Sound insulation acrylic plate structure optimization method based on deep learning

Abstract

The invention belongs to the field of architectural acoustic materials and structural acoustic design, and aims to solve the problems that an existing acrylic plate structure is weak in sound insulation capacity, depends on sound absorption materials or thickness superposition, cannot quickly design and optimize structural parameters and is uncontrollable in acoustic performance. By constructing an acrylic plate structure-performance database, concave-convex structure sound insulation physical constraint conditions and a structure performance prediction-reverse optimization prediction model, frequency customizable design is achieved through structure-performance accurate mapping, and a systematic technical solution is provided for a light, efficient, intelligent and controllable next-generation building sound insulation material. And the method has good practicability, expandability and industrial conversion prospect.

Description

Technical Field

[0001] This invention belongs to the field of architectural acoustic materials and structural acoustic design, and in particular relates to a method for optimizing the structure of residential sound insulation micro-topological corrugated acrylic panels based on deep learning. Background Technology

[0002] With the increasing density of modern urban dwellings and residents' growing demand for a higher quality of life, sound insulation materials are being used more and more widely in home and office environments. However, traditional sound insulation materials generally suffer from problems such as cumbersome installation, heavy weight, poor aesthetics, and limited lifespan, making it difficult to simultaneously meet the combined needs of "function + decoration." Acrylic (polymethyl methacrylate) sheets have become a commonly used material in modern interior decoration due to their good mechanical properties, optical transmittance, and decorative properties. However, due to their smooth surface and uniform internal structure, their ability to block airborne sound waves is limited, especially in dealing with common indoor noise in the speech frequency range (1-4kHz).

[0003] Traditional sound insulation techniques often rely on layering thick materials or using sound-absorbing sandwich panels to achieve sound insulation effects, resulting in large material volumes, complex construction, and a difficulty in balancing aesthetics and functionality. Especially in high-end home decoration and open-plan office spaces, users have higher requirements for material appearance, lightweight properties, and environmental friendliness. Addressing this challenge, how to improve the acoustic performance of materials through structural design while maintaining their thinness and aesthetics has become a technical bottleneck in material design.

[0004] Currently, some research is beginning to focus on methods that utilize surface structures to alter the sound wave propagation path. This involves introducing periodic textures and microstructures to create scattering, interference, and phase delay of sound waves, thereby improving the sound insulation performance of materials. However, the systematic application of these structural acoustic methods in thermoplastic materials such as acrylic remains relatively limited, particularly in terms of practical manufacturability, processing precision, and decorative compatibility, for which no mature solutions exist. Summary of the Invention

[0005] Against this backdrop, this invention proposes a deep learning-based method for optimizing the structure of residential sound-insulating micro-topological corrugated acrylic panels. By constructing a textured structure on the surface of a single-layer acrylic panel, passive sound insulation is achieved. This approach does not rely on traditional sound-absorbing materials or sandwich structures; it effectively suppresses mid-to-high frequency noise by altering sound wave propagation behavior solely through geometric changes. This invention provides a novel approach and solution for integrating structural acoustics with decorative materials. The acrylic panels optimized using the proposed method are lightweight, aesthetically pleasing, and easy to install, overcoming the limitations of existing acrylic materials in acoustic functionality and meeting the practical needs of modern indoor environments for high-performance, lightweight, and customized sound insulation materials.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention proposes a deep learning-based method for optimizing the structure of sound-insulating acrylic panels, comprising the following steps:

[0008] Step S1: Construct the acrylic sheet structure-performance database.

[0009] Specifically, an acrylic sheet structure-performance database was constructed using a 3:1 ratio of actual data to simulation data for the training and modeling of neural networks.

[0010] Step S2: Construct the physical constraints for sound insulation of the acrylic sheet with concave and convex structure.

[0011] Specifically, to break free from reliance on traditional porous sound-absorbing materials or sandwich filling materials, the interpretability of the network is enhanced by constructing physical constraints for sound insulation with concave and convex structures. This means that during the optimization process, physical mechanisms such as scattering, interference, impedance, and resonance are incorporated as constraints, allowing the network to maintain reasonable predictions even with unseen parameter combinations, thus avoiding the network relying solely on empirical fitting. Furthermore, in the subsequent back-optimization process, these physical constraints can serve as filtering conditions, preventing the output from being unmanufacturable or physically unreasonable.

