JP4869366B2 - Internal combustion engine knock detection device - Google Patents

Description

  The present invention relates to a knock detection device for an internal combustion engine, and more particularly to a device including a plurality of knock sensors.

  Patent Document 1 discloses a knocking detection device using a plurality of knock sensors. In this apparatus, an input circuit including a low-pass filter, an AD conversion circuit, and the like is provided in parallel corresponding to each of the plurality of knock sensors.

  Patent Document 2 discloses a signal processing circuit that processes output signals of two knock sensors. In this circuit, the output signals of the two knock sensors are input to the multiplexer, and the signal selected by the multiplexer is input to the amplifier.

JP-A-6-194250 JP 2007-88572 A

  The device disclosed in Patent Document 1 requires the same number of input circuits as the number of knock sensors, and the device disclosed in Patent Document 2 requires a multiplexer. That is, when the number of knock sensors is two or more, there is a problem that the configuration of the input circuit becomes complicated.

  The present invention has been made in consideration of this point, and an object of the present invention is to provide a knocking detection device that further simplifies the configuration of an input circuit when a plurality of knock sensors are used.

In order to achieve the above object, the invention according to claim 1 is directed to knock detection of an internal combustion engine comprising knock determination means for determining knock based on output signals of a plurality of knock sensors (11a, 11b) mounted on the internal combustion engine. The apparatus includes a single input circuit to which output signals of the plurality of knock sensors (11a, 11b) are input, and the knocking determination means determines knocking using the output signal (VS) of the input circuit. The plurality of knock sensors (11a, 11b) have a configuration equivalently represented by a parallel circuit of a resistor and a capacitor, and a pair of output terminals (each of the plurality of knock sensors (11a, 11b)) ( 111-114) are connected to a pair of input terminals (PI1, PI2) of the input circuit, and one of the pair of input terminals (PI1) is It is connected to the power supply via the anti (R1), the other of said pair of input terminals (PI2) is connected to ground, the knocking determining means, the output signal of the input circuit at predetermined crank angle intervals (VS) Frequency component analysis means for performing frequency component analysis, data storage means for storing the intensity of a plurality of frequency components obtained by the frequency component analysis as time series data (KMAP (j, i)), and the frequency component intensity Noise component calculating means for calculating time component data (NLMAP (j, i)) of the noise component based on the time-series data (KMAP (j, i)), and time-series data (KMAP ( j, i)) with time series data (NLMAP (j, i)) of the noise component, and time series data (JKM) of the frequency component intensity corrected by the noise correction means. And a PI (j, i)) binarizing means for binarizing the, you determine whether or not knocking has occurred based on time series data binarized (JKMAP (j, i)) It is characterized by that.

The invention according to claim 2, in device for detecting knocking in an internal combustion engine according to claim 1, wherein the engine has a plurality of cylinders ing from the first cylinder group and the second cylinder group, wherein the plurality of knock The sensors (11a, 11b) are provided corresponding to the first cylinder group and the second cylinder group.

According to the first aspect of the invention, the output signals of a plurality of knock sensors are input to a single input circuit, and knocking is determined. The plurality of knock sensors have a configuration equivalently represented by a parallel circuit of a resistor and a capacitor, and one set of output terminals of each of the plurality of knock sensors is connected to one set of input terminals of the input circuit. In addition, since one of the pair of input terminals is connected to the power supply through a resistor and the other of the pair of input terminals is connected to the ground, it is not necessary to provide an input circuit corresponding to each of the plurality of knock sensors. It is not necessary to provide a multiplexer. Therefore, the configuration of the input circuit can be simplified. In addition, frequency component analysis of the knock sensor output signal is performed at predetermined crank angle intervals, the intensity distribution of a plurality of frequency components obtained by this frequency component analysis is stored as time series data, and further, time series data of frequency component strength Based on the above, the time-series data of the noise component is calculated and the time-series data of the frequency component intensity is corrected by the time-series data of the noise component. Then, the corrected time series data is binarized, and knocking determination is performed based on the binarized time series data. One knock sensor output signal may be mixed with another knock sensor output signal and become a noise component due to the input of the output signals of multiple knock sensors to a single input circuit. It is possible to remove more noise components and perform more accurate determination. In addition, the binarized time series data shows changes in the intensity distribution as the engine rotates. By comparing this time series data with data corresponding to a specific change pattern when knocking occurs, knocking is performed. Further, the binarization reduces the data amount and simplifies the change pattern of the time series data, thereby reducing the memory capacity and increasing the calculation speed.

  According to the second aspect of the present invention, two knock sensors are provided corresponding to the first cylinder group and the second cylinder group, and knocking determination is performed using the output signals. Corresponding knocking determination can be performed.

It is a figure which shows the structure of the internal combustion engine and its control apparatus concerning one Embodiment of this invention. It is a figure which shows the attachment position of two knock sensors (11a, 11b), and the structure of a knock sensor signal input part. It is a wave form diagram which shows the output signal waveform of two knock sensors (11a, 11b). It is a figure for demonstrating sampling of a knock sensor output, and frequency component analysis. It is the figure which showed the intensity | strength data obtained by a frequency component analysis as a spectrum time series map and a binarization spectrum time series map. It is a figure which shows the binarization spectrum time series map of the seating noise of an intake valve. It is a figure which shows a binarization spectrum time series map in order to demonstrate a noise removal process. It is a figure which shows the binarization spectrum time series map in order to demonstrate the knocking determination using a master pattern map. It is a figure which shows an example of a weighting map. It is a flowchart of a knocking determination process. It is a figure which shows the map referred by the process of FIG. It is a flowchart of the binarization data map calculation process performed by the process shown in FIG. It is a flowchart of the binarization process performed by the process shown in FIG. It is a figure which shows the map referred by the process of FIG. It is a flowchart of the noise removal process performed by the process shown in FIG. It is a flowchart of the precision calculation process performed by the process shown in FIG. It is a figure for demonstrating the map referred by the process of FIG. It is a flowchart of the noise learning process performed by the process shown in FIG. It is a flowchart of the noise map update process performed by the process shown in FIG. It is a flowchart of the knocking determination process concerning the 2nd Embodiment of this invention. It is the figure which showed the intensity | strength data obtained by a frequency component analysis as a spectrum time series map and the noise component intensity | strength as a spectrum time series map. It is a figure which shows the map which binarized the spectrum time series map after the noise removal process, and the spectrum time series map. It is a flowchart of the data map calculation process performed by the process of FIG. It is a flowchart of the noise removal process performed by the process of FIG. It is a flowchart of the binarization process performed by the process of FIG. It is a flowchart of the noise learning process performed by the process of FIG. It is a figure for demonstrating the subject of the conventional knock determination method.

