Production method of composite sound absorption floor suitable for piano room rehearsal hall and studio

Abstract

The invention provides a production method of a composite sound-absorbing floor suitable for a piano room rehearsal hall and a studio, and belongs to the technical field of composite sound-absorbing floors. By collecting acoustic environment data of the piano room rehearsal hall or the studio, a key frequency and a decibel gathering area are accurately identified to form a sound gathering matrix; inputting the matrix into a pre-trained deep learning model to calculate an optimal structure parameter vector, designing a composite structure of a microporous UHPC plate, basalt cloth, a through hole ceramic plate and a cavity according to the optimal structure parameter vector, forming a seepage water discharge groove under the composite structure, realizing sound absorption, fire prevention, water seepage, dirt resistance, friction resistance and treading, determining parameters of each layer and keeping a proper interlayer spacing, and finally obtaining a composite structure of the microporous UHPC plate, the basalt cloth, the through hole ceramic plate and the cavity. Structural parameters are optimized through an actual measurement feedback mechanism to determine micro-pore UHPC: the pore diameter is 0.5 mm, the pore pitch is 15.6 mm, and the perforation rate is 8%, and finally batch production and sampling testing are carried out to ensure that the designed sound absorption effect is stably achieved, so that efficient customized sound absorption floor production aiming at a specific acoustic environment is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of composite sound-absorbing floors, and more specifically, relates to a production method of a composite sound-absorbing floor applicable to rehearsal halls and studios of music rooms. Background Art

[0002] As professional audio working environments, rehearsal halls and studios of music rooms have extremely high requirements for acoustic performance. Traditional sound-absorbing floor technologies mainly use porous materials (such as mineral wool boards and fiberglass), resonance sound-absorbing structures (such as perforated boards), and micro-perforated sound-absorbing boards, etc. to achieve acoustic regulation. These sound-absorbing floors are generally produced in prefabricated standard specifications, and the sound-absorbing effects in specific frequency bands are achieved through the combination of different materials, and are widely used in places such as concert halls and conference rooms.

[0003] However, traditional sound-absorbing floor designs are usually based on general acoustic parameter standards and are difficult to accurately adapt to the unique acoustic environments of specific rehearsal halls and studios of music rooms. In the prior art, the determination of the structural parameters of sound-absorbing floors mainly relies on the empirical judgment of professional acoustic engineers or is adjusted through limited experiments. This method lacks accurate analysis of the sound frequency spectrum distribution in the actual environment, resulting in a large deviation between the sound-absorbing effect and the actual requirements, especially when dealing with specific frequency aggregation regions.

[0004] For specific places such as rehearsal halls and studios of music rooms with extremely high requirements for acoustic performance, how to accurately design and optimize the structural parameters of the composite sound-absorbing floor based on actual acoustic environment data to achieve targeted sound-absorbing effects has become a technical problem to be solved urgently at present. That is to say, there is a technical problem in the prior art that an efficient sound-absorbing floor cannot be customized for the specific acoustic environment of rehearsal halls and studios of music rooms. Summary of the Invention

[0005] In view of this, the present invention provides a production method of a composite sound-absorbing floor applicable to rehearsal halls and studios of music rooms, which can solve the technical problem in the prior art that an efficient sound-absorbing floor cannot be customized for the specific acoustic environment of rehearsal halls and studios of music rooms.

[0006] The present invention is implemented as follows: The present invention provides a production method for a composite sound-absorbing floor suitable for a piano room, a rehearsal hall and a studio, comprising: collecting sound frequency samples in a piano room, a rehearsal hall and a studio, and establishing an initial acoustic environment database; analyzing the collected sound spectrum data, determining key frequency aggregation segments and decibel aggregation segments, and forming a sound aggregation matrix; inputting the sound aggregation matrix into a pre-trained sound-absorbing floor parameter model, and calculating a structural parameter vector required for an optimal sound absorption effect; determining microporous UHPC surface layer parameters according to the structural parameter vector; designing a basalt fiber middle layer according to low-frequency band data of the sound aggregation matrix; preparing a through-hole ceramic bottom plate according to mid-frequency band data of the sound aggregation matrix; assembling a microporous UHPC (ultra-high performance concrete) surface layer, a basalt fiber middle layer and a through-hole ceramic bottom plate; testing the performance of the composite sound-absorbing floor sample in a real acoustic environment and adjusting the structural parameter vector; and producing according to the optimized structural parameter vector.

[0007] Among them, the sound aggregation matrix refers to the sound energy aggregation in different frequency ranges expressed in matrix form after sampling and analyzing the acoustic environment. The rows of the matrix represent different frequency bands, the columns represent different decibel bands, and each element value represents the frequency or intensity of the frequency and decibel combination in the measurement environment.

[0008] Among them, the analysis of the collected sound spectrum data is specifically that, first, the sound spectrum data is grouped according to similarity through hierarchical clustering, and then the stratification results are optimized using the gray wolf hunting algorithm. By simulating the encirclement, pursuit and attack behaviors of the gray wolf, the precise positioning of the cluster center is achieved, thereby improving the accuracy of identifying sound frequency cluster segments and decibel cluster segments.

[0009] Among them, the microporous UHPC surface refers to the floor surface made of UHPC with tiny holes on the surface. UHPC material has extremely high strength and durability, and its microporous structure enables it to absorb medium and high frequency sound waves.

[0010] Among them, the structural parameter vector refers to a set of numerical values ​​that describe the structural characteristics of each component of the composite sound-absorbing floor, including micropore diameter, micropore spacing, micropore perforation rate, basalt fiber middle layer thickness, basalt fiber middle layer density, through-hole ceramic bottom plate thickness, through-hole diameter, through-hole rate and layer spacing.

[0011] Among them, the microporous UHPC surface parameters are determined according to the calculated structural parameter vector, including a micropore diameter range of 0.5 to 2.0 mm, a micropore spacing of 2.5 to 8.0 mm, and a micropore perforation rate of 5% to 25%.

[0012] Among them, the basalt fiber middle layer is designed based on the low-frequency band data of the sound aggregation matrix. Specifically, based on the low-frequency band data of the sound aggregation matrix, the thickness of the basalt fiber middle layer is designed to be 15 to 40 millimeters, and the density of the basalt fiber middle layer is adjusted to be between 45 and 120 kilograms per cubic meter.

[0013] Among them, the through-hole ceramic bottom plate is prepared according to the medium-frequency band data of the sound aggregation matrix. Specifically, according to the medium-frequency band data of the sound aggregation matrix, the through-hole ceramic bottom plate is prepared. The thickness of the through-hole ceramic bottom plate is 8 to 15 millimeters, the through-hole diameter is 3 to 8 millimeters, and the through-hole rate is 30% to 60%.

[0014] Among them, the structure of the pre-trained sound absorption floor parameter model is a deep learning model based on a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron. The input layer receives sound aggregation matrix data with a dimension of 300×200, and the output layer contains 9 neurons corresponding to 9 key parameters in the structure parameter vector respectively.

[0015] Among them, the training data set establishment step in the pre-training process of the sound absorption floor parameter model includes collecting the sound absorption performance test data of 500 composite sound absorption floor samples with different structural parameter combinations in a standard acoustic laboratory. The test frequency range is from 50 hertz to 20,000 hertz, divided into 300 frequency points, and the test decibel range is from 30 decibels to 110 decibels, divided into 200 decibel points.

[0016] Compared with the prior art, for a production method of a composite sound absorption floor applicable to a music practice room, rehearsal hall and studio, the proposed production method of the composite sound absorption floor realizes the precise production of a customized sound absorption floor for the acoustic environment of a specific music practice room and studio by constructing a closed-loop system of acoustic environment data acquisition - analysis - parameter optimization - structure design. This method first collects the sound spectrum data in the actual environment, uses hierarchical clustering and grey wolf hunting algorithm to accurately identify the key frequency and decibel aggregation areas, forms a sound aggregation matrix, and then calculates the optimal structural parameters through a pre-trained deep learning model.

[0017] Compared with the traditional design method relying on experience, the present invention realizes the direct mapping relationship from acoustic environment data to structural parameters, greatly improving the adaptability of the sound absorption floor to a specific environment. The three-layer composite structure design (microporous UHPC surface layer, basalt fiber middle layer, through-hole ceramic bottom plate) can achieve differential absorption of sound waves in different frequency bands, solving the problem that it is difficult for traditional single-structure sound absorption floors to process multi-band sounds simultaneously.

