JP2505445B2 - Ignition timing control device for internal combustion engine - Google Patents

Description

TECHNICAL FIELD The present invention relates to an ignition timing control device for an internal combustion engine, and more particularly to an improvement of a device for performing learning control.

(Prior Art) A device in which a learning function is added to an electronically controlled ignition timing control device has been proposed (see Japanese Patent Laid-Open No. 55-78168).
To explain this, the final ignition advance value Θ is obtained from the advance value (base advance value) Θ _M and the feedback factor detector (eg, knocking sensor) that approximates the target ignition timing (eg, MBT). The sum of the correction amount (feedback correction amount) Θ _K calculated based on the signal and the learning value Θ _L read out from the RAM is determined by Θ = Θ _M + Θ _L + Θ _K (1).

Here, the region where the learning value Θ _L is stored (learning region)
Is divided into small areas (hereinafter referred to as “areas”) with the main parameter of the operating condition as a coordinate axis, and the learning value is updated independently for each area. For example,
When the learning region is shown in FIG. 6, that number is the value of the learning value θ _L , which is stored for each predetermined division point (for example, 200 rpm, 60 mmHg) of the rotational speed N and the intake pipe pressure P.

A predetermined numerical value (for example, all zero or a calculated value) is set for Θ _L in the initial state, but thereafter, when the learning condition is satisfied, Θ _K is set as a new learning value and the learning value before updating is Will be updated.

In this way, if learning is performed for each area, it is possible to bring the ignition timing closer to the target value of the ignition timing with higher accuracy than in the case of not dividing the area. For example,
In the case of having a wide driving range such as an automobile engine, if a single learning value is used to cover the entire driving range, good accuracy will not be obtained in the driving range adopted as the learning value and in a driving range that can be approximated to this driving range. Even if it is obtained, the accuracy in the driving range that is far from this driving range cannot be guaranteed, or if the learned value averaged over all the driving ranges is adopted, the accuracy is unsatisfactory in any driving range. Because you will not be able to get it.

(Problems to be solved by the invention) By the way, when learning is performed independently for each area,
Knocking may occur, or the engine output may not be sufficiently drawn during a transition. This is due to the fact that the learning frequency is not the same for all areas. It is true that good control accuracy is guaranteed as long as learning is performed, but it seems that learning is rarely performed. In the area, the current state is not reflected on the learning value moment by moment, so if there is a difference from the state learned last time, the previous learning value is useless, and therefore despite having a learning function. However, this causes a situation in which accuracy is deteriorated.

For example, there is an octane number in the value that correlates to knocking, and it is assumed that fuel with a high octane number (high-octane gasoline) to ordinary fuel with a low octane number (regular gasoline).
When the value changes to, knocking is likely to occur as much as the octane number decreases, so the required ignition timing is shifted to the retard side. In this case, in an area where learning frequency is low and where knocking is likely to occur, it is necessary to update the learning value so that it becomes the value on the retard side according to this change in octane number, but immediately the learning value Is not updated, knocking occurs when driving conditions enter the area.

Conversely, if the engine output is changed from regular to high-octane gasoline to increase the engine output, the required ignition timing will shift to the advance side, but the learning value will not be updated accordingly, and the engine output will not be the best. At the time of desired acceleration, feedback control cannot be expected because it is in transition. For this reason, a deviation from the required ignition timing remains, resulting in a situation in which expensive high-octane gasoline is used but the power performance during acceleration is not improved.

The present invention has been made in view of such conventional problems, an area predicted to have a high learning frequency is a specific area, and an area predicted to have a low learning frequency is predetermined as an estimated area. If the learning value change amount in the specific area is large, if it is large, it is determined that the value related to the feedback factor has changed, and although the operating condition is not in the estimated area, the learning value change amount in the specific area It is an object of the present invention to provide an apparatus that updates the learning value of the estimated area by estimating from the above.

