Porous ceramic finite element model construction processing method and system for deep learning
By combining deep learning and finite element analysis to construct a finite element model of porous ceramics, the problem of balancing efficiency and accuracy in the preparation of porous ceramics has been solved, and efficient and accurate selection of design parameters and optimization of the performance of porous ceramics have been achieved.
Patent Information
- Application Number
- CN202511266360.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-05
- Publication Date
- 2025-12-16
AI Technical Summary
In existing porous ceramic preparation technologies, the optimization of design parameters is difficult to balance efficiency and precision, resulting in long research and development cycles, high costs, and material properties that do not meet diverse needs.
By combining deep learning and finite element analysis, a porous ceramic finite element model is constructed. The thermal conductivity is predicted by graph neural network and then finely screened using finite element analysis, thus achieving efficient and accurate selection of design parameters.
This significantly improves the efficiency and accuracy of thermal conductivity optimization in porous ceramics, provides a reliable porous ceramic preparation scheme, and ensures that material properties meet expectations.
Smart Images
Figure CN121148554A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of ceramic materials engineering, and in particular to a method and system for constructing and processing porous ceramic finite element models using deep learning. Background Technology
[0002] With the development of porous ceramic preparation technology, design parameter optimization has become a key foundation for improving material performance and production efficiency. Existing optimization methods rely solely on a single model for parameter screening, which either prolongs the R&D cycle and increases trial-and-error costs due to efficiency issues, or affects material performance due to insufficient accuracy. This makes it difficult to quickly adapt to diverse performance requirements and fails to meet the practical requirements of efficient and precise preparation of porous ceramics. Summary of the Invention
[0003] To address the aforementioned technical issues, this application provides a method and system for constructing and processing porous ceramic finite element models using deep learning. By integrating finite element and deep learning models, optimization efficiency is significantly improved while ensuring optimization accuracy, thus solving the problem of low efficiency or insufficient accuracy caused by traditional single models.
[0004] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide a method for constructing and processing a porous ceramic finite element model using deep learning, the method comprising: In the process of optimizing the design parameters of porous ceramics, a finite element analysis model is constructed to predict the thermal conductivity of porous ceramics. Based on the design parameter thresholds of porous ceramics, a sample dataset of thermal conductivity under different design parameters is obtained. A graph neural network is trained using the aforementioned thermal conductivity sample dataset to obtain a thermal conductivity prediction model; Based on the design parameter thresholds, with the aim of approximating the preset thermal conductivity, the thermal conductivity prediction model is used to coarsely screen the design parameters to obtain the initial design parameter space. Based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, the design parameters are finely screened using the finite element analysis model, and the optimal design parameters are output for the preparation of porous ceramics.
[0005] Secondly, embodiments of this application provide a deep learning-based finite element model construction and processing system for porous ceramics, the system comprising: The sample dataset construction module is used to build a finite element analysis model for predicting the thermal conductivity of porous ceramics during the optimization of design parameters. Based on the design parameter thresholds of porous ceramics, it analyzes and obtains sample datasets of thermal conductivity under different design parameters. The training prediction model module is used to train a graph neural network using the thermal conductivity sample dataset to obtain a thermal conductivity prediction model; The coarse screening parameter space module is used to coarsely screen the design parameters based on the design parameter thresholds, with the aim of approximating the preset thermal conductivity, and obtain the initial design parameter space using the thermal conductivity prediction model. The optimal parameter screening module is used to finely screen the design parameters based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, using the finite element analysis model, and output the optimal design parameters for porous ceramic preparation.
[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a deep learning-based method and system for constructing and processing finite element models of porous ceramics. By integrating finite element analysis with graph neural networks, it achieves efficient and accurate selection of design parameters in the optimization of thermal conductivity of porous ceramics. First, a finite element analysis model is constructed, and a thermal conductivity sample dataset is generated based on design parameter thresholds. Then, this dataset is used to train a graph neural network to obtain a thermal conductivity prediction model. Subsequently, this model is used to perform coarse selection of design parameters, and the selection range is dynamically adjusted based on its prediction accuracy to obtain an initial design parameter space. Finally, the finite element analysis model is used to precisely select within this space, outputting the optimal design parameters for porous ceramic preparation.
[0007] The technical solution of this application combines the high efficiency of graph neural networks with the high precision of finite element analysis. It adopts a multi-level screening and dynamic range adjustment strategy. First, graph neural networks are used to quickly and roughly screen parameters to narrow down the range. Then, finite element models are used for precise screening. This solves the problems of low efficiency of single finite element models and insufficient precision of single deep learning models. It achieves both efficiency and accuracy in thermal conductivity optimization and provides a reliable technical solution for the optimization of thermal conductivity of porous ceramics. Attached Figure Description
[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0009] Figure 1 A flowchart illustrating a deep learning-based method for constructing a finite element model of porous ceramics, provided in an embodiment of this application; Figure 2 This is a schematic diagram of a deep learning-based porous ceramic finite element model construction and processing system provided in an embodiment of this application.
[0010] The components represented by each number in the attached diagram are explained below: Sample dataset construction module 01, training prediction model module 02, coarse parameter space screening module 03, fine parameter screening module 04. Detailed Implementation
[0011] This application provides a deep learning-based method and system for constructing and processing porous ceramic finite element models. It addresses the technical problem that traditional parameter optimization methods in the prior art are difficult to balance efficiency and accuracy, resulting in difficulties in efficiently and accurately selecting optimal design parameters when optimizing the thermal conductivity of porous ceramics, and failing to meet the requirements for efficient preparation and performance optimization of porous ceramics.
