An intelligent stamping forming optimization method for an ultra-high strength steel roof cross beam

Abstract

The application discloses an intelligent stamping forming optimization method for an ultrahigh-strength steel roof cross beam, and belongs to the technical field of ultrahigh-strength steel stamping forming. The method comprises the following steps: in response to a stamping starting instruction, an initial stamping process chain containing space-time axis stamping nodes is constructed based on the geometric topology of the roof cross beam and batch material characteristics; when a trial stamping is performed at the first stamping node, stress wave propagation trajectories and die surface temperature distributions are collected; the two are mapped to a high-dimensional feature space to generate a plastic response feature tensor; the tensor is topologically aligned with a pre-constructed reference state grid and difference analysis is performed, and a stamping parameter field correction vector is output; based on the correction vector, closed-loop reconstruction is performed on all unexecuted nodes after the current node in the initial process chain, and an optimized stamping process sequence is generated. The method realizes real-time dynamic optimization of the ultrahigh-strength steel stamping process, and solves the problem of unstable forming quality caused by batch material fluctuations and die state changes.

Description

Technical Field

[0001] This invention relates to the field of ultra-high strength steel stamping technology, specifically to an optimized method for intelligent stamping of ultra-high strength steel top cover beams. Background Technology

[0002] Under the trend of lightweight automotive manufacturing, the stamping of ultra-high strength steel roof beams faces severe challenges. Ultra-high strength steel has a narrow plasticity window, and its forming quality is extremely sensitive to fluctuations in process parameters. In actual production, even using the same set of preset process parameters, differences in mechanical properties between different batches of material can lead to defects such as excessive springback and localized cracking in the formed parts. Existing stamping process optimization generally relies on offline simulation or trial-and-error adjustment modes. That is, before formal mass production, a set of fixed parameters is iteratively calculated using finite element software, and then locked after several trial stampings for fine-tuning before mass production. This open-loop control method with fixed parameters cannot perceive real-time changes in material properties during the stamping process, nor can it dynamically compensate for process parameters.

[0003] Offline simulation suffers from fundamental flaws. Its model accuracy is highly dependent on the accuracy of the constitutive equations, and the hardening behavior and damage evolution mechanisms of ultra-high-strength steel under complex stress states are difficult to describe precisely, leading to a persistent discrepancy between simulation results and actual forming conditions. Online adjustment modes, based on trial and error, heavily rely on operator experience and judgment, resulting in delayed response and limited adjustment accuracy. When batch material fluctuations occur, the process from defect detection to parameter correction often involves the generation of numerous scrapped parts. Furthermore, the stamping process is a continuous spatiotemporal evolution; the stress state experienced by the sheet metal changes dynamically at different closing stages of a single stamping stroke, and uniform fixed parameters cannot adapt to this dynamic characteristic.

[0004] More importantly, during continuous production, the die surface inevitably wears due to constant friction, leading to gradual changes in die clearance and surface friction coefficient. Existing optimization methods lack effective means to quantify the impact of die wear and cannot incorporate the real-time state of the die into the parameter correction system. Furthermore, the adjustment of stamping process parameters at each node of the process chain exhibits a coupling effect; parameter corrections at preceding nodes propagate along the process chain and produce attenuating effects. How to handle the coordinated correction of parameters at multiple nodes constitutes another technical challenge. Summary of the Invention

[0005] This paper presents an intelligent stamping forming optimization method for ultra-high strength steel top cover beams. During the stamping process of a single part, the method realizes online dynamic reconstruction of process parameters, automatically adapts to batch material performance fluctuations, and establishes a propagation and attenuation mechanism of stamping parameter correction along the process chain node sequence by real-time sensing of mold wear status and quantifying its modulation effect on parameter correction. This enables multi-node parameter collaborative optimization from the current node to the last node, maintaining the consistency of forming quality throughout the process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an intelligent stamping forming optimization method for ultra-high strength steel top cover beams, comprising: responding to a stamping start command, acquiring the geometric topology of the top cover beam and the batch material characteristics of the ultra-high strength steel; based on the geometric topology and the batch material characteristics, constructing an initial stamping process chain containing a spatiotemporal axis stamping node; when performing a trial stamping at the first stamping node of the initial stamping process chain, collecting multiple physical signal field data corresponding to that node, the physical signal field data including stress wave propagation trajectory and die surface temperature distribution; and combining the stress wave propagation trajectory and... The temperature distribution of the mold surface is mapped to a high-dimensional feature space, generating a plastic response feature tensor describing the current stamping state. The plastic response feature tensor is topologically aligned and difference-analyzed with a pre-constructed reference state grid, outputting a stamping parameter field correction vector. The reference state grid contains ideal plastic response features recorded in multiple historical stamping events and their corresponding stamping parameter adjustment records. Based on the stamping parameter field correction vector, closed-loop reconstruction is performed on the process parameters of all unexecuted stamping nodes after the current stamping node in the initial stamping process chain, generating an optimized stamping process sequence.

[0007] By discretizing the stamping stroke into a series of spatiotemporal axis stamping nodes and constructing a dynamically reconfigurable process chain, the previously experience-dependent overall mold trial is transformed into an adaptive optimization process based on local physical field feedback, greatly reducing the reliance on operator experience. At each node, only two types of highly sensitive physical field information need to be collected: stress wave propagation trajectory and die surface temperature distribution. This allows for the perception of deep plastic response characteristics such as material yield flow and interfacial friction state, eliminating the need for destructive sampling or offline testing and effectively shortening the debugging cycle. By fusing stress waves and temperature fields into a high-dimensional plastic response feature tensor, both forming energy changes and thermo-mechanical coupling effects can be captured simultaneously, achieving accurate characterization of the difficult-to-form properties of ultra-high-strength steel. Through topological alignment and difference analysis with a benchmark state grid composed of historical qualified manufacturing data, abstract feature deviations are directly mapped into executable stamping parameter field correction vectors, ensuring that every parameter adjustment is supported by sufficient process data and avoiding die damage and material waste caused by blind parameter tuning.

[0008] Preferably, when constructing the initial stamping process chain, the stamping stroke is divided into multiple continuous, equally spaced stroke segments according to the die closing distance. Each stroke segment corresponds to a spatiotemporal axis stamping node. An initial process parameter vector containing stamping speed, blank holder force, and die clearance values ​​is configured for each node, and all nodes are linearly linked in the order of stroke sequence. The stamping speed is configured to decrease proportionally from the first node to the last node. This decreasing gradient can naturally adapt to the change in the hardening rate of the material in the later stage of forming, providing a relatively smooth initial process reference for subsequent local corrections.

