Electric furnace smelting method of high-thermal-conductivity die steel

Abstract

The invention provides an electric furnace smelting method for high-thermal-conductivity die steel, which comprises the following steps: in the LF smelting process of a ladle furnace, electromagnetic stirring is carried out on a die steel raw material, the frequency is 6-10Hz, the current is 500-800A, electromagnetic stirring is loaded from 60 minutes of smelting in the LF furnace, and electromagnetic stirring is stopped when the LF furnace is finished. The method has the advantages that the high-thermal-conductivity die steel is smelted through the composite steelmaking technology of electromagnetic stirring and slag alkalinity regulation and control, extremely pure molten steel is obtained, secondary oxidation of the molten steel in the casting process is avoided through argon protection casting, the grade of non-metallic inclusions in the cast die casting steel ingot does not exceed 0.5 grade, and the casting quality is improved. And the heat conductivity is improved by 10-20%. The method is stable in technological process, high in operability and suitable for industrial large-scale production, and a reliable technical path is provided for performance improvement of high-end plastic die steel.

Description

Technical Field

[0001] This invention belongs to the technical field of mold steel in the metallurgical industry, and particularly relates to an electric furnace smelting method for high thermal conductivity mold steel. Background Technology

[0002] The mold steel manufacturing industry has developed a relatively mature smelting and processing system. Among them, the triple electric arc furnace (EAF) smelting process, consisting of an electric arc furnace (EAF), a ladle furnace (LF), and a vacuum degassing furnace (VD), has become the industry mainstream. This process is followed by forging or rolling forming, and then annealing and pre-hardening are performed to obtain the mold blank. Finally, the finished product is prepared through cavity machining, heat treatment, and precision machining. The core advantage of this process lies in the complementary functions of each furnace, achieving full-chain assurance from raw material melting to precise composition control.

[0003] Specifically, the EAF electric arc furnace serves as the starting point for smelting, with its core task being to efficiently melt scrap steel and complete preliminary dephosphorization, laying the compositional foundation for subsequent smelting. The LF ladle furnace undertakes the crucial functions of composition adjustment and deoxidation refining, achieving the required alloy element content by adding alloys. Simultaneously, it optimizes the purity of molten steel through operations such as creating high-basicity slag with CaO, surface diffusion deoxidation with SiFe powder slag, and final deoxidation with Al in the later stages, while reducing the number of non-metallic inclusions by utilizing their flotation characteristics in the molten steel. The VD furnace further reduces the oxygen content in the steel through vacuum degassing, supplementing the removal of non-metallic inclusions, ultimately constructing a complete smelting closed loop of "melting-refining-degassing".

[0004] A search of relevant patents in the fields of mold steel and steelmaking reveals that existing technology research and development mainly revolves around improving production efficiency, enhancing raw material adaptability, and optimizing key parameters, resulting in a number of targeted smelting solutions. For example, patent publication number CN108456763A proposes a production method based on a high proportion of recycled material. The electric arc furnace is equipped with 70-90% recycled material of the same steel grade. After melting and cleaning, 100-500 kg of CaO is added to create high-basicity slag for dephosphorization. Before tapping, the slag is thoroughly removed and aluminum-silicon deoxidizer and CaO are added, achieving efficient utilization of recycled material. CN107488813B targets ZW868 hot work die steel, specifying that after melting and cleaning in the electric arc furnace, slag is removed and auxiliary materials such as ferrosilicon are added. Aluminum deoxidation is performed during tapping, white slag is created in the LF furnace to finely adjust the composition, and the VD furnace maintains the process parameters at a vacuum degree of <67Pa for 18-35 minutes. CN106609314 targets H13 steel, adopting an 85% hot metal addition scheme to control the phosphorus content of the tapped steel to ≤0.003%. The refining stage is divided into adding calcium oxide and alumina in the early stage and adding specific high-silicon pre-melted slag in the later stage, combined with VD treatment at a vacuum degree of less than 100Pa for more than 25 minutes. These patented technologies have all achieved process optimization under specific steel grades or production conditions, but the core technology path has not deviated from the conventional three-stage process framework.

[0005] Currently, the control of non-metallic inclusions in die steel produced using mainstream electric arc furnace (EAF) processes has reached a relatively stable industry level. Data shows that die steel produced using conventional EAF+LF+VD processes generally maintains a non-metallic inclusion grade of 1.0 to 1.5, with inclusion types primarily being Class B (alumina) and Class D (spherical oxides). Specifically, Class B inclusions of grade 1.5 can exceed 184 μm in length, and the number of Class D inclusions of grade 1.5 can exceed nine in a single field of view. This level of control has become a typical characteristic of existing EAF smelting processes.