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0013] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0014] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0015] Example 1, as shown in the appendix Figure 1 As shown, this application provides a method for constructing and processing a porous ceramic finite element model using deep learning. The method includes the following steps: S110: In the process of optimizing the design parameters of porous ceramics, a finite element analysis model is constructed to predict the thermal conductivity of porous ceramics. Based on the design parameter thresholds of porous ceramics, a sample dataset of thermal conductivity under different design parameters is obtained. In this embodiment of the application, in the scenario of optimizing the design parameters of porous ceramics, in order to accurately obtain thermal conductivity sample data through the finite element model to support subsequent model training, it is necessary to first clarify the range of design parameters and generate a sample dataset.
[0016] Specifically, the design parameter thresholds are first determined, covering material ratio, porosity, pore size, pore distribution and specific surface area. These parameters directly affect the thermal conductivity of porous ceramics.
[0017] Meanwhile, within the determined design parameter threshold, a preset number (≤100) of design parameters are randomly selected, and the thermal conductivity value corresponding to each parameter is calculated using a finite element analysis model.
[0018] Furthermore, the design parameters are correlated with the corresponding thermal conductivity values to construct a thermal conductivity sample dataset, providing basic data support for the subsequent training of the graph neural network and ensuring the effectiveness of model training.
[0019] Step S110 in the method provided in this application embodiment includes: Obtain the design parameter thresholds for porous ceramics, where the design parameters include material ratio, porosity, pore size, pore distribution, and specific surface area. Within the design parameter threshold, a preset number of design parameters are randomly selected. Through the finite element analysis model, a number of thermal conductivity values corresponding to the design parameters are calculated. A thermal conductivity sample dataset is constructed by combining the design parameters, wherein the preset number is less than or equal to 100.
[0020] In this embodiment of the application, in order to construct a high-quality thermal conductivity sample dataset to support the subsequent training of the graph neural network, it is necessary to define the range of design parameters and obtain the corresponding thermal conductivity values through a finite element analysis model to ensure the validity and representativeness of the sample data.
[0021] Specifically, the design parameter thresholds for porous ceramics are first obtained. The core design parameters include material ratio, porosity, pore size, pore distribution, and specific surface area. These parameters directly affect the thermal conductivity of porous ceramics from the perspectives of material composition and microstructure.
[0022] Among them, material ratio refers to the proportional relationship between ceramic raw materials and various additives. Different ratios will change the matrix composition of the material, thus affecting the thermal conductivity. Porosity is the percentage of pore volume to the total volume of the material. As a weak link in thermal conductivity, the proportion of pores directly affects the efficiency of heat transfer.
[0023] In addition, pore size, i.e., the average diameter of the pores, can cause changes in the heat transfer path near the pores due to differences in pore size; pore distribution reflects the spatial distribution of pores in the material, and uniform or aggregated distribution has a significant difference in its impact on overall heat conduction; specific surface area is the total surface area per unit mass of material, and a larger specific surface area increases the interface for heat transfer, thereby affecting thermal conductivity.
[0024] Furthermore, a preset number of design parameters are randomly selected within the aforementioned design parameter thresholds, and this preset number does not exceed 100.
[0025] Specifically, during the selection process, random sampling is used to ensure the diversity of parameter combinations, covering scenarios with different material ratios and pore characteristics, and avoiding the one-sidedness of sample data.
[0026] Furthermore, each selected design parameter is input into the finite element analysis model, and the thermal conduction process of the porous ceramic is simulated and calculated through the model to obtain the thermal conductivity value corresponding to each design parameter.
[0027] The finite element analysis model is based on the theory of heat conduction and can accurately simulate the heat transfer path and efficiency inside porous ceramics under different design parameters, ensuring the accuracy of the thermal conductivity value calculation.
[0028] Specifically, the finite element analysis model is built on the ANSYS thermal analysis framework. It uses three-dimensional solid elements to divide the geometric model of porous ceramics. The matrix thermal conductivity corresponding to the material ratio, the equivalent thermal resistance coefficient corresponding to the porosity, the contact thermal resistance parameter related to the pore size, the mesh density adjustment coefficient corresponding to the pore distribution, and the interface heat exchange coefficient corresponding to the specific surface area are used as input parameters to input the heat conduction control equation.
[0029] Specifically, the governing equation for heat conduction is: ; in, It is the gradient operator, where k is thermal conductivity, T is temperature, Q is internal heat source density, ρ is density, and C is... p It is specific heat capacity. It is the rate of temperature change over time.
[0030] First, based on the pore distribution and pore size in the design parameters, a geometric mesh of the porous structure is constructed in the model. A finer mesh is used for the pore region to improve the simulation accuracy (e.g., the mesh size is 1 / 5 of the pore size), while a conventional mesh is used for the matrix region to balance computational efficiency (e.g., the mesh size is 1 / 2 of the pore size).
[0031] Furthermore, the thermal physical properties of the matrix material are set based on the material ratio, and the equivalent thermal conductivity correction coefficient is calculated according to the porosity to adjust the thermal conductivity of the matrix.
[0032] The formula for calculating the equivalent thermal conductivity correction factor is "correction factor = 1 - porosity × 0.8", where porosity is substituted in as a decimal.
[0033] For example, if the material ratio is alumina:zirconia = 6:4, the thermal conductivity of alumina can be set to 30 W / (m·K) and the thermal conductivity of zirconia to 2 W / (m·K). The average thermal conductivity of the matrix can then be calculated based on the ratio. When the porosity is 0.2, the equivalent thermal conductivity correction factor is 1 - 0.2 × 0.8 = 0.84. This factor is used to adjust the average thermal conductivity of the matrix to obtain the corrected matrix thermal conductivity.
[0034] Furthermore, thermal boundary conditions are set. That is, one side of the model is set as a constant-temperature heat source (temperature T1=300K) and the other side as a constant-temperature cold source (temperature T2=290K). The sides are adiabatic (heat flux density Q=0), and the heat flux density q under steady-state heat conduction is calculated by the iterative solver of the finite element analysis software.
[0035] Finally, the thermal conductivity value corresponding to the design parameter was calculated according to Fourier's law, and the single simulation calculation was completed.
