JPH0762461B2 - Knocking detection device for internal combustion engine - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a device for detecting a knocking state (hereinafter referred to as knock) occurring in an engine, and this device mainly detects the knocking state and ignition timing. Alternatively, it is used for a knock control device (hereinafter referred to as a knock control system) that controls knock control factors such as an air-fuel ratio and intake pressure.

[Conventional technology]

In general, a knock control system detects the vibration of the engine body caused by knock by a knock sensor, and advances or retards the ignition timing according to the detection result, so that the ignition timing is always near the knock limit. It controls and improves the engine output and fuel consumption.

In such a knock control system, as shown in, for example, Japanese Patent Laid-Open No. 58-7538, it is very important how to accurately detect the knock from the knock sensor.

Currently, mass-produced knock control systems mainly use knock sensors of the type that detect vibration of the engine body in terms of cost and reliability. Since a knock sensor of this type is affected by mechanical vibration noise of the engine, the SN ratio deteriorates in a high speed rotation range where the vibration noise increases, and thus it becomes very difficult to detect a high speed knock.

Further, the sensor output characteristics change due to manufacturing tolerance of the knock sensor and its change over time, and the vibration transmission characteristics of the engine change due to individual difference of the engine and its change over time, resulting in large variations in sensor output.

Therefore, it is difficult to detect a high-speed knock in consideration of the variation of the engine or the sensor and their changes over time, and the current situation is that knock control is substantially cut in the high-speed range.

In the low-medium speed range, even if a very large knock that damages the engine can be detected, the knocking noise during control varies greatly due to the above-mentioned variations and changes over time, which causes driver discomfort. There was also a case where a loud knocking sound that gave a sound was generated.

In order to solve the above-mentioned problems, it is a well-known fact that sensors aiming for a high SN ratio, such as in-cylinder pressure sensors and combustion light sensors, are being studied at the laboratory level. It is true that such sensors are now very costly and still have many reliability problems. However, even if the cost and reliability issues are resolved,
Sensor fabrication tolerances and aging issues remain.

In addition, since such a sensor does not receive mechanical vibration noise of the engine in principle, it is expected that the SN ratio in the high speed range will be improved, but the vibration noise generated by the engine is not limited to mechanical noise. There are also many pressure vibration noises and combustion light noises caused by combustion.

Since these combustion noises naturally change due to individual differences in the engine, basically variations in the engine,
Inevitably affected by changes over time. That is, the above problems cannot be fundamentally addressed even if such a sensor is used.

[Problems to be solved by the invention]

The present invention has been made in view of the above problems, and an object of the present invention is to provide a device for accurately detecting knock without being influenced by variations in a knock sensor or an engine, changes over time, and the like.

[Means for solving problems]

In order to achieve the above object, a first invention of the present invention relates to a knock sensor for detecting knock of an engine, and a signal from the knock sensor, for example, a rectified integral value corresponding to a maximum peak value of the signal. Of the knock intensity value V, and a frequency distribution of the knock intensity value V obtained when the knock intensity value V is subjected to multiple combustion cycle sampling, a predetermined cumulative percentage of the frequency of the knock intensity value V is obtained. Discriminating means for discriminating whether or not there is a substantially corresponding normal distribution within the range of points _{VL to VH, and} the range of the predetermined cumulative% points _{VL to VH} for discriminating whether or not the substantially lognormal distribution And distribution shape changing means for changing according to the engine conditions,
And a knock state detecting means for detecting an average knock state over a plurality of cycles according to the discrimination result of the discriminating means.

[Action]

Hereinafter, the outline of the present invention and the technical basis of the present invention will be described in detail.

In order to solve the above problems, the present inventors previously proposed several methods and devices in Japanese Patent Application No. 59-99898. In this Japanese Patent Application No. 59-99898, when the maximum peak value V _MAX in a predetermined section of the knock sensor is sampled in large numbers without knocking, one substantially lognormal distribution is formed and sampling is performed with knocking. In this case, it is noted that in this case, a combined distribution of two approximately lognormal distributions having different variances is obtained.

Here, in the above invention, the maximum crest value V _MAX of the sensor signal is required, and in order to detect the maximum crest value V _MAX , a peak hold circuit or a microcomputer for detecting a peak instead thereof (hereinafter referred to as a microcomputer Note) software, etc. are required. The maximum peak value of the knock control system currently in mass production or research and development
Since many systems do not directly detect V _MAX , in such a system, it is necessary to add a circuit for V _MAX detection or software for a microcomputer.

In view of this point, the present inventors do not necessarily have to detect the maximum peak value V _MAX by performing detailed experiments and devices after that, and the frequency component (generally 6
Although it is about 9 KHz, depending on the structure of the engine, it may exist on the lower frequency side or on the higher frequency side of 10 KHz or more. ) Including the amount indicating the strength of the knock vibration part corresponding to the maximum peak value, for example, a value obtained by integrating the knock vibration output, or the number of pulses in a pulse train obtained when the knock vibration output is compared with a predetermined level,
It has been discovered that the integrated value thereof or the effective value of the knock vibration output (hereinafter referred to as knock intensity value V) may be detected.

Furthermore, by comparing the distributive properties of these knock intensity values, we found more general properties common to these knock intensity values. The present invention utilizes the property common to the knock intensity values to provide a device that always detects an accurate knock state regardless of the type of knock sensor and the method of extracting the knock intensity value. Before describing the intensity value V and its properties,
First, I will briefly explain the nature of the maximum peak value V _{MAX that} was the basis of the discovery.

In FIG. 1, (1) is a knock sensor signal when only the frequency component of 6 to 9 KHz is extracted from the vibration of the engine body. When the maximum peak value V _MAX within this engine combustion section (for example, 10 ° to 90 ° ATDC) of this signal is sampled in multiple combustion cycles, a frequency distribution as shown in FIG. 2 is obtained. When this V _MAX is logarithmically transformed and the frequency distribution is drawn again, when sampling is performed in the absence of knock, it becomes a distribution close to a normal distribution as shown in Fig. (3), and knock has occurred to some extent. When sampling is performed in this state, the distribution deviates from the normal distribution on the high output side as shown in FIG.

In order to show this more clearly, a lognormal probability paper will be used for explanation. Generally, if the frequency distribution is a normal distribution, the cumulative frequency distribution will be an S-shaped curve having an inflection point at the cumulative 50% point as shown in FIG. The paper in which the S-shaped curve peculiar to the normal distribution is precisely graduated so that the S-shaped curves line up on a straight line when plotted is the normal probability paper as shown in FIG. Since such a normal probability paper (which is called a logarithmic normal probability paper when the characteristic value is a logarithmic value) itself is already known in the field of statistical analysis, a detailed description of the paper will be omitted. Only the properties of the paper which are directly related to the explanation of are described here.

The vertical axis of the normal probability paper, that is, the scale of cumulative frequency, becomes farther away from the 50% point as it goes to the scale at both ends with respect to the 50% point (for example, the interval between the 70% and 60% scales is 60 % And 50% longer than the graduation interval), but 50%
The scale is vertically symmetrical with respect to the scale. (For example, the 10% scale and the 90% scale are equidistant from the 50% scale, and in general, the α% scale and the (100−
The α)% scale is equidistant from the 50% scale. ) Another important property of normal probability paper is that the slope of the straight line when plotted on this paper matches the reciprocal of the standard deviation σ (ie 1 / σ). Therefore, it means that the distribution is greater as the straight line lays sideways as it sleeps.