Embodiments of the present invention will be described below with reference to the drawings.
[First Embodiment]
FIG. 1 is an overall configuration diagram of an internal combustion engine (hereinafter referred to as an “engine”) and a control device thereof according to an embodiment of the present invention. 3 is arranged. A throttle valve opening sensor 4 for detecting the throttle valve opening TH is connected to the throttle valve 3, and a detection signal of the sensor 4 is supplied to an electronic control unit (hereinafter referred to as “ECU”) 5.

  The fuel injection valve 6 is provided for each cylinder between the engine 1 and the throttle valve 3 and slightly upstream of the intake valve (not shown) of the intake pipe 2, and each injection valve is connected to a fuel pump (not shown). At the same time, it is electrically connected to the ECU 5 and the valve opening time of the fuel injection valve 6 is controlled by a signal from the ECU 5. Each cylinder of the engine 1 is provided with a spark plug 7, and the spark plug 7 is connected to the ECU 5. The ECU 5 supplies an ignition signal to the spark plug 7.

  An intake pressure sensor 8 for detecting the intake pressure PBA and an intake air temperature sensor 9 for detecting the intake air temperature TA are provided on the downstream side of the throttle valve 3. A cooling water temperature sensor 10 for detecting the engine cooling water temperature TW and two non-resonant knock sensors 11a and 11b are mounted on the main body of the engine 1. Detection signals of the sensors 8 to 11b are supplied to the ECU 5. As the first and second knock sensors 11a and 11b, for example, sensors capable of detecting vibrations in a frequency band from 5 kHz to 25 kHz are used.

An intake air flow rate sensor 13 for detecting the intake air flow rate GA is provided on the upstream side of the throttle valve 3 in the intake pipe 2, and the detection signal is supplied to the ECU 5.
A crank angle position sensor 12 that detects a rotation angle of a crankshaft (not shown) of the engine 1 is connected to the ECU 5, and a signal corresponding to the rotation angle of the crankshaft is supplied to the ECU 5. The crank angle position sensor 12 is a cylinder discrimination sensor that outputs a pulse (hereinafter referred to as “CYL pulse”) at a predetermined crank angle position of a specific cylinder of the engine 1, and relates to a top dead center (TDC) at the start of the intake stroke of each cylinder. A TDC sensor that outputs a TDC pulse at a crank angle position before a predetermined crank angle (every 180 degrees of crank angle in a four-cylinder engine) and one pulse (hereinafter referred to as “CRK”) with a constant crank angle cycle shorter than the TDC pulse (for example, a cycle of 6 °). The CYL pulse, the TDC pulse, and the CRK pulse are supplied to the ECU 5. These pulses are used for various timing controls such as fuel injection timing and ignition timing, and detection of engine speed (engine speed) NE.

  The engine 1 includes a first valve operating characteristic variable mechanism that continuously changes a valve lift amount and an opening angle (valve opening period) of an intake valve (not shown), and a crankshaft rotation angle of a cam that drives the intake valve. Is provided with a variable valve operation characteristic device 20 having a second valve operation characteristic variable mechanism that continuously changes the operation phase with reference to. The ECU 5 supplies the lift amount control signal and the operation phase control signal to the valve operation characteristic variable device 20, and controls the operation of the intake valve. The configurations of the first and second valve operating characteristic variable mechanisms are shown in, for example, Japanese Patent Application Laid-Open No. 2008-25418 and Japanese Patent Application Laid-Open No. 2000-227013.

  The ECU 5 shapes input signal waveforms from various sensors, corrects the voltage level to a predetermined level, converts an analog signal value into a digital signal value, etc., and a central processing unit (hereinafter referred to as “CPU”). A storage circuit (memory) for storing various calculation programs executed by the CPU, calculation results, and the like, an output circuit for supplying a drive signal to the fuel injection valve 6 and the spark plug 7, and the like.

  The engine 1 has # 1 to # 6 cylinders arranged as shown in FIG. 2A, and the first knock sensor 11a is provided corresponding to the first cylinder group consisting of # 1 to # 3 cylinders. The second knock sensor 11b is provided corresponding to the second cylinder group consisting of # 4 to # 6 cylinders.

  FIG. 2B is a circuit diagram illustrating a configuration of a knock sensor signal input unit included in the input circuit of the ECU 5, and a connection between the knock sensor signal input unit and the first and second knock sensors 11a and 11b. In this figure, knock sensors 11a and 11b are shown as equivalent circuits each consisting of a capacitor and a resistor. The first knock sensor 11a has a pair of output terminals 111 and 112, and the output terminals 111 and 112 are connected to the input terminals PI1 and PI2, respectively. The second knock sensor 11b has a pair of output terminals 113 and 114, and the output terminals 113 and 114 are also connected to the input terminals PI1 and PI2, respectively. That is, in this embodiment, the two knock sensors 11a and 11b are connected to the input terminals PI1 and PI2 in parallel.

  Input terminal PI2 is connected to ground. The input terminal PI1 is connected to the ground via the capacitor C1, is connected to the power supply VCC via the resistor R1, and is connected to the input terminal 122 of the amplifier circuit 121 via a series circuit of the resistor R2 and the capacitor C2. Yes.