(Means for Solving Problems) The present invention is, as shown in FIG. 1, a means 1 for setting a basic ignition timing (ADV0) according to an engine operating condition signal, and a feedback relating to the ignition timing of the engine. A means 2 for detecting a factor (for example, knocking occurrence), a means 3 for calculating a feedback amount (FB) for advancing / retarding the ignition timing ADV0 according to the detected value, and a plurality of learning regions with operating conditions as parameters. Means 5 for storing the learning value (LADV) for each area and storing the learned value (LA
DV), a means 6 for calculating the output ignition timing (ADV) by correcting the ignition timing ADV0 based on the read learning value LADV and the feedback value FB. Means 9 for determining whether or not there is an area, and means 10 for determining whether or not there is a learning condition,
In the ignition timing control device of the internal combustion engine, which comprises means 11 for updating the feedback amount FB as a new learning value LADV for each determined area when the learning condition is satisfied, the learning frequency is selected from the plurality of areas. Area 12 that is predicted to be high, a means 12 that presets an area that is predicted to have a low learning frequency as an estimated area, and if the learning conditions of the specific area are satisfied, the specific area before updating A new learning value (LADV) for the estimated area is determined according to the learning value (LADV _-1 ) and the feedback amount (FB) adopted as a new learning value for the specific area. While updating the learning value of the estimated area,
When the learning condition of the estimation area is satisfied, update of the learning value according to the learning value of the estimation area before updating and the feedback correction amount adopted as a new learning value of the estimation area is updated to the specific area. The means 13 which is not provided to the user are provided. In addition, "-1" attached to LADV means that it is a learning value before being updated.

(Operation) If the learning value of the specific area is updated based on the change in octane number, it may be necessary to update the learning value of the estimated area in some way. For example, in an estimated area where knocking is likely to occur (low rotation and high load range), if the octane number suddenly drops due to a change in fuel properties, knocking tends to occur depending on the drop. Since the learning value when the octane number is high is read, the learning value when the octane number is lowered is too advanced and knocking occurs.

In this case, in the specific area, the feedback amount FB is updated as the new learning value LADV and the learning value LADV _-1 before updating is updated. Therefore, the change in the learning value DADV (= LADV-LADV _-1 ) changes the octane number. It is determined that there is a change in the factors involved in equal knocking. Then, by estimating the value corresponding to the learning change amount as the new learning value in the estimation area, the learning value is updated although the learning condition in the estimation area is not satisfied.

When there is a large change in the learning value in the area where the learning frequency is predicted to be high, the learning value is also updated in the area (estimated area) where the learning frequency is predicted to be low. It will always be done. When learning is performed independently for each area, as a result, an area that is originally predicted to have a low learning frequency, such as an estimated area, occurs, but this estimated learning predicts that the learning frequency is low. The actual learning frequency in the area is increased. Therefore, even in an area where knocking is likely to occur but learning frequency is expected to be low, the occurrence of knocking that would occur if the decrease in octane number cannot be followed is avoided.

Also, when high-octane gasoline is used, the learning value is updated to the advance side in response to an increase in the octane number due to the substantial improvement in learning frequency, so even during acceleration when feedback control cannot be expected, high-octane gasoline is used. The high power performance peculiar to gasoline is fully brought out.

By the way, in order to determine the estimated area by actually driving, it is necessary to count the number of times feedbacked or updated for each area for a certain period of time, and it is necessary to judge whether it is an estimated area from the result. In the initial state before the judgment, the estimation learning cannot be performed.

On the other hand, in the present invention, if the specific area and the estimated area are set in advance before the start of driving, the estimated learning is performed even in the initial state before the elapse of a certain time.

Also, the retard amount with respect to the basic ignition timing is stored as a learning value for each area, and the learning value is increased by a predetermined value when knocking occurs, and the learning value is decreased by a predetermined value when knocking does not occur and the optimum value is obtained. The correction value is compared with the learning value of the area adjacent to the area to which the correction value belongs, and when the learning value of the adjacent area deviates from the correction value by the set value or more, the deviation amount is set. There is also one that modifies the learning value of the adjacent area at the same time so that the relationship between the area to which the modification value belongs and the area adjacent to it is switched (that is, the specific area and the estimated area are changed). Since it is not distinguished from the beginning), the correction result in the estimated area where the learning value is rarely corrected is reflected in the specific area, and There is a rather can degrade the accuracy of 習値 but does not occur such inconvenience in the present invention.

(Embodiment) FIG. 2 is a block diagram of a control system according to an embodiment of the present invention. In the figure, 21 to 25 are sensors provided in each part of the engine. Reference numeral 21 denotes a crank angle sensor which generates a signal for each unit angle (for example, 1 °) and a signal for each predetermined angle (for example, 180 ° for a four-cylinder engine). The engine speed N is obtained from the unit angle signal among these signals, and the crank angle position is determined from both signals.

