A composite material progressive damage prediction method based on LSTM-DNN

Abstract

The application provides a composite material progressive damage prediction method based on an LSTM-DNN, and comprises the following steps: generating a parameter distribution set of the composite material through a Monte Carlo method; developing ABAQUS through a Python script to simulate stress-strain curves and damage and failure evolution processes of the composite material under different conditions, and extracting time series data; constructing a series connection model combining an LSTM network and a DNN network to extract time series features and static features of the composite material and to fuse and predict stress sequences and damage factor sequences; training the model based on a back propagation algorithm, optimizing parameters through a loss function, and improving prediction accuracy by combining physical law constraints; and inputting new displacement and load data into the model to generate stress sequences and damage factor sequences, and predicting the progressive damage state and residual life of the composite material. The application models and predicts the time-dependent constitutive relation and progressive damage process of the composite material through the LSTM, and can improve the accuracy of the damage path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computational mechanics and material science, and particularly relates to a composite material progressive damage prediction method based on LSTM-DNN. BACKGROUND

[0002] Composite materials have been widely used in aerospace, automotive and energy industries due to their high strength and light weight. However, due to the complexity of the internal structure of composite materials, their failure behavior often exhibits a progressive nature. The progressive damage process usually involves fiber breakage, matrix cracking, and interfacial debonding, etc. and is accompanied by nonlinear characteristics. This complexity not only makes it difficult to characterize the damage mechanism of composite materials, but also poses a great challenge to material failure prediction. At present, although theoretical research and experimental testing have revealed the progressive damage characteristics of composite materials to some extent, experimental testing is costly and has limited prediction accuracy, which cannot meet the practical engineering requirements.

[0003] Progressive damage theory provides a new perspective for the study of the mechanical behavior of composite materials, which includes delamination, cracking and other phenomena under complex loading conditions, and attempts to accurately predict the material failure process through modeling methods. However, with the in-depth study of progressive damage theory, the complexity of the damage evolution process of composite materials gradually emerges. The diversity of failure modes, coupled nonlinear mechanical behavior and multi-scale characteristics further increase the difficulty of strength performance evaluation. This makes the traditional finite element method and continuous damage model face many problems in practical application: first, low computational efficiency. Traditional numerical methods rely on high-resolution meshing and complex nonlinear iteration, resulting in large consumption of computing resources and difficulty in meeting the time efficiency requirements of engineering. Second, insufficient accuracy. Under multi-axial loading conditions, traditional methods are difficult to effectively capture the influence of nonlinear load history on material damage behavior. Third, poor data correlation. Traditional methods fail to fully utilize the time correlation in experimental data, resulting in insufficient reliability and accuracy of damage prediction results.

[0004] The above problems seriously limit the application of traditional methods in composite material damage prediction, especially in complex engineering environments that require efficient and accurate prediction. It is urgent to find a more efficient and accurate solution. SUMMARY

[0005] The present application aims to at least partially solve one of the technical problems in the related art.

[0006] To this end, the first object of the present application is to provide a composite material progressive damage prediction method based on LSTM-DNN.

[0007] A second object of the present application is to provide a composite material progressive damage prediction device based on LSTM-DNN.

[0008] A third object of the present application is to provide an electronic device.

[0009] A fourth object of the present application is to provide a computer-readable storage medium.

[0010] A fifth object of the present application is to provide a computer program product.

[0011] To achieve the above objects, the first aspect of the present application provides a composite material progressive damage prediction method based on LSTM-DNN, comprising:

[0012] According to the geometric parameters, material parameters and load conditions of the composite material, the parameter range and distribution required for modeling are determined, and the parameter distribution set is generated by using the Monte Carlo method;

[0013] The parameter distribution set is input into ABAQUS for modeling calculation by using Python script for secondary development of ABAQUS, the stress-strain curve and damage-failure evolution process of the composite material under different parameter conditions are simulated, and the time series data of the composite material are extracted;

[0014] A series connection model combining LSTM network and DNN network is constructed, wherein the LSTM network is used to process the time series matrix and extract the time series features of the composite material, the DNN network is used to extract the features of the structural parameters and material strength parameters and fuse them with the time series features, and the stress sequence and damage factor sequence prediction results are output;

[0015] The preprocessed time series data, static structural parameters and material strength parameters are taken as input variables, the series connection model is trained based on the back propagation algorithm, the model parameters are optimized by using the loss function, the overfitting problem of the model is controlled by dynamically adjusting the learning rate and the regularization technique, and the final prediction model is generated after the training is completed;

[0016] Based on the trained model, new displacement and load data are input into the model to generate the stress sequence and damage factor sequence of the composite material, and the progressive damage state of the composite material structure is judged by the damage factor to predict the progressive damage state and remaining life of the composite material.

