DE102020103264B4 - Automated source identification from microphone array data - Google Patents

Abstract

Method for identifying real sources of noise on an object subject to a flow, with the steps of measuring sound pressures of sound emanating from the object subject to the flow, using a microphone array, with a Mach number of the flow being varied;- calculating a sound frequency and flow parameter-dependent spatial sound power total distribution from the measured sound pressures by beamforming and- identifying the real sound sources using spectral, parametric and spatial criteria from the spatial sound power total distribution, characterized in that the criteria include the spectral criterion that a spatial sound power distribution contributing to the sound power total distribution of a real sound source at different Frequencies of the sound occurs simultaneously - that the criteria include the parametric criterion that a sound power distribution contributing to the total sound power distribution ng of a real sound source with increasing Mach number of the inflow, and- that the criteria include the spatial criterion that a spatial sound power distribution of a real sound source contributing to the total sound power distribution calculated by beamforming is considered across all frequencies of the sound at which it occurs is a spatial normal distribution.

Description

TECHNICAL FIELD OF THE INVENTION

The invention relates to a method for identifying real sound sources on an object against which an inflow is flowing, from sound data measured with a microphone array. More specifically, the present invention relates to such a method with the steps: measuring sound pressures of sound emanating from the object being flown against, using a microphone array, with a Mach number of the flow being varied; Calculation of a sound frequency and flow parameter-dependent spatial sound power total distribution from the measured sound pressures by beamforming and identification of the real sound sources based on spectral, parametric and spatial criteria from the spatial sound power total distribution.

The real sound sources to be identified on the object against which the incoming flow flows are so-called aeroacoustic sources.

The total spatial sound power distribution calculated from the measured sound pressures by beamforming is also referred to as an acoustic source map or "dirty map".

STATE OF THE ART

For analysis, i. H. Identification and characterization of aeroacoustic sources are created with the help of microphone arrays and beamforming acoustic sound power total distributions, for example from aircraft or automobiles in wind tunnels or under real conditions. Due to the high dimensionality of the measurement conditions, the dimensions of which include not only spatial directions, time and frequency, but also the angle of attack and other flow parameters such as the Mach number, the Reynolds number of the flow, and the respective source type, such as monopole or dipole Extremely high expenditure of time and expert knowledge possible to analyze and interpret such total sound power distributions with regard to the aeroacoustic sources contributing to them.

During beamforming in the frequency domain, a total sound power distribution is calculated from a cross-spectral matrix of a microphone array, which displays the sound powers of all sound sources presumably contributing to the sound pressures measured with the microphone array as a function of location and frequency. In order to be able to characterize aeroacoustic sources, parameters such as the flow velocity or the Mach number are usually varied when measuring the sound pressure with the microphone array. Furthermore, vehicle or aircraft parameters such as a flap position, an angle of attack, rotor blade and side mirror configurations and the like can be changed in order to identify or minimize acoustically relevant problems, for example. The total sound power distributions are then evaluated manually and compared with each other. In the comparison, individual assumed aeroacoustic sources are analyzed under different inflow conditions and/or by considering typical source regions, such as e.g. B. the side, rear and front edges of existing objects compared.

The high dimensionality, i. H. the large number of parameters on which the sound power levels depend also forces experts to greatly reduce the amount of data during analysis and interpretation and to carry out the analysis in a geometry-driven manner. For example, the sound power levels are integrated over frequency and/or over location in order to understand global effects. After this, however, the total sound power distributions can only be interpreted qualitatively. An identification and characterization of all real sound sources is no longer possible.

In addition, experts characterize occurring sound powers by spatial allocation. So they assume a turbine noise when the sound power occurs at a turbine. With different parameters, however, different acoustic phenomena can dominate at the same location. It is often not possible to allocate the sound power that occurs in complex geometries.

In addition, artifacts, e.g. B. the effects of a point spread function (PSF) of beamforming, which can only be partially removed with deconvolution algorithms, and so there is a risk of incorrect interpretation as real sound sources. Not least for this reason, the total sound power distributions calculated by beamforming are referred to as "dirty maps".

A further disadvantage of the known interpretation of total sound power distributions is that parameter variations are normally only comparable within a data set of sound pressures with otherwise identical inflow conditions. In the wind tunnel tests, measurement series of wind tunnel tests must therefore be carried out in which, if possible, only one parameter is varied. For example, since different institutes specify different inflow velocities for wind tunnel tests, data sets of sound pressures measured on the same objects are rarely comparable.

From the DE 10 2008 053 302 A1 a device and a method for locating and separating a physical source in a wave field are known. Sensor signals are recorded by a number of sensors and are forwarded to a control/computer unit for evaluation. The control/computer unit shifts and/or processes the sensor signals to identify a first, for example strongest, source. Then it shifts the determined values of the sensor signals from the source to subtract from the totality of the sensor signals. The steps for identifying at least individual subsequent sources are then repeated. The steps for identification may include shifting in time, adjusting their amplitude and summing the sensor signals such that a maximum amplitude results for one of the strongest received first sources. The identification can take place in the time domain or, after transformation of the sensor signals, in the frequency domain.

Methods and devices for imaging acoustic sources are known from US 2006/0 256 975 A1, the sound of which is detected in a phase-sensitive manner using a microphone array. The distribution of the sound sources is determined by beamforming from the signals from the microphones. Beamforming effects are then removed from this distribution by deconvolution.

From the JP 2004 144 579 A it is known to optimize a distribution of sound pressure detection elements used for locating a switching source on a surface by using distributions of sound pressure across the sound pressure detection elements in an existing distribution to determine the locations of individual sources and these locations with the actual locations of the individual switching sources is compared.

From the CN 1 09 613 481 A a method for identifying sound sources by beamforming is known, which is adapted to the measurement environment in a wind tunnel by adapting the point transfer function on which beamforming is based to the actual point transfer function of the transmission of sound from a sound source with a known position to a microphone array in the wind tunnel.

A device for characterizing sound sources is known from US 2008/0 199 024 A1, which has a plurality of beam formers. The beamformers are each connected to a plurality of microphones that pick up sound emanating from a sound source at any position in a given space. Each of the beamformers determines a location and orientation of the sound source in the room. The positions and orientations are weighted based on the sound levels taken into account by the beamformer. The location and orientation with the highest weight is then assumed to be the true location and orientation of the sound source.

