Conical embedded tube type Helmholtz resonator structure parameter generation method and device

Abstract

The invention provides a conical embedded tube type Helmholtz resonator structure parameter generation method and device, and belongs to the field of mechanical engineering, and the method comprises the steps: obtaining sound absorption coefficient curves of a plurality of sample conical embedded tubes with known structure parameters; training the constructed forward prediction network by taking the structural parameters as input and the sequence extracted from the corresponding sound absorption coefficient curve as a label to obtain a trained forward prediction network; the constructed reverse design network and the trained forward prediction network are connected in series, a sequence extracted by an acoustic absorption coefficient curve is used as an output label and an input of the series network at the same time, forward prediction network parameters are fixed, and reverse design network parameters are trained; wherein the output of the reverse design network is the input of the forward prediction network, and the trained reverse design network is used for generating structural parameters. According to the method, the design steps are simplified, the design time is shortened, meanwhile, the design complexity is reduced, and the accuracy of structural parameter design is improved.

Description

Technical Field

[0001] The invention relates to the field of mechanical engineering, and in particular to a method and a device for generating structural parameters of a conical embedded tube Helmholtz resonator. Background Art

[0002] Nowadays, in the field of acoustics, the sound absorption effect of acoustic metamaterials in low-frequency and broadband environments has attracted extensive attention. They have shown a variety of structural types and a variety of structural parameters.

[0003] At present, many applications use genetic algorithms and other methods to optimize these parameters through continuous iterations to obtain the optimal parameter combination. However, genetic algorithms and other methods have problems such as slow convergence, local optimal traps, high computational costs, and high complexity, which affect the accuracy of the results and the efficiency of use. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a method and device for generating structural parameters of a conical embedded tube Helmholtz resonator.

[0005] The present invention provides a method for generating structural parameters of a conical embedded tube Helmholtz resonator, comprising: obtaining sound absorption coefficient curves of a plurality of sample conical embedded tubes with known structural parameters; taking the structural parameters as input and a sequence extracted from the corresponding sound absorption coefficient curve as a label, training a constructed forward prediction network to obtain a trained forward prediction network; connecting a constructed reverse design network and the trained forward prediction network in series to obtain a series network, taking the sequence extracted from the sound absorption coefficient curve as both an output label and an input of the series network, fixing the parameters of the forward prediction network, and training the parameters of the reverse design network; wherein the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for generating structural parameters.

[0006] According to a method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the present invention, the structural parameters include the inlet diameter, outlet diameter and cone neck length of the conical embedded tube, and the structural types of the forward prediction network and the reverse design network include fully connected neural networks.

[0007] According to a method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the present invention, after obtaining the sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters, the method further includes: extracting the sound absorption coefficient of each frequency point according to the sound absorption coefficient curve to obtain the sequence.

[0008] According to a method for generating structural parameters of a conical embedded tubular Helmholtz resonator provided by the present invention, after training the reverse design network parameters, the method further includes: obtaining a required sound absorption coefficient curve for the conical embedded tubular Helmholtz resonator whose structural parameters are to be determined, and extracting a sequence according to the required sound absorption coefficient curve; inputting the sequence into a structural parameter generation model to obtain the structural parameters to be determined; wherein the structural parameter generation model is a trained reverse design network.

[0009] According to a method for generating structural parameters of a conical embedded tubular Helmholtz resonator provided by the present invention, the method of obtaining a required sound absorption coefficient curve comprises: obtaining a required sound absorption coefficient value at a target frequency point; determining the sound absorption coefficients corresponding to a plurality of key frequency points based on the required sound absorption coefficient value at the target frequency point and in combination with the shape of the required sound absorption curve; and obtaining the required sound absorption coefficient curve by quadratic spline interpolation fitting according to the sound absorption coefficients corresponding to the target frequency point and the plurality of key frequency points.

[0010] According to a method for generating structural parameters of a conical embedded tubular Helmholtz resonator provided by the present invention, the training of the constructed forward prediction network includes: calculating the mean square error as a forward loss value based on the output sequence of the forward prediction network and the sequence extracted from the sound absorption coefficient curve, and updating the parameters of the forward prediction network based on the forward loss value.

[0011] According to a method for generating structural parameters of a conical embedded tubular Helmholtz resonator provided by the present invention, the training of the reverse design network parameters includes: calculating a mean square error as a reverse loss value based on an input sequence of the reverse design network and an output sequence of a forward prediction network; and updating the parameters of the reverse design network based on the reverse loss value.