[0018] Through the method of the present invention, the produced composite sound absorption floor can accurately match the specific acoustic requirements of a music practice room and studio, effectively solving the technical problem of being unable to customize an efficient sound absorption floor for a specific acoustic environment, and realizing the optimal matching between the acoustic environment and the sound absorption structure. Description of the Drawings

[0019] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0020] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] As Figure 1 shown, it is a flowchart of a production method of a composite sound-absorbing floor applicable to a rehearsal hall and a studio in a music room. The method includes the following steps:

[0022] S01. Collect sound frequency samples in the rehearsal hall and studio in the music room, obtain sound spectrum data through Fourier transform, record the sound frequency distribution and decibel value in the environment, and establish an initial acoustic environment database;

[0023] S02. Analyze the collected sound spectrum data by using hierarchical clustering analysis method combined with grey wolf hunting algorithm to determine the key frequency aggregation segments and decibel aggregation segments, and form a sound aggregation matrix;

[0024] S03. Input the sound aggregation matrix into a pre-trained sound-absorbing floor parameter model, and calculate the structural parameter vector required for the best sound-absorbing effect;

[0025] S04. According to the calculated structural parameter vector, determine the parameters of the microporous UHPC surface layer, including the microporous diameter range of 0.5 to 2.0 mm, the microporous spacing of 2.5 to 8.0 mm, and the microporous perforation rate of 5% to 25%;

[0026] S05. Design the thickness of the basalt fiber middle layer to be 15 to 40 mm according to the low-frequency band data of the sound aggregation matrix, and adjust the density of the basalt fiber middle layer between 45 and 120 kg / m³;

[0027] S06. Prepare a through-hole ceramic bottom plate according to the medium-frequency band data of the sound aggregation matrix. The thickness of the through-hole ceramic bottom plate is 8 to 15 mm, the through-hole diameter is 3 to 8 mm, and the through-hole rate is 30% to 60%;

[0028] S07. Assemble the microporous UHPC surface layer, the basalt fiber middle layer and the through-hole ceramic bottom plate according to the parameter ratio determined by the structural parameter vector, and keep the layer spacing of 5 to 25 mm;

[0029] S08. Test the performance of the composite sound-absorbing floor sample in a real acoustic environment, obtain the sound absorption coefficient curve, and adjust the structural parameter vector through a feedback mechanism to form the final production parameters;

[0030] S09. Carry out batch production according to the optimized structural parameter vector. Optionally, perform sampling acoustic performance test on each batch of products to ensure that the designed sound absorption effect is stably achieved.

[0031] Among them, the sound concentration matrix specifically refers to the sound energy concentration in different frequency ranges represented in matrix form after sampling and analyzing the acoustic environment. The rows of the matrix represent different frequency bands, the columns represent different decibel bands, and each element value represents the frequency or intensity of the frequency and decibel combination in the measurement environment.

[0032] Among them, the hierarchical cluster analysis method combined with the gray wolf hunting algorithm specifically refers to an acoustic data processing method. First, the sound spectrum data is grouped according to similarity through hierarchical clustering, and then the gray wolf hunting algorithm is used to optimize the stratification results. By simulating the encirclement, pursuit and attack behaviors of the gray wolf, the cluster center can be accurately located, thereby improving the accuracy of identifying sound frequency cluster segments and decibel cluster segments.

[0033] Among them, the microporous UHPC surface specifically refers to the floor surface made of UHPC with tiny holes on the surface. UHPC material has extremely high strength and durability, and its microporous structure enables it to absorb medium and high frequency sound waves.

[0034] Among them, the structural parameter vector specifically refers to a set of numerical values ​​that describe the structural characteristics of each component of the composite sound-absorbing floor, including key parameters such as micropore diameter, micropore spacing, micropore perforation rate, basalt fiber middle layer thickness, basalt fiber middle layer density, through-hole ceramic bottom plate thickness, through-hole diameter, through-hole rate and layer spacing.

[0035] The specific structure of the pre-trained acoustic floor parameter model is a deep learning model based on a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron. The input layer receives sound aggregation matrix data with a dimension of 300×200. The first convolutional layer uses 64 5×5 convolutional kernels to extract low-level acoustic features. The second convolutional layer uses 128 3×3 convolutional kernels to extract intermediate acoustic features. The third convolutional layer uses 256 3×3 convolutional kernels to extract high-level acoustic features. The fourth convolutional layer uses 512 2×2 convolutional kernels for feature fusion. After the convolutional result passes through global average pooling, it is input into the first fully connected layer, which contains 1024 neurons for feature dimensionality reduction. The second fully connected layer contains 512 neurons for feature transformation. The third fully connected layer contains 256 neurons that map to the output layer. The output layer contains 9 neurons corresponding to the micropore diameter, micropore spacing, micropore perforation rate, middle layer thickness of basalt fiber, middle layer density of basalt fiber, thickness of the through-hole ceramic bottom plate, through-hole diameter, through-hole rate, and layer spacing in the structural parameter vector respectively; The steps for establishing the training data set during the pre-training process of the acoustic floor parameter model specifically include collecting the acoustic performance test data of 500 composite acoustic floor samples with different structural parameter combinations in a standard acoustic laboratory. The test frequency range is from 50 Hz to 20,000 Hz, divided into 300 frequency points, and the test decibel range is from 30 dB to 110 dB, divided into 200 decibel points. At the same time, record the sound frequency distribution and decibel values in 25 typical music practice rooms, rehearsal halls, and studios to form a sound aggregation matrix. Establish a corresponding relationship between the structural parameter vector of the composite acoustic floor sample, the corresponding acoustic performance data, and the sound aggregation matrix to form a training data set containing 12,500 groups of samples; The steps for pre-training the acoustic floor parameter model specifically include using the stochastic gradient descent algorithm with a batch size of 64 to initialize the training of the model. The learning rate is set to 0.001 and the cosine annealing strategy is used for attenuation. Use the five-fold cross-validation method to determine the optimal model hyperparameters. The training iteration times are set to 10,000 times, and the performance of the validation set is evaluated every 1,000 times. Introduce an early stopping mechanism to stop training when the validation set error does not decrease for 5 consecutive evaluations. Calculate the difference between the predicted value and the true value through the mean squared error loss function, and optimize the model weights through the error backpropagation algorithm. Finally, on the validation set, the prediction error of each parameter in the structural parameter vector meets the accuracy requirement of being less than 5%.

[0036] The following describes the specific implementation manners of the above steps in detail.

[0037] The specific implementation of step S01 is to sample the sound in the rehearsal hall of the music room and the studio using high-precision acoustic measurement equipment. First, at least 9 evenly distributed sampling points are set in the space using omnidirectional pickups. Each sampling point collects sound samples for no less than 30 minutes, the sampling frequency is set to 48,000 Hz, and the quantization accuracy is 24 bits. Then, the fast Fourier transform algorithm is applied to the collected original waveform data to convert the time-domain signal into a frequency-domain signal, and the sound spectrum data in the range of 50 Hz to 20,000 Hz is obtained, with a frequency resolution of not less than 1 Hz. Next, the spectrum data is segmented and statistically analyzed according to 1 / 3 octave, and at the same time, the sound pressure level distribution in the range of 30 dB to 110 dB is recorded, and the dB resolution is set to 0.5 dB. Finally, the spectrum data and dB data of all sampling points are weighted and averaged to form the acoustic environment feature vector of this space, which is stored in the initial acoustic environment database. The purpose of this step is to establish an accurate digital representation of the acoustic environment and provide a data basis for the subsequent optimization of the sound absorption structure.

[0038] The specific implementation of step S02 is to perform multi-dimensional analysis and processing on the spectrum data in the initial acoustic environment database. First, the spectrum data is represented in the form of a 300×200-dimensional matrix, where 300 rows represent the frequency dimension and 200 columns represent the dB dimension. Then, the hierarchical clustering analysis method is used to perform initial clustering on the spectrum data. The specific method is to calculate the Euclidean distance between each frequency point as the similarity measure, and use the minimum variance method (Ward method) to construct a clustering tree. The clustering number threshold is set to 5 to 15, and the optimal number of clusters is determined by the elbow method. Then, the gray wolf hunting algorithm is introduced to optimize the clustering result. After optimization, the clustering centers are extracted as the key frequency aggregation segments, and the dB data is also subjected to hierarchical clustering and gray wolf algorithm optimization to obtain the key dB aggregation segments. Finally, the frequency aggregation segments and dB aggregation segments are combined to form a sound aggregation matrix, and each element value in the matrix represents the energy density of the frequency-dB combination. The role of this step is to accurately identify the key frequency characteristics in the acoustic environment and provide a targeted basis for the subsequent sound absorption structure design.

