Data-driven prediction method for fire retardant properties of fiber reinforced composites

Abstract

The present application relates to a kind of prediction method of data-driven based on the flame retardant performance of fiber reinforced composite material, the prediction method includes the following steps: S1) data set construction and pre-processing;S2) construct the deep neural network model for predicting the flame retardant performance of fiber reinforced composite material;The deep neural network model takes fiber parameter, flame retardant parameter and resin parameter as input variable, and flame retardant performance is output variable;S3) model verification and test;S4) flame retardant performance prediction application.The present application comprehensively considers fiber type, flame retardant type, resin type, raw material ratio and other characteristic variables for global optimization, the accuracy of the flame retardant performance prediction of fiber reinforced composite material of single flame retardant analysis can be improved by the synergy of multivariable, batch construction, calculation and processing of simulation model and result are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of fiber-reinforced composites, and particularly relates to a data-driven prediction method for the flame-retardant performance of fiber-reinforced composites. BACKGROUND

[0002] Fiber-reinforced composites are widely used in aerospace, energy and power, electronics, biomedicine and other fields due to their excellent performance, and have become a key basic material for lightweight and functional construction. In order to meet the actual application requirements, flame-retardant technology is usually applied in raw material formula or production process of fiber-reinforced composites, so as to improve the flame-retardant performance of the composites. In actual production, the flame-retardant performance of fiber-reinforced composites is affected by many factors such as fiber type, flame retardant type, resin type, raw material ratio and production process. Therefore, the prediction of the flame-retardant performance of fiber-reinforced composites has important guiding significance for industrial production.

[0003] Chinese patent CN103678875A discloses a prediction method for the dispersion behavior of flame retardants and flame retardant synergists in a polymer matrix, which comprises establishing a molecular model of the flame-retardant polymer material and a layer structure model of the polymer / inorganic flame retardant system or the polymer / inorganic flame retardant / flame retardant synergist system, obtaining the compatibility behavior of the polymer / inorganic flame retardant system or the polymer / inorganic flame retardant / flame retardant synergist system by using the Flory-Huggins model, calculating the binding energy and radial distribution function of the layer structure of the polymer / inorganic flame retardant system or the polymer / inorganic flame retardant / flame retardant synergist system by using the molecular dynamics method, so as to determine the influence of the flame retardant and the flame retardant synergist on the micro interaction of the polymer flame-retardant material. The phase distribution and dispersion behavior of the polymer / inorganic flame retardant system or the polymer / inorganic flame retardant / flame retardant synergist system are analyzed by using the dissipative particle dynamics method.

[0004] Chinese patent CN116312885A discloses a wood composite material flame-retardant performance prediction method and device, which quickly predicts the actual value of the flame-retardant performance of the wood composite material according to the existing structure data and production process data of the wood veneer and other conditions through a trained machine learning model.

[0005] Japanese patent JP2024035451A discloses a physical property prediction method and device for resin composition, which provides a physical property prediction method for resin composition capable of accurately predicting the elongation and tensile strength of resin composition, and a prediction device for predicting the physical properties of resin composition manufactured by using base polymer, flame retardant, flame retardant aid and other materials.

[0006] The prior art does not solve the problem of predicting the flame retardant performance of fiber-reinforced composites, and more attention is focused on the prediction of the dispersion behavior of flame retardants in the polymer matrix. SUMMARY

[0007] In view of the problems existing in the prior art, the purpose of the present application is to provide a method for predicting the flame retardant performance of fiber-reinforced composites.

[0008] One aspect of the present application provides a data-driven method for predicting the flame retardant performance of fiber-reinforced composites, comprising the following steps:

[0009] S1) Data set construction and preprocessing

[0010] Experimental data and material database information of glass fiber reinforced epoxy resin composites are collected, and standardized fiber parameters, flame retardant parameters and resin parameters are used as input variables, and the flame retardant performance is used as the output variable to construct a sample data set;

[0011] S2) Constructing a deep neural network model for predicting the flame retardant performance of fiber-reinforced composites;

[0012] S201) Neural network structure design: design a deep neural network model with full connection structure or a deep neural network model with multi-channel hybrid structure;

