A sheet metal stamping process parameter optimization method based on machine learning

Abstract

The application discloses a sheet metal stamping process parameter optimization method based on machine learning, and relates to the technical field of intelligent manufacturing. The method comprises the following steps: acquiring forming design information, material batch information, die forming information and stamping equipment constraint information of a to-be-stamped sheet metal part; dividing a deep-drawing risk area, a bending springback area and an edge forming area, and determining a material forming window, a die safety boundary and an equipment capacity boundary to form a regional initial forming deviation field; taking the regional initial forming deviation field and historical real forming deviation as a training basis, constructing a machine learning model for correcting sheet metal stamping forming deviation, learning the corresponding relationship between finite element simulation deviation and on-site stamping deviation in different forming areas, and forming a corrected forming deviation field adapted to current production conditions. The application forms the corrected forming deviation field, so that the candidate stamping parameter scheme is reversely selected around the deviation position, the deviation type and the deviation degree, and the regional pertinence of the parameter adjustment process is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method for optimizing sheet metal stamping process parameters based on machine learning. Background Technology

[0002] Sheet metal stamping is widely used in the manufacturing of electronic product structural parts, metal molds, mechanical equipment parts, and hardware stamping parts. The stamping quality is typically affected by material batch, mold condition, equipment operation, and process parameters. With the development of computer-aided engineering, finite element simulation, digital process planning, and machine learning technologies, the optimization of stamping process parameters is gradually shifting from experience-based setting to data-driven modeling. Finite element simulation can obtain the forming profile, drawing state, bending state, and springback trend in a virtual environment. Machine learning can use historical production data to learn the correlation between process parameters and forming quality, providing a computational basis for stamping process optimization.

[0003] Existing sheet metal stamping parameter optimization technologies typically use simulation or inspection results of the entire stamped part as the evaluation object, focusing on final quality indicators such as cracking, wrinkling, springback, or dimensional deviations. This makes it difficult to establish a stable correspondence between simulation deviations and actual deviations in different forming regions. When material batch fluctuations, die condition changes, or slider movement deviations coexist, existing machine learning methods easily mix forming offsets from different sources into a single quality deviation. This results in a lack of regionalized deviation criteria for selecting candidate stamping parameters, and the parameter schemes obtained through virtual verification may still have adaptation discrepancies with actual production conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a machine learning-based method for optimizing sheet metal stamping process parameters to solve the problem of difficulty in regionalizing and correcting virtual forming deviations and on-site stamping deviations.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a machine learning-based method for optimizing sheet metal stamping process parameters. The method includes: acquiring forming design information, material batch information, die forming information, and stamping equipment constraint information for the sheet metal part to be stamped; dividing the drawing risk area, bending springback area, and edge forming area; and determining the material forming window, die safety boundary, and equipment capability boundary to form a regionalized initial forming deviation field; using the regionalized initial forming deviation field and historical real forming deviations as training basis, constructing a machine learning model to correct sheet metal stamping forming deviations; learning the correspondence between finite element simulation deviations and on-site stamping deviations in different forming areas to form a corrected forming deviation field adapted to the current production conditions; and based on the corrected forming deviation field... The candidate stamping parameter scheme set is selected by reverse selection of deviation position, deviation type and deviation degree. Virtual forming verification is carried out on the candidate stamping parameter scheme set to extract the stroke correspondence between virtual stamping force change and virtual slider displacement change to obtain pressure-displacement process fingerprint. The pressure-displacement process fingerprint is matched with the correction forming deviation field to identify the forming offset source corresponding to material batch fluctuation, mold state change and slider movement deviation. Parameter adjustment schemes are screened within the material forming window, mold safety boundary and equipment capability boundary. The parameter adjustment schemes are input into the machine learning model for iterative verification to obtain iterative verification deviation. The iterative verification deviation is used to correct the correction forming deviation field and output the sheet metal stamping process parameter optimization scheme.

[0007] As a preferred embodiment of the sheet metal stamping process parameter optimization method based on machine learning described in this invention, the step of forming a regionalized initial forming deviation field is as follows: The forming design information is digitally analyzed to determine the forming structure area and the target forming reference, and to form a mapping relationship between the forming structure area; The mapping relationship of the forming structure region is associated with the material batch information, mold forming information and stamping equipment constraint information to determine the material forming window, mold safety boundary and equipment capability boundary corresponding to each forming structure region, and thus obtain the forming region data subject to forming constraints. Virtual forming pre-analysis is performed on the forming area data to obtain the virtual forming state, and the regional simulation deviation is obtained by comparing it with the target forming benchmark. The deep drawing risk area, bending springback area and edge forming area are divided according to the deviation position and forming structure area type to form a regionalized initial forming deviation field.

[0008] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the steps of constructing a machine learning model to correct sheet metal stamping forming deviations are as follows: Using the deep drawing risk area, bending springback area and edge forming area in the regionalized initial forming deviation field as nodes, and combining the material flow relationship, force transmission relationship and process influence relationship of adjacent areas, a forming area relationship diagram is constructed, and the regional simulation deviation is mapped to the forming area relationship diagram to obtain the regional simulation deviation characteristics. The historical true forming deviation of the stamped parts in the historical stamping production is obtained by comparing the forming state of the stamped parts with the target forming reference, and then collected according to the nodes in the forming area relationship diagram to form the true deviation characteristics of the area. By comparing the actual deviation characteristics of the region with the simulated deviation characteristics of the region region by region, the region deviation correction amount is obtained. Then, the region deviation training sample is obtained by associating it with material batch information, mold forming information and stamping equipment constraint information. A regional deviation residual correction model is constructed, which includes a regional feature input layer, a forming condition fusion layer, a regional relationship propagation layer, and a deviation residual output layer. The regional deviation residual correction model is trained using regional deviation training samples to obtain a machine learning model for correcting sheet metal stamping forming deviations.

[0009] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the step of forming a corrected forming deviation field adapted to the current production conditions is as follows: The machine learning model that inputs the regionalized initial forming deviation field into the machine learning model to correct sheet metal stamping forming deviation is used. The regional feature input layer extracts the regional type and deviation features. The forming condition fusion layer integrates material batch information, mold forming information and stamping equipment constraint information. The regional relationship propagation layer transmits the influence of adjacent regions. The deviation residual output layer outputs the predicted regional deviation correction amount of each forming region. The predicted regional deviation correction amount is backfilled into the regionalized initial forming deviation field to generate a corrected forming deviation field adapted to the current production conditions.

[0010] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the step of selecting a set of candidate stamping parameter schemes is as follows: Based on the corrected forming deviation field, the deviation position, deviation type and deviation degree corresponding to the deep drawing risk area, bending springback area and edge forming area are extracted and mapped to the material forming window, mold safety boundary and equipment capability boundary to obtain the candidate parameter retrieval space; A deviation reverse index table is established based on the candidate parameter retrieval space. The deviation type of each region is matched with the stamping parameters jointly defined by the stamping equipment constraint information, die forming information and material forming window through the deviation reverse index table, forming a set of candidate stamping parameter schemes.

[0011] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the steps for obtaining the pressure-displacement process fingerprint are as follows: The candidate stamping parameter scheme set is input into the virtual forming verification environment, the forming region data constrained by forming and the corrected forming deviation field are called, and the virtual stamping stroke is simulated for each candidate stamping parameter scheme to form candidate virtual forming verification data. Extract the virtual stamping force change and virtual slider displacement change corresponding to each candidate stamping parameter scheme from the candidate virtual forming verification data, and identify the contact start point, maximum forming load point, pressure holding stability point and unloading end point in the virtual stamping stroke to obtain stroke anchor point alignment data; Based on the stroke anchor point alignment data, the virtual punching force change and virtual slider displacement change are segmented, aligned, and normalized to obtain pressure-displacement alignment data. The stroke correspondence between the virtual punching force change and the virtual slider displacement change is extracted to obtain the pressure-displacement process fingerprint.

