A method and system for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys
By constructing physical metallurgical enhancement features and using a multi-model fusion method, the problem of predicting the yield strength of iron-based superalloys under complex conditions was solved, achieving high-precision prediction over a wide temperature range and supporting alloy performance evaluation and material screening.
Patent Information
- Application Number
- CN202611620693.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to accurately predict the yield strength of iron-based superalloys under complex temperature conditions, especially under different compositions, heat treatment processes, and temperature conditions, which limits the applicability and prediction accuracy of the models.
A method for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges was constructed. By acquiring data on chemical composition, heat treatment process, and test temperature, physical metallurgical enhancement characteristics were constructed. The model was trained using an extreme random tree regression model, a limit gradient boosting tree model, and a lightweight gradient boosting machine model. Finally, equal-weighted fusion was performed to form a predictive system for the yield strength of γ′-strengthened iron-based superalloys.
It achieves high-precision yield strength prediction in the range of 25-900℃, reduces the model's sensitivity to data noise, and improves the stability and reliability of predictions, making it suitable for alloy performance evaluation and candidate material screening.
Smart Images

Figure CN123120107A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of high-temperature alloy performance prediction and materials informatics technology, specifically relating to a method and system for predicting the cross-temperature range yield strength of γ′-strengthened iron-based high-temperature alloys. Background Technology
[0002] Iron-based superalloys possess excellent high-temperature strength, oxidation resistance, and microstructural stability, making them widely used in energy, aerospace, petrochemical, and high-temperature load-bearing components. Yield strength is a crucial mechanical property for evaluating the resistance of iron-based superalloys to plastic deformation, directly affecting their load-bearing capacity and safe operating range at different service temperatures. Since iron-based superalloys typically withstand complex temperature environments ranging from room temperature to high temperatures during actual service, accurately predicting the yield strength at different test temperatures is of great significance for alloy composition design, heat treatment process optimization, candidate material screening, and service performance evaluation.
[0003] The yield strength of iron-based superalloys is influenced by chemical composition, heat treatment process, testing temperature, and their interactions. Different alloying elements can affect material strength through mechanisms such as solid solution strengthening, precipitation strengthening, and grain boundary strengthening; solid solution treatment and multi-stage aging treatment can alter the microstructure and precipitate characteristics of the alloy; and as the testing temperature increases, dislocation motion, strengthening phase stability, and matrix softening behavior also change. Therefore, the yield strength of iron-based superalloys typically exhibits a complex nonlinear relationship with composition, process, and testing temperature, which is difficult to accurately describe using simple empirical formulas, thus limiting the applicability and prediction accuracy of models.
[0004] With the rapid development of materials databases and machine learning technologies, predicting the yield strength of iron-based superalloys using data-driven methods has become feasible. Machine learning models such as extreme random tree regression, limit gradient boosting tree, and lightweight gradient boosting machine models possess strong capabilities for nonlinear mapping and complex feature interaction modeling. However, existing studies often directly use the original chemical composition and process parameters as model inputs, failing to adequately mine the underlying physical metallurgical information. For example, the combined effect of γ′ formation and strengthening elements, the Ti / Al ratio, the ratios of γ′ precipitation strengthening elements to matrix elements and solid solution strengthening elements to matrix elements, and specific mass ratios such as Ni / (Fe+Ni) have not been fully characterized. Therefore, it is necessary to develop a high-precision prediction method for the cross-temperature range yield strength of iron-based superalloys that integrates physical metallurgical characteristics. Summary of the Invention
[0005] This invention aims to provide a method and system for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges, overcoming the limitations of traditional prediction methods under complex compositions, heat treatment processes, and varying test temperatures. First, the chemical composition, heat treatment process, test temperature, and yield strength data of the γ′-strengthened iron-based superalloy are preprocessed, and physical metallurgical enhancement features such as element atomic fractions, γ′ precipitation strengthening element combinations, solid solution strengthening element combinations, and element atomic ratios are constructed. During the model training phase, an extreme random tree regression model, a limit gradient boosting tree model, and a lightweight gradient boosting machine model are trained separately, and the prediction results of the three models are fused with equal weights of 1 / 3, 1 / 3, and 1 / 3, respectively. Finally, the model prediction accuracy is verified through cross-validation and multiple evaluation indicators. The data preprocessing, physical metallurgical enhancement feature construction, and three-model fusion prediction process are encapsulated into an executable program for predicting the yield strength of the γ′-strengthened iron-based superalloy at the target test temperature.
[0006] Specifically, the first aspect of the present invention provides a method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys, comprising: Step S1: Obtain chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys from the literature to form the original dataset of γ′-strengthened iron-based superalloys; Step S2: Perform data preprocessing on the original dataset. The data preprocessing includes unit unification, chemical composition verification, missing data processing, and duplicate sample processing to form a γ′-strengthened iron-based superalloy yield strength modeling dataset for model training and evaluation. Step S3: Construct physical metallurgical enhancement features based on the chemical composition data and the heat treatment process data, and combine the original features with the physical metallurgical enhancement features to form a model input feature set; Step S4: Based on the model input feature set, train three models respectively: extreme random tree regression model, extreme gradient boosting tree model, and lightweight gradient boosting machine model, and perform equal weighted fusion of the yield strength prediction values output by the three models to obtain the fused prediction value of the three models; Step S5: Use cross-validation to evaluate the fusion prediction values of the three models, calculate the model evaluation index and visualize the prediction results; after the evaluation is completed, retrain the three models using all the modeling data, and encapsulate the data preprocessing, physical metallurgical enhancement feature construction and three-model fusion prediction process into an executable program.
