Fan noise prediction method and system based on machine learning

Abstract

The invention relates to the technical field of aero-engines, in particular to a fan noise prediction method and system based on machine learning. The prediction method comprises the following steps: S1, inputting microphone array arrangement parameters, fan pneumatic and structural parameters and atmospheric environment parameters; s2, performing updating calculation on Heidmann model parameters under each specific directional angle to obtain frequency spectrums of fan forward noise and fan backward noise under the specific directional angle and OASPL frequency spectrum directivity; wherein the parameters of the Heidmann model comprise a directivity correction function F3; in the step S2, a fitting model of each noise component is generated based on a machine learning algorithm, and the directivity correction function F3 is updated through the fitting model of each noise component. According to the method, test data and machine learning are creatively combined to update the F3 function, the difference of directivity correction values caused by changes of working conditions and specific single-tone frequency can be effectively distinguished, and the model prediction precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of aeroengines, and more particularly to a fan noise prediction method and system based on machine learning. Background Art

[0002] Civil airworthiness standards have strict regulations on aircraft noise emissions, and the noise level of aircraft is also an important factor in market competition. Among them, aircraft engine noise accounts for a large contribution. With the development of modern high-bypass turbofan engines, the fan noise of the engine has gradually become the main noise source of the engine and even the entire aircraft.

[0003] Fan noise is the noise generated by the fan components of an aircraft engine. It is divided into fan single-tone noise, fan broadband and fan shock wave noise at high speed based on the generation mechanism and corresponding spectrum performance.

[0004] If the far-field noise level of the engine fan can be predicted quickly and accurately based on limited structural parameters and aerodynamic parameters in the early design stage, the cost of engine noise testing will be greatly reduced, which will play a key supporting role in airworthiness noise assessment.

[0005] In order to obtain the most accurate fan airworthiness noise prediction results in the shortest possible time, NASA developed the Heidmann fan noise engineering empirical prediction model (hereinafter referred to as the Heidmann model) in 1975, and integrated it into the ANOPP aircraft airworthiness noise assessment software to support aircraft airworthiness noise assessment work.

[0006] The Heidmann model is an empirical correlation prediction model for fan noise that has been gradually improved and developed under the guidance of aeroacoustic theory by conducting a large number of tests on engine fan components and combining the fan noise generation mechanism. The Heidmann model divides fan noise into five noise components for prediction. The fan noise components include:

[0007] Fan forward broadband noise, fan forward single-tone noise, fan forward shock wave noise (only occurs when the Mach number is greater than 1), fan rear broadband noise and fan rear single-tone noise;

[0008] The noise component of each part is obtained by a unified empirical formula:

[0009]

[0010] Where ΔT is the total temperature rise of the fan stage, ΔT0 is the reference total temperature rise of the fan stage (0.555K), m is the mass flow rate, and m0 is the reference mass flow rate (0.453kg / s);

[0011] F1 is the blade tip relative Mach number correction function, M TR is the relative Mach number of the fan rotor tip, (M TR ) D Design the blade tip relative Mach number for the fan;

[0012] F2 is the rotor-static distance correction function, and RSS is the ratio of the rotor-static distance to the fan blade chord length;

[0013] F3 is the directivity correction function, θ is the directivity angle;

[0014] F4 is the spectrum shape correction function, f is the TOB center frequency (24 frequencies from 50 Hz to 10000 Hz), f b is the blade passing frequency (BPF).

[0015] One-Third Octave Band (TOB) is a method of dividing the continuous frequency range of a narrowband spectrum into several frequency bands, and the frequency ratio of the upper and lower limits of the frequency band is That is, one-third octave frequency band. The sum of the sound energy in the corresponding frequency band gives the one-third octave spectrum.

[0016] The specific parameter settings of the F1, F2, F3, and F4 functions of the above five noise components are different, and the values ​​are derived from a large amount of experimental data and the most basic acoustic theory.

[0017] Based on the Heidmann framework and combined with the actual test data of each fan configuration, the parameters of the F1, F2, F3, and F4 functions of the five noise components can be updated to adapt to the preliminary evaluation of the far-field noise of the fan configuration and its derivative configurations.

[0018] Among them, the F3 function has a great influence on the overall shape of the spectrum and the directivity accuracy at each pointing angle. Ensuring the reasonable update setting of the F3 function is an important part of improving the prediction accuracy.

[0019] The conventional updating method of the F3 function in the prior art does not distinguish between working conditions. Therefore, in practical applications, the prediction accuracy of the model obtained by using this conventional correction method is limited, especially the directional prediction does not perform well in actual evaluation. Summary of the invention

[0020] The purpose of the present invention is to provide a fan noise prediction method and system based on machine learning to solve the problem of poor fan noise prediction accuracy of the existing Heidmann model.