[0012] As a preferred embodiment of the sheet metal stamping process parameter optimization method based on machine learning described in this invention, the steps for identifying the sources of forming offset corresponding to material batch fluctuations, die state changes, and slider movement deviations are as follows: The pressure-displacement process fingerprint is segmented and analyzed to extract the changes in the pressure slope, slider displacement hysteresis, and pressure-displacement coupling in the contact initiation segment, load rise segment, pressure holding and stabilization segment, and unloading segment, thereby obtaining the fingerprint abnormal segments. The fingerprint anomaly segment is matched with the deep drawing risk area, bending springback area and edge forming area in the correction forming deviation field to determine the deviation concentration area. Based on the material batch fluctuation, mold state change and slider movement deviation corresponding to the deviation concentration area, the forming offset source is determined.

[0013] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the steps of the parameter adjustment scheme are as follows: Based on the source of forming offset, the corresponding material forming window, mold safety boundary, or equipment capability boundary is called, and the deviation type and stamping parameter correspondence in the deviation reverse index table are filtered according to the source of forming offset to form a source constraint parameter set; By cross-filtering the source constraint parameter set and the candidate stamping parameter scheme set, the parameter combination that can reduce the degree of forming deviation in the deviation concentration area and meet the material forming window, mold safety boundary and equipment capacity boundary is retained to generate parameter adjustment scheme.

[0014] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the step of obtaining the iterative verification deviation is as follows: Write the stamping parameter items and stamping parameter values ​​corresponding to the parameter adjustment scheme into the production condition features, and associate the source constraint parameter set, deviation concentration area and pressure-displacement process fingerprint with the correction forming deviation field to form iterative verification input data; The iterative verification input data is input into the machine learning model for correcting sheet metal stamping forming deviation. The deviation prediction of each region after the parameter adjustment scheme is applied is performed and compared with the corrected forming deviation field to obtain the iterative verification deviation.

[0015] As a preferred embodiment of the machine learning-based sheet metal stamping process parameter optimization method of the present invention, the steps of outputting the sheet metal stamping process parameter optimization scheme are as follows: Based on the iterative verification deviation, locate the remaining deviation region in the correction forming deviation field that does not meet the preset passing threshold, and correct the correction forming deviation field according to the deviation type and degree corresponding to the remaining deviation region; The corrected forming deviation field is bound to the parameter adjustment scheme that meets the preset threshold and the pressure-displacement process fingerprint to generate a sheet metal stamping process parameter optimization scheme.

[0016] The beneficial effects of this invention are as follows: By constructing a machine learning model for correcting sheet metal stamping forming deviations, it is possible to learn the correspondence between the regionalized initial forming deviation field and the historical actual forming deviations, so that the drawing risk area, bending springback area and edge forming area can obtain deviation correction basis adapted to the current production conditions; by forming a corrected forming deviation field, the candidate stamping parameter schemes can be selected in reverse around the deviation position, deviation type and deviation degree, improving the regional targeting of the parameter adjustment process, and the corrected forming deviation field can be used for virtual forming verification and iterative verification, improving the adaptability of the sheet metal stamping process parameter optimization scheme to the current production conditions, and enhancing the forming stability and boundary controllability during the parameter adjustment process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a machine learning-based method for optimizing sheet metal stamping process parameters.

[0019] Figure 2 A flowchart for generating a correction forming deviation field.

[0020] Figure 3 A flowchart for generating pressure-displacement process fingerprints.

[0021] Figure 4 A flowchart for generating an optimized scheme for sheet metal stamping process parameters. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for optimizing sheet metal stamping process parameters based on machine learning, including the following steps: S1. Obtain the forming design information, material batch information, die forming information and stamping equipment constraint information of the sheet metal part to be stamped, divide the deep drawing risk area, bending springback area and edge forming area, and determine the material forming window, die safety boundary and equipment capability boundary to form a regionalized initial forming deviation field.

[0026] S1.1: Digitally analyze the forming design information to determine the forming structure area and the target forming datum, and form a mapping relationship between the forming structure areas; Specifically, the forming design information is digitally analyzed. Based on the graphic reference, hole coordinates, dimensional tolerances, bending angle annotations, drawing depth annotations, and edge contour annotations in the forming design information, different forming positions of the sheet metal part to be stamped are marked. The forming structure area, including the hole area, target size control area, bending line area, drawing surface area, and edge contour area, is determined. The target size, target contour, target bending angle, target drawing depth, and target edge contour are organized into the target forming reference. The correspondence between the forming structure area and the target forming reference is established, forming the forming structure area mapping relationship.

[0027] It should be noted that the forming design information comes from the CAD drawings, two-dimensional unfolded drawings, product engineering drawings, process drawings and product requirements of the sheet metal part to be stamped, including two-dimensional unfolded outline, three-dimensional forming outline, graphic reference, hole coordinates, bending line position, bending angle annotation, drawing surface range, drawing depth annotation, edge outline position, target dimension annotation, dimensional tolerance, target outline and key forming area annotation.

[0028] S1.2: Associate the mapping relationship of the forming structure region with the material batch information, mold forming information and stamping equipment constraint information to determine the material forming window, mold safety boundary and equipment capability boundary corresponding to each forming structure region, and obtain the forming region data subject to forming constraints; Specifically, taking the forming structure region determined in the forming structure region mapping relationship as the associated object, the plate thickness, plate thickness tolerance, yield strength, tensile strength, elongation, hardness, anisotropy parameters, surface roughness, surface coating state and incoming material flatness in the material batch information are mapped to the forming structure region to determine the material forming window corresponding to the forming structure region. The upper mold surface, lower mold surface, mold clearance, blank holder position, blank holder surface shape, cutting edge position, drawing bead position, and forming fillet radius in the mold forming information are mapped to the forming structure area of ​​the associated material forming window. The forming contact range, blank holder range, cutting edge range, material flow restriction range, and fillet transition range are defined to determine the mold safety boundary corresponding to the forming structure area. By mapping the equipment tonnage, slide stroke, slide speed range, slide bottom dead center position, holding time range, feeding step distance range, feeding speed range, blank holder force adjustment range, and equipment allowable load in the stamping equipment constraint information to the stamping actions involved in the forming structure area of ​​the associated material forming window and mold safety boundary, the equipment capacity boundary corresponding to the forming structure area is determined, and the forming area data under forming constraints is obtained.

[0029] It should be noted that the material batch information comes from the material quality certificate, material test report, enterprise material management record and plate batch label, including material grade, plate thickness, plate thickness tolerance, yield strength, tensile strength, elongation, hardness, anisotropy parameters, surface roughness, surface coating status, incoming material flatness, material batch number and material test record; The mold forming information comes from mold drawings, mold 3D models, mold process documents, mold ledgers and mold inspection records, including upper mold surface, lower mold surface, mold clearance, blank holder position, blank holder surface shape, cutting edge position, cutting edge wear, guide clearance, drawing bead position, forming fillet radius, unloading structure, mold closing height, cumulative number of mold uses and mold maintenance records; The constraint information of the stamping equipment comes from the stamping equipment parameter table, equipment manual, equipment operation record, equipment inspection record and equipment process setting interface, including equipment tonnage, slide stroke, slide speed range, slide bottom dead center position, holding pressure time range, feeding step distance range, feeding speed range, stamping cycle range, blank holder force adjustment range, equipment allowable load, equipment control accuracy and equipment alarm threshold.