[0036] Specifically, the formula for calculating thermal conductivity based on Fourier's law is as follows: ; Where L is the model thickness, A is the heat transfer area, and ΔT = T1 - T2, the thermal conductivity value corresponding to the design parameter is calculated, and a single simulation calculation is completed.
[0037] For example, when the design parameters are: material ratio (alumina:zirconia = 6:4), porosity 0.2, pore size 50 μm, uniform pore distribution, and specific surface area 10 m², ... 2 At / g, the finite element model first divides the pore region into meshes (size 10μm) according to 1 / 5 of the pore size, and divides the matrix region into meshes (size 25μm) according to 1 / 2 of the pore size, with a total of about 8000 elements.
[0038] Meanwhile, the average thermal conductivity of the matrix was calculated based on the proportions (30×0.6+2×0.4=18.8W / (m·K)). Based on the porosity of 0.2, the correction coefficient was calculated as 1-0.2×0.8=0.84, resulting in a corrected thermal conductivity of 18.8×0.84≈15.79W / (m·K).
[0039] Subsequently, with T1=300K, T2=290K, and lateral insulation, the heat flux density q=25W was calculated using an iterative solver; finally, substituting into Fourier's law, L=0.01m and A=0.0001m... 2Given ΔT = 10K, the thermal conductivity is calculated to be (25 × 0.01) / (0.0001 × 10) = 2.5 W / (m·K).
[0040] Finally, the design parameters are associated with their corresponding thermal conductivity values to construct a thermal conductivity sample dataset. This dataset uses the design parameters as input features and the thermal conductivity values as output labels, providing fundamental data support for the subsequent training of the graph neural network and ensuring that the model can learn the mapping relationship between the design parameters and thermal conductivity.
[0041] S120: Train a graph neural network using the thermal conductivity sample dataset to obtain a thermal conductivity prediction model; In this embodiment of the application, after obtaining the thermal conductivity sample dataset, in order to construct a model that can efficiently predict the thermal conductivity of porous ceramics, it is necessary to learn the mapping relationship between design parameters and thermal conductivity through a graph neural network.
[0042] Specifically, the thermal conductivity sample dataset is first divided into a sample training set and a sample test set according to a preset ratio, so as to be used for model training and performance verification respectively.
[0043] Meanwhile, using design parameters as input and thermal conductivity as supervision, the graph neural network is trained using a sample training set, so that the model continuously adjusts the parameters during the learning process until it can better fit the relationship between input and output, thereby obtaining a thermal conductivity prediction model.
[0044] Furthermore, the accuracy of the trained thermal conductivity prediction model is tested using a sample test set. By comparing the model's prediction results with the actual thermal conductivity values, the model's prediction accuracy is obtained to evaluate the model's prediction performance.
[0045] This step uses a graph neural network to learn and train on thermal conductivity sample data, constructing a model that can quickly predict the thermal conductivity of porous ceramics. This not only leverages the high efficiency of deep learning models but also provides a reliable basis for the preliminary screening of subsequent design parameters.
[0046] Step S120 in the method provided in this application embodiment includes: The thermal conductivity sample dataset is divided into a sample training set and a sample test set according to a preset ratio; Using design parameters as input and thermal conductivity as supervision, a graph neural network is trained using the sample training set, and a thermal conductivity prediction model is obtained after the data training is completed. The accuracy of the thermal conductivity prediction model is tested using the sample test set to obtain the model prediction accuracy.
[0047] In this embodiment of the application, in order to build a model that can efficiently predict the thermal conductivity of porous ceramics through graph neural networks, it is necessary to use a thermal conductivity sample dataset for training and testing to ensure that the model can accurately learn the mapping relationship between design parameters and thermal conductivity.
[0048] Specifically, the thermal conductivity sample dataset is first divided into a training set and a test set according to a preset ratio. The training set is used to learn the model parameters, and the test set is used to evaluate the model's generalization ability and prediction accuracy.
[0049] For example, a 7:3 ratio can be used, with 70% of the data serving as the training set for the model to learn potential patterns, and 30% of the data serving as the test set to verify the model's adaptability to new data.
[0050] Furthermore, the graph neural network is trained using design parameters (including material ratio, porosity, pore size, pore distribution, and specific surface area) as input features and the corresponding thermal conductivity value as a supervision signal.
[0051] In this process, the graph neural network extracts features and performs nonlinear mapping on the input design parameters, continuously adjusting the network weights and biases to make the predicted thermal conductivity value of the model output gradually approach the actual thermal conductivity value until the training process converges, thus obtaining a preliminary thermal conductivity prediction model.
[0052] During training, the mean squared error loss function is used to calculate the deviation between the predicted value and the actual value. The parameters are updated iteratively through the Adam optimizer. The initial learning rate is set to 0.001 and the batch size is 32. When the loss function value is lower than the preset threshold (such as 0.001) for 10 consecutive iterations, the training is considered to have converged, and a preliminary thermal conductivity prediction model is obtained.
[0053] For example, for a material ratio of alumina:zirconia = 6:4 and a porosity of 0.2, the model initially predicted a thermal conductivity of 2.3 W / (m·K). After multiple rounds of training, the predicted value was gradually adjusted to 2.5 W / (m·K), and the deviation from the actual value calculated by the finite element method was reduced to 0, which verified the accuracy of the thermal conductivity prediction model.
[0054] Finally, by using graph neural networks to extract features and perform nonlinear mapping of design parameters, combined with parameter optimization during the training process, we can achieve efficient prediction of the thermal conductivity of porous ceramics and obtain a thermal conductivity prediction model that reflects the correlation between design parameters and thermal conductivity.
