Method for pulsed electromagnetic force application pattern decision for hot forming of ultra-high strength steel

Abstract

The present application relates to the technical field of metal material hot working, in particular to a pulse electromagnetic force application mode decision method for super high strength steel hot forming. The method comprises obtaining geometric characteristic parameters of a part to be formed, initial performance parameters of a plate and process target performance parameters; inputting the above parameters into a pre-trained decision model; outputting pulse electromagnetic force application control parameters for the hot forming process from the decision model, the control parameters including a time sequence scheme, a spatial action mode and an energy intensity scheme of pulse electromagnetic force application; and controlling a pulse electromagnetic force generating device to perform corresponding operations according to the control parameters. The method realizes intelligent and global collaborative decision of pulse electromagnetic force application timing, spatial action position and energy size through an artificial intelligence model, can automatically generate an optimal application strategy matched with a specific part and process condition, and significantly improves the forming quality, microstructure performance and process consistency of super high strength steel hot formed parts.

Description

Technical Field

[0001] This invention relates to the field of hot working technology of metal materials, and specifically to a pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel. Background Technology

[0002] Hot stamping forming technology for ultra-high strength steel is a key process in lightweight sheet metal manufacturing. It achieves extremely high strength by forming the material in an austenitic state after heating and then rapidly quenching it. However, this process also suffers from problems such as relatively limited material fluidity at high temperatures, difficulty in filling complex features, uneven cooling phase transformation leading to springback, and differences in microstructure and properties. To improve these shortcomings, pulsed electromagnetic force-assisted technology has been introduced in recent years. By applying a transient high-intensity electromagnetic field during the forming process, the Lorentz force and magnetostrictive effect generated can promote material flow, refine the microstructure, and reduce springback. However, the introduction of pulsed electromagnetic force also brings significant complexity to process control: its final effect is highly dependent on the timing of application (at the optimal material plasticity or critical phase transformation point), spatial distribution (which areas of the part are affected), and energy intensity (pulse size, frequency, and waveform). Currently, determining process parameters heavily relies on engineers' experience and complex trial-and-error experiments, or can only optimize a single factor (such as energy level) to a limited extent. This current model struggles to meet the production demands of parts with varying geometries and performance requirements, and it fails to achieve coordinated global optimization of pulsed electromagnetic force across time, space, and energy dimensions. Consequently, the technology's potential remains untapped, and process stability and part consistency face challenges. Therefore, there is an urgent need for a method capable of intelligently, accurately, and automatically determining the optimal pulsed electromagnetic force application mode to unleash the full potential of this technology.

[0003] Therefore, existing technologies still need further development. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel, so as to solve the problems existing in the prior art.

[0005] To achieve the above-mentioned technical objectives, the present invention provides a pulsed electromagnetic force application mode decision-making method for hot forming of ultra-high strength steel, comprising: S1. Obtain the geometric feature parameters of the part to be formed, the initial performance parameters of the sheet metal, and the target performance parameters of the process. S2. Input the parameters into the trained decision model, and the decision model outputs control parameters for applying pulsed electromagnetic force to the thermoforming process; S3. The control parameters include the timing scheme, spatial action mode, and energy intensity scheme of the pulse electromagnetic force application, which are used to control the pulse electromagnetic force generating device to perform corresponding operations during the thermoforming process.

[0006] Specifically, obtaining the geometric feature parameters of the part to be formed includes: Based on the 3D CAD model of the part, the overall dimensions, cross-sectional shape change rate, minimum bending radius to sheet thickness ratio, and Gaussian curvature distribution of the feature region are extracted. The initial performance parameters of the sheet material include at least the grade, yield strength, tensile strength, elongation, and CCT curve characteristics of the austenitizing transformation temperature range. The target performance parameters of the process include at least the tensile strength range of each zone of the target part after forming, the allowable tolerance of springback angle, and the upper limit of the thickness reduction rate of the key thinning area.

[0007] Specifically, the training data for the decision model comes from historical thermoforming process experiments or experimentally calibrated finite element simulation data; each training data sample includes: an input vector and a corresponding output vector label; the input vector includes geometric feature parameters, initial performance parameters of the sheet material, historical temperature data of the simulated or measured process, and preset target performance parameters; the output vector label is the optimal pulse electromagnetic force application control parameter obtained after process optimization under the input conditions.

[0008] Specifically, the simulated or measured historical temperature data of the process includes: the temperature cooling curve of the plate from the heating furnace to the mold, the temperature field of the unsteady heat transfer process in the early stage of contact between the plate and the mold, and the cooling rate curve of each part under the influence of the mold cooling pipe during the pressure holding and quenching stage; the historical temperature data is used as the input of the decision model in the form of time series or feature values.

[0009] Specifically, the decision model is a machine learning model, specifically a deep neural network, a gradient boosting decision tree, or a support vector regression model; the model learns the nonlinear mapping relationship between the input vector and the output vector labels to establish a predictive mapping from process conditions to optimized control parameters.

[0010] Specifically, the input vector of the model needs to undergo feature engineering processing before being input into the model. This processing includes: The geometric feature parameters are dimensionless, the temperature history data is piecewise integrated to extract the cumulative thermal effect features, the CCT curve features in the initial performance parameters of the plate are extracted for key phase transition point temperature and time, and all features are normalized and spliced ​​into a unified feature vector.

[0011] Specifically, the training process of the decision model includes multi-objective optimization. Its loss function is configured to simultaneously minimize the errors in predicting temporal schemes, spatial action modes, and energy intensity schemes in the output vector labels. Multiple sub-losses are combined into a total loss through weighted summation for backpropagation and updating of model parameters.

[0012] Specifically, the timing scheme for applying the pulsed electromagnetic force is output by the decision model as a discrete-time sequence activation signal corresponding to the thermoforming process timeline; this sequence defines the specific start time and duration of applying the pulsed electromagnetic force in at least one of the following stages: sheet material transfer stage, initial contact with the mold, mold closing and loading stage, and pressure holding and quenching stage.

