A method for testing sound absorption performance of sound absorption material, electronic equipment and storage medium

Abstract

The application discloses the technical field of material testing, and discloses a sound absorption performance testing method for sound absorption material, electronic equipment and a storage medium, which comprises the following steps: after a recursive response matrix is constructed, the overall acoustic response signal is path decoupled to obtain a plurality of separated propagation path acoustic signals; a topological feature extraction network with a residual cross-layer error correction module is constructed to obtain an error correction topological feature map; a path coupling correlation matrix is constructed based on the error correction topological feature map and the recursive response matrix, the separated propagation path acoustic signals are feature mapped by using the matrix, and the acoustic contribution features of each layer of the sound absorption material to be tested are obtained; and the acoustic contribution features of the layers are input into a pre-trained convolution network prediction model to output the sound absorption coefficient prediction value of the sound absorption material to be tested; and the application is favorable for improving the reliability of sound absorption performance evaluation and structure optimization.

Description

Technical Field

[0001] This invention relates to the field of material testing technology, and more specifically, to a method, electronic device, and storage medium for testing the sound absorption performance of sound-absorbing materials. Background Technology

[0002] With the increasing demand for noise reduction performance in industrial and civilian sectors, composite sound-absorbing materials, due to their wide bandwidth and high-efficiency acoustic characteristics, are gradually becoming an important development direction in the field of noise control. Currently, the commonly used sound absorption performance testing methods in the industry mainly include the impedance tube method and the reverberation chamber method. Among them, the impedance tube method derives the sound absorption coefficient of the material by measuring the sound pressure ratio of the standing wave inside the tube to the incident wave. Although it is convenient to measure, it can only provide an overall macroscopic sound absorption index and cannot deeply reveal the complex sound wave propagation process inside the multilayer structure of composite materials and the coupling sound absorption characteristics between the layers of materials.

[0003] Meanwhile, since composite sound-absorbing materials are typically composed of layers of materials with different acoustic properties, the differences in local structure of each interface and the inhomogeneity of the manufacturing process lead to a non-uniform distribution of the curvature of the internal interfaces, further exacerbating the complexity and uncertainty of the internal sound wave propagation path. The aforementioned traditional methods lack effective means for identifying and finely correcting local non-uniform features, making the prediction and evaluation of sound absorption performance susceptible to interference from differences in the internal structure of the material and local irregularities, thus reducing the accuracy and reliability of performance predictions. Summary of the Invention

[0004] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method for testing the sound absorption performance of sound-absorbing materials, an electronic device, and a storage medium.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for testing the sound absorption performance of sound-absorbing materials, comprising:

[0007] The overall acoustic response signal of the sound-absorbing material under test is collected. Based on the path recursive response characteristics caused by the cumulative propagation of multi-interface reflections, a recursive response matrix is ​​constructed, and the overall acoustic response signal is decoupled to obtain multiple separate propagation path acoustic signals.

[0008] A topology feature extraction network with a residual cross-layer error correction module is constructed. The topology feature extraction network determines the error correction scale based on the local non-uniformity of the interface curvature in the topology of the sound-absorbing material under test, thereby obtaining an error-corrected topology feature map.

[0009] A path coupling correlation matrix is ​​constructed based on the error correction topology feature map and the recursive response matrix. This matrix is ​​then used to perform feature mapping on the acoustic signals of the separated propagation paths to obtain the acoustic contribution characteristics of each layer of the sound-absorbing material under test.

[0010] The acoustic contribution features of each layer are input into a pre-trained convolutional network prediction model, which outputs the predicted sound absorption coefficient of the sound-absorbing material to be tested.

[0011] An electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the method described in any of the preceding claims.

[0012] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the method described in any of the preceding claims.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] This invention constructs a path recursive response matrix based on the cumulative propagation of multi-interface reflections to perform path-level decoupling processing on the overall acoustic response signal. This separates the propagation path signals superimposed inside the composite sound-absorbing material into multiple independent propagation path acoustic signals, effectively overcoming the problem that existing overall testing methods such as impedance tube methods are difficult to distinguish the contributions of multi-path propagation, thereby improving the analytical capability of the sound wave propagation characteristics inside the composite structure.

[0015] This invention introduces a topology feature extraction network with a residual cross-layer error correction module and determines the error correction scale based on the local non-uniformity of the interface curvature. It performs differentiated correction processing on the local irregular features in the material topology, so that the topology feature extraction results can reflect the actual structural state of the material interface and reduce the interference of structural non-uniformity on acoustic feature analysis.

