Method for predicting hardenability of 8620H steel and iron material

Abstract

The invention relates to a method for predicting the hardenability of an 8620H steel and iron material, and belongs to the technical field of steel and iron material quality control methods. According to the technical scheme, on the basis of a 8620H steel and iron material hardenability data set, more than one machine learning algorithm is applied to perform model training, 8620H steel and iron material hardenability prediction models of the algorithms are constructed respectively, and key hardenability indexes J7.9 and J12.7 of steel are predicted; and a ten-fold cross validation method is adopted to evaluate the training model, evaluation indexes are used to measure the accuracy and reliability of each model, and an optimal hardenability prediction model is determined according to the accuracy and reliability. The method has the beneficial effects that the high-precision hardenability prediction model is constructed, the hardenability of the 8620H steel and iron material to be tested can be accurately predicted, and the method has the advantages of high precision, low cost and rapid implementation of material hardenability performance evaluation.

Description

Technical Field

[0001] The invention relates to a method for predicting the hardenability of 8620H steel material, belonging to the technical field of steel material quality control methods. Background Art

[0002] 8620H steel is a high-performance, low-carbon alloy steel. Due to its excellent mechanical properties and good processing characteristics, it is widely used in industries such as petroleum, chemical, automotive, aerospace, and machinery manufacturing. Hardenability is a key indicator for evaluating whether 8620H steel can obtain a uniform hardened layer during heat treatment. This performance indicator directly affects the service life and overall performance of the material. The level of hardenability is determined by many factors, including the chemical composition of the steel, the size of the austenite grains, and the degree of homogenization. The Jomini end quenching test, a traditional method for evaluating hardenability, determines the hardenability of steel by measuring the hardness gradient from the cooling end to the hot end.

[0003] Globally, controlling the hardenability of high-quality steel has become central to improving the quality and production efficiency of steel products. While the hardenability bandwidth of high-quality gear steel can be strictly controlled within an extremely narrow range internationally, domestic products still lag behind in this regard. Traditional hardenability prediction methods, such as empirical formulas and simplified analytical models, often fail to fully consider the complexity of material composition and the dynamic changes in microstructure in actual production, limiting the applicability and versatility of these models. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting the hardenability of 8620H steel material. By combining advanced data processing and machine learning technologies, the limitations of traditional prediction methods are overcome, and a high-precision hardenability prediction model is constructed. The method can accurately predict the hardenability of the 8620H steel material to be tested, has the advantages of high precision, low cost, and rapid implementation of material hardenability performance evaluation, and effectively solves the above-mentioned problems existing in the background technology.

[0005] The technical solution of the present invention is: a method for predicting the hardenability of 8620H steel material, comprising the following steps: Step S1: Collect relevant data of 8620H steel material from the production line, filter and clean the original data to form a hardenability data set of 8620H steel material; Step S2: Based on the 8620H steel material hardenability dataset, apply one or more machine learning algorithms to perform model training, and respectively construct 8620H steel material hardenability prediction models for each algorithm to predict the key hardenability indicators J7.9 and J12.7 of the steel; Step S3: The training models are evaluated using a ten-fold cross-validation method. The accuracy and reliability of each model are measured using evaluation indicators, and the optimal hardenability prediction model is determined accordingly.

[0006] The step S1 specifically includes the following steps: Step S11: Data collection: Collect production line data, including the steel's hardenability index J7.9 and J12.7 values, chemical element composition, and heat treatment process parameters; Step S12: Data preprocessing: missing value processing, outlier detection and data cleaning are performed on the initially collected data in sequence; ① Missing value processing: remove all records containing missing values ​​and null values; ② Outlier processing: Use box plots and histograms to detect and process outliers in the data; ③ Data cleaning: Process the data in the data set that have the same chemical composition and heat treatment process parameters but different J values.

[0007] The outlier detection in step S12 uses a box plot and histogram method to calculate the quartile and interquartile range of each data point. Based on the interquartile range, the outlier limit is defined as 1.5 times the IQR, and out-of-range outlier data points are removed. For the J7.9 and J12.7 hardness values, their normal ranges are determined and out-of-range data points are removed. The frequency distribution of the J7.9 and J12.7 values ​​is displayed using a histogram and a kernel density estimation curve to clarify the outlier threshold. Data cleaning includes: ① Group inspection: group data with the same chemical composition and process parameters into one group; ② Group data processing: if there are only two data in a group and the difference in J value exceeds 4 HRC, delete these two data; if the difference does not exceed 4 HRC, update the J value to the average of the two; ③ Multi-data group processing: for groups containing more than two data, calculate the median J value, eliminate the data with a difference of more than 4 HRC from the median, and then update the J value of the group with the average of the remaining data.