Reference numeral 22 denotes a sensor (for example, a hot wire type or a flop type) that detects the air flow rate Qa and is positioned as a main parameter of the operating condition together with the rotation speed N. In addition to Qa, the intake pipe pressure used in the conventional example may be used. On the other hand, the sub-parameters are the cooling water temperature Tw detected by the water temperature sensor 23, an ON / OFF signal from the idle switch 24 that outputs an ON signal when the throttle valve is fully closed, and the like.

25 is a sensor (knock sensor) that detects a value that correlates to knocking as a feedback factor.
It is composed of a piezoelectric element that outputs according to the vibration of the engine body. The signal from the sensor 25 is converted into a predetermined voltage value through a filter (bandpass filter) that passes only an output of a predetermined frequency corresponding to knocking, a rectifier, an integrator, and an amplifier, and this is input to a comparator. . Here, when knocking occurs, the voltage value (knocking level) KN input to the comparator becomes high. Therefore, KN is compared with the reference voltage (slice level) SL obtained from background noise. It can be determined that knocking has occurred when the SL is exceeded. However,
The method of detecting knocking is not limited to this.
The SL is obtained by looking up the table and the discrimination function performed by the comparator
Another method given to the program in the CPU will be described later.

Reference numeral 31 is an ignition device. For example, the full-transistor type is composed of a power transistor for interrupting the primary current of the ignition coil, an ignition coil for generating a high voltage by this interruption, an ignition plug for performing spark ignition by receiving the high voltage, and the like.

Reference numeral 30 denotes a control unit to which signals from the sensors 21 to 25 are input. The unit 30 corrects and calculates an optimum ignition timing (ignition advance value) based on a knocking signal, and converts this to an ignition signal. Output to the ignition device 31.

3 and 4 show an ignition advance value calculation routine executed in the CPU when the control unit 30 is mainly composed of a microcomputer. The same drawing is executed every time when knocking is detected (around 50 ° after the compression top dead center). The attached numbers are step numbers.

First, the final ignition advance value (output advance value) ADV is calculated as the sum of the basic advance value ADV0 and the feedback amount FB (step 58).

ADV = ADV0 + FB (2) where ADV0 in the first term is a basic value calculated according to the main parameters (Qa and N) and the sub-parameters (step
41). For example, ADV0 is obtained by referring to a three-dimensional map having Qa and N as parameters when the idle switch 24 is OFF, while ADV0 has only N as a parameter when the switch 24 is ON 2 It is obtained by referring to the dimension table. Furthermore, it is corrected according to the difference in Tw. A value that gives the crank angle before compression top dead center is used for ADV0.

Next, the FB of the second term also includes a learning value calculated here according to the feedback amount. Here, the feedback control based on the knocking signal is a control for tracking the ignition timing (knock limit advance value) at which knocking starts to occur when the ignition timing is advanced under a certain condition. In the control, knocking occurs. If not, the predetermined amount (δ
₂ ) The angle is gradually advanced, and when knocking occurs, the angle is retarded by a predetermined amount (δ ₁ ). Ie knocking level
Both KN and slice level SL are compared, and if KN ≧ SL, the value obtained by subtracting δ ₁ from the previous feedback amount (FB ₋₁ ) as the current feedback amount FB is determined that knocking has occurred. Conversely, if KN <SL, knocking has not occurred, so the value obtained by adding δ ₂ to the previous feedback amount (FB ₋₁ ) is set as the current feedback amount FB (steps 44 to 47, 44-46,48). Therefore, FB is a value including positive and negative. SL is obtained by referring to a two-dimensional table determined according to the rotation speed N (step 45). The feedback amount FB calculated this time is stored as FB _-1 for the next calculation (step 49).