[0016] This invention constructs a path coupling correlation matrix by fusing error-correcting topological feature maps and recursive response matrices, and maps the acoustic signals of the separated propagation paths based on this matrix to obtain the acoustic contribution characteristics of each layer. Furthermore, it combines a convolutional network prediction model to output the predicted value of the sound absorption coefficient, thereby quantifying the sound absorption contribution of each layer of the composite sound-absorbing material and providing a basis for material structure design and performance optimization. Attached Figure Description

[0017] Figure 1 A flowchart of a method for testing the sound absorption performance of sound-absorbing materials provided by the present invention;

[0018] Figure 2 A schematic diagram of the structure of an electronic device provided by the present invention;

[0019] Figure 3 This is a schematic diagram of the structure of a computer-readable storage medium provided by the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1

[0022] Please see Figure 1 As shown in the figure, this embodiment discloses a method for testing the sound absorption performance of sound-absorbing materials, the method comprising:

[0023] S101: Collect the overall acoustic response signal of the sound-absorbing material under test. Based on the path recursive response characteristics caused by the cumulative propagation of multi-interface reflections, construct the recursive response matrix and then perform path decoupling on the overall acoustic response signal to obtain multiple separate propagation path acoustic signals.

[0024] S102: Construct a topology feature extraction network with a residual cross-layer error correction module. The topology feature extraction network determines the error correction scale based on the local non-uniformity of the interface curvature in the topology of the sound-absorbing material under test, and obtains the error correction topology feature map.

[0025] S103: Construct a path coupling correlation matrix based on the error correction topology feature map and the recursive response matrix, and use this matrix to perform feature mapping on the acoustic signals of the separated propagation paths to obtain the acoustic contribution characteristics of each layer of the sound-absorbing material under test.

[0026] S104: Input the acoustic contribution features of each layer mentioned above into the pre-trained convolutional network prediction model, and output the predicted sound absorption coefficient of the sound-absorbing material to be tested.

[0027] Specifically, the overall acoustic response signal of the sound-absorbing material under test is first acquired using acoustic measurement equipment. It should be understood that the sound-absorbing material under test is composed of at least two layers of materials with different acoustic properties, such as a composite structure consisting of a foam layer, a sound-insulating cotton layer, and a damping plate layer. The method for acquiring the overall acoustic response signal specifically includes: emitting a preset sound wave signal using a loudspeaker on one side of the material under test; after the sound wave penetrates the material structure, it is simultaneously received on the other side of the material using a high-sensitivity microphone, thereby obtaining the overall acoustic response signal of the sound-absorbing material under test. The overall acoustic response signal includes the comprehensive response information of multiple propagation paths within the composite sound-absorbing structure.

[0028] Specifically, the construction of the recursive response matrix based on the path recursive response characteristics caused by multi-interface reflection cumulative propagation includes:

[0029] Collect multi-order reflection propagation signals from the interface of the sound-absorbing material under test, extract multi-order propagation delay difference features of the propagation path, and obtain a multi-order propagation delay feature set;

[0030] Specifically, to obtain the multi-order reflection propagation signals at each interface of the sound-absorbing material under test, a pulse acoustic detection technique with high spatiotemporal resolution is employed. Using a narrow pulse signal as the excitation source, the echo sequences reflected from each layer of the sound-absorbing material under test are time-domain gated and identified to obtain reflection propagation signals of different orders. The difference in the time range of the pulse response can effectively distinguish the multi-order reflection signals generated by sound waves at discontinuous interfaces within the layered structure, and the reflection signal response at each interface corresponds to a specific propagation order.

[0031] Furthermore, the extraction of the multi-order propagation delay difference features specifically includes: identifying the response time of each order of reflected propagation signal and determining the arrival time of each order signal response. (Unit: seconds) (Assigning propagation order numbers), and calculating multi-order propagation delay difference characteristics based on the arrival time differences between adjacent orders. The specific calculation formula is as follows: ;in, For the first Rank and number The delay difference between propagation paths, in seconds.

[0032] The multi-order propagation delay feature set is specifically represented as follows: in, This represents the maximum detectable propagation order at the interface of the material to be tested.

[0033] Based on the frequency concentration characteristics of the multi-order propagation delay feature set, the propagation mode association relationship between paths is determined, and the path mode association feature is obtained.

