Machine learning guidance laser forming ultrahigh strength and toughness steel design method based on fusion of physical and chemical characteristics

Abstract

The invention discloses a machine learning guidance laser forming ultrahigh strength and toughness steel design method based on fusion of physical and chemical characteristics, which combines the physical and chemical characteristics with Shapley additive interpretation and analysis, and overcomes the limitation that a traditional pure data driving model lacks theoretical support. The quantitative analysis of the influence between the input characteristics and the output performance is realized, the internal association between the physical and chemical characteristics and the material strengthening and toughening theoretical mechanism is disclosed, the contribution degree and the interaction effect of each alloy element and the heat treatment process parameters are determined, the theoretical basis of variable screening is provided for multi-objective optimization, and the optimization efficiency is improved. Therefore, the efficiency of obtaining the non-dominated solution set is remarkably improved. According to the method, the physical and chemical characteristic combination can be conveniently and rapidly determined, the relation between the input characteristic and the output performance can be quantified, and optimal design of the laser forming ultrahigh strength and toughness steel alloy components and the heat treatment process is achieved.

Description

Technical Field

[0001] This invention relates to the field of machine learning application technology and steel material design, and in particular to a machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the integration of physicochemical features. Background Technology

[0002] With the advancement of technology, the next generation of high-performance steels is continuously developing towards multi-alloying, lightweighting, and higher performance. To obtain ultra-high-strength steels with better strength-toughness matching and superior overall performance, researchers typically employ methods such as controlling alloy composition, improving preparation processes, and optimizing heat treatment regimes. However, this often requires extensive experimentation, leading to long development cycles and high costs. In recent years, data-driven methods, represented by machine learning, have provided new insights into the composition design, process optimization, and performance improvement of ultra-high-strength and tough steels.

[0003] However, directly establishing machine learning models relating material composition, processing, and performance faces certain bottlenecks. On the one hand, these models heavily rely on explicit mappings between chemical composition, processing parameters, and performance, requiring high-quality datasets. On the other hand, obtaining a large amount of effective experimental data to build a machine learning database presents a challenge. In the field of machine learning-guided development of high-performance materials, researchers have proposed introducing intermediate quantities into machine learning modeling methods to overcome the problem of insufficient information in small sample datasets. Currently, the acquisition of intermediate quantities can be mainly divided into three types: physical metallurgical model calculation, thermodynamic phase diagram calculation, and physicochemical characteristic quantity calculation. Among these, physical metallurgical models heavily rely on empirical models and have poor extrapolation; thermodynamic phase diagram calculations, due to the large number of alloying elements and high compositional dimensionality, lead to a surge in computational power requirements, significantly increasing costs while decreasing accuracy; while physicochemical characteristic quantity calculations are low-cost, can be directly obtained from the basic properties of elements, and have clear physical meanings, and have been applied to guide machine learning modeling for performance prediction of copper alloys, aluminum alloys, titanium alloys, and high-entropy alloys. It is worth noting that the physicochemical characteristics closely related to performance differ for different material systems and different preparation methods. Currently, ultra-high strength and toughness steel produced by laser additive manufacturing has been widely used in important sectors of the national economy such as aerospace, marine engineering and transportation due to its excellent comprehensive mechanical properties. However, there is still no machine learning method that integrates physical and chemical characteristics to guide the design of high-performance steel. Therefore, developing a laser-formed high-performance steel design method with low computational cost, high efficiency and clear physical meaning has become an urgent problem to be solved. Summary of the Invention

[0004] To overcome the shortcomings of existing machine learning methods in guiding the design of high-performance steel, such as unclear theoretical mechanisms, strong data dependence, and poor model interpretability, this invention proposes a machine learning-assisted laser-formed ultra-high strength and toughness steel design method that integrates physicochemical features. Unlike traditional modeling methods driven solely by composition-process-performance data, this invention significantly reduces dependence on large-scale datasets by introducing physicochemical features, thereby improving the prediction accuracy and generalization ability of the machine learning model. Furthermore, by combining the SHAPley Additive exPlanation (SHAP) analytical framework, the influence mechanism between feature variables and performance is quantitatively analyzed, effectively enhancing the theoretical interpretability and physical consistency of the model. Based on weight analysis and multi-objective optimization strategies, the synergistic effect of key alloying factors and heat treatment parameters is clarified, achieving composition design and heat treatment process optimization for ultra-high strength and toughness steel with low cost and high performance as the goals.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the fusion of physicochemical features is characterized by the following steps: Step S1: Construct the alloying factors corresponding to the composition of ultra-high strength steel based on physicochemical characteristics; Step S2: Screen out the key alloying factors that affect the ultimate tensile strength (UTS), yield strength (YS), and elongation (EL) from the alloying factors. Step S3: Construct a dataset consisting of key alloying factors, heat treatment process parameters, and tensile mechanical properties; and use the dataset for training and testing machine learning models to predict UTS, YS, and EL. Step S4: Use the Optuna visualization tool to optimize hyperparameters, employ a double early stopping strategy, and set the number of iterations to 500 to obtain the optimized machine learning models of UTS, YS, and EL. Step S5: Using SHAP analysis, analyze the optimized machine learning models of UTS, YS, and EL to determine the importance ranking of alloying elements and heat treatment processes; Step S6: Combining the SHAP analysis results with knowledge in the field of materials science, set the range of alloying elements and the range of heat treatment process parameters; Step S7: Based on the Multi-objective Optimization in Python (Pymoo) framework, the algorithm solves the problem under the guidance of the Pareto optimality principle. By setting the corresponding optimization parameters, the algorithm efficiently searches for the non-dominated solution set in the objective space to obtain the Pareto solution.

