JPH0617817B2 - Aircraft model identification method - Google Patents

Description

The present invention relates to an aircraft model identification method, and more particularly to an aircraft model identification method that utilizes sound pressure fluctuations of noise generated by an aircraft during flight or ground operation.

Aircraft noise accounts for a large proportion of noise pollution that affects the living environment. Since the establishment of the environmental standard for aircraft noise in 1973, the measurement and monitoring of aircraft noise have been frequently performed in line with its purpose. Among them, there are large-scale systems in many places around the airfield to monitor aircraft noise exposure over a long period of time, but in these, only noise caused by aircraft is specified. It has been recognized that it is necessary to classify and observe observations according to aircraft type and flight mode. However, although a method of automatically distinguishing between aircraft noise and other noise has been developed and put into practical use in recent years, an automatic identification method has not yet been established for the type and flight form of aircraft. The current situation.

This invention was made in consideration of the above circumstances,
The aircraft type and flight mode are identified using the noise generated by the aircraft itself, and the aircraft type that can also identify the aircraft type and flight mode by only acoustically measuring at the noise observation point. The purpose is to provide an identification method.

Hereinafter, the present invention will be described in detail with reference to embodiments.

In the aircraft model identification method of the present invention, first, a basic pattern is created for each group (hereinafter referred to as a model) classified by the type and flight mode of a known known aircraft. Next, in the actual identification process for the unknown aircraft to be identified, the weighted distance in the vector space of the reference pattern and the detection pattern by each of the characteristic parameters of the discrimination target is calculated to determine which aircraft the aircraft belongs to. Is to provide a method for identifying the aircraft model.

In the following examples, in order to clarify the content of the aircraft noise, the type of aircraft is changed to, for example, "Large 4-engine turbojet transport aircraft and model number", "Small twin-engine turbojet training machine and model number", "Large 4-engine It is classified into "Turboprop transport aircraft and model number", "Helicopter", "Other jet aircraft" and "Other propeller aircraft", and flight modes are classified into, for example, "takeoff" and "landing".

Further, in this embodiment, after the frequency band analysis value of the aircraft noise is made relative to the peak value of the level fluctuation before analysis and the time position of the peak, the characteristic parameter is extracted to obtain the detection pattern.

FIG. 1 is a flow chart showing an embodiment of the present invention. In this embodiment, model discrimination is performed in two stages.

In the first step, first, the characteristic parameters are extracted from the noise sample of the aircraft to be identified observed in step 1, and the sample pattern = (X ₁ , X ₂ , ...,
X _N ) is calculated, and in step 2, this sample pattern and a reference pattern of characteristic parameters prepared in advance for each model (total number M) ▲ ▼ = (Pi ₁ , Pi
₂ , ..., Pi _N ), i = 1 ,.
₁ (▲ ▼,) is calculated (the definition of the distance will be described later), and the smallest one is min {D ₁ (▲
▼,)}, min {D ₁ in step 3
If (▲ ▼,)} is also smaller than the preset identification threshold D ₀ , the sample is min {D
₁ (▲ ▼,)} is determined to belong to the model (step 4), and the process is terminated.

If D ₀ <min {D ₁ (▲ ▼,)}, then M
It is determined that the model does not belong to any of the individual models, and the process proceeds to the second step. First, in step 5, a sample pattern with characteristic parameters different from those in the first step is calculated, and then in step 6, the reference pattern ▲ ▼ (Jett) and ▲ ▼ (Propeller) are used to add a new distance D ₂ (▲
▼,) and D ₂ (▲ ▼,) are calculated, and the magnitudes of the calculated values are compared in step 7 to determine whether they are from another jet machine (step 8) or another propeller machine (step 9). To do. If necessary, a jet machine,
Whichever of the propeller planes is discriminated, it is classified according to the peak level of the loudness, and the discrimination procedure is ended.

FIG. 2 is a flow chart of another embodiment, and in this embodiment as well, the discrimination procedure is composed of two stages. In the first stage,
First, in step 11, the sample pattern Y is calculated from the noise sample of the aircraft to be identified, and in steps 12, the distances D ₂ (▲ ▼,) and D ₂ (▲).
▼,) is calculated, and the magnitude is compared in step 13, and the sample is compared with the jet machine (step 14).
Or a propeller machine (step 15).