[0034] Specifically, to determine the correlation between propagation modes between paths, spectral analysis is performed on the propagation signal at each order to obtain the spectral energy distribution characteristic function corresponding to each propagation path. ,in, For the first The spectral energy distribution function of each propagation path, For frequency variables.

[0035] Furthermore, the correlation between propagation modes is determined by calculating the similarity between the spectral energy distributions of different propagation paths. The specific calculation formula is as follows: in, For the first The path and the first Spectral similarity between paths This represents the number of frequency sampling points.

[0036] For example, the threshold for spectral similarity is set to 0.85, when the condition is met... At that time, the first Path and the Paths are determined to have a propagation pattern association, otherwise no association is found. The path pattern association features obtained through the above method are used to describe the pattern relationship information between propagation paths.

[0037] Clustering feature mapping is performed on the multi-order propagation delay feature set using path pattern association features, and a recursive response matrix is ​​formed based on the mapped clustering features;

[0038] Specifically, a path similarity matrix is ​​constructed based on path pattern association features. The matrix elements are defined as follows:

[0039] Further define the path similarity matrix The corresponding degree matrix Its diagonal elements are defined as: in, The total number of paths, These are the row numbers of the matrix.

[0040] Furthermore, construct the Laplace matrix. Specifically defined as: ;

[0041] Laplace matrix Perform eigenvalue decomposition and select the eigenvectors corresponding to the smallest eigenvalues ​​to form the clustering feature matrix. Clustering algorithms (such as K-means clustering) are used to analyze the cluster feature matrix. Clustering is performed to map the feature vectors in the multi-order propagation delay feature set to different clusters, thereby obtaining the path category classification results.

[0042] A recursive response matrix is ​​formed based on the above clustering mapping results. Specifically, it is expressed as: in, Indicates the first Delayed feature submatrices for each category, This represents the total number of path categories after clustering.

[0043] It should be noted that the recursive response matrix It can reveal the recursive response relationship between multiple reflection paths at different interfaces inside the material under test, providing a basis for subsequent path decoupling.

[0044] Specifically, the path decoupling of the overall acoustic response signal to obtain multiple separate propagation path acoustic signals includes:

[0045] Based on the sparse distribution characteristics of the recursive response matrix, sparse coding of the propagation path signal is performed to obtain the initial propagation path signal set;

[0046] Specifically, the overall acoustic response signal is represented as an observation signal vector. ,in Representing the time variable; the recursive response matrix Convert to path feature dictionary matrix ,in , To observe the dimension of the signal, This represents the path feature dimension.

[0047] It should be understood that the sparse distribution characteristic of the matrix refers to the matrix Only a portion of the elements are non-zero, which correspond to the actual propagation path features. The remaining elements are zero or close to zero, thus enabling the path features to have a separable representation.

[0048] Specifically, based on the aforementioned path feature dictionary matrix Construct a sparse representation model: in, Let be the sparse coefficient vector to be solved, representing the contribution coefficient of each propagation path to the overall acoustic response signal.

[0049] Furthermore, the sparse coefficient vector is obtained by solving the following optimization problem: st in, Representing vectors The number of non-zero elements in the middle.

[0050] Understandably, the orthogonal matching pursuit method is used to iteratively solve the above optimization problem. In each iteration, the dictionary column vector with the largest inner product with the residual signal is selected to gradually approximate the observed signal. until the residuals meet the set conditions. Thus, a set of sparse coefficient vectors is obtained. ,in This indicates the number of identified propagation paths, with specified conditions. The specific value is determined based on the real-time signal-to-noise ratio (SNR) of the test environment of the sound-absorbing material under test.

[0051] Based on the sparse coefficient vector and dictionary matrix The correspondence is used to reconstruct the initial propagation path signal set: in, Indicates the first The initial acoustic signal corresponding to each propagation path.

[0052] Based on the spectral energy difference characteristics of the path signals in the initial propagation path signal set, spectral domain adaptive matrix decomposition is performed to obtain the propagation path decoupling characteristics;

[0053] Specifically, for the initial propagation path signal set Each path signal Perform a frequency domain transformation to obtain its spectral representation. ,in Represents a frequency variable.

[0054] Furthermore, construct the path spectrum matrix. ,in: ;

[0055] It should be noted that the spectral energy difference characteristics are characterized by the differences in energy distribution of the spectrum of each path in different frequency ranges, and these differences are used to distinguish the propagation characteristics of different paths.