[0017] Optionally, the geometric parameters of the composite material include fiber layer angle and composite material thickness ; the material parameters of the composite material include fiber elastic modulus , Poisson's ratio , and matrix yield strength Interlaminar bonding strength ; the load condition of the composite material includes load history and strain rate .

[0018] Optionally, the Monte Carlo method is used to generate a combination of modeling parameters with random distribution, which conforms to the parameter range in actual engineering application, and is generated by the following way:

[0019] The fiber layer angle is randomly distributed in the range of 0° to 90°;

[0020] The thickness of the composite material randomly fluctuates within the range allowed by engineering;

[0021] The elastic modulus, Poisson's ratio, yield strength and bonding strength are randomly distributed according to the statistical law of material performance.

[0022] Optionally, the time series data includes displacement array , load history , deformation rate , damage factor .

[0023] Optionally, the preprocessing process of the time series data, structure parameters and material strength parameters includes:

[0024] The Kalman filter method is used to denoise the time series data , and the time series data after denoising is converted into a standard time series matrix , wherein, is the number of time steps;

[0025] The converted standard time series matrix and static structure parameters and material strength parameters are normalized, and the normalization formula is:

[0026]

[0027] , wherein, is the normalized value, and are the minimum and maximum values of the input variable respectively.

[0028] Optionally, the series connection model adopts three-layer LSTM unit, and a Dropout layer is added between each layer to prevent overfitting, and the single-step formula of each layer LSTM unit is:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] wherein, , , respectively represent the activation values of the forget gate, the input gate and the output gate, is the cell state, is the hidden layer state.

[0035] Optionally, the loss function of the series connection model is:

[0036]

[0037] wherein, , are the weight coefficients of different loss terms and , is the damage evolution constraint based on physical laws, and respectively are the predicted stress sequence and the damage factor sequence, and respectively are the simulated stress sequence and the damage factor sequence.

[0038] Optionally, the method further comprises:

[0039] calculating the damage factor of the composite material using a damage model weighted by fiber and matrix separation , and the formula is:

[0040]

[0041] wherein, and are the weight coefficients of fiber damage and matrix damage and , is the fiber damage variable, is the matrix damage variable;

[0042] the calculation formula of the fiber damage variable is:

[0043]

[0044] wherein, is the fiber damage critical variable, the maximum bearing capacity of the fiber;

[0045] the matrix damage variable The calculation formula is:

[0046]

[0047] In the formula, is the fracture strain of the matrix.

[0048] To achieve the above object, the second aspect of the present application proposes a composite material progressive damage prediction device based on LSTM-DNN, comprising:

[0049] A parameter distribution set generation module is configured to determine the parameter range and distribution required for modeling according to the geometric parameters, material parameters and load conditions of the composite material, and generate a parameter distribution set using the Monte Carlo method.

[0050] An extraction module is configured to use a Python script to develop ABAQUS, input the parameter distribution set into ABAQUS for modeling and calculation, simulate the stress-strain curve and damage evolution process of the composite material under different parameter conditions, and extract the time series data of the composite material.

[0051] A model construction module is configured to construct a series connection model combining an LSTM network and a DNN network, wherein the LSTM network is used to process the time series matrix and extract the time series features of the composite material, and the DNN network is used to extract the features of the structural parameters and material strength parameters and fuse them with the time series features to output the stress sequence and damage factor sequence prediction results.

[0052] A model training module is configured to use the preprocessed time series data, static structural parameters and material strength parameters as input variables, train the series connection model based on the back propagation algorithm, optimize the model parameters with a loss function, control the overfitting problem of the model through dynamic adjustment of the learning rate and regularization techniques, and generate a final prediction model after training.

[0053] A prediction module is configured to input new displacement and load data into the model based on the trained model, generate the stress sequence and damage factor sequence of the composite material, judge the progressive damage state of the composite material structure through the damage factor, and predict the progressive damage state and remaining life of the composite material.

[0054] To achieve the above object, the third aspect of the present application proposes an electronic device, comprising a processor and a memory in communication connection with the processor.

[0055] The memory stores computer execution instructions.

[0056] The processor executes the computer execution instructions stored in the memory to implement the method of any one of the first aspect.

[0057] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to implement the method according to any one of the first aspect.

[0058] To achieve the above object, the fifth aspect of the present application provides a computer program product, wherein the computer program is executed by a processor to implement the method according to any one of the first aspect.

