Convolutional neural network-based acoustic metamaterial plate structure design method and system

Abstract

The invention relates to the technical field of acoustics, and discloses an acoustic metamaterial plate structure design method and system based on a convolutional neural network, and the method comprises the steps: obtaining a first acoustic metamaterial plate and a first acoustic metamaterial plate model based on a first random geometric parameter; performing finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; respectively training the first convolutional neural network model under different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp; performing similarity calculation on the first apparent feature and each second apparent feature to obtain the similarity corresponding to each plate structure design source model; according to the target board structure design model, feature conversion is carried out on the target acoustic performance of the to-be-designed acoustic metamaterial and the environmental data, and target geometric parameters corresponding to the to-be-designed acoustic metamaterial are obtained. According to the invention, the structural design of the acoustic metamaterial plate can be accurately and efficiently realized.

Description

Technical Field

[0001] The present invention relates to the field of acoustic technology, and in particular to a method and system for designing an acoustic metamaterial plate structure based on a convolutional neural network. Background Art

[0002] Acoustic metamaterials are synthetic materials with special acoustic properties. Through their unique microstructure design, they can achieve special control of sound waves and have broad application prospects in the fields of noise control, acoustic imaging, acoustic sensors, etc. With the continuous expansion of application scenarios, the performance requirements of acoustic metamaterials are gradually increasing. How to design a metamaterial plate structure with excellent acoustic performance has become a hot topic in current research.

[0003] Traditional acoustic metamaterial plate structure design technology mostly relies on empirical formulas, physical experiments and finite element analysis, but this technology has many disadvantages. First, this design method is inefficient, and the design results are often limited to existing empirical data; second, different acoustic metamaterials require the construction and training of different models, which consumes a lot of computing resources and is difficult to accurately meet the complex and changing actual application needs; finally, traditional methods are difficult to consider the impact of environmental factors on acoustic performance, resulting in certain limitations in the design results in practical applications. Summary of the invention

[0004] In view of the above-mentioned shortcomings of traditional technologies, the present invention provides a method and system for designing an acoustic metamaterial plate structure based on a convolutional neural network.

[0005] In a first aspect, an embodiment of the present invention provides a method for designing an acoustic metamaterial plate structure based on a convolutional neural network, comprising: Designing the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model respectively; Performing an acoustic performance test on the first acoustic metamaterial plate to obtain target acoustic performance, and performing a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; Based on the target finite element model, data augmentation is performed on several groups of second random geometric parameters to obtain augmented data sets, and the first convolutional neural network model is trained at different timestamps according to the augmented data sets to obtain a plate structure design source model corresponding to each timestamp, wherein the first convolutional neural network model includes a convolutional neural network module and an attention mechanism module; Based on the second convolutional neural network model, respectively extract the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each of the timestamps of the acoustic metamaterial, and respectively calculate the similarity of the first apparent feature and each of the second apparent features to obtain the similarity corresponding to each of the plate structure design source models, wherein the second convolutional neural network model includes a plurality of convolutional layers; A target plate structure design model corresponding to the acoustic metamaterial to be designed is determined based on the similarity, and target acoustic performance and environmental data of the acoustic metamaterial to be designed are feature-converted according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0006] Preferably, the performing an acoustic performance test on the first acoustic metamaterial plate to obtain a target acoustic performance includes: Performing an acoustic performance test on the first acoustic metamaterial plate to obtain initial acoustic performance; Acquiring environmental data of the acoustic metamaterial, and performing calculation processing on the environmental data based on a preset environmental function to obtain an environmental factor; The initial acoustic performance is weightedly corrected using the environmental factors to obtain the target acoustic performance.

[0007] Preferably, the preset environment function is characterized by the following formula: in, Represents environmental factors, Represents environmental data, Indicates reference environment data, Indicates the ambient temperature, Indicates the ambient humidity. Indicates dust concentration, represents the atmospheric pressure, , represents the temperature weight, represents the humidity weight, represents the dust concentration weight, represents the atmospheric pressure weight, represents a dynamic adjustment function, Represents the dust concentration attenuation coefficient, represents the dust concentration adjustment factor, represents the pressure function, Representation parameters Parameters Importance ratio; The dynamic adjustment function is characterized by the following formula: in, , represents the temperature correction factor, represents the humidity attenuation coefficient; The pressure function is characterized by the following formula: in, and represents the correction factor for the high pressure area, Represents the attenuation coefficient of the low pressure area.

[0008] Preferably, performing finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model comprises: Discretizing the first acoustic metamaterial plate model to obtain an initial finite element model; Performing acoustic simulation on the initial finite element model to obtain comparative acoustic performance; calculating a performance deviation between the target acoustic performance and the comparative acoustic performance; The material properties and boundary conditions of the initial finite element model are iteratively adjusted based on the performance deviation until the performance deviation reaches a convergence state, thereby obtaining a target finite element model.

