CN115762685B - A sound-permeable layer and its optimization method for reducing the acoustic transmission loss at the interface of dissimilar materials - Google Patents

Abstract

The invention discloses a sound transmission layer for reducing acoustic transmission loss of a dissimilar material interface and an optimization method, and belongs to the field of optimization algorithms. The invention applies the concept of digital materials to disperse the whole optimization space into small material units, thereby parameterizing the whole design space and facilitating the optimization of the whole design space. The method adopts machine learning to obtain the prediction function of the relation between the digital material configuration and the average transmission loss, so that the prediction function is used for replacing finite element analysis in a genetic algorithm, and the calculation time is greatly saved. The method can design a more novel configuration, and is particularly suitable for designing the sound transmission layer in a low-frequency region.

Description

A sound-permeable layer and its optimization method for reducing the acoustic transmission loss at the interface of dissimilar materials

technical field

The invention belongs to the field of applied acoustics, and in particular relates to a sound-permeable layer and an optimization method for reducing the acoustic transmission loss at the interface of different materials.

Background technique

Acoustic transfer loss (STL), in the case of a non-dissipative medium, is mainly caused by the impedance mismatch of the acoustically conducting medium. The sound is reflected at the interface of dissimilar materials with mismatched impedance, resulting in the transmitted sound energy being less than the incident sound energy, and the energy loss caused by reflection is the acoustic transmission loss. Acoustic transmission loss is a critical issue in applications such as the design of acoustic sensors, transducers, and sound-absorbing materials. For example, the transmission loss between the sensor and the measured material will cause the measured sound pressure to be inaccurate; similarly, the energy obtained by the acoustic transducer will be significantly less than the input sound energy; The transmission loss caused by reflection at the interface will reduce the sound absorption coefficient of the material. An effective way to reduce the acoustic transmission loss is to lay an acoustic covering layer, that is, a sound-permeable layer, between two media with mismatched impedances. The sound-permeable layer is usually composed of composite materials or structures with gradual impedance changes, such as wedge-shaped structures, multi-layer structures with gradient impedance changes, and composite materials with gradient particle inclusions. This kind of composite material with gradual impedance makes the impedance transition evenly between the original impedance mismatched materials, thereby reducing the transmission loss, so it is also called the impedance matching layer.

Although these acoustic layers have been intensively studied, the optimal design of the acoustic layers is not very clear. In terms of theoretical analysis, existing theoretical methods, including transfer matrix method and non-uniform transmission line theory, are only applicable to some simple composite materials (for example, multilayer boards with discrete impedance, media with exponential impedance distribution). In terms of numerical calculation, the traditional numerical optimization calculation method is usually limited by human experience, optimization can only optimize the geometric parameters of a specific composite material model, and cannot explore other possible composite material models, so it is impossible to obtain the real (global ) optimal solution. In addition, traditional numerical calculation methods need to consume a lot of computing resources.

Contents of the invention

The problems existing in the prior art are: 1) The traditional optimal design method of the sound-permeable layer to reduce the acoustic transmission loss can only be aimed at specific composite material models, such as wedge-shaped structures and gradient inclusion structures, and cannot explore all possible material composite modes. As a result, the entire optimization space cannot be traversed, so that the global optimal solution cannot be obtained; 2) Traditional numerical optimization calculation methods (such as finite element, topology optimization) consume a lot of time. The purpose of the present invention is to solve the above-mentioned technical problems in the prior art, and provide a sound-permeable layer and an optimization method for reducing the acoustic transmission loss at the interface of different materials.

The concrete technical scheme that the present invention adopts is as follows:

In the first aspect, the present invention provides a method for optimizing a sound-transmitting layer for reducing the acoustic transmission loss of a dissimilar material interface. include:

S1. Taking a single periodic unit in the sound-transmitting layer as the optimization target, discretize the longitudinal section of the periodic unit into a binary matrix, and use different binary element values of the matrix to distinguish between the first medium material and the second medium material, where the first The medium material is the same as the material of the first medium layer or has a similar impedance, and the second medium material is the same as or has a similar impedance to the material of the second medium layer; a binary matrix is generated by randomization, and random sound-transmitting layer digital materials are generated in batches Model, each number in the model corresponds to a material unit, and the specific material of each unit is determined by the element value in the corresponding binary matrix;

S2. For each digital material model of the sound-transmitting layer generated in S1, use finite element simulation to obtain the STL curve of the acoustic transmission loss and the average value of the acoustic transmission loss ATL within the target frequency band;

