A bending method and system for multi-stage bending of sheet metal processing

Abstract

The application discloses a bending method and system for multi-section bending of sheet metal machining, and relates to the technical field of multi-section bending. The remaining bending area is determined according to the matching of multiple bending sections and the current bending position of the sheet metal part. Based on the next bending part, the corresponding bending requirements and the current bending offset content, the corresponding bending intelligent agent is constructed. The multi-level iteration is carried out on the bending damaged surface and the adjacent bending compensation surface. In the iteration process, multiple optimization areas are marked. The corresponding bending compensation mechanism is triggered in combination with the quality control system of the sheet metal part. The bending compensation content of each optimization area is marked, and multiple compensation parameters are determined by fusing the bending historical data. Based on the multiple compensation parameters and the corresponding optimization area, the bending data combination is constructed. The execution order of the multiple sub-bending combinations is marked along the analysis of the bending data combination, and the sequential repair of each optimization area is ensured.

Description

Technical Field

[0001] This invention relates to the technical field of multi-segment bending, and more particularly to a bending method and system for sheet metal processing for multi-segment bending. Background Technology

[0002] Sheet metal bending is a crucial process in sheet metal processing, widely used in automobile manufacturing, aerospace, and electronic equipment industries. As industrial products increasingly demand higher precision from sheet metal parts, traditional sheet metal bending technology has gradually revealed a series of technical bottlenecks that urgently need to be addressed.

[0003] Traditional bending processes typically rely on static procedures, executing bending actions sequentially according to a pre-defined drawing path. During actual processing, sheet metal parts undergo complex elastoplastic deformation, with each bending step resulting in positional deviations and accumulated deformation. Existing technologies often lack dynamic awareness of the "remaining bending area," making it difficult to match the current bending position with the planned trajectory in real time. This inability to accurately mark the current bending offset affects considerations for the next bending section, leading to lower accuracy of the damaged bending surface and adjacent bending compensation surfaces, and an inability to control the bending effect of the next bending section. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a bending method and system for sheet metal processing with multi-segment bending.

[0005] This invention provides a bending method for sheet metal processing with multi-segment bending, comprising:

[0006] During the bending process of sheet metal parts, multiple bending areas are determined based on the identification of the sheet metal part's processing drawings. Multiple bending segments are marked within these bending areas. The remaining bending areas are determined by matching these bending segments with the current bending position of the sheet metal part. The corresponding current bending offset is marked. This includes: real-time monitoring of the sheet metal part's bending process; acquiring the sheet metal part's processing drawings; dynamically identifying the processing drawings; determining multiple bending marks by combining the drawings' semantic features and geometric topology; determining the corresponding sub-bending areas by tracing each bending mark; and adjusting the bending process according to the logical sequence of each bending step. Within the sub-bending area, the corresponding bending segments are adaptively marked, and multiple bending segments are marked. Multiple bending data of the sheet metal part at the current moment are collected. Based on the multiple bending data and the current posture of the sheet metal part, the current state of the sheet metal part is determined. At the same time, the current bending position of the sheet metal part is marked. Dynamic matching is performed along the planned trajectory of multiple bending segments, the current bending position of the sheet metal part and the corresponding current state. Corresponding deviation constraint relationships are introduced for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, the current bending position of the sheet metal part is offset detected, and the corresponding current bending offset content is marked.

[0007] Traverse the remaining bending area to determine the next bending section. Based on the next bending section, the corresponding bending requirements, and the current bending offset, construct the corresponding bending agent and output the bending damage surface and the adjacent bending compensation surface based on the bending agent.

[0008] The bending damage surface and the adjacent bending compensation surface are iterated in multiple stages. During the iteration process, multiple areas to be optimized are marked. The corresponding bending compensation mechanism is triggered in combination with the sheet metal quality control system. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data.

[0009] Based on multiple compensation parameters and corresponding areas to be optimized, a bending data combination is constructed. Multiple sub-bending combinations are determined by parsing the bending data combination, and the execution order of the multiple sub-bending combinations is marked to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

[0010] This invention provides a sheet metal bending system for multi-segment bending, which is applied to the aforementioned sheet metal bending method for multi-segment bending; the sheet metal bending system for multi-segment bending includes:

[0011] The remaining bending area module is used to determine multiple bending areas based on the recognition of the sheet metal part's processing drawings during the bending process. Multiple bending segments are marked within these bending areas. The remaining bending area is determined by matching these bending segments with the current bending position of the sheet metal part. The corresponding current bending offset is also marked. This includes: real-time monitoring of the sheet metal part's bending process; acquiring the sheet metal part's processing drawings; dynamically recognizing the processing drawings; determining multiple bending marks by combining the semantic features and geometric topology of the drawings; determining the corresponding sub-bending areas by tracing each bending mark; and following the logic of the bending process. The corresponding bending segments are adaptively marked in each sub-bending area, and multiple bending segments are marked accordingly. Multiple bending data of the sheet metal part at the current moment are collected. The current state of the sheet metal part is determined based on the multiple bending data and the current posture of the sheet metal part. At the same time, the current bending position of the sheet metal part is marked. Dynamic matching is performed along the planned trajectory of multiple bending segments, the current bending position of the sheet metal part and the corresponding current state. Corresponding deviation constraint relationships are introduced for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, the current bending position of the sheet metal part is offset detected and the corresponding current bending offset content is marked.

[0012] The bending agent module is used to traverse the remaining bending area, determine the next bending part, construct the corresponding bending agent based on the next bending part, the corresponding bending requirements and the current bending offset, and output the bending damage surface and the adjacent bending compensation surface based on the bending agent.

[0013] The bending compensation module is used to perform multi-level iterations on the bent damaged surface and the adjacent bent compensation surface. During the iteration process, multiple areas to be optimized are marked. Combined with the sheet metal quality control system, the corresponding bending compensation mechanism is triggered. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data.

[0014] The repair module is used to construct a bending data combination based on multiple compensation parameters and the corresponding areas to be optimized, determine multiple sub-bending combinations by parsing the bending data combination, and mark the execution order of the multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] (1) During the bending process of sheet metal parts, multiple bending areas are determined based on the identification of the sheet metal parts processing drawings. Multiple bending segments are marked in the multiple bending areas. The remaining bending area is determined by matching the multiple bending segments with the current bending position of the sheet metal parts. The corresponding current bending offset content is marked. The remaining bending area is traversed to determine the next bending part. Based on the next bending part, the corresponding bending requirements and the current bending offset content, the corresponding bending intelligent body is constructed. The bending damaged surface and the adjacent bending compensation surface are output based on the bending intelligent body. The remaining bending area is introduced to control the next bending part. The next bending part, the corresponding bending requirements and the current bending offset content are fully considered, which improves the accuracy of the bending intelligent body and the accuracy of the bending damaged surface and the adjacent bending compensation surface.

[0017] (2) The bending damaged surface and the adjacent bending compensation surface are iterated in multiple stages. During the iteration process, multiple areas to be optimized are marked. The corresponding bending compensation mechanism is triggered in combination with the sheet metal quality control system. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating bending historical data. Based on multiple compensation parameters and the corresponding areas to be optimized, a bending data combination is constructed. Multiple sub-bending combinations are determined by parsing the bending data combination, and the execution order of multiple sub-bending combinations is marked to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect. A bending compensation mechanism is introduced to control multiple compensation parameters, improve the accuracy of multiple sub-bending combinations, and ensure the sequential repair of each area to be optimized in order to control the bending effect of the next bending part. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of a sheet metal bending method for multi-segment bending according to an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating step S11 in the sheet metal processing bending method for multi-segment bending according to an embodiment of the present invention.

[0020] Figure 3 This is a flowchart illustrating step S12 in the sheet metal processing bending method for multi-segment bending according to an embodiment of the present invention.

[0021] Figure 4 This is a flowchart illustrating step S13 in the sheet metal processing bending method for multi-segment bending according to an embodiment of the present invention.

[0022] Figure 5 This is a flowchart illustrating step S14 of the sheet metal bending method for multi-segment bending in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of the structural composition of a sheet metal bending system for multi-segment bending according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] Please see Figures 1 to 6 A bending method for sheet metal processing used in multi-segment bending, applicable to multi-segment bending scenarios; the bending method for sheet metal processing used in multi-segment bending includes:

[0026] Step S11: During the bending process of the sheet metal part, multiple bending areas are determined based on the identification of the sheet metal part's processing drawings. Multiple bending segments are marked in these multiple bending areas. The remaining bending area is determined by matching the multiple bending segments with the current bending position of the sheet metal part, and the corresponding current bending offset is marked.

[0027] Step S12: Traverse the remaining bending area to determine the next bending part. Based on the next bending part, the corresponding bending requirements and the current bending offset, construct the corresponding bending agent and output the bending damage surface and the adjacent bending compensation surface based on the bending agent.

[0028] Step S13: Perform multi-level iterations on the bent damaged surface and the adjacent bent compensation surface. During the iteration process, mark multiple areas to be optimized. Combine the sheet metal parts quality control system to trigger the corresponding bending compensation mechanism. In this bending compensation mechanism, mark the bending compensation content of each area to be optimized and integrate historical bending data to determine multiple compensation parameters.