[0030] S1.3: Perform virtual forming pre-analysis on the forming area data to obtain the virtual forming state, and compare it with the target forming benchmark to obtain the regional simulation deviation. Divide the deep drawing risk area, bending springback area and edge forming area according to the deviation position and forming structure area type to form a regionalized initial forming deviation field.

[0031] Specifically, based on the forming area data, in the virtual forming pre-analysis environment, the target forming benchmark is used as the forming target, the material forming window is used as the material deformation constraint, the mold safety boundary is used as the mold contact constraint, and the equipment capacity boundary is used as the stamping action constraint. Virtual forming simulation is performed on each forming structure area to obtain the virtual forming contour, virtual bending state, virtual drawing state, virtual edge state and virtual dimension state corresponding to each forming structure area, thus obtaining the virtual forming state. The virtual forming state is compared item by item with the target size, target contour, bending target state, deep drawing target state and edge contour target state in the target forming reference to obtain the regional simulation deviation. According to the deviation position of the regional simulation deviation and the forming structure region type corresponding to the deviation position, the deviation position corresponding to the deep drawing surface is divided into the deep drawing risk region, the deviation position corresponding to the bending line is divided into the bending springback region, and the deviation position corresponding to the edge contour is divided into the edge forming region, thus forming a regionalized initial forming deviation field.

[0032] S2. Based on the regionalized initial forming deviation field and historical real forming deviation, a machine learning model for correcting sheet metal stamping forming deviation is constructed. The model learns the correspondence between finite element simulation deviation and on-site stamping deviation in different forming areas, and forms a corrected forming deviation field that is adapted to the current production conditions. S2.1: Using the deep drawing risk area, bending springback area and edge forming area in the regionalized initial forming deviation field as nodes, and combining the material flow relationship, force transmission relationship and process influence relationship of adjacent areas, a forming area relationship diagram is constructed, and the regional simulation deviation is mapped to the forming area relationship diagram to obtain the regional simulation deviation characteristics. Specifically, the drawing risk area, bending springback area, and edge forming area in the regionalized initial forming deviation field are used as nodes. Areas with adjacent contact, continuous transition, or influence from the same virtual forming process are used as node connection objects. Based on the forming area data constrained by forming and the virtual forming pre-analysis process, the material flow relationship, force transmission relationship, and process influence relationship between adjacent areas are marked to form a forming area relationship diagram. The regional simulation deviation in the regionalized initial forming deviation field is mapped to the drawing risk area, bending springback area, or edge forming area in the forming area relationship diagram according to the deviation position. The deviation form of the regional simulation deviation relative to the target forming reference is determined as the deviation type, and the deviation magnitude between the virtual forming state and the target forming reference is determined as the deviation degree. The deviation position, deviation type, and deviation degree are retained to obtain the regional simulation deviation characteristics.

[0033] It should be noted that the deviation type represents the form of the deviation of the simulated area relative to the target forming datum. By comparing the virtual forming state with the target forming datum, if the deviation occurs in the drawing state, it is determined as insufficient drawing depth, excessive drawing depth, local contour shrinkage, or local contour expansion; if the deviation occurs in the bending state, it is determined as excessive bending angle, insufficient bending angle, outward springback, or inward springback; if the deviation occurs in the edge state, the deviation type can be determined as edge expansion, edge shrinkage, insufficient flange height, excessive flange height, or edge contour offset; if the deviation occurs in the dimensional state, the deviation type can be determined as dimension exceeding the upper limit, dimension below the lower limit, or hole position offset.

[0034] S2.2: Obtain the historical true forming deviation of the stamped parts in the historical stamping production process by comparing the forming state of the stamped parts with the target forming benchmark, and collect them according to the nodes in the forming area relationship diagram to form the true deviation characteristics of the area; Specifically, the forming state of stamped parts that have undergone quality inspection during historical stamping production is obtained. The actual contour, actual bending state, actual drawing state, actual edge state, and actual size state of the stamped parts in the forming state are compared with the target contour, bending target state, drawing target state, edge contour target state, and target size in the target forming datum to obtain the historical true forming deviation. Based on the node correspondence in the forming area relationship diagram, the historical true forming deviation is assigned to the corresponding drawing risk area, bending springback area, or edge forming area, and the deviation position, deviation type, and deviation degree in the corresponding node are retained to form the true deviation characteristics of the area.

[0035] S2.3: Compare the actual deviation characteristics of the region with the simulated deviation characteristics of the region region by region to obtain the regional deviation correction amount, and associate it with material batch information, mold forming information and stamping equipment constraint information to obtain regional deviation training samples; Specifically, using nodes in the forming region relationship diagram as comparison objects, the actual deviation features and simulated deviation features of the regions are matched one by one according to the deep drawing risk region, bending springback region, and edge forming region. Within the same region, registration is performed according to the deviation position, and the deviation type and degree at the corresponding deviation position are compared. When there is a change in the deviation type or a difference in the degree of deviation between the actual deviation features and the simulated deviation features of the region, the change in the deviation type and the difference in the degree of deviation are organized into the region deviation correction amount of the corresponding region. The region deviation correction amount is bound to the material batch information, mold forming information, and stamping equipment constraint information corresponding to the same region, so that the region deviation correction amount corresponds to the current material state, mold action state, and equipment action state, thus obtaining the region deviation training sample.

[0036] For the same forming area, the same deviation position, and the same deviation type, the expression for the area deviation correction amount is: ; in, Indicates the first Regional deviation correction amount for each forming area; Indicates the first The allowable deviation scale for each forming area comes from the dimensional tolerance, bending angle allowable deviation, drawing depth allowable deviation, hole position allowable deviation, or edge contour allowable deviation in the target forming datum. This indicates the true regional deviation under the same deviation location and type. Indicates the first Simulation deviation of the formed area under the same deviation position and the same deviation type; This represents the hyperbolic tangent function, used to compress the deviation difference to the range (−1,1); Indicates the forming area number; This represents a positive constant to prevent the denominator from being zero. Its dimensions are the same as the allowable deviation scale. This applies when the actual deviation, simulation deviation, and allowable deviation scales for the region have all been normalized to dimensionless values. Dimensionless small quantities are used, with values ​​ranging from 0.001 to 0.01.

[0037] It should be noted that when Bi is 0, not configured, or less than the upper limit of the repeatability error of the testing equipment, Bi is not directly used in the denominator calculation. Instead, the effective value of the upper limit of the repeatability error of the testing equipment, the allowable deviation scale of the same type of testing item, and the allowable deviation scale of the corresponding testing item in the product requirements is used as a substitute allowable deviation scale, and a substitute source mark is written into the regional deviation training sample. When the same forming area, the same deviation position, and the same deviation type lack historical real forming deviations, no corresponding regional deviation correction amount is generated, and the corresponding node is used as a node to be predicted in the model inference. When the detection field in the regional real deviation or regional simulation deviation is missing, the missing field is written into the missing mask. The missing field does not participate in the calculation of the regional deviation correction amount, and the remaining fields are still used to form regional deviation training samples according to the corresponding deviation position and the corresponding deviation type. When the regional real deviation and the regional simulation deviation are opposite in direction, the sign difference between the regional real deviation and the regional simulation deviation is retained, and the directional regional deviation correction amount is output through the hyperbolic tangent function. When the deviation types corresponding to the two are inconsistent, the deviation type change is written into the regional deviation training sample, and the deviation type corresponding to the regional real deviation is used as the supervision label, and the deviation type corresponding to the regional simulation deviation is used as the input label. The target forming reference allowable deviation is set based on the target dimension annotation, dimensional tolerance, target contour, bending angle annotation, drawing depth annotation, and edge contour requirements in the forming design information. It is confirmed in conjunction with product requirements, customer acceptance standards, stamping part inspection capabilities, and the forming fluctuation range of historical qualified stamping parts. This includes target dimension allowable deviation, hole position allowable deviation, edge contour allowable deviation, bending angle allowable deviation, and drawing depth allowable deviation. For example, the target dimension allowable deviation can be ±0.05mm to ±0.50mm, the hole position allowable deviation can be ±0.05mm to ±0.30mm, the edge contour allowable deviation can be ±0.10mm to ±1.00mm, the bending angle allowable deviation can be ±0.5° to ±2°, and the drawing depth allowable deviation can be ±0.10mm to ±0.80mm. When the allowable deviation in the target forming datum is symmetrical, a single allowable deviation value is used as the allowable deviation scale. When the allowable deviation in the target forming datum has an inconsistent upper and lower limit allowable deviation, the corresponding allowable deviation scale is selected according to the direction of the difference between the actual deviation and the simulated deviation in the region. If the actual deviation in the region deviates positively from the simulated deviation in the region, the upper limit allowable deviation scale is used. If the actual deviation in the region deviates negatively from the simulated deviation in the region, the lower limit allowable deviation scale is used. The term "regional bias correction" is a unified subject. In the training samples, the regional bias correction formed by the true regional bias characteristics and the simulated regional bias characteristics is called the sample regional bias correction. In the model output, the regional bias correction output by the bias residual output layer is called the predicted regional bias correction. The training loss is determined by the difference between the predicted regional bias correction and the sample regional bias correction.

