JPS61245039A - Knocking detecting device for internal combustion engine - Google Patents

Abstract

PURPOSE:To raise the detection accuracy of knocking by detecting a combustion pressure vibration at every one combustion cycle, and obtaining a physical quantity related to a combustion vibration energy of the time of non-knocking and the time of knocking. CONSTITUTION:A combustion pressure vibration is detected at every one combustion cycle by a cylinder internal pressure sensor 1, a prescribed knocking vibration is extracted by a band pass filter 5, and a physical quantity related to a combustion vibration energy of the time of non-knocking and the time of knocking is obtained by an operating means 17. Subsequently, by a smoothing means of a control unit 18, a smoothing value of at least one of these physical quantities is derived, and by comparing the physical quantity and this smoothing value, whether knocking has been generated or not is decided, and also, under a specified condition, the smoothing value of the physical quantity is learned and stored in a RAM 23. When an engine is operated under the specified condition, the smoothing value of the physical quantity which has been learned in advance is read out of the RAM 23, this learned smoothing value is replaced with an output of the smoothing means, and whether knocking has been generated or not is decided.

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention relates to a device for detecting knocking that occurs in an internal combustion engine of an automobile or the like, and particularly to a device that improves the accuracy of non-knocking detection during transient operation.

(Conventional technology) Non-king is mainly caused by compression ignition of unburned gas, and if it occurs violently, energy loss (reduction in output) occurs.
It is desirable to avoid this, as it may cause shock to the parts of the engine and other parts of the engine, as well as a decrease in fuel efficiency.

However, the mild non-king phenomenon itself does not have a negative effect on the engine, and even if non-king occurs by advancing the ignition timing, the fuel efficiency of the vehicle can be improved by increasing the combustion efficiency of the engine. From the viewpoint of improving fuel efficiency, allowing a moderate amount of knocking is preferable in order to obtain an operating state with optimal engine efficiency. Therefore, in order to increase the operating efficiency of the engine and suppress the non-king noise level to a predetermined value or less, it is necessary to control the non-king intensity to suit various operating conditions.

As a conventional non-king detection device that was used in a non-king control device that performs such control, for example, the
There is one such as that disclosed in No. 13749.

This device includes a knocking sensor such as a cylinder pressure sensor provided on the washer of the spark plug, a signal amplifier, a low-pass filter, an averaging circuit such as a knocking signal envelope processing circuit, and a comparison circuit. It is equipped with an ignition timing control circuit. And almost 5-6 kh of cylinder pressure signal
A specific frequency band component above z is detected as a knock signal, and the signal that has been subjected to peak hold processing, envelope processing, etc. is regarded as a signal corresponding to non-king energy, and is directly compared with a predetermined reference value. The knocking level is controlled by advancing or retarding the ignition timing based on the results.

(Problems to be Solved by the Invention) However, it is generally well known that the above-mentioned specific frequency band component is included in the output signal of the sensor even during non-knocking when no non-knocking occurs.

FIG. 10 shows an example of a power spectrum of cylinder pressure vibration of an internal combustion engine. This is an example of a 4-cylinder 1800cc engine operated at full load and 4800 RPM, but it has been confirmed that other engines have almost the same tendency. Therefore, the electric signal obtained by applying signal processing such as envelope processing and peak hold processing to the sensor output signal in a specific frequency band is regarded as the signal amount corresponding to the non-king energy, and this is directly compared with the standard comparison level. In conventional devices configured to do this, it may be difficult to determine the knock level.

Figure 10 shows a knock with a relatively large level, so the difference in power level between non-knocking A and knocking B is about 10 dB or more, but the so-called trace knock is considered to be the limit for the presence or absence of knock. Since this difference is approximately 2-3 dB under normal conditions, detection of such knocks is very difficult in conventional devices.