[0012] Specifically, the physical constraints for sound insulation of the concave-convex structure include the first scattering interference constraint, the second transmission loss constraint, the third reflection constraint, and the fourth periodic resonance constraint.

[0013] Step S3: Construct a structural performance prediction-reverse optimization prediction model to optimize the structural parameters of the acrylic sheet. First, use the structural performance prediction network with the physical constraints of the concave-convex structure sound insulation in step S2 to quickly estimate the sound insulation volume TL from the structural parameters. Then, using the target frequency fo and the minimum sound insulation requirement TLmin as conditions, optimize the structural parameters in reverse iteration through a genetic operator until the maximum number of iterations or the sound insulation requirement is met. Finally, output the optimal structural parameters and the trained prediction network. The physical constraints of the concave-convex structure sound insulation are set between the third and fourth fully connected layers.

[0014] Specifically, the structural performance prediction network is used to optimize the structural parameters; the reverse optimization network is used to perform reverse reasoning optimization of the structural parameters.

[0015] The workflow of the structural performance prediction-backward optimization prediction model includes the forward propagation process of the structural performance prediction network and the optimization process of the backward optimization module.

[0016] The forward propagation process of the structural performance prediction network includes:

[0017] Step S31: Input structural parameters. Use 10 structural parameters as input to the network, namely: bump unit height h, bump unit spacing d, bump unit diameter a, arrangement T, total texture coverage R, plate thickness H, material type M, and surface texture density f. s Target frequency band f, random disturbance δ.

[0018] Step S32: Use an adaptive normalization layer to adaptively normalize the input structural parameters.

[0019] Step S33: The normalized parameters are passed through the first fully connected layer to expand the dimension from 10 to 64.

[0020] Step S34: Apply the ReLU activation function to the output of the first fully connected layer and perform a Dropout operation with a Dropout rate of 0.2 to prevent overfitting.

[0021] Step S35: Pass the output of step S34 through a second fully connected layer to expand the dimension from 64 to 128.

[0022] Step S36: Perform batch normalization (BatchNorm) and apply the ReLU activation function to the output of the second fully connected layer.

[0023] Step S37: Input the output of step S37 into the self-attention mechanism and the multi-head attention mechanism to capture the dependencies between the input features.

[0024] Step S38: Pass the output of the previous step through a ResNet layer to enhance the network's depth and learning ability.

[0025] Step S39: Pass the output of the ResNet layer through a third fully connected layer, keeping the dimension at 128.

[0026] Step S40: Input the output of the third fully connected layer to the concave-convex structure sound insulation physical constraint module, and use the first scattering interference constraint, the second transmission loss constraint, the third reflection constraint and the fourth periodic resonance constraint to filter the parameters and obtain the parameters that simultaneously satisfy the above four constraint conditions.

[0027] Step S41: Input the output of step S40 into the fourth fully connected layer and the fifth fully connected layer in sequence to obtain the final predicted value of transmission loss TL.

[0028] Specifically, a composite loss function L is introduced. total =L MSE +λ r L res +λ e L energy +λ zL imp To improve prediction accuracy, where L MSE Mean square error, For resonance error, L energy In engineering, L is used for limiting and smoothing TL (Transmission Limiting) and punishing illogical abnormal oscillations. imp For impedance matching, d is the structural spacing, c is the speed of sound in air, and λ is the velocity of sound in air. r , λ e , λ z For weights.

[0029] The optimization process of the reverse optimization module is as follows:

[0030] Step S311: Set the network target, target frequency f0 and minimum sound insulation requirement TLmin≥x.

[0031] Step S312: Initialize the parameter group, initialize a set of structure parameters X1, X2, ..., Xn.

[0032] Step S313: Calculate the transmission loss TLi corresponding to the structural parameter Xi based on the structural performance prediction network.

[0033] Step S314: Fitness assessment, assess whether the transmission loss TLi corresponding to each structural parameter Xi meets the minimum sound insulation requirement TLmini≥x.