[0021] In order to achieve the above object, the present invention provides a fan noise prediction method based on machine learning, comprising the following steps:

[0022] Step S1, input microphone array layout parameters, fan aerodynamic and structural parameters, and atmospheric environment parameters;

[0023] Step S2, respectively updating and calculating the Heidmann model parameters at each specific directional angle, and obtaining the spectrum and OASPL spectrum directivity of the fan forward noise and the fan backward noise at the specific directional angle;

[0024] Wherein, the Heidmann model parameters include a directivity correction function F3;

[0025] In the step S2, a fitting model of each noise component is generated based on a machine learning algorithm, and the directivity correction function F3 is updated by the fitting model of each noise component.

[0026] In one embodiment, the Heidmann model parameters further include mechanical work related to temperature rise, specific work related to flow rate, blade tip relative Mach number correction function F1, rotor-to-static distance correction function F2, and spectrum shape correction function F4;

[0027] The fan forward noise includes fan forward broadband noise, fan forward single tone noise and fan forward shock wave noise;

[0028] The fan back-transmitted noise includes fan back-transmitted broadband noise and fan back-transmitted single-tone noise.

[0029] In one embodiment, the step S2 further comprises the following steps:

[0030] Step S21, calculating the mechanical work related to the temperature rise and the specific work related to the flow rate;

[0031] Step S22, respectively calculating the blade tip relative Mach number correction function F1 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0032] Step S23, respectively calculating the rotation-to-static distance correction function F2 of the fan forward broadband noise, the fan forward single-tone noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0033] Step S24, respectively calculating the directivity correction function F3 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0034] Step S25, respectively calculating the spectrum shape correction function F4 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0035] Step S26, calculating atmospheric attenuation and spherical distance diffusion correction parameters;

[0036] Step S27, summing the parameters obtained in steps S21 to S26 to obtain the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0037] Step S28, summing the energy of the fan forward broadband noise and the fan forward single-tone noise to obtain the spectrum and OASPL spectrum directivity of the fan forward noise at each directional angle, and summing the energy of the fan rearward broadband noise and the fan rearward single-tone noise to obtain the spectrum and OASPL spectrum directivity of the fan rearward noise at each directional angle.

[0038] In one embodiment, the step S28 further comprises:

[0039] When the Mach number is greater than 1, the fan forward broadband noise, fan forward single-tone noise and fan forward shock wave noise are summed to obtain the spectrum and OASPL spectrum directivity of the fan forward noise at each directional angle.

[0040] In one embodiment, the generating of the fitting model of each noise component based on the machine learning algorithm further comprises the following steps:

[0041] Input the noise time domain data under each working condition;

[0042] Divide the data into training set and test set;

[0043] Calculate the power spectrum of the time domain sample data collected by each audio acquisition device;

[0044] Perform single-tone broadband separation on the power spectrum to obtain broadband power spectrum and the first several axial frequency single-tone power spectra;

[0045] Extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum;

[0046] Select aerodynamic features and generate a feature matrix in combination with the pointing angle, and correspond the feature matrix to the extracted directivity correction data;

[0047] Based on a machine learning algorithm, the characteristic matrix and the directivity correction data are trained to generate a fitting model corresponding to the noise component. The fitting model is used to respond to the input characteristic matrix and output the SPL correction value of the corresponding noise component.

[0048] In one embodiment, the extraction of directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further includes the following steps:

[0049] The TOB-SPL data in matrix form is obtained according to the broadband power spectrum, the energy of the rows of the matrix is ​​summed, the maximum value is subtracted from the obtained data, the fan forward broadband noise data or the fan rearward broadband noise data is obtained and stored, and the directivity correction data of the fan forward broadband noise component or the fan rearward broadband noise component is extracted.

[0050] In one embodiment, the extraction of directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further includes the following steps:

[0051] According to the blade passing frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, the maximum value is subtracted from the obtained data, the fan front transmission single-tone noise data or the fan rear transmission single-tone noise data is obtained and stored, and the extraction of the directivity correction data of the fan front transmission single-tone noise component or the fan rear transmission single-tone noise component is completed.

[0052] In one embodiment, the extraction of directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further includes the following steps:

[0053] According to the inverse of the blade pass frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, and the maximum value is subtracted from the obtained data to obtain the fan forward shock wave noise data and store it, completing the extraction of the directivity correction data of the fan forward shock wave noise component.

[0054] In one embodiment, the noise time domain data is a plurality of groups of data, each group of data is a D*N matrix, D is the number of audio acquisition devices, and N is the total number of time domain samples of each audio acquisition device.

[0055] In one embodiment, the blade passing frequency multiples further include 1BPF, 2BPF and 3BPF.