[0013] Specifically, the spatial action mode corresponds to the collaborative working strategy of multiple independent control coils in the pulsed electromagnetic force generator; the spatial action mode parameters output by the decision model include the combination of coil numbers to be activated, the relative current phase difference of each activated coil, and the spatial superposition distribution description of the action area of ​​the pulsed magnetic field generated by each coil on the board, which are used to form a gradient magnetic field or a uniformly enhanced magnetic field in a specific area of ​​the board.

[0014] Specifically, the energy intensity scheme is output by the decision model as a set of pulse energy parameters that match the timing scheme and spatial action mode; for each activated coil, the set of pulse energy parameters independently defines the peak current density, pulse frequency, and waveform envelope characteristics of the applied pulse current during one or more activation periods.

[0015] Beneficial effects: The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel provided by this invention brings many significant benefits.

[0016] First, its core lies in the intelligent and automated application of pulsed electromagnetic force strategies through a data-driven AI decision-making model. The model can comprehensively consider the geometric complexity of the part, the initial properties of the material, the dynamic temperature history, and multi-dimensional target performance requirements, automatically outputting globally optimized control parameters. This completely eliminates reliance on human experience and significantly improves the efficiency and scientific rigor of process development.

[0017] Secondly, this invention achieves extremely precise control over the application process of pulsed electromagnetic force. The model-determined schemes encompass timing schemes accurate to the millisecond level, spatial action modes based on independent collaborative control of multiple coils, and differentiated energy intensity schemes. This collaborative decision-making regarding "when to apply, where to apply, and how to apply" allows the pulsed electromagnetic force to act like an "invisible hand," pre-treating at the end of material transport, providing auxiliary power when contact flow resistance increases sharply, and implementing structural impact at critical phase transition stages. This precise intervention at each stage maximizes the improvement of formability and microstructure.

[0018] Furthermore, this method greatly improves the adaptability of the process and the consistency of part quality. Faced with parts of different shapes, materials and performance requirements, the model can quickly generate customized optimal application strategies, ensuring that whether the parts are simple or complex, they can obtain uniform and qualified mechanical properties and geometric accuracy, significantly reducing the risk of defects such as springback, cracking and incomplete filling.

[0019] Finally, this invention encapsulates valuable process experience and knowledge into a reusable model, promoting the digital and intelligent upgrading of thermoforming processes. It lays a solid foundation for building process digital twins and realizing adaptive closed-loop control, and has significant industrial application value. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the pulsed electromagnetic force application mode decision-making method for hot forming of ultra-high strength steel provided in a specific embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0022] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0023] Please see Figure 1 This invention provides a pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel, comprising: S1. Obtain the geometric feature parameters of the part to be formed, the initial performance parameters of the sheet metal, and the target performance parameters of the process.

[0024] It should be further noted that, in the implementation of this method, the decision model is first constructed and trained offline. The model training is based on a dataset constructed by combining historical experiments and calibrated finite element simulations. Each sample in the dataset fully includes the input conditions (i.e., geometry, material, and target parameters) under specific working conditions, as well as the optimal output results (i.e., control parameters) determined through process optimization.

[0025] S2. Input the parameters into the trained decision model, and the decision model outputs control parameters for applying pulsed electromagnetic force to the thermoforming process.

[0026] It should be further noted that the model uses a deep neural network (DNN), with the number of input layer nodes matching the dimension of the feature vectors after feature engineering, and the number of output layer nodes matching the total dimension of the control parameters requiring decision-making. In online applications, for a new thermoforming task, the operator or automated system calls the trained model, inputs the specific parameters of the current task, and the model outputs a complete set of control instructions within milliseconds.

[0027] S3. The control parameters include the timing scheme, spatial action mode, and energy intensity scheme of the pulse electromagnetic force application, which are used to control the pulse electromagnetic force generating device to perform corresponding operations during the thermoforming process.

[0028] It should be further explained that the instruction set is sent directly to the multi-channel controller of the pulse electromagnetic force generator. The controller parses the instructions and converts them into precise trigger signals, phase control signals and power output signals for each independent coil, thereby automatically executing the optimal pulse electromagnetic force application mode determined during the forming process.

[0029] Understandably, this method deeply integrates artificial intelligence decision-making with advanced pulsed electromagnetic force technology, constructing an intelligent closed loop of "perception-decision-execution." It fundamentally changes the traditional reliance on "trial and error" or fixed parameter models in process development. Through the model's learning and internalization of massive amounts of process knowledge, it can quickly and automatically generate customized, globally optimized electromagnetic force application strategies for any new part. This achieves a paradigm shift in process design from "experience-driven" to "data and model-driven," significantly improving the consistency of performance and the efficiency of process development for ultra-high strength steel hot-formed parts.

[0030] Specifically, obtaining the geometric feature parameters of the part to be formed includes: extracting the overall dimensions, cross-sectional shape change rate, minimum bending radius to sheet thickness ratio, and Gaussian curvature distribution of the feature region based on the three-dimensional CAD model of the part; the initial performance parameters of the sheet material include at least the grade, yield strength, tensile strength, elongation, and CCT curve features of the austenitizing transformation temperature range; the target performance parameters of the process include at least the tensile strength range of each zone of the target part after forming, the allowable tolerance of springback angle, and the upper limit of the thickness reduction rate of the key thinning area.

[0031] It should be further explained that the specific extraction process of geometric feature parameters is as follows: First, import the part's 3D model in STEP or IGES format. The overall dimensions refer to the length L, width W, and height H of the enclosing cuboid of the model. The rate of change of cross-sectional shape is calculated as follows: A series of cross-sections are taken at fixed intervals (e.g., intervals 1-2 times the sheet thickness t) along the main forming direction of the part (usually the stamping direction). The perimeter P_i and area A_i of each cross-section are calculated. The rate of change of shape between adjacent cross-sections is then defined as... Ultimately, take all The average and maximum values ​​are used as features. The minimum bending radius R_min is obtained by analyzing all fillet features of the model and calculating the R_min / t ratio.

