JP6433340B2 - Signal analysis device and knocking detection device - Google Patents

Description

  The present invention performs an operation analysis of an object based on an original signal obtained by picking up a physical quantity such as a vibration accompanying a movement of an object that performs a periodic operation and a sound generated due to the vibration. The present invention relates to a signal analysis device and a knocking detection device that detects engine knocking as an example of the signal analysis device of the present invention.

  Conventionally, in engine ignition timing adjustment tests and the like, detection of engine knocking has been performed in order to find an optimal ignition timing and the like within a range where knocking does not occur.

  As a knocking detection method for such a purpose, conventionally, a change in the in-cylinder pressure of the engine with respect to time is measured, and knocking is detected based on the change in the in-cylinder pressure (see, for example, Patent Document 1).

  However, in order to measure the in-cylinder pressure of the engine, it is necessary to process the engine and attach a sensor to the inside of the engine, and preparation for measurement is difficult. Further, by processing the engine or mounting a sensor in the engine, the characteristics of the engine may change, and the accuracy of ignition timing adjustment and the like may be reduced. Also, after adjusting the ignition timing in a test using a test engine, an actual vehicle test is performed, but it is not possible to attach such a sensor to measure the in-cylinder pressure to the actual vehicle engine. Whether or not knocking has occurred is currently determined by a skilled person by his / her own hearing, and finally knocking is determined by the hearing of the skilled person. Also, even in the test using the test engine, the expert determines whether knocking has occurred or not, but the actual vehicle test even under conditions where the engine characteristics have changed and the test engine has not been knocked However, it has been pointed out that knocking occurs and there is a deviation in the knocking conditions.

  As a sensor for detecting knocking of an engine that does not require processing the engine, a so-called knock sensor is known (for example, see Patent Documents 2 to 5). The knock sensor may be attached to the outer surface of the engine, and the sensor can be easily attached as compared with the measurement of the in-cylinder pressure. However, this knock sensor detects strong knocking that may damage the engine, and it detects subtle knocking such as detecting whether knocking has started while adjusting the ignition timing little by little. Is difficult.

  In addition, attempts have been made to pick up vibrations and sounds associated with engine combustion and detect engine knock based on the signals (see, for example, Patent Documents 6 to 9). If knocking can be detected with sufficient accuracy from vibrations and sounds associated with engine combustion, engine processing is unnecessary, and it is expected that the deviation from knocking judgment by an expert's hearing is also eliminated.

  However, knocking detection with sufficient accuracy based on vibrations and sounds associated with engine combustion is possible only in the low rotation speed region, and it is not enough to cover the high rotation speed required in the engine test.

  Thus, although detection of knocking based on vibration and sound has been attempted, since the rotational speed does not reach the entire rotational speed required in the engine test, it has not been sufficiently put into practical use. .

  Also, not only in engine knocking detection, but in a wide field when performing motion analysis of an object based on signals obtained by picking up vibrations and sounds resulting from motion of the object performing periodic motion, It is desired to perform highly accurate motion analysis.

JP 2003-314349 A JP-A-4-339157 Japanese Patent Laid-Open No. 5-26721 JP-A-2005-83314 JP 2005-188297 A Japanese Patent Laid-Open No. 4-152239 WO2010 / 013663 Publication JP 2012-103157 A JP 2014-29121 A

  In view of the above circumstances, the present invention is based on a physical quantity such as vibration and sound, a signal analysis device capable of performing a motion analysis of an object with high accuracy, and knocking over the entire rotation speed required for an engine test, An object of the present invention is to provide a knocking detection device capable of detecting with sufficient accuracy over a wide rotation speed region required for an engine test based on the physical quantity.

The signal analysis device of the present invention that achieves the above object is
An original signal acquisition unit for acquiring an original signal representing a change with respect to time of a value of a physical quantity picked up by a sensor that picks up a physical quantity whose value is changed due to vibration associated with the movement of an object that performs a periodic operation;
A non-periodic signal that extracts the non-periodic signal in which the non-periodic component of the original signal is enhanced by suppressing the periodic component included in the original signal by applying an adaptive line enhancer to the original signal acquired by the original signal acquisition unit An extractor;
An angle information acquisition unit for acquiring angle information including origin information representing at least an origin of a periodic operation angle of the object;
Based on the angle information acquired by the angle information acquisition unit, a signal cutout unit that cuts out a signal within the cutout angle range from the non-periodic signal,
And a motion analysis unit that performs motion analysis of the object based on the signal cut out by the signal cutout unit.

  When performing high-accuracy analysis based on physical quantities such as vibration and sound, separate the characteristic components of the physical quantities resulting from the characteristic motion of the object from components other than the characteristic components. It is desirable to perform analysis with emphasis. In the signal analysis apparatus of the present invention, the adaptive line spectrum enhancer is used to extract a non-periodic signal in which a component caused by a characteristic operation is emphasized from the original signal, and the operation analysis is performed based on the non-periodic signal. Done. Therefore, highly accurate analysis is possible.

  Here, in the signal analysis device of the present invention, the object may be an engine, and the motion analysis unit may analyze whether or not engine knocking has occurred.

  The signal analyzing apparatus of the present invention is suitable for performing engine knocking analysis.

Moreover, the knocking detection device of the present invention that achieves the above-described object is
An original signal acquisition unit for acquiring an original signal representing a change with time of a value of a physical quantity picked up by a sensor that picks up a physical quantity whose value is changed due to vibration caused by engine combustion;
A non-periodic signal that extracts the non-periodic signal in which the non-periodic component of the original signal is enhanced by suppressing the periodic component included in the original signal by applying an adaptive line enhancer to the original signal acquired by the original signal acquisition unit An extractor;
An angle information acquisition unit for acquiring angle information including at least origin position information of the rotation angle of the engine;
Based on the angle information acquired by the angle information acquisition unit, from a non-periodic signal, a signal cutout unit that repeatedly cuts out a signal within the cutout angle range over a plurality of cycles of engine combustion
A plurality of first bispectrums are calculated based on each of a plurality of first signals for a plurality of cycles within a first period in which knocking does not occur, which are repeatedly cut out by the signal cutout unit, and Calculating a first normalized value obtained by normalizing an average value of the plurality of first bispectrums;
A second bispectrum is calculated based on a second signal for one cycle within a second period that is likely to cause knocking, which is cut out by the signal cutout unit, and a plurality of first bispectrums are calculated. Calculating a second normalized value obtained by normalizing an average value of a plurality of bispectrums obtained by combining the bispectrum and the second bispectrum;
A difference normalized value calculation unit that calculates a difference normalized value including a difference between the second normalized value and the first normalized value;
An index value calculation unit that calculates an index value serving as an index of the engine based on the difference normalized value calculated by the difference normalized value calculation unit;
And an information presentation unit for presenting information derived from the index value calculated by the index value calculation unit.

