An adaptive high-temperature damage model construction method and system

Abstract

The application provides a self-adaptive high-temperature damage model construction method and system. It belongs to the field of material science and engineering calculation technology. The method comprises the following steps: obtaining experimental data of a material in a high-temperature environment, pre-processing the collected data through a machine learning algorithm, and extracting key damage features; constructing a basic damage prediction model based on a deep learning framework, training the model using the pre-processed data, and through a reinforcement learning mechanism, allowing the basic damage prediction model to adaptively adjust internal parameters according to prediction errors; using a genetic algorithm to perform a global search on thermodynamic parameters in the model to find an optimal parameter combination, and combining real-time temperature monitoring data to dynamically adjust model parameters. Through the integration of deep learning, reinforcement learning, genetic algorithm and molecular dynamics simulation technology, the damage evolution process of the material in the high-temperature environment can be more accurately and comprehensively predicted.

Description

Technical Field

[0001] This invention proposes an adaptive high-temperature damage model construction method and system, belonging to the field of materials science and engineering calculation technology. Background Technology

[0002] Most existing high-temperature damage models are based on static mechanics theory, which makes it difficult to fully capture the complex damage mechanisms of materials under extreme conditions such as high temperature and high pressure, including thermal stress relaxation, creep fracture, and microstructure reconstruction. Furthermore, these models often lack adaptability and cannot adjust parameters according to the real-time state of the material during service, resulting in limited prediction accuracy. Therefore, developing an innovative model that can dynamically adapt to changes in the high-temperature environment and accurately predict material damage behavior is particularly urgent. Summary of the Invention

[0003] This invention provides an adaptive high-temperature damage model construction method and system to solve the problems mentioned in the background section above:

[0004] This invention proposes an adaptive high-temperature damage model construction method, the method comprising:

[0005] S1. Obtain experimental data of materials under high temperature environment, preprocess the collected data through machine learning algorithm, and extract key damage features;

[0006] S2. Construct a basic damage prediction model based on a deep learning framework, train the model using preprocessed data, and use a reinforcement learning mechanism to enable the basic damage prediction model to adaptively adjust its internal parameters according to the prediction error.

[0007] S3. Use a genetic algorithm to perform a global search for the thermodynamic parameters in the model, find the optimal parameter combination, and dynamically adjust the model parameters in combination with real-time temperature monitoring data;

[0008] S4. Based on molecular dynamics simulation technology, the microstructure evolution process of materials at high temperature is simulated, the simulation results are compared with experimental data, and the model parameters are further optimized through feedback mechanism.

[0009] S5. Select an independent validation dataset to validate the basic damage prediction model and evaluate its prediction performance. Based on the evaluation results, iteratively optimize the model.

[0010] This invention proposes an adaptive high-temperature damage model construction system, the system comprising:

[0011] Data acquisition module: Acquires experimental data of materials under high temperature environment, preprocesses the collected data through machine learning algorithm, and extracts key damage features;

[0012] Model building module: Based on the deep learning framework, a basic damage prediction model is built. The model is trained using preprocessed data, and through the reinforcement learning mechanism, the basic damage prediction model adaptively adjusts its internal parameters according to the prediction error.

[0013] Parameter adjustment module: Uses a genetic algorithm to perform a global search for the thermodynamic parameters in the model, finds the optimal parameter combination, and dynamically adjusts the model parameters in combination with real-time temperature monitoring data;

[0014] Process simulation module: Based on molecular dynamics simulation technology, it simulates the microstructure evolution process of materials at high temperatures, compares the simulation results with experimental data, and further optimizes the model parameters through a feedback mechanism;

[0015] Iterative optimization module: Select an independent validation dataset to validate the basic damage prediction model and evaluate its prediction performance. Based on the evaluation results, iteratively optimize the model.

[0016] The beneficial effects of this invention are as follows: By integrating deep learning, reinforcement learning, genetic algorithms, and molecular dynamics simulation techniques, it is possible to predict the damage evolution process of materials under high-temperature environments more accurately and comprehensively. This method combines experimental data with simulation results, enabling adaptive adjustment of damage model parameters under different temperatures and loading conditions, thereby improving the accuracy of damage prediction. In terms of data processing, machine learning algorithms are used to preprocess experimental data and extract key damage features. Furthermore, principal component analysis (PCA) and linear discriminant analysis (LDA) are used for feature selection and optimization, effectively removing noise and improving data validity and model learning performance. During training, reinforcement learning mechanisms and optimization algorithms (such as genetic algorithms) allow the damage prediction model to dynamically adjust its internal parameters based on prediction errors, achieving adaptive optimization. In addition, optimization techniques such as periodic error evaluation, learning rate decay strategies, early stopping, and grid search effectively avoid model overfitting and improve the model's generalization ability. Using molecular dynamics simulation techniques to simulate the microstructure of materials allows for a deeper understanding of the damage evolution mechanism at high temperatures, and feedback optimization is performed by comparing with experimental data, further improving the model's accuracy. This process helps identify the microscopic damage evolution of materials under high-temperature environments, providing a theoretical basis for materials design and engineering applications. By using a genetic algorithm to globally search for thermodynamic parameters, not only can the combination of thermodynamic parameters in the model be optimized, but the model parameters can also be dynamically adjusted based on real-time temperature monitoring data. This allows the prediction model to more accurately reflect changes in material performance under different operating conditions. Through multi-dimensional evaluation methods such as independent validation datasets, k-fold cross-validation, and feature importance analysis, the predictive performance of the model can be comprehensively analyzed, key features and parameters can be identified, and the model can be finely adjusted and continuously optimized, thereby ensuring its long-term stability and efficiency in practical applications. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the steps of the method described in this invention;

[0018] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] One embodiment of the present invention, such as Figure 1 As shown, an adaptive high-temperature damage model construction method is provided, the method comprising:

[0021] S1. Obtain experimental data of materials under high temperature environment, preprocess the collected data through machine learning algorithm, and extract key damage features;

[0022] S2. Construct a basic damage prediction model based on a deep learning framework, train the model using preprocessed data, and use a reinforcement learning mechanism to enable the basic damage prediction model to adaptively adjust its internal parameters according to the prediction error.

[0023] S3. Use a genetic algorithm to perform a global search for the thermodynamic parameters in the model, find the optimal parameter combination, and dynamically adjust the model parameters in combination with real-time temperature monitoring data;

[0024] S4. Based on molecular dynamics simulation technology, the microstructure evolution process of materials at high temperature is simulated, the simulation results are compared with experimental data, and the model parameters are further optimized through feedback mechanism.

[0025] S5. Select an independent validation dataset to validate the basic damage prediction model and evaluate its prediction performance. Based on the evaluation results, iteratively optimize the model.

[0026] The working principle of the above technical solution is as follows: Data is acquired from material experiments under high-temperature conditions. This data includes stress-strain curves (describing the deformation of the material under stress), temperature-time curves (describing temperature changes during the experiment), and microstructure images (showing the microstructure changes of the material at high temperatures). Noise and outliers are removed from the data to ensure accuracy and consistency. For missing data, interpolation or other algorithms are used to fill in the gaps to ensure data integrity. Key features that significantly influence material damage prediction, such as peak stress, strain rate, and temperature gradient, are extracted from the raw data. A basic damage prediction model is built based on deep learning frameworks (such as TensorFlow and PyTorch). This model can receive preprocessed data as input and output predictions of material damage. The model is trained using the preprocessed data to learn the features and patterns in the data. A reinforcement learning mechanism is introduced to enable the model to automatically adjust its internal parameters based on prediction errors to improve prediction accuracy. A genetic algorithm is used to perform a global search of the thermodynamic parameters in the model to find the optimal parameter combination. Genetic algorithms, by simulating natural selection and genetic mechanisms, can efficiently search the solution space; combined with real-time temperature monitoring data, model parameters are dynamically adjusted. This ensures that the model maintains high prediction accuracy under different temperature conditions; based on molecular dynamics simulation techniques, the microstructural evolution of materials at high temperatures is simulated. This helps to understand the damage mechanism of materials at high temperatures; comparing simulation results with experimental data, and further optimizing model parameters through a feedback mechanism, ensures that the model can more accurately reflect the actual damage situation of materials; independent validation datasets are selected to validate the basic damage prediction model. This allows for the evaluation of the model's predictive performance on unknown data; the model is iteratively optimized based on the validation results. If the prediction accuracy does not meet the requirements, the model structure, parameters, or data preprocessing methods are adjusted, and retraining and validation are performed until satisfactory prediction accuracy is achieved.

[0027] The effects of the above technical solution are as follows: Preprocessing the collected data using machine learning algorithms, including data cleaning, missing value imputation, and feature selection, effectively removes noise and outliers, ensuring data accuracy and consistency. Extracting key damage features helps the model better learn the features and patterns in the data, thereby improving prediction accuracy. This method builds the model entirely based on experimental data, avoiding the problems of relying on numerous assumptions and simplifications in traditional methods, thus improving the model's reliability and practicality. Through reinforcement learning, the basic damage prediction model can adaptively adjust its internal parameters based on prediction errors, continuously optimizing model performance. This adaptive adjustment capability allows the model to maintain high prediction accuracy when facing different high-temperature environments and material types. Using a genetic algorithm to perform a global search on the thermodynamic parameters in the model can find the optimal parameter combination, further improving the model's prediction accuracy. Dynamically adjusting model parameters in conjunction with real-time temperature monitoring data enables… The model can adapt to environmental changes in real time, improving its practicality and robustness. Based on molecular dynamics simulation technology, the microstructural evolution of materials at high temperatures is simulated, contributing to a deeper understanding of the damage mechanisms at these temperatures. This simulation technology provides strong theoretical support for model construction and optimization. Comparing simulation results with experimental data and using a feedback mechanism to further optimize model parameters allows the model to more accurately reflect the actual damage situation of materials. This feedback optimization mechanism helps improve the model's prediction accuracy and generalization ability. Selecting independent validation datasets to validate the basic damage prediction model allows for an objective evaluation of its predictive performance. This validation method helps identify problems and shortcomings in the model, providing direction for subsequent optimization. Iterative optimization based on validation results involves continuously adjusting the model structure, parameters, or data preprocessing methods until satisfactory prediction accuracy is achieved. This iterative optimization method enables continuous improvement and refinement of the model, enhancing its practicality and accuracy.

