A method for predicting mechanical properties of medium-low alloy cast steel

Abstract

The application relates to the technical field of metal material performance prediction, and discloses a middle-low alloy cast steel mechanical property prediction method, which comprises the following steps: collecting middle-low alloy cast steel sample data, establishing an original database of the middle-low alloy cast steel sample; uniformly expressing heat treatment process data according to preset rules, forming heat treatment process features, and integrating chemical component data and the heat treatment process features into original features of each piece of the middle-low alloy cast steel sample data; constructing physical information features according to the original features in the original database; training a middle-low alloy cast steel mechanical property prediction model by taking the physical information features or the original features and the physical information features as input variables; and finally, predicting the mechanical property of the middle-low alloy cast steel to be predicted by using the middle-low alloy cast steel mechanical property prediction model. The method can more accurately predict the mechanical property of the middle-low alloy cast steel, provides more efficient and accurate data guidance for the research and development of the middle-low alloy cast steel, and reduces the research and development cost.

Description

Technical Field

[0001] This invention relates to the field of metal material performance prediction technology, specifically to a method for predicting the mechanical properties of medium and low alloy cast steel. Background Technology

[0002] Medium and low alloy cast steels are widely used in heavy machinery, rail transportation, energy equipment, mining and metallurgy, and mold manufacturing due to their good formability, high comprehensive mechanical properties, and suitability for manufacturing large and complex load-bearing components. As the service conditions of engineering equipment continue to evolve towards heavier loads, impacts, wear, and higher reliability, higher requirements are being placed on the strength-plasticity balance of medium and low alloy cast steels. Among these, tensile strength and elongation, as key mechanical indicators characterizing the strength and plasticity of materials, are of great significance for material composition design, heat treatment system optimization, and engineering applications.

[0003] The development and performance optimization of existing low- and medium-alloy cast steels still largely rely on empirical trial and error, experimental screening, and traditional mechanistic analysis methods. While these methods can guide material design to some extent, they typically suffer from long development cycles, high experimental costs, and low efficiency, especially in multi-component systems and complex heat treatment conditions, making it difficult to achieve rapid evaluation and efficient optimization of material properties. In recent years, machine learning methods have provided a new technical path for material performance prediction; however, existing methods usually directly use elemental content and heat treatment parameters as input features, and their predictive ability largely depends on the coverage of training data. When the sample to be tested is not involved in model training and testing, or when faced with samples similar to existing low- and medium-alloy cast steel material systems but with different composition-process combinations, the model may still exhibit limited predictive reliability if it lacks physical information feature constraints and independent external verification. Meanwhile, during the austenitizing-cooling-tempering heat treatment process, the microstructure evolution and mechanical property response of low- and medium-alloy cast steels exhibit significant nonlinear characteristics. Directly using the original process parameters is usually insufficient to effectively characterize their intrinsic physical mechanisms, thus limiting the reliability and interpretability of model predictions.

[0004] Therefore, it is necessary to propose a prediction method that combines knowledge of the physical properties of medium and low alloy cast steel with machine learning models, which can improve the model's ability to predict the physical properties of medium and low alloy cast steel, such as strength and plasticity. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the present invention aims to provide a method for predicting the mechanical properties of medium and low alloy cast steel, which can predict the physical properties of medium and low alloy cast steel with high accuracy, provide data guidance for the research and development of medium and low alloy cast steel, and improve the research and development efficiency of medium and low alloy cast steel.

