JP2017090428A - Method and diagnostic system for extracting diagnostic signal with sound signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for accurately extracting a diagnostic signal by variably removing a noise signal with an input sound signal to fit the actual situations on the basis of the correlation with a signal in the input sound signal.SOLUTION: A diagnostic system according to one embodiment of the present invention, which extracts a diagnostic signal with a sound signal, includes: a signal separation unit for separating an input sound signal into a diagnostic signal interval and a noise signal interval; a noise removal parameter setting unit for setting a first parameter, which adjusts a noise attenuation degree on the basis of the correlation between a first signal of the diagnostic signal interval and a second signal of the noise signal interval; and a signal extraction unit for extracting a diagnostic signal by diminishing a noise signal included in the diagnostic signal interval on the basis of the first parameter.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a method and apparatus for extracting a diagnostic signal from a sound signal.

  The diagnostic system diagnoses the state of the mechanical device. Most diagnostic systems measure the vibration of the 10-1 kHz band generated at the diagnosis site, that is, the vibration to diagnose the state of the mechanical device. However, when the actual machine is driven, in other words, in a run-time environment, all the components of the machine are totally engaged, so diagnosis is made by the transported goods, work process, or joint and line movements. Since the vibration data of the part is affected and not only the diagnosis part but also vibration data of other parts can be included or its characteristics can be changed, it can be said that the reliability of the measured vibration data is very low. Therefore, in the vibration-based diagnosis system, after stopping the mechanical device, only the diagnosis part is operated to measure vibration. In this way, the vibration-based diagnosis system must always stop the mechanical device in order to reduce the influence of movements and environmental factors other than the diagnosis part, so it can accurately diagnose substantial problems that occur during driving. Hateful.

  Recently, sound-based diagnostic systems have been developed. The characteristics of vibration may vary depending on the transported object, work process, etc., but the sound generated when each joint of the mechanical device, the accelerator / decelerator, the motor bearing, etc. is operated includes characteristics specific to the part. Therefore, the sound-based diagnosis system can diagnose the state of the part with the mechanical device being driven even in the runtime environment.

  However, the sound-based diagnosis system can include not only a sound signal generated at a measurement site but also ambient environmental noise. Therefore, it is very important for the sound-based diagnosis system to accurately extract a signal generated at the diagnosis site from the input sound signal. However, conventional sound-based diagnosis systems mainly use noise reduction technology that simply filters an input signal using a low-pass filter or a high-pass filter. Such a filtering method can be used when a sound characteristic such as a human voice is known in advance or a sound with a clear characteristic is extracted. However, such a filtering method has a limit in accurately extracting only a sound generated at a diagnostic site among various similar sounds generated by an unspecified number of driven mechanical devices.

  On the other hand, even if we proceed with existing noise removal instead of simple filtering, the actual environmental factors and the actual special situation in the data are not considered at all, so the signal extracted by removing unnecessary signals Can be distorted. Therefore, conventional sound-based diagnostic systems have low reliability of the diagnostic signal extracted from the sound signal, and as a result, it is difficult to accurately diagnose the state of the diagnostic part.

  The problem to be solved by the present invention is to extract a diagnostic signal accurately by variably removing a noise signal from an input sound signal so as to match an actual situation based on a correlation of signals included in the input sound signal. It is to provide a method and apparatus.

  A diagnostic apparatus for extracting a diagnostic signal from a sound signal according to an embodiment of the present invention, comprising: a signal separating unit that separates an input sound signal into a diagnostic signal section and a noise signal section; a first signal of the diagnostic signal section; A noise removal parameter setting unit for setting a first parameter for adjusting the degree of noise attenuation based on the correlation of the second signal in the noise signal section; and a noise signal included in the diagnostic signal section based on the first parameter And a signal extraction unit for extracting a diagnostic signal.

  The noise removal parameter setting unit calculates a correlation coefficient between the first signal and the second signal, and adjusts the degree of attenuation of the second signal with the first signal based on the correlation coefficient. Parameters can be set.

  The noise removal parameter setting unit separates the first signal by frequency, extracts an energy change pattern according to time for each frequency of the first signal, separates the second signal by frequency, and extracts the second signal. An energy change pattern corresponding to time is extracted for each frequency of the first signal and the correlation coefficient for each frequency of the first signal and the second signal based on the energy change pattern extracted for each frequency of the first signal and the second signal. Can be calculated.

  The noise removal parameter setting unit may set the first parameter for each frequency based on the correlation coefficient for each frequency.

  The noise removal parameter setting unit may further set a second parameter for adjusting a noise attenuation level of the input sound signal based on an energy value ratio of signals included in the diagnostic signal section and the noise signal section. it can.

  The signal extraction unit may extract the diagnostic signal by using the first parameter and the second parameter to attenuate a noise signal included in the diagnostic signal section in the diagnostic signal section.

  The signal separation unit cuts the input sound signal into unit time, calculates a feature value of each unit time, compares the feature value of each unit time with a reference value, and converts the signal of each unit time into the noise signal section And the feature value is any one of a signal magnitude change value, an amplitude change value, an intensity change value, an energy value, and a spectrum value. There may be.