[0056] In one embodiment, the blade passes through the reciprocal of the frequency multiple, further including 1 / 8BPF, 1 / 4BPF and 1 / 2BPF.

[0057] In one embodiment, the training of the characteristic matrix and the directivity correction data further includes:

[0058] During training, grid search combined with k-fold cross validation was used for hyperparameter optimization.

[0059] In one embodiment, the aerodynamic features include:

[0060] Environmental pressure P amb , ambient humidity RH, relative conversion speed N corr, physical speed rpm, blade passing frequency BPF, blade tip Mach number Mr, outer duct pressure ratio Prb, flow tube total temperature T0, outer duct conversion flow

[0061] In order to achieve the above object, the present invention provides a fan noise prediction system based on machine learning, comprising:

[0062] a memory for storing instructions executable by a processor;

[0063] A processor is used to execute the instructions to implement any of the methods described above.

[0064] In order to achieve the above object, the present invention provides a computer-readable medium on which computer instructions are stored, wherein when the computer instructions are executed by a processor, any of the above methods is executed.

[0065] The present invention proposes a fan noise prediction method and system based on machine learning, which innovatively combines test data and machine learning to update the F3 function. It can effectively distinguish the differences in directivity correction values ​​caused by changes in working conditions and specific single-tone frequencies, and the model prediction accuracy is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The above and other features, properties and advantages of the present invention will become more apparent through the following description in conjunction with the accompanying drawings and embodiments, in which the same reference numerals always represent the same features, wherein:

[0067] Figure 1 A simplified flowchart of a fan noise prediction method according to a Heidmann framework according to an embodiment of the present invention is disclosed;

[0068] Figure 2 A detailed flow chart of a fan noise prediction method based on machine learning according to an embodiment of the present invention is disclosed;

[0069] Figure 3 A flow chart of a method for calculating the F3 function of each noise component according to an embodiment of the present invention is disclosed;

[0070] Figure 4 A schematic diagram of a training process of a fitting model according to an embodiment of the present invention is disclosed;

[0071] Figure 5 A comparison chart of the prediction results between this method and conventional methods is revealed. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the invention and are not used to limit the invention.

[0073] The original Heidmann fan noise prediction model, whose parameter settings of each part of the function are based on the fan configuration noise test data in the early years of the United States, is no longer applicable to the prediction of fan noise of new aircraft engines. Therefore, various research institutions have developed an updated version of the fan noise prediction model based on the Heidmann model and combined with the subsequent new fan configuration test data, that is, using the new test data to update the five links of the Heidmann model parameters, such as Hough1996, Kontos1996 and Krejsa2014 models. The fan noise prediction method based on machine learning proposed in the present invention is mainly aimed at the update of the F3 directivity correction function.

[0074] Figure 1 A simplified flow chart of a fan noise prediction method according to a Heidmann framework of an embodiment of the present invention is disclosed. Figure 1 As shown, the fan noise prediction method based on machine learning proposed in the present invention adopts the Heidmann framework and includes the following steps:

[0075] Step S1, input microphone array layout parameters, fan aerodynamic and structural parameters, and atmospheric environment parameters;

[0076] Step S2, respectively update and calculate the Heidmann model parameters at each specific directional angle to obtain the spectrum and OASPL spectrum directivity of the fan forward noise and fan rearward noise at the specific directional angle.

[0077] The overall sound pressure level (OASPL) is the total sound pressure level in the frequency band of interest obtained by superimposing the energy of all the sound pressure levels in the frequency band. The unit is decibel (dB), which can reflect the directivity of far-field noise.

[0078] Wherein, the Heidmann model parameters include a directivity correction function F3;

[0079] In the step S2, a fitting model of each noise component is generated based on a machine learning algorithm, and the directivity correction function F3 is updated by the fitting model of each noise component.

[0080] Figure 2 A detailed flow chart of a fan noise prediction method based on machine learning according to an embodiment of the present invention is disclosed. Figure 2 As shown, the step S2 further includes the following steps:

[0081] Step S21, calculating the mechanical work related to the temperature rise and the specific work related to the flow rate;

[0082] Step S22, respectively calculating the blade tip relative Mach number correction function F1 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0083] Step S23, respectively calculating the rotation-to-static distance correction function F2 of the fan forward broadband noise, the fan forward single-tone noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0084] Step S24, respectively calculating the directivity correction function F3 and the spectrum shape correction function F4 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0085] Step S25, respectively calculating the directivity correction function F3 and the spectrum shape correction function F4 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0086] Step S26, calculating atmospheric attenuation and spherical distance diffusion correction parameters;

[0087] Step S27, summing the parameters obtained in steps S21 to S26 to obtain the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise;

[0088] Step S28, summing the energy of the fan forward broadband noise and the fan forward single tone noise to obtain the spectrum and OASPL spectrum directivity of the fan forward noise at each directional angle;

[0089] The energy of the fan rear-transmitted broadband noise and the fan rear-transmitted single-tone noise is summed to obtain the spectrum and OASPL spectrum directivity of the fan rear-transmitted noise at each directional angle.