[0032] Furthermore, regarding the Gaussian distribution of curvature in the feature region, it is necessary to first identify local surfaces on the part with significant curvature changes (such as the bottom of a groove or the top of a boss) using curvature analysis tools. Then, uniform sampling points should be taken within this region, and the Gaussian curvature of each sampling point should be calculated. ,in It is the first principal curvature. It is the second principal curvature, and finally the mean of the Gaussian curvature of all sampling points in this region is calculated. Standard deviation Including positive ratios (convex surface) and negative ratios (saddle surface). The CCT curve characteristics of the sheet metal need to be extracted from CCT diagrams provided by the material supplier or experimentally measured. Key features include: austenitizing temperature. (Typically 900-950℃), the temperature at which martensite begins to transform. Martensite transformation end temperature And to avoid the critical cooling time for the formation of proeutectoid ferrite and pearlite. (For example, from the CCT map) The time required to cool to 550°C is less than The “each zone” in the process target parameters needs to be pre-divided according to the structural function and stress of the part. For example, the B-pillar part of the car is divided into the upper connection zone, the middle collision energy absorption zone and the lower mounting zone, and different tensile strength target ranges are set for each zone (such as upper >1500MPa, middle 1300-1500MPa).

[0033] Understandably, this deep, quantitative feature extraction method transforms the vague "geometric complexity," "hardenability," and "performance non-uniformity requirements" of parts into precise numerical features that the model can understand. For example, the rate of change of cross-sectional shape quantifies the intensity of material flow; R_min / t directly relates to cracking risk; the Gaussian curvature distribution reveals the geometric characteristics of local three-dimensional surfaces, affecting the effect of magnetic fields and stress distribution; and CCT features determine the cooling conditions required to achieve a complete martensitic transformation. These features are closely integrated with the professional physics background of hot forming of ultra-high strength steel, giving the input information of the AI ​​model clear technological significance. This ensures the physical rationality and domain professionalism of the model's decisions, preventing it from becoming a "black box" prediction lacking a physical basis.

[0034] Specifically, the training data for the decision-making model comes from historical thermoforming process experiments or experimentally calibrated finite element simulation data; each training data sample includes: The input vector and its corresponding output vector label; the input vector includes geometric feature parameters, initial performance parameters of the sheet material, historical temperature data of the simulated or measured process, and preset target performance parameters; the output vector label is the optimal pulse electromagnetic force application control parameter obtained after process optimization under the input conditions.

[0035] It should be further explained that the construction of the training dataset is fundamental to the success of the model, and this is accomplished through the following two collaborative approaches: (1) Collection and labeling of historical experimental data: Collect past experimental records of pulsed electromagnetic force assisted thermoforming from enterprises or laboratories. For each experiment, extract its input vector: the CAD model of the part (for extracting geometric features), the material certificate of the sheet metal (for obtaining initial performance), the temperature-time curve of the sheet metal measured by thermocouples at multiple points (temperature history), and the design performance requirements of the part (target parameters). The determination of the output vector label is crucial: it is necessary to select the set of pulsed electromagnetic force control parameters with the best comprehensive evaluation from multiple process parameter combinations recorded in this experiment (which may be obtained through experimental design method) as the "best" label. The comprehensive evaluation needs to be combined with the final inspection report of the part (hardness, strength, springback, thinning rate, forming defects), for example, setting a scoring function: ,in It is the measured tensile strength. It is the target strength. It's the rebound amount. It is the thinning rate. As weight, To set the maximum allowed limit. Select to enable. The highest set of control parameters is used as the output label under this experimental condition.

[0036] (2) Generation of calibrated finite element simulation data: Establish a high-fidelity multiphysics coupled simulation model. This model must include calculations of heat conduction, electromagnetic field, structural mechanics, and phase transition. It is necessary to calibrate key parameters in the simulation model, such as the contact thermal conductivity coefficient between the mold and the sheet metal, the constitutive model parameters of the material at high strain rates, and the electromagnetic coupling coefficient, by measuring the final part shape (through 3D scanning), key point hardness, and springback angle through forming experiments under at least three sets of different process parameters (such as different pressures and different electromagnetic force parameters), ensuring that the error between the simulation prediction and the experimental results is within 10%. After calibration, large-scale parametric simulation is performed using this simulation model. By using methods such as Latin hypercube sampling, tens of thousands of combination points are uniformly sampled in the input parameter space (geometry, material, target) and control parameter space (time, space, energy), and the simulation is run to calculate the output performance corresponding to each combination. Similarly, according to the preset target performance parameters and scoring function, the control parameter combination that makes the highest score is found for each set of input conditions, and it is used as the output label of the simulation sample.

[0037] Understandably, by combining a small amount of high-cost experimental data with a large amount of low-cost, high-efficiency calibration simulation data, a large-scale (e.g., tens of thousands to hundreds of thousands of samples) high-quality dataset covering a broad process window can be constructed. This dataset not only contains the "input-output" mapping relationship, but more importantly, its output labels are the "optimal solution" or "satisfactory solution" generated through a clear optimization process directly linked to the final part quality. This allows the decision model to learn the deep patterns hidden behind a large number of process parameters regarding "how to obtain the best results under given conditions," rather than simply fitting the data. This learning paradigm based on optimization results is the core of the model's ability to make effective decisions.

[0038] Specifically, the simulated or measured historical temperature data of the process includes: The temperature cooling curve of the sheet metal from the heating furnace to the mold, the temperature field of the unsteady heat transfer process in the early stage of contact between the sheet metal and the mold, and the cooling rate curve of each part under the influence of the mold cooling pipe during the pressure holding and quenching stage; the temperature history data are used as input to the decision model in the form of time series or feature values.