[0034] Step S315: Based on the fitness evaluation results, select structural parameters with higher fitness for crossover and mutation operations to generate new structural parameter sets.

[0035] Step S316: Repeat the above steps until the maximum number of iterations MaxGen = 100 is reached.

[0036] Step S317: Output the optimal structure parameter X: Output the final optimal structure parameter X.

[0037] Through the above steps, structural performance can be predicted and optimized to meet specific sound insulation requirements.

[0038] The present invention has the following advantages:

[0039] 1. Constructing an acrylic sheet structure-performance database to achieve a one-to-one correspondence between sheet structure and sound insulation performance: This invention establishes an acrylic sheet structure-performance database by custom-designing large-scale structural samples with different texture parameters (including height, spacing, shape, arrangement, and disturbance level), and obtaining transmission loss values ​​at target frequencies (500-5000Hz) through acoustic simulation or field testing. The samples include various texture arrangements such as periodic, honeycomb, and disturbed types, covering a height of 0.3-2.0 mm, a coverage rate of 30%-90%, and a texture density of 10-100 particles / cm². 2 With various structural configurations, comprehensive sample coverage and full expression of physical variables are achieved, providing a high-quality training data foundation for deep learning modeling.

[0040] 2. Introducing structural acoustic physical mechanisms, a deep learning prediction model with physical constraints on sound insulation due to concave-convex structures is designed: Addressing the phenomena of reflection, scattering, and phase delay of sound waves at different frequencies caused by concave-convex structures, this invention incorporates acoustic propagation laws into the neural network training process. This avoids outputting designs that are unmanufacturable or physically unreasonable, resulting in more accurate optimization results. Physical information embedding is used to regularize the network loss function, ensuring that the model's learning results conform to the physical constraints on sound insulation due to concave-convex structures, thus enhancing the network's structural understanding ability.

[0041] 3. Constructing a structural performance prediction-reverse optimization prediction model to achieve automated design and performance customization of personalized sound insulation panels. This invention further proposes a reverse optimization module that integrates a physical heuristic and data-driven collaborative mechanism. After inputting the target frequency band and transmission loss threshold, this module not only integrates conventional optimization algorithms (such as genetic algorithms, Bayesian optimization, and gradient guidance) for searching, but also introduces multi-objective Pareto front analysis, structural manufacturability constraint evaluation, and a local feature importance reweighting mechanism to achieve global-local joint optimization of concave-convex structural parameters (height, spacing, arrangement, etc.). The structural parameters output by the system can balance acoustic performance, manufacturing complexity, and material utilization, ultimately achieving an intelligent, rapid, and highly manufacturable customized sound insulation panel design scheme. Compared to traditional trial-and-error-based parameter adjustment methods and single optimization target search processes, this invention achieves a closed-loop innovation process from "structural performance - physical model - manufacturability trade-off - adaptive optimization strategy." This capability breaks through the traditional trial-and-error manual design method, significantly improving design efficiency and product diversity while ensuring performance. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an overall flowchart of the acrylic sheet structure optimization method proposed in this invention;

[0044] Figure 2 This is a flowchart of the structural performance prediction-backward optimization prediction model of the present invention;

[0045] Figure 3 The figure shows the experimental results of the optimization method proposed in this invention. Detailed Implementation

[0046] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0047] This invention proposes a deep learning-based method for optimizing the structure of sound-insulating acrylic panels, such as... Figure 1 As shown, it includes the following steps:

[0048] Step S1: Construct the acrylic sheet structure-performance database.

[0049] Specifically, an acrylic sheet structure-performance database was constructed using a 3:1 ratio of actual data to simulation data for the training and modeling of neural networks.

[0050] Specifically, when constructing the acrylic sheet structure-performance database, each set of structural data samples is designed to include the following physical parameters: unit height of the raised / lowered area h, unit spacing of the raised / lowered area d, unit diameter of the raised / lowered area a, arrangement T, total texture coverage R, sheet thickness H, material type M, transparent acrylic PMMA0, frosted acrylic PMMA1, and surface texture density f. s Target frequency band f, disturbance factor δ, experimental / simulation transmission loss value TL.

