An optimization method for broadband sound absorption structure of stereolithography based on machine learning

Abstract

The present invention discloses an optimization method for a broadband sound-absorbing structure based on machine learning. The steps are as follows: constructing a broadband sound-absorbing structure, including a neck porous structure and an internal cavity structure. In the neck porous structure, the first and second porous structures are symmetric about the center axis of the sound-absorbing structure and are not connected, with a sound wave incident channel in the middle, and cover plates are respectively arranged on both sides; the internal cavity structure is composed of the first, second, and third embedded partitions and a resonant cavity; using a commercial finite element solver to simulate and calculate the broadband sound-absorbing structure, and calculating the sound-absorbing performance of the broadband sound-absorbing structure; constructing a machine learning parameter optimization model, and optimizing the parameters of the sound-absorbing structure through a BP neural network and a genetic algorithm to obtain the optimal sound-absorbing structure parameters. The method of the present invention can achieve efficient broadband noise suppression, significantly improve the sound-absorbing performance, increase the effective bandwidth, and at the same time effectively reduce the design time and economic cost.

Description

Technical Field

[0001] The present invention relates to the field of machine learning and artificial intelligence technology, and in particular to a method for optimizing a photocuring broadband sound absorption structure based on machine learning. Background Art

[0002] With the acceleration of urbanization and the expansion of industrial production, noise pollution is becoming increasingly serious. The sound frequency range of industrial noise such as punching machines, grinders, compressors, etc. is from tens of hertz at low frequencies to thousands of hertz at high frequencies. The effective bandwidth is wide, and general low-frequency noise reduction is difficult to achieve a comprehensive sound absorption effect. Therefore, it is crucial to effectively absorb sound waves in a wide frequency range.

[0003] In recent years, common acoustic metamaterial sound-absorbing structures include Helmholtz resonant cavities, micro-perforated plate structures, and porous material structures, but they are difficult to effectively control wide-frequency noise when acting alone. Although some resonant sound-absorbing structures perform well in a wide frequency band, their actual effective sound absorption band is narrow, and there are problems with high time and economic costs in structural parameter design. Combining different sound-absorbing materials or structures to form a composite sound-absorbing structure, such as combining porous materials with Helmholtz resonators or micro-perforated plates, can give full play to their respective advantages and achieve sound absorption in a wider frequency range.

[0004] The noise reduction performance of sound-absorbing structures is strongly coupled to their geometric parameter design. Traditional optimization methods rely on manual trial and error and empirical adjustments, which are prone to local optimality and low computational efficiency when dealing with multivariable and nonlinear optimization problems. The data-driven method based on machine learning provides a new technical path to break through the global optimization bottleneck of complex acoustic systems through intelligent search strategies.

[0005] In addition, with the development of new manufacturing technologies such as 3D printing, new opportunities have emerged in the manufacture and application of acoustic metamaterials. Additive manufacturing methods based on stereolithography have been proven to be a reliable means to accurately manufacture acoustic metamaterials with complex geometric structures.

[0006] In summary, there is an urgent need for an efficient and accurate optimization method that can achieve ideal sound absorption effects within a larger frequency range, and at the same time, combined with advanced manufacturing technology, transform the research results on photocuring broadband sound absorption structures into practical applications. Summary of the invention

[0007] The problem to be solved by the present invention is to provide a method for optimizing a broadband sound absorption structure of a photocuring molding based on machine learning, which is used to achieve efficient broadband noise suppression, improve sound absorption performance, increase effective bandwidth, and reduce design time and economic cost.

[0008] The present invention adopts the following technical solution: An optimization method for a broadband sound-absorbing structure based on machine learning, comprising the following steps:

[0009] Step 1: Construct a broadband sound-absorbing structure, including at least one sound-absorbing unit. The sound-absorbing unit is a cylindrical cavity, including: a neck porous structure and an internal cavity structure. The neck porous structure is arranged at one end of the cylindrical cavity and consists of a first porous structure, a second porous structure, and a sound wave incident channel. The first porous structure and the second porous structure are not connected and are symmetric about the center axis of the sound-absorbing unit, with a sound wave incident channel in the middle. A first cover plate and a second cover plate are respectively arranged on both sides. The internal cavity structure consists of a first embedded partition, a second embedded partition, a third embedded partition, and a resonant cavity;

[0010] Step 2: Use a commercial finite element solver to simulate and calculate the broadband sound-absorbing structure: Regard the neck porous structure and the embedded partition structure as key narrow areas, and select a thermoviscous acoustic interface; Use the five-parameter JCA model to fix the porous medium parameters to describe the propagation behavior of sound waves in porous materials; Use a plane wave with an amplitude of 1 Pa as the excitation source to simulate incident harmonics, and calculate the sound absorption performance of the broadband sound-absorbing structure from 400 Hz to 1400 Hz;

[0011] Step 3: Construct a machine learning parameter optimization model, and optimize the parameters of the sound-absorbing unit through a BP neural network and a genetic algorithm; Use the sound absorption performance predicted by the BP neural network as the fitness value, and select the individual with the highest fitness through the genetic algorithm for further optimization to obtain the optimal sound-absorbing structure parameters;

[0012] Step 4: According to the optimized sound-absorbing structure parameters, by adjusting the thickness and porosity of the first porous structure and the second porous structure; and adjusting the distances of the first embedded partition, the second embedded partition, and the third embedded partition from the sound wave incident surface , , , adjust the acoustic impedance of the sound-absorbing unit to optimize the sound absorption performance of the broadband sound-absorbing structure.

