A method and system for real-time control of the gap for preventing wrinkles and cracks in aluminum disc punching

Abstract

The application discloses a kind of aluminum round piece stamping anti-wrinkle and crack clearance real-time regulation method and system, and the blank holder is divided into multiple independent adjustment regions along the circumference, and based on material parameter setting dynamic change global clearance reference curve along with stamping stroke, as the reference trajectory of feedforward control;Macroscopic geometry and microscopic strain characteristics are processed by a multimodal fusion model to generate multi-dimensional state data;Risk discrimination is carried out using a lightweight discrimination model to identify local areas at risk of wrinkling or cracking and risk types, and to generate clearance compensation for each adjustment region;According to the compensation, the independent fine-tuning of the corresponding actuator of each adjustment region is driven by closed-loop feedback control to dynamically correct the material flow resistance;The application realizes two-way dynamic regulation and high-frequency response of wrinkle area gap reduction and crack area gap increase through zoning layout and multi-modal intelligent sensing, and improves the stability of forming quality and process universality.

Description

Technical Field

[0001] This invention belongs to the field of stamping forming and intelligent control technology, specifically relating to a real-time gap control method and system for preventing wrinkles and cracks in aluminum disc stamping. Background Technology

[0002] In the existing technology, the forming quality control of sheet metal stamping usually relies on the process design of the drawing die structure and the empirical adjustment of the blank holder force. The requirements for suppressing wrinkling and cracking defects are met by constructing a finite element simulation model and carrying out parameter iteration. However, when the above-mentioned existing stamping defect control methods are applied to aluminum disc stamping, there are obvious shortcomings and applicability obstacles in controlling the differences in material flow in multiple regions and the risk of dynamic instability.

[0003] In practical applications, the drawing of aluminum discs involves multi-body contact and frictional constraints between the die, punch, and blank holder, resulting in highly nonlinear and non-uniform material flow. Due to the anisotropy of the aluminum material itself, the uniformity of sheet thickness, and fluctuations in lubrication conditions, aluminum discs are prone to instability and wrinkling in the flange area during stamping, while facing the risk of cracking in the sidewall area due to tensile stress concentration. Traditional equal blank holder force control and schemes based on the above static adjustment methods not only have the problems of response lag and inability to perceive the material state in real time, but also make it difficult to perform bidirectional dynamic control of "reducing gap in the wrinkling area and increasing gap in the cracking area" in different areas at the same time. This results in relatively weak discrimination and stability of defect suppression, and the overall universality of process parameters is also low.

[0004] It is evident that existing technologies often fail to address issues such as wrinkle suppression and crack prevention, dynamic response lag, and lack of refined zone control in complex aluminum disc stamping scenarios; these are the shortcomings of existing technologies.

[0005] In view of this, it is very necessary to provide a real-time gap control method and system for preventing wrinkles and cracks in aluminum disc stamping, so as to solve the above-mentioned defects in the prior art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies in complex stamping scenarios for aluminum discs, such as the inability to simultaneously suppress wrinkling and prevent cracking, lag in dynamic response, and lack of fine-grained zone control. This invention provides a real-time gap control method and system for preventing wrinkling and cracking during aluminum disc stamping, thereby solving the aforementioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides a method for real-time gap control in preventing wrinkles and cracks during aluminum disc stamping, comprising: Step S1: Divide the pressure ring into multiple independent adjustment areas along the circumference, and set a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. Step S2: During the stamping process, the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment area are collected simultaneously and processed by a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability. Step S3: Use a lightweight discriminant model based on a fully connected neural network to analyze and judge the risk of the multidimensional state data, identify local areas and risk types with wrinkling or cracking risks, and generate gap compensation amount for each adjustment area based on the global gap reference curve and risk type. Step S4: Based on the generated gap compensation amount, drive the actuators corresponding to each adjustment area to independently fine-tune the local gap and dynamically correct the material flow resistance.

[0008] By adopting the above technical solution, the intelligent perception and feature fusion of visual and mechanical signals are achieved through the synergistic effect of the pressure ring partition design and multimodal signal processing, which can accurately capture the risk of material flow instability. By using a lightweight discrimination model based on a fully connected neural network to perform risk discrimination and partition compensation calculation, bidirectional dynamic control of wrinkle reduction and crack increase in different regions can be achieved at the same time. High-precision fine-tuning is completed through the actuator and closed-loop feedback control, which effectively overcomes the shortcomings of traditional static control in response lag and inability to simultaneously suppress wrinkling and cracking, and improves the universality of process parameters and the stability of forming quality.

[0009] Preferably, in step S1, the global clearance reference curve defines the clearance range that dynamically changes with the stamping stroke, serving as a reference trajectory for feedforward control. The gap range includes the reference single-sided gap between the punch and die and the blanking gap between the blanking ring and the die. The global clearance reference curve varies with the stamping stroke, exhibiting a segmented convergence control strategy.

