Porous dehumidification material structure optimization method suitable for ultrasonic regeneration

Abstract

The invention provides a porous dehumidification material structure optimization method suitable for ultrasonic regeneration, and the method comprises the steps: constructing a parameterized resource library, generating a multi-shape porous structure based on a closed-loop generation algorithm, mapping the multi-shape porous structure to a three-dimensional entity through voxels, and subdividing the multi-shape porous structure to obtain a three-dimensional geometric model file; carrying out analogue simulation on the three-dimensional geometric model file based on a double-physics-field coupling method to obtain a three-dimensional model, applying a plurality of vibration excitations to any surface, collecting corresponding vibration displacement amplitudes, and obtaining a variable coefficient according to a preset algorithm; the fractal dimension, the porosity, the characteristic frequency and the variable coefficient are constructed into a training data set for training to obtain a deep neural network; and adopting a genetic algorithm to iteratively update the fractal dimension, the porosity and the characteristic frequency by taking variation coefficient minimization as a target until a preset termination condition is reached, obtaining a target fractal dimension, a target porosity and a target characteristic frequency, and obtaining the porous dehumidification material capable of realizing large-range efficient propagation of ultrasonic waves.

Description

Technical Field

[0001] This invention relates to the field of dehumidification materials, and more particularly to a method for optimizing the structure of porous dehumidification materials suitable for ultrasonic regeneration. Background Technology

[0002] Using power ultrasound to regenerate dehumidifying materials has been proven to be an effective method by numerous studies. An ultrasonic transducer applies ultrasound to porous dehumidifying materials, and due to the thermal and mechanical effects of ultrasound, the moisture in the dehumidifying materials is effectively removed. Traditional hot air regeneration technology leaves a lot of residual heat downstream of the air duct, resulting in energy waste. Using ultrasonic regeneration technology can reduce the temperature of the regeneration air, thereby improving energy utilization efficiency.

[0003] In existing technologies, the main ultrasonic material used is silicone. However, due to the high sound absorption coefficient of silicone, ultrasound attenuates very quickly in silicone, and ultrasonic energy can only be transmitted to a small part of the silicone, making it impossible to regenerate the remaining areas. Therefore, there is an urgent need for a porous dehumidifying material to achieve large-scale and efficient propagation of ultrasound. Summary of the Invention

[0004] This invention provides a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration. Its main purpose is to solve the problems of small propagation range and poor propagation efficiency of ultrasonic waves in porous dehumidifying materials in the prior art.

[0005] To achieve the above objectives, the present invention provides a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration, the method comprising:

[0006] Construct a parameterized resource library that includes geometric performance parameters, material property parameters, and operating condition parameters; the geometric performance parameters include fractal dimension and porosity;

[0007] A polymorphic porous structure with a porosity error below a preset threshold is generated based on a closed-loop generation algorithm. The structure is then mapped to a 3D solid and triangulated to obtain a 3D geometric model file. The 3D geometric model file is then simulated using a solid mechanics-pressure acoustic dual-physics coupling method to obtain a 3D model.

[0008] Multiple vibration excitations within the characteristic frequency range are applied to any face of the 3D model based on material property parameters and working condition parameters, and the corresponding vibration displacement amplitude is collected. The coefficient of variation representing the uniformity of vibration dispersion is obtained according to a preset algorithm.

[0009] The fractal dimension, porosity, feature frequency and coefficient of variation are used to construct a training dataset, and the initial deep neural network is trained on the training dataset to obtain a deep neural network.

[0010] A genetic algorithm is used to iteratively update the fractal dimension, porosity, and feature frequencies with the goal of minimizing the coefficient of variation through selection, mutation, and crossover until a preset termination condition is met, thereby obtaining the target fractal dimension, target porosity, and target feature frequencies. In each round of iterative update, the fractal dimension, porosity, and feature frequencies are input into a deep neural network and the coefficient of variation is output.

[0011] Optionally, the process of generating a polymorphic porous structure based on a closed-loop generation algorithm that satisfies a porosity error below a preset threshold, and obtaining a 3D geometric model file by mapping it to a 3D solid using voxels and performing triangulation includes:

[0012] Based on the porosity setting requirements, initial pore voxels are randomly selected from a pixel array of a specific number of rows and columns. The pore voxels are expanded to the surrounding area according to the preset guard erosion strategy until the target number of pore voxels is reached. Then, the surrounding area of ​​the pixel array is set as solid voxels to obtain a two-dimensional graphic.

[0013] After mapping multiple solid voxels in a two-dimensional graphic to three-dimensional units, a three-dimensional solid is obtained. Each face of the three-dimensional unit is triangulated to obtain 12 triangular facets to construct a triangular mesh. Then, a three-dimensional geometric model file is output through a preset file writing function.

[0014] The fractal dimension of a two-dimensional graphic is obtained using box counting.