[0039] The specific implementation of step S03 is to input the sound aggregation matrix into a pre-trained deep learning model for parameter calculation. First, the sound aggregation matrix is normalized so that all element values are mapped to the range of 0 to 1, reducing the impact of numerical differences on the model. Then, the normalized matrix is input into the pre-trained sound-absorbing floor parameter model, which adopts a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron. The input layer receives the sound aggregation matrix with a dimension of 300×200. The first convolutional layer uses 64 5×5 convolutional kernels and the ReLU activation function to extract low-level acoustic features; the second convolutional layer uses 128 3×3 convolutional kernels and the ReLU activation function to extract intermediate-level acoustic features; the third convolutional layer uses 256 3×3 convolutional kernels and the ReLU activation function to extract high-level acoustic features; the fourth convolutional layer uses 512 2×2 convolutional kernels and the ReLU activation function for feature fusion. A 2×2 max pooling operation is adopted between each convolutional layer to reduce the dimension of the feature map, and the batch normalization technique is applied to stabilize the training process. The output of the convolutional network is connected to the fully connected network after global average pooling. The first fully connected layer contains 1024 neurons for feature dimensionality reduction, uses the ReLU activation function and applies 50% Dropout to prevent overfitting; the second fully connected layer contains 512 neurons for feature transformation, also uses ReLU activation and 30% Dropout; the third fully connected layer contains 256 neurons mapped to the output layer, uses ReLU activation without Dropout; the output layer contains 9 neurons corresponding to the 9 parameters of the structural parameter vector, and uses a linear activation function to ensure continuous output values. After the forward propagation of the model is completed, the optimal structural parameter vector for a specific acoustic environment can be obtained. The purpose of this step is to transform complex acoustic features into specific structural parameters through a deep learning model, realizing the mapping from the acoustic environment to the physical structure.

[0040] The specific implementation of step S04 is to determine the specific parameters of the microporous UHPC surface layer according to the structural parameter vector. First, three parameters, namely micropore diameter, micropore spacing, and micropore perforation rate, are extracted from the structural parameter vector. Then, UHPC raw materials are prepared according to these parameters, and the mass ratio is: Portland cement 52.5%, mineral admixture 18.5%, fine aggregate 23.5%, high-range water reducer 1.5%, steel fiber 2.0%, and the water-cement ratio is controlled between 0.18 and 0.22. Next, the concrete is poured into a forming device with a microporous mold. The microporous mold adopts a detachable design. The micropore diameter is controlled within the range of 0.5 to 2.0 mm, which is determined according to the specific value of the structural parameter vector. The micropore spacing is set between 2.5 and 8.0 mm, and the micropore perforation rate is controlled between 5% and 25%. After forming, it is cured for 48 hours under standard conditions (temperature 20±2°C, relative humidity above 95%) and then demolded, followed by 7 days of standard curing. Finally, the formed microporous UHPC surface layer is cut and polished, with the thickness controlled between 5 and 12 mm, and the surface flatness deviation not exceeding 0.5 mm. The function of this step is to prepare a UHPC surface layer with a microporous structure, which is mainly used to absorb medium and high-frequency sound waves.

[0041] The specific implementation of step S05 is to design the basalt fiber middle layer based on the low-frequency data in the sound aggregation matrix. First, two parameters, namely the thickness and density of the basalt fiber middle layer, are extracted from the structural parameter vector. Then, basalt fibers with appropriate diameters are selected according to these parameters. The fiber diameter range is 7 to 13 μm, and the fiber length is 50 to 80 mm. Next, a basalt fiber mat is prepared by the air-laying process, controlling the random distribution of fiber directions. The fiber bonding uses thermosetting epoxy resin, and the resin addition amount is controlled between 3% and 7% of the fiber mass. During the forming process, the final density is controlled between 45 and 120 kg / m³ by adjusting the fiber feeding amount and compaction degree, and the specific value is determined according to the structural parameter vector. Finally, the basalt fiber mat is cut into the required shape and size, with the thickness controlled between 15 and 40 mm, and the thickness uniformity deviation not exceeding 2 mm. The purpose of this step is to prepare a basalt fiber middle layer with a specified thickness and density, which is mainly used to absorb low-frequency sound waves.

[0042] The specific implementation of step S06 is to prepare a through-hole ceramic bottom plate according to the frequency band data in the sound aggregation matrix. First, three parameters, namely the thickness of the through-hole ceramic bottom plate, the diameter of the through holes, and the through-hole rate, are extracted from the structural parameter vector. Then, a ceramic slurry is prepared, with the main components being 65% kaolin, 25% quartz, 8% feldspar, and 2% additive, and an appropriate amount of water is added to adjust it into a slurry with good fluidity. Next, a porous ceramic green body is prepared by the foaming method. Aluminum hydroxide is selected as the foaming agent, and the addition amount is 5% to 10% of the ceramic mass. The through-hole rate is adjusted between 30% and 60% by controlling the dosage of the foaming agent and the foaming time. After forming, it is dried at 80 to 100 °C for 24 hours, and then subjected to high-temperature sintering at 1100 to 1200 °C for 4 to 6 hours. Finally, the sintered ceramic bottom plate is cut and polished, with the thickness controlled between 8 and 15 mm, the diameter of the through holes controlled between 3 and 8 mm, and the surface flatness deviation not exceeding 0.8 mm. The function of this step is to prepare a ceramic bottom plate with a through-hole structure, which is mainly used to provide structural support and assist in absorbing medium-frequency sound waves.

[0043] The specific implementation of step S07 is to assemble three functional layers into a composite sound-absorbing floor according to the calculated spacing. First, the layer spacing parameter is extracted from the structural parameter vector. Then, height-adjustable support columns are installed on the through-hole ceramic bottom plate. The support columns are made of engineering plastics and designed as a lockable height-adjustable structure, with a height adjustment range of 5 to 25 mm. Next, a basalt fiber middle layer is placed on the support columns to ensure its flatness. Finally, a microporous UHPC surface layer is placed on the basalt fiber middle layer. A flexible fixing method is used between the microporous UHPC surface layer and the support columns, allowing for small displacements to adapt to temperature changes. After assembly, a stability test is carried out on the overall structure. It is required that under a uniform load of 200 kg per square meter, the surface deflection does not exceed 2 mm, and the connection strength between the support columns and each layer is not less than 0.8 MPa. The purpose of this step is to assemble the functional layers into an integrated structure according to the optimized design parameters, ensure an appropriate air gap between the layers, and form a complete acoustic system.

[0044] The specific implementation of step S08 is to test the performance of the composite sound-absorbing floor sample in an actual acoustic environment. First, the sound absorption coefficient of the composite sound-absorbing floor is tested in a standard reverberation chamber according to ISO 354 standard, and the test frequency range is six octave points from 125 Hz to 8000 Hz. Then, the composite sound-absorbing floor is installed in a part of the target rehearsal hall or studio of the piano room (about 20% of the total area), and the sound intensity method is used to measure the change of the indoor sound field before and after installation, including reverberation time, background noise level and sound field uniformity. Then, the test results are compared with the design target, and the deviation between the actual sound absorption effect and the expected target is calculated. If the deviation exceeds the preset threshold (the deviation of sound absorption coefficient is ±0.1 or the deviation of reverberation time is ±10%), the structural parameter vector is corrected according to the test results, and step S03 is returned to recalculate the optimized parameters. If the design target still cannot be achieved after three consecutive adjustments, the acoustic environment characteristics are re-evaluated or the design target is adjusted. The finally determined structural parameter vector is saved as production parameters. The function of this step is to verify the effectiveness of the design parameters through real environment tests and optimize the structural parameters through the feedback adjustment mechanism.

[0045] The specific implementation of step S09 is to carry out mass production according to the optimized structural parameter vector. First, a production parameter database is established, which contains information such as the structural parameter vector and the corresponding material formula, process parameters, etc. Then, production process control points are designed. Three quality control points are set in the preparation process of the microporous UHPC surface layer, namely raw material ratio, microporous forming and surface layer thickness; two quality control points are set in the preparation process of the basalt fiber middle layer, namely fiber distribution uniformity and forming density; three quality control points are set in the preparation process of the through-hole ceramic bottom plate, namely slurry uniformity, sintering temperature curve and through-hole rate. Then, a batch sampling inspection plan is formulated. 2% of the products in each batch are sampled, and the structural parameters of the sampled products are measured and the sound absorption performance is tested. The deviation between the test results and the design parameters does not exceed ±8%. If it is found that the performance of the batch of products does not meet the standard, all the products in this batch are inspected and graded, and the reasons are analyzed for process adjustment. Finally, a product file is established to record the production parameters, quality inspection data and applicable environment information of each batch of products, providing data support for subsequent product improvement. The purpose of this step is to ensure the stable quality of the mass-produced composite sound-absorbing floor products and continuously meet the design requirements.