[0007] The above-mentioned method for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges can cover multiple test temperature data from room temperature to high temperature conditions, such as 25℃, 650℃, and 700℃.
[0008] As a preferred embodiment, in the above-mentioned method for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges, the chemical composition data in step S1 includes the mass fractions of matrix elements, γ′ precipitation strengthening elements, solid solution strengthening elements, and grain boundary strengthening elements; more preferably, the chemical composition data also includes the mass fractions of other elements besides matrix elements, γ′ precipitation strengthening elements, solid solution strengthening elements, and grain boundary strengthening elements.
[0009] As a preferred embodiment, in the above-mentioned method for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges, the heat treatment process data in step S1 includes solution temperature, solution time, first-stage aging temperature, and first-stage aging time; more preferably, the heat treatment process data also includes at least one of second-stage aging temperature, second-stage aging time, third-stage aging temperature, and third-stage aging time. Solution treatment and first-stage aging are important heat treatment regimes commonly used in γ′-strengthened iron-based superalloys, significantly affecting the degree of recrystallization of the alloy matrix and the precipitation behavior of the γ′-strengthening phase. As essential parameters, they ensure the basic integrity of the process characteristics. Second-stage and third-stage aging are multi-stage aging regimes used for some alloys and are set as optional parameters. On the one hand, this adapts to a large number of data samples in the literature that only report single-stage aging processes, avoiding the discarding of effective data due to the lack of multi-stage aging parameters; on the other hand, when the sample contains multi-stage aging information, the model can capture the regulatory effect of multi-step aging on the size and distribution of precipitates, improving the accuracy of yield strength prediction for alloys that have undergone complex multi-stage aging treatment.
[0010] In the above-mentioned method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys, step S1 includes: The test temperature data is the temperature at which the γ′-strengthened iron-based superalloy was tested for yield strength. The yield strength data is the experimental yield strength value corresponding to the chemical composition data, the heat treatment process data, and the test temperature data.
[0011] As a preferred embodiment, in the above-mentioned method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys, step S2 includes: Step S21: Standardize the units and field names of the chemical composition data, heat treatment process data, test temperature data, and yield strength data from different data sources; Step S22: Verify the sum of element mass fractions for each sample. When Fe is the surplus element, calculate or verify the mass fraction of Fe based on the mass fractions of the other elements. Step S23: Remove samples with missing yield strength data and generate corresponding missing identifiers for missing heat treatment process data; Step S24: The samples with the same chemical composition data, heat treatment process data and test temperature data are identified as duplicate sample groups. The mean, median and range of yield strength within the duplicate sample group are calculated. The duplicate samples are merged or retained according to the degree of dispersion within the group. Step S25: Perform a consistency check on the preprocessed data to form a modeling dataset for model training and evaluation.
[0012] In some specific implementations, step S21: Unit and field unification: unify the element content in different documents to mass fraction, unify the temperature to degrees Celsius, unify the time to hours, unify the yield strength to MPa, and unify the field names in different data sources.
[0013] In some specific implementations, step S22: Chemical composition verification: Verify the sum of elemental mass fractions for each sample. When Fe is a surplus element and the literature does not directly provide the Fe content, calculate the Fe content based on the mass fractions of the remaining elements; when the literature provides the Fe content, retain the original Fe content and perform a consistency verification on the total elemental content.
[0014] In some specific implementations, step S23: missing data processing: remove samples with missing yield strength data or unqualified total chemical composition; for missing heat treatment process parameters, instead of directly filling in all data, generate corresponding missing identifiers and complete numerical filling in subsequent training sessions to reduce data leakage.
[0015] Multi-source literature datasets commonly suffer from data quality issues such as inconsistent units, discrepancies in element content recording, missing process parameters, and discrete samples from repeated experiments. Standardizing units and fields eliminates format differences between multi-source data; verifying element mass fractions identifies errors in literature records, ensuring physical consistency of composition data; selectively retaining missing heat treatment markers instead of global padding avoids data leakage issues in machine learning; merging or retaining duplicate samples based on their degree of dispersion balances sample quantity and data authenticity; and final consistency verification reduces the interference of dirty data on model training from the source, effectively improving the reliability and predictive stability of subsequent model training.
[0016] As a preferred embodiment, in the above-mentioned method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys, in step S3, The physical metallurgical enhancement characteristics include: the sum of the mass fractions of Al and Ti; the sum of the mass fractions of γ′ precipitation strengthening elements; the sum of the mass fractions of Mo and W; the mass fraction ratio of Ti to Al; the ratio of Ni to the total mass fraction of Fe and Ni; the total atomic fraction of γ′ precipitation strengthening elements; the total atomic fraction of solid solution strengthening elements; the atomic percentage of Ni in the matrix elements; the atomic ratio of γ′ precipitation strengthening elements to matrix elements; the atomic ratio of solid solution strengthening elements to matrix elements; the atomic ratio of Ti to Al; and the difference between the solid solution temperature and the first-stage aging temperature.