[0029] Step S14: Construct a bending data combination based on multiple compensation parameters and the corresponding areas to be optimized. Determine multiple sub-bending combinations by parsing the bending data combination and mark the execution order of the multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

[0030] refer to Figure 2 In step S11, the specific steps are as follows:

[0031] S111: Real-time monitoring of the bending process of sheet metal parts, collection of sheet metal part processing drawings, dynamic identification of the processing drawings, and determination of multiple bending marks by combining the semantic features and geometric topology of the drawings. The corresponding sub-bending areas are determined by tracing along each bending mark. The corresponding bending segments are adaptively marked in each sub-bending area according to the logical order of the bending process, and multiple corresponding bending segments are marked.

[0032] S112: Collect multiple bending data of the sheet metal part at the current moment, determine the current state of the sheet metal part based on the multiple bending data and the current posture of the sheet metal part, and mark the current bending position of the sheet metal part. Dynamically match the current bending position of the sheet metal part with the planned trajectory of multiple bending segments and the corresponding current state, and introduce corresponding deviation constraint relationships for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, perform offset detection on the current bending position of the sheet metal part and mark the corresponding current bending offset content.

[0033] In the embodiments of this application, the bending process of sheet metal parts is monitored in real time, the processing drawings of sheet metal parts are collected, the processing drawings are dynamically identified, and multiple bending marks are determined by combining the semantic features and geometric topology of the drawings. The corresponding sub-bending areas are determined by tracing along each bending mark. According to the logical order of the bending process, the corresponding bending segments are adaptively marked in each sub-bending area, and multiple corresponding bending segments are marked. This approach takes into account the overall consideration of tracing each bending mark and ensures the accuracy of the corresponding sub-bending areas.

[0034] At this point, the system does not simply perform image scanning, but uses dynamic image parsing to vectorize the processing drawings; it uses a semantic segmentation network to extract non-geometric information such as dimension annotations, tolerance grades, and bending direction symbols (such as upper / lower mold indicators) from the drawings to form a processing constraint knowledge graph; based on graph theory methods, it analyzes the connection relationships, angle relationships, and adjacency relationships of lines in the drawings, identifies the boundary contours and internal feature structures of the board material, and eliminates interference items such as dimension annotation lines; at the same time, it maps semantic features to geometric topological nodes, identifies topological edges with "bending attributes," and marks them with highlights in the digital model to distinguish bending lines from cutting lines.

[0035] The system employs a region growth method or topology partitioning method to trace back the specific sheet metal range affected by the bending line; it extends the search to both sides with the bending line as the boundary until it encounters a physical boundary or an adjacent bending influence area, thus defining the "force influence domain" of the bending action; the entire sheet metal part is discretized into several interconnected but spatially independent "sub-bending regions", each region containing the local geometric features and material properties required for that bending segment, providing a foundation for subsequent distributed control.

[0036] The system arranges the operation sequence within a region based on bending process logic rules such as "inside before outside," "short before long," and "avoiding interference." It introduces an interference verification method to simulate the spatial positional relationship between the workpiece, machine tool, and mold during the bending process and determine the optimal bending sequence. For different bending segments within the same sub-region, it adaptively assigns "segment ID" and "process priority weight" based on their length, angle, and sheet thickness. The system not only marks the position of the bending line but also marks the processing attributes of the segment (such as dead bends and free bends), forming an executable process link.

[0037] Specifically, assuming the workpiece to be processed is an A-shaped sheet metal bracket, the bracket consists of a base plate, two inclined vertical plates on both sides and a top connecting part, and the whole presents an inverted "V" shape outline, involving multiple continuous bends at different angles.

[0038] The system acquires CAD processing drawings of sheet metal brackets in real time. During the recognition process, it extracts the geometric topology and identifies the slanted outline of the two strokes forming the "A" shape and the horizontal outline at the bottom. At the same time, combined with the semantic features of the drawings, the system parses out that the top angle of the sheet metal bracket is 60° (high-precision tolerance requirement), while the bottom bend is 90° (ordinary tolerance). The system automatically identifies multiple key bending marks such as L1 (left vertical plate bending line), L2 (right vertical plate bending line), and L3 (bottom bending line) at the sharp corners at the top and the connection points on both sides of the bottom of the sheet metal bracket, filtering out the dimension annotation layer on the drawings.

[0039] Tracing along mark L1 (the bending line of the left vertical plate), the system extends left to the edge of the sheet metal and right to the vicinity of the central axis of symmetry of the sheet metal support structure, defining the "bending area of ​​the left vertical plate". Similarly, the "bending area of ​​the right vertical plate" is defined along L2, and the "bending area of ​​the bottom plate" is defined along L3. This division clarifies that when processing the bending area of ​​the left vertical plate, its deformation mainly occurs on the left side and will not physically interfere with the positioning reference of the bending area of ​​the right vertical plate.

[0040] Based on the logical sequence of the bending process, the system performs adaptive marking. Considering that the sheet metal support structure is prone to interference between the two sides of the sheet metal during bending (i.e., "interference"), and the travel limitation of the bending machine mold, the system determines that if the top A-shaped sharp corner is bent first, the workpiece cannot be placed into the machine tool opening when bending the sides later. Therefore, based on the "avoiding interference" logic, the system adaptively marks the bending section of the bottom plate bending area as the first priority section; the system marks the bending section of the left vertical plate bending area as the second priority section, and the right vertical plate bending area as the third priority section. At the same time, for the special sharp angle bending requirement at the top of the sheet metal support, the system marks the process attribute labels "requires special V-groove mold" and "small radius bending compensation" for this section.

[0041] Furthermore, multiple bending data points of the sheet metal part at the current moment are collected. Based on these multiple bending data points and the current posture of the sheet metal part, the current state of the sheet metal part is determined. Simultaneously, the current bending position of the sheet metal part is marked. Dynamic matching is performed along the planned trajectory of multiple bending segments, the current bending position of the sheet metal part, and the corresponding current state. Corresponding deviation constraints are introduced for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, the offset of the current bending position of the sheet metal part is detected, and the corresponding current bending offset is marked. This comprehensive consideration of multiple bending data points and the current posture of the sheet metal part ensures the accuracy of the current state of the sheet metal part.

[0042] At this point, high-precision grating rulers, angle encoders, and vision sensors integrated into the bending machine tool are used to synchronously collect multi-dimensional physical quantities in real time, including the actual thickness of the sheet metal, the current bending angle, the bending force change curve, and the springback amount. Using kinematic models and coordinate transformation matrices, the collected discrete data is mapped to the global coordinate system of the workpiece to calculate the six-degree-of-freedom attitude (position and orientation) of the sheet metal part in space. At the same time, combined with the material constitutive model, the internal stress distribution of the sheet metal is inferred, thereby constructing a "current state model" that includes geometric shape and mechanical state, providing data support for subsequent decision-making.

[0043] The system constructs a virtual "planned trajectory flow," aligning the planned trajectory generated in S111 with the real-time acquired current bending position in both time and space. By calculating Euclidean distance and angular deviation, it evaluates the degree of agreement between the current execution point and the theoretical node. It introduces "deviation constraint relationships," including geometric boundary constraints (e.g., the workpiece cannot exceed the machine tool travel), process capability constraints (e.g., maximum bending force), and accuracy constraints. The system employs a synchronous calibration method, calculating the deviation vector in real-time during the matching process. Once the actual position deviates from the planned trajectory by more than a threshold, calibration logic is immediately triggered, and the feasibility of subsequent paths is recalculated. Based on the calibrated position, the workpiece model is re-cut, precisely removing the completed bending portions, and the boundary topology of the "remaining bending area" is updated in real-time to prevent subsequent area positioning errors due to previous bending errors.

[0044] The "current actual bending position" is compared with the "theoretical bending position" using a differential method. This method not only detects translational offsets in spatial coordinates but also rotational offsets caused by slippage or springback of the sheet metal. The detected offsets are then structured and encoded to generate "current bending offset content." This content includes not only the offset value but also an offset type label (such as linear displacement, angular springback, sheet metal slippage), forming an offset feature vector containing vectorized features.

[0045] Specifically, the current process involves the first priority section identified in S111—the processing stage after the base plate is bent. Just as the base plate bending is completed and preparation begins for bending the left upright plate, the system collects current data via a sensor array. The vision sensor detects that the sheet metal support base plate has been bent to 90.3° (with slight over-bending), and the force sensor shows a rebound force of approximately 2.5kN. The system integrates this data to calculate the current "state" of the sheet metal support: the base plate is in a horizontal position, but the sheet metal support opening is slightly open, and due to the stretching during the base plate bending process, the length of the sheet metal on both upright plates is 0.2mm shorter than the theoretical value. At this point, the sheet metal part exhibits a specific spatial posture on the machine tool worktable.