[0006] Existing technologies and related patents for electric arc furnace steelmaking of die steel generally suffer from a core deficiency—a lack of specific control technologies for non-metallic inclusions. Whether it's a high-return-material production scheme, a dedicated process for specific steel grades, or a method for precise composition control, the level of non-metallic inclusion control is limited to the inherent capabilities of conventional electric arc furnace smelting processes, failing to develop a targeted system for inclusion suppression, modification, or efficient removal. This results in difficulty in overcoming existing bottlenecks in the purity of die steel, with a persistent presence of a certain number and size of Class B and Class D inclusions.

[0007] The presence of non-metallic inclusions significantly negatively impacts the performance of mold steels, especially the crucial thermal conductivity required for plastic mold steels. Alloying elements in steel already reduce thermal conductivity, and residual non-metallic inclusions further hinder the propagation of thermally conductive factors, exacerbating the deterioration of thermal conductivity. For plastic molds, insufficient thermal conductivity directly affects the cooling efficiency of injection molded parts, making it difficult to achieve the goal of shortening the injection cycle time, thus limiting the improvement of mold efficiency. Simultaneously, the presence of inclusions may also have potential adverse effects on the polishing performance, etching quality, and mechanical properties of mold steel, failing to fully meet the stringent requirements of high-end molds for comprehensive material performance.

[0008] In summary, the core deficiency of existing mold steel smelting technology lies in the lack of targeted and systematic control of non-metallic inclusions. Conventional triple processes and related patented technologies have not broken through the industry's normal level of inclusion control, resulting in the significant deterioration of key properties such as thermal conductivity due to residual Class B and Class D inclusions in the steel, which cannot meet the usage requirements of high-end plastic molds and other scenarios. Therefore, developing a smelting method for non-metallic impurity removal is an urgent problem to be solved in this field. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide an electric furnace smelting method for high thermal conductivity die steel.

[0010] To achieve the above objectives, this application adopts the following technical solution: The present invention provides an electric furnace smelting method for high thermal conductivity mold steel, which includes electromagnetic stirring of the mold steel raw material during the LF furnace smelting process, with a frequency of 6~10Hz and a current of 500~800A. The electromagnetic stirring is started 60 minutes after the LF furnace smelting begins and stopped when the LF furnace ends.

[0011] The smelting method also includes vacuum treatment in an air degassing furnace followed by electromagnetic stirring again, with a frequency of 6~10Hz, a current of 300~400A, and an electromagnetic stirring time of 10~15 minutes.

[0012] Furthermore, the complete smelting steps of the present invention are as follows: (1) The mold steel raw material is melted and cleaned in an electric arc furnace (EAF) and then dephosphorized to ≤0.005% before being tapped;

[0013] (2) The molten steel in the electric arc furnace is fed into the ladle furnace LF for refining. At the same time, 4~5 kg / ton of lime is added into the ladle furnace LF. After the molten steel and steel slag have completely entered the ladle furnace LF, 8~10 kg / ton of lime is added to make high basicity steel slag, so that the basicity of the steel slag reaches 4~5. After 60 minutes of smelting in the ladle furnace LF, the ferrosilicon reduces the slag basicity to 1.0~1.5. After 60 minutes of smelting in the ladle furnace, electromagnetic stirring is started.

[0014] (3) Enter the vacuum degassing furnace (VD) for vacuum degassing treatment. The vacuum degree should reach below 25 Pa and be maintained for 20-30 minutes. After the vacuum treatment in the VD furnace is completed, perform electromagnetic stirring again, and then turn on argon blowing and stirring for 5-10 minutes. The argon flow rate is 4-5 m / s. 3 / h, the smelting process ends;

[0015] (4) Enter the casting process and use argon gas to protect the casting mold to cast steel ingots.

[0016] Preferably, it also includes: during the LF smelting process in the ladle furnace, performing multi-dimensional dynamic monitoring of the mold steel raw materials, calculating the amount of lime to be added based on the multi-dimensional dynamic data, and adjusting the pH value of the steel slag in the ladle furnace to the target pH value;

[0017] During the LF smelting process in a ladle furnace, the die steel raw materials are dynamically monitored from multiple dimensions. Based on the multi-dimensional dynamic data, the amount of lime added is calculated to adjust the pH value of the steel slag in the ladle furnace to the target pH value, including:

[0018] Acquire multi-dimensional dynamic data of mold steel raw materials during the LF smelting process in a ladle furnace; the multi-dimensional dynamic data includes real-time pH value of steel slag, target pH value, steel slag characteristic data, lime characteristic data, smelting condition data, and environmental interference data; and align the multi-dimensional dynamic data.

[0019] The aligned multi-dimensional dynamic data is preprocessed, and the dimensions are unified based on the Z-score normalization formula. A coupling matrix of pH value, temperature, stirring intensity and environmental disturbance is constructed, and the fused feature value F is generated by weighted summation.

[0020] A dynamic pH prediction model is constructed based on an improved LSTM. The input is a fused feature value F and historical pH value sequence data, and the output is a pH prediction sequence for the next 10 minutes. A feedback model is constructed based on adaptive deviation correction. The basic deviation is corrected by multiple factors such as temperature, lime activity and stirring intensity to obtain the final effective deviation ΔpH and the deviation change rate. The output of the two models is obtained by dynamic weighted fusion of the two models to obtain the fused pH value sequence.