[0006] In a preferred embodiment, the physicochemical characteristics in step S1 include 81 types, which can be divided into six categories: atomic number, number of groups, Mendeleev number, electrochemical properties, and atomic size parameters; the composition of ultra-high strength steel includes the composition corresponding to ultra-high strength steel prepared by laser directed energy deposition and laser selective melting technology obtained through literature collection.

[0007] In a preferred embodiment, step S1, which involves constructing the alloying factor corresponding to the composition of ultra-high-strength steel based on physicochemical characteristics, includes the average alloying factor. f mi and variance alloying factor f vi The results are obtained by formulas (1) and (2) respectively: (1) (2) Where, represents the feature quantity associated with a single element. i Different characteristic quantities representing each element ( i =1, 2, …,81). j Corresponding to the different elements present in ultra-high strength steel, ( j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The percentage content of atoms.

[0008] In a preferred embodiment, step S2 includes: Step S21: Use the Extreme Gradient Boosting (XGBoost) algorithm to quantify and rank the contribution weights of the alloy factor. Step S22: Calculate any two alloying factors using formula (3). x and y Pearson correlation coefficient r : r=\frac {\sum{[({x}_{i}-{x}_{m})\times ({y}_{i}-{y}_{m})]}} {\sqrt {{({x}_{i}-{x}_{m})}^{2}\times \sqrt {{({y}_{i}-{y}_{m})}^{2}}}} (3) in, x i and y i Indicates the first i The values ​​of the two alloying factors corresponding to each sample. x m and y m These are the means of the two alloy factors in the attribute dataset, respectively; when | r When | > 0.95, it proves the alloy factor x and y There is a strong linear correlation between them, indicating that they provide redundant information in revealing the material property mechanism; in order to mitigate the effects of multicollinearity, factors with lower importance rankings are systematically eliminated, and the iterative elimination process continues until the top 10 alloy factors that have the most significant impact on performance are finally determined. Step S23: From the top 10 most significant alloy factors, select the optimal combination of alloy factors that achieves the highest performance prediction accuracy and define it as the key alloy factor; during the modeling process, randomly divide the data in the dataset into training and test sets in a 4:1 ratio; construct an exhaustive calculation model and use the ten-fold cross-validation method to optimize the established model parameters for ultimate tensile strength, yield strength, and elongation after fracture; the model accuracy is evaluated by the Mean Absolute Percentage Error (MAPE); Step S24: Determine the key alloying factors affecting UTS, YS, and EL based on the MAPE values ​​of the exhaustive calculation model.

[0009] In a preferred embodiment, the machine learning model in step S3 compares the fitting results of six algorithms, including XGBoost, Random Forest Regression (RFR), Gaussian Process Regression (GPR), Decision Tree (DT), K-Nearest Neighbors (KNN), and Support Vector Regression (SVR), using the mean absolute error (MAE) of the model trained by the ten-fold cross-validation method as the evaluation criterion, and adopts the algorithm with the lowest MAE value for model training and testing.

[0010] In a preferred embodiment, step S4 uses the SHAP analysis method to quantify the relationship between the input and output of the machine learning model; and uses weighted analysis to rank the importance of key alloy element composition and heat treatment process parameters.

[0011] This invention proposes a machine learning multi-objective optimization design method that combines physicochemical features. By combining SHAP analysis, the relationship between input and output is directly quantified, thereby improving the accuracy and generalization ability of the model and providing reliable support for related research.