[0046] Furthermore, the method of introducing the Grey Wolf Optimization (GWO) algorithm to optimize the clustering results takes the cluster centers as the optimization objectives after the preliminary hierarchical clustering analysis. First, the population of the GWO algorithm is constructed with a population size of 30 wolves, and each wolf is represented as a position vector with the same dimension as the number of frequency points. At initialization, the position of the α-wolf is set as the best cluster center obtained from hierarchical clustering, the positions of the β-wolf and δ-wolf are set as the second-best and third-best cluster centers respectively, and the positions of the remaining ω-wolves are randomly initialized near the cluster centers. Then, the hunting iteration process is executed. In each iteration, the surrounding coefficients A and C between all wolves and the target are calculated. A = 2a·r1 - a, C = 2·r2, where a linearly decreases from 2 to 0, and r1 and r2 are random vectors between 0 and 1. Next, the positions of each wolf are updated. The update formula for the position of the ω-wolf is X(t + 1) = X(t) - A·D, where D = |C·X p (t) - X(t)|, X p is the current optimal solution. In the specific process, the fitness value corresponding to the position of each wolf is evaluated in each iteration. The fitness function is defined as the weighted sum of the distances from each frequency point to this position, and the weight is the energy density of the frequency point. The α, β, and δ wolves are updated according to the fitness value to maintain the social hierarchy structure. The maximum number of iterations of the algorithm is set to 100, and an early stopping mechanism is introduced. When the improvement of the optimal solution is less than the threshold of 0.001 for 10 consecutive iterations, the algorithm terminates in advance. To ensure the global search ability, during the iteration process, when a > 1, the wolf pack is forced to move away from the current prey for exploration; when a < 1, the search range is reduced for exploitation. After final convergence, the position of the α-wolf is used as the optimized cluster center, and the belonging relationship of each frequency point is recalculated to form an optimized frequency aggregation segment. This method combines the adaptive search ability of the GWO algorithm with the structured analysis of hierarchical clustering, significantly improving the accuracy of frequency feature recognition.

[0047] Furthermore, the specific implementation of establishing the training dataset for the sound-absorbing floor parameter model is to construct a mapping relationship through systematic sampling and experimental testing. First, design the sampling space of the structural parameters, and set the value ranges for 9 key parameters: micropore diameter (0.5 - 2.0 mm, step size 0.1 mm), micropore spacing (2.5 - 8.0 mm, step size 0.5 mm), micropore perforation rate (5 - 25%, step size 2%), basalt fiber thickness (15 - 40 mm, step size 5 mm), basalt fiber density (45 - 120 kg / m³, step size 15 kg / m³), through-hole ceramic thickness (8 - 15 mm, step size 1 mm), through-hole diameter (3 - 8 mm, step size 0.5 mm), through-hole rate (30 - 60%, step size 5%), layer spacing (5 - 25 mm, step size 5 mm). Use the Latin hypercube sampling method to uniformly sample 500 groups of parameter combinations in the parameter space to ensure that the samples cover the parameter space and are evenly distributed. Then, make corresponding composite sound-absorbing floor samples according to the sampled parameters, and the sample size is 600×600 mm. Send the samples into the impedance tube test device, and measure the sound absorption coefficient curve according to the ISO 10534 - 2 standard, with the frequency range of 50 - 20000 Hz, divided into 300 frequency points. At the same time, measure the sound absorption performance at different sound pressure levels (30 - 110 dB, divided into 200 dB points) in the reverberation chamber according to the ISO 354 standard. Collect the sound spectrum data of 25 typical acoustic environments in parallel, including 10 piano rooms of different scales, 10 rehearsal halls of different uses, and 5 professional studios, and use the methods in the aforementioned S01 and S02 steps to obtain the sound aggregation matrix of each environment. Finally, establish the mapping relationship among the three. Each combination of structural parameters and 25 combinations of acoustic environments form 12500 groups of training samples, and each group of samples contains input features (sound aggregation matrix, dimension 300×200) and output labels (structural parameter vector, dimension 9). To enhance the generalization ability of the model, perform augmentation processing on the training data, including adding Gaussian noise (standard deviation is 5% of the original value), random masking (randomly mask 10% of the matrix elements), and parameter perturbation (add ±2% random offset to each element of the structural parameter vector). The final training dataset is stored in the TFRecord format for efficient reading and training by the deep learning model.

[0048] The following details the mathematical models or calculation processes involved in the present invention.

[0049] In step S01, the calculation process of sound sampling and spectrum data acquisition is specifically expressed as follows:

[0050]

[0051] Wherein, X(f) is the Fourier transform result of the signal x(t); x(t) is the time-domain sound signal; f is the frequency; t is the time; j is the imaginary unit.

[0052] Since the actual sampling is discrete, the discrete Fourier transform (DFT) is used for calculation, and its formula is:

[0053]

[0054] Wherein, X(k) is the spectral value at the k-th frequency point; x(n) is the sampling value at the n-th time point; N is the number of sampling points; k is the frequency sequence number, ranging from 0 to N - 1; n is the time sequence number, ranging from 0 to N - 1.

[0055] In actual calculation, the fast Fourier transform (FFT) algorithm is adopted, and its computational complexity is O(NlogN), which greatly improves the efficiency compared with the direct calculation of DFT of O(N 2 )

[0056] The obtained spectral data is subjected to 1 / 3 octave analysis, and the calculation formula is:

[0057]

[0058] Wherein, L 1 / 3 (i) is the sound pressure level of the i-th 1 / 3 octave band; B i is the frequency range of the i-th 1 / 3 octave band; X(f) is the spectral value at the frequency f; the center frequency f c (i) = 1000·10 (i-30) / 10 Hz, where i is the band number.

[0059] The calculation of the acoustic environment feature vector is the weighted average of each sampling point:

[0060]

[0061] Wherein, V env is the environmental feature vector; V p is the spectral vector of the p-th sampling point; w p is the weight coefficient, which is related to the sampling point position and is determined by room mode analysis, satisfying P is the total number of sampling points, not less than 9.

[0062] In step S02, the Euclidean distance calculation formula in hierarchical clustering analysis is:

[0063]

[0064] Wherein, d(i, j) is the distance between the frequency points i and j; v i (b) and vj (b) are the energy values of frequency points i and j in the b-th decibel segment; B is the number of decibel segments, set to 200.

[0065] The clustering distance calculation of the Ward method is as follows:

[0066]

[0067] In the formula, D(C i , C j ) is the distance between clusters C i and C j ; |C i | and |C j | are the number of elements in clusters C i and C j respectively; m i and m j are the centroids of clusters C i and C j respectively; ||m i - m j || is the Euclidean distance between the two centroids.

[0068] The position update calculation in the grey wolf optimization algorithm is specifically expressed as follows:

[0069]

[0070] In the formula, is the current position vector of the wolf; and are the position vectors of the α-wolf, β-wolf, and δ-wolf respectively; and are random vectors, and the calculation formula is where is a random vector in the interval [0, 1]; and are coefficient vectors, and the calculation formula is where is a vector that linearly decreases from 2 to 0 with the number of iterations, is a random vector in the interval [0, 1].

[0071] The fitness function of the wolf pack position is defined as:

[0072]

[0073] In the formula, is the fitness value of the position ; N is the total number of frequency points; w i is the weight of the i-th frequency point, which is proportional to its energy; is the distance from the i-th frequency point to the position The Euclidean distance; λ is the regularization coefficient, set to 0.01; is the regularization term, used to avoid overfitting, defined as that is, the square of the norm of the position vector.

[0074] The calculation formula for the sound aggregation matrix M is:

[0075] M = [m ij F×D ;

[0076] In the formula, M is the sound aggregation matrix; m ij is the energy density in the cross-region of the i-th frequency aggregation segment and the j-th decibel aggregation segment; F is the number of frequency aggregation segments, with a value range of 5 to 15; D is the number of decibel aggregation segments, with a value range of 3 to 8. The specific calculation is:

[0077]

[0078] In the formula, E(f, d) is the energy value at frequency f and decibel d; F i is the set of frequency points included in the i-th frequency aggregation segment; D j is the set of decibel points included in the j-th decibel aggregation segment; |F i | and |D j | are the number of elements in the sets F i and D j respectively.

[0079] In step S03, the normalization processing formula for the input sound aggregation matrix is:

[0080]

[0081] In the formula, M norm is the normalized sound aggregation matrix; M is the original sound aggregation matrix; M min and M max are the minimum and maximum values in the original matrix respectively.

[0082] In the deep learning model, the calculation formula for the convolution operation is:

[0083]

[0084] In the formula, F l，k (i, j) is the value at the position (i, j) of the k-th feature map in the l-th layer; K h and K w are the height and width of the convolution kernel respectively; W l，k (m, n) is the weight at the position (m, n) of the k-th convolution kernel in the l-th layer; F l-1 ​(i + m, j + n) is the value of the feature map at the (i + m, j + n) position in the (l - 1)-th layer; b l，k is the bias term of the k-th convolutional kernel in the l-th layer.

