A method for predicting the high-temperature oxidation behavior of FeCrAl alloys

Abstract

This invention proposes a method for predicting the high-temperature oxidation behavior of FeCrAl alloys, comprising: Step S1, data collection, acquiring multiple sets of oxidation behavior data, including four alloy materials and two experimental conditions, and dividing them into training and test sets proportionally; Step S2, model construction, using a CNN-LSTM hybrid architecture to extract local features of composition, temperature, and time, and model time series characteristics, with the input being composition / temperature / time and the output being oxidation weight gain, and embedding relevant equations for modeling; Step S3, hyperparameter optimization, optimizing model parameters using a Bayesian algorithm, combining L2 regularization and cross-validation to prevent overfitting, and evaluating model performance using R²; Step S4, testing and prediction, validating the model using the test set, and learning the coupling effect between elements through data to achieve high-precision prediction of the high-temperature steam oxidation behavior of FeCrAl alloys.

Description

Technical Field

[0001] This invention belongs to the field of materials science, specifically relating to the prediction of high-temperature oxidation behavior of FeCrAl alloys, and more particularly to a predictive model of the oxidation behavior of FeCrAl alloys in a high-temperature steam environment based on machine learning methods, which is applied to the performance evaluation and optimization of accident-resistant fuel (ATF) cladding materials in the nuclear energy field. Background Technology

[0002] The 2011 Fukushima nuclear accident exposed serious safety hazards of traditional zirconium alloy cladding materials in high-temperature steam environments, mainly manifested in violent oxidation reactions and the accompanying excessive hydrogen generation, posing a significant threat to reactor safety. Against this backdrop, FeCrAl alloy, due to its significantly superior high-temperature oxidation resistance (especially in the 600-1400℃ range) and extremely low hydrogen yield, is considered one of the candidate materials for accident-resistant fuel (ATF) cladding.

[0003] However, under typical accident conditions such as loss-of-coolant accidents (LOCA), FeCrAl alloy cladding must withstand the severe corrosion test of a high-temperature steam / oxygen mixed environment, and its oxidation behavior directly affects the structural integrity of nuclear fuel assemblies and reactor safety. Current research faces two major technical bottlenecks: First, high-temperature corrosion experiments are costly and time-consuming, making it difficult to fully cover the multi-dimensional parameter space of alloy composition, temperature, and time, thus limiting the completeness and coverage of experimental data; second, traditional parabolic oxidation models have limited accuracy in predicting complex operating conditions and cannot accurately reflect the influence of alloy composition coupling effects, dynamic temperature changes, and other factors on oxidation behavior.

[0004] In recent years, machine learning technology has been increasingly applied in alloy design and performance optimization. Although its application in oxidation kinetics is still immature, existing research has shown that it can accurately predict material corrosion kinetic parameters under specific boundary conditions. Therefore, developing data-driven, high-precision corrosion behavior prediction models to overcome the limitations of traditional experiments and models has become a crucial scientific problem to be solved in the study of high-temperature oxidation behavior of FeCrAl alloys. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method for predicting the high-temperature oxidation behavior of FeCrAl alloys, comprising the following steps: Step S1: Data Collection: Collect multiple oxidation behavior data of FeCrAl alloys in a high-temperature steam environment. Each oxidation behavior data point includes alloy chemical composition, corrosion weight gain, oxidation temperature, and oxidation time. Divide the oxidation behavior data into a training set and a test set at an 80:20 ratio. Step S2: Model Construction: Construct a hybrid deep learning prediction model. The model adopts an architecture combining convolutional neural networks and long short-term memory networks. The convolutional neural network is used to extract local features of alloy chemical composition, oxidation temperature, and oxidation time from the oxidation behavior data, while the long short-term memory network is used to model the oxidation time series features. The input of the deep learning prediction model is alloy chemical composition, oxidation temperature, and oxidation time, and the output is corrosion weight gain. The corrosion weight gain satisfies the following equation:

[0006]

[0007] Where ΔW represents the weight gain due to corrosion, and E C Here, T is the activation energy of the material corrosion reaction, R is the absolute temperature, and A is the gas constant.