[0008] The step S2 specifically includes: Step S21: feature extraction, extracting alloy component characteristic variables and corresponding target variables J7.9 and J12.7 from the 8620H steel material hardenability data set; Step S22: Algorithm selection: selecting a classic machine learning algorithm widely used in prediction tasks, including RF, XGBoost, and LightGBM; Step S23: Model construction, using the selected machine learning algorithm to construct a hardenability prediction model for 8620H steel material, which predicts the hardness of the end-quenched sample at different distances from the end based on the alloy composition.

[0009] The step S3 specifically includes: Step S31: cross validation, verifying the generalization ability of different models through ten-fold cross validation technology; Step S32: Performance evaluation, summarizing the results of each cross-validation, and evaluating the model performance by determining the coefficient R2, the root mean square error RMSE, and the proportion of predicted values ​​within the ±2 HRC bandwidth; Step S33: Model selection. Based on the performance evaluation results, the model with the highest R² value, the lowest RMSE value, and the largest proportion of predicted values ​​within the ±2 HRC bandwidth is selected as the final prediction model.

[0010] The beneficial effects of the present invention are as follows: by combining advanced data processing and machine learning technologies, the limitations of traditional prediction methods are overcome, and a high-precision hardenability prediction model is constructed, which can accurately predict the hardenability of the 8620H steel material to be tested, and has the advantages of high precision, low cost, and rapid implementation of material hardenability performance evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Schematic diagram of the process of threshold identification and outlier removal in J7.9 of the present invention; Figure 2 Schematic diagram of the process of threshold identification and outlier removal in J12.7 of the present invention; Figure 3 It is the frequency distribution diagram of J7.9 of the present invention; Figure 4 It is the frequency distribution diagram of J12.7 of the present invention; Figure 5 This is the J7.9 value prediction result diagram using the LightGBM model in the present invention; Figure 6 This is a graph showing the J7.9 value prediction results using the XGBoost model in the present invention; Figure 7 This is a graph showing the J12.7 value prediction results using the RF model of the present invention; Figure 8 This is a graph showing the J12.7 value prediction results using the LightGBM model in the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the invention implementation cases clearer, the technical solutions in the invention implementation cases will be clearly and completely described below in conjunction with the drawings in the implementation cases. Obviously, the implementation cases described are only a small part of the implementation cases of the present invention, rather than all the implementation cases. Based on the implementation cases in the present invention, all other implementation cases obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0013] A method for predicting the hardenability of 8620H steel material comprises the following steps: Step S1: Collect relevant data of 8620H steel material from the production line, filter and clean the original data to form a hardenability data set of 8620H steel material; Step S2: Based on the 8620H steel material hardenability dataset, apply one or more machine learning algorithms to perform model training, and respectively construct 8620H steel material hardenability prediction models for each algorithm to predict the key hardenability indicators J7.9 and J12.7 of the steel; Step S3: The training models are evaluated using a ten-fold cross-validation method. The accuracy and reliability of each model are measured using evaluation indicators, and the optimal hardenability prediction model is determined accordingly.

[0014] The step S1 specifically includes the following steps: Step S11: Data collection: Collect production line data, including the steel's hardenability index J7.9 and J12.7 values, chemical element composition, and heat treatment process parameters; Step S12: Data preprocessing: missing value processing, outlier detection and data cleaning are performed on the initially collected data in sequence; ① Missing value processing: remove all records containing missing values ​​and null values; ② Outlier processing: Use box plots and histograms to detect and process outliers in the data; ③ Data cleaning: Process the data in the data set that have the same chemical composition and heat treatment process parameters but different J values.

[0015] The outlier detection in step S12 uses a box plot and histogram method to calculate the quartile and interquartile range of each data point. Based on the interquartile range, the outlier limit is defined as 1.5 times the IQR, and out-of-range outlier data points are removed. For the J7.9 and J12.7 hardness values, their normal ranges are determined and out-of-range data points are removed. The frequency distribution of the J7.9 and J12.7 values ​​is displayed using a histogram and a kernel density estimation curve to clarify the outlier threshold. Data cleaning includes: ① Group inspection: group data with the same chemical composition and process parameters into one group; ② Group data processing: if there are only two data in a group and the difference in J value exceeds 4 HRC, delete these two data; if the difference does not exceed 4 HRC, update the J value to the average of the two; ③ Multi-data group processing: for groups containing more than two data, calculate the median J value, eliminate the data with a difference of more than 4 HRC from the median, and then update the J value of the group with the average of the remaining data.

[0016] The step S2 specifically includes: Step S21: feature extraction, extracting alloy component characteristic variables and corresponding target variables J7.9 and J12.7 from the 8620H steel material hardenability data set; Step S22: Algorithm selection: selecting a classic machine learning algorithm widely used in prediction tasks, including RF, XGBoost, and LightGBM; Step S23: Model construction, using the selected machine learning algorithm to construct a hardenability prediction model for 8620H steel material, which predicts the hardness of the end-quenched sample at different distances from the end based on the alloy composition.