[0012] The present invention also provides a device for generating structural parameters of a conical embedded tube Helmholtz resonator, comprising: a data input module, used to obtain sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; a first training module, used to use the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label, to train a constructed forward prediction network to obtain a trained forward prediction network; a second training module, used to connect the constructed reverse design network and the trained forward prediction network in series to obtain a series network, using the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fixing the forward prediction network parameters to train the reverse design network parameters; wherein the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for structural parameter generation.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for generating structural parameters of a conical embedded tube Helmholtz resonator as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating structural parameters of a conical embedded tube Helmholtz resonator as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for generating structural parameters of a conical embedded tube Helmholtz resonator as described in any one of the above is implemented.

[0016] The method and device for generating structural parameters of a conical embedded tubular Helmholtz resonator provided by the present invention determine a network model by jointly training a forward prediction network and a reverse design network. The obtained reverse design network can accurately establish a corresponding relationship between structural parameters and sound absorption coefficients, thereby simplifying the design steps, shortening the design time, reducing the design complexity, and improving the accuracy of structural parameter design. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a schematic flow chart of a method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the present invention;

[0019] Figure 2 It is a schematic diagram of the series network structure provided by the present invention;

[0020] Figure 3 It is a schematic diagram of the structure parameter generation model provided by the present invention;

[0021] Figure 4 This is one of the schematic diagrams of the structure of the conical embedded tube Helmholtz resonator provided by the present invention;

[0022] Figure 5 This is the second structural schematic diagram of the conical embedded tube Helmholtz resonator provided by the present invention;

[0023] Figure 6It is a schematic diagram comparing the prediction results of the neural network and the analytical solution under different parameters in the forward prediction network provided by the present invention;

[0024] Figure 7 It is a schematic diagram comparing the sound absorption coefficient curve predicted by the neural network provided by the present invention with the original curve;

[0025] Figure 8 It is a schematic diagram of a simulation model drawn by COMSOL provided by the present invention;

[0026] Fig. 9 It is a schematic diagram comparing the low-frequency sound absorption curves of finite element simulation and interpolation provided by the present invention;

[0027] Fig.10 It is a structural schematic diagram of a device for generating structural parameters of a conical embedded tube-type Helmholtz resonator provided by the present invention;

[0028] Fig.11 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] Combine the following Figure 1-Figure 11 The invention describes a method and device for generating structural parameters of a tapered embedded tube Helmholtz resonator.

[0031] The conical embedded tube Helmholtz resonator is composed of a conical embedded tube embedded in a square Helmholtz resonator. In the design of this sound absorber, the core sound absorbing component is a conical embedded tube. The calculation formula of the sound absorption coefficient α of this sound absorber is:

[0032]

[0033] Among them, Z NTHRs is the surface impedance, which is calculated as:

[0034]

[0035] Where Z0=ρ0C0 is the dielectric characteristic impedance of air, the speed of sound c0=340m / s, and the air density ρ0=1.21kg / m 3 .

[0036]

[0037] Where ω is the angular frequency, l a is the cone neck length, J n is the first kind Bessel function and the nth order Bessel function. σ x and k x are the porosity and perforation constant, respectively, which vary with the x position along the cone neck length and can be expressed as d x =d a1 +(d a2 -d a1 )x / l a , σ x =π(d x / 2) 2 / A. In the formula, i = 1, 2. R s =(2ωρ0η) 1 / 2 / 2 is the surface impedance, so the total end correction of the tapered neck can be Z end =R ext,1 +R ext,2 +X ext,1 +X ext,2 . Where R ext,1 and R ext,2 are the tapered neck inlet and outlet resistance corrections, respectively, ext,1 and X ext,2 The corresponding reactance correction. The acoustic impedance of the structure is expressed as Where S a =π(d a1 +d a2 ) 2 / 16 is the average cross-sectional area of ​​the tapered embedded tube, c c =ω / k c is the complex air velocity, ρ c =ρ0 / Ψ v is the complex air density, k c =k0[(γ-(γ-1)Ψ h ) / Ψ v ] 1 / 2 is the complex wave number. V is the volume of the cavity without the conical neck buried, V = c -S,S c =πr c 2 , V taper =π(l a -b c )[(r1+b) 2 +(r1+b)(r2+b)+(r2+b) 2 ] / 3, S in the impedance formula aa =π(d a1 2 +da1 d a2 +d a2 2 ) / 12.

[0038] Through the analysis of theoretical formulas, the key structural parameters of the conical embedded tube Helmholtz resonator can be defined.

[0039] Figure 1 FIG. 1 is a flow chart of a method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the present invention, such as Figure 1 As shown, the present invention provides a method for generating structural parameters of a conical embedded tube Helmholtz resonator, comprising:

[0040] 101. Obtain sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters.

[0041] In one embodiment, the structural parameters include the inlet diameter, outlet diameter and cone neck length of the conical embedded tube. That is, the inlet diameter, outlet diameter and cone neck length of the conical embedded tube are selected as key structural parameters. The present invention uses these structural parameters as input data of the neural network data set for model training, and this example is used for explanation in the following.

