Method for designing thin-film acoustic metamaterial by using weighted KNN algorithm
Through the weighted KNN algorithm, the thin film acoustic metamaterial cell was designed, and the polymorphic anti-resonance mode and orthogonal experiment were introduced, which solved the problems of poor low-frequency sound insulation effect and heavy structure in the existing technology, and achieved low-frequency wide frequency Improved sound insulation performance and reduced cost.
Patent Information
- Application Number
- CN202310498581.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-05
- Publication Date
- 2024-11-05
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing thin film acoustic metamaterials have problems such as single resonance mode in terms of noise reduction and sound insulation, resulting in discontinuity of the anti-resonance mode, insufficient effective sound insulation frequency, narrow STL bandwidth, thick structure, and high processing cost.
Thin film acoustic metamaterial cells are designed using weighted KNN algorithm, and polymorphic anti-resonance mode is introduced through local resonance theory and mechanical characteristics. Combined with orthogonal experiments, hierarchical analysis method and K-Fold cross-validation, the widening and isolation of sound insulation bandwidth is achieved. The increase in sound volume and the weight of factors affecting sound insulation performance is determined through classified prediction.
The design of low-frequency wide band sound insulation materials is realized, the accuracy and efficiency of sound insulation performance is improved, the thickness and processing cost of the structure are reduced, and the range of sound insulation frequency is broadened.
Smart Images

Figure CN118897982A_ABST
Abstract
Description
Technical field:
[0001] The present invention belongs to the field of acoustic metamaterials, and is particularly suitable for the field of sound insulation and noise reduction of thin film acoustic metamaterials (MAMs). The present invention aims to design a thin film acoustic metamaterial (MAM) unit cell using a weighted KNN algorithm. The invention is a method for helping researchers to accurately predict the sound insulation performance of a unit cell according to various parameters in the design of thin film acoustic metamaterials. Background technology:
[0002] In modern society, noise pollution generated by industrial production, construction, transportation and social life has caused many adverse effects on people and the surrounding environment. Traditional noise reduction and sound absorption materials such as mineral wool, glass wool, felt, wood wool sound-absorbing panels, etc. have good noise reduction effects in medium and high frequencies, but are limited by the classical wave propagation theory, that is, the higher the density of the material, the better the sound insulation and sound absorption effect. Therefore, the mass and volume of traditional materials are restricted by space, and the effect is not good in the field of low-frequency noise reduction. Acoustic metamaterials are a type of composite materials composed of artificial structures. Their double negative properties, namely negative equivalent mass density and negative equivalent modulus, can better meet the regulation requirements of medium and low frequency sound waves. In the past two decades, machine learning has gradually emerged and developed, and machine learning has been used as a tool in the structural design of acoustic metamaterials.
[0003] Acoustic metamaterials have three important properties: 1) composite materials with artificial structures; 2) anomalous properties that natural materials cannot possess; 3) artificial structures determine the final properties rather than the material itself. Under the double negative properties, sound waves of a specific frequency will be completely reflected when incident vertically on the surface of the material, and sound waves incident at a certain angle will also travel in a refraction path different from that of conventional materials. There are three main forms of acoustic metamaterial development: thin film type, curled space type and Helmholtz resonator. There are many achievements in the theory and application of these three directions. The material of the thin film structure is relatively soft, and a resonator is attached to the membrane. The sound wave propagates to the film to resonate. The resonator absorbs kinetic energy to generate movement, reducing the vibration amplitude of the film, thereby achieving the effect of reducing noise. Membrane-type acoustic metamaterials (MAM) can effectively control low-frequency noise and are small and thin in size, which broadens its application scenarios. It is a major research direction of acoustic metamaterials in recent years.
