Method for inhibiting temperature fluctuation of oriented silicon steel BAF furnace

Abstract

The invention discloses a method for inhibiting temperature fluctuation of an oriented silicon steel BAF furnace, which comprises the following steps: adding 0.005-0.01 wt% of Nb, and forming a nanoscale Nb (C, N) precipitate; a three-dimensional heat conduction model is established in a mode of linkage of infrared thermal imaging and electromagnetic induction heating in combination with a finite element method, and real-time monitoring and accurate control of a temperature field in the furnace are achieved. On the basis of a fuzzy PID hybrid control strategy, the optimal heating power and scanning path are predicted in combination with a convolutional neural network, and closed-loop control of local temperature compensation is achieved; a random forest model with the temperature gradient delta T, the Als content and the Nb precipitated phase density as input parameters is constructed, technological parameters are iteratively optimized in combination with a genetic algorithm, and dynamic self-adaptive adjustment of an annealing temperature curve is achieved. According to the method, uniform growth of Gaussian grains in the secondary recrystallization process is achieved, and the effects of reducing annealing temperature fluctuation, improving secondary recrystallization uniformity and optimizing grain orientation distribution are achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of oriented silicon steel production, in particular to an oriented silicon steel BAF furnace temperature fluctuation suppression method. BACKGROUND

[0002] As the core soft magnetic material of the power electronic industry, the magnetic performance of oriented silicon steel directly affects the energy efficiency of devices such as transformers and reactors. In the production process of oriented silicon steel, the temperature control of the high-temperature annealing link is particularly critical. Temperature fluctuations can significantly affect the secondary recrystallization behavior, leading to uneven grain orientation distribution and thus reducing the magnetic performance of the product. The traditional production process mainly faces two technical bottlenecks: one is the inherent inhibitor method, which requires high-temperature treatment above 1320℃, resulting in high energy consumption and easy grain coarsening; the other is the acquired inhibitor method, which reduces the processing temperature, but has problems such as complex nitriding process and uneven inhibitor distribution.

[0003] The existing temperature fluctuation suppression technology has obvious limitations: in terms of process control, there is a lack of accurate simulation of the thermodynamic behavior of precipitates; in terms of temperature compensation, it is difficult to realize real-time regulation of local temperature gradient; in terms of annealing curve optimization, an intelligent decision-making model coupled with multiple parameters has not been established. These problems result in the inability of existing methods to simultaneously meet the requirements of low-temperature processing, high-efficiency inhibition, and stable production, severely restricting the performance improvement and industrial application of high-magnetic oriented silicon steel.

[0004] In view of the above problems, the existing technology needs to be improved. SUMMARY

[0005] The purpose of the present application is to provide an oriented silicon steel BAF furnace temperature fluctuation suppression method, which has the effects of reducing annealing temperature fluctuation, improving secondary recrystallization uniformity, and optimizing grain orientation distribution.

[0006] The application provides an oriented silicon steel BAF furnace temperature fluctuation suppression method, and the technical scheme is as follows:

[0007] Step 1, molten steel preparation: prepare molten steel according to the following weight proportions, C: 0.02%-0.08%; Si: 3.0%-3.5%; Mn: 0.02%-0.18%; S: 0.005%-0.020%; Al: 0.020%-0.042%; N: 0.015%-0.040%; Sn: 0.08%-0.3%; Cu: 0.1%-0.9%; the balance is Fe;

[0008] Step 2, hot rolling: heat the billet to 1000-1100℃ and hot roll to a thickness of 1.5-2.5 times the original thickness;

[0009] Step 3, normalizing: normalizing the hot-rolled billet at 850-900℃;

[0010] Step 4, cold rolling: the normalized steel billet is cold rolled under the condition of reduction rate ≥ 20%;

[0011] Step 5, pickling: the cold-rolled steel billet is pickled with 2-4% hydrochloric acid;

[0012] Step 6, high temperature annealing: the pickled steel billet is high temperature annealed at 1150-1200℃;

[0013] Step 7, niobium element regulation: during high temperature annealing, 0.005-0.01wt% Nb is added to form nanoscale Nb(C, N) precipitates, the thermodynamic behavior of precipitated phase is simulated by Thermo-Calc software, the Nb content and precipitate size distribution are optimized by Zener pinning force formula to ensure the preferred growth of Gauss grains in the secondary recrystallization stage;

[0014] Step 8, temperature compensation control: infrared thermal imaging and electromagnetic induction heating are combined to establish a three-dimensional heat conduction model by finite element method, the temperature gradient at the edge is compensated by longitudinal magnetic induction heating device to realize closed-loop control of local temperature compensation;

[0015] Step 9, annealing temperature optimization: based on the fuzzy PID hybrid control strategy, the temperature fluctuation data is input into the convolutional neural network to predict the optimal heating power and scanning path, and the annealing temperature curve is optimized by the multivariate coupled machine learning model.

[0016] Further, the present application also proposes that step 7 comprises:

[0017] Step 701, simulate the thermodynamic behavior of precipitated phase by Thermo-Calc software:

[0018] Step 7011, load the database suitable for steel system in Thermo-Calc, input the steel composition, temperature range and pressure condition;

[0019] Step 7012, simulate the type of precipitated phase, its volume fraction (f) and precipitation temperature window under different Nb content by the equilibrium phase calculation module, simulate the precipitate size distribution by TC-PRISMA module, input the cooling rate and nucleation condition parameters to obtain the dynamic evolution data of precipitate average radius (r) and number density (Nv);

[0020] Step 7013, calculate the precipitated phase volume fraction f = ∑(V precipitated phase / V total) by the mole fraction, phase composition and other data output by Thermo-Calc;

[0021] Step 7014, determine the precipitate chemical stability and its interface energy with the matrix by combining precipitate phase sublattice composition analysis;

[0022] Step 702, optimization of Zener pinning force formula:

[0023] Step 7021, the Zener pinning force formula is: Z=3σf / 4r, where σ is the interface energy between the precipitate phase and the matrix, f is the volume fraction, and r is the average radius;

[0024] Step 7022, input the f, r, and σ parameters output by Thermo-Calc, and construct a multivariate model combined with the Zener formula to predict the optimal Nb content and process conditions;

[0025] Step 7023, use random forest or genetic algorithm to optimize the objective function (maximize Z value) to balance the needs of precipitate strengthening and grain refinement.