Now, when the distribution of the maximum peak value V _{MAX in} the state without knock is plotted on this lognormal probability paper, it becomes one straight line as shown by “without knock” in FIG. 2 (2).
This means that the maximum peak value V _MAX without knock forms one lognormal distribution.

On the other hand, the distribution of knocking occurring with a certain frequency (that is, a state in which knocking cycles and non-knock-safe are mixed at a certain ratio) has a distribution of “knocking” in FIG. As shown by, it becomes a polygonal line whose slope changes halfway.

That is, the lognormal distribution having a small variance formed on the relatively low output side and the lognormal distribution having a large variance formed on the relatively high output side are overlapped.

Now, the examples of the knock intensity values having the property that the knock intensity values V other than the maximum wave height value V _MAX can be approximated to one log-normal distribution in the state without knock are listed as follows.

The first is a value obtained by rectifying and integrating the vibration output in a predetermined section (for example, 10 ° to 90 ° ATDC). This value is obtained by changing the time constant of the integrator,
Alternatively, it represents the integrated value of the vibration output of the cycle, and in any case, it is a value corresponding to the vibration intensity of the cycle.

The second is the number of pulse trains generated when the vibration output in a predetermined section is compared with a certain threshold value (for example, a level slightly above the vibration center), or the integrated value of the pulse trains.
This value is also a value for the vibration intensity of the cycle. This intensity value is slightly mentally inferior because its range of change is limited compared to the range of the maximum peak value V _MAX , but it has the advantage of being easy to handle.

The third example is the effective value (RMS) of the vibration output in the predetermined section. In this RMS, the vibration level at time t is X (t), and the specific length of the predetermined time is T This value also corresponds to the vibration intensity of the cycle.

From the above three examples, the strength value V that generally corresponds to the vibration strength
For, it is natural and rational to think that the distribution in the state without knock is a nearly logarithmic distribution. For example, if a certain intensity value V can be approximated by a relation of approximately V = a · (V _MAX ) ⁿ (a, n is any real number) with respect to the maximum peak value V _MAX which is a member of the same intensity value, It is clear that log V = log a + n log V _MAX, and log V also has a normal distribution (that is, V is a logarithmic normal distribution) like log V _MAX .

Next, the present inventors investigated what kind of distribution difference the intensity values V have with and without knocking.

FIG. 3 shows an example of distribution when an in-cylinder pressure sensor is attached to one cylinder of the engine and the integral value ∫V given as the first example is extracted from this signal. Of the signal from the in-cylinder pressure sensor, only the frequency component (6 to 9 KHz) peculiar to knock is taken out by a bandpass filter, which is 10 ° after half-wave rectification.
It is a value that is integrated at a time constant of 5 msec from ATDC and taken in at a timing of 90 ° ATDC. As can be seen from FIG. 3, the distribution without knocking is represented by one straight line on the lognormal probability paper. That is, it forms one lognormal distribution. On the other hand, the distribution with knocking is certainly a straight line on the low value side, but at a certain point it is a curved curve. The difference on this distribution is compared with the difference on the maximum peak value V _MAX of the vibration of the engine body.

FIG. 4 is a distribution when the maximum peak value V _MAX of the vibration of the engine body is sampled under the same conditions as in FIG.
That is, the signal of the piezoelectric knock sensor mounted on the engine block is filtered by a bandpass filter having a center frequency of 7.5 MHz and a bandwidth Q = 20 dB, and the maximum peak value V _MAX between 10 ° to 90 ° ATDC of this signal is described above. This is the distribution when sampling is performed only for the cylinder to which the in-cylinder pressure sensor is attached. Since the data sampling is performed by the in-cylinder pressure sensor and the vibration sensor at the same time, the knock state is exactly the same.

Comparing FIG. 3 and FIG. 4 reveals the following.
First, the in-cylinder pressure sensor integrated value ∫V in FIG. 3 is
The bending is clearer than the maximum peak value V _MAX of the vibration sensor in FIG. 4, and the integral value ∫ V of the second in-cylinder pressure sensor is curved on the high output side. The point where the maximum peak value V _MAX of the vibration sensor is bent almost linearly. As will be described later in detail, this phenomenon means that the in-cylinder pressure sensor has a higher ability to distinguish between noise and knock, that is, the SN ratio is higher.

However, the most important point in the comparison between Fig. 3 and Fig. 4 is that even if the cumulative percentage of bending is ∫V, V _MAX
However, it is almost unchanged (about 70% in this example). This is the cumulative% equivalent to the bending point
Indicates that it can be a very effective and universal indicator for identifying knock states. This is because the cumulative% of bending points are almost the same, even though knock sensors of completely different types and different knock intensity values are taken out.

In fact, as will be described later, the cumulative% corresponding to this bending point is the percentage of non-knock cycles (noise cycles) in all cycles, in other words, the remaining% (10% in this example).
0-70 = 30%) indicates the knock cycle rate.

This will be described in detail with reference to FIG. Figure 1 (1)
In the figure, assuming that the SN ratio of knock detection is very high and the knock intensity value V of the knock cycle and the knock intensity value V of the non-knock cycle (noise cycle) can be completely separated, the frequency distribution corresponding to two knock states of large and small ( (Shown in the upper row) and the corresponding cumulative frequency distributions (shown in the lower row) on the normal probability paper. The horizontal axis is a logarithmic scale. In the upper frequency distribution, the solid line indicates that the ratio of noise to knock is 90:10, that is, the distribution of relatively small knock states in which 10% of the total is knock cycles, and the broken line indicates that 40% of the total is knock cycles. This is a relatively large knock state distribution. In actual engine knock, as the knock increases, the frequency of knock increases and the knock intensity value itself also increases. Therefore, the distribution corresponding to the knock cycle in the broken line distribution actually has a larger value. It shifts toward you. However, as the knocking state increases, the frequency of knocking will inevitably increase, so the essence will not be lost even if we discuss only the knocking frequency.

If the cumulative frequency distribution corresponding to each knock state in the upper row (that is, the distribution when the noise distribution and the knock distribution are regarded as one group and accumulated from the smaller value) is plotted on a lognormal probability paper, It looks like the bottom row. As can be seen from this figure, when viewed from the lower values, there is a substantially straight line up to a certain point, and a downward convex curve that sharply bends from that point. The cumulative% of the points where the bending occurs indicates the noise generation rate in any case. That is, the remaining% indicates the occurrence rate of knock.

Next, assume that knock and noise are very poorly distinguished under the same knock condition, that is, a sensor with a low SN ratio is used or a knock intensity value with a poor SN ratio is selected. The solid line in (2) of Fig. 5 represents the distribution sampled by a sensor with a good SN ratio when the knock occurrence frequency is 10% (noise and knock ratio 90:10).
The broken line shows the same knock condition, a sensor with a poor SN ratio, or
This is the distribution when a knock strength value with a poor SN ratio is selected.
Further corresponding cumulative distributions are shown in the lower row. From this figure, it can be seen that the curve is bent from a point in the distribution of the sensor having a good SN ratio, but when the SN ratio is relatively poor, it can be approximated by two straight lines bent from the middle. It can be seen that in both cases, the cumulative percentage of occurrence of bending is almost the same. The cumulative% also indicates the noise generation rate, and thus the remaining% indicates the noise generation rate.