[0038] S2.4: Construct a regional deviation residual correction model including a regional feature input layer, a forming condition fusion layer, a regional relationship propagation layer, and a deviation residual output layer. Use regional deviation training samples to train the regional deviation residual correction model to obtain a machine learning model for correcting sheet metal stamping forming deviations.

[0039] Specifically, the topology of the regional deviation residual correction model is determined by the forming region relationship diagram. The regional simulation deviation features are used as the node features of the regional feature input layer. The production condition features formed by material batch information, mold forming information, and stamping equipment constraint information are used as the fusion objects of the forming condition fusion layer. The material flow relationship, force transmission relationship, and process influence relationship are used as the propagation basis of the regional relationship propagation layer. The sample regional deviation correction amount is used as the supervision label of the deviation residual output layer. The regional feature input layer encodes the regional simulation deviation features. The forming condition fusion layer incorporates the production condition features corresponding to the same region into the encoded regional simulation deviation features. The regional relationship propagation layer transmits the regional deviation influence between adjacent nodes according to the forming region relationship diagram. The deviation residual output layer converts the transmitted regional correlation deviation features into the predicted regional deviation correction amount, thereby completing the construction of the regional deviation residual correction model. The regional simulation deviation features, material batch information, mold forming information, and stamping equipment constraint information from the regional deviation training samples are fed into the regional deviation residual correction model. The sample regional deviation correction amount in the regional deviation training samples is used as the supervision label. The difference between the predicted regional deviation correction amount and the sample regional deviation correction amount is used as the main loss. The training loss is formed by combining the consistency of deviation changes in adjacent regions and the out-of-bounds penalties of material forming window, mold safety boundary, and equipment capability boundary. The parameters of the regional feature input layer, forming condition fusion layer, regional relationship propagation layer, and deviation residual output layer are updated through backpropagation. When the decrease in training loss is less than the preset change threshold in at least three consecutive training rounds, the training loss is determined to have converged, and a machine learning model for correcting sheet metal stamping forming deviation is obtained.

[0040] It should be noted that the parameter adjustment scheme, source constraint parameter set, deviation concentration region, and pressure-displacement process fingerprint are not directly input into the regional deviation residual correction model as new model input channels. During the iterative verification phase, the slider speed, holding time, blank holder force, die clearance, feeding step distance, and slider bottom dead center position in the parameter adjustment scheme are written into the production condition features as updated values ​​of stamping equipment constraint information or die forming information. The deviation concentration region is written into the regional simulation deviation features as a node mask in the forming region relationship diagram. The source constraint parameter set is retained as a screening constraint for the material forming window, die safety boundary, and equipment capability boundary. The pressure-displacement process fingerprint is used for stroke consistency verification after model output. This ensures that the inputs from the training phase and the iterative verification phase maintain consistency in the regional simulation deviation features, production condition features, and forming region relationships. Figure 3 Consistency across all input objects; The change threshold is a criterion used to determine whether the training loss or validation loss of the regional deviation correction model has stabilized. When setting it, the regional deviation correction model is pre-trained using historical regional deviation training samples. The decrease in training loss between adjacent training rounds is recorded. The effective lower limit of improvement that continued training can bring is determined by combining the repeated measurement error of the stamping part detection forming state. The decrease in loss corresponding to the effective lower limit of improvement is then used as the change threshold. For example, when the training loss is represented by normalized loss, the change threshold can be set to a decrease of less than 0.001 to 0.01 for at least three consecutive training rounds. The smaller the repeated measurement error of the detection equipment, the smaller the change threshold.

[0041] S2.5: Input the regionalized initial forming deviation field into the machine learning model to correct the sheet metal stamping forming deviation. Extract the region type and deviation features through the region feature input layer. Integrate the material batch information, mold forming information and stamping equipment constraint information through the forming condition fusion layer. Transmit the influence of adjacent regions through the region relationship propagation layer. Output the predicted region deviation correction amount of each forming region through the deviation residual output layer. Specifically, the machine learning model for correcting sheet metal stamping forming deviations is input into the regionalized initial forming deviation field. The regional feature input layer reads the regional labels of the deep drawing risk area, bending springback area and edge forming area according to the nodes in the forming area relationship diagram, and determines the regional label as the region type. At the same time, it reads the deviation position, deviation type and deviation degree of the regional simulation deviation in the corresponding node to form regional deviation features. The forming condition fusion layer incorporates material batch information, mold forming information, and stamping equipment constraint information into the regional deviation features according to the same node correspondence, forming regional fusion features with production condition constraints; the regional relationship propagation layer propagates the influence of adjacent regions based on the material flow relationship, force transmission relationship, and process influence relationship in the forming region relationship diagram. Among them, the material flow relationship is transmitted according to the material flow direction between adjacent forming regions, the force transmission relationship is transmitted from the mold contact area to the adjacent forming region, and the process influence relationship is transmitted according to the action sequence in the virtual forming process; the deviation residual output layer receives the transmitted regional correlation deviation features and outputs the predicted regional deviation correction amount of the deep drawing risk area, bending springback area, and edge forming area under the corresponding deviation position and deviation type according to the node correspondence.

[0042] S2.6: Backfill the predicted regional deviation correction amount into the regionalized initial forming deviation field to generate a corrected forming deviation field adapted to the current production conditions.

[0043] Specifically, based on the node correspondence in the forming region relationship diagram, the predicted region deviation correction amount is mapped to the deep drawing risk area, bending springback area, or edge forming area; based on the deviation position and deviation type corresponding to the predicted region deviation correction amount, the corresponding region simulation deviation is determined in the regionalized initial forming deviation field, and the deviation degree of the corresponding region simulation deviation is corrected using the predicted region deviation correction amount; the corrected region simulation deviation is retained together with the material forming window, mold safety boundary, and equipment capability boundary of the corresponding region to form a corrected forming deviation field adapted to the current production conditions.