  The signal separation unit cuts the input sound signal into unit time, calculates a feature value of each unit time, and based on a feature value difference between the arbitrary unit time and the previous unit time of the arbitrary unit time, the arbitrary unit time Is separated into one of the noise signal section and the diagnostic signal section, and the feature value is a signal magnitude change value, an amplitude change value, an intensity change value, an energy value, and a spectrum value. Any one of them may be used.

  The diagnostic apparatus receives an input of a sound signal measured by at least two microphones attached to a diagnosis point, performs noise removal based on a physical phase difference of the measured sound signal, and inputs the input sound signal. May further include a sound signal input unit.

  The diagnostic apparatus according to another embodiment of the present invention is a method for extracting a diagnostic signal from a sound signal, cutting an input sound signal into unit time, calculating an energy value of each unit time, and a first unit time. Comparing the energy value difference between the second unit time and the critical value, and storing the signal of the first unit time in one of the diagnostic signal interval buffer and the noise signal interval buffer based on the comparison result; And a step of reducing a noise signal stored in a noise signal interval buffer with a signal stored in the diagnosis signal interval buffer and extracting a diagnosis signal, wherein the second unit time is equal to the first unit time. Previously unit time.

  The storing step updates the critical value based on the energy value difference between the signal stored in the diagnostic signal interval buffer and the signal stored in the noise signal interval buffer, and each unit based on the updated threshold value. A time signal may be stored in one of the diagnostic signal interval buffer and the noise signal interval buffer.

  The storing step stores the signal of the first unit time in the noise signal interval buffer when the energy value of the first unit time is smaller than the critical value compared to the energy value of the second unit time. And if the energy value of the first unit time is greater than the critical value compared to the energy value of the second unit time, the energy value of the first unit time is compared to the energy value of the second unit time. Determining whether the energy value is greater than the critical value, and when the energy value of the first unit time is greater than the critical value compared to the energy value of the second unit time, Storing the unit time signal in the diagnostic signal interval buffer; and the energy value of the first unit time is smaller than the critical value compared to the energy value of the second unit time. If the value may include the steps of storing a signal of said first unit time to the second time unit of the signal with the same buffer.

  The step of extracting the diagnostic signal is to determine a noise attenuation level in the first signal based on a correlation between the first signal stored in the diagnostic signal section buffer and the second signal stored in the noise signal section buffer. The method may include setting a parameter to be adjusted and extracting a diagnostic signal by attenuating the second signal with the first signal based on the parameter.

  A diagnostic apparatus according to still another embodiment of the present invention is a method for extracting a diagnostic signal based on a sound signal separated into a diagnostic signal section and a noise signal section, and includes a phase between the diagnostic signal section and the noise signal section. Calculating a relation number; setting a first parameter for adjusting a noise attenuation level of the diagnostic signal section based on the correlation coefficient; and including the diagnostic signal section based on the first parameter. And a diagnostic signal is extracted by attenuating the noise signal.

  The step of calculating the correlation coefficient includes separating a first time interval of the diagnostic signal interval by frequency, extracting an energy change pattern according to time for each frequency of the first time interval, and the noise signal interval. Separating the second time interval by frequency, extracting an energy change pattern according to time for each frequency of the second time interval, and extracting energy change for each frequency of the first time interval and the second time interval Calculating a correlation coefficient for each frequency of the first time interval and the second time interval based on a pattern.

  The step of setting the first parameter may set the first parameter for each frequency based on the correlation coefficient for each frequency.

  The second time interval may be a time interval adjacent to the first time interval.

  The diagnosis method may further include setting a second parameter for adjusting a noise attenuation level of the input sound signal based on an energy value ratio of signals included in the diagnosis signal section and the noise signal section. it can.

  The energy value ratio may be a signal to noise ratio of the input sound signal.

  The step of extracting the diagnostic signal may include extracting the diagnostic signal by attenuating a noise signal included in the diagnostic signal section in the diagnostic signal section using the first parameter and the second parameter. it can.

  According to the embodiment of the present invention, the diagnostic signal section can be accurately detected from the input sound signal. According to the embodiment of the present invention, it is possible to reduce the distortion of the diagnostic signal extracted by appropriately removing the noise signal from the input sound signal based on the correlation of the signals included in the input sound signal. According to the embodiment of the present invention, it is possible to diagnose the state of the diagnosis target device while driving the diagnosis target device based on the sound signal.