From the DE 197 36 021 A1 a method for determining acoustically relevant vibration sources that influence the noise level in the interior of a vehicle is known. The vehicle is flown against in a wind tunnel, and the vibrations occurring on the outer skin components of the vehicle under the influence of the relative wind are measured. A component of the outer skin of the vehicle, on which vibrations occur under the effect of the relative wind, is also specifically excited by a mechanical vibration exciter, and the resulting sound level in the interior of the vehicle is measured. The vibration excitation with the mechanical vibration exciter is used to determine the mechanical-acoustic transfer function of the exterior component in relation to the sound level in the interior and the mobility of the exterior component. The sound pressure in the interior of the vehicle caused by the airflow is approximately calculated by multiplying it by the vibration speeds that occur under the effect of the airflow. The vibrations occurring on the outer skin components are scanned with a laser beam or recorded by means of holographic examinations.

OBJECT OF THE INVENTION

The invention is based on the object of demonstrating a method for identifying real sound sources on an object subjected to an inflow, which can be automated and the results of which can be compared across different inflow parameters, in particular different inflow velocities.

SOLUTION

The object of the invention is achieved by a method for identifying real sound sources with the features of independent patent claim 1 . The dependent claims define preferred embodiments of the method according to the invention.

DESCRIPTION OF THE INVENTION

In the method according to the invention for identifying real sound sources on an object against which an inflow is flowing, the following steps are carried out: measuring sound pressures of sound emanating from the object in which the flow is blowing, using a microphone array, with a Mach number of the inflow being varied; Calculation of a sound frequency and flow parameter-dependent spatial sound power total distribution from the measured sound pressures by beamforming and identification of the real sound sources based on spectral, parametric and spatial criteria from the spatial sound power total distribution. The criteria include the spectral criterion that a spatial sound power distribution of a real sound source that contributes to the overall sound power distribution occurs at different frequencies simultaneously. The criteria include the parametric criterion that a sound power distribution of a real sound source that contributes to the total sound power distribution has increasing sound power levels as the Mach number of the inflow increases. The criteria also include the spatial criterion that a total spatial sound power distribution of a real sound source contributing to the total sound power distribution calculated by beamforming is a spatial normal distribution across all frequencies at which it occurs.

The considered sound power can be viewed directly or as a sound power level on a logarithmic scale. The sound power level is linearly dependent on the sound pressure level of the sound pressure measured at a fixed distance from the respective real sound source if at most linear attenuation occurs between the sound source and the location of the measurement. The sound power reflects the sound energy in watts radiated by the real sound source per unit of time. The sound power level Lw in decibels relates the sound power Pak according to: Lw= 10 Ig (P _ak /P ₀ ) dB to a reference value P ₀ = 10 ^-12 W standardized for airborne sound.

Based on the three criteria mentioned, an automated identification of the real sound sources of interest, i. H. from aeroacoustic sources, possible. For this purpose, the spatial sound power distributions that occur simultaneously at different frequencies can be marked in the total sound power distribution. It can then be checked for each of these marked spatial sound power distributions whether it has increasing sound powers with increasing Mach number of the inflow and whether the spatial sound power distributions are spatial normal distributions across all frequencies at which they occur.

To check whether this spatial criterion is met for a specific sound power distribution, the sound powers for each location must be integrated across the various frequencies at which the sound power distribution occurs, and the sound power distribution integrated across the frequency must then be checked to determine whether it is spatial normality qualified or not. For this purpose, the deviation from an ideal normal distribution can be quantified and compared with a limit value.

The spatial criterion of the normal distribution is not a criterion that applies to the actual spatial sound power distributions of real sound sources, but rather to the contributions of real sound sources to a total sound power distribution calculated by beamforming. In other words, the normal distributions, with which the spatial sound power distributions of real sound sources contribute to the total sound power distribution calculated by beamforming, are artifacts of beamforming that occur in this form with all real sound sources. It goes without saying that the normal distributions, with which the real sound sources contribute to the total sound power distribution, are only simple normals distributions around a point are real point sources of sound. Otherwise they are normal distributions around the dimensions of the real sound sources.

The above parametric criterion reflects that a real sound source becomes louder as the Mach number of the flow increases. This criterion applies to all aeroacoustic sources relevant here.

The spectral criterion mentioned above means that a real sound source occurs at different frequencies at the same time. Even if the real sound source has a dominant frequency, for example the natural frequency of a resonator, it also radiates sound power at other frequencies, for example the harmonics of the natural frequency and in the adjacent frequency bands.

When testing the method according to the invention, it was possible with the aid of the three criteria mentioned to automatically identify real sound sources from "dirty maps", i. H. without reducing the amount of data and without the help of an expert or other human intervention.

The result of the method according to the invention for identifying real sound sources can be a sound source map. Specifically, to create this sound source map, the sound powers of the spatial sound power distribution of an identified sound source at the respective frequency of the sound can be integrated over the location to form a sound power value, and this sound power value can be assigned to a spatial power focus of the spatial sound power distribution of the identified sound source at the respective frequency of the sound source map be assigned to sound. In other words, the sound power value for the respective frequency of the sound at the location of the power focus of the identified sound source is entered in the sound source map.

In order to be able to distinguish between real sound sources that are close to one another and possibly also overlapping one another in the method according to the invention, the additional spatial criterion can be applied that, with different Mach numbers of the inflow, overlapping sound power distributions that contribute to the total sound power distribution can be assigned to a real sound source.

When calculating the total sound power distribution from the measured sound pressures by beamforming, an acoustic point spread function, which is caused by the discrete arrangement of the microphones in the microphone array and whose expression depends on the respective microphone array, must often be compensated as completely as possible. In other words, the acoustic point spread function often deforms the total sound power distribution calculated only by beamforming to such an extent that it is impossible to identify real sound sources. To compensate for the acoustic point spread function, for example, a total raw sound power distribution initially calculated without taking the point spread function into account can be unfolded or processed in some other way with respect to the acoustic point spread function. As another example, a beamforming algorithm can be used to calculate the total sound power distribution that precompensates for the acoustic point spread function. Ideally, there is only one location per frequency of the sound emitted by the sound source in the calculated total sound power distribution for each actual location of a real sound source. With regard to these locations, the relevant spatial criterion is that a spatial sound power distribution of a real sound source that contributes to the total sound power distribution calculated by beamforming across all frequencies at which it occurs is a normal distribution. The compensation of the acoustic point spread function should therefore be carried out before identifying the real sound sources using the spatial criterion that a spatial sound power distribution of a real sound source contributing to the total sound power distribution calculated by beamforming is a spatial normal distribution across all frequencies at which it occurs , because this beamforming artifact may only be visible after compensating for the acoustic point spread function.