[0017] The original inputs are only single-element mass fractions and process parameters, which cannot directly characterize the physical metallurgical mechanism of the alloy. Mass ratio, atomic ratio, and elemental summation features can describe the compositional information related to γ′ phase formation and the level of solid solution strengthening; the proportion of matrix elements reflects the matrix stacking fault energy and matrix stability; the solid solution-aging temperature difference reflects the driving conditions of heat treatment on precipitation phase transformation. By transforming the above metallurgical mechanisms into model input features, prior material knowledge is introduced into machine learning, which makes up for the shortcomings of pure data-driven models in mining metallurgical mechanisms, reduces the difficulty of model learning, and improves the prediction accuracy and model interpretability under cross-composition and cross-temperature conditions.
[0018] In the above-mentioned method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys, The atomic fraction of element i is determined according to Equation I: Formula I; In the formula, x i For elements i atomic fraction, w i Let i be the mass fraction of element i. M i For elements i atomic weight n This represents the number of elements involved in the atomic fraction calculation.
[0019] in, j These are regular dummy variables. j Iterate through all elements involved in the calculation (from the 1st to the nth element), and assign each element its ( ) w j / M j Add them all together to get the total number of atoms.
[0020] In some specific embodiments, the matrix elements include Fe and Ni, the γ′ precipitation strengthening elements include Al, Ti, Nb, and Ta; and the solid solution strengthening elements include Cr, Mo, W, Nb, Ta, and V. In this case, the atomic ratio of the γ′ precipitation strengthening element to the matrix element is determined according to Formula II: Formula II; In the formula, The ratio of the strengthening element to the matrix element in the γ′ precipitation is given. The atomic fractions of Al, Ti, Nb, and Ta, respectively. These are the atomic fractions of Fe and Ni, respectively.
[0021] The atomic ratio of the solid solution strengthening element to the matrix element is determined according to Formula III: Formula III; In the formula, This represents the atomic ratio of the solid solution strengthening element to the matrix element. , These are the atomic fractions of Cr, Mo, W, Nb, Ta, and V, respectively. These are the atomic fractions of Fe and Ni, respectively.
[0022] In other specific embodiments, the γ′ precipitation strengthening element and the solid solution strengthening element may change, and the calculations for them can be adjusted accordingly with reference to Equations II and III.
[0023] As a preferred embodiment, in the above-mentioned method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys, step S4 includes: Step S41: Set multiple candidate parameters for the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model, and determine the target parameter combination for each model based on the prediction error of the candidate parameter combination in cross-validation. Step S42: Train the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model respectively using the target parameter combination to obtain the predicted yield strength values corresponding to the three models; Step S43: Perform equal-weighted fusion of the yield strength prediction values of the three models to obtain the fused prediction value of the three models.
[0024] The three types of tree models—extreme random tree regression, extreme gradient boosting tree, and lightweight gradient boosting machine—have different internal algorithmic logics, and individual models are prone to overfitting or underfitting. First, cross-validation is used to optimize the hyperparameters of each model, enabling each base model to reach its optimal performance. Then, an equal-weighted fusion strategy is used to integrate the outputs of the three models, which can offset the prediction bias caused by the algorithm of a single model and reduce the model's sensitivity to dataset noise. Compared with individual models, the fused model has a lower overall prediction error and better generalization performance.
[0025] As a preferred embodiment, in the above-mentioned method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys, step S5 includes: Step S51: Summarize the three-model fusion prediction results of each cross-validation test subset to form the cross-validation prediction value of each modeling sample; Step S52: Calculate the determination coefficient of the three-model fusion prediction results R 2 Root mean square error RMSE Mean absolute error MAE Used to evaluate the accuracy and error distribution of the three-model fusion prediction results; Step S53: Plot a scatter plot of measured values versus predicted values with the experimental yield strength value on the x-axis and the fused predicted value of the three models on the y-axis; Step S54: Encapsulate the data preprocessing, physical metallurgical enhancement feature construction, training results of the three models, and equal-weighted fusion prediction process into an executable file, providing an input interface for chemical composition, heat treatment process parameters, and target test temperature, and outputting the corresponding predicted yield strength value.
[0026] Calculate based on cross-validation results R 2 , RMSE , MAE Multiple metrics can be used to objectively quantify the model's predictive performance from multiple dimensions, including good fit and absolute error; actual measurements Predictive scatter plots provide a visual overview of the model's prediction bias in the high and low yield strength ranges, facilitating the identification of model failure zones; retraining on all data fully utilizes all existing sample information, maximizing the extraction of components within the dataset. process Performance correlation; the entire process is encapsulated into an executable program with an input interface, removing the barrier of algorithm code. Engineers only need to input the composition, process, and test temperature to complete the prediction, realize the rapid evaluation of alloy yield strength, directly serve the screening of candidate materials, and reduce the number of iterations of high-temperature alloy tests.
[0027] A second aspect of the present invention provides a system for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys, comprising: The data acquisition module is used to acquire chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys. The data preprocessing module is used to standardize units, verify chemical composition, handle missing data and duplicate samples in the acquired data to form a modeling dataset. The physical metallurgical enhancement feature construction module is used to construct physical metallurgical enhancement features based on the chemical composition data and the heat treatment process data, and combine them with the original features to form a model input feature set. The model training module is used to determine the parameters of the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model using cross-validation, and to train the corresponding models using the determined parameters. The fusion prediction module is used to perform equal-weighted fusion of the prediction results of the three models, namely the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model, to form a three-model fusion prediction result. The model evaluation module is used to summarize the cross-validation prediction results and calculate the evaluation index of the fusion prediction results of the three models; The result output module is used to output the predicted yield strength of the γ′-reinforced iron-based superalloy at the target test temperature.
