Neural network model-based auxiliary polyimide carbon fiber composite material structure design method

Abstract

The invention belongs to the field of polymer composite materials, and discloses a method for designing a polyimide carbon fiber composite material structure based on the assistance of a neural network model, which is characterized in that the polyimide carbon fiber composite material structure is designed based on the assistance of the neural network model, a compact functional structure is constructed, and a filler is used as an input feature of the neural network model. In combination with the target performance, the influence of material combination and thickness in a multi-layer or gradient structure on the performance is simulated through a neural network model, the influence of different filler types and content proportions on the comprehensive performance of the composite material is predicted, and design parameters are optimized in the structure design process; and an optimization scheme of intelligent regulation and control of process parameters and material preparation is obtained, so that the optimization of thermal performance and electromagnetic shielding performance is achieved while the excellent mechanical property of the material is maintained. A multi-layer or gradient structure design scheme can be generated, the comprehensive performance of the composite material is efficiently improved, the research and development time is shortened, and the method can be applied to large-scale intelligent production.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and advanced materials technology, specifically the deep integration of intelligent material design and composite materials engineering. This method, optimized by a neural network model, can accurately predict the optimal preparation conditions for composite materials and produce polyimide carbon fiber composites with integrated structure and function, promising broad application prospects in electronics, communications, military, and other fields. Background Art

[0002] Polyimide resin-based materials, as high-performance specialty engineering plastics, have attracted attention for their excellent mechanical properties, electrical insulation, and chemical stability. However, conventional polyimide resin-based composites lack multifunctionality. Therefore, the development of novel multifunctional polyimide resin-based carbon fiber composite films is of great significance, integrating the advantages of polyimide while overcoming its limitations.

[0003] In today's rapidly developing science and technology, electronic equipment and communication systems have become an indispensable part of daily life and industrial production. However, these systems are easily affected by external electromagnetic radiation and interference, resulting in abnormal equipment operation, even failures and data leakage. Therefore, the development of electromagnetic shielding materials with excellent mechanical properties, thermal stability and high efficiency has become an important research direction. Traditional electromagnetic shielding materials usually use metal or carbon-based materials. These materials have electromagnetic shielding properties to a certain extent, but have problems such as excessive density, easy oxidation or unstable conductivity. At the same time, harmful chemicals may be involved in the preparation process, affecting the environment. Therefore, the polyimide resin-based materials provided by the existing technology are difficult to have efficient electromagnetic shielding properties, stable mechanical properties and excellent electrical properties at the same time, and cannot meet the requirements of electronic equipment and communication systems for lightweight, processability and environmental protection of materials. Summary of the Invention

[0004] In response to the shortcomings of the above-mentioned background technology, the present invention provides a method for designing the structure of polyimide carbon fiber composite materials assisted by a neural network model, which aims to predict based on the neural network model, by constructing a dense functional structure, based on the excellent high thermal conductivity and high electrical conductivity of carbon fiber and the interaction of the electromagnetic shielding function enhancement layer, thereby simultaneously improving the mechanical strength, electromagnetic shielding performance and chemical stability of the composite material, and having efficient electromagnetic shielding performance, stable mechanical properties and excellent electrical properties, so as to meet the requirements of electronic equipment and communication systems for lightweight, processability and environmental protection of materials.

[0005] The present invention provides the following technical solutions: A method for designing a polyimide carbon fiber composite material structure assisted by a neural network model is characterized in that it is based on the neural network model to assist in designing a polyimide carbon fiber composite material structure, constructing a dense functional structure, taking the filler type and its mass fraction as input features of the neural network model, combining target performance including mechanical properties, thermal properties and electromagnetic shielding performance, simulating the influence of material combination and thickness on performance in a multilayer or gradient structure through the neural network model, predicting the influence of different filler types and content ratios on the comprehensive performance of the composite material, optimizing design parameters during the structural design process, obtaining intelligent control of process parameters and optimization schemes for material preparation, so that the material can achieve optimization of thermal properties and electromagnetic shielding performance while maintaining excellent mechanical properties, and specifically comprising the following steps: S1: Establish BP neural network model A back-propagation BP neural network model was established and pre-trained. This model was trained using a multi-layer feedforward neural network error back-propagation method. The network parameters were iteratively optimized to minimize the error between the network's actual output and the desired output. The input layer of the BP neural network model was used to input composite material preparation parameters, including temperature, time, component ratio, stirring speed, and magnetic particle content. The hidden layer was used to capture the nonlinear relationships between these parameters. The output layer provided the optimal combination of process parameters related to target performance, including thermal conductivity, mechanical strength, and electromagnetic shielding properties. S2: Predicting composite material preparation plans The target performance of the composite material designed for the direction is input into a pre-trained neural network model, and the model directionally predicts a preparation scheme for a polyimide resin-based electromagnetic shielding composite material, including directionally generating and outputting a predicted modified carbon fiber composite material scheme, a preparation scheme for an electromagnetic shielding performance enhancement layer, and a preparation scheme for a multifunctional layered polyimide resin-based electromagnetic shielding composite material; S3: Preparation of modified carbon fibers According to the directional prediction scheme, a modified carbon fiber composite material is prepared. The high thermal conductivity micron pitch-based carbon fiber is immersed in acetone and ultrasonically treated at a set temperature and time. After removing surface attachments, it is washed and dried; it is wet oxidized by a mixed solution of concentrated sulfuric acid and concentrated nitric acid to obtain a pretreated carbon fiber a; then, MXene and the pretreated carbon fiber are modified by monoamine oxidative polymerization, and mechanical stirring is carried out under nitrogen protection to obtain MXene-coated modified carbon fiber b; a surfactant, an organic ligand, a monoamine-modified micron pitch-based carbon fiber, and an organic solvent are mixed and added to a metal salt solution. The mixture is stirred for a set time and a hydrothermal reaction is carried out in a reactor. After the reaction, it is washed and centrifuged and dried. The obtained product is placed in a protective gas atmosphere and subjected to high-temperature treatment at a set temperature to obtain uniformly loaded nano-metal particles @MXene / micron pitch-based carbon fiber; Step S4: Preparing an electromagnetic shielding reinforcement layer Following the directional prediction scheme, an electromagnetic shielding enhancement layer was prepared. PBO microfibers were added to a mixture of trifluoroacetic acid and methanesulfonic acid and mechanically stirred to form a PBO nanofiber dispersion. MXene was ultrasonically dispersed at a set power and then mixed with the PBO nanofibers to form a stable dispersion. The PBO / MXene electromagnetic shielding enhancement layer was then prepared via a sol-gel conversion method.