[0028] In one embodiment of the present invention, S1 includes:

[0029] S11. Based on the thermal stability range of the material, set a series of temperature gradients, apply conditional loading, and set different loading rates and stress levels under each loading condition;

[0030] S12. At different temperature points and after loading stages, microstructure images were obtained by SEM, and the evolution process of micro-damage was recorded.

[0031] S13. Identify outliers using statistical methods and further verify the rationality of the data by combining the physical properties of the materials;

[0032] For time series data, such as temperature-time curves, linear interpolation or prediction interpolation based on time series models can be used.

[0033] For characteristic values ​​such as stress-strain curves, nearest-neighbor interpolation or prediction interpolation based on machine learning models is used.

[0034] S14. Extract the physical characteristics of the material, such as elastic modulus, yield strength, fracture strength, thermal characteristics such as coefficient of thermal expansion, thermal conductivity, heat capacity, and microstructural characteristics such as crack length, width, density, and dislocation distribution.

[0035] S15. Perform PCA analysis on the feature matrix, retain the first few principal components, and use LDA for feature selection in classification problems.

[0036] The working principle of the above technical solution is as follows: A series of temperature gradients are set according to the thermal stability range of the material. These temperature gradients cover multiple key temperature points from room temperature to the material's melting point. At each temperature point, different loading conditions are tested, such as static tension, compression, and fatigue loading. These loading conditions simulate various stress states that the material may encounter in practical applications. Under each loading condition, different loading rates and stress levels are set to investigate the material's response and damage under different stress conditions. Scanning electron microscopy (SEM) is used to acquire microstructural images of the material at different temperature points and after loading stages. By observing and analyzing these images, the evolution of microscopic damage in the material is recorded, such as crack formation, propagation, dislocation generation, and evolution. The experimental data is initially cleaned using statistical methods to identify and remove outliers. These outliers may be caused by experimental errors, instrument malfunctions, or data recording errors. The rationality of the data is further verified by combining the material's physical properties (such as the variation law of elastic modulus and yield strength). This helps ensure the accuracy and reliability of the data; for time-series data (such as temperature-time curves), linear interpolation or prediction interpolation based on time-series models is used to fill missing values ​​or smooth data; for characteristic data such as stress-strain curves, nearest-neighbor interpolation or prediction interpolation based on machine learning models is used to estimate missing or outlier data points; physical, thermal, and microstructural features related to material damage are extracted from experimental data; physical features include elastic modulus, yield strength, and fracture strength, which reflect the material's response under mechanical loading; thermal features include coefficient of thermal expansion, thermal conductivity, and heat capacity, which describe the material's thermal behavior at high temperatures; microstructural features include crack length, width, density, and dislocation distribution, which reveal the damage at the microscopic level of the material; principal component analysis (PCA) is performed on the extracted feature matrix to retain the first few principal components. PCA is a commonly used data dimensionality reduction technique that reduces data dimensionality and redundancy by projecting the original data onto a lower-dimensional space; in classification problems, linear discriminant analysis (LDA) is used for feature selection. LDA is a supervised dimensionality reduction method that maximizes inter-class distance and minimizes intra-class distance based on class labels, thereby selecting the features most useful for classification. By combining PCA and LDA, the model can be further simplified, prediction accuracy improved, and computational complexity reduced.

[0037] The effects of the above technical solution are as follows: Based on the material's thermal stability range, a series of temperature gradients were scientifically set, covering multiple key temperature points from room temperature to the material's melting point. This helps to comprehensively explore the material's damage behavior at different temperatures; at each temperature point, various loading condition experiments were conducted, such as static tension, compression, and fatigue loading, with different loading rates and stress levels set. This helps to gain a deeper understanding of the material's response and damage mechanisms under different stress states; scanning electron microscopy (SEM) was used to acquire microstructural images of the material at different temperature points and after loading stages, providing intuitive evidence of microscopic damage; the evolution process of microscopic damage was recorded in detail, helping to reveal the origin, development, and evolution of material damage; outliers were effectively identified based on statistical methods, ensuring the accuracy and reliability of the data; the rationality of the data was further verified by combining the material's physical properties (such as the variation law of elastic modulus and yield strength), which enhanced the data's credibility; for time series data (such as temperature-time curves), linear interpolation or time-based interpolation was used. Interpolation predictions from inter-sequence models help smooth data and fill in missing values. For characteristic data such as stress-strain curves, nearest-neighbor interpolation or machine learning-based predictive interpolation are used, which improves data completeness and accuracy. Physical, thermal, and microstructural features related to material damage are extracted from experimental data, covering multiple aspects of material damage. Principal component analysis (PCA) is performed on the feature matrix, retaining the first few principal components, which helps reduce data dimensionality and redundancy and improves the computational efficiency of the model. In classification problems, linear discriminant analysis (LDA) is used for feature selection, which helps select the most useful features for classification and improves the model's prediction accuracy.

[0038] In one embodiment of the present invention, S11 includes:

[0039] Based on the material's thermal stability range, an initial temperature gradient is set. Simultaneously, based on the material's effects at different temperatures, the temperature gradient is fine-tuned.

[0040] For each temperature gradient, multiple loading conditions were designed, and under each loading condition, the settings for loading rate and stress level were further refined;

[0041] The temperature control system ensures the accuracy and stability of the temperature during the experiment; during the heating process, the temperature changes are monitored in real time and fine-tuned as needed.

[0042] According to the experimental plan, appropriate loading devices and measuring equipment were selected. For static tensile and compression tests, a universal testing machine was used for loading and measurement; for fatigue loading tests, a fatigue testing machine was used for cyclic loading and measurement. Simultaneously, the accuracy and stability of the loading devices and measuring equipment were ensured to improve the accuracy of the experimental data.

[0043] During the experiment, a data acquisition system was used to record key data such as temperature, stress, and strain in real time. Based on the real-time monitoring data and experimental phenomena, feedback and adjustments were made to the experimental conditions.

[0044] Based on the preliminary experimental results and analysis, the experimental conditions were further optimized; and the experimental scheme was iteratively updated based on the adjustment and optimization of the experimental conditions.

[0045] The working principle of the above technical solution is as follows: Based on the thermal stability range of the material, starting from room temperature, a series of temperature gradients are set at certain temperature intervals (such as 50℃, 100℃, etc.) until the temperature approaches but does not exceed the melting point of the material. Considering the thermal expansion and thermal stress effects of the material at different temperatures, the initially set temperature gradients are fine-tuned. This helps to more accurately reflect the temperature environment of the material in real applications and optimize experimental conditions to capture the key damage behaviors of the material. For each temperature gradient, multiple loading conditions are designed, including static tension, compression, fatigue loading, etc., to simulate different stress conditions of the material in actual applications. Under each loading condition, the setting of loading rate and stress level is further refined. For example, different tensile rates and maximum tensile stresses are set in the static tensile test; different number of cycles, stress amplitude, and stress ratio are set in the fatigue loading test. A temperature control system is used to ensure the accuracy and stability of the temperature during the experiment. During the heating process, the temperature change is monitored in real time and fine-tuned as needed to maintain the predetermined temperature gradient. A suitable loading device (such as a universal testing machine, fatigue testing machine, etc.) is selected to ensure the accuracy and stability of the loading process. Simultaneously, according to the experimental plan, the loading device settings were adjusted to match the predetermined loading conditions; key data such as temperature, stress, and strain were recorded in real time through the data acquisition system. This data forms the basis for subsequent analysis of material damage behavior. Based on real-time monitoring data and experimental phenomena, the experimental conditions were fed back and adjusted. For example, if unstable temperature control or a malfunction in the loading device was detected, timely adjustments and repairs were made to ensure the accuracy and reliability of the experimental data. Based on the preliminary experimental results and analysis, the rationality and effectiveness of the experimental conditions were evaluated; parameters such as temperature gradient, loading rate, and stress level were adjusted according to material performance changes and damage mechanisms. This helps to more accurately simulate the stress conditions and performance changes of materials in practical applications. Based on the adjustment and optimization of experimental conditions, the experimental scheme was iteratively updated. Through continuous trial and improvement, the optimal combination of experimental conditions was found to ensure the accuracy and reliability of the experimental data.

[0046] The effects of the above technical solution are as follows: A preliminary temperature gradient is set based on the material's thermal stability range, and fine-tuned based on the material's thermal expansion and thermal stress effects at different temperatures, ensuring the scientific accuracy of the experimental temperature conditions; multiple loading conditions are designed for each temperature gradient, and the loading rate and stress level settings are refined, comprehensively simulating the stress conditions of the material in practical applications, improving the accuracy and practicality of the experiment; the temperature control system precisely controls the temperature during the experiment, monitoring temperature changes in real time and making fine-tuning, ensuring the accuracy and stability of the experimental temperature; appropriate loading devices and measuring equipment are selected, and their accuracy and stability are ensured, improving the accuracy and reliability of the experimental data; the data acquisition system records key data such as temperature, stress, and strain in real time, providing rich data support for subsequent analysis. The experimental conditions were adjusted based on real-time monitoring data and experimental phenomena, ensuring the smooth progress of the experiment and the accuracy of the experimental data. Based on the preliminary experimental results and analysis, the experimental conditions were further optimized, improving the accuracy and reliability of the experiment. On the basis of the adjustment and optimization of the experimental conditions, the experimental scheme was iteratively updated. Through continuous trial and improvement, the optimal combination of experimental conditions was found, ensuring the accuracy and reliability of the experimental data. The experimental results obtained through this technical solution have high practicality and guiding significance, providing strong data support and theoretical basis for the research, design, manufacturing, and application of materials. This technical solution not only improves the accuracy and reliability of the experiment but also promotes the development of materials science research, providing important methods and means for material performance evaluation, damage mechanism research, and lifetime prediction.

