A method for decomposing a noise source of an aeroengine component

Abstract

This invention discloses a method for decomposing noise sources in aero-engine components, comprising the following steps: S1, spatial sound source analysis; acquiring noise input at each measurement point, constructing a covariance matrix and performing singular value decomposition; obtaining the number of eigenvalues ​​reaching a set contribution threshold; S2, extracting broadband time-domain components and pure audio-domain components; removing discrete pure-tone periodic components to obtain stationary random broadband components; processing the total noise and broadband noise separately using a Hanning window-applied discrete Fourier transform and subtracting them to obtain the spectrum of the pure-tone noise components; S3, adaptive signal decomposition filtering; S4, sound source directivity identification; S5, sound source separation; S6, noise component identification; S7, amplitude calibration; S8, sound source merging. This invention can accurately decompose fan noise and jet noise from engine noise, enabling accurate assessment of the contribution of modern turbofan engine noise sources at different frequencies and angles.

Description

Technical Field

[0001] This invention relates to the field of aero-engine acoustics technology, and more specifically to a method for decomposing noise sources in aero-engine components. Background Technology

[0002] Engines are the primary noise source of aircraft, contributing up to 80% of the overall noise during takeoff and landing. Therefore, engine noise control is crucial in aeroacoustic design. Engine noise primarily originates from two sources: fan noise, stemming from the periodic interference of rotating turbines with the air or exhaust gases; and jet noise, arising from the turbulent pulsations created by the rapid mixing of high-speed jet streams with the surrounding medium when the jet enters a stationary or slow-moving airflow. Other noise sources include combustion noise and core engine noise. Because engine components work collaboratively, the noise generated by each component is difficult to separate through direct testing; therefore, appropriate engine noise source assessment methods must be designed.

[0003] Current engine noise source assessment methods mainly include cross-spectral frequency domain assessment methods based on power spectrum and cross-correlation between measurement channels. This method yields a continuous spectrum of fan noise and jet noise, and is the continuous spectrum-level decomposition method currently used by engine manufacturers. Another method is frequency domain assessment based on engine component noise prediction models, which can achieve 1 / 3 octave band level decomposition. For cross-spectral methods, due to errors in the calculation of self and cross-power spectra, the mechanical pure tone components in fan noise will significantly leak into the jet noise, affecting the accuracy of the results. Furthermore, due to the influence of frequency domain transformation parameters, the method has low resolution, making it difficult to meet the need for accurate noise contribution assessment. The method based on engine component noise prediction models not only requires extensive prior knowledge of engine design characteristics and accuracy limitations, but can only roughly obtain the noise sound pressure level at the center frequencies of each 1 / 3 octave band from 50 to 10000 Hz, failing to determine the accurate dominant frequency and corresponding amplitude.

[0004] Because existing methods have insufficient resolution and are difficult to apply universally to engines with different design features and operating conditions, it is necessary to develop an evaluation method that considers the generation mechanism and frequency distribution differences of the main sound sources of the engine, as well as the theory of maximum independence of sound sources, to refine the decomposition and contribution analysis of the noise sources of the main engine components. This method would provide a high-resolution and universally applicable experimental data analysis technique for carrying out low-noise design and noise airworthiness compliance verification of engines.

[0005] In existing technologies, methods such as the 3-microphone cross-spectral frequency domain method and the noise prediction model-based method are used to separate fan noise and jet noise. The 3-microphone cross-spectral frequency domain method assumes that the fan noise and other turbine rotation interference sound sources have strong frequency domain coherence because their sound source positions are fixed, while the jet noise source is constantly moving and therefore has almost no frequency domain coherence. When the directional measuring points arranged around the engine are sufficiently far apart to satisfy the above assumptions, the frequency domain coherence relationship of the attenuated fan noise at each measuring point and the energy conservation relationship of the total noise, fan noise and jet noise can be used to subtract the highly coherent fan noise from the total noise spectrum to obtain the spectra of the fan and jet noise at the measuring points. The noise prediction model-based method is based on a forward prediction engine noise source decomposition method. It uses the total noise spectrum obtained from ground static tests and the jet noise spectrum prediction based on parameters such as fan speed and number of blades. It also considers the characteristic that jet noise dominates at lower frequencies, while its contribution decreases significantly with increasing frequency in the hundreds of hertz range. Assuming the existence of a cutoff frequency below the blade passage frequency, it is assumed that jet noise dominates below this frequency, and that non-jet noise only includes the forward and backward propagation components of fan noise. By processing different frequency bands according to prior knowledge, the decomposed fan noise and jet noise are obtained.

[0006] For the 3-microphone cross-spectral frequency domain method, based on the generation mechanism of fan noise, the spectrum of fan noise exhibits distinct pure-tone noise components at the blade passage frequency, and these pure-tone noises are coherent across different measurement points. Theoretically, the coherence function value between measurement points at this frequency should be 1, i.e., 100% coherence. However, due to errors introduced during measurement and errors in estimating the autocorrelation and cross-correlation spectra, the actual coherence function value calculated at this frequency is between 0.8 and 0.9, resulting in an overall lower power spectrum of the fan noise than the actual value. Since jet noise is obtained by removing the calculated fan noise from the total noise based on the energy conservation relationship, the calculated jet noise not only has a higher amplitude of broadband components than the actual value but also exhibits higher pure-tone noise leakage. The noise directivity derived from this will inevitably be affected by the leakage pure tones, failing to reflect the actual noise directivity. Furthermore, as a frequency domain method that processes the power spectrum of the noise signal, its accuracy is inevitably constrained by the transformation parameters in the Welch power spectrum estimation method. Literature and data analysis consistently demonstrate that the sampling length of the original signal, the type of spectral windowing, the window length, and the overlap rate all significantly affect the resolution of the spectrum. If the selection is inappropriate, it will lead to a large number of error noises in the spectrum, resulting in low resolution of the decomposition results and making them unrecognizable.

[0007] For noise prediction model-based methods, developing a highly accurate noise prediction model requires extensive prior knowledge of the engine's design parameters (such as bypass ratio, fan diameter, rotor-to-stationary distance, and blade geometry) and ground static tests to derive the empirical parameters used in the model. Therefore, for each different engine model, or when design parameters are adjusted, it is necessary to conduct experimental analysis and recalculation step by step using the same methods, lacking universal applicability. Furthermore, since existing noise prediction models are all at the 1 / 3 octave band level, this method is insufficient to determine the accurate dominant noise frequency and amplitude.