Next, in the second step, for each of the jet machine and the propeller machine (only the jet machine side is shown in the figure), the sample pattern is calculated in step 16 and the distance D from the reference pattern prepared in advance is calculated in step 17.
₁ (▲ ▼,), i = 1, ..., M are calculated, and in step 18, min is performed as in the case of the embodiment of FIG.
If {D ₁ (▲ ▼,)} is smaller than the recognition threshold D ₀ , it is determined that the model belongs to the model corresponding to min {D ₁ (▲ ▼,)} (step 19), and if not, another jet machine is used. It is determined (step 20).

If determined to be other, if necessary, the sound level is classified according to the peak level of the loudness, and the process ends.

Here, the distance between the sample pattern and the reference pattern used in the above two embodiments will be described.
First, consider a vector space composed of N “feature parameters” extracted from sound pressure fluctuations of aircraft noise.

When the total number of models to be identified is M and the "reference pattern" of the i-th model is Pi, the distance D ₁ (▲) from the sample pattern X of the sample to be identified
▼,) It is defined as Here, Win is a weight.

In the above-described first and second embodiments, the reciprocal of the variance in the n-th component of the i group is used for this weight Win.

The distance D ₁ (▲ ▼,) defined here is the Mahalalanobis generalized distance used to measure the distance between multivariate normal populations (with equal covariance matrices), or in the simpler case Is similar to the weighted Euclidean distance, but is different in that the weight Win is given for each model and each feature pattern.

If weights are given to each model and characteristic parameters in this way, when the stability of a certain characteristic parameter of a certain model is not desirable, in other words, when the variance is very large, the reciprocal weight is very high. It is a small value and has little effect on the calculated distance.

On the contrary, if the model to which the machine belongs can be specified only by looking at the values of some of the characteristic parameters, all of the weights Win other than the model and the characteristic parameters are set to sufficiently small values, and they have almost no effect. You can say that.

In this way, the selection of the characteristic parameter that works effectively for each model has two advantages of shortening the calculation time and reducing the storage capacity of the reference pattern. It is considered to be useful when configuring the identification device used in the method of the present invention as a dedicated hardware device.

By using the distance weighted by the reciprocal Win of the variance for each model as described above for each component of the characteristic parameter, the difference in the reference pattern between the models is emphasized, which leads to the improvement of the discrimination ability.

Similar definitions are used for the distances D ₂ (▲ ▼,) and D ₂ (▲ ▼,), but they are based on different feature parameters.

Next, a method of selecting a characteristic parameter that is the basis of the identification process, and a method of creating a reference pattern and a detection pattern for each model, which includes the characteristic parameter will be described.

FIG. 3 shows an example of the arrangement of measurement points in a field survey conducted to observe the sound pressure fluctuation of aircraft noise for this purpose. In this example, one measurement point was provided on each extension of the runway 31 and immediately below the flight path about 1 km away from the ends of the runway 31 to observe aircraft noise during takeoff and landing. Noise measurement microphones 32a, 32 arranged at the measurement points
Although b was installed 1.2 m above the ground, the altitude to the aircraft was tens to hundreds of meters.

As a general rule, it is considered sufficient to create the reference pattern once, but in general, the place where noise is observed is often far from the runway 31, and the altitude from the microphones 32a and 32b for measuring noise to the aircraft is also high. It changes a lot. As will be described later, when the frequency identification of the noise from which the characteristic parameter is calculated changes significantly,
At that observation point, the aircraft noise will be acquired in advance and the reference pattern will be calculated. The height of the microphones 32a and 32b from the ground is not limited to 1.2 m and may be appropriately selected.

The signal level of the aircraft noise measurement signal obtained from the microphones 32a and 32b corresponds to the noise of five aircraft of the same model (for example, the noise during takeoff of a twin twin jet trainer), and five noise change curves are superposed. As shown in FIG. 4, while a known or unknown aircraft gradually approaches from a distant place, passes through the measurement point or its vicinity, and then moves to a distant place, as time passes, There is a transient change in which the signal level gradually increases from a low signal level to a peak level at or near the measurement point position and then decreases to a low signal level.