[0026] Further, the present application also proposes that in step 8, the temperature compensation control method of infrared thermal imaging and electromagnetic induction heating linkage includes:

[0027] Step 801, use an infrared thermal imager to collect real-time furnace temperature field data with an accuracy of ±2℃;

[0028] Step 802, establish a three-dimensional heat conduction model by finite element method to locate the local temperature difference abnormal area;

[0029] Step 803, use a longitudinal magnetic induction heating device to compensate for the temperature gradient at the edge by staggering the layout of the movable coil;

[0030] Step 804, based on the fuzzy PID hybrid control strategy, adjust the heating power according to the temperature error and change rate;

[0031] Step 805, input the temperature fluctuation data into a convolutional neural network to predict the optimal heating power and scanning path, and realize accurate compensation control of the temperature field.

[0032] Further, the present application also proposes that in step 9, the multivariate coupled machine learning model includes:

[0033] Step 901, establish a random forest model with temperature gradient ΔT, Als content, and Nb precipitate density as input parameters;

[0034] Step 902, iteratively optimize the process parameters by genetic algorithm, and train the model combined with historical production data;

[0035] Step 903, construct an Arrhenius equation correction term to predict the spatiotemporal evolution law of AlN volume fraction (f) and size (r);

[0036] Step 904, optimize the annealing curve through the JMatPro thermodynamic database to ensure the completeness and uniformity of the secondary recrystallization.

[0037] Further, the present application also proposes that step 6 comprises:

[0038] Step 601, heat to 1150-1200℃ and keep for 1-2 hours;

[0039] Step 602, naturally cool to 950-1000℃ and keep for 1-2 hours;

[0040] Step 603, heat to 1050-1100℃ and keep for 1-2 hours;

[0041] Step 604, naturally cool to room temperature.

[0042] Compared with the prior art, the present application provides an oriented silicon steel BAF furnace temperature fluctuation suppression method, which has the following beneficial effects:

[0043] 1. By adding 0.005-0.01wt% of Nb and forming nanoscale Nb(C,N) precipitates, grain boundary pinning strengthening at low temperature (about 1250-1300℃) is achieved, effectively inhibiting abnormal grain growth and improving the uniformity and completeness of secondary recrystallization, so that the Goss texture ratio reaches more than 95%.

[0044] 2. Using the infrared thermal imaging and electromagnetic induction heating linkage method, combined with the finite element method to establish a three-dimensional heat conduction model, the real-time monitoring and accurate control of the temperature field in the furnace are realized, and the transverse temperature fluctuation range is effectively reduced to within ±5℃.

[0045] 3. Based on the fuzzy PID hybrid control strategy, combined with the convolutional neural network to predict the optimal heating power and scanning path, the closed-loop control of local temperature compensation is realized, which significantly improves the accuracy and real-time performance of temperature fluctuation suppression.

[0046] 4. By constructing a random forest model with temperature gradient ΔT, Als content, and Nb precipitate density as input parameters, combined with genetic algorithm iterative optimization of process parameters, the dynamic adaptive adjustment of the annealing temperature curve is realized, effectively reducing the iron loss P17 to below 0.85W / kg and improving the magnetic induction B8 to above 1.88T. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The step flowchart of the oriented silicon steel BAF furnace temperature fluctuation suppression method of the present application;

[0048] Figure 2Flow chart of step 6 of the method for inhibiting temperature fluctuation of an oriented silicon steel BAF furnace according to the present application;

[0049] Figure 3 Flow chart of step 7 of the method for inhibiting temperature fluctuation of an oriented silicon steel BAF furnace according to the present application;

[0050] Figure 4 Flow chart of step 8 of the method for inhibiting temperature fluctuation of an oriented silicon steel BAF furnace according to the present application;

[0051] Figure 5 Flow chart of step 9 of the method for inhibiting temperature fluctuation of an oriented silicon steel BAF furnace according to the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. The components of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0053] It should be noted that similar reference numerals and letters represent similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0054] In the conventional existing high-temperature annealing process of oriented silicon steel, the uneven temperature field distribution and the insufficient stability of the inhibitor lead to fluctuation of magnetic properties. The traditional intrinsic inhibitor method relies on high-temperature solid solution to achieve pinning effect, but the excessively high annealing temperature easily causes grain coarsening and energy consumption rise, while the acquired inhibitor method of nitriding treatment has the problem of uneven distribution of precipitated phase. The insufficient accuracy of the heat conduction model and the lack of real-time control means lead to significant temperature gradient between the edge and the center during the annealing process, the grain growth orientation deviates from the preset Gaussian texture in the secondary recrystallization stage, and finally causes the increase of iron loss and the deterioration of magnetic induction.

[0055] For example, in a continuous annealing production line, the temperature field in the BAF furnace has a significant temperature drop in the edge region due to differences in radiation heat transfer efficiency and structural limitations of the furnace body. When the hot-rolled slab enters the annealing furnace, the temperature difference between the edge and center regions causes an imbalance in the dissolution-precipitation kinetics of nanoscale precipitates. The volume fraction of precipitates in the local region decreases, the Zener pinning force is insufficient, and the grain boundary migration rate increases, resulting in abnormal grain growth. At the same time, traditional PID control strategies are difficult to cope with nonlinear temperature fluctuations, and there is a time delay between thermal imaging data and heating power adjustment, which cannot real-time correct the transverse temperature gradient, ultimately leading to uneven magnetic domain structure on the finished plate surface.

[0056] If the above problems are not solved, the orientation selectivity and uniformity of secondary recrystallization of grains during annealing will continue to deteriorate, and the finished product magnetic induction and iron loss indicators cannot meet the design requirements of high energy efficiency transformers. Local stress concentration caused by abnormal temperature gradient will also exacerbate plate shape warping defects, reducing yield and product consistency. In addition, the traditional process has insufficient prediction accuracy for the thermodynamic behavior of inhibitors, resulting in production parameter adjustments relying on trial and error, increasing research and development cycle and quality control costs, and severely restricting the large-scale industrial application of high magnetic induction oriented silicon steel.

[0057] In the face of the above problems, the present application first studies the uniformity of the annealing temperature field and the stability of the inhibitor distribution. The high-temperature solid solution requirement of the traditional inhibitor system and the uniformity defect of the nitriding process indicate the need to develop a new type of low-temperature and high-efficiency inhibitor. In this regard, the present application considers introducing transition metal elements as the pinning phase core and optimizing the precipitation behavior through thermodynamic simulation. At the same time, to address the edge temperature gradient problem, a dynamic compensation mechanism is explored to eliminate the transverse heat conduction difference. In addition, the hysteresis of the traditional temperature control strategy prompts the present application to introduce real-time feedback and prediction algorithms to improve the regulation accuracy.