In order to characterize the real noise sources at the object over which the flow flows on the basis of the noise source map, in which the power of the noise sources at different frequencies and varied Mach number of the flow overflowing the object are entered, characteristics of various properties of the noise sources are determined. These various properties of the sound source are chosen from a group of properties that describe the acoustic properties of a real sound source. These include broadband self-similarity, Mach number scaling, tonality, tonal self-similarity, relative source motion, spatial extension, and shape of the spectrum. The features determined for each of the properties are each described by a single numerical value. The partial source map used as a basis could also have a different origin than the method according to the invention for identifying real sound sources. An overall sound source distribution, ie a so-called "dirty map" is unsuitable as a basis for the method according to the invention for characterizing real sound sources, since the real sound sources have not yet been identified and separated from the effects of beamforming and other artifacts that do not correspond to sound power of real sound sources are.

At least the numerical values for characteristics of three of the seven acoustic properties mentioned are calculated for the characterization of the real sound sources. Preferably, the numerical values relating to the characteristics of at least four, more preferably at least five, even more preferably at least six and most preferably all seven of said seven acoustic properties are calculated. Each of these properties has proven valuable for easily characterizing real-world sound sources by a few numerical values and distinguishing between different types of real-world sound sources.

As a feature of the broadband self-similarity of the respective sound source, a standard deviation of differences in the sound power of the sound source at different Mach numbers and at the same different frequencies of the sound can be calculated. Instead of or in addition to the different frequencies, different Strouhal numbers and/or Helmholtz numbers of the sound can also be taken into account. The Strouhal number Sr is defined according to Sr = fL/v as the quotient of the product of the frequency f of the sound and a variable L of the object in the flow on the one hand and the flow velocity v on the other. In contrast, the Helmholtz number He is defined as He = L/λ, where L is again a characteristic variable of the object in the flow and λ is the wavelength of the sound waves.

As a feature of the Mach number scaling of the respective sound source, an exponent of the Mach number of the inflow can be calculated, in which a squared sum of differences in the sound power of the sound source at different Mach numbers and at the same different frequencies of the sound assumes a minimum. Here too, instead of or in addition to the different frequencies, different Strouhal numbers and/or Helmholtz numbers of the sound can be taken into account.

Characteristics of the tonality of the respective sound source can be calculated specifically as a number and from relative sound powers, from interval widths and from sound frequencies of individual tones whose sound power exceeds broadband noise. Here, too, Strouhal numbers and/or Helmholtz numbers of the sound can be considered instead of or in addition to the frequencies. When calculating the tonality of the respective sound source, only individual tones that clearly stand out from noise are taken into account, i.e. H. only the identifiable peaks of the sound power spectrum, but basically independent of the width of the peaks.

Even more specifically, for the tonality of the respective sound source, in addition to the number, the parameters of gamma distribution functions of the relative sound powers, the interval widths and the frequencies of the sound of the individual tones with their sound power exceeding the broadband noise can be calculated. The distribution functions compensate for the fact that the number of individual tones varies from different sound sources and the individual tones have different interval widths, for example. The distribution function describes the characteristics of all individual tones regardless of their number. Again, the Strouhal numbers and/or Helmholtz numbers of the sound can be taken into account instead of or in addition to the frequencies. The gamma distribution functions are to be selected from the gamma distribution functions that are fully described by three independent parameters, shape, location and scale, and fitted to the relative sound power levels, interval widths and frequencies or Strouhal numbers or Helmholtz numbers of the individual tones.

The tonal self-similarity of the respective sound source can be calculated as the ratio of an overlap sum of overlaps of bandwidths of the individual tones at different Mach numbers to a total sum of the bandwidths of the individual tones. Depending on the consideration, the bandwidths can be determined in the frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound of the individual tones whose sound power exceeds the broadband noise. The tonal self-similarity thus indicates the relative extent to which the individual tones can be found in the spectra of the sound power at the different Mach numbers of the inflow.

The shape of the spectrum of the respective sound source can be calculated as the gradient parameter of a straight line fitted to the sound power of the sound source by linear regression over the frequency of the sound. The Strouhal number or Helmholtz number of the sound can also be taken into account here, but does not lead to any additional information. Alternatively or additionally, the shape of the spectrum of the respective sound source can be calculated as a coefficient of determination of the respective linear regression.

The relative source movement of the respective sound source can be calculated as the distance between different spatial positions of the sound source with different Mach numbers of the inflow, which is normalized to a difference between the different Mach numbers. Instead of the different Mach numbers, different frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound can also be considered. Then the distance between the different spatial positions is normalized to a difference between the different frequencies or Strouhal numbers or Helmholtz numbers of the sound. The spatial expansion of the respective sound source can be calculated specifically from parameters of a normal distribution of the spatial sound power distribution of the sound source. A spatial integral over the normal distribution of the spatial sound power distribution of the sound source can be used as the numerical value of the spatial extent.

The features of the sound sources determined by the method according to the invention for characterizing the sound sources can be normalized and orthogonalized. The sound sources can then be grouped according to their characteristics using a clustering method, in particular using a clustering method that does not require prior knowledge of the number of clusters, i. H. assigned to different groups. The clustering method can be a so-called density-based clustering method. With the help of the clustering method, groups of sound sources that are similar in each case with regard to their specific characteristics are automatically determined, with the sound waves of the various groups being dissimilar with regard to their specific characteristics. This, like calculating the numerical values for the various characteristics of the sound sources, does not require expert knowledge. Rather, the method according to the invention can be fully automated based on the above explanations, so that it runs without human intervention.

A largest of the various Mach numbers for which the sound pressures are measured with the microphone array in the method according to the invention or which are taken into account in the sound source map is regularly at least 5% larger than a smallest of the various Mach numbers. Often the largest is also at least 10% larger than the smallest of the various Mach numbers, and often the largest is at least 20% larger than the smallest of the various Mach numbers.

With regard to the various frequencies and/or Strouhal numbers and/or Helmholtz numbers considered, a largest one is regularly at least 10 times, typically at least 30 times and often at least 100 times larger than a smallest one. As a result, the sound is viewed in a broadband manner. All frequencies and the corresponding Strouhal numbers or Helmholtz numbers are often taken into account for which sound pressures are measured with the microphone array or for which data are contained in the sound source map.

Advantageous developments of the invention result from the patent claims, the description and the drawings.