[0010] By adopting the above technical solution, the global gap reference curve is used as the reference trajectory for feedforward control, and an ideal gap change trajectory is preset for the entire stamping process. The flow characteristics of the material at different forming stages are matched by a segmented convergence strategy, providing a stable reference for subsequent real-time closed-loop control and effectively reducing the initial control error.

[0011] Preferably, the control strategy for piecewise convergence specifically includes: During the pre-drawing stage, the blank holder clearance between the blank holder ring and the die is controlled to be within a loose range; In the main drawing section, the blank holder clearance between the blank holder ring and the die is controlled to converge from a loose range to a medium range. In the shaping section, the blank holder gap between the blank holder ring and the die is controlled to converge to the finishing range, while maintaining the set value of the reference single-sided gap of the punch and die.

[0012] By adopting the above technical solution, the flow resistance is reduced and cracking is prevented in the initial stage by setting the pressure gap in stages, the inflow rate and anti-wrinkle requirements are balanced in the main deformation stage, and the wrinkles are forced to be smoothed in the final stage to ensure accuracy. This achieves a deep fit between process parameters and the physical laws of material deformation.

[0013] Preferably, in step S2, the macroscopic geometric deformation characteristics are obtained through visual monitoring, including wrinkle height, edge tensile strain, and material inflow; the microscopic strain characteristics are obtained through mechanical strain monitoring, including material thinning rate.

[0014] By adopting the above technical solution, the material state is monitored from two dimensions, namely macroscopic geometric deformation and microscopic strain, through visual and mechanical sensing methods, respectively. This enables the comprehensive acquisition of multimodal data and provides a rich data foundation for the accurate identification of instability risks.

[0015] Preferably, in step S2: The wrinkling wave height is obtained by performing peak detection on the extracted contour line and calculating the height difference between the wave peak and the wave trough. The edge tensile strain and material inflow are calculated by tracking the displacement changes of preset marker points in the image; Wherein, the edge tensile strain is the relative deformation, and the calculation formula is:

[0016] in, For edge tensile strain, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; The material inflow rate is an absolute displacement, and the calculation formula is:

[0017] in, For material inflow, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; The material thinning rate is calculated based on the principle of constant volume or strain in the thickness direction.

[0018] By adopting the above technical solution, the specific calculation methods and physical meanings of each feature are clarified. In particular, the relative deformation and absolute displacement are distinguished, which ensures the quantifiability and accuracy of the feature data and provides standardized input for subsequent model processing.

[0019] Preferably, in step S2, the multimodal fusion model specifically includes: It receives macroscopic geometric feature vectors and microscopic strain feature vectors aligned with timestamps, and introduces stamping stroke stage identifiers as input; The self-attention layer automatically learns the dynamic weight allocation of different features when judging the risk of instability; The multimodal fusion model outputs a weighted fusion feature vector, which is then combined with spatial coordinate features by a splicing operator to generate multidimensional state data.

[0020] By adopting the above technical solution, a multimodal fusion model is introduced to replace the traditional simple splicing. Through the self-attention mechanism, the contribution of different features at different process stages is automatically learned, which significantly improves the intelligence level and anti-interference ability of feature fusion, and makes the generated multidimensional state data more accurately reflect the real instability risk.

[0021] Preferably, in step S3, a lightweight discriminant model based on a fully connected neural network is used to perform nonlinear mapping and pattern recognition on the multidimensional state data, and output the risk category and confidence level corresponding to each adjustment region. The risk categories include normal state, wrinkling-dominant state, cracking-dominant state, and combined risk state. Based on the global clearance reference curve and the calculated clearance compensation amount, the final clearance compensation output value of each adjustment region is determined.

[0022] By adopting the above technical solution, a lightweight discriminant model based on a fully connected neural network is used to realize nonlinear mapping and pattern recognition of multidimensional state data, which can accurately output the risk category and confidence level of each adjustment area. By defining normal, wrinkle-dominant, crack-dominant and composite risk states, a clear classification basis is provided for risk judgment under complex working conditions. The accuracy of the final gap compensation output value is ensured by superimposing the global gap reference curve and the compensation amount.

[0023] Preferably, when the output of the lightweight discrimination model is a wrinkling-dominant state, or the confidence level of the wrinkling-dominant state exceeds a first preset threshold, it is determined that there is a risk of wrinkling in the adjustment area, a compensation instruction to reduce the gap is generated, and a negative gap compensation amount is calculated. When the output of the lightweight discrimination model is a crack-dominant state or a composite risk state, or when the confidence level of the crack-dominant state exceeds the second preset threshold, it is determined that there is a crack risk in the adjustment area, a compensation command to increase the gap is generated, and the positive gap compensation amount is calculated. For adjustment regions determined to be in a normal state, control commands are output based on the global gap reference curve.