[0008] In summary, how to accurately separate fan noise and jet noise from engine noise is one of the important problems that urgently need to be solved in this field. Summary of the Invention

[0009] The purpose of this invention is to provide a method for decomposing noise sources in aero-engine components to address the shortcomings of existing technologies. This method can accurately decompose fan noise and jet noise from engine noise, enabling accurate assessment of the contribution of noise sources in modern turbofan engines at different frequencies and angles. It can also study the variation patterns of the frequency, amplitude, and directivity of major noise sources under different design and operating parameters of aero-engines, providing support for integrated aerodynamic-acoustic design and noise control measures. Furthermore, it can provide a means to verify the accuracy of engine noise prediction models based on the assessment results of the main engine noise sources. Finally, it can be used to solve the problem of mapping engine ground static test noise data to flight conditions in the noise airworthiness compliance certification of derivative aircraft.

[0010] This invention provides a method for decomposing noise sources in aero-engine components, comprising the following steps:

[0011] S1, Spatial sound source analysis; acquire noise input at each measuring point, construct the covariance matrix and perform singular value decomposition; obtain the number of eigenvalues ​​when the set contribution threshold is reached;

[0012] S2, extract the broadband time-domain component and the pure audio-domain component; remove the discrete pure tone periodic component to obtain the stationary random broadband component; use the discrete Fourier transform with Hanning window to process the total noise and broadband noise respectively and take the difference to obtain the spectrum of the pure tone noise component;

[0013] S3, adaptive signal decomposition filtering; decomposes the broadband time-domain signal at each measurement point into several components and a remainder term.

[0014] S4. Combine the broadband noise components, elements and remainders obtained at each measurement point in step S3 and perform singular value decomposition; and independently reconstruct the signal space at each measurement point to obtain the reconstructed signal.

[0015] S5, Sound source separation; the reconstructed signal is separated to obtain sound sources with the same number of feature values ​​as in step S1, and an estimate of the broadband component noise that constitutes the total broadband noise is obtained.

[0016] S6, determine the jet noise; using the broadband noise spectrum of the measuring point closest to the upstream fan inlet of the airflow channel as a benchmark, confirm that each component is either a broadband component of fan noise or jet noise.

[0017] S7, Amplitude Calibration; The estimation of the broadband portion of component noise is processed to obtain the unidentified component noise with accurate amplitude;

[0018] S8, Noise Component Identification: The energy of the discrete pure tone noise and the broadband component of the fan noise is summed to obtain the fan noise.

[0019] In the above-described method for decomposing noise sources in aero-engine components, step S1 optionally includes:

[0020] Obtain the noise at each measuring point;

[0021] Construct the covariance matrix C X Then, singular value decomposition is performed on the covariance matrix to determine its eigenvalues;

[0022] Arrange the eigenvalues ​​in descending order as λ1, λ2, ..., λ c ;

[0023] The ratio of effective components to total noise is determined based on the signal-to-noise ratio of background noise to total noise, and this value is set as the contribution threshold V. th ;

[0024] Based on the cumulative contribution of each eigenvalue to the total eigenvalue ∑λ TH The number of eigenvalues ​​*m* at which the threshold is reached is calculated. This value represents the number of sound sources in the space formed by the far-field microphone array, and also serves as a constraint on the maximum number of sound sources at each directional measurement point. The formulas for the covariance matrix and the cumulative contribution are as follows:

[0025] C X =X·X T =[λ1,λ2,…,λ c ] T ·[μ1,μ2,…,μ c (1)

[0026]

[0027] In the formula, [μ1,μ2,…,μ c ] represents the eigenvalues ​​[λ1, λ2, ..., λ] arranged in descending order. c The eigenvector corresponding to ].

[0028] In the above-described method for decomposing noise sources in aero-engine components, step S2 optionally includes:

[0029] Based on the periodicity of discrete pure tone components and considering the engine fan speed (RPM), the time-domain signal x at each measurement point is... q (t) is cut into segments of equal length; the formula for its length NPR is:

[0030] NPR = round(60 × F) S / RPM) (3)

[0031] Two adjacent segments y1 and y2 are extracted and cross-correlation analysis is performed to align the two signals; both y1 and y2 contain the same pure tone components. And stationary random signal components y1', y2';

[0032] After removing the periodic components of discrete pure tones, a stationary random broadband component x is obtained. q-b (t), calculate the amplitude z' of the stationary random broadband component; the formula for calculating the amplitude z' is,

[0033]

[0034]

[0035] The total noise and broadband noise are processed and subtracted using the discrete Fourier transform with a Hanning window to obtain the spectrum x of the pure tone noise component. qtones (f)

[0036] In the above-described method for decomposing noise sources in aero-engine components, step S3 optionally includes:

[0037] For the matrix of broadband time-domain signals at each measurement point, construct an unconstrained variational problem;

[0038] The stationary random broadband component at each measurement point is decomposed into a combination with K components and 1 remainder, wherein the number of components K is obtained based on the number of sound sources identified in step S1; the value of the penalty factor α satisfies the minimum value of the root mean square of the cross-correlation coefficient of the decomposed components.

[0039]

[0040] The constructed unconstrained variational problem is:

[0041]

[0042] Where k, q, and c have the meanings described above, {u k (t)} and {ω k} represents each component K itself and its corresponding center frequency, δ(t) is the Dirichlet function, λ is the Lagrange operator, and (δ(t)+j / πt)*P qk (t) is P qk (t) The spectrum after Hilbert transform, where * represents the convolution operator. For the gradient operator, exp(-jω) k t) is a continuous basic complex exponential function. This represents the corresponding analytical modulation signal, where ||·||2 is the 2-norm of the matrix.

[0043] Determine the values ​​of K and α according to the constraints in (6), and then assign the components... Center frequency Lagrange operators The initial value is set to 0, and the components are iteratively updated according to (8). Center frequency and Lagrange operators The value of is maintained until the termination condition described in (9) is met, thus achieving the adaptive decomposition specified in (6). At this point, the number of signal components and the center frequency between each measurement point remain consistent.

[0044]

[0045]

[0046] Where n is the number of iterations in the current iteration, and ε is the tolerance used to determine convergence, with a value of 10. -7 .

[0047] In the above-described method for decomposing noise sources in aero-engine components, step S4 optionally includes:

[0048] The broadband noise component x at each measurement point obtained in S3 q,b (t) and K components u qk (t), 1 remainder time-domain signal r q (t) are all rearranged into X according to the original order. q，new =[x qb (t),u q1 (t),u q2 (t)…,u qk (t)…,u qK (t),r q Singular value decomposition is performed on its covariance matrix [(t)].

[0049] Based on the contribution threshold t determined in S1, the number of feature values ​​S when the sum of the percentages of each feature value to the total feature values ​​reaches the cumulative contribution is calculated, thereby constraining the maximum number of decomposable sound sources at each measuring point.