[0028] Compared with existing technologies, this invention has significant advantages and beneficial effects. Through the above technical solution, this invention achieves considerable technological advancement and practicality, and has broad application value, possessing at least the following advantages: This invention uses the chemical composition, heat treatment process, and test temperature of γ′-strengthened iron-based superalloys as inputs to construct physical metallurgical enhancement features such as strengthening element combinations, solid solution strengthening element combinations, element atomic fractions, specific mass ratios or atomic ratios, and heat treatment temperature differences. This supplements information that is difficult to directly express using the original composition and process parameters. Three models—an extreme random tree regression model, a limit gradient boosting tree model, and a lightweight gradient booster model—each learn the nonlinear relationship between the input features and the yield strength. Equal-weighted fusion is used to reduce the impact of individual model structural differences on the prediction results. This method can be used for rapid prediction of the yield strength of γ′-strengthened iron-based superalloys at different test temperatures, providing a reference for alloy performance evaluation and candidate material screening.
[0029] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0030] Figure 1 This is a schematic flowchart of a method for predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges, according to the present invention.
[0031] Figure 2 (a) is a distribution diagram of yield strength data in Example 1; Figure 2 Figure (b) is a graph showing the relationship between test temperature and yield strength in Example 1.
[0032] Figure 3This is a scatter plot of the measured and predicted yield strength values for the lightweight gradient booster model (a), extreme random tree regression model (b), extreme gradient booster tree model (c), and three-model equal-weighted fusion model (d) in Example 1.
[0033] Figure 4 (a) is a summary diagram of the SHAP of the three-model equal-weight fusion model used in Example 1; Figure 4 In Figure (b), the global feature importance map is the three-model equal-weight fusion model used in Example 1.
[0034] Figure 5 This is a graphical user interface (GUI) diagram of the γ′-strengthened iron-based superalloy transtemperature range yield strength prediction system developed in this invention. The system is used to predict yield strength values after inputting elemental composition and process parameters. Detailed Implementation
[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0036] Example 1 This embodiment provides a method and system for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys. The prediction method flowchart is as follows. Figure 1 .
[0037] Step S1: Obtain chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys from the literature to form the original dataset of γ′-strengthened iron-based superalloys.
[0038] Step S11: Data Acquisition: Collect yield strength data of γ′-strengthened iron-based superalloys reported in publicly available literature. The mass fraction range of each alloying element in the γ′-strengthened iron-based superalloy is as follows: C 0-0.25%, Si 0-3.85%, Mn 0-1.91%, Cr 4.94-19.00%, Ni 23.60-43.50%, Mo 0-7.72%, Ti 0-5.31%, Al 0.04-4.34%, V 0-0.44%, Nb 0-2.93%, W 0-2.90%, B 0-0.10%, Ce 0-0.01%, Co 0-0.61%, Cu 0-2.84%, Ta 0-3.06%, Fe 34.17-56.55%.
[0039] Step S12: Heat treatment process data collection: Collect the solution temperature, solution time, first-stage aging temperature, first-stage aging time, second-stage aging temperature, second-stage aging time, third-stage aging temperature, and third-stage aging time corresponding to each yield strength data. Among them, the solution temperature is 900-1220 ℃, the solution time is 0.0167-8 h, the first-stage aging temperature is 600-830 ℃, the first-stage aging time is 1-1000 h, the second-stage aging temperature is 620-750 ℃, the second-stage aging time is 4-24 h, the third-stage aging temperature is 650-670 ℃, and the third-stage aging time is 8-20 h.
[0040] Step S13: Performance data acquisition: Record the target test temperature for yield strength testing of γ′-strengthened iron-based superalloys, as well as the experimental yield strength values corresponding to the chemical composition and heat treatment process. Figure 2 (a) is a distribution diagram of yield strength data in Example 1; Figure 2 Figure (b) shows the distribution of the relationship between test temperature and yield strength in Example 1. The target test temperature was 25-900 °C, and the experimental yield strength values corresponding to the chemical composition and heat treatment process were 230-1420 MPa. After subsequent data preprocessing, a total of 288 modeling samples were obtained in this example. The yield strength test temperatures of the samples covered 25-900 °C, including room temperature and multiple medium and high temperature test conditions. The constructed dataset covered multiple test temperature ranges from room temperature to high temperature. The model used the test temperature as one of the input features, and cross-validation was used within the above test temperature range to obtain the prediction results for each sample. The coefficient of determination was then used. R 2 Root mean square error RMSE The mean absolute error (MAE) is used to evaluate the overall predictive performance of the model under different test temperature conditions.
[0041] Step S2: Perform unit unification, chemical composition verification, missing data processing, and duplicate sample processing on the original dataset to form a γ′-strengthened iron-based superalloy modeling dataset.
[0042] Step S2 specifically includes the following steps: Step S21: Unit and field unification: Unify the element content in different documents to mass fraction, the temperature to degrees Celsius, the time to hours, the yield strength to MPa, and unify the field names in different data sources.
[0043] Step S22: Chemical composition verification: Verify the sum of elemental mass fractions for each sample. When Fe is a surplus element and the literature does not directly provide the Fe content, calculate the Fe content based on the mass fractions of the remaining elements; when the literature provides the Fe content, retain the original Fe content and perform a consistency verification on the total elemental content.