It is a schematic block diagram of the diagnostic apparatus which concerns on one Embodiment of this invention. 3 is a flowchart of a signal separation method according to an embodiment of the present invention. 3 is a flowchart of a signal separation method according to an embodiment of the present invention. It is an illustration of the input sound signal which concerns on one Embodiment of this invention. 3 is an example of a separated noise signal section according to an embodiment of the present invention. 3 is an example of a separated diagnostic signal section according to an embodiment of the present invention. It is a flowchart explaining the environment related parameter calculation method which concerns on one Embodiment of this invention. It is a flowchart explaining the correlation parameter calculation method which concerns on one Embodiment of this invention. 3 is an example of a noise signal section in a time-frequency domain according to an embodiment of the present invention. 3 is an example of a diagnostic signal interval in a time-frequency domain according to an embodiment of the present invention. It is a flowchart explaining the diagnostic signal extraction method which concerns on one Embodiment of this invention. 4 is an illustration of a time-frequency domain of a noise attenuated diagnostic signal according to an embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art to which the present invention pertains can easily implement the embodiments. However, the present invention can be embodied in a variety of different forms and is not limited to the embodiments described herein. In order to clearly describe the present invention in the drawings, unnecessary parts for the description are omitted, and similar parts are denoted by similar reference numerals throughout the specification.

  Throughout the specification, when a part “includes” a component, unless stated to the contrary, this means that the component may further include other components, unless otherwise stated. To do.

  FIG. 1 is a schematic configuration diagram of a diagnostic apparatus according to an embodiment of the present invention.

  Referring to FIG. 1, the diagnostic apparatus 100 extracts a diagnostic signal from a sound signal in which a noise signal (noise) is mixed, and diagnoses the state of a diagnostic point (diagnostic part) based on the diagnostic signal. The diagnosis point is an arbitrary position of the mechanical device, and may be, for example, a joint, an engine, a motor, an engine, a bearing, or the like. A microphone is attached to the diagnosis point or a microphone is provided around the diagnosis point. The diagnosis signal is a sound generated at the diagnosis point.

  The diagnosis apparatus 100 can extract various signals with sound signals of various frequencies, and will be described by taking an example of extracting a signal generated at a diagnosis point of a mechanical apparatus. In other words, the diagnostic device 100 extracts a sound that is emitted at a diagnostic point from sounds generated by various mechanical devices such as industrial automation equipment, automobiles, aircraft, and ships.

  The diagnostic apparatus 100 includes a sound signal input unit 110, a signal separation unit 130, a noise removal parameter setting unit 150, a signal extraction unit 170, and a determination unit 190. The signal separation unit 130 includes a noise signal section buffer 200 and a diagnostic signal section buffer 300.

  The sound signal input unit 110 receives an input of a sound signal measured by a microphone attached to a diagnosis point. The sound signal input unit 110 converts an analog sound signal into a digital sound signal and transmits the digital sound signal to the signal separation unit 130. The sound signal input unit 110 and the microphone can be connected wirelessly. The sound signal input unit 110 can transmit the sound signal to the signal separation unit 130 using wireless communication. This is because when a sound signal is transmitted by wired communication, a wire may be tangled or caught in a mechanical device, thereby hindering the operation of the mechanical device.

  The sound signal input unit 110 can remove initial noise depending on the characteristics and environment of the input sound signal. Here, since the diagnostic signal is not a human voice but a mechanical sound, the sound signal input unit 110 cannot remove noise using the characteristics of the human voice. This is because the conventional noise removal method for extracting a human voice regards motor sound, joint sound, etc. as noise.

  The sound signal input unit 110 removes initial noise using a physical phase difference between sound signals input to at least two microphones. Specifically, two microphones are provided at the diagnosis point. Two microphones are provided within a reference distance. Considering the speed of sound at room temperature, when the sound generated at the diagnosis point reaches each microphone, there is almost no phase difference between the sounds input to the two microphones. However, if the sound generated far away from the diagnosis point reaches each microphone due to refraction, diffraction, reflection, etc. of the sound according to environmental factors or distance, the phase difference between the sounds input to the two microphones Arise. The sound signal input unit 110 regards input signals having different phase differences from sounds input to the two microphones as noise and removes them. Thus, the sound signal that has undergone the initial noise removal includes sound information generated within a certain radius from the diagnosis point.

  The signal separator 130 separates the input sound signal transmitted from the sound signal input unit 110 into a diagnostic signal section and a noise signal section. The signal separation unit 130 can separate the input sound signal into a diagnostic signal section and a noise signal section based on an amplitude change of the input sound signal, an energy value, and the like. The signal separation unit 130 can store the separated signals in the noise signal section buffer 200 and the diagnostic signal section buffer 300.

  The noise removal parameter setting unit 150 sets various parameters used by the signal extraction unit 170 based on the data of the diagnostic signal section and the noise signal section. When all the noise signals are removed in the diagnostic signal section, there is a possibility that the signal becomes severely distorted and a signal indicating an abnormal sign is not extracted well. Therefore, the noise removal parameter setting unit 150 sets a parameter that determines how much the noise signal is to be reduced during noise removal. Noise removal parameters can include environment-related parameters and correlation parameters. Such a noise removal parameter can be said to be a damping factor.