[0059] The embodiments of the present application provide at least the following beneficial effects:

[0060] (1) By combining the time series feature extraction capability of the LSTM neural network and the progressive damage theory, the present application can dynamically capture the nonlinear damage behavior of the composite material under complex loading conditions, especially under complex working conditions such as multi-axial loading, and can more accurately predict the damage path and failure mode of the material, solving the problem that the traditional method is difficult to capture the nonlinear load effect.

[0061] (2) The present application combines the time series parameters and the non-time series parameters of the composite material by the memory capability of the LSTM for the time series data and the extraction capability of the DNN for the static features, fully excavates the time correlation and nonlinear relationship in the experimental data, and thus significantly improves the reliability and consistency of the prediction results.

[0062] (3) In addition, the present application combines the dynamic updating mechanism of the LSTM model, can adapt to various loading paths and complex composite material mechanical environment, and significantly improves the flexibility, meets the needs of diversified working conditions in engineering practice. At the same time, by introducing the physical law constraint based on the progressive damage theory in the loss function, the present application ensures that the prediction results not only conform to the data law, but also are consistent with the material mechanics theory, further enhancing the credibility of the model.

[0063] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter in the description of the application. BRIEF DESCRIPTION OF DRAWINGS

[0064] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings.

[0065] Figure 1 A flowchart of a composite material progressive damage prediction method based on LSTM-DNN provided by the embodiments of the present application;

[0066] Figure 2A network schematic diagram of a series connection model combining an LSTM network and a DNN network provided by an embodiment of the present application;

[0067] Figure 3 A structural schematic diagram of an LSTM unit provided by an embodiment of the present application;

[0068] Figure 4 A schematic diagram of displacement and external force load changes in a progressive damage process provided by an embodiment of the present application;

[0069] Figure 5 A schematic diagram of displacement and damage factor changes in a progressive damage process provided by an embodiment of the present application;

[0070] Figure 6 A structural schematic diagram of a composite material progressive damage prediction device based on LSTM-DNN provided by an embodiment of the present application. DETAILED DESCRIPTION

[0071] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0072] The present application aims to solve the problems of low calculation efficiency, insufficient prediction accuracy and poor data correlation in existing composite material progressive damage prediction, and proposes a composite material progressive damage prediction method based on a long short term memory network (LSTM).

[0073] As a time series modeling tool, LSTM can remember load history characteristics and dynamically update states, and has shown significant advantages in dealing with composite material nonlinear damage evolution problems. This method combines LSTM with progressive fracture theory, comprehensively evaluates the damage and evolution process of the material according to the composition and structure of the composite material, accurately predicts its failure mode, and further completely characterizes the damage mechanism of the composite material, thereby providing a new solution for modeling the damage behavior of the composite material.

[0074] In addition, the present application combines LSTM with a deep-learning neural network (DNN) to realize the organic integration of time series parameters and non-time series parameters. LSTM network is good at processing time series data, while DNN network can extract static features and realize nonlinear relationship modeling. This combination not only can significantly improve the prediction ability of the model under complex loading conditions, but also can ensure the efficiency, stability and accuracy of the neural network.

[0075] Figure 1 A flowchart of a composite material progressive failure prediction method based on LSTM-DNN provided by an embodiment of the present application is shown in FIG. 1. As shown in the figure, the method comprises the following steps: Figure 1

[0076] Step 101, according to the geometric parameters, material parameters and load conditions of the composite material, determine the parameter range and distribution required for modeling, and generate a parameter distribution set by using the Monte Carlo method.

[0077] In the embodiment of the present application, the geometric parameters of the composite material include the fiber layer angle and the thickness of the composite material , which are used to describe the structural characteristics of the composite material; the material parameters include the fiber elastic modulus , the Poisson's ratio , the base yield strength , the interlaminar bonding strength , which are used to characterize the mechanical properties of the composite material; the load conditions include the load history and the strain rate , which are used to simulate the stress condition and deformation speed of the composite material in the actual working environment.

[0078] In order to generate a combination of modeling parameters with random distribution, the Monte Carlo method is used in the embodiment of the present application to meet the distribution range requirements of the parameters in actual engineering applications. The specific implementation is as follows:

[0079] (1) The fiber layer angle is randomly selected within the range of 0° to 90° according to a certain probability distribution, which is used to reflect the performance change of the composite material under different layer design;

[0080] (2) The thickness of the composite material randomly fluctuates within the range allowed by engineering, which is used to simulate the influence of different structural thicknesses on the mechanical behavior;

[0081] (3) The elastic modulus, Poisson's ratio, yield strength and bonding strength are randomly generated according to the statistical performance law of the composite material, fully reflecting the diversity of the material properties of the composite material.