On the other hand, 3 stored in RAM as the learning value (LADV)
The content of the dimension map (learning map) is shown in FIG.
Area) ("00", "01", "10", "11"), and the values stored in different areas are different, but N and Qa are different in the same area. The same value that you are trying to store is stored. This is because the required ignition timing does not deviate so much even if the operating conditions are slightly different, and it is practically acceptable to consider the values to be substantially the same in the same area. Therefore, N and Qa at that time
From which is determined which area the learning value LADV of the determined area is obtained by referring to the map (steps 42 and 43). The RAM is a non-volatile memory, or a backup battery prevents the data from being erased even when the key switch is turned off, or when a volatile memory is used, a predetermined value (for example, 0 or a calculated value) is set when the engine is started. To be configured.

Next, areas other than the "00" area ("01", "10", "1"
1 ") immediately after entering (step
52-54). Immediately after entering the area means that the vehicle was in a transient state until then, so that the feedback amount calculated during the transition is not adopted. That is, although the feedback amount gives a precise value in a steady state as a characteristic of control, it becomes impossible to give an accurate value in a transient state. Therefore, when it is determined that it is immediately after, the learning value LADV read in step 43 is set again as FB instead of the feedback amount calculated based on the knocking signal (steps 52 to 5).
7). “LADV01” and “LADV1” shown in steps 55 to 57
"0" and "LADV11" represent learning values read from the "01", "10", and "11" areas, respectively. However, after the second time, the value of the feedback amount FB calculated based on the knocking signal is used as it is, assuming that it is in the steady state.

Then, the ignition advance value ADV (= ADV0 + F corrected by FB
B) is transferred to the ignition register of the interface (I / O) (step 59). In the register, the crank angle is AD
When it matches V, it outputs a signal that cuts off the primary current of the ignition coil.

Next, learning control, which basically consists of determining learning conditions and updating learning values. First, the learning condition is in principle when the feedback amount gives an accurate value, that is, in the steady state, and a condition considering other factors is added to this. For example, in a state near a steady condition with small load fluctuation and immediately before crossing the area (for example, from "01" to "10" area), or when a certain time or a certain number of revolutions have passed since the transition from the transient state to the steady state. Can be considered as a learning condition. Therefore, it is determined whether such a learning condition is satisfied by the driving condition at that time (step 50).

When the learning condition is satisfied, the learning value is updated (steps 50, 51). Here, the update of the learning value is the same as the conventional one. That is, the value of FB calculated based on the knocking signal in each area is used as it is as a new learning value LADV.
As long as the learning condition is satisfied, each area is rewritten independently.

However, in this case, it is predicted in advance that an area that is not easily updated and an area that is frequently updated will occur due to the difference in the learning frequency, and in the former area, inconvenience due to the update delay will occur, so the learning frequency is The learning result in the area predicted in advance as high is reflected in the area predicted in advance as the learning frequency is low. Specifically, among the areas shown in FIG. 5, the "01" area covering the low rotation and high load area is the area predicted in advance as having a low learning frequency.

It should be noted here that it is not difficult to drive in the "01" area, but it is difficult to satisfy the learning condition (especially steady state). For example, the operating conditions frequently enter this area due to full-open acceleration, but this is a transient state, and therefore the learning condition is not satisfied no matter how many times full-open acceleration is repeated.

Since this area is an operating area where knocking is particularly likely to occur, if there is a change in a factor related to combustion (eg octane number), some learning is required to reflect this. Therefore, the area in which any inconvenience does not occur even if the learning frequency is predicted to be low is excluded.

On the other hand, the areas predicted to have a high learning frequency in advance are the “10” area that covers the medium rotation high load area and the “11” area that is the high upturn high load area. This is because when the vehicle runs steadily, it falls into these areas. Therefore, since the learning value is frequently updated in these areas, it can be said that the current situation such as the octane number is best captured.

Now, assuming that the learning is performed in both the “01” area predicted to have a low learning frequency and the “10” or “11” area predicted to have a high learning frequency, both learning Considering how much the values deviate from each other, it seems that there is not much difference in the values between adjacent "01" and "10" areas. Value L
If ADV10 is known, it is estimated from this value LADV10 and the other "0
It is possible to determine the learning value LADV01 for the 1 "area.
This is what is called estimation learning. The rationale is that it is empirically known that the required ignition timing does not change greatly even if the operating conditions are slightly different.

Therefore, in this example, the "01" area is the estimated area,
When the area of “10” is set in advance as the specific area and the self learning is performed in the specific area, the estimated learning is performed in the estimated area according to the variation amount of the learning value. This is executed by the routine shown in FIG. Needless to say, the case where the “11” area is set as the specific area is also acceptable.