[0009] Preferably, performing data augmentation on a plurality of groups of second random geometric parameters based on the target finite element model to obtain an augmented data set comprises: Acquire several groups of second random geometric parameters, and design the acoustic metamaterial based on each group of the second random geometric parameters to obtain a corresponding second acoustic metamaterial plate model; Acquiring environmental data when designing each of the second acoustic metamaterial plate models, and performing calculation processing on each set of the environmental data to obtain corresponding environmental factors; Inputting each of the second acoustic metamaterial plate models into the target finite element model for acoustic simulation to obtain the acoustic performance corresponding to each of the second acoustic metamaterial plate models; The second random geometric parameters, the acoustic performance and the environmental factors corresponding to each of the second acoustic metamaterial plate models constitute an augmented data set.

[0010] Preferably, the first convolutional neural network model is trained respectively at different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp, including: Dividing the augmented data set into a training set and a test set according to a preset ratio; Based on the training set, the first convolutional neural network model is trained respectively at different timestamps to obtain a first plate structure design source model corresponding to each timestamp; Based on the test set, each of the first plate structure design source models is evaluated and optimized respectively to obtain a second plate structure design source model corresponding to each of the timestamps; Each of the second plate structure design source models is characterized as a plate structure design source model corresponding to the time stamp.

[0011] Preferably, the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature of the acoustic metamaterial corresponding to each of the timestamps are extracted based on the second convolutional neural network model, and the first apparent feature and each of the second apparent features are respectively calculated for similarity to obtain the similarity corresponding to each of the plate structure design source models, including: Acquiring a first surface image of the acoustic metamaterial to be designed and a second surface image of the acoustic metamaterial at each of the timestamps; Based on the second convolutional neural network model, feature extraction is performed on the first surface image and each of the second surface images respectively, so as to obtain a first surface feature of the acoustic metamaterial to be designed and a second surface feature of the acoustic metamaterial corresponding to each of the timestamps; Calculating the cosine similarity and the Pearson correlation coefficient between the first appearance feature and each of the second appearance features; The cosine similarity and the Pearson correlation coefficient are weightedly fused to obtain the similarity corresponding to each of the plate structure design source models.

[0012] Preferably, determining the target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity comprises: If the maximum value among the similarities is greater than a preset similarity threshold, the plate structure design source model corresponding to the maximum value is characterized as a first plate structure design model corresponding to the acoustic metamaterial to be designed; Using a particle swarm optimization algorithm to optimize the parameters of the first plate structure design model to obtain a second plate structure design model; The second plate structure design model is characterized as a target plate structure design model corresponding to the acoustic metamaterial to be designed.

[0013] Preferably, the step of performing feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed includes: Performing an acoustic performance test on the acoustic metamaterial to be designed to obtain the acoustic performance corresponding to the acoustic metamaterial to be designed; Acquiring environmental data when performing an acoustic performance test on the acoustic metamaterial to be designed, and performing computational processing on the environmental data to obtain an environmental factor corresponding to the acoustic metamaterial to be designed; Performing weighted correction on the acoustic performance based on the environmental factors to obtain the target acoustic performance corresponding to the acoustic metamaterial to be designed; The target plate structure design model is used to extract and map the target acoustic performance and the environmental factors to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0014] In a second aspect, an embodiment of the present invention provides an acoustic metamaterial plate structure design system based on a convolutional neural network, comprising: A design module, used for designing the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model; A finite element analysis module, configured to perform an acoustic performance test on the first acoustic metamaterial plate to obtain a target acoustic performance, and perform a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; A model building module, used to perform data augmentation on several groups of second random geometric parameters based on the target finite element model to obtain an augmented data set, and train the first convolutional neural network model at different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp, wherein the first convolutional neural network model includes a convolutional neural network module and an attention mechanism module; A similarity calculation module is used to extract the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each of the timestamps based on the second convolutional neural network model, and perform similarity calculation on the first apparent feature and each of the second apparent features to obtain the similarity corresponding to each of the plate structure design source models, wherein the second convolutional neural network model includes a plurality of convolutional layers; A geometric parameter determination module is used to determine a target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity, and perform feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0015] Compared with the prior art, the method and system for designing an acoustic metamaterial plate structure based on a convolutional neural network in an embodiment of the present invention have the following beneficial effects: the acoustic performance is corrected by fusing multi-scale environmental data features, providing a reliable basis for subsequent design; a large amount of reliable training data is generated by using data augmentation technology, and multiple plate structure design source models are obtained based on training of the augmented data set, which effectively improves the generalization ability and adaptability of the model; the target plate structure design model is determined by similarity calculation, which can quickly and accurately match the most suitable model for the acoustic metamaterial to be designed, thereby improving the design efficiency and accuracy; the geometric parameters of the acoustic metamaterial are accurately designed based on the convolutional neural network, which realizes the accurate mapping from performance and environmental data to structural geometric parameters, and provides strong technical support for the research and development and application of acoustic metamaterials. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of a flow chart of a method for designing an acoustic metamaterial plate structure based on a convolutional neural network according to an embodiment of the present invention; Figure 2 is a schematic diagram of a process for obtaining target acoustic performance and a target finite element model according to an embodiment of the present invention; Figure 3 is a schematic diagram of a process for similarity calculation according to an embodiment of the present invention; Figure 4 It is a structural schematic diagram of an acoustic metamaterial plate structure design system based on a convolutional neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0018] In the description of the present invention, it should be understood that the terms "first" and "second" etc. are used in the present invention to distinguish different objects rather than to describe a specific order.

