Foaming material porous structure parameter optimization method and system based on adversarial network

Abstract

The invention relates to the technical field of porous structure parameter optimization, and particularly discloses a foam material porous structure parameter optimization method based on an adversarial network, which comprises the following steps: S1, analyzing the application field and performance requirements of a foam material, determining an optimization target, and researching an existing foam material porous structure optimization method; s2, designing and constructing a system, designing a visual user interface, allowing a user to input parameters and check an optimization result, designing a network structure of a generator and a discriminator, and then selecting proper CAD software; s3, algorithm development and training: collecting parameter data of the foaming material, and training the adversarial network by using the collected data; and S4, system integration and testing: integrating the GUI, the adversarial network model and the CAD software into one system, and carrying out unit testing, integration testing and system testing. The porous structure of the foaming material is optimized by using GAN, and parameters such as the pore structure, the pore size and the density can be accurately adjusted, so that the physical performance of the material is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of porous structure parameter optimization, and in particular to a method and system for optimizing porous structure parameters of a foaming material based on an adversarial network. Background Art

[0002] With the rapid development of materials science, foam materials are widely used in the fields of construction, aerospace, and automotive industries due to their lightweight, efficient heat insulation, sound insulation, and cushioning properties. The porous structure of foam materials plays a vital role in their performance, among which the structural parameters such as pore distribution, size, and density directly affect the mechanical properties, thermal properties, and other functional performances of the material. Therefore, optimizing the porous structure parameters of foam materials has become a key technology to improve their performance.

[0003] At present, traditional foam material structure optimization methods require complex experimental steps and a long optimization cycle, and are limited by experimental conditions and operating experience.

[0004] Therefore, how to quickly optimize the pore structure of foaming materials while ensuring high efficiency and accuracy has become an urgent problem to be solved. Summary of the invention

[0005] The embodiment of the present invention provides a method and system for optimizing porous structure parameters of a foam material based on an adversarial network, so as to overcome the above-mentioned technical problems existing in the prior art.

[0006] A method for optimizing porous structure parameters of a foaming material based on an adversarial network comprises the following steps: S1: Analyze the application fields and performance requirements of foaming materials, determine the optimization targets, and study the existing optimization methods of porous structures of foaming materials, including template method, non-solvent induced phase separation method, freeze drying method and supercritical CO2 foaming method; S2: System design and architecture: design an intuitive user interface that allows users to input parameters and view optimization results, and design the network structure of the generator and discriminator to simulate and optimize the porous structure of the foaming material. Then select appropriate CAD software, including AutoCAD or Solidworks, for simulation and visualization of the porous structure. S3: Algorithm development and training, collect parameter data of foaming materials, including pore structure, pore size and density, and use the collected data to train the adversarial network to identify and optimize the porous structure parameters of foaming materials; S4: System integration and testing: Integrate the GUI, adversarial network model and CAD software into a system, and perform unit testing, integration testing and system testing to ensure stable operation of the system.

[0007] Preferably, in S3, parameter data of the foaming material, including pore structure, pore size and density, is collected, and the collected data is used to train the adversarial network, wherein the network structures of the generator and the discriminator are as follows: Generator network: It uses a multi-layer perceptron (MLP) structure. The input layer receives the user's initial parameters, and the output layer generates optimized porous structure parameters. Discriminator network: It adopts a convolutional neural network (CNN) structure. The input layer receives the porous structure parameters generated by the generator and the actual porous structure parameters, and the output layer evaluates the truth or falsity of the input parameters.

[0008] Preferably, the loss function of the generator network is: ; in, are the porous structure parameters generated by the generator; is the actual porous structure parameter; is the evaluation result of the discriminator on the parameters of the generated porous structure; is a hyperparameter for adjusting weights; The loss function of the discriminator network is: ; in, is the input porous structure parameter, is the real label, 1 represents real data, 0 represents generated data, is the size of the dataset.

[0009] Preferably, the generator network comprises 5 fully connected layers, the number of neurons in each fully connected layer is 128, 256, 512, 256, and 128 respectively, and the discriminator network comprises 5 convolutional layers, the convolution kernel size of each convolutional layer is 3x3, 5x5, 7x7, 5x5, and 3x3 respectively, and the number of convolution kernels is 32, 64, 128, 64, and 32 respectively; Each layer of the generator network uses the ReLU activation function, the output layer uses the tanh activation function, each layer of the discriminator network uses the LeakyReLU activation function, and the output layer uses the sigmoid activation function; The generator uses the Adam optimizer with a learning rate of 0.0002 and a momentum parameter of 0.5, and the discriminator uses the RMSProp optimizer with a learning rate of 0.0001.

[0010] Preferably, in step S4, the GUI, the adversarial network model and the CAD software are integrated into one system, and the specific steps are as follows: S4.1: Integration of GUI and adversarial network model: Use Python's Flask framework to build a web server, connect the user interface with the adversarial network model, and the user inputs the initial parameters of the foaming material through the web interface. The system passes the parameters to the generator to generate the optimized porous structure parameters. S4.2: Integration with CAD software: Use the API interface to pass the optimized porous structure parameters to the CAD software to generate a three-dimensional model of the porous structure. Specifically, use the COM API interface of AutoCAD to call the function of AutoCAD through a Python script to convert the optimized porous structure parameters into a three-dimensional model; S4.3: Unit testing: Perform unit testing on each component of the generator and discriminator to ensure that the function of each component is correct. The input and output functions of the generator are tested to ensure that the generator can correctly receive the initial parameters input by the user and generate the optimized porous structure parameters. S4.4: Integration test: Integrate the generator, discriminator and CAD software into a system and conduct integration test to ensure that the data transfer and functional collaboration between the components are correct, and test whether the porous structure parameters generated by the generator can be correctly transferred to the CAD software and generate a three-dimensional model; S4.5: System test: Perform system test on the whole system to ensure that the system is stable and reliable in actual use. Test whether the system can correctly generate the optimized porous structure parameters and display them in the result display module after the user inputs the parameters through the web interface.