[0046] The system marks the current bending position as "the R-angle connecting the base plate and the left vertical plate"; the system dynamically matches this position with the trajectory of the "left vertical plate bending section" planned by S111; the system finds that due to the deviation of the bending angle of the base plate (90.3° instead of 90°), the starting point of the bending of the left vertical plate has been displaced.

[0047] The system introduces a "sheet metal bracket contour symmetry constraint." It determines that if the original trajectory is not calibrated and bending is performed directly, the sheet metal bracket will become asymmetrical, and the top will not close properly. Therefore, the system triggers synchronous deviation calibration, recalculates the optimal bending entry point for the left vertical plate, and adjusts the position of the back gauge accordingly. Based on the new calibrated position, the system updates the remaining bending area, clearly defining the remaining processing range of the connection between the left vertical plate and the top, and excluding the already shaped base plate area.

[0048] The system performs depth offset detection on the current bending position of the sheet metal part. Compared with the theoretical model, the system finds that the sheet metal of the left vertical plate has a sliding offset in the X-axis direction relative to the machine tool mold (due to the thrust caused by the bending of the base plate), with an offset of 1.5mm. At the same time, there is an angle pre-offset of 0.3°. The system encapsulates the above information into "current bending offset content" and marks it as: {offset type: composite offset; linear displacement: (-1.5mm, 0, 0); angle offset: (0.3°); influence domain: root of the left vertical plate}. This key data will directly guide the bending intelligent body built in step S12, enabling it to calculate the damaged surface and the compensation surface in a targeted manner, ensuring the final forming accuracy of the sheet metal bracket.

[0049] refer to Figure 3 In step S12, the specific steps are as follows:

[0050] S121: Traverse the remaining bending area and optimize during the traversal to output the corresponding multiple unbent nodes. Combine the multiple unbent nodes with the material properties of the sheet metal part and the corresponding bending sequence to determine the next bending part.

[0051] S122: In the next bend section, mark the bend trajectory of the next bend section, load the corresponding bend requirements and current bend offset content into the bend trajectory, and combine the corresponding bend framework deep learning to build the corresponding bend agent. At this time, the bend agent has adaptive decision-making ability and performs corresponding bend planning for the next bend section.

[0052] S123: The bending agent dynamically responds to the dynamic bending of the sheet metal part and simulates the stress evolution process of the next bending part in the material forming process based on virtual space. Based on the identification of the stress evolution process, the corresponding damage content and compensation content are determined. At the same time, the corresponding bending damage surface is determined by tracing the damage content, and the adjacent bending compensation surface is determined based on the combination of the peripheral traversal mechanism of the bending damage surface and the compensation content.

[0053] In the embodiments of this application, the remaining bending area is traversed, and optimization is performed during the traversal to output multiple corresponding unbent nodes. The multiple unbent nodes are combined with the material properties of the sheet metal and the corresponding bending sequence to determine the next bending part. This method of combining multiple unbent nodes with the material properties of the sheet metal and the corresponding bending sequence to determine the next bending part is introduced.

[0054] At this point, the system constructs a directed acyclic graph (DAG) of the bending nodes based on the updated geometric model of the "remaining bending region" in S112. The traversal process is not a simple sequential scan, but adopts a heuristic search strategy to evaluate all potential bending nodes in the remaining region one by one.

[0055] A multi-objective evaluation function is introduced for optimization, with evaluation indicators including: maximizing fixture avoidance space, minimizing bending force stroke, and minimizing cumulative positioning error. The system calculates the fitness score of each candidate node and selects a set of candidate nodes that are executable under the current posture. The selected geometric entities are transformed into logical "unbent nodes", each carrying attributes such as position coordinates, bending angle, and required mold information, forming a task queue to be scheduled.

[0056] The system calls the material database to extract the material yield strength, elastic modulus, and minimum bending radius of the current sheet metal part. For high-springback materials (such as high-strength steel), the system prioritizes nodes that can reduce springback using boundary constraints. For easily cracked materials, nodes with uniform stress are prioritized. A "process sequence constraint matrix" is introduced to define the logical prerequisite relationships between nodes (e.g., some bends must be completed before others, otherwise interference will occur). The system verifies the process logic level of each unbent node. Combining the weight of the material properties' influence on forming quality with the mandatory constraints of the process sequence, the system performs a final sorting of the nodes in the task queue, locks the first node as the "next bending part," and outputs its complete processing feature description.

[0057] Specifically, the current state of the sheet metal bracket is as follows: the base plate bending has been completed (S112 stage ends). At this time, the sheet metal part is in an "L" shape or lying flat on the machine tool table (depending on the specific process). The system needs to decide whether to bend the left vertical plate, the right vertical plate or the top connecting part next.

[0058] The system traverses the "remaining bending area" (including the left upright plate, the right upright plate, and the top connecting part) defined by S112; the system detects that there are three potential unbent nodes in the remaining area: node N1 (bending line of the left upright plate), node N2 (bending line of the right upright plate), and node N3 (bending line of the top connecting part).

[0059] The system performs interference simulation optimization. If node N3 (top) is selected, since the top of the sheet metal support structure is relatively narrow and the current base plate has occupied the lower space, the machine tool slider is very likely to collide with the bent base plate when it descends. Therefore, the optimization method determines that the "spatial executability" score of node N3 is too low and it is removed. The system outputs node N1 (left vertical plate) and node N2 (right vertical plate) as valid unbent nodes. Both nodes have sufficient operating space in the current posture.

[0060] The system needs to determine which of N1 and N2 should be bent first as the "next bending part". Assuming that the sheet metal bracket is made of 304 stainless steel, it has a significant tendency to work harden and springback characteristics. The system analysis found that if N1 is bent first, the base plate can be used as a reference surface for strong clamping, which can effectively suppress the springback when N1 is bent. However, if N2 is bent first, the lack of rigid support will cause the sheet metal bracket structure to twist. According to the standard process logic matrix of the sheet metal bracket, it usually follows the principle of "difficult before easy" or "short side before long side" (depending on the specific clamping method). Assuming that the left upright plate (N1) has process holes that need to be positioned, the system determines that N1 is the logic pre-node.

[0061] Based on the requirements for material springback control and the process logic sequence, the system ultimately determines the left vertical plate (node ​​N1) as the "next bending section". The system then outputs detailed instructions for this section, including the bending angle (e.g., 45°), target position coordinates, and recommended mold model, providing a clear execution object for building the "bending agent" in step S12.

[0062] Furthermore, in the next bend section, the bending trajectory of the next bend section is marked, the corresponding bending requirements and the current bending offset are loaded into the bending trajectory, and combined with the corresponding bending framework deep learning to construct the corresponding bending agent. At this time, the bending agent has adaptive decision-making ability and performs corresponding bending planning for the next bend section, thus introducing the corresponding bending planning for the next bend section.

[0063] At this point, based on the geometric model, the centerline of the next bend is extracted as the reference trajectory and discretized into a high-density sequence of control points along the trajectory direction. The spatial coordinates, tangent vector, and curvature information of each control point are marked. The static "bending requirements" (such as target angle tolerance, minimum bending radius, and surface roughness requirements) and the dynamic "current bending offset content" (offset feature vector from S112) are associated with the attribute layer of the trajectory. Through data fusion technology, the offset is mapped to the control points of the trajectory, correcting the theoretical trajectory into an "initial working trajectory with deviation constraints", providing accurate input features for subsequent deep learning networks.

[0064] The system calls a pre-trained deep neural network framework (such as an improved CNN or Transformer architecture), which has been trained on massive historical bending data and has the ability to extract nonlinear features of the bending process; the input layer receives "trajectory features with deviation constraints", and the hidden layer performs multi-layer nonlinear transformations to extract the high-dimensional mapping relationship between material deformation, springback compensation and path optimization.

[0065] For a specific bending task, some layers of the network are frozen and fine-tuned to generate a dedicated "bending agent". This agent has a built-in state sensor, policy network and value evaluator, and has the ability to make adaptive decisions in dynamic environments (such as autonomously adjusting the indentation depth or bending speed according to the real-time offset).

[0066] The bending agent operates on a forward inference basis based on the current state. It does not only output a single action command, but generates a complete multi-level bending planning scheme. This plan includes: tool contact strategy (to avoid scratching the surface), segmented force curve (to control bending speed and holding time), springback prediction compensation, and a fuse mechanism for abnormal situations. The planning results are output in the form of a standardized instruction set, which directly guides the servo system of the physical machine tool.

[0067] Specifically, at the current moment, S121 has determined that the "left side panel" is the next bending section. This panel needs to be bent at a specific angle to support the top connection. S112 has also detected that the bending of the bottom plate has caused a linear displacement offset of 1.5mm.

[0068] The system marks the bending center line on the left upright plate, i.e., the reference bending trajectory T1; the system loads the "bending requirements" into T1: the target angle is 45°±0.5°, and the outer surface must be free of scratches; at the same time, the "current bending offset content" (1.5mm displacement offset in the negative X-axis direction) detected by S112 is injected into the trajectory attributes; the system generates the initial operation trajectory, clearly informing the subsequent modules: the starting point of the bending of the left upright plate needs to be offset to the left by 1.5mm to offset the dimensional shrinkage caused by the bending of the base plate, and to ensure that the overall span of the sheet metal bracket meets the design requirements.

