Aluminum alloy fatigue life prediction method considering fatigue attributes
Through the optimization of support vector machines and physical model parameters combined with graph neural network, genetic algorithm optimization, an integrated system is formed, which solves the problem of insufficient accuracy and adaptability of traditional aluminum alloy fatigue life prediction methods, and achieves efficient and accurate fatigue life prediction , enhancing application trust in the engineering field.
Patent Information
- Application Number
- CN202411377974.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-10
AI Technical Summary
Traditional aluminum alloy fatigue life prediction methods rely on experimental data, have low accuracy and lack adaptability, pure data-driven methods lack physical interpretation, difficult to gain trust in engineering practice, and it is difficult to obtain labeled data.
Combining graph neural network (GNN), support vector machine (GA-SVM) optimized by genetic algorithms and physical model parameter optimization, an integrated system is formed, and high-precision fatigue life prediction is achieved through data expansion, parameter optimization and physical model correction.
It significantly improves the accuracy and efficiency of fatigue life prediction, enhances the transparency and credibility of the model, reduces experimental costs, and is suitable for aviation, automobiles, high-speed rail and other fields.
Smart Images
Figure CN119296676A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of material science and data analysis technology, and specifically to an aluminum alloy fatigue life prediction system and a method for using the same. The method integrates a graph neural network (GNN), a support vector machine model (GA-SVM) optimized by a genetic algorithm, and optimization of physical model parameters, aiming to achieve accurate prediction and analysis of the fatigue life of aluminum alloys taking fatigue properties into consideration. The technology has significant advantages in the evaluation and optimization of the fatigue properties of aluminum alloys, and is mainly used in engineering fields such as aviation, automobiles, high-speed rail, and shipbuilding that have high requirements for the safe service performance of aluminum alloy materials. Background Art
[0002] As a widely used structural lightweight material, accurate prediction of fatigue life of aluminum alloy plays a vital role in ensuring the safety and reliability of engineering structures. Traditional fatigue life prediction methods are usually based on empirical physical models, and the determination of model parameters is highly dependent on a large number of experiments. Although these physical models provide a theoretical basis to a certain extent, the prediction accuracy is often low due to insufficient experimental data or data errors. In addition, physical models usually lack sufficient adaptability when facing complex working conditions, which limits their widespread promotion and effectiveness in practical applications.
[0003] With the advancement of machine learning technology, data-driven fatigue life prediction methods have become possible. However, although pure data-driven methods can process large-scale data and reveal complex relationships between data, they often lack the necessary physical explanations and are difficult to gain full trust in engineering practice. At the same time, most existing machine learning models rely on a large amount of labeled data for training, but it is very difficult to obtain this data in actual material testing.
[0004] Therefore, it is necessary to develop a new fatigue life prediction method that combines the advantages of machine learning and physical models. This method can provide both high-precision predictions and physically meaningful explanations to meet the needs of complex practical applications. This method will significantly improve the efficiency and accuracy of fatigue life prediction, reduce experimental costs, and enhance the transparency and credibility of the model. It has great practical application value in engineering fields such as aviation, automobiles, high-speed rail, and shipbuilding. Summary of the invention
[0005] Based on the above shortcomings of the existing technology, the present invention proposes a system combining graph neural network (GNN), genetic algorithm optimized support vector machine (GA-SVM), and physical model parameter optimization, which is specifically used to improve the accuracy of fatigue life prediction of aluminum alloys. This system effectively overcomes the limitations of the existing technology by integrating advanced machine learning methods with traditional physical theory analysis, and provides a more accurate and efficient fatigue life prediction method.
[0006] Technical implementation details: S1. Data and processing: 1. Experimental data collection: The fatigue performance of aluminum alloy is evaluated by the lifting and lowering method, and the fatigue life data of aluminum alloy under different stress levels are systematically collected, covering loading frequency, maximum stress, minimum stress and its corresponding fatigue life information. 2. Data expansion: The fatigue data is feature expanded using a graph neural network (GNN). The GNN model used contains three layers of graph convolutional layers, the feature output dimension of each layer is set to 64, and ReLU is used as the activation function. The GNN model is trained using the Adam optimizer, the learning rate is set to 0.001, the batch size is 32, and the training process is repeated 200 times to enhance the feature expression ability of the data.