[0011] Preferably, in S4.4 and S4.5, detailed log information is recorded, including input parameters of each test case, generated porous structure parameters, generated three-dimensional model and test results, and the log information is stored in a database to facilitate subsequent analysis and debugging.

[0012] Preferably, in S1, the template method is to control the pore structure of the foaming material by a template, and its key parameters include the pore size of the template, the arrangement of the template, the density of the template, the type and amount of the foaming agent; The non-solvent induced phase separation method uses a non-solvent to induce phase separation to form a porous structure. The key parameters include: the choice of solvent, the choice of non-solvent, the concentration of the solution, and the cooling rate; Freeze drying is to solidify the liquid by freezing, and then sublimate and dehydrate in a vacuum to form a porous structure. The key parameters include: freezing temperature, freezing time, sublimation temperature, and sublimation time; The supercritical CO2 foaming method uses supercritical CO2 as a foaming agent and controls the foaming process by changes in pressure and temperature. The key parameters include: CO2 pressure, temperature, foaming time, type and amount of foaming agent.

[0013] Accordingly, the present invention also provides a foam material porous structure parameter optimization system based on an adversarial network, comprising: User interface module: Use Python's Flask framework to build a web server and design an intuitive user interface to allow users to input the initial parameters of the foaming material and view the optimization results. This module interacts with the adversarial network model through HTTP requests, and the parameters entered by the user are passed to the adversarial network model module in JSON format; Adversarial network model module: including generator network and discriminator network, using multi-layer perceptron (MLP) and convolutional neural network (CNN) structures; CAD software module: Select AutoCAD or Solidworks as the simulation and visualization tool for the porous structure, use the API interface to pass the optimized porous structure parameters to the CAD software, and generate a three-dimensional model of the porous structure. Specifically, use the COM API interface of AutoCAD to call the function of AutoCAD through the Python script to convert the optimized porous structure parameters into a three-dimensional model; Data collection module: collects parameter data of foaming materials, including pore structure, pore size and density, through sensors and other data sources; Data storage module: stores the collected data in a database for training and testing adversarial network models. The database uses MySQL or SQLite. Test module: includes unit test, integration test and system test. The unit test module tests each component of the generator and discriminator to ensure that each component functions correctly. The integration test module integrates the generator, discriminator and CAD software into a system and performs integration test to ensure that the data transfer and functional collaboration between the components are correct.

[0014] Preferably, it also includes a real-time monitoring module, which collects the generated porous structure parameters in real time through sensors and dynamically adjusts the training parameters of the generator and the discriminator; The user interface module has user authority management function and multi-language support function. The user authority management function distinguishes the operation authority of different users, and the multi-language support function supports multi-language interfaces including Chinese and English.

[0015] Preferably, it also includes a cloud storage module, which stores the generated three-dimensional model and test results in the cloud and supports users to view and download data on different devices.

[0016] Beneficial Effects

[0017] 1. By using the generative adversarial network (GAN) to optimize the porous structure of the foam material, it is possible to accurately adjust parameters such as the pore structure, pore size, and density, thereby significantly improving the physical properties of the material, such as strength, thermal conductivity, sound insulation, etc., to meet the needs of different application fields.

[0018] 2. By combining the MLP and CNN structures, the generator and discriminator each take advantage of their own strengths in processing different data types. MLP is used in the generator to process continuous data input by the user (such as parameter settings), while CNN is used in the discriminator to process the spatial features of structural parameters. This combination makes model training more efficient and flexible.

[0019] 3. The optimization of the porous structure parameters of the foam material by adversarial network greatly improves the degree of automation of the design. The user only needs to input the initial parameters, and the system can automatically generate the optimization results that meet the requirements and convert them into a three-dimensional model, eliminating the tedious process of traditional manual design.

[0020] 4. Through detailed logging, the input, generated results, models and test output of each test process can be accurately tracked, making it easier to find and solve possible errors.

[0021] 5. By optimizing the combination of different methods, production efficiency can be improved and costs can be reduced without sacrificing performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 It is a flow chart of the steps of a method for optimizing the porous structure parameters of a foaming material based on an adversarial network; Figure 2 It is a system framework for optimizing the porous structure parameters of foam materials based on adversarial networks. Figure 1 ; Figure 3 It is a system framework for optimizing the porous structure parameters of foam materials based on adversarial networks. Figure 2 . DETAILED DESCRIPTION

[0023] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0024] See also Figure 1In an embodiment of the present invention, a method for optimizing porous structure parameters of a foaming material based on an adversarial network comprises the following steps: S1: Analyze the application fields and performance requirements of foaming materials, determine the optimization targets, and study the existing optimization methods of porous structures of foaming materials, including template method, non-solvent induced phase separation method, freeze drying method and supercritical CO2 foaming method; S2: System design and architecture: design an intuitive user interface that allows users to input parameters and view optimization results, and design the network structure of the generator and discriminator to simulate and optimize the porous structure of the foaming material. Then select appropriate CAD software, including AutoCAD or Solidworks, for simulation and visualization of the porous structure. S3: Algorithm development and training, collect parameter data of foaming materials, including pore structure, pore size and density, and use the collected data to train the adversarial network to identify and optimize the porous structure parameters of foaming materials; S4: System integration and testing: Integrate the GUI, adversarial network model and CAD software into a system, and perform unit testing, integration testing and system testing to ensure stable operation of the system.