S3. Convert the binary matrix of each sound-permeable layer digital material model into a binary vector v, take the binary vector v as the input of the training sample, and use the corresponding acoustic transmission loss average value ATL calculated in S2 as the label of the training sample, Build a training data set; use the training data set to carry out machine learning training, and obtain a prediction function for predicting the average ATL of the acoustic transfer loss according to the binary vector v;

S4. Using the training samples in the training data set as the initial population, and using the prediction function as the fitness function, optimize through a genetic algorithm to obtain the optimal binary vector v with the smallest average value of acoustic transfer loss ATL;

S5. Convert the optimal binary vector v into a digital material model of the sound-permeable layer to obtain the optimal configuration, and obtain the STL curve of the acoustic transmission loss and the average value of the acoustic transmission loss ATL within the target frequency band through finite element simulation, and then according to the finite The results of the meta-simulation judge whether the optimal configuration meets the expected optimization requirements. If so, the optimization is completed. If not, the training samples in the training data set are expanded, and then the machine learning training is performed again to generate a new prediction function, and the genetic algorithm is performed again. Optimize until the optimal configuration meets the optimization requirements;

S6. For the optimal configuration meeting the optimization requirements, smooth the boundary between the two dielectric materials to form the final configuration of the sound-transmitting layer.

As a preference of the first aspect above, the second medium layer is water, and the second medium material is hydrogel.

As a preference of the first aspect above, in the above S5, the practice of enlarging the training samples in the training data set is: adding the empirical configuration determined based on human experience in the original randomly generated training data set.

As a preference in the first aspect above, the empirical configuration includes a random gradient inclusion structure, a wedge-shaped gradient structure and a random wedge-shaped structure, each empirical configuration is a digital material model of a sound-transmitting layer, and also has a longitudinal section of a periodic unit The binary matrix formed after discretization;

In the digital material model of the sound-transmitting layer of the random gradient inclusion structure, the material type of each digital is randomly generated, but along the direction from the side where the second medium layer is located to the side where the first medium layer is located, the second medium appears in the digital The probability of the material decreases line by line, and the probability of the first medium material appearing in the digital gradually increases line by line;

In the digital material model of the sound-transmitting layer with a wedge-shaped gradient structure, the first medium material is in the form of a continuous wedge-shaped distribution with a relatively regular outer contour, and along the direction from the side where the second medium layer is located to the side where the first medium layer is located , the wedge width increases row by row.

In the digital material model of the sound-transmitting layer with a random wedge-shaped structure, the first medium material is disjointed as a whole and has an irregular wedge-shaped distribution, and the second medium material randomly appears inside the wedge, and along the side where the second medium layer is located In the direction to the side where the first medium layer is located, the number of digits appearing in the first medium material in a single row of digits increases row by row.

As a preference of the first aspect above, in the stochastic gradient inclusion structure, the probability growth curve of increasing gradient is in linear, polynomial, logarithmic or exponential form.

As a preference of the first aspect above, in the wedge-shaped gradient structure, the contour lines on both sides of the wedge-shaped distribution of the first dielectric material are straight lines, polynomial curves, logarithmic curves or exponential curves.

As a preference of the first aspect above, the machine learning model used in the machine learning training is a backpropagation artificial neural network, support vector machine, convolutional neural network, and linear regression.

As a preference of the first aspect above, the machine learning model adopts BP neural network.

As a preference of the first aspect above, the initial population is a part of training samples whose average value of acoustic transfer loss in the training data set is lower than a threshold.

As a preference of the first aspect above, the range of the target frequency band is 0-10 kHz.

In a second aspect, the present invention provides a sound-transmitting layer optimized according to the sound-transmitting layer optimization method described in any one of the above-mentioned first aspects, the sound-transmitting layer is the first medium layer and the second medium located at impedance mismatches A double-layer structure between layers, wherein the side in contact with the second dielectric layer is an impedance jumping layer, and the side in contact with the first dielectric layer is an intercalation layer, wherein the impedance jumping layer is completely composed of the first dielectric material; and the inclusion The main material of the layer is the second dielectric material, and there is a distribution area of the first dielectric material inside the main material, but the distribution area of the first dielectric material in the interlayer is completely surrounded by the second dielectric material.

Compared with the prior art, the present invention has the following beneficial effects:

1) The traditional optimization method can only optimize a small number of geometric parameters for a fixed configuration, but the method of the present invention uses the concept of digital materials to discretize the entire optimization space into small material units, thereby parameterizing the entire design space , which facilitates the optimization of the entire design space.

2) Compared with the traditional finite element calculation method combined with the genetic algorithm optimization method to traverse all possible digital material configurations, the method of the present invention first uses machine learning to obtain the prediction function of the relationship between the digital material configuration and the average transmission loss, so as to facilitate In the genetic algorithm, the prediction function is used instead of the finite element analysis, which greatly saves the calculation time.