[0013] Preferably, in Step 1, the first porous structure and the second porous structure are the same, and the bottom surface is an arc trapezoid; the first cover plate and the second cover plate are symmetric about the center axis of the sound-absorbing unit, are respectively arranged on both sides of the first porous structure and the second porous structure, and the first cover plate and the second cover plate do not cover the first porous structure and the second porous structure.

[0014] Preferably, the first embedded partition, the second embedded partition, and the third embedded partition are all semi-cylindrical, and the partitions have the same size and thickness, and are successively away from the sound wave incident channel, and the distance from the sound wave incident channel is not fixed;

[0015] The sound wave enters the sound absorption unit through the sound wave incident channel, bypasses the first embedded partition, the second embedded partition, and the third embedded partition in sequence, and exits at the sound wave exit surface at the other end of the cylindrical cavity.

[0016] Preferably, the sound absorption coefficient of the sound absorption unit , is calculated as follows:

[0017] ;

[0018] Among them, the reflection coefficient R is defined as , is the acoustic impedance of the sound absorption structure, is the characteristic impedance of air, is the density of air, is the speed of sound in air.

[0019] Preferably, the acoustic impedance of the sound absorption unit The specific expression is as follows:

[0020] ;

[0021] Among them, is the impedance of the embedded partition, is the impedance of the porous structure, is the area of the sound wave incident channel.

[0022] Preferably, a commercial finite element solver is used to simulate and calculate the broadband sound absorption structure, including the following sub-steps:

[0023] Step 2.1: Use the commercial finite element solver COMSOL Multiphysics to calculate the broadband sound absorption structure. The sound absorption unit is simulated using the pressure acoustics-frequency domain module, and the air domain range is defined by setting a hard boundary;

[0024] Step 2.2: Regard the neck porous structure and the embedded partition structure as key narrow areas, and select the thermoviscous acoustics interface; use the five-parameter JCA model to fix the porous medium parameters such as porosity, flow resistivity, and thermal conductivity to describe the propagation behavior of sound waves in porous materials; construct the geometry through the assembly module, and construct the required inlays through operations such as taking the difference set, stretching, and splitting; use finite element meshes, and to ensure accurate simulation accuracy, the mesh size is selected to be refined, and the range is for all entities; use a plane wave with an amplitude of 1 Pa as the excitation source to simulate the incident harmonic;

[0025] Step 2.3: After the plane-wave sound source enters from the porous structure at the neck on one side of the cavity, it continues to propagate in the resonant cavity and forms different sound propagation channels after passing through the embedded partition. The geometric dimensions of each sub-cavity correspond to different resonance frequencies, and the multi-level resonance peaks are superimposed to cover a wide frequency range, achieving wide-band sound absorption. During the simulation calculation process, parametric scanning is adopted, and the frequency points are input through an explicit list: the effective frequency range is limited to 400 Hz to 1400 Hz, the step size is 50 Hz, and there are 21 frequency points in total. The sound absorption coefficients at different frequency points are calculated respectively to adjust the good wide-band sound absorption effect.

[0026] Preferably, the BP neural network is constructed through three fully connected layers, and two hidden layers are set in the middle. The specific method is as follows:

[0027] Step 3.1.1: Data input: The data of the input layer includes the distances of the first embedded partition, the second embedded partition, and the third embedded partition from the sound wave incident surface. 、 、 ;

[0028] Step 3.1.2: Data fitting: Each hidden layer contains 64 neurons, and a ReLU activation function and a Dropout layer are also set after each hidden layer. The Dropout probability is 0.3.

[0029] Step 3.1.3: Data output: The output layer outputs 21 values, representing the sound absorption coefficients in the frequency band of 400 Hz to 1400 Hz. ;

[0030] Step 3.1.4: Network model training: Use Z-score for data preprocessing, use Huber Loss as the loss function, and adopt the Adam optimizer for training; the learning rate of the Adam optimizer is set to 0.001, and L2 regularization is applied during the optimization process, and the regularization coefficient is 1e-5.

[0031] Step 3.1.5: Network model verification: Use Backpropagation to calculate the gradient. The KFold cross-validation method is adopted during the verification process. Set the random seed, divide the data set into 5 subsets of equal size, 4 subsets are used to train the model, and 1 subset is used to verify the model to obtain the sound absorption performance parameters predicted by the optimized BP neural network.

