JP6491050B2 - Abnormal sound analyzer - Google Patents

Description

  The present invention relates to an abnormal noise analysis apparatus, and more particularly to an abnormal noise analysis apparatus that determines knocking of an engine in an engine test.

  Knocking (or knocking) is a phenomenon in which unburned gas is compressed by the combustion gas in the engine cylinder and self-ignited and rapidly burns and resonates. When knocking occurs, the combustion gas propagates heat by vibration. May damage the engine. Although it is necessary to delay the ignition timing in order to suppress knocking, there is a problem that if the ignition timing is delayed, there is a problem in that the fuel consumption is lowered. Therefore, it is necessary to accurately measure and determine the occurrence of knocking.

  Auditory evaluation is widely used as a method for determining knocking (see, for example, Patent Document 1). This is a method in which the tester actually listens to the engine operating sound, hears the magnitude and frequency of abnormal sounds (abnormal sounds different from normal sounds), and evaluates the knocking state. However, there is a problem that the auditory evaluation has a large variation due to the examiner and the test environment, and the reproducibility is low. In particular, whether or not weak sounds, high-frequency sounds, and instantaneous sounds can be heard depends largely on the individual's hearing ability and experience, and it is difficult to accurately identify them unless they are skilled. For this reason, the number of people who can perform auditory evaluation is limited. Moreover, since auditory evaluation is performed by a tester, there is a problem in that it cannot be applied to an automatic control system or an automatic adaptive measurement system.

  Therefore, in recent years, in order to quantitatively determine knocking, a method using in-cylinder pressure measurement (for example, see Patent Document 2) and a method using a vibration sensor (for example, see Patent Document 3) are widely used. These methods are methods for determining knocking based on in-cylinder pressure and engine vibration, and can perform quantitative knocking determination. However, these methods have a problem that calibration is required by auditory evaluation because the determination result does not match the result of auditory evaluation.

  From such a background, there is a demand for a method for measuring knocking by the same method as the human auditory function and determining knocking, and the present applicant has proposed Patent Document 4. According to Patent Document 4, the power of the background sound is obtained by performing the time mask process based on the signal immediately before and after the engine sound detection signal and the frequency mask process based on the signal in the adjacent frequency band. The power ratio of the background sound to the power of the background sound is calculated as the knock intensity, and the presence or absence of knocking is determined. Since this method performs time mask processing and frequency mask processing in the same way as the processing performed by human auditory sense, it can obtain results equivalent to auditory evaluation and perform quantitative knock determination by signal processing. Can do.

JP 61-142367 JP2000-110652A Japanese Patent No. 3054919 JP2010-252806

  However, although the method described above is measured by the same method as the human auditory function, there is a problem that the determination accuracy as an expert cannot be obtained. Specifically, there is a problem that an expert can determine when an unusual sound other than the knocking sound is generated or when the knocking sound is weak, but it cannot be reproduced.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide an abnormal sound analyzing apparatus that can accurately specify abnormal sounds such as knocking sounds.

In order to achieve the above object, the invention according to claim 1 comprises an abnormal sound analysis comprising a detection means for detecting a sound generated by a rotating body and a signal processing means for processing a signal obtained by the detection means. In the apparatus, the signal processing means compares the abnormal noise of the newest period with the abnormal noise of the previous period based on the angle and magnitude of the peak of the abnormal noise, thereby generating the abnormal noise generated periodically. The frequency of occurrence of the identified abnormal noise, the processing of removing the abnormal noise data when the occurrence frequency exceeds a threshold, and the abnormal noise based on the data after the removal And a process of calculating the size of.

  The inventor of the present invention can detect abnormal noise generated in a rotating body, such as abnormal noise (hereinafter referred to as “high frequency abnormal noise”) that occurs periodically and in synchronization with the rotating body, Obtained low-frequency allophones (hereinafter referred to as low-frequency allophones), and experts have learned that low-frequency allophones can be recognized without being affected by high-frequency allophones. . The present invention has been made on the basis of such knowledge. Since the abnormal noise generated periodically is identified and removed, and the magnitude of the abnormal noise is obtained, it is influenced by the periodic abnormal noise. It is possible to detect the occurrence of other abnormal noises. Therefore, for example, it is possible to accurately detect the occurrence of knocking noise from engine sound.

