JPH116457A - Engine air-fuel ratio control device - Google Patents

Abstract

(57) [Summary] [Problem] To normally accelerate the convergence of a learning value so that an air-fuel ratio does not become excessively rich or lean on the contrary to a value corresponding to a target stability. SOLUTION: Before a base learning value converges, a base learning value is selected, and after the base learning value converges, a normal learning value is selected by a selection means 25, and a feedback control condition based on the stability during lean operation is determined. When the condition is established, the air-fuel ratio correction means 27 corrects the initially set air-fuel ratio during lean operation with the selected learning value. When the feedback control condition based on the stability during the lean operation is satisfied and before the base learning value converges, the updating unit 30 first updates the base learning value,
When the feedback control condition based on the stability during the lean operation is satisfied and after the base learning value converges, the updating means 31 updates the normal learning value with the convergence value of the base learning value as an initial value.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an air-fuel ratio control device for an engine.

[0002]

2. Description of the Related Art At the same time as improving fuel efficiency of an engine, NO
In order to reduce x, the fuel supply is controlled so that the air-fuel ratio, which is the ratio of air and fuel, becomes a lean air-fuel ratio leaner than the stoichiometric air-fuel ratio. If the stability becomes worse than the control target value,
An engine operating method in which the air-fuel ratio is corrected to a rich side (or the ignition timing and the EGR rate are corrected to a side where combustion is stabilized) to ensure combustion stability is disclosed in Japanese Patent Application Laid-Open No. 58-16.
No. 0530 is proposed.

[0003]

By the way, in a region where there is a margin to the permissible stability limit, the stability exhibits a relatively flat characteristic with respect to the change in the air-fuel ratio, but the air-fuel ratio approaches the permissible stability limit. As the air-fuel ratio changes more steeply as the air-fuel ratio changes, overcorrection that goes beyond the target in the vicinity of the stability allowable limit is likely to occur, and the stability has exceeded the allowable limit due to this overcorrection. Hesitation and stumbling are inevitable. Therefore, in a conventional device that only performs feedback control based on stability during lean operation, the closer the control is to the vicinity of the stability limit, the more the response speed of feedback control is increased in order to suppress overcorrection and increase control stability. It is necessary to slow down, and in this case, the feedback control based on the stability may not be sufficiently performed due to a transient operation condition or the like.

Therefore, in order to enable control near the stability limit even in such an operation region, it is conceivable to use feedback control based on stability together with learning control. Explaining this, after the feedback control based on the stability is sufficiently performed in a steady state (the feedback control condition based on the stability) (see the left side of FIG. 2), the feedback correction amount of the air-fuel ratio (in FIG. (Indicated by a correction amount), control at the target stability becomes possible, and this feedback correction amount is stored as a learning value.
If the feedback control condition based on the degree of stability is not satisfied even during the lean operation (for example, at the time of gentle acceleration), this learning value is used as a correction amount (FIG. 2).
On the right side), control at the target stability can be performed without performing feedback control based on the stability. In FIG. 2, the ignition timing is ADV, and the air-fuel ratio is A.
/ F is abbreviated.

On the other hand, when the learning value becomes inappropriate due to some cause such as the influence of the purge gas, the response of the feedback control based on the stability is slow. During this time, the driver may feel uncomfortable due to unstable combustion.

In order to prevent the stability from being deteriorated when the learning value becomes inappropriate due to the influence of the purge gas or the like, a base learning value is introduced separately from the normal learning value. Later, the lean operation and the introduction of purge gas are prohibited, the air-fuel ratio feedback control is performed with the stoichiometric air-fuel ratio as the target value, and the base learning value is updated during the air-fuel ratio feedback control. 2) After the base learning value converges 3) At the time of lean operation, a learning value for making the air-fuel ratio richer is selected, and the air-fuel ratio is determined using the selected learning value. (See Japanese Patent Application Laid-Open No. 7-217470).
Applying the learning control in combination with the feedback control based on the stability described above. 1) Prohibiting the introduction of purge gas after starting, performing feedback control based on the stability, and updating the base learning value during the feedback control. 2) The introduction of purge gas is started after the convergence of the base learning value to update the normal learning value (hereinafter referred to as a normal learning value). 3) The base learning value is used before the convergence of the base learning value, and the base learning is performed again. After the convergence of the values, a normal learning value is selected, and the air-fuel ratio is open-controlled by using the selected learning value when the feedback control condition based on the stability during the lean operation is not satisfied.

However, the feedback control based on the stability requires the response speed to be slowed down as described above. Therefore, as described in Japanese Patent Laid-Open No. 7-217470, a normal learning value is set with the center value of the learning value as an initial value. It usually takes a long time for the learning value to converge,
During this time, the air-fuel ratio becomes excessively rich or leaner than the target stability equivalent value (when the air-fuel ratio becomes excessively richer than the target stability equivalent value, the deterioration of fuel efficiency and NOx When the amount of emission increases or the air-fuel ratio becomes excessively leaner than the value corresponding to the target stability, the drivability deteriorates due to unstable combustion.)

Therefore, the present invention sets the convergence value of the base learning value as the initial value of the normal learning value, thereby speeding up the convergence of the normal learning value, whereby the air-fuel ratio becomes richer than the target stability equivalent value. Or lean on the contrary.

[0009]

In the first invention, FIG.
As shown in the figure, means 21 for initially setting the air-fuel ratio at the time of the lean operation, and means 22 for storing the normal learning value LRDDML
Means 23 for storing a base learning value BSLDML;
Means 24 for determining whether or not the base learning value BSLDML has converged;
Means 25 for selecting the base learning value before the LDML converges and selecting the normal learning value after the base learning value converges
Means 26 for determining whether or not the condition for performing the feedback control based on the stability during the lean operation is provided. Based on the determination result, the learning value of the selected one when the feedback control condition based on the stability during the lean operation is not satisfied is obtained. Means for correcting the initially set air-fuel ratio during the lean operation.
Means for controlling the amount of fuel supplied to the engine so as to achieve the corrected air-fuel ratio during lean operation, means for detecting the stability FILDMP of the engine 29, and the stability during lean operation based on the determination result. When the feedback control condition is satisfied based on the base learning value BSLDML
Means 30 for updating the base learning value BSLDML so that the stability FILDMP of the engine coincides with the control target value LLSL of the stability before convergence, and feedback control based on the stability during lean operation based on the determination result. When the condition is satisfied and after the base learning value BSLDML converges, the normal learning value LRLDML is set such that the stability FILDMP of the engine matches the control target value LLSL of the stability with the convergence value of the base learning value BSLDML as an initial value. And means 31 for updating.

[0010] In a second aspect based on the first aspect, the normal learning value is a value for each learning area.

In a third aspect of the present invention, in the second aspect, the convergence value of the base learning value BSLDML is set as an initial value of the normal learning value in all the learning areas.

According to a fourth aspect of the present invention, in the second aspect, the base learning value BSLDML is composed of one data, the base learning value BSLDML is updated to a convergence value of the base learning value BSLDML, and the base learning value BSLDML is updated. Enrichment margin I according to the difference in driving conditions with each learning region
A value obtained by adding NTMGN is set as an initial value of the normal learning value for each learning area.