[0032] Preferably, the genetic algorithm is used to optimize the geometric parameters of the sound absorption structure. The specific processing is as follows:

[0033] Step 3.2.1: Set the initial population size and the number of genetic iterations, perform crossover on individuals using the two-point crossover operation, and set the crossover probability.

[0034] Step 3.2.2: Mutate the individuals using uniform mutation operation and set the mutation probability;

[0035] Step 3.2.3: Adopt tournament selection operation, set the tournament size, and adopt the elitist retention strategy. After each generation of evolution, retain the top 10% of the elite individuals and directly add them to the candidate pool of the next generation;

[0036] Step 3.2.4: Adopt a dynamic early stopping mechanism to record the historical optimal solution sequence. Terminate when the improvement amplitude is less than the threshold for several consecutive generations to balance the misjudgment risks of short-term fluctuations and long-term stagnation:

[0037] Step 3.2.5: Use the sound absorption performance predicted by the BP neural network as the fitness value, select the individual with the highest fitness for optimization, and obtain the optimal sound absorption structure parameters.

[0038] Preferably, the sound absorption unit is formed by three-dimensional printing and photocuring of heterogeneous materials; the first porous structure and the second porous structure are porous composite materials; the first embedded partition, the second embedded partition, and the third embedded partition are acoustic metamaterial partitions; the first cover plate and the second cover plate are photosensitive resin materials; the resonant cavity and the sound wave incident channel are composed of air.

[0039] Preferably, when the sound absorption structure is assembled from top to bottom, it is prepared by the stereolithography method and sealed with a thermally activated sealant or a film-like material.

[0040] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects:

[0041] 1. The sound absorption structure of the present invention is provided with the first and second porous structures at the neck. By adjusting the porosity and thickness of the porous material, the acoustic impedance of the neck can be optimized. At the same time, the introduction of the porous material can also broaden the sound absorption frequency band of the resonator, enabling it to achieve efficient sound absorption in a relatively wide frequency range, improving the broadband frequency response, and thus realizing the function of efficient broadband noise suppression.

[0042] 2. The sound absorption structure of the present invention is provided with three embedded partitions inside the cavity. By changing the position of the partitions, the propagation speed and resonance frequency of the sound wave are changed, enabling the sound absorption structure to achieve efficient sound absorption in a relatively wide frequency range. At the same time, the propagation path and resonance frequency of the sound wave in the cavity can be changed to achieve impedance matching, enabling the sound wave to enter the sound absorption structure to the maximum extent and be absorbed.

[0043] 3. The method of the present invention accurately simulates the physical properties of a broadband sound-absorbing structure through a commercial finite element solver, and optimizes the parameters of the sound-absorbing unit through a BP neural network and a genetic algorithm; the sound-absorbing performance predicted by the BP neural network is used as the fitness value, and the individual with the highest fitness is selected through the genetic algorithm for further optimization to obtain the optimal sound-absorbing structure parameters, reducing the time and economic costs in the traditional design method and improving the efficiency and accuracy of the design process. Description of the Drawings

[0044] Figure 1 Schematic diagram of the light-curing forming broadband sound-absorbing structure of the present invention;

[0045] Figure 2 Schematic diagram of the porous structure of the neck of the present invention;

[0046] Figure 3 Schematic diagram of the internal structure of the cavity of the present invention;

[0047] Figure 4 Schematic diagram of the sound pressure level diagram (frequency is 1000 Hz) calculated by the commercial finite element solver COMSOL Multiphysics for the sound-absorbing structure of the embodiment of the present invention;

[0048] Figure 5 Schematic diagram of the sound pressure diagram (frequency is 1000 Hz) calculated by the commercial finite element solver COMSOL Multiphysics for the sound-absorbing structure of the embodiment of the present invention;

[0049] Figure 6 Schematic diagram of the neural network model structure for predicting the sound-absorbing performance in the embodiment of the present invention;

[0050] Figure 7 Schematic diagram of the sound absorption coefficient result diagram simulated by the commercial finite element solver COMSOL Multiphysics for the sound-absorbing structure of the embodiment of the present invention;

[0051] Description of the reference numerals: 100 - sound-absorbing unit; 1 - first porous structure; 2 - second porous structure; 3 - sound wave incident channel; 4 - first embedded partition; 5 - second embedded partition; 6 - third embedded partition; 7 - resonance cavity; 8 - first cover plate; 9 - second cover plate; 10 - sound wave exit surface. Detailed Embodiments

[0052] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the application will be further elaborated in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments related to the present invention. All non-innovative embodiments made by other researchers in the field based on this embodiment fall within the protection scope of the present invention. At the same time, for the step numbers in the embodiments of the present invention, they are only set for the convenience of elaboration and explanation, and no limitation is imposed on the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0053] In an embodiment of the present invention, an optimization method for a light-curing forming broadband sound-absorbing structure based on machine learning includes the following steps:

[0054] Step 1: Construct a broadband sound-absorbing structure, including at least one sound-absorbing unit 100.