[0009] As a technical solution of this invention, when acquiring physical signal field data, multiple piezoelectric sensors and thermocouple sensors are pre-embedded on the punch and die of the stamping die. Acquisition is triggered synchronously when the stamping slide moves to the stroke position corresponding to the current node. The time-domain signal acquired by the piezoelectric sensors is reconstructed into a stress wave propagation trajectory, and the temperature values ​​acquired by the thermocouple sensors are interpolated to generate the temperature distribution of the die surface. The contour-following grid-like staggered arrangement of the sensors and the adaptive increase in sampling frequency as the stamping stroke progresses can, while ensuring data completeness, focus on capturing detailed features within the stroke segment where the forming force changes rapidly, providing high-resolution input signals for high-precision state identification.

[0010] Furthermore, when generating the plastic response characteristic tensor, a time-frequency domain transformation is performed on the stress wave propagation trajectory to obtain the attenuation spectrum of the stress wave energy on the frequency axis, and a gradient calculation is performed on the temperature distribution of the mold surface to obtain the temperature gradient vector field. These two are then normalized and fused into a three-dimensional tensor structure, with its three dimensions corresponding to spatial coordinates, the time axis, and the physical parameter type, respectively. This structured fusion method unifies the internal damage evolution of the material and the interfacial heat conduction behavior under the same representation framework. The coupling relationship between the components of the tensor directly reflects the multi-physics interaction during the stamping process, thereby significantly improving the sensitivity and specificity of state identification.

[0011] In constructing the benchmark state grid, this invention collects all historical plastic response feature tensors of qualified products from historical production batches. Each feature tensor is labeled with its corresponding historical stamping parameter vector, and clustered according to their distance from the features of standard qualified products. The cluster centers are used as benchmark state nodes, and the average value of all historical stamping parameter vectors within a cluster is stored as the associated correction benchmark value. This benchmark grid, formed based on manufacturing data bootstrapping, can continuously learn multivariate expressions of the ideal stamping state from historical experience, avoiding the shortcomings of traditional methods that rely on single samples or theoretical models to define standards, making the benchmark more statistically representative and robust.

[0012] The specific methods for topology alignment and difference resolution are as follows: The similarity values ​​between the current plastic response feature tensor and each node in the reference state grid are calculated. The node with the highest similarity is selected as the matching reference node. The deviation vector between their feature values ​​is calculated, and this deviation vector is directly used as the basis for the stamping parameter field correction vector. To make the correction amount more consistent with the actual equipment state, the deviation vector is input into a pre-trained parameter adjustment mapping model to obtain the initial correction component. A real-time wear factor calculated based on the cumulative number of stampings and the die surface hardness decay curve is obtained. This wear factor is used to weight and modulate the initial correction component to obtain the final stamping parameter field correction vector. The introduction of model mapping and wear compensation enables the same feature deviation to generate differentiated adjustment amounts that conform to the mechanical state at different die life stages, preventing the die from becoming unstable due to over-adjustment in the later stages.

[0013] In the closed-loop reconstruction, the initial process parameter vector of the first unexecuted stamping node after the current node is extracted. The stamping parameter field correction vector is then superimposed element-wise onto this vector to generate the first optimized process parameter vector, which is then used as a replacement. To simulate the propagation and attenuation of the stamping adjustment effect along the process, the plastic response feature tensor and the stamping parameter field correction vector are convolved to generate a propagation attenuation coefficient. This coefficient is then proportionally reduced and sequentially superimposed onto the initial process parameter vectors of the second and subsequent unexecuted nodes, thus obtaining the optimized stamping process sequence. Through this attenuation propagation mechanism, process nodes farther from the current adjustment point receive smaller correction amplitudes, which conforms to the physical law that the impact of local parameter adjustments on downstream strokes gradually weakens in stamping forming, ensuring the smoothness and consistency of the global process sequence.

[0014] As a preferred embodiment of the present invention, after optimizing the process sequence, dynamic self-verification of stamping parameters is also performed: during the execution of the optimized sequence, the actual stamping load of the current node is monitored in real time. Once it is found that the actual stamping load exceeds the theoretical load threshold of the corresponding node, the stamping process is immediately paused, and the operation of collecting physical signal field data is jumped to the current node as the new starting point. This online protection mechanism can detect process deviations in real time, automatically trigger a new round of optimization iteration, form a fully closed-loop intelligent stamping control process, suppress the fluctuation of forming quality within a very small range, thereby significantly improving the first-pass yield and production cycle stability of the top cover beam.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: During a single stamping stroke, stress wave propagation trajectories and die surface temperature distributions are acquired in real time via embedded sensors and fused into a high-dimensional tensor containing the current material plastic flow characteristics and interface state. After aligning this tensor with the baseline state of historically qualified parts, the calculated deviation vector directly reflects the direction and extent of the current batch material characteristics' deviation from the ideal state. Based on this deviation vector, a stamping parameter field correction vector is generated, enabling real-time quantitative perception and closed-loop compensation of the forming response of different batches of materials. This allows process parameters to be dynamically adjusted according to fluctuations in material properties, fundamentally avoiding large-scale forming defects caused by material fluctuations under fixed parameter modes.

[0016] When correcting parameters, a real-time wear factor calculated based on the cumulative number of stampings and the die surface hardness decay curve is introduced. Die wear alters the surface friction characteristics and heat transfer conditions. If wear is ignored and correction is based solely on plastic response deviation, systematic errors will occur. By using the real-time wear factor to weight and modulate the initial correction component, the final output stamping parameter field correction vector not only compensates for material property deviations but also simultaneously neutralizes the impact of interface state changes introduced by die wear on the forming results. This two-factor fusion correction method incorporates the cumulative service state of the die as a live variable into the decision-making loop of each stamping, solving the problem that traditional methods cannot quantify the impact of wear, leading to a gradual deterioration in forming quality in the later stages of die life.