Furthermore, in the case of the sensor installed in the spark plug washer mentioned above, the level of the sensor output signal during non-knocking may change due to changes in the mechanical vibration system due to the load acting on the sensor surface or tightening torque during installation. may change significantly, and this change may be larger than the change in the output signal due to these effects during knocking. Furthermore, according to experiments conducted by the applicant using several types of engines,
The sensor output fluctuation during non-knocking is 3 to 4 times the load at idling or fully open throttle, and is 8.
It has been confirmed that the speed increases by 2 to 3 times when the engine rotates from 00 PPM to 4800 RPM.

However, in conventional devices, signals with such large fluctuations are directly compared with a reference value, so in order to realize a device that can accurately determine knock and non-knock, it is necessary to check each engine and its operation. There was a problem in that a huge amount of experimentation was required for each condition and even for each sensor.

In addition, in a transient state where engine operating conditions suddenly change, the average value of vibration is affected by past vibration values, so
There is a problem in that it is difficult to accurately detect the knocking level during or immediately after a sudden change, and the accuracy of knocking detection decreases.

(Objective of the Invention) Therefore, the present invention detects combustion pressure vibration for each combustion cycle, obtains physical quantities related to combustion vibration energy during non-knocking and during knocking from the vibration, and
Find a smoothed value of at least one of these physical quantities,
By comparing the above physical quantity and this smoothed value, it is determined whether or not knocking has occurred, and the smoothed value of the above physical quantity is learned and memorized under specific conditions, and when the engine reaches the specific condition, the learned physical quantity is By performing the above comparison using the smoothed value of , the purpose is to accurately detect non-king regardless of the engine model, sensor installation state, operating conditions, etc., and to improve the detection accuracy.

(Structure of the Invention) The knocking detection device for an internal combustion engine according to the present invention, as shown in the basic conceptual diagram in FIG. The extraction means for extracting the knock vibration of the engine includes an operating state detecting means C for detecting the operating state of the engine, a specific condition determining means d for determining whether the engine is being operated under a predetermined specific condition, and an engine detecting means C for detecting the operating state of the engine; a calculation means e for calculating first and second physical quantities related to combustion vibration energy during non-knocking and knocking based on the output of the extraction means for each combustion cycle; and at least the first and second physical quantities. a smoothing means f that smoothes one in a predetermined manner; and when the engine is operated outside of specific conditions, the output of the smoothing means f is learned as corresponding to the operating state at that time, and the smoothing learning value is stored. It is determined whether or not knocking has occurred based on the output of the calculation means e and the output of the smoothing means f, and when the engine is operated under specific conditions, the first or first Read the smoothed learning value of the physical quantity of 2, and use this smoothed learning value as the smoothing hand vIt.
The present invention is provided with a knocking determination means for performing the above-mentioned determination by replacing the output of f, thereby accurately detecting knocking regardless of operating conditions and the like.

(Example) The present invention will be explained below based on the drawings.
This will be explained with reference to Figure 2.

As is clear from FIG. 10, there is a large difference in power level between knocking A and non-knocking B in the frequency components shown. Therefore, we focused on the frequency component in which the difference is particularly large, that is, in the example shown, 8 kHz to 17 kHz, and used a bandpass filter that uses that band as the passband to extract the frequency component from the sensor output. Take out the ingredients.

FIGS. 11(A) and 11(B) show waveforms of the sensor output after passing through a bandpass filter during knocking and non-knocking, respectively. Note that these represent the waveforms of high-frequency vibrations of the cylinder pressure.

Here, the time average value of power, that is, energy, in a certain specific band is determined. It is represented by a certain signal x (t) and is obtained as the time average of the square of the signal amplitude. Therefore, the absolute value of the signal shown in FIG. 11 is obtained. Since the right side of this equation 0 indicates the RMS (root mean square) of the signal x (t),
The left side can be considered to be a quantity indicating the power of the signal X(t), or at least a quantity that is monovalently correlated with the power. Although it is assumed here that the signal x (t) of the above equations 0 and 0 is simply a signal of one frequency, it is not practically supported even if it includes a plurality of frequency components.