[0056] Specifically, for the spectrum matrix Perform nonnegative matrix decomposition to obtain the basis matrix. With coefficient matrix : in, Represents the fundamental spectral mode matrix. Represents the path contribution coefficient matrix. For decomposition dimensions.

[0057] Understandably, the matrix Each column corresponds to the weight distribution of a propagation path across various fundamental spectral modes. This weight distribution is the propagation path decoupling feature: ;in, Indicates the first The decoupling feature vector of each path.

[0058] Based on the propagation path decoupling characteristics, the overall acoustic response signal is decoupled to obtain multiple separate propagation path acoustic signals;

[0059] Specifically, based on the propagation path decoupling feature set Combined with the basic spectral mode matrix Reconstruct the spectral signal for each path: ;in, Indicates the first Reconstructed spectral signal of propagation path.

[0060] Furthermore, an inverse frequency domain transform is performed on the reconstructed spectral signal to obtain the time domain signal: in, This indicates the inverse Fourier transform operation.

[0061] Through the above processing, multiple sets of acoustic signals from separate propagation paths are obtained: ;

[0062] It should be understood that sets Each element corresponds to an acoustic response signal of an independent propagation path within the composite sound-absorbing structure, thus achieving path-level separation of the overall acoustic response signal.

[0063] Specifically, the construction process of the topology feature extraction network with residual cross-layer error correction module includes:

[0064] The interface curvature of the topological structure of the sound-absorbing material under test is refined by sampling, and the local non-uniform distribution features of the interface curvature are extracted to form a local non-uniform feature map.

[0065] Specifically, firstly, a high-density spatial sampling is performed along the interface of the sound-absorbing material under test using a curvature sensing device to obtain the set of interface positions and corresponding curvature values. ,in, Sampling location, For position The local interface curvature value.

[0066] It is understandable that the local curvature region of an interface refers to the area within the interface structure where curvature changes are dense or curvature gradient changes are evident.

[0067] Furthermore, the sampled data is processed by interpolation fitting to obtain continuous local curvature distribution features, generating a local non-uniform feature map of two-dimensional spatial distribution. ,in, For interface position The local curvature value.

[0068] Based on the amplitude characteristics of regional curvature differences in the local non-uniform feature map, the local feature error correction scale of the residual cross-layer error correction module is determined, and the scale error correction mapping matrix is ​​obtained.

[0069] Specifically, based on the local non-uniform feature map obtained above For each local region, perform regional curvature difference analysis, calculate the amplitude characteristics of regional curvature difference, and obtain the characteristic values ​​of regional curvature difference. Among them, the regional curvature difference characteristic value Characterizing the first The degree of difference in curvature values ​​within a local region.

[0070] Preferably, the method for calculating the curvature difference characteristic value is defined as follows: ;in, For the first Number of sampling points in a local area For the first The curvature value of each sampling point For the first The mean of curvature values ​​within each region.

[0071] Furthermore, based on the regional curvature difference characteristic values The size determines the error correction scale factor for the corresponding local region. The error correction scale factor is used to guide the local error correction intensity of the residual cross-layer error correction module in the network.

[0072] Specifically, the method for determining the error correction scale factor is defined as follows: ;in, This represents the total number of local regions.

[0073] The scale correction mapping matrix is ​​obtained through the above process. : ;

[0074] It should be noted that the scale correction mapping matrix elements in This is used to clarify the mapping relationship between the local differences in interface curvature and the specific error correction scale of the residual cross-layer error correction module, so as to achieve compensation for the differentiated features of different regions.

[0075] The process of determining the scale-correction mapping matrix includes:

[0076] The local oscillation amplitude of the interface curvature in different spatial regions of the topological structure of the sound-absorbing material under test is collected to obtain the local oscillation characteristics of the interface curvature.

[0077] Specifically, the local oscillation feature of the interface curvature is used to characterize the local fluctuation of the interface curvature of the sound-absorbing material under test within a spatial region. First, a high-density spatial sampling method is used to collect curvature values ​​corresponding to multiple sampling points within each spatial region, and then the local oscillation amplitude characteristic value is calculated for each spatial region. .

[0078] For example, the characteristic value of local oscillation amplitude It can be calculated using the following formula: ;in, For the first The number of sampling points within a spatial region For the first in this region The curvature value of each sampling point This represents the average curvature value within that region.

[0079] It is understandable that the larger the characteristic value of the local oscillation amplitude, the more drastic the change in the interface curvature in the local area.