[0012] By adopting the above technical solution, the composition of ultra-high strength and toughness steel alloy with excellent comprehensive performance and the corresponding heat treatment process parameters can be quickly identified, which is low in cost and high in efficiency.

[0013] The beneficial effects and advantages of the present invention are as follows: (1) Compared with the prior art, the machine learning-guided laser forming ultra-high strength and toughness steel design method based on the integration of physical and chemical features proposed in the present invention transforms the material science theoretical mechanism into quantifiable physical and chemical features by introducing physical and chemical features and machine learning modeling, significantly reducing the dependence on data scale, enhancing the model's ability to capture the complex correlation between composition, process and performance, thereby greatly improving the accuracy of performance prediction.

[0014] (2) This invention combines physicochemical characteristics with SHAP interpretability analysis, which overcomes the limitations of traditional pure data-driven models that lack theoretical support, realizes the quantitative analysis of the influence between input characteristics and output performance, reveals the intrinsic relationship between physicochemical characteristics and theoretical mechanisms such as material toughening, clarifies the contribution and interaction effects of each alloy element and heat treatment process parameters, provides a theoretical basis for variable screening for multi-objective optimization, and thus significantly improves the efficiency of obtaining non-dominated solution sets.

[0015] The technical solution of the present invention will be further described in detail below with reference to the appendix and embodiments. Attached Figure Description

[0016] Figure 1 The main steps of a machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the integration of physicochemical features.

[0017] Figure 2 Screening process and results of key alloying factors.

[0018] Figure 3 Algorithm selection and machine learning model prediction.

[0019] Figure 4 SHAP analysis results.

[0020] Figure 5 Multi-objective optimization process.

[0021] Figure 6 The stretching curve corresponding to number 3 in the Pareto solution set.

[0022] Figure 7 The stretching curve corresponding to index 22 in the Pareto solution set.

[0023] Figure 8 The stretching curve corresponding to index 26 in the Pareto solution set. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the present invention will be further described below in conjunction with embodiments and results analysis; obviously, the described embodiments are only some embodiments of the present invention, not all embodiments, and cannot be used as a limitation on the scope of protection of the present invention. The encoding software and corresponding plug-ins used in the embodiments of the present invention are all obtained through conventional means.

[0025] This embodiment utilizes an interpretable machine learning method that integrates physicochemical features to guide the composition design and post-heat treatment process optimization of laser-formed ultra-high strength and toughness steel. It clarifies the key factors (including alloying elements and heat treatment process parameters) that have a combined effect on strength and toughness, and realizes the development of low-cost, high-performance laser-formed ultra-high strength and toughness steel. This provides technical support for the control of the microstructure and mechanical properties of laser additive manufacturing of ultra-high strength and toughness steel, as well as the preparation and repair of complex ultra-high strength and toughness steel structural components.

[0026] See Figure 1 As shown in Example 1, the machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the fusion of physicochemical features is implemented through the following steps: Step S1: Construct the alloy factor corresponding to the composition of ultra-high strength steel based on physicochemical characteristics; In a preferred embodiment, the physicochemical characteristics in step S1 include 81 types, which can be divided into six categories: atomic number, number of groups, Mendeleev number, electrochemical properties, and atomic size parameters; The composition of ultra-high strength steel includes the composition corresponding to ultra-high strength steel prepared by laser directed energy deposition and laser selective melting technology obtained through literature collection.

[0027] In step S1, the alloying factors corresponding to the composition of ultra-high strength steel are constructed based on physicochemical characteristics, including the mean alloying factor. f mi and variance alloying factor f vi The results are obtained by formulas (1) and (2) respectively: (1) (2) in, f ij This represents a characteristic quantity associated with a single element. i Different characteristic quantities representing each element ( i =1, 2,…, 81). j Corresponding to the different elements present in ultra-high strength steel. j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The percentage content of atoms.

[0028] Step S2: Screen out the key alloying factors that affect UTS, YS and EL from the alloying factors; Specifically, step S2 includes the following steps: Step S21: Use the XGBoost algorithm to quantify and rank the contribution weights of the alloy factors; Step S22: Calculate any two alloying factors using formula (3). x and y Pearson correlation coefficient r : r=\frac {\sum{[({x}_{i}-{x}_{m})\times ({y}_{i}-{y}_{m})]}} {\sqrt {{({x}_{i}-{x}_{m})}^{2}\times \sqrt {{({y}_{i}-{y}_{m})}^{2}}}} (3) in,x i and y i Indicates the first i The values ​​of the two alloying factors corresponding to each sample. x m and y m These are the means of the two alloy factors in the attribute dataset, respectively; when | r When | > 0.95, it proves the alloy factor x and y There is a strong linear correlation between them, indicating that they provide redundant information in revealing the material property mechanism; in order to mitigate the effects of multicollinearity, factors with lower importance rankings are systematically eliminated, and the iterative elimination process continues until the top 10 alloy factors that have the most significant impact on performance are finally determined. Step S23: From the top 10 most significant alloy factors, select the optimal combination of alloy factors that achieves the highest performance prediction accuracy and define it as the key alloy factor; during the modeling process, randomly divide the data in the dataset into training and test sets in a 4:1 ratio; construct an exhaustive calculation model and use the ten-fold cross-validation method to optimize the model parameters of the established UTS, YS, and EL respectively; the model accuracy is evaluated by MAPE; Step S24: Based on the MAPE value of the exhaustive calculation model, determine the key alloying factors that affect UTS, YS, and EL respectively.