[0017] When reconstructing parameters for multiple unexecuted nodes following the current stamping node, the plastic response characteristic tensor and the stamping parameter field correction vector are convolved to generate a propagation attenuation coefficient. The magnitude of the correction vector is then reduced node by node according to the coefficient ratio. The stamping process is a continuous deformation process; the impact of parameter changes in preceding stages on subsequent stages is not transmitted equally, but rather exhibits energy dissipation-like attenuation characteristics as the deformation history and constraints evolve. By using convolution to generate the propagation attenuation coefficient and applying it as a benchmark for node-by-node attenuation correction, a correction propagation mechanism consistent with the physical nature of the process is established. Differential compensation matching the degree of impact is applied to each downstream node in the process chain, rather than simply extrapolating the current node's correction amount to all subsequent nodes, effectively suppressing overcompensation or undercompensation phenomena that easily occur in multi-node coordinated adjustments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1This is a flowchart of the optimized method for intelligent stamping forming of ultra-high strength steel roof beams; Figure 2 This is a schematic diagram of the initial stamping process chain construction; Figure 3 This is a flowchart of the generation process for the plastic response feature tensor; Figure 4 This is a schematic diagram of the closed-loop reconstruction and dynamic self-verification process of the stamping process; Figure 5 It is a distribution map of the tensor clusters of historical plastic response characteristics; Figure 6 It is a dynamic comparison curve of stamping speed and stamping load. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] See Figure 1 This invention provides an intelligent stamping forming optimization method for ultra-high strength steel top cover beams. The method includes: responding to a stamping start command, acquiring the geometric topology of the top cover beam and batch material characteristics of the ultra-high strength steel; based on the geometric topology and batch material characteristics, constructing an initial stamping process chain containing spatiotemporal axis stamping nodes; during a trial stamping at the first stamping node of the initial stamping process chain, collecting multiple physical signal field data corresponding to that node, the physical signal field data including stress wave propagation trajectory and die surface temperature distribution; and combining the stress wave propagation trajectory and the die surface temperature... The distribution is mapped to a high-dimensional feature space to generate a plastic response feature tensor describing the current stamping state. The plastic response feature tensor is topologically aligned and difference-analyzed with a pre-constructed baseline state grid to output a stamping parameter field correction vector. The baseline state grid contains ideal plastic response features recorded in multiple historical stamping events and their corresponding stamping parameter adjustment records. Based on the stamping parameter field correction vector, closed-loop reconstruction is performed on the process parameters of all unexecuted stamping nodes after the current stamping node in the initial stamping process chain to generate an optimized stamping process sequence.

[0022] Example 1: In specific implementation, please refer to Figure 2In response to the stamping start command, after obtaining the geometric topology of the top cover beam and the batch material characteristics of the ultra-high strength steel, an initial stamping process chain containing the spatiotemporal axis stamping node is constructed based on the geometric topology and the batch material characteristics in the following manner.

[0023] Obtain the total closed stroke distance of the stamping die, and divide the total closed stroke distance into N consecutive stroke segments, where N is a positive integer. The number of stroke segments is determined according to the stamping stroke control accuracy requirements. Each stroke segment has an equal length, denoted as Δs. Each stroke segment corresponds to a time-space axis stamping node, forming a total of N time-space axis stamping nodes. The N time-space axis stamping nodes are numbered sequentially according to the stamping stroke progression order as the 1st time-space axis stamping node, the 2nd time-space axis stamping node, ..., the Nth time-space axis stamping node. The distance from the die closing position to the stroke start point corresponding to the i-th time-space axis stamping node is (i-1)·Δs.

[0024] An initial process parameter vector is configured for each spatiotemporal axis stamping node. This vector includes stamping speed, blank holder force, and die clearance values. For the i-th spatiotemporal axis stamping node, the initial process parameter vector is denoted as... , ,in This represents the stamping speed configured for the i-th spacetime axis stamping node, in millimeters per second; This represents the blank holder force assigned to the i-th spacetime axis stamping node, in kilonewtons; This represents the die clearance value configured for the i-th time-space axis stamping node, in millimeters; i is the sequence number of the time-space axis stamping node, i=1,2,…,N.

[0025] All spatiotemporal axis stamping nodes are linearly linked according to the sequence of stamping strokes to form an initial stamping process chain. The linear linking method is as follows: the first spatiotemporal axis stamping node is the first node of the initial stamping process chain, the second spatiotemporal axis stamping node is the subsequent node of the first spatiotemporal axis stamping node, and so on, until the Nth spatiotemporal axis stamping node is the last node of the initial stamping process chain. The nodes are connected through the direction of the stamping stroke to obtain an ordered node sequence, which constitutes the initial stamping process chain.

[0026] The stamping speed in the initial process parameter vector is configured to decrease proportionally from the first time-space axis stamping node to the last time-space axis stamping node, i.e., the stamping speed decreases geometrically as i increases. The specific configuration of the stamping speed is determined by the following formula: in, This represents the stamping speed of the i-th spatiotemporal axis stamping node; This represents the stamping speed of the first spacetime axis stamping node. The value is determined based on the start-up characteristics of the press and the initial yield strength of the ultra-high strength steel, and a specific value is selected from the range of 60% to 80% of the maximum allowable speed of the press. Indicates the common ratio of decreasing stamping speed. It is a real number, ranging from 0.85 to 0.95. The specific values ​​are determined based on the strain rate sensitivity coefficient and work hardening index in the batch material characteristics of ultra-high strength steel, so that the decrease in stamping speed with stroke is consistent with the adaptation requirement of the rate reduction due to the increase in flow stress during the plastic deformation process of ultra-high strength steel. Indicates the sequence number of the stamping node on the spacetime axis. Blankholder force. A constant value is set for all time-space axis stamping nodes. This constant value is calculated based on the projected area of ​​the flange region in the top cover beam geometry and the thickness of the ultra-high strength steel plate. (Die clearance value) All time-space axis stamping nodes are set with preset standard gap values, which are 1.1 to 1.2 times the thickness of the ultra-high strength steel plate.

[0027] Example 2: In practice, multiple physical signal field data corresponding to the first spatiotemporal axis stamping node of the initial stamping process chain are collected in the following way.

[0028] Multiple piezoelectric sensors and multiple thermocouple sensors are pre-embedded on the punch and die of the stamping die. The piezoelectric sensors are selected from piezoelectric ceramic sensors with a frequency response range covering 0.1 Hz to 500 kHz, used to sense stress wave signals generated during the stamping process. The thermocouple sensors are thin-film thermocouples with a response time constant of less than 5 milliseconds, used to sense the real-time temperature values ​​at various locations on the die surface. All piezoelectric sensors and all thermocouple sensors are arranged in a staggered, contoured grid pattern on the die surface. This staggered grid pattern means that sensing points are arranged at equal intervals along the meridian and parallel directions of the die surface. Piezoelectric sensors and thermocouple sensors alternate in both directions; that is, any adjacent node of a piezoelectric sensor is a thermocouple sensor, and any adjacent node of a thermocouple sensor is a piezoelectric sensor, thus forming a staggered contoured grid structure. The spacing between sensor points is determined based on the minimum radius of curvature in the geometric topology of the top cover beam, and half of the minimum radius of curvature is taken as the arc distance between adjacent sensor points.

[0029] When the stamping slide moves to the stroke position corresponding to the current time-space axis stamping node, all piezoelectric sensors and all thermocouple sensors are triggered to synchronously collect data. The triggering method is as follows: a displacement sensor is installed on the stamping slide, which monitors the stroke position of the stamping slide in real time. When the stamping slide reaches the mold closing position corresponding to the first time-space axis stamping node, the displacement sensor sends a trigger signal to the data acquisition controller. Upon receiving the trigger signal, the data acquisition controller sends a synchronous acquisition command to all piezoelectric sensors and all thermocouple sensors.