First of all, for my own learning, it is judged whether it is a "10" area, and when this is judged, the feedback amount
FB is used as it is as a new learning value LADV10 and the previous learning value stored in the "10" area is rewritten (step 51).
A, 51F).

Next, regarding the estimated learning, the variation amount DLADV10 DLADV10 = LADV10−FB (3) of the learning value in the “10” area is calculated, and this variation amount DLADV10 is compared with the predetermined amount (± δ ₃ ) ( Steps 51B, 51C). This is because it is desired to determine whether or not the octane number has changed with the change amount. For example, if the fuel properties do not change, LADV
The values of 10 and FB are equal, but if the engine is changed from high-octane to regular gasoline, knocking is likely to occur due to the decrease in octane number, so the required ignition timing deviates to the retard side by a predetermined amount (δ ₃ or more). , FB value is smaller than LADV10. Therefore, when DLADV ≧ δ ₃ , it can be determined that the gasoline has been changed to regular gasoline.

Then, when this is determined, a value obtained by subtracting a predetermined amount δ ₄ from the learning value LADV01 stored in the estimation area is set as a new learning value so that the learning value in the estimation area also corresponds to the regular gasoline. The learning value in the "01" area is rewritten (steps 51C and 52D). "10" corresponding to the shift of the required ignition timing to the retard side by changing the fuel property
This is because the learning value of the area has changed, and therefore it is necessary to change the "01" area where the driving condition is not so different from that of the area, in response to this property change.
The value of δ ₄ is DLADV1 if it can be considered that the driving conditions in both areas are practically the same.
The value 0 may be adopted as it is, or the value may be set according to the deviation of the operating conditions. This example adopts the latter. This is because the value of δ ₄ is determined in relation to the operating range covered by one area, and the wider the range is as shown in FIG. 5, the greater the deviation of the operating conditions. However, it goes without saying that the value of the operating point, which is finally determined by matching and in which knocking is most likely to occur, has priority.

Similarly, when DLADV ≦ −δ ₃ , it can be known that the gasoline has been changed to high-octane gasoline. In this case, therefore, the learning value in the estimated area is estimated to be advanced according to the high-octane gasoline. Learning value stored in the area
The learning value in the "01" area is rewritten using the value obtained by adding a predetermined amount δ _{4 to} LADV01 as a new learning value (steps 51C, 51).
E).

Therefore, if configured in this way, if there is a change in the factor (octane number) involved in knocking,
As far as the octane number is concerned, the learning value is updated in the estimated area at the same frequency as the specific area (steps 51C, 51D, 51F or 51C, 51E, 51F). Therefore, even if independent learning is rarely performed in the estimated area, this substantial improvement in learning frequency not only predicts that learning frequency is low, but also increases the octane value to the learning value in the area where knocking is likely to occur. Can be changed in response to changes in. That is, the present invention focuses on the fact that there is a known relationship between the specific area and the estimated area that the learning value can be estimated in accordance with the difference in the driving conditions between the two, and By using it, the deficiency of the independent learning function for each area is compensated, and the learning function is exerted more than ever.

As a result, the control accuracy of the knock limit control in the estimated area is improved without being affected by factors such as octane number that are involved in knocking, so that knocking can be prevented from occurring in the estimated area. Further, at the time of acceleration, the engine output in the estimated area can be sufficiently drawn out in accordance with the difference in fuel property, which improves the transient response.

Further, the specific area and the estimated area are not set each time based on the result of actual driving, but are set in advance, so that the initial performance is improved.

On the other hand, the number of times of feedback and the number of updates for each area are set for a certain time (or a certain number of rotations)
There is a method in which the area is counted and the estimation learning is performed on other areas based on the area having a larger count value.

With this device, the estimation learning cannot be started until a certain period of time elapses, and the estimation learning cannot be performed in the initial state because the counting must be performed again every time the initialization is performed. It should be noted that by holding the count value by battery backup, it is possible to perform estimation learning already without counting up at the beginning of the next operation, but even in this case, if the battery is removed, The count value will be lost, so you must count up again.

However, if the specific area and the estimated area are set in advance before starting operation as in this example, the estimated learning is performed even in the initial state before a certain period of time elapses or in the initial state immediately after returning to the battery backup state. Is possible.