1121ffl (A) is the sensor output signal after passing through a band-pass filter, and (B) is the crank angle for the range from 40" before top dead center (BTDC) to top dead center (
It is the value obtained by taking the absolute value of the signal in A) and integrating it, and (C)
is the value obtained by taking the absolute value of the signal in (A) and integrating it over the range from top dead center to 40 degrees after top dead center (ADTC). FIGS. 12(B) and 12(C) each show the cylinder pressure vibration energy within the crank angle range. That is, it is obtained by dropping the term 1/2T in the above equation 0. That is,
From 40° BTDC in Fig. 12 (B) to top dead center ('r
The section DC) shows the integral signal during non-knocking, and in this section the integral signal increases almost linearly, indicating that constant amplitude energy always exists regardless of the crank angle. Therefore, the relationship is established during non-knocking.

According to various experiments conducted by the applicant, the relationship shown in FIG. 12 can be considered to hold under almost all operating conditions. However, the integral interval (in this case, 40°BTDC)
~ top dead center) must be selected carefully.The reason is that vibrations of the spark plug are also detected, so depending on the selection of the integral interval, vibrations of intake and non-air valves when seated and unseated may be detected. This is because the relationship of equation 0 no longer holds true due to the influence of .

On the other hand, top dead center (TDC) ~ 40° AT in Fig. 12 (C)
The DC section indicates the integral output at the time of knocking. That is,
In this case, an increase in energy due to knocking will appear in the expansion stroke after top dead center.

In general, it is thought that human hearing determines the knock level based on the relative strength difference between the sound pressure level due to constantly occurring background noise and the sound pressure level due to non-king vibration. . Therefore, by directly comparing the energy of cylinder pressure vibrations during non-knocking and the energy of cylinder pressure vibrations during knocking, it is possible to detect the knocking level with high accuracy, which closely matches the sensory evaluation.

Furthermore, according to Equation 0, it can be considered empirically that knocking does not occur before top dead center, so the integral signal before top dead center is independent of whether or not non-king occurs after top dead center. First, it can be seen that this can be a predicted value of the vibration energy of the cylinder pressure in the expansion stroke after top dead center in the non-knocking state.

Focusing on this point, the rectified integral value of cylinder pressure vibration within a predetermined crank angle range before top dead center, and the predetermined crank angle within a predetermined crank angle range after top dead center or including the range before top dead center. By comparing the rectified integral value of the cylinder pressure vibration within the range, it is possible to directly compare the vibration energy of the cylinder pressure during non-knocking with the vibration energy of the cylinder pressure during the combustion stroke. This means that non-king detection can be achieved.

Here, in FIG. 12, if the rectified integral value of sensor vibration in a predetermined crank angle range before top dead center is N, and the rectified integral value of sensor vibration in a predetermined crank angle range after top dead center is S, then the above-mentioned It can be seen that the S/N ratio can be used as a highly accurate knocking detection parameter (hereinafter referred to as a knock parameter). However, the value of N changes for each cycle, and since the ignition timing is included in the integral interval of N, for example, when ignition noise is superimposed, the value of N becomes abnormally large, causing the correct knocking level. may not be detected. Therefore, instead of N, we calculate the moving average value N of N, and use the following formula
It is preferable to use the S/N represented by as the knock parameter.

N = −X N old + TX N new −
-------■η However, n: An integer value, for example n=4 to 16. Nold: Previous moving average value Nnew: Current moving average value However, as shown in Figure 13.14, N can vary depending on the operating conditions. Therefore, if N is simply calculated according to Equation 0 when operating conditions suddenly change, there is a risk that the correct non-king level cannot be calculated.

Especially when accelerating at full throttle immediately after starting the engine,
Since the value of N starts from zero, it is expected that the S/N will be an abnormally large value.

Therefore, the concept of learning control is introduced to prevent such problems.・In other words, by learning the value of N under steady operating conditions and storing it in nonvolatile memory, and immediately using that learned value as N when operating conditions suddenly change,
Efforts are being made to improve the accuracy of knocking detection.

Next, a first embodiment of the present invention based on the above background art will be described with reference to FIGS. 2 to 7. This embodiment is an example in which the present invention is applied to an ignition timing control device.

