An automobile seal positioning adjustment system based on industrial data processing

CN122883853APending Publication Date: 2026-10-09WUHU FROAD AUTOMOTIVE PARTS LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611142966.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-30
Publication Date
2026-10-09

AI Technical Summary

Technical Problem

在汽车密封件装配过程中,密封件通常为柔性粘弹性材料,在牵引、贴合和压入过程中容易受到拉伸、扭转、挤压以及轨迹曲率变化等因素影响,尤其在拐角过渡段和高曲率区域,容易出现回弹滞后、局部鼓包、翘边和压入不足等问题;现有方案多依赖视觉检测或位置偏差进行反馈修正,但仅依据已出现的外部形变进行控制,难以及时反映密封件内部尚未释放的应力状态,也难以适应高速装配节拍下应力沿密封件长度方向传播所带来的前导失稳风险;

Benefits of technology

本发明通过利用工业数据获取终端、边缘计算服务器和可编程逻辑控制器的协同处理,对轮廓点云数据、力矩数据和动作状态进行综合分析,并结合标准装配基准模型、预设粘弹性系数矩阵、虚拟张力-应变张量场、应力累积量、视觉扭曲观测值以及残余内应力梯度生成前馈补偿控制指令,能够将汽车密封件装配中的控制对象由已出现的外部位置偏差提升为对内部应力状态及其传播趋势的提前识别与干预,从而在待装配区域发生明显翘边、鼓包、回弹偏移和压入不足之前完成定位调整,相较于仅依赖视觉检测或位置偏差进行反馈修正的方式,具有更强的前瞻性、实时性和装配稳定性。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122883853A_ABST
    Figure CN122883853A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of automobile assembly automation and industrial data processing, in particular to an automobile sealing part positioning adjustment system based on industrial data processing; the automobile sealing part positioning adjustment system comprises an industrial data acquisition terminal, an edge computing server and a programmable logic controller; the industrial data acquisition terminal acquires torque data, action states and profile point cloud data of the automobile sealing part of an executing mechanism; the edge computing server acquires a three-dimensional model of a vehicle door as a standard assembly reference model, constructs a virtual tension-strain tensor field based on a preset viscoelastic coefficient matrix, the profile point cloud data and the torque data, calculates stress accumulation of a deformation stress wave based on the action states, carries out residual error analysis in combination with visual distortion observation values, and extracts a residual internal stress gradient; the programmable logic controller generates feedforward compensation control instructions based on the standard assembly reference model and the residual internal stress gradient and sends the feedforward compensation control instructions to the executing mechanism; the application has stronger foresight, real-time performance and assembly stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of automotive assembly automation and industrial data processing technology, specifically to an automotive sealing component positioning and adjustment system based on industrial data processing. Background Technology