  As a result of investigating the cause and phenomenon of knocking and trying various calculation methods that seem to be suitable for detection of knocking, the present inventors have completed the present invention. In the present invention, since the difference normalized value is calculated and an index value serving as an index for knocking is calculated based on the difference normalized value, knocking over a wide range of rotational speeds necessary for engine tests is performed. It can be detected with sufficient accuracy required for engine testing.

  According to the knock detection device of the present invention, the adaptive line spectrum calculator is applied to the original signal to extract the non-periodic signal in which the non-periodic component at the time of knocking is emphasized, and knocking is performed based on the non-periodic signal. Since detection is performed, knocking can be detected with higher accuracy.

In the knocking detection device of the present invention, the index value calculation unit may calculate the difference normalized value b_sub (f ₁ , f ₂ ) calculated by the difference normalized value calculation unit (where f ₁ and f ₂ are both frequencies). A characteristic amount representing a level including a secondary harmonic component is calculated from a value in the vicinity of f ₁ = f _{2, and} a frequency band in the vicinity of the natural frequency in at least one vibration mode of the engine is calculated. It is preferable that a representative value among the feature values is calculated, and a value derived from the representative value is used as an index value.

Knocking is a phenomenon in which a shock wave is generated by amplifying vibrations at the natural frequency (resonance frequency) of the engine due to abnormal combustion of the engine. This shock wave is accompanied by vibration at a frequency (frequency) that is twice the natural frequency (resonance frequency). When the physical quantity to be analyzed includes a second-order harmonic component composed of a certain frequency component and twice the frequency component, the second-order harmonic component is a line of bicoherence f ₁ = f ₂ . Appears as the above value. On the other hand, the noise component appears at a position deviating from the line of f ₁ = f ₂ . However, since the calculation involves an error, the feature amount is calculated from a value near f ₁ = f ₂ including the error. Here, the natural frequency (resonance frequency) of the engine can be known in advance by calculation, observation, or the like, but changes due to heat generated by engine combustion. For this reason, after calculating the feature value, the representative value is calculated based on the feature value in the frequency band including the fluctuation range of the natural frequency (resonance frequency). By using the value derived from the representative value thus calculated as an index value, knocking detection with higher accuracy can be performed.

Furthermore, in the knocking detection apparatus of the present invention, when calculating the representative value and using the value derived from the representative value as an index value, the index value calculation unit includes a plurality of first normalization calculation units. A plurality of first normalization values are calculated while sequentially replacing at least one first bispectrum of the bispectrums, and an average value μ and a standard deviation σ of the difference between the first normalization values are calculated. Calculate threshold value TR for determining whether knocking is present or not,
TR = α × σ + μ
Where α is calculated according to the coefficient,
It is preferable that the index value is calculated through a comparison operation for comparing the representative value and the threshold value.

  The above threshold value is a threshold value with which it is possible to determine whether or not knocking has occurred with a statistically significant probability. Therefore, whether or not knocking has occurred can be calculated as an index value having a statistically significant probability by performing a comparison operation that compares the representative value with the above threshold value.

  Further, the index value calculation unit further determines whether or not knocking has occurred for the specified number of cycles using the index value calculated through the comparison operation, and determines that knocking has occurred within the specified number of cycles. The knocking occurrence rate, which is the ratio of the number of cycles performed, is calculated, and the value of the average number of knocks per unit time calculated from the product of the knocking occurrence rate and the number of engine strokes per unit time from the rotational speed is calculated as a new value. More preferably, the index value is used.

  This new index value is an index value that has a high correlation with the knocking index in hearing.

  Here, in the knocking detection device of the present invention, it is preferable that the physical quantity to be detected is sound, and the sensor is a sensor that picks up sound emitted from the engine due to vibration accompanying engine combustion.

  By using sound as a detection target, knocking detection can be performed without contact with the engine.

  However, the knocking detection device of the present invention may be a sensor that detects engine vibration itself and the sensor picks up vibration itself accompanying engine combustion.

  Also in this case, it is only necessary to attach a sensor to the engine surface without machining the engine, and knocking detection can be easily performed.

  According to the present invention described above, the signal analysis device capable of performing the motion analysis of the object with higher accuracy, and knocking over the entire rotational speed required in the engine test can be performed with sufficient accuracy required in the engine test. A knocking detection device capable of detection is realized.

It is the figure which showed the outline | summary of the engine test system when the engine for a test which employ | adopted the knocking detection apparatus as one Embodiment of this invention was made into the test object. It is the figure which showed the outline | summary of the engine test system when the actual vehicle was made into the test object which employ | adopted the knocking detection apparatus as one Embodiment of this invention. It is a block diagram which shows the calculation content in the notebook PC shown in FIG. It is the block diagram which showed the adaptive line spectrum enhancer. It is the figure which showed an example of the input signal and output signal of an aperiodic signal extraction part. It is the figure which showed the flow of the process in the difference bicoherence calculation part shown in FIG. It is a conceptual diagram of the difference bicoherence calculation process in a difference bicoherence calculation part. It is the figure which showed an example of the calculation result of differential bicoherence. It is the block diagram which showed the calculation processing content in the index value calculation part shown in FIG. It is the figure which modeled and showed the feature-value calculation process. It is the figure which modeled and showed the representative value calculation process performed by the index value calculation part in FIG. It is explanatory drawing of the natural frequency (resonance frequency) of an engine. It is a figure showing knocking detection accuracy. It is the flowchart which showed the flow of the calculation in the preparatory stage at the time of measurement. It is the flowchart which showed the flow of the calculation at the time of measurement.

  Embodiments of the present invention will be described below. Here, a knocking analysis apparatus which is an example of the signal analysis apparatus of the present invention will be described.

  1 and 2 are diagrams showing an outline of an engine test system that employs a knocking detection device as one embodiment of the present invention.

  FIG. 1 shows an engine bench test for a test engine, and FIG. 2 shows an actual vehicle test for an actual vehicle.

  A test engine 1 shown in FIG. 1 is connected to a dynamo 2 and tests the engine under various operating conditions. An automobile 3 shown in FIG. 2 is an automobile equipped with an engine of the same type as the test engine shown in FIG. The automobile 3 is mounted on a test stand 4 for test operation. Various test conditions are reproduced on the test stand 4, and an operation test is performed under these various drive conditions.

  For example, the ignition timing adjustment test is performed with the test engine shown in FIG. 1. When the optimum ignition timing is determined, the automobile 3 is operated under the adjusted conditions, and whether or not the adjustment is good is confirmed. Done.