Further , according to the present invention, since the abnormal sound data whose occurrence frequency exceeds the threshold value is removed, only the high-frequency abnormal sound is removed. Therefore, even if there is periodicity, abnormal noise with low occurrence frequency is not removed. For example, when knock sound is generated several times periodically, it is possible to prevent the knock sound from being removed.

In addition, according to the present invention, since it is determined whether or not it has periodicity based on the newest period of abnormal noise, it is possible to accurately determine even when the abnormal sound occurrence condition changes every moment. it can.

Furthermore, according to the present invention, the noise is compared with the peak angle (the peak time converted into the rotation angle of the rotating body) and the magnitude, and it is determined whether or not the noise has periodicity. . Therefore, it is possible to reliably remove periodic abnormal sounds that are the same type in hearing.

  According to the present invention, the abnormal noise generated periodically is identified and removed, and then the magnitude of the abnormal noise is obtained. Therefore, the generation of other abnormal noise can be prevented without being affected by the periodic abnormal noise. For example, the occurrence of a knocking sound can be accurately determined from engine sounds.

The figure which shows typically the structure of the abnormal sound analyzer of this Embodiment. Diagram showing knocking determination flow Diagram showing data being processed by the processor Diagram showing high-frequency abnormal noise identification and removal flow

  A preferred embodiment of an abnormal sound analyzer according to the present invention will be described with reference to the accompanying drawings. An abnormal sound analysis apparatus 10 shown in FIG. 1 is an apparatus for determining knocking of the engine 12, and mainly includes a microphone 14, an angle sensor 16, a processor 18, and a display 20.

  The microphone 14 is a sensor for detecting engine sound, and is installed in the vicinity of the engine 12. The position of the microphone 14 is preferably able to accurately grasp the distance from the sound source. For example, the microphone 14 is installed at a position 40 cm from the sound source. Note that the microphone 14 only needs to obtain sound pressure, and for example, a vibration sensor may be used. In the case of a vibration sensor, the vibration sensor may be directly attached to the engine 12 and used in combination with a processing device that performs unit conversion.

  The microphone 14 is connected to the processor 18 via the amplifier 22 and the AD converter 24. The detection signal of the microphone 14 is amplified by the amplifier 22, converted into a digital signal by the AD conversion 24, and input to the processor 18.

  On the other hand, the angle sensor 16 is a sensor that detects the crank angle of the engine 12 and is connected to the processor 18 via the AD converter 28. Instead of installing the angle sensor 16, angle information may be received directly from a control means (ECU or the like) of the engine 12.

  The processor 18 is a means for performing various signal processing, and is connected to the display 20. The display device 20 is a device that displays a scatter diagram and a knocking determination result, which will be described later, and displays measurement conditions and the like as necessary. The scatter diagram and the knocking determination result may be output to the recording medium 28 or output to a printer (not shown) or the like.

  Next, various processes performed using the processor 18 will be described. FIG. 2 shows a knocking determination flow. In the flow of FIG. 2, first, the data obtained by the microphone 14 is separated into a plurality of frequency bands (step S11). The separation method is not particularly limited. For example, the separation is performed using a plurality of bandpass filters. Further, the bandwidth and the number of bands of the frequency band to be separated are not particularly limited, but may be determined as follows based on, for example, the balance between the detection accuracy of abnormal noise and the calculation load of the processor 18. First, the center frequency fc is determined using the definition of ERB (Equivalent Rectangular Bandwidth) expressed by the following equation.

式中のnはERBバンドを可聴領域に並べたときの低域側からの並び順を示す番号である。計測対象の周波数を4k〜20kHz程度であると考えnを27〜41の整数とする。このような中心周波数を持つ１５本のバンドに分離して処理を行うようにする。この時バンド幅は対数軸に対して等間隔となるようfc(n)に対してはfc(n-1/2)からfc(n+1/2)の範囲とする。なお、計測対象の音域によってｎの範囲を変えてもよい。

N in the formula is a number indicating the arrangement order from the low frequency side when the ERB bands are arranged in the audible region. The frequency to be measured is assumed to be about 4 to 20 kHz, and n is an integer of 27 to 41. The processing is performed by separating into 15 bands having such a center frequency. At this time, the bandwidth is in the range of fc (n−1 / 2) to fc (n + 1/2) for fc (n) so as to be equally spaced with respect to the logarithmic axis. Note that the range of n may be changed depending on the sound range to be measured.