[0047] In one embodiment of the present invention, S2 includes:

[0048] S21. Select a deep learning framework. If it is used to process time series data, choose the LSTM model, such as temperature-time curves and stress-strain curves. If it is used to process long series data, choose the Transformer model.

[0049] S22. Based on the physical properties of the material, design a loss function that includes multiple sub-terms, such as the elastic modulus prediction error and the yield strength prediction error, and add L1 or L2 regularization terms.

[0050] S23. Based on the data volume and model complexity, select the batch size, such as 64 or 128, and randomly shuffle the data before each training cycle; the data volume D is obtained using the following formula:

[0051]

[0052] Where n represents the number of features; f iThis represents the weight factor for processing the i-th feature, which is set according to the importance of the feature; d i The dimension of the i-th feature is represented; N_processes represents the total number of data points; N_0 represents a baseline number of data points used to adjust the impact of the amount of data on the results.

[0053] The model complexity C is obtained using the following formula:

[0054]

[0055] Where L represents the number of model layers; p1 represents the number of parameters in the l-th layer; α represents the activation function complexity factor, which has different values ​​for different activation functions; al represents the additional complexity weight of the activation function in the l-th layer; γ represents the layer number influence factor; L n This represents the baseline layer number, used to adjust the contribution of the layer number to the complexity.

[0056] S24. For fully connected layers and LSTM layers, use the Xavier initialization method; for layers with ReLU activation functions, use the He initialization method.

[0057] S25. When the model prediction error exceeds the preset threshold, adjust the model parameters to make the prediction result closer to the true value. During the model training process, dynamically adjust the model parameters according to the reward and punishment signals of the prediction result.

[0058] The working principle of the above technical solutions is as follows: LSTM models are suitable for processing time-series data, such as temperature-time curves and stress-strain curves. LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) that solves the gradient vanishing problem in long-series data processing by introducing memory units (CellState). LSTM can capture dependencies in long-term sequences, making it very suitable for processing time-series data. Transformer models are suitable for processing long-series data. Proposed by Google, the Transformer model is a neural network model based on a self-attention mechanism for processing sequential data. Compared to traditional recurrent neural network models, Transformer models have better parallel performance and shorter training time, and can capture long-distance dependencies. Combining the physical properties of materials, a loss function with multiple sub-terms is designed, such as elastic modulus prediction error and yield strength prediction error. These sub-terms reflect the model's accuracy in predicting different physical properties. L1 or L2 regularization terms are added to prevent overfitting, improving the model's generalization ability by constraining the complexity of model parameters. L1 regularization (Lasso regression) and L2 regularization (ridge regression) are two commonly used regularization methods. Choose an appropriate batch size based on the amount of data and model complexity, such as 64 or 128. The choice of batch size affects the model's training speed and convergence performance. Randomly shuffle the data before each training cycle; this helps the model learn the global features of the data and avoid getting trapped in local optima. Use the Xavier initialization method for fully connected layers and LSTM layers. The Xavier initialization method initializes weights based on the layer's input and output dimensions, helping to maintain consistent data distribution across layers. For layers with ReLU activation functions, use the He initialization method. The He initialization method considers the characteristics of the ReLU activation function and can more effectively maintain data distribution. When the model's prediction error exceeds a preset threshold, adjust model parameters (such as learning rate and weights) to make the prediction result closer to the true value. This can be achieved through optimization algorithms such as gradient descent. During model training, dynamically adjust model parameters based on the reward / penalty signals of the prediction results (such as accuracy and loss value). This can be achieved through methods such as reinforcement learning, enabling the model to better adapt to changes in the data.

[0059] The advantages of the above technical solution are as follows: The solution allows for flexible selection of deep learning frameworks. For example, the LSTM model is suitable for time series data (such as temperature-time curves and stress-strain curves), while the Transformer model is suitable for processing long-sequence data. This flexibility ensures that the model can optimally process different types of data. By selecting the most suitable deep learning framework, the model's predictive performance and accuracy can be significantly improved, thus better meeting the needs of practical applications. The design of the loss function incorporates the physical properties of the material, such as the prediction error of elastic modulus and the prediction error of yield strength, which helps the model more accurately capture the mechanical behavior of the material. By adding L1 or L2 regularization terms, overfitting of the model can be effectively prevented, improving the model's generalization ability. Selecting an appropriate batch size based on the amount of data and model complexity can balance the model's training speed and convergence effect, ensuring that the model reaches optimal performance within a reasonable time. Randomly shuffling the data before each training cycle helps the model learn the global features of the data and avoid getting trapped in local optima. The Xavier initialization method and the He initialization method are used to target different data types. Initializing weights for fully connected layers, LSTM layers, and ReLU activation functions helps maintain consistent data distribution across layers, thereby improving model training efficiency and performance. When the model's prediction error exceeds a preset threshold, adjusting model parameters helps bring the prediction closer to the true value, ensuring the model's prediction accuracy. During model training, dynamically adjusting model parameters based on reward / penalty signals from the prediction results further enhances the model's adaptability and accuracy. The comprehensive application of these techniques significantly improves the model's prediction accuracy, providing more reliable data support for materials science research. The flexibility and dynamism of this approach facilitate rapid model adjustment and optimization, shortening the materials research and development cycle and improving efficiency. This technical solution offers new ideas and methods for materials science research, contributing to technological innovation and industrial upgrading. The data volume calculation formula dynamically calculates the required data volume D based on the number of features, feature importance, feature dimensionality, total number of data points, and a baseline data point count. This method considers the impact of feature importance and dimensionality on data volume, while using a logarithmic function to smoothly adjust the data volume to adapt to datasets of different sizes. The model complexity formula is used to evaluate the model complexity C, considering the number of model layers L, the number of parameters per layer, the activation function complexity factor α, the additional complexity weight of each activation function, the layer number influence factor γ, and the baseline layer number L0. This evaluation method considers not only the model structure (number of layers and number of parameters per layer) but also the impact of activation function complexity and layer number on overall complexity, thus more accurately reflecting the model's complexity. Based on the data volume and model complexity, the batch size (e.g., 64, 128, etc.) can be reasonably selected.Larger batch sizes can accelerate the training process but may lead to increased memory consumption and less precise gradient updates; smaller batch sizes can improve the model's generalization ability but may increase training time. Randomly shuffling the data before each training cycle helps the model learn the global distribution of the data, avoids overfitting, and improves the model's generalization performance. These formulas and methods offer flexibility and adaptability, allowing adjustments based on different datasets and model structures. By adjusting parameters such as feature weight factors, the number of baseline data points, and activation function complexity factors, the calculation of data volume and model complexity can be optimized, resulting in better training effects and performance. Through reasonable data volume calculation and model complexity evaluation, as well as the selection and optimization of batch size, training efficiency can be improved, training time reduced, while maintaining or improving model performance. This is particularly important for large-scale datasets and complex models. In summary, the above formulas and methods for selecting data batch size have significant beneficial effects in the fields of machine learning and deep learning, enabling dynamic adjustment of data volume, accurate evaluation of model complexity, and optimization of batch size selection, thereby improving training efficiency and model performance.

[0060] In one embodiment of the present invention, S25 includes:

[0061] After each training cycle, the prediction error of the model on the validation set is calculated. The prediction performance of the model is evaluated by comparing the difference between the predicted value and the true value. The prediction error is calculated using the following formula:

[0062]

[0063] Among them, E total N represents the total prediction error on the validation set; N represents the number of samples on the validation set; M represents the number of other key physical properties besides the elastic modulus and yield strength; w e and processing w y These are the weights representing the prediction errors of elastic modulus (Elastic Modulus) and yield strength (Yield Strength). This represents the weight of the prediction error of other physical characteristics in the j-th processing; The elastic modulus prediction error for the i-th sample can be defined as follows: in Represents the true value. Indicates the predicted value; This represents the yield strength prediction error for the i-th sample, defined in the same way as... similar; This represents the prediction error of another physical characteristic of the i-th sample in the j-th processing, defined in the same way as the processing. similar;

[0064] Based on the magnitude and distribution of prediction errors, a strategy for adjusting model parameters is formulated; during model training, model parameters are updated using the gradient descent algorithm; and model parameters are dynamically adjusted based on the reward / penalty signals from the prediction results.

[0065] By using a learning rate decay strategy, a larger learning rate is set in the early stages of training to achieve rapid convergence; and as training progresses, the learning rate is gradually reduced to adjust the model parameters.

[0066] Add an L1 or L2 regularization term to the loss function and dynamically adjust the weight of the regularization term based on the model's predictive performance.

[0067] During model training, the changes in the loss function, the prediction error on the validation set, and the updates of the model parameters are monitored in real time. If the prediction error on the validation set does not decrease significantly after several consecutive training cycles, or shows an upward trend, an early stopping strategy is adopted to avoid model overfitting.

[0068] The model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) are tuned using grid search; the optimal hyperparameter combination is found by comparing the model performance under different combinations of hyperparameters.

[0069] The model's performance is evaluated on a test set. By comparing the performance of different models (such as LSTM, Transformer, etc.), the optimal model architecture is selected. Based on the evaluation results on the test set, the model is iteratively optimized.