[0017] The step S3 specifically includes: Step S31: cross validation, verifying the generalization ability of different models through ten-fold cross validation technology; Step S32: Performance evaluation, summarizing the results of each cross-validation, and evaluating the model performance by determining the coefficient R2, the root mean square error RMSE, and the proportion of predicted values ​​within the ±2 HRC bandwidth; Step S33: Model selection. Based on the performance evaluation results, the model with the highest R² value, the lowest RMSE value, and the largest proportion of predicted values ​​within the ±2 HRC bandwidth is selected as the final prediction model.

[0018] In practical applications, the present invention overcomes the limitations of traditional prediction methods and adopts advanced data processing and machine learning technologies to construct a new and efficient hardenability prediction method. First, the present invention uses three machine learning models, RF, XGBoost and LightGBM, to conduct in-depth research on the chemical composition and heat treatment process parameters of 8620H steel materials. By adopting a ten-fold cross-validation method, these models were systematically evaluated and a high-precision hardenability prediction model was successfully constructed, especially in predicting the hardenability index J7.9 and J12.7 values. The vast majority of the prediction results fall within the ±2 HRC error range. The specific steps are as follows: Step S1: Collect relevant data of 8620H steel materials from the production line, and implement methods including missing value processing, outlier detection and data cleaning to filter and clean the raw data to form a hardenability dataset of 8620H steel materials; Wherein, the step S1 specifically includes the following steps: Step S11: Data collection: Collect production line data from Hebei Iron and Steel Group Shigang Company. The data includes the steel's hardenability index (J7.9 and J12.7 values), chemical element composition (28 chemical elements such as C, Si, Mn, Cr, Mo, Ni, etc.), and heat treatment process parameters (normalizing temperature and time, quenching temperature and time) as the basis for subsequent analysis.

[0019] Step S12: Data preprocessing: The preliminarily collected data is processed with missing values, detected with outliers, and cleaned to improve data quality and consistency.

[0020] ① Missing value processing: remove all records containing missing values ​​and null values; ② Outlier handling: Boxplots and histograms were used to detect and handle outliers in the data. Specifically, the quartiles of each data point were calculated and the outlier threshold was defined as 1.5 times the interquartile range (IQR). Outlier data points were removed. For J7.9 and J12.7 hardness values, the normal range was determined and outliers were removed. Histograms and kernel density estimation (KDE) curves were used to display the frequency distribution of J7.9 and J12.7 values ​​and to clarify the outlier threshold. Figure 1 and Figure 2 The process of identifying and removing outliers by setting thresholds is shown. Data points exceeding these thresholds are identified as outliers and removed from the dataset. Figure 3 and Figure 4 The frequency distribution of J7.9 and J12.7 values ​​is presented using histograms and kernel density estimation (KDE) curves. These charts not only show the data distribution trend but also mark the boundaries of outliers with dotted lines. J7.9 values ​​are mainly concentrated around 33 HRC, while J12.7 values ​​are usually around 28 HRC.

[0021] ③ Data cleaning: For data in the data set with the same chemical composition and heat treatment process parameters but different J values, processing is performed; the specific steps include: 1) Group check: Data with the same chemical composition and process parameters are grouped together; 2) Small group data processing: If there are only two data in a group and the difference in J value exceeds 4 HRC, these two data are deleted; if the difference does not exceed 4 HRC, the J value is updated to the average of the two; 3) Multi-data group processing: For groups containing more than two data, the median J value is calculated, and the data with a difference of more than 4 HRC from the median are eliminated, and then the J value of the group is updated with the average of the remaining data.

[0022] Step S2: Based on the obtained 8620H steel material hardenability dataset, multiple machine learning algorithms including linear regression, K-nearest neighbor, random forest, and LightGBM are applied to perform model training. 8620H steel material hardenability prediction models of each algorithm are constructed to predict the key hardenability indicators J7.9 and J12.7 of the steel. Wherein, the step S2 specifically includes the following steps: Step S21: Feature extraction, extracting alloy component characteristic variables and corresponding target variables J7.9 and J12.7 from the 8620H steel material hardenability data set.

[0023] Step S22: Algorithm selection: select a classic machine learning algorithm that is widely used in prediction tasks, including RF, XGBoost, and LightGBM.

[0024] Step S23: Model construction, using the selected machine learning algorithm to construct a hardenability prediction model for 8620H steel material, which predicts the hardness of the end-quenched sample at different distances from the end based on the alloy composition.

[0025] Step S3: The trained models are evaluated using a ten-fold cross-validation method. Evaluation indicators such as R2, RMSE, and the proportion of predicted values ​​within the ±2 HRC bandwidth are used to measure the accuracy and reliability of each model, and the optimal hardenability prediction model is determined accordingly. Wherein, the step S3 specifically includes the following steps: Step S31: Cross-validation, verifying the generalization ability of different models through ten-fold cross-validation technology.