[0021] Based on the nonlinear mapping relationship between pH value and alkalinity, the difference ΔR between the target alkalinity and the current alkalinity is calculated. Combined with the SiO2 mass in steel slag, the effective CaO content of lime, and the moisture content correction, the theoretical lime requirement is obtained.

[0022] Based on the theoretical demand for lime, the lime is added dynamically in stages. The amount added in each stage is calculated, and the addition speed is controlled by a variable frequency screw feeder. The argon flow rate and stirring power are adjusted synchronously.

[0023] During the dosing process, the dosage is dynamically corrected. When the pH change rate exceeds the threshold, the lime dissolution efficiency is insufficient, or the deviation between the XRF detection result and the pH back-calculation result exceeds the threshold, the dosing is suspended and the dosage is recalculated and adjusted.

[0024] Set the maximum single addition amount and the boundary threshold of the total addition amount per furnace. Based on the steel slag temperature, stirring power and pH electrode status, construct an interlocking protection mechanism. In case of abnormality, trigger a pause in addition or emergency adjustment until the real-time pH value of the steel slag reaches the target pH value and stop the addition of lime.

[0025] Preferably, a dynamic pH prediction model is constructed based on an improved LSTM, taking into input data such as the fused feature value F and historical pH value sequences, and outputting a pH prediction sequence for the next 10 minutes; a feedback model is constructed based on adaptive deviation correction, correcting the basic deviation through multiple factors such as temperature, lime activity, and stirring intensity, to obtain the final effective deviation ΔpH and the deviation change rate, and obtaining the fused pH value sequence by dynamically weighting and fusing the outputs of the two models, including:

[0026] The improved LSTM-based dynamic pH prediction model includes an input layer, a normalization layer, a dual LSTM layer, an attention layer, a fully connected layer, and an output layer. The dual LSTM layer has 32 neurons in the first layer and 64 neurons in the second layer. The attention mechanism assigns a dynamic weight of 0.7 to the pH sequence and a stirring power of 0.3. The model outputs a predicted pH sequence for each minute within the next 10 minutes. and the confidence interval of the prediction error;

[0027] A feedback model is constructed based on adaptive deviation correction, and the final effective deviation ΔPH is calculated through an additive correction mechanism.

[0028] The outputs of the LSTM prediction model and the feedback model are weighted and fused to obtain a fused pH value sequence.

[0029] The weighted fusion formula is as follows: ; The predicted pH value after fusion at time t; Dynamic weights; It is a nonlinear time function; The pH value at time t is predicted by LSTM; This is the current measured pH value; This represents the current correction deviation.

[0030] In summary, by adopting the above technical solutions, this application achieves the following beneficial effects: This invention utilizes a composite steelmaking technology combining electromagnetic stirring and slag basicity control to smelt high thermal conductivity mold steel, obtaining extremely pure molten steel. Argon gas protection during casting prevents secondary oxidation of the molten steel, resulting in non-metallic inclusions in the cast ingots not exceeding grade 0.5, and improving thermal conductivity by 10-20%. In particular, the content of Class B and Class D inclusions is significantly reduced, effectively improving the purity of the molten steel and further enhancing the thermal conductivity and microstructure uniformity of the mold steel. The reasonable parameter matching of electromagnetic stirring in the LF and VD stages promotes the flotation and separation of inclusions, and the dynamic control of the transition from high-basicity slag to low-basicity slag enhances the slag's adsorption capacity for inclusions. This method has a stable process flow, strong operability, and is suitable for large-scale industrial production, providing a reliable technical path for improving the performance of high-end plastic mold steel. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments are described clearly and completely below. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0032] Example 1: After molten steel is removed from the 50-ton electric arc furnace (EAF) and dephosphorized to 0.003%, the molten steel is tapped into a 50-ton LF refining furnace via an eccentric bottom hole. Simultaneously, 200 kg of lime is added to the LF furnace ladle. After tapping from the EAF furnace, another 500 kg of lime is added to the LF furnace ladle to create a high-basicity slag with a basicity of 5. During the later stages of LF furnace smelting, ferrosilicon is added to reduce the slag basicity to 1.5. In the mid-to-late stages of LF furnace smelting (after 60 minutes), the electromagnetic induction coil attached to the ladle is energized for electromagnetic stirring at a frequency of 10 Hz and a current of 500 A. After the LF ladle furnace completes its refining process, the electromagnetic stirring is turned off, and the vacuum degassing process begins, achieving a vacuum of 25 Pa for 30 minutes. After the VD vacuum treatment is completed, the electromagnetic induction coil is energized again for electromagnetic stirring at a frequency of 10 Hz and a current of 300 A for 15 minutes. Then, argon blowing with weak stirring is initiated for 5 minutes at a flow rate of 4 m³ / s. 3 After the smelting process ends, the casting process begins, using argon protection to cast a 2-ton heavy-duty cast steel ingot. The non-metallic inclusion grades of the cast 2-ton ingot are: A (fine, grade 0); A (coarse, grade 0); B (fine, grade 0.5); B (coarse, grade 0.5); C (fine, grade 0); C (coarse, grade 0); D (fine, grade 0.5); D (coarse, grade 0.5), with a 10% improvement in thermal conductivity. Existing preparation methods for ingot non-metallic inclusion grades are: A (fine, grade 0); A (coarse, grade 0); B (fine, grade 1.5); B (coarse, grade 1.5); C (fine, grade 0); C (coarse, grade 0); D (fine, grade 1.5); D (coarse, grade 1.5).