According to a fifth aspect of the present invention, in any one of the first to fourth aspects, the base learning value includes a learning value during a purge cut, and the normal learning value includes a time during a purge cut and when the purge cut is not performed. It is a learning value.

In a sixth aspect of the present invention, in any one of the first to fifth aspects, the stability detection value is an engine rotation fluctuation amount.

[0015]

According to the first aspect of the present invention, when shifting from updating of the base learning value to updating of the normal learning value, the convergence value of the base learning value is used as an initial value instead of the center value of the learning value. Since the update of the learning value is started, the learning value becomes closer to the required value than when the updating of the normal learning value is started from the center value of the learning value. Thus, the convergence of the normal learning value can be accelerated, so that the air-fuel ratio does not become excessively rich or lean on the contrary to the value corresponding to the target stability.

In the case where the base learning value shifts to the normal learning value, even when the convergence value of the base learning value deviates from the central value of the learning value, the normal learning value is updated using the central value of the learning value as an initial value. In the conventional device that starts the process, the air-fuel ratio is switched stepwise at the time of transition from the base learning value to the normal learning value, thereby causing a torque step before and after the transition and giving the driver discomfort. , First
According to the invention, the air-fuel ratio step at the time of the transition from the base learning value to the normal learning value is eliminated, so that discomfort to the driver due to the torque step before and after the transition can be avoided.

In the second aspect, since the learning value is set to a value for each learning region, the control accuracy to the target stability is improved when not performing the feedback control based on the stability during the lean operation.

In the fourth invention, the value obtained by adding the enrichment margin corresponding to the difference in the operating conditions between the area in which the base learning value is updated and each of the learning areas of the normal learning value to the convergence value of the base learning value is usually used. Since the initial value of the learning value is set for each learning region, even if the driving region when the base learning value that is one data is updated and the learning region of the normal learning value in which the base learning value is stored are separated, At the beginning of the shift from the base learning value to the normal learning value, the air-fuel ratio does not deviate leaner than the surge limit.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In FIG. 1, reference numeral 1 denotes an engine body, and intake air is supplied from an air cleaner to a cylinder through an intake pipe 8. The fuel is supplied to a control unit (C / U in the figure) so that a predetermined air-fuel ratio is obtained according to the operating conditions.
The fuel is injected from the fuel injection valve 7 toward the intake port of the engine 1 based on the injection signal from the engine 2.

The control unit 2 has a Ref signal and a one-degree signal from a crank angle sensor 4 built in the distributor, an intake air amount signal from an air flow meter 6, and an O ₂ sensor 3 installed upstream of the three-way catalyst 10. Air-fuel ratio (oxygen concentration) signal, a coolant temperature signal from a water temperature sensor 15, a throttle valve opening signal from a throttle sensor, a gear position signal from a transmission gear position sensor, a vehicle speed signal from a vehicle speed sensor 16, and the like. And controls the lean air-fuel ratio and the stoichiometric air-fuel ratio according to the conditions while judging the operating state based on these.

A three-way catalyst 10 is provided in the exhaust pipe 9 to reduce NOx in exhaust gas and oxidize HC and CO with maximum conversion efficiency during operation at a stoichiometric air-fuel ratio. The three-way catalyst 10 oxidizes HC and CO when the air-fuel ratio is lean, but has a low NOx reduction efficiency. However, the more the air-fuel ratio shifts to the lean side, the smaller the amount of NOx generated. If the air-fuel ratio exceeds a predetermined air-fuel ratio, it can be reduced to the same level as that obtained by purifying with the three-way catalyst 10. At the same time, the lean air-fuel ratio The more the fuel efficiency is improved. On the other hand, when operating at a lean air-fuel ratio, combustion tends to be unstable depending on the operating conditions.

Therefore, the engine is operated at a lean air-fuel ratio in a predetermined operating range where the load is not so large,
At the same time, the engine stability (combustion fluctuation amount) is detected, and feedback control is performed based on the stability at the lean air-fuel ratio so that the stability matches the control target value.
Stability can be ensured.

In this case, in a region where there is a margin up to the permissible stability limit, the stability shows a relatively flat characteristic with respect to a change in the air-fuel ratio (or ignition timing). As the air-fuel ratio (or ignition timing) changes steeply as the stability limit is approached, the overcorrection that goes beyond the target in the vicinity of the stability tolerance easily occurs, as in the conventional device, If the stability exceeds the allowable limit due to this overcorrection, hesitation and tumbling cannot be avoided. Therefore, the closer to the stability tolerance limit, the more
In a conventional device that needs to reduce the response speed of feedback control based on stability in order to suppress overcorrection and enhance control stability, feedback control based on stability cannot be performed sufficiently under transient operating conditions or the like.

Therefore, in order to enable control in the vicinity of the stability limit even in such an operation range, it is conceivable to use feedback control based on stability together with learning control. Explaining this, after the feedback control based on the stability is sufficiently performed in the steady state (which is the feedback control condition based on the stability) (see the left side of FIG. 2), the target stability is determined by the feedback correction amount of the air-fuel ratio. And the feedback correction amount is stored as a learning value. When the feedback control condition based on the stability is not satisfied even during the lean operation (for example, during slow acceleration), this learning value is used as a correction amount (see the right side of FIG. 2), and the feedback based on the stability is performed. Even if the control is not performed, the control with the target stability can be performed.

However, on the other hand, when the learning value becomes inappropriate due to some effect such as the influence of the purge gas, the response of the feedback control based on the stability is slow, and the response until the stability converges to a good state. During this time, the driver may feel uncomfortable due to unstable combustion.

In order to prevent the stability from deteriorating when the learned value becomes inappropriate due to the influence of the purge gas, etc., the lean operation and the introduction of the purge gas are prohibited after the start, and the target stoichiometric air-fuel ratio is set. A conventional device that performs feedback control of the air-fuel ratio as a value, updates the base learning value during the air-fuel ratio feedback control, starts introduction of purge gas after the convergence of the base learning value, and updates the normal learning value. Therefore, this conventional device is applied to a system in which learning control is used in combination with the above-described feedback control based on stability, and the introduction of purge gas is prohibited after starting, and feedback control based on stability is performed. It is conceivable to update the base learning value and to start the introduction of purge gas after the convergence of the base learning value to update the normal learning value. It is.

However, as described above, the feedback control based on the stability needs to slow down the response speed, so that the normal learning value is updated with the center value of the learning value as an initial value as in the conventional device. Therefore, it usually takes a long time for the learning value to converge, during which time the air-fuel ratio becomes excessively rich or leaner than the value corresponding to the target stability.

To cope with this, in the first embodiment of the present invention, the convergence value of the base learning value is set as the initial value of the normal learning value.

The contents of the control executed by the control unit 2 will be described with reference to the following flowchart.

The flowchart of FIG. 3 is for calculating a target fuel-air ratio Tdml for controlling to a predetermined air-fuel ratio, and is executed at regular intervals (for example, every 10 ms) or in a background job. In the following flowchart, when the process is executed at regular time intervals or in a background job, the timing is simply described as 10 ms, and the control cycle is omitted.