In addition, even if it is predicted that the learning frequency is low,
The feedback amount FB is calculated in the “01” area,
Further, when the learning condition is satisfied, it goes without saying that the learning value of the "01" area is updated with this FB as a new learning value (steps 51A and 51G).

Also, the retard amount with respect to the basic ignition timing is stored as a learning value for each area, and the learning value is increased by a predetermined value when knocking occurs, and the learning value is decreased by a predetermined value when knocking does not occur and the optimum value is obtained. The correction value is compared with the learning value of the area adjacent to the area to which the correction value belongs, and when the learning value of the adjacent area deviates from the correction value by the set value or more, the deviation amount is set. There is also a known example in which the learning value of the adjacent area is simultaneously corrected and changed so that
In this publicly known example, the relationship between the area to which the correction value belongs and the area adjacent to it is exchanged (that is, the specific area and the estimated area are not distinguished from the beginning).
Therefore, the correction result in the estimation area in which the learning value is rarely corrected is reflected in the specific area, which may rather deteriorate the accuracy of the learning value in the specific area.

 The reason will be described in detail below.

In the present invention, in the example shown in FIG. 5, the low rotation and high load area (“01” area), which is often entered in a transient state such as acceleration / deceleration, but is rarely operated in a steady state, is an estimated area, It is possible to preset the medium rotation high load range (“10” area) and high rotation high load range (“11” area) that correspond to the operating conditions as specific areas, and conversely, the low rotation high load area is specified area. It is impossible to preset the medium rotation and high load area as the estimation area.

When the learning value of the medium rotation high load area or the high rotation high load area (specific area) is updated, the estimated learning based on the learning value is performed for the low rotation high load area. The accuracy of the learning value in the medium rotation high load area and high rotation high load area, which are often operated in a steady state, is feedback control → learning value correction → feedback control in which the corrected learning value is reflected →
The learning value is corrected → Feedback control that reflects the corrected learning value is repeated and the routine is gradually improved, and the learning value in the middle rotation high load area and high rotation high load area is improved. By repeating the estimation learning based on the low rotation high load region (estimation area), the accuracy of the learning value in the low rotation running load region is gradually improved.

However, it is said that the accuracy of the learning value in the low rotation and high load range is improved by performing the estimation learning based on the learning value in the medium rotation and high load range and the high rotation and high load range for the low rotation and high load range. Does not become the same as the accuracy of the learning value in the middle rotation high load area and high rotation high load area, and learning in the low rotation high load area is more accurate than the accuracy of the learning value in the middle rotation high load area and high rotation high load area. It goes without saying that the precision of the values drops.

On the other hand, according to the above-mentioned publicly known example, when the learning value in the medium rotation high load range (“10” area) is corrected in the four operation state sections shown in FIG. 5, it is the adjacent operation state section. Estimated learning is performed for three operating state categories: low rotation and high load range (“01” area), high rotation and high load range (“11” area), and low and medium load range (“00” area). When the learning value of the rotation high load area (“01” area) is corrected, the adjacent operation state classification of the medium rotation high load area (“10” area) and low medium load area (“00” area) Estimated learning is performed for two areas.

Corresponding this to the invention of the present application, in the above-mentioned known example, in particular, the estimation learning using the learning value of the low rotation high load region (estimated area) is performed in the medium rotation high load region (specific area).
Is also done for.

Here, after the learning value in the middle rotation high load area is frequently updated and the accuracy of the learning value in the middle rotation high load area is sufficiently improved, the learning value in the low rotation high load area happens to be updated. Considering the case described above, in the above-described known example, in this case as well, the estimation learning using the learning value of the low rotation high load region updated only once is performed for the medium rotation high load region.

However, in this case, since the learning value is updated only once in the low rotation and high load range, the learning value in the low rotation and high load range is not very accurate. The learning value in the low rotation and high load range, which is not so accurate, is reflected in the medium rotation and high load range in which the learning value is frequently updated and the accuracy is sufficiently improved in the above-mentioned known example. The accuracy of the learning value in the high rotation speed region is deteriorated.