  The environment-related parameter is a parameter that adjusts the degree of attenuation of the noise signal according to the measurement environment. The environment related parameter may be set based on a signal-to-noise ratio (SNR). The environment related parameter calculates a ratio for performing noise reduction based on the signal-to-noise ratio. That is, the diagnostic apparatus 100 determines whether to apply the set attenuation factor as it is or not, or to apply only a fixed ratio of the set attenuation factor. The environment-related parameter may be determined based on a ratio between the energy value of the diagnostic signal section stored in the diagnostic signal section buffer 300 and the energy value of the noise signal section stored in the noise signal section buffer 200. Noise removal is to extract the diagnostic signal accurately, but when the signal-to-noise ratio is lower than the reference value (for example, when talking with a small voice in front of a large speaker) Rather, since the diagnostic signal is removed, it is preferable not to remove the noise or to reduce the noise to a low level (for example, less than 10%). Alternatively, it is preferable not to remove noise even when the energy value of the noise signal section is very low (for example, when quiet), or to reduce the noise to a low level. Thus, the environment related parameter indicates the noise attenuation rate associated with the signal to noise ratio. For example, a noise attenuation rate of 0 may mean that noise is not removed (noise removal off), and an attenuation rate of 100% may mean that noise is completely removed (noise removal on).

  The correlation parameter is a parameter that adjusts the degree to which the noise signal included in the diagnostic signal section is attenuated. The correlation parameter can be set based on a correlation coefficient indicating how much the noise signal overlaps the diagnostic signal section 530. The noise removal parameter setting unit 150 calculates a correlation based on the energy change pattern of the diagnostic signal section and the energy change pattern of the noise signal section adjacent to the diagnostic signal section. The diagnostic signal section includes a noise signal, and when the entire noise signal is removed in the diagnostic signal section, distortion of the extracted signal becomes severe. Therefore, the noise removal parameter setting unit 150 determines the attenuation level of the noise signal in the diagnostic signal section based on the correlation parameter.

  The signal extraction unit 170 attenuates the noise signal in the diagnostic signal section based on the noise removal parameter and extracts the diagnostic signal. The signal extraction unit 170 can attenuate the noise signal in the noise signal section based on the environment-related parameter and reduce the noise signal in the diagnostic signal section based on the environment-related parameter and the correlation parameter. That is, the signal extraction unit 170 appropriately removes the noise signal in the diagnostic signal section and extracts a diagnostic signal with less distortion.

  The determination unit 190 diagnoses the state of the diagnosis part based on the signal extracted by the signal extraction unit 170.

  2 and 3 are flowcharts of a signal separation method according to an embodiment of the present invention, FIG. 4 is an illustration of an input sound signal according to an embodiment of the present invention, and FIG. FIG. 6 is an illustration of a separated noise signal section according to an embodiment, and FIG. 6 is an illustration of a separated diagnostic signal section according to an embodiment of the present invention.

  2 to 6, the signal separation unit 130 separates the input sound signal into a diagnostic signal section and a noise signal section. Referring to the input sound signal 400 of FIG. 4, the noise signal 410 and the diagnostic signal 430 have different signal sizes according to time. The signal magnitude can be expressed in terms of amplitude, strength, or energy value that the signal has. The noise signal included in the input sound signal is a signal that has been subjected to noise removal at first, or a signal that is generated at a point far from the diagnosis point, and thus the signal size is small like the noise signal 410. On the other hand, since the diagnostic signal is a signal generated near the diagnostic point, the amplitude change is larger than that of the noise signal like the diagnostic signal 430. This is similar to the characteristic that when sound data having the same energy value is input to the microphone in a quiet environment instead of a noise environment, the energy value is reduced by the distance.

  The signal separation unit 130 separates the input sound signal into a diagnostic signal section and a noise signal section based on the amplitude change or energy value of the signal section. In the noise signal section buffer 200, for example, a certain section of the input sound signal is stored as a noise signal section 510 as shown in FIG. 5, and the remaining section of the input sound signal is stored in the diagnostic signal section buffer 300 as shown in FIG. Can be stored as diagnostic signal interval 530.

  Referring to FIG. 2, the signal separator 130 receives an input sound signal from the sound signal input unit 110 (S110). An example in which the signal separation unit 130 receives an input of a sound signal as shown in FIG. 3 will be described.

  The signal separation unit 130 cuts the input sound signal in units of a fixed time, and calculates a feature value of the unit time (S120). The feature value is an index indicating the feature of the signal included in the unit time, and may vary depending on the signal processing domain such as the time domain and the frequency domain. For example, the feature value can be diversified such as a signal magnitude change value (amplitude change value), an intensity change value, an energy value, and a spectrum value. Here, the energy value of the signal will be described as the feature value.

  The signal separation unit 130 compares the energy value of each unit time with the reference value, and stores the signal of each unit time in the noise signal section buffer or the diagnostic signal section buffer (S130). For example, when the energy value of a certain unit time is larger than the reference value, the signal separation unit 130 stores the signal of this unit time in the diagnostic signal section buffer, and when the energy value of a certain unit time is smaller than the reference value, The signal separation unit 130 can store the unit time signal in the noise signal section buffer. The reference value can be set in various ways. For example, most of the input sound signals are input with a diagnostic signal while only a noise signal is initially input. Therefore, the signal separation unit 130 regards the initial signal of the input sound signal as a noise signal, and can determine the reference value based on the energy value of the initial signal.