[0058] In this regard, as shown in Figures 1-5 The present application proposes a method for suppressing temperature fluctuations in an oriented silicon steel BAF furnace, comprising:

[0059] Step 1, molten steel preparation: prepare molten steel according to the following weight proportions, C: 0.02%-0.08%; Si: 3.0%-3.5%; Mn: 0.02%-0.18%; S: 0.005%-0.020%; Al: 0.020%-0.042%; N: 0.015%-0.040%; Sn: 0.08%-0.3%; Cu: 0.1%-0.9%; the balance is Fe;

[0060] Step 2, hot rolling: heat the billet to 1000-1100℃ and hot roll to a thickness of 1.5-2.5 times the original thickness;

[0061] Step 3, normalizing: normalizing the hot-rolled billet at 850-900℃.

[0062] Step 4, cold rolling: the normalized steel billet is cold rolled under the condition of Reduction rate ≥ 20%;

[0063] Step 5, pickling: the cold-rolled steel billet is pickled with 2-4% hydrochloric acid;

[0064] Step 6, high temperature annealing: the pickled steel billet is high temperature annealed at 1150-1200℃;

[0065] Step 7, niobium element regulation: 0.005-0.01wt% Nb is added during high temperature annealing to form nanoscale Nb(C, N) precipitates. Thermo-Calc software is used to simulate the thermodynamic behavior of precipitates, and Zener pinning force formula is used to optimize Nb content and precipitate size distribution to ensure the preferred growth of Gaussian grains during secondary recrystallization stage.

[0066] Step 8, temperature compensation control: infrared thermal imaging and electromagnetic induction heating are used in combination with finite element method to establish a three-dimensional heat conduction model. The temperature gradient at the edge is compensated by longitudinal magnetic induction heating device to realize closed-loop control of local temperature compensation.

[0067] Step 9, annealing temperature optimization: based on the fuzzy PID hybrid control strategy, the temperature fluctuation data is input into the convolutional neural network to predict the optimal heating power and scanning path. The multivariate coupled machine learning model is used to optimize the annealing temperature curve.

[0068] Wherein, hot rolling to a thickness of 1.5-2.5 times the original thickness refers to the process of increasing the density and orientation of the material through plastic deformation. Specifically, it can be achieved by multi-pass rolling combined with temperature gradient control. This deformation range can balance the work hardening and recrystallization driving force to provide a reasonable initial organization for subsequent cold rolling.

[0069] Wherein, normalizing treatment at 850-900℃ refers to the heat treatment process of grain refinement by air cooling after austenitizing. Specifically, it can be achieved by using a segmented temperature control roller hearth normalizing furnace. This temperature range can effectively eliminate the hot rolling banded structure and improve the cold rolling workability.

[0070] Wherein, cold rolling under the condition of Reduction rate ≥ 20% refers to the process of introducing dislocation energy storage through large deformation rolling. Specifically, it can be achieved by using a four-high reversible mill combined with an online shape meter. This deformation threshold ensures sufficient energy storage to drive secondary recrystallization.

[0071] The nanoscale Nb(C, N) precipitates refer to composite precipitates with a size in the range of 10-50 nm, and can be realized by in-situ precipitation combined with rapid cooling process. The precipitates inhibit grain boundary migration through Zener pinning effect and control the secondary recrystallization process.

[0072] The Thermo-Calc software simulates the thermodynamic behavior of the precipitates, which refers to the process of calculating the phase equilibrium state based on the CALPHAD method. Specifically, the TCFE9 database can be used to load the composition parameters of the steel to realize the simulation. This simulation can predict the evolution law of precipitate type and volume fraction under different Nb contents.

[0073] The infrared thermal imaging and electromagnetic induction heating linkage refers to the cooperative control technology of optical temperature measurement and electromagnetic energy conversion. Specifically, a combination device of a scanning focal plane detector and a longitudinal magnetic yoke inductor can be used to realize this linkage mechanism, which realizes real-time monitoring and dynamic compensation of the temperature field.

[0074] The three-dimensional heat conduction model is established based on the finite element method, which refers to the method of using partial differential equation numerical solution to calculate the temperature distribution. Specifically, ANSYS Parametric Design Language can be used to write a multi-physics field coupling script to realize this model, which can accurately locate the abnormal temperature gradient area in the furnace.

[0075] The fuzzy PID hybrid control strategy refers to a composite control method combining fuzzy logic and proportional-integral-derivative algorithm. Specifically, an embedded controller with online rule base correction can be used to realize this strategy, which effectively solves the temperature tracking problem of nonlinear time-varying systems.

[0076] The convolutional neural network predicts the heating power and scanning path, which refers to a time series prediction technology based on deep learning. Specifically, historical production data sets can be trained using the ResNet architecture to realize this network, which optimizes the spatial distribution of heating parameters through feature extraction.

[0077] The core innovation of the present application is to build a multi-dimensional collaborative system of physical metallurgy and intelligent control. By regulating the grain boundary migration dynamics through the pinning effect of Nb(C, N) precipitates, combined with the joint optimization of three-dimensional heat conduction model and convolutional neural network, closed-loop control of temperature field and adaptive adjustment of annealing process are realized. This technical route that integrates material thermodynamic simulation, electromagnetic induction compensation and machine learning prediction breaks through the limitations of traditional single process parameter optimization, significantly improves production efficiency while ensuring stable magnetic properties.

[0078] The working process and principle of the present application are that the method solves the influence of temperature fluctuation on magnetic performance through multi-dimensional cooperative control technology. First, the molten steel is dosed, the specific proportion range of elements such as C and Si is limited, the balance of matrix strength and magnetic performance is ensured, and the contents of Sn and Cu are controlled to optimize the distribution of precipitated phases. Then the billet is heated to 1000-1100℃ for hot rolling, and the initial organization morphology is regulated by the combination of temperature and deformation amount. The billet after hot rolling is normalized at 850-900℃ to refine the grains and eliminate the hot rolling stress. The billet after normalizing is cold-rolled under the condition of Reduction rate≥20% to provide energy storage for subsequent recrystallization. After cold rolling, the billet is pickled with 2-4% hydrochloric acid to remove the surface oxide layer and avoid impurity interference.