[0006] Step S5: Preparing a multifunctional polyimide resin-based electromagnetic shielding composite material According to the scheme of directional prediction of the neural network model, a multifunctional layered polyimide resin-based electromagnetic shielding composite material is prepared, the modified carbon fiber obtained in step S3 and step S4 is mixed with the electromagnetic shielding reinforcement layer in proportion, the mixture is vacuum filtered and hot pressed, and then backfilled by polyimide resin impregnation and heat treated, finally preparing a polyimide resin-based electromagnetic shielding composite material with a multifunctional layered dense structure; Step S6: Composite material performance testing, feedback and optimization The actual performance of the prepared polyimide resin-based electromagnetic shielding composite material with multifunctional layers was tested to verify the accuracy of the prediction scheme, and then a multi-layer feedforward neural network model was used to intelligently optimize the process parameters in the prediction scheme.

[0007] Step S7: Multi-objective parameter collaborative optimization based on a neural network model is based on an improved non-dominated sorting genetic algorithm, with electromagnetic shielding effectiveness, tensile strength and thermal conductivity as optimization targets, and a multi-objective Pareto front search is performed on the process parameter combination output from step S2 to obtain the optimal parameter set that meets the preset performance weights; the parameter set includes: the fluctuation range of the magnetic particle content, the stirring rate gradient, and the hot pressing molding temperature-time coupling curve.

[0008] Compared with the prior art, the present invention has the following advantages and significant features: 1. In the structural optimization design of polyimide carbon fiber composite materials, the present invention constructs a neural network model that uses filler type and mass fraction as input features. Combined with target performance (such as mechanical properties, thermal properties, and electromagnetic shielding performance), the neural network model can effectively train and predict the impact of different filler types and contents on the overall material performance. Furthermore, in structural design, the neural network can also simulate the impact of material combinations and thicknesses on performance in multilayer or gradient structures. By optimizing design parameters, the material achieves optimal thermal and electromagnetic shielding properties while maintaining excellent mechanical properties. Combined with neural network-assisted optimization of the composite material preparation process, it is possible to explore and generate innovative multilayer or gradient structure design solutions, effectively improving the overall performance of the composite material, shortening R&D time, and achieving intelligent control of process parameters and optimization of material preparation, while also enabling large-scale intelligent production capabilities.

[0009] 2. This invention enhances the mechanical strength, electromagnetic shielding performance, and chemical stability of the composite material by constructing a dense functional structure, leveraging the excellent thermal and electrical conductivity of carbon fibers and the interaction between the electromagnetic shielding enhancement layer. Furthermore, by constructing a dense functional structure, leveraging the excellent electrical conductivity of modified carbon fibers and the interaction between the MXene electromagnetic shielding enhancement layer, the mechanical strength and electromagnetic shielding performance of the composite material are significantly enhanced. Furthermore, the production process requires only a small amount of experimentation, and with the support of neural network algorithm simulation, the optimal conditions for composite material preparation are identified. This intelligent process supports large-scale production, meets market demand for lightweight and high-performance materials, and provides a method for preparing carbon-carbon composite materials with excellent processability and environmental friendliness.

[0010] 3. The directional design and preparation method of the polyimide resin-based electromagnetic shielding composite material provided by the present invention lies in the intelligent modeling and multi-level structure design based on the back propagation neural network (BP neural network model), combined with experimental verification and parameter feedback optimization, to achieve accurate prediction of the composite material preparation process and coordinated performance regulation.

[0011] 4. The present invention utilizes treated high-thermal-conductivity micron pitch-based carbon fibers (CF) and then modifies the pretreated carbon fibers with MXene to produce a MXene / micron pitch-based carbon fiber uniformly loaded with nanometal particles. A PBO / MXene electromagnetic shielding enhancement layer is simultaneously prepared via a sol-gel conversion process. Finally, a polyimide resin-based carbon fiber electromagnetic shielding composite material is successfully prepared. This process utilizes a neural network model to intelligently optimize process parameters and assist in optimizing the experimental preparation process. The model identifies a specific formula for mixing the modified carbon fiber composite material and the electromagnetic shielding enhancement layer. This is followed by vacuum filtration and hot pressing. Finally, a polyimide resin-based composite film with high mechanical strength, excellent electromagnetic shielding performance, and high chemical stability is prepared by backfilling and heat treatment through impregnation with polyimide resin. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, other drawings obtained based on these drawings still fall within the scope of the present invention.