[0090] Furthermore, in the final energy summation stage of step S28, the noise component of the fan forward shock wave noise is added only when it is determined that the Mach number is greater than 1, and the fan forward broadband noise, the fan forward single-tone noise and the fan forward shock wave noise are energy-summed to obtain the spectrum of the fan forward noise at each directional angle and the OASPL spectrum directivity.

[0091] More specifically, the OASPL of the forward and backward propagation noise at this directional angle can be obtained by summing the total spectrum data of the forward and backward propagation noise in 24 frequency bands.

[0092] like Figure 1 and Figure 2 As shown in the figure, the Heidmann model can be used to calculate the SPL-TOB data of the above five parts of noise at various directional angles. On this basis, the spherical radius diffusion correction and atmospheric attenuation correction parameters are supplemented to form a complete fan noise empirical prediction process, which is called the Heidmann framework.

[0093] Sound pressure level (SPL) refers to a scale based on a logarithmic scale, with reference to a reference quantity representing the hearing threshold, to measure the effective sound pressure of sound, and the unit is decibel (dB).

[0094] SPL-TOB data refers to the sound pressure level in one-third octave bands.

[0095] The empirical formula for the noise component of each part of the Heidmann model is:

[0096]

[0097] Among them, step S21 corresponds to the 20log of the above formula 10 (ΔT / ΔT0)+10log 10 (m / m0(;

[0098] Step S22 corresponds to F1[(M TR ),(M TR ) D ];

[0099] Step S23 corresponds to F2 (RSS) of the above formula;

[0100] Step S24 corresponds to F3(θ) in the above formula;

[0101] Step S25 corresponds to the above formula

[0102] like Figure 1 and Figure 2 The method shown is the framework basis for fan noise prediction. For a fan of a new configuration of interest, its noise test data needs to be used to update the parameter part in steps S21 to S25.

[0103] The fan noise prediction method based on machine learning proposed in the present invention only focuses on the update of the directivity correction function F3, generates fitting models of various noise components based on the machine learning algorithm, and updates the directivity correction function F3 of the aircraft engine fan noise empirical model (Heidmann model) through the fitting models of various noise components. The directivity correction function F3 has a great influence on the overall shape of the spectrum and the directivity accuracy at each directional angle.

[0104] Figure 3 A flow chart of a method for calculating the F3 function of each noise component according to an embodiment of the present invention is disclosed, such as Figure 3 As shown, the basis for updating the F3 function is to first conduct noise tests on the relevant fan configurations, and then divide the collected test data into the above five parts (fan forward broadband noise, fan forward single-tone noise, fan forward shock wave noise, fan rear broadband noise and fan rear single-tone noise) noise components through multiple processing. Finally, the SPL-TOB data of these five parts of noise are used to calculate the directivity-corrected SPL data under each operating condition and each frequency.

[0105] The conventional F3 function update method is to obtain each directivity correction data array from the noise test data under each working condition, and make a statistical average of multiple working conditions and multiple frequencies for each of the five noise components, that is, the F3 function of each noise component will eventually be set to a fixed array summarized according to the test data, as shown in Table 1. Each directivity angle corresponds to a specific SPL value (corresponding to the fan forward broadband noise, fan forward single-tone noise, fan forward shock wave noise, fan rear broadband noise and fan rear single-tone noise, respectively), without distinguishing between working conditions. For example, the Hough1996, Kontos1996 and Krejsa2014 models developed based on the Heidmann framework, but this method does not consider the impact of changes in working conditions, and its versatility and fitting accuracy are low.

[0106] Table 1 Conventional F3 correction function (fan forward noise)

[0107]

[0108]

[0109] In fact, different working conditions and specific single-tone frequencies (TOB center frequencies related to the BPF multiples that the Heidmann framework focuses on) should have corresponding directivity correction data.

[0110] The fan noise prediction method based on machine learning proposed in the present invention adopts the fitting model of each noise component of the aircraft engine fan based on machine learning to update the F3 function, and combines the aerodynamic characteristic parameters related to the noise directivity (such as external converted flow rate, BPF, relative converted speed, directional angle, ambient pressure and humidity, etc.) and the directivity correction data SPL values ​​of the five parts of noise under various working conditions and frequencies, and trains the fitting models of the five noise components respectively, so as to realize automatic fitting of F3 data under various conditions. Compared with the traditional F3 update method, the prediction performance of the model can be effectively improved.