[0006] To solve the above problems, the technical solution adopted by the present invention is as follows: A method for predicting the mechanical properties of medium and low alloy cast steel, comprising: Collect sample data of medium and low alloy cast steel and establish a raw database of medium and low alloy cast steel samples. The raw database shall include at least chemical composition data, heat treatment process data and mechanical property data. According to the preset rules, the heat treatment process data is uniformly expressed to form heat treatment process characteristics, and the chemical composition data and heat treatment process characteristics are integrated into the original characteristics of each medium and low alloy cast steel sample data. Physical information features are constructed based on the original features in the original database; The collected sample data is divided into a training set and a validation set. The physical information features in the training set, or the original features and physical information features, are used as input variables, and the target physical performance data in the mechanical performance data in the training set are used as output variables to train the mechanical performance prediction model of low alloy cast steel. The heat treatment process data of the low-alloy cast steel sample to be predicted are uniformly expressed to form the heat treatment process characteristics to be predicted. The chemical composition data and heat treatment process characteristics of the low-alloy cast steel sample to be predicted are integrated into the original characteristics to be predicted of the low-alloy cast steel sample. The physical information features to be predicted are constructed based on the chemical composition data and heat treatment process data of the low-alloy cast steel to be predicted. The physical information features to be predicted, or the original features to be predicted and the physical information features to be predicted, are input into the mechanical property prediction model of medium and low alloy cast steel to obtain the predicted mechanical properties of the medium and low alloy cast steel.

[0007] The aforementioned method for predicting the mechanical properties of medium and low alloy cast steel includes physical information features constructed based on chemical composition data and process-side physical information features constructed based on heat treatment process data. The composition-side physical information features include carbon content, carbon equivalent, comprehensive characteristics of carbide-forming elements, atomic size mismatch, and equivalent electronegativity. The process-side physical information features include temperature characteristics, time characteristics, tempering parameters, and cooling intensity mapping characteristics.

[0008] The carbon equivalent CE in the above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel is calculated using the following formula:

[0009] in, , , , , and These are the mass fractions of elements C, Mn, Cr, Mo, V, Ni, and Cu, respectively. The comprehensive characteristics of carbide-forming elements The following formula is used to calculate:

[0010] in, , , , , and These represent the mass fractions of elements Cr, Mo, W, V, Nb, and Ti, respectively.

[0011] The above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel, wherein the atomic size mismatch... The following formula is used to calculate:

[0012] in, Let i be the atomic fraction of the i-th element. Let be the atomic radius of the i-th element. This is the weighted average atomic radius of all elements.

[0013] The above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel, wherein the equivalent electronegativity The following formula is used to calculate:

[0014] in, Let i be the atomic fraction of the i-th element. Let be the Pauling electronegativity of the i-th element.

[0015] The above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel uses S-shaped growth curve functions to construct the temperature characteristics for the final austenitizing stage temperature and the tempering stage temperature, respectively, and uses logarithmic functions to construct the time characteristics for the final austenitizing stage time and the tempering time, respectively.

[0016] The above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel, wherein the tempering parameters The following formula is used to calculate:

[0017] in, This is the absolute temperature for tempering. Tempering time, This is an empirical constant.

[0018] The aforementioned method for predicting the mechanical properties of medium and low alloy cast steel includes a cooling intensity mapping characteristic. The following formula is used to calculate:

[0019] in, The cooling method category is determined by classifying the cooling methods in the heat treatment process data, sorting the categorized cooling methods according to their cooling intensity, and assigning values ​​from weakest to strongest, and from smallest to largest. This represents the value of the cooling intensity category characteristic for the cooling method with the weakest cooling intensity. This represents the value of the cooling intensity category characteristic for the cooling method with the strongest cooling intensity.

[0020] The above-mentioned method for predicting the mechanical properties of medium and low alloy cast steel includes the following steps: dividing the collected sample data into a training set and a validation set, and using the composition-side physical information features and process-side physical information features in the training set as input variables to train the prediction model for the mechanical properties of medium and low alloy cast steel: Calculate the contribution of each physical information feature to the target physical performance, and filter the physical information features based on the contribution. The collected sample data is divided into a training set and a validation set. The physical information features that pass the screening in the training set, or the original features and the physical information features that pass the screening, are used as input variables to train the prediction model of the mechanical properties of low alloy cast steel.

[0021] The aforementioned method for predicting the mechanical properties of medium and low alloy cast steel uses an XGBoost model. During the training of this model, 5-fold cross-validation and a Bayesian optimization framework based on the Optuna framework are used for training and parameter tuning. During training, robust normalization is used to scale the input variables, and Box-Cox transformation is used to correct the distribution of the output variables.