[0039] It should be further explained that the acquisition and processing methods for historical temperature data are as follows: (a) Temperature-cooling curve during the transfer stage: During experimental measurement, K-type thermocouples were placed at different locations on the upper surface of the sheet metal at different distances from the edge and center, with a sampling frequency of 10Hz. The temperature change was recorded from the opening of the furnace door to the complete placement of the sheet metal onto the mold (usually 3-8 seconds). In simulation, this process can be simulated using thermal radiation and convection boundary conditions. Finally, the temperature-time curve of the center point of the sheet metal during this stage was extracted. And calculate its average cooling rate. ,in This refers to the furnace exit temperature (e.g., 930℃). The temperature at the instant before contact with the mold. For the transmission time. This average cooling rate is used as a key feature input to the model.

[0040] (b) Unsteady-state heat transfer temperature field in the initial contact phase: The temperature gradient is extremely large during this stage (0-0.5 seconds after mold closing). Experimental measurement is extremely difficult, and it mainly relies on calibrated simulations. From the simulation results, the temperature values ​​of the units on the sheet metal that first come into contact with the convex and concave fillets of the mold (usually located at the maximum drawing depth) are extracted at 0.1 seconds, 0.2 seconds, and 0.3 seconds after contact. Simultaneously, the maximum temperature difference between the sheet metal and the initial contact material (0-0.3 seconds) was calculated. These discrete temperature points and temperature differences serve as feature inputs to the model.

[0041] (c) Cooling rate curves for different zones during the holding-pressure quenching stage: Based on the part geometry and cooling pipe layout, the part is divided into 3-5 characteristic regions (e.g., flat zone, small rounded corner zone, large rounded corner zone, rib top, rib sidewall). Temperature-time curves of representative elements in each region during the holding-pressure stage (e.g., within 10 seconds) are obtained through simulation. From each curve, the most critical martensitic transformation zone is extracted (from...). Click Average cooling rate (point) For example, for region i, its cooling rate ,in and The temperature dropped to and At that moment. Ultimately, each region will be... and the area proportion of each region As a feature input model.

[0042] Understandably, extracting continuous temperature field data into the aforementioned feature values ​​with clear physical meaning, rather than directly inputting the massive original temperature sequence, greatly reduces the input dimensionality of the model, avoiding the "curse of dimensionality," while retaining the most critical thermal information affecting material phase transformation and plastic flow. For example, Influences surface oxidation and initial plasticity; the initial instantaneous temperature points and maximum temperature difference directly affect the material-mold bonding behavior and friction state; each zone... This directly determines the final martensite content and hardness of the region. The model learns the relationship between these characteristics and control parameters, which is essentially learning how to "compensate" or "optimize" the application strategy of electromagnetic force according to different thermal histories. For example, applying different modes of electromagnetic force in areas that cool rapidly to equalize the microstructure, or applying electromagnetic force to supplement plasticity when the temperature drops sharply at the initial stage of contact.

[0043] Specifically, the decision model is a machine learning model, specifically a deep neural network, a gradient boosting decision tree, or a support vector regression model; the model learns the nonlinear mapping relationship between the input vector and the output vector labels to establish a predictive mapping from process conditions to optimized control parameters.

[0044] It should be further noted that this implementation preferably uses a deep neural network (DNN) model due to its powerful representation capability for high-dimensional nonlinear mappings. The specific model construction, training, and hyperparameter settings are as follows: (1) Model Architecture: Construct a deep neural network with 5 fully connected layers. Number of nodes in the input layer The number of output layer nodes depends on the dimension of the feature vector after feature engineering; let's assume it's 120 dimensions. To control the total dimension of the parameters, we assume it to be 45 dimensions (e.g., 20 time-point states + 15 coil states and phases + 10 energy parameters). The hidden layer structure is as follows: the first hidden layer has 512 nodes and uses the ReLU activation function; the second hidden layer has 256 nodes and uses the ReLU activation function; the third hidden layer has 128 nodes and uses the ReLU activation function; and the fourth hidden layer has 64 nodes and uses the ReLU activation function. The output layer uses different activation functions depending on the output type: for binary outputs in the temporal scheme (applied / unapplied), the Sigmoid activation function is used, mapping the output value to the [0,1] interval. During training, a binary cross-entropy loss is used, and during inference, binarization is performed with a threshold of 0.5; for coil activation states (binary) in the spatial pattern, the Sigmoid function is also used; for continuous output values ​​such as phase difference and energy parameters, a linear activation function is used.

[0045] (2) Model Training: The total dataset was randomly divided into training, validation, and test sets in a ratio of 70%:15%:15%. The Adam optimizer was used, with its hyperparameters set as follows: initial learning rate... The exponential decay rate estimated by the first moment The exponential decay rate estimated by the second moment Batch gradient descent is employed, with a batch size of 64. The loss function used is the multi-objective weighted loss function defined in this invention. During training, the loss is calculated on the validation set after each epoch (one iteration of the training set). A learning rate decay strategy is used: if the validation set loss does not decrease for 5 consecutive epochs, the learning rate is multiplied by 0.5. An early stopping strategy is used: if the validation set loss does not decrease for 15 consecutive epochs, training is stopped and the model parameters with the lowest validation set loss are rolled back. The maximum number of total training epochs is set to 1000.

[0046] (3) Reasons for hyperparameter selection: The hidden layer structure of 512, 256, 128, and 64 was chosen to construct a funnel-shaped network that gradually extracts and compresses features. This structure has been proven effective in learning complex patterns in practice. An initial learning rate of 0.001 is a robust starting point commonly used by the Adam optimizer, ensuring fast convergence while minimizing oscillations. A batch size of 64 represents a good trade-off between computational efficiency and gradient estimation stability. The patience value for early stopping is set to 15 to give the model enough time to "struggle" and escape local optima while avoiding meaningless overfitting.