[0044] Step S23: Missing data processing: Remove samples with missing yield strength data or unqualified total chemical composition; for missing heat treatment process parameters, do not directly fill them in on all data, but generate corresponding missing labels and complete the numerical filling in each subsequent training session to reduce data leakage.
[0045] Step S24: Duplicate sample processing: Divide samples with the same chemical composition, heat treatment process and test temperature into the same duplicate sample group, calculate the mean, median and range of yield strength within the group, and merge or retain duplicate samples according to the degree of dispersion within the group.
[0046] Step S25: Data Consistency Verification: The data processed above is verified to form a modeling dataset for model training and evaluation. In this embodiment, a total of 288 modeling samples were obtained.
[0047] Step S3: Construct physical metallurgical enhancement features based on chemical composition data and heat treatment process data, and combine them with the original features to form the model input feature set.
[0048] First, the atomic fraction of each element is calculated based on its mass fraction and corresponding atomic weight. Then, the sum of the mass fractions of Al and Ti; the sum of the mass fractions of Al, Ti, Nb, and Ta; the sum of the mass fractions of Mo and W; the mass fraction ratio of Ti to Al; the ratio of Ni to the total mass fraction of Fe and Ni; the total atomic fraction of Al, Ti, Nb, and Ta; the total atomic fraction of Cr, Mo, W, Nb, Ta, and V; the atomic percentage of Ni in the matrix elements Fe and Ni; the total atomic ratio of Al, Ti, Nb, and Ta to the matrix elements Fe and Ni; the total atomic ratio of Cr, Mo, W, Nb, Ta, and V to the matrix elements Fe and Ni; the atomic ratio of Ti to Al; and the difference between the solution treatment temperature and the first-stage aging temperature are calculated. Finally, the above physical metallurgical enhancement characteristics are combined with the original chemical composition, heat treatment process, and test temperature characteristics to form the model input feature set.
[0049] The atomic fraction of element i is determined according to Equation I: Formula I; In the formula, x i For elements i atomic fraction, w i Let i be the mass fraction of element i. M i For elements i atomic weight n This represents the number of elements involved in the atomic fraction calculation.
[0050] The atomic ratios of strengthening elements Al, Ti, Nb, and Ta to the matrix elements are determined according to Formula II: Formula II; In the formula, The ratio of the strengthening element to the matrix element in the γ′ precipitation is given. The atomic fractions of Al, Ti, Nb, and Ta, respectively. These are the atomic fractions of Fe and Ni, respectively.
[0051] The atomic ratio of the solid solution strengthening element to the matrix element is determined according to Formula III: Formula III; In the formula, This represents the atomic ratio of the solid solution strengthening element to the matrix element. , These are the atomic fractions of Cr, Mo, W, Nb, Ta, and V, respectively. These are the atomic fractions of Fe and Ni, respectively.
[0052] Figure 4 (a) is a summary diagram of the SHAP of the three-model equal-weight fusion model used in Example 1; Figure 4 In Figure (b), the global feature importance map is the three-model equal-weight fusion model used in Example 1. Figure 4 Further, it is shown that in the three-model equal-weighted fusion model established in this embodiment, the combined total amount of Al, Ti, Nb, and Ta has the highest global feature importance, while single-element features such as Ti and Ni also contribute significantly. In addition, physical metallurgical reinforcement features such as the difference between solution treatment temperature and first-stage aging temperature, the Ni / (Fe+Ni) mass ratio, the Ti / Al mass ratio, and the atomic percentage of Ni in the Fe-Ni matrix also rank high in global importance. These results indicate that some of the constructed physical metallurgical reinforcement features can provide the model with information distinct from single-element content and original process parameters, and participate in the model's prediction of yield strength.
[0053] Step S4: Train the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model respectively, determine the model parameters through cross-validation, and construct the fusion prediction results of the three models; Step S4 specifically includes the following steps: Step S41: Parameter validation: Use two-fold cross-validation within each outer training set to evaluate the candidate parameter combinations of the three models.
[0054] Step S42: Model Parameter Selection: Preset candidate hyperparameter values for the Extremely Randomized Tree Regression model, the Ultimate Gradient Boosting Tree model, and the Lightweight Gradient Boosting Machine model respectively. Evaluate the predictive performance of each parameter configuration using cross-validation within the training data, and calculate the results based on the root mean square error. RMSE Determine the parameter configurations used for each model; After cross-validation evaluation, the parameters were re-determined using all modeling data. In this embodiment, the final parameters for the three models are shown in Table 1.
[0055] Table 1. Candidate hyperparameter values for the Extremely Random Tree Regression Model, the Limiting Gradient Boosting Tree Model, and the Lightweight Gradient Boosting Machine Model.
[0056] Step S43: Single model training and prediction: Using the target parameter combination, train the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model on the current outer training set respectively, and make predictions on the current outer test set to obtain the prediction results of the three single models.
[0057] Step S44: The prediction results of the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model are fused. Using 1 / 3, 1 / 3, and 1 / 3 equal-weight fusion as a benchmark, and employing non-equal-weight combinations such as 0.3 / 0.3 / 0.4, 0.3 / 0.4 / 0.3, and 0.4 / 0.3 / 0.3 for sensitivity comparison, the overall prediction performance of each scheme is similar. Considering model performance, the simplicity of the fusion method, and avoiding the introduction of additional weight optimization parameters, this embodiment adopts a 1 / 3, 1 / 3, and 1 / 3 equal-weight fusion method.