[0022] Compared with existing technologies, the beneficial effects of this invention are as follows: by training the mechanical property prediction model with physical information features that can further characterize the intrinsic physical mechanism of medium and low alloy cast steel, which are constructed from the original features including chemical composition data and heat treatment process characteristics, the reliability and interpretability of the model's prediction of the physical and mechanical properties of medium and low alloy cast steel are improved. The reliability of the model's prediction of sample data not covered by the training data is also improved, providing more rapid and reliable mechanical property evaluation results for the research and development of medium and low alloy cast steel, and further shortening the research and development cycle of medium and low alloy cast steel.

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for predicting the mechanical properties of low-alloy cast steel according to an embodiment of the present invention.

[0025] Figure 2 The R-values ​​of traditional models trained using only raw data and the models of this invention on independent test sets are compared. 2 Comparison chart.

[0026] Figure 3 This is a comparison chart of the RMSE of a traditional model trained using only raw data and the model of this invention on an independent test set.

[0027] Figure 4 This is a scatter plot comparing the EI predicted values ​​and experimental values ​​of the model in this embodiment of the invention on an independent test set.

[0028] Figure 5 This is a scatter plot comparing the UTS predicted values ​​and experimental values ​​of the model in this embodiment of the invention on an independent test set.

[0029] Figure 6 This is a comparison chart of the EI sample-by-sample experimental values ​​and predicted values ​​of the model in this embodiment of the invention on an independent test set.

[0030] Figure 7 This is a comparison chart of the UTS sample-by-sample experimental values ​​and predicted values ​​of the model in this embodiment of the invention on an independent test set.

[0031] Figure 8 This is a comparison chart of the UTS predicted values ​​and experimental values ​​of the model in this embodiment of the invention for samples outside the applicable domain.

[0032] Figure 9 This is a comparison chart of the EI predicted values ​​and experimental values ​​of the model in this embodiment of the invention for samples outside the applicable database. Detailed Implementation

[0033] The embodiments of the present invention are described in detail below, with reference to... Figure 1 The present invention provides a method for predicting the mechanical properties of medium and low alloy cast steel, comprising the following steps: Step 1: Collect sample data of low- and medium-alloy cast steel and establish a raw database of these samples. The raw database should include at least chemical composition data, heat treatment process data, and mechanical property data. It is understood that the sample data of low- and medium-alloy cast steel can be obtained from relevant publicly available literature. Chemical composition data should preferably be expressed as mass fractions, and the recorded constituent elements should at least include C, Si, Mn, Cr, Mo, and Ni. Further additions such as V, Nb, Ti, Al, and Cu can be made based on the completeness of the literature. The completed raw database can be shown in Table 1. Table 1. Original Database Parameters for Low Alloy Cast Steel

[0034] Step Two: The heat treatment process data is standardized according to preset rules to form heat treatment process features. The chemical composition data and heat treatment process features are then integrated into the original features of each low-alloy cast steel sample. Specifically, in this embodiment, the processing of each sample data is divided into three stages: normalizing, quenching, and tempering. The corresponding processing temperature, holding time, and cooling method are recorded for each stage. Here, T1, t1, and Cooling1 represent the temperature, holding time, and cooling method of the first stage of normalizing, respectively; T2, t2, and Cooling2 represent the temperature, holding time, and cooling method of the second stage of quenching, respectively; and Tt, tt, and Coolingt represent the temperature, holding time, and cooling method of the tempering stage, respectively. For samples that do not include a particular heat treatment stage, the corresponding table section is marked with " / ". Through the above standardization process, different heat treatment paths such as normalizing-tempering, quenching-tempering, and normalizing-quenching-tempering can be uniformly represented as standardized process data features suitable for subsequent physical information machine learning modeling. Furthermore, based on the presence or absence of normalizing and quenching stages, the sample heat treatment paths are categorized into normalizing-tempering, quenching-tempering, and normalizing-quenching-tempering types, and corresponding heat treatment path type identifiers are constructed. These, together with standardized process data features, constitute the standard heat treatment process features of the sample data. Specific processing rules are shown in Table 2. Table 2 Standardized Representation of Heat Treatment Process Data