[0047] Understandably, employing deep neural networks as the decision-making core can automatically extract complex, nonlinear mapping relationships from massive, high-dimensional process data without requiring manual specification of complex physical formulas or empirical rules. Once trained, the model can instantly (forward propagation calculations are completed within milliseconds) output decision schemes for new parts and process conditions. Compared to traditional optimization methods based on finite element simulation (which may require hours or even days of computation), this method improves online decision-making speed by several orders of magnitude, enabling real-time, online personalized process adjustments and laying the algorithmic foundation for realizing "digital twin"-driven intelligent thermoforming.

[0048] Specifically, the input vector of the model needs to undergo feature engineering processing before being input into the model. The processing includes: dimensionless processing of geometric feature parameters, piecewise integration of temperature history data to extract cumulative thermal effect features, extraction of key phase transition point temperature and time from the CCT curve features in the initial performance parameters of the plate, and normalization of all features before concatenating them into a unified feature vector.

[0049] It should be further explained that the specific steps and calculation formulas for feature engineering are as follows: (1) Dimensionless geometric feature: For the original dimensions extracted from the CAD model, divide by the nominal thickness of the sheet metal. For example, part length Convert to length-to-thickness ratio Minimum bending radius Transform into Rate of change of cross-sectional shape It is a ratio and requires no processing. The mean of a Gaussian distribution of curvature. and standard deviation Its dimensions are Multiply it by Dimensionless transformation yields and This decouples geometric features from the absolute dimensions of the parts, making it easier for the model to learn the similarity patterns of the shapes themselves.

[0050] (2) Extraction of features from historical temperature data: extracting features from temperature curves during the transmission phase. Calculate the cumulative thermal effect within the temperature range [Ar3, Ms]. Ar3 is the austenitizing critical temperature, and Ms is the martensite initiation temperature. The Ms point temperature can be taken. This integral value reflects the total thermal history experienced by the sheet metal in the critical temperature range before the phase change, containing more information than a single average cooling rate. For the temperature field at the initial contact stage, the average temperature of the entire sheet metal within 0.3 seconds after contact is calculated. and temperature standard deviation For the cooling rate during the pressure holding phase, the calculations for each zone are used directly. .

[0051] (3) CCT curve feature extraction: Read the values ​​directly from the CCT curve: martensite initiation temperature Martensitic transformation end temperature The critical cooling time for avoiding proeutectoid ferrite and pearlite. It is generally defined as the time required to cool from the austenitizing temperature to 550°C must be less than This value can be read directly from the curve or calculated through interpolation. For some advanced high-strength steels, it is also necessary to extract the bainite transformation start temperature. .

[0052] (4) Feature normalization: After the above processing, an initial feature vector containing all geometric, material, temperature and target features is obtained. Then, Z-score standardization is performed on each feature: .in, and These are features The mean and standard deviation are calculated over the entire training set. The standardized feature vectors. With a mean of 0 and a variance of 1, the data is fed into the neural network for training. Note that when applying the model online, new input data must be obtained from the training set. and Perform the same standardization transformation.

[0053] Understandably, this systematic feature engineering is a crucial step in injecting domain knowledge into AI models. Dimensionlessness enables the model to possess scale invariance, allowing it to generalize to parts of different sizes. Extracting integral features such as cumulative thermal effects from temperature curves captures the "cumulative" effect of the thermal process, making it more stable and representative than instantaneous values. Key phase transition points extracted from CCT curves directly input the core thermodynamic properties of the material into the model. Finally, Z-score normalization is a common and necessary practice for training stable deep neural networks. It ensures that all features are within a similar numerical range, preventing certain large numerical features (such as part length) from dominating gradient updates while small numerical features (such as curvature) are ignored, thereby accelerating convergence and improving the overall performance of the model.

[0054] Specifically, the training process of the decision model includes multi-objective optimization. Its loss function is configured to simultaneously minimize the errors in predicting temporal schemes, spatial action modes, and energy intensity schemes in the output vector labels. Multiple sub-losses are combined into a total loss through weighted summation for backpropagation and updating of model parameters.

[0055] It should be further explained that the multi-objective loss function The specific composition, calculation method of each sub-loss, and weight settings are as follows: The total loss function is defined as: in, , , These are the weights for temporal, spatial, and energy losses, respectively, and satisfy the following conditions: .

[0056] (1) Sub-loss of the timing scheme The timing scheme outputs a length of The sequence is given, where each position represents the probability of applying a pulse (Sigmoid output). This is a multi-label binary classification problem. A weighted binary cross-entropy loss is used: in, It is the true label (1 or 0) at the i-th time step. It is the probability predicted by the model. This is the total number of time steps (e.g., 20 seconds / 0.1 seconds = 200 steps). Since the timing of pulse application is sparse throughout the entire process cycle (most of the time is 0), different weights are assigned to positive and negative samples. and This helps resolve the class imbalance problem. A preferred setting is... , Give positive samples (apply pulses) slightly higher weights to encourage the model not to miss key application opportunities.

[0057] (2) Spatial Action Mode Sub-loss Spatial mode loss consists of two parts: coil activation state loss. and phase difference loss .

[0058] in, It is the balance coefficient, preferably 0.2. It also uses a weighted binary cross-entropy loss, and the calculation method is the same as... Similarly, targeting Each of the coils. This is the mean squared error loss for the phase difference prediction of the activated coils (true label 1). Assume that for the j-th activated coil, its true phase difference is... (Normalized to [0,1], corresponding to 0 to [1]) The model's predicted value is ,but in, This is the total number of activated coils in the sample.

[0059] (3) Energy intensity scheme sub-loss The energy parameters are continuous values, and a smooth L1 loss is used, which is less sensitive to outliers than MSE. For the k-th energy parameter, its true value is... The predicted value is ,but in, It is the total number of energy parameters.