[0004] The first thin-film acoustic metamaterial with negative dynamic mass density is composed of a rigid frame, an elastic membrane, and a mass block, and can produce a phenomenon of almost complete reflection of sound waves at a specific frequency, showing a strong sound attenuation capability. The structural unit of a flexible film with a rigid frame supporting a mass body has become a major design direction for thin-film metamaterials. Inspired by this, the bionic butterfly-shaped mass block thin-film acoustic metamaterial uses the resonance effect of the mass block to achieve ultra-high sound absorption in multiple frequency bands under ultra-wide frequencies, proving that thin-film acoustic metamaterials can be applied to various low-frequency noise reduction environments. In previous studies, computer simulations were performed on the parameters that affect the sound insulation performance of thin films. Physical software was used to simulate and test the prestress, thickness, edge length, and mass block distribution position of the thin film, and the effects of various parameters on the sound insulation performance were obtained. The effects of thin films of different mass and stiffness on the sound absorption effect were tested, and the results showed that the mass of the film had a significant effect on the sound absorption effect, and the film material with high density had better sound absorption effect. The factors affecting the low-frequency sound insulation of thin film acoustic metamaterials were studied in depth, and the influence of film parameters was studied from a theoretical perspective. The simulation test was used to verify that the analysis showed that small-sized thin film materials had better sound insulation effect and a wider sound insulation range, and the prestress applied to the film also had a certain effect on the sound insulation frequency peak. These studies have guiding significance for the preparation of thin film acoustic metamaterials, and also point out certain improvement directions for composite acoustic metamaterials, further promoting the practical progress of metamaterials.
[0005] However, existing thin-film acoustic metamaterials still have some drawbacks: in terms of noise reduction and sound insulation, the single resonance mode causes the anti-resonance mode to not exist continuously, the effective sound insulation frequency is not low enough, and the STL bandwidth is narrow; in terms of practicality, although the unit cells of existing thin-film metamaterials are relatively light and thin, the large-size structure is heavy, the rigid frame is difficult to support, and the processing cost is high.
[0006] Therefore, the present invention proposes a method for designing thin film acoustic metamaterials using a weighted KNN algorithm. The basic idea of this method is to design a lightweight thin film acoustic metamaterial (MAM) unit cell structure, study the sound transmission loss (STL), influencing factors and applications of the unit cell, and use weighted KNN to achieve classification and prediction of the sound insulation performance of the MAM unit cell. Firstly, according to the local resonance theory and mechanical properties, combined with the previous thin film acoustic metamaterials, an innovative design is carried out, and the multi-state anti-resonance mode is introduced to achieve the widening of the sound insulation bandwidth and the improvement of the sound insulation amount; secondly, the sound transmission loss (STL) and multi-state anti-resonance mode of the unit cell are analyzed; for the established MAM unit cell, the single influencing factor analysis of the sound insulation performance is carried out, and the influence of each factor on the sound insulation performance is obtained; then, the multi-influencing factor analysis of the sound insulation performance is carried out, and the factors with greater influence on the sound insulation performance are selected for orthogonal experiments to find the optimal solution and clarify the primary and secondary order of the factors affecting the sound insulation performance; then, the hierarchical analysis method is used to determine the specific weights of the factors affecting the sound insulation performance by constructing a judgment matrix and hierarchical total sorting; finally, the simulation data set is divided into three categories according to the sound insulation performance, and the weighted KNN algorithm classification model is used for classification prediction, and the K-Fold cross-validation is used to obtain the accuracy curve of the hyperparameter (k value in the KNN algorithm), and the k value with the highest classification accuracy is obtained, which proves the feasibility of using the weighted KNN algorithm to realize the performance prediction of thin film acoustic metamaterials. Summary of the invention:
[0007] The present invention provides a method for designing thin-film acoustic metamaterials in combination with a weighted KNN algorithm, which provides a new idea for the design of low-frequency broadband sound insulation materials. By using orthogonal experiments, hierarchical analysis method, weighted KNN algorithm, and K-Fold cross-validation, researchers can accurately predict the sound insulation performance of a unit cell according to various parameters in the design of thin-film acoustic metamaterials, and select appropriate parameters for design according to the sound insulation performance requirements.