[0013] Figure 1 is a SEM image of the composite material prepared in Example 1 of the present invention; Figure 2 This is a flow chart of optimizing BP neural network using genetic algorithm according to an embodiment of the present invention; Figure 3 It is a flowchart of the neural network model simulation of an embodiment of the present invention. DETAILED DESCRIPTION

[0014] To further illustrate the technical means and efficacy of the present invention to achieve its intended purpose, the present invention has conducted basic experiments to explore the optimal preparation method of the composite material under neural network simulation conditions. The following is a detailed description of an embodiment of the series of experiments: Example 1

[0015] The present embodiment provides a method for designing a polyimide carbon fiber composite material structure assisted by a neural network model. The method is based on the neural network model to assist in designing a polyimide carbon fiber composite material structure, construct a dense functional structure, use the filler type and its mass fraction as input features of the neural network model, combine the target performance including mechanical properties, thermal properties and electromagnetic shielding performance, simulate the influence of material combination and thickness on performance in a multilayer or gradient structure through the neural network model, predict the influence of different filler types and content ratios on the comprehensive performance of the composite material, optimize the design parameters during the structural design process, obtain an intelligent control scheme for process parameters and an optimization scheme for material preparation, so that the material can achieve optimization of thermal properties and electromagnetic shielding performance while maintaining excellent mechanical properties. The method specifically includes the following steps: S1: Establish BP neural network model A back-propagation BP neural network model was established and pre-trained. This model was trained using a multi-layer feedforward neural network error back-propagation method. The network parameters were iteratively optimized to minimize the error between the network's actual output and the desired output. The input layer of the BP neural network model was used to input composite material preparation parameters, including temperature, time, component ratio, stirring speed, and magnetic particle content. The hidden layer was used to capture the nonlinear relationships between these parameters. The output layer provided the optimal combination of process parameters related to target performance, including thermal conductivity, mechanical strength, and electromagnetic shielding properties. The steps for establishing the BP neural network model are as follows: ① Parameter initialization: Select from the sample space Group training data and initialize the distribution weights of test data , determine the prediction error threshold Φ and the number of weak predictors Initialize the weights from the input layer to the hidden layer, the hidden layer threshold, the weights from the hidden layer to the output layer, and the output layer threshold of the BP neural network; ② Weak predictor prediction: When training the t-th weak predictor, the sample parameters (including temperature, time, component ratio, stirring rate, and magnetic particle content) are input in a feedforward manner, the network weights are optimized through the back propagation algorithm, and the output of the training set is predicted to obtain g t (xi) and the absolute value of the prediction error e t (i), the calculation formula is: i=1,2,…,m where y i Actual performance parameters, including tensile strength, thermal conductivity, and electromagnetic shielding performance; ③ Calculate the error rate :Filter high error samples based on the error threshold Φ. The calculation formula is: in is an indicator function, which takes 1 when the condition is met and 0 otherwise; ④ Calculate the weak predictor weight α t : Dynamically adjust the contribution weight of the weak predictor according to the error rate. The calculation formula is: ; ⑤ Adjust the sample weight of the t+1th iteration: Increase the weight of high-error samples to optimize the model sensitivity. The calculation formula is: Where B t is the normalization factor, ensuring that the sum of the weights is 1; ⑥ Construct a strong predictor: Combine n weak BP neural network predictors into a strong prediction function h(x) through ensemble learning. Its output is the weighted voting result. The calculation formula is: The final model iteratively optimizes the mapping relationship between process parameters and performance, outputting the optimal component ratio and process conditions, including temperature, time, component ratio, stirring speed, and magnetic particle content; The pre-training step is to use the parameters of temperature, time, component ratio, stirring speed and magnetic particle content in the composite material preparation process as input feature data, combined with target performance data, including mechanical properties, thermal properties, and electromagnetic shielding performance, and import them into the BP neural network model for training, so that the pre-trained neural network can effectively predict the influence of different filler types and contents on the comprehensive performance of the material, and obtain a pre-trained neural network model.