  The amplifier circuit 121 includes a differential amplifier 123, a diode D1, resistors R3 to R6, and a capacitor C3. The diode D1 is connected between the input terminal 122 and the ground. The non-inverting input of the differential amplifier 123 is connected to the power supply VCC through the resistor R3 and is connected to the ground through the resistor R4. The inverting input of the differential amplifier 123 is connected to the input terminal 122 via the resistor R5, and a parallel circuit of the resistor R6 and the capacitor C3 is connected between the output of the differential amplifier 123 and the input terminal 122. The amplifier circuit 121 amplifies the input signal while attenuating frequency components higher than 25 kHz. The output signal of the amplifier circuit 121 is input to the subsequent A / D conversion circuit as the sensor output signal VS.

  FIG. 3 is a waveform diagram showing an example of a signal waveform of the sensor output signal VS. The cylinder numbers shown in FIG. 3 indicate cylinders in the combustion stroke, and “CAS” indicates an angle period during which sampling is performed for knocking determination described later. The sensor output signal VS is a signal obtained by adding the output signals of the first and second knock sensors 11a and 11b. Therefore, the output signal of the second knock sensor 11b is superimposed as noise on the detection signals corresponding to the cylinders # 1 to # 3 detected mainly by the first knock sensor 11a, and is mainly detected by the second knock sensor 11b. On the detection signals corresponding to the # 4 to # 6 cylinders, the output signal of the first knock sensor 11a is superimposed as noise.

  In this embodiment, as shown in FIG. 2B, the output terminals of the two knock sensors 11a and 11b are connected to a pair of input terminals PI1 and PI2, respectively, and the amplifier circuit 121 and its peripheral circuit are combined into one knock sensor. Since it is configured with a corresponding one, the configuration of the input circuit can be simplified. And it becomes possible to ensure determination accuracy by removing the influence of the output signal of the other knock sensor to be superposed by noise removal processing described later.

  Next, an outline of the knocking determination method in the present embodiment will be described. 4A shows the output signal waveform of the knock sensor 11, and FIG. 4B is an enlarged view of the waveform of the period TS in FIG. 4A. In the present embodiment, frequency component analysis by fast Fourier transform is performed on 50 data detected at a sampling period of 20 macroseconds. The result of the frequency component analysis is shown in FIG. The vertical axis of FIG. 4C is the energy intensity PS. In this embodiment, the energy intensity corresponding to the frequency (5, 6, 7,..., 24, 25 kHz) for each 1 kHz in the frequency band from 5 kHz to 25 kHz. PS is calculated every 6 degrees of crank angle.

  As a result, the time-series data of the frequency components is calculated as a two-dimensional matrix (hereinafter referred to as “spectrum time-series map”) as shown in FIG. In the spectrum time series map, the vertical direction is the frequency f [kHz], and the horizontal direction is the crank angle (crank angle after the top dead center at which the combustion stroke starts) CA [deg]. Here, “j” (j = 0 to 20) and “i” (i = 0 to 14) are used as the vertical and horizontal indexes, respectively, and the elements of the spectral time series map are used as the intensity parameter KMAP (j, i ) Is displayed. For example, the intensity parameter KMAP (0,0) is “34” at the lower left corner, the intensity parameter KMAP (0,14) is “31” at the lower right edge, and the intensity parameter KMAP (20,0) is “56” at the upper left edge. The intensity parameter KMAP (20, 14) corresponds to “30” in the upper right corner. The intensity parameter KMAP is a dimensionless quantity indicating the relative intensity.

  When the spectrum time series map of FIG. 5A is binarized, the binarized spectrum time series map of FIG. 5B is obtained. The binarized intensity parameter NKMAP (j, i) which is an element of the map of FIG. 5B is “1” or less than “50” when the intensity parameter KMAP (j, i) is “50” or more. Sometimes it takes “0”.

  The binarized spectrum time series map shown in FIG. 5B is a map obtained when knocking occurs as indicated by a broken line in FIG. On the other hand, the binarized spectrum time series map of the seating noise shown by the solid line in FIG. 27 is as shown in FIG. 6 and is clearly different from the map at the time of occurrence of knocking shown in FIG. Accurate knocking determination can be performed without erroneous determination of occurrence.

  In the present embodiment, a binarized spectrum time series map corresponding to noise (hereinafter simply referred to as “noise map”) is further based on the binarized spectrum time series map when it is determined that knocking has not occurred. The influence of noise is removed by subtracting the noise learning value generated by the learning calculation and the noise learning value on the noise map from the corresponding value on the binarized spectrum time series map calculated from the detection data.

  FIG. 7A is a diagram showing an example of a noise map obtained by learning calculation, and noise components appear in a band of 6 to 25 kHz at a timing of a crank angle of 6 degrees. FIG. 7B shows the binarized spectrum time series after the noise removal process is performed by subtracting the noise component shown in FIG. 7A from the binarized spectrum time series map shown in FIG. 5B. Show the map. By performing the noise removal process in this way, more accurate determination can be performed without the influence of noise.

  In the present embodiment, whether or not knocking has occurred is determined by comparing the binarized spectrum time series map obtained from the detection data with the master pattern map shown in FIG. . The master pattern map corresponds to a typical binary spectrum time series map obtained when knocking occurs.

  Specifically, the comparison is performed as follows. By integrating the product of the binarized intensity parameter NKMAP (j, i) on the binarized spectrum time series map and the master parameter MMAP (j, i) on the master pattern map, the integrated intensity value SUMAP is calculated. Then, a reference integrated value SUMM, which is an integrated value of the master parameter MMAP (j, i) itself, is calculated, and a precision PFIT (= SUMK / SUMM) is calculated as a ratio of the intensity integrated value SUMK to the reference integrated value SUMM. Then, when the relevance ratio PFIT exceeds the determination threshold value SLVL, it is determined that knocking has occurred.