[0013] The deep neural network model with full connection structure comprises an input layer, a plurality of full connection hidden layers and an output layer;

[0014] The deep neural network model with multi-channel hybrid structure comprises constructing a numerical feature channel and a category feature channel, and outputting through an output layer after weighted fusion of the double-channel features through an attention mechanism;

[0015] S202) Model training: using the constructed sample data set to train the deep neural network model, initializing the weights and biases before training, defining the loss function and introducing the thermodynamic constraint term to make the prediction results consistent with the material stability parameters, using the Adam optimization algorithm or introducing the gradient penalty of Wasserstein GAN for optimization, monitoring the loss value change during training and using the Early Stopping strategy to determine convergence;

[0016] S3) Model verification and testing, the prediction error and generalization ability of the model are evaluated by the validation set and the test set respectively;

[0017] S4) Flame retardant performance prediction application, input the parameters of the material to be tested into the trained deep neural network model, and output the prediction result of the flame retardant performance.

[0018] Further, the fiber parameters include fiber diameter and volume fraction, the flame retardant parameters include flame retardant type, decomposition temperature and volume fraction, and the resin parameters include resin type, resin content, viscosity and decomposition temperature; and the flame retardant performance includes limiting oxygen index, 12s vertical burning length and extinguishing time.

[0019] Further, the fiber volume fraction is 30-47 vol% of the total volume fraction of the composite material; the flame retardant volume fraction is 3-10 vol% of the total volume fraction of the composite material; and the resin content is 50-67 vol% of the total volume fraction of the composite material.

[0020] Further, in the flame retardant performance, the limiting oxygen index is 28-40%, the 12s vertical burning length is 1-15 cm, and the extinguishing time is 1-15 s.

[0021] Further, step S1) comprises the following steps:

[0022] S101) Collecting experimental data and material database information of fiber reinforced composite materials, the data including fiber parameters, flame retardant parameters, resin parameters and flame retardant performance data, and generating virtual samples based on thermodynamic model to enhance feature dimension;

[0023] S102) Standardizing and preprocessing the collected data, including normalizing numerical data, one-hot encoding or embedding layer mapping for text type data, and processing missing values and outliers, and performing dimensionality reduction when the data dimension exceeds the preset condition;

[0024] S103) Taking the standardized glass fiber, flame retardant and resin parameters as input variables and the flame retardant performance as output variable to construct a sample data set.

[0025] Further, the thermodynamic model introduced in S101) is the Kissinger equation, which is used to simulate the decomposition behavior of the flame retardant.

[0026] Further, the normalization method includes Z-score standardization method or Min-Max normalization method.

[0027] Further, in S201), the numerical feature channel extracts features through a fully connected layer, the category feature channel is processed through a bidirectional long short-term memory network (Bi-LSTM), and an attention mechanism is introduced to fuse the outputs of each channel.

[0028] Further, in S201), the category features in the category feature channel are mapped to low-dimensional dense vectors through an embedding layer and input to the Bi-LSTM network for sequence modeling.

[0029] Further, in step S202), the loss function adopted by the model training process includes a mean square error (MSE) and at least one thermodynamic constraint term, so that the predicted limiting oxygen index is positively correlated with the thermal decomposition temperature of the fiber or resin.

[0030] Further, in step S202), the model training process optimization algorithm adopts the gradient penalty strategy of Adam or Wasserstein GAN.

[0031] Further, in step S202), the model training process optimization algorithm adopts the Early Stopping strategy to control the training process.

[0032] Further, the training data is divided into a training set, a validation set and a test set in the proportions of 80%, 10% and 10% respectively.

[0033] Further, the number of neurons in the hidden layer of the DNN neural network model is 64 to 256 per layer, and the number of hidden layers is 3 to 5 layers.

[0034] Further, in step S3), the error includes a mean square error (MSE), a root mean square error (RMSE) and a determination coefficient (R²).

[0035] Further, in step S3), the generalization ability of the model is evaluated using the test set, and the model is optimized according to the results.

[0036] Further, the type of fiber in the fiber-reinforced composite material is one of glass fiber, basalt fiber, carbon fiber, aramid fiber, polyimide fiber and polyarylate fiber.