[0080] Spatial scale fusion mapping is performed based on the local oscillation characteristics of interface curvature, and the scale error correction mapping matrix is ​​determined based on the scale sensitivity of the fused region.

[0081] Specifically, based on the local oscillation amplitude characteristic values ​​obtained above First, spatial scale fusion mapping is performed, which involves fusing calculations based on the oscillation amplitude characteristics of spatially adjacent regions to obtain the fused regional scale sensitivity feature values. .

[0082] The specific fusion calculation formula is as follows: ;in, For the first A set of spatially adjacent regions of a region. This represents the number of adjacent regions in the set.

[0083] Furthermore, based on the fused regional scale sensitivity feature values... The relative size determines the scale correction mapping matrix. The final values ​​of each element in the matrix enable the elements in the scale correction mapping matrix to reflect the local sensitivity of the differences in the curvature region of the interface.

[0084] Specifically, the scale-correction mapping matrix The final determination method is as follows: ;in, The total number of spatial regions. For matrix The Each element represents the error correction scale for the corresponding region.

[0085] Cross-layer feature weight allocation is performed based on the scale error correction mapping matrix, and a topological feature extraction network with residual cross-layer error correction module is obtained through cross-layer feature iterative fusion.

[0086] Specifically, through the scale correction mapping matrix The scaling factor in the topology feature extraction network is used to assign differential weights to the cross-layer features of each local region.

[0087] Specifically, the cross-layer feature weight allocation process is represented as follows: ;in, To extract basic cross-layer feature weights in the network for topological feature extraction, The cross-layer feature weights are adjusted by the error correction scaling factor.

[0088] Preferably, the cross-layer features after weight allocation are fused through a multi-layer iterative fusion mechanism. That is, cross-layer feature transfer and fusion are achieved between adjacent feature layers by weighted summation. The specific implementation of the fusion process is as follows: ;in, For the first In the layer topological feature network, the first The feature vector of each node For nodes The set of adjacent nodes, To balance the fusion ratio between its own features and adjacent features, the value range was determined based on experimental data. For example, it can be set as .

[0089] Through the above cross-layer feature iterative fusion processing, a topology feature extraction network containing a residual cross-layer error correction module is obtained. The error correction scale is then determined based on the local interface curvature non-uniformity characteristics of the topology of the sound-absorbing material under test, and an accurate error-corrected topology feature map is output for subsequent path coupling correlation matrix construction.

[0090] Specifically, the process of constructing the path coupling association matrix includes:

[0091] Multi-channel feature convolution is performed on the error-corrected topological feature map and the recursive response matrix to form an initial coupled and correlated feature map;

[0092] Specifically, the error-corrected topological feature map and the recursive response matrix are used as multi-channel input features, respectively, and are fused through two-dimensional convolution operations to form an initial coupled and correlated feature map.

[0093] For example, the fusion process can be implemented using multi-channel two-dimensional convolution operations, where the input error-correcting topological feature map is denoted as the feature matrix. The recursive response matrix is ​​denoted as the characteristic matrix. Then the initial coupling-related feature mapping The calculation method is as follows: ;in, and These represent the number of channels in the error-correcting topological feature map and the recursive response matrix, respectively. For the error correction topological feature map, the first Channel characteristics, The first recursive response matrix Individual channel characteristics; and These are the corresponding convolution kernels. This is the bias term for the convolution operation. This represents a two-dimensional convolution operation. This represents a non-linear activation function, such as the ReLU function.

[0094] It should be noted that the multi-channel convolutional fusion method described above effectively combines topological spatial information with recursive response path delay information to obtain an initial coupling correlation feature map that can represent the spatial coupling relationship between paths. .

[0095] Based on the spatial energy distribution concentration characteristics of the path signals in the initial coupling correlation feature map, the local energy patterns of path interactions at the material interface are extracted.

[0096] The extraction process of the local energy modes of the material interface path interaction includes:

[0097] Local spatial clustering analysis is performed on the initial coupling correlation feature map to extract the energy concentration region features of the interface reflection path and obtain the initial energy pattern features;

[0098] Specifically, local spatial clustering analysis is used to map the initial coupled association features. Processing is performed to extract the spatial region where the reflected signal energy is concentrated in the material interface path interaction, thereby obtaining the initial energy mode characteristics. .

[0099] For example, the DBSCAN clustering algorithm can be used to analyze the feature mapping matrix. Cluster analysis was performed on the energy characteristics of local regions to calculate the energy density value of each local region. This allows for the identification of high-energy-density regions and the formation of initial energy pattern characteristics. ;in, The first characteristic corresponding to the initial energy mode Cluster regions, For position The energy density value at that location, A preset energy density threshold is used to determine the energy concentration area.