The advantages of features and combinations of several features mentioned in the description are merely exemplary and can have an effect alternatively or cumulatively without the advantages necessarily having to be achieved by embodiments according to the invention.

The following applies to the disclosure content - not the scope of protection - of the original application documents and the patent: Further features can be found in the drawings - in particular the illustrated geometries and the relative dimensions of several components to one another as well as their relative arrangement and operative connection. The combination of features of different embodiments of the invention or of features of different patent claims is also possible, deviating from the selected dependencies of the patent claims and is hereby suggested. This also applies to those features that are shown in separate drawings or are mentioned in their description. These features can also be combined with features of different patent claims. Likewise, features listed in the patent claims can be omitted for further embodiments of the invention, but this does not apply to the independent patent claims of the granted patent.

The features mentioned in the patent claims and the description are to be understood with regard to their number in such a way that exactly this number or a larger number than the number mentioned is present, without the need for an explicit use of the adverb “at least”. If, for example, an object is mentioned, this is to be understood in such a way that exactly one object, two objects or more objects are present. The features listed in the patent claims can be supplemented by other features or can be the only features that the respective method has.

The reference signs contained in the claims do not limit the scope of the subject-matter protected by the claims. They only serve the purpose of making the claims easier to understand.

character list

• 1 illustriert eine durch Beamforming und Kompensation der einer Punktspreizfunktion des Beamforming berechnete räumlichen Schallleistungsgesamtverteilung an einem umströmten Tragflügel.
• 2 illustriert eine auf Clustering basierte Identifikation realer Schallquellen in der Schallleistungsgesamtverteilung gemäß 1, wobei Raum- und Frequenzmittelpunkte der in verschiedenen Graustufen wiedergegebenen Cluster durch Sterne gekennzeichnet sind.
• 3 erläutert die Anwendung des räumlichen Kriteriums, dass eine zu der durch Beamforming berechneten Schallleistungsgesamtverteilung beitragende räumliche Schallleistungsverteilung einer realen Schallquelle über alle Frequenzen, bei denen sie auftritt, hinweg betrachtet eine räumliche Normalverteilung ist.
• 4 zeigt die Auftragung von numerischen Werten dreier als Beispiel ausgewählter Merkmale der identifizierten realen Schallquellen eines exemplarischen Datensatzes. Die ausgewählten Merkmale erfüllen die Bedingung, für verschiedene Quelltypen unterschiedliche numerische Werte zu liefern, so dass anhand der Gesamtheit der numerischen Werte auf den Quelltyp geschlossen werden kann.
• 5 zeigt breitbandig machzahlskalierte Spektren einer realen Schallquelle bei drei verschiedenen Machzahlen. Die Quelle skaliert tieffrequent über die Strouhalzahl des Schalls mit der Machzahl hoch 3,7 und hochfrequent über die Helmholtzahl des Schalls mit der Machzahl hoch 6,0.
• 6 zeigt mit dem erfindungsgemäßen Verfahren an einem Tragflügel eines Verkehrsflugzeugs identifizierte reale Schallquellen und deren mittlere Quellbewegungen mit der Machzahl.
• 7 illustriert die Bestimmung der Merkmale zu der Eigenschaft „Tonalität“ bei dem erfindungsgemäßen Verfahren.
• 8 zeigt das Ergebnis eines Clusterings auf Basis von Merkmalen identifizierter realer Schallquellen bei dem erfindungsgemäßen Verfahren.
• 9 ist ein Ablaufdiagramm der erfindungsgemäßen Verfahren zur Identifizierung und Charakterisierung realer Schallquellen.

The invention is further explained and described below with reference to preferred exemplary embodiments illustrated in the figures.

• 1 illustrates a total spatial sound power distribution on an airfoil calculated by beamforming and compensation of a point spread function of the beamforming.
• 2 illustrates a clustering-based identification of real sound sources in the total sound power distribution according to 1 , where spatial and frequency centers of the clusters reproduced in different shades of gray are marked by asterisks.
• 3 explains the application of the spatial criterion that a spatial sound power distribution of a real sound source that contributes to the total sound power distribution calculated by beamforming across all frequencies at which it occurs is a spatial normal distribution.
• 4 shows the plot of numerical values of three characteristics selected as examples of the identified real sound sources of an exemplary data set. The selected characteristics fulfill the condition of supplying different numerical values for different source types, so that the source type can be deduced from the total of the numerical values.
• 5 shows broadband Mach number-scaled spectra of a real sound source at three different Mach numbers. The source scales low-frequency via the Strouhal number of the sound with a Mach number to the power of 3.7 and high-frequency via the Helmholt number of the sound with a Mach number to the power of 6.0.
• 6 shows real sound sources identified with the method according to the invention on a wing of a commercial aircraft and their mean source movements with the Mach number.
• 7 illustrates the determination of the characteristics of the property "tonality" in the method according to the invention.
• 8th shows the result of clustering based on characteristics of identified real sound sources in the method according to the invention.
• 9 FIG. 12 is a flow chart of the methods according to the invention for identifying and characterizing real sound sources.

FIGURE DESCRIPTION

The methods according to the invention for identifying and characterizing real sound sources are based on conventional beamforming and unfolding methods, which are applied to sound pressures of sound emanating from an object being flown against, which are measured with a microphone array. Here, measured values of the sound pressures for different Mach numbers of the inflow of the object flown against are taken into account. The beamforming and unfolding methods used can be any method that uses the measured sound pressures to determine unknown sound sources on the object being flown against. These include direct methods such as CBF (conventional beamforming), FB (functional beamforming), deconvolution methods (DAMAS, DAMAS-2, DAMAS-2.1, DAMAS-C), data-dependent beamforming algorithms such as RAB (robust adaptive beamforming) and inverse Methods using the Tikhonov regularization of different norms (e.g. L_0, L_1,L_2) of the minimization problem ||p-Gq|| to solve. The beamforming and deconvolution methods calculate an acoustic frequency and flow parameter dependent spatial total sound power distribution. Real sound sources are identified from this total sound power distribution. To characterize the sound sources, numerical values of various selected features of the sound sources are calculated, and then the sound sources are grouped, in particular using a method of unsupervised learning, on the basis of their numerical values of the various features, i.e. combined into different groups of similar real sound sources. The calculated numerical values of the characteristics are independent of the flow velocity and thus make measurements at different Mach numbers directly comparable. Since all steps of the method according to the invention for identifying and characterizing real sound sources can be automated, the data from large series of measurements with many parameter variations can also be analyzed quickly and comprehensively. This also allows statistically secured statements about the individual real sound sources and sound source types, which have an increased informative value. In addition, the numerical values that characterize the real sound sources of a group, i.e. a type of sound source, are meta-information that can be anonymized in order to provide comparative data without revealing sensitive data related to the specific object in the flow.