[0024] By adopting the above technical solution, a clear risk discrimination logic and corresponding compensation direction are constructed through the risk category and confidence level output by the lightweight discrimination model. It can generate precise gap compensation instructions for different risk states and effectively achieve precise control of material flow resistance.

[0025] Preferably, in step S4, the actuator uses a closed-loop feedback control method to independently fine-tune each adjustment region; The actuator compares the final gap compensation output value of each adjustment area determined in step S3 with the actual gap value of each adjustment area fed back by the sensor to generate a position error signal for each adjustment area. Based on the position error signal, the control quantity is calculated to drive the actuator of the corresponding adjustment area, so that the actual gap value of each adjustment area approaches the final gap compensation output value determined in step S3.

[0026] By adopting the above technical solution, each adjustment region is independently fine-tuned through closed-loop feedback control, which realizes the accurate generation and processing of position error signals. This effectively drives the actuators of the corresponding adjustment regions, ensuring that the actual gap value approaches the target value, thereby significantly improving the accuracy and stability of gap control.

[0027] Secondly, this application also provides a real-time gap control system for preventing wrinkles and cracks during aluminum disc stamping, comprising: The physical structure partitioning module divides the pressure ring into multiple independent adjustment areas along the circumference, and sets a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. The multimodal state perception module synchronously collects the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment area during the stamping process, and processes them through a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability. The risk assessment and gap compensation module uses a lightweight assessment model based on a fully connected neural network to analyze and assess the risk of the multidimensional state data, identify local areas and risk types that have the risk of wrinkling or cracking, and generate the gap compensation amount for each adjustment area based on the global gap reference curve and risk type. The servo execution and closed-loop control module drives the actuators corresponding to each adjustment area based on the generated gap compensation amount, and independently fine-tunes the local gap to dynamically correct the material flow resistance.

[0028] The beneficial effects of this invention are as follows: Through the synergistic effect of the pressure ring partition design and the independent actuator, bidirectional dynamic control of wrinkle reduction and crack increase in different areas can be performed simultaneously, improving the accuracy of suppressing complex flow differences; by using a multimodal fusion model combining visual and mechanical sensing, and introducing a self-attention mechanism to dynamically allocate feature weights, the limitations of misjudgment by a single sensor are overcome, enabling accurate capture of material flow instability risks and providing a high-confidence quantitative basis for control decisions; by using a global gap reference curve that converges segmentally with the stamping stroke as the reference trajectory for feedforward control, and combining it with closed-loop feedback control to correct the gap in real time, the response lag is effectively reduced, allowing process parameters to deeply match the physical laws of the material at different deformation stages, improving the stability of forming quality and the universality of the process; by using risk category determination and confidence threshold logic based on a lightweight discrimination model, the discrimination logic of wrinkle and crack risks and the corresponding compensation direction are clarified, ensuring the safety of the forming process and avoiding defects caused by control logic conflicts.

[0029] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects.

[0030] Therefore, it is evident that the present invention has substantial features and progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for real-time gap control in the stamping of aluminum discs to prevent wrinkling and cracking.

[0033] Figure 2 This is a schematic diagram of a real-time gap control system for preventing wrinkles and cracks during aluminum disc stamping.

[0034] Among them, 1-physical structure partitioning module, 2-multimodal state perception module, 3-risk judgment and gap compensation module, and 4-servo execution and closed-loop control module. Detailed Implementation

[0035] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0036] Example 1: This embodiment provides a method for real-time gap control in aluminum disc stamping to prevent wrinkling and cracking, such as... Figure 1 As shown, it includes the following steps: Step S1: Divide the pressure ring into multiple independent adjustment areas along the circumference, and set a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. Step S2: During the stamping process, the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment area are simultaneously collected and processed through a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability; Step S3: Use a lightweight discriminant model based on a fully connected neural network to analyze and judge the risk of the multidimensional state data, identify local areas and risk types with wrinkling or cracking risks, and generate gap compensation amount for each adjustment area based on the global gap reference curve and risk type. Step S4: Based on the generated gap compensation amount, drive the actuators corresponding to each adjustment area to independently fine-tune the local gap and dynamically correct the material flow resistance.

[0037] Hereinafter, according to embodiments of the present invention, steps S1 to S4 will be specifically described.