[0035] Step 3: Construct physical information features based on the original features in the original database. In this embodiment, specifically, the physical information features include composition-side physical information features constructed based on chemical composition data, and process-side physical information features constructed based on heat treatment process features. The composition-side physical information features include at least carbon content (C), carbon equivalent (CE), and comprehensive features of carbide-forming elements. Atomic size mismatch and equivalent electronegativity Carbon content (C) is retained as a separate input variable to reflect the fundamental role of carbon content in steel on hardenability, matrix formation, carbide precipitation, and tempering response. Carbon equivalent (CE) comprehensively reflects the overall influence of elements such as C, Mn, Cr, Mo, V, Ni, and Cu on the hardening tendency and hardenability of steel; its calculation formula is as follows: (1) in, , , , , and These represent the mass fractions of elements C, Mn, Cr, Mo, V, Ni, and Cu, respectively.

[0036] Considering the phenomena of alloy carbide precipitation, tempering stability changes, and secondary hardening during the tempering process of medium and low alloy cast steel, the physical information characteristics of the composition are incorporated with the comprehensive characteristics of carbide-forming elements. This formula is used to comprehensively characterize the effects of carbide-forming elements such as Cr, Mo, W, V, Nb, and Ti on the evolution of tempering microstructure and mechanical property response. (2) in, , , , , and These represent the mass fractions of elements Cr, Mo, W, V, Nb, and Ti, respectively.

[0037] To further characterize the lattice effects and average chemical properties of multi-component alloy systems at the atomic scale, comprehensive physical descriptors such as atomic size mismatch and equivalent electronegativity are introduced, among which atomic size mismatch... The following formula is used to calculate: (3) in, The atomic fraction of the i-th element is calculated using the following formula: (4) in, Let i be the atomic weight of element i.

[0038] Let be the atomic radius of the i-th element. The weighted average atomic radius of all elements is calculated using the following formula: (5) Equivalent electronegativity The following formula is used to calculate: (6) in, Let be the Pauling electronegativity of the i-th element.

[0039] Process-side physical information features are used to characterize the nonlinear effects of heat treatment on microstructure and properties, mainly including temperature features, time features, tempering parameters, and cooling intensity mapping features. Among these, temperature features... The temperatures of the final austenitizing stage and the tempering stage were obtained by constructing S-shaped growth curves using the sigmoid function. The calculation formula is as follows: (7) Where T is the temperature of the corresponding stage. is the temperature scale parameter, and s is the temperature region characteristic parameter of the corresponding stage. The temperature parameters of each stage can be mapped to the 0-1 interval through the Sigmoid function, which is used to characterize the nonlinear effect of temperature on tissue transformation and performance response.

[0040] Time characteristics The calculation formulas are constructed by using logarithmic transformations on the final austenitizing stage time and the tempering stage time, respectively: (8) in, For the corresponding stage of heat preservation time, time characteristics It can characterize, to some extent, the nonlinear features of the effect of holding time on microstructure transformation or tempering process.

[0041] Tempering parameters are used to characterize the coupling effect of tempering temperature and tempering time, and to characterize the degree of tempering under the combined effect of tempering temperature and tempering time. The calculation formula is as follows: (9) in, This is the absolute temperature for tempering. Tempering time, This is an empirical constant.

[0042] To ensure the model considers the significant impact of the cooling path on the formation of the basic microstructure and subsequent tempering response of low- and medium-alloy cast steel after austenitization, the cooling methods were standardized and corresponding process characteristics were constructed. Specifically, the cooling methods described in the literature were uniformly categorized into furnace cooling, air cooling, oil cooling or polymer cooling, and water cooling. Based on the order of cooling intensity from weakest to strongest, the four cooling methods were assigned values ​​from low to high: furnace cooling, air cooling, oil cooling, polymer cooling, and water cooling were assigned values ​​of 1, 2, 3, and 4, respectively, forming the category characteristics of the cooling methods. And based on the characteristics of these cooling methods The values ​​are normalized and mapped to obtain the cooling intensity mapping characteristics. The calculation formula is shown below: (10) in, and These are the minimum and maximum values ​​of the cooling method category characteristic, respectively; that is, in this embodiment, , .