In other words, even if a sensor or knock strength value with a slightly poor SN ratio is selected, or if a sensor or knock strength value with a relatively high SN ratio is selected, the cumulative% at which bending occurs will be accurate. It is possible to indicate the occurrence rate of, and thus the knock state.

For example, when the noise distribution and the knock distribution are overlapped (that is, the S / N ratio is bad) like the frequency distribution of the broken line in Fig. 5 (2), the knock intensity value of each overlapped portion is increased. , It is impossible to individually identify whether this is noise or knock. In other words, it is impossible to discriminate at all by the conventional method in which the threshold value of a certain knock intensity value is set to a threshold value, and when the output is more than this, the knock is generated, and below this is noise. However, if attention is paid to the cumulative percentage of the bending points described here, the knocking state can be accurately detected even in such a case.

As described above, the knock state can be accurately detected by paying attention to the cumulative percentage of bending on the lognormal probability paper. Then, if the properties described above are carefully considered, this can be regarded as a more general property.

In other words, the essence of the fact that the distribution with knocks bends sharply on a log-normal probability paper is that the distribution has a relatively low output value, and most of them have a distribution estimated from only data that can be regarded as a noise group. On the other hand, the data on the high output side deviates from the estimated value in such a way that it does not belong to the population of the distribution. Then, by knowing from which point the distribution deviates, the state of knock can be accurately detected. That is, the distribution does not necessarily have to be approximated as a substantially logarithmic distribution.

Of course, a state in which all the sampled data are knocked, that is, an oversized knock state in which all cycles are knocked cannot be identified by the method described above. However, such an oversized knock state can be sufficiently discriminated by a conventional method (for example, a method of determining a knock when the knock intensity value exceeds a certain threshold value). It doesn't matter at all. Further, in such an oversized knock state, the dispersion of data becomes extremely large. Therefore, if a method of monitoring the dispersion is added, it is not necessary to use the conventional method in parallel.

Now, it is very easy to realize the method described above as a laboratory device (for example, a knock monitor measuring instrument). It is not necessary to explain that the function of sampling the data, the function of logarithmically converting the data, and the function of calculating the cumulative percentage of bending by revising the cumulative frequency distribution can be realized by using the microcomputer.

However, in order to apply the present invention to an actual knock control system, it is necessary to further devise.
That is, in the actual engine, it is necessary to detect the knocked state earlier, and moreover, it is necessary to put the control algorithm in the limited ROM and RAM of the vehicle-mounted microcomputer. For that purpose, it is desired to more easily obtain the cumulative percentage at which the bending occurs without intentionally sampling a large number of data and without performing logarithmic conversion.
The following is an example of such a scheme. Of course, various other schemes can be considered.

The method will be described below with reference to FIG.

FIG. 6 (1) shows the cumulative distribution of log V (illustrated on the lognormal probability paper in the upper row) and the frequency distribution at that time (illustrated in the lower row) when knock is smaller than the control target. . On the contrary, FIG. 2B shows the cumulative distribution and the frequency distribution in the knock state larger than the control target. Now, the state in which knock is occurring at a frequency of 10% in all cycles (hence the remaining 90% is a non-knock cycle) is the knock state of the control target. At this time, the distribution of the cumulative frequency in the knock state smaller than the target bends on the output side higher than the cumulative 90% point on the log-normal probability paper as shown in the upper part of Fig. (1). Therefore, it can be regarded as one straight line (that is, one lognormal distribution) on the output side lower than the 90% point.

Therefore, assuming a cumulative 50% point on the lognormal probability paper, that is, the median value M of the frequency distribution, consider two thresholds V _L and V _H that are vertically equidistant to this M on the logarithmic axis. . Equidistant on the logarithmic axis has a geometrical relation on the real axis (log V _H −log M = log M−−log V _L , that is, V _H / M
= M / V _L ). At this time, it is assumed that the lower threshold value V _L is set so that the knock intensity value falls below V _L by the frequency of the same numerical value as the control target knock occurrence frequency of 10%. That is, it is assumed that V _L is set at the cumulative 10% point. Then, when the knocking state is smaller than the target value as shown in FIG. 6 (1), the upper threshold V _H coincides with the cumulative 90% point due to the property of the lognormal probability paper. Therefore, when knocking is less than the target has a frequency of knock intensity values V is below the threshold V _L lower, the frequency of exceed the upper threshold V _H becomes degrees properly.
(10% in this case) However, in the knock condition larger than the target as shown in (2) of the figure, the curve bends at a percentage lower than 90% cumulative, so the upper threshold V _H becomes 90% cumulative. No match, corresponding to lower cumulative%. That is, this upper threshold
The frequency of exceeding V _H increases. (2) If it is shown in the upper diagram of the figure, the frequency increases by the amount of the double arrow. Therefore, when the knock is larger than the target, the frequency at which the knock frequency value V exceeds the upper threshold value V _H becomes greater than the frequency at which the knock frequency value V falls below the lower threshold value V _L.

From the above, the knock intensity value V is monitored for a predetermined cycle, and the relationship between the lower threshold value V _L and the upper threshold value V _H is proportional (M / V _L ≈V _H / M) and select this V
By but comparing the number N _H which exceeded the number N _L and an upper threshold value V _H which drops below the threshold V _L of the lower, the frequency distribution of the knock intensity value V is a predetermined cumulative% By determining whether or not the distribution is a logarithmic normal distribution within the range of the points V _{L to} V _H , (the frequency distribution of the knock intensity value V is a substantially logarithmic distribution within the range of the predetermined cumulative% points V _{L to} V _H. When the distribution is normal, as shown in FIG. 6 (1), the knock state is smaller than the target value, but the frequency distribution of the knock intensity value V has a predetermined cumulative percentage point _{VL to VH.}
It is possible to judge whether the knock state is larger or smaller than the target value because the knock state becomes larger than the target value as shown in FIG. it can.

The remaining challenge is to set V _L and V _H. The setting method will be described below.

Now, consider obtaining the approximate value V _M of the cumulative 50% point M without data sampling. Cumulative 50% point M
Is a point at which the probability that the knock intensity value V exceeds the M is equal to the probability that the knock intensity value V will fall. Therefore, each time the knock intensity value V is input, the value V _M is compared with a certain value V _M ,
Tara is greater towards and V, increases the V _M by [Delta] V _M, Tara small Conversely, if we successively update the V _M to reduce V _M by the same amount [Delta] V _M V _M is cumulative 50% point M Converge to. That is, the cumulative 50% point M can be obtained without sampling the data to obtain the distribution.

Next, consider obtaining the lower threshold V _L. V _L is cumulative
Since the point is 10%, the probability that the knock intensity value V will fall below V _L is 1/10, and conversely, the probability that it will exceed V _L is 9/10. That is, each probability is 1: 9. Therefore, let us consider obtaining the cumulative 50% point M by the same method as that used for obtaining. Every time the knock intensity value V is input, the V is compared with a certain value V _L, and if V is larger, V _L is Δ
It is increased by V _L, if we the V _L from smaller Conversely sequentially updates the V _L to decrease by 9 times the amount 9 · [Delta] V _L of [Delta] V _L, V _L is converged to a cumulative 10% point. Because the actual accumulation
This is because the expected value for increasing / decreasing V _L becomes 0 only at the 10% point (9/10 × ΔV _L −1 / 10 × 9 · ΔV _L = 0).
That is, by changing the ratio of the increase amount and the decrease amount,
In principle, the value can be converged to any cumulative% point.