[0037] Further, the type of resin in the fiber-reinforced composite material is one of polyethylene, polypropylene, polymethylpentene, polystyrene, polyamide and polycarbonate.

[0038] The application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the above prediction method.

[0039] The application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor implements the steps of the above prediction method when executing the computer program.

[0040] Advantages

[0041] Compared with the prior art, the flame-retardant performance of the fiber-reinforced composite material is analyzed by parameterized modeling, a flame-retardant performance analysis result database of the fiber-reinforced composite material is constructed, a model is constructed by combining data driving and introducing a neural network method, and the flame-retardant performance of the fiber-reinforced composite material is quickly and intelligently predicted. The flame-retardant performance prediction accuracy of the fiber-reinforced composite material by single flame retardant analysis is improved by comprehensively considering the global optimization of characteristic variables such as fiber types, flame retardant types, resin types, and raw material ratios, and the synergy of multiple variables. Batch construction, calculation and processing of simulation models and results are realized. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A topological structure diagram of a DNN neural network of the embodiment of the present application.

[0043] Figure 2 A model construction and application flowchart of the flame-retardant performance prediction method of the fiber-reinforced composite material of the present application. DETAILED DESCRIPTION

[0044] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. Note that the relative arrangement, numerical expressions, and numerical values of components and steps set forth in these embodiments are not limiting to the scope of the present application unless otherwise specifically stated. The following description of at least one exemplary embodiment is merely illustrative in nature and in no way limits the present application and its application or uses.

[0045] Techniques, methods, and equipment known to those of ordinary skill in the relevant art can not be discussed in detail, but where appropriate, the purpose of the present application is to provide a flame-retardant performance prediction method for fiber-reinforced composite materials, which uses data-driven methods to quickly and accurately predict the flame-retardant performance of fiber-reinforced composite materials, helps optimize material ratios and production processes, and improves production efficiency and product quality.

[0046] The techniques, methods, and equipment should be considered part of the specification. In all examples shown and discussed here, any specific values should be interpreted as merely exemplary, not as limiting. Therefore, other examples of exemplary embodiments can have different values.

[0047] The present application aims to provide a flame-retardant performance prediction method for fiber-reinforced composite materials, which uses data-driven methods to quickly and accurately predict the flame-retardant performance of fiber-reinforced composite materials, helps optimize material ratios and production processes, and improves production efficiency and product quality.

[0048] Embodiment 1

[0049] Specifically, in combination with Figure 1As shown, the application provides a method for predicting the flame retardant performance of a glass fiber reinforced polypropylene composite material, comprising the following steps:

[0050] Step S1) Data set construction and preprocessing

[0051] S101) Data collection

[0052] Experimental data of the glass fiber reinforced polypropylene composite material is collected, including parameters of glass fiber, parameters of flame retardant, parameters of resin, and flame retardant performance.

[0053] The parameters of glass fiber include fiber diameter and volume fraction; the parameters of flame retardant include flame retardant type, decomposition temperature and volume fraction; the parameters of resin include polypropylene brand, viscosity and decomposition temperature; the flame retardant performance includes limiting oxygen index, 12s vertical burning length and extinguishing time.

[0054] In some specific embodiments, the fiber diameter of the glass fiber reinforced polypropylene composite material is 11 μm, the volume fraction is 30-47 vol%; the flame retardant type of the flame retardant parameter is phosphorus flame retardant, the decomposition temperature is 250-350℃, the volume fraction is 5%-10 vol%; the resin parameter is polypropylene, the viscosity is 1.05-2.00 deciliter / gram, the decomposition temperature is 350-370℃; the limiting oxygen index of the flame retardant performance is in the range of 28%-35%, the 12s vertical burning length is 5cm-15cm, and the extinguishing time is 5s-15s.

[0055] In addition to experimental data, existing literature and patent data, tool books and public material databases (such as NIST, Polymer Database) need to be collected, and material attribute data such as fiber characteristics, thermal conductivity and flame retardant decomposition temperature are introduced to enhance the feature dimension. The decomposition behavior of different flame retardants is simulated by using thermodynamic model (such as Kissinger equation) to generate virtual samples.