[0043] Step 4: Divide the collected sample data into training and validation sets. Use the physical information features in the training set, or the original features and physical information features, as input variables, and use the target physical property data from the mechanical property data in the training set as output variables to train the mechanical property prediction model for low-alloy cast steel. In this embodiment, considering the good adaptability and stability of the XGBoost regression algorithm in handling complex nonlinear relationships, feature coupling effects, and small sample regression tasks, the XGBoost regression algorithm is selected to train the mechanical property prediction model for low-alloy cast steel. When the target physical property data to be predicted is tensile strength (UTS), physical information features are used as input variables for model training; when the target physical property data to be predicted is elongation (EI), the original features and physical information features are fused as input variables for model training. It is understood that, in some embodiments, when the target physical property data to be predicted is elongation EI, plasticity-related enhancement features can also be introduced into the input variables to further improve the performance of the model. These plasticity-related enhancement features include one or more of the following: comprehensive features of microalloying elements, comprehensive features of toughness alloying elements, total alloying degree features, comprehensive features of impurity elements, interaction terms of carbon content and carbide-forming element comprehensive features, the ratio of carbon equivalent to toughness alloying element comprehensive features, cooling intensity features after final austenitization, cooling intensity features during normalizing, interaction terms of cooling method and tempering temperature features, interaction terms of cooling intensity and tempering temperature features, and heat treatment path coding features.

[0044] In this embodiment, the model training process is implemented in a Python environment using the PyCharm software platform. All sample data is divided into training and test sets in an appropriate ratio, such as an 8:2 ratio. To further improve the model's predictive performance, 5-fold cross-validation is used to evaluate the model's generalization ability on the training set. The training set is divided into 5 mutually exclusive subsets. In each round, 4 subsets are selected as sub-training sets to train the model, and the remaining subset is used as the validation set to obtain out-of-fold prediction values. This process is repeated 5 times, so that each subset serves as a validation set once. After each training, the sample weights are determined based on the error between the out-of-fold prediction value corresponding to the selected validation set and the true performance value. Samples with larger prediction errors in the validation set are assigned lower training weights. Subsequently, in each round of cross-validation, the model is retrained using the corresponding sub-training sets and their sample weights to reduce the interference of noisy or discrete samples on the model parameter learning. A Bayesian optimization framework based on the Optuna framework is used to search for model parameters to obtain a better parameter combination.

[0045] In some embodiments, to further improve model performance, robust normalization is used to scale the input variables during training, and Box-Cox transformation is applied to correct the distribution characteristics of the output target variable to reduce the interference of outliers on the feature distribution. It is understood that if the input and output variables are scaled during training, the data input to the model must also be scaled in the same way when using the model for prediction, and the model's prediction results must be restored to the original performance scale.

[0046] In some embodiments, to further improve model performance, before training, the correlation between various features and the target physical performance data can be evaluated by analyzing sample data, i.e., the contribution of each feature to the target physical performance data. Based on the evaluation results, the constructed physical information features are then screened to ensure their suitability as input variables for model training. It is understood that screening methods such as mutual information scoring can be used to measure the degree to which input variables reduce the uncertainty of output variables, thereby screening physical information features and using the screened features as input variables to train the model.

[0047] Step 5: Standardize the heat treatment process data of the low-alloy cast steel sample to be predicted, forming the predicted heat treatment process characteristics. Integrate the chemical composition data and the predicted heat treatment process characteristics of the low-alloy cast steel sample into the predicted original characteristics of the low-alloy cast steel sample. It is understood that the method of standardizing the heat treatment process data of the low-alloy cast steel sample to be predicted needs to be consistent with that used during training.

[0048] Step Six: Construct the physical information features to be predicted based on the chemical composition data and heat treatment process data of the low-alloy cast steel to be predicted.

[0049] Step 7: Input the physical information features to be predicted, or the original features to be predicted and the physical information features to be predicted, into the mechanical property prediction model for medium and low alloy cast steel to obtain the predicted mechanical properties of the medium and low alloy cast steel. It is understandable that if the input variables undergo preprocessing such as scaling during model training, then the original features to be predicted and the physical information features to be predicted also need to undergo the same preprocessing during actual prediction. After obtaining the prediction results, the prediction results output by the model should be restored to the scale of the original performance data.

