Composite material fatigue stiffness degradation prediction method and device fused with machine learning
Through the convolutional neural network algorithm combined with experimental data, the accurate prediction of the fatigue stiffness degradation law of glass fiber composite materials in high temperature environments is achieved, and the difficulty of predicting the fatigue life of composite materials in the prior art is solved, and the accuracy of prediction is improved.
Patent Information
- Application Number
- CN202411434312.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2025-02-14
AI Technical Summary
There are difficulties in predicting the fatigue life of composite structures in high temperature environments, and the prior art is difficult to accurately predict the fatigue stiffness degradation pattern of glass fiber composites at different temperatures.
The convolutional neural network algorithm is used to construct the training set in combination with experimental data, which realizes the accurate prediction of the direction pull-tight fatigue stiffness degradation law of glass fiber composite fiber at different temperatures.
Through this method, the residual stiffness of the composite material at different temperatures, stress levels and fatigue cycles can be accurately predicted, which improves the accuracy of fatigue life prediction.
Smart Images
Figure CN119442845A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of high-temperature fatigue performance prediction of helicopter composite materials, and relates to a composite material fatigue stiffness degradation prediction method and device integrating machine learning. Background Art
[0002] The rotor is the core component for helicopter flight and control. Considering that the load environment of the rotor during flight is often very complex, the composite material structure is often accompanied by structural temperature rise during service due to its own viscoelastic effect and the influence of external environmental temperature. Therefore, the influence of thermal effect on structural fatigue performance has attracted much attention in practical engineering applications. High temperature environment may change the material properties of composite material structure, affect the degradation law of structural fatigue stiffness, and reduce the fatigue life of the structure. Summary of the invention
[0003] The purpose of the present invention is to propose a composite material fatigue stiffness degradation prediction method and device integrating machine learning, so as to accurately predict the fatigue stiffness degradation law of the composite material fiber direction of the glass fiber composite material structure at different temperatures.
[0004] Technical solution:
[0005] In a first aspect, a method for predicting high temperature fatigue stiffness degradation of composite materials integrating machine learning is provided, comprising:
[0006] Taking fiber direction composite laminates as the research object, the convolutional neural network algorithm was used to accurately predict the tension-tension fatigue stiffness degradation law of glass fiber composites in the fiber direction at different temperatures.
[0007] Furthermore, taking the fiber direction composite laminate as the research object, the convolutional neural network algorithm was used to achieve accurate prediction of the tensile-tensile fatigue stiffness degradation law of the fiber direction of glass fiber composites at different temperatures, including:
[0008] The stiffness degradation curves of the composite material at n different stress levels and different ambient temperatures are obtained through experiments;
[0009] The training set of the convolutional neural network is constructed with the composite material stiffness degradation data obtained from the experiment; different ambient temperatures, stress levels and the current number of fatigue cycles are used as the input of the training set, and the residual stiffness is used as the output of the training set;
[0010] The convolutional neural network is trained using a training set to obtain a network for predicting high temperature fatigue stiffness degradation of composite materials.
[0011] The target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials, and the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature is obtained;
[0012] According to the residual stiffness corresponding to different fatigue cycle numbers at the target stress level and the target ambient temperature, the parameters of the residual stiffness degradation model at the target stress level and the target ambient temperature are calculated, so as to obtain the residual stiffness degradation model at the target stress level and the target ambient temperature.
[0013] Furthermore, the residual stiffness degradation model is:
[0014]
[0015] Among them, ρ, λ and γ are the fitting parameters of the residual stiffness degradation model; E(0) represents the initial stiffness measured in the experiment, σ max is the critical failure stress measured in the test, εf is the critical failure stress measured in the test, n is the current number of fatigue cycles, N f is the nondestructive fatigue life under the current cyclic stress and stress ratio.
[0016] Furthermore, the target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials to obtain the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature, including:
[0017] First, the convolution layer is used to extract features of the input data, and the key features in the data are effectively captured with the help of convolution operations. Then, the data features are further screened using the pooling layer to extract more important features. Finally, the key features extracted by the convolutional neural network are integrated using the fully connected layer, and regression analysis is performed to output the residual stiffness of the composite material under different temperatures, stress levels, and fatigue cycle numbers.