  The signal separation unit 130 updates the reference value based on the total energy value of the signal stored in the noise signal section buffer (S140). The signal separator 130 may repeat the step (S130) of storing the sound signal in the noise signal interval buffer or the diagnostic signal interval buffer based on the updated reference value.

  Referring to FIG. 3, as for most input sound signals, only a noise signal is initially input, and a diagnosis signal may be input later. However, in some cases, it is difficult to predict the movement of the mechanical device, and in some cases, a diagnostic signal may be input at the initial stage of the input sound signal. In this case, since the diagnostic signal is also a kind of noise, the diagnostic apparatus 100 sees the initial signal and does not know whether it is a noise signal or a diagnostic signal. The signal separation unit 130 can separate the signals as follows in consideration of various situations of the input sound signal.

  The signal separation unit 130 cuts the input sound signal in units of a fixed time, and calculates an energy value during each unit time (S210).

  The signal separation unit 130 sets a reference value (E [t (0)]) and a critical value (A) (S220). The reference value (E [t (0)]) is a value for initial signal comparison. The reference value (E [t (0)]) and the critical value (A) are initial values, and values calculated by various methods can be set. For example, the reference value (E [t (0)]) may be an average energy value of the input sound signal or an average energy value of the initial input sound signal. The reference value (E [t (0)]) can be updated through a signal separation process.

  In the signal separation unit 130, the energy value (E [t (k)]) of the unit time [t (k)] is the energy value (E [t (k-1)) of the previous unit time [t (k-1)]. ]) Is smaller than the critical value (A) (Equation 1, E [t (k)] <E [t (k-1)]-A) is determined (S230).

  When Expression 1 (E [t (k)] <E [t (k−1)] − A) is “Yes”, the signal separation unit 130 converts the unit time [t (k)] signal into the noise signal section. Store in the buffer (S240).

  When Expression 1 (E [t (k)] <E [t (k−1)] − A) is “No”, the signal separation unit 130 determines the energy value (E [ t (k)]) is larger than the critical value (A) compared to the energy value (E [t (k-1)]) of the previous unit time [t (k-1)] (Equation 2, E [ t (k)]> E [t (k-1)] + A) is determined (S250).

  When Expression 2 (E [t (k)]> E [t (k−1)] + A) is “Yes”, the signal separation unit 130 converts the unit time [t (k)] signal into the diagnostic signal interval buffer. (S260).

  When Expression 2 (E [t (k)]> E [t (k−1)] + A) is “No”, the signal separation unit 130 converts the signal of unit time [t (k)] to the previous unit time [ It is stored in the same buffer as the signal of t (k-1)] (S270). The signal separation unit 130 sequentially separates n unit time signals.

  The signal separation unit 130 calculates the total energy value (noise section energy value) of the signal stored in the noise signal section buffer and the total energy value (diagnosis section energy value) of the signal stored in the vibration signal buffer (S280). .

  The signal separation unit 130 updates the critical value (A) based on the difference between the diagnostic interval energy value and the noise interval energy value (S290). The signal separation unit 130 can update the reference value (E [t (0)]) based on the noise section energy value.

  The signal separator 130 may repeat the step of storing the sound signal in the noise signal interval buffer or the diagnostic signal interval buffer based on the updated reference value and critical value.

  Various periods for updating the critical value can be set. For example, the critical value can be updated every unit time. When the critical value is updated every unit time, the noise signal and the diagnostic signal can be distinguished by the critical value updated every unit time even if the sound source has a non-constant energy value. Alternatively, the critical value can be updated after the separation for the entire unit time is completed.

  As described above, the signal separation unit 130 sets the reference value and the critical value as arbitrary values at the beginning, and updates the reference value or the critical value based on the separated signal interval, thereby making the noise signal interval and the diagnostic signal interval more accurate. Can be separated.

  FIG. 7 is a flowchart illustrating an environment-related parameter calculation method according to an embodiment of the present invention.

  Referring to FIG. 7, the noise removal parameter setting unit 150 calculates the total energy value (noise interval energy value) of the signal stored in the noise signal interval buffer 200 (S310).

  The noise removal parameter setting unit 150 calculates the total energy value (diagnosis interval energy value) of the signal stored in the diagnosis signal interval buffer 300 (S320).

  The noise removal parameter setting unit 150 calculates a signal-to-noise ratio based on the noise section energy value and the diagnosis section energy value (S330).

  The noise removal parameter setting unit 150 determines an environment-related parameter corresponding to the noise signal attenuation rate based on the signal-to-noise ratio (S340).

  FIG. 8 is a flowchart illustrating a correlation parameter calculation method according to an embodiment of the present invention. FIG. 9 is an example of a noise signal section in the time-frequency domain according to an embodiment of the present invention. FIG. 10 is an illustration of a diagnostic signal interval in the time-frequency domain according to an embodiment of the present invention.

  Referring to FIG. 8, the diagnostic signal section overlaps the diagnostic signal and the noise signal. The noise removal parameter setting unit 150 calculates a correlation coefficient indicating how much the diagnostic signal and the noise signal overlap in the diagnostic signal section, and determines the degree of attenuation of the noise signal in the diagnostic signal section based on the correlation coefficient. .