Here, the peak of the noise change curve is different depending on the aircraft model, mainly based on the difference of the noise source installed in the aircraft, and the noise change curve exceeds the predetermined signal level. The time in between (ie the transit time of the aircraft) varies primarily due to differences in the speed of the aircraft.

In the present invention, the characteristic pattern for discriminating the model by paying attention to the characteristics of the peak and the passage time of the noise change curve is first obtained by the frequency obtained from the noise change curve of the frequency band component included in the noise measurement signal. It is extracted from the component peak value level, its change amount, and the change at the peak time.

The reference pattern for the known aircraft and the detection pattern for the unknown aircraft are calculated by the following method.

That is, first, the noise measurement signals obtained from the microphones 32a and 32b are taken in through a wide band pass filter, and as shown in the noise change curve of "dB (A)" in FIG. Is obtained as noise change information representing the total level change of the noise generated each time the vehicle passes, and the peak level value and the time point of occurrence of the peak level are obtained.

The waveform diagram of FIG. 4 shows that the time-dependent changes in the detected signal levels of five aircraft of the same model are made relative to each other with respect to the time axis and the signal level as described later, and the five waveforms are superimposed on each other. Is shown.

Here, the noise change information is captured as significant information in a curve portion that exceeds a level 10 to 15 dB lower than the peak level over the range before and after the peak level.

In the case of this embodiment, a wide band pass filter having the A characteristic of JIS C1505-1988 was used, and thereby the signal level was re-evaluated so as to match the human auditory characteristic when listening as noise. It is designed to get information.

The number of samples to be acquired shall be an appropriate number for each aircraft type and flight mode to be identified. It is desirable that the number is sufficient to estimate the average and variance of the population of the feature parameters to be extracted, but for example, 5 to 10 is sufficient.

Second, the noise measurement signals obtained from the microphones 32a and 32b are separated into frequency component signals in real time by a plurality of narrow band pass filters, and f ₀ = 80 in FIG.
As shown in the noise change curves of Hz, 160Hz, 200Hz, 400Hz, 500Hz, 1000Hz, and 1250Hz, the noise change signal that occurs each time one aircraft passes is a signal of multiple frequency components included in the noise measurement signal. It is extracted as a change in level, and as a result, the type of noise source included in the change in noise (that is, the "engine sound of the jet engine" of the jet machine).
And "turbo sound", "propeller sound" and "engine sound" of propeller aircraft, "rotor blade sound" and "engine sound" of helicopter
Get information about).

In this embodiment, JI is used as a narrow band pass filter.
The S standard C1513-1983 1/3 octave band filter is applied, whereby the center frequency f ₀ is 80 Hz, 160 Hz, 200 Hz, 400 Hz, 500 Hz, 10 as shown in FIG.
The frequency components of the frequency bands of 00 Hz and 1250 Hz can be extracted.

Thirdly, the peak level value of the noise level change of the noise change information of the wideband bandpass filter and the peak occurrence time are held, and the peak level value is set to a predetermined reference level L.
_The level is adjusted so as to match with _A, and thereby the measured noise change information is relativized with respect to the signal level.

By performing relativization processing for this peak level value, it is possible to avoid pattern fluctuations due to changes in the distance between the aircraft and the observation point (which occur over time), and noise is reflected by the ground and other sources. It is possible to avoid the influence on the pattern even if it is put in the microphone.

With the fourth this, the noise change information in a broadband bandpass filter peak time point of generation of the change in the noise level is adjusted the time axis to a predetermined reference peak occurrence time T _A, thereby the noise change information the measurement time axis Is relativized.

Fifth, the signal levels and time bases of a plurality of frequency components are
Only the level adjustment amount and the time axis adjustment amount for the noise change information are adjusted, and thereby the relativization processing is performed for a plurality of frequency components.

Incidentally, the magnitude and time axis of the aircraft noise converted into the noise measurement signal by the microphones 32a and 32b are distributed based on the flight position and the transit time of the aircraft when passing on or near the measurement point. However, by performing such a relativization process, the dispersion of the measurement information can be suppressed, and as a result, as shown by the curve “dB (A)” in FIG. Curves) can be overlapped with each other so as to coincide with the reference peak level L _A and the peak occurrence time T _A at the peak level and the occurrence time.