In FIG. 2, reference numeral 1 denotes a cylinder pressure sensor (vibration detection means), and the cylinder pressure sensor 1 converts the combustion pressure in the cylinder into an electric charge using a piezoelectric element, and outputs an electric charge output S1. Specifically, the cylinder pressure sensor 1 is formed as a washer for a spark plug 3 screwed onto the cylinder head 2, as shown in detail in FIGS. 3(A) and 3(B). The spark plug 3 is pressed and fixed in the recess by the portion 3a.

The sensor output S1 is input to the charge amplifier 4,
Charge amplifier 4 is an operational amplifier OP.

A so-called charge-voltage conversion amplifier is constituted by resistors R1 and R2 and a capacitor C, converts the sensor output SL into a voltage signal S2, and outputs the voltage signal S2 to a bandpass filter (extraction means) 5. The bandpass filter 5 filters the frequency band corresponding to knocking vibration (for example, 6 KHz =
19 KHz) is allowed to pass through and is output as a signal S3 to the first and second signal processing circuits 6.7. Also,
The output from the crank angle sensor 8 is input to these first and second signal processing circuits 6.7, and the crank angle sensor 8 detects the crank angle of the engine. For example, in the case of a 6-cylinder engine, the crank angle the reference signal C1 every 0”,
A position signal C1 is output for every 1" of crank angle. The first signal processing circuit 6 is a circuit that detects vibration energy during non-knocking, and integrates the absolute value of the output signal S of the bandpass filter 5 as a reference. An absolute value integrator 9 that holds and resets its integral value in response to a signal C1, and a presettable counter 1 that is preset to a value corresponding to a predetermined crank angle and counts a position signal C8 in accordance with a reference signal Ci.
0 and 11, and this presettable counter 10 and 1
1, and a flip-flop 12 that controls the operation of the absolute value integrator 9 according to the output of the absolute value integrator 9.

Further, the second signal processing circuit 7 is a circuit that detects vibration energy at the time of knocking, and includes an absolute value integrator 13, a pre-safety pull counter 14, 15, and a pre-safety pull counter 14, which correspond to each section 9 to 12 of the first signal processing circuit 6, respectively. It is composed of a flip-flop 16.

In this embodiment, the present invention is applied to a six-cylinder engine, and the compression top dead center is set to 70 degrees after the crank angle reference position signal Ci is generated, and furthermore, the compression top dead center is set to 70 degrees before and after the compression top dead center.
If the vibration energy during non-knocking and knocking is detected between 0 degrees, the presettable counters 10 and 11 have crank angles of 30 degrees and 7 degrees, respectively.
0 degrees, and the presettable counters 14 and 15 are preset with values corresponding to crank angles of 70 degrees and 110 degrees, respectively. The first signal processing circuit 6, the second
The signal processing circuit 7 and the crank angle sensor 8 collectively constitute a calculation means I7.

The outputs S4, S of the absolute value integrator 9.13 and the output S6 of the presettable counter 15 are inputted to the control unit 18, and the output signal S7 from the operating state detection means 19 is further inputted to the control unit 18. Ru. The operating state detection means 19 detects parameters representing the operating state of the engine (for example, intake pipe pressure, throttle valve opening, starter motor operation, etc.) and outputs a signal S7.

The control unit 18 performs arithmetic processing for ignition timing control based on each input signal and outputs an ignition signal Sp corresponding to the optimum ignition timing to the ignition device 20.
generates a high voltage at this optimal ignition timing to ignite the air-fuel mixture.

The MPU 21 takes in necessary external data from the I10 interface U according to the program written in the ROM 22, and while exchanging data with the RAM 23, calculates and processes the processing values necessary for knocking control. The processed data is output to the I10 interface 24. The I10 interface includes an absolute value integrator 9.13 and a control unit 18.
The signal from and the interrupt signal S6 are input, and
The I10 interface 24 outputs an ignition signal Sp. The ROM 22 stores calculation programs for the MPU 21, and the RAM 23 is made of, for example, a non-volatile memory and stores data used in calculations in the form of a map, and retains the stored contents even after the engine is stopped.

FIG. 4 shows an example of signal waveforms of various parts at crank angles of 0 to 120 degrees.