  The noise removal parameter setting unit 150 converts the signal stored in the noise signal section buffer 200 into the time-frequency domain (S410). The signal 510 stored in the noise signal section buffer 200 is converted into the time-frequency domain 610 of FIG.

  The noise removal parameter setting unit 150 separates the noise section signal in the time-frequency domain for each frequency, and extracts an energy change pattern corresponding to the time for each frequency (S420). The noise removal parameter setting unit 150 stores an energy change pattern corresponding to the time for each frequency for each separated time section (time band). Referring to FIG. 9, in the time-frequency domain 610, the horizontal axis is time, and the vertical axis is frequency. That is, since the sound signal input in a certain time interval includes signals of various frequencies, the noise removal parameter setting unit 150 separates the signal by frequency. For example, referring to FIG. 9, the noise removal parameter setting unit 150 can extract an energy change pattern 710 corresponding to the time of the Nth frequency in a certain time interval. At this time, the energy change pattern 710 may be similar to a noise pattern according to time in the Nth frequency band where only noise exists.

  The noise removal parameter setting unit 150 converts the signal stored in the diagnostic signal section buffer 300 into the time-frequency domain (S430). The signal 530 stored in the diagnostic signal section buffer 300 is converted into the time-frequency domain 630 of FIG.

  The noise removal parameter setting unit 150 separates the diagnostic signal section in the time-frequency domain by frequency and extracts an energy change pattern according to time for each frequency (S440). The noise removal parameter setting unit 150 stores an energy change pattern corresponding to the time for each frequency for each separated time section (time band). Referring to FIG. 10, the noise removal parameter setting unit 150 can extract an energy change pattern 720 according to the time of the Nth frequency in a certain time interval, and the energy change pattern according to the time of the Kth frequency. 730 can be extracted.

  The noise removal parameter setting unit 150 compares the energy change patterns corresponding to the frequency-dependent times of the diagnostic signal section and the noise signal section, and calculates the correlation between the noise signal and the diagnostic signal for each frequency (S450). Specifically, the noise removal parameter setting unit 150 uses the energy change pattern according to the time of the noise signal section adjacent to the diagnostic signal section to detect the phase between the diagnostic signal and the noise signal existing in the diagnostic signal section. Estimate the number of relationships. For example, the Nth frequency pattern 720 of a certain diagnostic signal section is similar to the Nth frequency pattern 710 of an adjacent noise signal section, but the Kth frequency pattern 730 of the diagnostic signal section is adjacent. It may be different from the K-th frequency pattern (not shown) of the noise signal section. Accordingly, the correlation between the Nth frequency signal in a certain diagnostic signal section and the Nth frequency signal in the adjacent noise signal section is calculated to be high, and thereby, it is determined that the Nth frequency signal in the diagnostic signal section is close to the noise signal. . On the other hand, the diagnostic signal and the noise signal are mixed in the Kth frequency of a certain diagnostic signal section, and the correlation coefficient between the diagnostic signal and the noise signal included in the Kth frequency of the certain diagnostic signal section is based on the superposition rate. Is calculated.

  The noise removal parameter setting unit 150 calculates a correlation parameter for each frequency of the diagnostic signal section based on the correlation between the noise signal calculated for each frequency and the diagnostic signal (S460). The correlation parameter for each frequency is a parameter that adjusts the degree of attenuation of the noise signal included in the diagnostic signal section for each frequency. For example, if the pattern of the Nth frequency of a certain noise signal section and the Nth frequency of the diagnostic signal section match, this is almost a noise signal, so the correlation parameter of the Nth frequency is a correlation coefficient value that is almost similar to the noise signal. (A), whereby the attenuation factor for A is set by the ratio of the correlation coefficient calculated for the energy value of the Nth frequency. Even if the correlation coefficient value is not close to 100%, the energy value attenuation rate is set according to the ratio corresponding to the correlation coefficient value calculated as the energy value of the Nth frequency.

  FIG. 11 is a flowchart illustrating a diagnostic signal extraction method according to an embodiment of the present invention, and FIG. 12 is an example of a time-frequency domain of a noise-attenuated diagnostic signal according to an embodiment of the present invention. .

  The present invention is not a device for transmitting only the voice of the speaker to the other party through noise removal during a call, but is a device for diagnosing the state by detecting the sound of the section of the measurement site. Spectral data of the diagnostic signal section in which noise is attenuated is acquired without separately converting the signal section into audio data in the time domain again. After the acquisition, only the spectrum data of the diagnostic signal section is extracted without using the noise signal section separately.

  Referring to FIG. 11, the signal extraction unit 170 attenuates the noise signal included in the diagnostic signal section based on the noise removal parameter set or calculated by the noise removal parameter setting unit, and extracts only the diagnostic signal. .