[0085] The calculation formula of the ReLU activation function is:

[0086] ReLU(x) = max(0, x);

[0087] In the formula, ReLU(x) is the output of the ReLU function; x is the input value.

[0088] The calculation formula of batch normalization is:

[0089]

[0090] In the formula, is the value after normalization; x is the input value; μ B is the mean within the batch; is the variance within the batch; ∈ is a very small positive number to prevent division by zero, usually taking 10 -5 ; γ and β are learnable scaling and translation parameters; y is the final output of batch normalization.

[0091] The calculation formula of global average pooling is:

[0092]

[0093] In the formula, GAP(F) is the global average pooling result of the feature map F; H and W are the height and width of the feature map respectively; F(i, j) is the value of the feature map at the position (i, j).

[0094] The calculation formula of Dropout is:

[0095]

[0096] In the formula, y is the output after Dropout; x is the input; r is a random binary mask, following the Bernoulli distribution Bernoulli(1 - p), that is, each element is 1 with probability 1 - p and 0 with probability p; ⊙ represents element-wise multiplication; p is the Dropout rate, which is 0.5 for the first fully connected layer, 0.3 for the second, and 0 for the third.

[0097] During the training process, the loss function adopts the mean squared error (MSE):

[0098]

[0099] In the formula, MSE is the mean squared error loss; N is the number of samples; y i is the true label of the i-th sample; is the predicted value of the model for the i-th sample; λ is the regularization coefficient, set to 10 -4 ; W j is the j-th weight parameter in the model.

[0100] The Adam optimization algorithm is used to update the model parameters, and its calculation formula is:

[0101] m t = β1·m t-1 + (1 - β1)·g t ;

[0102]

[0103] In the formula, m t and v t are the first-order and second-order momenta respectively; g t is the current gradient; β1 and β2 are the momentum decay rates respectively, with values of 0.9 and 0.999; and are the momenta after bias correction; θ t is the updated parameter; η is the learning rate, with an initial value of 0.001; ∈ is a very small positive number to prevent division by zero, with a value of 10 -8 .

[0104] The learning rate is decayed using the cosine annealing strategy, and the calculation formula is:

[0105]

[0106] In the formula, η t is the learning rate at the t-th iteration; η min and η max are the minimum and maximum learning rates respectively, with values of 10 -5 and 0.001; t is the current iteration number; T is the total number of iterations, set to 10000.

[0107] Optionally, in step S05, the calculation formula for controlling the density of the middle layer of basalt fiber is:

[0108]

[0109] In the formula, ρ bf is the density of the middle layer of basalt fiber, with a range of 45 to 120 kg / m³; m fiber is the mass of basalt fiber; V panel is the volume of the middle layer; r resin is the resin addition ratio, with a range of 0.03 to 0.07.

[0110] Optionally, in step S06, the calculation formula for the via hole rate of the through-hole ceramic substrate is:

[0111]

[0112] In the formula, P ceramic is the through-hole rate, ranging from 30% to 60%; V holes is the through-hole volume; V total is the total volume of the bottom plate.

[0113] Optionally, in step S08, the sound absorption coefficient calculation formula of the composite sound absorption floor is:

[0114]

[0115] In the formula, α(f) is the sound absorption coefficient at frequency f; I absorbed (f) is the sound intensity absorbed; I incident (f) is the incident sound intensity.

[0116] Optionally, in actual measurement, the reverberation room method is used to calculate the sound absorption coefficient:

[0117]

[0118] In the formula, α st (f) is the statistical sound absorption coefficient at frequency f; V is the volume of the reverberation room, in cubic meters; S is the sample area, in square meters; T1(f) and T2(f) are the reverberation times of the reverberation room at frequency f before and after placing the sample, in seconds.

[0119] Optionally, in the process of optimizing the structural parameters, the parameter adjustment formula is:

[0120]

[0121] In the formula, P new is the adjusted structural parameter; P old is the structural parameter before adjustment; ΔP is the parameter adjustment step size, usually taking 5% of the parameter range; α target is the target sound absorption coefficient; α measured is the measured sound absorption coefficient; S p is the sensitivity coefficient, determined by experiment, ranging from 0.5 to 2.0.

[0122] Optionally, in the Latin hypercube sampling method, the parameter space division calculation is:

[0123]

[0124] In the formula, is the value of the i-th parameter in the j-th sample; is the j-th permutation value of the i-th parameter; A random number uniformly distributed in [0, 1]; n is the number of samples, set to 500; and are the minimum and maximum values of the i-th parameter respectively.

[0125] The selection of these equations and calculation processes is mainly based on physical acoustics principles and machine learning practices. The Fourier transform is used to convert the time-domain signal into the frequency domain, facilitating the analysis of the frequency characteristics of sound; the 1 / 3 octave analysis is more in line with the way the human ear perceives sound; the combination of hierarchical clustering and the grey wolf algorithm utilizes the structured characteristics of hierarchical clustering and the global search ability of the grey wolf algorithm to effectively identify key frequency features; the deep learning model uses a convolutional neural network to extract sound features, uses a fully connected network for parameter mapping, and techniques such as batch normalization and Dropout improve the training stability and generalization ability of the model; the Adam optimization algorithm combined with the cosine annealing learning rate strategy can effectively avoid local optima and accelerate convergence. The overall scheme realizes the automatic conversion from environmental requirements to specific designs by establishing a mapping relationship between acoustic environment characteristics and physical structure parameters, improving the accuracy and efficiency of the sound-absorbing floor design.

[0126] Specifically, the principle of the present invention is: The core technical principle of the present invention is based on the optimization of structural parameters driven by acoustic environment data and the design of a multi-layer composite sound-absorbing structure. First, the collected time-domain sound samples are converted into frequency-domain data through the Fourier transform to establish an initial acoustic environment database, which provides a data basis for subsequent structural parameter optimization. The hierarchical clustering combined with the grey wolf hunting algorithm is used to process the sound spectrum data, which can adaptively identify the key frequency and decibel aggregation regions in the environment, form a sound aggregation matrix, and realize the accurate extraction of acoustic environment characteristics.

[0127] At the parameter optimization level, the pre-trained sound-absorbing floor parameter model designed by the present invention adopts a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron, which can effectively learn the complex non-linear mapping relationship between the sound aggregation matrix and the optimal structural parameter vector. The model is pre-trained through a large-scale data set and has the ability to convert specific acoustic environment characteristics into sound-absorbing floor structural parameters, solving the problem that it is difficult to accurately match by traditional empirical design methods.

[0128] At the structural design level, the present invention adopts the principle of the synergistic effect of a three-layer composite structure: The microporous UHPC surface layer forms an array of Helmholtz resonators through the microporous structure, effectively absorbing medium and high-frequency sound waves; the basalt fiber middle layer uses the internal friction principle of porous materials to convert sound energy into heat energy, mainly targeting low-frequency sound waves; the through-hole ceramic bottom plate forms a large-scale resonance cavity to enhance the absorption of medium-frequency sound waves. The design of the spacing of the three layers is based on the principle of acoustic wave interference and phase superposition, further optimizing the overall sound-absorbing performance.

[0129] Through the actual measurement feedback mechanism, the present invention establishes a closed-loop optimization system from design to production to ensure that the final product can accurately match the design parameters. This technical principle based on data-driven, multi-layer composite structure and closed-loop optimization solves the core problem that traditional sound-absorbing floors cannot be precisely customized for specific acoustic environments from the mechanism, and realizes the directional optimization of sound-absorbing performance.

[0130] A specific Embodiment 1 of the present invention is provided below. The specific implementation manners of each step in this Embodiment 1 are described in detail as follows.

[0131] The specific implementation manner of step S01 is to sample the sound in the rehearsal hall of the music room and the studio using high-precision acoustic measurement equipment. First, at least 9 evenly distributed sampling points are set in the space using omnidirectional pickups, and each sampling point collects sound samples for no less than 30 minutes. The sampling frequency is set to 48000 Hz, and the quantization accuracy is 24 bits. Then, the fast Fourier transform algorithm is applied to the collected original waveform data to convert the time-domain signal into a frequency-domain signal, and its calculation formula is:

[0132]

[0133] In the formula, X(f) is the Fourier transform result of the signal x(t); x(t) is the time-domain sound signal; f is the frequency; t is the time; j is the imaginary unit.