[0030] Upon receiving the synchronization acquisition command, all piezoelectric sensors acquire time-domain signals at a preset sampling frequency, with each sensor outputting one time-domain signal channel. The time-domain signals acquired by all piezoelectric sensors are combined according to their spatial location to form a multi-channel synchronous time-domain signal set. This set of multi-channel synchronous time-domain signals is then conditioned and packaged to reconstruct the stress wave propagation trajectory. The stress wave propagation trajectory is stored as a two-dimensional matrix, where the row vectors correspond to the channel numbers of different piezoelectric sensors, and the column vectors correspond to the sampling time points.

[0031] Upon receiving the synchronization acquisition command, all thermocouple sensors acquire temperature values ​​at a preset sampling frequency, with each thermocouple sensor outputting a temperature value. The temperature values ​​acquired by all thermocouple sensors, along with the spatial coordinates of each sensor, are input into a spatial interpolation algorithm. The spatial interpolation algorithm employs Kriging interpolation, which performs mesh interpolation calculations on the three-dimensional curved surface model of the mold surface based on the position coordinates of all thermocouple sensors and the acquired temperature values. This generates a continuous temperature field distribution covering the entire mold surface area; this continuous temperature field distribution is the mold surface temperature distribution. The mold surface temperature distribution is stored in the format of three-dimensional curved surface mesh data, with each mesh node containing surface coordinates and a temperature value.

[0032] The sampling frequency adaptively increases as the stamping stroke progresses. Specifically, the total stamping stroke is divided into K consecutive stroke intervals, where K is a positive integer. Each stroke interval contains several consecutive spatiotemporal axis stamping nodes. The sampling frequency of both piezoelectric and thermocouple sensors remains constant within the same stroke interval. Between different stroke intervals, the sampling frequency increases progressively as the stroke interval moves forward. The sampling frequency configuration is determined by the following formula: in, This represents the sampling frequency of the m-th travel interval, in Hertz; This indicates the sampling frequency of the first travel interval. The value is taken as one-tenth of the upper limit of the frequency response range of the piezoelectric sensor, i.e., 50 kHz; This represents the sampling frequency increment factor. For real numbers, The value is taken as 1.15. The reason for 1.15 is that in the later stage of the stamping stroke, the proportion of high-frequency components of the stress wave signal increases, and the sampling frequency needs to be increased accordingly to meet the requirements of the sampling theorem. When the value is 1.15, the sampling frequency of adjacent travel intervals increases by 15%, which matches the average growth rate of the stress wave signal bandwidth as the travel changes. Indicates the sequence number of the travel interval. .

[0033] At the first time-space axis stamping node, the stamping slide has not yet started moving. At this time, the stroke interval number m is 1, and the sampling frequency is... As the stamping process enters the subsequent stroke range, the sampling frequency is sequentially adjusted according to... , Increasing.

[0034] Example 3: In specific implementation, please refer to Figure 3 The stress wave propagation trajectory and the temperature distribution on the mold surface are mapped to a high-dimensional feature space to generate a plastic response feature tensor describing the current stamping state, which is achieved in the following way.

[0035] A time-frequency domain transformation is performed on the stress wave propagation trajectory. The stress wave propagation trajectory is stored as a two-dimensional matrix in the form of multi-channel synchronous time-domain signals. Each row of the two-dimensional matrix corresponds to a channel signal of a piezoelectric sensor, and each column corresponds to a sampling time point. Discrete wavelet packet transform is performed on each row of channel signals in the two-dimensional matrix. The discrete wavelet packet transform uses the Daubechiesdb8 wavelet basis function, and the decomposition level is set to 5 levels. After performing the discrete wavelet packet transform, each channel signal obtains a wavelet packet energy value in the j-th sub-frequency band of the 5th level. The wavelet packet energy value is the sum of the squares of the wavelet packet coefficients in the corresponding sub-frequency band. The wavelet packet energy values ​​of all channel signals of all piezoelectric sensors in all sub-frequency bands of the 5th level are arranged according to the frequency band number to form an energy spectrum vector with a length of [missing information]. Each energy value is assigned to a frequency band in the energy spectral vector. The energy values ​​are then arranged sequentially along the frequency axis from low to high frequency, and the natural logarithm of each energy value is taken to obtain the attenuation spectrum of the stress wave energy on the frequency axis. The horizontal axis of the attenuation spectrum represents the frequency band number, and the vertical axis represents the wavelet packet energy value after taking the natural logarithm.

[0036] Gradient calculations are performed on the temperature distribution of the mold surface. The temperature distribution of the mold surface is stored as three-dimensional surface mesh data. Each mesh node contains surface coordinate values ​​and temperature values ​​in the mold surface coordinate system. The mold surface coordinate system adopts a two-dimensional orthogonal surface coordinate system with the principal curvature lines of the mold surface as coordinate axes. The coordinate axes are denoted as follows: direction and Direction. Regarding the temperature distribution on the mold surface... direction and Calculate the first-order partial derivatives in each direction. The first-order partial derivatives in the direction are calculated between adjacent grid nodes using a central difference scheme. The first-order partial derivatives in the direction are also calculated between adjacent grid nodes using the central difference scheme. The partial derivatives at each grid node are then calculated. The partial derivative values ​​of the direction and The partial derivatives of the directions are combined to form a two-dimensional vector, which represents the direction and rate of temperature change at the corresponding grid node. By traversing all grid nodes in the three-dimensional surface mesh data, a vector field composed of the two-dimensional vectors of all grid nodes is obtained, which is the temperature gradient vector field.

[0037] The attenuation spectral lines and temperature gradient vector field are normalized. The normalization method for the attenuation spectral lines is as follows: extract the ordinate values ​​of all frequency bands in the attenuation spectral lines, calculate the maximum and minimum values ​​of all ordinate values, and then apply the formula... The ordinate values ​​of each frequency band are linearly mapped to the interval [0,1], where This represents the ordinate value to be normalized. This represents the maximum value among all ordinate values. This represents the minimum value among all ordinate values. The normalization process for the temperature gradient vector field is as follows: extract the magnitude of the two-dimensional vector of all grid nodes in the temperature gradient vector field, calculate the maximum and minimum values ​​of all magnitudes, and then apply the formula... The magnitude of the 2D vector at each grid node is linearly mapped to the interval [0,1], while keeping the direction of the 2D vector unchanged. The normalized 2D vector is then reconstructed using the normalized magnitude and the original direction. This represents the magnitude of a two-dimensional vector to be normalized. This represents the maximum value among all two-dimensional vector magnitudes. This represents the minimum value among all two-dimensional vector magnitudes.