[0051] Specifically, the data storage method is to store all parameters (h,d,a,T,R,f) of each set of structural samples. sThe table header contains H, M, f, and δ; each sample is recorded on a separate line, showing the measured or simulated transmission loss TL value (dB); all data is saved in CSV format, with the filename indicating the test batch, acquisition method (measured / simulated), and timestamp; the database can be directly imported into modeling platforms such as Python and MATLAB, facilitating subsequent training and reverse optimization of the neural network.

[0052] In one embodiment, actual data is obtained through testing physical samples. Specifically, this includes pre-processing the physical sheet material, i.e., selecting transparent or frosted acrylic sheets and cutting them into 30cm × 30cm test samples according to a set thickness; during surface texture processing, using CNC milling equipment or laser engraving equipment, processing different shapes and parameters of raised and recessed textures on the sheet surface according to design drawings; periodically arranged samples are programmed using grid coordinates, while randomly disturbed samples are engraved after the program randomly disturbs the texture center coordinates; subsequently, hot pressing mold processing is performed, and for batch samples with consistent structures, steel molds are used for heating and forming to ensure consistent raised and recessed heights; finally, experimental testing is conducted: the processed samples are placed in a standard sound insulation test chamber (internal dimensions 1.2m × 1.2m × 1.0m), and with the aid of directional speakers and a microphone array, the transmission loss (TL) of the samples is tested at 250Hz intervals from 500Hz to 5000Hz using a frequency sweep method, and calculated using the sound pressure level difference method.

[0053] In one embodiment, simulation data is obtained through simulation experiments. The simulation sample acquisition process involves using the COMSOL Multiphysics Acoustic Module to establish a 3D structural modeling environment, setting the bump texture model according to parameters, and modeling the air layer / acrylic plate / receiving air layer as a layered domain; setting material parameters (PMMA sound velocity: 2730 m / s, density: 1180 kg / m³). 3 Set the boundary condition to Plane Wave Radiation Boundary and set the frequency scan through the FrequencyDomain Study module; extract the sound pressure P_out on the penetrating surface (back of the plate) and the sound pressure P_in on the incident surface in the simulation, and calculate the transmission loss TL(f) of the corresponding frequency band.

[0054] Step S2: Construct the physical constraints for sound insulation of the acrylic sheet with concave and convex structure.

[0055] Specifically, to break free from reliance on traditional porous sound-absorbing materials or sandwich filling materials, the interpretability of the network is enhanced by constructing physical constraints for sound insulation with concave and convex structures. That is, during the optimization process, physical mechanisms such as scattering, interference, impedance, and resonance are incorporated as constraints, allowing the network to maintain reasonable predictions even with unseen parameter combinations, thus avoiding the network learning only empirical fits. On the other hand, in the subsequent back-optimization process, physical mechanisms can serve as filtering conditions to avoid outputting designs that are unmanufacturable or physically unreasonable.

[0056] Specifically, the physical constraints for sound insulation in the concave-convex structure include the first scattering interference constraint. Because the concave-convex texture structure forms multiple scattering centers, when the incident sound wave interacts with the concave-convex unit, multipath scattering occurs, causing the sound wave energy to disperse in different directions, thereby reducing the sound energy in the transmission direction. For wavelength λ that is close to the size a of the concave-convex structure, the scattering cross-section σ mainly affects the sound wave intensity, and can be approximately expressed as:

[0057]

[0058] When a << λ, scattering is weak; when a ≈ λ, scattering is significantly enhanced, especially in the mid-to-high frequency range.

[0059] Specifically, the physical constraints for sound insulation in the concave-convex structure include a second transmission loss constraint. Because the concave-convex array causes sound waves to reflect off different textured unit surfaces, it creates a multi-path interference wave field with phase differences. Through the superposition of these phase differences, sound waves in certain frequency bands undergo destructive interference, leading to a decrease in transmitted wave energy. At this point, the total sound pressure of the superimposed sound waves P1 and P2 is:

[0060]

[0061] Wherein, the phase difference Δφ = φ2 - φ1 causes an interference effect, and the transmission loss can be expressed as:

[0062]

[0063] P incident The incident power is the sum of the incident and incident power. When multiple units are superimposed, a more complex interference is formed, and the overall intensity of the reflected wave can be significantly enhanced.