[0033] Example 2: After molten steel is removed from the 50-ton electric arc furnace (EAF) and dephosphorized to 0.005%, the molten steel is tapped and effluent flows from the EAF's eccentric bottom hole into a 50-ton LF refining furnace. Simultaneously, 300 kg of lime is added to the LF furnace ladle. After tapping from the EAF, another 400 kg of lime is added to the LF furnace ladle to create high-basicity slag, achieving a basicity of 4. During the later stages of LF furnace smelting, ferrosilicon is added to reduce the slag basicity to 1.0. In the mid-to-late stages of LF furnace smelting (after 60 minutes), the electromagnetic induction coil attached to the ladle is energized for electromagnetic stirring at a frequency of 6 Hz and a current of 800 A. After the LF ladle furnace completes its refining process, the electromagnetic stirring is turned off, and the vacuum degassing process begins, achieving a vacuum of 21 Pa for 20 minutes. After the VD vacuum treatment is completed, the electromagnetic induction coil is energized again for electromagnetic stirring at a frequency of 6 Hz and a current of 200 A for 10 minutes. Then, argon blowing with weak stirring is initiated for 10 minutes at an argon flow rate of 4.5 m³ / s. 3After the smelting process ends, the casting process begins, using argon gas protection to cast 5-ton heavy-duty cast steel ingots. The non-metallic inclusion grades of the cast 5-ton ingots are: A (fine, grade 0); A (coarse, grade 0); B (fine, grade 0.5); B (coarse, grade 0); C (fine, grade 0); C (coarse, grade 0); D (fine, grade 0.5); D (coarse, grade 0), with a 20% improvement in thermal conductivity. Existing preparation methods for ingot non-metallic inclusion grades are: A (fine, grade 0); A (coarse, grade 0); B (fine, grade 1.5); B (coarse, grade 1.5); C (fine, grade 0); C (coarse, grade 0); D (fine, grade 2.0); D (coarse, grade 1.5).

[0034] Example 3: After the molten steel in the electric arc furnace is cleaned and dephosphorized to 0.004%, it is tapped. The molten steel from the electric arc furnace flows into the LF refining furnace through the eccentric bottom hole. At the same time, 250 kg of lime is added to the ladle of the LF furnace. After the electric arc furnace tapping is completed, another 450 kg of lime is added to the ladle of the LF furnace to create high-basicity slag, with a slag basicity of 4.5. In the later stage of LF furnace smelting, ferrosilicon is added to reduce the slag basicity to 1.2. In the middle and late stage of LF furnace smelting (after 60 minutes), the electromagnetic induction coil attached to the ladle is energized for electromagnetic stirring at a frequency of 8 Hz and a current of 700 A. After the LF ladle furnace completes the refining task, the electromagnetic stirring is turned off and the VD process is initiated for vacuum degassing treatment, with a vacuum degree of 23 Pa and a holding time of 25 minutes. After the VD vacuum treatment is completed, the electromagnetic induction coil is energized again for electromagnetic stirring at a frequency of 8 Hz and a current of 260 A for 12 minutes. Then, argon blowing and weak stirring are started for 8 minutes at an argon flow rate of 5 m / s. 3 After the smelting process ends, the casting process begins, using argon protection to cast 5-ton heavy-duty cast steel ingots. The non-metallic inclusion grades of the cast 3-ton ingots are: A fine 0 grade; A coarse 0 grade; B fine 0.5 grade; B coarse 0 grade; C fine 0 grade; C coarse 0 grade; D fine 0.5 grade; D coarse 0.5 grade, with a 13% improvement in thermal conductivity. Existing preparation methods for ingot non-metallic inclusion grades are: A fine 0 grade; A coarse 0 grade; B fine 1.5 grade; B coarse 1.5 grade; C fine 0 grade; C coarse 0 grade; D fine 2.0 grade; D coarse 1.5 grade.