[0070] The working principle of the above technical solution is as follows: After each training cycle, the model is used to make predictions using a validation set, and the prediction errors of key physical properties (such as elastic modulus, yield strength, etc.) are calculated. By comparing the differences between the predicted values ​​and the true values, the model's prediction performance can be evaluated. This step is a crucial part of model performance monitoring, helping to identify model problems in a timely manner. Based on the magnitude and distribution of the prediction errors, a strategy for adjusting model parameters is formulated. If the prediction errors are generally large, it may indicate insufficient model complexity; in this case, increasing the number of network layers or neurons can be considered to improve the model's fitting ability. If the prediction errors are large within certain specific intervals, more detailed parameter adjustments may be needed for these intervals, such as adjusting the weight initialization method and learning rate. During model training, the model parameters are updated using the gradient descent algorithm. The gradient descent algorithm dynamically adjusts the model parameters based on the reward / penalty signal of the prediction results (i.e., the gradient of the prediction error), bringing the prediction results closer to the true values. This step is the core of model optimization; by continuously iterating and updating the parameters, the prediction error can be gradually reduced. The learning rate is one of the key factors affecting the model's training speed and effectiveness. In the early stages of training, setting a larger learning rate can accelerate the model's convergence speed. However, as training progresses, an excessively large learning rate may cause the model to oscillate around the optimal solution and fail to converge stably. Therefore, employing a learning rate decay strategy, gradually reducing the learning rate, helps the model to finely adjust parameters in the later stages of training, achieving better convergence results. Adding L1 or L2 regularization terms to the loss function can effectively prevent overfitting. The weights of the regularization terms can be dynamically adjusted based on the model's predictive performance. When the model's prediction error on the validation set is large, the weights of the regularization terms can be appropriately increased to limit the complexity of the model parameters; when the model's prediction error on the validation set is small, the weights of the regularization terms can be appropriately decreased to allow for more flexible adjustment of the model parameters. During model training, the changes in the loss function, the prediction error on the validation set, and the updates to the model parameters should be monitored in real time. If the prediction error on the validation set does not decrease significantly after several consecutive training epochs, or shows an upward trend, an early stopping strategy should be adopted to avoid model overfitting. The early stopping strategy is an effective method to prevent model overfitting; it can stop training in time before the model's performance begins to decline, thereby preserving the model's optimal performance. The model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) are tuned using grid search. Grid search is an exhaustive search method that tries all possible combinations of hyperparameters and compares the model performance under different combinations to find the optimal combination. Although this step is computationally intensive, it ensures that a globally optimal solution or a near-global optimal parameter combination is found. The model's performance is then evaluated on a test set, including metrics such as prediction accuracy and stability.By comparing the performance of different models (such as LSTM, Transformer, etc.), the optimal model architecture is selected. Based on the evaluation results on the test set, the model is iteratively optimized, including adjusting model parameters and optimizing network structure, to further improve the model's performance.

[0071] The effects of the above technical solution are as follows: After each training cycle, the prediction error of the model on the validation set is calculated and evaluated, which helps to identify problems in the model in a timely manner and take corresponding optimization measures; based on the magnitude and distribution of the prediction error, a fine-grained parameter adjustment strategy can be formulated, such as increasing model complexity or adjusting parameters for specific intervals, which can significantly improve the model's prediction performance; by gradually reducing the learning rate, it can be ensured that the model converges quickly in the early stage of training and that parameters can be adjusted more finely in the later stage, thereby improving the model's stability and accuracy; reasonable learning rate management helps to avoid the model oscillating around the optimal solution and reduces the risk of overfitting; adding L1 or L2 regularization terms to the loss function can effectively prevent the model from overfitting and improve the model's generalization ability; dynamically adjusting the weight of the regularization term according to the model's prediction performance can further optimize the model. Model performance is assessed through real-time monitoring of the loss function, prediction error on the validation set, and model parameter updates during training. This helps in timely identification and resolution of problems. An early stopping strategy is employed: training is stopped promptly when the prediction error on the validation set fails to decrease significantly or shows an upward trend after multiple training epochs to prevent overfitting. Grid search is used to fine-tune the model's hyperparameters, finding the optimal hyperparameter combination to improve performance. The performance of different models is evaluated on the test set, and the optimal model architecture is selected to ensure accuracy and stability in practical applications. Iterative optimization based on test set evaluation results continuously improves model performance. By continuously experimenting with new model architectures and parameter tuning strategies, the model can be continuously improved to adapt to different application scenarios and data characteristics. The above formula considers not only the two important physical properties of elastic modulus and yield strength, but also other possible key physical properties, thus enabling a more comprehensive evaluation of the model's predictive performance on different physical properties. Through weighting factors, the prediction errors of different physical properties can be weighted to varying degrees according to the needs of different application scenarios, thereby achieving customized evaluation of model performance. The prediction error in the formula is expressed as a relative error (i.e., the absolute value of the difference between the predicted value and the true value divided by the true value), which helps to compare physical properties with different dimensions and quantify the accuracy of model predictions. By calculating the total prediction error, the model's predictive performance on different physical properties can be clearly understood, thus guiding subsequent model optimization. For example, if the model's prediction error is found to be large on a certain physical property, the model structure or training strategy can be adjusted to improve the prediction accuracy of that property. Comprehensively considering the prediction errors of multiple physical properties and weighting and summing them helps to improve the reliability and stability of the model.Even if a model's prediction deviates on a certain characteristic, it can be compensated for to some extent by the prediction errors on other characteristics, thereby reducing the overall prediction error. The formula provides a unified framework for comparing and evaluating the performance of different models on the same validation set. By calculating the total prediction error, the prediction accuracy of different models on different physical characteristics can be intuitively compared, allowing for the selection of the better-performing model for subsequent applications. In summary, the above formula has significant beneficial effects in comprehensively evaluating model performance, providing flexibility and customizability, quantifying prediction errors, guiding model optimization, improving model reliability, and facilitating comparison and evaluation.

[0072] In one embodiment of the present invention, S3 includes:

[0073] S31. Convert the thermodynamic parameters into binary strings, where each bit represents a parameter value, and encode them using the real values ​​of the thermodynamic parameters.

[0074] S32. Combining multiple thermodynamic properties of the material, design a multi-objective fitness function and assign weights according to the importance of each property.

[0075] S33. By selecting temperature sensors, such as thermocouples and resistance temperature detectors, placed at key locations on the material, temperature data is collected in real time, and the collected temperature data is preprocessed.

[0076] S34. Based on real-time monitored temperature changes, design parameter update rules. For example, update the model parameters every time the temperature rises by a certain range. Set upper and lower limits for parameter adjustment based on the optimal parameter combination obtained from the genetic algorithm.

[0077] The working principle of the above technical solution is as follows: First, thermodynamic parameters (such as the coefficient of thermal expansion and thermal conductivity) are converted into binary strings. Each binary bit represents a parameter value, and these parameter values ​​are encoded using the real values ​​of the thermodynamic parameters. This encoding method ensures that each parameter can be uniquely represented as a binary string. A multi-objective fitness function is designed, combining multiple thermodynamic indicators of the material (such as prediction errors for the coefficient of thermal expansion and thermal conductivity). This function is used to evaluate the model's performance under different parameter combinations. Weights are allocated according to the importance of each indicator to ensure a balance in the performance of each thermodynamic indicator during optimization. Temperature sensors (such as thermocouples and resistance thermometers) are placed at key locations on the material to collect temperature data in real time. The collected temperature data is preprocessed, including filtering, denoising, and calibration, to ensure the accuracy and reliability of the data. Parameter update rules are designed based on the real-time monitored temperature changes. For example, the model parameters can be updated every time the temperature rises by a certain range. Simultaneously, upper and lower limits for parameter adjustment are set based on the optimal parameter combination obtained by the genetic algorithm to avoid excessive parameter adjustments that could lead to model instability. During the parameter update process, parameter values ​​can be gradually adjusted until the optimal parameter combination is found.

[0078] The advantages of the above technical solution are as follows: Converting thermodynamic parameters into binary strings, with each bit representing a parameter value, significantly improves parameter processing efficiency; binary encoding, with its discretization feature, accurately represents the parameter's value range, avoiding the precision loss that may occur with real-value encoding; using real-valued thermodynamic parameters ensures accuracy and authenticity; real-valued encoding also provides more possibilities for subsequent genetic algorithm operations (such as crossover and mutation), helping to find better parameter combinations; designing a multi-objective fitness function based on multiple thermodynamic indicators of the material (such as coefficient of thermal expansion and thermal conductivity) allows for comprehensive evaluation of model performance; this design helps find parameter combinations that perform well across multiple thermodynamic indicators, thus improving the overall model performance; weighting based on the importance of each indicator ensures a balance in the performance of each thermodynamic indicator during optimization; and the flexibility of weight allocation can be adjusted according to actual needs to adapt to different application scenarios. The key features of the data collection include: real-time temperature data acquisition via temperature sensors (such as thermocouples and RTDs) placed at critical locations on the material, ensuring data timeliness and accuracy; real-time data acquisition also provides strong support for subsequent model parameter updates; preprocessing the acquired temperature data (such as filtering and denoising) improves data reliability and stability; preprocessed data also reduces the impact of noise on model performance, thereby improving the model's prediction accuracy; designing parameter update rules based on real-time monitored temperature changes ensures the model can adapt to different temperature conditions; for example, updating model parameters every time the temperature rises by a certain range ensures stable performance during temperature changes; setting upper and lower limits for parameter adjustment based on the optimal parameter combination obtained from the genetic algorithm avoids model instability caused by excessive parameter adjustments; this measure also ensures stable performance during parameter updates, thereby improving model reliability and stability.

[0079] In one embodiment of the present invention, S31 includes:

[0080] Key parameters that significantly affect material properties, such as coefficient of thermal expansion, thermal conductivity, and heat capacity, are selected from a database of thermodynamic parameters. Based on the physical meaning of these parameters and experimental data, the effective range of values ​​for each parameter is determined.

[0081] For each selected thermodynamic parameter, a binary encoding scheme is designed; and the number of bits in the encoding is determined according to the parameter's value range and precision requirements; for example, if a parameter's value range is between 0 and 1, and it needs to be accurate to four decimal places, then a sufficiently long binary string can be used to represent this parameter.