The method according to the invention for identifying real sound sources meets the requirement of automatically identifying aeroacoustic sources as such and distinguishing them from beamforming and unfolding artifacts. With an inflow velocity, a real sound source is measured with the microphone array only as a realization of a continuous, physical mechanism. In the case of aeroacoustic sources, this physical mechanism changes monotonically as a function of the inflow velocity. This makes it possible to first identify and then characterize the real acoustic sources based on their change over the inflow velocity. In particular, the sound power of an aeroacoustic source depends on the frequency of the sound and the Mach number of the inflow. Sound sources that occur at the same location but whose sound power depends on other parameters such as angle of attack, Reynolds number differences and geometry variations are separate sound sources. The method according to the invention for identifying real sound sources makes use of the fact that real aeroacoustic sources exist in a wide frequency range or at least in several frequencies of the sound and become louder with increasing inflow speed. Furthermore, side lobes of a beamforming point spread function are frequency dependent and they migrate locally with increasing frequency of the sound, while the spatial position of a real aeroacoustic source varies only slightly. In this way, real sound sources can be identified using statistical methods from the total sound power distribution unfolded with regard to beamforming, i.e. a so-called “dirty map”. All points of the total sound power distribution that fulfill the spectral criterion that a spatial sound power distribution of a real sound source that contributes to the total sound power distribution occurs at different frequencies at the same time, and that fulfill the parametric criterion that a sound power distribution of a real sound source that contributes to the total sound power distribution with increasing Mach number of the flow has increasing sound power, are assigned to a sound power distribution of a real sound source using a density-based clustering method that does not require prior knowledge of the number of clusters, i.e. the number of aeroacoustic sources. The spatial criterion used is that a spatial sound power distribution of a real sound source that contributes to the total sound power distribution calculated by beamforming across all frequencies at which it occurs is a spatial normal distribution. This procedure enables an automated, precise identification of the aeroacoustic sources and their differentiation from beamforming artifacts such as noise or the point spread function. The focal points of the spatial sound source distributions of the individual real sound sources are then assigned for each sound frequency to the integrated sound power of the aeroacoustic source over the location. In this way, a spectrum is determined for each individual real sound source instead of over larger spatial integration areas without considering the sound source distribution. The spatial localization of the spatial sound sources is more precise than the resolution of the spatial total sound power distribution determined by beamforming, since the focal points of the sound power distributions are determined statistically.

Due to the probability density function of the normal distributions, even slightly overlapping real sound sources can be separated and the spatial expansion of the individual sound sources can be reconstructed.

In the 1 The spatial sound power distribution shown on an airfoil around a wing was calculated using "conventional beamforming" and "Clean-SC" from the sound pressures measured with a microphone array of sound emanating from the wing. In 1 is the frequency plotted against the extension of the wing in the x and y directions, and each location at which a sound power occurs occurs is marked in the space spanned in this way. The sound power density (Power Spectral Density = PSD) is displayed in dB in grayscale for each sound power that occurs.

In the same room as he is in 1 is stretched shows 2 the locations at which sound powers of potential real sound sources on the airfoil in the spatial sound power total distribution according to 1 appear. These locations are assigned to various potential real sound sources by clustering, ie cluster formation. The locations at which sound powers from a potential real sound source occur are shown in a shade of gray, and the respective spatial focus of these locations of a cluster, ie a potential real sound source, is indicated with a star.

3 explains the application of the spatial criterion that a spatial sound power distribution of a real sound source that contributes to the total sound power distribution calculated by beamforming across all frequencies at which it occurs is a spatial normal distribution when identifying individual real sound sources whose sound powers in 2 are included. The number of times sound power can be found at the corresponding coordinate over all frequencies is plotted logarithmically in greyscale at the top left. Level lines of the true distribution (dashed) and the fitted distribution (continuous) of the sound powers considered across all frequencies are shown at the top right. The real histogram values are plotted against the fitted distribution on the main axes of the distribution function at the bottom left. At the bottom right, the histogram value is plotted against the fitted distribution value for all points in the image section. In the case of ideal normal distributions, all points lie on the straight line shown. The distance to the center of the distribution is encoded in shades of gray. With the automatic application of the criterion explained here, it is checked whether a potential real sound source according to 2 maintains a specified maximum deviation of its considered spatial sound power distribution from a normal distribution.

• - Die Merkmale müssen geeignet sein, grundlegende Eigenschaften von aeroakustischen Quellen zu beschreiben.
• - Die Merkmale sollten sich möglichst wenig überschneiden.
• - Die Merkmale müssen automatisch bestimmbar sein.
• - Jedes Merkmal wird durch einen einzigen numerischen Wert beschrieben.

Every aeroacoustic source has acoustic properties to which features can be assigned that describe the sound source abstractly through meta information. In the method according to the invention for characterizing real sound sources, a selection of such properties and features is used that meet the following conditions:

• - The characteristics must be suitable for describing basic properties of aeroacoustic sources.
• - The characteristics should overlap as little as possible.
• - The characteristics must be automatically determinable.
• - Each characteristic is described by a single numerical value.

• - breitbandige Selbstähnlichkeit bei Frequenz, Helmholtzzahl oder Strouhalzahl des Schalls,
• - Machzahlskalierung bei Frequenz, Helmholtzzahl oder Strouhalzahl des Schalls,
• - Tonalität bei Frequenz, Helmholtzzahl oder Strouhalzahl des Schalls, ausgedrückt durch
  • • Anzahl der Einzeltöne,
  • • relative Tonlautstärke, Parameter „Shape“ einer an die Schallleistungen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • relative Tonlautstärke, Parameter „Location“ einer an die Schallleistungen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • relative Tonlautstärke, Parameter „Scale“ einer an die Schallleistungen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Frequenz der Einzeltöne, Parameter „Shape“ einer an die Frequenzen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Frequenz der Einzeltöne, Parameter „Location“ einer an die Frequenzen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Frequenz der Einzeltöne, Parameter „Scale“ einer an die Frequenzen der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Intervallbreite, Parameter „Shape“ einer an die Intervallbreiten der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Intervallbreite, Parameter „Location“ einer an die Intervallbreiten der Einzeltöne angefitteten Gammaverteilungsfunktion,
  • • Intervallbreite, Parameter „Scale“ einer an die Intervallbreiten der Einzeltöne angefitteten Gammaverteilungsfunktion,
• - tonale Selbstähnlichkeit bei Frequenz, Helmholtzzahl oder Strouhalzahl des Schalls,
• - relative Quellbewegung,
• - räumliche Ausdehnung, ausgedrückt durch die Parameter
  • • Kompaktheit,
  • • räumlicher Quelltyp,
• - und Form des Spektrums, ausgedrückt durch die Parameter
  • • Steigung der linearen Regression,
  • • Bestimmtheitsmaß der linearen Regression.