[0018] Furthermore, the stiffness degradation curves of the composite material at n different stress levels and different ambient temperatures are obtained through experiments, including:
[0019] Carry out tensile-tensile fatigue tests on glass fiber composite materials in the fiber direction at different temperatures, obtain the test stiffness degradation curves at different temperatures, and record the normalized fatigue cycle number n / N f , stress level, and normalized residual stiffness.
[0020] In the second aspect, a composite material high temperature fatigue stiffness degradation prediction device integrating machine learning is provided, comprising:
[0021] The processing module is used to study fiber-direction composite laminates and use the convolutional neural network algorithm to accurately predict the tension-tension fatigue stiffness degradation law of glass fiber composites in the fiber direction at different temperatures.
[0022] Furthermore, the processing module is specifically used for:
[0023] The stiffness degradation curves of the composite material at n different stress levels and different ambient temperatures are obtained through experiments;
[0024] The training set of the convolutional neural network is constructed with the composite material stiffness degradation data obtained from the experiment; different ambient temperatures, stress levels and the current number of fatigue cycles are used as the input of the training set, and the residual stiffness is used as the output of the training set;
[0025] The convolutional neural network is trained using a training set to obtain a network for predicting high temperature fatigue stiffness degradation of composite materials.
[0026] The target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials, and the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature is obtained;
[0027] According to the residual stiffness corresponding to different fatigue cycle numbers at the target stress level and the target ambient temperature, the parameters of the residual stiffness degradation model at the target stress level and the target ambient temperature are calculated, so as to obtain the residual stiffness degradation model at the target stress level and the target ambient temperature.
[0028] Furthermore, the residual stiffness degradation model is:
[0029]
[0030] Among them, ρ, λ and γ are the fitting parameters of the residual stiffness degradation model; E(0) represents the initial stiffness measured in the experiment, σ max is the critical failure stress measured in the experiment, ε f is the critical failure stress measured in the test, n is the current number of fatigue cycles, N f is the nondestructive fatigue life under the current cyclic stress and stress ratio.
[0031] Beneficial effects:
[0032] This patent aims to develop a composite material fatigue stiffness degradation prediction method and device that integrates machine learning, accurately predict the fatigue stiffness degradation law of the composite fiber direction of glass fiber composite structures at different temperatures, and provide a reference for the fatigue design of rotor composite structures. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1It is a flow chart for predicting fatigue residual stiffness of composite laminates;
[0034] Figure 2 is the scatter plot of the tension-tension residual stiffness in the fiber direction of the glass fiber laminate at 45°C;
[0035] Figure 3 It is the CNN convolutional neural network prediction result graph;
[0036] Figure 4 It is the degradation surface diagram of the tensile-tensile stiffness in the fiber direction of the glass fiber laminate at 45°C. DETAILED DESCRIPTION
[0037] The present invention provides a method for predicting high temperature fatigue stiffness degradation of composite materials by integrating machine learning. The fiber direction composite laminate is taken as the research object, and the convolutional neural network algorithm is used to realize the accurate prediction of the tensile-tensile fatigue stiffness degradation law of the fiber direction of glass fiber composite materials at different temperatures. The specific steps are as follows:
[0038] [1] proposed a residual stiffness degradation model for composite structures
[0039] Taking fiber-direction composite laminates as the research object and composite tension-tension fatigue tests at different temperatures as the model object, a residual stiffness degradation model of composite structures is proposed.
[0040] [2] Convolutional Neural Network Algorithm
[0041] In order to improve the distortion of the residual stiffness degradation model in the process of fitting the stiffness degradation law of composite materials at different stress levels and different ambient temperatures, the convolutional neural network algorithm is integrated with machine learning to achieve accurate prediction of the tensile-tensile fatigue stiffness degradation law of glass fiber composite materials in the fiber direction at different temperatures. Figure 1 .