3) More novel configurations can be designed by the method of the present invention, especially in the low frequency range, such as the design of the sound-permeable layer between water and other media, and a complex structure of impedance jumping is obtained in the (0-10kHz) range Type, compared with the traditional impedance matching configuration, this configuration reduces the average acoustic transmission loss by 31%.

Description of drawings

Fig. 1 is a schematic diagram of the generation method of the digital material model of the sound-permeable layer;

Fig. 2 is a schematic diagram of the finite element analysis method of the acoustic transmission loss and the calculation method of the average acoustic transmission loss;

Fig. 3 is a schematic diagram of artificial neural network;

Fig. 4 is a schematic diagram of genetic algorithm;

Figure 5 shows the generation effects of four different types of digital material configurations. The former is a completely random binary configuration, and the latter three are empirical configurations randomly generated based on human experience and using features that can reduce the STL value;

Fig. 6 is a flow chart of a sound-permeable layer optimization design method in an embodiment of the present invention;

Fig. 7 shows the configuration of the sound-permeable layer obtained by the optimized design in the embodiment of the present invention.

Detailed ways

The present invention will be further elaborated and illustrated below in conjunction with the accompanying drawings and specific embodiments. The technical features of the various implementations in the present invention can be combined accordingly on the premise that there is no conflict with each other.

In order to make the above objects, features and advantages of the present invention more comprehensible, specific implementations of the present invention will be described in detail below in conjunction with the accompanying drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described here, and those skilled in the art can make similar improvements without departing from the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below. The technical features in the various embodiments of the present invention can be combined accordingly on the premise that there is no mutual conflict.

In the description of the present invention, it will be understood that when an element is referred to as being "connected" to another element, it can be directly connected to the other element or indirectly, ie, intervening elements may be present. In contrast, when an element is referred to as being "directly" connected to another element, there are no intervening elements present.

In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinction and description, and cannot be interpreted as indicating or implying relative importance or implicitly indicating the quantity of indicated technical features . Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of these features.

The present invention provides a method for optimizing the sound transmission layer for reducing the acoustic transmission loss at the interface of dissimilar materials. The sound transmission layer is arranged between the first dielectric layer A and the second dielectric layer B whose impedances do not match. Due to the difference in impedance between the first dielectric layer A and the second dielectric layer B, the sound is reflected at the interface of dissimilar materials with mismatched impedance, resulting in the transmitted sound energy being less than the incident sound energy, and the energy loss caused by reflection is the acoustic Transmission loss. The purpose of setting the sound-permeable layer in the present invention is to make the impedance transition between the materials whose original impedances do not match, which can reduce the acoustic transmission loss, and reduce the transmission loss.

It should be noted that, since the sound-transmitting layer is generally continuous in a plane, and the structure therein exhibits periodic changes, it is not necessary to optimize the entire sound-transmitting layer in the present invention, but only need to extract the periodically changing units for optimization. . Since the configuration (also called structure) change of the sound-transmitting layer is mainly reflected in the longitudinal section, the present invention can optimize the configuration of the longitudinal section of a single periodic unit in the sound-transmitting layer.

In a preferred embodiment of the present invention, the method for optimizing the acoustic transmission layer for reducing the acoustic transmission loss at the interface of dissimilar materials includes the following steps:

S1. Taking a single periodic unit in the sound-transmitting layer as the optimization target, discretize the longitudinal section of the periodic unit into a binary matrix, and use different binary element values of the matrix to distinguish between the first medium material and the second medium material, where the first The medium material is the same as the material of the first medium layer or has a similar impedance, and the second medium material is the same as or has a similar impedance to the material of the second medium layer; a binary matrix is generated by randomization, and random sound-transmitting layer digital materials are generated in batches Each number in the model corresponds to a material unit, and the specific material of each unit is determined by the element value in the corresponding binary matrix.

It should be noted that the above binary matrix is constructed based on the physical model of the material sound transmission problem. As shown in Figure 1, the sound transmission layer is located between the medium material A and the medium material B, and the longitudinal section of a periodic unit of the sound transmission layer It can be discretized into a 0-1 binary matrix. The value of each matrix element in the binary matrix is a binary value, namely 0 or 1. In the present invention, 1 can be used to represent the first dielectric material, 0 can be used to represent the second dielectric material, and 0 can also be used to represent the first dielectric material. 1 represents the second dielectric material, which is not limited. Based on the binary matrix, the elements in the matrix can be correspondingly mapped according to their binary element values to form a unit filled with a specified medium material in the sound-transmitting layer. The present invention will refer to this unit as a digital, thus forming a numerical simulation. Digital material model of the acoustic layer. If the thickness of the sound-transmitting layer is H, and the binary matrix is a matrix of order n×m, then the height of a digit in the digital material model of the sound-transmitting layer is H/m.