[0050] To verify the effectiveness of this method, the same basic model was used to construct and train a standard XGBoost model (Raw-XGBoost model) with only the original features from the same sample data as input variables, and a PI-XGBoost model (PI-XGBoost model) trained using the method of this embodiment of the invention, which embeds physical information features, was compared. UTS and EI predictions were made on the same independent test set, separate from the training set, and compared with the true values ​​in the independent test set. The determination coefficient R was then used to determine the accuracy of the predictions. 2 The root mean square error (RMSE) and root mean square error (RMSE) are used as model evaluation metrics to assess the predictive performance of the two models. Among them, R... 2 R1 is used to characterize the goodness of fit between the model's predicted values ​​and the actual values; the closer the value is to 1, the better the model's fit. RMSE is used to characterize the magnitude of the model's prediction error; the smaller the value, the higher the model's prediction accuracy. The prediction performance of the two graphs on independent test sets is shown in Table 4. 2 The graphical comparison charts of RMSE and RMSE are as follows: Figure 2 and Figure 3 As shown: Table 3 Comparison of prediction performance of Raw-XGBoost and PI-XGBoost on independent test sets

[0051] From Table 3, Figure 2 and Figure 4 It can be seen that in UTS prediction, the PI-XGBoost model trained based on this method achieves higher R-values ​​on the independent test set. 2 The R² of the traditional Raw-XGBoost model is improved from 0.940 to 0.973, while the RMSE is reduced from 89.285 to 59.864. In EI prediction, the R² of the PI-XGBoost model trained based on this method on the independent test set is significantly improved. 2 The accuracy improved from 0.831 in the traditional Raw-XGBoost model to 0.869, while the RMSE decreased from 3.390 to 2.979, indicating that this method can significantly improve the model's ability to predict strength and plasticity.

[0052] To further clarify the one-to-one correspondence between the experimental true values ​​and model predictions for each test sample in the independent test set, the sample-by-sample prediction results of the PI-XGBoost model for the test samples are summarized in Tables 4 and 5. The test sample numbers in the tables are arranged in ascending order of experimental values, corresponding to... Figure 6 and Figure 7 The test sample numbers in the test samples are kept consistent.

[0053] It should be noted that in this embodiment, the number of valid test samples for the UTS target variable is 40 sets; since some literature records do not provide elongation data, the number of valid test samples for the E1 target variable is 39 sets. None of the test samples were used in model training, hyperparameter optimization, or robust weight determination.

[0054] Table 4 Comparison of experimental values ​​and PI-XGBoost predicted values ​​for independent UTS test samples

[0055] Table 5 Comparison of experimental values ​​and PI-XGBoost predicted values ​​for independent test samples.

[0056] Among them, NT: normalizing + tempering, QT: quenching + tempering, NQT: normalizing + quenching + tempering; Tables 4 and 5 further list the experimental values, predicted values, and errors for each sample in the independent test set, facilitating the verification of model prediction results at the individual sample level. Based on the data in Tables 4 and 5, the following plots are created: Figure 4 and Figure 5 The scatter plot showing the comparison between predicted and experimental values, and as shown in the figure... Figure 6 and Figure 7 The figure shows a comparison between the experimental and predicted values ​​for each sample. As can be seen from the figure, the predicted and experimental values ​​of the PI-XGBoost model trained based on the method of this invention are generally distributed along the ideal prediction line on the independent test set, indicating that the model has good fitting ability. The distribution of points in the UTS is more concentrated. While E1 still has a few deviation points due to the influence of casting defects, the dispersion of literature data, and the sensitivity of plasticity indicators, the overall trend and evaluation indicators are better than the traditional Raw-XGBoost model. Figure 6 and Figure 7 As shown, the prediction curve of the PI-XGBoost model of the present invention can follow the trend of the UTS and E1 experimental curves well, especially the high degree of fit in UTS prediction.