  Here, in the case of the engine bench test shown in FIG. 1 and the case of the actual vehicle test shown in FIG. 2, the sound sensor 5 is arranged near the engine 1. The sound sensor 5 is a sensor that picks up a sound caused by vibration accompanying combustion of the engine 1 and generates a sound pressure signal.

  The sound pressure signal picked up by the sound sensor 5 is input to the knocking detection device 10.

  An angle signal representing the rotation angle of the engine 1 is extracted from an ECU (Engine Control Unit) that controls the rotation of the engine 1.

  In the present embodiment, the angle signal includes a rotation pulse representing the origin of the rotation angle, one pulse per engine rotation, and a crank angle pulse, 720 pulses per engine rotation. In FIG. 2, the ECU 6 is shown separately from the automobile 2, but this is shown separately for easy understanding, and the ECU 6 in FIG. 2 is an element built in the automobile 3. .

  1 and 2, the sound pressure signal of the engine sound obtained by the sound sensor 5 and the angle signal from the ECU are input to the knocking detection device 10. As described above, the knocking detection device 10 of the present embodiment is commonly used for both the engine bench test (in the case of FIG. 1) and the actual vehicle test (in the case of FIG. 2).

  The knocking detection device 10 includes a data collection device 20 and a notebook personal computer (hereinafter simply referred to as “notebook PC”) 30. The data collection device 20 performs A / D conversion on the sound pressure signal from the sound sensor 5, temporarily stores it along with the angle signal, and then passes the data to the notebook PC 30. The notebook PC 30 receives data from the data collection device 20, executes calculation described later regarding knocking detection, and sends the calculation result to the data collection device 20. The data collection device 20 receives the calculation result from the notebook PC and sends it to the operation panel 40.

  The operation panel 40 is operated by the operator 50 during the engine bench test shown in FIG. 1 and adjusts the ignition timing and other various parameters of the engine 1. Various information is displayed on the monitor screen 40a of the operation panel 40 according to the parameter of the engine 1 to be adjusted. When adjusting the ignition timing of the engine 1, the calculation result regarding knocking sent from the knocking detection device 10 is displayed on the monitor screen 40a. The operator adjusts the ignition timing while viewing the display.

  In addition, the calculation result of the notebook PC 30 can be displayed on the display screen 30a of the notebook PC 30 itself.

  In the operation panel 40, the engine adjustment such as ignition timing adjustment is basically not performed during the actual vehicle test shown in FIG. In the operation panel 40, during the actual vehicle test, various information is displayed on the monitor screen 40a, and data on various states and changes of the engine are collected.

  FIG. 3 is a block diagram showing calculation contents in the notebook PC shown in FIG.

  The sound pressure signal acquisition unit 31 plays a role of acquiring a sound pressure signal picked up by the sound sensor 5 shown in FIGS. 1 and 2 and A / D converted by the data collection device 20. The sound pressure signal acquisition unit 31 is in charge of the data input port and the input interface circuit on the hardware of the notebook PC.

  Further, the angle signal acquisition unit 32 plays a role of acquiring an angle signal extracted from the ECU 6 shown in FIGS. 1 and 2 and passing through the data collection device 20. The angle signal acquisition unit 32 is also in charge of a data input port and an input interface in hardware.

  Here, the sound pressure signal acquisition unit 31 and the angle signal acquisition unit 32 correspond to examples of the original signal acquisition unit and the angle information acquisition unit according to the present invention. The sound pressure signal acquired by the sound pressure signal acquisition unit 31 corresponds to an example of the original signal referred to in the present invention.

  The non-periodic signal extraction unit 33 operates the adaptive line spectrum enhancer on the sound pressure signal acquired by the sound pressure signal acquisition unit 31, thereby suppressing the periodic component included in the sound pressure signal and Then, an aperiodic signal in which an aperiodic component including a sound pressure component due to vibration generated by knocking is emphasized is extracted.

  FIG. 4 is a block diagram showing an adaptive line spectrum enhancer.

In FIG. 4,
n is the sampling point number of the sound pressure signal obtained by the A / D conversion in the sound pressure signal acquisition unit 31,
d is a periodic component included in the sound pressure signal;
k is an aperiodic signal including sound pressure caused by knocking included in the sound pressure signal;
v is the observation noise,
r is an observation signal consisting of d + k + v,
u is a delayed signal of r,
y is the output signal of the adaptive filter,
e = k + v is a difference signal between the observation signal r and the output signal y of the adaptive filter.

  Here, in the adaptive filter constituting the adaptive line spectrum enhancer shown in FIG.

を初期値として、

Is the initial value,

の演算が繰り返し実行される。

This operation is repeatedly executed.

here,
w is the filter coefficient of the adaptive filter,
L + 1 is the number of taps of the adaptive filter,
η is the step gain,
β is the forgetting factor,
u is a delayed signal of r and is an input signal to the adaptive filter;
ε is the average power of the input signal u,
e is a difference signal between the observation signal r and the output signal y of the adaptive filter;
Because
w (n) and u (n) are respectively

で表わされる。

It is represented by

In the adaptive filter of the adaptive line spectrum enhancer shown in FIG. 4, by repeating the above calculation, as shown in FIG.
e (n) = k (n) + v (n) (5)
That is, a non-periodic signal obtained by removing the periodic signal d (n) from the observed signal r (n) is extracted.

  Here, knocking detection is performed based on the aperiodic signal thus extracted.

  The adaptive line spectrum enhancer itself is a conventionally known arithmetic unit, and further description of the adaptive line spectrum enhancer is omitted here.

  FIG. 5 is a diagram illustrating an example of an input signal and an output signal of the aperiodic signal extraction unit.

  Here, a pressure signal (A) representing an in-cylinder pressure change of the engine is also shown for easy understanding. The pressure signal (A) is a signal picked up by processing the engine and installing a pressure sensor in the cylinder.

  5B is an input signal to the non-periodic signal extraction unit 33 and is a sound pressure signal acquired by the sound pressure signal acquisition unit 31, and FIG. 5C is a non-periodic signal extraction unit. The non-periodic signal extracted from the non-periodic signal extraction unit 33.

  In FIG. 5, a time region T1 is a region where knocking does not occur, and a time region T2 is a region where occurrence of knocking is observed. The periodic component in the sound pressure signal shown in FIG. 5B is removed from the non-periodic signal in FIG. 5C, and a signal resulting from knocking is extracted.

  In FIG. 5A and FIGS. 5B and 5C, the time regions T1 and T2 are shown to be shifted in the time axis direction (horizontal axis direction). This is because a delay such as sound propagation time occurs between the timing at which the pressure in the cylinder is picked up by the pressure sensor and the timing at which the sound accompanying vibration due to the pressure change is picked up.