[0025] In one possible implementation, the optimization target is determined by analyzing different application fields of the foaming material (such as construction, automobiles or electronic products). The performance requirements of the foaming material are closely related to the porous structure parameters (such as pore structure, pore size, density, etc.). According to these requirements, a suitable optimization method is selected, including template method, non-solvent induced phase separation method, freeze drying method and supercritical CO2 foaming method.

[0026] Furthermore, the overall architecture and module design of the system are key. The system design includes an intuitive user interface (GUI) that allows users to input initial parameters and view optimization results. The user interface is designed so that non-professional users can easily interact with the system. At the same time, the network structure of the generator and the discriminator is the core of the optimization. The generator is responsible for generating the optimized porous structure parameters, while the discriminator evaluates whether these parameters conform to the real porous structure. The integration of CAD software (such as AutoCAD or Solidworks) is used to visualize and 3D model the optimized structural parameters, making it easier for users to understand and verify the optimization results.

[0027] Furthermore, the generative adversarial network (GAN) is trained by collecting relevant data of actual foaming materials (such as pore structure, pore size, density, etc.). The generator generates optimized porous structure parameters through training, while the discriminator is used to evaluate the difference between the generated structure and the real data. Through continuous iterative training, the network can identify and optimize the porous structure parameters of the foaming material. The adversarial training between the generator and the discriminator enables the system to provide more accurate optimization results, thereby improving the performance of the foaming material.

[0028] Furthermore, the user interface, generative adversarial network model and CAD software are integrated into a system and system testing is performed to ensure the stability and reliability of the entire system. The system integration can ensure smooth execution from user input to parameter optimization to visual modeling, and ensure that the functions of each module can run stably. In addition, system testing includes unit testing, integration testing and system-level testing to verify the interaction of each component and the overall performance of the system.

[0029] By using the generative adversarial network (GAN) to optimize the porous structure of the foam material, it is possible to accurately adjust parameters such as the pore structure, pore size, and density, thereby significantly improving the physical properties of the material, such as strength, thermal conductivity, and sound insulation, to meet the needs of different application fields. At the same time, the generative adversarial network reduces manual intervention and experience dependence through an automated optimization process, making the porous structure design of the foam material more intelligent and efficient. The user only needs to provide the initial parameters, and the system can automatically generate the optimized structure, simplifying the design process.

[0030] In an embodiment of the present invention, in S3, parameter data of the foaming material, including pore structure, pore size and density, is collected, and the collected data is used to train an adversarial network, wherein the network structures of the generator and the discriminator are as follows: Generator network: adopts a multi-layer perceptron (MLP) structure, the input layer receives the user's initial parameters, and the output layer generates optimized porous structure parameters; Discriminator network: adopts a convolutional neural network (CNN) structure, the input layer receives the porous structure parameters generated by the generator and the actual porous structure parameters, and the output layer evaluates the truth of the input parameters.

[0031] In one possible implementation, step S3 involves training a generative adversarial network (GAN) using the collected foam material parameter data to optimize the porous structure parameters of the material. Specifically, the generator adopts a multi-layer perceptron (MLP) structure, and its main task is to generate optimized porous structure parameters based on the input initial parameters. The MLP structure consists of multiple fully connected layers, each layer contains a number of neurons, and the neurons are connected by weighted connections. The input layer receives the initial parameters provided by the user (such as the initial setting of the pore structure, the pore size range, the material density, etc.), and these initial parameters reflect the user's demand for the foam material. Through multi-layer fully connected calculations, the output layer of the generator network will produce optimized porous structure parameters, such as the detailed size of the pore structure, the pore density, etc.

[0032] Furthermore, the discriminator adopts a convolutional neural network (CNN) structure, whose main task is to evaluate the authenticity of the porous structure parameters output by the generator. The CNN structure is particularly suitable for processing data with spatial structures, so it can efficiently compare the difference between the porous structure output by the generator and the actual porous structure. The input layer of the discriminator receives the porous structure parameters generated by the generator and the actual porous structure parameters (these actual parameters can be obtained from experimental data). The network extracts the features of these structural parameters through multi-layer convolution and pooling operations, and then outputs the true and false probabilities through the fully connected layer, and finally determines whether the generated parameters meet the actual physical and performance requirements.

[0033] The generator and the discriminator form an adversarial training framework. During the training process, the goal of the generator is to optimize its output so that the discriminator cannot distinguish the difference between the generated structure and the actual structure, that is, the generator hopes to "fool" the discriminator. The goal of the discriminator is to do its best to distinguish the parameters generated by the generator from the actual parameters. In this adversarial training, the generator continuously adjusts its network parameters through the back-propagation algorithm to generate parameters that are closer to the actual foam material structure. The discriminator improves its ability to distinguish between real and false data through training, thereby forcing the generator to continuously improve its output.

[0034] Through training of generative adversarial networks, the generator can generate high-quality porous structure parameters, which are verified by the discriminator to ensure that they meet the actual material performance requirements. This adversarial training improves the accuracy of the optimization results, allowing the generated porous structure to achieve the best physical properties (such as strength, thermal conductivity, sound insulation, etc.) while meeting performance requirements.

[0035] In an embodiment of the present invention, the loss function of the generator network is: ;in, are the porous structure parameters generated by the generator; is the actual porous structure parameter; is the evaluation result of the discriminator on the parameters of the generated porous structure; is a hyperparameter for adjusting weights; the loss function of the discriminator network is: ; in, is the input porous structure parameter, is the real label, 1 represents real data, 0 represents generated data, is the size of the dataset.