That is, FIG. 4(A) shows the reference signal C1 (120° signal in the case of a 6-cylinder engine) generated from the crank angle sensor 8, and FIG. 4(B) shows the position signal C1 generated every 1° of the crank angle. C) is the output signal S2 of the charge amplifier 4
, (D) are the output signals S1, (E
) is the output signal S of the flip-flop 12, (F) is the output signal S3 of the flip-flop 16, and (G) is the output of the absolute value integrator (hereinafter simply referred to as "integrator") 9 of the first signal processing circuit 6. Signals S4 and (H) indicate the output signal S of the integrator 13 of the second signal processing circuit 7, respectively.

Next, the operations of the first and second signal processing circuits 6.7 in FIG. 2 will be explained.

The reference signal S of 70 degrees before compression top dead center resets the integrators 9 and 13, and presettable counters (hereinafter simply referred to as "counters") 10.11 and 14.15
are each of the above values (30°, 70°, and 70°, 11
0″″) is preset. Each of these counters then starts counting the crank angle position signal C1, and first, the output signal of the counter 10 is inverted at a crank angle of 30 degrees. In response, the output signal S of the flip-flop 12 becomes low level (L) as shown in FIG. The integration of the output signal S is started.

At a crank angle of 70 degrees, the output of the counter 11 is inverted, so the output signal S of the flip-flop 12 becomes high level (H), and the integrator 9 holds the integrated value at that point. Furthermore, since the output of the counter 14 is inverted, the output signal S of the flip-flop 16 is
, becomes low level (L) as shown in FIG.

Further, at a crank angle of 110 degrees, the output of the counter 15 is inverted, so the output signal S of the flip-flop 16 becomes high level (H), and the integrator 13 holds the integrated value at that point. Therefore, the output signal S4 of the integrator 9 changes as shown in FIG. 4(G), and the output signal S of the integrator 13 changes as shown in FIG. 4(H).

In this way, the first and second signal processing circuits 6 in FIG.
．． 7, the integral value from 40 degrees before top dead center to top dead center,
In other words, the signal value corresponding to the first physical quantity related to vibration energy during non-knocking and the distance from top dead center to top dead center 1jiL4
It is possible to obtain an integral value up to 0 degrees, that is, a signal value corresponding to the second physical quantity related to the vibration energy at the time of knocking.

Note that the output SG of the counter 15 at a crank angle of 110 degrees is used as an interrupt signal to the control unit 18, and in response to this, the control unit 18 schedules the start of A/D conversion, which will be described later.

Next, the contents of the program executed by the control unit 18 will be explained.

FIG. 5.6 is a flowchart showing the knocking control and specific condition determination programs respectively written in the ROM 22, and P and PL7 in the figure indicate each step of the flowchart.

When the interrupt request signal S6 is issued, the control unit 18 starts executing the program shown in FIG. first,
At P, the output signal S4 of the integrator 9 is A/D converted and stored in the memory as the first physical quantity N related to the combustion vibration energy during non-knocking, and at P2, the first physical quantity N related to the combustion vibration energy during knocking is stored in the memory. As the second physical quantity S, the output signal Ss of the integrator 13 is A/D converted and stored in the memory.

Next, in P3, it is determined whether the engine is being operated under predetermined specific conditions. Note that the determination of the specific condition is executed by an interrupt processing program shown in FIG. 6, which will be described later. When the vehicle is not operated under specific conditions, the moving average value nu is calculated in accordance with the above-mentioned formula 0 in P4. Then, it is determined whether the learning conditions used to learn N are met in P4. In determining this learning condition, the flow has already flowed in accordance with the determination result that the specific condition is not met in step P□, so other conditions, mainly engine warm-up conditions, are used as the determination criterion. This takes into account the fact that it is not suitable for learning when the operating conditions of the engine are fluctuating. For example, in the case of an engine using hydraulic tappets, if air gets into the hydraulic system when it is cold, the seating vibration of the intake and exhaust valves will occur. This is because it becomes larger than after warm-up is completed.