[0064] Specifically, the physical constraints of the concave-convex structure for sound insulation include third reflection constraints. Based on the local impedance change mechanism, the acoustic impedance Z = ρc (medium density ρ, sound velocity c) determines the reflection and transmission characteristics of sound waves. The concave-convex structure leads to a non-uniform distribution of local acoustic impedance on the acrylic surface, forming local impedance matching and mismatch regions, thereby adjusting the sound wave reflection coefficient. Where Z1 is the air acoustic impedance, and Z2 is the equivalent acoustic impedance of the concave-convex structural region. The impedance difference enhances reflection and reduces transmission.

[0065] Specifically, the physical constraints for sound insulation in concave-convex structures include a fourth periodic resonance constraint. According to the periodic resonance effect, when the period d of the concave-convex arrangement approaches an integer fraction of the wavelength λ of the sound wave propagating in the material, Bragg scattering or periodic resonance occurs, causing significant reflection of sound waves in a specific frequency band. The condition is 2d sinθ=nλ, n=1,2,3..., where θ is the angle of incidence. Reflection is enhanced in the normal direction, forming a sound insulation band with a blocking frequency range.

[0066] Step S3: Construct a structural performance prediction-reverse optimization prediction model to optimize the structural parameters of the acrylic sheet. First, use the structural performance prediction network with the physical constraints of the concave-convex structure sound insulation in step S2 to quickly estimate the sound insulation volume TL from the structural parameters. Then, using the target frequency fo and the minimum sound insulation requirement TLmin as conditions, optimize the structural parameters in reverse iteration through a genetic operator until the maximum number of iterations or the sound insulation requirement is met. Finally, output the optimal structural parameters and the trained prediction network. The physical constraints of the concave-convex structure sound insulation are set between the third and fourth fully connected layers of the structural performance prediction network.

[0067] Specifically, the structural performance prediction network is used to optimize the structural parameters; the reverse optimization network is used to perform reverse reasoning optimization of the structural parameters.

[0068] Specifically, the structural performance prediction network includes, in sequence, an adaptive normalization layer, a first fully connected layer, a ReLU activation layer, a second fully connected layer, a BatchNorm+ReLU layer, a self-attention and multi-head self-attention layer, a ResNet layer, a third fully connected layer, a concave-convex structure sound insulation physical constraint module, a fourth fully connected layer, a fifth fully connected layer, and an output layer.

[0069] Because acoustic structure parameters vary significantly in physical dimensions and distribution scales (e.g., millimeters, percentages, frequencies), this invention designs an adaptive normalization mechanism at the front end of the prediction network to normalize the parameters. The adaptive normalization mechanism applies this mechanism to each input parameter x. i The adaptive normalization expression is:

[0070]

[0071] Where μ i σ i Let γ be the mean and standard deviation on the training set. i The learnable parameter represents the relative importance of the feature in that dimension. This mechanism is embedded after the network input layer, serving as a normalization layer and updated through gradient training, thereby achieving a fair representation of each parameter during the feature extraction process.

[0072] Specifically, the structural performance prediction network introduces residual connections (ResNet-style) and a Transformer submodule on top of the basic multilayer perceptron (MLP) structure to improve the model's ability to model the nonlinear relationship between complex structural parameters and acoustic response. The network structure input layer contains a 10-dimensional structural parameter vector; the activation function uses ReLU and adds a BatchNorm layer to accelerate convergence and improve stability; the Transformer module sets up two layers of Multi-Head and Self-Attention, where the Multi-Head Attention has a head of 4 and a hidden dimension of 128, uses LayerNorm and residual connections, and sets Linear and ReLU activation in the feedforward layer.

[0073] The workflow of the structural performance prediction-backward optimization prediction model includes the forward propagation process of the structural performance prediction network and the optimization process of the backward optimization module, as detailed below. Figure 2 As shown.