[0035] It also includes: during the LF smelting process in the ladle furnace, multi-dimensional dynamic monitoring of the mold steel raw materials, calculation of the amount of lime added based on the multi-dimensional dynamic data, and adjustment of the pH value of the steel slag in the ladle furnace to the target pH value;

[0036] During the LF smelting process in a ladle furnace, the die steel raw materials are dynamically monitored from multiple dimensions. Based on the multi-dimensional dynamic data, the amount of lime added is calculated to adjust the pH value of the steel slag in the ladle furnace to the target pH value, including:

[0037] Acquire multi-dimensional dynamic data of mold steel raw materials during the LF smelting process in a ladle furnace; the multi-dimensional dynamic data includes real-time pH value of steel slag, target pH value, steel slag characteristic data, lime characteristic data, smelting condition data, and environmental interference data; and align the multi-dimensional dynamic data.

[0038] The aligned multi-dimensional dynamic data is preprocessed, and the dimensions are unified based on the Z-score normalization formula. A coupling matrix of pH value, temperature, stirring intensity and environmental disturbance is constructed, and the fused feature value F is generated by weighted summation.

[0039] A dynamic pH prediction model is constructed based on an improved LSTM. The input is a fused feature value F and historical pH value sequence data, and the output is a pH prediction sequence for the next 10 minutes. A feedback model is constructed based on adaptive deviation correction. The basic deviation is corrected by multiple factors such as temperature, lime activity and stirring intensity to obtain the final effective deviation ΔpH and the deviation change rate. The output of the two models is obtained by dynamic weighted fusion of the two models to obtain the fused pH value sequence.

[0040] Based on the nonlinear mapping relationship between pH value and alkalinity, the difference ΔR between the target alkalinity and the current alkalinity is calculated. Combined with the SiO2 mass in steel slag, the effective CaO content of lime, and the moisture content correction, the theoretical lime requirement is obtained.

[0041] Based on the theoretical demand for lime, the lime is added dynamically in stages. The amount added in each stage is calculated, and the addition speed is controlled by a variable frequency screw feeder. The argon flow rate and stirring power are adjusted synchronously.

[0042] During the dosing process, the dosage is dynamically corrected. When the pH change rate exceeds the threshold, the lime dissolution efficiency is insufficient, or the deviation between the XRF detection result and the pH back-calculation result exceeds the threshold, the dosing is suspended and the dosage is recalculated and adjusted.

[0043] Set the maximum single addition amount and the boundary threshold of the total addition amount per furnace. Based on the steel slag temperature, stirring power and pH electrode status, construct an interlocking protection mechanism. In case of abnormality, trigger a pause in addition or emergency adjustment until the real-time pH value of the steel slag reaches the target pH value and stop the addition of lime.

[0044] In this embodiment, the steel slag characteristic data includes real-time steel slag quantity, real-time steel slag temperature, and real-time CaO / SiO2 ratio of the steel slag; the lime characteristic data includes real-time lime activity, lime particle size distribution, and lime moisture content; the smelting condition data includes LF furnace stirring power, argon flow rate, and sulfur content of molten steel in the preceding process; the environmental interference data includes furnace pressure and electrode arc intensity; wherein, the real-time CaO / SiO2 ratio of the steel slag is updated by combining the preceding XRF detection with the real-time pH value back-propagation state update; the real-time pH value is measured by an immersed zirconia pH electrode (temperature resistance 1700℃, response time <500ms); the steel slag fluidity is monitored by a piezoelectric vibration sensor (range 0-10mm / s², sampling frequency 10Hz) installed on the side wall of the ladle 20cm from the slag surface.

[0045] In this embodiment, the alignment of multi-dimensional dynamic data includes: using the pH value sampling time as the base timestamp (1Hz sampling), aligning other data to the same time series using linear interpolation, with the interpolation formula as follows: ;in, The data value after interpolation at time t; for The original data value at time t; t is the target time for interpolation; Environmental interference data is used for alignment after being denoised by a second-order Butterworth low-pass filter.

[0046] In this embodiment, the aligned multi-dimensional dynamic data is preprocessed, including:

[0047] The pH data was processed using a sliding window and Kalman filter: the window size was 10 sampling points, and samples were discarded. An outlier of 0.1, where For window mean; Complete implementation of Kalman filtering: State equation: ; This is the state estimate at time k; This is the state transition matrix; This is process noise, and it follows a normal distribution. For process noise covariance; observation equation: ; The observation value at time k; The observation matrix; To observe noise; To observe the noise covariance; update the equation: Outliers were removed from temperature, stirring power, and furnace pressure data using the 3σ criterion (σ is the standard deviation of a 20-second sliding window). Kalman gain; To estimate the covariance a priori; To estimate the covariance in the posterior timescale; (The unit matrix is ​​used). The moisture content of the lime is calibrated in real time by an infrared moisture sensor (accuracy ±0.1%). When the moisture content is >2%, a 150℃ preheating chamber is triggered to dry for 10 minutes.

[0048] In this embodiment, a pH-temperature-stirring intensity-environmental disturbance coupling matrix is ​​constructed, and a fusion feature value F is generated by weighted summation, including: constructing a four-dimensional coupling matrix from normalized data. ; Weighted summation formula: Weight Allocation and Update: Initial Weights Updated based on least squares method after each furnace cycle: constraint Weight update trigger condition: Forced update when the prediction error MAE > 0.08 for 5 consecutive heats; To optimize the total number of time points within the window; The predicted pH value at time t; The pH value is the actual value measured at time t.

