Additive manufacturing composite material high-temperature damage failure prediction method, system, equipment and medium
By constructing a graph neural network model based on experimental hysteresis data, the accuracy and efficiency problems of high-temperature damage and failure prediction of fiber-reinforced PEEK composites in traditional methods were solved, and rapid and accurate prediction of material life at high temperatures was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-04-03
AI Technical Summary
Traditional methods are difficult to accurately predict the damage and failure behavior of fiber-reinforced PEEK composites at high temperatures, especially due to the differences in stress response of specimens with different cross-sectional areas caused by differences in thermal expansion and interfacial bonding. Existing finite element analysis is computationally intensive and highly dependent on parameters.
A two-layer graph convolutional neural network regression model was constructed using experimental hysteresis data and graph neural network methods, through feature extraction and graph convolutional analysis, to predict high-temperature damage failure of fiber-reinforced PEEK composites.
It enables rapid and accurate prediction of the remaining life of fiber-reinforced PEEK composites at high temperatures, reducing modeling complexity and computational load, and improving prediction accuracy and generalization performance.
Smart Images
Figure CN121786599A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of composite material structural performance and intelligent life prediction technology, specifically to a method, system, equipment and medium for predicting high temperature damage and failure of additive manufacturing composite materials. Background Technology
[0002] Fiber-reinforced PEEK composites are widely used in aerospace, medical devices, and high-end structural components due to their high strength, high temperature resistance, and additive manufacturing capabilities. Under high-temperature cyclic loading, this material often exhibits significant hysteresis characteristics and damage accumulation behavior; the hysteresis curve can reflect the nonlinear deformation and fatigue degradation of the material. However, due to differences in thermal expansion and interfacial bonding between the fiber and the matrix, the stress response of specimens with different cross-sectional areas also varies significantly, making it difficult for traditional empirical life models to accurately predict their high-temperature failure behavior.
[0003] While existing finite element thermo-mechanical analysis or constitutive models can describe local responses, they are computationally intensive and highly parameter-dependent. Therefore, a high-temperature damage failure prediction method based on experimental hysteresis data and graph neural network feature extraction is needed to achieve efficient and accurate prediction of the remaining life of fiber-reinforced PEEK composites.
[0004] In recent years, deep learning methods, especially graph neural networks, have shown strong feature extraction capabilities in processing sequential or structured experimental curve data, providing a new modeling path for predicting the remaining life of composite materials.
[0005] Therefore, there is an urgent need for a new method that integrates experimental data with graph neural networks, using experimental hysteresis curves as input, to rapidly predict the damage or failure parameters of composite materials under a uniform high-temperature environment. Summary of the Invention
[0006] To address the shortcomings of existing technologies, the present invention aims to provide a method, system, equipment, and medium for predicting high-temperature damage and failure of additive manufacturing composite materials. By performing feature encoding and graph convolution analysis on the high-temperature hysteresis response of fiber-reinforced PEEK composite materials, the remaining life of the material can be predicted quickly and accurately, effectively reducing the amount of testing and modeling complexity.
[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0008] In a first aspect, the present invention proposes a method for predicting high-temperature damage failure of additive manufacturing composite materials, the key of which includes the following steps:
[0009] Step 1: Under uniform high temperature conditions, cyclic loading tests were conducted on fiber-reinforced PEEK composite material samples with different cross-sectional areas to obtain displacement-load hysteresis curves and their corresponding failure life data.
[0010] Step 2: Extract features and encode load positions from the hysteresis curve, and construct a node feature matrix;
[0011] Step 3: Treat the sampling points of the hysteresis curve as graph nodes, and construct a circular bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a circular graph structure dataset;
[0012] Step 4: Construct a two-layer graph convolutional neural network regression model based on the annular graph structure dataset, and input the node feature matrix for training;
[0013] Step 5: Use test samples to validate and evaluate the performance of the trained two-layer graph convolutional neural network regression model. If it meets the standard, proceed to step 6; otherwise, return to step 4 and train again.