These requirements are met by the following properties and associated characteristics:

• - broadband self-similarity in terms of frequency, Helmholtz number or Strouhal number of the sound,
• - Mach number scaling at frequency, Helmholtz number or Strouhal number of the sound,
• - Tonality at frequency, Helmholtz number or Strouhal number of the sound, expressed by
  • • number of single tones,
  • • Relative tone volume, "Shape" parameter of a gamma distribution function fitted to the sound power of the individual tones,
  • • Relative tone volume, "Location" parameter of a gamma distribution function fitted to the sound power of the individual tones,
  • • Relative tone volume, "Scale" parameter of a gamma distribution function fitted to the sound power of the individual tones,
  • • Frequency of the individual tones, "Shape" parameter of a gamma distribution function fitted to the frequencies of the individual tones,
  • • Frequency of the individual tones, "Location" parameter of a gamma distribution function fitted to the frequencies of the individual tones,
  • • Frequency of the individual tones, "Scale" parameter of a gamma distribution function fitted to the frequencies of the individual tones,
  • • Interval width, "Shape" parameter of a gamma distribution function fitted to the interval widths of the individual tones,
  • • Interval width, "Location" parameter of a gamma distribution function fitted to the interval widths of the individual tones,
  • • Interval width, "Scale" parameter of a gamma distribution function fitted to the interval widths of the individual tones,
• - tonal self-similarity in terms of frequency, Helmholtz number or Strouhal number of the sound,
• - relative source movement,
• - Spatial extent expressed by the parameters
  • • compactness,
  • • spatial source type,
• - and shape of the spectrum expressed by the parameters
  • • slope of the linear regression,
  • • Coefficient of determination of the linear regression.

In 4 shows the distributions of three exemplary features listed above for an exemplary data set with different types of aeroacoustic sources. Specifically, these are the characteristics “standard deviation of differences in the sound power of the respective sound source with different Mach numbers and with the same different Helmholtz numbers of the sound” of the property “broadband self-similarity”, “ratio of an overlap sum of overlapping bandwidths of the individual tones with different Mach numbers a total sum of the bandwidths of the individual tones determined when considering the "Strouhal numbers of the sound" of the property "tonal self-similarity" and "measure of linear regression" of the property "shape of the spectrum". From the plots it can be seen that none of the features on their own are suitable for distinguishing the different types of aeroacoustic sources. However, this is possible through their interaction.

The individual features are explained in more detail below.

Broadband self-similarity

A spectrum of an aeroacoustic source behaves partially self-similarly at different Mach numbers. The spectra of the sound power over the frequency, the Helmholtz number and/or the Strouhal number of the sound can be self-similar. The broadband self-similarity is the measure of the degree to which the spectra are self-similar at different Mach numbers for the respective type of frequency, i.e. frequency, Helmholtz number or Strouhal number of the sound. "Broadband" in this context indicates that the self-similarity is specified for the entire spectrum of sound power, or at least for a relevant part of the spectrum. The self-similarity of tonal portions of the spectrum is treated as a separate feature of tonal self-similarity. 5 shows broadband Mach number scaled spectra of an aeroacoustic source at three different Mach numbers. The sound power of the aeroacoustic source scales low-frequency via the Strouhal number of the sound with the Mach number raised to the power of 3.7 and high-frequency via the Helmholtz number with the Mach number raised to the power of 6.0.

The broadband self-similarity can be described, for example, by the uniformity of the change in sound power over the respective frequency in the spectrum. Mathematically, the broadband self-similarity Si with i={frequency, Helmholtz number, Strouhal number} with standard deviation σ is then Si = σ _i ((σ _M (L _W ( _i ,M))), where L _w is the sound power level and M is the Mach number is.

Mach number scaling

Mach number scaling describes how much the sound power of a sound source increases with increasing inflow velocity. The M ⁿ law is used for this, i.e. the sound power changes exponentially with the Mach number of the inflow velocity to the power n. In order to determine the numerical value n, the distance between the Mach number normalized spectra L̂ is minimized with L̂ = L - 10 log ₁₀ ( ^Mn ). In order to correctly determine the numerical value of the exponent n, the spectra must be self-similar, see 5 .

Relative Source Weaung

des Merkmals der mittlere Quellbewegung mit der Machzahl $${\langle \mathrm{\Delta }x\rangle }_M=\langle \frac{{\overrightarrow{x}}_{i+1}-{\overrightarrow{x}}_i}{M_{i+1}-M_i}\rangle i$$

An aeroacoustic source does not have to be spatially fixed via the Mach number, it can be identified at different locations with varying flow velocities. The relative source movement describes the extent to which the aeroacoustic source moves between different Mach numbers of the inflow. As a first approximation, the source only moves downstream as the Mach number increases. For i = N - 1 different Mach numbers is the numerical value ${〈\mathrm{\Delta }x〉}_M$

the characteristic of the mean source movement with the Mach number $${〈\mathrm{\Delta }x〉}_M=〈\frac{{\overrightarrow{x}}_{i+1}-{\overrightarrow{x}}_i}{M_{i+1}-M_i}〉i$$

des Merkmals der mittlere Quellbewegung mit der Frequenz $${\langle \mathrm{\Delta }x\rangle }_{ƒ}=\langle \frac{{\overrightarrow{x}}_{i+1}-{\overrightarrow{x}}_i}{{ƒ}_{i+1}-{ƒ}_i}\rangle i$$