[0044] S3. Based on the deviation position, deviation type and deviation degree in the corrected forming deviation field, a set of candidate stamping parameter schemes is selected in reverse. Virtual forming verification is carried out on the set of candidate stamping parameter schemes. The stroke correspondence between the virtual stamping force change and the virtual slide displacement change is extracted to obtain the pressure-displacement process fingerprint. S3.1: Based on the corrected forming deviation field, extract the deviation position, deviation type and deviation degree corresponding to the deep drawing risk area, bending springback area and edge forming area, and map them to the material forming window, mold safety boundary and equipment capability boundary to obtain the candidate parameter retrieval space; Specifically, based on the node correspondence in the forming region relationship diagram, the regional deviations corresponding to the drawing risk region, bending springback region, and edge forming region in the forming deviation field are located and corrected. Following the marking method of deviation position, deviation type, and deviation degree in the regional simulation deviation characteristics, the deviation position, deviation type, and deviation degree in the corrected forming deviation field are determined. Based on the forming region and deviation type to which the deviation position belongs, deviations in the drawing risk region are preferentially mapped to the material forming window, and the range of drawing-related stamping parameters is limited in conjunction with the die safety boundary. Deviations in the bending springback region are mapped to the die safety boundary, and the range of bending-related stamping parameters is limited in conjunction with the material forming window and equipment capability boundary. Deviations in the edge forming region are mapped to the die safety boundary and equipment capability boundary, limiting the range of edge forming-related stamping parameters. Among the stamping parameters limited by the stamping equipment constraint information and die forming information, the range of stamping parameters that have a preset mapping relationship with the deviation type and are within the corresponding boundary is retained, resulting in a candidate parameter retrieval space.

[0045] It should be noted that the preset mapping relationship refers to the correspondence between the deviation type and the adjustable stamping parameters. It is set based on historical stamping production records, die forming information, stamping equipment constraint information and verified process experience. For example, insufficient drawing depth corresponds to blank holder force, slide speed and lubrication status; excessive bending springback corresponds to die clearance, holding time and slide bottom dead center position; and edge contour offset corresponds to feeding step distance, cutting edge position and slide stroke.

[0046] S3.2: Based on the candidate parameter retrieval space, establish a deviation reverse index table. Through the deviation reverse index table, the deviation type of each region is matched with the stamping parameters jointly defined by the stamping equipment constraint information, die forming information and material forming window to form a set of candidate stamping parameter schemes.

[0047] Specifically, the search keys are the forming region, deviation type, and deviation degree of the deviation location; the search values ​​are the stamping parameter ranges defined by the stamping equipment constraint information and die forming information; and the valid conditions for the search values ​​are the material forming window, die safety boundary, and equipment capability boundary. This forms an index record in the deviation reverse index table of "forming region - deviation type - deviation degree - stamping parameter range". The deviation type and deviation degree of each region in the corrected forming deviation field are read, and the stamping parameter ranges corresponding to the same forming region and the same deviation type are found. The deviation degree is then compared with the corresponding allowable deviation upper limit in the target forming datum. The ratio is used as the deviation limit ratio; the allowable deviation limit is subtracted from the deviation degree to obtain the remaining allowable deviation, and the stamping parameter adjustment range shrinkage coefficient is determined accordingly; the intersection of the retrieved stamping parameter range with the material forming window, the mold safety boundary, and the equipment capacity boundary is taken to obtain the stamping parameter range after boundary constraints; the current process setting value is used as the adjustment reference value, and the stamping parameter adjustment range shrinkage coefficient is used to shrink the stamping parameter range after boundary constraints towards the adjustment reference value to obtain the shrinkage stamping parameter adjustment range; stamping parameter combinations that do not meet the three types of boundaries after shrinkage are eliminated to form a candidate stamping parameter scheme set.

[0048] It should be noted that the shrinkage coefficient of the stamping parameter adjustment range is set based on the degree of deviation, the upper limit of the corresponding allowable deviation in the target forming datum, the equipment control accuracy, and the process fluctuation range of historical qualified stamped parts. The ratio of the degree of deviation to the upper limit of the allowable deviation is used as the deviation occupancy ratio. When the degree of deviation does not exceed the upper limit of the allowable deviation, the shrinkage coefficient of the stamping parameter adjustment range decreases as the remaining allowable deviation decreases. When the degree of deviation exceeds the upper limit of the allowable deviation, the shrinkage coefficient of the stamping parameter adjustment range decreases as the deviation occupancy ratio increases, while retaining the lowest shrinkage lower limit. For example, the shrinkage coefficient of the stamping parameter adjustment range can be taken as 0.10 to 1.00.

[0049] S3.3: Input the set of candidate stamping parameter schemes into the virtual forming verification environment, call the forming area data constrained by forming and the corrected forming deviation field, perform virtual stamping stroke simulation on each candidate stamping parameter scheme, and form candidate virtual forming verification data; Specifically, in the virtual forming verification environment, the forming area data constrained by forming is used as the virtual forming object, the corrected forming deviation field is used as the regional deviation verification benchmark, and each set of candidate stamping parameter schemes in the candidate stamping parameter scheme set is loaded into the virtual stamping stroke. According to the stroke sequence of feeding positioning, slider approach, die contact, blank holder closure, forming loading, pressure holding and shaping, and unloading and springback, the stroke control parameters, blank holder constraint parameters, and die geometric constraint parameters corresponding to the feeding step distance, slider speed, blank holder force, pressure holding time, and die clearance are loaded at the corresponding stroke stages. Virtual forming simulation is performed on the deep drawing risk area, bending springback area, and edge forming area. The relationship between the virtual forming state and the corresponding regional deviation in the corrected forming deviation field is recorded at each stroke stage, and the changes in virtual stamping force and virtual slider displacement are recorded simultaneously. The virtual forming state, regional deviation changes, virtual stamping force changes, and virtual slider displacement changes corresponding to each set of candidate stamping parameter schemes are summarized to form candidate virtual forming verification data.

[0050] S3.4: Extract the virtual stamping force change and virtual slider displacement change corresponding to each candidate stamping parameter scheme from the candidate virtual forming verification data, and identify the contact start point, maximum forming load point, pressure holding stability point and unloading end point in the virtual stamping stroke to obtain stroke anchor point alignment data; Specifically, according to the candidate stamping parameter scheme number, the virtual stamping stroke record in the candidate virtual forming verification data is retrieved. Using the virtual stamping stroke position as the arrangement benchmark, the stamping force sampling value in the virtual stamping stroke record is continuously organized into virtual stamping force change, and the slider displacement sampling value in the virtual stamping stroke record is continuously organized into virtual slider displacement change. The stroke position where the virtual stamping force enters from the no-load benchmark and continuously rises and the virtual slider displacement reaches the mold contact position is determined as the contact start point. The stroke position where the virtual stamping force reaches its peak value is taken as the maximum forming load point. The stroke position where the virtual stamping force falls back to the no-load benchmark and the virtual slider displacement reaches the return release position is taken as the unloading end point. The stroke position where the virtual stamping force falls back to the no-load benchmark and the virtual slider displacement reaches the return release position is taken as the pressure holding stability point. The virtual stamping force change segment and virtual slider displacement change segment corresponding to the contact start point, maximum forming load point, pressure holding stability point, and unloading end point are marked to obtain stroke anchor point alignment data.

[0051] S3.5: Based on the stroke anchor point alignment data, the virtual punching force change and virtual slider displacement change are segmented, aligned, and normalized to obtain pressure-displacement alignment data. The stroke correspondence between the virtual punching force change and the virtual slider displacement change is extracted to obtain the pressure-displacement process fingerprint.