[0008] To achieve the above objectives, the present invention proposes a method for designing thin film acoustic metamaterials (MAMs) using a weighted KNN algorithm. First, based on the local resonance theory and mechanical properties, combined with the previous MAM unit cell, an innovative design is carried out, and the multi-state anti-resonance mode is introduced to achieve the widening of the sound insulation bandwidth and the improvement of the sound insulation amount; secondly, the sound transmission loss (STL) and multi-state anti-resonance mode of the unit cell are analyzed; for the established MAM unit cell, a single influencing factor analysis of the sound insulation performance is carried out to obtain the influence of each factor on the sound insulation performance; then, a multi-factor analysis of the sound insulation performance is carried out, and the factors that have a greater impact on the sound insulation performance are selected for orthogonal experiments to find the optimal solution and clarify the primary and secondary order of the factors affecting the sound insulation performance; then, the hierarchical analysis method is used to determine the specific weights of the factors affecting the sound insulation performance by constructing a judgment matrix and hierarchical total sorting; finally, the data set is divided into three categories according to the sound insulation performance, and the weighted KNN algorithm classification model is used for classification prediction, and the K-Fold cross-validation is used to obtain the accuracy curve of the hyperparameter (k value in the KNN algorithm), and the k value with the highest classification accuracy is obtained, which proves the feasibility of combining the weighted KNN algorithm to realize MAM performance prediction. Specifically, the method includes the following steps:
[0009] 1) MAM unit cell design. MAM is a local resonance material. The unit cell is composed of a thin film and a mass block. When the sound wave propagates to the unit cell, the vibration causes the elastic film to deform, and a natural frequency is generated in the process of restoring the deformation. When the sound wave frequency is close to the natural frequency of the resonance unit, the resonance unit and the sound wave are strongly coupled, a band gap appears, and the sound energy is consumed. The outside of the MAM is connected by a layer of rigid body to achieve localized resonance. According to the mechanical properties, the symmetrical cross-swing arm structure can realize a variety of anti-symmetrical vibration modes. Using multiple resonators distributed in a MAM unit cell in a reasonable way can achieve multi-state anti-resonance modes.
[0010] 2) Sound transmission loss (STL) and anti-resonance mode analysis. The sound transmission coefficient τ represents the sound insulation performance of the sound insulation structure, and the transmitted sound power (w t ) and the incident sound power (w), the sound intensity of the transmitted sound wave (I t ) to the sound intensity of the incident sound wave (I) or the transmitted sound pressure (p t ) and the incident sound pressure (p), the sound transmission coefficient τ is calculated as follows:
[0011]
[0012] Since the sound transmission coefficient τ is very small, the sound transmission loss is used to reflect the sound insulation performance of MAM, which is recorded as STL in dB. The calculation formula of the sound transmission coefficient STL is as follows:
[0013]
[0014] The vibration mode and sound transmission loss (STL) curve were obtained by modeling and simulation using COMSOL Multiphysics finite element software, and the sound insulation bandwidth, sound insulation peak and anti-resonance mode were analyzed.
[0015] 3) Analysis of single influencing factors of sound insulation performance. The relevant parameters of the film and the resonator affect the natural vibration frequency of the MAM. The lower limit of the STL bandwidth is determined by the equivalent parameter characteristics of the MAM as a whole, and the upper limit of the STL bandwidth is determined by the equivalent parameter characteristics of the sub-unit cell where the resonator is located. Based on the simulation model of the established MAM unit cell, the effects of film thickness, film prestress, swing arm thickness, and resonator mass (length, width, and height) on its sound insulation performance are studied through the sound insulation bandwidth and sound insulation peak.
[0016] 4) Multi-factor analysis and orthogonal test to clarify the order of importance of each factor. Orthogonal test is carried out on the factors with greater impact on sound insulation performance, and the examination index is set as the normalized bandwidth δ around the peak of the STL curve where the sound insulation is greater than 20dB N , δ N The calculation formula is as follows:
[0017]
[0018] where f e1 is the starting frequency, f s1 is the stop frequency, f b is the peak frequency. Select a suitable orthogonal table for design and use L n (t c ), where L is the code of the orthogonal array, n is the number of experiments, t is the number of levels of the factor, and c is the maximum number of factors. Determine the experimental plan, use COMSOL Multiphysics finite element software to simulate the results, analyze the experimental results, calculate the mean difference, determine the optimal plan, calculate the range, and clarify the order of importance of each factor.
[0019] 5) Assign weights to factors using the analytic hierarchy process. According to the order of importance of the factors in the orthogonal test, use the analytic hierarchy process to assign corresponding weights to different factors. Establish a hierarchical structure model, construct a judgment matrix for each level, perform a single ranking of the levels and test their consistency. Finally, perform a total ranking of the levels and test their consistency. Obtain the weights of each factor in the solution level to the target level.
[0020] 6) Weighted KNN algorithm realizes classification prediction. For the unknown category of the test sample, first find the k known category samples with the closest distance (Euclidean distance) to the test sample in the training set, and then count the number of times the category to which the k training samples belong appears. The category with the largest number of occurrences is the category of the test sample.(i) and x (j) The calculation formula of weighted distance is:
[0021]
[0022] n is the dimension of the sample data. Perform K-Fold cross validation on the sample data set, divide all samples into K sample subsets of equal size, traverse these K subsets in turn, use the current subset as the validation set each time, and all other samples as the training set, perform weighted KNN model training and accuracy evaluation, and finally use the average value of K evaluations as the final evaluation index to obtain the best classification model.