[0042] Generally speaking, when generating structural parameters through a neural network model, it can be understood that the sound absorption coefficient curve is used as input and the corresponding known structural parameters are used as labels for training. However, this is not the case in the present invention. Instead, a neural network is first trained to predict the sound absorption coefficient curve through structural parameters, which is referred to as a forward prediction network. The forward prediction network can be constructed using a fully connected neural network. Such a design enables the network to learn the relationship between structural parameters and sound absorption characteristics, and the forward prediction network obtained in 102 can predict the sound absorption coefficient under given structural parameters. Then, the reverse design network is obtained in combination with the step 103 to further determine the final structural parameters.

[0043] 102. Taking the structural parameters as input and the corresponding sound absorption coefficient curve extracted sequence as a label, the constructed forward prediction network is trained to obtain a trained forward prediction network.

[0044] At the output of the forward prediction neural network, the sound absorption coefficient curve will be used as label data to correspond to the structural parameters of the conical embedded tube Helmholtz resonator at the input end. In order to ensure the smooth progress of the neural network training, each sound absorption coefficient curve needs to be discretized when the data set is established.

[0045] In one embodiment, after obtaining the sound absorption coefficient curves of a plurality of sample conical embedded tubes with known structural parameters, the method further includes: extracting the sound absorption coefficient of each frequency point according to the sound absorption coefficient curve to obtain the sequence.

[0046] Considering the discretization of the sound absorption coefficient curve, the output layer can be the sound absorption coefficient of each frequency point in the target frequency band, such as the sound absorption coefficient of each frequency point increasing by 1Hz, that is, the sequence extracted from the sound absorption coefficient curve is the sound absorption coefficient of each frequency point increasing by 1Hz. Then, random sampling is performed in the data space to generate a data set. After the data set is constructed, the data set is divided. The data set can be divided into a training set, a test set, and a validation set in a ratio of 80%, 10%, and 10%.

[0047] The function of the forward prediction network is to predict the sound absorption coefficient of the conical embedded tube Helmholtz resonator. A standard fully connected neural network architecture can be used. The input layer receives three key structural parameters of the resonator, namely the inlet diameter, outlet diameter and tube length of the conical tube. Since a single conical embedded tube Helmholtz resonator is defined by these three parameters, the input layer of the neural network is equipped with three neuron units, and the output layer contains the sound absorption coefficient of each frequency point in the target frequency band with an increment of 1Hz. This design enables the network to predict the sound absorption coefficient under given structural parameters by learning the relationship between structural parameters and sound absorption characteristics.

[0048] The fitting relationship of the established forward prediction network is as follows: α n =f n (d a2 ,d a1 ,l a ). A hidden layer is included between the input layer and the output layer. Considering the training efficiency, the number of neurons in the hidden layer is set to a power of 2. The Relu activation function is used in the hidden layer to enhance the nonlinear expression ability of the model. The expression of the Relu function is as follows: f(x) = max(0,x). In the initialization stage of training, the weights and biases of all network layers are randomly set by truncated normal distribution. After the forward prediction network is established, the previously established data set is integrated into the program to start the training process of the neural network, which can be executed using the Tensorflow framework.

[0049] 103. The constructed reverse design network is connected in series with the trained forward prediction network to obtain a series network, and the sequence extracted from the sound absorption coefficient curve is used as both the output label and input of the series network. The parameters of the forward prediction network are fixed, and the parameters of the reverse design network are trained; wherein the output of the reverse design network is the input of the forward prediction network, and the trained reverse design network is used for structural parameter generation.

[0050] like Figure 2As shown in the figure, the purpose of the reverse design network is to quickly calculate the corresponding structural parameters by inputting the target sound absorption coefficient curve. The architecture is opposite to the forward prediction network. The input layer is used to input the sound absorption coefficient of each frequency point in the target frequency band with an increment of 1 Hz. The output layer contains 3 neurons, representing the inlet diameter, outlet diameter and tube length of the conical tube. In the design of the reverse design network, the order of neurons in the input layer and the output layer is consistent with the corresponding relationship in the forward prediction network. For an independent reverse design network, there is P n =g n (α1,α2,α3,.....α, n ), where n represents the number of neurons in the input layer, and function g(x) is the inverse function of function f(x). The reverse design network and the forward prediction network need to work together during the training process, and the characteristics of the forward prediction network may increase the complexity of the reverse design network training. To ensure the effectiveness of the training, the present invention optimizes the training effect by enhancing the depth and width of the network.