[0052] Specifically, based on the stroke anchor point alignment data, the same virtual stamping stroke is divided into contact loading segment, main forming segment, pressure holding and shaping segment, and unloading springback segment, using the contact start point, maximum forming load point, pressure holding and shaping point, and unloading springback point as segment boundaries. The virtual stamping force change and virtual slider displacement change within each segment are aligned according to the start and end points of the corresponding segment, and the change records of different sampling lengths within the segment are resampled at equal intervals to ensure that different candidate stamping parameter schemes have consistent data correspondence positions in the same stroke stage. Using the start point of the complete virtual stamping stroke as the normalization start point and the unloading end point as the normalization end point, the positions of each segment in the complete virtual stamping stroke are uniformly converted to the same stroke scale, while maintaining the positional correspondence of the contact start point, maximum forming load point, pressure holding and shaping point, and unloading end point under the same stroke scale, to obtain pressure-displacement aligned data. The correspondence between the virtual stamping force change and virtual slider displacement change at the same stroke position is extracted from the pressure-displacement aligned data to obtain the pressure-displacement process fingerprint.

[0053] S4. Match the pressure-displacement process fingerprint with the correction forming deviation field to identify the forming offset sources corresponding to material batch fluctuations, mold state changes and slider movement deviations, and screen parameter adjustment schemes within the material forming window, mold safety boundary and equipment capacity boundary. S4.1: The pressure-displacement process fingerprint is segmented and analyzed to extract the changes in the pressure slope, slider displacement hysteresis, and pressure-displacement coupling in the contact initiation segment, load rise segment, pressure holding and stabilization segment, and unloading segment, thereby obtaining the fingerprint abnormal segment. Specifically, using the contact start point, maximum forming load point, pressure holding stabilization point, and unloading end point in the stroke anchor point alignment data, the pressure-displacement process fingerprint is mapped to the contact start segment, load rise segment, pressure holding stabilization segment, and unloading segment. Within each segment, the change in punching pressure slope is extracted based on the change in virtual punching pressure at adjacent stroke positions in the pressure-displacement alignment data. Based on the difference in response position between the change in virtual slider displacement and the change in virtual punching pressure at the same stroke position, the hysteresis change in slider displacement is extracted. The pressure-displacement coupling change is extracted according to whether the change in virtual punching pressure and the change in virtual slider displacement are synchronously increased or stabilized. The stroke positions where the punching pressure slope changes abruptly, the slider displacement hysteresis increases, or the pressure-displacement coupling mismatch occurs are marked as abnormal positions, and the stroke range where adjacent stroke positions within the same segment are all abnormal positions is determined as the fingerprint abnormal segment.

[0054] It should be noted that when the virtual impact force changes from continuous increase to decrease, from stable to sudden increase or decrease within the same segment, or when a change direction inconsistent with the expected loading state of the segment appears between adjacent sampling points, it is determined to be a sudden change in the slope of the impact force; when the virtual impact force has entered the loading state corresponding to the current segment, but the virtual slider displacement at the same stroke position does not enter the corresponding displacement state synchronously, and the slider displacement response position moves backward relative to the virtual impact force response position, it is determined to be a lag increase in slider displacement; when the change direction of the virtual impact force and the change direction of the virtual slider displacement do not maintain the same progressive direction in the contact start segment and load increase segment, do not maintain a stable correspondence in the pressure holding and stabilization segment, or do not maintain a synchronous falling relationship in the unloading segment, it is determined to be a pressure-displacement coupling mismatch.

[0055] S4.2: Match the fingerprint abnormality section with the deep drawing risk area, bending springback area and edge forming area in the correction forming deviation field to determine the deviation concentration area, and determine the forming offset source based on the material batch fluctuation, mold state change and slider movement deviation corresponding to the deviation concentration area.

[0056] Specifically, based on the position of the fingerprint abnormal section in the virtual stamping stroke, the forming stage corresponding to the fingerprint abnormal section is determined, and the forming stage is mapped to the drawing risk area, bending springback area or edge forming area in the correction forming deviation field; within the corresponding forming area, the deviation of the area exceeding the allowable deviation of the target forming reference is taken as the candidate concentrated deviation, and the candidate concentrated deviations with continuous or adjacent deviation positions are merged to determine the deviation concentration area. The deviation type in the concentrated deviation area, the change in the stamping pressure slope, the change in slider displacement hysteresis, and the change in pressure-displacement coupling in the fingerprint anomaly section are compared with the material batch information, die forming information, and stamping equipment constraint information for judgment. If the concentrated deviation area is manifested as drawing depth, local contour shrinkage, or local contour expansion, accompanied by an overall change in the stamping pressure slope, then the forming offset is determined to be caused by material batch fluctuation. If the concentrated deviation area is manifested as bending angle deviation, springback outward expansion, springback inward contraction, or edge contour offset, and is concentrated in the die contact area, then the forming offset is determined to be caused by die state change. If the fingerprint anomaly section is manifested as increased slider displacement hysteresis or pressure-displacement coupling mismatch, and occurs simultaneously across multiple forming areas, then the forming offset is determined to be caused by slider movement deviation.

[0057] It should be noted that the overall change in the punching force slope refers to the virtual punching force change within the same fingerprint abnormality section, which continuously increases or decreases in the same direction relative to the corresponding segment reference pressure change, and the same direction change covers the main travel range of the fingerprint abnormality section. When only one candidate forming offset source exists, the candidate forming offset source is determined as the forming offset source. When there are two or more candidate forming offset sources, the corresponding information items are checked for whether they exceed the limits. For material batch fluctuations, the plate thickness, yield strength, tensile strength, elongation, and hardness are checked for whether they exceed the material forming window. For die condition changes, the die clearance, cutting edge wear, guide clearance, drawing bead position, and forming fillet radius are checked for whether they exceed the die safety boundary. For slider movement deviations, the slider stroke, slider speed, slider bottom dead center position, and holding time are checked for whether they exceed the equipment capacity boundary. When only one type exceeds the limit, the corresponding category is taken as the main forming offset source. When two or more types exceed the limit, the main forming offset source is determined in the order of slider movement deviation, material batch fluctuation, and die condition changes. The rest are taken as accompanying forming offset sources.

[0058] S4.3: Based on the source of forming offset, call the corresponding material forming window, mold safety boundary or equipment capability boundary, and filter the deviation type and stamping parameter correspondence in the deviation reverse index table according to the source of forming offset to form a source constraint parameter set; Specifically, a correspondence is established between the source of forming offset and its corresponding boundary. When the source of forming offset is material batch fluctuation, the material forming window is used as the parameter filtering boundary; when the source of forming offset is mold state change, the mold safety boundary is used as the parameter filtering boundary; when the source of forming offset is slider movement deviation, the equipment capacity boundary is used as the parameter filtering boundary; when the source of forming offset is a compound forming offset, the corresponding parameter filtering boundary is called first according to the main forming offset source, and then the corresponding parameter filtering boundary is called again according to the accompanying forming offset source. The deviation type corresponding to the deviation concentration area is found in the deviation reverse index table, the stamping parameter range corresponding to the deviation type is extracted, and the stamping parameter range is first intersected with the parameter filtering boundary corresponding to the main forming offset source to obtain the main source constraint parameter range; when there is an accompanying forming offset source, the main source constraint parameter range is intersected with the parameter filtering boundary corresponding to the accompanying forming offset source to obtain the compound source constraint parameter range; the stamping parameter range that exceeds the called parameter filtering boundary is removed, and the remaining stamping parameter range is merged according to the deviation position, deviation type, and deviation degree to form the source constraint parameter set.

[0059] S4.4: Cross-filter the source constraint parameter set and the candidate stamping parameter scheme set, retain the parameter combination that can reduce the degree of forming deviation in the deviation concentration area and meet the material forming window, mold safety boundary and equipment capability boundary, and generate parameter adjustment scheme.