[0023] Further, the specific steps of the above step 4) are as follows:
[0024] Step 4)-1: Starting state;
[0025] Step 4)-2: Clearly define the evaluation index as relative bandwidth δ N ;
[0026] Step 4)-3: Determine the factors that have a greater impact on sound insulation performance and select the factor level table;
[0027] Step 4)-4: Select a suitable orthogonal table L n (t c ) to design the header;
[0028] Step 4)-5: COMSOL Multiphysics finite element software simulation to obtain the orthogonal array results;
[0029] Step 4)-6: Calculate the mean difference K for each factor t i and range R i , i is the number of factors, t is the number of levels;
[0030] Step 4)-7: Determine the optimal solution based on the maximum mean difference of each factor, and judge whether the optimal solution is in the orthogonal table. If so, execute step 4)-9; otherwise, execute step 4)-8;
[0031] Step 4)-8: COMSOL Multiphysics finite element software simulation for verification test;
[0032] Step 4)-9: Obtain the optimal solution and clarify the priority order of each factor;
[0033] Step 4)-10: End state.
[0034] Further, the specific steps of the above step 5) are as follows:
[0035] Step 5)-1: Starting state;
[0036] Step 5)-2: Establish a hierarchical model, set the target layer (highest layer) to sound insulation performance (O); set the criterion layer (middle layer): film (C1), resonator (C2); set the bottom layer (scheme layer) to film prestress (P1), film thickness (P2), resonator length (P3), resonator width (P4), resonator height (P5);
[0037] Step 5)-3: Construct a judgment matrix between the target layer and the criterion layer, and perform hierarchical single sorting;
[0038] Step 5)-4: Calculate the consistency CR: CR = CI / RI, where CI = (λ max -n) / (n-1), λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, RI is found through the average random consistency index table, and it is determined whether CR is less than 0.1. If so, execute step 5)-6; otherwise, execute step 5)-5;
[0039] Step 5)-5: Adjust the judgment matrix between the target layer and the criterion layer;
[0040] Step 5)-6: Construct a judgment matrix between the criterion layer and the solution layer, and perform hierarchical single sorting;
[0041] Step 5)-7: Calculate the consistency CR: CR = CI / RI, where CI = (λ max -n) / (n-1), λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, RI is found through the average random consistency index table, and it is determined whether CR is less than 0.1. If so, execute step 5)-9; otherwise, execute step 5)-8;
[0042] Step 5)-8: Adjust the judgment matrix between the criterion layer and the solution layer;
[0043] Step 5)-9: Perform a total hierarchical ranking of the solution layer and the highest layer;
[0044] Step 5)-10: Check the consistency of the total hierarchical order. where α1, α2, ..., α j is the weight of each factor in the criterion layer, CI(j), RI(j) are the single sorting consistency index and average random consistency index in the criterion layer, j = 1, 2, 3, ..., n. Determine whether CR is less than 0.1, if so, execute step 5)-11; otherwise, execute step 5)-3;
[0045] Step 5)-11: Calculate the weight of each factor in the solution layer to the target layer;
[0046] Step 5)-12: End state.
[0047] Further, the specific steps of the above step 6) are as follows:
[0048] Step 6)-1: Starting state;
[0049] Step 6)-2: Import the data set and normalize the data;
[0050] Step 6)-3: Set the number of iterations K of K-fold cross validation;
[0051] Step 6)-4: Randomly divide the data set into K subsets according to the category ratio;
[0052] Step 6)-5: Traverse K subsets, each time use one subset as the test set, and the remaining K-1 subsets as the training set;
[0053] Step 6)-6: Set the value range of the hyperparameter k of weighted KNN;
[0054] Step 6)-7: Determine whether k is within the value range, if so, execute step 6)-8; otherwise, execute step 6)-16;
[0055] Step 6)-8: Determine whether the number of iterations is less than K, if so, execute step 6)-9; otherwise, execute step 6)-14;
[0056] Step 6)-9: Traverse the test set and calculate the weighted Euclidean distance between the test sample and the training set;
[0057] Step 6)-10: Sort the weighted Euclidean distances from small to large;
[0058] Step 6)-11: Count the categories of the k training samples with the smallest distance;
[0059] Step 6)-12: Obtain the classification of each sample in the validation set and obtain the accuracy;
[0060] Step 6)-13: number of iterations + 1;
[0061] Step 6)-14: Get the average of the K fold accuracy;
[0062] Step 6)-15: k+1;
[0063] Step 6)-16: Draw the accuracy curve of the hyperparameter k and select the hyperparameter k value with the highest accuracy;
[0064] Step 6)-17: End state. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 The present invention is a flowchart of a method for designing thin film acoustic metamaterials using a weighted KNN algorithm.