[0100] Preferably, the energy density threshold This can be obtained through statistical analysis of historical experimental data, such as by taking the feature matrix. The top 10 percentile of the internal energy density value is used as the threshold.

[0101] By utilizing the spatial distribution stability characteristics of the initial energy pattern, regional scale features are fused to obtain the local energy pattern of material interface path interaction.

[0102] Furthermore, based on the initial energy pattern characteristics obtained above... The spatial distribution stability characteristics are used for regional-scale fusion processing. The specific fusion method is as follows:

[0103] First, calculate the spatial stability eigenvalues ​​for each initial energy mode region. , defined as the reciprocal of the spatial variance of the energy density values ​​within that region, is expressed as: ;in, For the region The spatial variance of the internal energy density values ​​reflects the stability of the energy distribution in that region.

[0104] Secondly, utilizing stability eigenvalues By weighted fusion of regional energy patterns at different scales, the final local energy pattern of material interface path interaction is obtained. : ;in, The number of energy concentration regions. The fused local energy mode feature matrix reflects the locally stable energy concentration mode in the material interface path interaction, serving as a key basis for constructing the path coupling correlation matrix.

[0105] Multi-scale attention feature fusion processing is performed using the local energy patterns of path interactions at the material interface to construct a path coupling correlation matrix;

[0106] Specifically, based on the aforementioned local energy patterns of material interface path interactions. Multi-scale attention feature fusion processing is employed to form the final path coupling correlation matrix. .

[0107] It should be understood that multi-scale attention feature fusion aims to highlight the contribution of key regions in path interactions, thereby improving the accuracy of feature fusion. The specific process is as follows:

[0108] First, regarding local energy patterns Perform multi-scale spatial partitioning to obtain feature sets of sub-regions at different scales. For example, the feature matrix can be... The system divides the space into three scales: 2×2, 4×4, and 8×8, to capture spatial detail information at different scales.

[0109] Specifically, setting different scales for the first The feature matrices of the sub-regions are Then define the local attention weights. The calculation method is as follows: ;in, Indicates the first The mean of all energy pattern characteristic values ​​in each subregion, The total number of sub-regions at the corresponding scale, and the local attention weights. This reflects the weight of different sub-regions in their contribution to the overall acoustics.

[0110] Secondly, the local attention weights obtained above are used to analyze local energy pattern features at different scales. We perform weighted fusion separately to obtain the fused multi-scale attention feature matrix. The specific formula is as follows:

[0111] Furthermore, a cross-scale global attention mechanism is employed to process the feature matrix fused from different scales. Perform a re-fusion to determine the global attention weights for each scale. For example, it is defined as: ;in, The standard deviation of the feature values ​​within the fused feature matrix at each scale. Representing quantities at different scales, for example This corresponds to the three scales mentioned above: 2×2, 4×4, and 8×8.

[0112] Finally, based on the determined global attention weights The feature matrices fused at each scale are then weighted and fused again to form the final path coupling correlation matrix. :

[0113] It is understandable that the path coupling correlation matrix constructed through the above multi-scale attention feature fusion processing is... It comprehensively reflects the spatial energy distribution pattern and path coupling relationship of material interface path interaction, which can provide high-precision path association information for subsequent feature mapping processing of acoustic signals, thereby effectively supporting the accurate prediction of sound absorption performance.

[0114] Furthermore, the use of path coupling correlation matrix Set of acoustic signals for separate propagation paths The specific method for feature mapping is as follows: the acoustic contribution feature vectors of each layer are processed through matrix operations. (or ) obtained, among which The spatial energy pattern of interface path interaction has been integrated, thereby achieving precise decoupling mapping from path-level acoustic signals to the acoustic contributions of each material layer (foam layer, sound insulation cotton layer, damping plate layer, etc.).

[0115] Specifically, the training process of the convolutional network prediction model includes:

[0116] Acoustic contribution characteristic data of each layer of multiple standard sound-absorbing materials were collected to form an initial training sample set;

[0117] It should be noted that standard sound-absorbing materials refer to materials whose acoustic performance parameters have been precisely determined experimentally. For example, several representative sound-absorbing materials are selected, such as polyurethane foam, fiberglass wool, and damping boards of different densities. The acoustic contribution characteristic data of each material are collected and determined using the same processing methods described in S101 to S103 above, thereby constructing a complete initial training sample set. Each training sample specifically includes the independent acoustic contribution characteristics of each layer of each material, and the corresponding known overall sound absorption coefficient reference value.