However, this is not a problem when finding a cumulative percentage point such that the amount of increase and the amount of decrease are approximately equal, such as the cumulative 50% point, but it is far from the cumulative 50% point (for example, 10%).
% Or 90% point), the ratio of the amount of increase and the amount of decrease becomes too large, and hunting near the convergence value is a concern. Therefore, the present invention proposes another method for obtaining V _L.

Now, the knock intensity value V, a certain value _VL, and a predetermined cycle number C are continuously compared. And if knock intensity value V falls below the V _L at least once in the meantime, V _L was reduced by ΔV _L, the V _{L ΔV L} only if the better of all the while V continued around above the V _L It is considered that _VL is sequentially updated every predetermined number of cycles C by increasing the number of times by only. Since the amount of increase / decrease is the same, V _L converges on the point of the following probability relationship. That is, assuming that the probability that the knock intensity value V falls below the convergent value of V _L is P, (1-P) ^c = 1- (1-
P) converge to the point indicated by the relationship of ^c . V _{L at the} cumulative 10% point
In order to converge, the number of cycles c should be about 7 cycles with P = 0.1. Cumulative 50% by this method
The value of the cumulative% point far away from the point can be sequentially obtained with high accuracy.

Thus we have been able to determine the threshold V _L of cumulative 50% point V _M and lower, the upper threshold V _H, V _M / V _L = A as _{V H = V M · A (} = V _M ² / V _L ) can be easily obtained.

With the method described above, it becomes possible to easily obtain the cumulative percentage points of bending without data sampling and without logarithmic conversion.

〔Example〕

Hereinafter, the device of the present invention and its operation will be described in detail according to embodiments.

FIG. 7 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a 4-cylinder 4-cycle engine, 2 is a reference angle sensor for detecting a reference crank angle position (for example, top dead center) of the engine, and a crank angle sensor for generating an output signal at every constant crank angle of the engine. A distributor 3 having a built-in power supply is an igniter and a coil that receives an ignition timing control signal output from the control circuit 10 to cut off the power supply to the ignition coil.

4 is a knock sensor for detecting the vibration of the engine gorok corresponding to the knock phenomenon of the engine by a piezoelectric element or the like, 5 is an air flow meter for detecting the intake air amount of the engine, and outputting a signal corresponding thereto, 6 is A throttle valve connected to the throttle angle sensor, 7 is a fuel injection timing determined by the control circuit 10, and an injector for injecting fuel into the intake manifold based on the fuel injection timing, 8
Is a turbocharger for supercharging, 9 is an O ₂ sensor that generates an output signal according to whether the air-fuel ratio of exhaust gas is richer or leaner than the stoichiometric air-fuel ratio, and 10 is each of the above sensors Is a control circuit for controlling the ignition timing and the air-fuel ratio of the engine according to the state of the input signal from.

Next, the detailed configuration and operation of the control circuit 10 will be described with reference to FIG. In FIG. 8, reference numeral 10-1 is a central processing unit (CPU) for calculating the ignition timing and the fuel injection amount, which uses an 8-bit microprocessor. 10-2 is a read-only storage unit (ROM) for storing control programs and control constants necessary for calculation, and 10-3 is a CPU
Reference numeral 10-1 is a temporary storage unit (RAM) for temporarily storing operation data during operation according to a program. 10-4
And 10-5 are output signals of the reference angle sensor 2-1 and the crank angle sensor 2-2 built in the distributor 2 (a magnetic pickup is used in this embodiment).
Is a waveform shaping circuit for shaping the waveform.

10-6 is an interrupt control unit for causing the CPU to perform an interrupt process by an external signal or an internal signal, and 10-7 is configured so that the count value is increased by one every clock cycle that is the basic cycle of CPU operation. It is a 16-bit timer. The engine speed and the crank angle position are fetched into the CPU by the timer 10-7 and the interrupt controller 10-6 as follows. That is, the CPU reads the count value of the timer each time an interrupt is generated by the output signal of the reference angle sensor 2-1. Since the count value of the timer rises every clock cycle (for example, 1 μs), the difference between the count value at this time and the count value at the time of turning is calculated to calculate the timing of the reference angle sensor signal. The interval, that is, the time required for one revolution of the engine can be measured. In this way, the engine speed is obtained.

In addition, the crank angle position is determined based on the top dead center signal of the reference angle sensor 2-1 because the signal of the crank angle sensor 2-2 is output at every constant crank angle (for example, 30 ° CA). Can be known in 30 ° CA units. This crank angle signal for every 30 ° CA is used as a reference point for generating the ignition timing control signal.

10-8 is analog by switching multiple analog signals at appropriate times.
It is a multiplexer for leading to the digital converter (A / D converter) 10-9, and the switching timing is controlled by the control signal output from the output port 10-12. In this embodiment, an intake air amount signal from the air flow meter 5, a knock intensity value signal from the knock intensity detection circuit 10-10, etc. are input as analog signals. (In addition, the water temperature sensor signal,
Battery voltage etc. are input. ) 10-9 is A / for converting analog signal to digital signal
It is a D converter. Reference numeral 10-10 is a knock intensity value detection circuit which receives a signal from the knock sensor 4 and takes out a knock intensity value V for each cycle to form a knock intensity value detection signal. Ten
Reference numeral -11 is an input port for a digital signal, and a lint line signal from the O ₂ sensor 9, an idle signal and a power signal from the throttle sensor 6 are input to this port.

10-12 are output ports for outputting digital signals. From this output port, an ignition timing control signal for the gnita 3, a fuel injection control signal for the injector 7,
A control signal for the multiplexer 10-8 and a control signal for the knock intensity value detection circuit 10-10 are output. 10-13
Is a CPU bus, and the CPU puts control signals and data signals on this bus signal line to control peripheral circuits and send and receive data.

Next, the knock intensity value detection circuit 10-10 will be described with reference to FIG. In FIG. 9, 10-10-1 is a band pass filter, a high pass filter or the like for selecting and extracting only knock frequency components from the output signal of the knock sensor 4, 10-10
-2 is a rectifier for half-wave or full-wave rectification of the output of this filter, 10-10-3 is an integrator for integrating the output of the rectifier to obtain a knock intensity value for each cycle,
Reference numeral 10-10-4 is a gate circuit for controlling the integration interval of the integrator.

The operation of this knock intensity value detection circuit will be described with reference to FIG. FIG. 10 (1) shows the knock sensor output after passing through the filter, and FIG. 10 (2) shows the output obtained by half-wave rectifying the knock sensor output by the rectifier 10-10-2. (3) The figure shows the output signal of the gate circuit 10-10-4 which operates according to the control signal output from the output port 10-12 in the control circuit 10. Ie about 10
It rises at ゜ ATDC, and knock strength is CPU10-at about 90 ゜ ATDC
1 is a signal that falls immediately after being taken in, and this determines the integration section of the integrator 10-10-3. (4) The figure shows the integrator set and reset by the gate circuit 10-10-3
Is the output signal of.

The time constant of the integrator in this embodiment is adjusted to about 5 msec. The rising of the gate circuit starts integration of the signal after half-wave rectification (Fig. (2)), and the falling resets the integrated value. The gate circuit is CPU10-1
Falls immediately after the integrated value is fetched through the multiplexer 10-8 and the A / D converter 10-9, so that the integrated value for each cycle, that is, the knock intensity value, is successively fetched by the CPU.

Next, the knock detection and the operation of the knock control system will be described in detail with reference to the flowcharts of FIGS. 11 to 15.

