JP2764495B2 - Knock detection device for internal combustion engine - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knocking detection device for an internal combustion engine, and more particularly to an improved technique for detecting knocking occurrence from a detection signal of engine vibration.

[0002]

2. Description of the Related Art In an internal combustion engine, when knocking of a predetermined level or more occurs, not only does the output decrease, but also
Since the impact on the intake / exhaust valve and the piston is adversely affected by the impact, there is a device equipped with an ignition timing control device that detects the knocking and corrects the ignition timing so that the knocking can be avoided promptly. 58-1
05036, etc.).

Incidentally, knocking as abnormal combustion is observed as acoustic vibration of an engine block. Vibration of an internal combustion engine includes many mechanical vibrations caused by piston heads, gears, valve seats, and the like. Therefore, conventionally, the orthogonal analysis such as Fourier transform or Walsh-Hadamard transform is used, the spectrum analysis of the acoustic vibration is performed as a projection on a periodic signal, and a knock-specific frequency component is extracted to detect the presence or absence of knocking. I was

[0004]

However, a sudden signal such as knocking has the following characteristics.
In a conversion such as a Fourier transform for analyzing a periodic signal such as a sine wave, there is a problem that high-quality time / frequency analysis cannot be performed. 1) The signal waveform and the differential coefficient are often discontinuous.

2) The signal duration is short and the signal-to-noise ratio (S / N ratio) is poor. 3) Statistical properties are non-stationary. 4) The uncertainty relationship between the time resolution and the frequency resolution is uniformly determined. That is, even if the periodic signal used for the orthogonal transformation mathematically constitutes a perfect orthogonal transformation, the spectrum is estimated from the result of observing the signal having the above characteristics over a finite time, and the change of the spectrum is estimated. Tracking every moment is
It was difficult in principle. In particular, when the engine is running at high speed, the spectrum is estimated from a very small number of samples of the vibration signal, and deterioration of the estimation accuracy is in principle unavoidable due to the uncertainty relationship between the time resolution and the frequency resolution. Was.

Here, as one of the techniques for time / frequency analysis, wavelet transform (applied mathematics VOL.
1 NO. 3 SEP. 1991 Iwanami Shoten). This is a method of expanding a signal into a function system called a wavelet and performing time / frequency analysis using expansion coefficients. The wavelet function system is
Since the function is localized on the time axis and the frequency axis, information on the time and frequency of the signal can be obtained from the expansion coefficient. The wavelet function system is obtained by scaling and time shifting the basic wavelet function. Due to the nature of this functional system, the time resolution of the wavelet transform does not show a constant time / frequency resolution as in the case of the Fourier transform, as shown in FIGS. This is a technique suitable for analyzing a signal (a signal containing a high-frequency component) suddenly generated by such a property.

In view of the above, the application of the above wavelet transform to the analysis of the acoustic vibration of the knocking has been considered, but the following problem has occurred. That is, in the case of the non-orthogonal wavelet transform as a method of realizing the wavelet transform, there is a method of directly executing a convolution operation for each wavelet function, but in this method, the number of multiplications increases and decreases according to the scale of the wavelet, and a large The scale function increases the number of multiplications. Therefore, high-speed processing is required as in the case of application to knocking vibration analysis,
It has been difficult to apply the wavelet transform to applications that have large restrictions on software and hardware.

The present invention has been made in view of the above problems, and realizes a high-speed wavelet transform process, which can be applied to knocking detection, and can therefore perform knocking vibration analysis with high quality. The purpose is to provide.

[0009]

SUMMARY OF THE INVENTION Therefore, a knocking detection device for an internal combustion engine according to the present invention is a knocking detection device for an internal combustion engine which detects occurrence of knocking by analyzing engine vibration by wavelet transform, and is shown in FIG. It is configured as follows. In FIG. 1, a vibration sensor is attached to an engine body and outputs an analog detection signal corresponding to an engine vibration level. An analog signal from the vibration sensor is converted into a digital signal by an A / D converter. You.

[0010] The converted digital signal is processed by a frequency sampling filter having the same impulse response as a predetermined basic wavelet function, and knocking detection means detects knocking based on the processing result of the frequency sampling filter. Here, the frequency sampling filter can be configured by vertically connecting a comb filter and a predetermined number of resonators connected in parallel.

Further, the number of delay units constituting the comb filter is increased or decreased by tap setting, a predetermined number of resonators connected in parallel are vertically connected to respective taps, and filtering based on different numbers of delay units is performed. It is preferable to configure so that they are performed in parallel. Further, it is preferable to use a Gabor function or a Laplacian-Gaussian function as the predetermined basic wavelet function.