[0016] S2: Predicting composite material preparation plans The target performance of the composite material designed for the direction is input into a pre-trained neural network model, and the model directionally predicts a preparation scheme for a polyimide resin-based electromagnetic shielding composite material, including directionally generating and outputting a predicted modified carbon fiber composite material scheme, a preparation scheme for an electromagnetic shielding performance enhancement layer, and a preparation scheme for a multifunctional layered polyimide resin-based electromagnetic shielding composite material; S3: Preparation of modified carbon fibers During the preparation process of this step, a series of composite materials are prepared by pre-optimizing multiple key process parameters, including ultrasonic treatment time, reaction temperature, and stirring rate, through a neural network model. The neural network model is pre-trained by learning a large amount of experimental data. The model can predict the impact of these parameters on the properties of carbon fiber, including mechanical properties, thermal properties, and electromagnetic shielding properties. The optimal combination of preparation parameters is output based on the target performance requirements to ensure the effective loading of MXene on the carbon fiber surface and improve the overall performance of the material. According to the directional prediction scheme, a modified carbon fiber composite material is prepared. The high thermal conductivity micron pitch-based carbon fiber is immersed in acetone and ultrasonically treated at a set temperature and time. After removing surface attachments, it is washed and dried; it is wet oxidized by a mixed solution of concentrated sulfuric acid and concentrated nitric acid to obtain a pretreated carbon fiber a; then, MXene and the pretreated carbon fiber are modified by monoamine oxidative polymerization, and mechanical stirring is carried out under nitrogen protection to obtain MXene-coated modified carbon fiber b; a surfactant, an organic ligand, a monoamine-modified micron pitch-based carbon fiber, and an organic solvent are mixed and added to a metal salt solution. The mixture is stirred for a set time and a hydrothermal reaction is carried out in a reactor. After the reaction, it is washed and centrifuged and dried. The obtained product is placed in a protective gas atmosphere and subjected to high-temperature treatment at a set temperature to obtain uniformly loaded nano-metal particles @MXene / micron pitch-based carbon fiber; The carbon fiber is a high thermal conductivity asphalt-based carbon fiber with a thermal conductivity coefficient of ≥800 W / (m·K). The vacuum-filtered material is then hot-pressed at a temperature of 160-320°C, a pressure of 5-10 MPa, and a time of 1-3 hours.

[0017] The surfactant is one of cationic surfactant hexadecyltrimethylamine salt, anionic surfactant sulfonate, phosphate, and nonionic surfactant octylphenol polyoxyethylene ether; the monoamine monomer is one of dopamine, 5-hydroxytryptamine, aniline, ethylenediamine, and m-phenylenediamine.

[0018] The organic solvent is one of N,N-dimethylformamide (DMF), dimethyl sulfoxide (DMSO), and dimethylacetamide (DMAC); Step S4: Preparing an electromagnetic shielding reinforcement layer Following the directional prediction scheme, an electromagnetic shielding enhancement layer was prepared. PBO microfibers were added to a mixture of trifluoroacetic acid and methanesulfonic acid and mechanically stirred to form a PBO nanofiber dispersion. MXene was ultrasonically dispersed at a set power and then mixed with the PBO nanofibers to form a stable dispersion. The PBO / MXene electromagnetic shielding enhancement layer was then prepared via a sol-gel conversion method. The modified carbon fibers in steps S3 and S4 account for 40-60% by weight of the overall raw material formula, and the electromagnetic shielding performance enhancement layer accounts for 60-40% by weight of the overall raw material formula; the MXene dispersion process in steps S2 and S3 includes ultrasonic dispersion for 1-3 hours; Step S5: Preparing a multifunctional polyimide resin-based electromagnetic shielding composite material According to the scheme of directional prediction of the neural network model, a multifunctional layered polyimide resin-based electromagnetic shielding composite material is prepared, the modified carbon fiber obtained in step S3 and step S4 is mixed with the electromagnetic shielding reinforcement layer in proportion, the mixture is vacuum filtered and hot pressed, and then backfilled by polyimide resin impregnation and heat treated, finally preparing a polyimide resin-based electromagnetic shielding composite material with a multifunctional layered dense structure; Step S6: Composite material performance testing, feedback and optimization The actual performance of the prepared polyimide resin-based electromagnetic shielding composite material with a multifunctional layer was tested to verify the accuracy of the prediction scheme. Then, a multi-layer feedforward neural network model was used to intelligently optimize the process parameters in the prediction scheme. Specifically, a multi-layer feedforward neural network model based on the back-propagation algorithm is used to construct an intelligent production control model. The multi-layer feedforward neural network model is pre-trained with a large amount of historical experimental data, and the weights of the neurons are continuously adjusted to ensure high-precision process parameter prediction capabilities.

[0019] Step S7: Multi-objective parameter collaborative optimization based on a neural network model is based on an improved non-dominated sorting genetic algorithm, with electromagnetic shielding effectiveness, tensile strength and thermal conductivity as optimization targets, and a multi-objective Pareto front search is performed on the process parameter combination output from step S2 to obtain the optimal parameter set that meets the preset performance weights; the parameter set includes: the fluctuation range of the magnetic particle content, the stirring rate gradient, and the hot pressing molding temperature-time coupling curve.

[0020] Example 2 See attached Figure 2-3 The method for designing a polyimide carbon fiber composite material structure based on a neural network model provided in this embodiment is based on the above-mentioned embodiment 1 and further includes the following steps: 1. Process parameter prediction and optimization based on BP neural network 1) Neural network model construction and training A multi-layer feedforward BP neural network model was established. The input layer received the composite material preparation parameters (temperature, time, component ratio, stirring speed, magnetic particle content, etc.), the hidden layer captured the complex relationship between the parameters through a nonlinear activation function, and the output layer provided the optimal process parameter combination that matched the target performance (thermal conductivity, mechanical strength, and electromagnetic shielding performance).

[0021] 2) Using ensemble learning strategy to optimize the model: Parameter initialization: Select training data from the sample space, initialize network weights and thresholds, set the prediction error threshold Φ and the number of weak predictors.