In step 1, it is determined whether or not the condition is the lean operation condition by the flag #FLEAN. # FLEAN = 1
Is a lean operating condition, and # FLEAN = 0 is a non-lean operating condition. The determination of the lean operation condition is not described, but may be a known one disclosed in Japanese Patent Application Laid-Open No. 6-272591.

When the engine is in the lean operation condition, the process proceeds to step 2, and a map value of an air-fuel ratio thinner than the stoichiometric air-fuel ratio is obtained by referring to a predetermined map (MDMLL map) with the rotation speed Ne and the load Tp.

In step 3, the learning value is referred to. The reference to the learning value will be described with reference to the flowcharts of FIGS.

4 and 5, first, at step 11, the base learning value BSLDML is obtained from the flag #FBLLTD.
To see if has converged. As described below with reference to FIG.
# FBLLTD = 0 indicates that the base learning value BSLDML has not converged, and # FBLLTD = 1 indicates that the base learning value BSLDML has converged. Therefore, B
When the SLDML has not converged, the process proceeds from step 11 to step 15 in principle, and the base learning value BSLDML is calculated as follows.
In addition, after the BSLDML has converged, as long as the update of the stabilized fuel-air ratio correction coefficient LLDML, which will be described later with reference to FIG. 10, is permitted, in principle, step 11 in FIG. 4 and step 21 in FIG. Proceeding to 27, the normal learning value LRDDML (k) for each learning area is selected.

The base learning value BSLDML is shown in FIG.
As will be described later, this is one piece of data irrelevant to the operation area. The learning region of the normal learning value LRLDML (k) includes the rotation speed Ne and the load T as shown in FIG.
It is divided into a plurality of regions using p as a parameter, and independent learning values are stored for each learning region. LRL DML
K in (k) is a number assigned to distinguish each learning region from 0 to 15 shown in FIG.

However, under the following conditions (1) to (4), the reference learning values are slightly different.

<1> When the base learning value is not converged (FIG. 4
If it is in the BSLDML update area, but is not in a state in which the update of the stabilized fuel-air ratio correction coefficient LLDML (described later) is permitted, the process proceeds to step 18 from steps 12 and 14 and proceeds to BS.
The value obtained by adding the error margin LRMGN to LDML is included in the reference learning value.

Although the LLDML update is permitted, the BS
If it is not in the update region of the LDML, the process proceeds from the steps 12 and 13 to the steps 16 and 17, and the normal learning value LRLDML of the learning region to which the rotation speed Ne and the load Tp belong at that time.
(K) is selected, and this LRDDML (k) is included in the reference learning value.

If the LLDML update is not permitted and B
If it is not the update area of SLDML, Steps 12 and 1
The process proceeds from step 4 to step 19, where the central value of the learning value is 10
0% is included in the reference learning value.

<2> After the base learning value converges (when proceeding to step 21 and subsequent steps in FIG. 5) When the normal learning value LRLDML (k) at that time is larger than the value obtained by subtracting the predetermined value SLLDML # from BSLDML. (BSLDML-SLDDML # ≦ LR
If the LDML (k) is not in the LLDML update permission state, the process proceeds from step 22 or 23 to step 25,
The value obtained by adding the error margin LRMGN to LRDDML (k) is included in the reference learning value.

BSLDML-SLDLDL over DML #
When RLDML (k) decreases (BSLDML-
If SLDLML #> LRLDML (k) is not in the LLDML update permission state, the process proceeds from Steps 22 and 26 to Step 28 where the error margin LRM is added to BSLDML.
The value obtained by adding GN is included in the reference learning value.

In this case, the reason why the base learning value is selected without selecting the normal learning value is that the air-fuel ratio becomes too lean if the normal learning value is selected until this time. This is to avoid this.

, Margin of error LRMG
The reason for adding N is as follows.

The following two operating conditions 1) The first operating condition: when the learning value is updated while performing the feedback control based on the stability during the lean operation; 2) The second operating condition: the same lean operation Consider the case in which the initially set air-fuel ratio is corrected with the learning value stored in the storage means instead of the feedback control based on the stability. Note that the learning value at this time is exactly the learning value referred to according to FIG. 4 if the base learning value has not yet converged,
After the convergence of the base learning value, the learning value is referred to according to FIG. However, simply, it is sufficient to consider the base learning value before the convergence of the base learning value and the normal learning value after the convergence of the base learning value.

In this case, when the air-fuel ratio error (hereinafter simply referred to as the air-fuel ratio error) when the intake air amount is changed while keeping the target air-fuel ratio constant is uniform regardless of the operating conditions, or When there is no error (in an ideal state), the stability under the second operating condition matches the stability under the first operating condition (that is, the air-fuel ratio under the second operating condition The target stability is controlled), but the air-fuel ratio error depends on the operating conditions (for example, load and speed).
, The stability under the second operating condition deviates from the stability under the first operating condition by the difference between the air-fuel ratio errors between the two operating conditions. Due to this deviation, when the stability under the second operation condition becomes worse than the stability under the first operation condition (that is, the air-fuel ratio becomes richer than the target stability equivalent value), the combustion fluctuation increases. This increases the discomfort to the driver.

More specifically, the left side of FIG. 6 shows an example in which the air-fuel ratio error changes to the right with respect to the intake air amount (the air-fuel ratio error is positive in the region where the intake air amount is small, The air-fuel ratio error changes to the minus side as the intake air amount increases). In this case, the steady position (indicated by a black circle) corresponds to the above-described first operating condition, the gentle acceleration position (indicated by an open circle) corresponds to the above-described second operating condition, and the characteristic of the air-fuel ratio error is right. Because of the decrease, there is a difference in the air-fuel ratio error between the two positions. Even in the steady position, the stability is not performed based on the stability based feedback based on the stability due to the air-fuel ratio error if the feedback control based on the stability is not performed and the learning is not advanced. Since the control is performed, the stability is controlled to the target stability even if an air-fuel ratio error exists, as shown in the upper right part of FIG. 6, and the feedback correction amount at this time is stored as a learning value.

On the other hand, at a moderate acceleration position where the air-fuel ratio error becomes negatively larger than the steady-state position, the air-fuel ratio is shifted toward the lean side from the initially set air-fuel ratio by the difference of the air-fuel ratio error from the steady position. Therefore, if the initially set air-fuel ratio shifted by this difference is corrected using the learning value stored at the steady position as the correction amount, the air-fuel ratio becomes smaller than the target stability equivalent value as shown in the lower right part of FIG. It is too lean.

This means that the characteristic of the air-fuel ratio error rises to the right on the left side of FIG.
The same applies to the case where the slow acceleration position is also on the plus side and the slow acceleration position is more on the plus side than the steady position). In this case, the air-fuel ratio error with the first operating condition The air-fuel ratio under the second operating condition becomes excessively richer than the value corresponding to the target stability by the difference.