[0014] Step 6: Use a high-performance two-layer graph convolutional neural network regression model to predict the remaining life of additive manufacturing composite materials.
[0015] Furthermore, the acquisition of the displacement-load hysteresis curve and its corresponding failure life data in step 1 includes:
[0016] Step 1.1: Under a constant high temperature of 150℃, multiple groups of fiber-reinforced PEEK composite material samples were subjected to displacement-controlled cyclic loading tests until the samples failed due to fatigue.
[0017] Step 1.2: Record the displacement-load hysteresis curve and the corresponding failure cycle number for each group of samples;
[0018] Step 1.3: Measure the width b and thickness h of the sample, and calculate the cross-sectional area A = b × h of the sample as geometric input parameters;
[0019] Step 1.4: Select typical hysteresis loops of each group of experiments under high temperature conditions as model input samples. The number of sampling points for each curve is consistent, forming a sample dataset.
[0020] Furthermore, in step 2, each set of data in the node feature matrix includes displacement, load, cross-sectional area, and load percentile.
[0021] Furthermore, step 2, which involves feature extraction and load location encoding of the hysteresis curve and construction of the node feature matrix, includes:
[0022] Step 2.1: Rearrange the starting points of the two-dimensional data of each hysteresis curve, using the minimum or maximum value in the load column as the starting point, so that the loading direction of the hysteresis curves is consistent.
[0023] Step 2.2: Calculate the percentile ranking of the load data and normalize the loading stage;
[0024] Step 2.3: Calculate the cross-sectional area of the sample based on its geometric parameters;
[0025] Step 2.4: Combine displacement, load, cross-sectional area and load percentile to form a nodal feature matrix, and standardize the data.
[0026] Furthermore, in step 3, the sampling points of the hysteresis curve are treated as graph nodes, and a circular bidirectional edge is constructed between adjacent graph nodes according to the experimental sampling order to generate a circular graph structure dataset, including:
[0027] Step 3.1: Treat the equally spaced sampling points of the hysteresis curve as graph nodes and obtain the dataset for each graph node;
[0028] Step 3.2: Create two edges in opposite directions for each pair of adjacent nodes, and establish connections between the first and last nodes according to the experimental sampling order to generate a ring graph structure dataset.
[0029] Furthermore, the two-layer graph convolutional neural network regression model described in step 4 includes:
[0030] The input section is used to receive the node feature matrix for each sample;
[0031] The two-layer graph convolution part is used to extract local and global features at different scales from the input node feature matrix;
[0032] The global average pooling part is used to compress the node feature matrix obtained after two convolutional layers into a graph-level vector through the global average pooling operation.
[0033] The output section is used to map graph-level vectors to lifetime predictions using a linear regression layer.
[0034] Furthermore, the two-layer graph convolutional portion includes:
[0035] The first convolutional layer is used to map the input node feature matrix to a 128-dimensional latent space and uses ReLU activation and Dropout regularization to prevent overfitting.
[0036] The second convolutional layer is used to aggregate the node neighborhood features output by the first convolutional layer in order to capture the nonlinear dependencies between adjacent stages in the hysteresis curve.
[0037] Secondly, this invention proposes a high-temperature damage failure prediction system for additive manufacturing composite materials, used to implement the method described in the first aspect, including:
[0038] The data acquisition module is used to conduct cyclic loading tests on fiber-reinforced PEEK composite material samples with different cross-sectional areas under uniform high temperature conditions, and to obtain the displacement-load hysteresis curves and their corresponding failure life data.
[0039] The feature encoding module is used to extract features and encode load positions on the hysteresis curve, and to construct a node feature matrix.
[0040] The graph structure construction module is used to treat the sampling points of the hysteresis curve as graph nodes and construct a ring-shaped bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a ring graph structure dataset.