[0019] In the description of the present invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0020] like Figure 1 As shown, an embodiment of the present invention provides a method for designing an acoustic metamaterial plate structure based on a convolutional neural network, comprising the steps of: S1. Designing the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model; A set of random geometric parameters is taken as the first random geometric parameters, and the first acoustic metamaterial plate is manufactured according to the first random geometric parameters and the acoustic metamaterial. The geometric parameters include plate thickness, aperture size, hole spacing and arrangement. Specifically, the first random geometric parameters of this embodiment are plate thickness of 15 mm, aperture of 4 mm, hole spacing of 20 mm, and square arrangement.

[0021] In the modeling software, a three-dimensional geometric model of the first acoustic metamaterial plate, i.e., the first acoustic metamaterial plate model, is created according to the first random geometric parameters. Among them, the material properties of the acoustic metamaterial are input into the model so that the model can accurately simulate the response of the acoustic metamaterial plate under the action of sound waves, and the setting of boundary conditions will affect the propagation and emission of sound waves in the model, which is crucial to accurately simulate the acoustic performance. Specifically, the material properties include material density, elastic modulus, damping coefficient, material sound velocity and thermal expansion coefficient, and the boundary conditions include fixed boundaries, free boundaries and periodic boundaries. Furthermore, the material properties of this embodiment are material density 2.6g / cm 3 , elastic modulus 100GPa, damping coefficient 0.05, material sound velocity 3000m / s, thermal expansion coefficient 1*10 -5 / ℃.

[0022] S2. Performing an acoustic performance test on the first acoustic metamaterial plate to obtain target acoustic performance, and performing a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; Specifically, Figure 2 As shown, step S2 includes: S201, performing an acoustic performance test on the first acoustic metamaterial plate to obtain initial acoustic performance; The acoustic performance test of the first acoustic metamaterial plate is performed to obtain the initial acoustic performance at a specific frequency. Specifically, the initial acoustic performance includes sound absorption coefficient, acoustic impedance and acoustic transmission loss. In this embodiment, the acoustic performance test of the first acoustic metamaterial plate made by the first random geometric parameters is performed to obtain the initial acoustic performance at 1000 Hz, and the initial acoustic performance is a sound absorption coefficient of 0.45, an acoustic impedance of 1500+200ikg / (m 2 *s) and sound transmission loss 15dB.

[0023] S202, obtaining environmental data of the acoustic metamaterial, and performing calculation processing on the environmental data based on a preset environmental function to obtain an environmental factor; Environmental data is collected through an environmental sensor network, and the environmental data is input into a preset environmental function to obtain environmental factors.

[0024] Specifically, the following formula is used to characterize the preset environment function: in, Represents environmental factors, Represents environmental data, Indicates reference environment data, Indicates the ambient temperature, Indicates the ambient humidity. Indicates dust concentration, represents the atmospheric pressure, , represents the temperature weight, represents the humidity weight, represents the dust concentration weight, represents the atmospheric pressure weight, represents a dynamic adjustment function, Represents the dust concentration attenuation coefficient, represents the dust concentration adjustment factor, represents the pressure function, Representation parameters Parameters Importance ratio.

[0025] Furthermore, the following formula is used to characterize the dynamic adjustment function: in, , represents the temperature correction factor, Represents the humidity attenuation coefficient.

[0026] Furthermore, the following formula is used to characterize the pressure function: in, and represents the correction factor for the high pressure area, Represents the attenuation coefficient of the low pressure area.

[0027] In this embodiment, the environmental sensor network is used to collect the ambient temperature of 28°C (reference temperature 25°C), the ambient humidity of 55% (reference humidity 50%), and the dust concentration of 60μg / m 3 (Reference dust concentration 50μg / m 3 ), atmospheric pressure is 100.5kPa (reference atmospheric pressure 100kPa); take temperature weight 0.3, humidity weight 0.2, dust concentration weight 0.7, atmospheric pressure weight 0.4, temperature correction coefficient 0.1, humidity attenuation coefficient 0.05, dust concentration attenuation coefficient 0.9, dust concentration adjustment factor 60μg 2 / m6 , take the importance ratio , , , , , They are 4, 5, 3, 3.5, 4.5 and 2.5 respectively. The total calculated environmental factor is 1.12598.

[0028] S203, using environmental factors to perform weighted correction on the initial acoustic performance to obtain target acoustic performance; That is, the target acoustic performance is the product of the initial acoustic performance and the environmental factors.

[0029] Specifically, the target acoustic performance of this embodiment is a sound absorption coefficient of 0.507, an acoustic impedance of 1688.97+225.2ikg / (m 2 *s) and sound transmission loss 16.89dB.

[0030] Furthermore, if Figure 2 As shown, step S2 also includes: S204, discretizing the first acoustic metamaterial plate model to obtain an initial finite element model; The material properties and boundary conditions of the first acoustic metamaterial plate model are determined, and the first acoustic metamaterial plate model, material properties and boundary conditions are input into the finite element software for discretization to obtain an initial finite element model. Specifically, the finite element software divides the original continuous three-dimensional geometric model into a finite number of units according to certain rules, determines the connection points between the units, and assigns material properties and boundary conditions to these units and nodes, thereby converting the complex physical model into a finite element model capable of numerical calculation, and realizing the simulation analysis of the acoustic metamaterial plate.