[0056] S102) Data preprocessing

[0057] The data obtained in S101) is standardized and preprocessed, which includes normalizing the numerical type data, converting the text type data into digital form by one-hot encoding, or using embedding layer to map into low-dimensional dense vector instead of traditional one-hot encoding, capturing the potential correlation between materials, and comprehensively checking the missing values and abnormal values in the processed data. After processing, it can also be judged according to the data dimension of standardization processing, when the standardized data amount exceeds 1000 and the dimension exceeds 10, the dimension reduction processing is carried out.

[0058] For the numerical data of raw material ratio and flame retardant performance, a normalization method is used to convert data of different dimensions to the same scale so that the neural network model can learn effectively. In some embodiments, the normalization method uses Z-score standardization method or Min-Max normalization method.

[0059] The fiber type, flame retardant type, resin type, raw material ratio and flame retardant performance data are comprehensively checked to identify and process missing values and outliers. For the case of missing values in the collected data, mean filling or interpolation method is used to ensure data integrity. And by data denoising, outliers are removed. In some embodiments, data deviating more than 3 times the standard deviation from the mean is considered as an outlier and is removed. Dimensionality reduction processing uses principal component analysis (PCA) or t-SNE dimensionality reduction method to reduce data dimension.

[0060] S103) Constructing a data set

[0061] The standardized parameters of glass fiber, flame retardant parameters and resin parameters are used as variable parameters, and the flame retardant performance under different values of these variable parameters is used to construct a sample data set. The sample data set contains at least 200 sample data.

[0062] S2) DNN neural network model construction

[0063] S201) DNN neural network structure design

[0064] The deep neural network (DNN) structure includes input layer, hidden layer and output layer.

[0065] The feature variables received by the input layer are the variable parameters in the data set obtained in step S1). The hidden layer consists of 3-5 fully connected layers, each containing 64-256 neurons. In some embodiments, the hidden layer uses ReLU (Rectified Linear Unit) activation function to improve the nonlinear fitting ability of the model; in some embodiments, BatchNormalization is introduced to accelerate convergence and prevent gradient disappearance. The output layer is used to predict the flame retardant performance of glass fiber reinforced polypropylene composite material, and the flame retardant performance indicators are limiting oxygen index, 12s vertical burning length and extinguishing time.

[0066] On the basis of the above DNN neural network structure, a multi-channel hybrid DNN structure is further constructed, including a numerical feature processing channel and a category feature processing channel, and the features are fused and outputted.

[0067] Numerical feature processing channel: the numerical features in the data set obtained in step S1 are input into a fully connected layer (FC), including the fiber diameter and volume fraction of glass fiber, the decomposition temperature and volume fraction of flame retardant, the viscosity and decomposition temperature of resin, and other raw material ratio and physical property data. The fully connected layer is composed of 3 layers, and the number of neurons in each layer is 128, 64, and 32 respectively. The fully connected layer realizes deep feature extraction and nonlinear transformation of numerical features.

[0068] Category feature processing channel: for the category features in the data set, including glass fiber type, flame retardant type, polypropylene resin type, etc., the low-dimensional dense vectors are mapped through an embedding layer, and a bidirectional long short-term memory network (Bi-LSTM) is used for feature extraction processing. The network structure of the Bi-LSTM containing 2 layers of hidden layers is constructed, and the number of hidden units in each layer is 64. The sequence dependence relationship between different category features is captured through the Bi-LSTM network, so as to mine the synergistic effect between the flame retardant and the glass fiber and the resin.

[0069] Feature fusion and output: on the basis of the outputs of the numerical feature processing channel and the category feature processing channel, an attention mechanism is introduced.

[0070] The outputs of the fully connected layer and the Bi-LSTM are taken as the inputs of the attention mechanism. The outputs of different channels are dynamically weighted through the attention mechanism, and the feature combination that plays a key role in predicting the flame retardant performance (limiting oxygen index, 12s vertical burning length and extinguishing time) of the glass fiber reinforced epoxy resin composite material is automatically focused. Finally, the weighted and fused features are input into the output layer for flame retardant performance prediction.

[0071] S202) Model training

[0072] 80% of the sample data set is used as the training set, 10% is used as the validation set, and 10% is used as the test set. The parameters of glass fiber, flame retardant parameters, and resin parameters are used as input data, and the flame retardant performance is used as output data. With the help of the training set, the DNN neural network is trained.

