Intelligent generation and screening method of electroslag remelting slag system

Abstract

This invention discloses an intelligent generation and screening method for electroslag remelting slag systems, comprising: acquiring target remelting conditions; inputting a pre-trained candidate slag system generation model, which generates an initial set of candidate slag systems under target condition constraints by learning the probability distribution of historical slag system compositions; inputting the initial set into a pre-trained slag system performance prediction model, which outputs predicted thermal property parameters and predicted quality control effects for each candidate slag system by establishing a mapping between slag system composition and thermal properties and quality control effects; screening candidate slag systems that meet the constraints based on a preset set of metallurgical constraints; and sorting the screening results according to a comprehensive scoring function, outputting at least one set of slag system composition ranges and their predicted performance parameters that rank highest. Compared with existing technologies, this invention achieves intelligent generation and rapid screening of electroslag remelting slag systems by coordinating the generation and prediction models and combining metallurgical constraint screening with comprehensive scoring and sorting.

Description

Technical Field

[0001] This invention relates to the fields of metallurgical engineering and intelligent auxiliary design technology, and in particular to an intelligent generation and screening method for electroslag remelting slag systems. Background Technology

[0002] Electroslag remelting (ESR) is a secondary metallurgical process that uses a conductive slag to undergo Joule heating, gradually melting the consumable electrode, allowing the droplets to pass through a slag pool, and then redirecting and solidifying in a water-cooled crystallizer to form a high-quality steel ingot. In ESR, the slag not only acts as a conductive and heating medium but also participates in droplet heat and mass transfer, inclusion removal, droplet isolation from the atmosphere, and solidification microstructure control. Therefore, the composition of the slag and its properties such as liquidus temperature, viscosity, and conductivity have a significant impact on process stability and the final material quality. Current ESR slag system designs typically rely heavily on empirical slag formulation, existing recipes, limited experimental modifications, and gradual scale-up verification. This makes it difficult to achieve rapid adaptation between different target remelting materials, different process windows, and different quality control requirements. It also makes it difficult to simultaneously address multiple objectives such as liquidus temperature, viscosity, conductivity, volatilization loss, cost, inclusion control, cleanliness control, and microstructure uniformity improvement. This problem is particularly prominent for materials like hot work die steels, which are highly sensitive to cleanliness, inclusion levels, and microstructure uniformity.

[0003] A search revealed that application publication number CN117711522A discloses a design method, system, and electronic equipment for a five-element slag system used in electroslag remelting. This method pre-constructs a set of matching slag systems containing a large number of slag system compositions and their physical properties. Using the density, conductivity, optical basicity, calcium ion activity, viscosity, and other physical properties of the slag system to be optimized as screening criteria, it adaptively determines the screening range of each physical property parameter. It iteratively matches and calculates the average value of intermediate slag system compositions in the set of matching slag systems, gradually narrowing the candidate range until an optimal target slag system is obtained. However, this method is still a rule-based and numerical threshold-based interval shrinkage search. Its screening direction depends on the initial slag system to be optimized and the preset physical property parameter calculation formulas. It cannot actively learn the complex mapping relationships in historical data and lacks a multi-model collaborative mechanism for rapid virtual evaluation of candidate slag systems. It is difficult to flexibly adapt to different target remelting materials, different process windows, and multi-objective quality control requirements.

[0004] Therefore, how to achieve intelligent generation and rapid screening of electroslag remelting slag systems in order to shorten the slag system design cycle and reduce trial and error costs is a technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to overcome the problems of existing electroslag remelting slag system design processes, such as reliance on experience, high trial-and-error costs, long design cycles, and difficulty in simultaneously considering multiple objective thermal properties and quality control requirements. This invention proposes an intelligent generation and screening method for electroslag remelting slag systems. This method unifies the modeling of target remelting material categories, target steel chemical composition, process parameters, slag system composition, slag system thermal properties, and remelting product quality results. Through candidate slag system generation models, slag system performance prediction models, metallurgical constraint sets, and comprehensive scoring functions, it achieves rapid generation, screening, and ranking of candidate slag system composition ranges. For target remelting materials with high requirements for cleanliness, inclusion control, and microstructure uniformity improvement, corresponding quality indicators can be embedded as preferred quality-oriented targets in the candidate slag system generation, screening, and ranking process.