[0118] Specifically, for example, if the overall sound absorption coefficient of a certain polyurethane foam material is 0.75, then the corresponding training sample data includes the acoustic contribution feature sets corresponding to its foam layer, sound insulation cotton layer, and damping plate layer, respectively, as well as the overall sound absorption coefficient of 0.75. The size of the initial training sample set can, for example, reach hundreds to thousands to meet the needs of network training.

[0119] The initial training sample set is input into the convolutional network prediction model for the first iteration of training to obtain the preliminary prediction error distribution characteristics.

[0120] In the specific implementation process, the convolutional network prediction model adopts a convolutional neural network (CNN) structure. This model includes three convolutional layers with a kernel size of 3×3 and a stride of 1, two pooling layers, and two fully connected layers. The input is the acoustic contribution feature vector of each layer after S103 mapping, with a fixed dimension of 64. The output is a single-value absorption coefficient. During training, the Adam optimizer is used, and the initial learning rate is determined based on experimental data, for example, set to 0.001. During training, the acoustic contribution features of each layer of each material are used as input, and the overall absorption coefficient is used as the label output, performing forward propagation and backpropagation operations on the network model.

[0121] Specifically, after the first training iteration, statistical analysis is performed on the error between the network prediction results and the actual overall sound absorption coefficient to obtain preliminary prediction error distribution characteristics. For example, the absolute value of the error or the mean square error (MSE) can be used for quantitative characterization. ;in, Indicates the first The actual overall sound absorption coefficient of each training sample. This represents the predicted output of the network model. This represents the total number of training samples. This step allows us to evaluate the predictive performance of the network model during its initial training.

[0122] Based on the preliminary prediction error distribution characteristics, the regional error weights of the training sample set are optimized and adjusted, and iterative training is performed again to obtain the pre-trained convolutional network prediction model.

[0123] It should be understood that, in order to further improve the accuracy of model predictions, the regional error weights of the training sample set are optimized and adjusted based on the prediction error distribution characteristics obtained during the initial training. The specific operation is as follows:

[0124] First, the error distribution characteristics are statistically analyzed by region. For example, the error is divided into several intervals (e.g., 0-0.05, 0.05-0.10, 0.10-0.15, etc.), and the number and distribution of samples in each error interval are statistically analyzed.

[0125] Then, for sample regions with large errors (e.g., regions with errors greater than 0.10), the weights of the corresponding training samples are increased to improve the network's fitting accuracy for these regions; conversely, for regions with small errors, the sample weights are appropriately reduced to avoid model overfitting. The weight adjustment factor can be set, for example, between 1.2 and 1.5.

[0126] Furthermore, based on the training sample set after the weight adjustment, the convolutional network prediction model is iteratively trained again until the model prediction error reaches a predetermined stable convergence condition. For example, the stable convergence condition can be set as the prediction error fluctuation range after 10 consecutive iterations of training being within a set threshold (e.g., within 0.005).

[0127] Through the above training process, a pre-trained convolutional network prediction model that can accurately predict the overall sound absorption performance of the composite sound-absorbing structure is finally obtained.

[0128] Example 2

[0129] Please see Figure 2 As shown, this embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any of the sound absorption performance testing methods for sound-absorbing materials described above.

[0130] Since the electronic device described in this embodiment is the electronic device used to implement the sound absorption performance testing method for sound-absorbing materials in the embodiments of this application, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the sound absorption performance testing method for sound-absorbing materials described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any electronic device used by those skilled in the art to implement the sound absorption performance testing method for sound-absorbing materials in the embodiments of this application falls within the scope of protection of this application.

[0131] Example 3

[0132] Please see Figure 3 As shown, this embodiment discloses a computer-readable storage medium, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements any of the sound absorption performance testing methods for sound-absorbing materials described above.