[0060] (4) Reasons for weight selection: weight , , This reflects the relative importance of different control dimensions in the process. After multiple experimental verifications, an optimal setting is: , , This is because, in pulsed electromagnetic force-assisted thermoforming, the "timing" (whether it is applied when the material's plasticity is optimal or at a critical phase transformation point) and "location" (whether it is applied to areas that are difficult to fill or require reinforcement) are the primary factors determining the success or failure of the process. While fine-tuning the energy level is important, its effective window is relatively wide, and its optimal value depends on the correct timing and location. Therefore, assigning higher weights to temporal and spatial predictions in the loss function guides the model to prioritize learning these two aspects, achieving better overall control results in practice.

[0061] Understandably, this multi-objective loss function design essentially embeds the multi-objective trade-offs (timing, location, energy) from process optimization into the model's training objectives. During training, the model continuously adjusts its internal parameters through backpropagation and gradient descent to minimize this comprehensive loss. This means the model learns not to optimize any single aspect, but rather to find a balance among the three, ensuring that the overall set of control parameters output is closest to the "optimal" solutions validated in the training data. This guarantees the consistency and usability of the model's decision results, avoiding unreasonable outputs such as "correct timing but completely wrong location" or "accurate energy but missed timing."

[0062] Specifically, the timing scheme for applying the pulsed electromagnetic force is output by the decision model as a discrete-time sequence activation signal corresponding to the thermoforming process timeline; this sequence defines the specific start time and duration of applying the pulsed electromagnetic force in at least one of the following stages: sheet material transfer stage, initial contact with the mold, mold closing and loading stage, and pressure holding and quenching stage.

[0063] It should be further explained that the specific output format and application examples of the timing scheme are as follows: The model outputs a string of length... sequence ,in Temporal resolution of the sequence The timeout is typically set to 0.05 seconds or 0.1 seconds to balance the requirements of control precision and output dimensionality. The entire sequence covers the entire process from the sheet metal exiting the furnace (t=0) to the end of pressure holding and quenching (t=T_end, e.g., T_end=20 seconds). The sequence output by the model directly corresponds to the total enable signal of the pulsed electromagnetic force main power supply. For example, for the hot forming of a car door anti-collision beam (material 22MnB5, sheet thickness 1.5mm), the timing scheme output by the model might be analyzed as follows: the sequence is 0 for most of the time. There are three continuous intervals (pulse windows) with a value of 1: (a) First pulse window: corresponding to indices i from 60 to 65, i.e., absolute time t from 3.0 seconds to 3.25 seconds (assuming) This time point is at the end of the air stage where the sheet metal is transferred from the heating furnace to above the mold. At this time, the sheet metal temperature is approximately between 750℃ and 800℃, still in the austenitic region and with good plasticity. Applying a series of low-energy pulses (the specific energy is determined by the energy scheme) during this window aims to utilize the high-frequency mechanical vibration effect of electromagnetic force to cause the oxide scale on the surface of the sheet metal to vibrate and fall off, and to stimulate dislocation movement within the material, thus reserving a certain amount of "pre-plasticity" for the upcoming forming. At the same time, the slight vibration also helps the sheet metal to be placed more evenly on the mold.

[0064] (b) Second pulse window: corresponding to indices i from 102 to 110, i.e., absolute time t from 5.1 seconds to 5.5 seconds. This window precisely covers the initial stage when the upper die and sheet metal just begin to contact and continue to press down. At this time, the temperature of the part of the sheet metal in contact with the cold die drops sharply, and the fluidity deteriorates rapidly, making it a high-risk period for cracking and incomplete filling. Applying a pulse of medium energy in this window generates a strong Lorentz force (J×B force) that can provide the material with an additional driving force to "fill" deep into the die cavity, effectively compensating for the increase in flow stress caused by the temperature drop, and significantly improving the filling integrity of complex features (such as sharp fillets and ribs).

[0065] (c) The third pulse window: corresponding to indices i from 160 to 180, i.e., absolute time t from 8.0 seconds to 9.0 seconds. This window is located in the early to middle stage of the mold-holding quenching process. At this time, through mold cooling, the part temperature has dropped to near the martensite initiation transformation temperature Ms (approximately 380°C). Applying a series of high-energy, appropriately frequency pulses within this window can generate micro-area temperature rise due to Joule heating, delaying local cooling. Meanwhile, the strong magnetostrictive and Lorentz force impact effects can disrupt the lattice arrangement and increase nucleation sites in the early stages of martensite nucleation and growth, thereby significantly refining the final martensite lath bundle size. This can improve the strength and toughness of the part without changing the cooling medium.

[0066] Understandably, the multi-stage, precisely timed sequencing scheme determined by the model perfectly aligns with the different physical mechanisms required in the hot forming process of ultra-high strength steel. It no longer simply involves continuously or randomly applying pulses throughout the process, but rather uses pulse energy as a "surgical" intervention tool. During the "late transport stage," when the material has high plasticity and good flowability, it performs pretreatment to prevent oxidation and pre-plasticize; during the "early contact stage," when material flow resistance increases sharply, it provides strong assistance to maintain shape; and during the "early quenching stage," when phase transformation begins, it intervenes with impact to modify the microstructure. This sequencing strategy, deeply coupled with the physical state of the process, maximizes the unique role of pulsed electromagnetic force at different stages, achieving the superposition and optimization of process effects.

[0067] Specifically, the spatial action mode corresponds to the collaborative working strategy of multiple independent control coils in the pulsed electromagnetic force generator; the spatial action mode parameters output by the decision model include the combination of coil numbers to be activated, the relative current phase difference of each activated coil, and the spatial superposition distribution description of the action area of ​​the pulsed magnetic field generated by each coil on the board, which are used to form a gradient magnetic field or a uniformly enhanced magnetic field in a specific area of ​​the board.

[0068] It should be further explained that the spatial action mode controls the spatial distribution of electromagnetic force on the plate. Assume the pulsed electromagnetic force generator consists of an upper and lower coil array, each coil being independently controllable. The spatial mode parameters output by the model specifically include an activation mask vector, a phase difference vector, and a current amplitude scaling vector.