[0074] The forward propagation process of the structural performance prediction network includes:

[0075] Step S31: Input structural parameters. Use 10 structural parameters as input to the network, namely: bump unit height h, bump unit spacing d, bump unit diameter a, arrangement T, total texture coverage R, plate thickness H, material type M, and surface texture density f. s Target frequency band f, disturbance factor δ.

[0076] Step S32: Use an adaptive normalization layer to adaptively normalize the input structural parameters.

[0077] Step S33: The normalized parameters are passed through the first fully connected layer to expand the dimension from 10 to 64.

[0078] Step S34: Apply the ReLU activation function to the output of the first fully connected layer and perform a Dropout operation with a Dropout rate of 0.2 to prevent overfitting.

[0079] Step S35: Pass the output of step S34 through a second fully connected layer to expand the dimension from 64 to 128.

[0080] Step S36: Perform batch normalization (BatchNorm) and apply the ReLU activation function to the output of the second fully connected layer.

[0081] Step S37: Input the output of step S37 into the self-attention mechanism and the multi-head attention mechanism to capture the dependencies between the input features.

[0082] Step S38: Pass the output of the previous step through a ResNet layer to enhance the network's depth and learning ability.

[0083] Step S39: Pass the output of the ResNet layer through a third fully connected layer, keeping the dimension at 128.

[0084] Step S40: Input the output of the third fully connected layer to the concave-convex structure sound insulation physical constraint module, and use the first scattering interference constraint, the second transmission loss constraint, the third reflection constraint and the fourth periodic resonance constraint to filter the parameters and obtain the parameters that simultaneously satisfy the above four constraint conditions.

[0085] Step S41: Input the output of step S40 into the fourth fully connected layer and the fifth fully connected layer in sequence to obtain the final predicted value of transmission loss TL.

[0086] Specifically, a composite loss function L is introduced. total =L MSE +λ r L res +λ e L energy +λ z L imp To improve prediction accuracy, where L MSE Mean square error, For resonance error, L energy In engineering, L is used for limiting and smoothing TL (Transmission Limiting) and punishing illogical abnormal oscillations. imp For impedance matching, d is the structural spacing, c is the speed of sound in air, and λ is the velocity of sound in air. r , λ e , λ z For weights.

[0087] The optimization process of the reverse optimization module is as follows:

[0088] Step S311: Set the network target, target frequency f0 and minimum sound insulation requirement TLmin≥x.

[0089] Step S312: Initialize the parameter group, initialize a set of structure parameters X1, X2, ..., Xn.

[0090] Step S313: Calculate the transmission loss TLi corresponding to the structural parameter Xi based on the structural performance prediction network.

[0091] Step S314: Fitness assessment, assess whether the transmission loss TLi corresponding to each structural parameter Xi meets the minimum sound insulation requirement TLmini≥x.

[0092] Step S315: Based on the fitness evaluation results, select structural parameters with higher fitness for crossover and mutation operations to generate new structural parameter sets.

[0093] Step S316: Repeat the above steps until the maximum number of iterations MaxGen = 100 is reached.

[0094] Step S317: Output the optimal structure parameter X: Output the final optimal structure parameter X.

[0095] Through the above steps, structural performance can be predicted and optimized to meet specific sound insulation requirements.

[0096] Specifically, to achieve reverse reasoning of structural parameters—that is, to automatically generate structural parameters that meet the requirements based on the target frequency and target sound insulation performance (e.g., TL≥30dB)—a reverse optimization module for structural parameters is designed. The optimization objective is defined as:

[0097]

[0098] stf = f0, TL(f0, X) ≥ TL target

[0099] Where f(x) = f0, f0 is the target frequency, and TL target Let F(X) be the minimum required transmission loss value, and let F(X) be the trained performance prediction network. The main optimization tool is a Genetic Algorithm (GA).

[0100] Specifically, the genetic algorithm (GA) includes initializing the population and generating 100-200 sets of candidate structural parameters x. i Fitness assessment, using a prediction network for each group x i The transmission loss (TL) value is calculated as a fitness score. In the selection operation, parent structures with high fitness are retained and combined to form new structural parameters. Random perturbation parameters are added to increase exploratory mutation operations, and this process is repeated for 50–100 rounds to output the optimal structure. The optimization objective is to minimize the transmission loss (TL) to no less than 30 dB in a specified frequency range (e.g., 1–4 kHz), suitable for residential voice suppression. Each candidate structural parameter is evaluated by a trained prediction network to ensure the optimization results are practically feasible for manufacturing, avoiding the creation of unmanufacturable extreme structures.