[0055] The sound-absorbing unit 100 is a cylindrical cavity, as Figure 1 shown, and includes: a neck porous structure, an internal cavity structure, a first cover plate 8, and a second cover plate 9.

[0056] Among them, the neck porous structure is arranged at one end of the cylindrical cavity, as Figure 2 shown, and is composed of a first porous structure 1, a second porous structure 2, and a sound wave incident channel 3; the internal cavity structure, as Figure 3 shown, is composed of a first embedded partition 4, a second embedded partition 5, a third embedded partition 6, and a resonant cavity 7.

[0057] Furthermore, in the neck porous structure, the bottom surfaces of the first porous structure 1 and the second porous structure 2 are arc trapezoids, and the first porous structure 1 and the second porous structure 2 are not connected, with a sound wave incident channel 3 in between. The first porous structure 1 and the second porous structure 2 are axisymmetric about the center of the sound-absorbing unit 100.

[0058] The first cover plate 8 and the second cover plate 9 are axisymmetric about the center of the sound-absorbing unit 100. The first cover plate 8 and the second cover plate 9 are arranged on both sides of the tops of the first porous structure 1 and the second porous structure 2 and do not cover the first porous structure 1 and the second porous structure 2.

[0059] Furthermore, in the internal cavity structure, the first embedded partition 4, the second embedded partition 5, and the third embedded partition 6 are all semi-cylindrical, and the sizes of the multiple embedded partitions are exactly the same, successively away from the sound wave incident channel 3. The resonant cavity 7 and the sound wave incident channel 3 are composed of air.

[0060] When sound waves enter the first porous structure 1 and the second porous structure 2 from the sound wave incident surface, air vibrations will be generated in the necks. The resonance effect causes the energy of the sound waves to concentrate in the cavities and is converted into heat energy through friction and heat exchange with the cavity walls, thereby achieving the dissipation of sound energy.

[0061] The neck porous structure includes a resonant cavity entrance extending inward, and the porous materials of the first porous structure 1 and the second porous structure 2 have a large number of tiny pores, which can increase the contact area between the sound waves and the material surface, thereby enhancing the dissipation of sound energy.

[0062] When the sound waves pass through the neck porous structure, they are hindered and scattered by the porous material, the propagation speed slows down and the dissipation increases. The energy of the sound waves will be absorbed by the porous material and converted into heat energy through viscous force and heat conduction.

[0063] The porous material can also change the acoustic impedance of the neck. In this embodiment, when = (0.8 - 1.2) there will be a better sound absorption effect, and the sound waves are more likely to enter the resonator and be absorbed. By adjusting the porosity and thickness of the porous material, the acoustic impedance of the neck can be optimized. In addition, the introduction of the porous material can also broaden the sound absorption frequency band of the resonator, enabling efficient sound absorption in a relatively wide frequency range.

[0064] In the internal structure of the cavity, the sound absorption principle of the embedded partition structure is mainly based on the propagation and resonance effect of sound waves in a complex cavity. The first embedded partition 4, the second embedded partition 5, and the third embedded partition 6 jointly form a maze cavity. The sound waves pass through the sound wave incident channel 3 and successively bypass the first embedded partition 4, the second embedded partition 5, and the third embedded partition 6, and finally exit at the sound wave exit surface 10 at the other end of the cavity.

[0065] Due to the complex structure of the cavity, the propagation path of the sound waves is extended, increasing the contact time between the sound waves and the cavity wall, thereby enhancing the dissipation of sound energy.

[0066] In this embodiment, the sound wave propagation path in the maze cavity can be regarded as a series of acoustically resonant cavities connected in series. When = (0.8 - 1.2) there will be a better sound absorption effect, and the sound waves will form standing waves in the cavity, resulting in a significant enhancement of the sound pressure. This resonance effect causes the energy of the sound waves to concentrate in the cavity and is converted into heat energy through friction and heat exchange with the cavity wall, thereby achieving the dissipation of sound energy.

[0067] Specifically, the lengths and positions of the first embedded partition 4, the second embedded partition 5, and the third embedded partition 6 can adjust the sound wave propagation path and resonance frequency in the cavity. By changing the positions of the three embedded partitions in the internal structure of the cavity, the propagation speed and resonance frequency of the sound waves can be changed. This adjustment method enables the sound absorption structure to achieve efficient sound absorption within a relatively wide frequency range. At the same time, the sound wave propagation path and resonance frequency in the cavity can be changed, thereby achieving impedance matching, enabling the sound waves to enter the sound absorption structure to the greatest extent and be absorbed, rather than being reflected.

[0068] Furthermore, to place the practical application within the framework of basic acoustic theory, this embodiment studies the interaction between sound waves and the structure, which is quantitatively described by the key parameter of acoustic impedance.