[0050] Compare the number of sound sources m obtained in S1 with S; if m is less than S, take the smaller value. At each measurement point, the signal space X is reconstructed independently using the projection matrix. q ':

[0051]

[0052]

[0053]

[0054] In the formula, [μ1,μ2,…,μ K+2 ] represents the eigenvalues ​​[λ1, λ2, ..., λ] arranged in descending order. K+2 The eigenvector corresponding to ].

[0055] In the above-described method for decomposing noise sources in aero-engine components, step S5 optionally includes:

[0056] The reconstructed signal X at each measurement point q Based on the principle of maximum independence, we separate it from S. TP The same number of sound sources are used, and these sound sources are used as estimates of the broadband portion of the component noise that constitutes the total broadband noise. q (t):

[0057]

[0058] Among them, W q The solution matrix for each measurement point is obtained by solving the following method:

[0059] Establish the best estimate N based on each major noise component q The kurtosis κ(ω) of (t) is given by the objective function g(15) of equation (14):

[0060] κ(ω)=E{|N q (t)| 4}-2E 2 {|N q (t)| 2}-|E{|N q (t)| 2}| 2 / E 2 {|N q (t)| 2} (14)

[0061]

[0062] in, Let E be the gradient operator, and E{·} be the mathematical expectation.* This is the adjoint matrix.

[0063] For the objective function g, calculate the coefficients of its optimal step-size polynomial. Extracting the root μ of the optimal step-size polynomial z (z = 1, ..., 4); Select the root in the search direction that maximizes the absolute value of the kurtosis, and determine the optimal step size μ. opt =arg μ max|κ(ω+μg)|; based on the updated μ opt Update W(z) + =W(z-1)μ opt g then performs normalization on W(z). * =W(z) / ||W(z)||.

[0064] In the above-described method for decomposing noise sources in aero-engine components, step S6 optionally includes:

[0065] The broadband noise spectrum of the measuring point closest to the upstream fan inlet of the airflow channel x q-new-in (f) As a benchmark, examine its relationship with each component N. qi The spectrum N of (t) qi (f) Correlation coefficient, where higher values ​​represent the broadband component of fan noise, and lower values ​​represent jet noise; the test formula is:

[0066]

[0067] Where i is the amplitude at each frequency in the spectrum, and N q-fan (f) refers to the spectrum identified as broadband components of fan noise, N q-jet (f) refers to the spectrum identified as jet noise components.

[0068] In the above-described method for decomposing noise sources in aero-engine components, step S7 optionally includes:

[0069] Calculate N using Fourier transform (or power spectrum estimation method) q Each component N in (t) qi The amplitude spectrum N of (t) qi (f) Considering that the broadband noise components at the first-order blade passing frequency are all generated by the fan noise source, the component noise with accurate amplitude can be obtained. Taking fan noise as an example, its calculation formula is:

[0070]

[0071]

[0072] x q,fanb (t)=Rq,fan ·N q,fan (t) (19)

[0073] In the above-described method for decomposing noise sources in aero-engine components, optionally, in step S8, the formula for summing the discrete pure tone noise (component) and broadband component of the fan noise is:

[0074] x fan (f)=x qtones (f)+x qfanb (f) (20)

[0075] Where, x fan (f) represents the spectrum of the fan noise, x qtones (f) represents the spectrum of the pure tone noise component, x qfanb (f) is the spectrum of the broadband component of the fan noise.

[0076] In the above-described method for decomposing noise sources of aero-engine components, optionally, the noise input at each measuring point in S1 is the engine time-domain noise signal collected from at least two measuring points distributed on an arc with the geometric center of the engine as the center, a radius of 50m, and a height of ≥1.5m.

[0077] Compared with the prior art, the present invention has at least the following beneficial effects:

[0078] (1) Solving the problem of dual separation of amplitude and frequency of noise sources of major engine components: According to prior knowledge, the dominant frequency, spectral structure and amplitude characteristics of fan noise and jet noise are very different; while traditional methods either have the problem of excessive amplitude deviation, or rely on strong assumptions, or can only be processed within 1 / 3 octave band accuracy, none of which can achieve the dual separation of amplitude and frequency. The solution described in this application adopts a time-frequency domain combined processing method, which avoids the problem of excessive jet noise and achieves accurate amplitude separation; at the same time, since the time domain can achieve a frequency domain resolution no higher than its own, accurate frequency separation is achieved.

[0079] (2) Solve the problem of pure tone components of fan noise leaking into the noise spectrum of other components: By pre-extracting discrete pure tone components from the total noise and then starting the separation, not only can the discrete pure tone components be prevented from leaking into non-static interference sound sources such as jet noise, but the sound source separation process only needs to focus on making each component of broadband noise as independent as possible, which can improve the stability of the algorithm.

[0080] (3) Solving the problem of identifying the noise source of a component and its spectrum: In the existing methods, all the noise obtained by decomposition needs to be further judged based on prior knowledge to determine the source component. However, this method regards the spectrum as a function of frequency and amplitude, and quantifies the similarity of two spectra by testing the Pearson correlation coefficient. At the same time, it makes full use of the characteristics of fan noise propagating upstream of airflow and the dominant characteristics in front of the air intake, so as to realize the identification of noise sources of different components and their spectrum correspondence.

[0081] (4) Solving the problem of universal applicability for engines with different structural design characteristics: This application adopts a method of first global adaptive filtering and sound source number identification, and then separate each measurement point. It fully considers the extraction of correlation information contained in the fan sound between each channel, while taking into account the differences in the main components and spectrum structure of engine noise caused by engine design, speed conditions and other factors. Attached Figure Description

[0082] Figure 1 This is a flowchart of the steps in Embodiment 1 of the present invention;

[0083] Figure 2 Schematic diagram of the arrangement of far-field noise directivity measurement points;

[0084] Figure 3 This illustrates the contribution of each eigenvalue to the overall population.

[0085] Figure 4 To extract pure tone components and broadband components separately;

[0086] Figure 5 To decompose the signal into multiple components according to a preset method;

[0087] Figure 6 This refers to the process of identifying the sound source at the measurement point;

[0088] Figure 7 These are the two sound sources that have been separated.

[0089] Figure 8 To decompose the total noise under operating condition C, the dipole source spectrum and the monopole source spectrum are obtained;

[0090] Figure 9 The result of processing the total noise under condition C using the 3-microphone cross-spectrum method;

[0091] Figure 10 The results are the decomposition results for test conditions A and B;

[0092] Figure 11 The arrangement of far-field measurement points in the experimental example;

[0093] Figure 12 Directivity of some characteristic frequencies under test condition 1;

[0094] Figure 13 The noise decomposition results at each measuring point in Experiment 1 are shown.