[0082] For the above six key parameters of the fiber layer angle, the material thickness, the fiber elastic modulus, the Poisson's ratio, the load path and the strain rate, the Monte Carlo method is used to generate the parameter distribution set in the embodiment of the present application. During the generation process, the actual application value range and probability distribution of each parameter are fully considered, and 3000 groups of examples are constructed based on these parameter combinations.

[0083] ​Through this step, the application embodiment effectively generates a diversified and engineering-representative parameter distribution set, providing high-quality training data support for subsequent composite material damage prediction models. By comprehensively considering the diversity of parameters and their distribution characteristics, the generated parameter distribution set not only can truly reflect the stress behavior of the composite material, but also can provide solid foundation guarantee for damage prediction under complex working conditions.

[0084] Step 102, using Python script to develop ABAQUS, inputting the parameter distribution set into ABAQUS for modeling calculation, simulating the stress-strain curve and damage-damage evolution process of the composite material under different parameter conditions, and extracting the time series data of the composite material.

[0085] The application embodiment develops ABAQUS by using Python script, and inputs the parameter distribution set generated in step 101 into ABAQUS for modeling calculation. This process simulates the stress-strain curve and damage-damage evolution process of the composite material under different parameter conditions, and extracts the time series data of the composite material, providing comprehensive input data for subsequent LSTM modeling.

[0086] Specifically, first, a Python script is written to automatically generate the input files (inp files) required by ABAQUS according to the parameter distribution set generated in step 101. The input file contains key modeling information such as the geometric parameters, material parameters and load conditions of the composite material. Through the automatic processing of the script, the workload of manually generating the input file can be greatly reduced, while ensuring the accuracy and consistency of the input parameters.

[0087] Second, the Python script calls the ABAQUS background program to realize automatic modeling and calculation based on the multi-threading mechanism. Multi-threading calculation can significantly improve the running efficiency of ABAQUS simulation, so that the modeling and solving process of large-scale parameter combinations can be completed in a short time. This efficient calculation method is especially suitable for handling the multi-parameter and multi-condition simulation requirements involved in the application, providing strong support for large-scale data generation.

[0088] After the modeling and calculation are completed, ABAQUS simulation outputs the stress-strain curve and damage factor evolution process of the composite material under different parameter conditions. In order to comprehensively describe the damage behavior of the composite material, the application embodiment extracts the key time series data , including displacement array , load history , deformation rate , damage factor These time series data cover the mechanical state, deformation behavior, and damage evolution characteristics of composite materials during the loading process, and can fully reflect the stress process and failure mechanism of the materials.

[0089] Through the above steps, the application embodiment successfully generates composite material time series data sets covering multiple parameter conditions. These data not only multi-dimensionally represent the mechanical properties and damage evolution law of the composite material, but also provide a solid data foundation for subsequent deep learning model training and prediction. At the same time, through the automatic processing of Python scripts and the multi-threaded computer mechanism, the application embodiment greatly improves the efficiency of modeling and calculation, and exhibits the ability to handle large-scale simulation tasks.

[0090] Step 103, a series connection model combining LSTM network and DNN network is constructed, wherein the LSTM network is used to process the time series matrix and extract the time series features of the composite material, and the DNN network is used to extract the features of the structural parameters and material strength parameters and fuse them with the time series features to output the prediction results of the stress sequence and the damage factor sequence.

[0091] The application embodiment realizes accurate prediction of the stress sequence and the damage factor sequence of the composite material by constructing a series connection model combining LSTM network and DNN network. The LSTM network is used to process the time series data matrix and extract the time series features of the composite material; the DNN network is used to extract the features of the static structural parameters and material strength parameters of the composite material, and fuse them with the time series features, and finally output the predicted stress sequence and damage factor sequence.

[0092] Figure 2 The network schematic diagram of the series connection model combining LSTM network and DNN network provided by the application embodiment is shown. Specifically, the input of the model includes preprocessed time series data, structural parameters and material strength parameters, and the input variables are as follows:

[0093] : structural parameters, including fiber layer angle and composite material thickness;

[0094] : material strength parameters, including elastic modulus, Poisson's ratio, etc.

[0095] : time series data, including displacement array , load history , deformation rate , damage factor .