FIG. 11 is an interrupt routine for setting the ignition timing and the combustion injection time in the timer of the microcomputer. Interrupt 100
Therefore, in step 200, first, the basic ignition timing θ
_BSE and basic injection time τ _BSE are calculated. The basic ignition timing θ _BSE and the basic injection time τ _BSE are proportional to the engine load, which is the engine speed N and the engine addition Q / N (the intake air amount Q measured by the air flow meter divided by the engine speed N). ) Is stored in the ROM of the microcomputer as a two-dimensional map. Then, in step 200, at the same time, the ignition timing and the injection time are corrected by various sensor signals other than the knock sensor. For example, the ignition timing and the injection time are corrected by the water temperature, or the injection time is corrected by the battery voltage (considering the invalid injection time).

Next, at step 300, it is judged if the current operating condition is the knock control execution condition. For example, it is generally known that when the load of the engine is light, knocking hardly occurs, and if the light load is forcibly controlled to knock, the output, fuel consumption and the like are rather lowered. Therefore, in the present embodiment, the knock control is executed only when the engine load (Q / N can be determined) is a predetermined value or more. (Otherwise, it is conceivable to set a limit based on the engine speed.) If it is determined that the knock control execution condition is satisfied, in step 400, the ignition timing and the injection time are corrected by the knock control. The correction amount of the ignition timing by the knock control (in the present embodiment, the retard amount R from the basic ignition timing θ _BSE ) is calculated by another routine described later, but the final ignition timing θ is determined in this step 400.
Then, θ = θ _BSE −R is obtained.

Further, the correction amount of the fuel injection time is determined based on this retardation amount R. That is, when the retard amount R is large, the ignition timing is retarded, so the exhaust temperature may become too high. Therefore, in this case, it is necessary to prolong the injection time so as to make the air-fuel ratio rich. For example, the correction amount of the injection time may be determined so as to be proportional to the retard amount R.
The final ignition timing θ and the final injection time τ thus determined are set in the timer of the microcomputer in step 500. After being set, the program returns to the main routine (step 600).

FIG. 12 is a knock state detection and knock control routine which is the main object of the present invention. In the case of this embodiment, this interrupt routine is executed near the top dead center (TDC) of the engine for each ignition cycle.

When the interrupt routine is entered at step 700,
In the method as described in the operation of the invention, step 800 which constitutes a knock state detecting means for detecting an average knock state over a relatively large number of cycles is executed. This step 800
Details of will be described later. Next, in step 900, it is determined whether or not the current knock control execution condition is present,
If there is no execution condition, the process directly returns in step 1400.

If it is under the knock control execution condition, in step 1000, a knock determination level Vref for determining the presence or absence of knock for each cycle is calculated. Although the knock state can be separately detected by the method of the present invention, since a relatively multicycle period is required to detect it,
Considering the case where knocks occur frequently within a very short period of time, such as when the engine is suddenly accelerated, the conventional knock control system method of determining the knock for each cycle and retarding the ignition timing for each cycle is also used. I did it.

This knock determination level Vref is V _M obtained for each cylinder in step 800 (a value that converges to the median of the frequency distribution, details will be described later), and a constant K set for each cylinder (however, values also will be modified appropriately to an appropriate direction at a later step.) Vref by using the = K · V _M
It is created in the relationship of. Therefore, although the knock determination level differs for each cylinder, this allows the knocks of all the cylinders to be detected with high accuracy. The initial value of K is stored in the ROM as a two-dimensional map of engine speed N and cylinders.

Next, at step 1100, the knock intensity value V taken in as the value of the immediately preceding ignition cycle and the knock determination level Vref are compared in magnitude to make knock determination for each ignition. This is a step 1200 for calculating the retard amount R of the ignition timing according to the result (details will be described later). Next, in step 1300, it is determined whether or not the current knock determination level is appropriate, and this is always corrected in an appropriate direction (details will be described later). After this,
Return is made at step 1400.

Next, the step 800 of detecting the knock state will be described in detail with reference to FIG. In step 801,
The knock intensity value V taken this time is compared with the above-mentioned upper threshold value V _H corresponding to the cylinder, and if V> V _H ,
The number of times _NH is exceeded is incremented by one (step 802). In other cases, the routine proceeds to step 803, this time the knock intensity value V is compared with the lower threshold value _VL . If V <V _L , increase the number of rounds by one (step 80
4) Go to step 805. Otherwise, go directly to step 805. Then in step 805, the cumulative 50
The value V _{M of the} % point is updated. The update method is as described above. That is, as the exceeded current V _M which Nonnku intensity value V corresponding to the cylinder (V> V _M), the V _M is increased by [Delta] V _M, contrary to the V <the same [Delta] V _M and V _M if V _M Decrease by the amount. By doing so, V _M is constantly updated so as to always follow the cumulative 50% point.

Next, in step 806, which is the distribution shape changing means, it is checked whether or not the time to detect the knock state has been reached.
That is, the knock state is detected at predetermined intervals. In this embodiment, this interval is set to about 0.7 sec. There are two reasons why the interval is not the cycle number but the time interval. The first reason is that the knocking state is detected as close to human sensory evaluation as possible. Generally, when a person evaluates a knock state, it is determined by how many knocks occur in a certain time. In other words, the engine cycle progress is an unrelated concept for the driver of a car, and only the time between them is the evaluation standard that the driver can know. In this way, by detecting the knocking state on an hourly basis, it is possible to bring out the best output and fuel efficiency without making the driver of the vehicle uncomfortable.

The second reason relates to the subsequent updating of the thresholds V _L , V _H. As described above, in order to check whether the knock state is larger or smaller than the target value, it is necessary to set the lower threshold value V _L to the cumulative percentage point that is the same as the knock occurrence frequency in the target knock state. There is. However, in general, the target knock state is usually set smaller as the engine speed increases. Even if a relatively large amount of knock occurs in the low speed range of the engine, the engine will not be damaged and the output and fuel efficiency will be improved. However, in the high-speed range, preignition may be induced and the engine may be damaged even in a relatively infrequently knocked state. For this reason, it is necessary to suppress the knock state to a relatively small value in the high speed range. Therefore, the higher the speed, the lower the threshold
It is desirable to set V _L to a small percentage cumulative point. this
In order to set V _L to a small%, it is sufficient to increase the number C of updating cycles of V _L as already described in the operation. If the updating of V _L is executed every time, the number of cycles C during that time becomes larger as the speed becomes faster, so that the above-mentioned object can be achieved.

Next, in step 807, it is checked if the engine conditions have changed suddenly during the detection interval. That is, it is checked whether the engine speed and the engine load have changed by a predetermined value or more within 0.7 seconds. In this embodiment, the allowable change in engine speed is ± 30
0 rpm, ± 150 mmHg of allowable change of engine load in pressure conversion
Is set to. If it is determined in this step 807 that it is quasi-stationary, the knocking state is detected in step 808 which is the determining means.

The knocking state is detected by comparing the number of times N _H and N _L for each cylinder. That is, if N _H and N _L are almost equal, it is determined that the knocking state of the cylinder is smaller than the target value, and if it is larger than N _H and N _L even if the error amount is taken into consideration, it is higher than the target value. It is judged to be a big knock condition. The detected knock state is stored as a flag for each cylinder, and a knock state detection completion flag indicating that the knock state is detected is set.