When it is determined in P4 that the learning conditions are satisfied, the value of N is learned in P as corresponding to the driving state at that time, and the learning result is written in a predetermined table map to update the previous learned value. .

This table map is created by using, for example, intake pipe internal pressure MAP corresponding to load and engine speed R as parameters.
Created as a function of dimension. If the learning conditions are not met, jump to P6 and proceed to P7.

On the other hand, when the operation is performed under specific conditions in step P, the corresponding optimum learning value Nt is determined from the learning table map of N based on the operating state at that time.
Look up, set N-Nt in PKJ, and proceed to P7.

At P7, a knock parameter KN (KN-3/N) indicating the level of knock is calculated based on S and N, and at P, a slice level SL for determining the presence or absence of knock is calculated from the table map shown in FIG. Look up. Since the mechanical vibration of the engine tends to increase as the engine speed R increases, the relative S/N of the same knock level
becomes smaller as shown in the figure. This table map stores in advance the slice level SL at which the knock level becomes a trace at each engine speed R. Then,
The knock parameter KN is compared with the slice level SL at pH, and when KN≧SL, it is determined that knock has occurred and the PI2
The ignition timing is corrected to the retarded side by P, and an ignition signal Sp is output at a timing corresponding to the corrected ignition timing of the lever. On the other hand, when KN<SL, it is determined that there is no knock, and control is performed to the optimum advance angle value at P and 4 (for example, the current advance angle value is held or the advance angle is corrected), and the process proceeds to PI3.

In this way, the presence or absence of knock is determined by determining the knock parameter KN and slice level SL from the combustion vibrations during knock and non-knock within a predetermined crank angle range for each combustion stroke. Since KN is obtained by calculating K from the learned value, knocking can always be detected accurately regardless of the engine model, sensor installation status, operating status (particularly transient status), etc.

Therefore, the ignition timing can be optimally controlled according to the knocking detection results, thereby improving the combustion efficiency of the engine, particularly improving the torque characteristics, and improving the fuel efficiency and driving performance of the engine. .

Next, the specific condition determination program shown in FIG. 6 will be explained. This program is executed once every 10m5ec.

If the following conditions are met in each step of P21 to P Yukaku, proceed to P4, otherwise proceed to P□
Proceed to step 7. In other words, at P□, to time has elapsed after the engine was started, and at P, the amount of change Δ in the intake pipe internal pressure MAP
M (ΔM=1MAPnewMA P old l ) is less than the predetermined value ΔMO and enters a transient state, and a time period of tl has elapsed from immediately after the transition at P (immediately after ΔM≧ΔMo), the amount of change in engine speed R at P2Z ΔR(ΔR
=lRnew-Roldl) is less than a predetermined value of 680, and when t2 time has elapsed since ΔR≧ΔRo at PL'5, it is determined that the engine is not being operated under specific conditions. specific flag TF
Clear (TF=O). On the other hand, if even one of the conditions PjLl to P5 is not satisfied, it is determined that the vehicle is being operated under the specific conditions, and a specific flag TF is set (TF=0) in P27. The determination process in step P in the program described above is performed by determining this specific flag TF.

FIG. 8.9 is a diagram showing a second embodiment of the present invention, and the calculation of the knock parameter KN is different in this embodiment. In the program shown in FIG. 8, steps that perform the same processing as in the first embodiment are given the same numbers and their explanations are omitted, and steps that are different are given numbers in the 30s and above to explain the processing contents. .

When following the YES command at P after passing through P2, look up the corresponding optimal learning (ISt) from the learning table map of S (details will be described later in P31) based on the operating state at that time, and set it to 5t-3old at P3. Proceed to Pl.

On the other hand, when following the NO command at P, jump PH1, "Sz and proceed to P33. At Pl3, the knock parameter K
Calculate N according to the following formula 〇.

KN=S-3old ------■ Then P, O
, PII step processing is executed.

When KN<SL at pH, Pl4 calculates the current S according to the following equation, and PII5 determines whether the learning condition is met.