[0015] Optionally, the three-dimensional model is obtained by simulating the three-dimensional geometric model file based on the solid mechanics-pressure acoustic dual-physics coupling method, including:

[0016] The system receives 3D geometric model files through pre-set simulation software and performs solid mechanical field modeling on solid voxels and pressure acoustic field modeling on porous voxels.

[0017] Optionally, the process of applying multiple vibration excitations within a characteristic frequency range to any face of the 3D model based on material property parameters and operating condition parameters, and acquiring the corresponding vibration displacement amplitudes, includes:

[0018] One face of each 3D model is selected as the vibration surface for applying vibration excitation, and the face without excitation is the free face. Coupled conditions are set for the contact surface of the solid voxel and the porous voxel to represent the interaction between the solid and the air. The vibration excitation is a sinusoidal displacement excitation in the characteristic frequency range selected according to the material property parameters and the ambient temperature.

[0019] Multiple frequency points are selected at preset intervals within the characteristic frequency range, and simulation is performed on each frequency point. The vibration displacement amplitude at monitoring points uniformly set on each surface of the three-dimensional model is collected.

[0020] Optionally, a coefficient of variation representing the uniformity of vibration dispersion is obtained according to a preset algorithm, the expression of which is:

[0021]

[0022] Where x represents the vibration displacement amplitude of all monitoring points corresponding to a frequency point, and i represents the index of the frequency point. i This represents the vibration displacement amplitude of all monitoring points corresponding to frequency point i, and N represents the total number of frequency points.

[0023] Optionally, the process of training an initial deep neural network with a training dataset to obtain a deep neural network includes:

[0024] The training sample set is obtained from the training dataset according to the preset division ratio. The training sample set contains multiple samples, each of which contains fractal dimension, porosity and feature frequency. The samples in the training sample set are labeled with the true value of the coefficient of variation.

[0025] The initial deep neural network is trained using samples from the training sample set. The initial neural network includes an input layer, multiple hidden layers, a Fourier frequency domain smoothing filter layer, and an output layer. The input layer contains three neurons for inputting fractal dimension, porosity, and feature frequency, respectively. After the multiple hidden layers extract feature data through a nonlinear activation function, the Fourier frequency domain smoothing filter layer transforms the feature data to the frequency domain through Fourier transform to filter high-frequency oscillation noise. The output layer outputs the predicted value of the coefficient of variation.

[0026] A loss function is constructed using the true and predicted values ​​of the coefficient of variation. The Adam optimizer is used to iteratively update the parameters of the initial deep neural network with the goal of minimizing the loss function, until the preset number of iterations is reached, and then the deep neural network is obtained.

[0027] Optionally, the expression for the loss function is:

[0028] Loss = Loss MSE (input,recon)+λΣ||w||2;

[0029] Among them, Loss MSE λ represents the mean squared error loss, input represents the true value of the coefficient of variation, recon represents the predicted value of the output coefficient of variation, λ represents the regularization coefficient, and w represents the weighting coefficient.

[0030] Optionally, the process of iteratively updating the fractal dimension, porosity, and characteristic frequencies using a genetic algorithm with the objective of minimizing the coefficient of variation through selection, mutation, and crossover includes:

[0031] Generate multiple initial parameter combinations containing fractal dimension, porosity, and characteristic frequency according to the constraints. Input the initial parameter combinations into the deep neural network and output the coefficient of variation. Sort the coefficients of variation in ascending order and filter and retain the initial parameter combinations corresponding to the first preset number of coefficients of variation.

[0032] The initial parameter combination is updated through crossover and mutation. The updated parameter combination is then input into a deep neural network, and the output mutation coefficients are sorted. The parameter combinations corresponding to the first preset number of mutation coefficients are selected and retained until the preset termination condition is met, at which point the target fractal dimension, target porosity, and target feature frequency are obtained.

[0033] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the above-described method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration.

[0034] The beneficial effects of this invention are as follows:

[0035] This invention provides a method for optimizing the structure of porous dehumidifiers suitable for ultrasonic regeneration. It achieves a precise description of the porous dehumidifier through geometric performance parameters, material property parameters, and operating condition parameters. A closed-loop generation algorithm is used to strictly control porosity error, and a three-dimensional geometric model is generated by combining voxel mapping and triangulation, improving the accuracy of geometric parameter control and reducing porosity error. The coefficient of variation is introduced as a quantitative indicator of vibration uniformity, providing a clear target for optimization. A solid mechanics-pressure acoustic dual-physics coupling method is used for simulation to reconstruct the vibration interaction between the three-dimensional models and improve design accuracy. Deep neural networks and genetic algorithms are used for parameter combination and iterative updates, improving update efficiency and computational resource utilization. Based on multiple parameter combinations, porous dehumidifier designs can be implemented to meet the needs of various scenarios.

[0036] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0037] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0038] Figure 1This is a schematic flowchart of a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration, as described in an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the structure optimization method for porous dehumidifying materials suitable for ultrasonic regeneration, according to an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of a polymorphic perforated structure according to an embodiment of the present invention.