  The signal extraction unit 170 reduces the energy of the noise signal in the diagnostic signal section based on the environment-related parameter and the frequency-specific correlation parameter (S510). That is, when the noise signal is attenuated in the diagnostic signal interval, the signal extraction unit 170 reduces the energy value of the frequency according to the correlation ratio between the noise signal interval and the diagnostic signal interval for each frequency. If the correlation is large, in other words, if the value is close to 100%, it is almost similar to the noise signal, so this is regarded as a noise signal and greatly attenuated. If the correlation is small, that is, close to 0%. Since this can be seen as different from the noise signal, it is reduced by a small amount.

  However, since the correlation of the signals calculated theoretically with different operation types of the entire system, that is, the type of microphone and the input volume may not be suitable for various actual environments, the correlation coefficient is 100. Even if calculated as%, the maximum attenuation rate is set through a separate constant parameter without proceeding with 100% attenuation, thereby reducing the energy value. At this time, the environment set by the noise removal parameter setting unit Apply the parameters to calculate the final frequency energy decay rate.

  Even if the attenuation value set through the correlation is applied and the energy value obtained by applying the attenuation factor for each frequency is obtained, it can be said that the noise signal is attenuated within the diagnosis interval. If the ratio, ie the signal-to-noise ratio (SNR, here referred to as environment-related parameter (C)) is low, this can be seen as a case where the energy value of the diagnostic signal is very small or only the diagnostic signal is input. it can. In such a case, when the attenuation rate due to the correlation is applied as it is, the entire signal may be lost, distorted, or the remaining diagnostic signal may be diminished. Therefore, the environment parameter is applied as a means for supplementing such a situation to the attenuation rate of the energy value of each frequency calculated through the correlation coefficient. The higher the SNR, the closer the environment-related parameter (C) is to 1, and the lower the SNR, the closer to 0 the environmental parameter (C) is derived. By using this, if the environment-related parameter is applied to the attenuation rate for each frequency calculated by the correlation, the SNR is high, and the better, the value is closer to 1. Therefore, the attenuation of the frequency band calculated by the correlation The rate is applied as is, and the energy value is reduced. On the other hand, as the SNR is lower and worse, the environment-related parameter becomes closer to 0. Therefore, if the environment-related parameter is applied to the attenuation rate of the frequency band calculated by the correlation, a value close to 0% is calculated. Is done. Since an attenuation factor of 0% can mean no attenuation at all, the SNR is low in the end. If it is not good, the attenuation factor is 0%, that is, almost completely bypassed so that the input signal can be completely preserved. become.

  The signal extraction unit 170 outputs a diagnostic signal with noise attenuation (S520).

  If the energy value is attenuated for each frequency by the attenuation rate set by the correlation and the environmental parameter in the diagnostic signal section, the spectrum data from which the noise signal is removed in the diagnostic section signal as shown in FIG. Later, it is saved as a digital file for data diagnosis. Referring to FIG. 12, when comparing the time-frequency domain 650 of the diagnostic signal and the time-frequency domain 630 before noise attenuation, it can be seen that the time-frequency domain 650 shows the characteristics of the diagnostic signal more accurately.

  As described above, according to the embodiment of the present invention, it is possible to accurately detect a diagnostic signal section in which a movement of a diagnostic part has occurred in an input sound signal, and in particular, the signal-to-noise ratio is remarkably improved and the movement is accurately performed. Can detect the beginning and end of. According to the embodiment of the present invention, it is possible to reduce the distortion of the diagnostic signal extracted by appropriately removing the noise signal from the input sound signal based on the correlation of the signals included in the input sound signal. According to the embodiment of the present invention, it is possible to diagnose the state of the diagnosis target device while driving the diagnosis target device based on the sound signal.

  Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications of those skilled in the art using the basic concept of the present invention defined in the claims. And improvements are also within the scope of the present invention.

Claims (20)