  The signal cutout unit 34 shown in FIG. 3 refers to the angle signal acquired by the angle signal acquisition unit 32, and from the aperiodic signal (FIG. 5C) extracted by the aperiodic signal extraction unit 33. A signal within a specific angle range (time regions T1 and T2 shown in FIG. 5) is cut out repeatedly every period.

  Here, the crank angle pulse is counted in response to the generation of the rotation speed pulse of one rotation, the angle range where knocking can occur is determined based on the counted value, and the non-periodic signal shown in FIG. Among these, the non-periodic signal in the determined angle range (here, time regions T1, T2,... Shown in FIG. 5) is cut out. Specifically, here, an aperiodic signal within an angle range of about 90 degrees from the ignition position and the vicinity of TDC, including the ignition position and TDC (Top Dead Center), is cut out.

  Here, in the present embodiment, the cut-out angle range remains fixed even when the ignition timing is changed. However, the angle range of cutting may be changed according to the change in the ignition timing.

  Further, in this embodiment, the cut-out angle range is determined by counting the crank angle pulses of 720 pulses per rotation. However, based on only the rotation pulses of one rotation and one pulse, the engine is determined from the interval between adjacent rotation pulses. A rotation speed (time required for one rotation) is obtained, and a specific time region in the time required for one rotation may be set as a cut-out angle range. Alternatively, the cut-out angle range may be determined based on angle origin information by a rotation pulse or the like and an angle signal having a resolution lower than that of the crank angle pulse.

  The aperiodic signal cut out by the signal cutout unit 34 is input to the differential bicoherence calculation unit 35, and differential bicoherence is calculated.

  FIG. 6 is a diagram showing a flow of processing in the differential bicoherence calculation unit shown in FIG.

  The calculation process in the differential bicoherence calculation unit 35 is divided into a calculation process in a preparation stage before measurement and a calculation process in measurement. The upper part of FIG. 6 is a process in a preparation stage before measurement, and the lower part is a process at the time of measurement. Here, “first” is added to the signals and calculated values in the preparatory stage before measurement, such as “first signal”, “first...”, And the signals and calculated values at the time of measurement are The "second signal", "second ...", and so on are used to distinguish each other. However, the contents common to both the preparatory stage before measurement and the time of measurement are described without adding “first” and “second”, for example, simply as “signal”.

  In the preparatory stage before the measurement, the first signal cut out for each cycle of the engine rotation by the signal cutout unit 34 (see FIG. 3) in the first period in which knocking has not occurred is sequentially calculated as the difference. The bicoherence calculation unit 35 inputs the FFT (Fast Fourier Transform) calculation 351.

Here, the aperiodic signal cut out by the signal cutout unit 34 is
x (t) (6)
It expresses by. Here, t represents time.

This non-periodic signal x (t) is subjected to FFT processing, and frequency data F (f) (7)
Is calculated.

Then, bispectrum B (f ₁ , f ₂ ) = F (f ₁ ) F (f ₂ ) based on frequency data F (f) obtained by subjecting the cut signal x (t) to FFT processing. F * (f ₁ + f ₂ ) (8)
Is calculated 352. Here, f ₁ and f ₂ are frequencies, and * is a complex conjugate.

In the preparatory stage before the measurement, the first signal x (t) is cut out for each period and the FFT calculation 351 is performed for (n−1) periods, and the first bispectrum Bi ( f ₁ , f ₂ ), i = 1, 2,..., n−1 is calculated 352. Here, n is a positive integer, and an appropriate value (for example, n = 30) is determined in advance by a preliminary experiment or the like. Alternatively, n may be changed variously, and appropriate n may be determined by calculation of differential bicoherence statistical value calculation / evaluation 354 described later.

  Once (n-1) periods, ie, (n-1) first bispectrum calculations 352 have been performed, then using these (n-1) first bispectrums, A first bicoherence calculation 353 is performed.

  Here, in calculating the first bicoherence, the original bicoherence calculation formula is the calculation formula of the formula (9) for normalizing the bispectrum.

であるが、複素数の計算のため、条件によってはバイコヒーレンスの値が０から１に入らない場合が発生する。そこでバイコヒーレンスの定義に基づき、値が０から１に入るように修正した式（１０）の算出式

However, because of the calculation of complex numbers, there may occur a case where the value of bicoherence does not fall from 0 to 1 depending on conditions. Therefore, based on the definition of bicoherence, the calculation formula of formula (10) modified so that the value falls from 0 to 1

を用いる。ここでは、式（１０）を用いて算出した値ｂ（ｆ₁，ｆ₂）もバイコヒーレンスと称する。式（９）あるいは式（１０）で算出されたバイコヒーレンスｂ（ｆ₁，ｆ₂）は、式（９）を採用した場合であっても、式（１０）を採用した場合であっても、本発明にいう正規化値（第１の正規化値および第２の正規化値）の一例に相当する。

Is used. Here, the value b (f ₁ , f ₂ ) calculated using the equation (10) is also referred to as bicoherence. The bicoherence b (f ₁ , f ₂ ) calculated by the equation (9) or the equation (10) may be the case where the equation (9) is adopted or the case where the equation (10) is adopted. This corresponds to an example of a normalization value (first normalization value and second normalization value) according to the present invention.

Equation (9) or equation (10) is to normalize with averaging a plurality of bispectral _{B i (f 1, f 2} ). Here, in this embodiment, the expression (10) is adopted because the value b (f ₁ , f ₂ ) exceeds “1” according to the expression (9) which is the original bicoherence expression. On the other hand, when the expression (10) is adopted, the value b (f ₁ , f ₂ ) is strictly a value in the range of “0 to 1”, and the subsequent calculation is easy. However, this does not preclude the use of equation (9) as the equation for calculating the normalized value in the present invention.

  Furthermore, in this preparation stage, after (n−1) first bispectrums and the first bicoherence calculated from them are calculated, new (n−1) first bispectrums are calculated. Each time a bispectrum is input, a first bicoherence is calculated. Among the obtained first bicoherences, the first bicoherence having the smallest component on the line of f1 = f2 of the bicoherence and the (n-1) first bicoherences that are the basis of the first bicoherence. Are stored as reference bicoherence and bispectrum, and are used for calculation as n-1 first bispectrums and first bicoherence calculated from the bispectrum at the time of measurement. In addition, the bispectrum that is the basis of a plurality of bicoherences not selected as a reference may be retained in order to be reused in the statistical process of the next process.

  Next, when a new first signal cut out from one period is input, FFT calculation 351 is also performed for the first signal in the new one period, and first bispectrum calculation 352 is performed. . Then, based on the total (n) bispectrums obtained by adding the newly calculated one bispectrum to the (n-1) first bispectrums held, a new first Bicoherence is calculated.