[0031] S205, performing acoustic simulation on the initial finite element model to obtain comparative acoustic performance; The initial finite element model was acoustically simulated using standard sound waves to obtain comparative acoustic performance.

[0032] S206, calculating the performance deviation between the target acoustic performance and the comparative acoustic performance; Since the acoustic performance includes three evaluation indicators, namely, sound absorption coefficient, acoustic impedance and acoustic transmission loss, the performance deviation is represented by the weighted result of the deviation of each indicator. Specifically, the present embodiment uses the following formula to calculate the performance deviation: in, Indicates performance deviation, represents the sound absorption coefficient deviation, Indicates the acoustic impedance deviation, It should be noted that the weight coefficient of each indicator deviation can be adjusted according to actual conditions and is not limited to 2, 0.0001 and 0.1 set in this embodiment.

[0033] S207. Iteratively adjust the material properties and boundary conditions of the initial finite element model based on the performance deviation until the performance deviation reaches a convergence state, thereby obtaining a target finite element model.

[0034] The material properties and boundary conditions of the initial finite element model are iteratively adjusted based on the performance deviation until the performance deviation is less than a preset deviation threshold or no longer decreases, and the iterative adjustment is stopped to obtain a target finite element model. In this embodiment, the preset deviation threshold is 0.3.

[0035] S3. Based on the target finite element model, data augmentation is performed on several groups of second random geometric parameters to obtain augmented data sets, and the first convolutional neural network model is trained at different timestamps according to the augmented data sets to obtain a plate structure design source model corresponding to each timestamp; Specifically, step S3 includes: 1) obtaining several groups of second random geometric parameters, and designing the acoustic metamaterial based on each group of second random geometric parameters to obtain a corresponding second acoustic metamaterial plate model; Take several groups of random geometric parameters as several groups of second random geometric parameters. In the modeling software, create a three-dimensional geometric model of the corresponding second acoustic metamaterial plate, i.e., a second acoustic metamaterial plate model, according to each group of second random geometric parameters. Specifically, in this embodiment, 10 groups of second random geometric parameters are obtained, i.e., 10 second acoustic metamaterial plate models are constructed.

[0036] 2) Obtaining environmental data when designing each second acoustic metamaterial plate model, and performing calculations on each set of environmental data to obtain corresponding environmental factors; The environmental data when designing each second acoustic metamaterial plate model is collected through the environmental sensor network, and each set of environmental data is input into the preset environmental function to obtain the corresponding environmental factor. It should be noted that the process of calculating the environmental factor through the preset environmental function can be found in the previous text, and will not be repeated here.

[0037] 3) inputting each second acoustic metamaterial plate model into the target finite element model for acoustic simulation to obtain the acoustic performance corresponding to each second acoustic metamaterial plate model; 4) forming an augmented data set with the second random geometric parameters, acoustic performance and environmental factors corresponding to each second acoustic metamaterial plate model; The augmented data set obtained by the data augmentation technology in this embodiment includes 10 groups of augmented data, and each group of augmented data includes corresponding second random geometric parameters, acoustic performance and environmental factors.

[0038] Furthermore, step S3 also includes: 5) Divide the augmented data set into training set and test set according to the preset ratio; This embodiment uses the random forest algorithm to divide the augmented data set into a training set and a test set in an 8:2 ratio.

[0039] 6) Based on the training set, the first convolutional neural network model is trained at different timestamps to obtain the first plate structure design source model corresponding to each timestamp; The following is a detailed description of the first convolutional neural network model: The first convolutional neural network model includes an input layer, a fully connected network module, a convolutional neural network module, an attention mechanism module, and an output layer. The input layer is used to receive input data such as geometric parameters, acoustic performance, and environmental factors; the fully connected network module is used to map acoustic performance and environmental factors to a high-dimensional space and extract high-level features; the convolutional neural network module is used to reshape the high-level feature vector into a two-dimensional feature map and extract spatial features, capture the local dependencies between geometric parameters, and output the convolutional feature map; the attention mechanism module is used to enhance the key features in the convolutional feature map and suppress irrelevant features, capture the global dependencies between geometric parameters, and output the weighted feature vector; the output layer is used to map the weighted feature vector to the geometric parameter space and output the designed geometric parameters.

[0040] The training set is used to train the first convolutional neural network model at different timestamps, and the trained first convolutional neural network model is represented as a first plate structure design source model corresponding to each timestamp.

[0041] 7) Each first plate structure design source model is evaluated and optimized based on the test set to obtain a second plate structure design source model corresponding to each timestamp; The test set is used to evaluate the first plate structure design source model, and the model is optimized according to the evaluation results. The parameters of the model are continuously adjusted to gradually improve the performance of the model, and finally the target model, i.e., the second plate structure design source model, is obtained, so that it can output the design geometric parameters more accurately.

[0042] 8) Each second plate structure design source model is represented as a plate structure design source model corresponding to a timestamp.

[0043] This embodiment stores each plate structure design source model in the source model library according to the timestamp.