[0006] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a method for intelligent generation and screening of electroslag remelting slag systems is provided, comprising: Obtain the target remelting conditions, which include the target remelting material type, target molten steel or electrode composition, target electroslag remelting process conditions, and target metallurgical indicators; The target remelting condition is input into a pre-trained candidate slag system generation model, which generates an initial set of candidate slag systems under the constraint of the target remelting condition by learning the probability distribution of historical slag system composition. The initial set of candidate slag systems is input into a pre-trained slag system performance prediction model. The slag system performance prediction model establishes a mapping relationship between slag system composition, thermophysical parameters, and quality control effect, and outputs the predicted thermophysical parameters and predicted quality control effect for each candidate slag system. Based on a preset set of metallurgical constraints, candidate slag systems whose predicted performance satisfies the metallurgical constraints are selected from the initial set of candidate slag systems. The candidate slag systems are sorted according to the comprehensive scoring function, and the composition range of at least one candidate slag system with the highest ranking and its corresponding prediction performance parameters are output.

[0007] As a preferred technical solution, the method further includes: constructing an electroslag remelting slag system design association database based on historical sample data, wherein the historical sample data includes remelting material category data, molten steel or electrode composition data, electroslag remelting process parameter data, slag system composition data, slag system thermal property data, and remelting product quality result data.

[0008] As a preferred technical solution, the candidate slag system generation model and the slag system performance prediction model are pre-trained based on the electroslag remelting slag system design association database; the candidate slag system generation model adopts a generative model architecture, and the slag system performance prediction model adopts a regression model architecture.

[0009] As a preferred technical solution, the predicted thermophysical parameters include predicted liquidus temperature, predicted viscosity, and predicted electrical conductivity; the predicted quality control effect includes predicted non-metallic inclusion size, predicted non-metallic inclusion quantity, predicted cleanliness level, and predicted tissue uniformity evaluation of the remelted product.

[0010] As a preferred technical solution, the metallurgical constraint set includes: upper and lower limits of mass fraction of each slag component in the candidate slag system, constraint that the sum of mass fraction of each component is 100%, liquidus temperature constraint, viscosity constraint, electrical conductivity constraint, quality control target constraint, upper limit of volatilization loss constraint, upper limit of slag system cost constraint, and ratio constraint of basic slag system components and regulating components.

[0011] As a preferred technical solution, the liquidus temperature constraint is: the predicted liquidus temperature of the candidate slag system is 100℃~200℃ lower than the liquidus temperature of the target electrode.

[0012] As a preferred technical solution, the candidate slag system is based on CaF2-CaO-Al2O3, and further includes one or more regulating components selected from MgO, BaO, SiO2, and rare earth oxides.

[0013] As a preferred technical solution, the search domain of the candidate slag system generation model is limited to a multi-component system of CaF2-CaO-Al2O3-MgO-SiO2, and satisfies: , , , , And the sum of the mass fractions of each component is 100%.

[0014] As a preferred technical solution, the comprehensive scoring function is composed of a weighted sum of liquidus temperature matching item, viscosity matching item, conductivity matching item, quality control effect item, volatilization loss penalty item, and cost penalty item, and each item has an adjustable weight coefficient.

[0015] As a preferred technical solution, the target remelted material is hot work die steel; the hot work die steel is hot work die steel for die casting, hot work die steel for hot forging, hot work die steel for extrusion, or hot work die steel for hot stamping.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention obtains the target remelting conditions, utilizes a pre-trained candidate slag system generation model and slag system performance prediction model, and combines metallurgical constraint screening and comprehensive scoring ranking to achieve automated design from target condition input to candidate slag system output. It does not rely on human experience and repeated trial and error, and realizes intelligent generation and rapid screening of electroslag remelting slag systems.