It should be noted that the first dielectric material is preferably completely the same as the material of the first dielectric layer, and the same second dielectric material is preferably the same as the material of the second dielectric layer, but if the material impedance difference between the two is not large, Two different materials can then also be used. For example, if the second dielectric layer is water, since water cannot be used to construct the layer due to its liquid state, the second dielectric material can be replaced by hydrogel, because the impedance difference between hydrogel and water is very small. The material of the first medium also needs to be determined according to the material of the first medium layer to be actually simulated, and rubber, polyurethane (PU) and the like can be used.

S2. For each digital material model of the sound-transmitting layer generated in S1, use finite element simulation to obtain an acoustic transmission loss (STL) curve and an average acoustic transmission loss (ATL) within the target frequency band.

It should be noted that the calculation of the acoustic transmission loss (STL) curve within the target frequency band through finite element simulation belongs to the prior art. As shown in Figure 2, the flow of the finite element analysis method of the acoustic transmission loss and the calculation method of the average acoustic transmission loss is shown, and the construction of the finite element model is a key step. When building a finite element model based on the physical model of the sound transmission problem, it is necessary to set up a hard sound field boundary, a low reflection boundary, and then set a symmetrical boundary on both sides, and set a perfect matching layer from top to bottom within the range of the four boundaries. The B layer of the second dielectric material, the pressure acoustics and solid coupling boundary, the sound-permeable layer, and the A layer of the first dielectric material. The acoustic layer can be constructed by importing the aforementioned digital material model of the acoustic layer. Within the frequency band of interest, that is, within the target frequency band, the incident process of sound waves can be simulated through finite element simulation to obtain the STL curve composed of the STL values at different channels, and the STL values corresponding to all channels within the target frequency band By averaging, the ATL value can be obtained. For example, in Figure 2, the target frequency range of interest is the low-frequency range of 0 to 10 kHz, so this range is divided into 10 frequency bands, and the corresponding ATL value is obtained by averaging the STL values of the 10 frequency bands.

For each digital material model of the sound-transmitting layer mentioned above, the corresponding ATL value can be obtained through finite element simulation, and thus can be used as a sample for machine learning. Therefore, when generating digital material models of sound-transmitting layers in batches, the quantity needs to ensure that machine learning has enough samples for training.

S3. Convert the binary matrix of each sound-permeable layer digital material model into a binary vector v, take the binary vector v as the input of the training sample, and use the corresponding acoustic transmission loss average value ATL calculated in S2 as the label of the training sample, Construct a training data set; use the training data set to carry out machine learning training, and obtain a prediction function for predicting the average ATL of the acoustic transfer loss according to the binary vector v.

It should be noted that the conversion of binary matrix to binary vector is mainly for the convenience of inputting into machine learning models. The machine learning model that can be used in the present invention is not limited, and the backpropagation artificial neural network, support vector machine, convolutional neural network, linear regression, etc. can all be used as the machine learning model in the present invention. The training of the machine learning model belongs to the prior art, and no further description will be given here.

As shown in Figure 3, it shows a schematic diagram of constructing a prediction function using a machine learning model in the form of a BP neural network in an embodiment. The dimension of the input layer is the dimension of the binary vector v, which is 1*400, and the hidden layer has three layers. , the dimensions are 10-5-5, and the output layer dimension is 1, which is the ATL value. Therefore, the model obtained after the training of the machine learning model can be regarded as a prediction function f for predicting the average ATL of the acoustic transfer loss according to the binary vector v, that is, ATL=f(v).

S4. Using the training samples in the training data set as the initial population, and using the prediction function ATL=f(v) as the fitness function, optimize through the genetic algorithm to obtain the optimal binary vector v with the smallest average ATL of the acoustic transmission loss.

The genetic algorithm (Genetic Algorithm, referred to as GA) in the present invention is an adaptive random search heuristic algorithm, which is widely used in complex function system optimization, machine learning, system identification, fault diagnosis, classification system and other fields. As shown in Figure 4, the GA algorithm uses computer simulation operations to transform the problem-solving process into a process similar to the crossover and mutation of chromosomal genes in biological evolution. Individuals are screened based on fitness values to ensure that individuals with good fitness values have the opportunity to produce more sub-individuals in the next generation. Therefore, when using GA to solve specific problems, the choice of fitness value function has a great influence on the convergence and convergence speed of the algorithm. In the present invention, the above prediction function ATL=f(v) is recorded as the fitness function in the genetic algorithm optimization process. The final optimization goal is to minimize the fitness function, that is, to ensure the minimum average value of acoustic transmission loss ATL. After the above-mentioned optimal binary vector v is obtained through optimization, the binary vector v can be converted into a binary matrix form again, and then the optimal sound-transmitting layer digital material model can be formed.