[0055] S130: Based on the design parameter threshold, with the aim of approximating the preset thermal conductivity, the design parameters are coarsely screened using the thermal conductivity prediction model to obtain the initial design parameter space; In this embodiment of the application, after obtaining the thermal conductivity prediction model, in order to narrow the calculation range of the subsequent finite element analysis model and improve the efficiency of parameter optimization, it is necessary to use the model to perform a coarse screening of the design parameters in order to lock in the initial design parameter space that is more likely to approach the preset thermal conductivity.
[0056] Specifically, the design parameters are first enumerated within the design parameter threshold according to the preset parameter interval step size to generate an initial design parameter set covering the entire range, ensuring that no potential optimal parameters are missed.
[0057] Simultaneously, the initial design parameter set is input into the thermal conductivity prediction model, and the corresponding predicted thermal conductivity set is obtained through model analysis and calculation, thus quickly obtaining the thermal conductivity prediction results under various parameter combinations.
[0058] Furthermore, based on the preset thermal conductivity, the deviation between each predicted value and the preset value in the predicted thermal conductivity set is calculated one by one to obtain multiple thermal conductivity deviations, thereby measuring the degree of closeness between each design parameter and the target thermal conductivity.
[0059] Finally, based on these thermal conductivity deviations, the initial design parameter set is coarsely screened, and several better design parameters with smaller deviations are retained to construct the initial design parameter space, providing a targeted parameter range for the subsequent fine screening of the finite element analysis model.
[0060] This step, through efficient calculation and deviation screening of the thermal conductivity prediction model, reduces the number of parameters requiring fine calculation while ensuring coverage. It provides targeted parameter space support for the subsequent fine screening of the finite element model, ensuring the efficiency and accuracy of the parameter optimization process.
[0061] Step S130 in the method provided in this application embodiment includes: According to the preset parameter interval step size, the design parameters are enumerated within the design parameter threshold to obtain the initial design parameter set; Using the thermal conductivity prediction model, a predicted thermal conductivity set is obtained based on the initial design parameter set. Based on the preset thermal conductivity, the deviations of multiple predicted thermal conductivity values in the predicted thermal conductivity set are calculated, and multiple thermal conductivity deviations are output. Based on the multiple thermal conductivity deviations, the initial design parameter set is coarsely screened to obtain multiple better design parameters, and an initial design parameter space is constructed.
[0062] In this embodiment of the application, after obtaining the thermal conductivity prediction model, in order to achieve efficient screening of design parameters and focus on the parameter range that is more likely to approach the preset thermal conductivity, it is necessary to construct the initial design parameter space through systematic parameter enumeration, model prediction, deviation quantification and screening integration.
[0063] Specifically, firstly, design parameters such as material ratio, porosity, and pore size are enumerated within the design parameter threshold according to a preset parameter interval step size, forming an initial design parameter set. This step ensures comprehensive parameter coverage and provides sufficient basic data for subsequent screening.
[0064] The parameter interval step size can be flexibly set according to the characteristics of different design parameters. For example, the material ratio can be enumerated in 10% intervals (such as the ratio of alumina to zirconium oxide being adjusted sequentially from 1:9 to 9:1), the porosity can be enumerated from 0.1 to 0.5 in 0.05 intervals, and the pore size can be enumerated from 20μm to 100μm in 10μm intervals, etc.
[0065] Furthermore, the initial design parameter set is input into the thermal conductivity prediction model, which quickly analyzes the thermal conduction characteristics corresponding to each parameter combination and outputs a predicted thermal conductivity set, thus achieving a scientific evaluation of a large number of design parameters.
[0066] Furthermore, based on the preset thermal conductivity, multiple thermal conductivity deviations are obtained by calculating the absolute or relative deviation between each value in the predicted thermal conductivity set and the preset value, thereby quantifying the degree of matching between each design parameter and the target thermal conductivity.
[0067] For example, if the preset thermal conductivity is 2.5 W / (m·K), and the predicted thermal conductivity of a certain design parameter obtained by the thermal conductivity prediction model is 2.3 W / (m·K), then its absolute deviation is |2.3-2.5|=0.2 W / (m·K), and the relative deviation is (0.2 / 2.5)×100%=8%; the predicted thermal conductivity of another design parameter is 2.6 W / (m·K), and its absolute deviation is 0.1 W / (m·K), and the relative deviation is 4%. These deviation values can be used to intuitively determine that the latter matches the target thermal conductivity better.
[0068] Furthermore, based on these thermal conductivity deviations, the initial design parameter set is coarsely screened to obtain the corresponding optimal design parameters, thereby constructing the initial design parameter space.
[0069] The method provided in this application embodiment includes the step of "performing a coarse screening of the initial design parameter set based on the multiple thermal conductivity deviations to obtain multiple better design parameters" as follows: The parameter elimination ratio is configured according to the model prediction accuracy, wherein the parameter elimination ratio is the product of the ratio of the model prediction accuracy to the preset accuracy scalar and the initial parameter elimination ratio, and the initial parameter elimination ratio is 80%. Based on the multiple thermal conductivity deviations, according to the parameter elimination ratio, the initial design parameters with large thermal conductivity deviations are eliminated, resulting in multiple better design parameters.
[0070] In this embodiment of the application, in order to balance efficiency and reliability in the coarse screening stage, the screening rigor needs to be dynamically adjusted according to the performance of the thermal conductivity prediction model. By scientifically configuring the parameter elimination ratio, the better design parameters that are more likely to approach the preset thermal conductivity are accurately retained.
[0071] Specifically, the elimination ratio is first configured based on the model's prediction accuracy. The calculation of the elimination ratio is based on the model's prediction accuracy, allowing the screening intensity to be adaptively adjusted according to the model's precision.
[0072] The specific formula for calculating the parameter elimination ratio is "Parameter elimination ratio = Prediction accuracy of thermal conductivity prediction model / Preset accuracy scalar × Initial parameter elimination ratio". The initial parameter elimination ratio is 80%, and the preset accuracy scalar can be customized according to actual needs, for example, set to 85%.