[0096] The output variables of the model are:

[0097] : a sequence of external forces (stresses) representing the response of the composite material to the external forces during the step-by-step loading process;

[0098] : a sequence of damage factors representing the evolution of the damage of the composite material during the step-by-step loading process.

[0099] In the embodiments of the present application, the LSTM-DNN series connection model adopts three layers of LSTM units, each layer containing 128 neurons, and a Dropout layer is added between each layer to prevent overfitting. At the same time, the LSTM network adopts a multi-layer stacking structure, each layer containing 64 units, which can capture the complex dynamic evolution characteristics in the time series data.

[0100] Figure 3 The structure diagram of the LSTM unit provided in the embodiments of the present application is as follows, and the single-step formula of the LSTM unit is as follows:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106] In the formula, , , respectively represent the activation values of the forget gate, the input gate and the output gate, is the cell state, is the hidden layer state.

[0107] Finally, the time series features extracted by the LSTM are fused with the static features extracted by the DNN and , and input into a new DNN for further processing, and finally output the predicted stress sequence and the damage factor sequence .

[0108] In addition, before the model is constructed, the input data is preprocessed in the embodiments of the present application to ensure that the data features can be efficiently extracted by the network. The preprocessing process includes the following steps:

[0109] Firstly, the Kalman filtering method is used to perform noise reduction processing on the time series data to remove measurement noise and environmental interference, so as to improve the quality and reliability of the data. After the noise reduction processing, the time series data transformed into a standard time series matrix wherein, is the number of time steps.

[0110] Then, the transformed standard time series matrix and static structure parameters and material strength parameters are normalized, and the normalization formula is:

[0111]

[0112] wherein, is the normalized value, and are the minimum and maximum values of the input variable, respectively. Through normalization, the influence between different data dimensions and units can be eliminated, and a unified input data format is provided for network training.

[0113] In step 104, the preprocessed time series data, static structure parameters and material strength parameters are used as input variables, and a series connection model is trained based on a back propagation algorithm, a loss function is used to optimize model parameters, a learning rate is dynamically adjusted and a regularization technique is used to control the overfitting problem of the model, and a final prediction model is generated after training.

[0114] To ensure the efficiency of model training and the accuracy of prediction results, the embodiments of the present application first divide the data set reasonably. Specifically, the data set after sorting is divided into a training set, a validation set and a test set, with a ratio of 70%, 20% and 10% respectively. In the data division process, a random sampling method is used, but the distribution ratio of each type of damage mode in the three subsets is ensured to be similar, so as to avoid the problem of insufficient model generalization ability caused by uneven data distribution. Through this strategy, the prediction ability of the model under different working conditions can be effectively improved.

[0115] In the model training stage, the embodiments of the present application use configuration parameters such as batch size and training iteration number , so that the model can fully learn the feature information in the data. To further improve the training efficiency and optimization effect, the present application selects the Adam optimizer as the optimizer, sets the initial learning rate to 0.001, and dynamically adjusts the learning rate according to the loss value of the validation set to avoid the problem of gradient disappearance or gradient explosion. At the same time, by recording the loss value and accuracy of the training set and the validation set, and drawing the learning curve, the performance change of the model can be intuitively displayed. This way not only helps to optimize the training process, but also quickly locates potential problems in training.

[0116] To evaluate the network performance and avoid overfitting, the embodiments of the present application use the validation set to evaluate the loss function and accuracy of the model after the end of each training cycle. If the loss value of the validation set does not decrease for several cycles or the accuracy does not improve, the training is terminated in advance. This early stopping strategy can effectively prevent the occurrence of model overfitting problem, improve the training efficiency while ensuring the generalization ability of the model.

[0117] In addition, the embodiments of the present application further control the model overfitting phenomenon by combining L2 regularization technology and Dropout layer. L2 regularization applies constraints to model parameters by adding a weight decay term to the loss function, thereby avoiding overfitting caused by excessively large parameters. And the Dropout layer reduces the dependence between neurons by randomly discarding part of the activation values of neurons, thereby enhancing the robustness of the model from the network structure.

[0118] Before training, the embodiments of the present application have carried out detailed arrangement and classification on the input data, to ensure that the model can accurately capture the damage evolution characteristics of the composite material under different working conditions. Specifically, the following aspects are included:

[0119] (1) Classify and label the stress distribution data, and perform detailed processing according to different stress concentration regions, stress directions and other characteristics, so as to better identify and extract stress features during model training;

[0120] (2) Normalize the damage factor evolution time series data to unify the data scale and eliminate the influence of data dimension, thereby providing support for feature extraction of deep learning model;

[0121] (3) Accurately record the damage mode according to the macroscopic damage performance of the composite material (such as crack propagation path, fracture morphology, deformation characteristics, etc.). These records provide rich label information for model training, and also provide an important basis for subsequent prediction result evaluation.