In order to speed up the convergence of the genetic algorithm, when selecting the initial population, the training samples in the training data set can be screened to remove some samples with high ATL values and retain some samples with small ATL values.

S5. Convert the above-mentioned optimal binary vector v into the digital material model of the sound-permeable layer to obtain the optimal configuration, and obtain the acoustic transmission loss STL curve and the average value of the acoustic transmission loss ATL within the target frequency band through finite element simulation, and then according to the finite element The simulation results judge whether the optimal configuration meets the expected optimization requirements, and if so, the optimization is completed; if not, the training samples in the training data set are expanded, and then the machine learning training is performed again to generate a new prediction function, and the genetic algorithm is optimized again , until the optimal configuration meets the optimization requirements.

It should be noted that the finite element simulation in this step is the same as that in S2 above, and will not be repeated here. Its purpose is to judge whether the optimal configuration meets the expected optimization requirements. The expected optimization requirements here need to be determined according to the actual situation. The STL curve that the sound-transmitting layer is expected to finally meet and the ATL value that needs to be satisfied can be preset in advance. The specific value is not limited here.

However, if it is found after the finite element simulation that the optimal configuration cannot meet the expected optimization requirements, it may be that the number of samples of the digital material model of the sound-transmitting layer generated in the initial batch is not enough, and the optimization process has fallen into a local optimal solution, and there is no Find the global optimal solution. Therefore, in this case, it is necessary to expand the training samples in the training data set, and then perform machine learning training again to generate a new prediction function f, and perform the genetic algorithm optimization of S4 and S5 again, and then judge whether it meets expectations again. This process is repeated continuously until the optimal configuration meets the expected optimization requirements.

It should be noted that when expanding the training samples in the training data set, the method of random generation can continue to be used to generate the digital material model of the sound-transmitting layer and obtain the labels through finite element simulation. The corresponding binary matrix is randomly generated, and may not necessarily conform to the configuration form required to reduce ATL.

Therefore, as a preferred method of the embodiment of the present invention, when expanding the training samples in the training data set, the method adopted is not to randomly generate, but to generate some empirical configurations in advance based on the experience summarized by humans, and then perform these empirical configurations. After the ATL value is obtained by finite element simulation, it is added to the original randomly generated training data set in the form of a sample. The empirical configuration form added here is not limited, and can be selected according to relevant research or actual experiments in the prior art. In this example, three empirical configurations are given, including stochastic gradient inclusion structure, wedge-shaped gradient structure and random wedge-shaped structure. The binary matrix formed after discretization of the profile, the dimension of the binary matrix is the same as that of the randomly generated binary matrix. The specific forms of these three empirical configurations are described in detail below.

As shown in the first picture in Figure 5, compared with the completely randomly generated digital material model of the sound-transmitting layer, since its binary matrix is randomly generated, the type of medium material in each digital is also disorderly.

As shown in the second picture in Figure 5, in the digital material model of the sound-permeable layer with random gradient inclusion structure, the material type of each digital is randomly generated, that is, each digital is randomly set to be the first medium material or the second medium However, macroscopically speaking, along the direction from the side where the second medium layer is to the side where the first medium layer is located, the probability of the second medium material appearing in the digits decreases row by row, and the probability of the first medium material appearing in the digits gradually decreases. Row gradient increments. From a macro point of view, along the direction from the side where the second medium layer is located to the side where the first medium layer is located, the probability of the second medium material appearing in the numbers decreases line by line, and the number of the first medium material numbers appearing in each line of numbers Constantly increasing, but the location of its occurrence is uncertain. In this stochastic gradient inclusion structure, since the probability of the first dielectric material appearing in the digital needs to increase step by row, it is necessary to set a corresponding probability growth curve to control the probability of the first dielectric material appearing in each row in actual implementation. Optional The probability growth curve of is in linear, polynomial, logarithmic or exponential form, which is not limited, and different models can be generated as richly as possible to facilitate global optimization.