[0051] The inverse design network contains three hidden layers, and the number of neurons in each hidden layer is a power of 2. The network weights are initialized using a truncated normal distribution with a standard deviation set to 0.1.

[0052] However, the reverse design network cannot form an independent and complete network, because there may be two (or more) sets of data (x1, R1) and (x2, R2) in the training set, where x represents the structural parameters, R represents the sound absorption coefficient, R1=R2, and x1≠x2. That is, after inputting R1(R2) and obtaining the output value x, there will be two or more different labeled values ​​x. j (j=1,2,…), different loss functions are calculated, which causes the network to fail to converge. After integrating the forward prediction network, the loss function of the entire network becomes the Euclidean distance between the predicted value and the labeled value of the sound absorption coefficient curve, which eventually makes the network converge. Therefore, it is necessary to integrate the forward prediction network and the reverse design network so that the two can work together.

[0053] Integrate the forward prediction network and the reverse design network: After completing the construction of the reverse design network, load the trained forward prediction network model and place it after the reverse design network. Since the front end of the forward prediction network has been trained to obtain a matrix of a certain dimension, which contains neurons in the output layer of the reverse design network, the forward prediction network and the reverse design network can be directly connected in series to form an autoencoder model. In this configuration, the output layer of the reverse design network directly corresponds to the input layer of the forward prediction network. At the same time, the output result of this layer of the trained model is the structural parameters of the required conical embedded tube Helmholtz resonator.

[0054] When training the entire deep learning model, the first half of the reverse design network needs to be trained separately first, so the second half of the forward prediction network is fixed so that it is not trained.

[0055] After the parameters are configured successfully, the network training begins. After iterations, the neural network model converges. So far, the neural network model established for the structural design of a single conical tube Helmholtz resonator has completed the construction and training stages.

[0056] For the target conical embedded tube whose structural parameters need to be determined, the trained neural network model can be used to design the structural parameters of the conical embedded tube Helmholtz resonator. Since the required design result is the output data of the output layer in the inverse design network, a special output port can be set in the model to capture the output data of the middle layer, which is the structural parameters of the conical embedded tube Helmholtz resonator calculated by the neural network.

[0057] The method for generating structural parameters of a conical embedded tubular Helmholtz resonator of the present invention determines a network model by jointly training a forward prediction network and a reverse design network. The obtained reverse design network can accurately establish a corresponding relationship between structural parameters and sound absorption coefficients, thereby simplifying the design steps, shortening the design time, reducing the design complexity, and improving the accuracy of structural parameter design.

[0058] In one embodiment, after the reverse design network parameters are trained, the method further includes: obtaining a required sound absorption coefficient curve for a conical embedded tube Helmholtz resonator whose structural parameters are to be determined, and extracting a sequence according to the required sound absorption coefficient curve; inputting the sequence into a structural parameter generation model to obtain the structural parameters to be determined; wherein the structural parameter generation model is a trained reverse design network.

[0059] For the conical embedded tube Helmholtz resonator whose structural parameters are to be determined, the required sound absorption coefficient curve is first determined. The sound absorption coefficient curve can be reasonably adjusted based on the previous sound absorption coefficient curve, such as the sound absorption coefficient curve optimized for the low-frequency part.

[0060] First, the required sound absorption coefficient curve is generated by the quadratic spline interpolation method, and then the generated curve is discretized into data points according to the frequency, and these data points are serialized and saved to form a target input data set with predetermined sound absorption characteristics. The trained reverse design network model is loaded, and the preprocessed input data is input into the network to obtain the structural parameters to be determined.

[0061] After the training of the series network model is completed, since the required structural parameters of the conical embedded tube Helmholtz resonator are extracted from the middle layer, in order to reduce the storage space occupied, the present invention can only discard the second half of the entire neural network model and only retain the reverse design network of the first half, such as Figure 3 As shown, only the first half of the network is used during use, which reduces the storage space occupied and achieves the purpose of lightweight.

[0062] In addition, the embodiments of the present invention can draw a corresponding sound absorption coefficient curve according to the sound absorption effect, and then obtain precise structural parameters that meet the requirements. For example, a specific conical embedded tube Helmholtz resonator can be customized according to the low-frequency sound absorption effect to achieve perfect sound absorption of sound waves at low frequencies while ensuring structural accuracy.

[0063] In one embodiment, obtaining the required sound absorption coefficient curve includes: obtaining a required sound absorption coefficient value at a target frequency point; determining the sound absorption coefficients corresponding to a plurality of key frequency points based on the required sound absorption coefficient value at the target frequency point and in combination with the shape of the required sound absorption curve; and obtaining the required sound absorption coefficient curve by quadratic spline interpolation fitting according to the sound absorption coefficients corresponding to the target frequency point and the plurality of key frequency points.