[0029] Step S3: Construct a dataset consisting of key alloying factors, heat treatment process parameters, and tensile mechanical properties; and use the dataset for training and testing machine learning models to predict UTS, YS, and EL. Specifically, the heat treatment process parameters and tensile mechanical properties in step S3 are obtained from literature collection regarding the heat treatment process parameters and tensile mechanical properties corresponding to ultra-high strength steel prepared by laser directional energy deposition and laser selective melting technology. The heat treatment process parameters include solution temperature, solution time, tempering temperature, and tempering time, and the tensile mechanical properties include UTS, YS, and EL.

[0030] In step S3, the dataset consisting of key alloy factors, heat treatment process parameters and tensile mechanical properties is used to train and test machine learning models of UTS, YS and EL. For example, the data in the dataset is randomly divided into training set and test set in a 4:1 ratio, or the dataset is randomly divided into training set and test set in a 7:3 ratio.

[0031] Specifically, in step S3, the machine learning model compares the fitting results of six algorithms, including XGBoost, GPR, DT, KNN, and SVR, and uses the MAE corresponding to the model trained by the ten-fold cross-validation method as the evaluation standard. The XGBoost algorithm with the lowest MAE value is used for model training and testing.

[0032] Step S4: Use the Optuna visualization tool to optimize hyperparameters, employ a double early stopping strategy, and set the number of iterations to 500 to finally obtain the optimized machine learning models of UTS, YS, and EL. Specifically, in step S4, the Optuna visualization tool is used to optimize the hyperparameters of the XGBoost-fitted model. The corresponding hyperparameter optimization ranges are: n_estimators: (50, 500), max_depth: (2, 12), learning_rate: (0.01, 0.3), subsample: (0.7, 0.8), colsample_bytree: (0.7, 0.8), gamma: (0, 5), min_child_weight: (1, 6), reg_alpha: (0, 5), reg_lambda: (0.5, 5). The model's determination coefficient R is returned through an embedded loop structure. 2 As an evaluation metric, a dual early stopping strategy is adopted to improve computational efficiency, and the number of iterations is set to 500 to ensure a refined search for hyperparameters.

[0033] Step S5: Using SHAP analysis, analyze the optimized machine learning models of UTS, YS, and EL to determine the importance ranking of alloying elements and heat treatment processes; Specifically, using the SHAP analysis method, the relationship between the input and output of the machine learning model based on the XGBoost algorithm is quantitatively analyzed; through weight analysis, it is determined that the alloying elements that can synergistically optimize tensile strength and elongation are Si, Cu, C, Al, Mn, and Cr, and the heat treatment process is a one-step tempering heat treatment process.

[0034] Step S6: Combining the SHAP analysis results with knowledge in the field of materials science, set the range of alloying elements and the range of heat treatment process parameters; Specifically, the alloy element ranges (wt.%) are as follows: C 0.05~0.30%, Cr 12.0~16.0%, Ni 1.0~4.0%, Si 0.3~0.8%, Mn 0.5~1.5%, Al 0.2~0.8%, Cu 0.3~1.2%; the heat treatment process parameters are as follows: tempering temperature 400~600 °C, tempering time 2~12h.

[0035] Step S7: Based on the multi-objective optimization Pymoo framework, the solution is obtained under the guidance of the Pareto optimality principle. By setting the corresponding optimization parameters, the algorithm efficiently searches for non-dominated solution sets in the objective space to obtain Pareto solutions.

[0036] This invention guides the composition design and process optimization of laser-formed ultra-high strength and toughness steel by integrating interpretable machine learning methods based on physicochemical features. It quantifies the influence mechanism of alloying elements and process parameters in synergistically improving strength and toughness, and realizes the development of low-cost, short-process laser-formed ultra-high strength and toughness steel with excellent comprehensive mechanical properties. It provides a new approach to solving the bottleneck problems of existing laser additive manufacturing of ultra-high strength steel, which requires large amounts of expensive elements such as Ni and complex multi-step post-heat treatment processes.