[0134] Since the actual sampling is discrete, the discrete Fourier transform (DFT) is used for calculation, and its formula is:

[0135]

[0136] In the formula, X(k) is the spectral value at the k-th frequency point; x(n) is the sampling value at the n-th time point; N is the number of sampling points; k is the frequency serial number, ranging from 0 to N - 1; n is the time serial number, ranging from 0 to N - 1.

[0137] The obtained spectral data undergoes 1 / 3 octave analysis, and the calculation formula is:

[0138]

[0139] In the formula, L 1 / 3 (i) is the sound pressure level of the i-th 1 / 3 octave band; B i is the frequency range of the i-th 1 / 3 octave band; X(f) is the spectral value at the frequency f; the center frequency f c (i) = 1000·10 (i-30) / 10 Hz, where i is the band number.

[0140] The calculation of the acoustic environment feature vector is the weighted average of each sampling point:

[0141]

[0142] In the formula, V env is the environmental feature vector; V p is the spectrum vector of the p-th sampling point; w p is the weight coefficient, which is related to the sampling point position and is determined by room mode analysis, satisfying P is the total number of sampling points, not less than 9. The purpose of this step is to establish an accurate digital representation of the acoustic environment and provide a data basis for subsequent optimization of the sound absorption structure.

[0143] The specific implementation of step S02 is to perform multi-dimensional analysis and processing on the spectrum data in the initial acoustic environment database. First, the spectrum data is represented in the form of a 300×200-dimensional matrix, where 300 rows represent the frequency dimension and 200 columns represent the decibel dimension. Then, the hierarchical clustering analysis method is used to perform initial clustering on the spectrum data, and the Euclidean distance is calculated as the similarity metric:

[0144]

[0145] In the formula, d(i, j) is the distance between frequency points i and j; v i (b) and v j (b) are the energy values of frequency points i and j in the b-th decibel segment respectively; B is the number of decibel segments, set to 200.

[0146] The Ward method is used to construct a clustering tree, and its clustering distance is calculated as:

[0147]

[0148] In the formula, D(C i , C j ) is the distance between clusters C i and C j ; |C i | and |C j | are the number of elements in clusters C i and C j respectively; m i and m j are the centroids of clusters C i and C j respectively; ||m i -m j || is the Euclidean distance between the two centroids.

[0149] Then, the gray wolf hunting algorithm is introduced to optimize the clustering result, and the position update is calculated as follows:

[0150]

[0151] In the formula, is the current position vector of the wolf; and are the position vectors of the α-wolf, β-wolf, and δ-wolf respectively; and are random vectors, and the calculation formula is where is a random vector in the interval [0, 1]; and are coefficient vectors, and the calculation formula is where is a vector that linearly decreases from 2 to 0 with the number of iterations, is a random vector in the interval [0, 1].

[0152] The fitness function of the wolf pack position is defined as:

[0153]

[0154] In the formula, is the fitness value of the position ; N is the total number of frequency points; w i is the weight of the i-th frequency point, which is proportional to its energy; is the Euclidean distance from the i-th frequency point to the position ; λ is the regularization coefficient, set to 0.01; is the regularization term, used to avoid overfitting, and is defined as that is, the square of the norm of the position vector.

[0155] Finally, the frequency aggregation segment and the decibel aggregation segment are combined to form a sound aggregation matrix:

[0156] M = [m ij F×D ;

[0157] In the formula, M is the sound aggregation matrix; m ij is the energy density in the intersection area of the i-th frequency aggregation segment and the j-th decibel aggregation segment; F is the number of frequency aggregation segments, with a value range of 5 to 15; D is the number of decibel aggregation segments, with a value range of 3 to 8. The specific calculation is:

[0158]

[0159] In the formula, E(f, d) is the energy value at frequency f and decibel d; F i is the set of frequency points included in the i-th frequency aggregation segment; D j is the set of decibel points included in the j-th decibel aggregation segment; |F i | and |D j | are the sets F respectively​i and D j The number of elements of. The function of this step is to accurately identify the key frequency characteristics in the acoustic environment and provide a targeted basis for the subsequent design of the sound absorption structure.

[0160] The specific implementation of step S03 is to input the sound aggregation matrix into a pre-trained deep learning model for parameter calculation. First, perform normalization on the sound aggregation matrix:

[0161]

[0162] In the formula, M norm is the normalized sound aggregation matrix; M is the original sound aggregation matrix; M min and M max are the minimum and maximum values in the original matrix respectively.

[0163] Then input the normalized matrix into the pre-trained sound absorption floor parameter model, which adopts a hybrid architecture of four-layer convolutional neural network and three-layer fully connected perceptron. The calculation formula for the convolution operation is:

[0164]

[0165] In the formula, F l，k (i, j) is the value of the kth feature map in the lth layer at the position (i, j); K h and K w are the height and width of the convolution kernel respectively; W l，k (m, n) is the weight of the kth convolution kernel in the lth layer at the position (m, n); F l-1 (i + m, j + n) is the value of the feature map in the (l - 1)th layer at the position (i + m, j + n); b l，k is the bias term of the kth convolution kernel in the lth layer.

[0166] Apply the ReLU activation function:

[0167] ReLU(x) = max(0, x);

[0168] In the formula, ReLU(x) is the output of the ReLU function; x is the input value.

[0169] The calculation formula for batch normalization is:

[0170]

[0171] In the formula, is the normalized value; x is the input value; μ B is the mean within the batch; is the variance within the batch; ∈ is a very small positive number to prevent division by zero, usually taking 10-5 ; γ and β are learnable scaling and translation parameters; y is the final output of batch normalization.

[0172] The calculation formula for global average pooling is:

[0173]

[0174] In the formula, GAP(F) is the global average pooling result of the feature map F; H and W are the height and width of the feature map respectively; F(i, j) is the value of the feature map at the position (i, j).

[0175] The calculation formula for Dropout is:

[0176]

[0177] In the formula, y is the output after Dropout; x is the input; r is a random binary mask, following the Bernoulli distribution Bernoulli(1 - p), that is, each element is 1 with probability 1 - p and 0 with probability p; ⊙ represents element-wise multiplication; p is the Dropout rate, which is 0.5 for the first fully connected layer, 0.3 for the second, and 0 for the third.

[0178] The model output is the optimal structural parameter vector for a specific acoustic environment, containing 9 parameters. The purpose of this step is to transform complex acoustic features into specific structural parameters through a deep learning model, realizing the mapping from the acoustic environment to the physical structure.

[0179] The specific implementation of step S04 is the same as the foregoing and will not be elaborated here.

[0180] The specific implementation of step S05 is to design the basalt fiber middle layer based on the low-frequency band data in the sound aggregation matrix. First, extract two parameters, the thickness and density of the basalt fiber middle layer, from the structural parameter vector, and then select basalt fibers with appropriate diameters. The fiber diameter ranges from 7 to 13 microns, and the fiber length is 50 to 80 millimeters. The basalt fiber mat is prepared by the air-laying process, and the density control calculation formula is:

[0181]

[0182] In the formula, ρ bf is the density of the basalt fiber middle layer, ranging from 45 to 120 kilograms per cubic meter; m fiber is the mass of the basalt fiber; V panel is the volume of the middle layer; r resin is the resin addition ratio, ranging from 0.03 to 0.07.

[0183] Finally, cut the basalt fiber felt into the required shape and size, with the thickness controlled between 15 and 40 millimeters. The purpose of this step is to prepare a basalt fiber middle layer with a specified thickness and density, which is mainly used for absorbing low-frequency sound waves.

[0184] The specific implementation of step S06 is to prepare a through-hole ceramic bottom plate according to the frequency band data in the sound aggregation matrix. First, extract three parameters, namely the thickness of the through-hole ceramic bottom plate, the through-hole diameter, and the through-hole rate, from the structural parameter vector, and then prepare a ceramic slurry. The main components are 65% kaolin, 25% quartz, 8% feldspar, and 2% additive. Use the foaming method to prepare a porous ceramic blank. The calculation formula for the through-hole rate is:

[0185]

[0186] In the formula, P ceramic is the through-hole rate, with a range of 30% to 60%; V holes is the through-hole volume; V total is the total volume of the bottom plate.

[0187] After forming, dry it at 80 to 100 °C for 24 hours, then perform high-temperature sintering at 1100 to 1200 °C. Finally, cut and polish the sintered ceramic bottom plate, with the thickness controlled between 8 and 15 millimeters and the through-hole diameter controlled between 3 and 8 millimeters. The function of this step is to prepare a ceramic bottom plate with a through-hole structure, which is mainly used for providing structural support and assisting in absorbing medium-frequency sound waves.

[0188] The specific implementation of step S07 is the same as the foregoing and will not be elaborated here.