[0060] Specifically, using the range of stamping parameters in the source constraint parameter set as the screening condition, each group of candidate stamping parameter schemes in the candidate stamping parameter scheme set is matched item by item, and candidate stamping parameter schemes with consistent stamping parameter items and stamping parameter values ​​falling within the corresponding range of the source constraint parameter set are retained; based on the candidate virtual forming verification data, the retained candidate stamping parameter schemes are mapped to the deviation concentration area, and candidate stamping parameter schemes that make the remaining deviation of the deviation concentration area relative to the target forming datum greater than the deviation before correction are eliminated; The remaining candidate stamping parameter schemes are verified for material forming window, mold safety boundary and equipment capability boundary. Candidate stamping parameter schemes that meet the three types of boundaries are retained, and the candidate stamping parameter scheme with the smallest remaining deviation in the deviation concentration area is determined as the parameter adjustment scheme.

[0061] S5. Input the parameter adjustment scheme into the machine learning model for iterative verification, obtain the iterative verification deviation, and use the iterative verification deviation to correct the forming deviation field, and output the sheet metal stamping process parameter optimization scheme.

[0062] S5.1: Write the stamping parameter items and stamping parameter values ​​corresponding to the parameter adjustment scheme into the production condition features, and associate the source constraint parameter set, deviation concentration area and pressure-displacement process fingerprint with the correction forming deviation field to form iterative verification input data; Specifically, the parameter adjustment scheme is mapped to the stamping parameter items and value ranges of the source constraint parameter set, retaining the correspondence between the filtered stamping parameter ranges in the source constraint parameter set and the parameter adjustment scheme. Based on the deviation location, type, and degree in the deviation concentration region, the parameter adjustment scheme is mapped to the deviation concentration region to be corrected. Furthermore, based on the stroke stage and fingerprint abnormality segment in the pressure-displacement process fingerprint, the parameter adjustment scheme is mapped to the pressure-displacement process fingerprint. The stamping parameter items and values ​​involved in the parameter adjustment scheme are written into the production condition features of the corresponding forming region. The deviation concentration region is written into the node mask in the forming region relationship diagram. The source constraint parameter set is written into the constraint fields corresponding to the material forming window, mold safety boundary, and equipment capability boundary. The pressure-displacement process fingerprint is written into the stroke consistency verification field after model output, and all are loaded into the corresponding forming region in the correction forming deviation field to form iterative verification input data.

[0063] S5.2: Input the iterative verification input data into the machine learning model for correcting sheet metal stamping forming deviation, predict the deviation of each region after the parameter adjustment scheme is applied, and compare it with the corrected forming deviation field to obtain the iterative verification deviation.

[0064] Specifically, the iterative verification input data is input into the machine learning model for correcting sheet metal stamping forming deviations. The regional feature input layer reads the corrected forming deviation field and deviation concentration area from the iterative verification input data. The forming condition fusion layer fuses the material batch information, mold forming information, and stamping equipment constraint information corresponding to the parameter adjustment scheme. The regional relationship propagation layer transmits the deviation impact of the parameter adjustment scheme on adjacent regions according to the forming region relationship diagram. The deviation residual output layer outputs the predicted deviation correction amount corresponding to each forming region after the parameter adjustment scheme is applied. The deviation degree of the corresponding forming region in the corrected forming deviation field is updated according to the predicted deviation correction amount to obtain the iterative predicted forming deviation field. Based on the node correspondence in the forming region relationship diagram, the iteratively predicted forming deviation field and the corrected forming deviation field are compared under the same forming region, the same deviation position, and the same deviation type. After the model output, the changes in the fingerprint abnormal section in the pressure-displacement process fingerprint are called to verify the stroke consistency. The difference in the degree of deviation and the difference in stroke fingerprint before and after the parameter adjustment scheme are obtained. The difference in the degree of deviation and the difference in stroke fingerprint are used as the iterative verification deviation.

[0065] S5.3: Based on the iterative verification deviation, locate the remaining deviation region in the correction forming deviation field that does not meet the preset passing threshold, and correct the correction forming deviation field according to the deviation type and degree corresponding to the remaining deviation region; Specifically, according to the node correspondence in the forming region relationship diagram, the iterative verification deviation is mapped to the deep drawing risk area, bending springback area, and edge forming area in the corrected forming deviation field; the deviation degree difference and stroke fingerprint difference in the iterative verification deviation are compared with the preset passing threshold, and combined with the fingerprint abnormal segment corresponding to the stroke fingerprint difference, the forming region where the deviation position that does not meet the preset passing threshold is located is located, and the remaining deviation area is determined; the deviation position, deviation type, and deviation degree corresponding to the remaining deviation area are read, the direction of increase or decrease correction of the regional deviation is determined according to the deviation type, and the correction magnitude of the regional deviation is determined according to the difference between the deviation degree and the preset passing threshold. The regional deviations under the same forming region, the same deviation position, and the same deviation type in the corrected forming deviation field are corrected to form the corrected corrected forming deviation field.

[0066] It should be noted that the passing threshold is set based on the target forming datum allowable deviation, the forming fluctuation range of historical qualified stamped parts, the repeatability error of the testing equipment, the stable fluctuation range of the pressure-displacement process fingerprint, and the product accuracy level. The target forming datum extracts the target dimension allowable deviation, bending angle allowable deviation, drawing depth allowable deviation, edge contour allowable deviation, and hole position allowable deviation. The range of deviation differences and the fluctuation range of the pressure-displacement process fingerprint under the same deviation type in historical qualified stamped parts are statistically analyzed. Combined with the repeatability error of the testing equipment, minor differences that cannot be stably identified are eliminated. The upper limit of the deviation difference and the upper limit of the stroke fingerprint difference that can distinguish between qualified and unqualified forming states are set as the passing threshold. For example, the deviation difference is represented by the normalized target forming datum allowable deviation, and the passing threshold can be between 0.05 and 0.20. The stroke fingerprint difference is represented by the normalized difference of the pressure-displacement process fingerprint, and the passing threshold can be between 0.03 and 0.15.

[0067] S5.4: Bind the corrected forming deviation field with the parameter adjustment scheme that meets the preset threshold and the pressure-displacement process fingerprint to generate a sheet metal stamping process parameter optimization scheme.

[0068] Specifically, the deviation position, deviation type, and deviation degree of each forming region in the corrected forming deviation field are mapped to the stamping parameter items, stamping parameter values, and applicable forming regions in the parameter adjustment scheme that meets the preset pass threshold, forming a deviation-parameter correspondence record. According to the stroke stage, fingerprint abnormal section, and pressure-displacement correspondence in the pressure-displacement process fingerprint, the parameter adjustment scheme that meets the preset pass threshold is mapped to the verification basis in the virtual stamping stroke. The corrected forming deviation field, the parameter adjustment scheme that meets the preset pass threshold, and the pressure-displacement process fingerprint are bound by forming region to generate a sheet metal stamping process parameter optimization scheme.

[0069] In summary, this invention constructs a machine learning model for correcting sheet metal stamping forming deviations. This model learns the correspondence between the regionalized initial forming deviation field and historical actual forming deviations, enabling the drawing risk area, bending springback area, and edge forming area to obtain deviation correction basis adapted to the current production conditions. By forming a corrected forming deviation field, candidate stamping parameter schemes are selected in reverse based on the deviation position, deviation type, and deviation degree, improving the regional targeting of the parameter adjustment process. Furthermore, the corrected forming deviation field is used for virtual forming verification and iterative verification, improving the adaptability of the sheet metal stamping process parameter optimization scheme to the current production conditions and enhancing the forming stability and boundary controllability during the parameter adjustment process.