[0037] Specifically, for ease of understanding, see Specific Implementation 2 below, which introduces a machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the fusion of physicochemical features.

[0038] Step S1: Construct the alloying factors corresponding to the composition of ultra-high strength steel based on physicochemical characteristics; The physicochemical characteristics of step S1 include 81 types, which can be divided into six categories: atomic number, number of groups, Mendeleev number, electrochemical properties, and atomic size parameters; the composition of ultra-high strength steel includes the composition corresponding to ultra-high strength steel prepared by laser directed energy deposition and laser selective melting technology obtained through literature collection.

[0039] In step S1, the alloying factors corresponding to the composition of ultra-high strength steel are constructed based on physicochemical characteristics, including the mean alloying factor. f mi and variance alloying factor f vi The results are obtained by formulas (1) and (2) respectively: (1) (2) in, f ij This represents a characteristic quantity associated with a single element. i Different characteristic quantities representing each element ( i =1, 2,…, 81).j Corresponding to the different elements present in ultra-high strength steel. j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The percentage content of atoms.

[0040] Step S2: Screening out key alloying factors affecting UTS, YS, and EL from the alloying factors; specifically, Figure 2 The screening process and results for key alloying factors are presented. The top 15 alloying factors, ranked by importance, are as follows: Figure 2 As shown in (a)-(f), A8-1 and S21-1 exhibit a strong linear correlation in constructing the UTS prediction model, while A5-1, C3-1, and S22-1 also show a similar strong linear correlation. For the constructed YS prediction model, C16-2 and A8-1 show a significant linear correlation, while S22-1, C16-2, E11-1, E14-1, and A5-1 constitute another group of alloy factors with strong linear correlations. For the constructed EL prediction model, C16-2 and A8-1 show a strong linear correlation, while E9-1, A8-1, and C11-1 form another group of strongly linearly correlated alloy factors. By eliminating the lower-ranked alloy factors in the strongly correlated groups, the top 10 alloy factors were finally selected for exhaustive calculation and analysis, as detailed in Table 1.

[0041] Table 1. Ten alloying factors selected for exhaustive calculations

[0042] Figure 2 (g)-(i) show the results of exhaustive screening, indicating that the error level can be minimized by combining 2 to 5 alloying factors. With 3 alloying factors, the combination of S17-2, C12-1, and C14-1 minimizes the MAPE value of the UTS prediction model to 7.6%. The YS prediction model, by using E11-1 and S17-2 as key alloying factors, achieves a MAPE value of 9.0%. The EL prediction model, by introducing E9-1 and C9-1 as key alloying factors, achieves the best prediction performance, with a minimum MAPE value of 20.9%. This invention identifies {S17-2, C12-1, C14-1}, {E11-1, S17-2}, and {E9-1, C9-1} as the key alloying factors affecting the UTS, YS, and EL of the material, respectively. Table 2 lists these key alloying factors and their corresponding physicochemical characteristics.

[0043] Table 2 Key alloying factors obtained through exhaustive calculations

[0044] Specifically, the key alloying factor S17-2 in step S2 is calculated using formulas (1) and (2): (1) (2) in, f ij This represents a characteristic quantity associated with a single element. i The S17 feature values ​​corresponding to each element are shown in Table 3 (see the row containing S17 for details). j These correspond to the different elements present in ultra-high strength steel (specifically C, Cr, Ni, Si, Mn, Co, Mo, Al, Nb, V, Cu, Ti, and Fe). j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The atomic percentage content. At this point, the atomic percentage content calculated according to formula (1) is... f mi This refers to the key alloying factor S17-1, which will be calculated... f mi Substituting into formula (2) to calculate f vi This is the key alloying factor S17-2.

[0045] Specifically, the key alloying factor C14-1 in step S2 is calculated using formula (1): (1) in, f ij This represents a characteristic quantity associated with a single element. i The C14 feature values ​​corresponding to each element are shown in Table 3 (see the row containing C14 for details). j These correspond to the different elements present in ultra-high strength steel (specifically C, Cr, Ni, Si, Mn, Co, Mo, Al, Nb, V, Cu, Ti, and Fe). j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The atomic percentage content. At this point, the atomic percentage content calculated according to formula (1) is... f mi This is the key alloying factor C14-1.

[0046] Table 3. Values ​​of 81 physicochemical characteristic constants for different elements.