[0095] Figure 14 For the directivity of some characteristic frequencies under test condition 2;

[0096] Figure 15 For the directivity of some characteristic frequencies under test condition 2;

[0097] Figure 16 The noise decomposition results at each measuring point in Experiment 2 are shown.

[0098] Figure 17 The results show the noise decomposition at each measuring point in Experiment 3. Detailed Implementation

[0099] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0100] To address the problems raised in the background art, the reason why existing technologies cannot accurately decompose fan noise and jet noise from engine noise is mainly due to the following specific problems: (1) the dual separation of amplitude and frequency of noise sources (jet noise and fan noise) of the main engine components; (2) the leakage of pure tone components of fan noise into the noise spectrum of other components; (3) the identification and correspondence between component noise sources and their spectra; and (4) the problem of universal applicability to engines with different structural design characteristics. Solving any of the above problems can help to accurately decompose fan noise and jet noise from engine noise. To solve one or more of the above problems, the present invention proposes the following embodiments.

[0101] Example 1

[0102] Please refer to Figure 1 This embodiment proposes a method for decomposing noise sources in aero-engine components, characterized by the following steps:

[0103] S1, Spatial sound source analysis; acquire noise input at each measurement point, construct the covariance matrix and perform singular value decomposition; obtain the number of eigenvalues ​​when the set contribution threshold is reached.

[0104] S2, extract the broadband time-domain component and the pure audio-domain component; remove the discrete pure tone periodic component to obtain the stationary random broadband component; use the discrete Fourier transform with Hanning window to process the total noise and broadband noise respectively and take the difference to obtain the spectrum of the pure tone noise component.

[0105] S3, adaptive signal decomposition filtering; decomposes the broadband time-domain signal at each measurement point into several components and a remainder term.

[0106] S4, Sound source directivity identification: Combine the broadband noise components, elements and remainders obtained at each measurement point in step S3 and perform singular value decomposition; and independently reconstruct the signal space at each measurement point to obtain the reconstructed signal;

[0107] S5, Sound source separation; the reconstructed signal is separated to obtain sound sources with the same number of feature values ​​as in step S1, and an estimate of the broadband component noise that constitutes the total broadband noise is obtained.

[0108] S6, determine the jet noise; using the broadband noise spectrum of the measuring point closest to the upstream fan inlet of the airflow channel as a benchmark, confirm that each component is either a broadband component of fan noise or jet noise.

[0109] S7, Amplitude Calibration; The estimation of the broadband portion of component noise is processed to obtain the unidentified component noise with accurate amplitude.

[0110] S8, Sound source merging; sum the energy of the discrete pure tone noise and the broadband component of the fan noise to obtain the fan noise.

[0111] The above-mentioned time-frequency domain combined processing method avoids the problem of excessive jet noise and achieves accurate amplitude separation. Simultaneously, since the time domain can achieve a resolution no higher than its own frequency domain, accurate frequency separation is achieved. By pre-extracting discrete pure tone components from the total noise before separation, not only can the leakage of discrete pure tone components into non-static interference sources such as jet noise be avoided, but the source separation process only needs to focus on maximizing the independence of each component of the broadband noise, thus improving the stability of the algorithm. Treating the spectrum as a function of frequency and amplitude, the similarity of different spectra is quantified using the Pearson correlation coefficient test. It also fully utilizes the characteristics of fan noise propagating upstream of the airflow and the dominance of features in front of the intake duct to identify noise sources from different components and their corresponding spectral relationships. The method of first global adaptive filtering and sound source number identification, followed by separation at each measurement point, fully considers the extraction of correlation information contained in the fan noise between channels, while also taking into account the differences in the main components and spectral structure of engine noise caused by factors such as engine design and operating speed.

[0112] Example 2

[0113] This embodiment is a more detailed description of the steps based on Embodiment 1.

[0114] Please refer to Figure 1 and Figure 2 In this embodiment, for an arc-shaped directional measuring point array with radius R and c microphones (c>2 and arranged at equal intervals to uniformly cover the upstream and downstream of the engine airflow channel), such as... Figure 2The measured engine time-domain noise signal X(t) = [x1(t), x2(t), ... x2(t)] with a sampling rate of Fs (Fs≥22400Hz) is obtained. q (t)…,x c (t)] T , where x q (t) represents the uniformly sampled continuous acoustic signal in the time domain measured at each measuring point, while simultaneously acquiring the number of blades B and fan speed RPM of the engine under test.

[0115] S1. Spatial Sound Source Analysis: Constructing the Covariance Matrix C X =X·X T and C X Perform singular value decomposition to determine its eigenvalues, and arrange them in descending order as λ1, λ2, ..., λ c The ratio of effective components to total noise is determined based on the signal-to-noise ratio of background noise to total noise, and this value is set as the contribution threshold V. th ;Based on the cumulative contribution of each eigenvalue to the total eigenvalue ∑λ th The number of eigenvalues ​​m when the threshold is reached is calculated. This value is the number of sound sources in the space formed by the far-field microphone array, and it is also the constraint condition for the maximum number of sound sources at each directional measurement point.

[0116] C X =X·X T =[λ1,λ2,…,λ c ] T ·[μ1,μ2,…,μ c (1)

[0117]

[0118] In the formula, [μ1,μ2,…,μ c ] represents the eigenvalues ​​[λ1, λ2, ..., λ] arranged in descending order. c The eigenvector corresponding to ].

[0119] S2. Extracting broadband time-domain and pure audio-domain components: Based on the periodicity of discrete pure tone components, considering the engine fan speed (RPM), the time-domain signal x at each measurement point is extracted. q (t) is cut into segments of equal length, and the length NPR is calculated as shown in (3). Two adjacent segments y1 and y2 are extracted and cross-correlation analysis is performed to determine the phase shift Δτ between them. The number of sampling points r, which is the same as the number of leading points, is borrowed from the leader to align the two signals. At this time, y1 and y2 both contain the same pure tone components. The stationary random signal components y1' and y2' are obtained. After removing the discrete pure tone periodic components from y1 and y2, the stationary random broadband component is obtained, whose mean square value of amplitude z' is twice the actual value, as shown in (4) and (5). Repeat the above steps until all equal-length sub-segments have been calculated, and the broadband component x with the sampling rate unchanged and the number of sampling points reduced to 1 / 2 of the original signal can be obtained. qb (t). Then, by applying the Discrete Fourier Transform with a Hanning window to process the total noise and broadband noise separately and taking the difference, the spectrum x of the pure tone noise component can be obtained. qtones (f), where q represents the number of the microphone measurement point to be processed. The round operator indicates that the segment length is rounded to the nearest integer.