[0073] Before the training work is carried out, the weights and biases of the DNN neural network are initialized. The loss function and optimization algorithm are defined, wherein the mean square error (MSE) is used as the loss function to measure the error between the predicted flame retardant performance of the model and the actual optimal flame retardant performance. In order to improve the physical rationality of the model, a thermodynamic constraint term can be introduced into the loss function, such as making the predicted limiting oxygen index (LOI) of the model and the stability index such as the thermal decomposition temperature of the fiber or the resin maintain a positive correlation.

[0074] The optimization process adopts the stochastic gradient descent (SGD) algorithm and its variants as the optimization algorithm, such as using the Adam optimization algorithm to adaptively adjust the learning rate and accelerate the model convergence speed; or introducing the gradient penalty matrix of Wasserstein GAN to force the model to generate a predicted distribution that conforms to the flame retardant mechanism, and improve the physical consistency of the predicted results generated by the model with the flame retardant mechanism.

[0075] The training set is divided into several batches, each batch containing a certain number of samples, such as 32 samples per batch. In each training step, the gradient is calculated from the current batch of data, and the weights and biases of the neural network are updated. During the training process, monitoring is performed, and the loss values of the training set and the validation set are calculated regularly, and the loss curve is plotted. Model convergence judgment, when the model is considered to have reached a state of convergence, stop training. The model convergence judgment can use the Early Stopping strategy, for example, if the validation set loss function does not decrease for 10 consecutive rounds, stop training.

[0076] S3) Model verification and testing

[0077] S301) Model performance evaluation

[0078] The trained DNN neural network is tested for effectiveness using the validation set, and the prediction error of the model on the validation set is calculated. If the error exceeds the preset range, check if the data is abnormal, adjust the model structure and retrain.

[0079] Calculate the mean square error (MSE), root mean square error (RMSE), coefficient of determination (R²), etc.

[0080] The DNN neural network model that passes the verification is used to perform prediction verification on the test set, and the model prediction results are compared with the actual test results to evaluate the accuracy and generalization ability of the model. If the model performs worse than expected on the prediction set, further optimize the model, and the optimization methods include increasing the amount of sample data, adjusting the model structure or optimizing the training algorithm.

[0081] S4) Flame retardant performance prediction application

[0082] S401) Data input

[0083] Obtain the information of the fiber type, flame retardant type, resin type, raw material ratio, etc. of the fiber reinforced composite material to be tested, and input it into the trained DNN neural network model.

[0084] S402) Result output:

[0085] The model outputs the prediction result of the limiting oxygen index of the glass fiber reinforced polypropylene composite material to be tested, so as to provide a decision basis for material research and development personnel and production enterprises in material ratio design and production process optimization. Enterprises can select the optimal material formula and production process according to the prediction result, reduce the production cost, and improve the flame retardant performance and quality of the product.

[0086] In summary, the prediction method of the present application has at least the following advantages over the prior art:

[0087] 1) The present application builds a DNN neural network model, taking fiber types, flame retardant types and other parameters as inputs. After training and learning, the model can quickly give the flame retardant performance prediction result, greatly shortening the research and development cycle. In actual production, the material formula and process may need to be adjusted in real time due to various factors. The model of the present application can quickly process new input parameters and timely predict the adjusted flame retardant performance, providing immediate support for production decision-making.

[0088] 2) The prior art has poor prediction accuracy when analyzing the influence of a single flame retardant on the flame retardant performance of fiber reinforced composite materials, because it is difficult to comprehensively consider the interaction of multiple factors. The present application considers the synergistic effect of fiber types, resin types, raw material ratios and other factors with flame retardants through parameterized modeling, so that the model can capture the complex physical and chemical relationships inside the material.

[0089] The method of the present application can be used for prediction of other different fiber reinforced thermoplastic composite materials in addition to glass fiber reinforced polypropylene composite materials. Fiber reinforced composite materials are a kind of high-performance materials composed of fibers and matrix materials through a specific process. Among them, the reinforcing fiber bears the main load and endows the material with high strength, high modulus and other characteristics; the matrix material bonds the reinforcing fiber together, so that the composite material has a certain shape, protects the fiber from external environment, and transmits stress when the material is stressed. The fibers include carbon fiber, glass fiber, natural fiber, ceramic fiber and other different fibers. The matrix material includes resin matrix and the like.