[0111] Machine learning is an engineering-oriented programming paradigm based on statistical analysis. It can find statistical structures from examples and then find rules to automate tasks. Its system is trained by data rather than explicitly written by programs. The "learning" here refers to the automatic search process of finding better data representations.

[0112] The present invention proposes a fan noise prediction method and system based on machine learning, which adopts a machine learning algorithm to update the F3 function of the empirical model of aircraft engine fan noise, and innovatively applies the LGBM regression algorithm to the fitting of the directivity-corrected SPL data of the Heidmann model. Moreover, by adjusting the BPF item in the characteristic parameter accordingly, the directivity-corrected data of the single-tone frequencies concerned by the Heidmann framework: 1 / 8BPF, 1 / 4BPF, 1 / 2BPF, 1BPF, 2BPF, and 3BPF can be further accurately fitted, instead of generally correcting all BPF-related single tones with shock wave F3 and single tone F3.

[0113] Generating fitting models for each noise component based on a machine learning algorithm further includes the following steps:

[0114] Input the noise time domain data under each working condition;

[0115] Divide the data into training set and test set;

[0116] Calculate the power spectrum of the time domain sample data collected by each audio acquisition device;

[0117] Perform single-tone broadband separation on the power spectrum to obtain broadband power spectrum and the first several axial frequency single-tone power spectra;

[0118] Extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum;

[0119] Select aerodynamic features and generate a feature matrix in combination with the pointing angle, and correspond the feature matrix to the extracted directivity correction data;

[0120] Based on a machine learning algorithm, the characteristic matrix and the directivity correction data are trained to generate a fitting model corresponding to the noise component. The fitting model is used to respond to the input characteristic matrix and output the SPL correction value of the corresponding noise component.

[0121] Furthermore, the extraction of directivity correction data is completed according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum, further comprising the following steps:

[0122] The TOB-SPL data in matrix form is obtained according to the broadband power spectrum, the energy of the rows of the matrix is ​​summed, the maximum value is subtracted from the obtained data, the fan forward broadband noise data or the fan rearward broadband noise data is obtained and stored, and the directivity correction data of the fan forward broadband noise component or the fan rearward broadband noise component is extracted.

[0123] Furthermore, the extraction of directivity correction data is completed according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum, further comprising the following steps:

[0124] According to the blade passing frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, the maximum value is subtracted from the obtained data, the fan front transmission single-tone noise data or the fan rear transmission single-tone noise data is obtained and stored, and the extraction of the directivity correction data of the fan front transmission single-tone noise component or the fan rear transmission single-tone noise component is completed.

[0125] The blade passing frequency multiples further include 1BPF, 2BPF and 3BPF.

[0126] Furthermore, the extraction of directivity correction data is completed according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum, further comprising the following steps:

[0127] According to the inverse of the blade pass frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, and the maximum value is subtracted from the obtained data to obtain the fan forward shock wave noise data and store it, completing the extraction of the directivity correction data of the fan forward shock wave noise component.

[0128] The reciprocal of the blade passing frequency multiple further includes 1 / 8BPF, 1 / 4BPF and 1 / 2BPF.

[0129] In this embodiment, the aerodynamic features include:

[0130] P amb Ambient pressure, RH ambient humidity, N corr Relative conversion speed, rpm physical speed, BPF blade passing frequency, Mr blade tip Mach number, Prb outer duct pressure ratio, T0 flow tube total temperature, External conversion flow.

[0131] It can be understood as updating F3(θ) to the following expression:

[0132]

[0133] Updated F3 function

[0134] It is not a specific function formula, but a fitting model based on machine learning saved in .pkl format. A fitting model is trained for each of the five noise components of the fan, such as Figure 3 As shown, in represents 'inlet' imported noise, i.e. forward noise; ex represents 'exhaust' export noise, i.e. backward noise; BB represents 'broadband' broadband noise; 'Tone' represents single tone noise, which will be further divided into 1BPF, 2BPF and 3BPF according to the prediction method set by the Heidmann framework; 'Shock' represents shock wave noise, which will be further divided into 1 / 8BPF, 1 / 4BPF and 1 / 2BPF according to the prediction method set by the Heidmann framework.

[0135] Figure 4 A schematic diagram of a training process of a fitting model according to an embodiment of the present invention is disclosed, such as Figure 4 As shown, the training method of the fitting model of the fan forward noise is explained by taking the use of the fan forward noise test data as an example.