[0041] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0043] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0044] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0045] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0046] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0047] Figure 1 This is a flowchart illustrating a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration, according to an embodiment of the present invention. This application provides a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration, comprising the following steps S101 to S105:

[0048] Step S101: Construct a parameterized resource library including geometric performance parameters, material property parameters, and operating condition parameters; the geometric performance parameters include fractal dimension and porosity.

[0049] Step S102: Generate a polymorphic porous structure with a porosity error below a preset threshold based on a closed-loop generation algorithm. Obtain a three-dimensional geometric model file by mapping voxels to a three-dimensional solid and performing triangulation. Simulate the three-dimensional geometric model file based on the solid mechanics-pressure acoustic dual-physics field coupling method to obtain a three-dimensional model.

[0050] Step S103: Apply multiple vibration excitations within the characteristic frequency range to any face of the three-dimensional model according to the material property parameters and working condition parameters, collect the corresponding vibration displacement amplitude, and obtain the coefficient of variation to represent the uniformity of vibration dispersion according to the preset algorithm.

[0051] Step S104: Construct a training dataset using fractal dimension, porosity, feature frequency, and coefficient of variation, and use the training dataset to train the initial deep neural network to obtain the deep neural network.

[0052] Step S105: A genetic algorithm is used to iteratively update the fractal dimension, porosity, and feature frequency with the goal of minimizing the coefficient of variation through selection, mutation, and crossover until a preset termination condition is met, thereby obtaining the target fractal dimension, target porosity, and target feature frequency; wherein, in each round of iterative update, the fractal dimension, porosity, and feature frequency are input into a deep neural network and the coefficient of variation is output.

[0053] In step S101, the fractal dimension of the parameterized resource library is in the range of 1.5 to 2.8, the porosity is in the range of 30% to 70%, and the fractal dimension (D) is... f Porosity () represents the degree of self-similarity of porous structures. The data represents the proportion of pores within the structure; material property parameters include the elastic modulus E(T); the operating condition dataset includes excitation type, characteristic frequency range, and ambient temperature range. The excitation type is a sinusoidal excitation with an amplitude of 0.1 N, the characteristic frequency range is 15 kHz to 30 kHz, and the ambient temperature range is 25 to 80 degrees Celsius; the material property is a temperature-dependent material property, expressed as: E(T) = E0(1 - c1(TT)). ref )-c2(TT ref ) 2 Where E0 represents the room temperature reference elastic modulus, c1 and c2 represent the material property degradation coefficients, and T ref The reference temperature is indicated; furthermore, the deviation between the actual porosity value and the preset requirement value is verified in real time using stereochemical methods.

[0054] In step S102, the parameterized resource library, combined with the closed-loop generation algorithm, achieves a fractal porous structure with a porosity error within ±0.1%. In this embodiment of the invention, the process of generating a polymorphic porous structure with a porosity error below a preset threshold based on the closed-loop generation algorithm, and obtaining a three-dimensional geometric model file by mapping voxels to a three-dimensional entity and performing triangulation includes steps S1021 to S1023:

[0055] Step S1021: Based on the porosity setting requirements, initial pore voxels are randomly selected from a specific number of rows and columns of the pixel array. The pore voxels are expanded to the surrounding area according to the preset guard erosion strategy until the target number of pore voxels is reached. Then, the surrounding area of ​​the pixel array is set as solid voxels to obtain a two-dimensional graphic.

[0056] Step S1022: After mapping multiple solid voxels in the two-dimensional graphic to three-dimensional units, a three-dimensional solid is obtained. Each face of the three-dimensional unit is triangulated to obtain 12 triangular facets to construct a triangular mesh. Then, the three-dimensional geometric model file is output through a preset file writing function.

[0057] Step S1023: Obtain the fractal dimension of the two-dimensional graphic using box counting.

[0058] Specifically, the n*n pixel array is divided into sub-blocks based on porosity requirements, and initial void voxels are randomly seeded. The preset guard erosion strategy adopts a 4-connected guard erosion strategy to ensure that the voids form only a single connected domain. The voids are hollowed out point by point from the initial void voxel outwards, and the bwlabel function is called to detect the number of connected domains. If a break occurs, the process is reversed. The void elements at the edge are forcibly converted into solid voxels to generate a two-dimensional graphic with a void-free boundary structure, and the number of void voxels is adjusted at the same time. By generating cuboids of preset length and height at the position of solid voxels, the solid voxels are mapped to cubic three-dimensional units in 3D space and then spliced ​​into a complete three-dimensional entity. Triangulation discretizes the surface of the three-dimensional entity into triangular facets. A three-dimensional entity contains six faces, each face is divided into two triangular facets, and a three-dimensional entity has 12 triangular facets that form a triangular network. The three-dimensional geometric model file (STL) is exported through stlwrite. STL is a standard file format used for 3D printing and CAD design, which stores the vertex information and normal vector information of the triangular facets.