[0069] (1) Activate coil number combination: The model outputs a value with a length equal to the total number of coils. binary vector ,in This indicates that coil j needs to be activated at the current time. This indicates that the coil is not activated. For example, for a system with 12 coils (6 on top and 6 on the bottom), the model might output: This indicates that coils 2, 3, and 5 of the upper mold and coils 9 and 11 of the lower mold are activated. The coil numbers and physical positions are predefined; for example, coils 1-6 correspond to the left-to-right and front-to-back sections of the upper mold, and coils 7-12 correspond to the lower mold.

[0070] (2) Relative current phase difference of each activated coil: For all activated coils (i.e. (The coil), the model outputs a phase value. This corresponds to the phase shift of the current waveform (multiplied by) (Obtaining radians). The phase difference determines the relative timing relationship of the pulsed magnetic fields generated by each coil. For example, if coil 2... coil 3 coil 5 This means that the current pulses of the three coils are sequentially out of phase by 1 / 4 of a cycle. Coils in phase (phase difference of 0) generate magnetic fields that are superimposed, enhancing the overall field strength in that region; coils out of phase (phase difference of 0.5) generate magnetic fields in opposite directions in adjacent regions, thus creating a strong magnetic field gradient. A gradient magnetic field will generate a gradient force. (in (where is the magnetic susceptibility). This force points in the direction of increasing magnetic field strength and can be used to actively "pump" material from a weak magnetic field region to a strong magnetic field region. It is particularly effective for supplementing local material deficiencies (such as at the root of draw beads).

[0071] (3) Description of the spatial superposition distribution of the action area: This is achieved by assigning a current amplitude scaling factor to each active coil. To achieve this. Assume the system's maximum allowable peak current is... Then the peak current actually output by coil j is The model outputs different... This is used to control the distribution of the spatial magnetic field strength. For example, for a localized protrusion on a part that requires special reinforcement, the coils directly above and below it (such as coil 2 and coil 9) can be instructed to operate at a high ratio (e.g., ... ) works, while the auxiliary coils around it operate at a lower ratio (such as This process creates a focused, strong magnetic field region at the protrusion. Combined with phase control, a rotating magnetic field or a traveling wave magnetic field can also be generated to induce specific flow patterns in the material.

[0072] Understandably, this independent and coordinated control of multiple coils endows this method with unprecedented spatial modulation capabilities. It can dynamically "draw" the required magnetic field distribution map based on the real-time requirements of the part's three-dimensional shape. For example, when forming a complex part with a deep cavity and reinforcing ribs, the model can decide to activate multiple in-phase, high-intensity coils in the bottom region of the deep cavity to generate a uniform, strong magnetic field, providing a powerful "ejection force" to assist in filling; simultaneously, it can activate a set of alternating-phase coils in the sidewall region of the ribs to generate a gradient magnetic field, guiding the material from the flat area to the ribs and preventing excessive local thinning. This "on-demand allocation and precise force application" spatial mode enables pulsed electromagnetic force technology to proactively adapt to and compensate for the non-uniformity of material flow caused by the geometric complexity of the part, which is key to achieving high-quality forming of complex ultra-high-strength steel parts.

[0073] Specifically, the energy intensity scheme is output by the decision model as a set of pulse energy parameters that match the timing scheme and spatial action mode; for each activated coil, the set of pulse energy parameters independently defines the peak current density, pulse frequency, and waveform envelope characteristics of the applied pulse current during one or more activation periods.

[0074] It should be further explained that the energy intensity scheme defines specific electrical parameters for each activated coil during each activation period. These parameters, together with the physical characteristics of the coil (such as the number of turns and cross-sectional area), determine the magnitude and shape of the electromagnetic force acting on the plate. The model outputs a three-dimensional parameter table, the dimensions of which are [number of activation periods] × [number of activated coils] × [number of energy parameters].

[0075] (1) Peak current density This is the most crucial energy parameter, measured in A / mm². It directly determines the amplitude of the current inside the coil, and consequently, the strength of the generated magnetic field. and Lorentz force density The model makes a decision for each coil in each time period based on the objective. Values. For example, a higher value might be required for the "initial contact with assisted flow" window. (e.g., 300-500 A / mm²) to generate sufficient driving force for plastic deformation; for the "refinement of microstructure during quenching" window, medium but high frequency may be required. (e.g., 150-250 A / mm²) to generate a sustained impact effect; for the "transfer stage preprocessing" window, a lower value can be used. (e.g., 50-100A / mm²).

[0076] (2) Pulse frequency The unit is Hz, which determines the number of pulses applied per second. The choice of frequency is closely related to the target physical mechanism. Low-frequency (1-20Hz), high-energy single large pulses have a strong impact effect and are suitable for providing a large "push" at a specific moment (such as the moment of contact). Medium- and high-frequency (20-200Hz) pulse trains can generate continuous vibration or vibration effects, which are beneficial for promoting dynamic recrystallization of materials, grain breakage, or stress homogenization. For example, in the phase transformation refinement stage, a frequency of 50-100Hz is often chosen to apply multiple impacts in a short period of time. The model will decide on a frequency for each coil-time combination based on the duration of the timing window and the target.

[0077] (3) Waveform envelope characteristics of a single pulse: This defines the shape of the single pulse current as a function of time. Common waveforms include: a) Rectangular wave: rises quickly, has concentrated energy, and produces a large electromagnetic impact, but requires high-performance power supply and coil.

[0078] b) Attenuated sine wave (damped oscillating wave): generated by capacitor discharge, the frequency is determined by the LC circuit, the pulse width is short and the peak value is high, and it is suitable for occasions that require instantaneous strong impact.

[0079] c) Trapezoidal wave: Features a rising edge, a flat top, and a falling edge. The force application rate can be adjusted by controlling the rise / fall time, resulting in a smoother and more controllable force. The model can output waveform type encoding, as well as key parameters such as rise time. Flat top duration descent time For example, for assisted flow, the model might output a "trapezoidal wave." , , This allows for smooth loading and unloading of forces, avoiding impact damage to the material; for microstructure refinement, it may output a "damped sine wave with an oscillation frequency..." decay time constant ", to generate high-frequency oscillation impact.