[0079] The core process lies in high-temperature annealing, niobium element regulation, temperature compensation control and annealing temperature optimization. High-temperature annealing is carried out at 1150-1200℃ to lay a foundation for subsequent precipitated phase regulation. In the process of high-temperature annealing, 0.005-0.01wt% of Nb is added to form nanoscale Nb(C,N) precipitates. The thermodynamic behavior of precipitated phases is simulated by Thermo-Calc software, the Nb content and precipitate size distribution are optimized by combining Zener pinning force formula, the grain boundary migration is inhibited by using pinning effect, and the preferred orientation of secondary recrystallization is ensured.

[0080] Temperature compensation control adopts infrared thermal imaging and electromagnetic induction heating linkage, establishes a three-dimensional heat conduction model by combining finite element method, compensates the temperature gradient at the edge by longitudinal magnetic induction heating device, realizes closed-loop control of local temperature compensation, and solves the problem of edge temperature fluctuation in traditional methods. Annealing temperature optimization introduces a hybrid control strategy based on fuzzy PID, inputs the temperature fluctuation data into convolutional neural network to predict the optimal heating power and scanning path, optimizes the annealing temperature curve through the multivariate coupled machine learning model, realizes the dynamic optimization of the annealing curve, and breaks through the limitations of traditional single parameter regulation.

[0081] As a preferred embodiment, the scheme of the present application is implemented as follows:

[0082] First, the molten steel is dosed according to the following weight fraction: C 0.05%, Si 3.2%, Mn 0.10%, S 0.010%, Al 0.030%, N 0.025%, Sn 0.15%, Cu 0.5%, and the balance is Fe. The prepared molten steel is cast into a billet.

[0083] The billet is heated to 1050℃ and hot-rolled to a thickness of 2 times the original thickness. The billet after hot rolling is normalized at 880℃ for 2 hours. The billet after normalizing is cold-rolled under the condition of Reduction rate 25%. After cold rolling, the billet is pickled with 3% hydrochloric acid.

[0084] The pickled steel billet is placed in a BAF furnace and high-temperature annealing is performed at 1180℃. During high-temperature annealing, 0.008wt% of Nb is added. The thermodynamic behavior of the Nb(C,N) precipitate phase is simulated by Thermo-Calc software, and the optimal Nb content and precipitate size distribution are determined by combining the Zener pinning force formula.

[0085] An infrared thermal imager is used to monitor the temperature field distribution in the furnace in real time, and a three-dimensional heat conduction model is established by combining the finite element method. Through the longitudinal magnetic induction heating device, the detected edge temperature gradient is compensated in real time. Based on the fuzzy PID control strategy, the collected temperature fluctuation data are input into the convolutional neural network to predict the optimal heating power and scanning path.

[0086] Through a multivariate coupled machine learning model, a random forest model is constructed with temperature gradient, Als content and Nb precipitate density as input parameters. Combined with genetic algorithm iterative optimization of process parameters, the annealing temperature curve is dynamically and adaptively adjusted.

[0087] Through the above scheme, the present application realizes the grain boundary pinning strengthening under low temperature conditions, effectively suppresses the abnormal growth of grains, and improves the uniformity and completeness of secondary recrystallization. By using the linkage mode of infrared thermal imaging and electromagnetic induction heating, real-time monitoring and accurate control of the temperature field in the furnace are realized, and the transverse temperature fluctuation range is effectively reduced. Based on the fuzzy PID hybrid control strategy, combined with convolutional neural network prediction, closed-loop control of local temperature compensation is realized, which significantly improves the accuracy and real-time performance of temperature fluctuation suppression. By constructing a multivariate coupled machine learning model, the annealing temperature curve is dynamically and adaptively adjusted, which effectively reduces the iron loss and improves the magnetic induction. By using in-situ EBSD to analyze the grain boundary migration trajectory during annealing, and combining with the characterization of precipitates to verify the pinning effect, the stability and uniformity of the pinning inhibitor are ensured, which provides a reliable guarantee for the stable production of high-magnetic-induction oriented silicon steel.

[0088] In some of the above schemes of the present application, there are problems such as the inability to dynamically simulate the thermodynamic behavior of the precipitate phase, the difficulty in quantifying the influence of the volume fraction and size distribution of the precipitate phase on the pinning force, and the lack of quantitative analysis of the interfacial energy and the stability of the precipitate phase, which leads to unstable pinning effect and low efficiency of process parameter optimization.

[0089] To this end, the present application further proposes a method of simulating the behavior of the precipitate phase by thermodynamic software during high-temperature annealing and optimizing the pinning force by machine learning algorithm.

[0090] The thermodynamic simulation part loads the steel system database and inputs the steel composition, temperature range and pressure condition, simulates the type of precipitated phase, its volume fraction and precipitation temperature window under different niobium contents through the equilibrium phase calculation module. In combination with the TC-PRISMA module for simulating the size distribution of the precipitated phase, the input cooling rate parameter is controlled in the range of 0.5-2.0℃ / s, and the nucleation activation energy is set to 150-250kJ / mol, so as to obtain the dynamic evolution data of the precipitate average radius in the range of 5-50nm. The phase field coupling algorithm is used when converting the molar fraction to the volume fraction, and the CALPHAD method is used for analyzing the interfacial energy of the sublattice composition, and the value is between 0.5-1.5J / m 2 The characteristic variables of the random forest algorithm in the Zener model optimization include the temperature gradient, the niobium content and the cooling rate, and the objective function is set as the dual-objective optimization of the maximum pinning force and the minimum grain size standard deviation.

[0091] Specifically, in the high-temperature annealing stage, the mapping relationship between the steel composition and the process condition is established through the thermodynamic software, the volume fraction data output by the equilibrium phase calculation module and the size distribution data of the precipitated phase generated by the TC-PRISMA module form a three-dimensional parameter space. When the input niobium content is 0.005-0.01wt%, the volume fraction of the precipitated phase fluctuates in the range of 0.02%-0.08%, and the average radius decreases exponentially with the increase of the cooling rate. The interfacial energy calculation adopts the atomic interaction parameter in the sublattice model, and the chemical stability of the precipitate is determined by the principle of minimizing the Gibbs free energy. The Zener pinning force model establishes a quantitative relationship between the thermodynamic output parameters and the pinning strength, and sets the volume fraction weight coefficient to 0.6, the interfacial energy coefficient to 0.3 and the average radius coefficient to 0.1 in the multivariate model. The crossover probability is set to 0.8 and the mutation probability is set to 0.05 in the genetic algorithm iteration process, and after 50-100 generations of evolution, the optimal process parameter combination is obtained, which increases the pinning force by 15-30% and reduces the grain size dispersion to within 10%.