[0057] To further examine the predictive ability of the PI-XGBoost model trained by the method of this invention for samples outside the applicable database, statistical analysis was performed on the sample data in the original database used for model training, and the main feature range of the database is shown in Table 6: Table 6. Main Feature Range of Low- and Medium-Alloy Cast Steel Database

[0058] It should be noted that the range listed in Table 6 is only one of the basic conditions for determining the applicable domain of the model. It is also necessary to combine the heat treatment path and cooling method categories with existing categories in the database, as well as the similarity between the sample after it is converted into PI physical information features and the training sample in the PI feature space for judgment.

[0059] Subsequently, based on the above judgment conditions, experimental samples that met the model's applicable domain but were not included in the database were selected as verification objects. The selected samples could be existing medium-low alloy cast steel material samples, or newly designed or prepared medium-low alloy cast steel material samples. The nominal chemical composition of the selected samples outside the model's applicable domain was first given, and a composition table was formed. In this verification experiment, two samples were selected for verification; detailed compositions are shown in Tables 7 and 8. Table 7. Main elemental composition (wt%) of Sample 1

[0060] Table 8. Main elemental composition (wt%) of Sample 2

[0061] Subsequently, based on the compositional characteristics and research needs of the validation samples outside the database within the model's applicable domain, a corresponding heat treatment regime was formulated, and a heat treatment process table was generated. It is understood that the heat treatment process table here records heat treatment parameters according to the standardized method described above in this invention, namely, recording the treatment temperature, holding time, and cooling method for the normalizing, quenching, and tempering stages respectively. Simultaneously, depending on whether the validation sample includes normalizing and quenching stages, its heat treatment path is identified, and the final austenitization and cooling parameters before tempering are determined according to the aforementioned rules.

[0062] After heat treatment of the two samples, room temperature tensile tests were performed. Based on the recorded engineering stress-strain curves, the true values ​​of the tensile strength (UTS) and elongation (E1) of the specimens were measured and calculated. At least three valid samples were tested under each heat treatment condition, and the arithmetic mean of the final results was used as the true value for the validation sample. The specific data are shown in Tables 9 and 10. Table 9. Heat treatment process and mechanical properties of Sample 1

[0063] Table 10 Heat treatment process and mechanical properties of Sample 2

[0064] The chemical composition and heat treatment data of the two samples were standardized and their physical information features were constructed using the same process described above. These features were then input into the PI-XGBoost model trained using the method of this invention to obtain the predicted UTS and EI values ​​for samples outside the database within the applicable domain of the two models. Finally, the predicted UTS and EI values ​​for samples outside the database within the applicable domain of the two models were compared with the actual UTS and EI values ​​obtained from testing the actual samples to evaluate the predictive performance of the PI-XGBoost model for samples outside the database within the applicable domain. The comparison results are shown in Tables 11 and 12, and plotted based on the comparison results as follows: Figure 8 and Figure 9 The comparison chart shown: Table 11 Detailed results of tensile strength prediction for samples outside the database

[0065] Table 12 Detailed results of out-of-database sample extension rate prediction

[0066] From Table 11, Table 12, Figure 8 and Figure 9 The comparison results show that the model obtained by the method of this invention has an average relative error of only 0.58% in predicting the tensile strength of the samples, with a maximum relative error not exceeding 1.11%. Elongation, as a plasticity index highly sensitive to microstructure, is much more difficult to predict than strength indexes. By further integrating the original composition, original process parameters, and plasticity-related enhancement features on the basis of physical information characteristics, the maximum relative error was controlled within 10%. These results demonstrate that the method of this invention has a certain predictive adaptability to samples outside the database within the model's applicable domain. It can provide reliable theoretical guidance for the composition design and process optimization of cast steel materials, reduce the number of experiments, shorten the R&D cycle, and lower R&D costs.

[0067] It should be noted that in the description of this invention, any descriptions of orientation, such as up, down, front, back, left, right, etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings. They are only for the purpose of facilitating the description of this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or operated in a specific orientation, and should not be construed as a limitation of this invention.