[0059] Furthermore, the box counting method uses square boxes of fixed size ε to cover the two-dimensional image and counts the minimum number of boxes N(ε). It calculates log(1 / ε) and logN(ε) and performs first-order polynomial fitting, and the slope is the fractal dimension.

[0060] In this embodiment of the invention, the simulation of a three-dimensional geometric model file based on the solid mechanics-pressure acoustic dual-physics field coupling method to obtain a three-dimensional model includes: receiving the three-dimensional geometric model file through preset simulation software and performing solid mechanics field modeling on solid voxels and pressure acoustic field modeling on pore voxels.

[0061] Specifically, solid mechanical field modeling sets temperature-dependent material properties to describe the vibration and stress-strain behavior of solids, including but not limited to elastic modulus, density, and Poisson's ratio; pressure acoustic field modeling sets acoustic medium properties to describe sound wave transmission and pressure changes within voids.

[0062] In step S103, in this embodiment of the invention, the process of applying multiple vibration excitations within a characteristic frequency range to any face of the three-dimensional model according to material property parameters and working condition parameters, and acquiring the corresponding vibration displacement amplitude, includes steps S1031 to S1033:

[0063] Step S1031: Select one face of each three-dimensional model as the vibration surface for applying vibration excitation, and the face without excitation is the free surface. Set coupling conditions for the contact surface of the solid voxel and the porous voxel to represent the interaction between the solid and the air. The vibration excitation is a sinusoidal displacement excitation in the characteristic frequency range selected according to the material property parameters and the ambient temperature.

[0064] Step S1032: Select multiple frequency points within the characteristic frequency range at preset intervals and perform simulation on each frequency point to collect the vibration displacement amplitude at monitoring points uniformly set on each surface of the three-dimensional model.

[0065] Specifically, an excitation is applied to one side of a 3D model to induce regular vibrations. The 3D unit is a solid, and the voxel locations contain air pores. Coupling conditions are set at the interface between the solid and the air. The vibration of the 3D unit drives the air within the pores to generate sound waves. The pressure changes generated by the sound wave transmission react on the 3D unit and change its vibration state, realizing the interaction of two physical fields. The fractal dimension, porosity, characteristic frequency, and vibration displacement amplitude constitute a four-dimensional original dataset, and a deep neural network is trained based on the fractal dimension, porosity, characteristic frequency, and coefficient of variation. At least 20 frequency points are selected at equal intervals within the 15kHz to 30kHz frequency band to simulate corresponding frequency scenes, and the vibration displacement amplitude at the set monitoring points is collected. The detection points are evenly distributed on the surface of the 3D model.

[0066] Furthermore, the coefficient of variation (CV) is calculated based on the characteristic frequency and the corresponding vibration displacement amplitude. A smaller CV indicates better vibration dispersion uniformity and more efficient ultrasonic wave propagation. In this embodiment of the invention, the CV representing vibration dispersion uniformity is obtained according to a preset algorithm. The expression for the preset algorithm is:

[0067]

[0068] Where x represents the vibration displacement amplitude of all monitoring points corresponding to a frequency point, and i represents the index of the frequency point. i This represents the vibration displacement amplitude of all monitoring points corresponding to frequency point i, and N represents the total number of frequency points.

[0069] In step S104, the deep neural network includes a Fourier frequency domain smoothing filter layer to suppress high-frequency oscillations and achieve an average error of less than or equal to 5% in the predicted value of the coefficient of variation.

[0070] In this embodiment of the invention, the process of training an initial deep neural network with a training dataset to obtain a deep neural network includes steps S1041 to S1043:

[0071] Step S1041: Obtain a training sample set from the training dataset according to a preset partitioning ratio. The training sample set contains multiple samples, each of which contains fractal dimension, porosity, and feature frequency. The samples in the training sample set are labeled with the true value of the coefficient of variation.

[0072] Step S1042: Train the initial deep neural network using samples from the training sample set. The initial neural network includes an input layer, multiple hidden layers, a Fourier frequency domain smoothing filter layer, and an output layer. The input layer contains three neurons for inputting fractal dimension, porosity, and feature frequency, respectively. After the multiple hidden layers extract feature data through a nonlinear activation function, the Fourier frequency domain smoothing filter layer transforms the feature data to the frequency domain through Fourier transform to filter high-frequency oscillation noise. The output layer outputs the predicted value of the coefficient of variation.

[0073] Step S1043: Construct a loss function using the true value and predicted value of the coefficient of variation. With the goal of minimizing the loss function, use the Adam optimizer to iteratively update the parameters of the initial deep neural network according to the preset learning rate until the preset number of iterations is reached to obtain the deep neural network.

[0074] Specifically, the training dataset is divided into training, validation, and test sets in a ratio of 7:1.5:1.5. The fractal dimension, porosity, and feature frequencies in the training set are used as training samples, and the coefficient of variation in the training set is used as labels to train the initial deep neural network. The hidden layer contains three layers with 64, 32, and 16 neurons respectively, and the output layer contains one neuron to output the predicted value of the coefficient of variation. The Fourier frequency domain smoothing filter layer is a Fourier space and smoothing filter layer. After the feature data is transformed into the frequency domain, a variable window exponential weighted smoothing filter method is used to suppress high-frequency oscillations.