[0008] The weight gain due to oxidation is positively correlated with oxidation time, satisfying the following relationship;

[0009] ΔW = K P t n

[0010] Among them, K P is the material corrosion rate constant; t is the corrosion time, and n is the time exponent;

[0011] Step S3: Hyperparameter Optimization: The deep learning prediction model constructed in Step S2 is trained using the training set from Step S1. Bayesian optimization is used to adjust the model's hyperparameters, and L2 regularization and cross-validation are introduced to prevent overfitting. The evaluation equation for the deep learning prediction model is as follows:

[0012]

[0013] Among them, R 2 Here, y is the performance determination coefficient, n is the number of data points, and y is the coefficient of determination. i For experimental values, y i pre For the predicted value, y i ave The average value is the experimental value; Step S4 Test Prediction: The optimized deep learning prediction model is tested using the test set in Step S1, and evaluated using the evaluation equation in Step S3.

[0014] Furthermore, the alloys included in the oxidation behavior data are Cr, Al, Mo, and Nb.

[0015] Furthermore, analytical constraints are added in step S2, including: when Cr > 25 wt%, the alloy becomes more brittle and its corrosion resistance cannot be improved; when Cr < 7 wt%, a stable oxide film cannot be formed, and the material has low corrosion resistance; the Al content in the range of 2–6.5 wt% affects the corrosion weight gain of FeCrAl alloy; when Al < 2 wt%, a continuous Al2O3 oxide film cannot be formed; when Al > 7%, the material is prone to brittle fracture, and its corrosion resistance is difficult to improve.

[0016] The technical solution of the present invention has the following significant advantages over the prior art:

[0017] (1) Improve prediction accuracy and efficiency: By introducing an integrated learning algorithm to model the oxidation weight gain of FeCrAl alloys with different alloy compositions under different oxidation temperatures and oxidation times, it is possible to predict their high-temperature oxidation behavior efficiently and accurately. Compared with traditional empirical formulas or physical models, it has a higher fitting degree and generalization ability.

[0018] (2) It helps to understand the influence mechanism of key factors on oxidation behavior: This invention uses algorithms to interpret the output of machine learning models, which can quantify the influence of various alloying elements (such as Cr, Al, Mo, Nb, etc.) and environmental parameters (temperature, time, etc.) on oxidation behavior, thereby understanding the intrinsic relationship between key factors and oxidation performance.

[0019] (3) Save experimental resources and development cycle: The prediction model can effectively reduce the dependence on a large number of high-temperature and long-cycle experiments, significantly reduce the time and cost required for research and development of FeCrAl oxidation-resistant alloys, and is particularly suitable for applications with stringent material performance requirements such as nuclear energy. Attached Figure Description

[0020] Figure 1 This is a comparison between the training set data and the calculation results of the prediction model.

[0021] Figure 2 This is a comparison between the test set data and the calculation results of the prediction model. Detailed Implementation

[0022] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] This invention proposes a method for predicting the high-temperature oxidation behavior of FeCrAl alloys, comprising the following steps: Step S1: Data collection; collecting 195 data points on the oxidation behavior of FeCrAl alloys in a high-temperature water vapor environment, each data point including alloy chemical composition (wt.%), corrosion weight gain (in mg / dm³). 2 The oxidation behavior data were divided into training and testing sets in an 80:20 ratio, and included the following alloys: Cr, Al, Mo, and Nb. The entire dataset contained two experimental conditions (oxidation temperature and oxidation time) and four alloy element characteristics (Cr, Al, Mo, Nb, excluding Fe).