[0101] Experimental results are as follows Figure 3As shown in the figure. This invention verifies the accuracy of the prediction model through experiments. Several sets of transmission loss data for convex-concave acrylic panels were collected in a sound insulation test chamber and compared with the output of the prediction network. The results show that within the 500–4000 Hz frequency band, the average error between the predicted and measured values ​​is less than 5%, and the correlation coefficient reaches 0.92. This result demonstrates the high-precision prediction capability and good generalization performance of the network of this invention.

[0102] The structural optimization network structure proposed in this application realizes an end-to-end modeling process of "structural input → performance prediction → structural reverse optimization". With a structural performance prediction network that embeds the physical constraints of sound insulation of concave and convex structures, as well as a reverse optimization module, it effectively improves the design efficiency and sound insulation reliability of acrylic sheet concave and convex structures, and meets the material design requirements under multi-frequency band and complex structures.

[0103] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0105] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0107] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0108] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0109] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0110] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0111] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for optimizing the structure of sound-insulating acrylic panels based on deep learning, characterized in that: The structural optimization method includes the following steps: Step S1: Construct the acrylic sheet structure-performance database; Specifically, an acrylic sheet structure-performance database was constructed according to a 3:1 ratio of actual data to simulation data for use in neural network training and modeling. Step S2: Construct the physical constraints for sound insulation of the acrylic sheet with concave and convex structure, including the first scattering interference constraint, the second transmission loss constraint, the third reflection constraint, and the fourth periodic resonance constraint; Step S3: Construct a structural performance prediction-reverse optimization prediction model to optimize the structural parameters of the acrylic sheet. First, a structural performance prediction network with the physical constraints of the concave-convex structure sound insulation from Step S2 is used to quickly estimate the sound insulation volume TL from the structural parameters. Then, using the target frequency fo and the minimum sound insulation requirement TLmin as conditions, the structural parameters are iteratively optimized using a genetic operator until the maximum number of iterations or the sound insulation requirement is met. Finally, the optimal structural parameters and the trained prediction network are output. The physical constraints of the concave-convex structure sound insulation are set between the third and fourth fully connected layers of the structural performance prediction network. Specifically, the structural performance prediction network is used to optimize the structural parameters; the reverse optimization network is used to perform reverse reasoning optimization of the structural parameters. Introducing a composite loss function L into the structural performance prediction network total =L MSE +λ r L res +λ e L energy +λ z L imp To improve prediction accuracy, where L MSE Mean square error, For resonance error, L energy In engineering, L is used for limiting and smoothing TL (Transmission Limiting) and punishing illogical abnormal oscillations. imp For impedance matching, d is the structural spacing, c is the speed of sound in air, and λ is the velocity of sound in air. r , λ e , λ z For weights.

2. The deep learning-based sound insulation acrylic panel structure optimization method according to claim 1, wherein when constructing the acrylic panel structure-performance database, each set of structural data samples is designed to include the following physical parameters: unit height h of concave / convex surfaces, unit spacing d of concave / convex surfaces, unit diameter a of concave / convex surfaces, arrangement T, total texture coverage R, panel thickness H, material type M, transparent acrylic PMMA0, frosted acrylic PMMA1, and surface texture density f. s Target frequency band f, random disturbance δ, experimental / simulation transmission loss value TL; Specifically, the data storage method is to store all parameters (h,d,a,T,R,f) of each set of structural samples. s H, M, f, δ) serve as the data header; each sample is recorded on a separate line, showing the measured or simulated transmission loss TL value.