[0120] NPR = round(60×F) S / RPM) (3)

[0121]

[0122]

[0123] S3. Adaptive signal decomposition filtering: For the matrix x of the broadband time-domain signal at each measurement point... b (t)=[x 1,b (t),x 2,b (t),…x q,b (t)…,x c,b (t)] T Construct an unconstrained variational problem, and convert the time-domain signal x at each measurement point into a time-domain signal x. q,b (t) is simultaneously decomposed into a combination with K components and 1 remainder term using the same parameters, where the number of components K that can be decomposed is obtained based on the number of sound sources m identified in S1. The value of the penalty factor α satisfies the minimum value of the root mean square of the cross-correlation coefficients of the decomposed components, as shown in (6):

[0124]

[0125] In the formula, q is the measurement point number, with a value range of [1, c], k is the component number, with a value range of [1, K], and P q,k (t) represents the time-domain signal of each component decomposed, r q (t) represents the remainder term, and cov is the covariance operator.

[0126] The constructed unconstrained variational problem is:

[0127]

[0128] Where k, q, and c have the meanings described above, {u k (t)} and {ωk} represents each component K itself and its corresponding center frequency, δ(t) is the Dirichlet function, λ is the Lagrange operator, and (δ(t)+j / πt)*P qk (t) is P qk (t) The spectrum after Hilbert transform, where * represents the convolution operator. For the gradient operator, exp(-jω) k t) is a continuous basic complex exponential function. This represents the corresponding analytical modulation signal, where ||·||2 is the 2-norm of the matrix.

[0129] Determine the values ​​of K and α according to the constraints in (6), and then assign the components... Center frequency Lagrange operators The initial value is set to 0, and the components are iteratively updated according to (8). Center frequency and Lagrange operators The value of is maintained until the termination condition described in (9) is met, thus achieving the adaptive decomposition specified in (6). At this point, the number of signal components and the center frequency between each measurement point remain consistent.

[0130]

[0131]

[0132] Where n is the number of iterations in the current iteration, and ε is the tolerance used to determine convergence, with a value of 10. -7 .

[0133] S4. Sound source directivity identification: Identify the broadband noise component x at each measurement point obtained in S3. q,b (t) and K components u qk (t), 1 remainder time-domain signal r q (t) are all rearranged into X according to the original order. q，new =[x qb (t),u q1 (t),u q2 (t)…,u qk (t)…,u qK (t),r q [t], perform singular value decomposition on its covariance matrix, and determine the contribution threshold V determined in S1. th The number of eigenvalues, S, is calculated when the sum of the percentages of each eigenvalue to the total eigenvalues ​​reaches the cumulative contribution, thus constraining the maximum number of decomposable sound sources at each measurement point. Simultaneously, the number of sound sources m obtained in S1 is compared with S, and the smaller value is taken as the dimension of the reconstructed signal, i.e., the actual number of sound sources S at that measurement point. TPAt each measurement point, the signal space X is reconstructed using a projection matrix. q ':

[0134]

[0135]

[0136]

[0137] In the formula, [μ1,μ2,…,μ K+2 ] represents the eigenvalues ​​[λ1, λ2, ..., λ] arranged in descending order. K+2 The eigenvector corresponding to ].

[0138] S5. Sound source separation: Reconstructing the signal X at each measurement point. q Based on the principle of maximum independence, we separate it from S. TP For a given number of sound sources, determine which sound source is the best estimate N of the broadband component noise that constitutes the total broadband noise at each measurement point. q (t):

[0139]

[0140] Among them, W q The solution matrix for each measurement point is obtained by solving the following method:

[0141] Establish the best estimate N based on each major noise component q The kurtosis κ(ω) of (t) is given by the objective function g(15) of equation (14):

[0142]

[0143]

[0144] in, Let E be the gradient operator, and E{·} be the mathematical expectation. * This is the adjoint matrix.

[0145] For the objective function g, calculate the coefficients of its optimal step-size polynomial. Extracting the root μ of the optimal step-size polynomial z (z = 1, ..., 4); Select the root in the search direction that maximizes the absolute value of the kurtosis, and determine the optimal step size μ. opt =arg μ max|κ(ω+μg)|; based on the updated μ opt Update W(z) + =W(z-1)μ opt g then performs normalization on W(z). *=W(z) / ||W(z)||.

[0146] By performing the above steps until the convergence condition is met, W can be realized. q Solve for it.

[0147] S6. Noise Component Identification: Based on the properties of matrix multiplication, N q (t) has a spectral structure that is completely consistent with the component noise components, but its amplitude differs from the actual component noise broadband source.

[0148] The fan noise component contributes the most to the sound source at the fan inlet, and the mixed noise obtained from far-field measurements mainly includes jet noise and fan noise. Therefore, the broadband noise spectrum of the measurement point closest to the upstream fan inlet of the airflow channel is x q-new-in (f) Using Fourier transform (or power spectrum estimation method) as a benchmark, calculate N. q Each component N in (t) qi The amplitude spectrum N of (t) qi (f), examine its relationship with each component N. qi The spectrum N of (t) qi (f) correlation coefficient, where the higher value is the broadband component of fan noise and the lower value is the jet noise, as in (16).

[0149]

[0150] Where i is the amplitude at each frequency in the spectrum, and N q-fan (f) refers to the spectrum identified as broadband components of fan noise, N q-jet (f) refers to the spectrum identified as jet noise components.

[0151] S7. Amplitude Calibration: Further, based on the principle of energy conservation, that is, the sum of the acoustic energy of all broadband noise is equal to the total broadband component x qb (t) sound energy Based on the principle that the broadband noise components at the first-order blade passage frequency are all generated by the fan noise source, the corresponding amplitude restitution coefficient R (with fan noise x) can be calculated. q,fanb Taking the solution of (t) as an example, and so on, the component noise with accurate amplitude is obtained, such as (17)-(19).

[0152]

[0153]

[0154] x q,fanb (t)=R q,fan ·N q,fan (t) (19)

[0155] Where R q,fan N is the fan amplitude recovery coefficient. q,fan (t) represents the demixed component identified as a broadband component of fan noise.

[0156] S8. Sound source merging: Since the discrete pure tone components can all be determined to originate from the fan's rotation-to-static interference noise, the energy of the discrete pure tone components and broadband components of the fan noise are summed to finally obtain completely separated fan noise and jet noise, as shown in (20).

[0157] x fan (f)=x qtones (f)+x qfanb (f) (20)

[0158] Where, x fan (f) is the spectrum of the fan noise obtained by Fourier transform, x qtones (f) is the spectrum obtained by Fourier transform of the pure tone noise component, x qfanb (f) is the spectrum obtained by Fourier transform of the broadband components of the fan noise.