  FIG. 8B shows the product (KMAP (j, i) of the intensity parameter KMAP (j, i) shown in FIG. 7B and the master parameter MMAP (j, i) on the master pattern map shown in FIG. j, i) × MMAP (j, i)). In this example, the product map of FIG. 8B is exactly the same as the original map shown in FIG. 7B, and the precision PFIT is “0.863”.

  In the actual knocking determination process to be described later, weighting is performed according to the engine operating state when calculating the intensity integrated value SUMK and the reference integrated value SUMM. An example of a weighting map in which weighting parameters WMAP (j, i) for performing this weighting are set is shown in FIG. The weighting map shown in FIG. 9 includes a parameter value from a crank angle of 12 degrees to 72 degrees for a component at a frequency of 6 kHz, a parameter value from a crank angle of 36 degrees to 66 degrees for a component at a frequency of 10 kHz, and a frequency of 11 kHz, 13 kHz, 15 kHz, For the 20 kHz component, weights are set for the parameter values of the crank angle from 12 degrees to 30 degrees.

  The weighting by the weighting map as shown in FIG. 9 is performed in order to compensate for the change of the characteristic with respect to the frequency of the binarized spectrum time series map depending on the engine operating state. By performing weighting according to the engine operating state, it is possible to make an accurate determination regardless of the change in the engine operating state.

  The sampling period of knock sensor output signal VS in this embodiment is 20 microseconds, and frequency component analysis is performed using 50 sampling values. Thereby, frequency components up to 25 kHz can be calculated at 1 kHz intervals. The time TD6 required for the crankshaft to rotate 6 degrees is 1 millisecond when the engine speed NE is 1000 rpm, and 50 sampling values are obtained. For example, when the engine speed NE is 2000 rpm, the time TD6 is 500 microseconds and the number of sampling data is 25. Therefore, the frequency component analysis uses the sampling value for the crank angle of 12 degrees detected immediately before. And every 6 degrees of crank angle.

  The frequency component analysis (Fast Fourier Transform) is performed in a process (not shown), and the intensity value STFT (k) (k = 0 to (KN−1)) for each frequency obtained as a result of calculation is obtained every 6 degrees of crank angle. It is stored in a storage circuit (RAM) of the ECU 5. KN is the number of intensity values obtained in the crank angle range from 6 degrees to 90 degrees after top dead center, and is 315 (the number of data on the map shown in FIGS. 5 to 8) in this embodiment.

  FIG. 10 is a flowchart of the process for performing the knocking determination by the above-described method, and this process is executed by the CPU of the ECU 5 in synchronization with the generation of the TDC pulse.

  In step S11, the binarized data map calculation process shown in FIG. 12 is executed, and the above-described binarized spectrum time series map is calculated. In step S12, the noise removal process shown in FIG. 15 is executed, and a process of removing noise components is performed using a binarized noise map as illustrated in FIG.

  In step S13, the relevance ratio calculation process shown in FIG. 16 is executed, and the relevance ratio PFIT is calculated using the binarized spectrum time series map from which the noise component has been removed and the master pattern map.

  In step S14, an SLVL map set as shown in FIG. 11, for example, according to the engine speed NE and the intake pressure PBA is searched, and a determination threshold SLVL is calculated. For the engine speed NE and the intake pressure PBA other than the grid points set in the SLVL map, the determination threshold SLVL is calculated by interpolation calculation.

  In step S15, it is determined whether or not the relevance ratio PFIT calculated in step S13 is larger than the determination threshold value SLVL. If the answer is affirmative (YES), it is determined that knocking has occurred, and the knocking flag FKNOCK is set. “1” is set (step S16).

  If PFIT ≦ SLVL in step S15, it is determined that knocking has not occurred, and the knocking flag FKNOCK is set to “0” (step S17). Next, the noise learning process shown in FIG. 18 is executed, and the binarized noise map (see FIG. 7A) is updated.

FIG. 12 is a flowchart of the binarized data map calculation process executed in step S11 of FIG.
In step S21, the crank angle index i, the frequency index j, and the memory address index k are all initialized to “0”. In step S22, it is determined whether or not the frequency index j is larger than a value obtained by subtracting “1” from the frequency data number JN (21 in the present embodiment). Initially, the answer is negative (NO), so the process proceeds to step S23 to determine whether or not the crank angle index i is larger than the value obtained by subtracting “1” from the crank angle data number IN (15 in this embodiment). To do.

Initially, the answer to step S23 is also negative (NO), so the memory address index k is calculated by the following equation (1) (step S24). Therefore, the memory address index k changes from 0 to (JN × IN−1) as the crank angle index i and the frequency index j increase.
k = i + j × IN (1)

  In step S25, the intensity parameter KMAP (j, i) is set to the intensity value STFT [k] stored in the memory, and in step S26, the binarization process shown in FIG. 13 is executed. Next, the crank angle index i is incremented by “1” (step S27).

  In step S31 of FIG. 13, the BLVL map shown in FIG. 14 is searched according to the engine speed NE and the intake pressure PBA, and the binarization threshold BLVL is calculated. The BLVL map is set so that the binarization threshold BLVL increases as the engine speed NE increases, and the binarization threshold BLVL increases as the intake pressure PBA increases. The set value indicated by the line L1 is applied in the range from the first predetermined intake pressure PBA1 (for example, 53 kPa (400 mmHg)) to the second predetermined intake pressure PBA2 (for example, 80 kPa (600 mmHg)). Lines L2 and L3 correspond to a third predetermined intake pressure PBA3 (for example, 93 kPa (700 mmHg)) and a fourth predetermined intake pressure PBA4 (for example, 107 kPa (800 mmHg)), respectively.