A diagnostic device for extracting a diagnostic signal with a sound signal,
A signal separator for separating the input sound signal into a diagnostic signal section and a noise signal section;
A noise removal parameter setting unit for setting a first parameter for adjusting a degree of noise attenuation based on a correlation between the first signal of the diagnostic signal section and the second signal of the noise signal section;
A signal extractor for attenuating a noise signal included in the diagnostic signal section based on the first parameter and extracting a diagnostic signal;
Diagnostic device including The noise removal parameter setting unit includes:
The correlation coefficient between the first signal and the second signal is calculated, and the first parameter for adjusting the degree of attenuation of the second signal with the first signal is set based on the correlation coefficient. The diagnostic device according to 1. The noise removal parameter setting unit includes:
Separating the first signal according to frequency, extracting an energy change pattern according to time according to the frequency of the first signal;
Separating the second signal according to frequency, extracting an energy change pattern according to time according to the frequency of the second signal;
The diagnostic apparatus according to claim 2, wherein a correlation coefficient for each frequency of the first signal and the second signal is calculated based on an energy change pattern extracted for each frequency of the first signal and the second signal. The noise removal parameter setting unit includes:
The diagnostic apparatus according to claim 3, wherein the first parameter is set for each frequency based on the correlation coefficient for each frequency. The noise removal parameter setting unit includes:
The diagnostic apparatus according to claim 2, further comprising a second parameter for adjusting a noise attenuation level of the input sound signal based on a ratio of energy values of signals included in the diagnostic signal section and the noise signal section. The signal extraction unit
The diagnostic apparatus according to claim 5, wherein the diagnostic signal is extracted by reducing a noise signal included in the diagnostic signal section in the diagnostic signal section using the first parameter and the second parameter. The signal separator is
The input sound signal is cut into unit times, a feature value for each unit time is calculated, a feature value for each unit time is compared with a reference value, and a signal for each unit time is compared between the noise signal interval and the diagnostic signal interval. Separated into one of them,
The diagnostic apparatus according to claim 1, wherein the feature value is any one of a signal magnitude change value, an amplitude change value, an intensity change value, an energy value, and a spectrum value. The signal separator is
The input sound signal is cut into unit times, a feature value of each unit time is calculated, and the signal of the arbitrary unit time is converted to the noise signal based on a feature value difference between the arbitrary unit time and the previous unit time of the arbitrary unit time. Separating into one of the section and the diagnostic signal section,
The diagnostic apparatus according to claim 1, wherein the feature value is any one of a signal magnitude change value, an amplitude change value, an intensity change value, an energy value, and a spectrum value. A sound signal that receives input of a sound signal measured by at least two microphones attached to a diagnosis point and performs noise removal based on a physical phase difference of the measured sound signal to generate the input sound signal The diagnostic apparatus according to claim 1, further comprising an input unit. A method in which a diagnostic device extracts a diagnostic signal using a sound signal,
Cutting the input sound signal into unit time and calculating the energy value of each unit time;
The energy value difference between the first unit time and the second unit time is compared with the critical value, and the signal of the first unit time is converted into one of the diagnostic signal interval buffer and the noise signal interval buffer based on the comparison result. Saving, and
Attenuating the noise signal stored in the noise signal interval buffer with the signal stored in the diagnostic signal interval buffer and extracting the diagnostic signal; and
The diagnostic method, wherein the second unit time is a unit time before the first unit time. The storing step includes
The critical value is updated based on the energy value difference between the signal stored in the diagnostic signal interval buffer and the signal stored in the noise signal interval buffer, and the signal of each unit time is diagnosed based on the updated critical value. The diagnostic method according to claim 10, wherein the diagnostic method is stored in any one of a signal interval buffer and the noise signal interval buffer. The storing step includes
Storing the signal of the first unit time in the noise signal interval buffer when the energy value of the first unit time is smaller than the critical value compared to the energy value of the second unit time;
If the energy value of the first unit time is greater than the critical value compared to the energy value of the second unit time, the energy value of the first unit time is greater than the critical value compared to the energy value of the second unit time. Determining whether the value is greater than the value;
Storing the first unit time signal in the diagnostic signal interval buffer when the energy value of the first unit time is greater than the critical value compared to the energy value of the second unit time;
When the energy value of the first unit time is smaller than the critical value compared to the energy value of the second unit time, the signal of the first unit time is the same buffer as the signal of the second unit time. To save the stage,
The diagnostic method of Claim 10 containing these. Extracting the diagnostic signal comprises:
Setting a parameter for adjusting the degree of noise attenuation in the first signal based on the correlation between the first signal stored in the diagnostic signal section buffer and the second signal stored in the noise signal section buffer;
Reducing the second signal with the first signal based on the parameter and extracting a diagnostic signal;
The diagnostic method of Claim 10 containing these. A diagnostic device extracts a diagnostic signal based on a sound signal separated into a diagnostic signal section and a noise signal section,
Calculating a correlation coefficient between the diagnostic signal section and the noise signal section;
Setting a first parameter for adjusting a noise attenuation level of the diagnostic signal section based on the correlation coefficient;
Extracting a diagnostic signal by attenuating a noise signal included in the diagnostic signal section based on the first parameter;
A diagnostic method comprising: Calculating the correlation coefficient comprises:
Separating the first time interval of the diagnostic signal interval by frequency and extracting an energy change pattern according to time for the frequency of the first time interval;
Separating the second time interval of the noise signal interval by frequency, and extracting an energy change pattern according to time for each frequency of the second time interval;
Calculating a correlation coefficient for each frequency of the first time interval and the second time interval based on an energy change pattern extracted for each frequency of the first time interval and the second time interval;
The diagnostic method of Claim 14 containing these. Setting the first parameter comprises:
The diagnostic method according to claim 15, wherein the first parameter is set for each frequency based on the correlation coefficient for each frequency.   The diagnostic method according to claim 15, wherein the second time interval is a time interval adjacent to the first time interval. The method according to claim 14, further comprising: setting a second parameter for adjusting a noise attenuation level of the input sound signal based on a ratio of energy values of signals included in the diagnostic signal section and the noise signal section. Diagnosis method.   The diagnostic method according to claim 18, wherein the energy value ratio is a signal-to-noise ratio of the input sound signal. Extracting the diagnostic signal comprises:
The diagnostic method according to claim 18, wherein the diagnostic signal is extracted by reducing a noise signal included in the diagnostic signal section in the diagnostic signal section using the first parameter and the second parameter.

Images