The source can also be identified at different frequencies at different points in space. For i = N - 1 different frequencies is the numerical value ${〈\mathrm{\Delta }x〉}_{ƒ}$

the characteristic of mean source movement with frequency $${〈\mathrm{\Delta }x〉}_{ƒ}=〈\frac{{\overrightarrow{x}}_{i+1}-{\overrightarrow{x}}_i}{{ƒ}_{i+1}-{ƒ}_i}〉i$$

des Merkmals der mittlere Quellbewegung mit der Machzahl aufgetragen. 6 shows real sound sources identified with the method according to the invention on a wing of a commercial aircraft. Each sound source is numbered and given the number is the magnitude of the numerical value ${〈\mathrm{\Delta }x〉}_M$

of the characteristic of the mean swelling motion plotted against the Mach number.

tonality

The tonality describes which frequency components are present in a sound power distribution of a real sound source compared to broadband noise. A single tone contained in a sound power distribution has several characteristics: the frequency or center frequency, the interval width, the absolute sound power and the relative sound power compared to broadband noise. Due to the fact that a sound power distribution can contain several tonal components with corresponding characteristics, but a single value is required as a characteristic, the characteristics are described by a statistical distribution. This distribution can be described independently of the number of random samples, i.e. the number of individual tones, using the mean value and standard deviation. To describe the tonality of each aeroacoustic source using features, the number of individual tones contained (single value) and the relative sound power of the individual tones, the interval width of the individual tones and the frequencies at which the individual tones occur (distribution function) are used. Among other things, gamma distribution functions are suitable here, which can be completely described by numerical values of three independent parameters, Shape, Location and Scale.

7 shows the spectrum of the sound power density at three different Mach numbers as well as the local maxima of the sound power density found therein, which are interpreted as individual tones. The relative sound powers (bottom left), the interval widths (bottom middle) and the occurring Strouhal numbers (bottom right) are shown as histograms or their approximated gamma distribution functions.

Tonal self-similarity

Dabei kann zur Berechnung der tonalen Selbstähnlichkeit eine untere Schranke der Schallleistung der Einzeltöne festgelegt werden, so dass nur prägnante Einzeltöne in diese Berechnung einfließen. Analog zur breitbandigen Selbstähnlichkeit muss die tonale Selbstähnlichkeit separat für alle Frequenztypen, das heißt, Frequenz, Helmholtzzahl und Strouhalzahl des Schalls, geprüft werden.The tonal self-similarity describes the self-similarity of the tonal parts of the spectra of the sound power distribution of a real sound source. To do this, the intervals of the determined individual tones are compared. A single tone is self-similar if the spectra at other Mach numbers in the same subinterval Î _i have a single tone. The tonal self-similarity St then results with $S_t=\frac{{\displaystyle {\sum }_i{\widehat{I}}_i}}{{\displaystyle {\sum }_i{I}_i}}.$

In order to calculate the tonal self-similarity, a lower limit of the sound power of the individual tones can be defined so that only concise individual tones are included in this calculation. Analogous to broadband Self-similarity, tonal self-similarity must be checked separately for all frequency types, that is, frequency, Helmholtz number and Strouhal number of the sound.

spatial extent

mit $$a=\frac{co{s}^2\theta }{2{\sigma }_{x_1}^2}+\frac{si{n}^2\theta }{2{\sigma }_{x_2}^2},b=-\frac{sin2\theta }{4{\sigma }_{x_1}^2}+\frac{sin2\theta }{4{\sigma }_{x_2}^2},c=\frac{si{n}^2\theta }{2{\sigma }_{x_1}^2}+\frac{co{s}^2\theta }{2{\sigma }_{x_2}^2}.$$

The spatial expansion describes the spatial distribution of the sound power. In this way, for example, punctiform aeroacoustic sources can be distinguished from linear aeroacoustic sources. The spatial expansion of a sound source is estimated from parameters of the normal distribution of the sound power distribution. In a two-dimensional view, for example, according to $$ƒ\left(x_1,x_2\right)=Aexp\left(-\left(a{\left(x_1-\widehat{x_1}\right)}^2+2b\left(x_1-\widehat{x_1}\right)\left(x_2-\widehat{x_2}\right)+c{\left(x_2-\widehat{x_2}\right)}^2\right)\right),$$

With $$a=\frac{cO{s}^2\mathrm{<figure-callout\; id="2"\; label="\theta "\; filenames="DE102020103264B4\_0012.png,DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{2{\sigma }_{x_1}^2}+\frac{si{n}^2\mathrm{<figure-callout\; id="2"\; label="\theta "\; filenames="DE102020103264B4\_0012.png,DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{2{\sigma }_{x_2}^2},b=-\frac{sin2\mathrm{<figure-callout\; id="4"\; label="\theta "\; filenames="DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{4{\sigma }_{x_1}^2}+\frac{sin2\mathrm{<figure-callout\; id="4"\; label="\theta "\; filenames="DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{4{\sigma }_{x_2}^2},c=\frac{si{n}^2\mathrm{<figure-callout\; id="2"\; label="\theta "\; filenames="DE102020103264B4\_0012.png,DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{2{\sigma }_{x_1}^2}+\frac{cO{s}^2\mathrm{<figure-callout\; id="2"\; label="\theta "\; filenames="DE102020103264B4\_0012.png,DE102020103264B4\_0016.png"\; state="\{\{state\}\}">\theta </figure-callout>}}{2{\sigma }_{x_2}^2}.$$

The standard deviations σ _i are orthogonal. The ratio of the standard deviations thus indirectly reflects the type of sound source (point or line source), the integral over the normalized normal distribution function (with A=1) provides information about the compactness of the source.

shape of the spectrum

Spectrum shape is an abstract feature that describes the shape of the spectrum over the frequency of the sound. A linear regression over the frequency y(f) = af + b is suitable for this. The slope parameter a reflects the mean increase or decrease in the sound power level, while the R ² value, ie the coefficient of determination of the linear regression, describes how well the data follow the regression.

The numerical values of the individual features are used for clustering (unsupervised learning) of the real sound sources. For this purpose, the features are first normalized and orthogonalized, for example using standard methods known to those skilled in the art, such as “whitening” and “principal component analysis”. Since the characteristics of different aeroacoustic sources scatter to different extents, density-based methods, for example Gaussian Mixture, are particularly suitable for clustering, for example using standard methods known to the person skilled in the art, such as “Gaussian Mixture”. The result of the clustering is an automatically created list of the real sound sources, grouped according to their characteristics. So shows 8th the result of a clustering with n=6 clusters for different angles of attack of an inflow of a wing. The numbers and associated shades of gray correspond to the groups of real sound sources, with -1 indicating no real sound source. The clustering method used had no information about the location of the sound sources or the angle of attack. Clustering allows individual conspicuous sound sources to be identified immediately. Furthermore, the characteristics can be statistically evaluated within a group with mean and standard deviation and thus have an increased meaningfulness. At the same time, different measurements, configurations and objects can be automatically compared with each other. Information about the source location or the parameter configuration, for example the angle of attack, can be deliberately omitted so that the sound sources are grouped solely on the basis of their acoustic properties.