As shown in the third picture in Figure 5, in the digital material model of the sound-transmitting layer with a wedge-shaped gradient structure, the first medium material is in a continuous piece with a relatively regular wedge-shaped distribution on the outside, and along the side where the second medium layer is located In the direction to the side where the first dielectric layer is located, the width of the wedge increases row by row. Compared with the previous two configurations, the distribution of dielectric materials in this configuration is no longer chaotic, but relatively regular. In the wedge-shaped gradient structure, the contour lines on both sides of the wedge-shaped distribution of the first dielectric material can be straight lines, polynomial curves, logarithmic curves or exponential curves specified based on experience, which is not limited, and can be generated as richly as possible. model for global optimization.

As shown in the fourth picture in Figure 5, in the digital material model of the sound-transmitting layer with a random wedge-shaped structure, the first medium material is distributed in a wedge-shaped discontinuous and irregular outer contour as a whole, and the second medium material randomly appears inside the wedge , and along the direction from the side where the second medium layer is located to the side where the first medium layer is located, the number of numbers appearing in the first medium material in a single line of numbers increases line by line. Compared with the previous wedge-shaped gradient structure, the difference of this random wedge-shaped structure is that the outer contour is no longer specified according to experience, but randomly generated, but the overall shape is still wedge-shaped, but because it is randomly generated, the inside of the wedge-shaped area is also A second dielectric material may appear where the wedge-shaped distribution of the first dielectric material is no longer completely continuous.

S6. For the optimal configuration meeting the optimization requirements, smooth the boundary between the two dielectric materials to form the final configuration of the sound-transmitting layer.

It should be noted that the purpose of boundary smoothing in this step is to consider the technical requirements of actual material processing, because material boundaries with too discrete boundaries are not achievable in actual processing. After smoothing the boundary between the two dielectric materials, the boundary can be guaranteed to be relatively smooth, which is beneficial to the processing of the sound-transmitting layer.

In addition, after the final configuration of the sound-transmitting layer is obtained, before the actual processing of the sound-transmitting layer, the final configuration can also be simplified to a certain extent to remove some structural features that do not meet the requirements of the processing technology, or further optimize The features in the obtained final configuration are subjected to mechanism research to confirm which structural features are key features and which are non-key features, so as to further adjust the actual processing configuration.

In the following, the method for optimizing the sound transmission layer for reducing the acoustic transmission loss at the interface of dissimilar materials shown in S1-S6 above is applied to a specific example to demonstrate its specific implementation process and technical effect.

Example

In this embodiment, the process of optimizing the sound transmission layer for reducing the acoustic transmission loss at the interface of different materials is shown in Figure 6, and the specific steps are as follows:

Step 1: For two media A and B with mismatched impedance and a given thickness H of the sound-transmitting layer, generate random digital material models of the sound-transmitting layer in batches based on the concept of digital materials, and the sound-transmitting layer is the first medium material A A composite material composed of a second dielectric material B. In this embodiment, the media on both sides of the sound-transmitting layer are water and rubber respectively, so the media materials constituting the sound-transmitting layer are respectively selected as acrylamide hydrogel and polyurethane (PU). The generation method is as follows: First, randomly generate a series of n×m-order binary matrices M, 1 and 0 in the matrix represent dielectric material A and dielectric material B respectively; then, according to the According to the corresponding relationship, a digital material model of the sound-permeable layer is generated, where each number represents a material unit, and the height of each material unit is H/m.

Step 2: For the digital material configuration of each sound-transmitting layer generated in batches, use finite element software to simulate and calculate the acoustic transmission loss (STL) in the frequency band concerned, and calculate the average value of the acoustic transmission loss in the frequency band ATL. In this embodiment, the concerned frequency range is the low frequency range of 0-10 kHz, which is divided into 10 frequency bands, and the corresponding ATL value can be obtained by averaging the STL values of the 10 frequency bands on the STL curve.

The third step: convert the binary matrix of each sound-transmitting layer digital material model into a binary vector v, together with the calculated ATL value, construct a training sample, and use all training samples as a training data set to train the BP neural network, thereby The trained model is obtained as a prediction function between the binary vector v and the ATL value, ie ATL=f(v).

Step 4: Screen out a part of the binary vector v with a lower ATL value from the training data set, and use the corresponding training sample as the initial population. By setting the appropriate crossover rate, mutation rate and convergence criterion, ATL=f(v) The genetic algorithm is optimized for the fitness function, and the lowest ATL prediction value and the corresponding binary vector v are obtained. In this embodiment, the genetic algorithm is implemented directly using the optimization toolkit in MATLAB, the solver is ga-Genetic Algorithm, and the crossover probability and mutation probability in the genetic operator are set to 0.8 and 0.01, respectively.