[0069] For the bending task of the left upright plate, the system launches a "bending frame deep learning model," which is built based on a database of bending characteristics of 304 stainless steel. The input layer reads the corrected trajectory features, and the hidden layer analyzes the springback trend of the material when bent at 45° (304 stainless steel usually has a large springback). The system instantiates a "dedicated agent for bending the left upright plate," which determines through a deep learning network that due to a 1.5mm offset, the top of the sheet metal support will not be able to close if the bending is done in a conventional manner. Therefore, the agent activates its adaptive decision-making ability, prepares to implement an overbending strategy to offset the springback, and adjusts the fixture position to accommodate the offset.

[0070] The bending agent outputs a detailed bending plan for the left upright plate: the agent decides to correct the theoretical bending angle of 45° to the actual bending angle of 47.2° (with a predicted springback of 2.2°), and plans a force application curve of "slow contact - fast bending - pressure holding and springback" to reduce stress concentration; considering the appearance requirements of the sheet metal bracket, the agent specifically marks the location of the bending damage surface in the plan, instructs step S13 to focus on the micro-cracks in this area, and reserves compensation space; the processing of the left upright plate of the sheet metal bracket no longer relies on rigid program code, but is planned in real time by an agent that "understands the process, knows the deviation, and can make decisions", ensuring high-precision forming under complex deviation environment.

[0071] Therefore, the bending agent dynamically responds to the dynamic bending of the sheet metal part and simulates the stress evolution process of the next bending section during material forming based on virtual space. It identifies the corresponding damage and compensation content based on the identification of this stress evolution process. Simultaneously, it traces the damage content to determine the corresponding bending damage surface and determines the adjacent bending compensation surface based on the combination of the peripheral traversal mechanism and the compensation content. This comprehensive consideration of the peripheral traversal mechanism and the compensation content ensures the accuracy of the adjacent bending compensation surface. Furthermore, it introduces the remaining bending area to control the next bending section, fully considering the next bending section, the corresponding bending requirements, and the current bending offset, thus improving the accuracy of the bending agent and the accuracy of the bending damage surface and adjacent bending compensation surface.

[0072] At this point, the bending agent receives displacement, force, and velocity signals uploaded by the machine tool sensors in real time through the data interface, driving the virtual model and physical entity to move synchronously, achieving millisecond-level dynamic response. In the virtual space, the agent calls the elastoplastic finite element solver or a trained proxy model. Based on the material's constitutive equation (stress-strain relationship), it simulates the distribution and flow of the internal stress field of the sheet metal part during bending. The system focuses on monitoring the change of the "stress concentration factor," recording the stress redistribution process from elastic deformation to plastic deformation, and generating a stress cloud map that changes over time to predict potential failure risks caused by material anisotropy or thickness inhomogeneity.

[0073] The process involves identifying the corresponding damage and compensation content based on the stress evolution process, and then performing feature extraction and quantitative evaluation on the simulation results. For the damage content, the system sets thresholds based on material mechanical properties (such as maximum allowable stress and ultimate strain). During the stress evolution process, if the equivalent stress in a certain region exceeds the yield strength and approaches the fracture limit, or the strain gradient exceeds the safety boundary of the material forming limit diagram (FLD), it is determined that the region has a damage risk. Damage content includes: microcrack initiation tendency, excessive thinning, and surface tearing risk. For the compensation content, the agent uses reverse engineering logic to deduce compensation strategies for the identified damage risks. Compensation content includes: overbending angle correction (to offset springback), variable flux pressure strategy (to adjust pressure distribution), or local annealing process parameters (to release stress), aiming to control the damage risk within an acceptable range.

[0074] The mapping from abstract data to specific geometric entities constructs the optimization object for step S13. At this point, in the bent damaged surface, the identified "damaged content" is mapped back to the CAD model surface of the sheet metal part using a geometric topology approach. The system traces the projection range of the high-stress area in three-dimensional space and accurately delineates the "bent damaged surface". This surface region not only includes geometric boundaries but also is associated with specific damage type labels.

[0075] Using the damaged bending surface as the seed surface, a perimeter traversal method (such as a K-nearest neighbor-based mesh search) is initiated to expand the search outward to adjacent geometric surfaces. Combined with the determined "compensation content", the system analyzes the structural rigidity and material allowance of the surrounding area, and selects adjacent areas that have the capacity to bear the compensation and do not affect the overall assembly. These are marked as "nearby bending compensation surfaces", which constitutes the basis for the "neighbor-based compensation" process optimization.

[0076] Specifically, at the current moment, the bending agent is performing the bending task of the "left side plate" determined by S121. The connection between the plate and the base plate is designed with a process hole for positioning. The bending radius is small (R2), and the material is cold-rolled steel plate that is prone to cracking. As the machine tool slider is pressed down, the left side plate begins to bend. The bending agent captures the bending force curve in real time and drives the digital twin model of the left side plate in virtual space.

[0077] The intelligent agent simulated the stress evolution during the bending angle from 0° to the target 45°. The system focused on simulating the R-angle region where the left vertical plate connects to the bottom plate. Due to the notch effect caused by the process hole in this area, the simulation showed that as the bending angle increases, the stress lines in this area rapidly become denser, forming a high-gradient stress concentration field, and the peak stress is close to the tensile strength of the material.

[0078] The intelligent agent scans and identifies the results of the virtual simulation in real time. In the virtual calculation, the system identifies that when the bend reaches 43°, the equivalent stress at the edge of the process hole on the left vertical plate exceeds the safety threshold, and determines that there is a "risk of tensile tearing damage" in this area (damage content). At the same time, it predicts that the thickness of the plate at this point will be reduced to 80% of the original thickness (excessive thinning).

[0079] To address the risk of tearing, the agent calculates a compensation strategy: stress needs to be dispersed by adjusting the stress state of the area before bending to the correct position. Therefore, the "compensation content" is determined to be: reduce the bending speed of the area by 10% and introduce reverse micro-stress support (achieved by adjusting the pressure of the hydraulic pad under the mold) to increase local material flowability and reduce the tendency to stretch.

[0080] The system geometricizes the identification results to prepare for subsequent repairs. Within the damaged surface, following the location of the "tensile tear damage risk," the system traces along the 3D model of the sheet metal bracket, precisely marking a strip-shaped area extending 5mm to the left from the edge of the process hole as the "bending damage surface," which is currently in a critical failure state. Simultaneously, the system initiates a perimeter traversal mechanism to check the neighborhood of the "bending damage surface." The system finds that below the damaged surface is the base plate plane, and above it is the vertical plate web. Since the base plate has already been bent and cannot be altered, while the vertical plate web area above it has sufficient material allowance, combined with the compensation content of "reverse micro-stress support," the system marks the vertical plate web area above the damaged surface (approximately 10mm wide) as the "adjacent bending compensation surface." The system will apply a specific leveling or correction process to this "adjacent bending compensation surface" to indirectly release the residual stress of the "bending damage surface," eliminate potential cracking risks, and ensure the quality integrity of the left vertical plate of the sheet metal bracket.

[0081] refer to Figure 4 In step S13, the specific steps are as follows:

[0082] S131: Perform multi-level iterative coupling analysis based on geometry and stress distribution on the bent damaged surface and the adjacent bent compensation surface, and output multiple iteration results at different levels during the iteration process. Multiple point cloud data are determined based on the selection of multiple iteration results. Multiple point cloud data form the corresponding region boundary, and the corresponding optimization region is determined by combining the corresponding multiple surface optimization factors, so as to mark multiple optimization regions.

[0083] S132: Collect the quality control system of sheet metal parts, determine multiple quality control elements based on the detection of the quality control system, dynamically match multiple quality control elements and each area to be optimized, and determine the corresponding bending compensation elements. Match the corresponding bending compensation mechanism based on the deep learning of each bending compensation element and the remaining bending area of ​​the sheet metal parts.

[0084] S133: Real-time monitoring of the execution of the bending compensation mechanism, marking the bending compensation content of each area to be optimized, forming a corresponding compensation gradient based on the priority of each bending compensation content, and integrating historical bending data to trigger multi-dimensional parameter optimization to determine the optimal multiple compensation parameters.

[0085] In the embodiments of this application, a multi-level iterative coupling analysis based on geometry and stress distribution is performed on the bent damaged surface and the adjacent bent compensation surface. During the iteration process, multiple iteration results at different levels are output. Multiple point cloud data are determined based on the selection of multiple iteration results. The multiple point cloud data form the corresponding region boundary. The corresponding optimization region is determined by combining the corresponding multiple surface optimization factors to mark the multiple optimization regions. This approach takes into account the overall consideration of selecting multiple iteration results and ensures the accuracy of multiple point cloud data.