[0064] For example, if the required sound absorption coefficient curve exceeds 0.95 at 200 Hz, the sound absorption coefficient at the target frequency point 200 Hz can be set to 1. The shape of the required sound absorption curve can be a maximum value at 200 Hz, with both sides gradually decreasing. According to the shape of the required target sound absorption curve, a series of key frequency points are selected, and a sound absorption coefficient curve that meets the design requirements is constructed through quadratic spline interpolation technology.

[0065] In one embodiment, the structure types of the forward prediction network and the reverse design network are include Fully connected neural network, such as Figure 2 shown.

[0066] In one embodiment, the training of the constructed forward prediction network includes: calculating a mean square error as a forward loss value according to an output sequence of the forward prediction network and a sequence extracted from a sound absorption coefficient curve, and updating parameters of the forward prediction network according to the forward loss value.

[0067] For the forward prediction network, during the model training process, the mean square error (MSE) is used as the loss function to guide the optimization process of the network parameters. The definition of MSE is as follows: in Represents the target value, that is, the characteristic sound absorption coefficient value, α i Represents the predicted value of the neural network at this time, and n represents the number.

[0068] At the same time, the Adam optimization algorithm can be used to dynamically adjust the gradient of the model to ensure the convergence speed of the neural network. The update rule of each update is: where θ represents all learning parameters, and γ is the learning rate, and each learning parameter update has During the training process, the small batch gradient descent strategy is adopted. Specifically, the total data set is divided into multiple small batches, each small batch contains several groups of data, and only one batch of data is updated at a time. At this point, the forward prediction network is built.

[0069] In one embodiment, the training of the reverse design network parameters includes: calculating a mean square error as a reverse loss value based on an input sequence of the reverse design network and an output sequence of a forward prediction network; and updating the parameters of the reverse design network based on the reverse loss value.

[0070] The mean square error (MSE) is also used as the loss function. The mean square error calculation at this time is the distance between the output and the target after the reverse design network and the forward prediction network, that is, in Represents the target value, that is, the characteristic sound absorption coefficient value, α i represents the predicted value of the neural network at this time, and n represents the number of neurons. Function f(x) and function g(x) represent the decoder and encoder respectively.

[0071] At the same time, the Adam optimization algorithm is used in the training process, the learning rate is set to 0.001, and combined with the small batch gradient descent technology, several groups of data are processed in each batch, so as to achieve accurate update of network parameters and avoid oscillation. During the training process of the entire model, the only data set used is the label data of the forward prediction network, that is, the discretized sound absorption coefficient curve data set, so the whole process can be regarded as unsupervised learning. The data set of the structural parameters of the conical embedded tubular Helmholtz resonator is only used when training the forward prediction network. Although there is no involvement of structural parameter data in the reverse design network, since the trained forward prediction network is used in this process, the structural parameter data has actually played a role in the training of the reverse design network.

[0072] The following is an explanation with specific examples:

[0073] A conical Helmholtz resonator with an inlet diameter and an outlet diameter between 1 and 5 mm and a cone neck length of 1 mm to 50 mm is used as an example to establish a data set. The schematic diagram of the conical embedded tube Helmholtz resonator structure is shown in Figure 4 and Figure 5As shown. Through the analysis of the theoretical model, the three key structural parameters of the conical Helmholtz resonator are obtained: the inlet diameter, outlet diameter and cone neck length of the conical embedded tube, where the inlet diameter and outlet diameter are between 1-5mm, and the cone neck length is from 1mm to 50mm. These structural parameters will be used as input data of the neural network data set for model training. At the output end of the neural network, the sound absorption coefficient curve will be used as label data to correspond to the structural parameters of the conical embedded tube Helmholtz resonator. In order to ensure the smooth progress of the neural network training, each sound absorption coefficient curve needs to be discretized when the data set is established.

[0074] Random sampling is performed in the preset data space to generate a set of 50,000 groups of data, forming a 50,000×3 input matrix. Subsequently, based on the analytical solution, the sound absorption coefficient curve is discretized, and data is collected every 1Hz in the frequency range of 100-300Hz to obtain 200 characteristic frequencies, and a 50,000×200 label matrix is ​​constructed to ensure the consistency of each set of input data with the corresponding label data. After the data set is constructed, the data set is divided. According to the ratio of 80%, 10%, and 10%, the data set is divided into a training set, a test set, and a validation set. That is, the training set contains 40,000 groups of data, and the test set and validation set each contain 5,000 groups of data.