When such relativization processing is performed, the curve f ₀ = 80 in FIG.
As shown in Hz, 160Hz, 200Hz, 400Hz, 500Hz, 1000Hz and 1250Hz, it is possible to effectively suppress the dispersion of measurement information by overlapping the five curves for the peak level and the peak occurrence point for the frequency component. it can.

Sixth, the peak level values of the plurality of frequency components thus relativized with respect to the signal level and time axis are held, and the plurality of peak level values are obtained as basic characteristic parameters. As a result, the reference pattern can be formed by the peak level values of the multiple frequency components of the known aircraft type, and the detection pattern can be formed by the peak level values of the multiple frequency components of the unknown aircraft type. Can be formed.

The peak level values for the plurality of frequency components forming the reference pattern and the detection pattern thus formed are
If this is expressed as a characteristic curve diagram with the horizontal axis as frequency, it can be expressed as the array shape of points as shown in FIG. Therefore, if the array shape of points as shown in FIG. 5 is obtained for each model, it is possible to represent the type of noise source possessed by each type of aircraft by the difference in the shape.

At each point in FIG. 5, “white circle” indicates the peak level (dB) of the frequency component of each frequency band, and “the line segment that crosses the white circle in the vertical direction” indicates the average level of the observed values (the center point of the white circle). The standard deviation (variation range) from the position is shown.

In the case of this embodiment, the aircraft noise identification system obtains peak level values of a plurality of frequency components as a basic feature parameter, and uses this to obtain a next model identification feature parameter.

First, for each model, i = 1, 2 ... Lth 1/3 each
The octave band frequency component obtained from the filter peak level value Li (i = 1,2 ...... L) , the deviation ΔLi = Li from a relative criterion level value L _A used when the relative treatment as described above _{-L A (i = 1,2 ...... L} ) dB
Is obtained, and the model is identified using this deviation ΔLi as a characteristic parameter. Thus, a relatively small amount of data can be used to make the similarity determinations described above with respect to FIGS.

In the case of this embodiment, the fluctuation of the frequency component due to the Doppler effect occurs as a change within the 1/3 octave band.

Secondly, as shown in FIG. 6, i = 1, 2 for each model.
... Output signal of Lth 1/3 octave band filter (that is, waveforms JD1, JD2, JD in FIG. 6)
The signal level of the frequency component indicated by 3 is a predetermined level ΔL (for example, ΔL = 5 to 10) from the peak level value.
The time over which the threshold level lower by dB) is exceeded is obtained as the duration Di (i = 1, 2 ... L), and the model is identified using this duration Di as a characteristic parameter.

Here, the change in the magnitude of the aircraft noise observed varies depending on the model, flight mode, sound frequency, positional relationship between the aircraft and the observation point, and the like. So, flight form, sound frequency,
If the positional relationship between the aircraft and the observation point is determined, the change in the magnitude of the observed noise can be made unique to the model.

The speed of change also depends on the moving speed of the aircraft,
The degree is small before and after takeoff and landing.

After all, as shown in FIG. 6, the length of the time D1, D2, D3 ... When the signal level of each frequency component exceeds the threshold level lower by a predetermined level .DELTA.L from the peak level is known as the detection pattern. By determining the degree of similarity with the model reference pattern, the aircraft model to be observed can be identified.

Third, for each model, and Natsuta time T _A output signal, that noise change information wideband bandpass filter) is the peak level as shown in FIG. 7, 1/3-octave-band filter output signal (i.e. the Waveform KD in Fig. 7
1), KD2, KD3, low frequency components, medium-height frequency components, and high-frequency components, which are representative examples of high-frequency components.
Obtains a time difference Δti = _{Ti-T A} with (i = 1,2 ...... L), the identification of the model by using the peak level occurrence time difference as a feature parameter.

Here, the main source of aircraft noise is the engine injection noise, which has noise components over a wide frequency range from low frequencies to high frequencies.

The way the injection noise is generated differs depending on the structure and form of the engine, and the difference appears as a difference in the direction in which individual frequency components are strongly radiated and a difference in magnitude.

For example, a sharp metal sound of a jet engine has a property of being extremely strongly transmitted to the front of the engine as a sound of a fan, and the frequency component of this factor in the observed sound becomes maximum at an early time.