[0136] Input the noise time domain data under each working condition, wherein the noise time domain data is T groups of data, each group of data is a D*N matrix, D is the number of audio acquisition devices, N is the total number of time domain samples of each audio acquisition device, and preferably, the audio acquisition device is a microphone;

[0137] Divide 80% of the data into a training set and 20% into a test set (including airworthiness conditions, such as N corr =79.9% corresponding to the overflight condition);

[0138] Calculate the power spectrum for each row of data, that is, the N time domain sample data collected by each audio acquisition device;

[0139] Perform single-tone broadband separation on the power spectrum to obtain broadband power spectrum and the first 100 axial frequency single-tone power spectra;

[0140] According to the broadband power spectrum, the TOB-SPL data is obtained, which is a 24*D matrix. The energy of the rows of the matrix is ​​summed to obtain D data. The maximum value is subtracted to obtain the F3_in_BB array, which is the broadband noise data of the fan forward transmission.

[0141] According to the single-tone power spectrum, extract the Vth, 2Vth, and 3Vth shaft frequency single-tone data (V is the number of blades), calculate the SPL data respectively, that is, the SPL data under 1BPF, 2BPF, and 3BPF, and get 3*D data. Subtract the maximum value of each row of data to get 3 F3_in_Tone arrays, that is, the fan forward transmission single-tone noise data;

[0142] According to the single-tone power spectrum, extract the 1 / 8V, 1 / 4V, and 1 / 2V shaft frequency single-tone data (V is the number of blades), calculate the SPL data respectively, that is, the SPL data under 1 / 8BPF, 1 / 4BPF, and 1 / 2BPF, and get 3*D data. Subtract the maximum value of each row of data to get 3 F3_in_Shock arrays, that is, the fan forward shock wave noise data;

[0143] Continue the above extraction operation until all 0.8T*D rows of training data are processed. This completes the extraction of directivity correction data, and the data corresponding to each component is stored in the form of column vectors;

[0144] After feature importance analysis, 9 aerodynamic features are finally selected. Combined with the feature of pointing angle, the feature matrix required for model training is obtained. The feature matrix should correspond one by one according to the working conditions and the pointing correction data stored. Among them, the 9 aerodynamic features include the ambient pressure P amb , ambient humidity RH, relative conversion speed N corr , physical speed rpm, blade passing frequency BPF, blade tip Mach number Mr, outer duct pressure ratio Prb, flow tube total temperature T0, outer duct conversion flow

[0145] Based on the machine learning algorithm, the lightgbm module in Python is used to train the above feature matrix and the corresponding directivity correction data.

[0146] During training, grid search combined with k-fold cross validation is used to optimize hyperparameters and generate a fitting model corresponding to the noise component. The joblib module is used to save the final optimized and trained model in the .pkl format.

[0147] In this embodiment, the machine learning algorithm is the LGBM machine learning algorithm, which is implemented using the lightgbm module in Python.

[0148] Light Gradient Boosting Machine (LGBM), developed by Microsoft, is an open source, fast, distributed, high-performance lightweight gradient boosting machine learning framework based on decision trees. Its main idea is to use weak classifiers (decision trees) for iterative training to obtain the optimal model algorithm.

[0149] The LGBM machine learning algorithm framework has the advantages of fast training speed and low memory usage. It is one of the best algorithm frameworks for predicting small and medium-sized structure data. According to tests, the training of data within 100 working conditions takes less than one minute, and the response time of the trained model to give prediction results is in milliseconds, which meets the needs of fast prediction of engineering experience models. And the trained model can be saved in .pkl format for easy calling.

[0150] So far, we have obtained Figure 3 The three directivity correction data shown are the fan forward broadband noise fitting model F3_in_BB.pkl, the fan forward single tone noise fitting model F3_in_Tone.pkl, and the fan forward shock wave noise fitting model F3_in_Shock.pkl.

[0151] In this embodiment, grid search combined with k-fold cross validation is used to perform hyperparameter optimization during training.

[0152] Cross-validation can solve the problem of insufficient data set size and parameter tuning. There are three main methods: simple cross-validation (HoldOut test), k-fold cross-validation, and bootstrap.

[0153] In order to solve the shortcomings of simple cross-validation, k-fold cross-validation includes the following steps:

[0154] First, all samples are divided into k sample subsets of equal size;

[0155] Traverse these k subsets in turn, each time taking the current subset as the validation set and all other samples as the training set to train and evaluate the model;

[0156] Finally, the average of the k evaluation indicators is taken as the final evaluation indicator.

[0157] The training of fan back-propagation noise data is similar, and the following results are obtained: Figure 3 The fan rear-transmitted broadband noise fitting model F3_ex_BB.pkl and the fan rear-transmitted single-tone noise fitting model F3_ex_Tone.pkl are shown. So far, using the noise test data collected from the first batch of tests on the new configuration fan, five fitting models F3_in_BB.pkl, F3_in_Tone.pkl, F3_in_Shock.pkl, F3_ex_BB.pkl, and F3_ex_Tone.pkl have been trained to predict the directivity correction data, thereby realizing the update and improvement of the directivity correction function F3.