[0066] Figure 2 for Figure 1 Schematic diagram of the basic structure of the MAM unit cell design.
[0067] Figure 3 for Figure 2 COMSOL Multiphysics finite element software simulation model of the MAM unit cell.
[0068] Figure 4 for Figure 1 Analysis of sound transmission loss (STL) curve.
[0069] Figure 5 for Figure 1 Analysis of antiresonance modes in structures.
[0070] Figure 6 for Figure 1 Analysis of single influencing factors of medium sound insulation performance.
[0071] Figure 7 for Figure 1 A flow chart showing the order of importance of the factors that affect the experiment.
[0072] Figure 8 for Figure 1 Flowchart for assigning weights to factors in the AHP process.
[0073] Fig. 9 for Figure 8 Middle hierarchy model.
[0074] Fig.10 for Figure 1 Flowchart of the weighted KNN algorithm for classification prediction.
[0075] Fig.11 for Fig.10 Schematic diagram of the iterations of K-fold cross validation. DETAILED DESCRIPTION
[0076] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.
[0077] A method for designing a thin film acoustic metamaterial using a weighted KNN algorithm is characterized by comprising the following steps.
[0078] S 1 designs the MAM unit cell structure. According to the theory of local resonance materials, when sound waves propagate to the unit cell, the vibration causes the elastic film to deform, and a natural frequency is generated in the process of restoring the deformation. When the sound wave frequency is close to the natural frequency of the resonance unit, a band gap appears, consuming the sound energy. According to the mechanical properties, the symmetrical cross-swing arm structure can realize a variety of anti-symmetrical vibration modes. Using multiple resonators distributed in a MAM unit cell in a reasonable way can realize multi-state anti-resonance modes;
[0079] S2 performs sound transmission loss (STL) and anti-resonance mode analysis. The sound transmission coefficient τ represents the sound insulation performance of the sound insulation structure. Since the sound transmission coefficient τ is very small, the sound transmission loss (STL) is used to reflect the sound insulation performance of the MAM. The vibration mode and sound transmission loss (STL) curve are obtained through modeling and simulation with finite element software, and the sound insulation bandwidth, sound insulation peak, and anti-resonance mode are analyzed;
[0080] S3 analyzes the single influencing factors of sound insulation performance. The relevant parameters of the film and resonator affect the natural vibration frequency of the MAM. Based on the simulation model of the established MAM unit cell, the effects of film thickness, film prestress, swing arm thickness, and resonator mass on its sound insulation performance are studied through the sound insulation bandwidth and sound insulation peak:
[0081] S4 conducts multi-factor analysis and orthogonal tests to clarify the order of importance of each factor. Orthogonal tests are conducted on factors that have a significant impact on sound insulation performance. The test index is set as the normalized bandwidth of sound insulation greater than 20dB around the peak of the STL curve. A suitable orthogonal table is selected for design, and the test plan is determined. The simulation results are analyzed through finite element software, the mean difference is calculated, the optimal solution is determined, the range is calculated, and the order of importance of each factor is clarified;
[0082] S5 uses the analytic hierarchy process to assign factor weights. According to the order of priority of the factors in the orthogonal test, the analytic hierarchy process is used to assign corresponding weights to different factors. A hierarchical structure model is established, and a judgment matrix of each level is constructed. A single hierarchical sort is performed and its consistency is tested. Finally, a total hierarchical sort is performed and its consistency is tested. The weight of each factor in the solution layer to the target layer is obtained;
[0083] S6 performs weighted KNN algorithm to achieve classification prediction. For the unknown category of the test sample, first obtain the k known category samples with the closest distance to the test sample in the training set, and then count the number of times the category to which the k training samples belong appears, and the category with the largest number of appearances is the category of the test sample. The sample data set is subjected to K-Fold cross validation, and the weighted KNN model is trained and the accuracy is evaluated. Finally, the average value of K evaluations is used as the final evaluation index to obtain the best classification model.
[0084] Figure 1 A flowchart for designing thin film acoustic metamaterials using the weighted KNN algorithm.