Therefore, when the feedback control based on the stability in the lean operation is not established, the learning value is shifted toward the side where the torque generated by the engine increases by the difference between the operating condition at that time and the air-fuel ratio error between the learning representative point. Therefore, an error margin LRMGN is introduced. Learning value (BSLDM
L, LRLDML) are values in which the air-fuel ratio goes to the rich side as this value increases, and the error margin LRMGN is the air-fuel ratio error between the two operating conditions (the first operating condition and the second operating condition described above). Since LRMGN is given a value of 0 or a positive value, adding LRMGN shifts the initially set air-fuel ratio during lean operation to the rich side.

More specifically, as shown in FIG. 7, the LRMGN indicates the load Tp and the learning representative point (the stationary position or the first position shown in the left side of FIG. 6) when the feedback control based on the stability is not performed even during the lean operation. The absolute value | Tp-LRTP | of the difference from the load LRTP) is a parameter. When | Tp-LRTP | is 0, the minimum value is 0, and the value increases as | Tp-LRTP | It is.

As shown on the left side of FIG. 6, (i) when the air-fuel ratio error at the steady position is on the minus side and the air-fuel ratio error at the slow acceleration position is more negative than the steady position, When the learning value is referred to as the correction amount at the position of the gentle acceleration, the air-fuel ratio exceeds the target stability equivalent value by the difference of the air-fuel ratio error from the steady position and becomes excessively lean. Conversely, (ii) when the air-fuel ratio error at the steady position is on the minus side and the air-fuel ratio error at the slow acceleration position is minus and smaller than the steady position (not shown), the slow acceleration is not performed. When the correction value is set by referring to the learning value at the position, the air-fuel ratio is richer than the target stability equivalent value by the difference of the air-fuel ratio error from the steady position (that is, the side where the combustion fluctuation falls). Therefore, in this case, discomfort to the driver is not increased. Therefore, it is only in the above (i) that the air-fuel ratio must be shifted to the lean side in order to eliminate the discomfort to the driver due to the increase in the combustion fluctuation. However, the control unit cannot distinguish between the case (i) and the case (ii).
Also in the case of (ii), in order to shift the air-fuel ratio to the rich side, the absolute value of the difference between Tp and LRTP is used as a parameter as shown in FIG.

When the cases (i) and (ii) can be distinguished in advance, the air-fuel ratio is shifted to the lean side only in the case (i), and conversely, in the case (ii). Is shifted to the rich side, the air-fuel ratio can be controlled to a value corresponding to the target stability also in the case of (ii).

The stationary position (learning representative point) shown on the left side of FIG. 6 is set by selecting a condition in which the load in the lean operation region is medium. The reason is as follows. If the error margin is large, the air-fuel ratio changes greatly. Therefore, it is desirable that the error margin be as small as possible. Also, the difference in the air-fuel ratio error between the two positions (the steady position and the slow acceleration position) is large. Since it is necessary to increase the error margin as much as possible, when the steady position is in the middle of the load in the lean operation region, the difference in the air-fuel ratio error between the two positions (therefore, the error margin) is minimized. It is presumed that it becomes.

In the case where the air-fuel ratio error changes according to the operating conditions, the intake air amount is shown on the left side of FIG. 6 as a representative of the operating conditions. However, the air-fuel ratio error may change depending on the engine speed. As shown in FIG. 7, the error margin also uses the absolute value | Ne−LRNE | of the difference between the rotation speed Ne and the rotation speed LRNE of the learning representative point as a parameter, as shown in FIG. 7, and | Ne−LRNE | is large. I see LRMGN
Is increased.

In this way, the error margin LRMGN
In the second operating condition, the learning value is shifted to the rich side by the difference of the air-fuel ratio error between the second operating condition and the first operating condition, so that the first operating condition and the second operating condition Even if there is an air-fuel ratio error difference between the two, the air-fuel ratio can be controlled to a value corresponding to the target stability. For example, FIG. 8 rewrites the characteristic on the right side of FIG. 6 again. When the learning value is referred to as the correction amount at the time of gentle acceleration (second operating condition), the air-fuel ratio is set at the steady state (the second operation condition). (One operating condition) and the air-fuel ratio error exceeds the value corresponding to the target stability and excessively leans. In the present embodiment, however, the difference between the two operating conditions during slow acceleration and steady state is obtained. Because the air-fuel ratio error is shifted to the rich side by the difference (see the right side of FIG. 8), the air-fuel ratio is controlled to the target stability equivalent value even if the air-fuel ratio error exists between the two operating conditions. You can.

When the learning value is referred to in this manner, the process returns to step 4 in FIG. 3, and the reference learning value is multiplied by the map fuel-air ratio under the lean operation condition, that is, Tdml = Mdml × reference learning value. The target fuel-air ratio Tdml is calculated by the equation (1).

Since the reference learning value increases as the engine rotation fluctuation increases during feedback control based on the stability, the target fuel-air ratio Tdml increases as the combustion fluctuation increases, that is, the target fuel-air ratio becomes It shifts to the rich side. When the engine is not in the lean operation condition, the stoichiometric air-fuel ratio or a map fuel-air ratio of a value higher than the stoichiometric air-fuel ratio is referred to in accordance with a predetermined map (MDMLS map) in the same manner as the above-mentioned MDMLL map. In step 6, the fuel-air ratio Mdml is directly entered into Tdml.

Although not shown here, the air-fuel ratio is gently switched to prevent a sudden change in the torque and to ensure the stability of the driving performance. The fuel-air ratio correction coefficient DML may be obtained by performing a damper operation.
No. 272591).

Next, the flowchart of FIG. 9 is for detecting the stability, and is executed for each input of the Ref signal independently of FIG. The Ref signal is a signal generated every 180 degrees of the crank angle in the case of a four-cylinder engine and every 120 degrees of the crank angle in the case of a six-cylinder engine.

In steps 31 and 32, the Ref signal period T
REF is measured and converted to an engine speed R0 by multiplying the reciprocal of the REF by a constant K1. In step 33, band reject filter processing is performed to remove noise generated at a specific frequency due to an error in the crank angle sensor. Subsequently, a band-pass filter process for extracting a vehicle resonance frequency component that is perceived as vibration by the human body is performed in step 3
In step 4, the absolute value processing is performed in step 35, and the result is filtered in step 36 by the filter output F.
Let it be ILOUT. By the band-pass filter processing, the DC component and the high-frequency component are removed from the time-series data of the engine speed, and only the fluctuation component of the vehicle resonance frequency band is left. Also, the magnitude of the fluctuation is extracted by the absolute value processing. Therefore, the filter output FILOUT
Represents the magnitude of the fluctuation component included in the vehicle resonance frequency band in the engine rotation fluctuation.

In step 37, this filter output FIL
OUT is subjected to a weighted average process, and the result is referred to as a stability index F
ILDMP.

The flowcharts in FIGS. 10 and 11 are for calculating the stabilized fuel-air ratio correction coefficient LLDML as the stability correction amount, and are executed independently of FIG. However,
It is assumed that the stability index FILDMP has been obtained according to FIG. 9 immediately before executing the operation of step 42 in FIG.