  In step S32, it is determined whether or not the strength parameter KMAP (j, i) is greater than the binarization threshold BLVL. If the answer is affirmative (YES), the binarization strength parameter NKMAP (j, i) Is set to "1" (step S33). On the other hand, if KMAP (j, i) ≦ BLVL in step S32, the binarization strength parameter NKMAP (j, i) is set to “0” (step S34).

  Returning to FIG. 12, while the answer to step S23 is negative (NO), steps S24 to S27 are repeatedly executed. When the crank angle index i exceeds (IN-1), the process proceeds to step S28, where the crank angle index i is increased. To “0”, the frequency index j is incremented by “1”, and the process returns to step S22.

  While the answer to step S22 is negative (NO), steps S23 to S28 are repeatedly executed, and when the frequency index j exceeds (JN-1), this process ends.

  The intensity value STFT [k] obtained as a result of the frequency component analysis stored in the memory by the processing of FIG. 12 is converted into the spectrum time-series map format shown in FIG. The parameter KMAP (j, i) is binarized, and the binarization strength parameter NKMAP (j, i) is calculated. That is, the binarized spectrum time series map shown in FIG. 5B is generated.

FIG. 15 is a flowchart of the noise removal process executed in step S12 of FIG.
In step S41, both the crank angle index i and the frequency index j are initialized to “0”. In step S42, it is determined whether or not the frequency index j is larger than a value obtained by subtracting “1” from the frequency data number JN. Since this answer is negative (NO) at first, the process proceeds to step S43, and it is determined whether or not the crank angle index i is larger than a value obtained by subtracting “1” from the crank angle data number IN.

Since the answer to step S43 is negative (NO) at first, the binarized intensity parameter NKMAP (j, i) is corrected by the following equation (2) to calculate the corrected binarized intensity parameter JKMAP (j, i). To do. NNMAP (j, i) in Expression (2) is a binarization noise parameter on the binarization noise map updated by the learning process.
JKMAP (j, i) = NKMAP (j, i) −NNMAP (j, i) (2)

  In step S45, it is determined whether or not the corrected binarization strength parameter JKMAP (j, i) is a negative value. If the answer is negative (NO), the process immediately proceeds to step S47. When JKMAP (j, i) <0, JKMAP (j, i) is set to “0” (step S46), and the process proceeds to step S47.

  In step S47, the crank angle index i is incremented by “1”, and the process returns to step S43. While the answer to step S43 is negative (NO), steps S44 to S47 are repeatedly executed. When the crank angle index i exceeds (IN-1), the process proceeds to step S48, and the crank angle index i is set to “0”. At the same time, the frequency index j is incremented by "1", and the process returns to step S42. While the answer to step S42 is negative (NO), steps S43 to S48 are repeatedly executed, and when the frequency index j exceeds (JN-1), this process ends.

  The corrected binarized intensity parameter JKMAP (j, i) from which noise has been removed is obtained by the process of FIG. 15 (see FIG. 7B).

FIG. 16 is a flowchart of the precision calculation processing executed in step S13 of FIG.
In step S51, a master pattern map is selected according to the engine speed NE and the intake pressure PBA. In step S52, a weighting map is selected according to the engine speed NE and the intake pressure PBA. The weighting map is provided in order to compensate for the change of the characteristic with respect to the frequency of the binarized spectrum time series map depending on the engine operating state. When the engine speed NE or the intake pressure PBA (engine load) changes, the temperature in the combustion chamber changes, and the binarized spectrum time series map changes. Therefore, by selecting the master pattern map and the weighting map according to the engine speed NE and the intake pressure PBA, it is possible to make an accurate determination regardless of changes in the engine operating state.

  In the present embodiment, as shown in FIG. 17, nine master pattern maps and nine weighting maps are set in advance corresponding to nine engine operation regions defined by the engine speed NE and the intake pressure PBA. In step S51, one of the nine master pattern maps is selected, and in step S52, one of the nine weighting maps is selected. In FIG. 17, for example, the low rotation region is a region where the engine speed NE is 2000 rpm or less, the middle rotation region is a region from 2000 rpm to 4000 rpm, and the high rotation region is a region exceeding 4000 rpm. The low load area is, for example, an area where the intake pressure PBA is 67 kPa (500 mmHg) or less, the medium load area is an area from 67 kPa to 93 kPa (700 mmHg), and the high load area is an area exceeding 93 kPa.

  In step S53, both the crank angle index i and the frequency index j are initialized to “0”, and the intensity integrated value SUMK and the reference integrated value SUMM are initialized to “0”. The intensity integrated value SUMK and the reference integrated value SUMM are updated in step S57, which will be described later, and are applied to the calculation of the relevance ratio PFIT in step S60.

  In step S54, it is determined whether or not the frequency index j is larger than the value obtained by subtracting “1” from the frequency data number JN. Since this answer is negative (NO) at first, the process proceeds to step S55, where it is determined whether or not the crank angle index i is larger than the value obtained by subtracting “1” from the crank angle data number IN.

Initially, the answer to step S55 is also negative (NO), so the process proceeds to step S56, and the weighting master parameter MMW and the weighting product parameter KMW are calculated by the following equations (3) and (4). WMAP (j, i) in the following formula is a weighting parameter set in the weighting map. The weighted product parameter KMW is obtained by weighting the product of the master parameter MMAP (j, i) and the corrected binarized intensity parameter JKMAP (j, i) by the weighting parameter WMAP (j, i).
MMW = MMAP (j, i) × WMAP (j, i) (3)
KMW = MMAP (j, i) × JKMAP (j, i) × WMAP (j, i) (4)

In step S57, the weighted master parameter MMW and the weighted product parameter KMW are integrated by the following formulas (5) and (6) to calculate the reference integrated value SUMM and the intensity integrated value SUMK.
SUMM = SUMM + MMW (5)
SUMK = SUMK + KMW (6)