[0024] Step S2: Model Construction; Construct a hybrid deep learning prediction model. In this embodiment, the model adopts a hybrid architecture that combines a convolutional neural network (CNN) with a long short-term memory network (LSTM). The network includes an input layer, a hidden layer, and an output layer. The input layer receives six feature data, including the mass fractions of Cr, Al, Mo, and Nb, oxidation temperature, and oxidation time. The hidden layer is divided into two levels: the first level hidden layer contains 36 neurons, and the second level hidden layer contains 14 neurons, which are used to perform deep mapping on the extracted features. The output layer is a single neuron that outputs the predicted value of corrosion weight gain.

[0025] The collaborative workflow of this hybrid architecture first extracts local features from the six input feature data through a CNN layer, focusing on capturing the local correlation features between alloy chemical compositions and between compositions and environmental parameters (such as temperature and time). Then, the feature sequence extracted by the CNN is input into an LSTM layer, which models the dynamic changes in the time dimension during oxidation (such as the gradual effect of oxidation time accumulation on corrosion weight gain), achieving the fusion of local features and time-series features. Finally, the corrosion weight gain prediction result is obtained through mapping via a fully connected layer. The input to the deep learning prediction model is alloy chemical composition, oxidation temperature, and oxidation time, and the output is corrosion weight gain. The corrosion weight gain satisfies the following equation:

[0026]

[0027] Where ΔW represents the weight gain due to corrosion, and E C Here, T is the activation energy of the material corrosion reaction, R is the absolute temperature, and A is the gas constant.

[0028] The weight gain due to oxidation is positively correlated with oxidation time, satisfying the following relationship;

[0029] ΔW = K P t n

[0030] Among them, KP denoted as the material corrosion rate constant; t represents the corrosion time, and n is the time exponent. Furthermore, during model building, constraints on the proportions of each material component need to be added. When Cr > 25 wt%, the alloy becomes more brittle and corrosion resistance cannot be improved; when Cr < 7 wt%, a stable oxide film cannot be formed, resulting in low corrosion resistance. Al content in the range of 2–6.5 wt% affects the corrosion weight gain of FeCrAl alloys; when Al < 2 wt%, a continuous Al2O3 oxide film cannot be formed; when Al > 7%, the material is prone to brittle fracture, and corrosion resistance is difficult to improve. The coupling effects among the four elements Cr, Al, Mo, and Nb are automatically captured by the model through learning from the training data, without the need for manual pre-setting of coupling relationships, ensuring that the model can reflect the synergistic influence of multiple factors.

[0031] Step S3: Hyperparameter Optimization; The deep learning prediction model constructed in Step S2 is trained using the training set from Step S1. Bayesian optimization is used to adjust the model's hyperparameters. Optimized hyperparameters include, but are not limited to, the kernel size of the CNN layers, the number of hidden units in the LSTM layers, the learning rate, the batch size, and the activation functions of the network layers. Through iterative search using the Bayesian optimization algorithm, the hyperparameter combination that minimizes the model's training error is found. To prevent overfitting due to the large amount of data and numerous parameters, an L2 regularization term is introduced into the fully connected layers of the model. By applying a penalty term to the network weights, the excessively large values ​​of the weight parameters are limited. Simultaneously, combined with 5-fold cross-validation, the training set is further divided into 5 subsets. Four subsets are used alternately as the training subset, and one subset as the validation subset. Through multiple training and validation iterations, the model parameters are optimized, enhancing the model's robustness and generalization ability. The deep learning prediction model evaluation equation is as follows:

[0032]

[0033] Among them, R 2 Here, y is the performance determination coefficient, n is the number of data points, and y is the coefficient of determination. i For experimental values, y i pre For the predicted value, y i ave This represents the average of the experimental values.