S = -X S old + -X S new
・－－－－■! ! h However, 5old: Previous moving average value News If the current SP boar satisfies the learning conditions, use 's&' to learn the value of 100 as corresponding to the driving state at that time, and use the learning result as a predetermined value. After writing in the table map and updating the previous learning value, proceed to P.4. The format of the table map is the same as the & table map in the first embodiment. If the learning conditions are not met, jump to P11+ and proceed to P14.

In this example, S-8 is used as the knock parameter.
is adopted and compared with the slice level SL to determine the presence or absence of a knock. This method has the following advantages over the knock parameter S/H in the first embodiment.

As shown in FIG. 9, it can be assumed that the knock detection parameters have a substantially linear relationship with the knock level. On the other hand, since there are variations in S and N due to mechanical vibrations and differences between cylinders, variations may occur in this linear relationship. This variation affects knock detection accuracy. For example, when S/N is used as the knock parameter KN, variations in mechanical vibration with respect to S give variations in the position of the offset point A of the straight line AB in the figure. Further, variations in mechanical vibration with respect to N not only cause variations in the position of the offset point A of the straight line AB, but also cause variations in the slope of the straight line AB. In this case, N
Since there is no strong correlation in the mechanical vibration between and S,
It can be said that the S/N has a large variation due to mechanical vibration.

On the other hand, when S-8 is used as the knock parameter KN, since only S is used, the variation due to mechanical vibration is smaller than S/N. Also, S and S
Since the difference in S is calculated, it has the effect of absorbing variations in mechanical vibration with respect to S to some extent, and the variations are further reduced. Therefore, s-g has the advantage of higher knock detection accuracy than S/N.

(Effects) According to the present invention, non-knocking can be accurately detected regardless of the engine model, sensor mounting state, operating conditions, etc., and the knocking detection accuracy can be improved.

[Brief explanation of the drawing]

FIG. 1 is a basic conceptual diagram of the present invention, FIGS. 2 to 7 are a first embodiment of the present invention, FIG. 2 is a circuit configuration diagram thereof, and FIG. 3 (A
) is a sectional view showing the installed state of the cylinder pressure sensor, FIG. 3(B) is a plan view of only the cylinder pressure sensor, and the fourth [ff1
(A) to (H) are waveform diagrams showing the signal waveforms of each part.
Figure 6 is a flowchart showing the program for knocking control, Figure 6 is a flowchart showing the program for determining specific conditions, Figure 7 is a diagram showing the relationship between engine speed and slice level, and Figures 8 and 9 are diagrams showing the present invention. the second of
FIG. 8 is a flowchart showing the non-king control program, FIG. 9 is a diagram showing the relationship between knock level and parameter level, and FIG. 10 is a flowchart showing the non-king control program. A line diagram showing an example of the power spectrum, Figures 11 (A) and (B) are waveform diagrams of signals corresponding to crank angle to explain the technology for knocking and non-knocking, and Figure 13 shows engine speed and signal waveform diagrams. FIG. 14 is a diagram showing the relationship between the engine load and the integral value level. 1-- Cylinder pressure sensor (vibration detection means), 5-- band pass filter (extraction means), 17-- calculation means, 18-- control unit (specific condition determination means, smoothing means) , storage means, knock determination means, 19.---operation state detection means.

Claims (1)

[Scope of Claims] a) vibration detection means for detecting engine combustion pressure vibration; b) extraction means for extracting predetermined knock vibration from the output of the vibration detection means; c) operation for detecting the operating state of the engine. d) specific condition determining means for determining whether or not the engine is operated under a predetermined specific condition; and e) non-knock and knock detection means based on the output of the extracting means for each combustion cycle of the engine. f) a smoothing means for smoothing at least one of the first and second physical quantities in a predetermined manner; and g) when the engine is running. storage means for learning the output of the smoothing means when it is operated outside of specific conditions as corresponding to the operating state at that time and storing the learned smoothing value; h) storage means for storing the smoothing learning value; When the engine is operated under specific conditions, the smoothed learning value of the first or second physical quantity is read out from the storage means, and this smoothed learning value is outputted from the smoothing means. A knocking detection device for an internal combustion engine, comprising: knock determination means that performs the above determination in place of .