[0085] Figure 2 Schematic diagram of the basic structure of a MAM unit cell.
[0086] Figure 3 This is the simulation model of the MAM unit cell.
[0087] Figure 4 This is the sound transmission loss (STL) curve analysis.
[0088] Figure 5 For anti-resonance mode analysis.
[0089] Figure 6 This is a single factor analysis on sound insulation performance.
[0090] Figure 7 A flowchart to clarify the order of importance of various factors in orthogonal experiments. The specific steps are as follows:
[0091] Step 1: Starting state; Step 2: Clearly define the test indicator as relative bandwidth δ N ; Step 3: Determine the factors that have a greater impact on sound insulation performance and select the factor level table; Step 4: Select a suitable orthogonal table L n (t c ) to design the table header; Step 5: COMSOL Multiphysics finite element software simulation to obtain the orthogonal table results; Step 6: Calculate the mean difference K of each factor t i and range R i , i is the number of factors, t is the number of levels; Step 7: Determine the optimal solution based on the maximum mean difference of each factor, and judge whether it is in the orthogonal table. If so, execute step 9; otherwise, execute step 8; Step 8: COMSOL Multiphysics finite element software simulation for verification test; Step 9: Obtain the optimal solution and clarify the order of importance of each factor; Step 10: End state.
[0092] Figure 8 Flowchart for assigning factor weights to the AHP method. The specific steps are as follows:
[0093] Step 1: Starting state; Step 2: Establish a hierarchical model, set the target layer (highest layer) to sound insulation performance (O); set the criterion layer (middle layer): film (C1), resonator (C2); set the bottom layer (scheme layer) to film prestress (P1), film thickness (P2), resonator length (P3), resonator width (P4), resonator height (P5); Step 3: Construct a judgment matrix between the target layer and the criterion layer, and perform hierarchical single sorting; Step 4: Calculate the consistency CR: CR = CI / RI, where CI = (λ max -n) / (n-1), λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, RI is found through the average random consistency index table, and it is judged whether CR is less than 0.1. If so, execute step 6; otherwise, execute step 5; Step 5: adjust the judgment matrix between the target layer and the criterion layer; Step 6: construct the judgment matrix between the criterion layer and the solution layer, and perform hierarchical single sorting; Step 7: calculate the consistency CR: CR = CI / RI, where CI = (λ max -n) / (n-1), λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, RI is found through the average random consistency index table, and it is judged whether CR is less than 0.1. If so, execute step 9; otherwise, execute step 8; Step 8: adjust the judgment matrix between the criterion layer and the solution layer; Step 9: perform a total hierarchical ranking of the solution layer and the highest layer; Step 10: perform a consistency check on the total hierarchical ranking, where α1, α2, ..., α j is the weight of each factor in the criterion layer, CI(j), RI(j) are the single sorting consistency index and average random consistency index in the criterion layer, j = 1, 2, 3, ..., n. Determine whether CR is less than 0.1, if so, execute step 11; otherwise, execute step 3; Step 11: Calculate the weight of each factor in the solution layer to the target layer; Step 12: End state.
[0094] Fig. 9 It is a hierarchical model.
[0095] Fig.10 The flowchart for implementing classification prediction for the weighted KNN algorithm is as follows:
[0096] Step 1: Starting state; Step 2: Import the data set and normalize the data; Step 3: Set the number of iterations K of K-fold cross validation; Step 4: Randomly divide the data set into K subsets according to the category ratio; Step 5: Traverse the K subsets, each time using one subset as the test set and the remaining K-1 subsets as the training set; Step 6: Set the value range of the hyperparameter k of weighted KNN; Step 7: Determine whether k is within the value range, if so, execute step 8; otherwise, execute step 16; Step 8: Determine whether the number of iterations is less than K, if so, then Execute step 9; otherwise, execute step 14; Step 9: traverse the test set and calculate the weighted Euclidean distance between the test sample and the training set; Step 10: Sort the weighted Euclidean distance from small to large; Step 11: Count the categories of the k training samples with the smallest distance; Step 12: Obtain the classification of each sample in the validation set and obtain the accuracy; Step 13: Number of iterations + 1; Step 14: Obtain the average accuracy of K folds; Step 15: k+1; Step 16: Draw the accuracy curve of the hyperparameter k and select the hyperparameter k value with the highest accuracy; Step 17: End state.
[0097] Fig.11 Schematic diagram of the K-fold cross validation iteration.