[0189] The specific implementation of step S08 is to test the performance of the composite sound-absorbing floor sample in an actual acoustic environment. First, test the sound absorption coefficient of the composite sound-absorbing floor in a standard reverberation chamber according to the ISO 354 standard:

[0190]

[0191] In the formula, α(f) is the sound absorption coefficient at frequency f; I absorbed (f) is the absorbed sound intensity; I incident (f) is the incident sound intensity.

[0192] In actual measurement, the reverberation chamber method is used to calculate the sound absorption coefficient:

[0193]

[0194] In the formula, α st(f) is the statistical sound absorption coefficient at frequency f; V is the volume of the reverberation chamber in cubic meters; S is the area of the sample in square meters; T1(f) and T2(f) are the reverberation times of the reverberation chamber at frequency f before and after placing the sample respectively, in seconds.

[0195] Then install the composite sound absorption floor in a partial area of the target space, and use the sound intensity method to measure the change of the indoor sound field before and after installation. Compare the test results with the design target. If the deviation exceeds the preset threshold, correct the structure parameter vector according to the test results:

[0196]

[0197] In the formula, P new is the adjusted structure parameter; P old is the structure parameter before adjustment; ΔP is the parameter adjustment step size, usually taking 5% of the parameter range; α target is the target sound absorption coefficient; α measured is the measured sound absorption coefficient; S p is the sensitivity coefficient, determined by experiment, with a range of 0.5 to 2.0.

[0198] The function of this step is to verify the effectiveness of the design parameters through real environment tests and optimize the structure parameters through the feedback adjustment mechanism.

[0199] The specific implementation of step S09 is the same as the foregoing and will not be elaborated here.

[0200] In the whole implementation plan, the establishment of the training data set uses the Latin hypercube sampling method to evenly cover the parameter space:

[0201]

[0202] In the formula, is the value of the i-th parameter in the j-th sample; is the j-th permutation value of the i-th parameter; is a random number uniformly distributed in [0, 1]; n is the number of samples, set to 500; and are the minimum and maximum values of the i-th parameter respectively.

[0203] During the model training process, the mean squared error loss function is adopted:

[0204]

[0205] In the formula, MSE is the mean squared error loss; N is the number of samples; y i is the true label of the i-th sample; is the predicted value of the model for the i-th sample; λ is the regularization coefficient, set to 10-4 ; W j is the jth weight parameter in the model.

[0206] The parameter update adopts the Adam optimization algorithm:

[0207] m t =β1·m t-1 +(1-β1)·g t ;

[0208]

[0209] In the formula, m t and v t are the first-order and second-order momentum respectively; g t is the current gradient; β1 and β2 are momentum decay rates, with values ​​of 0.9 and 0.999 respectively; and is the momentum after deviation correction; θ t is the updated parameter; η is the learning rate, the initial value is 0.001; ∈ is a very small positive number to prevent division by zero, the value is 10 -8 .

[0210] The learning rate is decayed using the cosine annealing strategy:

[0211]

[0212] Where η t is the learning rate of the tth iteration; η min and η max are the minimum and maximum learning rates, respectively, and the value is 10 -5 and 0.001; t is the current iteration number; T is the total iteration number, which is set to 10000.

[0213] In order to better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: In a newly built rehearsal hall project, researchers used the composite sound-absorbing floor of the present invention for design and application to solve the problems of poor sound absorption, easy dust accumulation, difficult cleaning, and poor dirt resistance of the traditional rehearsal hall floor, thus solving the pain points of rehearsal hall floor use over the years. This example uses the piano rehearsal hall as the main application scenario, and verifies the effectiveness of the present invention through acoustic environment sampling, design optimization, and field application.

[0214] First, the researchers sampled the acoustic environment of different types of rehearsal halls in the conservatory of music. Twelve evenly distributed sampling points were set in the piano rehearsal hall using an omnidirectional pickup. At each sampling point, a 45-minute sound sample was collected, with the sampling frequency set at 48,000 Hz and the quantization precision at 24 bits. After the collection, the time-domain signal was converted into a frequency-domain signal through the fast Fourier transform algorithm, and the sound spectrum data in the range of 50 Hz to 20,000 Hz was obtained.

[0215] Through 1 / 3 octave analysis, the researchers obtained the sound pressure level distribution in different frequency bands in the rehearsal hall, as shown in Table 1:

[0216] Table 1 1 / 3 octave sound pressure level distribution in the piano rehearsal hall

[0217] Center frequency (Hz) Sound pressure level (dB) Standard deviation (dB) 63 78.4 4.2 125 82.6 3.8 250 85.3 3.1 500 79.8 2.5 1000 76.2 2.3 2000 73.5 2.0 4000 68.7 1.8 8000 62.4 1.5 16000 54.6 1.3

[0218] Based on the collected acoustic data, the researchers analyzed the spectrum data using hierarchical clustering analysis combined with the grey wolf hunting algorithm. The initial clustering used the Ward method to construct a clustering tree, with the number of clusters set at 8. Then, it was optimized through the grey wolf hunting algorithm, with the population size set at 30 and the number of iterations at 85. Finally, the optimized frequency aggregation segments and decibel aggregation segments were obtained, forming a sound aggregation matrix.

[0219] The sound aggregation matrix was input into a pre-trained sound-absorbing floor parameter model for calculation, and a structural parameter vector was obtained. The model adopted a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron. The first convolutional layer used 64 5×5 convolutional kernels, the second convolutional layer used 128 3×3 convolutional kernels, the third convolutional layer used 256 3×3 convolutional kernels, and the fourth convolutional layer used 512 2×2 convolutional kernels. The fully connected layers contained 1024, 512, and 256 neurons respectively, and the output layer contained 9 neurons corresponding to the structural parameters. The model was trained using the stochastic gradient descent algorithm with a batch size of 64, a learning rate of 0.001 and using the cosine annealing strategy, and the number of iterations was 7500 times.

[0220] Based on the calculation results, the researchers determined the parameters of the microporous UHPC surface layer as follows: the micropore diameter is 0.5 mm, the micropore spacing is 15.6 mm, and the micropore perforation rate is 8%. At the same time, they also determined that the thickness of the basalt fiber middle layer is 28 mm and the density is 85 kg / m³; the thickness of the through-hole ceramic bottom plate is 10 mm, the through-hole diameter is 5 mm, and the through-hole rate is 45%; the layer spacing is 20 mm.

[0221] According to the determined parameters, the researchers prepared composite sound-absorbing floor samples. First, UHPC was formulated with the following proportions: Portland cement 52.5%, mineral admixtures (silica fume and fly ash) 18.5%, quartz sand 23.5%, polycarboxylate superplasticizer 1.5%, steel fibers (length 12 mm, diameter 0.2 mm) 2.0%, and water-cement ratio 0.19. The concrete was poured into a precision mold with evenly distributed micropores, cured for 48 hours in an environment with a temperature of 20 ± 1°C and a relative humidity of over 95% after molding, and then demolded and subjected to standard curing for 7 days. After curing, the surface was polished, and the final thickness was controlled at 8 mm.

[0222] The middle layer of basalt fiber uses basalt fibers with a diameter of 10 μm and a length of 65 mm, which are prepared into a fiber mat through an air-laying process. The fiber bonding uses epoxy resin, and the addition amount is 5% of the fiber mass. By controlling the fiber feeding amount and compaction degree, the final density reaches 85 kg / m³, and the thickness is 28 mm.

[0223] The through-hole ceramic bottom plate uses a slurry prepared with 65% kaolin, 25% quartz, 8% feldspar, and 2% additives, and a porous green body is prepared by the foaming method. The addition amount of the foaming agent is 7% of the ceramic mass. After the green body is dried at 90°C for 24 hours, it is subjected to high-temperature sintering at 1150°C for 5 hours. The thickness of the sintered ceramic bottom plate is 10 mm, the through-hole diameter is 5 mm, and the through-hole rate is 45%.

[0224] The assembly of the three-layer structure is fixed with highly adjustable engineering plastic support columns. The height of the support columns is set at 20 mm to ensure precise control of the layer spacing. A water seepage discharge groove is designed at the bottom, with a groove width of 10 mm, a depth of 15 mm, and a spacing of 300 mm, distributed in a grid pattern and connected to the drainage system. The seepage groove is made of stainless steel, corrosion-resistant and easy to clean.