[0038] The normalized decay spectrum and the normalized temperature gradient vector field are fused into a three-dimensional tensor structure, which is the plastic response characteristic tensor describing the current stamping state. The three dimensions of the three-dimensional tensor structure correspond to the spatial coordinate axis, the time axis, and the physical parameter type axis, respectively. The spatial coordinate axis dimension uses the discrete grid node index of the mold surface coordinate system as its coordinate values, determined by the grid node positions of the temperature gradient vector field. The length of the spatial coordinate axis dimension is the total number of grid nodes on the mold surface, denoted as . The time axis dimension is established by mapping the frequency band indices of the attenuation spectral lines to the equivalent time delay. The corresponding equivalent time delay is ,in This is the highest effective frequency of the stress wave signal. The stretching factor, The value is set to 3. The reason for this value is that, under the 5-level decomposition structure of the discrete wavelet packet transform, the bandwidth of the 5th level sub-band is 1 / 32 of the original sampling frequency. The time delay resolution needs to match this bandwidth. When the stretching factor... When the value is 3, the step size of the equivalent time delay can cover the typical time range of stress wave propagation back and forth within the mold surface.

[0039] The physical parameter type axis dimension contains two parameter channels. The first parameter channel stores the normalized attenuation spectral line values, and the second parameter channel stores the gradient magnitude at the corresponding spatial coordinate position in the normalized temperature gradient vector field. For the p-th grid node in the spatial coordinate axis dimension and the equivalent time delay position corresponding to the frequency band index j in the time axis dimension of the three-dimensional tensor structure, the element values ​​of the plastic response characteristic tensor in the two channels of the physical parameter type axis are determined by the following formula: in, This represents the element value of the plastic response feature tensor at index p on the spatial coordinate axis, index j on the time axis, and index c on the physical parameter type axis; p is the index value of the spatial coordinate axis, ranging from 1 to c. The integer; j is the index value of the time axis dimension, corresponding to the frequency band number, j is an integer from 1 to 32; c is the index value of the physical parameter type axis, c=1 represents the attenuation spectral line parameter channel, c=2 represents the temperature gradient vector field parameter channel. This represents the ordinate value of the normalized attenuation spectral line in the j-th frequency band; Indicates the spectral broadening index. It is a real number, ranging from 0.5 to 1.5. The specific value is determined based on the grain size distribution dispersion in the batch material characteristics of ultra-high strength steel. When the grain size distribution is relatively uniform... Taking a value close to 1.0, when the grain size distribution is relatively wide... Values ​​less than 1.0 are used to enhance the representation of high-frequency attenuation features in tensors; This represents the two-dimensional vector magnitude of the normalized temperature gradient vector field at the p-th grid node.

[0040] Through the above process, the plastic response feature tensor is generated. The plastic response feature tensor is stored as a three-dimensional array, with a dimension of [missing value]. The plastic deformation characteristics contained in the stress wave propagation trajectory and the temperature distribution of the die surface under the current stamping state are encoded into a unified high-dimensional data structure.

[0041] Example 4: In practice, the plastic response feature tensor is topologically aligned and the difference is analyzed with the pre-constructed reference state mesh to output the stamping parameter field correction vector, which is achieved in the following way.

[0042] The pre-constructed baseline state mesh is generated by collecting all historical plastic response feature tensors corresponding to the top cover beams of products deemed qualified in historical production batches. The historical plastic response feature tensors are generated using the same tensor construction method as the aforementioned plastic response feature tensors; that is, each historical plastic response feature tensor includes a spatial coordinate axis dimension, a time axis dimension, and a physical parameter type axis dimension, all with a size of [missing value]. For each historical plastic response feature tensor, its corresponding historical stamping parameter vector is labeled. This vector contains historical stamping speed, historical blank holder force, and historical die clearance values. The format of the historical stamping parameter vector is as follows: ,in This represents the k-th historical stamping speed. This represents the k-th historical pressure edge force. This represents the k-th historical die clearance value, where k is the sequence number of the historical stamping event.

[0043] All labeled historical plastic response feature tensors are clustered according to their distance from the standard qualified product feature. The standard qualified product feature is determined as follows: from all historical plastic response feature tensors, the plastic response feature tensor corresponding to the batch whose historical stamping parameter vectors completely meet the design nominal parameter requirements is selected as a reference tensor. The arithmetic mean of all reference tensors at their corresponding element positions is taken to obtain a standard qualified product feature tensor, denoted as [missing information]. Standard qualified product characteristic tensor The dimension of the historical plastic response feature tensor is the same as that of the historical plastic response feature tensor. For each historical plastic response feature tensor, the historical plastic response feature tensor and the standard qualified product feature tensor are calculated. The Euclidean distance between the historical plastic response feature tensor and the standard qualified product feature is used as the distance value. All labeled historical plastic response feature tensors and their distance values ​​are used as input to a clustering algorithm. The clustering algorithm uses K-means clustering, with the number of clusters set to Q, where Q is a positive integer greater than 1. The value of Q is determined based on the total number of historical stamping events and the granularity requirements of stamping parameter adjustments. The value of Q is the square root of the total number of historical stamping events, rounded up. After clustering, Q clusters are obtained, each containing several historical plastic response feature tensors and their corresponding historical stamping parameter vectors.

[0044] The feature vector of each cluster center is used as a baseline state node. The feature vector of a cluster center is the tensor obtained by taking the arithmetic mean of the historical plastic response feature tensors within that cluster at the corresponding element positions, denoted as . , where q is the cluster number, and q is an integer from 1 to Q. Each baseline state node corresponds to a feature vector of a cluster center. The average value of all historical stamping parameter vectors within the cluster is stored as the associated corrected baseline value for the baseline state node. The associated corrected baseline value is a stamping parameter vector, denoted as... ,in This represents the average of all historical stamping speeds within the q-th cluster. This represents the average historical edge pressure force within the q-th cluster. This represents the average of all historical mold gap values ​​within the q-th cluster. All baseline state nodes and their associated corrected baseline values ​​constitute a pre-constructed baseline state mesh.

[0045] In some embodiments, the baseline state grid is stored as a relational mapping table, where each record corresponds to a cluster number q, and the record content includes the feature vector of the cluster center. and associated correction benchmark value .

[0046] The specific method for topologically aligning the plastic response feature tensor with the pre-constructed baseline state grid is as follows: The similarity value between the plastic response feature tensor and each baseline state node in the baseline state grid is calculated. The similarity value uses a cosine similarity metric based on the tensor inner product. The plastic response feature tensor... Expand into a one-dimensional vector The feature vector of the cluster center of each baseline state node. Expand into a one-dimensional vector Plastic response characteristic tensor Similarity value with the q-th baseline state node Calculated using the following formula: in, Represents the characteristic tensor of plastic response The similarity value with the q-th baseline state node. The value range of is [-1, 1]; Represents the characteristic tensor of plastic response Expanded one-dimensional vector The value at the d-th element position; The feature vector representing the cluster center of the q-th baseline state node Expanded one-dimensional vector The value at the d-th element position; d is the element index of the one-dimensional vector, and d takes integer values ​​from 1 to D; D represents the total length of the one-dimensional vector, and D takes values ​​of... .