In step S58, the crank angle index i is incremented by “1”, and the process returns to step S55. While the answer to step S55 is negative (NO), steps S56 to S58 are repeatedly executed. When the crank angle index i exceeds (IN-1), the process proceeds to step S59, and the crank angle index i is set to “0”. At the same time, the frequency index j is incremented by “1”, and the process returns to step S54. While the answer to step S54 is negative (NO), steps S55 to S59 are repeatedly executed. When the frequency index j exceeds (JN-1), the process proceeds to step S60, and the precision PFIT is calculated by the following equation (7). Is calculated.
PFIT = SUMK / SUMM (7)

FIG. 18 is a flowchart of the noise learning process executed in step S18 of FIG.
In step S71, the crank angle index i, the frequency index j, the addition learning parameter LK, and the subtraction learning parameter LM are all initialized to “0”. In step S72, it is determined whether or not the frequency index j is larger than a value obtained by subtracting “1” from the frequency data number JN. Since this answer is negative (NO) at first, the process proceeds to step S73, where it is determined whether or not the crank angle index i is larger than a value obtained by subtracting "1" from the crank angle data number IN.

  Initially, the answer to step S73 is also negative (NO), so the process proceeds to step S74 to determine whether or not the binarized intensity parameter NKMAP (j, i) is equal to the binarized noise parameter NNMAP (j, i). To do. If the answer is affirmative (YES), the process immediately proceeds to step S80.

  If the answer to step S74 is negative (NO) and the binarization intensity parameter NKMAP (j, i) is different from the binarization noise parameter NNMAP (j, i), the binarization intensity parameter NKMAP (j, i It is determined whether i) is larger than the binarized noise parameter NNMAP (j, i) (step S75). If the answer is affirmative (YES), the addition learning parameter LK is set to “1”, and the subtraction learning parameter LM is set to “0” (step S76). On the other hand, when NKMAP (j, i) <NNMAP (j, i), the addition learning parameter LK is set to “0” and the subtraction learning parameter LM is set to “1” (step S77).

In step S78, the addition learning parameter LK and the subtraction learning parameter LM are corrected by the following equations (8) and (9). DSNOISE in the equations (8) and (9) is a noise learning coefficient set to 0.1, for example.
LK = DSNOISE × LK (8)
LM = DSNOISE × LM (9)

In step S79, the addition learning parameter LK and the subtraction learning parameter LM are applied to the following equation (10) to update the noise parameter NMAP (j, i). The binarized noise parameter NNMAP (j, i) is calculated by binarizing the noise parameter NMAP (j, i) in the process of FIG.
NMAP (j, i) = NMAP (j, i) + LK−LM (10)

  In step S80, the crank angle index i is incremented by “1”, and the process returns to step S73. While the answer to step S73 is negative (NO), steps S74 to S80 are repeatedly executed. When the crank angle index i exceeds (IN-1), the process proceeds to step S81, and the crank angle index i is set to “0”. At the same time, the frequency index j is incremented by “1”, and the process returns to step S72. While the answer to step S72 is negative (NO), steps S73 to S81 are repeatedly executed. When the frequency index j exceeds (JN-1), the process proceeds to step S82, and the noise map update process shown in FIG. 19 is performed. Execute.

  In step S91 of FIG. 19, both the crank angle index i and the frequency index j are initialized to “0”. In step S92, it is determined whether or not the frequency index j is larger than a value obtained by subtracting “1” from the frequency data number JN. Since the answer to step S93 is negative (NO) at first, the process proceeds to step S93, where it is determined whether or not the crank angle index i is larger than a value obtained by subtracting “1” from the crank angle data number IN.

  Initially, the answer to step S93 is also negative (NO), so the process proceeds to step S94 to determine whether or not the noise parameter NMAP (j, i) is larger than a noise binarization threshold NLVL (for example, 0.8). If this answer is affirmative (YES), the binarized noise parameter NNMAP (j, i) is set to “1” (step S95). On the other hand, if NMAP (j, i) ≦ NLVL, the binarized noise parameter NNMAP (j, i) is set to “0” (step S96).

  In step S97, the crank angle index i is incremented by "1", and the process returns to step S93. While the answer to step S93 is negative (NO), steps S94 to S97 are repeatedly executed. When the crank angle index i exceeds (IN-1), the process proceeds to step S98, and the crank angle index i is set to “0”. At the same time, the frequency index j is incremented by "1", and the process returns to step S92. While the answer to step S92 is negative (NO), steps S93 to S98 are repeatedly executed, and when the frequency index j exceeds (JN-1), this process ends.

  If NKMAP (j, i)> NNMAP (j, i) in step S75 in FIG. 18, LK = 0.1 and LM = 0 in step S78, and the noise parameter NMAP (j, i, i) is incremented by "0.1". On the other hand, if NKMAP (j, i) <NNMAP (j, i) in step S75, LK = 0 and LM = 0.1 in step S78, and the noise parameter NMAP (j, i is determined by equation (10). ) Is decremented by "0.1". When the noise parameter NMAP (j, i) is larger than the noise binarization threshold NLVL in the processing of FIG. 19, the binarization noise parameter NNMAP (j, i) is set to “1”, while the noise parameter NMAP When (j, i) is equal to or less than the noise binarization threshold NLVL, the binarization noise parameter NNMAP (j, i) is set to “0”.

  The binarization noise map is updated according to the binarization intensity parameter NKMAP (j, i) when it is determined that knocking has not occurred by the processing of FIGS. 18 and 19, for example, the intake noise of the intake valve As shown in the figure, noise that occurs constantly is reflected in the binarized noise map. Further, since the output signals of the two knock sensors 11a and 11b are input to a single input circuit, one knock sensor output signal may be mixed with other knock sensor output signals and become a noise component. However, such a noise component is also reflected in the binarized noise map. As a result, it is possible to make a highly accurate determination without the influence of noise.