On the contrary, jet noise is strongly radiated obliquely backward, so that a loud noise is observed after the aircraft has passed over the sky.

Since the degree differs depending on the model, the model can be identified by observing the difference in the time when the peak of each signal component of noise appears as a detection pattern.

After all, as shown in Fig. 7, the volume of the observed sound changes depending on the model (engine type), the frequency of the sound, and the positional relationship between the aircraft and the observation point, and it is observed most strongly in each frequency band. The time of day is not limited to the time when the distance between the aircraft and the observation point becomes shortest.

Fourth, for the jet aircraft and the propeller aircraft (and helicopter), the peak level value Li of the frequency component obtained from the i = 1, 2 ... Lth 1/3 octave band filter.
Of (i = 1, 2 ... L), the difference between two peak level values Li and Li + 1 extracted from adjacent frequency bands in a predetermined frequency range in the frequency axis direction is obtained, and the absolute value of the difference is obtained. The sum of squares of is calculated and is used as a characteristic parameter to discriminate between a jet aircraft and a propeller aircraft (and helicopter).

For example, the first-order difference ΔL (1) However, it can be obtained by j = j ₁ , j ₂ ... (4) 1 ≦ j ₁ , j ₂ ... (5).

As described above, when the sum of squares of the difference between the peak values of the frequency components adjacent to each other in the frequency axis direction is obtained and used as the characteristic parameter, the jet machine and the propeller machine (and the helicopter) can be easily distinguished.

In general, the noise spectrum of a jet machine makes a comparatively gentle spectrum change in the frequency axis direction, whereas the noise of a propeller machine (and helicopter) is mainly generated from the rotating part (that is, propeller, swivel blade, etc.). Therefore, the spectrum changes drastically in the frequency axis direction (that is, it increases by a specific frequency component). Therefore, if the difference information indicating the change in the peak value of the adjacent frequency components is used as the characteristic parameter, the jet machine and the propeller machine (and the helicopter) can be reliably identified.

According to the above configuration, the distance calculation for discrimination is weighted by the variance for each model and each feature parameter, so that the difference in features between model groups can be emphasized and clarified. Selection of parameters can be performed.

Incidentally, all parameters may be sorted by model, and parameters that have good data cohesion and small variance for a specific model and large variance for other models may be appropriately extracted.

On the other hand, conventionally, only the level value of the spectrum is simply used as the characteristic parameter, and the relativization process or the process of obtaining the peak value of the frequency component from the predetermined frequency band as in the above-described embodiment is performed. For not doing
There is a problem in that the configuration and processing time of the entire model identification system becomes large because the feature parameters are not well organized, differences in aircraft types and flight forms are difficult to extract, and the number of feature parameters used is large.

Table 1 shows another example of the reference pattern of the characteristic parameters created for the example of the airfield shown in FIG. In this example, which uses a total of 10 patterns, all the flying aircraft are classified into 7 models shown in Table 2, other jet aircraft, and other propeller aircraft. A reference pattern and weighting factors were created.
Then, based on this, the result of trying the model discrimination processing using the observation data for a total of 5 days is shown in Table 3 as an identification errata. In Table 3, Out is the identification result by the method of the present invention, and In represents the true model. Also, T / O indicates takeoff and L / D indicates landing. Despite the fact that the number of feature parameters used for discrimination is only 10 in this way, a very good discrimination ability can be obtained.

In the embodiment of FIG. 4, the peak level value of the frequency component is obtained by selecting the 1/3 octave band filter having the center frequency f ₀ of f ₀ = 80 Hz to 125 Hz. Range 25Hz-8k
Multiple frequency bands may be set between Hz as necessary.

Further, for example, when there is a possibility that the low frequency component has a characteristic such as a transportation machine having a large turbofan type engine, it is sufficient to include up to that band.

Further, in the above-described embodiment, the 1/3 octave band filter is used to obtain the frequency component signal, but the present invention is not limited to this, and a wide frequency band such as a 1/1 octave band filter is used. You may also use these.