[0092] As a preferred embodiment, the scheme of the application is implemented as follows:

[0093] Load the TCFE9 database in the Thermo-Calc software, input the steel composition as C 0.05%, Si 3.2%, Mn 0.1%, S 0.01%, Al 0.03%, N 0.025%, Sn 0.2%, Cu 0.5%, Nb 0.008%, and set the temperature range as 800-1300℃ and the pressure as 1 atmosphere. Simulate the types of precipitated phases, their volume fractions and precipitation temperature windows under different Nb contents by the equilibrium phase calculation module. Combine the TC-PRISMA module to simulate the size distribution of the precipitated phases, input the cooling rate as 5℃ / s, and the nucleation condition parameter as 10℃ of supercooling. Obtain the dynamic evolution data of the average radius and number density of the precipitates.

[0094] Calculate the volume fraction of the precipitated phases f = ∑(V precipitated phase / V total) by the mole fraction and phase composition data output by Thermo-Calc. Determine the chemical stability of the precipitates and the interface energy σ between the precipitates and the matrix as 0.5J / m 2 .

[0095] Substitute the f = 0.002, r = 5nm and σ = 0.5J / m 2 parameters output by Thermo-Calc into the Zener pinning force formula Z = 3σf / 4r to construct a multivariate model and predict the optimal Nb content and process conditions. Use the random forest algorithm to optimize the objective function, balance the requirements of precipitated phase strengthening and grain refinement, and obtain the optimal Nb content as 0.008%.

[0096] Through the above technical solutions, the present application realizes accurate simulation and parameter optimization of the thermodynamic behavior of the precipitated phases. Through the equilibrium phase calculation and TC-PRISMA module of the Thermo-Calc software, the dynamic evolution data of key parameters such as the type, volume fraction and size distribution of the precipitated phases are obtained. Combined with the Zener pinning force formula and machine learning algorithm, a multivariate model is established to realize accurate matching of the Nb content and the precipitate parameters. Thus, the problem that the precipitate parameters are difficult to accurately match the pinning requirements in the regulation of niobium elements is solved, and the stability of the pinning effect and the efficiency of process parameter optimization are improved.

[0097] In some of the above schemes of the present application, the traditional temperature monitoring means has insufficient accuracy, resulting in inaccurate positioning of local temperature difference abnormal areas, the edge temperature gradient compensation mechanism lacks dynamic response capability, the power adjustment of the electromagnetic induction heating device lags behind the temperature change, and closed-loop control cannot be realized, so that the temperature field compensation effect is limited.

[0098] To this end, the application further proposes an infrared thermal imaging and electromagnetic induction heating linkage temperature compensation control method, which includes real-time acquisition of furnace temperature field data, establishment of a three-dimensional heat conduction model to locate abnormal areas, use of a movable coil to compensate for the temperature gradient at the edge, application of a hybrid control strategy to adjust the heating power, and prediction of the optimal heating parameters through a neural network.

[0099] Among them, the temperature field data acquisition uses an infrared thermal imager, with a monitoring accuracy controlled within ±2℃, ensuring the accuracy of data acquisition. The three-dimensional heat conduction model is constructed by the finite element method, which can dynamically simulate the temperature distribution state in the furnace, and combined with temperature monitoring data, it can identify local temperature abnormal areas. The movable coil uses a longitudinal magnetic field induction heating device, which adjusts the spatial distribution of the electromagnetic field through staggered layout to compensate for the temperature gradient formed by the difference in heat conduction at the edge. The hybrid control strategy combines fuzzy logic and PID algorithm to dynamically adjust the output power of the induction coil according to the real-time temperature error and its rate of change. Temperature fluctuation data is processed through a convolutional neural network to extract the spatial distribution characteristics of the temperature field, predict the optimal heating power parameters and coil scanning path, and form a closed-loop control loop.

[0100] Specifically, the infrared thermal imager acquires the temperature field distribution data in the furnace at a frequency of 30 frames per second, and inputs the temperature data into the finite element model for real-time calculation. When the model output shows that there is a temperature fluctuation of more than ±8℃ in the edge area, the movable coil of the longitudinal magnetic field induction device is adjusted to the target area according to the coordinate positioning, and the coil spacing is controlled within the range of 50-80mm. The fuzzy PID controller receives the temperature error signal, and when the absolute value of the error exceeds 3℃, the fuzzy reasoning module is started, and the duty cycle adjustment instruction is output to the power regulation unit. The convolutional neural network model is trained based on historical temperature fluctuation data set, the input layer receives the temperature field data 5 seconds before and after the current time, and the output layer generates the heating power adjustment curve and coil movement path planning for the next 10 seconds. This closed-loop system updates the control parameters every 200 milliseconds, so that the temperature fluctuation amplitude is stabilized within ±5℃.

[0101] As a preferred embodiment, the scheme of the application is implemented as follows:

[0102] The infrared thermal imaging and electromagnetic induction heating linkage temperature compensation control method includes the following steps:

[0103] First, the infrared thermal imager is used to acquire the temperature field data in the furnace in real time. The infrared thermal imager uses a non-cooled microbolometer with a wavelength range of 8-14μm, a temperature measurement range of 0-1500℃, and a temperature measurement accuracy of ±2℃. The thermal imager is installed on the top of the furnace and scans the temperature field in the furnace through a quartz observation window at a frequency of 10Hz.

[0104] Secondly, a three-dimensional heat conduction model was established by finite element method. The geometric model of the furnace body was constructed using ANSYS software, and the mesh was divided into hexahedral elements with a total number of about 1 million. The thermal physical properties of the materials were input according to the performance of the actual used refractory and steel belt. The boundary conditions included heat loss of furnace wall, radiation heat transfer, etc. The temperature field distribution in the furnace was simulated by transient thermal analysis, and the calculation time step was 1 s. According to the simulation results, the local temperature difference abnormal area was located.

[0105] Thirdly, the longitudinal magnetic induction heating device was used to compensate the temperature gradient at the edge. The longitudinal magnetic induction heating device was composed of a movable coil, which was wound with water-cooled copper pipes with an inner diameter of 100 mm, an outer diameter of 150 mm, and a height of 200 mm. The coil was driven by a servo motor and could move in X, Y, and Z directions with a maximum stroke of 500 mm. Through the staggered layout, the magnetic field action range was optimized to realize accurate compensation of the temperature gradient at the edge.