[0073] For example, when the prediction accuracy of the thermal conductivity prediction model is low at 70%, the parameter elimination ratio decreases accordingly, i.e., 70% / 85%×80%≈65.88%, meaning fewer design parameters are eliminated, thereby expanding the screening range to avoid missing potential optimal parameters; when the prediction accuracy is high at 90%, the parameter elimination ratio increases accordingly, i.e., 90% / 85%×80%≈84.71%, thus eliminating more parameters with large deviations and achieving more accurate preliminary screening.
[0074] Furthermore, based on the obtained thermal conductivity deviations, the initial design parameter set is filtered according to a pre-configured parameter elimination ratio. Specifically, the thermal conductivity deviations are first sorted in descending order, and the number of parameters to be eliminated is determined based on the elimination ratio. For example, when the initial design parameter set contains 100 parameters and the elimination ratio is 80%, the 80 parameters with the largest deviations are eliminated, and the 20 parameters with smaller deviations are retained as the optimal design parameters.
[0075] For example, if the prediction accuracy of the thermal conductivity prediction model is 85%, and the preset accuracy scalar is 85%, then the parameter elimination ratio is 85% / 85%×80%=80%. If the initial design parameter set has 50 parameters, and the thermal conductivity deviation ranges from 0.1W / (m·K) to 2.0W / (m·K), then according to the elimination ratio of 80%, 40 parameters with large deviations need to be eliminated, and finally the 10 design parameters with the smallest deviations are retained as the better design parameters.
[0076] This initial design parameter set coarse screening step can dynamically adjust the screening range according to the model performance, and ensure the objectivity of the screening through quantitative deviation ranking, providing a high-quality parameter foundation for the subsequent construction of the initial design parameter space.
[0077] Furthermore, after obtaining several optimal design parameters, these parameters need to be integrated to construct an initial design parameter space. This space forms an ordered range of parameter combinations by clarifying the inherent relationships between the various optimal design parameters.
[0078] Specifically, for different design parameters such as material ratio, porosity, and pore size, their value distributions among the optimal design parameters are statistically analyzed to determine the effective range of each parameter.
[0079] For example, the proportion of alumina in the material composition is concentrated between 60% and 80%, the porosity is mainly distributed between 0.2 and 0.4, and the pore size is mostly in the range of 40μm to 60μm. These ranges together constitute the boundary of the initial design parameter space.
[0080] At the same time, considering the mutual influence between parameters, such as the synergistic effect of porosity and pore size on heat conduction, this correlation must be preserved when constructing the space to ensure that the combination of parameters in the space has practical physical meaning and to avoid contradictions or unreasonable combinations.
[0081] For example, for a material ratio of alumina:zirconia = 6:4 and a porosity of 0.2, in the optimal design parameters, the corresponding pore size is mostly between 40μm and 60μm, and the specific surface area is mainly distributed around 8m². 2 / g-12m 2 Within the range of / g, the pores are uniformly distributed. Therefore, in the initial design parameter space, the value ranges of these parameters will be interrelated, forming a cooperative subspace.
[0082] By constructing an initial design parameter space, the scope of subsequent fine screening is effectively narrowed, providing a targeted and reasonable parameter combination for the finite element analysis model, thereby further improving the overall optimization efficiency while ensuring optimization accuracy.
[0083] S140: Based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, the design parameters are finely screened using the finite element analysis model, and the optimal design parameters are output for the preparation of porous ceramics.
[0084] In this embodiment of the application, after obtaining the initial design parameter space, in order to further improve the accuracy of parameter optimization, it is necessary to use the finite element analysis model to perform fine calculation and screening of the better design parameters in order to determine the optimal design parameters that best meet the preset thermal conductivity requirements.
[0085] Specifically, the thermal conductivity of several optimal design parameters within the initial design parameter space is first calculated using a finite element analysis model.
[0086] Among them, the finite element analysis model, with its high-precision simulation capability, can meticulously simulate the heat conduction process inside porous ceramics, thereby obtaining the optimal thermal conductivity value corresponding to each optimal design parameter, providing reliable data for subsequent accurate selection.
[0087] Furthermore, using a preset thermal conductivity as a benchmark, the deviations between multiple optimal thermal conductivity values and the preset value are calculated, resulting in multiple secondary thermal conductivity deviations. These secondary thermal conductivity deviations clearly measure the actual closeness of each optimal design parameter to the target thermal conductivity.
[0088] Ultimately, the optimal design parameter corresponding to the minimum secondary thermal conductivity deviation was determined as the optimal design parameter. This parameter can approximate the preset thermal conductivity to the greatest extent and can be used to guide the preparation of porous ceramics.
[0089] This step, through high-precision calculations and secondary deviation screening using the finite element analysis model, achieves detailed optimization of design parameters within the parameter range locked by the coarse screening, providing a scientific and reliable parameter basis for the preparation of porous ceramics.
[0090] Step S140 in the method provided in this application embodiment includes: Using the finite element analysis model, multiple optimal thermal conductivity values of multiple optimal design parameters within the initial design parameter space are calculated; Based on the preset thermal conductivity, deviations are calculated for the multiple optimal thermal conductivity values to obtain multiple secondary thermal conductivity deviations, and the optimal design parameter corresponding to the smallest secondary thermal conductivity deviation is set as the optimal design parameter.
[0091] In this embodiment of the application, after obtaining the initial design parameter space, it is necessary to perform fine calculations and screening on the better design parameters to determine the optimal design parameters that can approximate the preset thermal conductivity to the greatest extent, so as to provide accurate guidance for the preparation of porous ceramics.
[0092] Specifically, the thermal conductivity of several optimal design parameters within the initial design parameter space is first calculated using a finite element analysis model.