In step 41 of FIG. 10, it is determined whether or not the updating of the stabilized fuel-air ratio correction coefficient LLDML is permitted. The specific contents for this are shown in FIG. This determination is made by checking the contents of the steps of FIG. 11 one by one, and when all the items are satisfied, L
Allow update of LDML, and if even one goes wrong, LLD
Update of ML is not permitted.

Step 51: The flag # FLEAN = 1 (lean operation condition). Step 52: The absolute value | ΔTp | of the variation of the load Tp from before 50 ms is equal to or less than a predetermined value LLDTP #. : The absolute value | ΔNe | of the amount of change of the rotational speed Ne from 50 ms before is less than or equal to a predetermined value LLDNE #. Step 54: 50 ms of the throttle valve opening TVO
The absolute value | ΔTVO | of the amount of change from before is a predetermined value LLDT
VO # or less. Step 55: No cylinder is cutting fuel. Step 56: No abnormality (abbreviated as NG in the figure) in the crank angle sensor. Step 57: Flag # FCNST = 1 (during slow acceleration or steady operation). ), The update of LLDML is permitted in step 58, and if not, the flow proceeds to step 59 and the update of LLDML is not permitted.

The above steps 51 to 57 are conditions for accurately updating the LLDML (including the detection of the stability). In the method of estimating the combustion fluctuation from the rotation fluctuation using the correlation between the combustion fluctuation and the rotation fluctuation as in the present embodiment,
When the engine speed fluctuates due to acceleration / deceleration, there is a concern that the amount of rotation change due to the acceleration / deceleration cannot be separated from the rotation fluctuation caused by combustion fluctuation, and the magnitude of combustion fluctuation is estimated to be larger than actual. The above steps 52, 53, 5
According to 4, 57, in an unsteady state (at the time of sudden acceleration or sudden deceleration), L
The update of the LDML is prohibited. Step 5
The flag #FCNST of No. 7 is based on a comparison between the vehicle speed change amount ΔVSP and a predetermined value LLCDVH, and ΔVSP <LLCDVH.
This flag is set to “1” at the time of slow acceleration or steady state (see FIG. 12), and is set to “0” at ΔVSP ≧ LLCDVH (at the time of rapid acceleration).

When it is determined whether or not the update of LLDML is permitted in this way, the process returns to step 41 of FIG. 10, and the process returns to step 42 only when the update of LLDML is permitted.
Proceed to the following.

In step 42, the stability index FILDM
P (already obtained in step 37 in FIG. 9) is read, in step 43, a stability target value LLST is obtained by referring to a predetermined map from the load and the number of revolutions, and using these values, in step 44, LLDML (New) = LLDML (old) + GLLFB # × (FILDMP−LLSL) (2) where LLDML (new): stabilized fuel-air ratio correction coefficient after update LLDML (old): stabilized fuel-air ratio before update Correction coefficient GLLFB #: Gain The stable fuel-air ratio correction coefficient (initial setting to 100% at start-up) LLDML is updated by the equation.

From the equation (2), the stabilized fuel-air ratio correction coefficient LLD
ML is obtained by setting the stability index FILDMP to the stability target value LLS.
The value increases as the value exceeds T, that is, as the combustion deteriorates.

The gain GLLFB # in the equation (2) is a value that determines the response speed of the feedback control based on the stability, and is a relatively small value in order to suppress overcorrection and increase the control stability, similarly to the conventional device. (The response speed of the feedback control has been reduced.)

In step 45, the lower limiter LDMLM
N and the upper limiter LDMLMX are read, and step 4
At 6, the LLDML limiter process is performed. Here, as shown in FIG. 13, the base learning value BSLDML
Is subtracted from the lower limit limiters LDMLMN and BSLDML to a predetermined value LRC
The value obtained by adding HG # is defined as the upper limiter LDMLMX.

In step 47 of FIG. 10, the learning value is updated. This will be described with reference to the flowchart of FIG.

First, at step 71, a learning counter (initialized to 0 at startup) CBSLDM and a predetermined value NBSLDM
# Compare. The learning counter CBSLDM is for measuring the number of times the base learning value BSLDML has been updated. If CBSLDM <NBSLDM #, it is determined that the base learning value BSLDML has not converged, and the routine proceeds to step 72, where "0" is set to the flag #FBLDTD.
The flag # FBLLTD = 0 indicates that BSLDML has not converged. On the other hand, #FBLL
TD = 1 indicates that BSLDML has converged as described later.

In steps 73 and 74, whether the load Tp is in a predetermined range (LLCTPL ≦ Tp ≦ LLCTPH) or not, and if the rotational speed Ne is in a predetermined range (LLCNEL ≦ Ne ≦)
If both are in the predetermined area, it is determined in step 75 that "1" is set in the flag #FLRLLC. The flag # FLRLLC = 1 indicates that it is in the base learning value update area (see FIG. 15). On the other hand, when either the load or the rotational speed is not in the predetermined region, the flag # FLRLLC = 0 (described later).

In steps 76 and 77, the stabilized fuel-air ratio correction coefficient LLDML at that time is used as a base learning value BSLDM.
L, and the learning counter CBSLDM is incremented.

That is, when the engine enters the lean operation state for the first time after the engine is started and the feedback control based on the stability is started, if the base learning value BSLDML has not converged, before the normal learning value LRLDML (k) is updated. Since the purge is cut to update the base learning value BSLDML, the base learning value BSLDML converges to the required value of the stabilized fuel-air ratio correction coefficient LLDML determined only by the characteristics of the engine without being affected by the purge. After the base learning value BSLDML converges, the base learning value is not updated after that, and the process proceeds to the update of the normal learning value LRLDML (k).

However, since the purge must be continuously cut until the base learning value BSLDML converges, it is necessary to converge the base learning value BSLDML in as short a time as possible. Therefore, the base learning value BSLDM
L does not have a learning map like the normal learning value LRDDML (k), and is composed of one data regardless of operating conditions.

On the other hand, since the same base learning value BSLDML is updated and referred to irrespective of the operating conditions, if the operating conditions at the time of updating the base learning value BSLDML differ greatly from those at the time of reference, the air-fuel ratio at the time of reference is different. There is a concern that the control accuracy will deteriorate. However, the base learning value BSLD
The base learning value BSLDML is referred to when the ML is updated and the operating conditions are significantly different from each other when the base learning value becomes unstable due to the influence of the introduction of the purge gas (specifically, after the base learning value converges, the BSLDML- SLLDM
When L #> LRLDML (k) and the LLDML update is not permitted, BSDLML is referred to in step 28 of FIG. 5), and the reference period is also temporary, so there is no major problem.

The base learning value BSLDML is stored in the backup memory. BS updated during driving this time
The LDML is held as it is until the next operation. The initial setting value of BSLDML is 100% which is the center value of the learning value.

At step 78, the base learning value BSLDM
The same value of the stabilized fuel-air ratio correction coefficient LLDML as in L is entered in the normal learning values LRLDML (k) of all the learning regions. The learning area of the normal learning value LRLDML (k) is
As shown in FIG. 16, the engine speed is divided into a plurality of regions by using the rotation speed Ne and the load Tp as parameters, and independent learning values are stored for each learning region. K of normal learning value LRLDML (k)
Is a number assigned to distinguish each learning region from 0 to 15 shown in FIG. Therefore, the normal learning values L of all the learning regions are obtained by the operation of step 78.
The convergence value of the base learning value is stored in RLDML (k) as an initial value.