  The following processing from step S12 to step S18 is performed for each separated frequency band. First, a delay correction process is performed in order to achieve synchronization between the separated frequency bands (step S12). This is because each of the above-described bandpass filters has a unique delay, and can be synchronized by applying a correction delay for each band.

  Next, the signal is squared and the sound power is calculated (step S13). In obtaining the root mean square, the average width is preferably set as small as possible so that the peak is not erased by the averaging calculation, and is set to, for example, five bands of the band center frequency. Thereby, as shown in FIG. 3A, data in which the power on the vertical axis fluctuates greatly is obtained.

  Next, the auditory sensation A characteristic correction value is applied (step S14). The auditory sensation A characteristic correction value is a numerical value representing the approximate audible characteristic of a person, and can be approximated to an output similar to the audibility of a person. Next, the frequency mask effect is corrected (step S15). Although the method is not particularly limited, for example, considering that the high frequency band is affected by the low frequency band, a value of 0.0631 times the frequency band one lower is added. As another method, a loudness calculation chart may be used.

  Next, the time mask effect is corrected (step S16). The method is not particularly limited, but for example, considering that the auditory nerve responds fast and slows down, a primary delay filter of 0.005 sec when the volume is increased, and a primary delay filter of 0.043 sec when the volume is decreased Through. As another method, when the sound is reduced, it may be approximated that the power of the background sound is attenuated at a constant attenuation rate. Thereby, as shown in FIG.3 (b), the data which rises rapidly and falls slowly are obtained. This mimics the level of human auditory nerve activity.

  Next, an abnormal sound level is calculated (step S17). Here, the abnormal sound level expresses a human sense of abnormal sound (that is, how a person feels the abnormal sound). First, as shown in FIG. 3B, the abnormal sound level is calculated as a part where the level of the auditory nerve activity is locally maximized (in the figure, a white circle part and a part where the abnormal sound level peak point later). Then, a portion that becomes a minimum immediately before that (a portion that is a black circle in the figure and later becomes a starting point of an abnormal sound level) is searched. Then, the difference from the minimum portion to the maximum portion is obtained, and 0 is set when the maximum portion is exceeded. Thereby, the abnormal sound level as shown in FIG. By obtaining the abnormal sound level, the starting point of the abnormal sound and the peak point of the abnormal sound become clear. For example, in FIG. 3C, the abnormal sound C has a start point time C1 and a peak value C2. The time is converted into a crank angle. In the conversion to the crank angle, the signal from the angle sensor 16 is synchronized by performing a delay correction process based on the distance between the microphone 14 and the sound source.

  Next, the high frequency abnormal noise is identified and removed (step S18). That is, an abnormal noise that occurs at a high frequency in synchronization with the cycle of the engine 12 is specified, and the abnormal noise data is removed. As a result, data in which high-frequency abnormal noise is removed and low-frequency abnormal noise such as knock noise remains is obtained. The flow of specifying and removing high frequency abnormal noise will be described later.

  Next, the sum total of all frequency bands is obtained (step S19). That is, the abnormal sound level in the entire frequency band is obtained by adding the abnormal sound level obtained for each frequency band. Then, it is determined that knocking has occurred when the calculated abnormal sound level value exceeds the threshold value (step S20). When knocking occurs, it is notified to the operator by displaying it on the display device 20 or generating a pseudo sound.

  FIG. 4 shows a flow of identification and removal of high frequency abnormal noise. The processor 18 is inputted in advance with the number N of periods to be referred to, the threshold value P of occurrence frequency, and the number N1 of reference abnormal sounds.

  As shown in FIG. 4, first, a reference abnormal sound (hereinafter referred to as a reference abnormal sound) is set (step S21). As the reference abnormal noise, among the abnormal noises of the newest period, N1 (for example, five) abnormal sounds having a large peak are automatically selected. Note that the method of selecting the reference noise is not limited to this, and for example, the highest peak noise may be set as the reference noise for each predetermined crank angle.