Further, the first differential bicoherence is calculated by subtracting the reference bicoherence selected in the previous stage for each frequency (f ₁ , f ₂ ) from the first bicoherence. This is repeated each time a new first signal cut out from a new period is input,
A large number (here, m) of first differential bicoherences are calculated.

  Here, a new first signal cut out from one cycle is used as input. However, in order to shorten the calculation time, a bispectrum that is the source of a plurality of bicoherences that were not selected as a reference in the previous stage is used as input. It may be reused and m first differential bicoherences may be calculated.

  After that, processing of statistical value calculation / evaluation 354 of differential bicoherence is performed.

  In this differential bicoherence statistical value calculation / evaluation 354, after sequentially calculating the differential bicoherence, for each of the plurality of differential bicoherences and for each of one or more vibration modes of interest of the engine, A representative value in the frequency band near the natural frequency of the vibration mode is calculated. The calculation of the representative value is an algorithm that is also performed at the time of measurement, and the description of the algorithm for calculating the representative value is left to the description at the time of measurement described later. When a representative value for each vibration mode is calculated for each differential bicoherence, an average value and a standard deviation of the representative values are calculated for each vibration mode and for each number of cycles. These average values and standard deviations are input to the index value calculation unit 36. Also, here, the difference value between the representative values calculated from the difference coherence used for calculating the average value and the standard deviation of the representative values is calculated, and the difference value is stabilized within a predetermined small value range. Therefore, it is evaluated that knocking has not occurred. Each differential bicoherence can be retained and used for each rotation speed (r / min).

  As for the value of n described above, various first bicoherences obtained by changing the value of n are calculated, and evaluation similar to the above is performed based on the various first bicoherences. The value of n may be determined by selecting the value of n where the standard deviation is most stable.

  Next, calculation processing at the time of measurement in the differential bicoherence calculation unit 35 will be described.

  The following processing is performed for the second period during which knocking may occur.

  Here, as in the preparation stage, the non-periodic signal cut out by the signal cutout unit 34 (here, referred to as “second signal” because of measurement) is subjected to the FFT calculation 355 to obtain the second frequency. Data F (f) is calculated. Next, a second bispectrum calculation 356 is performed based on the second frequency data F (f).

Furthermore, this time, in the preparation stage, the first bispectrum B _i , i = 1, 2 for (n−1) periods calculated within the first period during which it is known that knocking has not occurred. ,..., N−1 and the second bispectrum B_T for one period calculated this time are combined to perform a second bicoherence calculation 357 based on the bispectrum for a total of n periods. Further, a differential bicoherence calculation 358 is performed by subtracting the _first bicoherence calculated in the preparation stage for each frequency (f ₁ , f ₂ ) from the _second bicoherence.

  The differential bicoherence here is

ここで、ｂ＿ｓｕｂ（ｆ₁，ｆ₂）は、差分バイコヒーレンス
Ｂ_i，ｉ＝１，２，・・・，ｎ−１は、準備段階で算出しておいた（ｎ−１）周期分の第１のバイスペクトル
Ｂ＿Ｔは、計測時において今回算出された第２のバイスペクトル
ｂ（ｆ₁，ｆ₂）は、準備段階で算出しておいた第１のバイコヒーレンス
である。

Here, b_sub (f ₁ , f ₂ ) is differential bicoherence B _i , i = 1, 2,..., N−1 is calculated for (n−1) periods calculated in the preparation stage. The first bispectrum B_T is the second bispectrum b (f ₁ , f ₂ ) calculated this time at the time of measurement, and is the first bicoherence calculated in the preparation stage.

  Here, this expression (11) is an expression when the expression (10) is used in the calculation of the bispectrum. The first bicoherence and the second bicoherence may be calculated using equation (9), and the differential bicoherence thereof may be calculated.

  Even during this measurement, when a new second signal cut out from a new one period is input, the FFT calculation 355 is also performed on the new second signal in the new one period, and a new new signal is generated. A second bispectral calculation 356 is performed. This time, the first bispectrum for (n-1) periods calculated in the preparation stage is left as it is, and the first bispectrum for the (n-1) periods is used as the first bispectrum calculated last time. A new second bicoherence calculation 357 is performed based on a bi-spectrum for n periods obtained by adding the second bispectrum newly calculated this time instead of the second bispectrum. Further, the first bicoherence calculated in the preparation stage is subtracted from the newly calculated second bicoherence, and a new differential bicoherence calculation 358 is performed. This is repeatedly executed, and new differential bicoherence is sequentially calculated.

  FIG. 7 is a conceptual diagram of differential bicoherence calculation processing in the differential bicoherence calculation unit.

  Here, the differential bicoherence calculation process which is the core of the calculation in the differential bicoherence calculation unit 35 illustrated in FIG. 3 described with reference to FIG. 6 will be described.

  Here, in the preparatory stage before measurement, signal extraction, FFT processing, and bispectrum calculation are performed for each (n-1) period, and (n-1) bispectrums are calculated (I). . Further, a first bicoherence is calculated from the (n−1) bispectrums (II).

Here, bicoherence is a function expressed by variables of _two frequencies f ₁ and f ₂ , as expressed by the above formula (9) or formula (10). Therefore, in FIG. 7B and (D) described later, the horizontal axis is represented by frequency f ₁ and the vertical axis is represented by frequency f ₂ . This bicoherence is symmetric with respect to the diagonal 92 of f ₁ = f ₂ , and therefore, here, the bicoherence value is calculated only in the region of f ₁ ≧ f ₂ (hatched region).

  Further, at the time of measurement, a signal within the extraction range is extracted from the original signal for one period, and FFT processing and bispectrum calculation are performed (III). A second bicoherence is calculated based on a total of n bispectrums, which is a sum of (n-1) bispectrums of (n-1) periods calculated previously (IV). . Further, the first bicoherence calculated in the preparation stage is subtracted from the second bicoherence to calculate the differential bicoherence (V).

  FIG. 8 is a diagram illustrating an example of a calculation result of differential bicoherence.

FIG. 8 shows the differential bicoherence b_sub (f ₁ , f ₂ ) calculated by the above calculation for a signal obtained by superimposing a 1 kHz sine wave signal and a 2 kHz sine wave signal. Both the horizontal axis and the vertical axis are frequencies. For example, the horizontal axis is f ₁ and the vertical axis is f ₂ .

In FIG. 8, a large value a appears on the diagonal line of f ₁ = f ₂ and at positions f ₁ and f ₂ = 1 kHz. Thus, the differential bicoherence b_sub (f ₁ , f ₂ ) is a quadratic order composed of a certain frequency component (here, f) and twice the frequency component (2f) in the signal to be analyzed. when there is a change in the harmonic components, the diagonal of f ₁ = f _2, a large value at the position of f ₁ = f ₂ = f is the property that would appear.