Step 5: Convert the binary vector v of the lowest ATL prediction value obtained by the genetic algorithm into a digital material configuration to obtain the optimal configuration, and use the finite element method to calculate the acoustic transmission loss STL curve of the optimal configuration, and calculate Mean ATL. Then judge whether the STL curve and ATL value of the optimal configuration meet the optimization expectations. If yes, the optimization ends. Otherwise, the training data set needs to be expanded to add some configurations generated based on human experience. Compared with the completely random configuration, the empirical configuration has a wider range of ATL values, includes more features, and is conducive to machine learning to obtain a more accurate prediction model. Then repeat the finite element simulation of the second step and the optimization process of the third to fifth steps.

The empirical configuration adopted in this embodiment includes the stochastic gradient inclusion structure, wedge-shaped gradient structure and random wedge-shape structure shown in Fig. It can be in the form of linear, polynomial, logarithmic, exponential, etc.; in the wedge-shaped gradient structure, wedge-shaped structures with different gradients can be used, including linear, parabolic, exponential, etc., and the parameters of curvature changes can be randomly generated; in addition, random wedge In the configuration material, the outer contour of the wedge block is randomly generated, and the inner part has a probability to generate vacancies in the thickness direction.

Step 7: After obtaining the optimal configuration that meets expectations, smooth the edges of the material interface on the optimal configuration to form the final configuration.

As shown in Figure 7, the left figure shows the final configuration of the sound-transmitting layer optimized by the above-mentioned sound-transmitting layer optimization method in this embodiment. The sound-transmitting layer is located between the first dielectric layer and the second dielectric layer with mismatched impedances. A double-layer structure in between, wherein the side in contact with the second dielectric layer is an impedance jumping layer, and the side in contact with the first dielectric layer is an intercalation layer, wherein the impedance jumping layer is completely composed of the first dielectric material A; and the intercalation layer The second dielectric material B is used as the main material, and there is a distribution area of the first dielectric material A inside the main material, but the distribution area of the first dielectric material A in the interlayer is completely surrounded by the second dielectric material B.

It can be seen that the configuration optimized by the optimization method of the present invention is not consistent with the configuration obtained by artificial experience, and instead forms a layer of impedance jump layer on the side in contact with the second dielectric layer, which violates the artificial The concept of impedance gradients needs to be guaranteed empirically. Compared with the traditional impedance matching configuration, this novel complex configuration of impedance jumping has better performance, especially in the low frequency range (0-10kHz). Compared with the traditional matching configuration of impedance gradient , and its average acoustic transmission loss was further reduced by 31%.

In addition, in this embodiment, the distribution area of the material B in the interlayer in the final configuration after the above-mentioned edge smoothing is approximately elliptical. Based on the final configuration, structural feature analysis can be carried out to extract the main structural features, filter the secondary structural features, and then simplify the main structural features, and verify the simplified optimal configuration through finite elements, so as to obtain final configuration. For example, as shown in the right figure in Fig. 7, the elliptical material B distribution area can be simplified into a circle, and the impedance jump layer can be smoothly formed into a smooth layer body. By parameterizing various parameters, including the overall sound transmission layer Thickness D, width W, impedance jump layer thickness D ₂ , radius d of material B area, distance h between the bottom of material B area and the bottom of the sound-transmitting layer, and the ratio of W, D ₂ , d, h to D w _r , t _r , d _r , _hr , you can continue to explore these parameters, judge the mechanism of these parameters affecting the final performance, and whether there is a possibility of further optimization of this configuration.

The above-mentioned embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. Various changes and modifications can be made by those skilled in the relevant technical fields without departing from the spirit and scope of the present invention. Therefore, all technical solutions obtained by means of equivalent replacement or equivalent transformation fall within the protection scope of the present invention.

Claims (10)

1. The method for optimizing the sound transmission layer for reducing the acoustic transmission loss of the dissimilar material interface is characterized by comprising the following steps of:

s1, discretizing a longitudinal section of a periodic unit into a binary matrix by taking a single periodic unit in an acoustic transmission layer as an optimization target, and respectively representing a first dielectric material and a second dielectric material by different binary element values of the matrix, wherein the first dielectric material and the material of the first dielectric layer have the same impedance, and the second dielectric material and the material of the second dielectric layer have the same impedance; generating a binary matrix through randomization, generating a random sound-transmitting layer number code material model in batches, wherein each number in the model corresponds to a material unit, and the specific material of each unit is determined by the element value in the corresponding binary matrix;

s2, obtaining an acoustic transmission loss curve and an acoustic transmission loss average value in a target frequency band range by utilizing finite element simulation for each sound transmission layer number code material model generated in the S1;

s3, converting a binary matrix of each sound transmission layer number code material model into a binary vector, taking the binary vector as input of a training sample, and taking the corresponding acoustic transmission loss average value calculated in the S2 as a label of the training sample to construct a training data set; performing machine learning training by using a training data set to obtain a prediction function for predicting an acoustic transmission loss average value according to the binary vector;