[0049] In this embodiment, based on the nonlinear mapping relationship between pH value and alkalinity, the difference ΔR between the target alkalinity and the current alkalinity is calculated. Combined with corrections based on the SiO2 mass in the steel slag, the effective CaO content of the lime, and the moisture content, the theoretical lime requirement is obtained, including:

[0050] pH-Alkalinity-Lime Amount Correlation Model: Alkalinity Mapping Relationship: initial parameters ; For steel slag basicity; target basicity Current alkalinity Alkalinity difference Mass of SiO2 in steel slag ; XRF detection values Inversely derived from pH value Weighted fusion: ; The mass fraction of SiO2 is inferred from the pH value; The initial SiO2 mass (kg); Initial alkalinity; This is the current alkalinity value; ; The current total mass of steel slag (kg); The final SiO2 mass fraction used; The mass fraction of SiO2 detected by XRF; The mass fraction of SiO2 is inferred from the pH value; the effective CaO mass of lime. Theoretical lime content ; (A represents lime activity, and 80 is the standard value); ; This represents the theoretical lime requirement (kg). This is the activity correction factor; This is the moisture content correction factor; The required mass of CaO (kg); This represents the mass fraction of CaO in lime.

[0051] In this embodiment, the phased dynamic dosing includes:

[0052] Phase 1: Rapid Alkali Supplementation Phase (First 2 Minutes), Dosage Where P is the stirring power; This is a correction factor for the stirring power; control method: variable frequency screw feeder (maximum feeding speed 30kg / s, accuracy ±1%), feeding speed 10-20kg / s; synchronously increase argon flow rate by 10% (baseline flow rate 50L / min), stirring power maintained. ±5%; Phase 2: Fine-tuning phase (5 minutes in the middle) Dosage ; For predicting confidence coefficients; ; The half-width of the LSTM prediction error confidence interval; liquidity determination: vibration sensor amplitude. >5mm / If insufficient liquidity is detected after 10 seconds, the injection will be paused and the temperature increased by 10. ,treat <3mm / Recovery; Phase 3: Stable convergence phase (last 3 minutes), dosage ; ; The unit is per second, representing the rate of change of pH value; Stability coefficient; Convergence criterion: pH value is collected every 30 seconds, twice consecutively. Dosing stops at 0.01 / s. Total dosage constraint: ; .

[0053] In this embodiment, the dynamic correction and protection mechanism includes: a sudden deviation response priority rule (executed in sequence): safety priority: when the steel slag temperature > 1650℃ or the pH electrode signal is lost, the addition is suspended and argon gas strong blowing (flow rate 100L / min) is started to cool down; dissolution efficiency priority: lime dissolution efficiency. ; To improve the electrical conductivity of steel slag; extend the current stage time by 20% and reduce the feeding rate to 50%; prioritize pH loss control. ;or When necessary, suspend the addition of feed. Subsequent dosage coefficient × 0.7; Subsequent dosage coefficient × 1.2; Cross-validation: CaO / SiO2 ratio detected by XRF every 2 minutes. The value derived from pH Compare: ; This represents the relative deviation between XRF and pH inferred alkalinity. For XRF detection of alkalinity value; The alkalinity value is derived from the pH value; if ≥0.15, adjust model parameters: ; The corrected parameter 'a'; for The sign function; ; The corrected b parameter; The corrected c parameter; Interlocking protection mechanism: Temperature protection: Stirring power reduced by 20% when T>1620℃, addition suspended when T>1650℃; Electrode protection: When the standard deviation of the pH signal over the 10-second sliding window is >0.15, switch to the backup pH electrode, and the weight of the main electrode reduced to 30%; Emergency mode: When both pH electrodes fail, switch to alkalinity control mode: based on XRF detection. and The amount of lime is calculated directly, and the addition rate is reduced to 5 kg / s.

[0054] The working principle and beneficial effects of the above technical solution are as follows: Through multi-dimensional dynamic monitoring and data processing, the amount of lime added can be accurately calculated, and the pH value of steel slag can be precisely adjusted to the target value, thereby improving the smelting quality of mold steel; by using an improved LSTM to construct a dynamic pH value prediction model and an adaptive deviation correction feedback model, accurate prediction and deviation correction of future pH values ​​can be achieved, providing a scientific basis for lime addition; lime is dynamically added in stages based on theoretical demand, combined with a variable frequency screw feeder to control the addition speed, and argon flow rate and stirring power are adjusted synchronously to adapt to changes in the smelting process; real-time monitoring is conducted during the addition process, and the addition amount is dynamically corrected according to the pH value change rate, lime dissolution efficiency, and test result deviations to ensure the addition effect; a boundary threshold for the addition amount is set, and a chain protection mechanism is constructed to promptly suspend addition or make emergency adjustments in case of abnormalities, ensuring the safety and stability of the smelting process.