[0159] Based on the above content, this embodiment has at least the following beneficial effects:

[0160] (1) Solve the problem of dual separation of amplitude and frequency of noise sources in major engine components:

[0161] Prior knowledge indicates that fan noise and jet noise exhibit significant differences in their dominant frequencies, spectral structures, and amplitude characteristics. Traditional methods either suffer from excessive amplitude deviation, rely on strong assumptions, or can only process within a 1 / 3 octave band accuracy, failing to achieve the dual separation of amplitude and frequency. The solution described in this application employs a time-frequency domain combined processing method, avoiding the problem of excessively high jet noise and achieving accurate amplitude separation. Simultaneously, since the time domain can achieve a resolution no higher than its own frequency domain resolution, accurate frequency separation is achieved.

[0162] (2) Solve the problem of pure tone components of fan noise leaking into the noise spectrum of other components: By pre-extracting discrete pure tone components from the total noise and then starting the separation, not only can the discrete pure tone components be prevented from leaking into non-static interference sound sources such as jet noise, but the sound source separation process only needs to focus on making each component of broadband noise as independent as possible, which can improve the stability of the algorithm.

[0163] (3) Solving the problem of identifying the noise source of a component and its spectrum: In the existing methods, all the noise obtained by decomposition needs to be further judged based on prior knowledge to determine the source component. However, this method regards the spectrum as a function of frequency and amplitude, and quantifies the similarity of two spectra by testing the Pearson correlation coefficient. At the same time, it makes full use of the characteristics of fan noise propagating upstream of airflow and the dominant characteristics in front of the air intake, so as to realize the identification of noise sources of different components and their spectrum correspondence.

[0164] (4) Solving the problem of universal applicability for engines with different structural design characteristics: This application adopts a method of first global adaptive filtering and sound source number identification, and then separate each measurement point. It fully considers the extraction of correlation information contained in the fan sound between each channel, while taking into account the differences in the main components and spectrum structure of engine noise caused by engine design, speed conditions and other factors.

[0165] Experimental Examples

[0166] Given the difficulty in obtaining the noise generated by each component of the engine through testing, in order to achieve quantitative description, an example of the use of anechoic chamber test data decomposition is adopted to illustrate the beneficial effects of this application:

[0167] In the experiment, a dipole sound source with specific pure audio frequency components was generated using flow around a cylinder. Sounds with low correlation at each measurement point were used as monopole sound sources, thus constructing a hybrid sound source with both monopole and dipole components. Three different incoming flow velocities were used in the experiment, representing three different operating conditions, as shown in Table 1. During measurement, the microphone array was placed in the near field of the sound source at a radius of 1 m, with a sampling rate of 48000 Hz, a sampling duration of 3 s, and a total of 39 measurement channels.

[0168] Table 1. Operating parameters of the anechoic chamber during various tests.

[0169]

[0170] Select microphone measurement point 1 in working condition C to demonstrate the decomposition process of the target sound source by this method:

[0171] For the noise time-domain signal obtained from the microphone array measurement, calculate its covariance matrix and the contribution of each eigenvalue, such as... Figure 3 Based on the background noise analysis results during data measurement, the feature value threshold for sound source identification was set to 85%. It can be seen that the cumulative contribution of the first two feature values ​​is 87.26%, so the number of sound sources is 2. This result is consistent with the original setting of sound sources in the experiment, which shows that this method is capable of identifying the maximum number of sound sources in space.

[0172] Furthermore, since the actual physical sound source is not the engine sound source, the length of the signal segment is specified as 900 based on the sampling rate and sampling length. At this point, the original signal at each measurement point is divided into 160 equal-length segments. Through the steps described in this method to extract the broadband time-domain component and the discrete pure audio-domain component, the spectra of both contained in the original signal are as follows: Figure 5 .

[0173] Based on the sound source identification results, and following the parameter selection principles of this method, the time-domain broadband components separated in the previous step are decomposed using K=2 and α=50 to minimize the mean square value of the cross-correlation coefficients of each component (at which point the value is 0.066), and the cross-correlation values ​​between each component and the remainder term r are all less than 0.1. Figure 6 .

[0174] For the original broadband components, the covariance matrix of each component and the recombined matrix of the remainder is calculated to obtain the eigenvalues. The contributions of the first two eigenvalues ​​are 82.82% and 17.01%, respectively, totaling 99.83%. The contributions of the other two eigenvalues ​​are orders of magnitude different and can be ignored. Therefore, it can be concluded that there are two sound sources mixing at the measurement point. Figure 7 .

[0175] Based on this, further source separation is performed, resulting in two components with different spectral structures. The spectra of the two components after amplitude recovery are as follows: Figure 8 As shown.

[0176] Since the sound source is located in the near field and its spatial distribution pattern differs from that of the engine, an equivalent method is adopted, which measures the similarity between the two and the dipole and monopole sources in the mixed noise components by using the Pearson correlation coefficient of the spectrum.

[0177] Table 2 Components and Sound Source Similarity

[0178]

[0179] The Pearson correlation coefficients in the table show that source 1 has a high similarity to the dipole source and should originate from the broadband portion of the dipole source; source 2 has a high similarity to the monopole source and should also belong to the monopole source. Combining the pure tone component and broadband component corresponding to the dipole source, the final results for the dipole and monopole sources are as follows: Figure 8 As shown in (a) and (b), the amplitude and frequency of the discrete pure tone components of the dipole source in the decomposition results are accurately identified, and the average difference between the broadband components and the original amplitude is <3dB. Analysis shows that the abrupt frequency changes are due to errors in the calculation process, and the influence of these frequencies can be effectively eliminated using a moving average method.

[0180] The dipole sound source obtained after cross-spectral processing under the same operating conditions is as follows: Figure 9As shown in (a), the monopole sound source is as follows: Figure 9 As shown in (b).

[0181] In the monopole sound source obtained by cross-spectrum processing, the frequency corresponding to the discrete pure tone showed obvious amplitude leakage, with a leakage of 27dB, and the amplitude of the broadband component was 9dB higher than the expected amplitude. It can be seen that the method in this application solves the technical problem of the leakage of discrete pure tone components into the noise spectrum of other components.

[0182] To verify the general applicability of the method described in this application, the same treatment was applied to other operating conditions, and the results are as follows. Figure 10 As shown in (a)-(d), the decomposition results show that the pure tone characteristic frequencies and amplitudes are correctly decomposed, and the amplitude error of the broadband components is less than 3%, thus proving that this method has universal applicability to noise sources with different characteristic frequencies.

[0183] Practical application examples: analyzing static test noise data for different operating conditions and different engines.