[0041] The model building and training module is used to construct a two-layer graph convolutional neural network regression model based on the ring graph structure dataset, and to train it by inputting the node feature matrix;
[0042] The model evaluation module is used to validate and evaluate the performance of the trained two-layer graph convolutional neural network regression model using test samples.
[0043] The lifetime prediction module is used to predict the remaining lifetime of additive manufacturing composite materials using a performance-compliant two-layer graph convolutional neural network regression model.
[0044] Thirdly, the present invention highlights a computer device comprising a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the method as described in the first aspect.
[0045] Fourthly, the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, implements the steps of the method described in the first aspect.
[0046] The significant effects of this invention are:
[0047] This invention achieves rapid prediction of the remaining life of fiber-reinforced PEEK composites at high temperatures by integrating high-temperature hysteresis test data with graph convolutional neural networks. This method eliminates the need for complex simulations, effectively extracts hysteresis features, and considers the effects of cross-sectional area and loading stage. It boasts high prediction accuracy and strong generalization performance, significantly improving the efficiency and intelligence level of high-temperature composite material life assessment. Attached Figure Description
[0048] Figure 1 This is a flowchart of the method described in this invention;
[0049] Figure 2 This is a graph showing the relationship between the number of failure cycles and stress range of PEEK composite material under high temperature conditions according to the present invention;
[0050] Figure 3 The hysteresis curve of the fiber-reinforced PEEK composite material under high-temperature test conditions of the present invention;
[0051] Figure 4 This is a diagram of the graph convolution operator structure used in this invention;
[0052] Figure 5 This is a schematic diagram of the structure of the convolutional neural network in this invention;
[0053] Figure 6 This is a schematic diagram of the prediction results of the convolutional neural network in this invention;
[0054] Figure 7 This is a schematic diagram of the system described in this invention;
[0055] Figure 8 This is a schematic block diagram of the device described in this invention. Detailed Implementation
[0056] The specific embodiments and working principles of the present invention will be further described in detail below with reference to the accompanying drawings.
[0057] Example 1:
[0058] like Figure 1 As shown in the figure, this invention provides a method for predicting high-temperature damage failure of additive manufacturing composite materials, and the specific steps are as follows:
[0059] Step 1: Under uniform high temperature conditions, cyclic loading tests were conducted on fiber-reinforced PEEK composite material samples with different cross-sectional areas to obtain displacement-load hysteresis curves and their corresponding failure life data.
[0060] In some embodiments, the fiber-reinforced PEEK displacement-load hysteresis curve and its corresponding failure life data are obtained through the following steps:
[0061] Step 1.1: Under constant high temperature of 150℃, multiple groups of fiber-reinforced PEEK composite material samples were subjected to displacement-controlled cyclic loading tests. During the test, the loading rate and temperature were kept stable to avoid the influence of temperature drift on the hysteresis mode and life calibration until the samples failed due to fatigue.
[0062] Step 1.2: Record the displacement-load hysteresis curve and the corresponding failure cycle number for each group of samples; that is, record the complete cycle number for each sample during the test, and finally obtain its failure cycle number. As a baseline lifespan label. For example... Figure 2 As shown, the relationship between the number of failure cycles and the stress range of PEEK composites under high temperature conditions is illustrated.
[0063] Step 1.3: Before testing, measure the width b and thickness h of the sample using calipers / microscopy, and calculate the cross-sectional area A = b × h of the sample as the geometric input parameter for the subsequent model. For automatic reading and traceability in the future, it is recommended to use "Dataset A" as the directory name, where A is the numerical cross-sectional area.
[0064] Displacement is synchronously acquired at a uniform sampling rate in each loop. With load This forms the initial hysteresis loop data. For ease of subsequent learning, it is recommended that each hysteresis loop be presented in a separate table with two columns: [displacement, load]. The code defaults to reading the first two columns of Excel data in this format. data;
[0065] Step 1.4: Select typical hysteresis loops from each group of experiments under high temperature conditions as model input samples, and ensure that the sampling points N = 500 points to ensure that the input dimensions of different samples are consistent, thus forming a sample dataset.