[0036] In a possible implementation, in the method for optimizing porous structure parameters of foamed materials based on adversarial networks, the generator and discriminator networks are closely linked together through a loss function to achieve optimization of the porous structure parameters. Specifically, the loss functions of the generator and the discriminator define their training objectives and evaluation criteria. The goal of the generator is to continuously improve the quality of the generated porous structure parameters by minimizing its loss function, and the goal of the discriminator is to maximize its loss function, thereby improving its ability to distinguish true and false data.

[0037] Specifically, the generator continuously approaches the real porous structure by minimizing the structural difference part in the loss function, ensuring that the optimization results meet the actual physical and performance requirements. At the same time, through the evaluation results of the discriminator, the generator can further reduce the probability of generating false data, thereby ensuring the authenticity and availability of the parameters. At the same time, by optimizing the loss function of the discriminator, the discriminator can gradually improve its ability to identify the difference between generated data and real data. This not only enhances the training effect of the adversarial network, but also further promotes the improvement of the generator in data quality and parameter optimization.

[0038] In an embodiment of the present invention, the generator network includes 5 fully connected layers, and the number of neurons in each fully connected layer is 128, 256, 512, 256, and 128 respectively; the discriminator network includes 5 convolutional layers, and the convolution kernel sizes of each convolutional layer are 3x3, 5x5, 7x7, 5x5, and 3x3 respectively, and the number of convolution kernels is 32, 64, 128, 64, and 32 respectively; each layer of the generator network uses the ReLU activation function, and the output layer uses the tanh activation function; each layer of the discriminator network uses the LeakyReLU activation function, and the output layer uses the sigmoid activation function; the generator uses the Adam optimizer with a learning rate of 0.0002 and a momentum parameter of 0.5; the discriminator uses the RMSProp optimizer with a learning rate of 0.0001.

[0039] In one possible implementation, the generator network adopts a 5-layer fully connected layer structure, and the number of neurons between layers changes layer by layer, specifically 128, 256, 512, 256, and 128. Such a structural design can gradually expand the representation capacity of the network, and use the tanh activation function in the final output layer so that the generated output range conforms to the predetermined structural parameter range (such as porosity, density, etc.). The generator transmits signals within each layer through the ReLU activation function to ensure that the network can capture complex porous structure data through nonlinear transformations. The generator optimizes the neural network layer by layer, continuously reducing the difference between the generated data and the real data, thereby accurately generating the porous structure parameters of the foam material that meet the actual physical requirements. By using the tanh activation function as the output layer of the generator, the output parameters are ensured to be within a reasonable range, which is crucial for the generation of complex porous structures.

[0040] Furthermore, the discriminator uses 5 convolution layers, and the size and number of convolution kernels of each convolution layer are 3x3, 5x5, 7x7, 5x5, 3x3, and the number of convolution kernels is 32, 64, 128, 64, 32. Through this convolution structure, the discriminator can effectively extract the spatial features of the input data, so as to judge whether the input data comes from the real foaming material parameters or the false data generated by the generator. Each convolution layer uses the LeakyReLU activation function to solve the "dead neuron" problem that may be caused by the traditional ReLU activation function, so as to maintain the expression ability of the network. The output layer uses the sigmoid activation function to output a probability value between 0 and 1, indicating the probability that the input data is real data. The discriminator extracts spatial features in the data through the convolutional neural network, and can effectively identify the subtle differences between the generated data and the real data. The different sizes and numbers of convolution kernels enable the discriminator to extract features at different scales, thereby improving the accuracy and robustness of the discrimination. At the same time, the use of the LeakyReLU activation function helps avoid the "dead neuron" phenomenon and ensures that each layer of the discriminator can learn effectively.

[0041] Furthermore, the generator uses the Adam optimizer with a learning rate set to 0.0002 and a momentum parameter of 0.5. The Adam optimizer has good convergence when dealing with nonlinear optimization problems, and can help the generator quickly and effectively find the optimal porous structure parameters. The discriminator uses the RMSProp optimizer with a learning rate set to 0.0001. The RMSProp optimizer performs well in dealing with non-stationary objective functions, and can prevent the discriminator from excessive oscillation during training, thereby improving its ability to distinguish between true and false data. The Adam optimizer effectively copes with fluctuations in gradient descent and improves the efficiency of generator training by adaptively adjusting the learning rate; the RMSProp optimizer avoids oscillations in the discriminator during training by adjusting the root mean square of the gradient, making the adversarial training process more stable.

[0042] In the embodiment of the present invention, in step S4, the GUI, the adversarial network model and the CAD software are integrated into one system, and the specific steps are as follows: S4.1: Integration of GUI and adversarial network model: Use Python's Flask framework to build a web server, connect the user interface with the adversarial network model, and the user inputs the initial parameters of the foaming material through the web interface. The system passes the parameters to the generator to generate the optimized porous structure parameters. S4.2: Integration with CAD software: Use the API interface to pass the optimized porous structure parameters to the CAD software to generate a three-dimensional model of the porous structure. Specifically, use the COM API interface of AutoCAD to call the function of AutoCAD through a Python script to convert the optimized porous structure parameters into a three-dimensional model; S4.3: Unit testing: Perform unit testing on each component of the generator and discriminator to ensure that the function of each component is correct. The input and output functions of the generator are tested to ensure that the generator can correctly receive the initial parameters input by the user and generate the optimized porous structure parameters. S4.4: Integration test: Integrate the generator, discriminator and CAD software into a system and conduct integration test to ensure that the data transfer and functional collaboration between the components are correct, and test whether the porous structure parameters generated by the generator can be correctly transferred to the CAD software and generate a three-dimensional model; S4.5: System test: Perform system test on the whole system to ensure that the system is stable and reliable in actual use. Test whether the system can correctly generate the optimized porous structure parameters and display them in the result display module after the user inputs the parameters through the web interface.