  Note that the precision PFIT may be calculated using the binarized intensity parameter NKMAP as it is without performing the noise removal process (FIG. 10, step S12). When noise removal processing is not performed, there is a high possibility of being affected by noise. However, as described with reference to FIG. 27, it is possible to determine knocking by distinguishing from noise.

  As described above in detail, in the present embodiment, the fast Fourier transform operation (frequency component analysis) of the output signal of the knock sensor 11 is performed at an interval of 6 degrees of the crank angle, and the intensity of the frequency component of 5 kHz to 25 kHz obtained as a result. A spectral time series map which is time series data of is generated. That is, the elements of the spectrum time series map are stored in the memory as intensity parameters KMAP (j, i) that are two-dimensional array data. Then, by binarizing the intensity parameter KMAP (j, i), the binarized intensity parameter NKMAP (j, i) is calculated, and knocking is performed based on the binarized intensity parameter NKMAP (j, i). It is determined whether it has occurred. Since the binarization intensity parameter NKMAP (j, i) reflects the change in frequency component distribution accompanying the engine rotation, the binarization intensity parameter NKMAP (j, i) and the specific change pattern when knocking occurs The occurrence of knocking can be accurately determined by comparing the master parameter MMAP (j, i) on the master pattern map corresponding to. Also, by binarizing the intensity parameter KMAP (j, i) obtained as a result of frequency component analysis, the amount of data is reduced and the change pattern of the time-series data is simplified. Speed can be increased.

  Since the binarization strength parameter NKMAP (j, i) (JKMAP (j, i)) shows a change pattern close to the master parameter MMAP (j, i), there is a high possibility that knocking has occurred. By calculating a parameter indicating the similarity (correlation) between the binarized intensity parameter NKMAP (j, i) (JKMAP (j, i)) and the master parameter MMAP (j, i), an accurate determination can be made. It can be carried out.

  In the present embodiment, the relevance ratio PFIT is used as a parameter indicating this similarity (correlation), and it is determined that knocking has occurred when the relevance ratio PFIT exceeds the determination threshold value SLVL. By using the relevance ratio PFIT, the similarity (correlation) between the binarized strength parameter NKMAP (j, i) (JKMAP (j, i)) and the master parameter MMAP (j, i) is a relatively simple calculation. Therefore, it is possible to accurately evaluate and make an accurate determination.

  Further, based on the binarized intensity parameter NKMAP (j, i), time series data NNMAP (j, i) of the noise component is calculated, and the binarized intensity parameter NKMAP (j, i) is time series data of the noise component. Is corrected by the noise parameter NNMAP (j, i), and knocking determination is performed based on the corrected binarized intensity parameter JKMAP (j, i). Therefore, it becomes possible to perform more accurate determination by removing the influence of the noise component that regularly appears like the seating noise described above and the other knock sensor output signal that is superimposed.

  Also, the binarization strength parameter NKMAP (j, i) (JNKMAP (j, i)) and the master parameter MMAP (j, i) are multiplied by a weighting parameter WMAP (j, i) set according to the frequency. Then, the precision PFIT is calculated. Since the frequency component that increases when knocking occurs is known in advance, the determination accuracy can be improved by giving a large weight to a parameter corresponding to a frequency in the vicinity of the frequency.

  In the present embodiment, the ECU 5 constitutes a knock determination means, a frequency component analysis means, a data storage means, a binarization means, a noise component calculation means, and a noise correction means. Specifically, steps S12 to S18 in FIG. 10 correspond to knocking determination means, and step S11 (processing in FIG. 12) corresponds to data storage means and binarization means. Step S18 corresponds to noise component calculation means, and step S12 corresponds to noise correction means.

[Second Embodiment]
In the present embodiment, the knocking determination process (FIG. 10) in the first embodiment is changed to the process shown in FIG. Except for the points described below, the second embodiment is the same as the first embodiment.

  In the processing shown in FIG. 20, first, noise correction processing is performed on the intensity parameter KMAP before binarization, the correction intensity parameter JKMAPI is calculated (step S12a), the correction intensity parameter JKMAPI is binarized, and correction is performed. A binarized intensity parameter JKMAP is calculated (step S11b).

  FIG. 21A shows an example of a spectrum time-series map in the present embodiment, and FIG. 21B is a diagram showing an example of a noise map in the present embodiment, and a noise learning value that is not binarized. Is set. The noise removal process is performed by subtracting the noise map value of FIG. 21B from each map value of the spectrum time series map of FIG. 21A, and the corrected spectrum time series map shown in FIG. can get. Then, by binarizing the corrected spectrum time series map of FIG. 22A, a corrected binarized intensity parameter map shown in FIG. 22B is obtained.

  FIG. 23 is a flowchart of the data map calculation process executed in step S11a of FIG. In this process, step S26 of the binarized data map calculation process shown in FIG. 12 is deleted. That is, only the process of converting the intensity value STFT (k) into a spectral time-series map format is performed.

  FIG. 24 is a flowchart of the noise removal process executed in step S12a of FIG. In this process, steps S44 to S46 of the process shown in FIG. 15 are changed to steps S44a to S46a, respectively.

In step S44a, the correction strength parameter JKMAPI is calculated by the following equation (2a). NLMAP (j, i) in Expression (2a) is a noise parameter on the noise map (FIG. 21B) updated by the learning process.
JKMAPI (j, i) = KMAP (j, i) −NLMAP (j, i) (2a)

  In step S45a, it is determined whether or not the correction strength parameter JKMAPI (j, i) is a negative value. If the answer to step S45a is negative (NO), the process immediately proceeds to step S47. If JKMAPI (j, i) <0, JKMAPI (j, i) is set to “0” (step S46a), and the process proceeds to step S47.

  The corrected spectrum time series map shown in FIG. 22A is obtained by the processing of FIG.

  FIG. 25 is a flowchart of the binarization process executed in step S11b of FIG. In this process, steps S32 to S34 in the binarization process shown in FIG. 13 are changed to steps S32a to 34a, respectively, and steps S41 to S43, S47 and S48 for binarizing all map values (FIG. 24). Step for performing the same processing as the step shown in FIG.