[0055] A dynamic pH prediction model is constructed based on an improved LSTM. Inputting fused feature value F and historical pH value sequences, it outputs a predicted pH value sequence for the next 10 minutes. A feedback model is constructed based on adaptive deviation correction, correcting the basic deviation through multiple factors such as temperature, lime activity, and stirring intensity to obtain the final effective deviation ΔpH and the deviation change rate. The outputs of the two models are then fused using dynamic weighting to obtain the fused pH value sequence, including:

[0056] The improved LSTM-based dynamic pH prediction model includes an input layer, a normalization layer, a dual LSTM layer, an attention layer, a fully connected layer, and an output layer. The dual LSTM layer has 32 neurons in the first layer and 64 neurons in the second layer. The attention mechanism assigns a dynamic weight of 0.7 to the pH sequence and a stirring power of 0.3. The model outputs a predicted pH sequence for each minute within the next 10 minutes. and the confidence interval of the prediction error;

[0057] A feedback model is constructed based on adaptive deviation correction, and the final effective deviation ΔPH is calculated through an additive correction mechanism.

[0058] The outputs of the LSTM prediction model and the feedback model are weighted and fused to obtain a fused pH value sequence.

[0059] The weighted fusion formula is as follows: ; The predicted pH value after fusion at time t; Dynamic weights; It is a nonlinear time function; The pH value at time t is predicted by LSTM; This is the current measured pH value; This represents the current correction deviation.

[0060] In this embodiment, the model input and structure are as follows: Input data: fused feature value F (1-dimensional), historical 5-minute pH sequence (300 points × 1Hz), lime activity A, and real-time CaO / SiO2 ratio of steel slag. Environmental disturbance index Network structure: Input layer (304-dimensional) → LayerNorm normalization → LSTM1 (32 units, tanh activation) → Dropout (0.2) → LSTM2 (64 units, tanh activation) → Attention layer: , These are trainable weights; This is the i-th hidden state; To fuse feature values; Attention score for the i-th feature; Fully connected layer (128 units, ReLU) → Output layer (10 units, linear activation); Loss function: ; This is the total loss function; This is the mean square error term; Mean absolute error; To predict the rate of change of pH value; This represents the actual rate of change in pH value.

[0061] In this embodiment, the adaptive deviation correction feedback model includes: basic deviation: Temperature correction: (304 stainless steel calibration value, H13 mold steel is 0.0035); Activity correction: ( Time (80 / A) Upper limit 1.5); Stirring intensity correction: kW; Compensation Mechanism: Low Temperature Compensation: At time C Low stirring compensation: At kW, Deviation change rate: (10-second difference).

[0062] In this embodiment, the dual-model fusion mechanism includes: dynamic weights: Let be the fusion weight at time t; Basic weights; The attenuation coefficient; The absolute value of the rate of change of deviation; when hour, Decrease by 10% every 0.1 seconds, with a lower limit of 0.3; Non-linear time function: ( Unit: minutes ; for Time-based fusion of predicted values, This represents the current correction deviation (dimensionless).

[0063] The working principle and beneficial effects of the above technical solution are as follows: The improved LSTM dynamic pH prediction model has a reasonable structural design. Combined with the attention mechanism, it assigns dynamic weights to different factors and can output an accurate pH prediction sequence and error confidence interval for the next 10 minutes, providing a reliable basis for subsequent operations. The feedback model based on adaptive deviation correction calculates the final effective deviation through a multi-factor additive correction mechanism, which can effectively reduce prediction errors and improve prediction accuracy. The weighted fusion formula is used to dynamically weight and fuse the outputs of the LSTM prediction model and the feedback model, fully combining the advantages of both, so that the fused pH sequence is more in line with the actual situation, and the ability to predict and control the pH changes of steel slag is improved.

[0064] 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 disclosed 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-described technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for electric furnace smelting of high thermal conductivity die steel, characterized in that, The smelting method includes electromagnetic stirring of the mold steel raw material during the LF furnace smelting process, with a frequency of 6~10Hz and a current of 500~800A. The electromagnetic stirring is started 60 minutes after the LF furnace smelting begins and stopped when the LF furnace ends.

2. The method according to claim 1, characterized in that, The smelting method includes vacuum treatment in a vacuum degassing furnace (VD) followed by electromagnetic stirring at a frequency of 6-10 Hz, a current of 300-400 A, and a stirring time of 10-15 minutes.

3. The method according to claim 1 or 2, characterized in that, The method includes the following steps: (1) The mold steel raw material is melted and cleaned in an electric arc furnace (EAF) and then dephosphorized to ≤0.005% before being tapped; (2) The molten steel in the electric arc furnace is fed into the ladle furnace LF for refining. At the same time, 4~5 kg / ton of lime is added into the ladle furnace LF. After the molten steel and steel slag have completely entered the ladle furnace LF from the electric arc furnace, 8~10 kg / ton of lime is added to make high basicity steel slag, so that the basicity of the steel slag reaches 4~5. After smelting in the ladle furnace LF for 60 minutes, the ferrosilicon reduces the slag basicity to 1.0~1.