[0098] In summary, the purpose of the present invention is to design a thin film acoustic metamaterial (MAM) unit cell using a weighted KNN algorithm. Analyzing the final evaluation index and obtaining the best classification model can help researchers accurately predict the sound insulation performance of the unit cell according to various parameters in the design of thin film acoustic metamaterials, and select appropriate parameters for design according to the sound insulation performance requirements.
Claims
1. A method for designing a thin film acoustic metamaterial (MAM) using a weighted KNN algorithm, characterized in that: Firstly, according to the local resonance theory and mechanical properties, combined with the previous MAM unit cell, an innovative design is carried out, and the multi-state anti-resonance mode is introduced to achieve the widening of the sound insulation bandwidth and the improvement of the sound insulation amount; secondly, the sound transmission loss (STL) and multi-state anti-resonance mode of the unit cell are analyzed; for the established MAM unit cell, the single influencing factor of the sound insulation performance is analyzed to obtain the influence of various factors on the sound insulation performance; After that, a multi-factor analysis of sound insulation performance was conducted, and the factors that had a greater impact on the sound insulation performance were selected for orthogonal experiments to find the optimal solution and clarify the order of priority of the factors affecting the sound insulation performance. Then, the analytic hierarchy process was used to construct a judgment matrix and perform a hierarchical total ranking to determine the specific weights of the factors affecting the sound insulation performance. Finally, the data set was divided into three categories according to the sound insulation performance, and the weighted KNN algorithm classification model was used for classification prediction. The K-Fold cross-validation was used to obtain the accuracy curve of the hyperparameter (the k value in the KNN algorithm), and the k value with the highest classification accuracy was obtained, proving the feasibility of combining the weighted KNN algorithm to realize MAM performance prediction. The method includes the following steps: 1) MAM unit cell design. MAM is a local resonance material. The unit cell is composed of a thin film and a mass block. When the sound wave propagates to the unit cell, the vibration causes the elastic film to deform, and a natural frequency is generated in the process of restoring the deformation. When the sound wave frequency is close to the natural frequency of the resonance unit, the resonance unit and the sound wave are strongly coupled, a band gap appears, and the sound energy is consumed. The outside of the MAM is connected by a layer of rigid body to achieve localized resonance. According to the mechanical properties, the symmetrical cross-swing arm structure can realize a variety of anti-symmetrical vibration modes. Using multiple resonators distributed in a MAM unit cell in a reasonable way can realize multi-state anti-resonance modes; 2) Sound transmission loss (STL) and anti-resonance mode analysis. The sound transmission coefficient τ represents the sound insulation performance of the sound insulation structure, and the transmitted sound power (w t ) and the incident sound power (w), the sound intensity of the transmitted sound wave (I t ) to the sound intensity of the incident sound wave (I) or the transmitted sound pressure (p t ) and the incident sound pressure (p), the sound transmission coefficient τ is calculated as follows: Since the sound transmission coefficient τ is very small, the sound transmission loss is used to reflect the sound insulation performance of MAM, which is recorded as STL in dB. The calculation formula of the sound transmission coefficient STL is as follows: The vibration mode and sound transmission loss (STL) curves were obtained by modeling and simulation using COMSOL Multiphysics finite element software, and the sound insulation bandwidth, sound insulation peak, and anti-resonance mode were analyzed. 3) Analysis of single influencing factors of sound insulation performance. The relevant parameters of the film and the resonator affect the natural vibration frequency of the MAM. The lower limit of the STL bandwidth is determined by the equivalent parameter characteristics of the MAM as a whole, and the upper limit of the STL bandwidth is determined by the equivalent parameter characteristics of the sub-unit cell where the resonator is located. Based on the simulation model of the established MAM unit cell, the effects of film thickness, film prestress, swing arm thickness, and resonator mass (length, width, and height) on its sound insulation performance are studied through the sound insulation bandwidth and sound insulation peak value; 4) Multi-factor analysis and orthogonal test to clarify the order of importance of each factor. Orthogonal test is carried out on the factors with greater impact on sound insulation performance, and the examination index is set as the normalized bandwidth δ around the peak of the STL curve where the sound insulation is greater than 20dB N , δ N The calculation formula is as follows: where f e1 is the starting frequency, f s1 is the stop frequency, f b is the peak frequency. Select a suitable orthogonal table for design and use L n (t c ), where L is the code of the orthogonal array, n is the number of experiments, t is the number of levels of the factor, and c is the maximum number of factors. Determine the experimental plan, simulate the results through COMSOL Multiphysics finite element software, analyze the experimental results, calculate the mean difference, determine the optimal plan, calculate the range, and clarify the order of importance of each factor; 5) Assign weights to factors using the analytic hierarchy process. According to the order of importance of the factors in the orthogonal test, use the analytic hierarchy process to assign corresponding weights to different factors. Establish a hierarchical structure model, construct a judgment matrix for each level, perform a single ranking of the levels and test their consistency. Finally, perform a total ranking of the levels and test their consistency. Obtain the weights of each factor in the solution level to the target level; 6) Weighted KNN algorithm realizes classification prediction. For the unknown category of the test sample, first find the k known category samples with the closest distance (Euclidean distance) to the test sample in the training set, and then count the number of times the category to which the k training samples belong appears. The category with the largest number of occurrences is the category of the test sample. (i) and x (j) The calculation formula of weighted distance is: n is the dimension of the sample data. Perform K-Fold cross validation on the sample data set, divide all samples into K sample subsets of equal size, traverse these K subsets in turn, use the current subset as the validation set each time, and all other samples as the training set, perform weighted KNN model training and accuracy evaluation, and finally, take the average value of K evaluations as the final evaluation index to obtain the best classification model.
2. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 1), the MAM unit cell structure is designed. According to the theory of local resonance materials, when sound waves propagate to a unit cell, the vibration causes the elastic film to deform, and a natural frequency is generated in the process of restoring the deformation. When the sound wave frequency is close to the natural frequency of the resonance unit, a band gap appears, consuming the sound energy. According to the mechanical properties, the symmetrical cross-arm structure can realize a variety of anti-symmetrical vibration modes. Using multiple resonators distributed in a MAM unit cell in a reasonable way can realize multi-state anti-resonance modes.
3. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 2), sound transmission loss (STL) and anti-resonance mode analysis are performed. The sound transmission coefficient τ represents the sound insulation performance of the sound insulation structure. Since the sound transmission coefficient τ is very small, the sound transmission loss (STL) is used to reflect the sound insulation performance of the MAM. The vibration mode and sound transmission loss (STL) curve are obtained by modeling and simulation using finite element software, and the sound insulation bandwidth, sound insulation peak, and anti-resonance mode are analyzed.
4. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 3), a single factor influencing the sound insulation performance is analyzed. The relevant parameters of the film and the resonator affect the natural vibration frequency of the MAM. Based on the simulation model of the established MAM unit cell, the effects of film thickness, film prestress, swing arm thickness, and resonator mass on its sound insulation performance are studied through the sound insulation bandwidth and sound insulation peak.
5. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 4), multi-factor analysis is performed, and orthogonal tests are used to clarify the order of importance of each factor. Orthogonal tests are performed on factors that have a greater impact on sound insulation performance. The examination index is set as the normalized bandwidth of sound insulation around the peak of the STL curve greater than 20dB. A suitable orthogonal table is selected for design, and the test plan is determined. The simulation results are analyzed through finite element software, the mean difference is calculated, the optimal plan is determined, the range is calculated, and the order of importance of each factor is clarified.
6. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 5), the analytic hierarchy process is used to assign weights to the factors. According to the order of priority of the factors in the orthogonal test, the analytic hierarchy process is used to assign corresponding weights to different factors. A hierarchical structure model is established, and a judgment matrix of each level is constructed. The hierarchical single sorting is performed and its consistency is tested. Finally, the hierarchical total sorting is performed and its consistency is tested. Get the weight of each factor in the solution layer to the target layer.
7. The method for designing thin film acoustic metamaterials using a weighted KNN algorithm according to claim 1, characterized in that: In step 6), the weighted KNN algorithm is used to achieve classification prediction. For the unknown category of the test sample, firstly, the k known category samples with the closest distance to the test sample are obtained in the training set, and then the number of occurrences of the category to which the k training samples belong is counted, and the category with the largest number of occurrences is the category of the test sample. The sample data set is subjected to K-Fold cross validation, and the weighted KNN model is trained and the accuracy is evaluated. Finally, the average value of the K evaluations is used as the final evaluation index to obtain the best classification model.
Citation Information
Cited By
High-dimensional metamaterial acoustic performance prediction method and system based on space mapping multi-fidelity modeling
CN119763740A
Acoustic metamaterial structure with pre-tension thin film and design method
CN121565128A