[0047] The baseline state node with the highest similarity value is selected as the matching baseline node, and the cluster number corresponding to the matching baseline node is denoted as . The condition is satisfied: for all values ​​of q, the following holds true. .

[0048] Calculate the deviation vector between the eigenvalues ​​of the plastic response characteristic tensor and the eigenvalues ​​of the matched reference node. The eigenvalues ​​of the plastic response characteristic tensor are derived from the plastic response characteristic tensor. Extraction is performed by calculating the plastic response feature tensor. The average vector along the spatial coordinate axis is obtained by taking the arithmetic mean of the U elements along the spatial coordinate axis. The two-dimensional matrix is ​​expanded into a one-dimensional vector of length 64, denoted as the plastic response eigenvector. The feature values ​​of the matching benchmark nodes are derived from the feature vectors of the cluster centers of the matching benchmark nodes in the same way. Extracting from this, we obtain a one-dimensional vector of length 64, denoted as the baseline feature vector. Calculate the eigenvectors of the plastic response. With reference eigenvectors The difference vector between them, i.e. ,in The length is 64. The deviation vector is... Deviation vector Each component represents the offset of the plastic response feature tensor from the eigenvalue of the matching reference node in the corresponding dimension. (Obstacle vector) This serves as the basis for the stamping parameter field correction vector.

[0049] The specific method for outputting the stamping parameter field correction vector is as follows: The deviation vector... The input is a pre-trained parameter adjustment mapping model. The core architecture of the parameter adjustment mapping model is a four-layer fully connected feedforward neural network. This four-layer fully connected feedforward neural network consists of one input layer, two hidden layers, and one output layer. The input layer contains 64 neurons, each receiving a bias vector. The system has 64 components. The first hidden layer contains 128 neurons, and the second hidden layer contains 64 neurons. The activation function of each neuron in the hidden layer is a hyperbolic tangent function. The output layer contains 3 neurons, corresponding to the stamping speed adjustment, blank holder force adjustment, and die clearance adjustment, respectively. The activation function of each neuron in the output layer is a linear function. The input layer and the first hidden layer are fully connected, the first hidden layer and the second hidden layer are fully connected, and the second hidden layer and the output layer are fully connected. The weight matrix and bias vector between adjacent layers are determined through training.

[0050] The training steps for the parameter adjustment mapping model are as follows: Collect paired data between historical plastic response feature tensors recorded in historical stamping events and the corresponding actual stamping parameter adjustment values. For each training sample, the historical deviation vector calculated using the aforementioned deviation vector extraction method is used as the model input, and the three-dimensional vector composed of the stamping speed adjustment value, blank holder force adjustment value, and die clearance adjustment value recorded in the historical stamping event is used as the model output label. The training dataset contains no fewer than 500 training samples. During training, the mean squared error loss function is used, the Adam optimizer is used, the initial learning rate is set to 0.001, the batch size is set to 32, and the number of training iterations is set to 2000 rounds. After training, the weight matrix and bias vector of the four-layer fully connected feedforward neural network are fixed to obtain the trained parameter adjustment mapping model. The parameter adjustment mapping model outputs a three-dimensional vector, which is the initial correction component, denoted as... .

[0051] The real-time wear factor of the current stamping node is obtained. This factor is calculated based on the cumulative number of stamping operations and the die surface hardness decay curve. The die surface hardness decay curve is pre-calibrated through offline wear experiments, which record the decrease in Rockwell hardness of the die material under different cumulative stamping operations. The cumulative number of stamping operations is denoted as... The initial surface hardness of the mold is denoted as The actual surface hardness of the mold measured under the current cumulative number of stamping cycles is denoted as . Real-time wear factor The calculation formula is Real-time wear factor It is a dimensionless numerical value, ranging from 0 to 1. The closer the value is to 1, the lower the degree of mold wear. The closer the value is to 0, the higher the degree of mold wear. (Cumulative stamping count) The data is obtained through a counter on the stamping production line, indicating the completion of each stamping operation. The value is incremented by 1.

[0052] Using real-time wear factor For the initial correction component Weighted modulation is performed to obtain the final ramming parameter field correction vector. The weighted modulation method is element-wise multiplication, and the ramming parameter field correction vector is denoted as... ,Right now Through the above process, the topological alignment and difference analysis of the plastic response feature tensor with the pre-constructed reference state mesh are completed, and the final stamping parameter field correction vector is output.

[0053] See Figure 5 In the figure, the horizontal axis represents the "first principal component," and the vertical axis represents the "second principal component," both being the two principal feature components obtained after dimensionality reduction of the plastic response feature tensor using Principal Component Analysis (PCA). The figure shows scatter points for five different clusters, each labeled with a different symbol: cluster 1 is marked with a circle, cluster 2 with a triangle, cluster 3 with a square, cluster 4 with a diamond, and cluster 5 with a star. Each cluster corresponds to a set of dimensionality-reduced data points of the historical plastic response feature tensor, reflecting its distribution in the feature space.

[0054] As shown in the figure, the five clusters exhibit relatively clear separation in the two-dimensional feature space of the first and second principal components. This indicates that after classifying the historical plastic response feature tensor using the K-means clustering algorithm, the cluster centers can effectively represent the typical forms of various plastic response features. Specifically, the samples of cluster 1 are mainly concentrated in the intervals of the first principal component 0 to 3 and the second principal component -1 to 4; cluster 2 is mainly distributed in the intervals of the first principal component 6 to 10 and the second principal component 0 to 5; the samples of cluster 3 are located between the first principal component 3 to 7 and the second principal component 6 to 10; cluster 4 is distributed in the intervals of the first principal component 0 to 3 and the second principal component 7 to 11; and cluster 5 is mainly concentrated in the intervals of the first principal component 7 to 11 and the second principal component 7 to 11.

[0055] This figure illustrates the results of feature extraction and cluster analysis of the historical plastic response feature tensor during the pre-construction of the baseline state grid in Example 4. The five clusters correspond to different baseline state nodes. The distribution differences between the clusters indicate significant differences in the plastic response states represented by each node. This facilitates subsequent topological alignment and difference resolution between the plastic response feature tensor and the baseline state grid through similarity calculation, thereby accurately outputting the stamping parameter field correction vector. The clustering structure shown in the figure provides a rich and diverse training sample base for the parameter adjustment mapping model, supporting the model's intelligent optimization and adjustment of stamping parameters under different plastic response states.