[0080] A complete energy intensity scheme example is as follows: During the timing window B (t=5.1-5.5s), for the activated upper mode coils 2, 3, and 5: ① Coil 2 (located directly above the center protrusion of the part): , The waveform is a trapezoidal wave ( ), phase .

[0081] ② Coil 3 (located at the raised edge): , The waveform is a trapezoidal wave (same as coil 2), with phase... .

[0082] ③ Coil 5 (located on the other side of the protrusion): , The waveform is a trapezoidal wave (same as coil 2), with phase... This setup aims to create a strong magnetic field region with superimposed phase at the center, providing the primary forming force; and to create a secondary strong magnetic field with a certain phase difference at the edges, guiding the material to flow from both sides towards the center.

[0083] Understandably, the refined and differentiated control of energy intensity schemes allows the "dosage" of pulsed electromagnetic force to be perfectly matched with the spatiotemporal distribution. The model not only decides "when and where" to apply force, but also "how much force and how" to apply it. By independently adjusting the current density, frequency, and waveform of each coil, optimal energy allocation can be achieved, applying high energy at locations and times requiring strong action, and using lower energy at less important locations or times. This not only maximizes the efficiency of electromagnetic energy use and reduces overall energy consumption, but more importantly, it allows for extremely precise "programming" of the material's stress and heat history. This enables more accurate control of the material's plastic deformation path and phase transformation process, providing unprecedented technological means to achieve "customization" and "functional gradient" of part performance.

[0084] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0085] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0086] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0087] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0088] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel, characterized in that, Includes the following steps: S1. Obtain the geometric feature parameters of the part to be formed, the initial performance parameters of the sheet metal, and the target performance parameters of the process. S2. Input the parameters into the trained decision model, and the decision model outputs control parameters for applying pulsed electromagnetic force to the thermoforming process; S3. The control parameters include the timing scheme, spatial action mode, and energy intensity scheme of the pulse electromagnetic force application, which are used to control the pulse electromagnetic force generating device to perform corresponding operations during the thermoforming process.

2. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 1, characterized in that, The process of obtaining the geometric feature parameters of the part to be formed includes: Based on the 3D CAD model of the part, the overall dimensions, cross-sectional shape change rate, minimum bending radius to sheet thickness ratio, and curvature Gaussian distribution of the feature region are extracted. The initial performance parameters of the sheet material include at least the grade, yield strength, tensile strength, elongation, and CCT curve characteristics of the austenitizing transformation temperature range. The target performance parameters of the process include at least the tensile strength range of each zone of the target part after forming, the allowable tolerance of springback angle, and the upper limit of the thickness reduction rate of the key thinning area.

3. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 1 or 2, characterized in that, The training data for the decision model comes from historical thermoforming process experiments or experimentally calibrated finite element simulation data. Each training data sample includes: an input vector and a corresponding output vector label; the input vector includes geometric feature parameters, initial performance parameters of the sheet material, historical temperature data of the simulated or measured process, and preset target performance parameters; the output vector label is the optimal pulse electromagnetic force application control parameter obtained after process optimization under the input conditions.

4. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 3, characterized in that, The simulated or measured historical temperature data of the process includes: the temperature cooling curve of the plate from the heating furnace to the mold, the temperature field of the unsteady heat transfer process in the early stage of contact between the plate and the mold, and the cooling rate curve of each part under the influence of the mold cooling pipe during the pressure holding and quenching stage; the historical temperature data is used as the input of the decision model in the form of time series or feature values.

5. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 4, characterized in that, The decision model is a machine learning model, specifically a deep neural network, a gradient boosting decision tree, or a support vector regression model; the model learns the nonlinear mapping relationship between the input vector and the output vector labels to establish a predictive mapping from process conditions to optimized control parameters.

6. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 5, characterized in that, The input vector of the model needs to undergo feature engineering before being input into the model. The processing includes: The geometric feature parameters are dimensionless, the temperature history data is piecewise integrated to extract the cumulative thermal effect features, the CCT curve features in the initial performance parameters of the plate are extracted for key phase transition point temperature and time, and all features are normalized and spliced ​​into a unified feature vector.

7. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 6, characterized in that, The training process of the decision model includes multi-objective optimization. Its loss function is configured to simultaneously minimize the errors in predicting temporal schemes, spatial action modes, and energy intensity schemes in the output vector labels. Multiple sub-losses are combined into a total loss through weighted summation for backpropagation and updating of model parameters.

8. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to any one of claims 5-7, characterized in that, The timing scheme for applying the pulsed electromagnetic force is output by the decision model as a discrete time sequence activation signal corresponding to the thermoforming process timeline. This sequence defines the specific start time and duration of applying the pulsed electromagnetic force in at least one of the following stages: sheet material transfer stage, initial contact with the mold, mold closing and loading stage, and pressure holding and quenching stage.

9. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 8, characterized in that, The spatial action mode corresponds to the collaborative working strategy of multiple independent control coils in the pulse electromagnetic force generator; the spatial action mode parameters output by the decision model include the combination of coil numbers to be activated, the relative current phase difference of each activated coil, and the spatial superposition distribution description of the action area of ​​the pulse magnetic field generated by each coil on the board, which are used to form a gradient magnetic field or a uniformly enhanced magnetic field in a specific area of ​​the board.

10. The pulsed electromagnetic force application mode decision method for hot forming of ultra-high strength steel according to claim 9, characterized in that, The energy intensity scheme is output by the decision model as a set of pulse energy parameters that match the temporal scheme and spatial action mode; The pulse energy parameter set independently defines, for each activated coil, the peak current density, pulse frequency, and waveform envelope characteristics of the applied pulse current during one or more activation periods.

Images