[0007] S2, Genetic Algorithm Optimized Support Vector Machine Model (GA-SVM): 1. Parameter optimization: The parameters of the support vector machine are optimized through genetic algorithm. The specific optimized parameters include the regularization parameter (C), which is set in the range of 0.1 to 10000; the selection range of the kernel function gamma is ['scale', 'auto', 0.001, 0.01, 0.1, 1, 10]; and the epsilon parameter, which is set in the range of 0.001 to 0.5. The configuration of the genetic algorithm is: the population size is 50, the number of iterations is 40 generations, the crossover probability is set to 0.5, and the mutation probability is 0.2. The genetic algorithm uses a roulette selection strategy, the crossover operation uses a two-point crossover, and the mutation operation is a uniform mutation. 2. Model training and evaluation: After the support vector machine model training is completed after parameter optimization, the mean square error (MSE) and the coefficient of determination (R²) are used as evaluation indicators to test the prediction performance of the model to ensure its accuracy and robustness. S3. Physical model parameter optimization: 1. Detailed explanation of the model formula. Through the stress concentration coefficient K t , surface roughness coefficient C s The SWT model is modified, and the improved physical equation for SWT fatigue life prediction is: (1) (2) in q is the notch sensitivity coefficient, b is the fatigue strength index, c is the fatigue ductility index. These fatigue property parameters are predicted by the predicted values of machine learning. 2. Parameter determination Based on the output of the SVM model, the parameters in the physical model are optimized. q , b and c , the goal is to minimize the difference between the predicted lifetime and the experimental lifetime.
[0008] S4. System verification and application: 1. Result verification: The prediction results of the model are evaluated by calculating the mean square error (MSE) and the coefficient of determination (R²) to verify the accuracy and reliability of the model. These indicators can effectively measure the accuracy of the model prediction. MSE is used to measure the difference between the predicted value and the true value, and R² is used to measure the ability of the model to explain data variation. 2. Parameter adjustment and optimization: Based on the prediction error of the model, the model parameters are adjusted. Through continuous iterative optimization process, the performance of the model is gradually improved to improve the accuracy and stability of the prediction. This process may involve adjusting the parameters of the genetic algorithm, the hyperparameters of the support vector machine, and the structure and training parameters of the graph neural network to ultimately achieve the ideal prediction results.
[0009] Beneficial effects of the technical solution: 1. Improve processing efficiency and accuracy: This invention significantly improves the processing efficiency and prediction accuracy of fatigue life prediction of aluminum alloy considering fatigue properties by integrating graph neural network (GNN) and genetic algorithm optimized support vector machine (GA-SVM). Compared with the traditional manual analysis method, this integrated machine learning method reduces human intervention and proposes a data-driven fatigue property parameter acquisition method to ensure the consistency and repeatability of the prediction results. In particular, the genetic algorithm can quickly find the optimal model settings when adjusting the parameters of the support vector machine through the natural selection mechanism, which greatly improves the speed and accuracy of the prediction.
[0010] 2. Enhanced data analysis capabilities: Through the application of graph neural networks, the present invention demonstrates powerful capabilities in data expansion and feature extraction. GNN can deeply analyze the complex characteristics of aluminum alloy fatigue life data to generate a more comprehensive data set. This enhances the expressiveness of the data, provides basic support for more accurate fatigue life prediction, and helps engineers and researchers better understand the performance of aluminum alloys under different working conditions and make more accurate material performance evaluations and application decisions.
[0011] 3. Ease of operation: The system is developed based on a modern computing platform, has a user-friendly interface, and supports automated data processing and parameter optimization. Users do not need to have a deep algorithm background to perform efficient predictive analysis. The entire system has a simple operation process and supports batch processing, which greatly reduces the difficulty and time cost of operation, and is very suitable for wide application in industrial production and scientific research.
[0012] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following is a detailed description in combination with the preferred embodiments and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings, and the principles and operation processes of the present invention, as well as other aspects, features and advantages of the present invention are explained through examples. In the referenced drawings, the detailed workflow and result display of the automated metal aluminum alloy fatigue life prediction system provided by the present invention are shown through different charts.