[0022] Weak predictor training: By feeding forward the input sample parameters and backpropagating the optimized weights, the predicted output gt(xi) and the absolute value of the error et(i) are calculated (formula: et(i)=|y i -gt(x i )|, where y i are actual performance parameters such as tensile strength, thermal conductivity, and electromagnetic shielding performance).

[0023] Dynamic weight adjustment: filter high error samples based on error rate and dynamically assign weak predictor weights α t , and iteratively update the sample weights to enhance the model sensitivity, and finally construct a strong prediction function h(x) through weighted voting.

[0024] 3) Process parameter prediction and solution generation Input the target performance requirements (such as electromagnetic shielding effectiveness ≥ 25 dB, tensile strength ≥ 100 MPa), and the pre-trained model will output the modified carbon fiber preparation plan, electromagnetic shielding reinforcement layer design plan, and the component ratio and process conditions of the multifunctional layered composite material (such as temperature gradient, stirring rate, magnetic particle content, etc.).

[0025] 2. Oriented preparation of modified carbon fiber 1) Surface treatment and functional modification The high thermal conductivity micron pitch-based carbon fiber is immersed in acetone for ultrasonic cleaning to remove surface impurities and then dried; and a concentrated sulfuric acid / nitric acid mixture is used for wet oxidation treatment to obtain a pretreated carbon fiber a; MXene was coated on the surface of carbon fiber by monoamine oxidative polymerization, and mechanically stirred under nitrogen protection to form MXene-modified carbon fiber b; 2) Nanometal particle loading Surfactants, organic ligands, monoamine-modified carbon fibers and organic solvents are mixed, and a metal salt solution is added for a hydrothermal reaction. After centrifugal drying, the mixture is subjected to high-temperature heat treatment in a protective atmosphere to produce nano-metal particles @MXene / micron pitch-based carbon fibers, thereby improving interfacial bonding strength and conductivity.

[0026] Neural network-assisted optimization: The model predicts key parameters such as ultrasonic treatment time, hydrothermal temperature, and stirring rate to ensure uniform MXene loading and nanoparticle dispersion.

[0027] 3. Construction of electromagnetic shielding performance enhancement layer 1) PBO / MXene composite structure design PBO microfibers were dispersed in a trifluoroacetic acid / methanesulfonic acid mixture and mechanically stirred to obtain a PBO nanofiber dispersion; MXene was ultrasonically dispersed and mixed with the PBO nanofibers to form a stable suspension.

[0028] The PBO / MXene electromagnetic shielding reinforcement layer was prepared by the sol-gel conversion method, and the high conductivity of MXene and the dielectric properties of PBO fibers were used to synergistically improve the electromagnetic wave absorption efficiency.

[0029] 4. Integrated preparation of multifunctional hierarchical composite materials The modified carbon fibers from step S2 and the electromagnetic shielding reinforcement layer from step S3 are mixed in proportion and vacuum filtered to form a preform. After hot pressing, the preform is impregnated and backfilled with a polyimide resin. The resin matrix is ​​cured through a gradient heat treatment (200-350°C) to form a multifunctional hierarchical structure consisting of a conductive filler layer, an electromagnetic shielding layer, and a reinforcement framework, achieving synergistic enhancements in electromagnetic shielding, mechanical strength, and thermal conductivity.

[0030] 5. Performance Testing and Closed-Loop Optimization Performance verification and feedback mechanism The actual properties of composite materials such as electromagnetic shielding effectiveness (SE), tensile strength (TS), and thermal conductivity (k) were tested, and the model accuracy was verified by comparing the predicted results.

[0031] The neural network model training set is updated based on the test data, and the network weights are dynamically adjusted through the back-propagation algorithm to form a closed-loop optimization system and continuously improve the process parameter prediction capabilities.

[0032] In this process, the application of genetic algorithm is divided into the following key steps: In experimental design, the first step is to screen the experimental conditions and determine the required environment and materials. Next, the experimental parameters are adjusted by adjusting various variables, such as temperature, time, and material ratios, to determine the optimal experimental conditions. During this process, these parameters can be continuously optimized to define the parameters used in the experimental process. These parameters consist of a series of composite materials, each representing a possible structure for a BP neural network. Here, the experimental parameters are the structural parameters of the neural network, including the number of hidden layers and the number of nodes in each layer. This process continues until the optimal combination is found, and termination conditions are set for the experiment, such as when performance improvement reaches a certain standard or the number of experiments reaches a set value. Subsequently, the corresponding material structure is designed based on these optimized parameters, and performance evaluation is conducted to ensure that the material's performance meets the expected requirements, thereby driving further improvements to the subsequent experimental design.

[0033] After completing the preliminary experimental results, the resulting data was fed into a BP neural network model for training. Each experimental parameter, such as the number of hidden layers and nodes, corresponds to the structure of the neural network and is defined as an individual. To evaluate the predictive effectiveness of each network structure on material properties, we used cross-validation to train the neural network model on a training set and then judged the network's performance based on the prediction error on the validation set. The smaller the error, the higher the network's fitness.

[0034] Next, a multi-layer neural network algorithm is used to optimize the network structure. The parameters of the initial network model are randomly generated and encoded as real numbers representing each individual in the population. A genetic algorithm is used to select highly compatible experimental parameters. Random pairing and crossover operations are then performed to generate "new materials" with corresponding properties. Parameter adjustment operations are also introduced to maintain diversity in the preparation process, optimizing the network structure generation by generation. This process continues until a predetermined termination condition is reached, such as a maximum number of generations or a fitness threshold.