If CBSLDM <NBSLDM # and the base learning value update area continues, steps 75 to 78 are repeated until CBSLDM ≧ NBS
When LDM # is reached, the process proceeds to step 80 to indicate that the base learning value BSLDML has converged, and the flag #FBL
"1" is set to LTD, and thereafter, the process proceeds to steps 81 and 82, and the normal learning value LRLDML (k) is updated for each learning region. When the rotation speed Ne and the load Tp belong to the five learning regions shown in FIG. 16, the stabilized fuel-air ratio correction coefficient LLDML is inserted only into the normal learning value LRDDML (5) of the determined five learning regions. .

Here, the learning region of the normal learning value LRDDML (k) is obtained by dividing the entire operation region into a plurality of regions using the rotation speed Ne and the load Tp as parameters as shown in FIG. Have independent learning values stored therein. LRDDML (k) represents a learning value for each learning region. K in LRDDML (k) is a number assigned to distinguish each learning region. When there are 16 learning regions, k can take an integer from 0 to 15, while CBSLDM <
If the NBSLDM # is not the update area of the base learning value BSLDML, the process proceeds from step 73 or 74 to step 7.
The process proceeds to 9, and “0” is set to the flag #FBLDTD.

At this time, the operations of steps 81 and 82 are further performed. By this operation, the value of LLDML enters the normal learning value, and as a result, LLDML also enters the reference learning value (see steps 11, 12, 13, 16, and 17 in FIG. 4). That is, in the flowchart of FIG. 14, since the process of once storing the stabilized fuel-air ratio correction coefficient LLDML as a learning value is always performed, the step 8 is also performed at this time.
It is proceeding to the operations of 1, 82. 14, 4, and 5, as a result, the learning value simultaneously enters the reference learning value, and the air-fuel ratio correction uses only the reference learning value (see steps 3 and 4 in FIG. 3). It works in the same way as the feedback control. Before the convergence of BSLDML (that is, the LLDML update permission state is
When it is not in the update area of L), the normal learning value is used as a dummy.

The flowchart of FIG. 17 shows the control operation for calculating and outputting the fuel injection pulse width using the target fuel-air ratio Tdml obtained in this way, and is executed after FIG.

First, in step 81, using the target fuel-air ratio Tdml, a target fuel-air ratio equivalent amount Tfbya is calculated by the following equation: Tfbya = Dml + Ktw + Kas (3)

Here, Ktw is the fuel increase according to the cooling water temperature, and Kas is the fuel increase immediately after starting. In addition,
The unit of Tfbya in Expression (3) is%, but the unit of Tfbya in Expression (4) described later is an anonymous number.

Next, in step 82, the output of the air flow meter is A / D converted, linearized, and the intake air flow rate Q
Is calculated. Then, in step 83, from the intake air flow rate Q and the engine speed Ne, a basic injection pulse width Tp at which a stoichiometric air-fuel ratio can be obtained is obtained as Tp = K × Q / N. K is a constant.

Then, in step 84, based on this Tp, one fuel injection pulse width Ti in the sequential injection is calculated by the following equation: Ti = (Tp + Kathos) × Tfbya × (α + αm−1) × 2 + Ts (4) Is calculated.

Here, Kathos is a transient correction amount, α is an air-fuel ratio feedback correction coefficient, αm is an air-fuel ratio learning correction coefficient calculated based on α, and Ts is an actual valve opening after the injection valve receives an injection signal. This is an invalid pulse width for compensating for an operation delay before the operation. Also, since equation (4) is an equation in sequential injection (injection is performed once for every two revolutions of the engine in four cylinders in accordance with the ignition order of each cylinder),
There is a number 2 before Ts. However, at the time of the lean condition, α and αm are fixed to predetermined values.

As described above, the basic injection pulse width T
p is used as the engine load. Although this Tp is used here for simplicity, a weighted average value of the Tp in consideration of the intake pipe volume can be used in place of the Tp.

Next, in steps 85 and 86, the fuel cut is judged. In steps 87 and 88, if the fuel cut condition is satisfied, the invalid injection pulse width Ts is stored in the output register, otherwise Ti is stored in the output register. In preparation for injection at a predetermined injection timing according to the output of

The flowchart of FIG. 18 is for controlling the opening and closing of the purge valve, and is executed independently of FIGS. 3, 10, and 17.

The opening / closing control of the purge valve is performed in steps 91 and 9
By checking the contents of 2, 93 and 94 one by one, the purge valve is opened when all of the items are satisfied, and the purge valve is closed when even one is not satisfied.

Step 91: The idle switch is not ON. Step 92: The cooling water temperature Tw is within a predetermined range (TWCPL).
≦ Tw ≦ TWCPH). Step 93: When the load Tp is in a predetermined region (TPCPL ≦ T)
Step 94: flag # FBLLTD = 1 (BS)
(After convergence of LDML) Sometimes, at step 95, the purge valve is opened; otherwise, the routine goes to step 96, at which the purge valve is closed. That is, the base learning value BSLDML is updated during the purge cut. The update of the base learning value BSLDML at the time of purging is prohibited because only the air-fuel ratio error due to the variation in the flow characteristics of the components (fuel injection valve or air flow meter) related to the basic air-fuel ratio or the deterioration over time due to the base learning value BSLDML. Because we want to absorb.

Here, the operation of the present embodiment will be described. As shown in FIG. 19, after a purge cut has been performed for a while after the start of the lean operation, purge gas is introduced from the canister, and during the purge, the operating condition shifts from the learning region 1 to the learning region 2, and the learning region 1 Suppose you come back to. However, the base learning value BSLDML has a central value of 10%.
0% is the normal learning value LRDDML for each learning area.
It is assumed that (k) contains 100% of the central value in all the learning regions. Note that the learning areas 1 and 2 in FIG. 19 merely mean that the learning areas are different, and the normal learning values of the learning areas 1 and 2 are LRLDML (1), LRL
It is not DML (2).

Base learning value BSLDML, normal learning value L
Since the RLDML (k) and the stabilized fuel-air ratio correction coefficient LLDML are all 100%, the air-fuel ratio (and ignition timing) at the time of the lean operation is initially set so as to be always in a stable state. At the start timing of the lean operation, the air-fuel ratio is richer than the target air-fuel ratio (the stability index FILDMP becomes smaller than the stability target value LLSL), so that the base learning value is updated to a smaller side.
Immediately before the timing of t2, the base learning value converges to the value of a, and the air-fuel ratio at this time matches the target air-fuel ratio (see the solid line in the third row).

On the other hand, in the present embodiment, when updating the base learning value, unlike the conventional device, the same value as the value of the stabilized fuel-air ratio correction coefficient LLDML stored in the base learning value is applied to all the learning regions. Although not shown, the normal learning values of the learning areas 1 and 2 follow the same change as the base learning value, and a is set immediately before the timing of t2.
Value.