[0090] The present application can be a system, a method and / or a computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon, which cause a processor to implement aspects of the present application.

[0091] A computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0092] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0093] Computer readable program instructions for carrying out operations of the present application can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate array (FPGA), or programmable logic array (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present application.

[0094] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0095] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0096] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0097] The flow diagrams and the block diagrams in the drawings are presented to illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flow diagrams and the block diagrams can represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logic functions. In some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow diagrams, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and

[0098] Embodiments of the application have been described above, and the description is intended to be illustrative of the embodiments of the application and not exhaustive. Numerous modifications and adaptations thereof will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The scope of the application is defined by the appended claims.

Claims

1. A data-driven method for predicting the flame retardant properties of fiber-reinforced composite materials, characterized in that, The prediction method includes the following steps: S1) Dataset construction and preprocessing; Experimental data and material database information of fiber-reinforced composite materials were collected, and a sample dataset was constructed using standardized fiber parameters, flame retardant parameters and resin parameters as input variables and flame retardant performance as output variable. S2) Construct a deep neural network model for predicting the flame retardant properties of fiber-reinforced composites; S201) Neural Network Architecture Design: Design a fully connected deep neural network model or a multi-channel hybrid deep neural network model; A fully connected deep neural network model consists of an input layer, multiple fully connected hidden layers, and an output layer. A deep neural network model with a multi-channel hybrid structure includes constructing numerical feature channels and class feature channels, and then outputting the dual-channel features through a weighted fusion of attention mechanisms and outputting them through an output layer. S202) Model Training: The deep neural network model is trained using the constructed sample dataset. Before training, the weights and biases are initialized, the loss function is defined and a thermodynamic constraint term is introduced to ensure that the prediction results are physically consistent with the material stability parameters. The Adam optimization algorithm or the gradient penalty of Wasserstein GAN is introduced for optimization. During the training process, the change of loss value is monitored and the Early Stopping strategy is used to determine convergence. S3) Model validation and testing: The model's prediction error and generalization ability are evaluated using validation and test sets, respectively. S4) Flame retardant performance prediction application: Input the parameters of the material to be tested into a trained deep neural network model, and output the prediction results of its flame retardant performance. Step S1) includes the following steps: S101) Collect experimental data and material database information of fiber-reinforced composite materials. The data includes fiber parameters, flame retardant parameters, resin parameters and flame retardant performance data. Based on the thermodynamic model, virtual samples are generated to enhance the feature dimension. S102) Standardize and preprocess the collected data, including normalizing numerical data, one-hot encoding or embedding layer mapping of text data, and handling missing and outlier values. Dimensionality reduction is performed when the data dimension exceeds the preset conditions. S103) uses standardized fiber, flame retardant and resin parameters as input variables and flame retardant performance as output variable to construct a sample dataset; In S201), the numerical feature channel extracts features through a fully connected layer; the categorical feature channel in S201) is processed through a bidirectional long short-term memory network. In step S202), the loss function used in the model training process includes mean square error and at least one thermodynamic constraint term, so that the predicted limiting oxygen index is positively correlated with the thermal decomposition temperature of the fiber or resin.

2. The prediction method according to claim 1, characterized in that, The fiber parameters include fiber diameter and volume fraction; the flame retardant parameters include flame retardant type, decomposition temperature and volume fraction; the resin parameters include resin type, resin content, viscosity and decomposition temperature; and the flame retardant properties include limiting oxygen index, 12s vertical burning length and extinguishing time.

3. The prediction method according to claim 1, characterized in that, The fiber-reinforced composite material contains one of the following fiber types: glass fiber, basalt fiber, carbon fiber, aramid fiber, polyimide fiber, and polyarylate fiber.

4. The prediction method according to claim 1, characterized in that, The resin in the fiber-reinforced composite material is one of polyethylene, polypropylene, polymethylpentene, polystyrene, polyamide, and polycarbonate.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the prediction method according to any one of claims 1-4.

6. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the prediction method according to any one of claims 1-4.

Images