[0106] Then, the heating power was adjusted based on the fuzzy PID hybrid control strategy. The input variables of the fuzzy controller were temperature error e and error change rate ec, and the output variables were the adjustment amounts of PID parameters Kp, Ki, and Kd. The fuzzy rule base contained 49 if-then rules. The PID controller used incremental algorithm and dynamically adjusted PID parameters according to the fuzzy reasoning results to output the heating power adjustment amount.

[0107] Finally, the temperature fluctuation data was input into the convolutional neural network to predict the optimal heating power and scanning path. The convolutional neural network had a 5-layer structure, including 2 convolutional layers, 2 pooling layers, and 1 fully connected layer. The input was a 128×128 pixel temperature field image, and the output was the heating power and scanning path parameters. The network was trained by Adam optimizer with a learning rate of 0.001, a batch size of 32, and an iteration number of 10000. According to the prediction results, the heating power and coil moving trajectory were adjusted in real time to realize accurate compensation control of the temperature field.

[0108] By the technical solution, high-precision real-time monitoring and accurate control of the temperature field in the furnace are achieved. The infrared thermal imaging technology provides high-precision temperature field data of ±2℃, laying an accurate data foundation for subsequent modeling and control. The three-dimensional heat conduction model constructed by the finite element method can accurately simulate the dynamic distribution of the temperature field in the furnace and effectively locate the local temperature difference abnormal area. The movable coil staggered layout of the longitudinal magnetic induction heating device optimizes the magnetic field action range, and realizes accurate compensation of the temperature gradient at the edge. The fuzzy PID hybrid control strategy combines the robustness of fuzzy logic and the accuracy of PID control, dynamically adjusts the heating power according to the real-time change rate of the temperature error, and improves the system response speed. The convolutional neural network uses its spatial feature extraction capability to realize intelligent prediction of the optimal heating power and scanning path, forming a closed-loop control mechanism for temperature field compensation. This multi-dimensional collaborative control method significantly improves the stability and precision of temperature compensation, effectively suppresses the temperature fluctuation in the furnace, and provides a stable temperature environment for the uniform secondary recrystallization of oriented silicon steel.

[0109] In some of the above schemes of the present application, traditional machine learning models are difficult to effectively couple the temperature gradient, chemical composition and precipitated phase evolution, etc. Multivariable parameters, resulting in the lack of accurate prediction of the dynamic behavior of the AlN phase in the optimization of the annealing curve, and failing to realize the integrity and uniformity control of the secondary recrystallization combined with the thermodynamic database.

[0110] To this end, the present application further proposes a machine learning model coupled with multiple variables, specifically including: establishing a random forest model with temperature gradient ΔT, Als content, and Nb precipitated phase density as input parameters; iteratively optimizing process parameters through genetic algorithm, and training the model combined with historical production data; constructing an Arrhenius equation correction term to predict the spatiotemporal evolution law of AlN volume fraction and size; optimizing the annealing curve through the JMatPro thermodynamic database to ensure the integrity and uniformity of the secondary recrystallization.

[0111] Wherein, the value range of the temperature gradient ΔT is limited to 50-150℃ / min, the Als content is controlled in the interval of 0.020%-0.042%, and the Nb precipitated phase density is set to 1×10^20-5×10^21 / m 3 . The crossover probability of the genetic algorithm is set to 0.6-0.8, the mutation probability is adjusted to 0.01-0.05, and the evolution number is optimized to 200-500 times. The activation energy parameter Q of the Arrhenius equation correction term is corrected to 180-220kJ / mol, and the pre-exponential factor A is adjusted to 1×10^12-5×10^14s^-1. The phase equilibrium calculation module in the JMatPro thermodynamic database is called, and the input parameters include the temperature field distribution gradient, the alloy element diffusion coefficient and the interface energy data.

[0112] Specifically, the temperature gradient ΔT and the Als content are normalized by feature engineering, and the Gini coefficient is used as the splitting criterion when the decision tree node of the random forest model splits. In the iteration process of genetic algorithm, the heating rate and holding time of the annealing curve are coded as chromosomes. The fitness function is defined as the weighted product of the secondary recrystallization completeness and the magnetic induction B8 value. The Arrhenius equation correction term is embedded in the neural network hidden layer, and the error rate of the predicted value of the AlN volume fraction and the transmission electron microscope measured data is controlled within ± 3%. The phase transition driving force data output by the JMatPro database is converted into temperature compensation coefficients, which are used to dynamically adjust the gradient change slope of the annealing curve. In the model training process, the temperature fluctuation spectrum characteristics in the historical production data are extracted as joint input parameters in time domain and frequency domain, and the best heating power adjustment mode is matched through convolution kernel operation. The temperature field reconstruction module maps the optimized annealing curve into a three-dimensional heat flow distribution diagram, which forms a closed loop feedback with the power output of the electromagnetic induction heating device. As a result, the matching error between the AlN phase size evolution rate and the secondary recrystallization grain boundary migration speed is reduced to less than 5%, and the grain orientation deviation angle is controlled within 3°.

[0113] As a preferred embodiment, the scheme of the application is implemented as follows:

[0114] The multivariate coupled machine learning model includes the following steps:

[0115] Firstly, a random forest model is established with the temperature gradient ΔT, the Als content and the Nb precipitated phase density as input parameters. This model uses the decision tree integration method to improve the prediction accuracy and generalization ability through the combination of multiple decision trees. In the input parameters, the temperature gradient ΔT can be measured by a thermocouple array, the Als content is determined by a spectral analyzer, and the Nb precipitated phase density is obtained by transmission electron microscopy observation and statistics.

[0116] Secondly, the process parameters are iteratively optimized by genetic algorithm, and the model is trained combined with historical production data. The genetic algorithm uses binary coding, the population size is set to 100, the crossover probability is 0.8, and the mutation probability is 0.1. In the iteration process, the minimum temperature fluctuation and the maximum magnetic performance are used as the objective function, and the parameter combination is continuously optimized through selection, crossover and mutation operations. The historical production data includes temperature curves, Als content changes, Nb precipitated phase evolution and magnetic performance indicators of the final product under different process conditions.

[0117] Further, a modified Arrhenius equation is constructed to predict the spatiotemporal evolution of the volume fraction (f) and size (r) of AlN. The modified equation introduces temperature-dependent diffusion coefficient and interfacial energy, and considers the coupling between AlN precipitation kinetics and temperature field. The specific expression is: f(t, T) = f0exp(-Q / RT)(1-exp(-kt^n)), where Q is the apparent activation energy, R is the gas constant, k is the rate constant, and n is the time index. r(t, T) is derived through the LSW theory.