[0075] In this embodiment of the invention, the expression for the loss function is:

[0076] Loss = Loss MSE (input,recon)+λΣ||w||2;

[0077] Among them, Loss MSE λ represents the mean squared error loss, input represents the true value of the coefficient of variation, recon represents the predicted value of the output coefficient of variation, λ represents the regularization coefficient, and w represents the weighting coefficient.

[0078] Furthermore, the learning rate was set to 0.001, and the number of iterations was set to 500; the loss function consisted of mean squared error loss and regularization loss; the regularization coefficient λ = 0.001.

[0079] In step S105, a genetic algorithm is used to optimize the fractal dimension, porosity, and characteristic frequency to obtain a porous dehumidifying material design scheme that meets the requirement of uniform ultrasonic wave propagation. In this embodiment of the invention, the process of using a genetic algorithm to iteratively update the fractal dimension, porosity, and characteristic frequency with the goal of minimizing the coefficient of variation through selection, mutation, and crossover includes steps S1051 to S1052:

[0080] Step S1051: Generate multiple initial parameter combinations including fractal dimension, porosity and characteristic frequency according to the constraints. Input the initial parameter combinations into the deep neural network and output the coefficient of variation. Sort the coefficients of variation in ascending order and filter and retain the initial parameter combinations corresponding to the first preset number of coefficients of variation.

[0081] Step S1052: Obtain updated parameter combinations by crossover and mutation of the initial parameter combinations. Input the updated parameter combinations into the deep neural network and output the mutation coefficients. Sort the parameters and filter and retain the parameter combinations corresponding to the first preset number of mutation coefficients until the preset termination condition is reached to obtain the target fractal dimension, target porosity and target feature frequency.

[0082] Specifically, the constraints are: fractal dimension within the range of 1.5–2.8, porosity within the range of 30%–70%, and characteristic frequency within the range of 15kHz–30kHz. The coefficient of variation output by the parameter combination after inputting it into the deep neural network is used as the fitness function to measure the vibration dispersion uniformity of a parameter combination. The smaller the coefficient of variation, the larger the fitness function, and the parameter combination with the smaller coefficient of variation is retained. Crossover and mutation operations are performed on the parameter combination to generate new parameter combinations, and the coefficient of variation output by the deep neural network is input. The crossover and mutation operations are stopped after reaching the preset number of iterations or the convergence condition to obtain the target parameter combination with the smallest coefficient of variation. The target parameter combination includes the target fractal dimension, target porosity, and target characteristic frequency. Based on the target parameter combination, a porous dehumidifying material with good ultrasonic propagation effect is designed. Furthermore, the porous dehumidifying material is a silica gel-based dehumidifying material, which can improve the propagation uniformity and energy utilization efficiency of ultrasonic waves in the material and can be used for ultrasonic regeneration in the 15kHz–30kHz ultrasonic frequency band.

[0083] On the other hand, embodiments of the present invention also provide a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the above-described method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration.

[0084] The present invention will now be described with reference to a specific embodiment:

[0085] Figure 2 This is a schematic diagram illustrating the structural optimization method for porous dehumidifying materials suitable for ultrasonic regeneration, as described in an embodiment of the present invention. The present invention proposes a structural optimization method for porous dehumidifying materials suitable for ultrasonic regeneration.

[0086] Step 1: Construct a parameterized resource library for fractal porous structures.

[0087] 1.1 Definition of Geometric Performance Parameters: Fractal dimension and porosity are the core design parameters. The fractal dimension is denoted as D. f , a continuous variable characterizing the self-similarity of porous structures, ranging from 1.5 to 2.8; porosity is denoted as A continuous variable characterizing the proportion of pores within the structure, ranging from 30% to 70%; the deviation between the calculated porosity and the preset value is verified in real time using stereoscopic methods, with a value ≤3%.

[0088] Material property parameters include temperature-dependent material properties, including elastic modulus E(T), density ρ, and Poisson's ratio υ; the elastic modulus varies with ambient temperature, expressed as E(T) = E0(1-c1(TT)). ref )-c2(TT ref ) 2Where E0 represents the room temperature reference elastic modulus, c1 and c2 represent the material property degradation coefficients, and T ref This indicates a reference temperature to ensure that the material properties match the actual working conditions.

[0089] The operating condition dataset covers the target application scenarios, including vibration excitation type, characteristic frequency range, and ambient temperature range. The vibration excitation type is sinusoidal excitation with an amplitude of 0.1N, the characteristic frequency range is 15kHz to 30kHz, and the ambient temperature range is 25 to 80℃. Each operating condition scenario corresponds to specific load conditions and material property parameters, providing standardized input for subsequent simulations.