[0044] S4, extracting the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each timestamp of the acoustic metamaterial based on the second convolutional neural network model, and performing similarity calculation on the first apparent feature and each second apparent feature, respectively, to obtain the similarity corresponding to each plate structure design source model; Specifically, Figure 3 As shown, step S4 includes: S401, obtaining a first surface image of the acoustic metamaterial to be designed and a second surface image of the acoustic metamaterial at each time stamp; The first surface image of the acoustic metamaterial to be designed and the second surface image of the acoustic metamaterial at each time stamp are obtained by a high-definition camera. At the same time, the acoustic metamaterial to be designed and the acoustic metamaterial at each time stamp are spectrally analyzed to obtain the first spectral feature of the acoustic metamaterial to be designed and the second spectral feature of the acoustic metamaterial at each time stamp. In addition, the first hardness and the first volume density of the acoustic metamaterial to be designed, as well as the second hardness and the second volume density of the acoustic metamaterial at each time stamp are obtained respectively. It can be understood that the acoustic metamaterial to be designed and the acoustic metamaterial belong to different materials of the same type.

[0045] S402, performing feature extraction on the first surface image and each second surface image based on the second convolutional neural network model, to obtain a first surface feature of the acoustic metamaterial to be designed and a second surface feature of the acoustic metamaterial corresponding to each timestamp; The second convolutional neural network model is described in detail below: The second convolutional neural network model is used to process the surface image to obtain the image surface features. Specifically, the second convolutional neural network model is composed of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, a second fully connected layer and an output layer connected in series in sequence. The input layer preprocesses the surface image to obtain a surface enhanced image; the first convolutional layer uses gray level co-occurrence matrix and local binary pattern to extract the texture features of the surface enhanced image; the second convolutional layer uses adaptive median filtering and edge detection to extract the roughness and surface defect features of the surface enhanced image; the third convolutional layer further extracts the high-level texture and structural features of the surface enhanced image; the fourth convolutional layer uses a three-dimensional reconstruction algorithm to extract the thickness uniformity features of the surface enhanced image; the first fully connected layer uses a color histogram and color moment to extract RGB features; the second fully connected layer uses a highlight detection algorithm and a reflectivity algorithm to calculate and extract the reflectivity features; the output layer receives the image processing results of the fully connected layer and splices the image features to output the image surface features.

[0046] The image surface features, first spectral features, first hardness and first volume density of the acoustic metamaterial to be designed are feature spliced ​​to obtain the first apparent features of the acoustic metamaterial to be designed, and the image surface features, second spectral features, second hardness and second volume density of the acoustic metamaterial are feature spliced ​​to obtain the second apparent features of the acoustic metamaterial corresponding to each timestamp.

[0047] S403, calculating the cosine similarity and Pearson correlation coefficient of the first appearance feature and each second appearance feature; S404, weighted fusion is performed on the cosine similarity and the Pearson correlation coefficient to obtain the similarity corresponding to each plate structure design source model.

[0048] It can be understood that the similarity is actually a comprehensive similarity, which is obtained by weighted fusion of the cosine similarity and the Pearson correlation coefficient of the first appearance feature and the second appearance feature.

[0049] S5. Determine a target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity, and perform feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0050] Specifically, step S5 includes: 1) If the maximum value among the similarities is greater than a preset similarity threshold, the plate structure design source model corresponding to the maximum value is characterized as the first plate structure design model corresponding to the acoustic metamaterial to be designed; When the highest comprehensive similarity is greater than the preset similarity threshold, the corresponding plate structure design source model in the source model library is taken as the first plate structure design model. The preset similarity threshold is set to 0.7 in this embodiment. On the contrary, if it is determined that there is no plate structure design source model that can be transferred for learning of the acoustic metamaterial to be designed, the above data augmentation step needs to be performed to continue to increase the training data to enrich the source model library.

[0051] 2) Using the particle swarm optimization algorithm to optimize the parameters of the first plate structure design model to obtain the second plate structure design model; Before optimizing the parameters of the first plate structure design model, it is necessary to determine the optimization objective function. The following is an explanation of the construction process of the optimization objective function: i) Obtain geometric parameter deviation The environmental data of the acoustic metamaterial to be designed is collected through the environmental sensor network, and the environmental data is input into the preset environmental function to obtain the environmental factor. Another set of random geometric parameters is taken as the third random geometric parameters, and the acoustic metamaterial to be designed is made into an acoustic metamaterial board according to the third random geometric parameters, and the acoustic performance of the acoustic metamaterial board is tested to obtain the acoustic performance. The acoustic performance and the environmental factor are input into the first board structure design model to obtain the geometric parameters, and the geometric parameter deviation of the third random geometric parameters and the geometric parameters is calculated.

[0052] ii) Determine the optimization objective function The optimization objective function is determined according to the geometric parameter deviation. Specifically, the optimization objective function is characterized by the following formula: in, represents the optimization objective function, represents the deviation function, Indicates geometric size deviation, represents the size weight, and represents the size penalty coefficient, represents the geometric parameters, represents the third random geometric parameter, represents the penalty weight, represents the quality weight, represents the similarity weight, Represents the structural equivalent stress related to geometric size deviation, represents the allowable stress of the structure, represents the weight of acoustic metamaterial per unit area corresponding to the geometric parameters, represents the weight of the acoustic metamaterial per unit area corresponding to the third random geometric parameter, represents the volume density of the acoustic metamaterial to be designed, represents the unit area, The plate thickness represents the geometric parameters, The plate thickness represents the third random geometric parameter, represents the Pearson correlation coefficient between the third random geometric parameter vector and the geometric parameter vector.