[0158] For unknown data outside of model training (test data sets), we only need to input the characteristic matrix parameters (9 aerodynamic parameters + pointing angle) into the corresponding fitting model to obtain the SPL value that should be corrected for each noise component. In addition, this fan configuration does not need to be equipped with microphones in the subsequent testing phase, so the noise level can be more accurately assessed. The same is true for its derivative models, which can greatly save testing costs.

[0159] Compared with the fixed array obtained by the conventional F3 function update method as shown in Table 1, the fan noise prediction method and system based on machine learning proposed in the present invention combines noise test data and machine learning to update the F3 function in the original Heidmann model, and innovatively uses the LGBM algorithm to carry out statistical learning fitting analysis, so that the F3 function of each noise component becomes a function of aerodynamic characteristic parameters and pointing angle. The Heidmann model after updating the F3 part in this way has a significant improvement in the noise prediction accuracy of the new configuration fan of concern.

[0160] Figure 5 This is a comparison chart of the prediction results of the Hedimann model after updating by the method of the present invention and the conventional method, as shown in Figure 5 Under a certain test overflight condition shown in the figure, the fan forward noise prediction performance of the Heidmann model with the F3 function updated by this method and the Heidmann model with the F3 function updated by the traditional method are compared. On average, the OASPL directivity error can be reduced by 3dB at each angle, and it is obvious that the overall trend fitting performance is better.

[0161] The present invention proposes a fan noise prediction system based on machine learning. The fan noise prediction system based on machine learning may include an internal communication bus, a processor, a read-only memory (ROM), a random access memory (RAM), a communication port, and a hard disk. The internal communication bus can realize data communication between components of the fan noise prediction system based on machine learning. The processor can make judgments and issue prompts. In some embodiments, the processor can be composed of one or more processors.

[0162] The communication port can realize data transmission and communication between the machine learning-based fan noise prediction system and external input / output devices. In some embodiments, the machine learning-based fan noise prediction system can send and receive information and data from the network through the communication port. In some embodiments, the machine learning-based fan noise prediction system can transmit and communicate data with external input / output devices in a wired form through the input / output terminal.

[0163] The fan noise prediction system based on machine learning may also include different forms of program storage units and data storage units, such as hard disks, read-only memories (ROMs) and random access memories (RAMs), which can store various data files used for computer processing and / or communication, as well as possible program instructions executed by the processor. The processor executes these instructions to implement the main part of the method. The results of the processor processing are transmitted to an external output device through a communication port and displayed on a user interface of the output device.

[0164] For example, the implementation process file of the above-mentioned fan noise prediction system based on machine learning can be a computer program, which is saved in a hard disk and can be recorded in a processor for execution to implement the method of the present invention.

[0165] When the implementation process file of the fan noise prediction method based on machine learning is a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described herein can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0166] The present invention proposes a fan noise prediction method and system based on machine learning, which innovatively combines test data and machine learning to update the F3 function. It can effectively distinguish the differences in directivity correction values ​​caused by changes in working conditions and specific single-tone frequencies, and the model prediction accuracy is effectively improved.

[0167] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0168] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0169] Those skilled in the art will appreciate that information, signals, and data may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips cited throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or optical particles, or any combination thereof.

[0170] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0171] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or executed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0172] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0173] The above embodiments are provided for persons familiar with the art to implement or use the present invention. Personnel familiar with the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited to the above embodiments, but should be the maximum scope of the innovative features mentioned in the claims.

Claims

1. A fan noise prediction method based on machine learning, characterized in that: The following steps are involved: Step S1, input microphone array layout parameters, fan aerodynamic and structural parameters, and atmospheric environment parameters; Step S2, respectively updating and calculating the Heidmann model parameters at each specific directional angle, and obtaining the spectrum and OASPL spectrum directivity of the fan forward noise and the fan backward noise at the specific directional angle; Wherein, the Heidmann model parameters include a directivity correction function F3; In the step S2, a fitting model of each noise component is generated based on a machine learning algorithm, and the directivity correction function F3 is updated by the fitting model of each noise component.