[0090] Figure 3 This is a schematic diagram of a polymorphic porous structure according to an embodiment of the present invention. 1.2 The closed-loop generation algorithm generates a fractal porous structure by accurately determining the two-dimensional controllable porosity, three-dimensional solidification, and fractal dimension, including: on an n×n pixel array, the target porosity... The error is controlled within ±0.1%; it ensures that the pores exist in only a single connected domain in the 4-connectivity sense; it automatically generates 3D STL solid files corresponding to pixels, which can be directly used for 3D printing and numerical simulation; it adopts a strict box counting strategy of "not magnifying pixels" to give a fractal dimension D consistent with the solid. f .

[0091] The parametric resource library for fractal porous structures is constructed by coupling three main functional modules in sequence:

[0092] (1) Uniform porous structure generation module.

[0093] Input: Array size n, target porosity The size of the sub-block is `blockSize`, and the random seed is `seed`.

[0094] Sub-block partitioning: Divide the n×n array into n / blockSize 2 The number of holes to be excavated is distributed evenly among the individual blocks according to their porosity.

[0095] Random seeding: Randomly select candidate positions within each sub-block to form a candidate list cand.

[0096] 4-Connectivity Guard Erosion: Hollow out point by point and call the bwlabel function to detect the number of connected components, perform PID feedback, and backtrack if a break occurs to ensure that only a single solid phase is retained in the end.

[0097] Forced boundary closure: Forces surrounding pixels to be solid, forming a non-through boundary, which facilitates the application of subsequent fluid or mechanical boundary conditions.

[0098] Output: A binary logic matrix P, where true represents solid and false represents pores.

[0099] (2) 3D solidification module.

[0100] Input: P, number of actual solid blocks n true Array side length, single block height, and case name name .

[0101] Voxel to Entity Mapping: For a pixel P(i,j)==true, generate a cuboid with length L=length / n and height at position (i,j).

[0102] Vertex-face pre-allocation: 8×n pre-allocations in one go true vertices and 12×n true This uses a triangular facet to avoid the overhead of MATLAB's dynamic expansion.

[0103] Local coordinate calculation: Generate the coordinates of 8 vertices according to the right-hand rule, and decompose them into 12 triangular patches according to the STL format requirements.

[0104] Triangulation object construction: calling triangulation generates a closed manifold.

[0105] STL file writing: Call stlwrite to export STL files, which can be directly read by 3D printing slicing software or finite element preprocessors.

[0106] (3) Fractal dimension measurement module.

[0107] Input: A two-dimensional binary matrix BW with the same size as P.

[0108] Multi-scale scanning: Using a preset scale sequence scales = [2,3,4,6,8,12,24], strict box counting is performed on the BW without "magnifying pixels". That is, for each s×s block, if any point in the block is true, the count is incremented by 1.

[0109] Double logarithmic fitting: Calculate log(1 / ε) and logN(ε) and perform first-order polynomial fitting. The slope is the box count dimension D.

[0110] Output: D, scales, and N; optionally, a log-log plot can be drawn to visualize the results.

[0111] Step 2: Create multiphysics simulation and original dataset. Based on the COMSOL multiphysics coupling simulation approach, conduct vibration performance simulation of fractal porous structures.

[0112] Physics field and boundary condition settings: Solid mechanics is used for the solid part of the skeleton, and pressure acoustics is used for the porous part.

[0113] Boundary conditions are set: a specified displacement excitation is applied to one side of the structure, with the amplitude set to A0, while the rest of the structure is set to free; the solid-gas interface adopts acoustic-structural coupling.

[0114] Within the characteristic frequency range of 15kHz to 30kHz, m frequency points are selected at equal intervals, where m≥20, to ensure complete frequency coverage. Characteristic frequency modal simulation is performed on each fractal porous structure sample. 100 monitoring points are uniformly distributed on the surface of the structure, and vibration displacement amplitude data of the monitoring points at each frequency point are collected to form a four-dimensional original dataset including fractal dimension, porosity, characteristic frequency, and vibration displacement amplitude.

[0115] Mesh accuracy control: The mesh cell size of the STL file is set to ≤0.1mm. Local mesh refinement is performed on the vibration-sensitive areas of the structure, including the edges of pores and structural connection parts. The mesh cell size of the refined area is 1 / 2 of that of the regular area to ensure the calculation accuracy of key areas in subsequent simulations.

[0116] Step 3: Quantitative Evaluation of Vibration Dispersion Uniformity. A vibration kinetic energy dispersion uniformity evaluation system based on the coefficient of variation (CV) is constructed to achieve quantitative characterization of performance. The CV calculation formula is as follows: CV = Where x is the amplitude of the vibration displacement at all monitoring points at a certain characteristic frequency. The CV value is the average of the squared amplitudes of vibration displacements at all monitoring points at the same frequency, representing the concentration trend of vibration kinetic energy distribution. The larger the CV value, the more significant the difference in vibration distribution at that frequency, and the worse the uniformity of vibration dispersion. For each porous structure sample, the CV value at each frequency point in the range of 15kHz to 30kHz is calculated, and the CV values ​​are added to the four-dimensional original dataset to form a standardized training sample library including fractal dimension, porosity, characteristic frequency, and CV value.