[0047] Step S3: Construct a dataset consisting of key alloying factors, heat treatment process parameters, and tensile mechanical properties; and use the dataset for training and testing machine learning models to predict UTS, YS, and EL. Specifically, the data distribution in the dataset is shown in Table 4.

[0048] Table 4 Data distribution of the dataset

[0049] Specifically, Figure 3 Tables (a)-(c) present the preliminary results of algorithm selection. By comparing the fitting results of six algorithms, including XGBoost, FRF, GPR, DT, KNN, and SVR, the MAE value obtained from the ten-fold cross-validation of the corresponding models was used as the evaluation criterion (see Table 5). The results show that the XGBoost algorithm exhibits the lowest MAE values ​​for all three mechanical performance indicators in the ten-fold cross-validation, specifically: UTS 125.0 MPa, YS 138.0 MPa, and EL 2.5%. These results indicate that the XGBoost algorithm has the highest prediction accuracy in the evaluation models, and therefore it was selected to build machine learning models for predicting UTS, YS, and EL, respectively.

[0050] Table 5. Mean Absolute Error (MAE) of Machine Learning Models for Six Algorithms

[0051] Step S4: Use the Optuna visualization tool to optimize hyperparameters, employ a double early stopping strategy, and set the number of iterations to 500 to finally obtain the optimized machine learning models of UTS, YS, and EL. Specifically, the Optuna visualization tool was used for hyperparameter optimization. The corresponding hyperparameter ranges were: n_estimators: (50, 500), max_depth: (2, 12), learning_rate: (0.01, 0.3), subsample: (0.7, 0.8), colsample_bytree: (0.7, 0.8), gamma: (0, 5), min_child_weight: (1, 6), reg_alpha: (0, 5), reg_lambda: (0.5, 5). An embedded loop structure was used to return the R-values ​​of the model. 2 As evaluation metrics, a dual early stopping strategy was used, with 500 iterations. The final UTS, YS, and EL model training results are as follows: Figure 3 As shown in (d)-(f), it can be seen that the R of the UTS model training set... 2 The value is 0.914, which is the R-value of the YS model training set. 2 The value is 0.939, the R-value of the EL model training set. 2 The value is 0.886. The R-value for each model's training set... 2 The value of its test set R 2 The values ​​are basically consistent, with differences of less than 0.05, indicating that the constructed UTS, YS, and EL models all exhibit good prediction accuracy and generalization ability.

[0052] Step S5: Using SHAP analysis, analyze the optimized machine learning models of UTS, YS, and EL to determine the importance ranking of alloying elements and heat treatment process parameters; Specifically, the SHAP method is used to quantify and analyze the relationship between the input and output of a machine learning model based on the XGBoost algorithm. The results are as follows: Figure 4 As shown in (a)-(c), each point in each numerical graph represents an independent sample, and the color corresponds to the SHAP value of the key alloying factor. Positive values ​​indicate that the element has a positive impact on material properties, while negative values ​​indicate a negative impact. When the SHAP value approaches zero, it indicates that the element has a small impact on material properties.

[0053] Based on the weighted calculation method, the influence coefficient of the key alloy factor is calculated according to the slope of the best-fit line between the key alloy factor value and its corresponding SHAP value. K ni And the element ranking weight coefficients associated with each key alloying factor. Z ni The results are shown in Table 6. Figure 4(a)-(c) visually demonstrate the clear correlation between the model inputs (alloy factor and heat treatment process parameters) and the corresponding tensile mechanical properties, where Figure 4 (a1)-(c2) respectively rank the elements corresponding to the key alloying factors of UTS, YS and EL by importance, so as to systematically evaluate the influence of each element on tensile mechanical properties. Figure 4 (d) shows the ranking of the importance of elements affecting the overall tensile mechanical properties of the alloy. The lower the ranking value, the greater the impact on the overall properties.

[0054] Table 6 Influence coefficients of key alloying factors K ni and element sorting weight coefficient Z ni

[0055] Step S6: Combining the SHAP analysis results with knowledge in the field of materials science, set the range of alloying elements and the range of heat treatment process parameters; Considering the high prices of Ti and Nb, combined with Figure 4 Based on the ranking of element importance affecting the overall mechanical properties of the alloy in (d), Si, Cu, and Al were selected as candidate alloying elements. Furthermore, Mn is commonly found in carbon steel, and the Cr content was set above 12 wt.% to ensure corrosion resistance. Meanwhile, Ni was introduced as a candidate element to balance the phase fraction; the specific element selection range is shown in Table 7. Figure 4 (e) shows the comprehensive normalized SAP values ​​for heat treatment, indicating that tempering temperature and tempering time are the main heat treatment factors affecting the comprehensive mechanical properties. Therefore, one-step tempering was selected as the post-heat treatment process, and the specific parameter ranges are detailed in Table 7.