[0017] 2. The candidate slag system generation model of this invention is responsible for exploring the formulation space, the slag system performance prediction model is responsible for quickly and virtually evaluating the thermophysical parameters and quality control effects of each formulation, and the metallurgical constraint set is responsible for eliminating invalid formulations that do not meet the process feasible domain. The three work together to realize the intelligent process of electroslag remelting slag system from automatic generation and rapid evaluation to compliance screening.

[0018] 3. This invention, through a conditional generation mechanism, makes the generated candidate slag systems more consistent with the requirements of the target material category, steel composition, process parameters and metallurgical indicators, thereby improving the pertinence and effectiveness of the generated formula and avoiding the generation of a large number of ineffective formulas.

[0019] 4. The candidate slag systems output by this invention are presented in the form of composition range rather than a single fixed composition point. This output format fully considers the raw material fluctuations and operational deviations in industrial production, making the recommended slag system formulation more engineering-tolerant and feasible.

[0020] 5. The comprehensive scoring function of this invention can flexibly adjust the scoring orientation according to the quality control focus of the target remelted material by adjusting the weight coefficients of each item, so that the screening results are more in line with the actual application needs. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating the structure of the associated database for the electroslag remelting slag system of this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] Example 1: like Figure 1 As shown, this invention proposes an intelligent generation and screening method for electroslag remelting slag systems, the method comprising: Step S1: Collect historical sample data during the electroslag remelting process and construct an electroslag remelting slag system design association database; the historical sample data includes at least remelting material category data, molten steel or electrode composition data, electroslag remelting process parameter data, slag system composition data, slag system thermal property data, and remelting product quality result data. Step S2: Perform feature encoding on the input data in Step S1, construct the input feature vector, and establish a metallurgical constraint set; the metallurgical constraint set includes at least the upper and lower limits of slag component mass fraction constraints, liquidus temperature constraints, viscosity constraints, electrical conductivity constraints, and quality control target constraints. Step S3: Based on the electroslag remelting slag system, design an association database to train a slag system performance prediction model, so as to establish a mapping relationship between the input feature vector and the thermal properties of the slag system and the quality results of the remelting products; Step S4: Based on the electroslag remelting slag system design association database, train the candidate slag system generation model; after training, generate an initial candidate slag system set according to the target remelting material category, target molten steel or electrode composition, target electroslag remelting process conditions, and target metallurgical indicators input by the user. Step S5: Input the initial candidate slag system set into the slag system performance prediction model to obtain the predicted liquidus temperature, predicted viscosity, predicted conductivity and predicted quality control effect of each candidate slag system, and screen each candidate slag system according to the metallurgical constraint set. Step S6: Sort the selected candidate slag systems according to the comprehensive scoring function, and output at least one set of candidate slag system composition ranges and their corresponding predicted performance parameters that meet the requirements of the target electroslag remelting process; Step S7: Update the electroslag remelting slag system design association database based on laboratory test melting results or industrial production results, and perform incremental training or iterative optimization on the slag system performance prediction model and candidate slag system generation model.

[0024] The following provides a detailed description of each step of the present invention.

[0025] Step S101: Construct an association database for the design of electroslag remelting slag systems.