  Knocking is a kind of resonance phenomenon of the engine. When knocking occurs, a shock wave is generated in the cylinder, and the vibration and sound of the engine includes a certain natural frequency (resonance frequency) component and its harmonics. Wave components (especially double frequency components) appear. Here, focusing on this point, differential bicoherence has been successfully adopted for the detection of knocking.

The differential bicoherence b_sub (f ₁ , f ₂ ) calculated by the differential bicoherence calculation unit 35 in FIG. 3 is input to the index value calculation unit 36.

  FIG. 9 is a block diagram showing the calculation processing contents in the index value calculation unit shown in FIG.

  In this index value calculation unit 36, threshold calculation 361 is performed in a preparatory stage before measurement. Other arithmetic processing is performed during measurement.

In the threshold calculation 361, the average value μ and the standard deviation σ of the representative values for each vibration mode of interest of the engine are received from the differential bicoherence calculation unit 35, and the threshold TR is
TR _m = α _m × σ _m + μ _m (12)
However,
TR _m is the threshold value of vibration mode m α _m is the coefficient set for each vibration mode m σ _m is the standard deviation of the representative value of vibration mode m in the state where knocking has not occurred μ _m is knocking It is calculated according to the average value of the representative values of the vibration mode m in the state where it is not.

  Further, the index value calculation unit 36 performs a feature amount calculation 362 and a representative value calculation 363 at the time of measurement. Here, the feature amount represents the amount of change of the secondary harmonic component at each frequency between the preparation stage and the measurement time, and the representative value is a representative feature amount for each vibration mode and for each number of cycles. It represents a value.

  FIG. 10 is a diagram showing a model of the feature amount calculation process.

As described above, when the differential bicoherence b_sub (f ₁ , f ₂ ) is calculated, the signal to be analyzed changes to a second harmonic component (frequency f and 2f) (from the first period in which knocking does not occur). When there is a change), a large value appears at a position of f ₁ = f ₂ = f on the diagonal line.

Therefore, in the example shown in FIG. 10A, data on the diagonal line f ₁ = f ₂ of the differential bicoherence b_sub (f ₁ , f ₂ ) is extracted, and the extracted data is at f ₁ = f ₂ = f. It is regarded as a feature quantity.

In FIG. 10B, the average value of the data in the 3 × 3 region centered on the point on the diagonal line f ₁ = f ₂ is calculated, and the average value is a feature when f ₁ = f ₂ = f. It is taken as a quantity.

Here, although it is a 3 × 3 region, the region of f ₁ > f ₂ with the diagonal line of f ₁ = f ₂ of the differential bicoherence b_sub (f ₁ , f ₂ ) as the symmetry axis due to the property of bicoherence And the same data appear in the region of f ₁ <f ₂ . Therefore, here, an average value of six data in the region of f1 ≧ f2 out of the 3 × 3 region is calculated, and the average value is set as a feature amount of the frequency f ₁ = f ₂ = f.

Theoretically, a numerical value representing the magnitude of the second harmonic component appears on the diagonal line of f ₁ = f ₂ , but it may also appear at a slightly deviated position due to a calculation error or the like. Therefore, in the example of FIG. 10B, the average value of the data in the 3 × 3 region is adopted as the feature amount. Although an average value is used here, the maximum value in the 3 × 3 region may be used as the feature amount at the frequency f ₁ = f ₂ = f at the center of the region. In addition, any amount representing the level of change in the secondary harmonic component can be adopted as the feature amount.

  Here, the series of processes from the extraction of the non-periodic signal to the calculation of the feature amount is repeatedly executed for each number of cycles of the engine.

  FIG. 11 is a diagram showing a model of representative value calculation processing executed by the index value calculation unit in FIG.

  FIG. 11A conceptually shows a number of differential bicoherence calculation results for each number of cycles over a large number of cycles.

In the differential bicoherence calculation unit 35 in FIG. 3, differential bicoherence b_sub (f ₁ , f ₂ ) is calculated for each number of cycles within a predetermined required cycle number range, and the index value calculation unit 36 The feature amount is calculated for each number of cycles. The feature quantity for each number of cycles is a function of the frequency f.

  FIG. 11B is a diagram in which feature quantities for each number of cycles are arranged side by side on a two-dimensional plane with the frequency and the number of cycles as variables. The vertical axis is the feature quantity.

  The plurality of rectangles extending in the frequency direction indicate that the feature amount in one row in the rectangle is a feature amount as a function of the frequency f calculated for one cycle number.

  FIG. 11C shows the same feature quantity distribution as FIG. 11B. However, FIG. 11C shows an ellipse extending in the cycle number direction for several frequencies. That is, here, attention is paid to the feature quantity with the number of cycles as a variable at some specific frequencies. These several specific frequencies correspond to the natural frequency (resonance frequency) of the engine.

  FIG. 12 is an explanatory diagram of the natural frequency (resonance frequency) of the engine.

  The natural frequency (resonance frequency) fp, q of the engine is

ただし、Ｃは音速
Ｂｒは、ボア径
ｐ，ｑは、振動モード
Ρp,qは、振動モード（ｐ，ｑ）における定数
を表わしている。

Where C is the speed of sound
Br is the bore diameter
p and q are vibration modes
Ρp, q represents a constant in the vibration mode (p, q).

As shown in FIG. 12, for example, when the vibration mode is the (1, 0) mode, the constant _{1,0 1,0} = 1.841, and the natural frequency (resonance frequency) when the bore diameter Br is 87 mm is f _1,0 = 6712 Hz. The same applies to other vibration modes (p, q). The “in-cylinder vibration mode” diagram illustrates how the gas in the bore vibrates.

  Here, the (3, 0) mode is shown in ascending order of the natural frequency (resonance frequency) fp, q. Analytically, there are higher order vibration modes. However, what is practically important in engine tests are vibration modes within the range shown here.

  Returning to FIG. 11, the description will be continued.

  A long ellipse shown in FIG. 11C encloses a feature quantity with the number of cycles as a variable in each vibration mode. The natural frequency (resonance frequency) of the engine is expressed by Equation (13), and the temperature in the engine changes depending on the operating condition of the engine, and the sound speed C changes accordingly. There are also various error factors such as calculation errors.

  Therefore, the frequency band of the search is determined for each vibration mode, the maximum value of the feature amount in the frequency band is searched for each number of cycles, and the searched maximum value is represented by the number of cycles of the vibration mode. Value. However, whether or not knocking of the vibration mode has occurred in the number of cycles, such as calculating the average value of a plurality of peak values of feature values in the frequency band instead of using the maximum value as a representative value. Any value may be used as a basis for calculating an index value as an index.