[0068] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first" or "second" is mentioned, it is only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0069] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0070] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting the mechanical properties of medium and low alloy cast steel, characterized in that, include: Collect sample data of medium and low alloy cast steel and establish a raw database of medium and low alloy cast steel samples. The raw database shall include at least chemical composition data, heat treatment process data and mechanical property data. According to the preset rules, the heat treatment process data is uniformly expressed to form heat treatment process characteristics, and the chemical composition data and heat treatment process characteristics are integrated into the original characteristics of each medium and low alloy cast steel sample data. Physical information features are constructed based on the original features in the original database; The collected sample data is divided into a training set and a validation set. The physical information features in the training set, or the original features and physical information features, are used as input variables, and the target physical performance data in the mechanical performance data in the training set are used as output variables to train the mechanical performance prediction model of low alloy cast steel. The heat treatment process data of the low-alloy cast steel sample to be predicted are uniformly expressed to form the heat treatment process characteristics to be predicted. The chemical composition data and heat treatment process characteristics of the low-alloy cast steel sample to be predicted are integrated into the original characteristics to be predicted of the low-alloy cast steel sample. The physical information characteristics to be predicted are constructed based on the chemical composition data and heat treatment process data of the low-alloy cast steel to be predicted; The physical information features to be predicted, or the original features to be predicted and the physical information features to be predicted, are input into the mechanical property prediction model of medium and low alloy cast steel to obtain the predicted mechanical properties of the medium and low alloy cast steel.

2. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 1, characterized in that, The physical information features include composition-side physical information features constructed based on chemical composition data and process-side physical information features constructed based on heat treatment process data. The composition-side physical information features include carbon content, carbon equivalent, comprehensive characteristics of carbide-forming elements, atomic size mismatch, and equivalent electronegativity. The process-side physical information features include temperature features, time features, tempering parameters, and cooling intensity mapping features.

3. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The carbon equivalent CE is calculated using the following formula: in, , , , , and These are the mass fractions of elements C, Mn, Cr, Mo, V, Ni, and Cu, respectively. The comprehensive characteristics of carbide-forming elements The following formula is used to calculate: in, , , , , and These represent the mass fractions of elements Cr, Mo, W, V, Nb, and Ti, respectively.

4. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The atomic size mismatch The following formula is used to calculate: in, Let i be the atomic fraction of the i-th element. Let be the atomic radius of the i-th element. This is the weighted average atomic radius of all elements.

5. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The equivalent electronegativity The following formula is used to calculate: in, Let i be the atomic fraction of the i-th element. Let be the Pauling electronegativity of the i-th element.

6. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The temperature characteristics are constructed by using S-shaped growth curve functions for the final austenitizing stage temperature and the tempering stage temperature, respectively, and the time characteristics are constructed by using logarithmic functions for the final austenitizing stage time and the tempering time, respectively.

7. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The tempering parameters The following formula is used to calculate: in, This is the absolute temperature for tempering. Tempering time, This is an empirical constant.

8. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 2, characterized in that, The cooling intensity mapping feature The following formula is used to calculate: in, The cooling method category is determined by classifying the cooling methods in the heat treatment process data, sorting the categorized cooling methods according to their cooling intensity, and assigning values ​​from weakest to strongest, and from smallest to largest. This represents the value of the cooling intensity category characteristic for the cooling method with the weakest cooling intensity. This represents the value of the cooling intensity category characteristic for the cooling method with the strongest cooling intensity.

9. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 1, characterized in that, The steps of dividing the collected sample data into a training set and a validation set, and using the composition-side physical information features and process-side physical information features in the training set as input variables to train the mechanical property prediction model for low-alloy cast steel include: Calculate the contribution of each physical information feature to the target physical performance, and filter the physical information features based on the contribution. The collected sample data is divided into a training set and a validation set. The physical information features that pass the screening in the training set, or the original features and the physical information features that pass the screening, are used as input variables to train the prediction model of the mechanical properties of low alloy cast steel.

10. The method for predicting the mechanical properties of medium and low alloy cast steel according to claim 1, characterized in that, The mechanical property prediction model for medium and low alloy cast steel is an XGBoost model. During the training process of the mechanical property prediction model for medium and low alloy cast steel, the model is trained and its parameters are tuned through 5-fold cross-validation and a Bayesian optimization framework based on the Optuna framework. During the training process, a robust normalization method is used to scale the input variables, and a Box-Cox transformation is used to correct the distribution of the output variables.

Images