[0056] Table 7. Range of input variables (elemental wt.%) and heat treatment process parameters based on SHAP analysis

[0057] Step S7: Based on the multi-objective optimization Pymoo framework, the solution is obtained under the guidance of the Pareto optimality principle. By setting the corresponding optimization parameters, the algorithm efficiently searches for non-dominated solution sets in the objective space to obtain Pareto solutions.

[0058] Specifically, see Figure 5 Step S7 includes: Step S71: Construct a multi-objective optimization problem. Based on the Pareto optimality principle, the material performance optimization problem is constructed as a multi-objective optimization problem. Three key objective functions are determined: maximizing UTS, maximizing YS, and maximizing EL. The elemental composition of Si, Cu, C, Al, Mn, and Cr, as well as process parameters, are selected as decision variables, including the alloy element content and heat treatment process parameters, and the value range of each variable is set. Step S72: Configure NSGA-III algorithm parameters. Set the initial population size to 1000 to ensure sufficient exploration of the design space; set the maximum number of iterations to 500 to ensure sufficient convergence; use a simulated binary crossover operator with a crossover probability of 0.7; use a polynomial mutation operator with a mutation probability of 0.2. Generate uniformly distributed reference points in the 3D target space based on the sampling reference point method to guide the population to evolve uniformly towards the Pareto front; Step S73: Perform optimization. Initiate the multi-objective optimization iterative process: In each generation, evaluate the three key function values ​​of all individuals in the population using the XGBoost machine learning model. Perform non-dominated sorting based on Pareto dominance, dividing the solutions into different frontier levels. Terminate optimization when there is no significant improvement in the Pareto front for 100 consecutive generations or when the maximum number of iterations is reached; Step S74: Extract the Pareto optimal solution set. By identifying all non-dominated solutions, extract the Pareto optimal solution set from the final population to form an approximate Pareto front. Select representative Pareto optimal solutions and record their corresponding material composition, process parameters, and predicted performance.

[0059] Specifically, the multi-objective optimization results obtained, i.e. the corresponding partial Pareto solutions, are shown in Table 8.

[0060] Table 8. Composition (wt.%), heat treatment process parameters, and tensile mechanical properties corresponding to multi-objective optimization.

[0061] Specifically, number 3 was selected for experimental verification, and the corresponding tensile curve is as follows: Figure 6 As shown, the measured tensile mechanical properties of the sample are: UTS 1706 MPa, YS 1512 MPa, and EL 14.7%. A novel Fe-14.92Cr-1.8Ni-1.2Mn-0.59Si-0.4Al-0.31Cu-0.1C ultra-high strength and toughness steel was successfully designed using the method of this invention. After laser forming, a one-step tempering heat treatment at 530℃ for 3 hours exhibited excellent comprehensive mechanical properties.

[0062] Specifically, number 22 was selected for experimental verification, and the corresponding tensile curve is as follows: Figure 7As shown, the measured tensile mechanical properties of the sample are: UTS 1689 MPa, YS 1373 MPa, and EL 18.3%. A novel Fe-15.23Cr-3.2Ni-1.1Mn-0.75Cu-0.61Si-0.41Al-0.18C ultra-high strength and toughness steel was successfully designed using the method of this invention. After laser forming, a one-step tempering heat treatment at 480℃ for 3 hours exhibited excellent comprehensive mechanical properties.

[0063] Specifically, number 26 was selected for experimental verification, and the corresponding tensile curve is as follows: Figure 8 As shown, the measured tensile mechanical properties of the sample are: UTS 1608 MPa, YS 1359 MPa, and EL 19.7%. A novel Fe-12.39Cr-1.6Ni-0.97Mn-0.66Cu-0.34Si-0.3Al-0.23C ultra-high strength and toughness steel was successfully designed using the method of this invention. After laser forming and a one-step tempering heat treatment at 530℃ for 6 hours, it exhibited excellent comprehensive mechanical properties.