[0043] In a possible implementation, a web server is built using Python's Flask framework to form a front-end interface for user interaction. The GUI provides a concise and intuitive operation method, through which the user inputs the initial parameters of the foaming material. These parameters include the physical properties of the porous structure such as porosity and density. The Flask framework passes the user's input to the adversarial network model at the back end through an HTTP request. The generator accepts these initial parameters and generates optimized porous structure parameters based on the rules learned during the training process.

[0044] Furthermore, the optimized porous structure parameters are passed to the CAD software and connected to the COM API interface of AutoCAD through the API interface. Using Python scripts, the optimized parameters generated by the generator can be converted into a format acceptable to CAD, thereby generating a three-dimensional model of the porous structure.

[0045] Furthermore, unit tests are performed on each component of the generator and discriminator to ensure that each module functions correctly. In particular, the input and output of the generator are verified to ensure that it can correctly receive the initial parameters input by the user and generate the optimized porous structure parameters.

[0046] Furthermore, the generator, discriminator and CAD software are integrated and tested to test whether the data transmission and collaboration between the components are correct. In particular, it is verified whether the porous structure parameters generated by the generator can be successfully transferred to the CAD software and the three-dimensional model can be correctly generated.

[0047] Furthermore, in the system test, the stability and reliability of the entire system are verified. After the user inputs the initial parameters through the web interface, can the system correctly generate the optimized porous structure parameters and display the results through the display module? During the test, we not only focus on whether the optimization function is stable, but also on the performance of the system in actual applications.

[0048] The integration of GUI, adversarial network model and CAD software makes data flow and functional collaboration closer, greatly improving the work efficiency of the entire optimization process. The optimization results of the generator are seamlessly passed to the CAD software through the API interface, realizing the automation and visualization of the design. At the same time, through the integrated system, the generator can accurately generate porous structure parameters that conform to physical laws based on the optimization results of adversarial training. Through GUI input, users can flexibly adjust the design parameters, making this method highly adjustable and adaptable in practical applications.

[0049] In an embodiment of the present invention, in S4.4 and S4.5, detailed log information is recorded, including the input parameters of each test case, the generated porous structure parameters, the generated three-dimensional model and the test results, and the log information is stored in a database to facilitate subsequent analysis and debugging.

[0050] In a possible implementation, in steps S4.4 and S4.5, detailed log information is recorded to ensure the traceability and debuggability of the entire system. In these two steps, when the system performs integration testing and system testing, it not only verifies the functions and collaboration of each module, but also records the detailed information of each test case.

[0051] Specifically, whenever the user inputs initial parameters through the GUI, the input information will be recorded as a log entry. The input parameters will be passed to the adversarial network model as the basis for training and optimization. Furthermore, the adversarial network model generates optimized porous structure parameters based on the input parameters, and these results will also be recorded in the log. The log will record each optimized parameter generated and ensure that the effect of each optimization can be traced during subsequent tests. Furthermore, the process of converting the optimized porous structure parameters into a three-dimensional model through CAD software will also be recorded. The generation process of each three-dimensional model and its results will become part of the log for subsequent verification and analysis. Furthermore, each test (including integration testing and system testing) will generate a test result, and success or failure will be recorded. All log information is stored in the database through the API interface to ensure the integrity and traceability of the test data.

[0052] Through detailed logging, the inputs, generated results, models, and test outputs of each test process can be accurately tracked, making it easier to find and resolve possible errors.

[0053] In an embodiment of the present invention, in S1, the template method is to control the pore structure of the foaming material through the template, and its key parameters include the pore size of the template, the arrangement of the template, the density of the template, the type and amount of the foaming agent; the non-solvent induced phase separation method is to use a non-solvent to cause phase separation to form a porous structure, and the key parameters include: the choice of solvent, the choice of non-solvent, the concentration of the solution, and the cooling rate; the freeze-drying method is to solidify the liquid by freezing, and then sublimate and dehydrate in a vacuum to form a porous structure, and the key parameters include: freezing temperature, freezing time, sublimation temperature, and sublimation time; and the supercritical CO2 foaming method is to use supercritical CO2 as a foaming agent to control the foaming process by changes in pressure and temperature, and the key parameters include: CO2 pressure, temperature, foaming time, type and amount of foaming agent.

[0054] In one possible embodiment, the porous structure of the foamed material is optimized by a plurality of different preparation methods, such as template method, non-solvent induced phase separation method, freeze drying method and supercritical CO2 foaming method. Each preparation method has its own specific key parameters, which together affect the final porous structure performance.

[0055] Specifically, the template method controls the pore structure of the foamed material by prefabricating the template. The key parameters of the template, such as the pore size, arrangement, and density, directly determine the size, shape, and distribution of the pores in the material. The template method is suitable for controlling the geometric morphology of the pore structure and is suitable for applications that require high-precision pore arrangement. The key parameters of the template method include the template pore size, arrangement, template density, and the type and amount of foaming agent. These parameters determine the porosity and pore size distribution of the final foamed material, thereby affecting its physical properties, such as mechanical strength, air permeability, and thermal conductivity.

[0056] Furthermore, the non-solvent induced phase separation method uses a non-solvent to induce the separation of different phases in the solution to form a porous structure. The key parameters include solvent selection, non-solvent selection, solution concentration, and cooling rate. The compatibility of the solvent and the non-solvent determines the efficiency of phase separation, while the solution concentration and cooling rate affect the size and uniformity of the pores. This method is suitable for generating materials with high specific surface area and controllable pore structure, and is widely used in fields that require high surface area and good mechanical properties, such as biomedical materials and thermal insulation materials.