[0075] The purpose of the forward prediction network is to predict the sound absorption coefficient of the conical embedded tube Helmholtz resonator. It adopts a standard fully connected neural network architecture. The input layer receives three key structural parameters of the resonator, namely the inlet diameter, outlet diameter and tube length of the conical tube. Since a single conical embedded tube Helmholtz resonator is defined by these three parameters, the input layer of the neural network is equipped with three neuron units, and the output layer contains 200 neuron units, corresponding to the sound absorption coefficient of each frequency point in the frequency range of 100Hz to 300Hz, increasing by 1Hz. This design enables the network to predict the sound absorption coefficient under given structural parameters by learning the relationship between structural parameters and sound absorption characteristics.

[0076] The fitting relationship of the established forward prediction network is as follows: α n =f n (d a2 ,d a1 ,l a), where n = 1, 2, 3, ..., 200. There are four hidden layers between the input layer and the output layer. These four hidden layers are equipped with 128, 256, 512, and 512 neurons respectively. The Relu activation function is used in the hidden layer to enhance the nonlinear expression ability of the model. The expression of the Relu function is as follows: f(x) = max(0, x). In the initialization stage of training, the weights and biases of all network layers are randomly set through truncated normal distribution. In the model training process, the mean square error (MSE) is selected as the loss function to guide the optimization process of network parameters. The definition of MSE is as follows in Represents the target value, that is, the characteristic sound absorption coefficient value, α i Represents the predicted value of the neural network at this time. At the same time, the Adam optimization algorithm is used to dynamically adjust the gradient of the model to ensure the convergence speed of the neural network. The update rule for each time is: where θ represents all learning parameters, and γ is the learning rate, and each learning parameter update has At the same time, the learning rate is set to 0.001.

[0077] During the training process, a small batch gradient descent strategy was adopted. Specifically, the total data set of 50,000 groups was divided into multiple small batches, each small batch contained 32 groups of data, and only one batch of data was updated at a time. At this point, the forward prediction network was built.

[0078] After that, the previously established data set is integrated into the program to start the neural network training process. This process is performed using the Tensorflow framework. After 100 iterations, the forward prediction network successfully reaches a convergence state, and the final mean square error value is less than 0.0001, indicating that the network has excellent performance. For input data and label data, there are After 100 iterations, the forward prediction network successfully reached a convergence state.

[0079] Then, we use actual cases to verify. After normalizing the actual case data, we input it into the network and run it. Before drawing the sound absorption coefficient curve, we perform a denormalization process. The formula is y=y(y max -y min )+y min Finally, the sound absorption coefficient curve calculated by the network is obtained, and then the analytical solution is compared with the predicted result, such as Figure 6 As shown, for the four curves, it is found that the two are highly consistent, verifying that the forward prediction network of a single conical embedded tube Helmholtz resonator has been successfully constructed.

[0080] The purpose of the reverse design network is to quickly calculate the corresponding structural parameters by inputting the target sound absorption coefficient curve. The architecture is opposite to the forward prediction network. The input layer contains 200 neurons, corresponding to the sound absorption coefficients of each frequency point in the frequency range of 100Hz-300Hz. The output layer contains 3 neurons, representing the inlet diameter, outlet diameter and tube length of the conical tube. In the design of the reverse design network, the order of neurons in the input layer and the output layer is consistent with the corresponding relationship in the forward prediction network. For an independent reverse design network, there is P n =g n (α1,α2,α3,......,α 200 ), and the function g(x) is the inverse function of the function f(x).

[0081] The reverse design network and the forward prediction network need to work together during the retraining process, and the characteristics of the forward prediction network may increase the complexity of the reverse design network training. To ensure the effectiveness of the training, the present invention optimizes the training effect by enhancing the depth and width of the network. The reverse design network contains three hidden layers, and the number of neurons in each hidden layer is a power of 2, specifically 512 neurons in the first layer, 128 neurons in the second layer, and 64 neurons in the third layer, and the input layer contains 200 neurons. The network weights are initialized using a truncated normal distribution, and the standard deviation is set to 0.1.

[0082] After completing the construction of the reverse design network, load the trained forward prediction network model and place it after the reverse design network. Since the front end of the forward prediction network has been trained to obtain a 3x128 weight matrix and a 1x128 bias matrix, where 3 represents the number of neurons in the output layer of the reverse design network and 128 represents the number of neurons in the first layer of the forward prediction network, the forward prediction network can be directly connected in series with the reverse design network to form an autoencoder model. In this configuration, the output layer of the reverse design network directly corresponds to the input layer of the forward prediction network. At the same time, the output result of this layer is the required conical embedded tube Helmholtz resonator structure parameters. When training the entire deep learning model, the first half of the reverse design network needs to be trained separately, so the second half of the forward prediction network is fixed so that it is not trained. The mean square error (MSE) is also used as the loss function. At this time, the mean square difference calculation is the distance between the output and the target after the reverse design network and the forward prediction network, that is:

[0083]

[0084] in, Represents the target value, that is, the characteristic sound absorption coefficient value, α i Represents the predicted value of the neural network at this time. Function f(x) and function g(x) represent the decoder and encoder respectively.