Semi-supervised learning methods are also possible, in which a data set with known and labeled standard sound sources, e.g. cavity, cylinder in flow, free jet, etc., or objects, side mirrors, flap, rudder, etc., is added to the data to be analyzed. Similar sources are then directly assigned to these known aeroacoustic sources.

9 gives an overview of the implementation of the method according to the invention. They start with the step of measuring the sound pressures of sound emanating from the object being flown against, using a microphone array, with the Mach number of the flow being varied. After a total sound power distribution has been calculated, at least the parametric criterion of the sound powers increasing with the Mach number, the spectral criterion of the occurrence of a real sound source at different frequencies and the spatial criterion of the spatial normal distribution in the total sound power distribution calculated by beamforming to identify the real sound sources. A sound source map is created for the identified real sound sources. Based on the sound source map, numerical values of the various features relating to the various properties of the real sound sources are calculated in order to characterize them. Based on the calculated numerical values, the sound sources are clustered into different groups.

Claims (15)

Method for identifying real sources of sound on an object against which an inflow is flowing, with the steps of measuring sound pressures of sound emanating from the object in which the flow is blowing, using a microphone array, with a Mach number of the inflow being varied; - Calculation of a sound frequency and flow parameter-dependent spatial sound power total distribution from the measured sound pressures by beamforming and - Identification of the real sound sources using spectral, parametric and spatial criteria from the spatial sound power total distribution, characterized in that - the criteria include the spectral criterion that one of the The spatial sound power distribution of a real sound source that contributes to the total sound power distribution occurs simultaneously at different frequencies of the sound, - that the criteria include the parametric criterion that a sound power distribution of a real sound source that contributes to the total sound power distribution shows increasing sound power levels with increasing Mach number of the inflow, and - that the criteria include the spatial Criterion include that a spatial S contributing to the total sound power distribution calculated by beamforming sound power distribution of a real sound source across all sound frequencies at which it occurs is a spatial normal distribution. procedure after claim 1 , characterized in that sound powers of the spatial sound power distribution of an identified sound source at the respective frequency of the sound are integrated to form a sound power value which is assigned in a sound source map to a spatial power focus of the spatial sound power distribution of the identified sound source at the respective frequency of the sound. Method according to one of the preceding claims, characterized in that the criteria include the spatial criterion that with different Mach numbers of the inflow contributing to the overall sound power distribution and spatially overlapping sound power distributions are to be assigned to a real sound source. Method according to one of the preceding claims, characterized in that the sound frequency and flow parameter-dependent spatial total sound power distribution is calculated from the measured sound pressures by compensating for the acoustic point spread function. procedure after claim 2 or one of the claims 3 and 4 , as far as this on claim 2 are back-related, characterized in that the sound power levels of the sound sources at different frequencies of the sound and varied Mach number of the flow flowing over the object are entered in the sound source map and that, for the characterization of the real sound sources on the object over which the flow flows, on the basis of the sound source map, characteristics of different Properties of the sound sources are determined, - the various properties of the sound sources being selected from a group of acoustic properties which include - broadband self-similarity, - Mach number scaling, - tonality - tonal self-similarity, - relative source movement, - spatial expansion, - as well as a form of the spectrum, and - wherein the characteristics determined for each of the properties are each described by a single numerical value. procedure after claim 5 , characterized in that the features of the broadband self-similarity of the respective sound source are calculated as standard deviations of differences in the sound power of the sound source - at different Mach numbers and - at the same different frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound. procedure after claim 5 or 6 , characterized in that the features of the Mach number scaling of the respective sound source are calculated as exponents of the Mach number of the inflow, in which a square sum of differences in the sound power of the sound source - at different Mach numbers and - at the same different frequencies and / or Strouhal numbers and / or Helmholtz numbers of sound assumes a minimum. Procedure according to one of Claims 5 until 7 characterized in that the characteristics of the tonality of the respective sound source are calculated from a number, from relative sound powers, from interval widths and from frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound of individual tones whose sound power exceeds broadband noise. procedure after claim 8 , characterized in that the characteristics of the tonality of the respective sound source are calculated as the number and as parameters of gamma distribution functions of the relative sound powers, the interval widths and the frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound with their sound power over the broadband individual tones that go beyond hissing noise. Procedure according to one of Claims 8 and 9 , characterized in that the characteristics of the tonal self-similarity of the respective sound source are calculated as ratios of overlapping sums of overlapping bandwidths of the individual tones at different Mach numbers to total sums of the bandwidths of the individual tones, the bandwidths in the frequencies and/or Strouhal numbers and/or Helmholtz numbers of sound of the individual tones whose sound power exceeds the broadband noise can be determined. Procedure according to one of Claims 5 until 10 , characterized in that the features of the shape of the spectrum of the respective sound source are calculated - as a gradient parameter of a straight line fitted to the sound power of the sound source by linear regression over the frequency or Strouhal number or Helmholtz number of the sound and - as a coefficient of determination of the linear regression. Procedure according to one of Claims 5 until 11 , characterized in that the characteristics of the relative source movement of the respective sound source are calculated as distances between different spatial positions of the sound source at different Mach numbers of the flow or at different frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound, which indicate a difference between the different Mach numbers of the inflow or the various frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound are normalized. Procedure according to one of Claims 5 until 12 , characterized in that the features of the spatial expansion of the respective sound source are calculated from the parameters of a normal distribution of the spatial sound power distribution of the sound source, with optional spatial compactness of the sound source as an integral over the normalized normal distribution and an acoustic source type of the sound source as a ratio of orthogonal standard deviations of the normal distribution is calculated. Procedure according to one of Claims 5 until 13 , characterized in that the characteristics of the sound sources are normalized and orthogonalized and the sound sources are then grouped according to their characteristics using a clustering method, which optionally does not require prior knowledge of the number of clusters. Procedure according to one of Claims 1 until 14 , characterized in that a largest is greater by at least 5% or 10% or 20% than a smallest of the various Mach numbers and/or - that a largest is at least 10 times or 30 times or 100 times larger than a smallest of the different frequencies and/or Strouhal numbers and/or Helmholtz numbers of the sound.

Images