 16a, 16b therefore ensures the tendency to make structurally-related corrections in the factor for the structural correction (and therefore only in a certain map area) adaptively and to have an effect (multiplicative) on the entire map in the same way Capture changes through the global factor. This tendency is further reinforced and supported by the different grading (Δ value for FSA and FGA per learning cycle) mentioned above, based on the assumption that global changes are relativized by the gross grading in the ability of the structural factor to be adopted and be suppressed.

The circuit breaker 17 provided in FIG. 1 at the output of the low pass 13 serves to implement the general prohibition on adaption.

A further, particularly advantageous embodiment of the invention consists in dynamically improving the adaptation of the mixture control, specifically in that, for example, the averaging (via the low-pass filter 13) of the manipulated variable X'r of the lambda controller 11, the formation of structural and global Factors FSA and FGA as well as, for example, the counting of the settling delay and the minimum averaging time in the steps of the steps of the manipulated variable of the lambda controller (see FIG. 3), where the number of these passes is the number of probe passes which is an operating parameter (actual value) Internal combustion engine is, corresponds.

Compared to a previously used time grid, there is therefore the advantageous adaptation of the adaptation dynamically to the operating parameter λ, in other words, the adaptation does not run asynchronously in a time grid. The respective factors can be optimally determined by the dynamic adaptation - this also results in a self-locking design of the adaptation, because if the controller runs against one of its actuating stops, no further adaptation takes place because the manipulated variable no longer performs jumps and in this case is implausible is seen.

In this variant of the present invention remain fundamentally identical formation of the factors that prohibit structural and global adaptation and the additional prohibition for structural adaptation after a start is defined for a predefinable number of x learning cycles of the global factor FGA. During these x learning cycles, opening the switch 15 results in a structural ban - the weighting of the filter output Ya is also retained for the global factor for these learning cycles.

After the x learning cycles have been completed, adaptation takes place in parallel, globally and structurally according to the weighting k and (k-1). The limitation to the 3% correction per FSA learning cycle mentioned above can, however, be omitted, so that any multiple of 3% per learning cycle can be changed in the structural factor FSA - the change levels for the global factor FGA always remain 8 times or 16 times finer.

It is essential that, as can be seen from the diagram of FIG. 3, the manipulated variable of the lambda controller 11 with permitted adaptation (structure and / or global) is processed with the value in the low-pass filter that results after the jump in the manipulated variable , ie at the jump times t = i; i + 1; i + 2 .... After expiry of n _{s of} such jumps in the manipulated variable, which follow an adaptation area change, the
Low pass 13 too

${\text{Ya (i + n}}_{\text{s}}{\text{) = Xr '(i + n}}_{\text{s}}\text{).}$

In other words, by the number of times you entered low pass 13 is released (set to initial value) each time a new adaptation area expires, the further values then being determined in accordance with the recursive low pass formula given below:

The recursion takes place after each manipulated variable jump. The adoption of the correction value for the adaptation factor (formation of FSA and FGA) determined with the above equations is only permitted when at least m _s recursion steps of the filter 13 have been completed.

If the conditions n ≧ n _s and m ≧ m _s for the time (adaptation area change) are then met, the current filter output value Ya (j) via the weighting levels k and (1-k) and corresponding quantization at 14a and 14b as correction value for processed the new formation of the factors FSA and FGA, as shown in the block diagram of FIG. 1. This also explains the switches in FIG. 1, namely a first switch 18, which closes when the condition n ≧ n _s is met and acts on the low-pass filter 13 with the output of the limiting block (limited manipulated variable of the lambda controller) - and of the switch 19 which forwards the output value of the filter for parallel processing and adaptation when the condition m ≧ m _{s is} met.

At the time of an adaptation area change, n and m are set to zero; 3 it is assumed that n _s = 4.

The following table shows the adjustable sizes as well as their range limits which are recognized as advantageous - it is not to be regarded as restrictive.

By switching off (counting) probe passes or jumps at the manipulated variable of the lambda controller, the adaptation automatically becomes self-locking when the controller runs against one of its actuating stops. It makes sense to evaluate this case as implausible due to the large selectable adjustment and adaptation range and, at least as a first option, not to adapt any further.

All new features specified in the description and in the claims can be essential to the invention, alone or in combination.

Claims (10)

1. Method for adapting the mixture control in internal-combustion engines, in which a pilot-control quantity dependent on the load and speed of the internal-combustion engine for the fuel proportioning is influenced by at least one adaptively variable correction quantity (structural adaptation, global adaptation) and the formation of the correction quantities starts from the output signal of a lambda controller (11), characterised in that, for the adaptation of the correction quantity, the output signals of the lambda controller obtained immediately after a jump of the lambda-probe signal are filtered under predetermined operating conditions by means of a recursive low-pass formula, the recursion taking place at each set-variable jump, and in that other previous output signals of the lambda controller are not used for this adaptation of the correction quantity.
2. Method according to Claim 1, characterised in that the recursion commences only after a predetermined number (n_s) of jumps of the lambda-probe signal (set-variable jumps) following a change of operating characteristic (change of adaptation area) has elapsed, and in that, at the start of the recursion, the initial value (Ya(i+n_s)) of a low-pass filter (13) is set to a predetermined value (Xr'(i+n_s)), and in that the take-over of the correction value produced by the low-pass filter to a particular adaptation factor (FSA, FGA) takes place only after a predetermined number (m_s) of recursion steps of the filter (13) is completed.
3. Method according to Claim 2, characterised in that the low-pass values are determined according to the recursive low-pass formula $\text{Ya(j) = (b.Ya(j-1)+Xr'(j))/(b+1).}$

   
4. Method according to one of Claims 1 to 3, characterised in that the correction is made by means of a structural factor (FSA) and a global factor (FGA) and a structural and/or global correction is made only when a minimum correction threshold is exceeded.
5. Method according to one of Claims 1 to 4, characterised in that the structural correction is limited to a predetermined value per learning cycle.
6. Method according to one of Claims 1 to 5, characterised in that the structural and global corrections have a differing grading.
7. Device for adapting the mixture control in internal-combustion engines, in which a pilot-control quantity dependent on the load and speed of the internal-combustion engine for the fuel proportioning is influenced by at least one adaptively variable correction quantity (structural adaptation, global adaptation), and the formation of the correction quantities starts from the output signal of a lambda controller (11), characterised in that there are means (13) which, for the adaptation of the correction quantity, filter the output signals of the lambda controller, obtained immediately after a jump of the lambda-probe signal, by means of a recursive low-pass formula under predetermined operating conditions, the recursion taking place at each set-value jump, and do not use other previous output signals of the lambda controller for this adaptation of the correction quantity.
8. Device according to Claim 7, characterised in that there are counting and switching means (18, 19) which, whilst ensuring a dynamic matching of the adaptation to an actual operating characteristic (λ), execute the transmission of the value (Xr') obtained after the jump of the set variable to the low-pass filter (13) causing its averaging, only when a predetermined number (n_s) of set-variable jumps or probe passes has elapsed.
9. Device according to Claim 8, characterised in that the take-over of the correction value (Ya) determined by the low-pass filter to the adaptation factor (FSA, FGA) by means of counters and switch means (19) is cleared only when a predetermined number (m_s) of recursion steps of the filter (13) averaging the set variable of the lambda controller (11) is completed.
10. Device according to one of Claims 7 to 9, characterized in that there are weighting stages (16a, 16b) which feed the correction quantity (Ya) produced by the low-pass filter (13) to the change-stage blocks (14a, 14b) for the structural factor (FSA) and global factor (FGA) differently (weighting k and (k-1) respectively).

Images