[0085] At the same time, the Adam optimization algorithm is used in the training process, the learning rate is set to 0.001, and the small batch gradient descent technology is combined to process 32 sets of data in each batch, thereby achieving accurate updating of network parameters and avoiding oscillations.

[0086] After the parameters are configured successfully, the network training begins. After 200 iterations, the neural network model converges and eventually reaches a mean square error of less than 0.0002. The model performance is then verified using actual data. First, a specific sound absorption coefficient curve is selected from the test set and normalized. After that, the processed data is input into the neural network model for in-depth analysis. After the model is decoded and output, a denormalization operation is performed to restore the data to its original scale and compare it with the original input curve. Figure 7 As shown, the two are highly consistent. So far, the neural network model established for the structural design of a single conical tube Helmholtz resonator has completed the construction and training stages.

[0087] The trained neural network model is used to design the structural parameters of the conical embedded tube Helmholtz resonator. First, the target sound absorption coefficient curve is generated by the quadratic spline interpolation method, and then the generated curve is discretized into 200 data points according to the frequency, and these data points are serialized and saved to form a target input data set with predetermined sound absorption characteristics. The neural network model is loaded, and the preprocessed input data is input into the network to obtain the structural parameters.

[0088] After completing the preliminary preparations, the prediction network is used to design the structural parameters of the conical Helmholtz resonator that meets specific sound absorption performance requirements. First, the sound absorption coefficient of the target sound absorption coefficient curve at 200 Hz must exceed 0.95. Subsequently, according to the shape of the predetermined target sound absorption curve, a series of key frequency points are selected, and a sound absorption curve that meets the design standards is constructed through quadratic spline interpolation technology. Next, the curve is discretized, starting from 100 Hz, and a sound absorption coefficient value is collected every 1 Hz until a set of 200 data points is formed. After these data points are normalized, they are input into the neural network to generate the corresponding structural parameters, that is, the inlet diameter of the conical tube is 4.98 mm, the outlet diameter is 1.10 mm, and the length is 1.26 mm.

[0089] The structural parameters of the obtained conical embedded tube Helmholtz resonator were verified by finite element simulation method and the model was constructed using COMSOL6.0 software. The analytical model is as follows: Figure 8 As shown. Fig. 9 It can be seen that the sound absorption coefficient curve obtained by simulation is highly consistent with the target curve, which verifies the correctness of the deep learning model of the present invention.

[0090] The following is a description of a device for generating structural parameters of a tapered embedded tubular Helmholtz resonator provided by the present invention. The device for generating structural parameters of a tapered embedded tubular Helmholtz resonator described below and the method for generating structural parameters of a tapered embedded tubular Helmholtz resonator described above can be referred to each other.

[0091] Fig.10 is a schematic diagram of the structure of the device for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the present invention, such as Fig.10 As shown, the structure parameter generation device of the conical embedded tube Helmholtz resonator includes: a data input module 1001, a first training module 1002 and a second training module 1003. The data input module 1001 is used to obtain the sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; the first training module 1002 is used to train the constructed forward prediction network with the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label to obtain the trained forward prediction network; the second training module 1003 is used to connect the constructed reverse design network and the trained forward prediction network in series to obtain a series network, use the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fix the forward prediction network parameters to train the reverse design network parameters; the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for structure parameter generation.

[0092] In one embodiment, the device also includes a processing module for obtaining a required sound absorption coefficient curve for a conical embedded tube Helmholtz resonator whose structural parameters are to be determined, and extracting a sequence based on the required sound absorption coefficient curve; inputting the sequence into a structural parameter generation model to obtain the structural parameters to be determined; wherein the structural parameter generation model is a trained inverse design network.

[0093] The device embodiments provided in the embodiments of the present invention are for implementing the above-mentioned method embodiments. For specific processes and detailed contents, please refer to the above-mentioned method embodiments, which will not be repeated here.

[0094] The implementation principle and technical effects of the conical embedded tubular Helmholtz resonator structural parameter generating device provided in the embodiment of the present invention are the same as those of the aforementioned conical embedded tubular Helmholtz resonator structural parameter generating method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the conical embedded tubular Helmholtz resonator structural parameter generating device, reference may be made to the corresponding contents in the aforementioned conical embedded tubular Helmholtz resonator structural parameter generating method embodiment.