[0082] Establish the mapping relationship between binary encoding and real values ​​of thermodynamic parameters; analyze the influence of different encoding bit lengths on the accuracy of parameter representation; evaluate whether the encoding accuracy meets the preset requirements by comparing experimental data and encoded parameter values;

[0083] The designed binary encoding scheme is applied to the encoding and decoding process of actual thermodynamic parameters. The accuracy and reliability of the encoding scheme are evaluated by comparing the original data and the encoded data.

[0084] Based on the test results, the coding scheme is optimized for performance. The coding scheme is continuously iterated and optimized based on actual application needs and feedback.

[0085] The working principle of the above technical solution is as follows: Key parameters that significantly affect material properties are selected from a database of thermodynamic parameters. These parameters typically include the coefficient of thermal expansion, thermal conductivity, and heat capacity, which have a significant impact on the material's thermal stability, thermal conductivity, and thermal response. Based on the physical meaning and experimental data of these parameters, the effective value range for each parameter is determined. For example, the coefficient of thermal expansion may have a stable value within a specific temperature range, while thermal conductivity may be affected by factors such as material composition and structure. For instance, assuming that the coefficient of thermal expansion of a certain material ranges from 1 × 10⁻⁵ (1 / ℃) at room temperature, the thermal conductivity ranges from 1 to 5 (W / m·K) at room temperature, and the heat capacity ranges from 500 to 1000 (J / kg·℃) at room temperature, these value ranges will serve as the basis for subsequent coding scheme design. A binary coding scheme is designed for each selected thermodynamic parameter. The number of bits in the code is determined based on the parameter's value range and accuracy requirements. For example, if a parameter's value ranges from 0 to 1 and requires accuracy to four decimal places, a sufficiently long binary string can be used to represent this parameter. The specific number of bits can be determined through calculation. For instance, a 16-bit binary number can represent integers between 0 and 65535. If 0 to 65535 is linearly mapped to 0 to 1 (while retaining four decimal places), the required precision can be achieved. Encoding schemes can use fixed-length or variable-length binary strings to represent different thermodynamic parameters. Fixed-length encoding schemes are suitable for parameters with relatively stable value ranges, while variable-length encoding schemes can be dynamically adjusted according to the parameter's value range. During encoding, the real values ​​of the thermodynamic parameters need to be quantized and converted into binary strings. Quantization can be performed using uniform or non-uniform quantization, depending on the parameter's value range and distribution characteristics. During decoding, the binary string needs to be restored to the real values ​​of the thermodynamic parameters. This restoration process can be achieved through inverse quantization, that is, converting the binary string into the corresponding real value according to the quantization rules and mapping relationships used during encoding. The impact of different encoding bit lengths on the parameter representation precision can be determined through theoretical analysis and experimental verification. Generally, the more bits in the encoding, the higher the precision of the parameter representation, but the computational complexity and storage requirements will also increase accordingly. The accuracy of the encoding scheme can be evaluated by comparing experimental data with the encoded parameter values. If the difference between the encoded parameter values ​​and the experimental data is within an acceptable range, the encoding scheme is considered to have sufficient accuracy. The encoding scheme is then evaluated to determine if it meets the preset precision requirements (e.g., four decimal places). If the encoding scheme cannot meet the preset precision requirements, it needs to be redesigned or the number of bits adjusted. The designed binary encoding scheme is then applied to the encoding and decoding of actual thermodynamic parameters.By comparing the original data and the encoded data, the accuracy and reliability of the encoding scheme can be evaluated. Based on the test results, the performance of the encoding scheme can be optimized. This includes adjusting the encoding bit depth, improving the encoding algorithm, or introducing new encoding techniques. Continuous optimization of the encoding scheme can improve its accuracy and reliability and meet the needs of practical applications. Based on practical application needs and feedback, the encoding scheme can be continuously iterated and optimized. This includes collecting more experimental data, analyzing the performance bottlenecks of the encoding scheme, and exploring new encoding techniques and methods. Through continuous iteration and optimization, it can be ensured that the encoding scheme remains in its optimal state.

[0086] The advantages of the above technical solution are as follows: By selecting key parameters that significantly affect material properties, such as the coefficient of thermal expansion, thermal conductivity, and heat capacity, from the material's thermodynamic parameter library, the amount of data processed subsequently can be greatly reduced, improving the efficiency of parameter processing. Simultaneously, these key parameters can more accurately reflect the thermodynamic properties of the material, providing more valuable information for subsequent analysis and optimization. Using a binary encoding scheme to encode thermodynamic parameters fully utilizes the advantages of computers in processing binary data, improving the speed and efficiency of data processing. By determining the number of bits in the encoding based on the parameter's value range and precision requirements, it can be ensured that the encoded data accurately represents the original parameter value, avoiding information loss and precision degradation. The number of bits in the encoding scheme can be dynamically adjusted according to the parameter's value range and precision requirements. This flexibility allows the encoding scheme to adapt to different parameters and precision requirements, improving its adaptability. By establishing a mapping relationship between binary encoding and the real values ​​of thermodynamic parameters, the accuracy and reliability of the encoding and decoding process can be ensured. Simultaneously, this mapping relationship helps analyze the impact of different encoding bit lengths on parameter representation accuracy, providing strong support for encoding scheme optimization. By comparing experimental data and encoded parameter values, the accuracy of the encoding scheme can be evaluated to see if it meets the preset requirements. If it does not meet the requirements, the encoding scheme can be adjusted and optimized to improve encoding accuracy. This accuracy evaluation and optimization process ensures the accuracy and reliability of the encoding scheme in practical applications, providing strong support for subsequent model parameter updates and genetic algorithm optimization. Performance optimization of the encoding scheme can be performed based on test results, including adjusting the encoding bit length, improving the encoding algorithm, or introducing new encoding techniques. These optimization measures can further improve the efficiency and accuracy of the encoding scheme. Furthermore, through continuous iteration and optimization, the encoding scheme can be made more suitable for practical application needs and feedback, improving its practicality and reliability. This technical solution provides strong support for the analysis and optimization of the thermodynamic properties of materials. By accurately representing and processing thermodynamic parameters, the research and development process of materials can be accelerated, and the performance and quality of materials can be improved. This technical solution can also promote technological innovation and development in related fields. For example, in industries such as energy, aerospace, and automobiles, there are high requirements for the thermodynamic properties of materials. By adopting this technical solution, technological progress and innovative development in these industries can be promoted.

[0087] In one embodiment of the present invention, step S4 includes:

[0088] S41. Based on the chemical composition of the material, determine the types and quantities of atoms in the simulation, and set relevant simulation parameters, such as initial temperature, pressure, and boundary conditions. Then, based on the material's properties, select the potential function, such as the Lennard-Jones potential or the Morse potential; process...

[0089] S42. Temperature is controlled during the simulation process using a Nose-Hoover hot tub, and pressure is controlled during the simulation process using the Parrinello-Rahman method.

[0090] S43. Based on the simulation accuracy and computational efficiency, the time step is selected, and microstructure images are periodically output during the simulation to observe the evolution of micro-damage; the simulation accuracy is obtained using the following formula:

[0091]

[0092] Where T represents the number of physical properties being compared (e.g., elastic modulus, yield strength, etc.); p i e represents the simulated predicted value of the i-th physical property; i R represents the experimental measurement value of the i-th physical property. i w represents the acceptable error range or reference value for the i-th physical characteristic being processed. i This represents the weighting factor for the i-th physical characteristic, reflecting the importance of that characteristic in the overall accuracy assessment;

[0093] The computational efficiency is obtained using the following formula:

[0094]

[0095] Where E represents computational efficiency; T r T represents the actual time required to complete the simulation; p This represents the simulated versus ideal or expected completion time; C represents the computational resource consumption coefficient.

[0096] S44. Compare the simulated microstructure images with the experimental microstructure images to analyze the similarities and differences in micro-damage.

[0097] S45. Compare the simulated physical properties (such as elastic modulus, yield strength, etc.) with the experimental data, and adjust the model parameters using gradient descent based on the differences between the simulation results and the experimental data.

[0098] The working principle of the above technical solution is as follows: Based on the chemical composition of the material, first determine the types and numbers of atoms involved in the simulation. This is the foundation of the simulation because different atoms have different physical and chemical properties, which will directly affect the simulation results. Next, set the initial temperature, pressure, and boundary conditions for the simulation. These parameters will determine the physical state of the simulation environment, thus affecting the microstructure and properties of the material. The selection of the initial temperature should be based on the actual working environment of the material or the research objective. The pressure setting should consider the mechanical properties of the material and the influence of the external environment. The selection of boundary conditions should reflect the actual constraints of the material, such as free boundaries, fixed boundaries, or periodic boundaries. Based on the characteristics of the material, select an appropriate potential function to describe the interactions between atoms. The potential function is a mathematical model describing the interatomic interaction forces in the simulation and is crucial to the accuracy of the simulation results. The Lennard-Jones potential is suitable for describing the interactions between inert gas atoms; the Morse potential is often used to describe the interactions between metal atoms. When selecting a potential function, its applicability, accuracy, and computational efficiency should be considered. For example, suppose we are studying the microscopic damage process of a certain metallic material. Based on the chemical composition of the material, we determined that the atoms involved in the simulation were metal atoms, and set the initial temperature to room temperature (e.g., 300K), the pressure to atmospheric pressure, and the boundary conditions to periodic boundaries. When selecting the potential function, we chose the Morse potential to describe the interactions between metal atoms. The temperature during the simulation was controlled using a Nose-Hoover thermostat. The Nose-Hoover thermostat is a commonly used temperature control method that restricts atomic motion by introducing a non-Newtonian term into the equation of motion, thus keeping the temperature fluctuating around the expected value. The pressure during the simulation was controlled using the Parrinello-Rahman method. The Parrinello-Rahman method is a pressure control method that simultaneously considers changes in atomic position and simulation cell size; it adapts to changes in internal pressure by adjusting the size and shape of the simulation cells, thereby maintaining pressure stability during the simulation. An appropriate time step was selected based on the simulation's accuracy and computational efficiency. The time step is the increment in time progression during the simulation, determining the simulation's precision and computational load. Smaller time steps improve accuracy but increase computational cost, while larger time steps improve computational efficiency but may decrease accuracy. Therefore, a trade-off between accuracy and efficiency must be struck when selecting the time step. During the simulation, microstructure images are periodically output. These images reflect the material's microstructure and damage evolution during the simulation. Observing these images allows us to understand the material's micro-damage mechanisms, damage propagation processes, and the impact of damage on material properties. The simulated microstructure images are then compared with experimentally obtained images.This comparison verifies the accuracy and reliability of the simulation results; by comparing the similarities and differences in microstructures, the effectiveness of the simulation method can be evaluated, providing a basis for subsequent optimization and improvement; the simulated physical properties (such as elastic modulus, yield strength, etc.) are compared with experimental data. This comparison further verifies the accuracy of the simulation results and evaluates the simulation method's ability to predict material properties; based on the differences between the simulation results and experimental data, model parameters are adjusted using optimization algorithms such as gradient descent. These parameters include the potential function parameters, initial conditions, boundary conditions, etc.; by adjusting these parameters, the difference between the simulation results and experimental data can be reduced, improving the simulation's precision and accuracy; based on the accuracy of the simulation results, the initial conditions and potential function are adjusted. These adjustments further optimize the simulation process, improving the reliability and accuracy of the simulation results; for example, if a large difference is found between the simulation results and experimental data, the potential function can be changed or the initial conditions adjusted.