[0042] [3] Conduct fatigue tests
[0043] The fiber direction tensile fatigue test of glass fiber composite materials was carried out at different temperatures to obtain the experimental stiffness degradation curves at different temperatures, and the normalized fatigue cycle number n / Nf, stress level, and normalized residual stiffness were recorded.
[0044] [4] Prediction of residual stiffness degradation model integrating machine learning
[0045] The fatigue test results are input into the residual stiffness degradation model integrated with machine learning to achieve accurate prediction of the residual stiffness of composite structures at different temperatures and stress levels.
[0046] The present invention is further described in detail below with reference to examples.
[0047] [1] proposed a residual stiffness degradation model for composite structures
[0048] Taking the fiber-direction composite laminate as the research object and the composite tension-tension fatigue test at different temperatures as the model object, a residual stiffness degradation model of the composite structure is proposed, as follows:
[0049]
[0050] Among them, ρ, λ and γ are experimental fitting parameters. E(0) represents the initial stiffness, σ max is the critical stress for failure, ε f is the critical stress for failure, n is the number of current fatigue cycles, N f is the nondestructive fatigue life under the current cyclic stress and stress ratio.
[0051] [2] Convolutional Neural Network Algorithm
[0052] In order to improve the distortion of the residual stiffness degradation model in the process of fitting the stiffness degradation law of composite materials at different stress levels and different ambient temperatures, machine learning (i.e., convolutional neural network algorithm) is integrated to achieve accurate prediction of the tensile-tensile fatigue stiffness degradation law of glass fiber composite materials in the fiber direction at different temperatures. Figure 1 .
[0053] First, the training set is constructed using the fiber direction tensile fatigue residual stiffness degradation data obtained from the experiment, in which different ambient temperatures, stress levels and current fatigue cycles are used as training set inputs, and the residual stiffness is used as the training set output. Secondly, the convolution layer is used to extract features from the input data, and the key features in the data are effectively captured with the help of convolution operations. Then, the data features are further screened using the pooling layer to extract more important features. Finally, the key features extracted by the convolutional neural network are integrated using the fully connected layer, and regression analysis is performed to output the residual stiffness of the composite material under different temperatures, stress levels and fatigue cycles.
[0054] [3] Conduct fatigue tests
[0055] Carry out tensile-tensile fatigue tests on glass fiber composite materials in the fiber direction at different temperatures, obtain the test stiffness degradation curves at different temperatures, and record the normalized fatigue cycle number n / N f , stress level, and normalized residual stiffness.
[0056] Taking the tensile-tensile fatigue residual stiffness of glass fiber composite materials in the fiber direction at 45°C as an example, the scatter plot of the variation of residual stiffness with the number of fatigue test cycles is obtained as follows: Figure 2 .
[0057] The residual stiffness degradation function expression of the test result is given as follows:
[0058]
[0059] The residual stiffness degradation function can give the stiffness degradation law of composite materials at different temperatures.
[0060] [4] Prediction of residual stiffness degradation model integrating machine learning
[0061] The fatigue test results are input into the residual stiffness degradation model integrated with machine learning to further predict the residual stiffness of the composite structure at different temperatures, such as Figure 3 As shown in the figure, the root mean square error (RMSE) of the stiffness degradation prediction result based on the convolutional neural network is 0.045983, the mean absolute error (MAE) is 0.039383, and the mean relative percentage error (MAPE) is 4.7545%.
[0062] at the same time, Figure 4 The predicted envelope surface under different stress levels is given. The fusion of convolutional neural network algorithm can realize the accurate prediction of the tensile-tensile fatigue stiffness degradation law of glass fiber composite materials in the fiber direction at different temperatures, and can realize accurate evaluation of stiffness degradation under different stress levels and temperatures.
Claims
1. A method for predicting high temperature fatigue stiffness degradation of composite materials by integrating machine learning, characterized in that: include: Taking fiber direction composite laminates as the research object, the convolutional neural network algorithm was used to accurately predict the tension-tension fatigue stiffness degradation law of glass fiber composites in the fiber direction at different temperatures.