[0066] The code has enforced a 500-point constraint: samples with fewer than 500 points are skipped to ensure data consistency. Figure 3 As shown, the hysteresis curve of a sample of fiber-reinforced PEEK composite material under high-temperature test conditions is displayed.
[0067] The dataset constructed in step 1.4, where the data obtained from each sample in the experiment is the number of cycles, must be subtracted from the number of failed cycles obtained in step 1.2 to obtain the remaining lifetime of the dataset. That is, for each selected hysteresis loop, its current cycle number is recorded. If predicting total lifespan, with As a label; if the remaining lifespan is predicted, As a label.
[0068] Step 2: Extract features and encode load positions from the hysteresis curve, and construct a node feature matrix;
[0069] During implementation, the loading stage process should be characterized by load percentile normalization, and displacement, load and cross-sectional area should be combined to form standardized four-dimensional input features.
[0070] That is, in some specific implementations, step 2, which involves sampling, standardizing, and percentile encoding the hysteresis curve, includes the following steps:
[0071] Step 2.1: Rearrange the starting points of the two-dimensional data of each hysteresis curve, using the minimum or maximum value in the load column as the starting point, so that the loading direction of the hysteresis curves is consistent.
[0072] In practice, during the feature extraction and load location encoding stages, the two-dimensional data of each hysteresis curve are first processed. (representing the first) The starting points of the sampled points (displacement and load) are rearranged, using the minimum or maximum value in the load column as the starting point, so that the loading direction of the curves is consistent. This is denoted as:
[0073]
[0074] in, or .
[0075] Step 2.2: Load data To calculate percentile rankings, the loading phase is normalized to the interval [0,1]:
[0076]
[0077] in, This indicates the percentile code for the load. This represents the number of sampling points.
[0078] Step 2.3: Calculate the cross-sectional area of the sample based on its geometric parameters.
[0079]
[0080] And add a constant column for each sampling point. .
[0081] Step 2.4: Combine displacement, load, cross-sectional area, and load percentile to form a four-dimensional nodal feature matrix:
[0082]
[0083] Perform mean-variance standardization on each feature:
[0084]
[0085] in, and These are the mean and standard deviation of the features in the training set, respectively.
[0086] After standardization, each hysteresis curve sample is transformed into a node feature matrix with consistent structure and clear physical meaning, providing standardized input for subsequent loop graph construction and GCN learning.
[0087] Step 3: Treat the sampling points of the hysteresis curve as graph nodes, and construct a circular bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a circular graph structure dataset;
[0088] It should be noted that when constructing the loop diagram in this step, the node order should be consistent with the experimental sampling order, and bidirectional connecting edges should be established between adjacent sampling points to form a loop topology to ensure the sequential continuity of the hysteresis curves and the integrity of information propagation. If it is necessary to strengthen the mechanical connection between non-adjacent stages, skip edges can be appropriately introduced to expand the feature receptive field. Therefore:
[0089] In some implementations, the specific methods for generating the annular graph structure dataset that can be used for graph convolution learning in step 3 include:
[0090] Step 3.1: In the graph structure construction stage, firstly, the equally spaced sampling points of the hysteresis curve are regarded as nodes, and the dataset of each graph node is obtained;
[0091] The implementation process can be summarized as follows: each node represents the mechanical state at a time step, consisting of four characteristics: displacement. Load Cross-sectional area and load percentile Thus, a sample can be represented as a set of nodes:
[0092]
[0093] Step 3.2: To reflect the sequential continuity of the hysteresis curve, bidirectional adjacency connections are established between nodes. Specifically, for each pair of adjacent nodes... and Establish two edges in opposite directions. and And establish connections between the first and last nodes according to the experimental sampling order. and This creates a circular bidirectional graph structure. This structure ensures that the model can propagate information throughout the entire loading-unloading cycle, fully preserving the periodic characteristics of the hysteresis curve.