[0012]

According to this configuration, the analog detection signal of the engine vibration is converted into a digital signal, and the digital signal is processed by the frequency sampling filter. here,
The wavelet function is a function localized on the time axis, and the expansion coefficient corresponding to the wavelet function system that has been time-shifted on the same scale must match the output of the acyclic filter (Finite Impulse Response filter). Therefore, the basic wavelet transform is executed by the above-described frequency sampling filter, which is one of the methods for realizing the acyclic filter.

The frequency sampling filter can be constructed by connecting a comb-shaped filter and a predetermined number of resonators connected in parallel vertically, and by setting taps on the comb-shaped filter, the number of delay units can be reduced. Thus, parallel processing of filtering based on different numbers of delay units (wavelet functions with different scales) can be performed by the tap setting.

[0014]

Embodiments of the present invention will be described below. In FIG. 2 showing a hardware configuration of an embodiment, a knock sensor (vibration sensor) 1 is attached to a cylinder block (main body) of an internal combustion engine (not shown), and an analog voltage having a waveform corresponding to engine vibration is provided by a built-in piezoelectric element. Output a signal.

An analog voltage signal output from the knock sensor 1 is converted by an A / D converter 2 into numerical data (digital signal) at a predetermined sampling frequency (for example, 50 KHz). Then, the numerical data is processed by the frequency sampling filter 3 (digital filter).

The processing result of the frequency sampling filter 3 is used to determine the ignition timing (ignition advance value) ADV from an ignition plug 6 provided facing the combustion chamber of the internal combustion engine, and to the ignition circuit 5
The control unit 4 is input to a control unit 4 with a built-in microcomputer for controlling based on an ignition signal sent to the microcomputer. The control unit 4 as a knocking detecting means determines whether knocking has occurred based on the processing result of the frequency sampling filter 3, When knocking occurs, the ignition timing ADV is retarded and the occurrence of knocking is quickly avoided.

Here, a detailed configuration of the frequency sampling filter 3 will be described. The frequency sampling filter 3 is, as shown in the conceptual diagram of FIG.
Resonator consisting of a predetermined number of basic resonators connected in parallel with 11
12 is a digital filter formed by connecting the twelfth and twelfth. The comb filter 11 is composed of N delay units 13 connected in series, two multipliers 14 and 15, and an adder 16, and the data delayed by the N delay units 13 is multiplied by a multiplier. 14
, The data bypassing the delay unit 13 and the output of the multiplier 14 are added by an adder 16, and the output of the adder 16 is further multiplied by a multiplier coefficient A by a multiplier 15. Output.

The resonator 12 includes two multipliers 17, 18
, An adder 19 and a delay unit 20, a plurality of basic resonators are connected in parallel, and the outputs of the multipliers 17 in each basic resonator are added to each other by adders 21a, 21b,. , The sum of the outputs of all the basic resonators is finally output. In the basic resonator, the multiplier 18 adds the output of the comb filter 11 and the output of the adder 19 to a predetermined multiplication coefficient exp (j · 2
π / N · k _i) was added by the adder 19 and the value delayed by the further delay device 20 multiplies, to output is multiplied by a predetermined multiplier coefficient H _i by the multiplier 17 to the output of the adder 19 It has become.
The setting of the multiplication coefficients of the multipliers 17 and 18 will be described later in detail.

In this embodiment, the digital signal of the engine vibration is analyzed by a discrete wavelet transform which is more suitable for analyzing a signal generated suddenly than an orthogonal transform based on a periodic function such as a Fourier transform. Then, the presence or absence of occurrence of knocking is detected. The discrete wavelet transform is a method of expanding a signal into a wavelet function system and performing time / frequency analysis of the signal by using expansion coefficients. The wavelet function is obtained by scale conversion and time shift of a basic wavelet function g (t). Can be

The wavelet function g _{a, b} (t) of scale a and time shift b is

[0021]

(Equation 1)

Is defined as Basic wavelet function g
For (t), a function localized on the time axis and the frequency axis is selected. Therefore, the wavelet function g _{a, b} (t) is also a function localized on the time axis and the frequency axis. 3 and 4
Shows the time / frequency distribution of the wavelet function g _{a, b} (t) when the scale is set to 1 and when the scale is contracted to α smaller than 1 as shown in FIG. Function scale expansion
It can be seen that frequency characteristics can be changed by compression.