  In step S32a, it is determined whether or not the corrected intensity parameter JKMAPI (j, i) is greater than the binarization threshold BLVL. If the answer is affirmative (YES), the corrected binarization intensity parameter JKMAP (j, i i) is set to "1" (step S33a). On the other hand, if JKMAPI (j, i) ≦ BLVL in step S32a, the corrected binarization strength parameter JKMAP (j, i) is set to “0” (step S34a).

  With the processing in FIG. 25, the corrected binarized intensity parameter map shown in FIG. 22B is obtained from the map shown in FIG.

  FIG. 26 is a flowchart of the noise learning process executed in step S18a of FIG. In this process, steps S74 to S78 and S82 in the process shown in FIG. 18 are deleted, and step S79 is changed to step S79a.

In step S79a, the noise parameter NLMAP (j, i) is updated. Specifically, the intensity parameter KMAP (j, i) obtained when knocking is not detected is applied to the following equation (31) to update the noise parameter NLMAP (j, i). The noise parameter NLMAP (j, i) on the right side is a map value before update, and DSNOISEa is an annealing coefficient set to “0.5”, for example.
NLMAP (j, i) = NLMAP (j, i) × DSNOSEa
+ KMAP (j, i) × (1-DSNOSEa) (31)

The smoothing coefficient DSNOISEa in the equation (31) may be changed in the range of 0 to 1 according to the engine operating state as described below.
1) When the engine speed NE suddenly increases (for example, the difference (NE (m) -NE (m-1)) between the current value NE (m) of the engine speed and the previous value NE (m-1). If it is 200 rpm or more), the smoothing coefficient DSNOISEa is changed to a smaller value. Thereby, the weight of the current value KMAP (j, i) of the intensity parameter is increased, and the learning speed of the noise parameter NLMAP (j, i) is increased. The “m” is a discretization time discretized in the execution cycle of the process of FIG.

  2) When the valve opening timing CAI of the intake valve changes rapidly (for example, when the absolute value of the difference between the current value CAI (m) and the previous value CAI (m-1) of the valve opening timing is 5 degrees or more) Changes the annealing coefficient DSNOISEa to a smaller value.

  3) When the lift amount LFT of the intake valve changes abruptly (for example, when the absolute value of the difference between the current value LFT (m) of the lift amount and the previous value LFT (m-1) is 0.5 mm or more) The annealing coefficient DSNOISEa is changed to a smaller value.

  According to the present embodiment, since the map value (NLMAP) of the noise map is updated using the intensity parameter KMAP before binarization, the noise parameter NLMAP () with higher accuracy than the first embodiment is used. j, i) is obtained. Although the output signals of the two knock sensors 11a and 11b are input to a single input circuit, one knock sensor output signal may be mixed with another knock sensor output signal and become a noise component. It is possible to perform more accurate determination by removing such noise components.

  In this embodiment, step S11a in FIG. 20 corresponds to data storage means, step S12a corresponds to noise correction means, step S18a corresponds to noise component calculation means, and step S11b corresponds to binarization means. Steps S13 to S17 correspond to knocking determination means.

  The present invention is not limited to the embodiment described above, and various modifications can be made. For example, the sampling period of the knock sensor output and the crank angle interval for performing frequency component analysis are not limited to those described above (20 microseconds, 6 degrees), and can be changed within a range in which the object of the present invention is achieved. . Further, the binarized spectrum time series map (configured by a matrix of 21 rows × 15 columns in the above-described embodiment) can be similarly changed.

  In the embodiment described above, the relevance ratio PFIT is calculated by multiplying the binarization strength parameter NKMAP (j, i) and the master parameter MMAP (j, i) by the weighting parameter WMAP (j, i). The calculation may be performed without multiplying the parameter WMAP (j, i), that is, without weighting.

  In the above-described embodiment, the determination threshold SLVL, the binarization threshold BLVL, the master pattern map, and the weighting map are calculated or selected according to the engine speed NE and the intake pressure PBA. It may be fixed to a value or one map.

  The present invention can also be applied to knock detection of a marine vessel propulsion engine such as an outboard motor having a crankshaft as a vertical direction.

1 Internal combustion engine 5 Electronic control unit (knocking determination means, frequency component analysis means, data storage means, binarization means, noise component calculation means)
11 Knock sensor

Claims (2)

In a knock detection device for an internal combustion engine comprising knock determination means for determining knock based on the output signals of a plurality of knock sensors mounted on the internal combustion engine,
A single input circuit to which output signals of the plurality of knock sensors are input;
The knocking determining means determines knocking using an output signal of the input circuit,
The plurality of knock sensors have a configuration equivalently represented by a parallel circuit of a resistor and a capacitor, and a pair of output terminals of each of the plurality of knock sensors is connected to a pair of input terminals of the input circuit. , One of the pair of input terminals is connected to a power source via a resistor, and the other of the pair of input terminals is connected to a ground ,
The knocking determination means includes
Frequency component analysis means for performing frequency component analysis of the output signal of the input circuit at a predetermined crank angle interval;
Data storage means for storing the intensity of a plurality of frequency components obtained by the frequency component analysis as time-series data;
Noise component calculation means for calculating time series data of noise components based on the time series data of the frequency component intensity,
Noise correction means for correcting the time-series data of the frequency component intensity with the time-series data of the noise component;
Binarization means for binarizing the time series data of the frequency component intensity corrected by the noise correction means,
Device for detecting knocking in an internal combustion engine, characterized that you determine whether or not knocking has occurred based on time series data binarized.   The engine has a plurality of cylinders including a first cylinder group and a second cylinder group, and the plurality of knock sensors are provided corresponding to the first cylinder group and the second cylinder group. The knock detection device for an internal combustion engine according to claim 1.

Images