On the other hand, in the present invention, after the learning value in the middle rotation high load range is frequently updated and the accuracy of the learning value in the middle rotation high load range is sufficiently improved, it happens that the low rotation high load range ( When the learning value of the estimated area) is updated, the estimated learning using the learning value of the low rotation high load area is not performed for the medium rotation high load area (specific area). That is, the present invention increases the accuracy of the learning value by repeating the original learning for the specific area, while increasing the accuracy of the learning value by repeating the estimation learning exclusively for the estimation area, and in that case, By comparison, the learning value in the estimation area is inevitably less accurate, so the learning result in the estimation area is prevented from reaching the specific area, and therefore the learning result in the estimation area is affected by the learning result in the specific area. The accuracy of the learning value is not deteriorated.

Lastly, in this example, the knock limit control is described as the feedback control of the ignition timing, but the present invention is not limited to this, and control to the MBT and the like are conceivable, and the same applies to these.

(Effects of the Invention) As described above, according to the present invention, in the ignition timing control device in which learning is performed independently for each area based on the signal for detecting the feedback factor, the learning frequency is high among a plurality of areas. The predicted area is a specific area,
An area that is predicted to have a low learning frequency is set in advance as an estimated area, and when a learning condition for a specific area is satisfied, a new learning value for the estimated area is determined according to the learning value change amount of the area. Then, while updating the learning value of the estimated area with the determined learning value, when the learning condition of the estimated area is satisfied, the learning value of the estimated area before updating and the new learning of the estimated area Since the learning value is not updated according to the feedback correction amount used as a value for a specific area, the learning frequency can be increased for areas predicted to have a low learning frequency. This prevents knocking even in areas where knocking is likely to occur but learning frequency is expected to be low, and at the time of acceleration, sufficient engine output according to the fuel property is provided. In addition to being able to demonstrate the performance, the initial performance is improved compared to the case where the estimated area is determined by actually driving, and when learning is performed in an area regardless of the learning frequency, the learning result is used as the learning result. Compared with the case of reflecting in the area adjacent to the separated area, the accuracy of the learning value of the specific area is not deteriorated due to the influence of the learning result in the estimated area.

[Brief description of drawings]

FIG. 1 is a conceptual block diagram of the present invention, FIG. 2 is a block diagram of a control system of one embodiment of the present invention, and FIGS. 3 and 4 show the contents of calculation executed in the CPU of this embodiment. The flow chart, FIG. 5, is a characteristic diagram showing the learning region of this embodiment. FIG. 6 is a characteristic diagram showing a learning area of a conventional example. 1 ... Basic ignition timing setting means, 2 ... Feedback factor detecting means, 3 ... Feedback amount calculating means, 5 ...
Learning value storage means, 6 ... Learning value reading means, 7 ... Output ignition timing calculation means, 9 ... Area discrimination means, 10 ... Area setting means, 13 ... Estimated learning means, 21 ... Crank angle sensor, 22 …… Air flow rate sensor, 23 …… Water temperature sensor, 24
Idle switch, 25 knock sensor, 30 control unit, 31 ignition device.

Claims (1)

(57) [Claims] 1. A means for setting a basic ignition timing according to an engine operating condition signal, a means for detecting a feedback factor related to the engine ignition timing, and an advance / retardation of the ignition timing according to the detected value. A means for calculating the feedback amount, a means for dividing the learning region into a plurality of areas using the operating condition as a parameter, a means for storing a learned value for each area, a means for reading the stored learned value, and a read learning Means for calculating the output ignition timing by correcting the ignition timing based on the value and the feedback value; means for determining which area the vehicle is in based on the operating condition signal; and determination of whether or not the learning condition is satisfied. And a means for updating the feedback amount as a new learning value for each of the determined areas when a learning condition is satisfied. In the case where the learning condition of the specific area is satisfied, means for presetting an area predicted to have a high learning frequency from the plurality of areas as a specific area and an area predicted to have a low learning frequency as an estimated area In addition, a new learning value of the estimated area is determined according to the learning value of the specific area before updating and the feedback amount adopted as a new learning value of the specific area, and estimation is performed using the determined learning value. While updating the learning value of the area, when the learning condition of the estimated area is satisfied, the learning value of the estimated area before updating and the feedback correction amount adopted as a new learning value of the estimated area are set. An ignition timing control device for an internal combustion engine, comprising: means for not updating a learning value according to a specific area.