[0014] like Figures 1 to 5 As shown, the system provided by the present invention uses advanced data analysis technology and algorithm optimization physical model to process complex experimental data, thereby improving the efficiency of data processing and the accuracy of prediction. The following are the specific steps for using the system: S1. Data processing: Obtain specific experimental data through experiments, and expand the data through the GNN network to obtain the data set to be processed.
[0015] S2. Genetic algorithm optimization of random forest: Genetic algorithm is used to automatically adjust the key parameters of random forest and obtain the optimal parameters of SVM. Figure 2 The chart shows the fitness changes during the genetic algorithm optimization process. This chart shows the best fitness and average fitness during the genetic algorithm operation, reflecting the algorithm optimization effect and convergence, thereby verifying the effectiveness and stability of the algorithm. Then the optimized random forest network is used for training to complete the fatigue life prediction. Figure 3 The comparison of the predicted value and the actual value of SVM in the figure is shown in Figure 2, as well as the mean square error (MSE) and the coefficient of determination (R 2 ) to evaluate the model and obtain R 2 =0.94, MSE =9.8×10 7 , indicating that the SVM prediction model has good accuracy.
[0016] S3, physical model parameter optimization: Based on the prediction results of the support vector machine, the predicted value SN curve is obtained ( Figure 4 ), and then the L-BFGS-B algorithm is used to fit the fatigue property parameters of the improved SWT fatigue life physical equation to obtain the parameters q = 0.82, b = - 0.95 and c = -0.69, thus determining and improving the physical equation for predicting fatigue life of aluminum alloy considering fatigue properties.
[0017] S4. Result verification and optimization: Figure 5 The comparison results between the prediction model prediction results and the experimental data are shown to verify the accuracy of the prediction model. The data distribution characteristics in the figure are: the fatigue life model prediction values and the actual values are mostly near the 45-degree line, which proves that the fatigue life prediction accuracy is high.
[0018] S5. Report generation: Generate detailed reports of prediction results and parameter optimization.
[0019] The above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and changes can be made without departing from the principle of the present invention, and these improvements and changes are also regarded as the protection scope of the present invention.
Claims
1. A method for predicting fatigue life of aluminum alloy, characterized in that The following steps are involved: S1. Data collection: Obtain fatigue life data of aluminum alloy under different stress levels through experiments; S2. Data expansion: Use graph neural network (GNN) to expand the collected data. The GNN model contains at least three layers of graph convolution layers, and the output feature dimension of each layer is set to 64, and the ReLU activation function is used for nonlinear transformation; S3. Genetic algorithm optimization of support vector machine model: The parameters of the support vector machine (SVM) model are optimized by genetic algorithm. The optimized parameters include: regularization parameter (C), the range is set to 0.1 to 10000; kernel function gamma, the selection range is ['scale', 'auto', 0.001, 0.01, 0.1, 1,10]; epsilon parameter, the optimization range is 0.001 to 0.5; the optimization goal is to minimize the mean square error (MSE) of the model to improve the prediction accuracy; S4. Determination of physical model parameters: The key fatigue property parameters in the physical model are optimized by using the data feedback of the support model prediction output. The physical equation is the modified SWT parameter equation including: （1） （2） in q is the notch sensitivity coefficient, b is the fatigue strength index, c is the fatigue ductility index. These fatigue property parameters are predicted by machine learning prediction values; S5. System verification: Compare the prediction results with the experimental data and use the R 2 and MSE values, and compare the model prediction values with the actual values to verify the prediction accuracy and reliability of the system.
2. The method according to claim 1, characterized in that During the training of the graph neural network (GNN) model, the Adam optimizer is used for parameter update, the learning rate is set to 0.001, the batch size is set to 32, and the training cycle is not less than 100 cycles.
3. The method according to claim 1, characterized in that The specific implementation of the genetic algorithm includes the following parameters: the population size is set to 50, the selection strategy is the roulette method, the crossover probability is 0.5, the mutation probability is 0.2, and the genetic algorithm iteration generation is 40 generations.
4. The method according to claim 1, characterized in that The optimization step of the fatigue property parameters of the physical equation is to fit the physical model formula through the L-BFGS-B algorithm, and the optimization goal is to minimize the difference between the fatigue life predicted by the model and the fatigue life measured by the experiment.