[0026] Step S32: Performance evaluation, summarize the results of each cross-validation, and evaluate the model performance by determining the coefficient R2, the root mean square error RMSE, and the proportion of predicted values ​​within the ±2 HRC bandwidth.

[0027] Step S33: Model selection. Based on the performance evaluation results, the model with the highest R² value, lowest RMSE value, and the largest proportion of predicted values ​​within the ±2 HRC bandwidth was selected as the final prediction model. The XGBoost model performed best when predicting J7.9, achieving 95.40% accuracy and an R² value of 0.876. When predicting J12.7, the RF and LightGBM models achieved prediction accuracies exceeding 95%. The LightGBM model achieved the best prediction performance, with an R² value of 0.902, an RMSE of 0.579, and 98.03% of predictions within the ±2 HRC range.

[0028] The above description is merely an embodiment of the present invention and does not constitute any form of limitation to the present invention. The present invention may also have other forms of embodiments based on the above structures and functions, which are not listed here one by one. Therefore, any simple modification, equivalent changes, and modifications made to the above embodiments by any person skilled in the art in accordance with the technical essence of the present invention without departing from the scope of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for predicting the hardenability of 8620H steel material, characterized in that The following steps are involved: Step S1: Collect relevant data of 8620H steel material from the production line, filter and clean the original data to form a hardenability data set of 8620H steel material; Step S2: Based on the 8620H steel material hardenability dataset, apply one or more machine learning algorithms to perform model training, and respectively construct 8620H steel material hardenability prediction models for each algorithm to predict the key hardenability indicators J7.9 and J12.7 of the steel; Step S3: The training models are evaluated using a ten-fold cross-validation method. The accuracy and reliability of each model are measured using evaluation indicators, and the optimal hardenability prediction model is determined accordingly.

2. The method for predicting the hardenability of 8620H steel material according to claim 1, wherein: The step S1 specifically includes the following steps: Step S11: Data collection: Collect production line data, including the steel's hardenability index J7.9 and J12.7 values, chemical element composition, and heat treatment process parameters; Step S12: Data preprocessing: missing value processing, outlier detection and data cleaning are performed on the initially collected data in sequence; ① Missing value processing: remove all records containing missing values ​​and null values; ② Outlier processing: Use box plots and histograms to detect and process outliers in the data; ③ Data cleaning: Process the data in the data set that have the same chemical composition and heat treatment process parameters but different J values.

3. The method for predicting the hardenability of 8620H steel material according to claim 2, characterized in that: The outlier detection in step S12 uses a box plot and histogram method to calculate the quartile and interquartile range of each data point. Based on the interquartile range, the outlier limit is defined as 1.5 times the IQR, and out-of-range outlier data points are removed. For the J7.9 and J12.7 hardness values, their normal ranges are determined and out-of-range data points are removed. The frequency distribution of the J7.9 and J12.7 values ​​is displayed using a histogram and a kernel density estimation curve to clarify the outlier threshold. Data cleaning includes: ① Group inspection: group data with the same chemical composition and process parameters into one group; ② Group data processing: if there are only two data in a group and the difference in J value exceeds 4 HRC, delete these two data; if the difference does not exceed 4 HRC, update the J value to the average of the two; ③ Multi-data group processing: for groups containing more than two data, calculate the median J value, eliminate the data with a difference of more than 4 HRC from the median, and then update the J value of the group with the average of the remaining data.

4. The method for predicting the hardenability of 8620H steel material according to claim 1, wherein: The step S2 specifically includes: Step S21: feature extraction, extracting alloy component characteristic variables and corresponding target variables J7.9 and J12.7 from the 8620H steel material hardenability data set; Step S22: Algorithm selection: selecting a classic machine learning algorithm widely used in prediction tasks, including random forest (RF), extreme gradient boosting (XGBoost), and lightweight gradient boosting machine (LightGBM); Step S23: Model construction, using the selected machine learning algorithm to construct a hardenability prediction model for 8620H steel material, which predicts the hardness of the end-quenched sample at different distances from the end based on the alloy composition.

5. The method for predicting the hardenability of 8620H steel material according to claim 1, characterized in that: The step S3 specifically includes: Step S31: cross validation, verifying the generalization ability of different models through ten-fold cross validation technology; Step S32: Performance evaluation, summarizing the results of each cross-validation, and evaluating the model performance by determining the coefficient R2, the root mean square error RMSE, and the proportion of predicted values ​​within the ±2 HRC bandwidth; Step S33: Model selection. Based on the performance evaluation results, the model with the highest R² value, the lowest RMSE value, and the largest proportion of predicted values ​​within the ±2 HRC bandwidth is selected as the final prediction model.

Images