According to the wavelet function g _{a, b} (t), the discrete wavelet transform for the signal s (t) is

[0024]

(Equation 2)

Is defined as Here, T indicates a sampling period, and S (a, b) indicates an expansion coefficient of the signal s (t) with respect to the wavelet function g _{a, b} (t). When the time resolution and the frequency resolution of the wavelet transform are compared with the Fourier transform, the results are as shown in FIGS. 5 and 6. In the wavelet transform, since the low-frequency component is observed over a long time, its frequency distribution can be understood in detail. You can,
On the other hand, since the high-frequency component is observed only for a short time, the frequency resolution is deteriorated, but the time resolution is improved.

Sudden signals, such as knocking, contain high frequency components due to the nature of their occurrence and, of course, have a short duration. In addition to the information on the frequency, information on its occurrence and duration is also important. Therefore, the wavelet transform is a time / frequency analysis method more suitable for analyzing a knocking signal than an orthogonal transform based on a periodic function such as a Fourier transform.

Here, the wavelet function shown in Equation 1 is a function localized on the time axis. Therefore, the expansion coefficient corresponding to the wavelet function system that has been time-shifted on the same scale is a non-recursive filter (Finite I
In this embodiment, as described above, the frequency sampling filter 3 is used as a method for realizing the acyclic filter.

The impulse response of the frequency sampling filter 3 is

[0029]

(Equation 3)

## EQU1 ## A delay unit number N of the comb filter, by determining appropriately the coefficient k _i of the resonator, the impulse response h and (nT) and a wavelet function g (nT),

[0031]

(Equation 4)

If the relationship is expressed by the following equation, the output y (nT) of the frequency sampling filter 3 is

[0033]

(Equation 5)

## EQU1 ## This is because the expansion coefficient S (1, n of the wavelet function of scale a = 1 and shift b = nT
T). Therefore, the multiplier coefficient of the multiplier 18 is set to exp
(J · 2π / N · k i) and then, also, the multiplication factor of the multiplier 17 by a H _i, are those that can achieve equal frequency sampling filter 3 of the basic wavelet function and the impulse response, this embodiment Then, a filter having such characteristics is referred to as a basic filter.

In the frequency sampling filter 3,
By changing the number N of delay units of the comb filter 11 and the multiplication coefficient A of the multiplier, different scales of the wavelet function can be realized. In the frequency sampling filter 3, with respect to the number N of delay units of the comb filter 11,

[0036]

(Equation 6)

In order to realize a wavelet transform at a different scale a _i having the relationship of, the N that determines the number of delay units of the comb filter is changed to N _i , and the multiplication coefficient A is

[0038]

(Equation 7)

May be changed to Further, as shown in the conceptual diagram of FIG. 7, taps are set in the middle of N series of delay units connected in series with one comb filter, and each tap is multiplied in FIG. vessel 15 after the construction of resonators each resonator R _i to R _k
, It is possible to perform the wavelet transform on a plurality of different scales a _{i to} a _k by using one comb filter in common. Note that N _i to N _k indicating the number of delay units are set in accordance with the respective scales according to the relationship of Equation (6).

According to the frequency sampling filter 3 described above, the number of times of multiplication executed between one sample at all scales (a _{i to} a _k ) is equal, and the processing is faster than the case where the convolution operation is directly executed by equation (2). Becomes possible. Furthermore, in the case of configuring as hardware, since the configuration is simple, LSI layout can be easily performed.
Further, since the parallelism is high, parallel processing by a plurality of signal processors can be easily realized.

Here, a specific embodiment applied experimentally as knocking detection will be described. First, the Gabor function was selected for the basic wavelet function g (t). This Gabo
The r function is known as a function with the smallest spread on the time / frequency axis,

[0042]

(Equation 8)

In this embodiment, T = √
A Gabor function with 2/10, ω ₀ = 12π / 128 T was used. This function is approximated by an impulse response of a basic filter in which the number N of delay units of the comb filter is 128 and the number of basic resonators constituting the resonator 12 connected in parallel is 7.

Therefore, the Gabor function is subjected to a Fourier transform by applying a square window of 128 points, seven frequencies having the largest Fourier coefficients are selected, and coefficients k _i (i = 0 _{. .} 6) was determined. Further, the coefficient H _i (i = 0 _... 6) corresponding to the coefficient k _i is made to coincide with the Fourier coefficient, and a basic wavelet function approximated by a filter is obtained as shown in FIG.