Next, at step 809, the upper threshold value V _H and the lower threshold value V _L are updated. The update method is as follows. First, V _H and V _L have a relationship of V _H / V _M = V _M / V _L as described in the action. Let this geometric ratio be A (V _H / V _M ＝ V _M / V _L ＝
By updating A) and A, V _H and V _L can be updated. Since V _L = V _M / A, if the knock intensity value V is lower than V _L even once during this detection interval (that is, if the N _L counted in step 804 is 1 or more). Increases V _L by Δ by increasing A by ΔA
V _L is increased by ΔV _L by decreasing V _L and decreasing A by ΔA when N _L = 0. This allows V _L to be set to the required cumulative percentage point. And the upper side of the threshold V _H can be made of V _H = A · V _M. In this way, V _H and V _L are updated for each cylinder.

Next, in step 810, N _H and N _L are cleared (N _H = 0, N _L =
After 0), the counter that creates the detection interval of 0.7 sec is cleared (step 811).

Next, with reference to FIG. 14, a method for calculating the ignition retard amount R in step 1200 of FIG. 12 will be described in detail.

First, if in step 1201 the knock intensity value V exceeds the knock determination level K · V _M, the retard amount from the basic ignition timing is regarded as the cycle is knocking cycles R
Is increased by ΔR ₁ (step 1202). This ΔR ₁ is also used for general knock control, and is usually 0.
Although it is about 5 ° to 2 ° CA, it is set to 1 ° CA in this embodiment. Next, at step 1203, the counter for the number of knock determinations is incremented. This counts the number of times regarded as a knock cycle, and is used as a judgment material for the knock judgment level correction described later.

In step 1201, if the V ≦ K · V _M considered non knock cycle proceeds to step 1204. Step 12
In 04, the ignition timing is corrected in the advance direction by decreasing R by ΔR ₁ only when the non-knock cycle is continued for a predetermined time (generally about 1 second). In other cases, the current retardation amount R is retained as it is.

Next, at step 1205, it is checked whether or not the above-described average knock state is detected by the knock state detection completion flag at step 808 in FIG. If not yet detected (ie during the detection interval mentioned above)
The process proceeds to step 1213, but if the detection completion flag is set, the size of the knocked state is checked in step 1206. That is, if the flag that the knock state is larger than the target value is set, the retard amount R is increased by ΔR ₂ in step 1207. This ΔR ₂ is different from ΔR ₁ in which the knock is determined and retarded for each cycle described above, and it is a correction amount for correcting the ignition timing depending on the average state of the knocks, so a value smaller than ΔR ₁ , for example, 0.5 ° CA
The degree is good. As a result, the retard amount R is slowly corrected in accordance with the average knock state, and is also corrected in small increments centering on this to quickly avoid knock depending on whether or not knock occurs in each cycle.

When the retard amount R is set for each cylinder (that is, in the case of the ignition timing control for each cylinder), the retard amount R of the cylinder may be increased or decreased according to the knock state of the cylinder as described above. In the case of uniform ignition timing control for all cylinders, there is only one retard angle amount R. At this time, the knock state of each cylinder may be comprehensively determined and increased or decreased. For example, in the case of a four-cylinder engine, it is possible to determine that the knocking state is large when the knocking state is large in two or more cylinders. Alternatively, instead of having the above N _L and N _H for each cylinder, one N
It is also possible to directly know the knock state of the whole by counting the number with _L and N _H.

Even in this case, it is desirable to have two thresholds V _L and V _H for each cylinder, to perform comparison for each cylinder and use only the comparison result (that is, only N _L and N _H ) for all cylinders. if you sacrifice some accuracy of detection of the state, it is possible to commonly V _L, a V _H on all cylinders.

Next, at step 1208, the retard amount R is set to the maximum retard amount limit Rm.
Compare with ax. Generally, the retard amount R is usually limited to avoid excessive ignition timing fluctuation due to knock control and to prevent exhaust temperature rise.
This limit value Rmax was a predetermined fixed value. However, in this example, this is learned by the average knock state. That is, the retard amount R is kept within the range of the minimum retard amount limit Rmin and the maximum retard amount limit Rmax so that the ignition timing is not unnecessarily advanced or conversely retarded. Then, we will learn how to absorb changes in seasons, differences in gasoline properties, and engine variations. If the ignition timing is too retarded (that is, if R becomes too large), the exhaust temperature rises, which causes a problem. However, if the engine is knocking too often, the engine must be retarded further. May be damaged. In this case, engine protection is prioritized. It is also possible to detect this state and turn on the abnormal lamp.

Step 1208 is a step for determining whether or not the control range of the retard amount R should be corrected in the retard direction. That is, when R is required to be larger than the maximum retardation amount Rmax and the average knocking state is large, Rmax is set to ΔRmax (for example, about 0.2 ° CA) in step 1209.
And at the same time increase Rmin to ΔRmin (also 0.2 ° C
By increasing only A), the limit range of R is corrected in the more retarded direction. This control range is preferably, for example, about 5 ° to 7 ° CA (Rmax-Rmin≈5 ° CA to 7 ° CA).

If it is determined in step 1206 that the knocked state is smaller than the target, the reverse operation described above is performed. That is, in step 1210, R is decreased (that is, corrected in the advance direction) due to the knocked state,
At 212, the control range of the retard amount R is corrected in the advance direction.

Thus, the process proceeds to step 1213, where the retard amount R is finally obtained.
Limit. That is, when R> Rmax, R = Rmax, and when R <Rmin, R = min.

Next, the process of automatically correcting the knock determination level used for the knock determination in each cycle in step 1300 of FIG. 12 in the proper direction will be described with reference to FIG.

First, in step 1301, it is checked by the knock state detection completion flag whether or not a knock state is detected. If it is detected, the knock state is checked in step 1302. If it is determined that the knock state is larger than the target, the number of knock determinations for each cycle so far is checked in step 1303 (the number of determinations is counted in step 1203 in FIG. 14).

When the number of knock determinations is less than or equal to the predetermined value, it is determined that knock determination is possible but it is too high, and K is decreased by ΔK to lower the knock determination level (step 1304). That is, when the knock state is large and the number of knock determinations is small, the knock determination level is too high and the knock determination is erroneous. In this case, if the number of knock determinations is relatively large, it means that the knock determination has been normally performed.
Hold as is.

If it is determined in step 1302 that the knock state is smaller than the target, it is checked in step 1305 whether the number of knock determinations is too large. If the number of knock determinations is too large, K is increased by ΔK in order to correct the knock determination level to a high value (step 1306).
That is, when the knock state is small and the number of knock determinations is too large, the knock determination level is too low and the noise cycle is erroneously determined as knock. If the number of knock determinations is small, this is only because the engine is not actually knocked, so the suitability of the knock determination level is unknown (may be appropriate or may be too high). Therefore, in this case, K is held as it is.

The knock determination level is corrected for each cylinder.
That is, K corresponding to each cylinder is corrected in an appropriate direction based on the knock state for each cylinder and the number of knock determinations for each cylinder. Of course, it is also possible to correct K as one common K without having it for each cylinder in the same way as described in the method of calculating the retardation amount R. Further, the determination of the number of knock determinations (steps 1303 and 1305) may be performed as "0" or "1" as if there was or was not a knock determination during that time.

Next, in step 1307, the knock determination count is cleared, and subsequently, the knock state detection completion flag is also cleared (step 1308).

As described above, knock control is executed.