[0086] At this point, the system establishes a multi-scale analysis architecture of "macro-mesh-micro". The macro level focuses on the overall bending angle and springback trend; the mesh level focuses on the curvature changes in the bending fillet area; and the micro level analyzes the surface texture and microcrack tendency. The "geometric morphology data" characterizing shape deviation is coupled with the "stress distribution data" characterizing the internal stress state. In each iteration, the system simulates applying a virtual correction force to the compensation surface and observes the geometric recovery and stress release of the damaged surface. The iteration process continues until the convergence criterion is met (e.g., the stress change rate is less than a threshold). The system outputs iteration snapshots at different levels, including the corrected mesh model, residual stress distribution cloud map, and deformation displacement field, as the basic data for subsequent screening.

[0087] Multi-dimensional evaluation functions (such as quality score, stress concentration, and geometric deviation value) are introduced to score the results of each iteration. The system selects the iteration scheme with the best comprehensive score or specific index (such as the most thorough stress elimination). Among the selected optimal iteration schemes, the coordinates of grid nodes that meet the "optimization threshold" are extracted. These nodes are usually located at positions where the geometric deviation exceeds the tolerance zone or the stress gradient is too large, forming a discrete point cloud set with spatial coordinates. The discrete point cloud is spatially topologically reconstructed using convex hull or alpha shape methods to generate a closed polygon boundary. This boundary clearly defines the contour of the area that "requires manual / machine intervention" and excludes non-critical areas.

[0088] The system matches the region boundaries with a library of "surface optimization factors," which include surface flatness defects, angle deviations, local material thinning, and work hardening brittleness. Through mapping analysis, it determines the dominant defect type within each boundary region. Based on the boundary enclosure results and dominant defect attributes, the system finally delineates the "regions to be optimized." Each region is assigned a unique ID and associated with specific optimization attribute labels (such as "first region - angle correction" and "second region - stress release"). These labeled regions will serve as direct inputs to trigger the compensation mechanism in subsequent steps of S13.

[0089] Specifically, in step S123, a "bending damage surface" (risk of tearing) and its adjacent "adjacent bending compensation surface" (web area of ​​the vertical plate) were marked near the process holes on the left vertical plate. The system initiated a multi-level iterative coupled analysis on the "bending damage surface" and "adjacent bending compensation surface" of the left vertical plate. The first iteration (macro): simulated applying an overall downward pressure to the adjacent bending compensation surface. The analysis results showed that the stress on the damaged surface decreased by 10%, but the overall span dimension of the sheet metal bracket exceeded the tolerance by 0.5mm. The second iteration (micro): the strategy was adjusted to apply local lateral thrust and a small amount of tension to the compensation surface. The analysis showed that the stress concentration factor of the damaged surface decreased significantly, and the overall dimension remained stable. The system output the results under the second-level iteration scheme: the high stress points in the damaged surface area disappeared, and the geometry was restored to 99% of the theoretical curvature.

[0090] The system filters and extracts data models from the iterative output. In the second-level iteration results, the system selects the mesh nodes with the best stress release effect and the nodes whose geometric shape has significantly recovered. In the 3D model, these nodes are represented as a ring of discrete points around the edge of the process hole and a set of elongated discrete points on the compensation surface. At the same time, using the convex hull method, the discrete points at the edge of the process hole are bounded into an irregular polygonal boundary (covering the potential microcrack zone). Meanwhile, the discrete points on the compensation surface are bounded into a rectangular boundary. These two boundaries precisely define the geometric range of which subsequent compensation operations need to be performed.

[0091] The system combines specific defect attributes to make a final qualitative assessment of the enclosed area. For the irregular boundary area around the process hole, the system, based on previous simulation data, determines that the dominant factors are "residual tensile stress concentration" and "surface microcrack tendency". For the rectangular boundary area on the compensation surface, the dominant factor is determined to be "local flatness deviation" (as a side effect of compensation stress). The system finally marks two specific areas to be optimized: Area M1 (located on the damaged surface): marked as "high stress release area", which needs to be leveled or stress-relieved in step S132; Area M2 (located on the compensation surface): marked as "geometric shape correction area", which needs to be slightly leveled in the subsequent bending compensation mechanism to eliminate the bulging deformation caused by compensation stress.

[0092] Furthermore, the quality control system of sheet metal parts is collected, and multiple quality control elements are determined based on the detection of this quality control system. Dynamic matching is performed along multiple quality control elements and each area to be optimized, and corresponding bending compensation elements are determined. Based on the deep learning of each bending compensation element and the remaining bending area of ​​the sheet metal parts, the corresponding bending compensation mechanism is matched, which is compatible with the overall consideration of the detection of the quality control system and ensures the accuracy of multiple quality control elements.

[0093] At this point, the system reads the product quality specification documents corresponding to the sheet metal parts (such as ISO drawing requirements, internal quality inspection standards, and customer-specific technical agreements) through the data interface. This is not just document reading, but the extraction of structured and unstructured quality knowledge. Using natural language processing (NLP) or rule engines, specific "quality control elements" are decoupled from the system. These elements include: dimensional tolerance grades (such as IT13), geometric tolerance requirements (such as flatness and perpendicularity), surface quality limits (such as scratch depth threshold), and material performance indicators (such as hardness range). These elements constitute the "benchmark anchor points" for subsequent determination of defect severity and selection of compensation strategies.

[0094] The system constructs a mapping matrix and compares the attributes of the "area to be optimized" marked by S131 (such as stress concentration and geometric deviation) with the "quality control elements" one by one; for example, it matches the measured angular deviation of the area with the tolerance zone and matches the risk of surface microcracks with the surface quality requirements.

[0095] Based on the matching results, the system filters out the key indicators that must be intervened and transforms them into "bending compensation elements". These elements not only include the target repair value (such as "the angle needs to be adjusted back by 0.5°"), but also the constraints (such as "the compensation process must not generate new scratches"). It defines the input boundaries and output targets of the compensation operation.

[0096] The system calls a pre-trained deep neural network model (such as a reinforcement learning policy network), with input variables including "bending compensation elements" (target state) and "remaining bending area state" (current constraint environment). The network model is trained based on historical big data and can predict the combined effects of different compensation methods. The model outputs the optimal "bending compensation mechanism," which is not a single action but a set of logically rigorous combination strategies, such as the "over-bending + local leveling" combination mechanism, the "variable pressure springback compensation" mechanism, or the "thermal-assisted stress release" mechanism. Based on the reasoning results, the system activates the corresponding process logic module.

[0097] Specifically, S131 has marked two areas to be optimized on the left upright plate: area M1 (edge ​​of process hole, high stress / micro-crack risk) and area M2 (web compensation area, flatness deviation). The system needs to decide which strategy to use to repair these areas while ensuring the overall quality of the sheet metal bracket. The system collects the quality control system documents of the sheet metal bracket and analyzes the key quality control elements for the left upright plate, including: Element Q1: Bending angle tolerance is ±0.5° (corresponding to the opening size of the sheet metal bracket); Element Q2: The surface must not have cracks or scratches with a depth greater than 0.1mm (appearance requirement); Element Q3: Web flatness is less than 0.2mm (assembly mating surface requirement); Element Q4: The material yield strength must be kept within the standard range (to avoid work hardening failure).

[0098] The system dynamically matches the regions M1 and M2 to be optimized with Q1-Q4. For region M1 (high stress / microcrack risk): the system matches it with elements Q2 (surface quality) and Q4 (material properties). It is found that the stress concentration at M1 has threatened the material properties, and the microcrack tendency violates Q2. For region M2 (flatness out of tolerance): the system matches it with element Q3 (flatness). It is found that the bulging deformation of M2 has exceeded the tolerance zone. At the same time, the bending compensation elements for M1 are determined to be: "eliminate tensile stress concentration" + "repair surface microcracks (depth <0.05mm)"; the bending compensation elements for M2 are determined to be: "flatness correction to within 0.1mm" + "keep bending angle unchanged".

[0099] The system combines the bending compensation elements with the constraints of the remaining bending area (the top connection part has not yet been bent) and uses a deep learning model to match the optimal mechanism. The model analysis suggests that if M1 is directly mechanically ground, the dimensions will be damaged; if M2 is directly subjected to strong pressure leveling, it will cause an overall angle change; considering that the remaining bending area (top) can provide reverse support force during subsequent processing.

[0100] For M1, the system matches a "local pulse stress redistribution mechanism". The strategy of this mechanism is to use an ultrasonic impact tool to perform high-frequency micro-forging on the M1 area, redistributing the residual tensile stress into compressive stress, thereby eliminating the tendency to crack and meeting the requirements of Q2 and Q4. For M2, the system matches a "multi-point micro-shaping compensation mechanism". The strategy of this mechanism is to use a CNC hydraulic pad to apply fixed-point reverse pressure to the M2 area, and perform synchronous correction in combination with the clamping force of the next bending process.

[0101] Therefore, by real-time monitoring of the execution of the bending compensation mechanism, marking the bending compensation content of each area to be optimized, forming a corresponding compensation gradient based on the priority of each bending compensation content, and integrating historical bending data to trigger multi-dimensional parameter optimization, the optimal multiple compensation parameters are determined. This approach is compatible with the overall consideration of integrating historical bending data to trigger multi-dimensional parameter optimization, ensuring the accuracy of the optimal multiple compensation parameters.