  Next, by comparing an abnormal noise having a period before the reference abnormal noise with the reference abnormal noise, an abnormal noise that has been generated in synchronization with the reference abnormal noise (hereinafter referred to as a synchronous abnormal noise) is specified (step S22). ). Specifically, first, the magnitude and crank angle of the peak of the reference abnormal sound are obtained, and these are used as reference values. Compared with the noise of the previous period, there is a peak within a predetermined angle range (for example, within ± 5 degrees) with respect to the reference value of the crank angle, and a predetermined ratio with respect to the reference value of the peak An abnormal noise that is (for example, 10%) or more is specified as a synchronous abnormal noise. At this time, when there are a plurality of candidates for synchronous noise, the one having a small difference from the reference value of the crank angle is determined as synchronous noise. The synchronous abnormal noise is specified between the period immediately before the reference abnormal noise and the period before (N-1). That is, the reference noise and the synchronous noise are set from N consecutive cycles.

  Next, the occurrence frequency of the reference noise and the synchronous noise is obtained (step S23). Specifically, the total number of reference abnormal sounds and synchronous abnormal sounds is obtained, and the occurrence frequency is obtained by dividing by the reference period N. For example, when the number N of cycles to be referred to is 25, the synchronous abnormal noise is 19, and the reference abnormal noise is 1, the occurrence frequency is 80%.

  Next, it is determined whether or not to remove the abnormal sound according to the obtained occurrence frequency. That is, if the determined occurrence frequency exceeds the threshold value P, the abnormal sound data is removed (step S24). The data is removed from the abnormal sound level shown in FIG. 3C from the start point of the abnormal sound to the peak point. As a result, abnormal noise data periodically generated at a high frequency is removed, and sudden (or even low frequency abnormal noise data) data remains. Therefore, since knock data remains, knocking can be accurately determined in step S20 of FIG.

  As described above, according to the present embodiment, the knocking sound is determined after removing the high-frequency abnormal noise, so that knocking can be accurately determined without being affected by the high-frequency abnormal noise. Further, according to the present embodiment, since the abnormal sound data is removed when the abnormal sound occurrence frequency exceeds the threshold value, the abnormal sound data having a low occurrence frequency remains even if periodic. Become. Therefore, for example, even when a knocking sound is periodically generated only several times, the knocking sound data remains without being removed, so that the knocking can be accurately determined.

  In the embodiment described above, it is preferable that the threshold value P can be changed. If the threshold value P is too large, the ability to remove high frequency abnormal noise will be low, and the effect will hardly be obtained. On the other hand, if the threshold value P is too small, a part of the low-frequency abnormal noise is removed, so that the determination accuracy of the abnormal noise decreases. Therefore, it is preferable to obtain an appropriate value while changing the threshold value P. Specifically, the threshold value P is changed from a larger value to a smaller value in order, and a threshold value P at which abnormal sound data that does not affect the sense of hearing disappears sufficiently and knock data does not disappear is obtained. Thereby, it is possible to set the threshold value P at which high-frequency abnormal noise can be removed and knock sound can be determined systematically.

  In the embodiment described above, a data narrowing process based on the crank angle may be added. Specifically, the data is narrowed down with respect to the abnormal sound level obtained in step S17 at the crank angle which is a condition for occurrence of knocking. At that time, only the abnormal sound data starting within the range of the generation condition is extracted, and the abnormal sound data starting outside the range is removed. As a result, most of the abnormal sound data unrelated to knocking is removed. By performing the above-described high-frequency abnormal noise removal processing on such data, the high-frequency abnormal noise data unrelated to knocking is further removed. Therefore, knocking can be determined more accurately by processing the data after removal.

  DESCRIPTION OF SYMBOLS 10 ... Knock determination apparatus, 12 ... Engine, 14 ... Microphone, 16 ... Angle sensor, 18 ... Processor, 20 ... Display device, 22 ... Amplifier, 24 ... AD converter, 26 ... AD converter, 28 ... Recording medium

Claims (1)

In an abnormal sound analysis apparatus comprising a detection means for detecting a sound generated by a rotating body, and a signal processing means for processing a signal obtained by the detection means,
The signal processing means includes
Processing to identify periodically occurring abnormal noise by comparing the abnormal noise of the newest period with the abnormal noise of the previous period based on the angle and magnitude of the peak of the abnormal noise;
Calculating the frequency of occurrence of the specified abnormal noise, and removing the abnormal noise data when the frequency of occurrence exceeds a threshold ;
A process of calculating the size of the abnormal sound based on the data after the removal;
An abnormal sound analyzing apparatus characterized by

Images