3. The sound insulation acrylic panel structure optimization method based on deep learning according to claim 1, wherein the first scattering interference constraint is: when the wavelength λ is close to the size a of the concave-convex structure, the scattering cross section σ mainly affects the sound wave intensity, which can be approximately expressed as: in, When a << λ, scattering is weak; when a ≈ λ, scattering is significantly enhanced. The second transmission loss constraint is: the total sound pressure level of the superimposed two sound waves P1 and P2 is: The phase difference Δφ = φ2 - φ1 causes an interference effect, and the transmission loss can be expressed as: P incident The incident power; when multiple units are superimposed, a more complex interference is formed, and the overall intensity of the reflected wave can be significantly enhanced. The third reflection constraint is: sound wave reflection coefficient Where Z1 is the air acoustic impedance and Z2 is the equivalent acoustic impedance of the concave-convex structure region; the impedance difference enhances reflection and reduces transmission. The fourth periodic resonance constraint is as follows: According to the periodic resonance effect, when the period d of the concave-convex arrangement is close to an integer fraction of the wavelength λ of the sound wave propagating in the material, Bragg scattering or periodic resonance is generated, causing the sound wave in a specific frequency band to be significantly reflected; the condition is 2dsinθ=nλ, n=1,2,3..., where θ is the incident angle. Reflection is enhanced in the normal direction, forming the sound insulation band blocking frequency band.

4. The deep learning-based sound insulation acrylic panel structure optimization method according to claim 1, wherein the workflow of the structural performance prediction-backward optimization prediction model includes the forward propagation process of the structural performance prediction network and the optimization process of the backward optimization module.

5. The deep learning-based sound-insulating acrylic panel structure optimization method according to claim 4, wherein the forward propagation process of the structural performance prediction network includes: Step S31: Input structural parameters. Use 10 structural parameters as input to the network, namely: bump unit height h, bump unit spacing d, bump unit diameter a, arrangement T, total texture coverage R, plate thickness H, material type M, and surface texture density f. s Target frequency band f, random disturbance δ; Step S32: Use an adaptive normalization layer to adaptively normalize the input structural parameters; Step S33: Pass the normalized parameters through the first fully connected layer to expand the dimension from 10 to 64; Step S34: Apply the ReLU activation function to the output of the first fully connected layer and perform a Dropout operation with a Dropout rate of 0.2 to prevent overfitting; Step S35: Pass the output of step S34 through a second fully connected layer to expand the dimension from 64 to 128; Step S36: Batch normalize the output of the second fully connected layer and apply the ReLU activation function; Step S37: Input the output of step S37 into the self-attention mechanism and the multi-head attention mechanism to capture the dependencies between the input features; Step S38: Pass the output of the previous step through a ResNet layer to enhance the network's depth and learning ability; Step S39: Pass the output of the ResNet layer through the third fully connected layer, keeping the dimension at 128; Step S40: Input the output of the third fully connected layer to the concave-convex structure sound insulation physical constraint module, and use the first scattering interference constraint, the second transmission loss constraint, the third reflection constraint and the fourth periodic resonance constraint to filter the parameters and obtain the parameters that simultaneously satisfy the above four constraint conditions. Step S41: Input the output of step S40 into the fourth fully connected layer and the fifth fully connected layer in sequence to obtain the final predicted value of transmission loss TL.

6. The deep learning-based sound insulation acrylic panel structure optimization method according to claim 5, wherein the adaptive normalization layer for each input parameter x i The adaptive normalization expression is: Where μ i σ i Let γ be the mean and standard deviation on the training set. i is a learnable parameter that represents the relative importance of the feature in this dimension.

7. The optimization process of the inverse optimization module in the deep learning-based sound insulation acrylic panel structure optimization method according to claim 4 is as follows: Step S311: Set the network target, target frequency fo and minimum sound insulation requirement TLmin≥x; Step S312: Initialize the parameter group, initialize a set of structural parameters X1, X2, ..., Xn; Step S313: Calculate the transmission loss TLi corresponding to the structural parameter Xi based on the structural performance prediction network; Step S314: Fitness assessment, assess whether the transmission loss TLi corresponding to each structural parameter Xi meets the minimum sound insulation requirement TLmin≥x; Step S315: Based on the fitness evaluation results, select structural parameters with higher fitness for crossover and mutation operations to generate new structural parameter sets; Step S316: Repeat the above steps until the maximum number of iterations MaxGen = 100 is reached; Step S317: Output the optimal structure parameters X.

Images