[0117] Step 4: Construct and train the neural network prediction model. The deep neural network is based on an improvement of the Fourier leaf space-smoothing filter layer, used to achieve accurate prediction of CV values.

[0118] 4.1 The network structure design adopts a fully connected deep neural network, consisting of an input layer, three hidden layers, and an output layer. The input layer contains three neurons, corresponding to the fractal dimension, porosity, and feature frequency, respectively. The output layer contains one neuron, corresponding to the CV value. The number of neurons in the hidden layers are 64, 32, and 16, respectively. A Fourier space-smoothing filter layer is embedded in the last layer of the network to receive the prediction data output from the fully connected layer. High-frequency prediction oscillations are processed using a variable window exponential weighted smoothing filter method. The window weights range from 0.2 to 0.9 and are dynamically adjusted according to the degree of data oscillation. The filtered data is then processed through matrix operations, and the concatenated matrix is ​​as follows: M1 and M3 are edge filtering matrices, M2 is the intermediate filtering matrix, and the final output is the CV prediction value.

[0119] 4.2 Model Training and Accuracy Validation. The standardized training sample library was divided into training, validation, and test sets in a ratio of 7:1.5:1.5. The Adam optimizer was used with a learning rate of 0.001 and 500 iterations. The total loss function consisted of mean squared error (MSE) and a regularization term, expressed as Loss = Loss MSE (input,recon)+λΣ||w||2; where λ=0.001 is the regularization coefficient and w is the network weight; after training, the average absolute error of the CV value prediction on the test set is ≤5%, ensuring that the model can accurately learn the nonlinear mapping relationship between the input parameters and the CV value.

[0120] Step 5: Intelligent optimization of porous structure parameters. Based on the predictive model and intelligent algorithm, parameter optimization is achieved to achieve optimal vibration dispersion uniformity.

[0121] 5.1 Optimization Objectives and Constraints. The core objective is to minimize the CV value, which corresponds to optimal vibration dispersion uniformity. The constraint is the fractal dimension D. f Porosity is between 1.5 and 2.8. Between 30% and 70%, the characteristic frequency f is between 15 kHz and 30 kHz.

[0122] 5.2 Optimization Algorithm and Iteration Process. A genetic algorithm is used, with the CV value predicted by the neural network as the fitness function. The algorithm parameters include a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.05. The optimization process terminates when the CV value change rate is less than 3% after three consecutive iterations, or when the number of iterations reaches 100. The optimal parameter combination that satisfies the constraints is output and denoted as D. f , f and the corresponding minimum CV value provide precise design parameters for the actual fabrication of porous structures.

[0123] In summary, this invention provides a method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration. The method includes: constructing a parameterized resource library and generating a polymorphic porous structure with a porosity error below a preset threshold based on a closed-loop generation algorithm; obtaining a three-dimensional geometric model file by mapping voxels to a three-dimensional entity and performing triangulation; simulating the three-dimensional geometric model file using a solid mechanics-pressure acoustic dual-physics coupling method to obtain a three-dimensional model; applying multiple vibration excitations within a characteristic frequency range to any face of the three-dimensional model and collecting the corresponding vibration displacement amplitudes; obtaining the coefficient of variation according to a preset algorithm; constructing a training dataset using the fractal dimension, porosity, characteristic frequency, and coefficient of variation to train an initial deep neural network to obtain a deep neural network; and using a genetic algorithm to iteratively update the fractal dimension, porosity, and characteristic frequency through selection, mutation, and crossover with the goal of minimizing the coefficient of variation, until a preset termination condition is reached to obtain the target fractal dimension, target porosity, and target characteristic frequency.

[0124] This invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the above-described method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration.

[0125] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration, characterized in that, The method includes: Construct a parameterized resource library that includes geometric performance parameters, material property parameters, and operating condition parameters; the geometric performance parameters include fractal dimension and porosity; A polymorphic porous structure with a porosity error below a preset threshold is generated based on a closed-loop generation algorithm. The structure is then mapped to a 3D solid and triangulated to obtain a 3D geometric model file. The 3D geometric model file is then simulated using a solid mechanics-pressure acoustic dual-physics coupling method to obtain a 3D model. Multiple vibration excitations within the characteristic frequency range are applied to any face of the 3D model based on material property parameters and working condition parameters, and the corresponding vibration displacement amplitude is collected. The coefficient of variation representing the uniformity of vibration dispersion is obtained according to a preset algorithm. The fractal dimension, porosity, feature frequency and coefficient of variation are used to construct a training dataset, and the initial deep neural network is trained on the training dataset to obtain a deep neural network. A genetic algorithm is used to iteratively update the fractal dimension, porosity, and feature frequencies with the goal of minimizing the coefficient of variation through selection, mutation, and crossover until a preset termination condition is met, thereby obtaining the target fractal dimension, target porosity, and target feature frequencies. In each round of iterative update, the fractal dimension, porosity, and feature frequencies are input into a deep neural network and the coefficient of variation is output.

2. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The process of generating a polymorphic porous structure with a porosity error below a preset threshold based on a closed-loop generation algorithm, and obtaining a 3D geometric model file by mapping voxels to a 3D solid and performing triangulation, includes: Based on the porosity setting requirements, initial pore voxels are randomly selected from a pixel array of a specific number of rows and columns. The pore voxels are expanded to the surrounding area according to the preset guard erosion strategy until the target number of pore voxels is reached. Then, the surrounding area of ​​the pixel array is set as solid voxels to obtain a two-dimensional graphic. After mapping multiple solid voxels in a two-dimensional graphic to three-dimensional units, a three-dimensional solid is obtained. Each face of the three-dimensional unit is triangulated to obtain 12 triangular facets to construct a triangular mesh. Then, a three-dimensional geometric model file is output through a preset file writing function. The fractal dimension of a two-dimensional graphic is obtained using box counting.

3. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The three-dimensional model is obtained by simulation of the three-dimensional geometric model file based on the solid mechanics-pressure acoustic dual-physics coupling method, including: The system receives 3D geometric model files through pre-set simulation software and performs solid mechanical field modeling on solid voxels and pressure acoustic field modeling on porous voxels.

4. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The process of applying multiple vibration excitations within a characteristic frequency range to any face of a 3D model based on material property parameters and operating condition parameters, and acquiring the corresponding vibration displacement amplitudes, includes: One face of each 3D model is selected as the vibration surface for applying vibration excitation, and the face without excitation is the free face. Coupled conditions are set for the contact surface of the solid voxel and the porous voxel to represent the interaction between the solid and the air. The vibration excitation is a sinusoidal displacement excitation in the characteristic frequency range selected according to the material property parameters and the ambient temperature. Multiple frequency points are selected at preset intervals within the characteristic frequency range, and simulation is performed on each frequency point. The vibration displacement amplitude at monitoring points uniformly set on each surface of the three-dimensional model is collected.

5. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The coefficient of variation representing the uniformity of vibration dispersion is obtained according to a preset algorithm. The expression of the preset algorithm is as follows: Where x represents the vibration displacement amplitude of all monitoring points corresponding to a frequency point, and i represents the index of the frequency point. i This represents the vibration displacement amplitude of all monitoring points corresponding to frequency point i, and N represents the total number of frequency points.

6. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The process of training an initial deep neural network using a training dataset to obtain a deep neural network includes: The training sample set is obtained from the training dataset according to the preset division ratio. The training sample set contains multiple samples, each of which contains fractal dimension, porosity and feature frequency. The samples in the training sample set are labeled with the true value of the coefficient of variation. The initial deep neural network is trained using samples from the training sample set. The initial neural network includes an input layer, multiple hidden layers, a Fourier frequency domain smoothing filter layer, and an output layer. The input layer contains three neurons for inputting fractal dimension, porosity, and feature frequency, respectively. After the multiple hidden layers extract feature data through a nonlinear activation function, the Fourier frequency domain smoothing filter layer transforms the feature data to the frequency domain through Fourier transform to filter high-frequency oscillation noise. The output layer outputs the predicted value of the coefficient of variation. A loss function is constructed using the true and predicted values ​​of the coefficient of variation. The Adam optimizer is used to iteratively update the parameters of the initial deep neural network with the goal of minimizing the loss function, until the preset number of iterations is reached, and then the deep neural network is obtained.

7. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The expression for the loss function is: Loss=Loss MSE (input,recon)+λΣ||w||2; Among them, Loss MSE λ represents the mean squared error loss, input represents the true value of the coefficient of variation, recon represents the predicted value of the output coefficient of variation, λ represents the regularization coefficient, and w represents the weighting coefficient.

8. The method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration according to claim 1, characterized in that, The process of using a genetic algorithm to iteratively update fractal dimension, porosity, and characteristic frequencies with the goal of minimizing the coefficient of variation involves selection, mutation, and crossover. Generate multiple initial parameter combinations containing fractal dimension, porosity, and characteristic frequency according to the constraints. Input the initial parameter combinations into the deep neural network and output the coefficient of variation. Sort the coefficients of variation in ascending order and filter and retain the initial parameter combinations corresponding to the first preset number of coefficients of variation. The initial parameter combination is updated through crossover and mutation. The updated parameter combination is then input into a deep neural network, and the output mutation coefficients are sorted. The parameter combinations corresponding to the first preset number of mutation coefficients are selected and retained until the preset termination condition is met, at which point the target fractal dimension, target porosity, and target feature frequency are obtained.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for optimizing the structure of porous dehumidifying materials suitable for ultrasonic regeneration as described in any one of claims 1 to 8.

Images