Further, in the above-mentioned embodiment, the reference relativization information is
Although a wide band filter having characteristics is used, the present invention is not limited to this, and has C characteristics (JIS standard, C1505-1988).
It is also possible to use the one having the characteristic of drooping in the low frequency band and the high frequency band. With this configuration, it is possible to suppress the disturbance generated in the relatively low frequency band and the high frequency band based on the weather condition or the like.

As is clear from the above description, the present invention can automatically and quickly identify the type and flight form of an aircraft only from the acoustic measurement of the noise generated by the aircraft, and its effect is There is a great one.

[Brief description of drawings]

FIG. 1 is a flow chart of an embodiment of the present invention, FIG. 2 is a flow chart of another embodiment, FIG. 3 is a point arrangement diagram of an example of noise measurement, and FIG. 4 is a feature parameter extraction feature. For the purpose of explanation, a diagram in which aircraft noise frequency analysis results are overlaid, FIG. 5 is a diagram showing an example of relative levels of each frequency, and FIG. 6 is a signal waveform diagram used for explaining characteristic parameters of duration. FIG. 7 is a signal waveform diagram for explaining characteristic parameters of time.

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Akinori Yokota, 3-20-41 Higashimoto-cho, Kokubunji-shi, Tokyo Inside Kobayashi Institute of Physical Sciences (72) Susumu Shimizu 3--20-41 Higashimoto-cho, Kokubunji, Tokyo Kobayashi Institute of Science (56) References JP-A-56-158400 (JP, A) JP-A-55-7749 (JP, A) JP-B-41-20126 (JP, B1)

Claims (2)

[Claims] (A) Obtaining a reference measurement signal that represents a change in aircraft noise while an aircraft of a known type passes through a measurement point or its vicinity, and determine a frequency signal component included in the reference measurement signal. By extracting for a plurality of frequency bands of, a plurality of reference frequency component signals representing changes in the frequency signal component while the aircraft of the known model passes through the measurement point or its vicinity, the plurality of reference frequencies The reference frequency component peak time at which the component signal peaks is held as the reference patterning information, and the reference passage time at which the reference measurement signal peaks is held, and the plurality of reference frequency component peak times and the reference passage time are held. (B) Detection of changes in aircraft noise while an aircraft of unknown model to be measured passes through the measurement point or its vicinity. Obtaining the measurement signal, by extracting the frequency signal component contained in the detection measurement signal for the predetermined plurality of frequency bands,
Obtaining a plurality of detection frequency component signals representing changes in the frequency signal components while the aircraft of the unknown type passes through the measurement point or its vicinity, and the detection frequency component peak time at which the plurality of detection frequency component signals become peaks Respectively hold as detection patterning information, hold the detection passage time at which the detection measurement signal reaches a peak, and obtain the time difference between the plurality of detection frequency component peak times and the detection passage time as a detection pattern, (c) An aircraft characterized by determining that the unknown model is the known model corresponding to the reference pattern when the degree of similarity is within a predetermined range by comparing the detection pattern with the reference pattern Model identification method. 2. (a) Obtaining a reference measurement signal that represents a change in aircraft noise while an aircraft of a known type passes through a measurement point or its vicinity, and determine a frequency signal component included in the reference measurement signal. By extracting for a plurality of frequency bands of, a plurality of reference frequency component signals representing changes in the frequency signal component while the aircraft of the known model passes through the measurement point or its vicinity, the plurality of reference frequencies For the threshold signal level that is reduced by a predetermined signal level from the signal level where the component signal peaks, the time during which each of the reference frequency component signals exceeds the threshold signal level is obtained as a reference pattern, and (b) is measured. A measurement signal representing the change in aircraft noise while an unknown type of aircraft should pass through the measurement point or its vicinity is included in the detection measurement signal. Frequency signal components by extracting the said predetermined plurality of frequency bands,
Obtaining a plurality of detection frequency component signals that represent changes in the frequency signal components while the aircraft of the unknown type passes through the measurement point or its vicinity, and determine a predetermined level from the signal level at which the plurality of detection frequency component signals peak. Only the signal level is reduced for the threshold signal level, the time during which each of the detection frequency component signals exceeds the threshold signal level is obtained as a detection pattern, and (c) by comparing the detection pattern with the reference pattern. An aircraft model identification method, characterized in that when the degree of similarity is within a predetermined range, the unknown model is determined to be the known model corresponding to the reference pattern.