[0094] Therefore, the entire ring graph structure dataset can be represented by a set of adjacency relations. Represented as:
[0095]
[0096] in, The node feature matrix and the edge set are given. It describes the physical proximity relationships in the loading process.
[0097] Step 4: Construct a two-layer graph convolutional neural network regression model based on the ring graph structure dataset, and train it by inputting the node feature matrix to obtain a mapping model that can predict the remaining lifetime of the sample.
[0098] In this embodiment, as Figure 4As shown, the graph convolution operator structure used in this invention has C input channels and F output features. The graph structure is shared between layers, and Yi represents the label.
[0099] For example Figure 5 As shown, the two-layer graph convolutional neural network regression model adopts a two-layer GCN convolutional structure. It extracts hysteresis features at different scales through local aggregation and global pooling. The activation function is ReLU combined with Dropout regularization to prevent overfitting. The optimizer can be Adam, and the learning rate is controlled at around 1e-3. An early stopping mechanism on the validation set is set during training to ensure model convergence and stability.
[0100] That is, the two-layer graph convolutional neural network regression model includes:
[0101] The input section is used to receive the node feature matrix for each sample;
[0102] The two-layer graph convolution part is used to extract local and global features at different scales from the input node feature matrix;
[0103] The global average pooling part is used to compress the node feature matrix obtained after two convolutional layers into a graph-level vector through the global average pooling operation.
[0104] The output section is used to map graph-level vectors to lifetime predictions using a linear regression layer.
[0105] Preferably, the two-layer graph convolutional portion includes:
[0106] The first convolutional layer is used to map the input node feature matrix to a 128-dimensional latent space and uses ReLU activation and Dropout regularization to prevent overfitting.
[0107] The second convolutional layer is used to aggregate the node neighborhood features output by the first convolutional layer in order to capture the nonlinear dependencies between adjacent stages in the hysteresis curve.
[0108] Based on the above structure, the overall structure of the two-layer graph convolutional neural network regression model is "input + two layers of GCN and global pooling + linear regression output", used to map high-temperature hysteresis data to remaining lifetime. Specifically, the input part receives the node feature matrix of each sample. The input includes displacement, load, cross-sectional area, and load percentile. Subsequently, two layers of graph convolution are used to extract local and global features at different scales: the first layer maps the input to a 128-dimensional latent space and uses ReLU activation and Dropout to prevent overfitting; the second layer further aggregates the node neighborhood features to capture the nonlinear dependencies between adjacent stages in the hysteresis curve. The node feature matrix obtained after the two convolution layers is... Compressed into graph-level vectors using global average pooling. This represents the comprehensive characteristics of the entire hysteresis loop. Finally, a linear regression layer is used to map these 128-dimensional features into lifetime predictions. .
[0109] During the training phase, the mean squared error loss function is used, combined with the Adam optimizer and early stopping mechanism to ensure that the model converges stably with a limited number of samples.
[0110] The model has a simple structure and a moderate number of parameters, which can effectively capture the cyclic degradation law of the hysteresis curve of fiber-reinforced PEEK composite materials and achieve high-precision prediction of the remaining life at high temperature.
[0111] Step 5: Use test samples to perform lifetime prediction and visualization evaluation on the trained two-layer graph convolutional neural network (GCN) regression model to verify the model's accuracy and generalization ability. If the performance meets the standard, proceed to step 6; otherwise, return to step 4 to train again until the standard is met.
[0112] In some implementations, the specific process of step 5, which involves lifetime prediction and visualization evaluation of the test samples, is as follows:
[0113] After model training, a comprehensive performance evaluation and practical application verification are performed to ensure that the prediction results meet the requirements in terms of accuracy and physical consistency. First, the model is used to perform inference predictions using test set samples. The hysteresis curve of each sample is input into the trained graph convolutional network to obtain the predicted lifetime value. Subsequently, the predicted results were compared with the experimentally measured lifetime. Compare and calculate evaluation indicators.