2. The fan noise prediction method based on machine learning according to claim 1, characterized in that: The Heidmann model parameters also include mechanical work related to temperature rise, specific work related to flow rate, blade tip relative Mach number correction function F1, rotor-to-static distance correction function F2, and spectrum shape correction function F4; The fan forward noise includes fan forward broadband noise, fan forward single tone noise and fan forward shock wave noise; The fan back-transmitted noise includes fan back-transmitted broadband noise and fan back-transmitted single-tone noise; The step S2 further comprises the following steps: Step S21, calculating the mechanical work related to the temperature rise and the specific work related to the flow rate; Step S22, respectively calculating the blade tip relative Mach number correction function F1 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise; Step S23, respectively calculating the rotation-to-static distance correction function F2 of the fan forward broadband noise, the fan forward single-tone noise, the fan rearward broadband noise and the fan rearward single-tone noise; Step S24, respectively calculating the directivity correction function F3 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise; Step S25, respectively calculating the spectrum shape correction function F4 of the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise; Step S26, calculating atmospheric attenuation and spherical distance diffusion correction parameters; Step S27, summing the parameters obtained in steps S21 to S26 to obtain the fan forward broadband noise, the fan forward single-tone noise, the fan forward shock wave noise, the fan rearward broadband noise and the fan rearward single-tone noise; Step S28, summing the energy of the fan forward broadband noise and the fan forward single-tone noise to obtain the spectrum and OASPL spectrum directivity of the fan forward noise at each directional angle, and summing the energy of the fan rearward broadband noise and the fan rearward single-tone noise to obtain the spectrum and OASPL spectrum directivity of the fan rearward noise at each directional angle.

3. The fan noise prediction method based on machine learning according to claim 2, characterized in that: The step S28 further comprises: When the Mach number is greater than 1, the fan forward broadband noise, fan forward single-tone noise and fan forward shock wave noise are summed to obtain the spectrum and OASPL spectrum directivity of the fan forward noise at each directional angle.

4. The fan noise prediction method based on machine learning according to claim 1, characterized in that: The method of generating a fitting model for each noise component based on a machine learning algorithm further comprises the following steps: Input the noise time domain data under each working condition; Divide the data into training set and test set; Calculate the power spectrum of the time domain sample data collected by each audio acquisition device; Perform single-tone broadband separation on the power spectrum to obtain broadband power spectrum and the first several axial frequency single-tone power spectra; Extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum; Select aerodynamic features and generate a feature matrix in combination with the pointing angle, and correspond the feature matrix to the extracted directivity correction data; Based on a machine learning algorithm, the characteristic matrix and the directivity correction data are trained to generate a fitting model corresponding to the noise component. The fitting model is used to respond to the input characteristic matrix and output the SPL correction value of the corresponding noise component.

5. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The step of extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further comprises the following steps: The TOB-SPL data in matrix form is obtained according to the broadband power spectrum, the energy of the rows of the matrix is ​​summed, the maximum value is subtracted from the obtained data, the fan forward broadband noise data or the fan rearward broadband noise data is obtained and stored, and the directivity correction data of the fan forward broadband noise component or the fan rearward broadband noise component is extracted.

6. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The step of extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further comprises the following steps: According to the blade passing frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, the maximum value is subtracted from the obtained data, the fan front transmission single-tone noise data or the fan rear transmission single-tone noise data is obtained and stored, and the extraction of the directivity correction data of the fan front transmission single-tone noise component or the fan rear transmission single-tone noise component is completed.

7. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The step of extracting the directivity correction data according to the axial frequency single-tone data corresponding to the broadband power spectrum and / or the single-tone power spectrum further comprises the following steps: According to the inverse of the blade pass frequency multiples, the single-tone power spectrum corresponding to the shaft frequency single-tone data is extracted, the SPL data is calculated, and the maximum value is subtracted from the obtained data to obtain the fan forward shock wave noise data and store it, completing the extraction of the directivity correction data of the fan forward shock wave noise component.

8. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The noise time domain data is a plurality of groups of data, each group of data is a D*N matrix, D is the number of audio acquisition devices, and N is the total number of time domain samples of each audio acquisition device.

9. The fan noise prediction method based on machine learning according to claim 6, characterized in that: The blade passing frequency multiples further include 1BPF, 2BPF and 3BPF.

10. The fan noise prediction method based on machine learning according to claim 7, characterized in that: The reciprocal of the blade passing frequency multiple further includes 1 / 8BPF, 1 / 4BPF and 1 / 2BPF.

11. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The training of the characteristic matrix and the directivity correction data further includes: During training, grid search combined with k-fold cross validation was used for hyperparameter optimization.

12. The fan noise prediction method based on machine learning according to claim 4, characterized in that: The aerodynamic features include: P amb Ambient pressure, RH ambient humidity, N corr Relative conversion speed, rpm physical speed, BPF blade passing frequency, Mr blade tip Mach number, Prb outer duct pressure ratio, T0 flow tube total temperature, External conversion flow.

13. A fan noise prediction system based on machine learning, comprising: a memory for storing instructions executable by a processor; A processor, configured to execute the instructions to implement the method according to any one of claims 1 to 12.

14. A computer-readable medium having computer instructions stored thereon, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1 to 12 is performed.

Images