[0053] The particle swarm optimization algorithm is used to optimize the hyperparameters of the first plate structure design model, generate two chaotic sequences at two different scales, and perform chaotic mapping on the population, as shown below: in, Indicates The chaos value of the dimension, Indicates the first chaotic sequence corresponding to The chaos value of the dimension, The corresponding first The chaos value of the dimension, and represents the scale weight, Indicates The chaos value of the dimension, represents chaotic random numbers, Represents the control parameter, which is used to adjust the chaotic sequence distribution.

[0054] Calculate the fitness of the particle and determine it according to the particle fitness Dimension The optimal position of the particle at the iteration and the optimal position of the population , update the particle position and velocity as follows: in, Indicates that the particle Dimension Iterate the position update, Indicates that the particle Dimension Iteration position, Indicates that the particle Dimension Iteration update speed, Indicates that the particle Dimension Iteration speed, Indicates Dynamically adjust weights for each iteration, represents the golden sine coefficient, represents the maximum number of iterations, represents the Levy flight coefficient, represents the individual learning factor, represents the social learning factor, express Random numbers, and express Random numbers, represents the Levy flight strategy, represents the adaptive flight step length, represents the regularization coefficient, represents the current population vector, represents the current population vector mean, represents the Euclidean norm.

[0055] Dynamically adjust weights The population diversity index and convergence speed are introduced and calculated by the fuzzy inference system as follows: in, , and represents a constant coefficient, Indicators of population diversity The membership function of Indicates the convergence speed The membership function of represents the number of particles, Indicates that the population Dimensionally The average position of the iterations, The convergence speed represents the change of the global optimal solution in the last five generations. Represents the objective function.

[0056] The elite reverse mutation strategy is used to mutate the global unchanged population for five consecutive iterations, and the quantum tunnel crossing mechanism is used to update the particles that fall into the local optimum, as shown below: in, Indicates that the particle mutation is in dimension The position after mutation, Indicates that the particle is trapped in the local optimum Dimensionally The updated position after iterations, represents the truncation function, and express Variable lower and upper bounds for the dimensional search space, represents the reverse learning coefficient, represents the Cauchy random distribution function, represents the standard deviation of particle positions, and express Uniform random number.

[0057] The fitness of the particles is calculated to determine the optimal position of the particles and the optimal position of the population, and the particle position and speed are updated. The iteration is repeated until the optimization objective function is minimized and the iteration is stopped. The optimal model parameters are determined and the second plate structure design model is output.

[0058] This embodiment optimizes the size weight in the objective function Pick , penalty weight Take 0.1, quality weight Take 0.2, similarity weight Take 0.1, size penalty coefficient and Take 2 and 1.5; set the number of particles in the particle population N to 50, the maximum number of iterations K to 100, and the scale weight and Take 0.6, 0.4, chaotic random numbers Take 0.5, control parameter Take 0.8; and update the particle position and velocity, and repeat the iteration (when k=20, When 1.2 and k=40, When 0.8 and k=60, When 0.65 and k=65, is 0.64, k=6, The iteration is stopped when the optimization objective function reaches the minimum and no longer changes. The model parameters corresponding to the 65th iteration are taken as the optimal model parameters and the second plate structure design model is output.

[0059] 3) characterizing the second plate structure design model as a target plate structure design model corresponding to the acoustic metamaterial to be designed; Furthermore, step S5 further includes: 4) Conducting acoustic performance tests on the acoustic metamaterial to be designed to obtain the acoustic performance corresponding to the acoustic metamaterial to be designed; 5) Obtaining environmental data when the acoustic performance test of the acoustic metamaterial to be designed is carried out, and performing calculations on the environmental data to obtain environmental factors corresponding to the acoustic metamaterial to be designed; 6) Perform weighted correction on acoustic performance based on environmental factors to obtain the target acoustic performance corresponding to the acoustic metamaterial to be designed; It should be noted that the execution process of steps 4) to 6) can refer to steps S201 to S203 described above, and will not be repeated here.

[0060] 7) The target plate structure design model is used to extract and map the target acoustic performance and environmental factors to obtain the target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0061] It can be understood that the convolutional neural network module and the attention mechanism module in the target board structure design model are both feature extraction. Specifically, this embodiment uses the target acoustic performance (sound absorption coefficient 0.55, acoustic impedance 1700+250ikg / m 2*s and acoustic transmission loss 10dB) and environmental factor 1.089 were input into the target plate structure design model for feature extraction and mapping, and the target geometric parameters (plate thickness 5mm, hole diameter 2mm, hole spacing 10mm, hexagonal arrangement) were obtained.