[0056] Example 5: In specific implementation, please refer to Figure 4 The process parameters of all unexecuted stamping nodes after the current stamping node in the initial stamping process chain are reconstructed in a closed loop to generate an optimized stamping process sequence, which is achieved in the following way.

[0057] Extract the position index of the first unexecuted stamping node after the current stamping node. In the initial stamping process chain, each time-space axis stamping node has a unique sequence number. Let the current time-space axis stamping node that has performed a trial stamp be the [number]. If there are multiple time-space axis stamping nodes, then the position index of the first unexecuted stamping node after the current stamping node is: Based on location index Obtain the initial process parameter vector corresponding to the unexecuted stamping node, denoted as .

[0058] The stamping parameter field correction vector is compared with the initial process parameter vector. Element-by-element superposition is performed to generate a first optimized process parameter vector. The first optimized process parameter vector is denoted as Using the first optimized process parameter vector Replace the first step in the initial stamping process chain Initial process parameter vector corresponding to each time-space axis stamping node .

[0059] After replacing the parameters for the first unexecuted stamping node, the updated stamping process chain is used as an intermediate process sequence. In the intermediate process sequence, the first... The process parameter vectors of the nodes have been updated to the first optimized process parameter vector, while the process parameter vectors of the remaining spatiotemporal axis stamping nodes that have not yet been executed remain the initial process parameter vectors.

[0060] A propagation attenuation coefficient is generated by convolving the plastic response feature tensor with the stamping parameter field correction vector. (Plastic response feature tensor) For one The three-dimensional tensor, in which The length of the spatial coordinate axis dimension is equal to the total number of mesh nodes on the mold surface. 32 represents the length of the time axis dimension, and 2 represents the length of the physical parameter type axis dimension. First, the plastic response characteristic tensor... Global average pooling is performed along the spatial coordinate axis and the physical parameter type axis to obtain a time series vector of length 32, denoted as . , The Middle The calculation method for each element is as follows: ,in For the index of the spatial coordinate axis dimension, Indexes for the axis dimensions of physical parameter types. For the time axis dimension index, Take integers from 1 to 32. Stamping parameter field correction vector. The length is , The stamping parameter field correction vector As a one-dimensional convolution kernel, for time series vectors Perform a one-dimensional convolution operation with a stride of 1 to obtain a length of... The convolution result sequence. The propagation attenuation coefficient is calculated by the following formula: in, This represents the propagation attenuation coefficient, which is a scalar value. Indicates the position index of the convolution result sequence; Represents the stamping parameter field correction vector The element index; Represents the stamping parameter field correction vector The One component; Represents a time series vector In the The element value at each position; 3 represents the length of the stamping parameter field correction vector, with a value of 3; 32 represents the length of the time axis dimension, i.e., the time series vector. Length. Propagation attenuation coefficient. The magnitude of the value reflects the degree of attenuation of the influence of the plastic response characteristics of the current stamping state on subsequent process parameters.

[0061] Based on the propagation attenuation coefficient Scaling down the stamping parameter field correction vector proportionally This yields a decay correction vector, denoted as... , The attenuation correction vector This is superimposed on the initial process parameter vector of the second unexecuted stamping node following the current stamping node. Its initial process parameter vector is The second optimized process parameter vector is obtained by superimposing the parameters. And replace the first step in the initial stamping process chain with the second optimized process parameter vector. The initial process parameter vector of each time-space axis stamping node.

[0062] And so on, for the first stamping node after the current stamping node... The position index of the unperformed stamping node is: The attenuation correction vector is generated by multiplying the previous attenuation correction vector by the propagation attenuation coefficient. ,Right now The attenuation correction vector With the Initial process parameter vector for each spatiotemporal axis stamping node The optimized process parameter vectors are superimposed and replaced with the superimposed optimized process parameter vectors. This process is continued until all unexecuted stamping nodes after the current stamping node are traversed. All replaced spatiotemporal axis stamping nodes are then arranged in their original order to form the optimized stamping process sequence.

[0063] After generating the optimized stamping process sequence, a dynamic self-verification of stamping parameters is performed. The specific method of dynamic self-verification of stamping parameters is as follows: During the execution of the optimized stamping process sequence, the actual stamping load of the current execution node is monitored in real time. The actual stamping load is collected in real time by force sensors installed on the press slide or die, and the force sensors output the stamping load force value at the current moment. The theoretical load threshold of the corresponding node in the optimized stamping process sequence is obtained. The theoretical load threshold is predetermined by numerical calculation based on the blank holder force value, die clearance value, and stamping speed value in the optimized process parameter vector of that node, combined with the material constitutive model of ultra-high strength steel, and stored in the process sequence. The actual stamping load is compared with the theoretical load threshold. If the actual stamping load exceeds the theoretical load threshold, the stamping process is immediately paused. The current execution node is used as the new starting point to re-acquire multiple physical signal field data corresponding to the current execution node, including stress wave propagation trajectories and die surface temperature distribution. The process then proceeds to generate plastic response characteristic tensors, perform topology alignment and difference analysis, and reconstruct the closed-loop stamping process chain, regenerating an optimized stamping process sequence for this node and subsequent nodes. If the actual stamping load does not exceed the theoretical load threshold, the next node of the optimized stamping process sequence is executed.

[0064] See Figure 6In the figure, the horizontal axis represents the stamping node number i on the space-time axis, ranging from 1 to 200, reflecting the node sequence of the stamping process according to the stroke segment. The left vertical axis is the stamping speed in millimeters per second, and the right vertical axis is the stamping pressure in kilonewtons.

[0065] The solid curve represents the trend of the initial stamping speed as the stamping node number changes. The figure shows that the initial stamping speed starts from 80 mm / s at the first node, decreases exponentially, and then drops rapidly to near zero after about the 50th node, and then remains constant at zero, which is in line with the design principle of the stamping speed decreasing in a geometric series in Example 1.

[0066] The dashed curve represents the change in stamping speed after optimization. The optimized stamping speed shows a significant negative fluctuation at node number 70, and then quickly recovers to a value close to zero. This indicates that when the process parameters after the current stamping node are adjusted based on the closed-loop reconstruction method in Example 5, a local reduction in stamping speed occurs, reflecting the dynamic correction effect of the stamping parameter field correction vector on the stamping speed.

[0067] The dotted-line curve represents the actual stamping load, showing a slow upward trend overall, with a peak at approximately node number 150, followed by a slight decline. This peak exceeds the theoretical load threshold (dotted-line curve), triggering the dynamic self-verification mechanism of the stamping parameters. This embodies the closed-loop control process described in Example 5, which involves pausing the stamping process and re-acquiring physical signals for parameter adjustment.