[0099] The effects of the above technical solution are as follows: By determining the type and number of atoms in the simulation based on the chemical composition of the material, and setting precise simulation parameters (such as initial temperature, pressure, and boundary conditions) and selecting appropriate potential functions, this technical solution can accurately simulate the microstructure of the material. This helps to deeply understand the microscopic properties of the material and provides strong support for material design and optimization. By controlling the temperature and pressure during the simulation process using the Nose-Hoover hot bath and the Parrinello-Rahman method, respectively, this technical solution can ensure that the simulation environment fluctuates stably within the set range. This helps to reduce the uncertainty in the simulation process and improve the reliability and accuracy of the simulation results. By selecting an appropriate time step and periodically outputting microstructure images during the simulation, this technical solution can efficiently observe the evolution process of micro-damage. This helps to promptly detect potential damage in the material and provides an important basis for material life prediction and maintenance. By comparing the simulated microstructure images with the experimental microstructure images, and comparing the differences between the simulated physical properties and experimental data, this technical solution can verify the simulation results. The accuracy of the simulation results is improved, which helps ensure the effectiveness of the simulation method and provides a reliable guarantee for subsequent materials research and application. Based on the difference between the simulation results and experimental data, the model parameters are adjusted using optimization algorithms such as gradient descent, and the initial conditions and potential functions of the simulation are adjusted according to the accuracy of the simulation results. This technical solution can continuously improve the accuracy of the simulation. This helps to further optimize the simulation method, improve the reliability and accuracy of the simulation results, and provide stronger support for the research and application of materials science. This technical solution can be used to study the relationship between the microstructure and properties of materials, revealing the microscopic damage mechanism and evolution law of materials. This helps to provide a scientific basis for the design, preparation, and optimization of materials, and promotes the development of materials science. In engineering applications, this technical solution can be used to evaluate the durability and reliability of materials and predict the lifespan and performance changes of materials. This helps to provide important references for engineering design and maintenance, ensuring the safety and reliability of engineering projects. This technical solution can also be used for the development and research of new materials, predicting their performance in practical applications by simulating the microstructure and properties of new materials. This helps to accelerate the research and development process of new materials and promote the development of the new materials industry. The above formulas are used to evaluate the accuracy and computational efficiency of the simulation, and the time step and periodic output of microstructure images are selected based on these evaluation results during the simulation process.The simulation accuracy calculation formula comprehensively evaluates the overall accuracy of the simulation by considering the differences between the simulated predicted values ​​and experimental measurements of multiple physical characteristics, combined with the acceptable error range and weighting factor of each characteristic. The weighting factor reflects the importance of each physical characteristic in the overall accuracy evaluation, making the evaluation results more in line with the needs of practical applications. Through accuracy evaluation, problems in the simulation can be identified in a timely manner, such as large differences between predicted and experimental measurements, allowing for corresponding improvements and enhancing the reliability of the simulation. The accuracy evaluation results can serve as an important basis for selecting the time step. When the simulation accuracy is low, the time step can be appropriately reduced to improve accuracy; conversely, when the simulation accuracy is high, the time step can be appropriately increased to improve computational efficiency. The computational efficiency evaluation formula quantitatively evaluates the computational efficiency of the simulation by comparing the actual time required to complete the simulation with the expected completion time, combined with the computational resource consumption coefficient. The computational efficiency evaluation results can serve as a basis for optimizing the allocation of computational resources. When the computational efficiency is low, increasing computational resources or optimizing the algorithm can be considered to improve efficiency; conversely, when the computational efficiency is high, resources can be rationally allocated to perform more simulation or analysis work. By periodically evaluating computational efficiency, bottlenecks in the simulation process can be identified promptly, such as insufficient computational resources or inefficient algorithms, allowing for targeted improvements. The combined application of accuracy and computational efficiency evaluation formulas can improve simulation accuracy while optimizing computational efficiency, resulting in higher-quality simulation results. Regularly outputting microstructure images during the simulation, combined with the accuracy and computational efficiency evaluation results, enables more accurate observation and analysis of the evolution of microscopic damage, providing strong support for research in materials science and engineering.

[0100] In one embodiment of the present invention, S41 includes:

[0101] The elemental composition of the material is determined by chemical composition analysis (such as mass spectrometry, energy dispersive spectroscopy, etc.); based on the elemental composition, the types of atoms that need to be included in the simulation are identified, and the proportion of each type of atom is determined.

[0102] Based on the actual working environment of the material, set the initial temperature, pressure and boundary conditions for the simulation;

[0103] Choose an appropriate potential function based on the crystal structure and interatomic interaction characteristics of the material; for commonly used potential functions such as Lennard-Jones potential and Morse potential, their parameters (such as well depth, equilibrium distance, etc.) need to be adjusted according to the characteristics of the material.

[0104] By changing the initial conditions (such as temperature and pressure) and the potential function parameters, the effects of these changes on the simulation results are analyzed, and the parameters that have a significant impact on the simulation results are identified.

[0105] Validate the simulation using known experimental data or theoretical results; compare the simulation results with experimental data or theoretical predictions to assess the accuracy and reliability of the simulation; adjust the simulation parameters based on the validation results.

[0106] The initial positions of atoms are determined based on the crystal structure and density of the material; initial velocities are assigned to atoms based on the initial temperature and the principles of statistical mechanics.

[0107] The boundary conditions are refined to ensure the accuracy and stability of the simulation results; processing

[0108] Select appropriate simulation software and algorithms based on the complexity of the simulation and available computing resources; allocate computing resources (such as CPU, GPU, memory, etc.) reasonably based on the scale of the simulation and computational requirements.

[0109] The working principle of the above technical solution is as follows: Mass spectrometry, energy dispersive spectroscopy, and other analytical methods are used to determine the elemental composition of the material. These methods infer the elemental composition by measuring the energy of light or particles emitted or absorbed by the material at specific wavelengths. Atom type and quantity determination: Based on the elemental composition, the types of atoms to be included in the simulation are identified, and the proportion of each type of atom is calculated. This is usually based on the chemical formula or molecular formula of the material. For example, if the material is an aluminum alloy, the main atomic type is aluminum (Al) and other possible alloying elements (such as copper Cu, magnesium Mg, etc.). Through chemical composition analysis, the content of each element can be determined, thereby determining the proportion of each type of atom in the simulation. The initial temperature, pressure, and boundary conditions of the simulation are reasonably set according to the actual working environment of the material. The selection of the initial temperature needs to consider the phase transition point and thermal stability of the material to ensure that no unexpected phase transitions occur during the simulation. The initial pressure should reflect the stress state of the material in real applications. The boundary conditions are reasonably set according to the specific problem being simulated (such as tension, compression, shear, etc.) to simulate real physical processes; and an appropriate potential function is selected based on the crystal structure of the material and the interaction characteristics between atoms. For commonly used potential functions such as the Lennard-Jones potential and the Morse potential, their parameters (such as well depth and equilibrium distance) need to be adjusted according to the material characteristics to ensure the accuracy of the simulation results. For example, for metallic materials, the Morse potential may be a suitable choice because it can well describe the interactions between metallic atoms. When adjusting parameters, the potential function parameters can be optimized by comparing the simulation results with known experimental data or theoretical predictions; the simulation can be verified using known experimental data or theoretical results. The accuracy and reliability of the simulation can be evaluated by comparing the differences between the simulation results and experimental data or theoretical predictions; based on the verification results, the simulation parameters (such as initial conditions and potential function parameters) can be adjusted to improve the accuracy and reliability of the simulation; the initial positions of atoms are generated according to the crystal structure and density of the material. This usually involves the calculation of the lattice constant and the arrangement of atoms in the lattice; initial velocities are assigned to atoms according to the initial temperature and statistical mechanics principles. This usually involves the relationship between temperature and kinetic energy and the application of the Maxwell-Boltzmann distribution; the boundary conditions are refined to ensure the accuracy and stability of the simulation results. For example, under periodic boundary conditions, it is necessary to ensure that the interactions of atoms at the boundary are consistent with those of the interior atoms to avoid the influence of boundary effects on the simulation results. Appropriate simulation software and algorithms should be selected based on the complexity of the simulation and computational resources. This involves evaluating the simulation scale, computational requirements, and the performance of the software and algorithms; for example, large-scale molecular dynamics simulations may require the use of high-performance computing clusters and parallel algorithms to improve computational efficiency. When selecting simulation software, factors such as functionality, ease of use, and compatibility with other tools need to be considered.