[0069] For the calculation of the sound absorption coefficient of the acoustic metamaterial under normal incidence conditions in this embodiment, the following formula is used:

[0070] ;

[0071] where the reflection coefficient R is defined as , represents the acoustic impedance of the sound absorption structure, is the characteristic impedance of air. Specifically, the density of air is 1.21 kg / m³, and the speed of sound in air is 343 m / s.

[0072] Specifically, the overall acoustic impedance of the sound absorption structure has the following specific expression:

[0073] ;

[0074] where represents the impedance of the embedded partition part, represents the impedance of the cavity;

[0075] The impedance calculation formula for the cavity impedance is as follows:

[0076] ;

[0077] ;

[0078] ;

[0079] where: is the porosity of the porous neck; is the thickness of the first porous structure 1 and the second porous structure 2; is the flow resistance; is the angular frequency; is the imaginary unit; is the dynamic viscosity of air; d is the width of the acoustic wave incident channel 3, w is the width of the first porous structure 1 and the second porous structure 2;

[0080] The impedance of the embedded partition part The calculation formula is as follows:

[0081] ;

[0082] wherein, is the impedance of the th cavity; is the wave number in air; is the length of the th embedded partition from the acoustic wave incident channel 3; is the thickness of the first embedded partition 4, the second embedded partition 5, and the third embedded partition 6.

[0083] In this embodiment, when =(0.8 - 1.2) , the sound absorption structure has a better sound absorption effect; when = , the sound absorption effect is the best.

[0084] Step 2, Use a commercial finite element solver to perform simulation calculations on the broadband sound absorption structure.

[0085] To verify the sound absorption effect, in this embodiment, a commercial finite element solver COMSOL Multiphysics is used to perform simulation calculations on the broadband sound absorption structure, and the specific method is as follows:

[0086] First, build the geometry and add materials. In the commercial finite element solver COMSOL Multiphysics, build the geometric body through the geometry module, and build the required chimera through operations such as taking the difference set, stretching, and splitting.

[0087] According to the structural requirements, set three semicircular partitions with adjustable length from the incident surface; and add a porous sound absorption structure at the neck, fix the porous medium parameters such as porosity and flow resistivity, and the material of the internal cavity of the structure is air according to the required requirements, and its density and other relevant coefficients are automatically added by the system.

[0088] Then, add multi-physics fields and construct a mesh. The entire sound-absorbing unit uses the "pressure acoustics, frequency domain" module; the range of the air domain is defined by setting a hard boundary.

[0089] The porous structure of the neck and the Helmholtz resonator are regarded as key narrow areas, and the thermo-viscous acoustic interface is selected; the five-parameter JCA model (the Johnson-Champoux-Allard of five parameters) is used to fix the porous medium parameters such as porosity and flow resistivity to describe the propagation behavior of sound waves in porous materials; a plane wave with an amplitude of 1 Pa is used as the excitation source to simulate incident harmonics. A finite element mesh is constructed. To ensure accurate simulation accuracy, the mesh size is selected to be refined, and the range is for all entities.

[0090] Then, conduct a pressure acoustics frequency domain study and output the results. Add a parametric sweep, conduct simulations with multiple modifications to the baffle length, and collect data sets; set and record the lengths of three embedded baffles from the sound wave incident surface. 、 、 , calculate their sound absorption performance from 400 Hz to 1400 Hz, and record the sound absorption performance at a step of 50 Hz. Record the sound absorption coefficients at 21 frequencies corresponding to 3 geometric parameters in each group to form a complete data set; the data set contains geometric parameters (input features) and sound absorption coefficients (target outputs) for the training and validation of subsequent machine learning neural network models.

[0091] The sound absorption results simulated by the above commercial finite element solver COMSOL Multiphysics are as Figure 4 and Figure 5 shown. Figure 4 represents the distribution of the sound pressure level under the condition of a frequency of 1000 Hz. The sound pressure level gradually decreases from the center of the sound source outward, conforming to the law of energy attenuation of sound wave propagation; and the sound pressure level covers multiple frequency bands in the range of 88 - 98 dB, indicating that the sound absorption structure has the ability to regulate in a wide frequency band. Figure 5 represents the distribution of the sound pressure under the condition of a frequency of 1000 Hz. After the plane wave sound source enters from the neck porous material on one side of the cavity, the sound wave enters the improved Helmholtz resonator and continues to propagate. Different sound propagation channels are formed by the embedded baffles, thereby adjusting the sound absorption effect. The distribution of the sound pressure level and the sound pressure shows obvious stratification in space, especially in the resonant cavity part, where the sound energy is more concentrated.