[0122] In the training stage, the embodiments of the present application optimize the parameters of the series connection model based on the back propagation algorithm. The model loss function includes the loss function from the stress prediction sequence and the loss function from the damage factor sequence prediction, wherein the loss function of the damage factor comes from two parts, data error and theoretical error. That is, the loss function of the series connection model is:

[0123]

[0124] In the formula, , are the weight coefficients of different loss terms and , is the damage evolution constraint based on physical law, and are the predicted stress sequence and damage factor sequence respectively, and are the simulated stress sequence and damage factor sequence respectively. By comprehensively considering the error terms of the three parts, the final loss function can effectively guide the training process of the model, so that it not only meets the data distribution characteristics, but also meets the requirements of physical constraints.

[0125] For the simulated damage factor in the loss function, the embodiments of the application use the damage model weighted by the separation of fibers and matrix to calculate the loss factor of the composite material , the formula is:

[0126]

[0127] In the formula, and are the weight coefficients of fiber damage and matrix damage, and , is the fiber damage variable, is the matrix damage variable.

[0128] The calculation formula of the fiber damage variable is:

[0129]

[0130] In the formula, is the fiber damage critical variable, the maximum bearing capacity of the fiber;

[0131] The calculation formula of the matrix damage variable is:

[0132]

[0133] In the formula, is the fracture strain of the matrix.

[0134] After training, the final prediction model is generated by the embodiments. The model can accurately predict the stress response and damage evolution of the composite material in the gradual loading process, and provides efficient and reliable technical support for the composite material structure design and performance evaluation in engineering application. The accuracy of the model prediction result can be referred to Figure 4 and Figure 5 .

[0135] Step 105, based on the trained model, input new displacement and load data into the model to generate stress sequence and damage factor sequence of the composite material, and judge the progressive failure state of the composite material structure through the damage factor to predict the progressive failure state and the remaining life of the composite material.

[0136] The embodiment of the application inputs new displacement and load data into the trained prediction model to generate a stress sequence and a damage factor sequence of the composite material, and further realizes accurate prediction of the progressive damage state and the remaining life of the composite material.

[0137] Specifically, the new input data includes displacement data and load history of the composite material in actual working conditions. After the data is preprocessed (such as normalization processing) in the same way as the training data, it is input into the trained series connection model. The model generates a stress sequence and a damage factor sequence of the composite material in the loading process through the learned time series feature and static feature fusion capability.

[0138] It can be understood that the generated damage factor sequence is an important basis for judging the progressive damage state of the composite material structure. By analyzing the change trend of the damage factor, the damage accumulation speed, the local damage distribution and the overall damage degree of the composite material in the loading process can be clearly identified. In combination with the progressive damage theory, the critical value of the damage factor is further used to judge the failure state of the composite material. When the damage factor reaches the set critical value, it can be determined that the composite material structure has entered an irreversible damage stage.

[0139] In addition, through joint analysis of the stress sequence and the damage factor sequence , the embodiment can also predict the remaining life of the composite material. Specifically, by combining the damage factor growth curve generated by the model with the loading conditions and working condition characteristics, the remaining time required for the composite material to go from the current state to complete failure can be accurately calculated, thereby providing an important reference basis for structural design, life evaluation and safety monitoring in engineering practice.

[0140] Through this step, the embodiment of the application realizes a complete prediction path from stress response to damage evaluation, and provides a systematic solution for modeling the progressive damage of the composite material. The prediction result can not only accurately represent the mechanical behavior and damage evolution law of the composite material in the loading process, but also can provide data support and theoretical guidance for structural design optimization and material performance improvement.

[0141] In order to realize the above-mentioned embodiment, the application further provides a composite material progressive damage prediction device based on LSTM-DNN. Figure 6 A structural schematic diagram of a composite material progressive damage prediction device based on LSTM-DNN provided by the embodiment of the application. As Figure 6 shown, the device comprises:

[0142] The parameter distribution set generation module 100 is configured to determine a parameter range and distribution required for modeling according to geometric parameters, material parameters and load conditions of the composite material, and generate a parameter distribution set by using a Monte Carlo method.

[0143] The extraction module 200 is configured to use a Python script to perform secondary development on ABAQUS, input the parameter distribution set into ABAQUS for modeling and calculation, simulate stress-strain curves and damage and failure evolution processes of the composite material under different parameter conditions, and extract time series data of the composite material.