  FIG. 11D shows representative values with the number of cycles as a variable in each vibration mode (p, q). This representative value is a value in the range of 0 to 1, and the closer to 1, the stronger the knocking is occurring.

  In the index value calculation unit 36 in FIG. 3, the representative value calculation 363 with the number of cycles as a variable is performed as described above.

  In the index value calculation unit 36, after the representative value is calculated for each vibration mode m = (p, q) and for each number of cycles as described above, the comparison operation 364 and the index value calculation 365 are performed. .

In the comparison operation 364,
CA _{s, m} = b_sub _{s, m} -TR _m (14)
However,
CA _{s, m} is a difference value for each vibration mode m and for each cycle number s,
b_sub _{s, m} is a representative value for each vibration mode m and for each cycle number s,
TR _m is an average value of representative values in a state where knocking does not occur for each vibration mode m.

  In the index value calculation 365, using the difference value of the equation (14),

が算出される。

Is calculated.

here,
index _s is an index value for each cycle number s,
K _{s, m} is a coefficient that takes a value of “1” when CA _{s, m} is positive and “0” when it is negative.

  That is, here, for each number of cycles, an index value corresponding to the magnitude of the secondary harmonic component and the number of appearances in a plurality of vibration modes is calculated.

  In addition, using the calculated index value, the knocking occurrence rate, which is the ratio of occurrence of knocking in the specified number of cycles, and the product of the number of engine strokes per unit time from the rotational speed are calculated per unit time. The value of the average number of knocks can also be used as an index value. This index value is highly correlated with the knocking index in the auditory sense.

  The index value calculated in this way is input to the information presentation unit 37 in FIG.

  The information presenting unit 37 presents information derived from the index value calculated by the index value calculating unit 36, for example, a bar graph having a length or a color according to the size of the index value. Specifically, for example, image information representing such a bar graph is created and sent to the operation panel 40 shown in FIGS. On the operation panel 40, an image based on the transmitted image information is displayed on the monitor screen 40a and is used for observation by the operator 50. Or this information presentation part 37 displays an image also on the display screen 30a of the notebook PC 30 as needed.

  The information derived from the index value by the information presentation unit 37 is not limited to the graph as described above, and may be information that contributes to the occurrence of knocking by the operator 50 and the evaluation of the degree of knocking. Further, the information presentation method is not limited to display. For example, the index value is divided into multiple stages such as no knocking, weak knocking, moderate knocking, strong knocking, etc. by threshold processing, and the current stage of the knocking is displayed or informed by sound. May be. Furthermore, the index value does not need to be a value calculated by the equations (14) and (15), and may be a value that is a knocking index calculated based on the representative value.

  FIG. 13 is a diagram illustrating knocking detection accuracy.

  FIG. 13A is an explanatory diagram, and FIG. 13B is a comparative example, which is obtained by extracting the sound pressure signal (see FIG. 5B) before extracting the aperiodic signal without extracting the aperiodic signal. FIG. 13C is a diagram showing knocking detection accuracy when calculated based on a signal, and FIG. 13C is an example in which an aperiodic signal is extracted and the extracted aperiodic signal (see FIG. 5C) is cut out. It is the figure which showed the knocking detection precision when computing based on the signal obtained in this way.

  Here, knocking detection based on a pressure signal (see FIG. 5A) obtained by measuring the in-cylinder pressure of the engine is also performed.

  FIG. 13A shows the calculation of the match rate in a contingency table in order to analyze whether the predicted value matches the output value of the determination. At that time, when the result is pressure data and the prediction is the determination result of the sound data, positive and negative examples of the result and the prediction are 4 from TP, TN, FP, and FN from the initials of True, False, Positive, and Negative. Express as a condition.

であって、
Ｄ:テストデータ集合
Ｋｐ:圧力データでノックだと判断された集合
Ｎｐ:圧力データでノックではないと判断された集合
Ｋｓ:音データでノックだと判断された集合
Ｎｓ:音データでノックではないと判断された集合
とする。
ＴＰ、ＴＮ、ＦＰ、ＦＮの４条件に対して

Because
D: Test data set Kp: Set determined to be knocked by pressure data Np: Set determined to be not knocked by pressure data Ks: Set determined to be knocked by sound data Ns: Not knocked by sound data A set determined to be
For 4 conditions of TP, TN, FP, FN

を表わしている。

Represents.

  Comparing FIG. 13B and FIG. 13C, both TP and TN are higher in the example of FIG. 13C than in the comparative example of FIG. 13B. This indicates that the detection accuracy of knocking is improved by introducing the non-periodic signal extraction unit 33 shown in FIG.

  FIG. 14 and FIG. 15 are flowcharts showing the flow of operations described so far.

  Here, FIG. 14 is a flowchart showing a flow of calculation in the preparation stage at the time of measurement, and FIG. 15 is a flowchart showing a flow of calculation in the measurement.

  Here, the operating conditions of the engine are kept constant, and the processing shown in FIGS. 14 and 15 is executed under the constant operating conditions.

  First, the calculation in the preparatory stage before measurement shown in FIG. 14 will be described.

  The engine is operated at a certain rotation speed, and a sound pressure signal and an angle signal (crank angle pulse and rotation pulse) are acquired within a first period in which knocking is not expected (step) S101), an aperiodic signal is extracted by a filtering process using an adaptive line spectrum enhancer (see FIG. 4) (step S102). Then, a signal within the cut angle range determined based on the angle signal is cut out from the non-periodic signal (step S103).

  Next, the cut out aperiodic signal is subjected to FFT processing (step S104), and a first bispectrum (see equation (8)) is calculated (step S105). Here, steps S101 to S105 are repeated (n-1) times (step S106), and the first bicoherence (see equation (10)) is calculated based on the obtained bispectrum (step S107).

  The first bicoherence is calculated m times sequentially using the first newly cut out first signal (step S108), and it is evaluated whether or not knocking has occurred (step S109). ). When it is evaluated that knocking has not occurred (step S110), a threshold value TR (see equation (12)) is calculated (step S111), and the calculation process before measurement is terminated.

  As a result of the evaluation in step S109, when it cannot be determined that knocking has not occurred (step S110), the process returns to step S101, and the processes in steps S101 to S109 are repeated.

  Next, calculation processing at the time of measurement will be described with reference to FIG.

  At the time of measurement, the engine is operated at one rotational speed that is the same as before measurement, and the sound pressure signal and angle signal (crank angle pulse and rotation pulse) are acquired in the same manner as before measurement (FIG. 14). (Step S201), an aperiodic signal is extracted (Step S202), and an aperiodic signal within the cut-out angle range is cut out (Step S203).