[0064] The machine learning-guided laser-formed ultra-high strength and toughness steel design method based on the fusion of physicochemical features in this embodiment successfully obtained the steel composition and corresponding heat treatment process parameters that combine ultra-high strength and high toughness.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been described above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the fusion of physicochemical features, characterized in that, The method is implemented through the following steps: Step S1: Construct the alloying factors corresponding to the composition of ultra-high strength steel based on physicochemical characteristics; Step S2: Screen out the key alloying factors that affect the ultimate tensile strength, yield strength and elongation after fracture from the alloying factors; Step S3: Construct a dataset consisting of key alloying factors, heat treatment process parameters, and tensile mechanical properties; and use the dataset for training and testing machine learning models to predict ultimate tensile strength, yield strength, and elongation after fracture. Step S4: Use the Optuna visualization tool to optimize hyperparameters, employ a double early stopping strategy, and set the number of iterations to 500. Finally, obtain the optimized machine learning models for ultimate tensile strength, yield strength, and elongation after fracture. Step S5: Using the Shapley Additive exPlanation (SHAP) analysis method, analyze the optimized machine learning model of ultimate tensile strength, yield strength and elongation after fracture, so as to determine the importance ranking of alloying elements and heat treatment process parameters. Step S6: Combining the SHAP analysis results with knowledge in the field of materials science, set the range of alloying elements and the range of heat treatment process parameters; Step S7: Based on the multi-objective optimization framework, the solution is obtained under the guidance of the Pareto optimality principle; by setting the corresponding optimization parameters, the algorithm efficiently searches for the non-dominated solution set in the objective space to obtain the Pareto solution.

2. The method according to claim 1, characterized in that, The physicochemical characteristics in step S1 include 81 types, which can be divided into six categories: atomic number, number of groups, Mendeleev number, electrochemical properties, and atomic size parameters; the composition of ultra-high strength steel includes the composition corresponding to ultra-high strength steel prepared by laser directed energy deposition and laser selective melting technology obtained through literature collection.

3. The method according to claim 1, characterized in that, In step S1, the alloy factors corresponding to the composition of ultra-high strength steel, constructed based on physicochemical characteristics, include mean alloy factor and variance alloy factor, which are calculated using formulas (1) and (2), respectively: (1) (2) in, f ij This represents a characteristic quantity associated with a single element. i These represent the different characteristic quantities of each element. i =1, 2, …, 81 j Corresponding to the different elements present in ultra-high strength steel. j =1, 2, …, n , n The number of constituent elements of ultra-high strength steel; α j Represents element j The percentage content of atoms.

4. The method according to claim 1, characterized in that, Step S2 includes: Step S21: Use the extreme gradient boosting algorithm to quantify and rank the contribution weights of the alloy factor; Step S22: Calculate any two alloying factors using formula (3). x and y Pearson correlation coefficient r : r=\frac {\sum{[({x}_{i}-{x}_{m})\times ({y}_{i}-{y}_{m})]}} {\sqrt {{({x}_{i}-{x}_{m})}^{2}\times \sqrt {{({y}_{i}-{y}_{m})}^{2}}}} (3) in, x i and y i Indicates the first i The values ​​of the two alloying factors corresponding to each sample. x m and y m These are the means of the two alloying factors in the dataset, respectively; when |r When | > 0.95, it proves the alloy factor x and y There is a strong linear correlation between them, indicating that they provide redundant information in revealing the material property mechanism; in order to mitigate the effects of multicollinearity, factors with lower importance rankings are systematically eliminated, and the iterative elimination process continues until the top 10 alloy factors that have the most significant impact on performance are finally determined. Step S23: From the top 10 most significant alloy factors, select the optimal combination of alloy factors that achieves the highest performance prediction accuracy and define it as the key alloy factor; during the modeling process, randomly divide the data in the dataset into training and test sets in a 4:1 ratio; construct an exhaustive calculation model and use the ten-fold cross-validation method to optimize the parameters of the established models for ultimate tensile strength, yield strength, and elongation after fracture; the model accuracy is evaluated by the mean absolute percentage error. Step S24: Based on the average absolute percentage error value of the exhaustive calculation model, determine the key alloying factors that affect the ultimate tensile strength, yield strength, and elongation after fracture.

5. The method according to claim 1, characterized in that, The heat treatment process parameters and tensile mechanical properties in step S3 were obtained through literature collection, including the heat treatment process parameters and corresponding tensile mechanical properties of ultra-high strength steel prepared by laser directional energy deposition and laser selective melting technology.

6. The method according to claim 1, characterized in that, In step S3, the machine learning model compares the fitting results of six algorithms, including extreme gradient boosting, random forest regression, Gaussian process regression, decision tree, K-nearest neighbor algorithm and support vector regression. The mean absolute error of the model trained by the ten-fold cross-validation method is used as the evaluation criterion, and the algorithm with the lowest mean absolute error is used for model training and testing.

7. The method according to claim 1, characterized in that, In step S5, the SHAP analysis method is used to quantify the relationship between the input and output of the machine learning model; and the importance ranking of key alloy element composition and heat treatment process parameters is obtained through weight analysis.

Images