[0092] In addition, the image shows the spatial distribution characteristics of the sound pressure level under specific conditions (such as a frequency of 1000 Hz), reflecting the working principle of the sound-absorbing structure in this embodiment: when air enters the cavity through the small holes in the porous neck, the vibration of the air column in the neck resonantly couples with the volume of the cavity, and the sound energy is converted into heat energy through viscous friction and heat dissipation, absorbing sound waves of specific frequencies. At the same time, the dispersion of the porous structure expands the sound-absorbing frequency band; the partition design of the partition further enhances the reflection and interference of sound waves, prolongs the sound wave propagation path, and improves the energy dissipation efficiency; at the same time. The geometric dimensions of each sub-cavity correspond to different resonance frequencies, and the multi-level resonance peaks are superimposed to cover a wide frequency range, thus achieving broadband sound absorption. This layered sound pressure level distribution is a direct manifestation of precisely controlling the sound energy distribution by adjusting the geometric structure and propagation path, verifying the design principle of adjusting the sound-absorbing effect in this embodiment.

[0093] Step 3: Construct a machine learning parameter optimization model, and optimize the parameters of the sound-absorbing unit through the BP neural network and genetic algorithm.

[0094] To further improve the sound-absorbing performance of the sound-absorbing structure, this embodiment adopts an optimization method of machine learning, and performs machine learning optimization on the sound-absorbing performance of the structure based on the BP neural network and genetic algorithm.

[0095] First, determine a BP neural network model, as Figure 6 shown, and construct the network structure through three fully connected layers.

[0096] Specifically, both hidden layers contain 64 neurons, and a ReLU activation function is set behind each hidden layer to enhance the non-linear fitting ability. In addition, a Dropout layer is set behind each hidden layer, and the Dropout probability is 0.3 to prevent overfitting. Finally, 21 values are generated in the output layer, representing the sound absorption coefficients in the frequency band from 400 Hz to 1400 Hz.

[0097] To optimize the network performance, the BP neural network model is trained and verified according to the dataset collected in step 2. In the training process of the neural network, data preprocessing is first performed using Z-score, the Huber Loss is used as the loss function, and the Adam optimizer is used for training. The learning rate of the Adam optimizer is set to 0.001, and L2 regularization is applied during the optimization process, with a regularization coefficient of 1e-5 to improve the robustness of the model. The gradient is calculated using Backpropagation; in the model verification process, the KFold cross-validation method is adopted, a random seed is set, and the dataset is divided into 5 subsets of equal size; 4 of the subsets are used to train the model, and the remaining 1 subset is used to verify the model, and iterative training is performed until the convergence condition is met.

[0098] Furthermore, based on the fitness function, the genetic algorithm is used to optimize the geometric parameters of the sound-absorbing structure.

[0099] Specifically, the fitness and individuals are initialized, the initial population size is set to 100; the number of genetic iterations is set to 100 generations, the two-point crossover operation is used to cross the individuals, the uniform mutation operation is used to mutate the individuals, and the tournament selection operation is adopted; the elite retention strategy is adopted, and the top 10% of the elite individuals are retained after each generation of evolution and directly added to the next-generation candidate pool to prevent high-quality genes from being lost during crossover and mutation; the dynamic early stopping mechanism is adopted, the historical optimal solution sequence is recorded, and the process is terminated when the improvement amplitude is less than the threshold for 20 consecutive generations to balance the misjudgment risks of short-term fluctuations and long-term stagnation.

[0100] Specifically, for the crossover operation: the two-point crossover operation (cxTwoPoint) is used to cross the individuals, and the crossover probability is 0.5. By randomly selecting two crossover points and the gene expressions of the offspring individuals and are updated as follows:

[0101] ;

[0102] ;

[0103] where and are the gene sequences of the two parent individuals, is the part from the start position of the sequence to before the crossover point ; is the part of the sequence between the crossover points and ; is the part of the sequence from the crossover point to the end of the sequence.

[0104] Specifically, for the mutation operation: the individuals are mutated through the Gaussian mutation operation (mutGaussian), and the mutation probability is 0.3. For the individual , each gene undergoes Gaussian mutation with a probability of 0.3:

[0105] ;

[0106] where is the Gaussian distribution noise with a mean of 0 and a variance of , is the individual before mutation, is the individual after mutation.

[0107] Specifically, the selection operation: Through the tournament selection operation (selTournament), with a tournament size of 3, randomly select 3 individuals from the population { }, calculate their fitness values , and select the individual with the highest fitness:

[0108] ;

[0109] Among them, The fitness value of the i-th individual.

[0110] During the optimization process, the sound absorption performance predicted by the neural network model is used as the fitness value, and the genetic algorithm selects the individual with the highest fitness for further optimization to obtain the optimal structural parameters.

[0111] By accurately simulating the physical properties of the broadband sound absorption structure, the corresponding sound absorption parameters are obtained, thereby further verifying and optimizing the optimized geometric parameters obtained by the neural network model and the genetic algorithm to ensure that the obtained sound absorption structure has better sound absorption performance.

[0112] Step 4. According to the optimized sound absorption structure parameters, by adjusting the thickness and porosity of the first porous structure 1 and the second porous structure 2; and adjusting the distances of the first embedded partition 4, the second embedded partition 5, and the third embedded partition 6 from the sound wave incident surface , , , the acoustic impedance of the sound absorption unit is adjusted , and the sound absorption performance of the broadband sound absorption structure is optimized.