[0068] The dotted curve represents the theoretical load threshold. As the stamping process progresses, the load threshold gradually increases, indicating that the theoretical load threshold dynamically matches the comprehensive parameters of stamping speed, blank holder force, and die clearance, thus demonstrating the plastic deformation characteristics of ultra-high strength steel.

[0069] Overall, the data in the figure demonstrates the dynamic optimization of the stamping process based on the real-time plastic response feature tensor and the stamping parameter field correction vector in Example 5. This ensures that the actual stamping load is always controlled within the theoretical threshold range, and depicts the entire process of stamping speed optimization and adjustment and closed-loop self-verification response when the load is abnormal. This verifies the effectiveness and safety of the closed-loop reconstruction of the stamping process chain in this invention.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An optimized method for intelligent stamping forming of ultra-high strength steel roof beams, characterized in that, The method includes: S1: In response to the stamping start command, obtain the geometric topology of the top cover beam and the batch material characteristics of the ultra-high strength steel, and construct an initial stamping process chain containing the spatiotemporal axis stamping node based on the geometric topology and the batch material characteristics. S2: When performing a trial stamping at the first stamping node of the initial stamping process chain, collect multiple physical signal field data corresponding to that node. The physical signal field data includes stress wave propagation trajectory and mold surface temperature distribution. S3: Map the stress wave propagation trajectory and the mold surface temperature distribution to a high-dimensional feature space to generate a plastic response feature tensor describing the current stamping state; S4: Perform topological alignment and difference analysis between the plastic response feature tensor and a pre-constructed reference state grid to output a stamping parameter field correction vector. The reference state grid contains ideal plastic response features and their corresponding stamping parameter adjustment records recorded in multiple historical stamping events. S5: Based on the stamping parameter field correction vector, perform closed-loop reconstruction on the process parameters of all unexecuted stamping nodes after the current stamping node in the initial stamping process chain to generate an optimized stamping process sequence.

2. The optimized method for intelligent stamping forming of ultra-high strength steel roof beams according to claim 1, characterized in that, The specific method for constructing an initial stamping process chain containing a spacetime axis stamping node in step S1 is as follows: The stamping stroke is divided into multiple continuous equally spaced stroke segments according to the die closing distance, and each stroke segment corresponds to a time-space axis stamping node; An initial process parameter vector is configured for each spatiotemporal axis stamping node, and the initial process parameter vector includes stamping speed, blank holder force and die clearance value; All time-space axis stamping nodes are linearly linked according to the order of the stamping stroke to form the initial stamping process chain.

3. The optimized method for intelligent stamping forming of ultra-high strength steel top cover beams according to claim 2, characterized in that, The stamping speed in the initial process parameter vector is configured to decrease proportionally from the first node to the last node.

4. The optimized method for intelligent stamping forming of ultra-high strength steel roof beams according to claim 1, characterized in that, The specific method for collecting multiple physical signal field data corresponding to the node in step S2 is as follows: Multiple piezoelectric sensors and thermocouple sensors are pre-embedded in the punch and die of the stamping die; When the stamping slide moves to the stroke position corresponding to the current time-space axis stamping node, all sensors are triggered to collect data synchronously. The time-domain signals collected by all piezoelectric sensors are reconstructed into the stress wave propagation trajectory, and the temperature values ​​collected by all thermocouple sensors are interpolated to generate the temperature distribution of the mold surface.

5. The optimized method for intelligent stamping forming of ultra-high strength steel roof beams according to claim 4, characterized in that, The piezoelectric sensor and thermocouple sensor are arranged in a staggered, contoured grid pattern on the mold surface, and the sampling frequency adaptively increases as the stamping stroke progresses.

6. The optimized method for intelligent stamping forming of ultra-high strength steel top cover beams according to claim 1, characterized in that, The specific method for generating a plastic response feature tensor describing the current stamping state in step S3 is as follows: Perform a time-frequency domain transformation on the stress wave propagation trajectory to obtain the attenuation spectrum of the stress wave energy on the frequency axis; Gradient calculations are performed on the temperature distribution of the mold surface to obtain a temperature gradient vector field; After normalizing the attenuation spectral lines and the temperature gradient vector field, they are fused into a three-dimensional tensor structure. The three dimensions of this three-dimensional tensor structure correspond to spatial coordinates, time axis, and physical parameter type, respectively.

7. The optimized method for intelligent stamping forming of ultra-high strength steel roof beams according to claim 1, characterized in that, The method for generating the pre-constructed reference state grid in step S4 is as follows: Collect all historical plastic response characteristic tensors corresponding to the top cover beams of qualified products from historical production batches; For each historical plastic response feature tensor, label its corresponding historical stamping parameter vector, which includes historical stamping speed, historical blank holder force, and historical die clearance value; All labeled historical plastic response feature tensors are clustered according to their distance from the features of standard qualified products. The feature vector of each cluster center is used as a reference state node, and the average value of all historical stamping parameter vectors in the cluster is stored as the associated correction reference value of the reference state node.

8. The optimized method for intelligent stamping forming of ultra-high strength steel top cover beams according to claim 1, characterized in that, The specific method for performing topological alignment and difference resolution between the plastic response feature tensor and a pre-constructed baseline state grid in step S4 is as follows: Calculate the similarity value between the plastic response feature tensor and each reference state node in the reference state grid; The baseline state node with the highest similarity value is selected as the matching baseline node; Calculate the deviation vector between the eigenvalues ​​of the plastic response characteristic tensor and the eigenvalues ​​of the matching reference node, and use this deviation vector as the stamping parameter field correction vector.

9. The optimized method for intelligent stamping forming of ultra-high strength steel top cover beams according to claim 8, characterized in that, The specific method for outputting the stamping parameter field correction vector is as follows: The deviation vector is input into a pre-trained parameter adjustment mapping model, which outputs an initial correction component. Obtain the real-time wear factor of the current stamping node, which is calculated based on the cumulative number of stampings and the die surface hardness decay curve; The initial correction component is weighted and modulated using the real-time wear factor to obtain the final stamping parameter field correction vector.

10. The optimized method for intelligent stamping forming of ultra-high strength steel roof beams according to claim 1, characterized in that, The specific method for performing closed-loop reconstruction of the process parameters of all unexecuted stamping nodes after the current stamping node in the initial stamping process chain in step S5 is as follows: Extract the position index of the first unexecuted stamping node after the current stamping node, and obtain the initial process parameter vector corresponding to the unexecuted stamping node based on the position index; The stamping parameter field correction vector is superimposed element by element with the initial process parameter vector to generate a first optimized process parameter vector. Replace the corresponding initial process parameter vector in the initial stamping process chain with the first optimized process parameter vector.

Images