[0225] After the composite sound-absorbing floor samples were prepared, the researchers tested their sound absorption coefficients in a standard reverberation chamber according to ISO 354 standard, and the results are shown in Table 2:

[0226] Table 2 Test results of sound absorption coefficients of composite sound-absorbing floors

[0227] Frequency (Hz) Sound absorption coefficient Standard deviation 125 0.28 0.03 250 0.42 0.02 500 0.65 0.02 1000 0.79 0.01 2000 0.82 0.01 4000 0.76 0.02 8000 0.70 0.02

[0228] Meanwhile, the researchers also tested other performance indicators of the composite sound-absorbing floor, and the results are shown in Table 3:

[0229] Table 3 Test results of performance of composite sound-absorbing floors

[0230]

[0231]

[0232] After the test, the researchers installed a composite sound-absorbing floor in the newly built piano rehearsal hall of the music college, with an installation area of 120 square meters. The reverberation time before installation was 2.3 seconds, and after installation, it was 1.6 seconds, reaching the ideal reverberation time range for a piano rehearsal hall. At the same time, the researchers conducted a 90-day usage tracking survey and collected usage data, as shown in Table 4:

[0233] Table 4 Tracking Survey of the Usage of Composite Sound-Absorbing Floor (90 days)

[0234] Evaluation item Score (1 - 10) Main feedback Acoustic effect 9.2 The reverberation time is moderate, the sound quality is clear, and there is no sound color variation Cleaning convenience 8.7 The surface is not easy to accumulate dust, water stains are easy to wipe, and no special cleaning agent is required Dirt resistance 9.0 No obvious stains are seen on the surface within 90 days, and chalk marks are easy to clean Using comfort 8.5 Good foot feeling, stable stepping, no obvious vibration Water seepage treatment 9.3 A small amount of water stains quickly penetrate and drain, without generating ponding Scratch resistance 8.8 There are only slight scratches at the piano casters, and the overall surface is intact Impact resistance 9.1 It can withstand slight collisions of musical instruments without damage and has good reliability Odor evaluation 9.5 No peculiar smell, does not affect the indoor air quality Overall satisfaction 9.0 The comprehensive performance is excellent, solving many problems in the use of traditional floors

[0235] Through this embodiment, the researchers compared the performance differences between the traditional floor materials in piano rehearsal halls and the composite sound-absorbing floor of the present invention. The floors of traditional music rehearsal halls usually use materials such as wooden floors, carpets, or rubber floors, which have obvious deficiencies in terms of acoustic performance, durability, and maintenance costs. Although wooden floors have good aesthetics and foot feeling, their sound-absorbing performance is limited and they are prone to moisture and deformation; carpets have good sound-absorbing effects but are not dirt-resistant, difficult to clean, and have a short lifespan; rubber floors are durable but have poor sound-absorbing performance and will harden and crack over time.

[0236] The composite sound-absorbing floor of the present invention realizes excellent sound-absorbing performance through the precise combination of a microporous UHPC surface layer, a basalt fiber middle layer, and a through-hole ceramic bottom plate. Especially in the medium and high-frequency bands (500 - 4000 Hz), its sound-absorbing effect is significantly better than that of traditional materials. At the same time, the microporous UHPC surface layer provides excellent wear resistance, dirt resistance, and scratch resistance. The basalt fiber middle layer provides low-frequency sound absorption and vibration isolation functions, and the through-hole ceramic bottom plate provides structural support and ventilation. The design of the water seepage discharge groove solves the problem of easy water accumulation of traditional floor materials, greatly improving the maintenance efficiency and service life of the rehearsal hall.

[0237] The hierarchical clustering analysis method adopted in the present invention combines the grey wolf hunting algorithm and the deep learning model, breaking through the limitations of traditional acoustic design that only relies on empirical formulas or simple models. It can accurately identify the acoustic characteristics of a specific space and automatically generate optimal structural parameters, significantly improving the design accuracy and efficiency. At the same time, the application of microporous UHPC solves the problems of insufficient strength and poor durability of traditional sound-absorbing materials, and the use of basalt fiber avoids the disadvantages of easy aging and flammability of organic fibers.

[0238] Through practical applications, it is proved that the composite sound-absorbing floor of this Embodiment 2 has excellent sound-absorbing performance, extremely high durability, excellent cleanliness, and excellent comfort, effectively solving the problems of short service life, high maintenance cost, and unstable acoustic performance of traditional rehearsal hall floor materials, and providing a practical, durable, and high-performance floor solution for music education venues.

[0239] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 5, 6, and 7 below.

[0240] Table 5 Variable Explanation Table (First Part)

[0241]

[0242]

[0243] Table 6 Variable Explanation Table (Second Part)

[0244]

[0245]

[0246] Table 7 Variable Explanation Table (Third Part)

[0247]

[0248] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios, characterized in that: include: Collect sound frequency samples from piano rehearsal rooms and studios to build an initial acoustic environment database; Analyze the collected sound spectrum data, determine the key frequency clustering segments and decibel clustering segments, and form a sound clustering matrix; input the sound clustering matrix into the pre-trained sound-absorbing floor parameter model to calculate the structural parameter vector required for the best sound absorption effect; Determine the microporous UHPC surface parameters based on the structural parameter vector; The basalt fiber middle layer is designed based on the low-frequency data of the sound aggregation matrix; Prepare a through-hole ceramic base plate according to the frequency band data of the sound focusing matrix; Assemble the microporous UHPC surface layer, basalt fiber middle layer and through-hole ceramic bottom plate; test the performance of the composite sound-absorbing floor sample in a real acoustic environment and adjust the structural parameter vector; and produce according to the optimized structural parameter vector.

2. The method for producing a composite sound-absorbing floor suitable for piano rehearsal rooms and studios according to claim 1, characterized in that: The sound concentration matrix refers to the sound energy concentration in different frequency ranges expressed in matrix form after sampling and analyzing the acoustic environment. The rows of the matrix represent different frequency bands, the columns represent different decibel bands, and each element value represents the frequency or intensity of the frequency and decibel combination in the measurement environment.

3. The method for producing a composite sound-absorbing floor suitable for piano rehearsal rooms and studios according to claim 2, characterized in that: The specific analysis of the collected sound spectrum data is as follows: first, the sound spectrum data is grouped according to similarity through hierarchical clustering, and then the stratification results are optimized using the gray wolf hunting algorithm. By simulating the encirclement, pursuit and attack behaviors of the gray wolf, the cluster center can be accurately located, thereby improving the accuracy of identifying sound frequency cluster segments and decibel cluster segments.

4. The method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios according to claim 3, characterized in that: The microporous UHPC surface refers to the floor surface made of UHPC with tiny holes on the surface. UHPC material has extremely high strength and durability, and its microporous structure enables it to absorb medium and high frequency sound waves.

5. The method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios according to claim 4, characterized in that: The structural parameter vector refers to a set of numerical values ​​that describe the structural characteristics of each component of the composite sound-absorbing floor, including micropore diameter, micropore spacing, micropore perforation rate, basalt fiber middle layer thickness, basalt fiber middle layer density, through-hole ceramic bottom plate thickness, through-hole diameter, through-hole rate and layer spacing.

6. The method for producing a composite sound-absorbing floor suitable for piano rehearsal rooms and studios according to claim 5, characterized in that: According to the calculated structural parameter vector, the microporous UHPC surface layer parameters are determined, including a micropore diameter range of 0.5 to 2.0 mm, a micropore spacing of 2.5 to 8.0 mm, and a micropore perforation rate of 5% to 25%.

7. The method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios according to claim 6, characterized in that: The basalt fiber middle layer is designed based on the low-frequency band data of the sound concentration matrix. Specifically, the thickness of the basalt fiber middle layer is designed to be 15 to 40 mm, and the density of the basalt fiber middle layer is adjusted to be between 45 and 120 kilograms per cubic meter.

8. The method for producing a composite sound-absorbing floor suitable for piano rehearsal rooms and studios according to claim 7, characterized in that: The through-hole ceramic base plate is prepared according to the mid-frequency band data of the sound focusing matrix. Specifically, the through-hole ceramic base plate is prepared according to the mid-frequency band data of the sound focusing matrix. The thickness of the through-hole ceramic base plate is 8 to 15 mm, the through-hole diameter is 3 to 8 mm, and the through-hole rate is 30% to 60%.

9. The method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios according to claim 8, characterized in that: The structure of the pre-trained sound-absorbing floor parameter model is a deep learning model based on a hybrid architecture of a four-layer convolutional neural network and a three-layer fully connected perceptron. The input layer receives 300×200-dimensional sound aggregation matrix data, and the output layer contains 9 neurons corresponding to the 9 key parameters in the structural parameter vector.

10. The method for producing a composite sound-absorbing floor suitable for piano rooms, rehearsal halls and studios according to claim 9, characterized in that: The training data set establishment step in the pre-training process of the sound-absorbing floor parameter model includes collecting sound absorption performance test data of 500 composite sound-absorbing floor samples with different structural parameter combinations in a standard acoustic laboratory. The test frequency range is 50 Hz to 20,000 Hz, divided into 300 frequency points, and the test decibel range is 30 decibel to 110 decibel, divided into 200 decibel points.

Images