  Next, the cut out aperiodic signal is subjected to FFT processing (step S204), and a second bispectrum (see equation (8)) is calculated (step S205).

Next, a total of n cycles including the second bispectrum for one cycle calculated in step S205 and the first bispectrum for (n-1) cycles calculated in step S105 of FIG. A second bicoherence (see equation (10)) is calculated based on the minute bispectrum (step S206). Further, the first bicoherence calculated in step S107 of FIG. 14 is subtracted for each frequency (f ₁ , f ₂ ) from the second bicoherence calculated in step S206 to obtain differential bicoherence (formula (See (11)) is calculated (step S207). Further, an index value (see formula (15)) is calculated using the calculated differential bicoherence and the threshold value calculated in step S111 of FIG. 14 (step S208), and information derived from the index value is presented. (Step S209).

  At the time of measurement, the above processing is repeatedly executed during the measurement period (second period).

  Here, the processing shown in FIGS. 14 and 15 is performed for each operating condition while changing the operating condition of the engine.

  In this embodiment, as described above, knocking is detected by extracting a non-periodic signal from the sound pressure signal and extracting a change in the secondary harmonic component included in the non-periodic signal. It is possible to detect even weak knocking that has been almost wiped out by various noises and cannot be detected, or knocking in the high rotation speed region of the engine.

  Here, the original signal is obtained by picking up the sound generated due to the vibration of the engine, but instead of picking up the sound, the vibration of the engine itself is picked up to obtain the original signal. May be.

  Although the example in which the present invention is applied to the knocking detection device has been described here, the signal analysis device of the present invention is not limited to the knocking detection device, and is generated due to the vibration of an object that performs a periodic operation. The present invention can be widely applied to the case where the sound is picked up and the motion of the object is analyzed.

1 Engine 2 Dynamo 3 Car 4 Test stand 5 Sound sensor 6 ECU
DESCRIPTION OF SYMBOLS 10 Knocking detection apparatus 20 Data collection apparatus 30 Notebook type personal computer 30a Display screen 31 Sound pressure signal acquisition part 32 Angle signal acquisition part 33 Aperiodic signal extraction part 34 Signal extraction part 35 Differential bicoherence calculation part 36 Index value calculation part 37 Information presentation unit 40 Operation panel 40a Monitor screen 50 Operator

Claims (8)

An original signal acquisition unit for acquiring an original signal representing a change with respect to time of a value of a physical quantity picked up by a sensor that picks up a physical quantity whose value is changed due to vibration associated with the movement of an object that performs a periodic operation;
By applying an adaptive line spectrum enhancer to the original signal acquired by the original signal acquisition unit, the periodic component contained in the original signal is suppressed, and an aperiodic signal in which the aperiodic component of the original signal is enhanced is extracted. An aperiodic signal extraction unit;
An angle information acquisition unit for acquiring angle information including origin information representing at least an origin of a periodic operation angle of the object;
Based on the angle information acquired by the angle information acquisition unit, a signal extraction unit that extracts a signal within the extraction angle range from the non-periodic signal,
A signal analysis apparatus comprising: a motion analysis unit that performs a motion analysis of the object based on the signal cut out by the signal cutout unit.   The signal analysis apparatus according to claim 1, wherein the object is an engine, and the operation analysis unit analyzes whether or not knocking of the engine occurs. An original signal acquisition unit for acquiring an original signal representing a change with time of a value of a physical quantity picked up by a sensor that picks up a physical quantity whose value is changed due to vibration caused by engine combustion;
By applying an adaptive line spectrum enhancer to the original signal acquired by the original signal acquisition unit, the periodic component contained in the original signal is suppressed, and an aperiodic signal in which the aperiodic component of the original signal is enhanced is extracted. An aperiodic signal extraction unit;
An angle information acquisition unit for acquiring angle information including at least origin position information of the rotation angle of the engine;
Based on the angle information acquired by the angle information acquisition unit, a signal cutout unit that repeatedly cuts out a signal within a cutout angle range over a plurality of cycles of engine combustion from the non-periodic signal;
Based on each of a plurality of first signals for a plurality of cycles within a first period in which knocking does not occur repeatedly cut out by the signal cutout unit, a plurality of first bispectrums are calculated, Calculating a first normalized value obtained by normalizing an average value of the plurality of first bispectrums;
A second bispectrum is calculated based on a second signal for one cycle within a second period that is likely to cause knocking and is cut out by the signal cutout unit, and the plurality of first Calculating a second normalized value obtained by normalizing an average value of a plurality of bispectrums combining the bispectrum and the second bispectrum;
A difference normalized value calculation unit that calculates a difference normalized value including a difference between the second normalized value and the first normalized value;
An index value calculation unit that calculates an index value serving as an index of the engine based on the difference normalized value calculated by the difference normalized value calculation unit;
An knocking detection apparatus comprising: an information presentation unit that presents information derived from the index value calculated by the index value calculation unit. The index value calculation unit, f ₁ of the differential normalized value calculated difference normalization value calculated in section _{b_sub (f 1, f 2)} ( where, represents the frequency both f _1, f ₂ is) = A feature value representing a level including a second-order harmonic component is calculated from a value in the vicinity of f _2, and the feature value in the frequency band near the natural frequency of the engine in at least one vibration mode is calculated. 4. The knocking detection apparatus according to claim 3, wherein a representative value is calculated and a value derived from the representative value is used as the index value. The index value calculator causes the difference normalized value calculator to calculate a plurality of the first normalized values while sequentially replacing at least one first bispectrum of the plurality of first bispectrums. , Calculating an average value μ and a standard deviation σ of differences between the plurality of first normalized values, and determining a threshold value TR for determining whether knocking is present or not,
TR = α × σ + μ
Where α is calculated according to the coefficient,
5. The knocking detection apparatus according to claim 4, wherein the index value is calculated through a comparison operation for comparing the representative value and the threshold value.   The index value calculation unit further determines whether or not knocking has occurred for the specified number of cycles using the index value calculated through the comparison operation, and determines that knocking has occurred in the specified number of cycles. Calculate the knocking rate that is the ratio of the number of cycles, and calculate the value of the average number of knocks per unit time calculated from the product of the knocking rate and the number of engine strokes per unit time from the rotation speed as a new indicator. 6. The knocking detection device according to claim 5, wherein the knocking detection device is a value.   The knocking detection device according to any one of claims 3 to 6, wherein the sensor is a sensor that picks up a sound emitted from the engine due to vibration caused by engine combustion.   The knocking detection device according to any one of claims 3 to 6, wherein the sensor is a sensor that picks up vibrations associated with engine combustion.

Images