[0038] In step S1, the core task is to divide the blank holder ring of the stamping die into multiple independent adjustment areas along the circumference, and set a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. In step S1, it is necessary to complete the logical division of the blank holder ring adjustment area, the analysis of the material parameters of the target aluminum disc, and the definition of the ideal clearance value range covering the entire stamping stroke, so as to provide a physical carrier and digital reference benchmark for subsequent real-time monitoring and dynamic adjustment. In this embodiment, the pressure ring is a key force transmission component connecting the press slide and the die, mainly used to control the resistance of material flowing into the die. To solve the problem that traditional integral pressure rings cannot cope with uneven circumferential flow, anisotropy, and uneven lubrication leading to local wrinkling, step S1 involves a partitioned design for the pressure ring. Based on the diameter of the aluminum disc, the pressure ring is divided into 6 to 12 independent adjustment areas. Each adjustment area is mechanically independent, and its bottom or side is equipped with a precision micro-motion actuator. This design, which decouples the traditional integral rigid constraint into a distributed flexible constraint, allows each adjustment area to independently perform gap fine-tuning. The micro-motion actuator specifically includes a servo motor with a ball screw, a wedge mechanism, or a piezoelectric ceramic stack.

[0039] Furthermore, step S1 also sets a global clearance reference curve that dynamically changes with the stamping process based on the material parameters of the target aluminum disc; the global clearance reference curve serves as a reference trajectory for feedforward control, defining the law governing the change of the clearance between the blank holder and the die with the stamping stroke under ideal conditions, specifically including: Obtain the material parameters of the aluminum disc to be stamped, including aluminum alloy grades representing different yield strengths and hardening indices, as well as geometric parameters including the initial thickness t of the aluminum disc; based on the above material parameters, set the reference single-sided clearance Z between the punch and die and the initial blank holder clearance between the blank holder ring and the die. The theoretical range of the reference single-sided clearance Z between the punch and die is 1.05t to 1.20t. For softer aluminum materials or shallow drawing conditions, the lower limit of 1.05t is taken to ensure forming accuracy, while for harder aluminum materials or deep drawing conditions, the upper limit of 1.20t is taken to prevent tearing. The initial blank holder clearance between the blank holder ring and the die is also considered. The value is set to 1.1t to 1.3t as the starting point for dynamic control.

[0040] In the embodiments of this application, the global clearance reference curve that dynamically changes with the stamping stroke S is specifically manifested as a segmented convergence control strategy, where the stamping stroke S represents the percentage of the slide stroke to the total stroke; in the pre-drawing section where 0%≤S<30%, the initial blank holder clearance is controlled. In a relatively loose range, the flange area is large and the material has good fluidity. The main purpose is to reduce flow resistance and prevent the material from cracking due to excessive tensile stress in the initial stage. In the main drawing section (30% ≤ S < 80%), the initial blank holder gap is controlled. The material flow linearly converges from the loose range to the medium range. This stage is characterized by the most intense material flow and is also the peak period for wrinkling. Dynamic stability is achieved by gradually tightening the gap to balance the material inflow rate and anti-wrinkle requirements. In the shaping section (80%≤S≤100%), the edge clamping gap is controlled. The process converges to the finishing range while maintaining the set value for the single-sided clearance Z of the punch and die reference. The main purpose of this stage is to force the sidewalls to flatten, eliminate residual wrinkles, and ensure the final dimensional accuracy and roundness of the workpiece.

[0041] In the embodiments of this application, the loose range, medium range, and fine range of the blanking gap are defined according to the initial thickness t of the aluminum disc: The relaxed range is set to 1.2-1.5t, preferably 1.3t, for the pre-drawing section, aiming to significantly reduce material flow resistance and ensure a smooth material start-up; The medium range is set to 1.1-1.2t, which is suitable for the main drawing section. It serves as a dynamic convergence area for the transition from the loose range to the finishing range, and is used to balance the material inflow rate and anti-wrinkle requirements. The finishing range is set to 1.05-1.1t, which is compatible with the lower limit of the reference single-sided clearance Z between the punch and die; the finishing range is used for the forming section, which aims to force the smoothing of sidewall wrinkles through a smaller clearance, so as to ensure the final dimensional accuracy of the workpiece.

[0042] Thus, step S1 completes the transformation from a physical mold structure to a partitioned adjustable structure, and generates a global gap reference curve in the digital space to guide the entire process; step S1 establishes the action space and reference trajectory for subsequent real-time control, providing a stable and consistent environmental basis and constraint boundary for multimodal state perception in step S2 and gap compensation decision in step S3.

[0043] In step S2, the core task is to simultaneously collect the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment region, and fuse them to generate multidimensional state data reflecting the risk of material flow instability. Step S2 requires spatiotemporal alignment and fusion of visual monitoring data and mechanical strain monitoring data to form a quantitative description of the current material flow state, thereby providing a high-confidence input signal for subsequent risk decisions. In this embodiment of the application, in order to achieve accurate capture of the stamping state of aluminum discs, step S2 constructs a real-time monitoring component covering the entire field of view and key points. The real-time monitoring component includes a high-speed industrial camera arranged above the mold and an array of piezoelectric sensors and strain sensors distributed on the blank holder or die. After the stamping stroke is started, the status data of the adjustment area is collected synchronously through the real-time monitoring component. The status data includes visual monitoring data and mechanical strain monitoring data. In order to ensure the synchronization of multi-source data in the time domain, hardware triggering or high-precision timestamp alignment technology is used to ensure that the visual image frame and the sensor sampling point are strictly corresponding. The visual monitoring data is used to reflect macroscopic geometric deformation characteristics. Specifically, it is obtained by taking pictures of the flange receiving line and side wall contour with a high-speed camera and analyzing the original pixel matrix through image processing algorithms. The data includes wrinkling wave height, edge tensile strain and material inflow, which intuitively reflects whether the material has undergone macroscopic instability, that is, the occurrence of wrinkling. The wrinkling wave height is calculated by performing peak detection on the extracted contour line and calculating the height difference between the wave peak and the wave trough. The calculation expression is as follows:

[0044] in, The set of pixel positions representing each point on the contour line is converted into physical height after calibration by camera intrinsic and extrinsic parameters. The edge tensile strain With material inflow It is obtained by tracking the displacement changes of preset marker points in the image; The edge tensile strain is a relative deformation, used to characterize the degree of tensile deformation at the edge of the sheet metal. The calculation formula is as follows:

[0045] in, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; Material inflow This is an absolute displacement, used to characterize the actual flow distance of the blank in the flange area into the mold cavity. The calculation formula is:

[0046] in, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; The mechanical strain monitoring data is used to reflect microscopic strain characteristics. Specifically, strain gauges attached to the surface of the mold cavity or the edge of the disc are used to monitor the plastic deformation of the material in real time. The raw signals are then converted and obtained through strain acquisition cards or digital image correlation algorithms, including the material thinning rate. It is used to warn of internal damage that is not visible to the naked eye, i.e., the risk of cracking.

[0047] The material thinning rate is calculated based on the principle of constant volume or strain in the thickness direction, and the calculation formula is as follows:

[0048] in, For the initial thickness, True strain in the thickness direction; Furthermore, macroscopic geometric deformation characteristics and microscopic strain characteristics are fused together to generate multidimensional state data that reflects the risk of material flow instability. In this embodiment of the application, in order to achieve more accurate feature association and noise suppression, step S2 adopts a multimodal fusion model based on the attention mechanism to realize feature fusion; Specifically, the timestamp-aligned macroscopic geometric feature vectors With micro-strain eigenvectors As the input sequence, the current stamping stroke stage identifier is introduced. The input is then fed into the trained multimodal fusion model; The multimodal fusion model automatically learns the dynamic weight allocation of different features when judging instability risk through a self-attention layer. This process is constrained by prior process knowledge, namely, strengthening the monitoring of material inflow rate in the main drawing section where material flow is intense. The focus is on enhancing the material thinning rate in the shaping segment. This increases the level of attention, thereby overcoming the limitations of traditional fixed-weight fusion in complex working conditions; The multimodal fusion model outputs a high-dimensional feature vector after weighted fusion, which is then integrated using the concatenation operator ⊕ to combine spatial coordinate features. This maps the feature data to its corresponding physical location, generating multidimensional state data S(t):

[0049] in, This represents the fused feature vector output by the model. Includes the position codes of each adjustment zone ; It should be noted that the multimodal fusion model is trained through supervised learning using historical stamping data. The training dataset contains no fewer than 500 stamping samples, in which the ratio of wrinkled, cracked, and qualified samples is no less than 1:1:2; thus, it has stronger anti-interference ability and more accurate instability prediction ability than simple mathematical splicing; the historical stamping data includes samples marked as wrinkled, cracked, or qualified. Thus, step S2 completes the conversion from material deformation to state data. By using multimodal signal fusion, the problem of high misjudgment rate of a single sensor is overcome, ensuring the effective operation of the subsequent "monitoring-analysis-adjustment" closed loop.

[0050] In step S3, the core task is to use a lightweight discriminant model based on a fully connected neural network to analyze and judge the risk of the multidimensional state data, identify local areas and risk types with wrinkling or cracking risks, and generate gap compensation amounts for each adjustment area based on the global gap reference curve and risk type.

[0051] The global gap reference curve As a feedforward controller, it defines the ideal trajectory of the blank holder clearance as a function of the stamping stroke S under risk-free conditions; under most normal operating conditions, it will directly output the global clearance reference curve. The value serves as a control command, thereby ensuring that the loose range of the pre-drawing section, the convergent transition of the main drawing section, and the finishing range of the shaping section are strictly executed; this design internalizes the physical laws of material deformation into the default behavior of the control system, reducing the correction burden of feedback control. The multidimensional state data S(t) is analyzed and risk-discriminated to identify risk types and locations. Specifically, a pre-built lightweight discrimination model is used to perform nonlinear mapping and pattern recognition on the high-dimensional feature combinations in the multidimensional state data, outputting the risk category and confidence level corresponding to each adjustment region. The risk category is uniquely determined by the distribution pattern of the multidimensional state data in the fused feature space, rather than by a single geometric or mechanical feature. The risk categories include normal state, wrinkling-dominated state, cracking-dominated state, and composite risk state. Furthermore, the lightweight discrimination model is implemented using a classifier structure based on a fully connected neural network, including a feature normalization layer, a nonlinear mapping layer, and a risk classification layer; The feature normalization layer standardizes the input multidimensional state data S(t) to eliminate the interference of different feature units on model training. The specific expression is as follows:

[0052] Where S(t) represents multidimensional state data; and These are the mean and standard deviation vectors of the corresponding features in the training dataset, respectively; The nonlinear mapping layer performs a nonlinear transformation on the normalized feature vector through one or more fully connected layers. For the l-th hidden layer, its forward propagation calculation formula is:

[0053] in, The output of the previous layer is the first layer's input. ; and These are the weight matrix and the bias vector, respectively. ( ) represents a nonlinear activation function; This process achieves a nonlinear mapping from the fused feature space to the risk feature space, and automatically learns the high-order coupling relationship between macroscopic geometric features and microscopic strain features.

[0054] The risk classification layer uses the Softmax function as the classifier, and the output of the last hidden layer is used as the classifier. The probability distribution is mapped to four preset risk categories; its calculation formula is:

[0055] Where j∈{1,2,3,4} corresponds to the normal state, wrinkle-dominant state, crack-dominant state and composite risk state, respectively. and Let P be the weight vector and bias scalar corresponding to the j-th class; the maximum value in the output vector P is the risk class determined by the lightweight discriminant model, and this maximum value itself is the corresponding confidence level.

[0056] The parameter set of the model ={ , , , The dataset is obtained through supervised learning training using historical stamping datasets. The optimization objective is typically to minimize the cross-entropy loss function.

[0057] Where M is the number of training samples. Let i be the true label of the i-th sample; The historical stamping dataset contains a large number of stamping samples collected under different process parameters. Each sample contains corresponding multidimensional state data S(t) and real risk labels determined by manual annotation or based on high-precision offline simulation. The stamping samples include wrinkling-dominant state samples, cracking-dominant state samples, composite risk state samples, and normal state samples. The wrinkling-dominant state sample corresponds to a working condition where macroscopic instability is obvious but microscopic damage does not exceed limits. Its characteristic combination is manifested by wrinkling wave height. Significantly higher than normal levels, while the material thinning rate Below the critical thinning rate threshold, the workpiece surface shows obvious wrinkles but no signs of necking or microcracks; the crack-dominant state corresponds to a condition of severe micro-thinning but no obvious wrinkling, and its characteristic combination is manifested in the material thinning rate. Exceeding the threshold for the thinning rate, and the wrinkling wave height Maintaining at a low level, the thickness of the workpiece decreases sharply in local areas but does not form obvious surface wrinkles; the composite risk state sample simultaneously exhibits significant wrinkling characteristics and high thinning rate characteristics, i.e., wrinkling wave height. With material thinning rate All exceeded the corresponding thresholds, indicating that the workpiece had both macroscopic instability and wrinkles, as well as the risk of breakage due to local overstretching; the wrinkling wave height and material thinning rate of the normal state sample were within the safe range, and the forming quality was good.

[0058] The historical stamping dataset is used to train a lightweight discriminative model for supervised learning, enabling the model to learn the distribution boundaries of different risk patterns in the fused feature space of multidimensional state data.

[0059] When the lightweight discrimination model outputs a wrinkling-dominant state, or the confidence level of the wrinkling-dominant state exceeds a first preset threshold, the sector is determined to have a wrinkling risk; during this process, the wrinkling wave height... As a key feature factor affecting the model output, the maximum allowable wave height threshold in this example is equivalent to the first preset threshold after model training or calibration. When the lightweight discrimination model outputs a crack-dominant state or a combined risk state, or when the confidence level of the crack-dominant state exceeds a second preset threshold, the sector is determined to have a crack risk; during this process, the material thinning rate... As a key feature factor affecting the model output, the corresponding limit thinning rate threshold in this example is equivalent to the second preset threshold after model training or calibration.

[0060] Furthermore, the first preset threshold and the second preset threshold can be calibrated through grid search, Bayesian optimization, or performance evaluation based on the validation set to achieve a balance between false positive rate and false negative rate.

[0061] Furthermore, based on the different risk types identified above, and according to the global gap baseline curve... Calculate the specific clearance compensation amount based on the current stamping stroke S. For sector i that is determined to have a risk of wrinkling, a compensation command to reduce the gap is generated, at which point the compensation amount is... The calculation logic is a negative adjustment:

[0062] Where k is the gain coefficient, set to 0.02-0.05mm, used to adjust the response intensity; the compensation command to reduce the gap will drive the actuator to reduce the pressure gap in that area. This increases the resistance to material flow and inhibits wrinkling; For sector i identified as having a risk of cracking, a compensation command to increase the gap is generated; the compensation amount... The calculation logic is a positive adjustment:

[0063] Where k' is the gain coefficient, the compensation command to increase the gap will drive the actuator to increase the pressure gap in this region. This reduces material flow resistance, promotes material inflow, and prevents tearing. For areas where no risk was detected, the global gap reference curve set in step S1 is used. It outputs instructions to maintain the current gap or make minor adjustments to ensure the stability of the overall process.