[0102] At this time, the system uses the machine tool PLC and external analysis instruments to collect key physical quantities in real time during the execution of the compensation mechanism, such as the hydraulic cylinder pressure change curve, the mold pressing displacement, and the material deformation rate; the monitoring logic focuses on determining whether the actual execution action deviates from the logic framework planned in S132.

[0103] For each area to be optimized as determined by S131, the system updates its status label in real time; the static "to be optimized" label is dynamically updated to "compensating in progress" status, and the compensation operation type corresponding to each area (such as pressure correction, position offset, rebound compensation) is recorded in detail to form a real-time compensation operation log, providing data support for subsequent priority determination.

[0104] The system assigns a weighted score to the bending compensation content based on the matching degree of the quality control elements in S132, usually following the principle of "structural safety takes precedence over geometric accuracy, and critical dimensions take precedence over appearance quality"; for example, compensation involving material fracture risk has the highest priority, and compensation involving flatness deviation has the second highest priority.

[0105] Based on priority ranking, the system constructs an ordered execution sequence model, namely "compensation gradient". This model specifies the intensity and time window of the compensation force. High-priority regions enjoy greater resource weight (such as greater correction force and longer holding time), while the compensation parameters of low-priority regions need to be adaptively fine-tuned based on the correction of high-priority regions to prevent physical interference between different compensation actions.

[0106] The system searches the bending history database and filters out process cases similar to the current sheet metal part in terms of material grade, thickness, and bending angle. It extracts successful compensation parameters and lessons learned from failures in historical records to form a prior knowledge base. A multi-objective optimization function is constructed using either genetic or particle swarm optimization methods. Input variables include real-time monitoring data, compensation gradient constraints, and historical prior parameters. The optimization objectives are set as: minimizing geometric deviation, homogenizing residual stress, and minimizing processing time. The system searches within the solution space and outputs the optimal parameter combination that satisfies all constraints, such as precise over-bending angle values ​​(accurate to 0.01°), optimal holding time (accurate to milliseconds), and reverse force magnitude, directly guiding the machine tool servo system's execution.

[0107] Specifically, S132 has matched a "pulse stress redistribution mechanism" for the edge of the process hole on the left vertical plate (region M1) and a "multi-point micro-shaping mechanism" for the web plate (region M2); the system monitors the compensation execution process of the left vertical plate of the sheet metal bracket in real time; when the compensation mechanism is activated, the system collects the pressure data of the hydraulic cushion and the position data of the grating ruler at a frequency of 100Hz; the system marks in real time in the virtual model: the stress relief task of region M1 (edge ​​of process hole) is currently being performed, and its compensation content is marked as "high frequency micro-forging"; at the same time, region M2 (web plate) is marked as being in standby mode, waiting for M1 to be completed before performing flatness correction.

[0108] The system prioritizes and weights the compensation actions for M1 and M2 based on the nature of the defects. The system determines that region M1 has a risk of microcracks and belongs to the "material failure type defect", so its priority is set to P1 (highest). Region M2 only has flatness deviation and belongs to the "geometric deviation type defect", so its priority is set to P2 (second highest). Based on this, the system forms a compensation gradient sequence: the first echelon (P1) concentrates resources to solve the stress concentration problem of M1 and allocates high-frequency impact heads to work at maximum rated power; the second echelon (P2) uses the hydraulic push rod of the lower die of the machine tool to perform micron-level shaping of M2 after the stress of M1 is released. This gradient design avoids the overall structural distortion caused by forcibly leveling M2 before the stress of M1 is eliminated.

[0109] The system retrieved historical data and found that the average springback of the same batch of 304 stainless steel was 2.3° when bent at a 45° angle. Furthermore, an additional 10% pulse energy was needed at the process holes to completely eliminate stress. Based on real-time monitoring data (current angle deviation +0.8°) and historical priors, the system triggered a multi-dimensional parameter optimization method and concluded that if compensation was based solely on the historical average, the top opening size of the sheet metal bracket would be too large. The final optimal compensation parameter combination was determined as follows: For area M1: the pulse impact frequency was set to 150Hz, and the duration was 3.5 seconds (0.5 seconds longer than the historical average to ensure complete stress elimination); For area M2: the hydraulic cushion reverse lifting height was 0.12mm, and the holding time was 2 seconds. This set of parameters ensured that while eliminating the risk of cracking, the overall contour accuracy of the sheet metal bracket was strictly controlled within the tolerance range.

[0110] refer to Figure 5 In step S14, the specific steps are as follows:

[0111] S141: Input multiple compensation parameters and corresponding regions to be optimized into the corresponding combination space, and reconstruct the corresponding matching relationship in the combination space. Trigger deep matching of multiple compensation parameters and corresponding regions to be optimized along the reconstructed matching relationship to construct the final bending data combination. At the same time, perform dynamic analysis on the final bending data combination to form multiple bending combination sequences. Determine the corresponding sub-bending combination based on the identification of each bending combination sequence, and mark the execution order of multiple sub-bending combinations in combination with process constraints and the principle of optimal time efficiency.

[0112] S142: Based on the execution order, the sequential repair of each region to be optimized is triggered, and the corresponding surface adjustment content is marked during the repair process. Each surface adjustment content and the corresponding sub-bending combination are loaded into the corresponding closed-loop control path to output a multi-dimensional closed-loop control loop. The bending effect of the next bending part is monitored in real time. If the bending effect of the next bending part does not meet the preset bending effect, the closed-loop control path is triggered again until the bending effect of the next bending part meets the preset bending effect.

[0113] In the embodiments of this application, multiple compensation parameters and corresponding regions to be optimized are input into a corresponding combination space, and the corresponding matching relationship is reconstructed in the combination space. The depth matching of multiple compensation parameters and corresponding regions to be optimized is triggered along the reconstructed matching relationship to construct the final bending data combination. At the same time, the final bending data combination is dynamically parsed to form multiple bending combination sequences. The corresponding sub-bending combination is determined according to the identification of each bending combination sequence, and the execution order of multiple sub-bending combinations is marked in combination with process constraints and the principle of optimal time efficiency. This approach takes into account the overall consideration of the identification of each bending combination sequence and ensures the accuracy of the corresponding sub-bending combination.

[0114] At this point, the system allocates a multi-dimensional "combination space" in memory, with its coordinate axes representing the geometric space dimension (position and range of the region to be optimized) and the process parameter dimension (pressure, displacement, and time). The system uses the discrete compensation parameters output by S133 (such as "pressure value P" and "displacement value D") and the regions to be optimized determined by S131 (such as "region M1" and "region M2") as input vectors. Using topological mapping, the original linear list structure is broken, and a mesh matching relationship based on spatial adjacency and parameter coupling is established. For example, if the compensation parameters of two adjacent regions conflict (such as one needs to be raised and the other needs to be lowered), the system automatically reconstructs the association attributes of the two in this space and marks them as "mutually exclusive" or "cooperative", laying the logical foundation for subsequent deep matching.

[0115] Following the reconstructed relational network, the system uses graph neural networks or depthwise traversal to semantically fuse parameters and regions. The system verifies the feasibility of parameters (e.g., whether the pressure exceeds the equipment limit) and calculates whether the expected deformation field after the parameters are applied covers the entire region to be optimized. After a successful match, the system packages the "regional geometric information" and "compensation process parameters" to form a standard bending data combination. Each combination is an independent data package containing: target region ID, operation type (e.g., shaping, springback compensation), specific parameter values, and expected execution effect. This serves as a bridge connecting the digital model and the physical machine tool.

[0116] The system logically decomposes the constructed bending data combination and generates multiple bending combination sequences according to the execution time sequence. Each sequence represents a processing step. Within the sequence, sub-bending combinations are further subdivided. For example, compensation for a large area is divided into two sub-combinations: "coarse adjustment" and "fine adjustment". A multi-objective optimization approach is introduced, comprehensively considering process constraints (e.g., stress must be eliminated before geometric shaping to avoid work hardening leading to fracture) and the principle of optimal time efficiency (e.g., reducing the number of mold changes, reducing the number of sheet metal flips, and merging similar operations in adjacent areas). The system assigns a unique time sequence label to each sub-bending combination and generates the final execution queue.

[0117] Specifically, the optimization area M1 (edge ​​of the process hole on the left vertical plate) needs to be stress-relieved (parameters: pulse frequency 150Hz, time 3.5s); the optimization area M2 (web of the left vertical plate) needs to be flattened (parameters: hydraulic cushion lifts 0.12mm, pressure holding 2s); at the same time, the system planning also needs to complete the bending of the right vertical plate (belonging to the remaining bending area).

[0118] The system inputs the geometric coordinates of M1 and M2 and their respective compensation parameters into the multidimensional combination space; the system recognizes that M1 (process hole) and M2 (web) are spatially adjacent; in the combination space, the system reconstructs the matching relationship between the two as "stress-geometric coupling constraint"; that is, the stress relief operation of M1 will slightly affect the flatness of M2, so the parameters of the two need to be spatially coordinated and matched, rather than set independently.