[0093] Similarly, this finite element analysis model is based on the theory of heat conduction. By constructing a three-dimensional geometric model, dividing into fine meshes, and setting material properties and boundary conditions, it can accurately simulate the heat transfer process inside porous ceramics, thereby obtaining the optimal thermal conductivity value corresponding to each optimal design parameter.
[0094] The calculation process of the finite element analysis model needs to be combined with the detailed characteristics of specific design parameters. For example, for the optimal design parameters with a material ratio of alumina:zirconia = 6:4, porosity of 0.2 and pore size of 50μm, the finite element analysis model will divide the mesh into different densities according to the pore distribution characteristics, set the matrix thermal conductivity according to the material ratio, adjust the equivalent thermal resistance according to the porosity, and obtain the corresponding optimal thermal conductivity value by iteratively solving the heat conduction control equation.
[0095] Furthermore, using a preset thermal conductivity as a benchmark, the absolute or relative deviations between multiple optimal thermal conductivity values and the preset value are calculated, resulting in multiple secondary thermal conductivity deviations. These secondary thermal conductivity deviations, compared to the thermal conductivity deviations output by the thermal conductivity prediction model, can more accurately measure the actual closeness of each optimal design parameter to the target thermal conductivity.
[0096] For example, if the preset thermal conductivity is 2.5 W / (m·K), and the optimal thermal conductivity value calculated by the finite element analysis model for a certain optimal design parameter is 2.45 W / (m·K), then its second absolute deviation is |2.45-2.5|=0.05 W / (m·K); the optimal thermal conductivity value of another optimal design parameter is 2.52 W / (m·K), and its second absolute deviation is 0.02 W / (m·K). By comparison, it can be seen that the latter has a higher degree of matching with the preset thermal conductivity.
[0097] Finally, the optimal design parameter corresponding to the minimum secondary thermal conductivity deviation is set as the optimal design parameter. This optimal design parameter can meet the preset thermal conductivity requirements to the greatest extent and can be directly used to guide the preparation process of porous ceramics, ensuring that the prepared porous ceramics have the expected thermal conductivity performance.
[0098] For example, the initial design parameter space contains 10 optimal design parameters. After calculation by the finite element analysis model, the optimal thermal conductivity values obtained are 2.43 W / (m·K), 2.46 W / (m·K), 2.48 W / (m·K), 2.49 W / (m·K), 2.50 W / (m·K), 2.51 W / (m·K), 2.53 W / (m·K), 2.55 W / (m·K), 2.57 W / (m·K), and 2.60 W / (m·K), respectively, and the preset thermal conductivity is 2.5 W / (m·K).
[0099] After calculating the second absolute deviation, the values of the second absolute deviation are 0.07 W / (m·K), 0.04 W / (m·K), 0.02 W / (m·K), 0.01 W / (m·K), 0 W / (m·K), 0.01 W / (m·K), 0.03 W / (m·K), 0.05 W / (m·K), 0.07 W / (m·K), and 0.10 W / (m·K).
[0100] Among them, the optimal design parameters corresponding to a thermal conductivity value of 2.50 W / (m·K) are: alumina:zirconia = 6:4 material ratio, porosity 0.2, pore size 50 μm, uniform pore distribution, and specific surface area 10 m². 2 The second absolute deviation of ( / g) is the smallest, which is 0 W / (m·K). Therefore, this parameter is set as the optimal design parameter for the preparation of porous ceramics.
[0101] This step leverages the high-precision calculation capabilities of the finite element analysis model to precisely optimize design parameters within the range of parameters locked by coarse screening. This not only compensates for insufficient model prediction accuracy but also avoids the inefficiency of fine calculation of all parameters, effectively balancing optimization efficiency and accuracy, making the optimization of the thermal conductivity of porous ceramics more efficient and precise.
[0102] The method provided in this application embodiment further includes, after "obtaining the thermal conductivity prediction model": Multiple optimal design parameters and multiple optimal thermal conductivity values are used as incremental sample data and divided into incremental training set and incremental test set; The thermal conductivity prediction model is incrementally learned and tested using the incremental training set and incremental test set, respectively, to obtain the prediction accuracy of the optimized thermal conductivity prediction model and the quadratic model. Based on the prediction accuracy of the quadratic model, a quadratic parameter elimination ratio is configured. The optimized thermal conductivity prediction model and the quadratic parameter elimination ratio are used for coarse parameter screening in the subsequent optimization process. The thermal conductivity prediction model can be iteratively optimized using incremental data, and the parameter elimination ratio can be iteratively updated.
[0103] In this embodiment of the application, in order to continuously improve the performance of the thermal conductivity prediction model and enable it to play a more accurate and efficient role in the subsequent parameter optimization process, it is necessary to iteratively optimize the model through incremental learning and update the parameter elimination ratio accordingly to adapt to changes in model performance.
[0104] Specifically, the optimal design parameters obtained from the initial screening and their corresponding optimal thermal conductivity values are first used as incremental sample data.
[0105] These incremental sample data were calculated using a finite element analysis model, possessing high accuracy and relevance, and providing valuable supplementary information for model optimization.
[0106] Furthermore, the incremental sample data is divided into an incremental training set and an incremental test set according to a certain ratio, for example, an 8:2 ratio, with 80% of the data used for incremental learning of the model and 20% of the data used for performance testing of the optimized model.
[0107] Furthermore, the existing thermal conductivity prediction model is incrementally learned using an incremental training set. During the learning process, the existing thermal conductivity prediction model adjusts its network parameters based on the new sample data, thereby better capturing the mapping relationship between design parameters and thermal conductivity and improving the model's prediction accuracy.
[0108] Meanwhile, the incrementally learned thermal conductivity prediction model was tested using an incremental test set to evaluate its prediction performance on new data, obtain the prediction accuracy of the secondary model, and use this to measure the effect of the model optimization.
[0109] Furthermore, based on the prediction accuracy of the quadratic model, the elimination ratio of the quadratic parameters is configured.