[0035] Once the optimal network structure parameters are obtained, this structure is used in subsequent model training to further improve its predictive performance. Through this optimization process, the neural network model can effectively predict material properties, such as changes in compressive strength after impact. Ultimately, this optimized network not only improves prediction accuracy but also significantly reduces calculation time, meeting the needs of structural reliability and immediate performance assessment.

[0036] Implementation of the neural network model: To achieve intelligent optimization of these process parameters, a multi-layer feedforward neural network (MLP) model was employed. The input layer contains process parameters (such as temperature, time, component ratio, stirring speed, MXene content, and magnetic particle content), while the hidden layer captures the nonlinear relationships between these parameters. The output layer then generates the optimal combination of process parameters associated with target performance (such as thermal performance, mechanical strength, and electromagnetic shielding).

[0037] The model uses a backpropagation algorithm and is trained using extensive historical experimental data, continuously adjusting the weights of its neurons to ensure high-precision predictions of process parameters. During production, real-time process data collected can be used to update and adjust the model, dynamically optimizing the manufacturing process and enabling intelligent production control.

[0038] Example 3 The method for designing a polyimide carbon fiber composite material structure based on a neural network model provided in this embodiment is a specific application of the aforementioned first and second embodiments, and further includes the following steps: 1. Material Preparation and Model Pre-training (1) Prepare raw materials, which include (by weight percentage)

[0039] BP neural network model construction: Input parameters: preparation temperature (200-350°C), hydrothermal reaction time (1-6 h), stirring rate (100-500 rpm), MXene content (20-80 wt%), magnetic particle content (20-80 wt%); Output targets: electromagnetic shielding effectiveness (SE ≥ 25 dB), tensile strength (TS ≥ 100 MPa), thermal conductivity (k ≥ 5 W / m·K); Training process: Parameter initialization: 100 groups of samples were selected from historical experimental data, the error threshold Φ was set to 0.05, and the number of weak predictors n was set to 10; Weak predictor training: Through iterative optimization of the backpropagation algorithm, high-error samples are screened and weights are dynamically adjusted to ultimately construct a strong prediction function h(x); Model validation: The reserved test set was used to verify the model accuracy, with a prediction error of ≤5%.

[0040] Output prediction data and composite material preparation 1) High thermal conductivity micron pitch-based carbon fibers (CF) are wet-oxidized to obtain pretreated carbon fibers. MXene and the obtained pretreated carbon fibers are modified by dopamine oxidative polymerization. The modified carbon fibers coated with dopamine MXene are mixed with a surfactant, an organic ligand, dopamine-modified MXene, and an organic solvent under mechanical stirring and nitrogen protection. The mixture is then added to a metal salt solution and stirred for 1 hour. A hydrothermal reaction is then carried out in a reactor. After the reaction, the product is washed and centrifuged and dried under a protective gas atmosphere at high temperature for 2 hours to obtain uniformly loaded nano-metal particles@MXene / micron pitch-based carbon fibers. 2) 1 g of commercial PBO microfibers was added to 100 mL of a 1:1 trifluoroacetic acid / methanesulfonic acid solution and mechanically stirred for 12 h to obtain a yellow PBO nanofiber dispersion. MXene was added to 100 mL of a 1:1 trifluoroacetic acid / methanesulfonic acid solution and ultrasonically dispersed for 1 h. The two were then mixed and mechanically stirred for 12 h to form a stable dispersion. A PBO / MXene electromagnetic shielding performance enhancement layer was prepared through a uniform gelation process using a sol-gel conversion method. 3) In step S3, based on the neural network model, polyimide carbon fiber composite material is assisted: The modified carbon fiber composite material and the electromagnetic shielding performance enhancement layer described in step S1 and step S2 are prepared according to the following raw material formula (weight percentage): 50% modified carbon fiber composite material and 50% electromagnetic shielding performance enhancement layer. The above materials are vacuum filtered and subsequently hot-pressed, wherein the hot-pressing temperature is 280°C, the pressure is 10 MPa, and the time is 2 hours. A composite material precursor is obtained by hot-pressing, and finally, it is impregnated with a polyimide resin and backfilled and heat-treated to obtain a polyimide resin-based electromagnetic shielding composite film.

[0041] 5. Performance Testing and Model Feedback Optimization Performance testing

[0042] Model closed-loop optimization Feedback the measured performance to the BP neural network model, update the training set and retrain; After three rounds of iterations, the error of SE value predicted by the model was reduced to 2%, and the error of tensile strength was ≤3%.

[0043] See attached Figure 1 The neural network model-assisted polyimide carbon fiber composite material structure design method provided in this embodiment is based on the methods and tools provided in the aforementioned embodiments 1 to 3, and specifically prepares polyimide carbon fiber composite materials with different target properties. The neural network model learns the experimental parameters obtained from past experiments, including process parameter data from different series of experiments, thereby having predictive capabilities. The following embodiments 1 to 7 specifically demonstrate the process of prediction and experimental verification of the neural network model. When the neural network model is specifically applied to the development of new materials with various specific target properties, it can achieve maximum optimization of electromagnetic shielding performance while ensuring mechanical properties.