In the above embodiment, the average knock state is detected, and thereby the ignition timing is manipulated, the ignition timing control range is modified, and the knock determination level for the knock determination for each cycle is modified. Can all be effective alone. For example, for detection of knocking state, this can be directly applied to a knock monitoring device or a gasoline octane number detecting device. As an example applied to an actual automobile, it can also be used to display a warning light when knocking is likely to occur abnormally due to gasoline properties or some other cause.

Regarding the operation of the ignition timing, in the above embodiment, the ignition timing is operated for each ignition according to the knock determination result for each cycle, and the ignition timing is operated in a longer cycle according to the average knock state. The double loop with is adopted. This is because the transient state of the engine in which a sudden change in knock state is expected is taken into consideration. However, since the average knock state can be calculated within, for example, 1 second as in the present embodiment, it is possible to correct only the knock state without modifying the ignition timing for each ignition. In this case, the knock determination level for determining the presence or absence of knock for each cycle is not necessary.

It is also possible to configure a simple knock avoidance system that detects the knock state for a longer period of time, and switches the ignition timing characteristic to two or more stages using a plurality of maps. .

Further, although the ignition timing is treated as a factor for controlling the knock in the above-mentioned embodiment, this is due to the knock control factors such as the air-fuel ratio, the intake pressure (supercharging pressure such as turbo), the EGR, the anti-knock agent, etc. It can be widely applied. In this case, an immediate knock control factor such as ignition timing is controlled according to the knock determination result for each cycle, and an average knock occurrence such as boost pressure is generated according to the detection result of the average knock state. Various arrangements such as controlling the factors that influence the state are also possible. Of course, only one of the ignition timing, air-fuel ratio, intake pressure, EGR, anti-knock agent, etc. may be controlled.

Further, regarding the correction of the control range of the knock control factor such as the ignition timing, it is possible to perform only this according to the knock state. That is, the control can be performed only according to the knock determination result for each cycle, and only the control range can be corrected according to the average knock state.

Further, regarding the modification of the knock determination level, it is not always necessary to commonly use the signal to be determined to be knocked for each cycle and the knock intensity value for detecting the knock state (that is, the signal integration value in the embodiment). For example, the maximum peak value V _MAX of the signal may be used as the signal to be judged to be knocked, and the integral value ∫V may be used to detect the average knock state. In this case, it is not necessary to directly know the value of V _MAX that becomes the signal to be judged. That is, using a comparison circuit, etc., V _MAX
Only information about whether is greater or less than the knock decision level is needed. However, it is more efficient to use them in common as in the above embodiment.

Further, in the above embodiment, the knock intensity value is set by setting the relationship between the two threshold values V _L and V _{H in} a substantially equal relationship with the median value M (50% point of the cumulative distribution) of the distribution. It is determined whether or not the frequency distribution of V is a substantially log-normal distribution within a range of predetermined cumulative percentage points _{VL to VH.}
Since some error may occur in the approximate value V _M that should converge to, it is better to add a slight offset D such as V _L = V _M / A, V _H = (A + D) ・ V _M In some cases, it may be convenient to determine whether or not the frequency distribution of the intensity values V is a substantially log-normal distribution within the range of the predetermined cumulative% points _{VL to VH} . Since this offset D can be used as an element for finely adjusting the target knocking sound, either a positive sign or a negative sign can be taken, but essentially a positive value is desirable for D. This is because, in order to obtain V _M , the same amount ΔV _{M is} increased / decreased to converge to the median value M according to the knock intensity value V, but even if it converges, the median value M fluctuates by the same amount. Therefore, if you look at this on the logarithmic axis, the amount that deviates to the low side inevitably increases. Therefore, it is desirable to make D positive in the sense of correcting this. Or in order to correct this from the beginning, raise V _{M a} little more than 50% cumulative (for example, cumulative
It is possible to converge toward the position of 55%). In this case, the ratio of increase / decrease of V _M may be set to 55:45, for example.

Further, in the above embodiment, the lower threshold value _VL is corrected by feedback so that it becomes the cumulative percentage point of the same value as the knock occurrence frequency in the target knock state,
A fixed value may be used when the distribution of knock intensity values can be predicted to some extent or when it is not necessary to detect the knock state with such accuracy. Therefore, in this case, V _H is also a fixed value, and it is not particularly necessary to find the median value M of the distribution. However, it is clear that the above embodiment is far superior.

In the above embodiment, the average knock state is detected by comparing the number of times the upper threshold value V _H is exceeded and the number of times the lower threshold value V _L is decreased. It is also possible to detect the knock state.

That is, the cumulative α% point when accumulated, Lα, the cumulative α% point when accumulated from the higher value, Hα, and the median value V _{M of the} distribution are calculated from the lower value. , V _M / Lα and Hα / V _H are examined, it is possible to determine whether the knock state is larger or smaller than the target knock occurrence frequency α%.

That is, in the small knock state in which the knock occurrence frequency is less than α%, Lα, V _M , and Hα are aligned on a log normal probability paper. Therefore, in this case, Lα and Hα are V _M
And V α / L α and _H α / V _M are almost equal to each other.

On the contrary, in a large knock state in which the knock occurrence frequency is higher than α%, this geometric relationship is broken, and Hα / V _H is more V _M / L α
Get bigger. As will be apparent from the disclosure of the operation part of the invention, Hα and Lα are determined by comparing knock intensity value V with Lα and Hα for a predetermined number of cycles C, and V is Lα for a predetermined number of times. If it is lower than the above, Lα is decreased by ΔLα, and if it is less than the predetermined number of times, Lα is decreased by ΔLα.
Lα and Hα are increased by Lα, and at the same time, when V exceeds Hα by the predetermined number of times or more, Hα is increased by ΔHα, and when V is less than the predetermined number of times, Hα is decreased by ΔHα. Converge to the cumulative percentage point.

Further, in the above-mentioned embodiment, the method and the apparatus which do not use the logarithmic converter are shown, but the use of the logarithmic converter or the software for the logarithmic conversion instead thereof is allowed in consideration of the constraint conditions such as the cost. Then, various methods and devices can be considered. For example, it is also possible to detect the knock state by the magnitude relationship between the average value and the median value of the logarithmic conversion value log V of the knock intensity value V. That is,
When the knocked state is small, the distribution of log V is almost normal, and the median and average of the normal distribution match. However, when the knock state becomes large, the average value becomes larger than the median value, so that the knock state can be identified.

Further, in the above-described embodiment, the rectification in the predetermined section of the sensor signal and the integral value ∫V are used as the knock intensity value V, but other values corresponding to the maximum peak value may be used. This has already been described in detail in the operation of the invention.

Further, in the above embodiment, the type of sensor that detects the vibration of the engine block is used as the knock sensor,
As described in detail above, a microphone, an in-cylinder pressure sensor, a combustion light sensor, or the like can be used.

Also, as already explained, the assumed distribution is not limited to a lognormal distribution, and by applying an appropriate conversion such as a binomial distribution or a gamma distribution, it can be treated as a lognormal distribution for practical purposes. With respect to such a distribution, the above embodiment can be applied as it is.

〔The invention's effect〕

As described above, in the present invention, the frequency distribution when the knock intensity V corresponding to the maximum wave height value is subjected to multi-cycle sampling is a substantially log-normal distribution within the range of the predetermined cumulative percentage points _{VL to VH.} Since the average knock state is detected by determining whether or not it is true, the true knock state can be detected very accurately without being affected by the variations of the knock sensor or engine, which have been difficult in the past, and changes over time. Not only can it be detected, but the range of the predetermined cumulative percentage points V _{L to} V _H that determines whether or not it is a substantially lognormal distribution can be changed according to the engine condition, so that a knock that is adapted to the engine condition can be obtained. There is an excellent effect that the state can be detected.