[0110] Specifically, the calculation method for the secondary parameter elimination ratio is similar to that for the parameter elimination ratio. It also uses the model prediction accuracy as the core basis, and multiplies the ratio of the model prediction accuracy to the preset accuracy scalar by the initial parameter elimination ratio.
[0111] The resulting secondary parameter elimination ratio can be dynamically adjusted based on the performance of the optimized model. When the prediction accuracy of the secondary model is high, the secondary parameter elimination ratio is increased accordingly to more rigorously screen parameters; when the prediction accuracy of the secondary model is low, the secondary parameter elimination ratio is decreased to retain more potential excellent parameters.
[0112] For example, if the prediction accuracy of the quadratic model is 90%, the preset accuracy scalar is 85%, and the initial parameter elimination ratio is 80%, then the quadratic parameter elimination ratio is 90% / 85%×80%≈84.71%; if the prediction accuracy of the quadratic model is 75%, then the quadratic parameter elimination ratio is 75% / 85%×80%≈70.59%.
[0113] Furthermore, the optimized thermal conductivity prediction model and the secondary parameter elimination ratio are used for coarse parameter screening in the subsequent optimization process. This iterative optimization method continuously improves the performance of the thermal conductivity prediction model, and the parameter elimination ratio is dynamically updated accordingly, making the entire parameter optimization process more efficient and providing more reliable support for the optimization of design parameters for porous ceramics.
[0114] This process first uses a graph neural network to efficiently screen and narrow down the parameter range, and then leverages a finite element analysis model for high-precision calculation and optimization of the design parameters. It combines the advantages of graph neural networks to compensate for the insufficient prediction accuracy of traditional models with the efficiency gains from detailed calculations of all parameters through finite element analysis, effectively balancing optimization efficiency and accuracy, making the optimization of the thermal conductivity of porous ceramics more efficient and precise.
[0115] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes a deep learning-based finite element model construction method for porous ceramics. First, in the optimization of porous ceramic design parameters, a finite element analysis model is constructed. Based on design parameter thresholds such as material ratio and porosity, no more than 100 parameters are randomly selected, and their corresponding thermal conductivity values are calculated to obtain a sample dataset. This dataset is then proportionally divided into a training set and a test set. A graph neural network is trained to obtain a thermal conductivity prediction model, and its accuracy is tested. Subsequently, using this model, parameters are enumerated at preset intervals to generate an initial design parameter set. The predicted thermal conductivity and its deviation from the preset value are calculated. Combined with the model accuracy, a parameter elimination ratio is configured to coarsely screen and obtain optimal design parameters, thus constructing an initial design parameter space. Finally, the thermal conductivity of the parameters within this space is calculated using the finite element model, and the deviation is calculated again. The design parameter corresponding to the minimum second-order thermal conductivity deviation is used as the optimal design parameter for preparation. Simultaneously, the optimal design parameters and thermal conductivity are used as incremental data to iteratively optimize the model and parameter elimination ratio, improving the efficiency and accuracy of subsequent screening.
[0116] The method provided in this application, through the technical solution of "sample generation - model training - coarse screening and focusing - fine screening and optimization", significantly improves the efficiency of the overall optimization process while ensuring the optimization accuracy, making the thermal conductivity optimization of porous ceramic materials more efficient and accurate.
[0117] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of a deep learning-based porous ceramic finite element model construction and processing method provided in Embodiment 1, this application also provides a deep learning-based porous ceramic finite element model construction and processing system, specifically including: The sample dataset construction module 01 is used to construct a finite element analysis model for predicting the thermal conductivity of porous ceramics during the optimization of design parameters of porous ceramics. Based on the design parameter thresholds of porous ceramics, it analyzes and obtains sample datasets of thermal conductivity under different design parameters. Training prediction model module 02 is used to train a graph neural network using the thermal conductivity sample dataset to obtain a thermal conductivity prediction model; The coarse screening parameter space module 03 is used to coarsely screen the design parameters based on the design parameter thresholds, with the aim of approximating the preset thermal conductivity, and obtain the initial design parameter space by using the thermal conductivity prediction model. The optimal parameter screening module 04 is used to finely screen the design parameters based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, using the finite element analysis model, and output the optimal design parameters for porous ceramic preparation.
[0118] In one embodiment, the sample dataset construction module 01 is further configured to: Obtain the design parameter thresholds for porous ceramics, where the design parameters include material ratio, porosity, pore size, pore distribution, and specific surface area. Within the design parameter threshold, a preset number of design parameters are randomly selected. Through the finite element analysis model, a number of thermal conductivity values corresponding to the design parameters are calculated. A thermal conductivity sample dataset is constructed by combining the design parameters, wherein the preset number is less than or equal to 100.
[0119] In one embodiment, the training prediction model module 02 is further configured to: The thermal conductivity sample dataset is divided into a sample training set and a sample test set according to a preset ratio; Using design parameters as input and thermal conductivity as supervision, a graph neural network is trained using the sample training set, and a thermal conductivity prediction model is obtained after the data training is completed. The accuracy of the thermal conductivity prediction model is tested using the sample test set to obtain the model prediction accuracy.
[0120] In one embodiment, the coarse screening parameter space module 03 is also used for: According to the preset parameter interval step size, the design parameters are enumerated within the design parameter threshold to obtain the initial design parameter set; Using the thermal conductivity prediction model, a predicted thermal conductivity set is obtained based on the initial design parameter set. Based on the preset thermal conductivity, the deviations of multiple predicted thermal conductivity values in the predicted thermal conductivity set are calculated, and multiple thermal conductivity deviations are output. Based on the multiple thermal conductivity deviations, the initial design parameter set is coarsely screened to obtain multiple better design parameters, and an initial design parameter space is constructed.