[0110] The advantages of the above technical solution are as follows: Firstly, determining the elemental composition of the material through chemical composition analysis, and accordingly identifying the types and proportions of atoms required in the simulation, accurately reflects the material's microstructure. Secondly, setting initial temperature, pressure, and boundary conditions based on the material's actual working environment, and selecting appropriate potential functions and adjusting their parameters, allows for more accurate simulation of the material's physical and chemical behavior. Thirdly, changing initial conditions and potential function parameters, and analyzing their impact on simulation results, identifies parameters that significantly affect the simulation results, thereby optimizing the simulation settings. Fourthly, using known experimental data or theoretical results to verify the simulation allows for the evaluation of its accuracy and reliability, and adjusting simulation parameters based on the verification results further improves the reliability of the simulation results. Fifthly, selecting appropriate simulation software and algorithms, and rationally allocating computing resources (such as CPU, GPU, memory, etc.) based on the complexity of the simulation and available computational resources, ensures efficient operation of the simulation process while avoiding resource waste. Finally, refining the boundary conditions further enhances the simulation. For example, ensuring that the interactions of atoms at the boundary under periodic boundary conditions are consistent with those of the internal atoms can reduce the impact of boundary effects on simulation results and enhance the stability and accuracy of the simulation results. This technical solution provides a powerful tool for materials science research, enabling in-depth exploration of the relationship between the microstructure and properties of materials, and providing a scientific basis for the design and development of new materials. Simultaneously, this technical solution can also provide reliable simulation results for engineering applications, guiding material selection and performance optimization, and improving product quality and reliability. In the field of materials development, this technical solution can accelerate the research and development process of new materials, predicting material performance and behavior through simulation, and reducing experimental costs and time. In engineering applications, this technical solution can assess the durability and reliability of materials, predict performance changes under specific working environments, and provide important references for engineering design and maintenance. In the field of scientific research, this technical solution can reveal the microscopic damage mechanisms and evolution laws of materials, providing a theoretical basis for the optimization and improvement of material performance.

[0111] In one embodiment of the present invention, step S5 includes:

[0112] S51. Divide the original dataset into training set, validation set and test set according to the time sequence; use the early data as the training set, the middle data as the validation set and the later data as the test set.

[0113] S52. Check the consistency of data distribution among the training set, validation set and test set, and select evaluation metrics according to the model's prediction target, such as accuracy, recall, F1 score and mean squared error (MSE).

[0114] S53, and through k-fold cross-validation, comprehensively evaluate the predictive performance of the model, analyze the influence of model parameters on the prediction results, identify key parameters, and analyze the importance of features to the prediction results to identify key features;

[0115] S54. Based on the analysis results, adjust the hierarchical structure, activation function, loss function, etc. of the model, and optimize the feature selection and extraction strategy.

[0116] S55. Set multiple convergence criteria, such as the prediction performance improvement being less than a set threshold after multiple consecutive iterations, or the model parameter changes tending to stabilize. When the set convergence criteria are met, stop the iteration and output the final model.

[0117] The working principle of the above technical solution is as follows: The original dataset is divided into training, validation, and test sets to ensure the independence of model training, validation, and testing, thereby obtaining a more accurate model performance evaluation. The dataset is divided according to time sequence to ensure uniform data distribution. Early data is used as the training set for model learning; mid-term data is used as the validation set to adjust model parameters and select the best model; and late-term data is used as the test set to evaluate the final model performance. It is ensured that the data distribution among the training, validation, and test sets is consistent to avoid bias in model performance evaluation due to differences in data distribution. Appropriate evaluation metrics are selected based on the model's prediction objective, such as accuracy, recall, F1 score (for classification problems), or mean squared error (MSE, for regression problems). These metrics objectively reflect the model's predictive performance. The training set is further divided into k complementary subsets. Each time, k-1 subsets are selected as training data, and the remaining subset is used as validation data to evaluate the model. This process is repeated k times, each time selecting a different subset as validation data. The average of the k test metrics is then calculated as an estimate of the model's accuracy. K-fold cross-validation is used to comprehensively evaluate the model's predictive performance. Simultaneously, the influence of model parameters on the prediction results is analyzed to identify key parameters; the importance of features to the prediction results is also analyzed to identify key features. Based on the analysis results of k-fold cross-validation, the model's hierarchical structure, activation function, and loss function are adjusted to improve predictive performance. Feature selection and extraction strategies are optimized, selecting features more important to the prediction results for model training to improve efficiency and accuracy. Multiple convergence criteria are set, such as the improvement in prediction performance being less than a set threshold after multiple iterations, and model parameter changes stabilizing. These criteria are used to determine whether the model has converged to the optimal solution. When the set convergence criteria are met, the iteration process stops, and the final model is output. This avoids the model overfitting problem caused by overtraining.

[0118] The effects of the above technical solution are as follows: By dividing the training, validation, and test sets according to time sequence and ensuring the uniformity of data distribution, data leakage and overfitting problems can be effectively avoided, enabling the model to better generalize to unseen data; using early data as the training set, mid-term data as the validation set, and late-term data as the test set helps simulate data flow and changes in the real world, making the model more adaptable to practical application scenarios; checking the consistency of data distribution among the training, validation, and test sets ensures that the model's performance is consistent and reliable across different datasets; selecting appropriate evaluation metrics based on the model's prediction objectives, such as accuracy, recall, F1 score, and mean squared error (MSE), can more accurately measure the model's performance and provide strong support for model optimization; and k-fold cross-validation can improve the predictive ability of the model. It enables comprehensive evaluation, reducing performance evaluation bias caused by improper data partitioning; it analyzes the importance of model parameters and features, helping to identify key parameters and features and providing important basis for model optimization; it adjusts the model's hierarchical structure, activation function, and loss function based on the analysis results of k-fold cross-validation, which can significantly improve model performance; it optimizes feature selection and extraction strategies, reducing the impact of redundant features on model performance and improving model efficiency and accuracy; it sets multiple convergence criteria, such as the prediction performance improvement being less than a set threshold after multiple consecutive iterations, and the model parameter changes tending to stabilize, ensuring that the model stops iterating at an appropriate time and avoiding overfitting caused by overtraining; when it stops iterating and outputs the final model, it ensures that the model has achieved optimal performance on the given dataset, providing reliable prediction results for practical applications.

[0119] One embodiment of the present invention, such as Figure 2 As shown, an adaptive high-temperature damage model construction system is provided, the system comprising:

[0120] Data acquisition module: Acquires experimental data of materials under high temperature environment, including stress-strain curves, temperature-time curves, and microstructure images. The collected data is preprocessed using machine learning algorithms, including data cleaning, missing value imputation and feature selection, and key damage features are extracted.

[0121] Model building module: Based on the deep learning framework, a basic damage prediction model is built. The model is trained using preprocessed data, and through the reinforcement learning mechanism, the basic damage prediction model adaptively adjusts its internal parameters according to the prediction error.

[0122] Parameter adjustment module: Uses a genetic algorithm to perform a global search for the thermodynamic parameters in the model, finds the optimal parameter combination, and dynamically adjusts the model parameters in combination with real-time temperature monitoring data;

[0123] Process simulation module: Based on molecular dynamics simulation technology, it simulates the microstructure evolution process of materials at high temperatures, compares the simulation results with experimental data, and further optimizes the model parameters through a feedback mechanism;

[0124] Iterative optimization module: Select an independent validation dataset to validate the basic damage prediction model and evaluate its prediction performance. Based on the evaluation results, iteratively optimize the model.

[0125] The working principle of the above technical solution is as follows: Data is acquired from material experiments under high-temperature conditions. This data includes stress-strain curves (describing the deformation of the material under stress), temperature-time curves (describing temperature changes during the experiment), and microstructure images (showing the microstructure changes of the material at high temperatures). Noise and outliers are removed from the data to ensure accuracy and consistency. For missing data, interpolation or other algorithms are used to fill in the gaps to ensure data integrity. Key features that significantly influence material damage prediction, such as peak stress, strain rate, and temperature gradient, are extracted from the raw data. A basic damage prediction model is built based on deep learning frameworks (such as TensorFlow and PyTorch). This model can receive preprocessed data as input and output predictions of material damage. The model is trained using the preprocessed data to learn the features and patterns in the data. A reinforcement learning mechanism is introduced to enable the model to automatically adjust its internal parameters based on prediction errors to improve prediction accuracy. A genetic algorithm is used to perform a global search of the thermodynamic parameters in the model to find the optimal parameter combination. Genetic algorithms, by simulating natural selection and genetic mechanisms, can efficiently search the solution space; combined with real-time temperature monitoring data, model parameters are dynamically adjusted. This ensures that the model maintains high prediction accuracy under different temperature conditions; based on molecular dynamics simulation techniques, the microstructural evolution of materials at high temperatures is simulated. This helps to understand the damage mechanism of materials at high temperatures; comparing simulation results with experimental data, and further optimizing model parameters through a feedback mechanism, ensures that the model can more accurately reflect the actual damage situation of materials; independent validation datasets are selected to validate the basic damage prediction model. This allows for the evaluation of the model's predictive performance on unknown data; the model is iteratively optimized based on the validation results. If the prediction accuracy does not meet the requirements, the model structure, parameters, or data preprocessing methods are adjusted, and retraining and validation are performed until satisfactory prediction accuracy is achieved.