[0058] Step S5: Summarize the prediction results of each outer test set, calculate the model evaluation index, and perform prediction result visualization, physical metallurgical enhancement feature ablation verification, and model encapsulation. Step S5 specifically includes the following steps: Step S51: Summarize the prediction results of cross-validation: Combine the prediction results of each outer test set in the five outer cross-validations according to the order of the modeling samples, so that each modeling sample obtains a prediction value generated by the model that was not trained using this sample.
[0059] Step S52: Evaluation index calculation: Calculate the coefficients of determination for the extreme random tree regression model, the extreme gradient boosting tree model, the lightweight gradient boosting machine model, and the weighted fusion result of the three models respectively. R 2 Root mean square error RMSE Mean absolute error MAE In this embodiment, the result of the equal-weighted fusion of the three models... R2 0.8070, RMSE 86.304 MPa MAE The value is 59.020 MPa. Compared with the three single models, the result of the three-model equal-weighted fusion is... R 2 Highest, RMSE Lowest and MAE The lowest value indicates that the equal-weighted fusion of the three models can achieve good overall prediction performance on the dataset of this embodiment. The equal-weighted fusion of the three models is the preset final prediction output scheme of this embodiment.
[0060] The evaluation metrics for the three single-model and three-model equal-weight fusion results are shown in Table 2.
[0061] Table 2 Evaluation results of three single-model and three-model equal-weight fusion models
[0062] Step S53: Visualize the prediction results: Plot a scatter plot of measured values and predicted values with the experimental yield strength value on the x-axis and the predicted value on the y-axis, and set a reference line to show that the experimental value and the predicted value are equal. Figure 3 The results of the extreme random tree regression model, the extreme gradient boosting tree model, the lightweight gradient boosting machine model, and the weighted fusion of the three models are presented respectively.
[0063] Step S54: Model Encapsulation: Retrain the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model using all modeling samples, retaining the equal weight fusion relationship of the three models, and encapsulate the data preprocessing, physical metallurgical enhancement feature construction, and three-model fusion prediction process into an executable program. After the user inputs the chemical composition, heat treatment process parameters, and target test temperature of the γ′-strengthened iron-based superalloy to be predicted, the system generates model input features according to the same rules as in the training phase and outputs the predicted yield strength value of the alloy at the target test temperature.
[0064] This embodiment also provides a system for predicting the cross-temperature range yield strength of γ′-strengthened iron-based superalloys. Figure 5 This is a graphical user interface (GUI) diagram of the prediction system. The prediction system is applied to the aforementioned method for predicting the transtemperature range yield strength of a γ′-strengthened iron-based superalloy, including: The data acquisition module is used to acquire chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys. The data preprocessing module is used to standardize units, verify chemical composition, handle missing data and duplicate samples in the acquired data to form a modeling dataset. The physical metallurgical enhancement feature construction module is used to construct physical metallurgical enhancement features based on chemical composition data and heat treatment process data, and combine them with the original features to form the model input feature set. The model training module is used to determine the parameters of the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model using cross-validation, and to train the corresponding models using the determined parameters. The fusion prediction module is used to perform equal-weighted fusion of the prediction results of three models: extreme random tree regression model, extreme gradient boosting tree model, and lightweight gradient boosting machine model, to form a fusion prediction result of the three models. The model evaluation module is used to summarize the cross-validation prediction results and calculate the evaluation index of the three-model fusion prediction results; The results output module is used to output the predicted yield strength of the γ′-strengthened iron-based superalloy at the target test temperature.
[0065] Comparative Example 1 CN118352002A discloses a method for predicting the yield strength of γ′ phase-strengthened cobalt-based superalloys. The embodiments of this patent application collected data on 402 γ′ phase-strengthened cobalt-based superalloys, using alloy composition, solution temperature, solution time, aging temperature, aging time, and test temperature as input parameters. The method compared various algorithms, including nearest neighbor algorithm, random forest algorithm, gradient boosting algorithm, and support vector machine algorithm, ultimately selecting a single gradient boosting algorithm to establish the yield strength prediction model. The coefficient of determination on the test set is... R 2 The value is 0.91, and the average relative error is 6.4%.
[0066] Compared with the prior art embodiments described above, the improvement of this embodiment is not merely in changing the machine learning algorithm, but also in further optimizing the model input features and the prediction model construction method. First, the prior art mainly uses raw parameters such as alloy composition, single-stage aging process parameters, and test temperature to directly perform machine learning modeling. However, this application, based on the raw chemical composition, heat treatment process parameters, and target test temperature, further constructs physical metallurgical enhancement features such as precipitation strengthening element combinations, solid solution strengthening element combinations, element mass ratios, atomic fractions and atomic ratios, and solid solution-aging temperature differences. This allows the model to directly utilize the combination information related to γ′ phase formation, solid solution strengthening elements, and matrix composition, rather than entirely relying on the model to learn the aforementioned metallurgical relationships from the content of a single element.
[0067] Secondly, the aforementioned existing technologies select a single gradient boosting model as the final prediction model after comparing multiple algorithms. This application, however, establishes an extreme random tree regression model, a limiting gradient boosting tree model, and a lightweight gradient boosting machine model, and then performs equal-weighted fusion of the prediction results of the three models. Under the same modeling data and evaluation conditions as in Example 1, the equal-weighted fusion result of the three models... R 2 0.8070, RMSE 86.304 MPa MAE It is 59.020 MPa, R 2 More than three single models, and RMSE and MAE The scores are all lower than those of three single models, indicating that equal-weighted fusion can reduce the dependence of the final prediction results on the structure of a single model and obtain better overall prediction performance.