[0144] The model construction module 300 is configured to construct a series connection model combining an LSTM network and a DNN network, wherein the LSTM network is configured to process a time series matrix and extract time series features of the composite material, and the DNN network is configured to extract features of structural parameters and material strength parameters and fuse the features with the time series features, and output stress sequence and damage factor sequence prediction results.

[0145] The model training module 400 is configured to use preprocessed time series data, static structural parameters and material strength parameters as input variables, train the series connection model based on a back propagation algorithm, optimize model parameters by using a loss function, control overfitting of the model by using a dynamic adjustment learning rate and a regularization technique, and generate a final prediction model after training.

[0146] The prediction module 500 is configured to input new displacement and load data into the model based on the trained model, generate stress sequence and damage factor sequence of the composite material, and judge a progressive failure state of the composite material structure by using the damage factor, and predict the progressive failure state and residual life of the composite material.

[0147] In order to achieve the above-mentioned embodiments, the present application further provides an electronic device, comprising a processor and a memory connected with the processor; the memory stores computer execution instructions; and the processor executes the computer execution instructions stored in the memory to realize the method provided by the foregoing embodiments.

[0148] In order to achieve the above-mentioned embodiments, the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to realize the method provided by the foregoing embodiments.

[0149] In order to achieve the above-mentioned embodiments, the present application further provides a computer program product, comprising a computer program, and the computer program is executed by a processor to realize the method provided by the foregoing embodiments.

[0150] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the present application comply with relevant laws and regulations and do not violate public order and good customs.

[0151] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and not shared or sold outside these legitimate uses. In addition, such collection / sharing should be carried out after the user's informed consent is received, including but not limited to informing the user to read the user agreement / user notice before the user uses the function, and signing the agreement / authorization including authorization of relevant user information. In addition, any necessary steps should be taken to protect and secure access to such personal information data and ensure that other people with access to personal information data comply with their privacy policies and processes.

[0152] The present application contemplates providing an implementation in which users can selectively block the use of, or access to, personal information data. That is, the disclosure contemplates providing users with control to prevent or limit others from accessing or using personal information data. Once personal information data is no longer needed, it can be deleted or rendered unreadable, e.g., by limiting access to the data and / or by blocking or deleting the data. Additionally, in some examples, such personal information data is removed from the user prior to collection.

[0153] In the foregoing various embodiment descriptions, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples, without contradiction.

[0154] In addition, the terms "first", "second", etc. are used only for the purpose of description and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "plurality" is at least two, for example, two, three, etc., unless otherwise explicitly specified.

[0155] Any processes or methods described in the flowcharts or otherwise described herein can be understood as representing modules, segments, or portions of code that include one or more executable instructions for implementing specific logical functions or steps, and the various embodiments of the application can include additional or fewer steps performing the same or equivalent functions as those shown or discussed, in different orders, including substantially simultaneous execution of the functions described with respect to particular steps, and the like.

[0156] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing the logic function, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can specifically include the following, which are non-exhaustive list: electrical connection (electrical device having one or more wires), portable computer diskette (magnetic device), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fiber device, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable medium upon which the program is printed, as the program can be electronically captured, for example, via the optical scanner of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0157] It should be understood that portions of the application can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As such, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0158] Those skilled in the art can understand that all or part of the steps of the method carried out by the above-mentioned embodiments can be instructed by a program to the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0159] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0160] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

[0161] It should be understood that the steps can be reordered, added or deleted using the various forms of flow shown above. For example, each step described in the present application can be executed in parallel, sequentially or in different order, as long as the desired results of the technical solution of the present application can be achieved, and this is not limited herein.

[0162] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and replacements can be made according to design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A progressive failure prediction method for composite materials based on LSTM-DNN, characterized in that, Includes the following steps: Based on the geometric parameters, material parameters, and load conditions of the composite material, the range and distribution of parameters required for modeling are determined, and the Monte Carlo method is used to generate the parameter distribution set. The parameter distribution set was input into ABAQUS for modeling and calculation using Python scripts to simulate the stress-strain curves and damage-failure evolution process of composite materials under different parameter conditions, and to extract the time series data of composite materials. A cascaded model combining LSTM and DNN networks is constructed. The LSTM network is used to process the time series matrix and extract the time series features of the composite material, while the DNN network is used to extract the features of structural parameters and material strength parameters and fuse them with the time series features to output the predicted results of stress sequence and damage factor sequence. The preprocessed time series data, static structural parameters, and material strength parameters are used as input variables. The cascaded model is trained based on the backpropagation algorithm. The model parameters are optimized with a loss function. The overfitting problem of the model is controlled by dynamically adjusting the learning rate and regularization techniques. After training, the final prediction model is generated. Based on the trained model, new displacement and load data are input into the model to generate stress sequence and damage factor sequence of composite material. The progressive failure state of composite material structure is determined by the damage factor, and the progressive failure state and remaining life of composite material are predicted. The loss function of the concatenated model is: In the formula, , The weighting coefficients for different loss terms and , It is a damage evolution constraint based on physical laws. and These are the predicted stress sequence and damage factor sequence, respectively. and These are the simulated stress sequence and damage factor sequence, respectively. This represents the number of time steps.