s4, taking training samples in the training data set as an initial population, taking the prediction function as an fitness function, and optimizing through a genetic algorithm to obtain an optimal binary vector with the minimum acoustic transmission loss average value;

s5, converting the optimal binary vector into a sound transmission layer number material model to obtain an optimal configuration, obtaining an acoustic transmission loss curve and an acoustic transmission loss average value in a target frequency band range through finite element simulation, judging whether the optimal configuration meets expected optimization requirements according to finite element simulation results, if so, completing optimization, if not, expanding training samples in a training data set, then performing machine learning training again to generate a new prediction function, and performing genetic algorithm optimization again until the optimal configuration meets the optimization requirements;

s6, smoothing the boundary between the two dielectric materials aiming at the optimal configuration meeting the optimization requirement to form the final configuration of the sound-transmitting layer.

2. The method for optimizing an acoustic transmission layer for reducing interface acoustic transmission loss of a dissimilar material according to claim 1, wherein the second dielectric layer is water, and the second dielectric material is hydrogel.

3. The method for optimizing a sound-transmitting layer for reducing acoustic transmission loss of a dissimilar material interface according to claim 1, wherein in S5, the method for expanding training samples in a training data set is as follows: an empirical configuration determined from an artificial summary experience is added to the original randomly generated training dataset.

4. The method for optimizing a sound-transmitting layer for reducing acoustic transmission loss of a dissimilar material interface according to claim 3, wherein the empirical configurations comprise a random gradient inclusion structure, a wedge-shaped gradient structure and a random wedge-shaped structure, each empirical configuration is a sound-transmitting layer code material model and also comprises a binary matrix formed by discretizing a longitudinal section of a periodic unit;

in the sound transmission layer number code material model of the random gradient inclusion structure, the material type of each code is randomly generated, but the probability of the second medium material in the code is gradually decreased line by line along the direction from the side of the second medium layer to the side of the first medium layer, and the probability of the first medium material in the code is gradually increased line by line gradient;

in the sound-transmitting layer number material model of the wedge-shaped gradient structure, the whole first dielectric material is in continuous sheet and has regular wedge-shaped distribution of outer outline, and the wedge-shaped width gradually increases along the direction from the side of the second dielectric layer to the side of the first dielectric layer;

in the sound transmission layer number code material model of the random wedge-shaped structure, the whole first dielectric material is in wedge-shaped distribution with non-connected pieces and irregular outer contours, the second dielectric material randomly appears in the wedge-shaped interior, and the number of the numbers of the first dielectric material appearing in a single line number is gradually increased line by line along the direction from the side of the second dielectric layer to the side of the first dielectric layer.

5. The method for optimizing an acoustically transparent layer for reducing interfacial acoustic transmission loss of a dissimilar material according to claim 4, wherein said probability growth curve of gradient increment in said random gradient inclusion structure is in the form of a linear, polynomial, logarithmic or exponential.

6. The method for optimizing a sound-transmitting layer for reducing acoustic transmission loss of a dissimilar material interface according to claim 4, wherein contour lines on two sides of the first dielectric material distributed in a wedge-shaped gradient structure are straight lines, polynomial curves, logarithmic curves or exponential curves.

7. The method for optimizing a sound transmission layer for reducing acoustic transmission loss of a dissimilar material interface according to claim 1, wherein a machine learning model adopted in the machine learning training is a counter-propagating artificial neural network, a support vector machine, a convolutional neural network or a linear regression.

8. The method for optimizing an acoustic transmission layer for reducing acoustic transmission loss at a dissimilar material interface of claim 7, wherein said initial population is a portion of training samples in a training dataset having an average value of acoustic transmission loss below a threshold value.

9. The method for optimizing an acoustic transmission layer for reducing acoustic transmission loss of a dissimilar material interface according to claim 1, wherein the target frequency band range is 0-10 khz.

10. The sound-transmitting layer optimized by the sound-transmitting layer optimizing method according to any one of claims 1 to 9, wherein the sound-transmitting layer has a double-layer structure between a first dielectric layer and a second dielectric layer with unmatched impedance, wherein the side contacting the second dielectric layer is an impedance jump layer, and the side contacting the first dielectric layer is a sandwich layer, wherein the impedance jump layer is entirely made of a first dielectric material; the second dielectric material is used as a main material of the sandwich layer, a first dielectric material distribution area is arranged in the main material, and the first dielectric material distribution area in the sandwich layer is completely surrounded and wrapped by the second dielectric material.