The scale a _{i of the} basic filter is

[0046]

(Equation 9)

Wavelets at 50 scales from i = −30 to 20 are performed by tapping the comb filter as described above. However, on the scale shown in Equation 9, the number of delay units _Ni is not an integer due to Equation 6. Therefore,

[0048]

[Equation 10]

The number of delay units was approximated by an integer. Here, Q [•] means rounding of a numerical value below a decimal point. In the above setting, FIG. 9 shows the frequency characteristics of the basic wavelet function with the scale a = 1 and the wavelet function with the minimum scale (a = 0.125), and the wavelet function with the basic wavelet function and the maximum scale (a = 4). FIG. 10 shows the frequency characteristics of the function. As is clear from these figures, the frequency resolution of the wavelet transform is high in the low frequency range and low in the high frequency range.

By the way, as described above, the wavelet transform is performed using the 50 wavelet functions of different scales. When such a wavelet transform is realized by direct convolution, the number of complex multiplications per sample is 8,147. . On the other hand, according to the embodiment using the frequency sampling filter described above (an embodiment in which the number of delay units is increased or decreased by tapping the comb filter), the number of complex multiplications is 750, and the direct convolution operation is performed. Wavelet transformation can be performed with 1/10 or less of the number of times of execution. Therefore, the analysis by the wavelet transform can be performed at a high speed, and therefore, the application to the knocking detection requiring the high-speed processing can be performed.By analyzing the knocking detection by using the wavelet transform, even at the time of high rotation, A high-quality knocking detection device capable of detecting knocking occurrence with high time resolution can be realized.

Next, specific results of the wavelet transform during abnormal combustion (when knocking occurs) and during normal combustion are shown in FIG. 11 and FIG. FIGS. 11 (a), (b) and (c) show the case where the vibration of the internal combustion engine during typical normal combustion is observed,
FIG. 11 (a) shows the vibration waveform, FIG. 11 (b) shows the amplitude of the wavelet transform, and FIG. 11 (c) shows the phase of the wavelet transform.

On the other hand, FIGS. 12 (a), (b) and (c) show cases where abnormal combustion (knocking) has occurred, and FIG. FIG. 12B shows the amplitude of the wavelet transform, and FIG. 12C shows the phase of the wavelet transform. The target internal combustion engine is a V-type eight-cylinder engine for automobiles, and the engine is operated at a constant speed of 1200 rpm, and the ignition timing is forcibly advanced to generate abnormal combustion. Was.

FIG. 11B showing the amplitude of the wavelet transform.
In FIG. 12 (b), the amplitude is represented by a change in shading from black to white, where black indicates the minimum amplitude and white indicates the maximum amplitude. Similarly, FIG. 11 shows the phase of the wavelet transform.
12 (c) and FIG. 12 (c), the change from black to white is from 0 to 2π
Represents the phase change to. Note that the sampling period shown in FIGS. 11 and 12 is between 1.4 ms and 5.4 ms after the ignition of the air-fuel mixture.

When the amplitude of the wavelet is compared between the normal combustion and the abnormal combustion (when knocking occurs), it is found that the components having a scale of 2 or less increase in both time shifts from b = 120 to b = 160. I understand, of course,
Since this portion occurs at the same time regardless of the occurrence of abnormal combustion, it is predicted that the vibration is not caused by abnormal combustion but is mechanical vibration called slap noise.

On the other hand, between b = 50 and b = 100 samples, the scale is 2 ¹ only when abnormal combustion occurs.
Expansion coefficient of 2 ⁰ is increased from. The amplitude of the 2 ^-2 2 ^-3 of the scale is increased time within this period. FIG. 11 shows the phase of the wavelet transform.
According to (c) and FIG. 12 (c), when the phase of the expansion coefficient is compared between b = 50 and b = 100, when abnormal combustion occurs, the phase change from 0 to 2π is 2 ^0. from contrast and clearly appeared in 2 ¹ scale, not appear the changes in the case of normal combustion.

[0056] From the above comparison, the number of revolutions 1200 rpm, the vibration of the abnormal combustion (knocking) from scale 2 ¹ 2
It was found that detection was possible from 2.4 ms to 3.4 ms after the mixture was ignited by the expansion coefficient of the wavelet function from ⁰ to 2 ^-2 to 2 ^-3 . Therefore, if the above experiment is performed at a plurality of rotation speeds, it is possible to identify a time when vibration due to abnormal combustion appears in each rotation speed range, and use this as a determination condition when the engine is actually operated. Of abnormal combustion can be detected as a large change in the amplitude and phase of the wavelet at the specific time, separately from mechanical vibration. Therefore, if the format of the knocking detection is programmed in the control unit 4 in advance, it is possible to detect the presence or absence of the occurrence of knocking based on the amplitude and phase in the expansion coefficient of the wavelet transform output from the frequency sampling filter 3. .