[0118] Finally, the annealing curve is optimized by the JMatPro thermodynamic database to ensure the completeness and uniformity of the secondary recrystallization. The JMatPro software is used to calculate the phase equilibrium state at different temperatures to obtain the dissolution temperature and volume fraction of AlN, Nb(C, N) and other precipitated phases. Based on these thermodynamic data, combined with the kinetic model, a multi-stage annealing curve is designed to achieve accurate control of the precipitated phase.

[0119] Through the above technical solutions, the present application realizes accurate prediction of the dynamic behavior of the AlN phase, effectively coupling the temperature gradient, chemical composition and precipitated phase evolution, and other multi-variable parameters. This method significantly improves the accuracy and reliability of the annealing curve optimization, and provides strong support for the uniformity control of grain growth during the secondary recrystallization process. At the same time, by integrating the thermodynamic database, the dynamic regulation of phase equilibrium during the annealing process is ensured, further optimizing the magnetic properties and microstructure uniformity of the oriented silicon steel.

[0120] In some of the above schemes of the present application, a high-temperature annealing process is proposed to realize the preferred growth of grains in the secondary recrystallization stage. However, when performing a conventional high-temperature annealing, due to the improper matching of the heating and cooling rates and the holding time, it may lead to uneven distribution of precipitated phases or hindered grain boundary migration, thereby causing negative effects of temperature fluctuations on the grain size and stability of precipitates.

[0121] To this end, the present application further proposes the following technical solutions: in the high-temperature annealing stage, first heat the material to a temperature in the range of 1150-1200℃ and maintain for 1-2 hours, then naturally cool to a temperature in the range of 950-1000℃ for 1-2 hours of holding, then heat again to a temperature in the range of 1050-1100℃ and maintain for the same holding time, and finally cool to room temperature in a natural cooling manner.

[0122] The temperature regulation is divided into four stages: an initial high-temperature stage, an intermediate cooling stage, a secondary heating stage, and a final cooling stage. In the initial high-temperature stage, by controlling the temperature in the range of 1150-1200℃, the full recrystallization of austenite grains is promoted, and sufficient thermal activation conditions are provided for the stable nucleation of precipitates. The holding time is set to 1-2 hours to ensure that the precipitates in the matrix reach the critical size and form a uniform distribution. In the intermediate cooling stage, natural cooling is used to gently reduce the temperature gradient to the range of 950-1000℃, which can avoid the stress concentration at the grain boundary caused by rapid cooling, while retaining the interface bonding strength between the precipitates and the matrix. In the secondary heating stage, the temperature is raised to 1050-1100℃, which can activate the driving force required for grain boundary migration, but is lower than the peak temperature in the initial high-temperature stage, effectively inhibiting the coarsening or dissolution of precipitates. The final natural cooling process maintains the size stability of the grain structure by slowly releasing heat.

[0123] Specifically, the holding treatment in the initial high-temperature stage promotes the formation of dispersed distribution of AlN and Nb(C,N) precipitates in the matrix, providing a basis for subsequent grain pinning. The natural cooling rate in the intermediate cooling stage is controlled in the range of 15-20℃ / min, avoiding thermal expansion mismatch between the precipitates and the matrix through thermodynamic equilibrium. In the secondary heating stage, the temperature difference is used to drive the grain boundary migration, while the Zener pinning force and the grain boundary migration speed reach a dynamic balance. In the final cooling stage, the overall temperature of the material is uniformly reduced by natural heat dissipation, eliminating local residual stress. The four-stage temperature path matches the evolution law of precipitates and the grain growth kinetics in each temperature range, so that the grain size deviation is controlled within ±5μm during the secondary recrystallization process, the average radius of precipitates is stabilized in the range of 20-50nm, and the Gaussian texture ratio is improved to more than 95%.

[0124] As a preferred embodiment, the scheme of the present application is implemented as follows:

[0125] The high-temperature annealing process includes the following steps:

[0126] First, the pickled steel billet is placed in an annealing furnace, and the temperature is raised to 1180℃ at a rate of 5℃ / min, and held for 1.5 hours. In this stage, the grains are fully recrystallized and a stable precipitate distribution is formed.

[0127] Next, the temperature is reduced to 975℃ by natural cooling, and held for 1.5 hours. This step avoids the destruction of the interface bonding between the precipitates and the matrix due to thermal stress caused by rapid cooling, and provides thermodynamic equilibrium conditions for subsequent heating.

[0128] Then, the temperature is raised to 1075℃ at a rate of 3℃ / min, and the temperature is kept for 1.5 hours. This stage activates the driving force of grain boundary migration required for secondary recrystallization, while avoiding the coarsening or dissolution of precipitates caused by excessively high temperature.

[0129] Finally, the temperature is reduced to room temperature in a natural cooling manner. The stability of the grain structure is maintained through the slow heat release process.

[0130] During the entire annealing process, an infrared thermal imager is used to monitor the temperature distribution in the furnace in real time, and local temperature compensation is performed in combination with the electromagnetic induction heating device to ensure that the temperature control accuracy of each stage is within ±2℃.

[0131] Through the above technical solutions, the present application effectively suppresses the temperature fluctuations in the high-temperature annealing process. The staged control annealing process accurately matches the evolution law of the precipitates and the grain growth kinetics in each temperature interval, effectively suppressing the temperature fluctuations caused by thermal shock or insufficient holding during a single temperature rise and fall process. The multi-stage temperature control strategy significantly improves the uniformity and completeness of secondary recrystallization, and improves the magnetic properties of oriented silicon steel. At the same time, by optimizing the matching of the temperature rise and fall rate and the holding time, the problems of uneven distribution of precipitates and obstruction of grain boundary migration are avoided, and the stability of product quality is improved.

[0132] In some of the above schemes of the present application, the temperature fluctuations present in the high-temperature annealing process will cause the secondary recrystallization of the grains to be uneven, affecting the stability of the magnetic properties of the oriented silicon steel. To this end, the present application further proposes a high-temperature annealing process including staged temperature control, specifically including four key temperature control steps.