At time t2 when the base learning value has converged, the routine shifts to updating of the normal learning value and the purge is started.
The air-fuel ratio temporarily becomes richer than the target air-fuel ratio, and the stability is improved by that amount (the stability index FILD is returned again).
MP becomes smaller than the stability target value LLSL), so that the normal learning value of the learning area 1 is a convergence value of the base learning value.
The learning value is updated to a smaller value, and is limited to b, which is the lower limit of the learning value, from the timing of t3 (see the solid line in the third row). As described above, in the present embodiment, when the transition from the update of the base learning value to the update of the normal learning value is performed in the same learning region 1, the convergence value of the base learning value is used instead of the central value of the learning value of 100%. A
Since the update of the normal learning value is started with the initial value as the initial value, the normal learning value becomes closer to the required value than when the update of the normal learning value is started from 100% which is the center value of the learning value. The convergence of the values is accelerated, so that the air-fuel ratio does not become excessively rich or lean on the contrary to the value corresponding to the target stability.

In order to maintain the target air-fuel ratio with respect to the change in the purge gas concentration shown in the drawing, the learned value usually has to change abruptly as required, but the gain of the above equation (2) is Since the value is set to a relatively small value in order to suppress the overcorrection and enhance the control stability, a response delay occurs in the normal learning value with respect to the required value.

During the period from t4 to t5, the operating condition is not steady (the LLDML update is not permitted), so that the control based on the stability is performed when the feedback control condition is not satisfied (shown by open control in the figure). Then, either the base learning value or the normal learning value is referred to in accordance with the flow shown in FIGS. In the period from t4 to t5, BSLDML-SLDDML #> LRLDML (k)
Therefore, the flow proceeds to steps 22, 26, and 28 in FIG. 5, and a value obtained by adding the error margin LRMGN to a, which is the convergence value of the base learning value BSLDML, is referred to (third).
(See the solid line in the column).

When switching from the learning area 1 to the learning area 2 at the timing of t5, the normal learning value of the learning area 1 becomes b
, And the process proceeds to update the normal learning value of the learning area 2. At this time, the normal learning value of the learning area 2 contains a which is the same value as the convergence value of the base learning value BSLDML.

However, during the period from t5 to t6, since the operating condition is not steady (the LLDML update is not permitted), the control is performed when the feedback control condition based on the stability is not satisfied. Therefore, the learning value is referred to. In the period from t5 to t6, BSLD
Since ML-SLLDML # <LRLDML (k), the flow proceeds to steps 22, 23, and 25 in FIG. 5, and the error margin LRM is added to a, which is the initial value of the normal learning value of the learning area 2.
The value to which GN has been added is referred to (see the solid line in the third row).

The normal learning value of the learning area 2 is updated from the timing of the steady state t6 with the value of a, which is the convergence value of the base learning value, as an initial value. Since the fuel ratio becomes rich, the normal learning value in the learning area 2 is updated to be smaller than a, and matches the required value at the timing of t7 (see the solid line in the third row). As described above, the value of a, which is the convergence value of the base learning value, is also set as the initial value of the normal learning value of the learning region 2. Than when starting to update the learning value,
Normally, the learning value becomes closer to the required value, and the air-fuel ratio is not unnecessarily made rich. As a result, an increase in NOx can be suppressed.

The normal learning value c at the timing of t8 when the operating condition is not steady (the LLDML update is no longer permitted) is held as the normal learning value of the learning region 2, and the stability is maintained from t8 to t9. The process proceeds to the control when the feedback control condition is not satisfied based on the learning value, and the learning value is referred to in accordance with the flow charts of FIGS. In this case, BSL
FIG. 5 from DML-SLL DML # <LRL DML (k)
Steps 22, 23, and 25 are referred to, and the value obtained by adding the error margin LRMGN to c, which is the value of the normal learning value of the learning area 2 held at the timing of t8, is referred to (the third stage). Solid line).

Returning to the learning area 1 at the timing of t9,
The process proceeds to the update of the normal learning value of the learning area 1. At this time, the normal learning value of the learning area 1 contains the value of b, which is the normal learning value of the learning area 1 held at the timing of t4.

However, during the period from t9 to t10, the operating conditions are not steady (the LLDML update is not permitted).
Therefore, the learning value is referred to according to the flow of FIGS. 4 and 5 by the control when the feedback control condition based on the stability is not satisfied. In this case, BSLDML-SLLDM
From L #> LRL DML (k), steps 22 and 2 in FIG.
6 and 28, the value obtained by adding the error margin LRMGN to the initial value a of the base learning value is referred to (third).
(See the solid line in the column).

From the timing of steady time t10, t4
The normal learning value of the learning area 1 is updated with the lower limiter value b held at the timing of (1) as an initial value. t
Since the purge gas concentration at t10 is smaller than the purge gas concentration at the timing of timing 4, when the update of the normal learning value of the learning region 1 is restarted at the timing of t10, the air-fuel ratio is reduced by the decrease in the purge gas concentration. Becomes lean at once. Since the leanness reduces the stability (the stability index FILDMP becomes larger than the stability target value LLSL), the side where the normal learning value of the learning region 1 becomes larger than the value of b (the side where the air-fuel ratio is made rich). , And coincides with the required value at the timing of t11 (see the solid line in the third row).

In the third row of FIG. 19, LRMGN has an open margin, SLLDML # has an open learning value selection getter, and BSLDML-SLLD has an open learning value selection slice level.
ML #.

As described above, in the present embodiment, when the transition from the update of the base learning value to the update of the normal learning value is performed in the same learning region 1, the base learning value is not changed from 100% which is the center value of the learning value. Since the update of the normal learning value starts with the convergence value of the initial value as the initial value, the central value of the learning value of 10
Since the learning value becomes closer to the required value than when the normal learning value is updated from 0%, the convergence of the normal learning value is suppressed even if the normal learning value update speed is originally low. Therefore, the air-fuel ratio does not become excessively rich or lean on the contrary to the target stability equivalent value.

Further, in the case where the base learning value shifts from the base learning value to the normal learning value in the same learning region, even when the convergence value of the base learning value deviates from 100%, 100% which is the center value of the learning value is set as the initial value. In the conventional device for starting the update of the normal learning value, the air-fuel ratio is switched in a stepwise manner when shifting from the base learning value to the normal learning value, thereby causing a torque step before and after the shift and causing discomfort to the driver. However, in the present embodiment, it is possible to eliminate the air-fuel ratio step at the transition from the base learning value to the normal learning value, thereby avoiding discomfort to the driver due to the torque step before and after the transition. can do.

The flowchart of FIG. 20 is the second embodiment, and corresponds to FIG. 14 of the first embodiment. Only the step 101 is different from FIG. The same steps as those in FIG. 14 are denoted by the same step numbers.

Since the base learning value is not the value for each learning region but one data, the more the driving region at the time of updating the base learning value and the learning region of the normal learning value in which the base learning value is stored, the further apart the learning region becomes. The accuracy as the initial value of the learning value is deteriorated. Therefore, in the second embodiment, at the beginning of the transition from the base learning value to the normal learning value, the base learning value is set so that the air-fuel ratio does not deviate leaner than the surge limit due to the influence of the deterioration of the accuracy of the initial value. The value obtained by adding a predetermined enrichment margin to the convergence value is set as the initial value of the normal learning value.