In the above embodiment, the Gabor function is used as the basic wavelet function, but instead of this Gabor function, Laplacian-Gaussian is used.
an) A function may be used. FIG. 13 shows the Laplacian-Gaussian function used in the design of the basic filter.
14 shows a Laplacian-Gaussian function as a basic wavelet function approximated by a basic filter.

FIGS. 15 and 16 show the amplitude of the wavelet transform during normal combustion and abnormal combustion (when knocking occurs) when the Laplacian-Gaussian function is used as a basic wavelet function.
As is apparent from FIGS. 15 and 16, the Laplacian-
When the Gaussian function is used as the basic wavelet function, the frequency resolution is low, and the part where the amplitude of the wavelet transform is characteristically large due to the occurrence of abnormal combustion is less clear than when the Gabor function is used as the basic wavelet function. However, the change in the amplitude characteristic accompanying the abnormal combustion occurs sharply, and the time resolution is improved. For this reason,
There is an advantage that the time change of the amplitude of the knock component can be clearly read.

[0059]

As described above, according to the present invention, the time and the time suitable for analysis of a sudden signal such as knocking vibration are obtained.
Wavelet transform, which is a frequency analysis method, can be processed at a higher speed than when direct convolution operation is performed, and it is particularly applicable to knocking detection that requires high-speed processing. There is an effect that knocking can be detected with high quality even at the time of high rotation.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a basic configuration of the present invention.

FIG. 2 is a system schematic diagram showing a hardware configuration of an embodiment of the present invention.

FIG. 3 is a diagram showing a time / frequency distribution of a wavelet function.

FIG. 4 is a diagram showing a time / frequency distribution of a wavelet function.

FIG. 5 is a diagram showing time / frequency resolution in wavelet transform.

FIG. 6 is a diagram showing time / frequency resolution in Fourier transform.

FIG. 7 is a conceptual diagram showing a filter for realizing a wavelet transform on a plurality of different scales.

FIG. 8 is a diagram showing a basic wavelet function approximated by a basic filter.

FIG. 9 is a diagram showing frequency characteristics of a minimum-scale wavelet function.

FIG. 10 is a diagram showing frequency characteristics of a maximum-scale wavelet function.

FIG. 11 is a diagram illustrating wavelet transform characteristics during normal combustion.

FIG. 12 is a diagram illustrating wavelet transform characteristics during abnormal combustion (when knocking occurs).

FIG. 13 is a diagram illustrating a Laplacian-Gaussian function used for setting a basic filter.

FIG. 14 is a diagram illustrating a Laplacian-Gaussian function as a basic wavelet function approximated by a basic filter.

FIG. 15 is a diagram illustrating amplitude characteristics during normal combustion of a wavelet transform using a Laplacian-Gaussian function.

FIG. 16 is a diagram illustrating amplitude characteristics of wavelet transform using a Laplacian-Gaussian function during abnormal combustion (when knocking occurs).

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Knock sensor 2 A / D converter 3 Frequency sampling filter 4 Control unit 5 Ignition circuit 6 Ignition plug 11 Comb filter 12 Resonator

──────────────────────────────────────────────────続 き Continuation of front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G01H 17/00 G01M 15/00

Claims (5)

(57) [Claims] 1. A knocking detection device for an internal combustion engine which detects occurrence of knocking by analyzing engine vibration by wavelet transform, comprising: a vibration sensor attached to an engine body for outputting an analog detection signal corresponding to the engine vibration level. An A / D converter for converting an analog detection signal from the vibration sensor into a digital signal; and a digital filter for processing the digital signal converted by the A / D converter, comprising a predetermined basic wavelet function and an impulse. A knocking detection device for an internal combustion engine, comprising: a frequency sampling filter having the same response; and knocking detection means for detecting occurrence of knocking based on a processing result of the frequency sampling filter. 2. The knocking detection device for an internal combustion engine according to claim 1, wherein said frequency sampling filter is constituted by connecting a comb filter and a predetermined number of resonators connected in parallel in a vertical direction. . 3. A method in which the number of delay devices constituting the comb filter is increased or decreased by tap setting, a predetermined number of resonators connected in parallel are vertically connected to respective taps, and filtering based on different numbers of delay devices is performed. 3. The knocking detection device for an internal combustion engine according to claim 2, wherein the knocking detection device is configured to perform the knocking in parallel. 4. The method according to claim 1, wherein the predetermined basic wavelet function is G
4. The knocking detection device for an internal combustion engine according to claim 1, wherein the knocking detection device is an abor function. 5. The knocking detection device for an internal combustion engine according to claim 1, wherein said predetermined basic wavelet function is a Laplacian-Gaussian function.