2. The method according to claim 1, characterized in that Taking fiber-direction composite laminates as the research object, the convolutional neural network algorithm is used to accurately predict the degradation law of tensile-tensile fatigue stiffness of glass fiber composites in the fiber direction at different temperatures, including: The stiffness degradation curves of the composite material at n different stress levels and different ambient temperatures are obtained through experiments; The training set of the convolutional neural network is constructed with the composite material stiffness degradation data obtained from the experiment; different ambient temperatures, stress levels and the current number of fatigue cycles are used as the input of the training set, and the residual stiffness is used as the output of the training set; The convolutional neural network is trained using a training set to obtain a network for predicting high temperature fatigue stiffness degradation of composite materials. The target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials, and the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature is obtained; According to the residual stiffness corresponding to different fatigue cycle numbers at the target stress level and the target ambient temperature, the parameters of the residual stiffness degradation model at the target stress level and the target ambient temperature are calculated, so as to obtain the residual stiffness degradation model at the target stress level and the target ambient temperature.
3. The method according to claim 2, characterized in that The residual stiffness degradation model is: Among them, ρ, λ and γ are the fitting parameters of the residual stiffness degradation model; E(0) represents the initial stiffness measured in the experiment, σ max is the critical failure stress measured in the test, εf is the critical failure stress measured in the test, n is the current number of fatigue cycles, N f is the nondestructive fatigue life under the current cyclic stress and stress ratio.
4. The method according to claim 2, characterized in that: The target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials to obtain the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature, including: First, the convolution layer is used to extract features of the input data, and the key features in the data are effectively captured with the help of convolution operations. Then, the data features are further screened using the pooling layer to extract more important features. Finally, the key features extracted by the convolutional neural network are integrated using the fully connected layer, and regression analysis is performed to output the residual stiffness of the composite material under different temperatures, stress levels, and fatigue cycle numbers.
5. The method according to claim 2, characterized in that: The stiffness degradation curves of composite materials at n different stress levels and different ambient temperatures are obtained through experiments, including: Carry out tensile-tensile fatigue tests on glass fiber composite materials in the fiber direction at different temperatures, obtain the test stiffness degradation curves at different temperatures, and record the normalized fatigue cycle number n / N f , stress level, and normalized residual stiffness.
6. A composite material high temperature fatigue stiffness degradation prediction device integrating machine learning, characterized in that: include: The processing module is used to study fiber-direction composite laminates and use the convolutional neural network algorithm to accurately predict the tension-tension fatigue stiffness degradation law of glass fiber composites in the fiber direction at different temperatures.
7. The device according to claim 6, characterized in that The processing module is specifically used for: The stiffness degradation curves of the composite material at n different stress levels and different ambient temperatures are obtained through experiments; The training set of the convolutional neural network is constructed with the composite material stiffness degradation data obtained from the experiment; different ambient temperatures, stress levels and the current number of fatigue cycles are used as the input of the training set, and the residual stiffness is used as the output of the training set; The convolutional neural network is trained using a training set to obtain a network for predicting high temperature fatigue stiffness degradation of composite materials. The target ambient temperature, target stress level and different fatigue cycle numbers are input into the network for predicting high temperature fatigue stiffness degradation of composite materials, and the residual stiffness corresponding to the target stress level and different fatigue cycle numbers at the target ambient temperature is obtained; According to the residual stiffness corresponding to different fatigue cycle numbers at the target stress level and the target ambient temperature, the parameters of the residual stiffness degradation model at the target stress level and the target ambient temperature are calculated, so as to obtain the residual stiffness degradation model at the target stress level and the target ambient temperature.
8. The device according to claim 7, characterized in that The residual stiffness degradation model is: Among them, ρ, λ and γ are the fitting parameters of the residual stiffness degradation model; E(0) represents the initial stiffness measured in the experiment, σ max is the critical failure stress measured in the test, εf is the critical failure stress measured in the test, n is the current number of fatigue cycles, N f is the nondestructive fatigue life under the current cyclic stress and stress ratio.