[0184] Through such Figure 11 The far-field directivity measurement points shown in (a) and (b) were used to measure the static noise generated by two engines with different bypass ratios under three operating conditions, obtaining spectrum and directivity information. The experimental sampling rate was 32768 Hz.

[0185] Table 3. Test case information for different operating conditions and different engines.

[0186]

[0187] Since the original values ​​of the specific amplitude and key frequencies of the spectrum in this experiment are not convenient to display directly, the frequency scale unit of each group of experiments is the blade passage frequency and its higher harmonics. The spectrum structure has not changed and is still sufficient to reflect the characteristics of the noise source of the engine components.

[0188] First, the 1 / 3 octave band spectrum of the noise time-domain signal obtained at each measurement point from Experiment 1 was calculated. Based on this, the directivity diagrams of the total engine noise at the first three blade passing frequencies and some frequencies were plotted, as shown below. Figure 12 As shown in (a)-(d).

[0189] Depend on Figure 12Under test condition 1, the frequency band containing the discrete pure tone characteristic frequency of the engine fan noise, i.e., the passing frequencies of the first three blades, exhibits a significant difference in sound pressure level upstream of the intake duct compared to downstream, with a maximum difference of up to 20 dB. Similar characteristics are observed in a portion of the frequency band above 2000 Hz; based on prior knowledge, fan noise dominates in these frequencies. Conversely, for the frequency band containing the center frequency below 500 Hz, the downstream sound pressure level is significantly higher than the upstream sound pressure level; based on the jet noise generation mechanism, it can be determined that jet noise is dominant in these frequencies.

[0190] The measurement data from Experiment 1, with a sampling duration of 10 seconds, were processed using the method provided in this application. Due to the large number of measurement points, the following points were selected: a 10° measurement point closest to the air intake, a 50° measurement point where fan noise was dominant, a 90° measurement point located in the middle of the engine, a 150° measurement point where jet noise was dominant, and a 170° measurement point closest to the exhaust nozzle outlet. The resulting fan noise spectrum and jet noise spectrum are shown below. Figure 13 As shown by the medium gray line, the total noise is as follows: Figure 13 As shown by the black line in the middle.

[0191] Because the same coordinate start and end values ​​and scale were used for all graphs, the relative magnitudes of the noise components can be directly compared. At the 10° measurement point, only one primary sound source was calculated, which can be identified as fan noise based on spectral structure observation and spectral similarity calculations combined with prior knowledge. At other angles, jet noise contributes in a way that can be identified from the spectrum. According to the decomposition results obtained by the method in this application, jet noise's contribution in the low-frequency band (0 to several hundred Hz) gradually increases with increasing angle; fan noise dominates the high-frequency band in this experiment. This conclusion is consistent with the pointer diagram and prior knowledge combined with the sound generation mechanism, proving that the method proposed in this application solves the dual separation problem of amplitude and frequency of the main engine component noise sources in engine component noise source decomposition. It also overcomes the leakage problem of pure tone noise in practical applications and successfully solves the correspondence problem between component noise sources and their spectra.

[0192] To verify the ability of the method described in this application to solve the problem of universal applicability of sound source decomposition in engine components, the data from Experiments 2 and 3 were further processed using the same method, and their directivity diagrams are shown below. Figure 14 As shown in Figure 15, the component noise decomposition results are as follows: Figure 16 As shown in -17.

[0193] Experiment 3 used a turbofan engine with a high bypass ratio (B>10) and a low speed. Therefore, its fan noise should be dominant. The low-frequency noise component of the engine (which was determined by prior knowledge to be jet noise) did not show obvious backward propagation characteristics, while the passing frequency of each blade and the high-frequency band (fan noise) still maintained a high forward pointing characteristic.

[0194] Due to the increase in rotational speed, based on the directivity diagram of Experiment 2 and prior knowledge, it is known that the amplitude of the jet noise will be higher than that of Experiment 1, the dominant frequency will hardly change significantly, and the angle at which it makes an effective contribution upstream of the intake duct will be earlier than that of Experiment 1; as for the fan noise, its amplitude and characteristic frequency will be significantly increased. Figure 16 The decomposition results shown and the directional change patterns they reflect prove the correctness of the method described in this application.

[0195] like Figure 17 As shown, in Experiment 3, the fan noise significantly dominates the noise generated by the engine under this operating condition, while the amplitude of the jet noise is significantly lower, differing from the total noise by nearly 20 dB at high frequencies. However, upstream of the intake duct, at angles such as 10° and 50°, the contribution of jet noise to the total noise is less than 10%, insufficient to be reflected in the frequency spectrum and thus automatically removed by the method, consistent with the information provided by the directivity diagram.

[0196] Therefore, based on the above application examples, it can be considered that the method described in this application solves the problem of universal applicability required in the engine noise source decomposition method.

[0197] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method for decomposing noise sources in aero-engine components, characterized in that: Including steps, S1, Spatial sound source analysis; acquire noise input at each measuring point, construct the covariance matrix and perform singular value decomposition; obtain the number of eigenvalues ​​when the set contribution threshold is reached; S2, extract the broadband time-domain component and the pure audio-domain component; remove the discrete pure tone periodic component to obtain the stationary random broadband component; use the discrete Fourier transform with Hanning window to process the total noise and broadband noise respectively and take the difference to obtain the spectrum of the pure tone noise component; S3, adaptive signal decomposition filtering; decomposes the broadband time-domain signal at each measurement point into several components and a remainder term. S4, Sound source directivity identification; combine the broadband noise components, elements and remainders obtained at each measurement point in step S3 and perform singular value decomposition; and independently reconstruct the signal space at each measurement point to obtain the reconstructed signal; S5, Sound source separation; the reconstructed signal is separated to obtain sound sources with the same number of feature values ​​as in step S1, and an estimate of the broadband component noise that constitutes the total broadband noise is obtained. S6, Noise Component Identification: Using the broadband noise spectrum of the measurement point closest to the upstream fan inlet of the airflow channel as a benchmark, identify each component as a broadband component of fan noise or jet noise, including... The broadband noise spectrum of the measuring point closest to the upstream fan inlet of the airflow channel. As a benchmark, it is tested against each component. N qi ( t ) spectrum The correlation coefficient is used to determine the frequency response, where higher values ​​represent the broadband component of fan noise and lower values ​​represent jet noise. The test formula is: in, i The amplitude at each frequency point in the spectrum. The spectrum identified as broadband components of fan noise. The spectrum identified as jet noise components; S7, Amplitude Calibration; The estimation of the broadband portion of component noise is processed to obtain the unidentified component noise with accurate amplitude; S8, Sound source merging; sum the energy of the discrete pure tone noise and the broadband component of the fan noise to obtain the fan noise.