[0126] The effects of the above technical solution are as follows: Preprocessing the collected data using machine learning algorithms, including data cleaning, missing value imputation, and feature selection, effectively removes noise and outliers, ensuring data accuracy and consistency. Extracting key damage features helps the model better learn the features and patterns in the data, thereby improving prediction accuracy. This method builds the model entirely based on experimental data, avoiding the problems of relying on numerous assumptions and simplifications in traditional methods, thus improving the model's reliability and practicality. Through reinforcement learning, the basic damage prediction model can adaptively adjust its internal parameters based on prediction errors, continuously optimizing model performance. This adaptive adjustment capability allows the model to maintain high prediction accuracy when facing different high-temperature environments and material types. Using a genetic algorithm to perform a global search on the thermodynamic parameters in the model can find the optimal parameter combination, further improving the model's prediction accuracy. Dynamically adjusting model parameters in conjunction with real-time temperature monitoring data enables… The model can adapt to environmental changes in real time, improving its practicality and robustness. Based on molecular dynamics simulation technology, the microstructural evolution of materials at high temperatures is simulated, contributing to a deeper understanding of the damage mechanisms at these temperatures. This simulation technology provides strong theoretical support for model construction and optimization. Comparing simulation results with experimental data and using a feedback mechanism to further optimize model parameters allows the model to more accurately reflect the actual damage situation of materials. This feedback optimization mechanism helps improve the model's prediction accuracy and generalization ability. Selecting independent validation datasets to validate the basic damage prediction model allows for an objective evaluation of its predictive performance. This validation method helps identify problems and shortcomings in the model, providing direction for subsequent optimization. Iterative optimization based on validation results involves continuously adjusting the model structure, parameters, or data preprocessing methods until satisfactory prediction accuracy is achieved. This iterative optimization method enables continuous improvement and refinement of the model, enhancing its practicality and accuracy.

[0127] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing an adaptive high-temperature damage model, characterized in that, The method comprises: S1, obtaining experimental data of the material under high temperature environment, preprocessing the collected data through machine learning algorithm, and extracting key damage characteristics; the experimental data includes stress-strain curve, temperature-time curve and microstructure image; S2, constructing a basic damage prediction model based on a deep learning framework, training the model using the preprocessed data, and adjusting the internal parameters of the basic damage prediction model according to the prediction error through a reinforcement learning mechanism; S3, using genetic algorithm to perform global search on the thermodynamic parameters in the model, finding the optimal parameter combination, and dynamically adjusting the model parameters combined with real-time temperature monitoring data; S4, based on molecular dynamics simulation technology, simulating the microstructure evolution process of the material under high temperature, comparing the simulation results with the experimental data, and further optimizing the model parameters through feedback mechanism; S5, selecting an independent verification data set to verify the basic damage prediction model, evaluating its prediction performance, and iteratively optimizing the model according to the evaluation results; The S2 comprises: S21, selecting a deep learning framework, selecting an LSTM model if used for processing time series data, and selecting a Transformer model if used for processing long sequence data; S22, combining the physical properties of the material, designing a loss function containing multiple sub-items, and adding L1 or L2 regularization term; S23, selecting batch size according to data volume and model complexity, and randomly shuffling data before each training cycle; S24, using Xavier initialization method for fully connected layer and LSTM layer, and using He initialization method for ReLU activation function layer; S25, when the model prediction error is greater than the preset threshold, adjusting the model parameters to make the prediction results closer to the true value, and dynamically adjusting the model parameters according to the reward and punishment signal of the prediction result during the model training process; The S3 comprises: S31, converting the thermodynamic parameters into binary strings, each bit representing a parameter value, and using the real value of the thermodynamic parameters for encoding; S32, combining multiple thermodynamic indicators of the material, designing a multi-objective fitness function, and assigning weights according to the importance of each indicator; S33, selecting temperature sensors arranged at key positions of the material to collect real-time temperature data, and preprocessing the collected temperature data; S34, according to the real-time monitored temperature change, designing parameter updating rule, setting the upper and lower limits of parameter adjustment according to the optimal parameter combination obtained by genetic algorithm.

2. The method of claim 1, wherein the high temperature damage model is constructed adaptively. The S1 comprises: S11, according to the thermal stability range of the material, setting temperature gradient, loading conditions, and setting different loading rates and stress levels under each loading condition; S12, after different temperature points and loading stages, obtaining microstructure images through SEM, and recording the evolution process of micro damage; S13, identifying outliers based on statistical methods, and further verifying the rationality of the data combined with the physical properties of the material; S14, extract the physical characteristics, thermal characteristics and microstructure characteristics of the material; S15, perform PCA analysis on the feature matrix, retain the principal components, and use LDA for feature selection in the classification problem.

3. The adaptive high-temperature damage model construction method according to claim 2, characterized in that, The S11 comprises: According to the thermal stability range of the material, the initial temperature gradient setting is carried out, and at the same time, the temperature gradient is fine-tuned based on the effect of the material at different temperatures; For each temperature gradient, design multiple loading conditions, and further refine the setting of loading rate and stress level under each loading condition; Through the temperature control system, the accuracy and stability of the temperature in the experimental process are controlled; in the heating process, the change of temperature is monitored in real time, and fine-tuning is carried out according to the need; During the experiment, the key data are recorded in real time through the data acquisition system, and the experimental conditions are fed back and adjusted according to the real-time monitoring data and experimental phenomena; According to the results and analysis of the preliminary experiment, the experimental conditions are further optimized; on the basis of experimental condition adjustment and optimization, the experimental scheme is iteratively updated.

4. The method of claim 1, wherein the method further comprises: The S25 comprises: After each training cycle, the prediction error of the model on the validation set is calculated, and the prediction performance of the model is evaluated by comparing the difference between the predicted value and the true value; According to the size and distribution of the prediction error, the strategy of adjusting the model parameters is formulated; in the model training process, the model parameters are updated through the gradient descent algorithm; according to the reward and punishment signal of the prediction result, the model parameters are dynamically adjusted; Through the learning rate decay strategy, set the learning rate to quickly converge in the early stage of training; and gradually reduce the learning rate to adjust the model parameters as the training proceeds; Add L1 or L2 regularization term to the loss function, and dynamically adjust the weight of the regularization term according to the prediction performance of the model; In the model training process, the change of loss function, the prediction error on the validation set and the update of model parameters are monitored in real time. If the prediction error on the validation set does not decrease significantly after continuous multiple training cycles, or shows an upward trend, the early stopping strategy is adopted to avoid model overfitting; Optimize the hyperparameters of the model through grid search; find the optimal combination of hyperparameters by comparing the performance of the model under different combinations of hyperparameters; Evaluate the performance of the model on the test set, select the optimal model architecture by comparing the performance of different models, and iteratively optimize the model according to the evaluation results on the test set.

5. The method of claim 1, wherein the method further comprises: The S31 comprises: Screen out the key parameters that significantly affect the performance of the material from the thermodynamic parameter library of the material, and determine the effective value range of each parameter according to the physical meaning and experimental data of these parameters; For each screened thermodynamic parameter, design a binary coding scheme; and the number of bits of the coding is determined according to the value range and accuracy requirement of the parameter; Establish the mapping relationship between binary coding and real value of thermodynamic parameter; analyze the influence of different coding bit number on parameter representation accuracy; evaluate whether the coding accuracy meets the preset requirement by comparing the experimental data and the coded parameter value; The designed binary encoding scheme is applied to the encoding and decoding process of actual thermodynamic parameters. By comparing the original data and the encoded data, the accuracy and reliability of the encoding scheme are evaluated. According to the test results, the performance of the encoding scheme is optimized, and the encoding scheme is iterated and optimized according to the actual application requirements and feedback.

6. The method of claim 1, wherein the method further comprises: The S4 comprises: S41, according to the chemical composition of the material, the type and number of atoms in the simulation are determined, and the related parameters of the simulation are set, and according to the characteristics of the material, the potential function selection is carried out; S42, the temperature in the simulation process is controlled by Nose-Hoover heat bath, and the pressure in the simulation process is controlled by Parrinello-Rahman method; S43, according to the accuracy and calculation efficiency of the simulation, the time step selection is carried out, and during the simulation process, the microstructure image is output regularly, and the evolution process of micro damage is observed; S44, the microstructure image obtained by simulation is compared with the microstructure image obtained by experiment, and the similarity and difference of micro damage are analyzed; S45, the difference between the physical characteristics obtained by simulation and the experimental data is compared, and the model parameters are adjusted by gradient descent according to the difference between the simulation results and the experimental data.

7. The method of claim 1, wherein the method further comprises: The S5 comprises: S51, according to the time sequence, the original data set is divided into training set, validation set and test set, the early data is used as training set, the middle data is used as validation set, and the late data is used as test set; S52, the data distribution consistency between the training set, the validation set and the test set is checked, and the evaluation index is selected according to the prediction target of the model; S53, and through k-fold cross-validation, the prediction performance of the model is comprehensively evaluated, the influence degree of the model parameters on the prediction results is analyzed, and the key parameters are identified; the importance of the features to the prediction results is analyzed, and the key features are identified; S54, according to the analysis results, the model is adjusted, and the feature selection and extraction strategy is optimized; S55, set multiple convergence standards, when the set convergence standard is met, stop iteration, and output the final model.

8. A system for implementing the adaptive high temperature damage model construction method according to claim 1, characterized in that, The system comprises: Data acquisition module: acquiring experimental data of materials in high temperature environment, preprocessing the collected data by machine learning algorithm, and extracting key damage features; Model construction module: based on deep learning framework, the basic damage prediction model is constructed, the model is trained by using the preprocessed data, and the basic damage prediction model is adjusted by reinforcement learning mechanism according to the prediction error; Parameter adjustment module: using genetic algorithm to search the optimal parameter combination in the model, combining with real-time temperature monitoring data, dynamically adjusting the model parameters; Process simulation module: based on molecular dynamics simulation technology, the microstructure evolution process of materials in high temperature is simulated, the simulation results are compared with the experimental data, and the model parameters are further optimized through feedback mechanism; Iterative optimization module: select independent verification dataset, verify the basic damage prediction model, and evaluate its prediction performance, and according to the evaluation result, the model is iteratively optimized.

Images