The part different from the first embodiment will be specifically described. In step 101 of FIG. 20, a subroutine is executed. This subroutine will be described with reference to the flowchart of FIG.

In step 111, 0 is entered in the area number k. In step 112, the enrichment margin INTMGN (k) for this area is obtained with reference to the map shown in FIG.
The value obtained by adding (k) to the stabilized fuel-air ratio correction coefficient LLDML in step 113 is used as the enrichment margin INTM
Normal learning value LRLD in the same region as when referring to GN (k)
Put in ML (k).

As shown in FIG. 22, the enrichment margin INT
The MGN (k) region is divided in the same manner as the learning region of the normal learning value shown in FIG. 16 and the region number is also the same as the normal learning value, and each region has an independent value. I
This is because the magnitude of NTMGN (k) needs to be determined based on the difference in operating conditions between the area where the base learning value is updated and each learning area of the normal learning value.

In step 114, k and 15 are compared, and k is determined.
If ≤15, k is incremented in step 115, and the operations in steps 112 and 113 are repeated. Eventually, when k> 15, the subroutine of FIG. 21 ends. Before performing the operations of steps 112 and 113 for all the learning areas, the control interval of 10 m in FIG.
s does not elapse.

As described above, in the second embodiment, the enrichment margin corresponding to the difference in the operating conditions between the area in which the base learning value is updated and each learning area of the normal learning value is added to the convergence value of the base learning value. Since the added value is set as the initial value of the normal learning value, the driving region when the base learning value, which is one data, is updated is separated from the learning region of the normal learning value in which the base learning value is stored. However, the air-fuel ratio does not deviate leaner than the surge limit at the beginning of the transition from the base learning value to the normal learning value.

Expression (2) above (expression for updating LLDML)
Has been described using the integral control method, but a proportional control method or a proportional integral method may be used.

In the embodiment, the base learning value BSLDM
L, the description has been given of the case where the means for storing the normal learning value LRLDML (k) is a backup memory (for example, a backup RAM), but may be a simple RAM. Even in this case, after the base learning value converges during the lean operation and the feedback control based on the stability, if the feedback control based on the stability is stopped even during the lean operation, the convergence value of the base learning value is set to the initial value. By starting the update of the normal learning value, the convergence of the normal learning value can be accelerated.

In the embodiment, the normal learning value is set to a value for each learning region, but may be constituted by one data.

Finally, in FIG. 1, reference numeral 11 denotes lean NOx.
It is a catalyst. In the control using this catalyst, when it is necessary to once release NOx adsorbed to the limit of the occluded substance of this catalyst, HC, C, which are unburned components in the exhaust gas,
When the amount of O is equal to all NOx (NO released from the storage material)
The air-fuel ratio is enriched so as to exceed the required amount for reducing both x and NOx in the exhaust gas without excess and deficiency, and then immediately returns to the stoichiometric air-fuel ratio at a predetermined recovery speed. The description is omitted because it is not directly related (see Japanese Patent Application No. 7-101149).

[Brief description of the drawings]

FIG. 1 is a control system diagram of a first embodiment.

FIG. 2 is a characteristic diagram for explaining the principle of stability learning control.

FIG. 3 is a flowchart for explaining calculation of a target fuel-air ratio Tdml.

FIG. 4 is a flowchart for describing reference to a learning value.

FIG. 5 is a flowchart illustrating reference to a learning value.

FIG. 6 is a characteristic diagram illustrating an influence on a stability learning control due to a difference in an air-fuel ratio error.

FIG. 7 is a characteristic diagram of an error margin LRMGN.

FIG. 8 is a characteristic diagram for explaining an effect of an error margin LRMGN.

FIG. 9 is a flowchart illustrating detection of stability.

FIG. 10 is a flowchart for explaining calculation of a stabilized fuel-air ratio correction coefficient LLDML.

FIG. 11 is a flowchart illustrating a determination of permission to update a stabilized fuel-air ratio correction coefficient LLDML.

FIG. 12 is a waveform chart for explaining a flag #FCNST.

FIG. 13 is a flowchart illustrating the calculation of a learning limiter.

FIG. 14 is a flowchart illustrating updating of a learning value.

FIG. 15 is an explanatory diagram of an update area of a base learning value.

FIG. 16 is a region diagram of a normal learning value.

FIG. 17 is a flowchart for explaining calculation of a fuel injection pulse width and its output.

FIG. 18 is a flowchart for explaining purge valve control.

FIG. 19 is a waveform chart for explaining the operation of the first embodiment.

FIG. 20 is a flowchart illustrating updating of a learning value according to the second embodiment.

FIG. 21 is a flowchart illustrating a subroutine of FIG. 20;

FIG. 22 shows a margin for enrichment INTMG of the second embodiment.
It is an area figure of N (k).

FIG. 23 is a diagram corresponding to the claims of the first invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Engine body 2 Control unit 4 Crank angle sensor 6 Air flow meter 7 Fuel injection valve

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FI F02M 25/07 550 F02M 25/07 550R

Claims (6)

[Claims] A means for initially setting an air-fuel ratio during lean operation; a means for storing a normal learning value; a means for storing a base learning value; and a means for determining whether or not the base learning value has converged. Means for selecting the base learning value before the base learning value converges from the determination result, and the normal learning value after the base learning value converges, and a condition for performing feedback control based on stability during lean operation. Means for determining whether or not the air-fuel ratio initially set during lean operation is corrected with the selected learning value when the feedback control condition based on the stability during lean operation is not satisfied based on the determination result. Means for controlling the amount of fuel supplied to the engine so as to achieve the corrected air-fuel ratio during lean operation; means for detecting the stability of the engine; and Means for updating the base learning value such that the stability of the engine matches the control target value of the stability when the feedback control condition based on the stability during the engine operation is satisfied and before the base learning value converges. When the feedback control condition based on the stability at the time of lean operation is satisfied based on the determination result and after the base learning value converges, the stability of the engine is controlled with the convergence value of the base learning value as an initial value. Means for updating the normal learning value so as to coincide with a target value. 2. The air-fuel ratio control device for an engine according to claim 1, wherein the normal learning value is a value for each learning region. 3. The engine air-fuel ratio control device according to claim 2, wherein a convergence value of the base learning value is set as an initial value of a normal learning value in all learning regions. 4. The method according to claim 1, wherein the base learning value is composed of one data, and a convergence value of the base learning value is set according to a difference in an operating condition between a region for updating the base learning value and each learning region of the normal learning value. 3. The engine air-fuel ratio control device according to claim 2, wherein a value obtained by adding the enrichment margin is set as an initial value of the normal learning value for each learning region. 5. The method according to claim 1, wherein the base learning value is a learning value during a purge cut, and the normal learning value is a learning value including a time during a purge cut and a time when a purge cut is not performed. An air-fuel ratio control device for an engine according to any one of the preceding claims. 6. The air-fuel ratio control device for an engine according to claim 1, wherein the detected stability value is an amount of rotation fluctuation of the engine.