[0114] In the specific implementation process, the model performance is comprehensively evaluated by indicators such as mean absolute error (MAE), mean square error (MSE), and coefficient of determination (R²), which quantitatively measures the prediction accuracy and fitting degree. Relevant charts are used to verify the reliability and physical consistency of the prediction. At the same time, generalization verification is carried out on samples with different cross-sectional areas to ensure that the model can accurately capture the coupled effects of geometric differences and loading stages on lifespan.
[0115] To further verify the reliability of the model, the output prediction results can be plotted as follows: Figure 6 The scatter plot shows the comparison between the predicted and actual values.
[0116] Step 6: Input the displacement-load hysteresis curve of the additive manufacturing composite material to be predicted in real time, and use a performance-compliant two-layer graph convolutional neural network regression model to predict its remaining life.
[0117] Example 2:
[0118] like Figure 7 As shown, this embodiment of the invention provides a high-temperature damage failure prediction system for additive manufacturing composite materials, comprising:
[0119] The data acquisition module is used to conduct cyclic loading tests on fiber-reinforced PEEK composite material samples with different cross-sectional areas under uniform high temperature conditions, and to obtain the displacement-load hysteresis curves and their corresponding failure life data.
[0120] The feature encoding module is used to extract features and encode load positions on the hysteresis curve, and to construct a node feature matrix.
[0121] The graph structure construction module is used to treat the sampling points of the hysteresis curve as graph nodes and construct a ring-shaped bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a ring graph structure dataset.
[0122] The model building and training module is used to construct a two-layer graph convolutional neural network regression model based on the ring graph structure dataset, and to train it by inputting the node feature matrix;
[0123] The model evaluation module is used to validate and evaluate the performance of the trained two-layer graph convolutional neural network regression model using test samples.
[0124] The lifetime prediction module is used to predict the remaining lifetime of additive manufacturing composite materials using a performance-compliant two-layer graph convolutional neural network regression model.
[0125] Example 3:
[0126] like Figure 8 As shown, an embodiment of the present invention provides a computer device, including a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, it implements the steps of the method described in Embodiment 1.
[0127] Example 4:
[0128] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.
[0129] In summary, this invention proposes a method for predicting high-temperature damage and failure of additive manufacturing composite materials based on high-temperature hysteresis test data and graph convolutional neural networks. This method fully integrates experimental physical laws with data-driven modeling principles. Through hysteresis curve resampling, load percentile encoding, and ring graph structure construction, it achieves a structured expression of damage characteristics of fiber-reinforced PEEK composite materials under high-temperature service conditions. Furthermore, a two-layer GCN regression model is used to establish a nonlinear mapping relationship between hysteresis characteristics and service life. Through the above modeling and learning methods, without complex simulations, it can efficiently and accurately predict the remaining service life of fiber-reinforced composite materials under high-temperature environments. The simulation results are highly consistent with the experimental data, significantly improving the efficiency and intelligence level of high-temperature composite material service life assessment.
[0130] The technical solution provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make several improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention.
Claims
1. A method for predicting high-temperature damage failure of additive manufacturing composite materials, characterized in that, Includes the following steps: Step 1: Under uniform high temperature conditions, cyclic loading tests were conducted on fiber-reinforced PEEK composite material samples with different cross-sectional areas to obtain displacement-load hysteresis curves and their corresponding failure life data. Step 2: Extract features and encode load positions from the hysteresis curve, and construct a node feature matrix; Step 3: Treat the sampling points of the hysteresis curve as graph nodes, and construct a circular bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a circular graph structure dataset; Step 4: Construct a two-layer graph convolutional neural network regression model based on the annular graph structure dataset, and input the node feature matrix for training; Step 5: Use test samples to validate and evaluate the performance of the trained two-layer graph convolutional neural network regression model. If it meets the standard, proceed to step 6; otherwise, return to step 4 and train again. Step 6: Use a high-performance two-layer graph convolutional neural network regression model to predict the remaining life of additive manufacturing composite materials.
2. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 1, characterized in that, Step 1, which involves obtaining the displacement-load hysteresis curve and its corresponding failure life data, includes: Step 1.1: Under a constant high temperature of 150℃, multiple groups of fiber-reinforced PEEK composite material samples were subjected to displacement-controlled cyclic loading tests until the samples failed due to fatigue. Step 1.2: Record the displacement-load hysteresis curve and the corresponding failure cycle number for each group of samples; Step 1.3: Measure the width b and thickness h of the sample, and calculate the cross-sectional area A = b × h of the sample as geometric input parameters; Step 1.4: Select typical hysteresis loops of each group of experiments under high temperature conditions as model input samples. The number of sampling points for each curve is consistent, forming a sample dataset.
3. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 1, characterized in that, In step 2, each set of data in the node feature matrix includes displacement, load, cross-sectional area, and load percentile.
4. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 1 or 3, characterized in that, Step 2, which involves feature extraction and load location encoding of the hysteresis curve and construction of the node feature matrix, includes: Step 2.1: Rearrange the starting points of the two-dimensional data of each hysteresis curve, using the minimum or maximum value in the load column as the starting point, so that the loading direction of the hysteresis curves is consistent. Step 2.2: Calculate the percentile ranking of the load data and normalize the loading stage; Step 2.3: Calculate the cross-sectional area of the sample based on its geometric parameters; Step 2.4: Combine displacement, load, cross-sectional area and load percentile to form a nodal feature matrix, and standardize the data.
5. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 1, characterized in that, Step 3 describes treating the sampling points of the hysteresis curve as graph nodes and constructing a circular bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a circular graph structure dataset, including: Step 3.1: Treat the equally spaced sampling points of the hysteresis curve as graph nodes and obtain the dataset for each graph node; Step 3.2: Create two edges in opposite directions for each pair of adjacent nodes, and establish connections between the first and last nodes according to the experimental sampling order to generate a ring graph structure dataset.
6. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 1, characterized in that, The two-layer graph convolutional neural network regression model mentioned in step 4 includes: The input section is used to receive the node feature matrix for each sample; The two-layer graph convolution part is used to extract local and global features at different scales from the input node feature matrix; The global average pooling part is used to compress the node feature matrix obtained after two convolutional layers into a graph-level vector through the global average pooling operation. The output section is used to map graph-level vectors to lifetime predictions using a linear regression layer.
7. The method for predicting high-temperature damage failure of additive manufacturing composite materials according to claim 6, characterized in that, The two-layer graph convolutional part includes: The first convolutional layer is used to map the input node feature matrix to a 128-dimensional latent space and uses ReLU activation and Dropout regularization to prevent overfitting. The second convolutional layer is used to aggregate the node neighborhood features output by the first convolutional layer in order to capture the nonlinear dependencies between adjacent stages in the hysteresis curve.
8. A high-temperature damage failure prediction system for additive manufacturing composite materials, characterized in that, include: The data acquisition module is used to conduct cyclic loading tests on fiber-reinforced PEEK composite material samples with different cross-sectional areas under uniform high temperature conditions, and to obtain the displacement-load hysteresis curves and their corresponding failure life data. The feature encoding module is used to extract features and encode load positions on the hysteresis curve, and to construct a node feature matrix. The graph structure construction module is used to treat the sampling points of the hysteresis curve as graph nodes and construct a ring-shaped bidirectional edge between adjacent graph nodes according to the experimental sampling order to generate a ring graph structure dataset. The model building and training module is used to construct a two-layer graph convolutional neural network regression model based on the ring graph structure dataset, and to train it by inputting the node feature matrix; The model evaluation module is used to validate and evaluate the performance of the trained two-layer graph convolutional neural network regression model using test samples. The lifetime prediction module is used to predict the remaining lifetime of additive manufacturing composite materials using a performance-compliant two-layer graph convolutional neural network regression model.
9. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.