[0133] Among them, the first stage is to raise the temperature to 1150-1200℃ and keep it for 1-2 hours, which can promote the full progress of the primary recrystallization, for example, under the condition of 1180℃ and 1.5 hours, the grain size can reach 5-8 microns. The second stage is to naturally cool to 950-1000℃ and keep it for 1-2 hours, which can make the Nb(C,N) precipitates uniformly distributed at the grain boundaries by slowly cooling, and the average size of the precipitates can be controlled in the range of 20-50 nanometers when kept at 980℃. The third stage is to raise the temperature to 1050-1100℃ again and keep it for 1-2 hours, which can activate the pinning effect, for example, the grain boundary migration rate can be reduced to below 0.5μm / s under the condition of 1080℃ and holding. The fourth stage is to naturally cool to room temperature to complete the annealing, and the cooling process maintains a temperature gradient of 10-15℃ / min to prevent stress concentration.

[0134] Specifically, the staged temperature control mechanism forms a temperature fluctuation buffer zone through two temperature rises and two temperature holding, in which the first high temperature holding ensures the full recrystallization of the base body, and the intermediate temperature stage promotes the stabilization of the precipitated phase. The temperature gradient in the natural cooling stage is controlled within ±8℃ / m, effectively reducing the influence of thermal stress on grain orientation. When synergized with the nano-Nb(C,N) precipitated phase in the pre-process, this temperature curve can increase the secondary recrystallized grain size distribution concentration by more than 30%. By precisely controlling the holding time of each stage, for example, limiting the second stage holding time to 1.5 hours, the grain growth kinetics and inhibitor stability can be balanced, ultimately achieving the technical effect of reducing iron loss to 0.85 W / kg.

[0135] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for suppressing temperature fluctuation of an oriented silicon steel BAF furnace, characterized by, The method comprises: Step 1, molten steel ingredient: according to the following weight fraction, molten steel is prepared, C: 0.02%-0.08%; Si: 3.0%-3.5%; Mn: 0.02%-0.18%; S: 0.005%-0.020%; Al: 0.020%-0.042%; N: 0.015%-0.040%; Sn: 0.08%-0.3%; Cu: 0.1%-0.9%; the balance is Fe; Step 2, hot rolling: the billet is heated to 1000-1100℃, and hot rolling is carried out to the thickness of 1.5-2.5 times of the original thickness; Step 3, normalizing: the hot-rolled billet is normalized at 850-900℃; Step 4, cold rolling: the normalized billet is cold rolled under the condition that the Reduction rate is greater than or equal to 20%; Step 5, pickling: the cold-rolled billet is pickled with 2-4% hydrochloric acid; Step 6, high temperature annealing: the pickled billet is high temperature annealed at 1150-1200℃; Step 7, niobium element regulation: 0.005-0.01wt% Nb is added during high temperature annealing to form nanoscale Nb(C, N) precipitates, the thermodynamic behavior of precipitated phase is simulated by Thermo-Calc software, the Nb content and precipitate size distribution are optimized by Zener pinning force formula to ensure the preferred growth of Gauss grains in the secondary recrystallization stage; Step 8, temperature compensation control: infrared thermal imaging and electromagnetic induction heating are linked, a three-dimensional heat conduction model is established by combining finite element method, the temperature gradient of the edge is compensated by longitudinal magnetic induction heating device to realize closed-loop control of local temperature compensation; Step 9, annealing temperature optimization: based on the mixed control strategy of fuzzy PID, the temperature fluctuation data is input into convolutional neural network to predict the optimal heating power and scanning path, and the annealing temperature curve is optimized by multivariate coupled machine learning model.

2. The method of claim 1, wherein the method is characterized by: The step 7 comprises: Step 701, simulating the thermodynamic behavior of precipitated phase by Thermo-Calc software: Step 7011, loading the database suitable for steel system in Thermo-Calc, inputting the composition of steel grade, temperature range and pressure condition; Step 7012, simulating the type of precipitated phase, its volume fraction (f) and precipitation temperature window under different Nb content by equilibrium phase calculation module, combining TC-PRISMA module to simulate the size distribution of precipitated phase, inputting cooling rate and nucleation condition parameters to obtain the dynamic evolution data of precipitate average radius (r) and number density (Nv); Step 7013, calculating the volume fraction f of precipitated phase according to the mole fraction and phase composition data output by Thermo-Calc, f = ∑(V precipitated phase / V total); Step 7014, determining the chemical stability of precipitate and the interface energy between precipitate and matrix by combining precipitated phase sublattice composition analysis; Step 702, optimization of Zener pinning force formula: Step 7021, the Zener pinning force formula is: Z = 3σf / 4r, wherein, σ is the interface energy between precipitated phase and matrix, f is the volume fraction, and r is the average radius; Step 7022, input the f, r, σ parameters output by Thermo-Calc into the Zener formula to build a multivariate model and predict the optimal Nb content and process conditions; Step 7023, use random forest or genetic algorithm to optimize the objective function (maximize Z value) to balance the strengthening of precipitated phases and the demand for grain refinement.

3. The method of claim 1, wherein the method is characterized by: In step 8, the temperature compensation control method of infrared thermal imaging combined with electromagnetic induction heating includes: Step 801, use an infrared thermal imager to collect real-time furnace temperature field data with an accuracy of ±2℃; Step 802, establish a three-dimensional heat conduction model by finite element method to locate the local temperature difference abnormal area; Step 803, use a longitudinal magnetic induction heating device to compensate for the temperature gradient at the edge by movable coil staggered layout; Step 804, based on the fuzzy PID hybrid control strategy, adjust the heating power according to the temperature error and change rate; Step 805, input the temperature fluctuation data into a convolutional neural network to predict the optimal heating power and scanning path, and realize accurate compensation control of the temperature field.

4. The method of claim 1, wherein the method is characterized by: In step 9, the multivariate coupled machine learning model includes: Step 901, establish a random forest model with temperature gradient ΔT, Als content, and Nb precipitated phase density as input parameters; Step 902, use genetic algorithm to iteratively optimize process parameters and train the model based on historical production data; Step 903, construct the Arrhenius equation correction term to predict the spatiotemporal evolution law of AlN volume fraction (f) and size (r); Step 904, optimize the annealing curve through the JMatPro thermodynamic database to ensure the integrity and uniformity of secondary recrystallization.

5. The method of claim 1, wherein the method is characterized by: In step 6, it includes: Step 601, heat to 1150-1200℃ and hold for 1-2 hours; Step 602, naturally cool to 950-1000℃ and hold for 1-2 hours; Step 603, heat to 1050-1100℃ and hold for 1-2 hours; Step 604, naturally cool to room temperature.

Images