2. The method according to claim 1, characterized in that, The geometric parameters of the composite material include fiber layup angles. and composite material thickness The material parameters of the composite material include fiber elastic modulus. Poisson's ratio Base yield strength Interlayer bonding strength The load conditions of the composite material include load history. and strain rate .

3. The method according to claim 2, characterized in that, The Monte Carlo method is used to generate a combination of modeling parameters with a random distribution that conforms to the parameter range in practical engineering applications, and is generated in the following way: The fiber layup angle is randomly distributed within the range of 0° to 90°; The thickness of the composite material fluctuates randomly within the allowable range of engineering requirements; The elastic modulus, Poisson's ratio, yield strength, and bond strength are randomly distributed according to the statistical laws of material properties.

4. The method according to claim 3, characterized in that, The time series data Including displacement array Load history Deformation rate Damage factors .

5. The method according to claim 3, characterized in that, The preprocessing of the time series data, structural parameters, and material strength parameters includes: The time series data were processed using the Kalman filter method. Denoising is performed on the time series data, and the data after denoising is then processed. Transform into a standard time series matrix ; The transformed standard time series matrix, static structural parameters, and material strength parameters are normalized using the following formula: in, Normalized value and These are the minimum and maximum values ​​of the input variables, respectively.

6. The method according to claim 5, characterized in that, The cascaded model uses a three-layer LSTM unit, with a Dropout layer added between each layer to prevent overfitting. The single-step formula for each LSTM unit is as follows: In the formula, , , These represent the activation values ​​of the forget gate, input gate, and output gate, respectively. In cellular state, This is the hidden layer state.

7. The method according to claim 6, characterized in that, Also includes: The loss factor of the composite material was calculated using a fiber-matrix separation-weighted damage model. The formula is: In the formula, and The weighting coefficients for fiber damage and matrix damage are as follows: , For fiber damage variables, For matrix damage variables; Fiber damage variables The calculation formula is: In the formula, It is the critical variable for fiber damage. Maximum load-bearing capacity of the fiber; matrix damage variables The calculation formula is: In the formula, The fracture strain of the matrix is ​​denoted as .

8. A progressive failure prediction device for composite materials based on LSTM-DNN, characterized in that, include: The parameter distribution set generation module is used to determine the range and distribution of parameters required for modeling based on the geometric parameters, material parameters and load conditions of the composite material, and to generate the parameter distribution set using the Monte Carlo method. The extraction module is used to perform secondary development of ABAQUS using Python scripts. It inputs the parameter distribution set into ABAQUS for modeling and calculation, simulates the stress-strain curves and damage-failure evolution process of composite materials under different parameter conditions, and extracts the time series data of composite materials. The model construction module is used to construct a cascaded model combining an LSTM network and a DNN network. The LSTM network is used to process the time series matrix and extract the time series features of the composite material, while the DNN network is used to extract the features of structural parameters and material strength parameters and fuse them with the time series features to output the predicted results of stress sequence and damage factor sequence. The model training module is used to train the cascaded model based on the backpropagation algorithm, using preprocessed time series data, static structural parameters and material strength parameters as input variables. The model parameters are optimized with a loss function, and the overfitting problem of the model is controlled by dynamically adjusting the learning rate and regularization techniques. After training, the final prediction model is generated. The prediction module is used to input new displacement and load data into the model based on the trained model, generate stress sequence and damage factor sequence of composite material, and determine the progressive failure state of composite material structure through damage factor, and predict the progressive failure state and remaining life of composite material. The loss function of the concatenated model is: In the formula, , The weighting coefficients for different loss terms and , It is a damage evolution constraint based on physical laws. and These are the predicted stress sequence and damage factor sequence, respectively. and These are the simulated stress sequence and damage factor sequence, respectively. This represents the number of time steps.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-7.

Images