[0113] In this embodiment, in the sound absorption unit , the entire sound absorption unit is high; the width of the sound wave incident channel of the porous neck is , the width of the porous neck is w = 8mm, and the porosity is 0.5; the thickness of the porous structure is ; the thickness of the embedded partition is ; based on the machine learning model for improving the sound absorption performance, adjust the lengths , , of the embedded partition from the sound wave incident channel 3, and when , , , the broadband noise reduction effect of the described sound absorption structure is relatively significant.

[0114] Such as Figure 7As shown, the sound absorption structure of this embodiment exhibits good sound absorption effect, with a generally high sound absorption coefficient, a wide effective absorption bandwidth, and a relatively significant sound absorption and noise reduction effect, meeting the design requirements.

[0115] It can be seen that in the entire optimization process of the light-curing forming broadband sound absorption structure based on machine learning, the present invention finally obtains a sound absorption structure that not only meets the requirements of broadband sound absorption performance but also has realizability through broadband sound absorption structure design, structure optimization based on BP neural network and genetic algorithm, finite element simulation, and light-curing forming technology. It can achieve efficient broadband noise suppression, significantly improve the sound absorption performance, increase the effective bandwidth, and at the same time effectively reduce the design time and economic cost.

[0116] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for optimizing broadband sound absorption structures of photocuring based on machine learning, characterized in that: The steps include: Step 1, constructing a broadband sound absorption structure, comprising at least one sound absorption unit, wherein the sound absorption unit is a cylindrical cavity, comprising: a neck porous structure, and a cavity internal structure; the neck porous structure is arranged at one end of the cylindrical cavity, and is composed of a first porous structure (1), a second porous structure (2), and a sound wave incident channel (3); the first porous structure (1) and the second porous structure (2) are not connected, and are symmetrical along the central axis of the sound absorption unit, with the sound wave incident channel (3) in the middle, and a first cover plate (8) and a second cover plate (9) are arranged on both sides; the cavity internal structure is composed of a first embedded baffle (4), a second embedded baffle (5), a third embedded baffle (6), and a resonant cavity (7); Step 2: Use a commercial finite element solver to simulate and calculate the broadband sound absorption structure, consider the neck porous structure and the embedded partition structure as the key narrow area, and select the thermoviscous acoustic interface; use the five-parameter JCA model to fix the porous medium parameters and describe the propagation behavior of sound waves in porous materials; use plane waves as the excitation source, simulate the incident harmonics, and calculate the sound absorption performance of the broadband sound absorption structure from 400Hz to 1400Hz; Step 3: Build a machine learning parameter optimization model, and optimize the sound absorption unit parameters through BP neural network and genetic algorithm; take the sound absorption performance predicted by BP neural network as the fitness value, select the individual with the highest fitness through genetic algorithm for further optimization, and obtain the optimal sound absorption structure parameters; Step 4: According to the optimized sound absorption structure parameters, the thickness of the first porous structure (1) and the second porous structure (2) are adjusted. and porosity ; and adjusting the distance between the first embedded baffle (4), the second embedded baffle (5), and the third embedded baffle (6) and the sound wave incident surface , , , adjust the acoustic impedance of the sound absorbing unit , optimize the sound absorption performance of broadband sound absorption structure.

2. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 1, characterized in that: In step 1, the first porous structure (1) and the second porous structure (2) have the same structure, and the bottom surface is an arc-shaped trapezoid; The first cover plate (8) and the second cover plate (9) are symmetrical along the central axis of the sound absorption unit and are respectively arranged on both sides of the first porous structure (1) and the second porous structure (2); and the first cover plate (8) and the second cover plate (9) do not cover the first porous structure (1) and the second porous structure (2).

3. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 2, characterized in that: The first embedded baffle (4), the second embedded baffle (5), and the third embedded baffle (6) are all semi-cylindrical, and have the same size and thickness. They are successively farther away from the sound wave incident channel (3), and the distances from the sound wave incident channel (3) are not fixed; The sound wave enters the sound absorbing unit through the sound wave incident channel (3), bypasses the first embedded baffle (4), the second embedded baffle (5), and the third embedded baffle (6) in sequence, and then exits from the sound wave exit surface (10) at the other end of the cylindrical cavity.

4. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 1, characterized in that: The sound absorption coefficient of the sound absorption unit is , calculated as follows: ； Among them, the reflection coefficient R Defined as , is the acoustic impedance of the sound absorbing unit, is the characteristic impedance of air, is the density of air, is the speed of sound in air.

5. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 4, characterized in that: The acoustic impedance of the sound absorbing unit , the expression is as follows: ； in, is the impedance of the embedded partition, is the impedance of the porous structure, is the area of ​​the sound wave incident channel; The impedance of the porous structure , the expression is as follows: ； ； ； in, is the porosity of the first porous structure and the second porous structure; is the thickness of the first porous structure and the second porous structure; is the flow resistance; is the angular frequency; is an imaginary unit; is the dynamic viscosity of air; d is the width of the sound wave incident channel, w is the width of the cavity neck structure; The impedance of the embedded partition , the expression is as follows: ； in, For the The impedance of a cavity; is the wave number in the air; For the The length of each embedded partition from the sound wave incident channel; It is the thickness of the first embedded partition, the second embedded partition, and the third embedded partition.