[Brief description of drawings]

FIG. 1 is a diagram showing the maximum value of the knock signal and its distribution state, FIG. 2 is a diagram showing the cumulative distribution of the maximum value of the knock signal and this distribution on a log-normal probability paper, and FIG. 3 is a diagram showing the knock signal. FIG. 4 is a diagram showing the cumulative distribution of the integrated value on a log-normal probability paper, and FIG. 4 is a diagram showing the cumulative distribution of the maximum value of the knock signal on the log-normal probability paper in the same knock state as in FIG. 3, and FIG. FIG. 6 is a diagram showing distribution states of knock intensity values of knock cycles and non-knock cycles, FIG. 6 is a diagram for comparing distributions of knock intensity values when knocks are smaller and larger than a target, and FIG. FIG. 8 is a detailed configuration diagram of a control circuit in FIG. 7, FIG. 9 is a detailed configuration diagram of a knock intensity value detection circuit in FIG. 8, and FIG. 10 is a knock intensity. 11 to 15 are signal waveform diagrams used to explain the operation of the value detection circuit. Kicking is a flowchart showing the procedure of detection and control of the knock. 1 ...... engine, 3 ...... igniter, ignition coil, 4 ...... knock sensor, 7 ...... injector, 8 ...... turbocharger, 9 ...... O ₂ sensor, 10 ...... control circuit, 10-1 ...... CPU, Ten
-2 ... ROM, 10-3 ... RAM, 10-10 ... Knock intensity value detection circuit, 10-10-2 ... Rectifier, 10-10-3 ... Integrator.

Claims (13)

[Claims] 1. A knock sensor for detecting a knock of an engine, a knock intensity value detecting means for extracting a knock intensity value V corresponding to a maximum peak value of this signal from a signal of the knock sensor, and a knock intensity value V. the obtained when multi-cycle sampling, frequency distribution of the knock intensity value V is determined whether or not the substantially log-normal distribution in the range of the knock intensity value predetermined cumulative percentage point of the frequency of V V _L ~V _H Determining means, and the predetermined cumulative% for determining whether or not the logarithmic normal distribution
Knocking state detecting means for detecting an average knocking state over a plurality of cycles according to the discrimination result of the discriminating means, and a distribution shape changing means for changing in accordance with the engine condition in the range of points _{VL to VH.} A knocking detection device for an internal combustion engine, comprising: 2. The knock intensity value is a value obtained by rectifying and integrating a knock vibration output including a frequency component peculiar to the knock within a predetermined section, or when the knock vibration output is compared with a predetermined level within the predetermined section. The knocking detection device according to claim 1, which is the number of pulses of the obtained pulse train, the integrated value of the pulse train, or the effective value of the knock vibration output. 3. The knocking detection device according to claim 1, wherein the approximately lognormal distribution is a lognormal distribution. 4. The determination means is set to a relatively low first threshold value V _L and a relatively high value within a range of possible values of the knock intensity value V. A second threshold value V _H, and compares the knock intensity value V with the threshold values V _L and V _H over a plurality of cycles to obtain a knock intensity value V
Counts the number of times that the first threshold V _L falls below the second threshold V _H , and the number of times the knock intensity value V exceeds the second threshold V _H.
The knocking detection device according to any one of claims 1 to 3, wherein the frequency distribution is substantially discriminated based on the magnitude relationship between the count values. 5. The first threshold value V _L is higher than the cumulative 50% point M with respect to the cumulative 50% point M of the frequency distribution obtained when the knock intensity value V is subjected to multi-cycle sampling. The knocking detection device according to claim 4, wherein the second threshold value V _H is set to a smaller side and the second threshold value V _H is set to a larger side than the point M. 6. The discriminating means has means for estimating the cumulative 50% M points by the value of the knock intensity value V inputted one after another, and with respect to the estimated value V _{M of the} point M, The first
Of the threshold value V _L and the second threshold value V _H are substantially equal (V _M / V _L + D≈V _H / V _M ), or V _H is set to a value slightly larger than the ratio. The knocking detection device according to claim 5, wherein the knocking detection device has a relationship (V _M / V _L + D = V _H / V _M , D> 0). 7. The estimated value V _M is a variable that is sequentially updated in each cycle, and is set to V every time a knock intensity value V is input.
By comparing V _M with V _M , if V is larger, V _M is increased by ΔV _M , and if it is smaller, V _M is decreased by ΔV _M , so that V _M becomes the cumulative 50% point M. The first threshold value V _L using this value of V _M and a predetermined geometrical ratio A is changed so as to converge to V _L = V _M / A, the second threshold value
V _H is V _H = V _M · A or V _H = V _M · (A + D) (D> 0)
Claim 5 is characterized in that it is set in the relationship of
The knocking detection device according to item 6 or 7. 8. the geometric number A and variables will be updated every predetermined number of cycles C, and during this cycle number C said knock intensity value V and the first threshold value V _L (= V _M / Compare with A),
When the knock intensity value V is lower than V _L by a predetermined number of times or more, the geometrical ratio A is slightly increased to correct V _L to the lower side, and when it is less than the predetermined number, V _L is increased. The geometrical ratio A is slightly decreased to correct the direction, and the probability that the knock intensity value V falls below the first threshold value V _L is substantially equal to the desired probability. The knocking detection device according to claim 7, wherein the number A is adjusted. 9. The probability that the knock intensity value V substantially falls below the first threshold value _VL by changing the number of cycles C according to engine conditions by the distribution shape changing means. 9. The knocking detection device according to claim 8, characterized in that is changed according to engine conditions. 10. The cycle number C takes a larger value as the engine speed increases, whereby the probability that the knock intensity value falls below the first threshold value V _L is substantially equal to the engine speed. 10. The knocking detection device according to claim 9, wherein the knocking detection device is configured such that the knocking detection device gradually decreases as it does not rise. 11. The knocking detection device according to claim 10, wherein the cycle number C is set so that the cycle elapsed time becomes substantially constant. 12. The knocking strength value V, to which the discriminating means successively inputs the value V _M of the cumulative 50% point M of the frequency distribution obtained when the knocking strength value V is subjected to multi-cycle sampling.
And the knock intensity value V, the desired cumulative α when cumulative from the lower value
And a means for estimating a cumulative α% point Lα corresponding to%, and a means for estimating a cumulative% point Hα when the knock intensity value V is accumulated from the higher value, V _M / Lα Hα / V _M
The knocking detection device according to any one of claims 1 to 3, wherein the frequency distribution is substantially determined based on a magnitude relationship with 13. The cumulative percentage points Lα, Hα are updated every predetermined number of cycles C, and the predetermined number of cycles C
Between the knock intensity value V and Lα and Hα, Lα is decreased by ΔLα when V is lower than Lα by a predetermined number of times or less, and Lα is decreased by ΔLα.
By increasing Lα, at the same time, when the knock intensity value V exceeds Hα by the predetermined number of times or more, Hα is increased by ΔHα, and when the knock intensity value V is less than the predetermined number of times, Hα is decreased by ΔHα. 13. The knocking detection device according to claim 12, wherein the knocking detection device is converged to a point.