[0119] The system performs deep matching along the reconstructed coupling relationship; if stress relief is performed first in M1, the flatness error of M2 will be reduced by 20%; based on this, the system deeply binds the regional data of M1 with the "pulse parameter" to construct the bending data combination C1; binds the regional data of M2 with the "hydraulic jacking parameter" and adds a correction coefficient (the jacking height is adjusted to 0.1mm to offset the deformation caused by the operation of M1) to construct the bending data combination C2; thus forming data packets C1 and C2 containing precise operation instructions, which are no longer isolated data points, but logically related execution entities.

[0120] The system dynamically analyzes the bending tasks of C1, C2 and the right vertical plate; the system analyzes three main sequences: sequence A (left vertical plate compensation), sequence B (right vertical plate bending), and sequence C (overall verification); for sequence A, sub-bending combination A1 (M1 stress elimination) and sub-bending combination A2 (M2 plane correction) are analyzed.

[0121] According to process constraints, stress concentration at M1 must be eliminated first, otherwise subsequent stress will cause tearing at M1; therefore, A1 is marked as the first priority; the optimal efficiency is determined as follows: A2 (M2 correction) and A1 are on the same side plate and use the same set of fixture systems; to reduce auxiliary time, the system marks A2 as the second priority and executes it immediately after A1; since the left side plate (sequence A) can be used as a rigid reference to support the bending of the right side plate (sequence B) after processing, the system marks sequence B as the third priority; the system outputs an execution instruction stream with clear timing labels: [Execution sequence 1: A1 - Pulse stress elimination] -> [Execution sequence 2: A2 - Hydraulic shaping] -> [Execution sequence 3: B - Right side plate bending], this sequence ensures that the sheet metal bracket achieves optimal processing efficiency while ensuring quality.

[0122] Furthermore, based on this execution order, the sequential repair of each region to be optimized is triggered, and the corresponding surface adjustment content is marked during the repair process. Each surface adjustment content and the corresponding sub-bending combination are loaded into the corresponding closed-loop control path to output a multi-dimensional closed-loop control loop. The bending effect of the next bending part is monitored in real time. If the bending effect of the next bending part does not meet the preset bending effect, the closed-loop control path is triggered again until the bending effect of the next bending part meets the preset bending effect. This introduces a bending compensation mechanism to control multiple compensation parameters, improve the accuracy of multiple sub-bending combinations, and ensure the sequential repair of each region to be optimized in order to control the bending effect of the next bending part.

[0123] At this point, based on the timing tag generated by S141, the system sends a sequence of instructions to the servo and hydraulic systems of the bending machine via the PLC (Programmable Logic Controller). Each region to be optimized is activated sequentially according to the logic of "stress release first, geometric shaping later". While physical repair is being performed, the system calculates the deformation tensor of the sheet metal in real time based on the digital twin model. The system captures the geometric surface displacement (such as Z-axis settlement and X-axis slip) and topological changes caused by every minute movement, and marks the difference between the theoretical model and the actual model as "surface adjustment content" in real time. This not only records how much has been repaired, but also records the dynamic response characteristics during the repair process.

[0124] The system uses real-time "surface adjustment content" as feedback signals and process parameters (such as pressure and displacement) in the "sub-bending combination" as feedforward signals, both of which are loaded into the "closed-loop control path." A multi-dimensional closed-loop control loop including position loop, speed loop, and torque loop is constructed. The position loop ensures the precise pressing depth of the die; the torque loop adjusts the bending force in real time to adapt to uneven material thickness; and the speed loop optimizes the impact at the moment of contact. This loop realizes a millisecond-level cycle of "perception-decision-execution" and can dynamically fine-tune the machine tool movements based on the real-time feedback of surface adjustment content.

[0125] Using a laser displacement sensor or vision measurement system, the geometric shape and physical state after repair are collected in real time, and the deviation vector between the "measured value" and the "preset value" (theoretical model) is calculated. A convergence criterion function is introduced. If it is determined that "the bending effect does not meet the preset" (such as the deviation is greater than the tolerance zone, or the residual stress still exceeds the standard), the system triggers the abnormal handling logic.

[0126] Instead of terminating the processing, the system automatically adjusts the control gain or target setpoint of the closed-loop control path, triggering a new round of micro-repair cycle. This process iterates until the deviation vector converges to zero or enters the allowable tolerance range, at which point the system outputs a "processing complete" signal, achieving truly intelligent adaptive processing.

[0127] Specifically, the system needs to perform final repairs on the left upright plate (including the M1 stress concentration area and the M2 flatness deviation area), and then immediately complete the bending of the right upright plate; according to the instruction of S141, the system triggers the sub-bending combination A1 (for the edge of the M1 process hole); the ultrasonic impact head is activated to perform high-frequency tapping on the edge of the process hole; during the tapping process, the system monitors the micron-level extension of the plate surface through the sensor, and marks the surface adjustment content at this time as: "The surface compressive stress layer is completed, and the geometric position is slightly moved 0.02mm in the positive X-axis direction"; the system then records this change to provide a benchmark update for subsequent processes.

[0128] The system triggers sub-bending combination A2 (for the flatness of the web of M2); the system loads the sub-bending combination parameters of M2 (target lifting amount 0.1mm) and the previously marked surface adjustment content (reference offset 0.02mm) into the control path; the system outputs a multi-dimensional closed-loop control loop: force loop: the hydraulic cushion begins to lift upwards; position loop: the grating ruler monitors the lifting height in real time; feedback logic: when a sudden increase in lifting force is detected (meaning material hardening or reaching the position), the system automatically reduces the lifting speed to prevent overshoot.

[0129] After the left upright plate is repaired, the system continues to bend the right upright plate (as the next bending part); after the right upright plate is bent, the vision system immediately scans the overall outline of the sheet metal bracket; the system finds that due to the slight extension during the repair of the left upright plate, the total span dimension of the sheet metal bracket after the right upright plate is bent is 0.3mm smaller than the preset value, which does not meet the preset tolerance requirement of ±0.1mm.

[0130] The system determines that the bending effect is not up to standard, does not unload the material, and directly triggers the closed-loop control path again; the system calculates the difference and instructs the machine tool to apply a small amount of "reverse tension compensation" to the right vertical plate or adjust the position of the die back gauge; after the compensation action is completed, the system measures again and finds that the total span dimension deviation has been corrected to 0.05mm, meeting the preset bending effect; the system stops iterating, outputs a qualified signal, and the sheet metal bracket processing at this stage is successfully completed; the entire bending process is no longer a one-time open-loop execution, but a dynamic closed-loop system with self-sensing and self-correcting capabilities, ensuring the final forming quality of the sheet metal bracket under complex working conditions.

[0131] Please see Figure 6 , Figure 6 This is a schematic diagram of the structural composition of a sheet metal bending system for multi-segment bending according to an embodiment of the present invention; the sheet metal bending system for multi-segment bending is applied to the above-mentioned sheet metal bending method for multi-segment bending; the sheet metal bending system for multi-segment bending includes:

[0132] The remaining bending area module 21 is used to determine multiple bending areas based on the recognition of the sheet metal part's processing drawings during the bending process. Multiple bending segments are marked within these bending areas. The remaining bending area is determined by matching these multiple bending segments with the current bending position of the sheet metal part. The corresponding current bending offset is marked. This includes: real-time monitoring of the sheet metal part's bending process; acquiring the sheet metal part's processing drawings; dynamically recognizing the processing drawings; determining multiple bending marks by combining the drawing's semantic features and geometric topology; determining the corresponding sub-bending areas by tracing each bending mark; and determining the bending process logic. The algorithm adaptively marks the corresponding bending segments in each sub-bending area and marks multiple corresponding bending segments; it collects multiple bending data of the sheet metal part at the current moment, determines the current state of the sheet metal part based on the multiple bending data and the current posture of the sheet metal part, and marks the current bending position of the sheet metal part. It dynamically matches the planned trajectory of multiple bending segments, the current bending position of the sheet metal part and the corresponding current state, and introduces the corresponding deviation constraint relationship for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, it performs offset detection on the current bending position of the sheet metal part and marks the corresponding current bending offset content.

[0133] The bending agent module 22 is used to traverse the remaining bending area, determine the next bending part, construct the corresponding bending agent based on the next bending part, the corresponding bending requirements and the current bending offset content, and output the bending damage surface and the adjacent bending compensation surface based on the bending agent.

[0134] The bending compensation module 23 is used to perform multi-level iteration on the bending damaged surface and the adjacent bending compensation surface. During the iteration process, multiple areas to be optimized are marked. Combined with the quality control system of sheet metal parts, the corresponding bending compensation mechanism is triggered. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data.