[0062] The embodiment of the present invention is a method for designing an acoustic metamaterial plate structure based on a convolutional neural network. The method corrects the acoustic performance by fusing multi-scale environmental data features, thereby providing a reliable basis for subsequent design. The method uses data augmentation technology to generate a large amount of reliable training data, and obtains multiple plate structure design source models based on the augmented data set training, thereby effectively improving the generalization ability and adaptability of the model. The target plate structure design model is determined by similarity calculation, so that the most suitable model can be quickly and accurately matched for the acoustic metamaterial to be designed, thereby improving the design efficiency and accuracy. The method accurately designs the geometric parameters of the acoustic metamaterial based on the convolutional neural network, thereby realizing the accurate mapping from performance and environmental data to structural geometric parameters, thereby providing strong technical support for the research and development and application of acoustic metamaterials.

[0063] Based on the above-mentioned convolutional neural network-based acoustic metamaterial plate structure design method, Figure 4 As shown, an embodiment of the present invention provides an acoustic metamaterial plate structure design system based on a convolutional neural network, comprising: A design module 1 is used to design the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model; Finite element analysis module 2, used for performing an acoustic performance test on the first acoustic metamaterial plate to obtain target acoustic performance, and performing a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; A model building module 3 is used to perform data augmentation on several groups of second random geometric parameters based on the target finite element model to obtain an augmented data set, and train the first convolutional neural network model at different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp, wherein the first convolutional neural network model includes a convolutional neural network module and an attention mechanism module; A similarity calculation module 4 is used to extract the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each time stamp of the acoustic metamaterial based on the second convolutional neural network model, and perform similarity calculation on the first apparent feature and each second apparent feature to obtain the similarity corresponding to each plate structure design source model, wherein the second convolutional neural network model includes a plurality of convolutional layers; The geometric parameter determination module 5 is used to determine the target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity, and perform feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain the target geometric parameters corresponding to the acoustic metamaterial to be designed.

[0064] It should be noted that each module in the above-mentioned acoustic metamaterial plate structure design system based on convolutional neural network can be fully or partially implemented by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of an acoustic metamaterial plate structure design system based on convolutional neural network, please refer to the definition of an acoustic metamaterial plate structure design method based on convolutional neural network in the above text. The two have the same functions and effects, which will not be repeated here.

[0065] In summary, the embodiment of the present invention is a method and system for designing an acoustic metamaterial plate structure based on a convolutional neural network. The method and system correct the acoustic performance by fusing multi-scale environmental data features, thereby providing a reliable basis for subsequent design. The data augmentation technology is used to generate a large amount of reliable training data, and multiple plate structure design source models are obtained based on the augmented data set training, thereby effectively improving the generalization ability and adaptability of the model. The target plate structure design model is determined by similarity calculation, so that the most suitable model can be quickly and accurately matched for the acoustic metamaterial to be designed, thereby improving the design efficiency and accuracy. The geometric parameters of the acoustic metamaterial are accurately designed based on the convolutional neural network, thereby realizing the accurate mapping from performance and environmental data to structural geometric parameters, thereby providing strong technical support for the research and development and application of acoustic metamaterials.

[0066] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A method for designing acoustic metamaterial plate structure based on convolutional neural network, characterized in that: include: Designing the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model respectively; Performing an acoustic performance test on the first acoustic metamaterial plate to obtain target acoustic performance, and performing a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; Based on the target finite element model, data augmentation is performed on several groups of second random geometric parameters to obtain augmented data sets, and the first convolutional neural network model is trained at different timestamps according to the augmented data sets to obtain a plate structure design source model corresponding to each timestamp, wherein the first convolutional neural network model includes a convolutional neural network module and an attention mechanism module; Based on the second convolutional neural network model, respectively extract the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each of the timestamps of the acoustic metamaterial, and respectively calculate the similarity of the first apparent feature and each of the second apparent features to obtain the similarity corresponding to each of the plate structure design source models, wherein the second convolutional neural network model includes a plurality of convolutional layers; A target plate structure design model corresponding to the acoustic metamaterial to be designed is determined based on the similarity, and target acoustic performance and environmental data of the acoustic metamaterial to be designed are feature-converted according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

2. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The performing of an acoustic performance test on the first acoustic metamaterial plate to obtain a target acoustic performance includes: Performing an acoustic performance test on the first acoustic metamaterial plate to obtain initial acoustic performance; Acquiring environmental data of the acoustic metamaterial, and performing calculation processing on the environmental data based on a preset environmental function to obtain an environmental factor; The initial acoustic performance is weightedly corrected using the environmental factors to obtain the target acoustic performance.

3. The method for designing an acoustic metamaterial plate structure according to claim 2, characterized in that: The preset environment function is characterized by the following formula: in, Represents environmental factors, Represents environmental data, Indicates reference environment data, Indicates the ambient temperature, Indicates the ambient humidity. Indicates dust concentration, represents the atmospheric pressure, , represents the temperature weight, represents the humidity weight, represents the dust concentration weight, represents the atmospheric pressure weight, represents a dynamic adjustment function, Represents the dust concentration attenuation coefficient, represents the dust concentration adjustment factor, represents the pressure function, Representation parameters Parameters Importance ratio; The dynamic adjustment function is characterized by the following formula: in, , represents the temperature correction factor, represents the humidity attenuation coefficient; The pressure function is characterized by the following formula: in, and represents the correction factor for the high pressure area, Represents the attenuation coefficient of the low pressure area.

4. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The step of performing finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model includes: Discretizing the first acoustic metamaterial plate model to obtain an initial finite element model; Performing acoustic simulation on the initial finite element model to obtain comparative acoustic performance; calculating a performance deviation between the target acoustic performance and the comparative acoustic performance; The material properties and boundary conditions of the initial finite element model are iteratively adjusted based on the performance deviation until the performance deviation reaches a convergence state, thereby obtaining a target finite element model.

5. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The step of performing data augmentation on a plurality of groups of second random geometric parameters based on the target finite element model to obtain an augmented data set includes: Acquire several groups of second random geometric parameters, and design the acoustic metamaterial based on each group of the second random geometric parameters to obtain a corresponding second acoustic metamaterial plate model; Acquiring environmental data when designing each of the second acoustic metamaterial plate models, and performing calculation processing on each set of the environmental data to obtain corresponding environmental factors; Inputting each of the second acoustic metamaterial plate models into the target finite element model for acoustic simulation to obtain the acoustic performance corresponding to each of the second acoustic metamaterial plate models; The second random geometric parameters, the acoustic performance and the environmental factors corresponding to each of the second acoustic metamaterial plate models constitute an augmented data set.

6. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The first convolutional neural network model is trained respectively at different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp, including: Dividing the augmented data set into a training set and a test set according to a preset ratio; Based on the training set, the first convolutional neural network model is trained respectively at different timestamps to obtain a first plate structure design source model corresponding to each timestamp; Based on the test set, each of the first plate structure design source models is evaluated and optimized respectively to obtain a second plate structure design source model corresponding to each of the timestamps; Each of the second plate structure design source models is characterized as a plate structure design source model corresponding to the timestamp.

7. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The first surface feature of the acoustic metamaterial to be designed and the second surface feature of the acoustic metamaterial corresponding to each of the timestamps are extracted based on the second convolutional neural network model, and the first surface feature and each of the second surface features are respectively calculated for similarity to obtain the similarity corresponding to each of the plate structure design source models, including: Acquiring a first surface image of the acoustic metamaterial to be designed and a second surface image of the acoustic metamaterial at each of the timestamps; Based on the second convolutional neural network model, feature extraction is performed on the first surface image and each of the second surface images respectively, so as to obtain a first surface feature of the acoustic metamaterial to be designed and a second surface feature of the acoustic metamaterial corresponding to each of the timestamps; Calculating the cosine similarity and the Pearson correlation coefficient between the first appearance feature and each of the second appearance features; The cosine similarity and the Pearson correlation coefficient are weightedly fused to obtain the similarity corresponding to each of the plate structure design source models.

8. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The step of determining a target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity comprises: If the maximum value among the similarities is greater than a preset similarity threshold, the plate structure design source model corresponding to the maximum value is characterized as a first plate structure design model corresponding to the acoustic metamaterial to be designed; Using a particle swarm optimization algorithm to optimize the parameters of the first plate structure design model to obtain a second plate structure design model; The second plate structure design model is characterized as a target plate structure design model corresponding to the acoustic metamaterial to be designed.

9. The method for designing an acoustic metamaterial plate structure according to claim 1, characterized in that: The step of performing feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed includes: Performing an acoustic performance test on the acoustic metamaterial to be designed to obtain the acoustic performance corresponding to the acoustic metamaterial to be designed; Acquiring environmental data when performing an acoustic performance test on the acoustic metamaterial to be designed, and performing computational processing on the environmental data to obtain an environmental factor corresponding to the acoustic metamaterial to be designed; Performing weighted correction on the acoustic performance based on the environmental factors to obtain the target acoustic performance corresponding to the acoustic metamaterial to be designed; The target plate structure design model is used to extract and map the target acoustic performance and the environmental factors to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

10. An acoustic metamaterial plate structure design system based on convolutional neural network, characterized in that: include: A design module, used for designing the acoustic metamaterial based on the first random geometric parameter to obtain a first acoustic metamaterial plate and a first acoustic metamaterial plate model; A finite element analysis module, configured to perform an acoustic performance test on the first acoustic metamaterial plate to obtain a target acoustic performance, and perform a finite element analysis on the first acoustic metamaterial plate model based on the target acoustic performance to obtain a target finite element model; A model building module, used to perform data augmentation on several groups of second random geometric parameters based on the target finite element model to obtain an augmented data set, and train the first convolutional neural network model at different timestamps according to the augmented data set to obtain a plate structure design source model corresponding to each timestamp, wherein the first convolutional neural network model includes a convolutional neural network module and an attention mechanism module; A similarity calculation module is used to extract the first apparent feature of the acoustic metamaterial to be designed and the second apparent feature corresponding to each of the timestamps based on the second convolutional neural network model, and perform similarity calculation on the first apparent feature and each of the second apparent features to obtain the similarity corresponding to each of the plate structure design source models, wherein the second convolutional neural network model includes a plurality of convolutional layers; A geometric parameter determination module is used to determine a target plate structure design model corresponding to the acoustic metamaterial to be designed based on the similarity, and perform feature conversion on the target acoustic performance and environmental data of the acoustic metamaterial to be designed according to the target plate structure design model to obtain target geometric parameters corresponding to the acoustic metamaterial to be designed.

Images