[0095] Fig.11 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig.11As shown, the electronic device may include: a processor (processor) 1101, a communication interface (Communications Interface) 1102, a memory (memory) 1103 and a communication bus 1104, wherein the processor 1101, the communication interface 1102, and the memory 1103 communicate with each other through the communication bus 1104. The processor 1101 can call the logic instructions in the memory 1103 to execute the method for generating structural parameters of a conical embedded tube Helmholtz resonator, which includes: obtaining sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; using the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label, training the constructed forward prediction network to obtain the trained forward prediction network; connecting the constructed reverse design network and the trained forward prediction network in series to obtain a series network, using the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fixing the forward prediction network parameters to train the reverse design network parameters; wherein the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for structural parameter generation.

[0096] In addition, the logic instructions in the above-mentioned memory 1103 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0097] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the above methods, the method comprising: obtaining sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; using the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label, training the constructed forward prediction network to obtain the trained forward prediction network; connecting the constructed reverse design network and the trained forward prediction network in series to obtain a series network, using the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fixing the forward prediction network parameters to train the reverse design network parameters; wherein the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for structural parameter generation.

[0098] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the method for generating structural parameters of a conical embedded tube Helmholtz resonator provided by the above-mentioned methods, the method comprising: obtaining sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; using the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label, training the constructed forward prediction network to obtain the trained forward prediction network; connecting the constructed reverse design network and the trained forward prediction network in series to obtain a series network, using the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fixing the forward prediction network parameters to train the reverse design network parameters; wherein the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used for structural parameter generation.

[0099] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0100] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating structural parameters of a conical embedded tube Helmholtz resonator, characterized in that: include: Obtaining sound absorption coefficient curves of multiple sample conical embedded tubes with known structural parameters; Taking the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label, the constructed forward prediction network is trained to obtain a trained forward prediction network; The constructed reverse design network and the trained forward prediction network are connected in series to obtain a series network, the sequence extracted from the sound absorption coefficient curve is used as both the output label and input of the series network, and the parameters of the forward prediction network are fixed to train the parameters of the reverse design network; Among them, the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used to generate structural parameters.

2. According to the method for generating structural parameters of a conical embedded tube Helmholtz resonator according to claim 1, the structural parameters include the inlet diameter, outlet diameter and cone neck length of the conical embedded tube, and the structural types of the forward prediction network and the reverse design network include fully connected neural networks.

3. The method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to claim 1, characterized in that: After obtaining the sound absorption coefficient curves of a plurality of sample conical embedded tubes with known structural parameters, the method further includes: The sound absorption coefficient of each frequency point is extracted according to the sound absorption coefficient curve to obtain the sequence.

4. The method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to claim 1, after training the reverse design network parameters, further comprising: For a conical embedded tube Helmholtz resonator whose structural parameters are to be determined, a required sound absorption coefficient curve is obtained, and a sequence is extracted according to the required sound absorption coefficient curve; Inputting the sequence into a structural parameter generation model to obtain structural parameters to be determined; Wherein, the structural parameter generation model is a trained inverse design network.

5. The method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to claim 4, characterized in that: The obtaining of the required sound absorption coefficient curve comprises: Obtain the required value of the sound absorption coefficient at the target frequency point; Based on the required value of the sound absorption coefficient at the target frequency point and in combination with the shape of the required sound absorption curve, the sound absorption coefficients corresponding to multiple key frequency points are determined; According to the sound absorption coefficients corresponding to the target frequency point and multiple key frequency points, the required sound absorption coefficient curve is obtained by quadratic spline interpolation fitting.

6. The method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to claim 1, characterized in that: The training of the constructed forward prediction network includes: According to the output sequence of the forward prediction network and the sequence extracted from the sound absorption coefficient curve, a mean square error is calculated as a forward loss value, and the parameters of the forward prediction network are updated according to the forward loss value.

7. The method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to claim 1, characterized in that: The training of the reverse design network parameters includes: According to the input sequence of the reverse design network and the output sequence of the forward prediction network, the mean square error is calculated as the reverse loss value; The parameters of the reverse design network are updated according to the reverse loss value.

8. A device for generating structural parameters of a conical embedded tube Helmholtz resonator, characterized in that: include: A data input module is used to obtain sound absorption coefficient curves of a plurality of sample conical embedded tubes with known structural parameters; A first training module is used to train the constructed forward prediction network using the structural parameters as input and the corresponding sequence extracted from the sound absorption coefficient curve as a label to obtain a trained forward prediction network; The second training module is used to connect the constructed reverse design network and the trained forward prediction network in series to obtain a series network, using the sequence extracted from the sound absorption coefficient curve as both the output label and input of the series network, and fixing the parameters of the forward prediction network to train the reverse design network parameters; Among them, the output of the reverse design network is the input of the forward prediction network, and the reverse design network after training is used to generate structural parameters.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating structural parameters of a tapered embedded tube Helmholtz resonator according to any one of claims 1 to 7 is implemented.

Images