[0044] The following is the prediction and preparation of a series of target performance materials based on the neural network model, and then the mechanical strength, electromagnetic shielding performance and thermal conductivity of the prepared polyimide carbon fiber composite materials are tested. The results are shown in the following table.

[0045]

[0046] Attachment Figure 1The SEM image of the polyimide carbon fiber composite material prepared in Example 1 is shown. The microstructural characteristics of this material are as follows: the image shows that the composite material is interwoven to form a complex network structure at the nanometer-micrometer scale, which helps to improve the electromagnetic shielding performance. In terms of electromagnetic shielding, the good electrical conductivity of carbon fiber enables its interwoven network to effectively reflect, scatter and absorb electromagnetic waves, forming a shielding layer; in terms of thermal conductivity, the network structure provides an efficient path for heat conduction, and heat can be quickly transferred along the paths where the fibers are connected to each other; and in terms of mechanical properties, the high strength characteristics of carbon fiber and the stable structure formed by the interwoven network between them enable the composite material to effectively disperse stress through the synergistic effect between fibers when subjected to external forces, thereby enhancing the overall mechanical properties of the composite material.

[0047] Furthermore, within the scope of the present invention, other embodiments obtained by optimizing the composite material preparation process with the aid of a neural network model, selecting different preparation processes, components, and their ratios based on the preparation conditions predicted by demand, can also achieve the intended technical effects of the present invention. Therefore, these embodiments are not listed one by one.

[0048] The above description is merely a preferred embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any technical features that are the same or similar to those of the above embodiments of the present invention are within the scope of protection of the present invention.

Claims

1. A method for designing a polyimide carbon fiber composite material structure based on a neural network model, characterized in that: It is based on the neural network model to assist in the design of polyimide carbon fiber composite material structure, construct a dense functional structure, take the filler type and its mass fraction as the input features of the neural network model, combine the target performance including mechanical properties, thermal properties and electromagnetic shielding performance, and simulate the influence of material combination and thickness on performance in multi-layer or gradient structure through the neural network model. It predicts the influence of different filler types and content ratios on the comprehensive performance of composite materials, optimizes design parameters during the structural design process, and obtains intelligent control of process parameters and optimization scheme of material preparation, so that the material can achieve optimization of thermal performance and electromagnetic shielding performance while maintaining excellent mechanical properties.

2. The method for designing a polyimide carbon fiber composite material structure based on a neural network model according to claim 1, characterized in that: The specific steps include: S1: Establish BP neural network model A back-propagation BP neural network model was established and pre-trained. This model was trained using a multi-layer feedforward neural network error back-propagation method. The network parameters were iteratively optimized to minimize the error between the network's actual output and the desired output. The input layer of the BP neural network model was used to input composite material preparation parameters, including temperature, time, component ratio, stirring speed, and magnetic particle content. The hidden layer was used to capture the nonlinear relationships between these parameters. The output layer provided the optimal combination of process parameters related to target performance, including thermal conductivity, mechanical strength, and electromagnetic shielding properties. S2: Predicting composite material preparation plans The target performance of the composite material designed for the direction is input into a pre-trained neural network model, and the model directionally predicts a preparation scheme for a polyimide resin-based electromagnetic shielding composite material, including directionally generating and outputting a predicted modified carbon fiber composite material scheme, a preparation scheme for an electromagnetic shielding performance enhancement layer, and a preparation scheme for a multifunctional layered polyimide resin-based electromagnetic shielding composite material; S3: Preparation of modified carbon fibers According to the directional prediction scheme, a modified carbon fiber composite material is prepared, wherein the high thermal conductivity micron pitch-based carbon fiber is immersed in acetone and ultrasonically treated at a set temperature and time, and then washed and dried after removing surface attachments; and the pretreated carbon fiber a is obtained by wet oxidation treatment with a mixed solution of concentrated sulfuric acid and concentrated nitric acid; Then, MXene and pretreated carbon fibers are modified by monoamine oxidative polymerization and mechanically stirred under nitrogen protection to obtain MXene-coated modified carbon fibers b. A surfactant, an organic ligand, a monoamine-modified micron pitch-based carbon fiber, and an organic solvent are mixed and added to a metal salt solution. The mixture is stirred for a set time and subjected to a hydrothermal reaction in a reactor. After the reaction, the mixture is washed and centrifugally dried. The obtained product is placed in a protective gas atmosphere and subjected to high-temperature treatment at a set temperature to obtain uniformly loaded nano-metal particles @MXene / micron pitch-based carbon fibers. Step S4: Preparing an electromagnetic shielding reinforcement layer Following the directional prediction scheme, an electromagnetic shielding enhancement layer was prepared. PBO microfibers were added to a mixture of trifluoroacetic acid and methanesulfonic acid and mechanically stirred to form a PBO nanofiber dispersion. MXene was ultrasonically dispersed at a set power and then mixed with the PBO nanofibers to form a stable dispersion. The PBO / MXene electromagnetic shielding enhancement layer was then prepared via a sol-gel conversion method. Step S5: Preparing a multifunctional polyimide resin-based electromagnetic shielding composite material According to the scheme of directional prediction of the neural network model, a multifunctional layered polyimide resin-based electromagnetic shielding composite material is prepared, and the modified carbon fiber obtained in step S3 and step S4 is mixed with the electromagnetic shielding reinforcement layer in proportion. The mixture is vacuum filtered and hot-pressed, and then backfilled by polyimide resin impregnation and heat treated to finally prepare a polyimide resin-based electromagnetic shielding composite material with a multifunctional layered dense structure.