6. The method for optimizing broadband sound absorption structure of photocuring based on machine learning according to claim 5, characterized in that: The simulation calculation of the broadband sound absorption structure is carried out using a commercial finite element solver, which includes the following sub-steps: Step 2.1, the broadband sound absorption structure is calculated using the commercial finite element solver COMSOL Multiphysics. The sound absorption unit is simulated using the pressure acoustics-frequency domain module, and the air domain is defined by setting hard boundaries; Step 2.2: Consider the porous neck structure and embedded baffle structure as the key narrow area and select the thermoviscous acoustic interface; The five-parameter JCA model is used to fix the porous media parameters, including: porosity , flow resistance , thermal conductivity, describing the propagation behavior of sound waves in porous materials; constructing geometric bodies through assembly modules, and constructing required chimeras through difference sets, stretching, and splitting operations; using finite element meshes, selecting and refining the mesh size, and the range is for all entities; using a plane wave with an amplitude set to 1Pa as the excitation source to simulate the incident harmonics; Step 2.3, after the plane wave sound source is incident from the porous structure of the neck on one side of the cavity, it enters the resonant cavity and continues to propagate, forming different sound propagation channels through three embedded partitions. The geometric dimensions of each sub-cavity correspond to different resonant frequencies. After the multi-level resonance peaks are superimposed, a wide frequency range is covered to achieve broadband sound absorption; Step 2.4, perform simulation calculations through parametric scanning, and input frequency points through an explicit list; limit the effective frequency range to 400 Hz to 1400 Hz, with a step size of 50 Hz, for a total of 21 frequency points; calculate the sound absorption coefficients at different frequency points respectively, and adjust the broadband sound absorption effect.

7. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 6, characterized in that: The BP neural network is constructed by three fully connected layers with two hidden layers in the middle. The specific method is as follows: Step 3.1.1, data input: The input layer data includes the distances of the first embedded partition, the second embedded partition, and the third embedded partition from the sound wave incident surface. , , ; Step 3.1.2, data fitting: Each hidden layer contains 64 neurons, and each hidden layer is also set with a ReLU activation function and a Dropout layer, with a Dropout probability of 0.3; Step 3.1.3, data output: The output layer outputs 21 values, representing the sound absorption coefficient in the frequency range of 400Hz to 1400Hz ; Step 3.1.4, network model training: use Z-score for data preprocessing, use Huber Loss as the loss function, and use Adam optimizer for training; the learning rate of Adam optimizer is set to 0.001, and L2 regularization is applied during the optimization process, and the regularization coefficient is 1e-5; Step 3.1.5, network model verification: Backpropagation is used to calculate the gradient. The KFold cross-validation method is used in the verification process. A random seed is set and the data set is divided into 5 subsets of equal size. Four subsets are used to train the model and one subset is used to verify the model. The sound absorption performance parameters predicted by the optimized BP neural network are obtained.

8. The method for optimizing broadband sound absorption structure of photocuring based on machine learning according to claim 7, characterized in that: The genetic algorithm is used to optimize the geometric parameters of the sound absorbing structure. The specific method is as follows: Step 3.2.1, set the initial population size, the number of genetic iterations, use the two-point crossover operation to cross individuals, and set the crossover probability; Step 3.2.2, use uniform mutation operation to mutate individuals and set the mutation probability; Step 3.2.3, use the tournament selection operation, set the tournament size, adopt the elite retention strategy, retain the elite individuals after each generation of evolution, and add them to the next generation candidate pool; Step 3.2.4: Use a dynamic early stopping mechanism to record the historical optimal solution sequence. When the improvement of several consecutive generations is less than the threshold, the solution is terminated to balance the misjudgment risk of short-term fluctuations and long-term stagnation: Step 3.2.5: Use the sound absorption performance predicted by the BP neural network as the fitness value, select the individual with the highest fitness for optimization, and obtain the optimal sound absorption structure parameters.

9. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 1, characterized in that: The sound absorbing unit is formed by three-dimensional printing and photocuring of heterogeneous materials; the first porous structure (1) and the second porous structure (2) are porous composite materials; the first embedded partition (4), the second embedded partition (5) and the third embedded partition (6) are acoustic metamaterial partitions; the first cover plate (8) and the second cover plate (9) are photosensitive resin materials; and the resonant cavity (7) and the sound wave incident channel (3) are composed of air.

10. The method for optimizing broadband sound absorption structure of stereolithography based on machine learning according to claim 1, characterized in that: When the sound absorbing structure is assembled from top to bottom, it is prepared by a stereolithography method and sealed by a heat-activated sealant or a film-like material.

Images