[0026] This step forms the basis for subsequent model training and intelligent design. The collected data should cover as many target remelting material categories, a wide range of process parameters, and various slag compositions as possible to ensure sufficient diversity and representativeness in the database. Specifically, such as... Figure 2 As shown, the historical sample data includes: Target remelting material category data: hot work die steel, such as hot work die steel for die casting, hot work die steel for hot forging, hot work die steel for extrusion, or hot work die steel for hot stamping; Target molten steel or electrode composition data: mass fraction of major elements in the steel, including C, Si, Mn, Cr, Mo, V, Ni, Al, Ti, etc.; Electroslag remelting process parameter data: including current (I), voltage (U), melting rate (V) m ), slag volume (M) slag ), slag pool depth (h) slag The specific meanings and symbols of the electroslag remelting process parameters, including the type of protective atmosphere (e.g., air, argon, nitrogen) and crystallizer cooling conditions (e.g., water flow rate, water temperature), are shown in Table 1. Table 1 Slag system composition data: including the components of each slag system and their mass fraction. The typical basic system is CaF2-CaO-Al2O3, and may also include adjustment groups such as MgO, BaO, SiO2, and rare earth oxides. The sum of the mass fractions of each component is 100%. Slag thermophysical property data: including liquidus temperature (T L The data can include viscosity (η), electrical conductivity (σ), density (ρ), surface tension (γ), etc., which can be obtained through experimental measurement. Remelted product quality results data include total oxygen content, total sulfur content, non-metallic inclusion size, non-metallic inclusion quantity, inclusion type distribution, cleanliness level, and microstructure uniformity evaluation. For materials such as hot work die steel that are sensitive to cleanliness and microstructure uniformity, non-metallic inclusion size, non-metallic inclusion quantity, cleanliness level, and microstructure uniformity evaluation are preferred as quality control indicators.

[0027] The data is stored in the database in a uniform format, and outliers are cleaned and missing values ​​are processed, such as by filling in the median or deleting them.

[0028] Step S201: Feature encoding and metallurgical constraint set establishment.

[0029] Feature encoding: In order for machine learning models to process various types of input data, feature encoding of the raw data is required, specifically including: Categorical variables, such as remelted material category and protective atmosphere type, use one-hot encoding or embedding. For example, the hot work die steel category can be encoded as a binary vector of length 4, corresponding to four applications: die casting, hot forging, extrusion, and hot stamping. Continuous variables, such as the content of each element in steel composition, process parameters (current, voltage, etc.), slag composition, etc., are standardized by Z-score or normalized by Min-Max to make their mean 0 and standard deviation 1, or scaled to the [0,1] interval to eliminate the influence of dimensions. Feature construction: For process parameters, construct derived features, such as power density = I×U / electrode cross-sectional area, melting rate and current ratio, etc., to enhance the expressive power of the model; Finally, all the encoded feature vectors are concatenated to form an input feature vector X, whose dimension is usually between 50 and 150.

[0030] Establish a metallurgical constraint set: The metallurgical constraint set Ω is a set of criteria established based on metallurgical principles and engineering experience for screening candidate slag systems. This set includes at least the following constraints: Slag system component mass fraction upper and lower limit constraints Ω1: The mass fraction of each component (such as CaF2, CaO, Al2O3, etc.) should be within a reasonable range in engineering. For example, CaF2 is usually 15~70%, CaO is 0~40%, Al2O3 is 20~45%, MgO is 0~5%, SiO2 is 0~10%, etc. The constraint Ω2 that the sum of the mass fractions of all components must be 100% means that the sum of the mass fractions of all components must equal 100%. Liquidus temperature constraint Ω3: The predicted liquidus temperature of the candidate slag system should be 100~200℃ lower than the liquidus temperature of the target electrode to ensure good slag skin formation and ingot surface quality. Viscosity constraint Ω4: The predicted viscosity should be within the target window, for example, 0.5~3 Pa·s at 1700℃; Conductivity constraint Ω5: The predicted conductivity should be within the target window, for example, 1~5 S / cm at 1700℃; Quality control target constraint Ω6: The predicted cleanliness control effect, inclusion control effect, or tissue homogeneity improvement effect should not be lower than the preset threshold or better than the original electrode level; Volatilization loss limit constraint Ω7: The volatilization loss of easily volatile components (such as CaF2) in the slag system shall not exceed the set value; Slag system cost upper limit constraint Ω8: The prices of each component are different, but the total cost of the slag system does not exceed the preset target; The ratio constraint of basic slag system components and regulating components Ω9: For example, the mass fraction ratio of Al2O3 to CaO is controlled at 0.9~1.3, preferably 1.0~1.2; the SiO2 content can be selected as low cleanliness type branch (0~3%) or Si balance regulating type branch (5~10%) according to the steel grade requirements.