[0135] Repair module 24 is used to construct a bending data combination based on multiple compensation parameters and the corresponding area to be optimized, determine multiple sub-bending combinations by parsing the bending data combination, and mark the execution order of multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

[0136] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A bending method for sheet metal processing used in multi-segment bending, characterized in that, include: During the bending process of sheet metal parts, multiple bending areas are determined based on the identification of the sheet metal part's processing drawings. Multiple bending segments are marked within these bending areas. The remaining bending areas are determined by matching these bending segments with the current bending position of the sheet metal part. The corresponding current bending offset is marked. This includes: real-time monitoring of the sheet metal part's bending process; acquiring the sheet metal part's processing drawings; dynamically identifying the processing drawings; determining multiple bending marks by combining the drawings' semantic features and geometric topology; determining the corresponding sub-bending areas by tracing each bending mark; and adjusting the bending process according to the logical sequence of each bending step. Within the sub-bending area, the corresponding bending segments are adaptively marked, and multiple bending segments are marked. Multiple bending data of the sheet metal part at the current moment are collected. Based on the multiple bending data and the current posture of the sheet metal part, the current state of the sheet metal part is determined. At the same time, the current bending position of the sheet metal part is marked. Dynamic matching is performed along the planned trajectory of multiple bending segments, the current bending position of the sheet metal part and the corresponding current state. Corresponding deviation constraint relationships are introduced for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, the current bending position of the sheet metal part is offset detected, and the corresponding current bending offset content is marked. Traverse the remaining bending area to determine the next bending section. Based on the next bending section, the corresponding bending requirements, and the current bending offset, construct the corresponding bending agent and output the bending damage surface and the adjacent bending compensation surface based on the bending agent. The bending damage surface and the adjacent bending compensation surface are iterated in multiple stages. During the iteration process, multiple areas to be optimized are marked. The corresponding bending compensation mechanism is triggered in combination with the sheet metal quality control system. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data. Based on multiple compensation parameters and corresponding areas to be optimized, a bending data combination is constructed. Multiple sub-bending combinations are determined by parsing the bending data combination, and the execution order of the multiple sub-bending combinations is marked to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

2. The bending method for sheet metal processing for multi-segment bending according to claim 1, characterized in that, The process of traversing the remaining bending area to determine the next bending section, constructing a corresponding bending agent based on the next bending section, the corresponding bending requirements, and the current bending offset, and outputting the bent damage surface and adjacent bending compensation surfaces based on the bending agent includes: Traverse the remaining bending area and optimize during the traversal to output multiple unbent nodes. Combine the multiple unbent nodes with the material properties of the sheet metal part and the corresponding bending sequence to determine the next bending section.

3. The bending method for sheet metal processing for multi-segment bending according to claim 2, characterized in that, The process of traversing the remaining bending area to determine the next bending section, constructing a corresponding bending agent based on the next bending section, the corresponding bending requirements, and the current bending offset, and outputting the bent damage surface and adjacent bending compensation surfaces based on the bending agent, also includes: In the next bend, the bending trajectory of the next bend is marked, the corresponding bending requirements and the current bending offset are loaded into the bending trajectory, and the corresponding bending framework deep learning is combined to build the corresponding bending agent. At this time, the bending agent has adaptive decision-making ability and performs corresponding bending planning for the next bend. The bending agent responds dynamically to the dynamic bending of the sheet metal part and simulates the stress evolution process of the next bending part in the material forming process based on virtual space. Based on the identification of the stress evolution process, the corresponding damage content and compensation content are determined. At the same time, the corresponding bending damage surface is determined by tracing the damage content, and the adjacent bending compensation surface is determined based on the combination of the peripheral traversal mechanism of the bending damage surface and the compensation content.

4. The bending method for sheet metal processing for multi-segment bending according to claim 1, characterized in that, The process involves multi-level iterations of the damaged bending surface and adjacent bending compensation surfaces. During the iteration process, multiple areas to be optimized are marked. Combined with the sheet metal parts' quality control system, a corresponding bending compensation mechanism is triggered. In this mechanism, the bending compensation content for each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data, including: A multi-level iterative coupled analysis based on geometry and stress distribution is performed on the bent damaged surface and the adjacent bent compensation surface. During the iteration process, multiple iteration results at different levels are output. Multiple point cloud data are determined based on the selection of multiple iteration results. Multiple point cloud data form the corresponding region boundary. Combined with the corresponding multiple surface optimization factors, the corresponding optimization region is determined to mark multiple optimization regions.

5. The bending method for sheet metal processing for multi-segment bending according to claim 4, characterized in that, The process involves multi-level iterations of the damaged bending surface and adjacent bending compensation surfaces. During the iteration process, multiple areas to be optimized are marked. A corresponding bending compensation mechanism is triggered based on the sheet metal part's quality control system. This mechanism marks the bending compensation content for each area to be optimized and integrates historical bending data to determine multiple compensation parameters. The process also includes: The quality control system of sheet metal parts is collected, and multiple quality control elements are determined based on the detection of the quality control system. Dynamic matching is performed along multiple quality control elements and each area to be optimized, and the corresponding bending compensation elements are determined. The corresponding bending compensation mechanism is matched according to the deep learning of each bending compensation element and the remaining bending area of ​​the sheet metal parts. The execution of the bending compensation mechanism is monitored in real time, the bending compensation content of each area to be optimized is marked, the corresponding compensation gradient is formed based on the priority of each bending compensation content, and the multi-dimensional parameter optimization is triggered by integrating historical bending data to determine the optimal multiple compensation parameters.

6. The bending method for sheet metal processing for multi-segment bending according to claim 1, characterized in that, The process involves constructing a bending data combination based on multiple compensation parameters and corresponding areas to be optimized, determining multiple sub-bending combinations by parsing this bending data combination, and marking the execution order of the multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending section meets the preset bending effect, including: Multiple compensation parameters and corresponding regions to be optimized are input into the corresponding combination space, and the corresponding matching relationship is reconstructed in the combination space. The deep matching of multiple compensation parameters and corresponding regions to be optimized is triggered along the reconstructed matching relationship to construct the final bending data combination. At the same time, the final bending data combination is dynamically parsed to form multiple bending combination sequences. The corresponding sub-bending combination is determined according to the identification of each bending combination sequence, and the execution order of multiple sub-bending combinations is marked in combination with process constraints and the principle of optimal time efficiency.

7. The bending method for sheet metal processing for multi-segment bending according to claim 6, characterized in that, The process of constructing a bending data combination based on multiple compensation parameters and corresponding areas to be optimized, determining multiple sub-bending combinations by parsing the bending data combination, and marking the execution order of the multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect, also includes: Based on this execution order, the repair of each region to be optimized is triggered sequentially. During the repair process, the corresponding surface adjustment content is marked. Each surface adjustment content and the corresponding sub-bending combination are loaded into the corresponding closed-loop control path to output a multi-dimensional closed-loop control loop. The bending effect of the next bending part is monitored in real time. If the bending effect of the next bending part does not meet the preset bending effect, the closed-loop control path is triggered again until the bending effect of the next bending part meets the preset bending effect.

8. A bending system for sheet metal processing for multi-segment bending, characterized in that, The sheet metal bending system for multi-segment bending is applied to the sheet metal bending method for multi-segment bending as described in any one of claims 1-7; the sheet metal bending system for multi-segment bending comprises: The remaining bending area module is used to determine multiple bending areas based on the recognition of the sheet metal part's processing drawings during the bending process. Multiple bending segments are marked within these bending areas. The remaining bending area is determined by matching these bending segments with the current bending position of the sheet metal part. The corresponding current bending offset is also marked. This includes: real-time monitoring of the sheet metal part's bending process; acquiring the sheet metal part's processing drawings; dynamically recognizing the processing drawings; determining multiple bending marks by combining the semantic features and geometric topology of the drawings; determining the corresponding sub-bending areas by tracing each bending mark; and following the logic of the bending process. The corresponding bending segments are adaptively marked in each sub-bending area, and multiple bending segments are marked accordingly. Multiple bending data of the sheet metal part at the current moment are collected. The current state of the sheet metal part is determined based on the multiple bending data and the current posture of the sheet metal part. At the same time, the current bending position of the sheet metal part is marked. Dynamic matching is performed along the planned trajectory of multiple bending segments, the current bending position of the sheet metal part and the corresponding current state. Corresponding deviation constraint relationships are introduced for synchronous deviation calibration, thereby delineating the remaining bending area. At the same time, the current bending position of the sheet metal part is offset detected and the corresponding current bending offset content is marked. The bending agent module is used to traverse the remaining bending area, determine the next bending part, construct the corresponding bending agent based on the next bending part, the corresponding bending requirements and the current bending offset, and output the bending damage surface and the adjacent bending compensation surface based on the bending agent. The bending compensation module is used to perform multi-level iterations on the bent damaged surface and the adjacent bent compensation surface. During the iteration process, multiple areas to be optimized are marked. Combined with the sheet metal quality control system, the corresponding bending compensation mechanism is triggered. In this bending compensation mechanism, the bending compensation content of each area to be optimized is marked, and multiple compensation parameters are determined by integrating historical bending data. The repair module is used to construct a bending data combination based on multiple compensation parameters and the corresponding areas to be optimized, determine multiple sub-bending combinations by parsing the bending data combination, and mark the execution order of the multiple sub-bending combinations to trigger the sequential repair of each area to be optimized until the bending effect of the next bending part meets the preset bending effect.

Images