[0057] Furthermore, freeze drying solidifies the liquid by freezing, and dehydrates it by sublimation in a vacuum, eventually forming a porous structure. The key parameters include freezing temperature, freezing time, sublimation temperature, and sublimation time. The freezing temperature and freezing time control the pore formation process, while the sublimation temperature and sublimation time determine the size and distribution of the pores. This method can produce materials with high porosity and good mechanical properties, and is particularly suitable for preparing large-pore, porous, and lightweight materials.

[0058] Furthermore, the supercritical CO2 foaming method uses supercritical CO2 as a foaming agent to control the foaming process under high pressure and temperature. Key parameters include CO2 pressure, temperature, foaming time, type and amount of foaming agent. Supercritical CO2 can dissolve in polymers under high pressure and expand rapidly when the pressure is released to form a uniform pore structure. This method is suitable for preparing lightweight foamed materials with good mechanical properties, and because CO2 has a low environmental impact, it is suitable for the preparation of green and sustainable materials.

[0059] Different methods optimize the shape, distribution and size of pores by controlling key parameters, thereby improving the mechanical properties, air permeability, thermal conductivity and other properties of the material. For example, the template method can accurately control the pore size and pore arrangement, while the non-solvent induced phase separation method can regulate the uniformity of pores at the molecular level. Since each method involves the adjustment of a series of key parameters, the pore structure of porous materials can be customized according to the needs of different applications. This high degree of customizability makes these foaming materials widely used in different fields, such as thermal insulation materials, structural materials, filter materials, sound-absorbing materials, etc. At the same time, by optimizing the combination of different methods, the production efficiency can be improved and the cost can be reduced without sacrificing performance. For example, the supercritical CO2 foaming method can greatly reduce the need for the use of harmful chemical solvents while being environmentally friendly, thereby reducing environmental pollution during the production process. In particular, the application of the supercritical CO2 foaming method can reduce the emission of harmful gases, which is in line with the trend of green manufacturing. In addition, other methods such as freeze-drying and non-solvent induced phase separation do not involve toxic chemicals and meet the requirements of environmental protection.

[0060] See also Figure 2 Based on the same inventive concept, an embodiment of the present invention provides a foam material porous structure parameter optimization system based on an adversarial network, including: a user interface module: a Web server is built using the Flask framework of Python, and an intuitive user interface is designed to allow the user to input the initial parameters of the foam material and view the optimization results. The module interacts with the adversarial network model through HTTP requests, and the parameters input by the user are passed to the adversarial network model module in JSON format; an adversarial network model module: including a generator network and a discriminator network, using a multi-layer perceptron (MLP) and a convolutional neural network (CNN) structure; a CAD software module: AutoCAD or Solidworks is selected as a simulation and visualization tool for the porous structure, and the optimized porous structure parameters are passed to the CAD software using an API interface to generate a three-dimensional model of the porous structure, specifically using AutoCAD's COM API interface, calls AutoCAD functions through Python scripts, and converts the optimized porous structure parameters into a three-dimensional model; data collection module: collects parameter data of foaming materials, including pore structure, pore size and density, through sensors and other data sources; data storage module: stores the collected data in the database for training and testing of adversarial network models, and the database uses MySQL or SQLite; test module: includes unit testing, integration testing and system testing. The unit testing module tests each component of the generator and discriminator to ensure that each component functions correctly. The integration testing module integrates the generator, discriminator and CAD software into a system for integration testing to ensure that the data transfer and functional collaboration between components are correct.

[0061] In a possible embodiment, the user interface module is the front-end part of the system, and a web server is built using the Flask framework of Python, with an intuitive and easy-to-use interface. The user inputs the initial parameters of the foaming material through this interface, such as the porosity, pore size, density and other information of the material. The interface design is concise and intuitive, allowing users to easily input data and view optimization results. This module interacts with the adversarial network model module of the back end through HTTP requests, and the parameters input by the user are passed to the adversarial network model module in JSON format. In this process, the user can not only input the initial parameters, but also view the optimized parameters and the corresponding porous structure three-dimensional model. Further, the joint action of the generator and the discriminator enables the system to optimize the pore structure that meets the design requirements in the continuous training process, thereby achieving the goal of optimal performance. Through adversarial training, the generator gradually improves the ability to generate porous structure parameters, and the discriminator continuously improves the accuracy of judgment of these parameters.

[0062] Furthermore, the CAD software module realizes the visualization of the optimized porous structure parameters. The system supports AutoCAD or Solidworks as a 3D modeling tool, and passes the optimized porous structure parameters to the CAD software through the API interface. Specifically, the COM API interface of AutoCAD is used to interact with AutoCAD through Python scripts to convert the porous structure parameters optimized by the adversarial network into a 3D model. This module ensures that the optimization results can not only be accurately calculated, but also displayed in a graphical way, which is convenient for engineers to further analyze and evaluate the materials.

[0063] Furthermore, the data collection module is responsible for collecting the actual parameter data of the foaming material in real time, including pore structure, pore size, density, etc. Through sensors and other data sources, the module can continuously obtain the key performance indicators of the foaming material during the experiment or production process. These data will provide feedback to the system to further improve the adversarial network model and ensure that the optimization results are more in line with actual application needs.

[0064] Furthermore, in order to conduct effective training and testing, the data storage module stores all the collected data in the database. The database uses MySQL or SQLite to ensure efficient access and management of data. Through the data storage module, rich sample data can be provided for the training of the adversarial network model, and it is also convenient for the model to be traced back and updated.