2. The method for decomposing noise sources in aero-engine components according to claim 1, characterized in that: Step S1 includes, Obtain the noise at each measuring point; Constructing the covariance matrix C X Then, singular value decomposition is performed on the covariance matrix to determine its eigenvalues; Arrange the feature values ​​in descending order as follows λ 1, λ 2, …, λ c ; The ratio of effective components to total noise is determined based on the signal-to-noise ratio of background noise to total noise, and this value is set as the contribution threshold. V th ; Based on the cumulative contribution of each eigenvalue to the total eigenvalue ∑ λ th Solve for the number of eigenvalues ​​when the threshold is reached. m This value represents the number of sound sources in the space formed by the far-field microphone array, and is also a constraint on the maximum number of sound sources at each directional measurement point. The formulas for the covariance matrix and cumulative contribution are as follows: In the formula, The eigenvalues ​​are sorted in descending order. The corresponding eigenvector.

3. The method for decomposing noise sources in aero-engine components according to claim 2, characterized in that: Step S2 includes, Based on the periodicity of the discrete pure tone components, the sampling rate of the microphone used for measurement is considered. Fs and engine fan speed RPM The time-domain signal at each measuring point x q ( t Cut into segments of equal length; Its length NPR The formula is: The round operator rounds the length of a sub-section to the nearest integer. Extract two adjacent sub-series y 1 and y 2. Perform cross-correlation analysis to align the two signal segments; y 1 and y Both contain the same pure tone components. , and stationary random signal components y 1', y 2'; After removing the periodic components of discrete pure tones, stationary random broadband components are obtained. x q,b ( t ), calculate the amplitude of the stationary random broadband component. z '; Amplitude z The formula for calculating ' is, By using the discrete Fourier transform with a Hanning window to process the total noise and broadband noise separately and then subtracting them, the spectrum of the pure tone noise component is obtained.

4. The method for decomposing noise sources in aero-engine components according to claim 3, characterized in that: Step S3 includes, For the matrix of broadband time-domain signals at each measurement point, construct an unconstrained variational problem; The stationary random broadband components at each measurement point x q,b ( t Decomposed into having K The combination of one component and one remainder, in which the number of components is decomposed. K The value is obtained based on the number of sound sources in step S1; penalty factor α The value satisfies the minimum value of the root mean square of the cross-correlation coefficient of the components obtained by decomposition. In the formula, q The measurement point number is [1, ...]. c ], k The component index is [1, ...]. K ], P q,k ( t () represents the time-domain signal of each component decomposed. r q ( t ) represents the remainder term, and cov is the covariance operator; The constructed unconstrained variational problem is: in, k , q and c The meaning is as stated above, { u k ( t )}and{ ω k } are each component K It itself and its corresponding center frequency, δ ( t ) is the Dirichlet function, λ is the Lagrange operator, ( δ ( t )+ j / πt ) P qk ( t )for P qk ( t The spectrum after Hilbert transform. For convolution operators, t is the gradient operator, exp(- jω k t ) is a continuous basic complex exponential function. This represents the corresponding analytical modulation signal. Let be the 2-norm of the matrix; Determine according to the constraints in (6) K and α The value of the component Center frequency Lagrange operators The initial value is set to 0, and the components are iteratively updated according to (8). Center frequency and Lagrange operators The value of is maintained until the termination condition described in (9) is met, thus achieving the adaptive decomposition specified in (6). At this point, the number of signal components and the center frequency between each measurement point remain consistent. in, n This represents the iteration number. ε The tolerance used to determine convergence is set to 10. -7 .

5. The method for decomposing noise sources in aero-engine components according to claim 1, characterized in that: Step S4 includes, The broadband noise component at each measurement point obtained in S3 x q,b ( t )and K Each component u qk ( t ), 1 remaining term time-domain signal r q ( t All were rearranged according to their original order. X q，new = [ x qb ( t ), u q1 ( t ), u q2 ( t …, u qk ( t …, u qK ( t ), r q ( t Singular value decomposition is performed on its covariance matrix; Based on the contribution threshold determined in S1 V th Calculate the number of eigenvalues ​​whose sum of the percentages of each eigenvalue to the total number of eigenvalues ​​reaches the cumulative contribution. S This constrains the maximum number of decomposable sound sources at each measuring point; The number of sound sources obtained in S1 m and S Comparing the two, the smaller one is taken as the dimension of the reconstructed signal, i.e., the total number of sound sources. S TP At each measurement point, the signal space is reconstructed independently using a projection matrix. X q ’ : In the formula, The eigenvalues ​​are sorted in descending order. The corresponding eigenvector.

6. The method for decomposing noise sources in aero-engine components according to claim 1, characterized in that: Step S5 includes, Reconstructed signal at each measurement point X q ’ Based on the principle of maximum independence, separate from... S TP For sound sources of the same number, determine which sound source is the best estimate of the broadband component noise that constitutes the total broadband noise at each measurement point. N q ( t ): in, W q The solution matrix for each measurement point is obtained by solving the following method: Establish optimal estimates based on each major noise component N q ( t ) kurtosis κ ( ω ), as in equation (14), the objective function g (15): in, For gradient operators, E {·} represents the mathematical expectation, [·] The adjoint matrix; For the objective function g Calculate the coefficients of its optimal step-size polynomial. Extracting the roots of the optimal step-size polynomial μ z ( z =1, …,4); Select the root in the search direction that maximizes the absolute value of the kurtosis, and determine the optimal step size. ; According to the updated μ opt renew And perform normalization ; Perform the above steps until the convergence condition is met.

7. The method for decomposing noise sources in aero-engine components according to claim 1, characterized in that: Step S7 includes, Calculate using short-time Fourier transform N q ( t Each component in ) N qi ( t amplitude spectrum By considering the characteristic that the broadband noise components at the first-order blade passage frequency are all generated by the fan noise source, the accurate amplitude of the component noise broadband components can be obtained, using fan noise as an example. For example, the calculation formula is as follows: in R q,fan This is the fan amplitude recovery factor. N q,fan ( t ) is the demixed component identified as a broadband component of fan noise.

8. The method for decomposing noise sources in aero-engine components according to claim 1, characterized in that: In step S8, the formula for summing the discrete pure tone noise and the broadband component of the fan noise is: in, The spectrum of fan noise obtained by Fourier transform is shown below. The spectrum is obtained by Fourier transforming the pure tone noise component. The spectrum is obtained by Fourier transform of the broadband components of the fan noise.

9. The method for decomposing noise sources of aero-engine components according to any one of claims 1-8, characterized in that: The noise input at each measuring point in S1 is the engine time-domain noise signal collected from at least two measuring points distributed on an arc with the geometric center of the engine as the center, a radius of 50m, and a height of ≥1.5m.

Images