[0031] The metallurgical constraint set used in this invention, including constraint numbers, expressions, and their meanings, is shown in Table 2: Table 2 These constraints can be stored in the form of numerical boundaries, inequalities, or logical rules for easy subsequent programmatic calls.

[0032] Step S301: Train the performance prediction model for slag systems.

[0033] The slag performance prediction model Mp is a supervised learning model. Its input is an input feature vector X (containing material type, steel composition, process parameters, and slag composition), and its output is the predicted thermophysical properties (liquidothermal temperature, viscosity, electrical conductivity, etc.) and predicted quality control effects (total oxygen, inclusions, cleanliness level, etc.) of the candidate slag system. The model architecture can be random forest, gradient boosting tree (such as XGBoost, LightGBM), support vector machine, or deep neural network. In this embodiment, due to the limited data volume and the complex nonlinear relationship between the features and the output, random forest or gradient boosting tree models are preferred because they have better robustness and interpretability.

[0034] Training process: The database constructed in step S101 is divided into an 80% training set and a 20% test set. The model is trained using the training set, and hyperparameters such as the number of trees, maximum depth, and learning rate are adjusted through cross-validation. The model's prediction accuracy is evaluated using the test set, such as the root mean square error (RMSE) and coefficient of determination (R²). 2 In this embodiment, LightGBM is used as the prediction model, with 500 trees, a maximum depth of 6, and a learning rate of 0.05. Liquidus temperature, viscosity, conductivity, and cleanliness score are used as output targets. After training, model Mp can quickly and cost-effectively predict the performance and expected quality of new slag formulations.

[0035] Step S401: Train the candidate slag generation model.

[0036] The candidate slag system generation model Mg is used to generate diverse initial candidate slag system compositions that conform to the basic laws of slag systems under the constraints of given target remelting conditions. The model can be a generative model, such as a variational autoencoder (VAE), a diffusion generative model, a genetic optimization model, or a Bayesian optimization model. In this embodiment, a variational autoencoder is preferred, which can learn the low-dimensional latent variable distribution of slag system composition and can controllably generate new composition vectors.

[0037] To prevent the formation of obviously unreasonable slag systems, such as negative compositions or compositions exceeding process boundaries, an engineered composition search space can be pre-defined. For example, for the CaF2-CaO-Al2O3-MgO-SiO2 system, the following limitations apply: , , , , , Furthermore, the sum of the mass fractions of each component must be 100%. This search domain combines the experimental coverage range of publicly available hot work die steel ESR slag and the range of common industrial formulations. The symbols and mass fraction window representations of each slag component in the candidate slag system are shown in Table 3. Table 3 For the preferred embodiment of hot work die steel for die casting, the range of candidate slag system composition is further limited to: , , , , , This preferred window, centered on the nominal composition of the publicly disclosed hot work die steel standard slag 31.5 / 29.5 / 33.5 / 1.5 / 3, is more suitable for the ESR process of die casting die steel.

[0038] Training process: The slag composition data from the database is used as training samples. For a VAE, the encoder maps the composition vector to the latent variable mean and variance, and the decoder reconstructs the composition vector from the latent variables. The loss function includes reconstruction error and KL divergence. In this embodiment, the candidate slag composition generation model uses a VAE with a hidden layer dimension of 8. Both the encoder and decoder are three-layer fully connected networks with ReLU activation. After training, given target conditions (inputting as condition vectors to the model or used to adjust the sampling region of the latent variable space), the model can sample from the latent variable space and generate an initial set of candidate slag compositions via the decoder.

[0039] To facilitate understanding of the meaning of each input parameter, intermediate variable, and output parameter in this invention, the relevant parameters and their symbols are shown in Table 4: Table 4 Step S501: Performance prediction and screening of candidate slag systems.