[0065] Furthermore, the testing module ensures that each part of the system works as expected. The testing module is divided into unit testing, integration testing and system testing: Unit testing is used to independently test each component of the generator and discriminator to ensure that each component functions correctly. For example, it checks whether the generator can generate a reasonable pore structure based on the input parameters, and whether the discriminator can correctly evaluate the generated porous structure parameters.

[0066] Integration testing is used to combine the generator, discriminator and CAD software modules together for integration testing to ensure that they can work together correctly. Integration testing verifies the correctness of data transfer and functional collaboration between different modules.

[0067] System testing is used to comprehensively test the entire system to ensure that the system can operate stably under different input conditions and generate optimization results that meet actual application requirements.

[0068] See also Figure 3 In an embodiment of the present invention, a real-time monitoring module is also included, which collects the porous structure parameters generated by sensors in real time and dynamically adjusts the training parameters of the generator and the discriminator; and the user interface module has a user authority management function and a multi-language support function. The user authority management function distinguishes the operation permissions of different users, and the multi-language support function supports multi-language interfaces including Chinese and English.

[0069] In a possible implementation, the user authority management function and multi-language support function of the real-time monitoring module and the user interface module enhance the flexibility, operability and international adaptability of the system. Specifically, the real-time monitoring module collects and generates porous structure parameters (such as porosity, pore size, density, etc.) through sensors, and passes these data to the adversarial network model module. The function of this module is to dynamically adjust the training parameters of the generator and the discriminator during the operation of the system to adapt to changes in actual production or experimental conditions, thereby continuously optimizing the generation results. The real-time data collected by the sensor is sent to the data storage module for subsequent analysis and model updating. At the same time, the module works closely with the adversarial network model module to provide real-time feedback on the actual parameter changes of the current porous structure, ensuring that the training of the generator and the discriminator can respond to changes in the external environment or experimental data in a timely manner. This dynamic adjustment ensures that the optimization results are more in line with actual needs.

[0070] Furthermore, the user rights management function allows system administrators to control the rights of different users, such as restricting certain users to only view the optimization results but not to modify the generated porous structure parameters, or restricting certain users to only access certain functional modules. Administrators can set different levels of access rights according to their needs. The rights management function is directly associated with the user interface module to ensure that only authorized users can perform certain operations (such as modifying parameters, starting optimization algorithms, accessing sensitive data, etc.), thereby protecting the security and integrity of system data. Administrators control access to core functions such as real-time monitoring and model training through the rights management function to ensure the compliance and operability of the system.

[0071] Furthermore, the multi-language support function supports different languages ​​(such as Chinese, English, etc.) through interface switching, so that international users can operate the system according to their own language habits, improving the user experience. The language selection of the user interface is implemented by the front-end interface module, and users can freely choose the language to operate on the interface. At the same time, the multi-language support function is closely integrated with the user interface module to ensure that the system can dynamically display the interface content according to the language selected by the user and pass the corresponding operation instructions to the back end. The feedback information, optimization results and parameters of the generated model within the system can be adjusted and displayed according to the language selection, so that the system has good global adaptability.

[0072] In an embodiment of the present invention, a cloud storage module is also included. The cloud storage module stores the generated three-dimensional model and test results in the cloud, and supports users to view and download data on different devices.

[0073] In one possible implementation, the cloud storage module is responsible for storing the generated 3D model and test results in the cloud to ensure data security, accessibility, and sharing. Users can view and download data anytime, anywhere on different devices through a network connection. This module supports data interaction with other modules of the system (such as the optimization module, user interface module, real-time monitoring module, etc.). When the optimization algorithm generates a new 3D model or conducts a new experimental test, the cloud storage module automatically uploads the results to the cloud for archiving.

[0074] The cloud storage module is tightly integrated with other modules (such as the real-time monitoring module, model generation module, and user interface module). When the optimization process is completed, the generated 3D model and test data will be uploaded to the cloud storage, and users can view and manage them through the user interface module. The cloud storage module also synchronizes data with the system's database to ensure data consistency and accuracy.

[0075] The cloud storage module ensures that the 3D models and test results generated by the system can be stored remotely, avoiding the capacity limitations of local storage and the risk of data loss. Even if the local system fails, the cloud data can still be safely stored and restored at any time, improving the reliability of the data and the fault tolerance of the system.

[0076] The optional implementation modes of the embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation modes. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical scheme of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention.

[0077] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0078] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. In addition, various different implementation methods of the embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the ideas of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for optimizing porous structure parameters of foaming materials based on an adversarial network, characterized in that: The steps include: S1: Analyze the application fields and performance requirements of foaming materials, determine the optimization targets, and study the existing optimization methods of porous structures of foaming materials, including template method, non-solvent induced phase separation method, freeze drying method and supercritical CO2 foaming method; S2: System design and architecture: design an intuitive user interface that allows users to input parameters and view optimization results, and design the network structure of the generator and discriminator to simulate and optimize the porous structure of the foaming material. Then select appropriate CAD software, including AutoCAD or Solidworks, for simulation and visualization of the porous structure. S3: Algorithm development and training, collect parameter data of foaming materials, including pore structure, pore size and density, use the collected data to train the adversarial network, identify and optimize the porous structure parameters of foaming materials; S4: System integration and testing: Integrate the GUI, adversarial network model and CAD software into a system, and perform unit testing, integration testing and system testing.

2. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 1, characterized in that: In S3, parameter data of the foaming material, including pore structure, pore size and density, are collected. The collected data is used to train the adversarial network, where the network structures of the generator and the discriminator are as follows: Generator network: It uses a multi-layer perceptron MLP structure. The input layer receives the user's initial parameters, and the output layer generates optimized porous structure parameters. Discriminator network: It adopts the convolutional neural network CNN structure. The input layer receives the porous structure parameters generated by the generator and the actual porous structure parameters, and the output layer evaluates the truth or falsity of the input parameters.

3. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 1, characterized in that: The loss function of the generator network is: ； in, are the porous structure parameters generated by the generator; is the actual porous structure parameter; is the evaluation result of the discriminator on the parameters of the generated porous structure; is a hyperparameter for adjusting weights; The loss function of the discriminator network is: ； in, is the input porous structure parameter, is the real label, 1 represents real data, 0 represents generated data, is the size of the dataset.

4. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 3, characterized in that: The generator network comprises 5 fully connected layers, and the number of neurons in each fully connected layer is 128, 256, 512, 256, and 128 respectively; the discriminator network comprises 5 convolutional layers, and the convolution kernel size of each convolutional layer is 3x3, 5x5, 7x7, 5x5, and 3x3 respectively, and the number of convolution kernels is 32, 64, 128, 64, and 32 respectively; Each layer of the generator network uses the ReLU activation function, the output layer uses the tanh activation function, each layer of the discriminator network uses the LeakyReLU activation function, and the output layer uses the sigmoid activation function; The generator uses the Adam optimizer with a learning rate of 0.0002 and a momentum parameter of 0.5, and the discriminator uses the RMSProp optimizer with a learning rate of 0.0001.

5. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 1, characterized in that: In step S4, the GUI, adversarial network model, and CAD software are integrated into a system. The specific steps are as follows: S4.1: Integration of GUI and adversarial network model: Use Python's Flask framework to build a web server, connect the user interface with the adversarial network model, and the user inputs the initial parameters of the foaming material through the web interface. The system passes the parameters to the generator to generate the optimized porous structure parameters. S4.2: Integration with CAD software: Use the API interface to pass the optimized porous structure parameters to the CAD software to generate a three-dimensional model of the porous structure. Specifically, use the COM API interface of AutoCAD to call the function of AutoCAD through a Python script to convert the optimized porous structure parameters into a three-dimensional model; S4.3: Unit testing: Perform unit testing on each component of the generator and the discriminator, testing the input and output functions of the generator so that the generator correctly receives the initial parameters input by the user and generates the optimized porous structure parameters; S4.4: Integration test: Integrate the generator, discriminator and CAD software into a system and conduct integration test to test whether the porous structure parameters generated by the generator can be correctly transferred to the CAD software and generate a three-dimensional model; S4.5: System test: Perform system test on the whole system to test whether the system can correctly generate optimized porous structure parameters and display them in the result display module after the user inputs parameters through the web interface.

6. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 5, characterized in that: In S4.4 and S4.5, detailed log information is recorded, including the input parameters of each test case, the generated porous structure parameters, the generated three-dimensional model, and the test results. The log information is stored in the database to facilitate subsequent analysis and debugging.

7. The method for optimizing porous structure parameters of foamed materials based on adversarial networks according to claim 1, characterized in that: In S1, the template method is to control the pore structure of the foaming material through the template, and its key parameters include the pore size of the template, the arrangement of the template, the density of the template, the type and amount of the foaming agent; The non-solvent induced phase separation method uses a non-solvent to induce phase separation to form a porous structure. The key parameters include: the choice of solvent, the choice of non-solvent, the concentration of the solution, and the cooling rate; Freeze drying is to solidify the liquid by freezing, and then sublimate and dehydrate in a vacuum to form a porous structure. The key parameters include: freezing temperature, freezing time, sublimation temperature, and sublimation time; The supercritical CO2 foaming method uses supercritical CO2 as a foaming agent and controls the foaming process by changes in pressure and temperature. The key parameters include: CO2 pressure, temperature, foaming time, type and amount of foaming agent.

8. A foam material porous structure parameter optimization system based on an adversarial network, characterized in that: include: User interface module: Use Python's Flask framework to build a web server and design an intuitive user interface to allow users to input the initial parameters of the foaming material and view the optimization results. This module interacts with the adversarial network model through HTTP requests, and the parameters entered by the user are passed to the adversarial network model module in JSON format; Adversarial network model module: including generator network and discriminator network, using multi-layer perceptron MLP and convolutional neural network CNN structure; CAD software module: Select AutoCAD or Solidworks as the simulation and visualization tool for the porous structure, use the API interface to pass the optimized porous structure parameters to the CAD software, and generate a three-dimensional model of the porous structure. Specifically, use the COM API interface of AutoCAD to call the function of AutoCAD through the Python script to convert the optimized porous structure parameters into a three-dimensional model; Data collection module: collects parameter data of foaming materials, including pore structure, pore size and density, through sensors and other data sources; Data storage module: stores the collected data in a database for training and testing adversarial network models. The database uses MySQL or SQLite. Test module: includes unit test, integration test and system test. The unit test module tests each component of the generator and discriminator. The integration test module integrates the generator, discriminator and CAD software into a system for integration test.

9. The foam material porous structure parameter optimization system based on adversarial network according to claim 8, characterized in that: It also includes a real-time monitoring module, which collects the generated porous structure parameters in real time through sensors and dynamically adjusts the training parameters of the generator and the discriminator; The user interface module has user authority management function and multi-language support function. The user authority management function distinguishes the operation authority of different users, and the multi-language support function supports multi-language interfaces including Chinese and English.

10. The foam material porous structure parameter optimization system based on adversarial network according to claim 9, characterized in that: It also includes a cloud storage module, which stores the generated 3D models and test results in the cloud and supports users to view and download data on different devices.

Images