[0068] Furthermore, this application simultaneously characterizes multi-stage heat treatment process parameters, including first-stage, second-stage, and third-stage aging, thus enabling it to adapt to data on γ′-strengthened iron-based superalloys employing single-stage or multi-stage aging regimes. In summary, the technical advantages of this application are mainly reflected in: improving the characterization ability of the original data on alloy strengthening mechanisms through physical metallurgical enhancement characteristics; enhancing the applicability to complex heat treatment regimes through multi-stage heat treatment parameters; and reducing reliance on a single model and improving overall prediction performance through equal-weighted fusion of multiple models.
[0069] Comparative Example 2 CN117637053A discloses a high-temperature alloy composition design method based on materials genome engineering technology. Its embodiments are based on 120 powder high-temperature alloy composition, test conditions and yield strength data, construct element property related features, and use support vector machine to predict the yield strength of candidate nickel-based high-temperature alloys at 20 ℃ and 750 ℃.
[0070] Compared with the existing technologies, this application focuses on predicting the yield strength of γ′-strengthened iron-based superalloys across temperature ranges. It uses chemical composition, solution treatment and multi-stage aging process parameters, and the target test temperature as basic inputs, and further constructs physical metallurgical enhancement features such as precipitation strengthening element combinations, solution strengthening element combinations, mass ratio, atomic fraction and atomic ratio, and the solution-aging temperature difference. Feature ablation results show that after adding enhancement features, the limiting gradient boosting tree model... R 2 It increased from 0.7660 to 0.7992. RMSE The decrease from 95.020 MPa to 88.017 MPa indicates that this feature construction can improve the yield strength prediction performance.
[0071] Furthermore, the 288 modeling samples in Embodiment 1 of this application cover multiple test temperature conditions ranging from 25 to 900 °C, and employ equal-weighted fusion of three models: extreme random tree, extreme gradient boosting tree, and lightweight gradient boosting machine. Under the same evaluation conditions, the fused model achieves... R 2 =0.8070、 RMSE =86.304 MPa, the overall prediction performance is better than the three single models.
[0072] Therefore, compared with the material design process of CN117637053A, which mainly aims at "candidate composition generation - thermodynamic and experimental screening - performance prediction - composition optimization", the advantages of this application are: it does not require the pre-construction of a large-scale candidate composition space and the completion of high-throughput thermodynamic and diffusion multi-node screening, and can directly predict the yield strength based on the chemical composition, heat treatment process and target test temperature of existing γ′-strengthened iron-based superalloys; at the same time, through the physical metallurgical enhancement characteristics and weighted fusion of multiple models oriented towards the strengthening mechanism of γ′-strengthened iron-based superalloys, it improves the characterization ability and comprehensive prediction performance of complex process states and multiple temperature conditions.
[0073] Comparative Example 3 To verify the influence of physical metallurgical reinforcement features on the predictive performance of γ′-strengthened iron-based superalloys across temperature ranges, Comparative Example 3 did not construct the physical metallurgical reinforcement features of step S3, but only used the original chemical composition, heat treatment process parameters, and target test temperature as model inputs, while keeping the modeling samples, data preprocessing methods, cross-validation division, model parameter selection methods, and evaluation indicators consistent with Example 1.
[0074] Taking the extreme gradient boosting tree model as an example, compared to Example 3 which only uses the original features, the coefficient of determination of the model is... R 2 The root mean square error is 0.7660. RMSE The value is 95.020 MPa; In Example 1, after adding physical metallurgical enhancement features, under the same evaluation conditions, the limiting gradient boosting tree model... R 2 Increased to 0.7992, RMSE It decreased to 88.017 MPa.
[0075] The results show that, under the condition that the modeling samples and evaluation methods are kept consistent, constructing physical metallurgical enhancement features consisting of precipitation strengthening element combinations, solid solution strengthening element combinations, element mass ratios, element atomic fractions and atomic ratios, and heat treatment temperature differences can improve the model's ability to characterize the nonlinear relationship between the composition, heat treatment process, and yield strength of γ′-strengthened iron-based superalloys, thereby improving the prediction performance.
[0076] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys, characterized in that, include: Step S1: Obtain chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys from the literature to form the original dataset of γ′-strengthened iron-based superalloys; Step S2: Perform data preprocessing on the original dataset. The data preprocessing includes unit unification, chemical composition verification, missing data processing, and duplicate sample processing to form a γ′-strengthened iron-based superalloy yield strength modeling dataset for model training and evaluation. Step S3: Construct physical metallurgical enhancement features based on the chemical composition data and the heat treatment process data, and combine the original features with the physical metallurgical enhancement features to form a model input feature set; Step S4: Based on the model input feature set, train three models respectively: extreme random tree regression model, extreme gradient boosting tree model, and lightweight gradient boosting machine model, and perform equal weighted fusion of the yield strength prediction values output by the three models to obtain the fused prediction value of the three models; Step S5: Use cross-validation to evaluate the fusion prediction values of the three models, calculate the model evaluation index, and visualize the prediction results; after the evaluation is completed, retrain the three models using all the modeling data, and encapsulate the data preprocessing, physical metallurgical enhancement feature construction, and three-model fusion prediction process into an executable program.
2. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 1, characterized in that, The temperature range across the temperature spectrum is 25-900℃.
3. The method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys according to claim 1, characterized in that, In step S1, the chemical composition data includes the mass fractions of matrix elements, γ′ precipitation strengthening elements, solid solution strengthening elements, and grain boundary strengthening elements. In step S1, the heat treatment process data includes solution temperature, solution time, first-stage aging temperature, and first-stage aging time.
4. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 1, characterized in that, Step S2 includes: Step S21: Standardize the units and field names of the chemical composition data, heat treatment process data, test temperature data, and yield strength data from different data sources; Step S22: Verify the sum of element mass fractions for each sample. When Fe is the surplus element, calculate or verify the mass fraction of Fe based on the mass fractions of the other elements. Step S23: Remove samples with missing yield strength data and generate corresponding missing identifiers for missing heat treatment process data; Step S24: The samples with the same chemical composition data, heat treatment process data and test temperature data are identified as duplicate sample groups. The mean, median and range of yield strength within the duplicate sample group are calculated. The duplicate samples are merged or retained according to the degree of dispersion within the group. Step S25: Perform a consistency check on the preprocessed data to form a modeling dataset for model training and evaluation.
5. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 1, characterized in that, In step S3 The physical metallurgical enhancement characteristics include: the sum of the mass fractions of Al and Ti; the sum of the mass fractions of γ′ precipitation strengthening elements; the sum of the mass fractions of Mo and W; the mass fraction ratio of Ti to Al; the ratio of Ni to the total mass fraction of Fe and Ni; the total atomic fraction of γ′ precipitation strengthening elements; the total atomic fraction of solid solution strengthening elements; the atomic percentage of Ni in the matrix elements; the atomic ratio of γ′ precipitation strengthening elements to matrix elements; the atomic ratio of solid solution strengthening elements to matrix elements; the atomic ratio of Ti to Al; and the difference between the solid solution temperature and the first-stage aging temperature.
6. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 5, characterized in that, The matrix elements include Fe and Ni; The γ′ precipitation strengthening elements include Al, Ti, Nb, and Ta; The solid solution strengthening elements include Cr, Mo, W, Nb, Ta, and V.
7. The method for predicting the transtemperature range yield strength of γ′-reinforced iron-based superalloys according to claim 6, characterized in that, The atomic fraction of element i is determined according to Equation I: Formula I; In the formula, x i For elements i atomic fraction, w i Let i be the mass fraction of element i. M i For elements i atomic weight n The number of elements involved in the atomic fraction calculation; The atomic ratio of the γ′ precipitation strengthening element to the matrix element is determined according to Formula II: Formula II; In the formula, To enhance the atomic ratio of the element to the matrix element, The atomic fractions of Al, Ti, Nb, and Ta, respectively. These are the atomic fractions of Fe and Ni, respectively. The atomic ratio of the solid solution strengthening element to the matrix element is determined according to Formula III: Formula III; In the formula, The ratio of the atomic number of the solid solution strengthening element to the atomic number of the matrix element. , These are the atomic fractions of Cr, Mo, W, Nb, Ta, and V, respectively. These are the atomic fractions of Fe and Ni, respectively.
8. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 1, characterized in that, Step S4 includes: Step S41: Set multiple candidate parameters for the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model, and determine the target parameter combination for each model based on the prediction error of the candidate parameter combination in cross-validation; Step S42: Train the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model respectively using the target parameter combination to obtain the predicted yield strength values corresponding to the three models; Step S43: Perform equal-weighted fusion of the yield strength prediction values of the three models to obtain the fused prediction value of the three models.
9. The method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to claim 1, characterized in that, Step S5 includes: Step S51: Summarize the three-model fusion prediction results of each cross-validation test subset to form the cross-validation prediction value of each modeling sample; Step S52: Calculate the determination coefficient of the three-model fusion prediction results. R 2 Root mean square error RMSE Mean absolute error MAE ; Step S53: Plot a scatter plot of measured values versus predicted values with the experimental yield strength value on the x-axis and the fused predicted value of the three models on the y-axis; Step S54: Encapsulate the data preprocessing, physical metallurgical enhancement feature construction, training results of the three models, and equal-weighted fusion prediction process into an executable file, providing an input interface for chemical composition, heat treatment process parameters, and target test temperature, and outputting the corresponding predicted yield strength value.
10. A prediction system applicable to the method for predicting the transtemperature range yield strength of γ′-strengthened iron-based superalloys according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire chemical composition data, heat treatment process data, test temperature data, and yield strength data of γ′-strengthened iron-based superalloys. The data preprocessing module is used to standardize units, verify chemical composition, handle missing data and duplicate samples in the acquired data, and form a modeling dataset. The physical metallurgical enhancement feature construction module is used to construct physical metallurgical enhancement features based on the chemical composition data and the heat treatment process data, and combine them with the original features to form a model input feature set; The model training module is used to determine the parameters of the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model using cross-validation, and to train the corresponding models using the determined parameters. The fusion prediction module is used to perform equal-weighted fusion of the prediction results of the three models, namely the extreme random tree regression model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model, to form a three-model fusion prediction result. The model evaluation module is used to summarize the cross-validation prediction results and calculate the evaluation index of the fusion prediction results of the three models; The result output module is used to output the predicted yield strength of the γ′-reinforced iron-based superalloy at the target test temperature.