[0040] Perform the following operations sequentially on the initial candidate slag system set generated in step S401: Performance prediction involves combining the composition vector of each candidate slag system with fixed target remelting material type, steel composition, process parameters, etc., to form a complete input feature vector, which is then fed into the slag system performance prediction model Mp trained in step S301. The model outputs the predicted liquidus temperature corresponding to the candidate slag system. Predicting viscosity Predicted conductivity And predicting the effectiveness of quality control (e.g., predicting total oxygen content or cleanliness score); preferably, it can further obtain predictions of non-metallic inclusion control effects and predictions of tissue homogeneity improvement effects; Metallurgical constraint screening: The predicted values ​​are compared with the metallurgical constraint set Ω established in step S201. A candidate slag system is retained only if all predicted properties of a candidate slag system simultaneously satisfy all constraints in Ω; otherwise, it is eliminated. For example, if the predicted liquidus temperature of a candidate slag system does not meet the condition of being 100-200°C lower than the electrode liquidus temperature, or the predicted viscosity exceeds the target window, or the predicted cleanliness effect is lower than the threshold, then the candidate slag system is eliminated.

[0041] Step S601: Output comprehensive score and ranking.

[0042] To further optimize the selection of the best solution, a comprehensive scoring function is adopted. To perform scoring and ranking, the multi-objective scoring function designed in this embodiment includes the following sub-items: Liquidus temperature matching item This reflects the degree to which the predicted value deviates from the center of the target window; the smaller the value, the better.

[0043] Viscosity matching item This reflects the degree to which the predicted viscosity deviates from the center of the target window; the smaller the value, the better.

[0044] Conductivity matching term This reflects the degree to which the predicted conductivity deviates from the center of the target window; the smaller the value, the better.

[0045] Quality control effectiveness items This reflects the predicted cleanliness and inclusion control effects. For some indicators (such as cleanliness score), a higher score is better, so a negative sign is used in the actual scoring function.

[0046] Volatilization loss penalty This reflects the risk of volatilization loss. It is calculated based on the content of high volatile components in the slag system, and the lower the value, the better.

[0047] Cost penalty items This reflects the cost of the slag system, estimated based on the market prices of each component; the lower the value, the better.

[0048] The comprehensive scoring function can be written as: ;in, to These are weighting coefficients, all of which are non-negative. In the embodiment targeting hot work die steel for die casting, the weighting coefficients are set as follows: =0.3, =0.2, =0.15, =0.25, =0.05, =0.05 to highlight the importance of cleanliness control; the weighting coefficient can be adjusted according to the quality control focus of the specific target remelted material.

[0049] Sorting and Output: Calculate the comprehensive score for each candidate slag system. The candidate slag systems are sorted in ascending order of scores, and the composition range of at least one of the top-ranked candidate slag systems is output. This means that a quality score range is given for each component. For example, for the top N candidate slag systems, the minimum and maximum values ​​of each component are calculated to obtain the recommended composition range. Simultaneously, the predicted performance parameters (predicted liquidus temperature, predicted viscosity, predicted conductivity, predicted cleanliness level, etc.) corresponding to the candidate slag system are output. The recommended format for the candidate slag system output is shown in Table 5. Table 5 Compared to outputting a single, fixed formula, outputting a range of compositions offers greater industrial robustness, adapting to raw material fluctuations and operational deviations during the production process, and facilitating large-scale engineering implementation.

[0050] Step S701: Iterative optimization of the model.

[0051] To continuously improve the prediction accuracy and generation quality of the model, this invention also includes a model update mechanism. The user selects one or more recommended candidate slag systems from the output of step S601, conducts laboratory melting tests or small-scale industrial experiments, measures the thermal properties (liquidotherm temperature, viscosity, electrical conductivity) of the actual slag system, and detects the quality results of the remelted steel ingot (total oxygen content, inclusion size and quantity, cleanliness level, microstructure uniformity, etc.). The new data obtained from the above experimental verification (including input conditions and measured results) is used as a new sample and backfilled into the electroslag remelting slag system design association database constructed in step S101. Periodically or when a certain amount of new data has been accumulated, the incremental training or complete retraining of the slag performance prediction model Mp and the candidate slag generation model is triggered, enabling the model to learn from the latest practices, continuously adapt to new materials, new processes and more stringent quality requirements, and continuously improve prediction accuracy and generation quality.