5. Electromagnetic stirring is carried out after smelting in the ladle furnace for 60 minutes.

4. The method according to claim 3, characterized in that, The process also includes step (3): Vacuum degassing in a vacuum degassing furnace (VD) until the vacuum level reaches below 25 Pa, and the process is maintained for 20-30 minutes. After the vacuum degassing in the VD furnace is completed, electromagnetic stirring is performed again, followed by argon blowing and stirring for 5-10 minutes at an argon flow rate of 4-5 m / s. 3 / h, the smelting process ends; (4) Enter the casting process and use argon gas to protect the casting mold to cast steel ingots.

5. The method according to claim 3, characterized in that, Also includes: During the LF smelting process in the ladle furnace, the mold steel raw materials are dynamically monitored in multiple dimensions. Based on the multi-dimensional dynamic data, the amount of lime added is calculated to adjust the pH value of the steel slag in the ladle furnace to the target pH value. During the LF smelting process in a ladle furnace, the die steel raw materials are dynamically monitored from multiple dimensions. Based on the multi-dimensional dynamic data, the amount of lime added is calculated to adjust the pH value of the steel slag in the ladle furnace to the target pH value, including: Acquire multi-dimensional dynamic data of mold steel raw materials during the LF smelting process in a ladle furnace; the multi-dimensional dynamic data includes real-time pH value of steel slag, target pH value, steel slag characteristic data, lime characteristic data, smelting condition data, and environmental interference data; and align the multi-dimensional dynamic data. The aligned multi-dimensional dynamic data is preprocessed, and the dimensions are unified based on the Z-score normalization formula. A coupling matrix of pH value, temperature, stirring intensity and environmental disturbance is constructed, and the fused feature value F is generated by weighted summation. A dynamic pH prediction model is constructed based on an improved LSTM. The input is a fused feature value F and historical pH value sequence data, and the output is a pH prediction sequence for the next 10 minutes. A feedback model is constructed based on adaptive deviation correction. The basic deviation is corrected by multiple factors such as temperature, lime activity and stirring intensity to obtain the final effective deviation ΔpH and the deviation change rate. The output of the two models is obtained by dynamic weighted fusion of the two models to obtain the fused pH value sequence. Based on the nonlinear mapping relationship between pH value and alkalinity, the difference ΔR between the target alkalinity and the current alkalinity is calculated. Combined with the SiO2 mass in steel slag, the effective CaO content of lime, and the moisture content correction, the theoretical lime requirement is obtained. Based on the theoretical demand for lime, the lime is added dynamically in stages. The amount added in each stage is calculated, and the addition speed is controlled by a variable frequency screw feeder. The argon flow rate and stirring power are adjusted synchronously. During the dosing process, the dosage is dynamically corrected. When the pH change rate exceeds the threshold, the lime dissolution efficiency is insufficient, or the deviation between the XRF detection result and the pH back-calculation result exceeds the threshold, the dosing is suspended and the dosage is recalculated and adjusted. Set the maximum single addition amount and the boundary threshold of the total addition amount per furnace. Based on the steel slag temperature, stirring power and pH electrode status, construct an interlocking protection mechanism. In case of abnormality, trigger a pause in addition or emergency adjustment until the real-time pH value of the steel slag reaches the target pH value and stop the addition of lime.

6. The method according to claim 5, characterized in that, A dynamic pH prediction model is constructed based on an improved LSTM. Inputting fused feature value F and historical pH value sequences, it outputs a predicted pH value sequence for the next 10 minutes. A feedback model is constructed based on adaptive deviation correction, correcting the basic deviation through multiple factors such as temperature, lime activity, and stirring intensity to obtain the final effective deviation ΔpH and the deviation change rate. The outputs of the two models are then fused using dynamic weighting to obtain the fused pH value sequence, including: The improved LSTM-based dynamic pH prediction model includes an input layer, a normalization layer, a dual LSTM layer, an attention layer, a fully connected layer, and an output layer. The dual LSTM layer has 32 neurons in the first layer and 64 neurons in the second layer. The attention mechanism assigns a dynamic weight of 0.7 to the pH sequence and a stirring power of 0.

3. The model outputs a predicted pH sequence for each minute within the next 10 minutes. and the confidence interval of the prediction error; A feedback model is constructed based on adaptive deviation correction, and the final effective deviation ΔPH is calculated through an additive correction mechanism. The outputs of the LSTM prediction model and the feedback model are weighted and fused to obtain a fused pH value sequence. The weighted fusion formula is as follows: ; The predicted pH value after fusion at time t; Dynamic weights; It is a nonlinear time function; The pH value at time t is predicted by LSTM; This is the current measured pH value; This represents the current correction deviation.