3. The method for designing a polyimide carbon fiber composite material structure based on a neural network model according to claim 2, characterized in that: The following steps are also included: Step S6: Composite material performance testing, feedback and optimization The actual performance of the prepared polyimide resin-based electromagnetic shielding composite material with multifunctional layers was tested to verify the accuracy of the prediction scheme, and then a multi-layer feedforward neural network model was used to intelligently optimize the process parameters in the prediction scheme.

4. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 3, characterized in that: The following steps are also included: Step S7: Multi-objective parameter collaborative optimization based on a neural network model is based on an improved non-dominated sorting genetic algorithm, with electromagnetic shielding effectiveness, tensile strength and thermal conductivity as optimization targets, and a multi-objective Pareto front search is performed on the process parameter combination output from step S2 to obtain the optimal parameter set that meets the preset performance weights; the parameter set includes: the fluctuation range of the magnetic particle content, the stirring rate gradient, and the hot pressing molding temperature-time coupling curve.

5. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 2, characterized in that: The steps of establishing the BP neural network model in step S1 are as follows: ① Parameter initialization: Select from the sample space Group training data and initialize the distribution weights of test data , determine the prediction error threshold Φ and the number of weak predictors Initialize the weights from the input layer to the hidden layer, the hidden layer threshold, the weights from the hidden layer to the output layer, and the output layer threshold of the BP neural network; ② Weak predictor prediction: When training the t-th weak predictor, the sample parameters (including temperature, time, component ratio, stirring rate, and magnetic particle content) are input in a feedforward manner, the network weights are optimized through the back propagation algorithm, and the output of the training set is predicted to obtain g t (xi) and the absolute value of the prediction error e t (i), the calculation formula is: i=1,2,…,m where y i Actual performance parameters, including tensile strength, thermal conductivity, and electromagnetic shielding performance; ③ Calculate the error rate :Filter high error samples based on the error threshold Φ. The calculation formula is: in is an indicator function, which takes 1 when the condition is met and 0 otherwise; ④ Calculate the weak predictor weight α t : Dynamically adjust the contribution weight of the weak predictor according to the error rate. The calculation formula is: ; ⑤ Adjust the sample weight of the t+1th iteration: Increase the weight of high-error samples to optimize the model sensitivity. The calculation formula is: Where B t is the normalization factor, ensuring that the sum of the weights is 1; ⑥ Construct a strong predictor: Combine n weak BP neural network predictors into a strong prediction function h(x) through ensemble learning. Its output is the weighted voting result. The calculation formula is: The final model iteratively optimizes the mapping relationship between process parameters and performance, outputting the optimal component ratio and process conditions, including temperature, time, component ratio, stirring speed, and magnetic particle content; The pre-training step is to use the parameters of temperature, time, component ratio, stirring speed and magnetic particle content in the composite material preparation process as input feature data, combined with target performance data, including mechanical properties, thermal properties, and electromagnetic shielding performance, and import them into the BP neural network model for training, so that the pre-trained neural network can effectively predict the influence of different filler types and contents on the comprehensive performance of the material, and obtain a pre-trained neural network model.

6. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 2, characterized in that: In the preparation process of step S3, a series of composite materials are prepared by pre-optimizing multiple key process parameters, including ultrasonic treatment time, reaction temperature, and stirring rate, through a neural network model; the neural network model is pre-trained by learning a large amount of experimental data, and the obtained pre-trained model can predict the influence of these parameters on the properties of carbon fiber, including mechanical properties, thermal properties, and electromagnetic shielding properties, and output the optimal combination of preparation parameters according to the target performance requirements to ensure the effective loading of MXene on the carbon fiber surface.

7. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 3, characterized in that: The step S6 specifically adopts a multi-layer feedforward neural network model based on the back propagation algorithm to construct an intelligent production control model. The multi-layer feedforward neural network model is pre-trained with a large amount of historical experimental data, and the weights of the neurons are continuously adjusted to ensure high-precision process parameter prediction capabilities.

8. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 2, characterized in that: The carbon fiber in step S3 is a high thermal conductivity asphalt-based carbon fiber, and the thermal conductivity of the carbon fiber is ≥800 W / (m·K); the material after vacuum filtration is subjected to hot pressing treatment, and the hot pressing temperature is 160~320℃, the pressure is 5-10MPa, and the time is 1-3h.

9. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 2, characterized in that: The surfactant in step S3 is one of the cationic surfactant cetyltrimethylamine salt, the anionic surfactant sulfonate, phosphate, and the nonionic surfactant octylphenol polyoxyethylene ether; the monoamine monomer is one of dopamine, 5-hydroxytryptamine, aniline, ethylenediamine, and m-phenylenediamine.

10. The method for designing a polyimide carbon fiber composite material structure based on a neural network model as claimed in claim 2, characterized in that: The organic solvent in step S3 is one of N,N-dimethylformamide, dimethyl sulfoxide, and dimethylacetamide; The weight proportion of the modified carbon fiber in steps S3 and S4 to the overall raw material formula is in the range of 40-60%, and the weight proportion of the electromagnetic shielding performance enhancement layer to the overall raw material formula is in the range of 60-40%.

Images