[0052] This invention trains a candidate slag system generation model and a performance prediction model based on historical data. After the user inputs the target remelting conditions, a candidate slag system is generated. After performance prediction, metallurgical constraint screening and comprehensive scoring and ranking, the composition range of the candidate slag system and its predicted performance are output. This realizes intelligent generation and multi-objective optimization of slag systems, shortens the design cycle and reduces trial and error costs.

[0053] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent generation and screening of electroslag remelting slag systems, characterized in that, include: Obtain the target remelting conditions, which include the target remelting material type, target molten steel or electrode composition, target electroslag remelting process conditions, and target metallurgical indicators; The target remelting condition is input into a pre-trained candidate slag system generation model, which generates an initial set of candidate slag systems under the constraint of the target remelting condition by learning the probability distribution of historical slag system composition. The initial set of candidate slag systems is input into a pre-trained slag system performance prediction model. The slag system performance prediction model establishes a mapping relationship between slag system composition, thermophysical parameters, and quality control effect, and outputs the predicted thermophysical parameters and predicted quality control effect for each candidate slag system. Based on a preset set of metallurgical constraints, candidate slag systems whose predicted performance satisfies the metallurgical constraints are selected from the initial set of candidate slag systems. The candidate slag systems are sorted according to the comprehensive scoring function, and the composition range of at least one candidate slag system with the highest ranking and its corresponding prediction performance parameters are output.

2. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The method further includes: constructing an electroslag remelting slag system design association database based on historical sample data, wherein the historical sample data includes remelting material category data, molten steel or electrode composition data, electroslag remelting process parameter data, slag system composition data, slag system thermal property data, and remelting product quality result data.

3. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The candidate slag system generation model and the slag system performance prediction model are pre-trained based on the electroslag remelting slag system design association database; the candidate slag system generation model adopts a generative model architecture, and the slag system performance prediction model adopts a regression model architecture.

4. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The predicted thermophysical parameters include predicted liquidus temperature, predicted viscosity, and predicted electrical conductivity; the predicted quality control effects include predicted non-metallic inclusion size, predicted non-metallic inclusion quantity, predicted cleanliness level, and predicted tissue uniformity evaluation of the remelted products.

5. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The metallurgical constraint set includes: upper and lower limits of mass fraction of each slag component in the candidate slag system, constraint that the sum of mass fraction of each component is 100%, liquidus temperature constraint, viscosity constraint, electrical conductivity constraint, quality control target constraint, upper limit of volatilization loss constraint, upper limit of slag system cost constraint, and ratio constraint of basic slag system components and regulating components.

6. The intelligent generation and screening method for electroslag remelting slag systems according to claim 5, characterized in that, The liquidus temperature constraint is that the predicted liquidus temperature of the candidate slag system is 100℃ to 200℃ lower than the liquidus temperature of the target electrode.

7. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The candidate slag system is based on CaF2-CaO-Al2O3 and further includes one or more regulating components selected from MgO, BaO, SiO2, and rare earth oxides.

8. The intelligent generation and screening method for electroslag remelting slag system according to claim 7, characterized in that, The search domain of the candidate slag system generation model is limited to a multi-component system of CaF2-CaO-Al2O3-MgO-SiO2, and satisfies: , , , , And the sum of the mass fractions of each component is 100%.

9. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The comprehensive scoring function is composed of a weighted sum of liquidus temperature matching, viscosity matching, conductivity matching, quality control effect, volatilization loss penalty, and cost penalty, and each sub-item has an adjustable weight coefficient.

10. The intelligent generation and screening method for electroslag remelting slag systems according to claim 1, characterized in that, The target remelted material is hot work die steel; the hot work die steel is hot work die steel for die casting, hot work die steel for hot forging, hot work die steel for extrusion, or hot work die steel for hot stamping.

Images