[0121] In one embodiment, the optimal parameter screening module 04 is further configured to: Using the finite element analysis model, multiple optimal thermal conductivity values of multiple optimal design parameters within the initial design parameter space are calculated; Based on the preset thermal conductivity, deviations are calculated for the multiple optimal thermal conductivity values to obtain multiple secondary thermal conductivity deviations, and the optimal design parameter corresponding to the smallest secondary thermal conductivity deviation is set as the optimal design parameter.
[0122] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0123] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0124] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.
Claims
1. A method for constructing and processing a finite element model of porous ceramics using deep learning, characterized in that, The methods include: In the process of optimizing the design parameters of porous ceramics, a finite element analysis model is constructed to predict the thermal conductivity of porous ceramics. Based on the design parameter thresholds of porous ceramics, a sample dataset of thermal conductivity under different design parameters is obtained. A graph neural network is trained using the aforementioned thermal conductivity sample dataset to obtain a thermal conductivity prediction model; Based on the design parameter thresholds, with the aim of approximating the preset thermal conductivity, the thermal conductivity prediction model is used to coarsely screen the design parameters to obtain the initial design parameter space. Based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, the design parameters are finely screened using the finite element analysis model, and the optimal design parameters are output for the preparation of porous ceramics.
2. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 1, characterized in that, Based on the design parameter thresholds for porous ceramics, a sample dataset of thermal conductivity under different design parameters was obtained, including: Obtain the design parameter thresholds for porous ceramics, where the design parameters include material ratio, porosity, pore size, pore distribution, and specific surface area. Within the design parameter threshold, a preset number of design parameters are randomly selected. Through the finite element analysis model, a number of thermal conductivity values corresponding to the design parameters are calculated. A thermal conductivity sample dataset is constructed by combining the design parameters, wherein the preset number is less than or equal to 100.
3. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 2, characterized in that, A graph neural network is trained using the aforementioned thermal conductivity sample dataset to obtain a thermal conductivity prediction model, including: The thermal conductivity sample dataset is divided into a sample training set and a sample test set according to a preset ratio; Using design parameters as input and thermal conductivity as supervision, a graph neural network is trained using the sample training set, and a thermal conductivity prediction model is obtained after the data training is completed. The accuracy of the thermal conductivity prediction model is tested using the sample test set to obtain the model prediction accuracy.
4. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 3, characterized in that, Based on the aforementioned design parameter thresholds, and with the aim of approximating a preset thermal conductivity, the thermal conductivity prediction model is used to coarsely screen the design parameters to obtain an initial design parameter space, including: According to the preset parameter interval step size, the design parameters are enumerated within the design parameter threshold to obtain the initial design parameter set; Using the thermal conductivity prediction model, a predicted thermal conductivity set is obtained based on the initial design parameter set. Based on the preset thermal conductivity, the deviations of multiple predicted thermal conductivity values in the predicted thermal conductivity set are calculated, and multiple thermal conductivity deviations are output. Based on the multiple thermal conductivity deviations, the initial design parameter set is coarsely screened to obtain multiple better design parameters, and an initial design parameter space is constructed.
5. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 4, characterized in that, Based on the aforementioned multiple thermal conductivity deviations, the initial design parameter set is coarsely screened to obtain several optimal design parameters, including: The parameter elimination ratio is configured according to the model prediction accuracy, wherein the parameter elimination ratio is the product of the ratio of the model prediction accuracy to the preset accuracy scalar and the initial parameter elimination ratio, and the initial parameter elimination ratio is 80%. Based on the multiple thermal conductivity deviations, according to the parameter elimination ratio, the initial design parameters with large thermal conductivity deviations are eliminated, resulting in multiple better design parameters.
6. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 1, characterized in that, Based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, the design parameters are finely screened using the finite element analysis model to output the optimal design parameters, including: Using the finite element analysis model, multiple optimal thermal conductivity values of multiple optimal design parameters within the initial design parameter space are calculated; Based on the preset thermal conductivity, deviations are calculated for the multiple optimal thermal conductivity values to obtain multiple secondary thermal conductivity deviations, and the optimal design parameter corresponding to the smallest secondary thermal conductivity deviation is set as the optimal design parameter.
7. The method for constructing and processing a porous ceramic finite element model using deep learning according to claim 1, characterized in that, The process of obtaining the thermal conductivity prediction model further includes: Multiple optimal design parameters and multiple optimal thermal conductivity values are used as incremental sample data and divided into incremental training set and incremental test set; The thermal conductivity prediction model is incrementally learned and tested using the incremental training set and incremental test set, respectively, to obtain the prediction accuracy of the optimized thermal conductivity prediction model and the quadratic model. Based on the prediction accuracy of the quadratic model, a quadratic parameter elimination ratio is configured. The optimized thermal conductivity prediction model and the quadratic parameter elimination ratio are used for coarse parameter screening in the subsequent optimization process. The thermal conductivity prediction model can be iteratively optimized using incremental data, and the parameter elimination ratio can be iteratively updated.
8. A deep learning-based finite element model construction and processing system for porous ceramics, characterized in that, The system is used to execute the deep learning-based porous ceramic finite element model construction processing method according to any one of claims 1-7, the system comprising: The sample dataset construction module is used to build a finite element analysis model for predicting the thermal conductivity of porous ceramics during the optimization of design parameters. Based on the design parameter thresholds of porous ceramics, it analyzes and obtains sample datasets of thermal conductivity under different design parameters. The training prediction model module is used to train a graph neural network using the thermal conductivity sample dataset to obtain a thermal conductivity prediction model; The coarse screening parameter space module is used to coarsely screen the design parameters based on the design parameter thresholds, with the aim of approximating the preset thermal conductivity, and obtain the initial design parameter space using the thermal conductivity prediction model. The optimal parameter screening module is used to finely screen the design parameters based on the initial design parameter space, with the aim of approximating the preset thermal conductivity, using the finite element analysis model, and output the optimal design parameters for porous ceramic preparation.