[0064] Final gap output value Determined by both the baseline curve value and the compensation amount:

[0065] Thus, step S3 completes the conversion from risk identification to control commands, and outputs a set of partition gap compensation commands containing the target gap values ​​of each sector, which serves as the direct input to the servo actuator in step S4, ensuring the effective operation of the "monitoring-analysis-adjustment" closed loop.

[0066] In step S4, the core task is to drive the actuators corresponding to each adjustment area according to the generated gap compensation amount, to independently fine-tune the local gap, and dynamically correct the material flow resistance. Step S4 needs to convert the digital control command into physical displacement and complete the action execution within a millisecond response time to ensure the real-time performance and stability of the closed-loop control.

[0067] In this embodiment, to achieve high-precision, high-frequency response control of the gap between the pressure rings, step S4 deploys an execution component. This execution component corresponds one-to-one with the adjustment areas defined in step S1, specifically including a servo drive unit and a precision transmission mechanism, as well as a grating ruler or displacement sensor for feedback of actual displacement. The combination of the servo drive unit and the precision transmission mechanism, and the design concept of independent zone control of the pressure rings, both adopt a segmented pressure ring device or a circumferentially distributed execution scheme. The servo drive unit includes a servo driver and a servo motor; the precision transmission mechanism includes a ball screw or a wedge mechanism. Receive the final gap compensation value of each sector calculated in step S3. ,Will The actual gap currently fed back by the sensor The comparison is performed to generate a position error signal; The position error signal is amplified by the servo driver, driving the servo motor to rotate. The motor converts the rotational motion into linear displacement through a ball screw mechanism or wedge mechanism, pushing the corresponding adjustment area to move vertically. For areas at risk of wrinkling, the actuator pushes the sector downward to reduce the edge clamping gap. For areas at risk of cracking, the actuator retracts upwards to increase the pressure clearance. The entire execution process completes response and positioning within 10ms to 50ms, with the actual gap after adjustment provided in real time via a grating ruler. This forms a local closed loop of instruction-execution-feedback, ensuring that the gap adjustment accuracy reaches the micrometer level; its control logic is expressed as:

[0068] in, For output control quantity, For gap error, , , These are the adjustment parameters for the PID controller.

[0069] Based on the position error signal and the calculated control quantity Drive the servo motor to operate, so that the actual gap value Approaching the final gap compensation output value .

[0070] Thus, step S4 completes the final implementation from digital instructions to physical gap adjustment. Through high-frequency servo execution and precision micro-motion, the material flow resistance is dynamically corrected. The output of step S4 directly affects the material deformation process. Together with the monitoring in step S2 and the decision-making in step S3, it forms a complete real-time closed loop of monitoring-analysis-adjustment, ensuring the anti-wrinkle and anti-crack stability of the aluminum disc stamping process.

[0071] Example 2: This embodiment provides a real-time gap control system for preventing wrinkles and cracks during aluminum disc stamping, such as... Figure 2 As shown, it includes: The physical structure partitioning module 1 divides the pressure ring into multiple independent adjustment areas along the circumference, and sets a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. The multimodal state perception module 2 synchronously collects the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment area during the stamping process, and processes them through a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability; The risk discrimination and gap compensation module 3 uses a lightweight discrimination model based on a fully connected neural network to analyze and discriminate the multidimensional state data, identify local areas and risk types with wrinkling or cracking risks, and generate gap compensation amounts for each adjustment area based on the global gap reference curve and risk type. The servo execution and closed-loop control module 4 drives the actuators corresponding to each adjustment area based on the generated gap compensation amount, and independently fine-tunes the local gap to dynamically correct the material flow resistance.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0073] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0074] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0075] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0076] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0077] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0078] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0079] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0080] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for real-time gap control in aluminum disc stamping to prevent wrinkling and cracking, characterized in that, Includes the following steps: Step S1: Divide the pressure ring into multiple independent adjustment areas along the circumference, and set a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. Step S2: During the stamping process, the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment area are collected simultaneously and processed through a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability; Step S3: Use a lightweight discriminant model based on a fully connected neural network to analyze and judge the risk of the multidimensional state data, identify local areas and risk types with wrinkling or cracking risks, and generate gap compensation amount for each adjustment area based on the global gap reference curve and risk type. Step S4: Based on the generated gap compensation amount, drive the actuators corresponding to each adjustment area to make independent fine adjustments and dynamically correct the material flow resistance.