[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing sheet metal stamping process parameters based on machine learning, characterized in that, include: Obtain forming design information, material batch information, die forming information and stamping equipment constraint information of the sheet metal part to be stamped, divide the deep drawing risk area, bending springback area and edge forming area, and determine the material forming window, die safety boundary and equipment capability boundary to form a regionalized initial forming deviation field; Based on the regionalized initial forming deviation field and historical real forming deviation, a machine learning model for correcting sheet metal stamping forming deviation is constructed. The model learns the correspondence between finite element simulation deviation and on-site stamping deviation in different forming areas, and forms a corrected forming deviation field that is adapted to the current production conditions. Based on the deviation position, deviation type and deviation degree in the corrected forming deviation field, a set of candidate stamping parameter schemes is selected in reverse. Virtual forming verification is carried out on the set of candidate stamping parameter schemes. The stroke correspondence between the virtual stamping force change and the virtual slide displacement change is extracted to obtain the pressure-displacement process fingerprint. By matching the pressure-displacement process fingerprint with the correction forming deviation field, the source of forming offset corresponding to material batch fluctuations, mold state changes and slider motion deviations is identified, and parameter adjustment schemes are screened within the material forming window, mold safety boundary and equipment capacity boundary. The parameter adjustment scheme is input into the machine learning model for iterative verification to obtain the iterative verification deviation. The iterative verification deviation is then used to correct the forming deviation field and output the sheet metal stamping process parameter optimization scheme.

2. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 1, characterized in that, The steps for forming the regionalized initial forming deviation field are as follows: The forming design information is digitally analyzed to determine the forming structure area and the target forming reference, and to form a mapping relationship between the forming structure area; The mapping relationship of the forming structure region is associated with the material batch information, mold forming information and stamping equipment constraint information to determine the material forming window, mold safety boundary and equipment capability boundary corresponding to each forming structure region, and thus obtain the forming region data subject to forming constraints. Virtual forming pre-analysis is performed on the forming area data to obtain the virtual forming state, and the regional simulation deviation is obtained by comparing it with the target forming benchmark. The deep drawing risk area, bending springback area and edge forming area are divided according to the deviation position and forming structure area type to form a regionalized initial forming deviation field.

3. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 2, characterized in that, The steps for constructing the machine learning model to correct sheet metal stamping forming deviations are as follows: Using the deep drawing risk area, bending springback area and edge forming area in the regionalized initial forming deviation field as nodes, and combining the material flow relationship, force transmission relationship and process influence relationship of adjacent areas, a forming area relationship diagram is constructed, and the regional simulation deviation is mapped to the forming area relationship diagram to obtain the regional simulation deviation characteristics. The historical true forming deviation of the stamped parts in the historical stamping production is obtained by comparing the forming state of the stamped parts with the target forming reference, and then collected according to the nodes in the forming area relationship diagram to form the true deviation characteristics of the area. By comparing the actual deviation characteristics of the region with the simulated deviation characteristics of the region region by region, the region deviation correction amount is obtained. Then, the region deviation training sample is obtained by associating it with material batch information, mold forming information and stamping equipment constraint information. A regional deviation residual correction model is constructed, which includes a regional feature input layer, a forming condition fusion layer, a regional relationship propagation layer, and a deviation residual output layer. The regional deviation residual correction model is trained using regional deviation training samples to obtain a machine learning model for correcting sheet metal stamping forming deviations.

4. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 3, characterized in that, The steps for forming a corrected forming deviation field adapted to the current production conditions are as follows: The machine learning model that inputs the regionalized initial forming deviation field into the machine learning model to correct sheet metal stamping forming deviation is used. The regional feature input layer extracts the regional type and deviation features. The forming condition fusion layer integrates material batch information, mold forming information and stamping equipment constraint information. The regional relationship propagation layer transmits the influence of adjacent regions. The deviation residual output layer outputs the predicted regional deviation correction amount of each forming region. The predicted regional deviation correction amount is backfilled into the regionalized initial forming deviation field to generate a corrected forming deviation field adapted to the current production conditions.

5. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 4, characterized in that, The steps for selecting the set of candidate stamping parameter schemes are as follows: Based on the corrected forming deviation field, the deviation position, deviation type and deviation degree corresponding to the deep drawing risk area, bending springback area and edge forming area are extracted and mapped to the material forming window, mold safety boundary and equipment capability boundary to obtain the candidate parameter retrieval space; A deviation reverse index table is established based on the candidate parameter retrieval space. The deviation type of each region is matched with the stamping parameters jointly defined by the stamping equipment constraint information, die forming information and material forming window through the deviation reverse index table, forming a set of candidate stamping parameter schemes.

6. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 5, characterized in that, The steps to obtain the pressure-displacement process fingerprint are as follows: The candidate stamping parameter scheme set is input into the virtual forming verification environment, the forming region data constrained by forming and the corrected forming deviation field are called, and the virtual stamping stroke is simulated for each candidate stamping parameter scheme to form candidate virtual forming verification data. Extract the virtual stamping force change and virtual slider displacement change corresponding to each candidate stamping parameter scheme from the candidate virtual forming verification data, and identify the contact start point, maximum forming load point, pressure holding stability point and unloading end point in the virtual stamping stroke to obtain stroke anchor point alignment data; Based on the stroke anchor point alignment data, the virtual punching force change and virtual slider displacement change are segmented, aligned, and normalized to obtain pressure-displacement alignment data. The stroke correspondence between the virtual punching force change and the virtual slider displacement change is extracted to obtain the pressure-displacement process fingerprint.

7. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 6, characterized in that, The steps for identifying the sources of forming offset corresponding to material batch fluctuations, mold state changes, and slider movement deviations are as follows: The pressure-displacement process fingerprint is segmented and analyzed to extract the changes in the pressure slope, slider displacement hysteresis, and pressure-displacement coupling in the contact initiation segment, load rise segment, pressure holding and stabilization segment, and unloading segment, thereby obtaining the fingerprint abnormal segments. The fingerprint anomaly segment is matched with the deep drawing risk area, bending springback area and edge forming area in the correction forming deviation field to determine the deviation concentration area. Based on the material batch fluctuation, mold state change and slider movement deviation corresponding to the deviation concentration area, the forming offset source is determined.

8. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 7, characterized in that, The steps for adjusting the filtering parameters are as follows: Based on the source of forming offset, the corresponding material forming window, mold safety boundary, or equipment capability boundary is called, and the deviation type and stamping parameter correspondence in the deviation reverse index table are filtered according to the source of forming offset to form a source constraint parameter set; By cross-filtering the source constraint parameter set and the candidate stamping parameter scheme set, the parameter combination that can reduce the degree of forming deviation in the deviation concentration area and meet the material forming window, mold safety boundary and equipment capacity boundary is retained to generate parameter adjustment scheme.

9. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 8, characterized in that, The steps to obtain the iterative verification deviation are as follows: Write the stamping parameter items and stamping parameter values ​​corresponding to the parameter adjustment scheme into the production condition features, and associate the source constraint parameter set, deviation concentration area and pressure-displacement process fingerprint with the correction forming deviation field to form iterative verification input data; The iterative verification input data is input into the machine learning model for correcting sheet metal stamping forming deviation. The deviation prediction of each region after the parameter adjustment scheme is applied is performed and compared with the corrected forming deviation field to obtain the iterative verification deviation.

10. The method for optimizing sheet metal stamping process parameters based on machine learning as described in claim 9, characterized in that, The optimization scheme for the output sheet metal stamping process parameters includes the following steps: Based on the iterative verification deviation, locate the remaining deviation region in the correction forming deviation field that does not meet the preset passing threshold, and correct the correction forming deviation field according to the deviation type and degree corresponding to the remaining deviation region; The corrected forming deviation field is bound to the parameter adjustment scheme that meets the preset threshold and the pressure-displacement process fingerprint to generate a sheet metal stamping process parameter optimization scheme.

Images