[0034] Step S4: Test Prediction; Input the test set data divided in Step S1 into the optimized deep learning prediction model according to the feature format consistent with the training set. The model automatically outputs the corrosion weight gain prediction value corresponding to each test data point; Use the R² coefficient of determination evaluation equation in Step S3 to quantitatively compare the predicted value with the experimental value of the test set, and calculate the R² value of the test set; At the same time, draw a scatter plot of the predicted value and the experimental value (e.g., ... Figure 2As shown in the figure, the fitting effect of the model is intuitively displayed. The closer R² is to 1, the better the predictive performance of the FeCrAl alloy high-temperature oxidation behavior prediction model, and the more accurately the model can reflect the intrinsic relationship between alloy composition, temperature, time and corrosion weight gain.

[0035] Figure 1 and Figure 2 The figures show a comparison between the training set data and the test set data, and the results calculated by the prediction model. As can be seen from the figures, the correlation coefficient R... 2 All values ​​are above 0.996, indicating that the high-temperature oxidation behavior prediction model for FeCrAl alloy established in this invention not only has excellent fitting effect on training data, but also has extremely strong generalization ability. It can accurately predict the high-temperature oxidation behavior of FeCrAl alloy under new working conditions that have not been trained, effectively breaking through the technical bottlenecks of limited coverage of traditional experiments and insufficient prediction accuracy of traditional models.

[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the high-temperature oxidation behavior of FeCrAl alloys, characterized in that, Includes the following steps: Step S1 Data Collection: Collect multiple oxidation behavior data of FeCrAl alloy in a high-temperature water vapor environment. The oxidation behavior data includes alloy chemical composition, corrosion weight gain, oxidation temperature and oxidation time. Divide the oxidation behavior data into training set and test set in a ratio of 80:

20. Step S2: Model Construction: Construct a hybrid deep learning prediction model. This model employs an architecture combining convolutional neural networks (CNNs) and long short-term memory (LSTM) networks. The CNNs extract local features of the alloy chemical composition, oxidation temperature, and oxidation time from the oxidation behavior data, while the LTM networks model the oxidation time series features. The inputs to the deep learning prediction model are the alloy chemical composition, oxidation temperature, and oxidation time, and the output is the corrosion weight gain. The corrosion weight gain satisfies the following equation: Where ΔW represents the weight gain due to corrosion, and E C Here, T is the activation energy of the material corrosion reaction, R is the absolute temperature, and A is the gas constant. The weight gain due to oxidation is positively correlated with oxidation time, satisfying the following relationship; ΔW = K P t n Among them, K P is the material corrosion rate constant; t is the corrosion time, and n is the time exponent; Step S3 Hyperparameter Optimization: The deep learning prediction model constructed in Step S2 is trained using the training set described in Step S1. The model hyperparameters are adjusted using a Bayesian optimization algorithm, and an L2 regularization term and cross-validation method are introduced to prevent overfitting. The evaluation equation for the deep learning prediction model is as follows: Among them, R 2 The performance determination coefficient is y, where n is the number of data points. i For experimental values, y i pre For the predicted value, y i ave This represents the average of the experimental values. Step S4 Test Prediction: Test the optimized deep learning prediction model using the test set described in step S1, and evaluate it using the evaluation equation described in step S3.

2. The method for predicting the high-temperature oxidation behavior of FeCrAl alloys according to claim 1, characterized in that, The alloys included in the oxidation behavior data are Cr, Al, Mo, and Nb.

3. The method for predicting the high-temperature oxidation behavior of FeCrAl alloys according to claim 2, characterized in that, In step S2, analytical constraints are added, including: When Cr > 25 wt%, the alloy becomes more brittle and its corrosion resistance cannot be improved. When Cr < 7 wt%, a stable oxide film cannot be formed, and the material has low corrosion resistance. The Al content in the range of 2–6.5 wt% affects the corrosion weight gain of FeCrAl alloy. When Al < 2 wt%, a continuous Al2O3 oxide film cannot be formed. When Al > 7%, the material is prone to brittle fracture, and its corrosion resistance is difficult to improve.

Images