2. The method according to claim 1, characterized in that, In step S1, the global clearance reference curve defines the clearance range that changes dynamically with the stamping stroke, serving as a reference trajectory for feedforward control. The gap range includes the reference single-sided gap between the punch and die and the blanking gap between the blanking ring and the die. The global clearance reference curve varies with the stamping stroke, exhibiting a segmented convergence control strategy.

3. The method according to claim 2, characterized in that, The control strategy for piecewise convergence specifically includes: During the pre-drawing stage, the blank holder clearance between the blank holder ring and the die is controlled to be within a loose range; In the main drawing section, the blank holder clearance between the blank holder ring and the die is controlled to converge from a loose range to a medium range. In the shaping section, the blank holder gap between the blank holder ring and the die is controlled to converge to the finishing range, while maintaining the set value of the reference single-sided gap of the punch and die.

4. The method according to claim 1, characterized in that, In step S2, the macroscopic geometric deformation characteristics are obtained through visual monitoring, including wrinkle height, edge tensile strain, and material inflow; the microscopic strain characteristics are obtained through mechanical strain monitoring, including material thinning rate.

5. The method according to claim 4, characterized in that, In step S2, the wrinkling wave height is obtained by performing peak detection on the extracted contour line and calculating the height difference between the wave peak and the wave trough. The edge tensile strain and material inflow are calculated by tracking the displacement changes of preset marker points in the image; Wherein, the edge tensile strain is the relative deformation, and the calculation formula is: in, For edge tensile strain, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; The material inflow rate is an absolute displacement, and the calculation formula is: in, For material inflow, The initial spacing between the marked points. This refers to the real-time spacing of the marker points during the stamping process; The material thinning rate is calculated based on the principle of constant volume or strain in the thickness direction.

6. The method according to claim 5, characterized in that, In step S2, the multimodal fusion model specifically includes: It receives macroscopic geometric feature vectors and microscopic strain feature vectors aligned with timestamps, and introduces stamping stroke stage identifiers as input; The self-attention layer automatically learns the dynamic weight allocation of different features when judging the risk of instability; The multimodal fusion model outputs a weighted fusion feature vector, which is then combined with spatial coordinate features by a splicing operator to generate multidimensional state data.

7. The method according to claim 1, characterized in that, In step S3, a lightweight discriminant model based on a fully connected neural network is used to perform nonlinear mapping and pattern recognition on the multidimensional state data, and output the risk category and confidence level corresponding to each adjustment region. The risk categories include normal state, wrinkling-dominant state, cracking-dominant state, and combined risk state.

8. The method according to claim 7, characterized in that, When the output of the lightweight discrimination model is a wrinkle-dominant state or the confidence level of the wrinkle-dominant state exceeds the first preset threshold, it is determined that there is a risk of wrinkling in the adjustment area, a compensation instruction to reduce the gap is generated, and the negative gap compensation amount is calculated. When the output of the lightweight discrimination model is a crack-dominant state or a composite risk state, or when the confidence level of the crack-dominant state exceeds the second preset threshold, it is determined that there is a crack risk in the adjustment area, a compensation command to increase the gap is generated, and the positive gap compensation amount is calculated. For adjustment areas determined to be in a normal state, control commands are output based on the global gap reference curve; Based on the global clearance reference curve and the calculated clearance compensation amount, the final clearance compensation output value of each adjustment region is determined.

9. The method according to claim 8, characterized in that, In step S4, the actuator uses a closed-loop feedback control method to independently fine-tune each adjustment region; The actuator compares the final gap compensation output value of each adjustment area determined in step S3 with the actual gap value of each adjustment area fed back by the sensor to generate a position error signal for each adjustment area. Based on the position error signal, the control quantity is calculated to drive the actuator of the corresponding adjustment area, so that the actual gap value of each adjustment area approaches the final gap compensation output value determined in step S3.

10. A real-time gap control system for preventing wrinkles and cracks during aluminum disc stamping, characterized in that, include: The physical structure partitioning module divides the pressure ring into multiple independent adjustment areas along the circumference, and sets a global clearance reference curve that dynamically changes with the stamping stroke based on the material parameters of the target aluminum disc. The multimodal state perception module simultaneously collects the macroscopic geometric deformation characteristics and microscopic strain characteristics of each adjustment region during the stamping process, and processes them through a multimodal fusion model to generate multidimensional state data reflecting the risk of material flow instability. The risk assessment and gap compensation module uses a lightweight assessment model based on a fully connected neural network to analyze and assess the risk of the multidimensional state data, identify local areas and risk types that have the risk of wrinkling or cracking, and generate the gap compensation amount for each adjustment area based on the global gap reference curve and risk type. The servo execution and closed-loop control module drives the actuators corresponding to each adjustment area based on the generated gap compensation amount, and independently fine-tunes the local gap to dynamically correct the material flow resistance.

Images