[0133] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters, weights, and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0134] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0135] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0136] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0137] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0138] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0140] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0141] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for testing the sound absorption performance of sound-absorbing materials, characterized in that, include: The overall acoustic response signal of the sound-absorbing material under test is collected. Based on the path recursive response characteristics caused by the cumulative propagation of multi-interface reflections, a recursive response matrix is ​​constructed, and the overall acoustic response signal is decoupled to obtain multiple separate propagation path acoustic signals. A topology feature extraction network with a residual cross-layer error correction module is constructed. The topology feature extraction network determines the error correction scale based on the local non-uniformity of the interface curvature in the topology of the sound-absorbing material under test, thereby obtaining an error-corrected topology feature map. A path coupling correlation matrix is ​​constructed based on the error correction topology feature map and the recursive response matrix. This matrix is ​​then used to perform feature mapping on the acoustic signals of the separated propagation paths to obtain the acoustic contribution characteristics of each layer of the sound-absorbing material under test. The acoustic contribution features of each layer are input into a pre-trained convolutional network prediction model, which outputs the predicted sound absorption coefficient of the sound-absorbing material to be tested.

2. The method according to claim 1, characterized in that, The recursive response matrix is ​​constructed based on the path recursive response characteristics caused by multi-interface reflection cumulative propagation, including: Collect multi-order reflection propagation signals from the interface of the sound-absorbing material under test, extract multi-order propagation delay difference features of the propagation path, and obtain a multi-order propagation delay feature set; Based on the frequency concentration characteristics of the multi-order propagation delay feature set, the propagation mode association relationship between paths is determined, and the path mode association feature is obtained. Clustering feature mapping is performed on the multi-level propagation delay feature set using path pattern association features, and a recursive response matrix is ​​formed based on the mapped clustering features.

3. The method according to claim 1, characterized in that, The path decoupling of the overall acoustic response signal to obtain multiple separate propagation path acoustic signals includes: Based on the sparse distribution characteristics of the recursive response matrix, sparse coding of the propagation path signal is performed to obtain the initial propagation path signal set; Based on the spectral energy difference characteristics of the path signals in the initial propagation path signal set, spectral domain adaptive matrix decomposition is performed to obtain the propagation path decoupling characteristics; Based on the propagation path decoupling characteristics, the overall acoustic response signal is decoupled to obtain multiple separate propagation path acoustic signals.

4. The method according to claim 1, characterized in that, The construction process of the topology feature extraction network with residual cross-layer error correction module includes: The interface curvature of the topological structure of the sound-absorbing material under test is refined by sampling, and the local non-uniform distribution features of the interface curvature are extracted to form a local non-uniform feature map. Based on the amplitude characteristics of regional curvature differences in the local non-uniform feature map, the local feature error correction scale of the residual cross-layer error correction module is determined, and the scale error correction mapping matrix is ​​obtained. Cross-layer feature weight allocation is performed based on the scale-correction mapping matrix, and a topological feature extraction network with residual cross-layer error correction module is obtained through cross-layer feature iterative fusion.

5. The method according to claim 1, characterized in that, The process of determining the scale-correction mapping matrix includes: The local oscillation amplitude of the interface curvature in different spatial regions of the topological structure of the sound-absorbing material under test is collected to obtain the local oscillation characteristics of the interface curvature. Spatial scale fusion mapping is performed based on the local oscillation characteristics of interface curvature, and the scale error correction mapping matrix is ​​determined based on the scale sensitivity of the fused region.

6. The method according to claim 1, characterized in that, The process of constructing the path coupling association matrix includes: Multi-channel feature convolution is performed on the error-corrected topological feature map and the recursive response matrix to form an initial coupled and correlated feature map; Based on the spatial energy distribution concentration characteristics of the path signals in the initial coupling correlation feature map, the local energy patterns of path interactions at the material interface are extracted. We utilize the local energy patterns of path interactions at material interfaces to perform multi-scale attention feature fusion processing and construct a path coupling correlation matrix.

7. The method according to claim 6, characterized in that, The extraction process of the local energy modes of the material interface path interaction includes: Local spatial clustering analysis is performed on the initial coupling correlation feature map to extract the energy concentration region features of the interface reflection path and obtain the initial energy pattern features; By utilizing the spatial distribution stability characteristics of the initial energy pattern, regional scale features are fused to obtain the local energy pattern of material interface path interaction.

8. The method according to claim 1, characterized in that, The training process of the convolutional network prediction model includes: Acoustic contribution characteristic data of each layer of multiple standard sound-absorbing materials were collected to form an initial training sample set; The initial training sample set is input into the convolutional network prediction model for the first iteration of training to obtain the preliminary prediction error distribution characteristics. Based on the preliminary prediction error distribution characteristics, the regional error weights of the training sample set are optimized and adjusted, and iterative training is performed again to obtain the pre-trained convolutional network prediction model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method described in any one of claims 1 to 8.

Images