[0002] With the development of automotive manufacturing automation and flexible assembly technology, the automatic pressing of automotive seals in parts such as car doors has become an important part of final assembly production. In order to achieve stable sealing and high-quality assembly, higher requirements are placed on positioning adjustment and dynamic control during the assembly process. In the assembly of automotive seals, the seals are usually made of flexible viscoelastic materials. During traction, bonding, and pressing, they are easily affected by factors such as stretching, torsion, extrusion, and changes in trajectory curvature. Especially in corner transition sections and high curvature areas, problems such as rebound lag, local bulging, edge lifting, and insufficient pressing are prone to occur. Existing solutions mostly rely on visual inspection or positional deviation feedback correction. However, control is based solely on the external deformation that has already occurred, which makes it difficult to reflect the stress state that has not yet been released inside the seal in a timely manner. It is also difficult to adapt to the risk of leading instability caused by stress propagating along the length of the seal under high-speed assembly. Therefore, it is crucial to improve the automatic assembly accuracy of automotive seals, reduce the amount of secondary correction displacement, and ensure assembly quality by comprehensively processing the contour shape data, stress data, and motion status during the assembly process, identifying the internal stress accumulation and propagation trend of the seal in advance, and generating corresponding feedforward compensation control strategies to complete the positioning adjustment before significant deviations occur in the assembly area. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides an automotive seal positioning and adjustment system based on industrial data processing. Specifically, the technical solution of the present invention includes: Industrial data acquisition terminals, edge computing servers, and programmable logic controllers; The industrial data acquisition terminal acquires torque data and motion status from external actuators, and acquires contour point cloud data of automotive seals, and sends the contour point cloud data, torque data and motion status to the edge computing server; The edge computing server acquires the 3D model of the car door as a standard assembly reference model, and based on a pre-configured preset viscoelastic coefficient matrix characterizing the physical properties of the automotive seals, the preset viscoelastic coefficient matrix includes the relaxation time constant, shear modulus and elastic recovery coefficient of the corresponding discrete segment of the automotive seals. A virtual tension-strain tensor field is constructed by combining contour point cloud data and torque data. Based on the action state and the virtual tension-strain tensor field, the dynamic stress transmission characteristics in the virtual tension-strain tensor field are extracted as deformation stress waves. The stress accumulation of the deformation stress waves is calculated, and the contour point cloud data is spatially registered and aligned with the standard assembly datum model. Based on the registration and alignment results, the point cloud coordinate space deviation is extracted from the contour point cloud data as the visual distortion observation value. Residual analysis is performed on the stress accumulation and the visual distortion observation value to extract the residual internal stress gradient and send the residual internal stress gradient to the programmable logic controller. The programmable logic controller (PLC) is connected to the actuator to generate feedforward compensation control commands based on the standard assembly reference model and residual internal stress gradient, and sends the feedforward compensation control commands to the actuator to realize real-time positioning adjustment and internal stress compensation control during the assembly process of automotive seals.

[0004] Preferably, the edge computing server defines the discrete segments of the automotive seal as multiple virtual nodes, maps the contour point cloud data and the torque data to the initial energy state of each virtual node, and constructs a spring-damped directed graph composed of virtual nodes based on the initial energy state and a preset viscoelastic coefficient matrix as a virtual tension-strain tensor field to improve the characterization accuracy of the stress deformation characteristics during the assembly of the automotive seal.

[0005] Preferably, the edge computing server applies a wave propagation dissipation operator containing a damping attenuation coefficient to the virtual tension-strain tensor field. Based on the wave propagation dissipation operator, it calculates the propagation speed and propagation phase of the deformation stress wave generated by the actuator's action along the automotive seal. Based on the propagation speed and propagation phase, it predicts the stress accumulation in the assembly area of ​​the automotive seal on the assembly trajectory at future time, so as to achieve early prediction of stress propagation trends.

[0006] Preferably, residual analysis is performed on the accumulated stress and the observed visual distortion to extract the residual internal stress gradient, including: converting the accumulated stress into a stress level score and the observed visual distortion into a distortion score; calculating the deviation between the stress level score and the distortion score as the fitting residual; when the fitting residual is greater than a preset residual threshold, the fitting residual is determined as the residual internal stress gradient; when the fitting residual is not greater than the preset residual threshold, the residual internal stress gradient is determined as zero to reduce the impact of small disturbances on the control results.

[0007] Preferably, the programmable logic controller calculates the reverse compensation energy required to offset the deformation stress wave based on the residual internal stress gradient, converts the reverse compensation energy into the fine-tuning speed offset and pose offset of the actuator according to the preset energy-kinematic mapping relationship, and generates feedforward compensation control instructions with timestamps based on the fine-tuning speed offset and pose offset to improve the timeliness of compensation and positioning stability during the assembly process.

[0008] Preferably, the programmable logic controller is further configured to: extract the position of the current action state of the actuator on the preset target assembly trajectory based on the timestamp as the assembly progress, determine whether the remaining time span between the current assembly progress of the actuator and the preset key trajectory point in the standard assembly reference model is less than or equal to the preset time advance; if it is less than or equal to the preset time advance, control the actuator to perform a reverse micro-torsion action to release internal stress; otherwise, maintain the current motion state of the actuator.

[0009] Preferably, the industrial data acquisition terminal includes an industrial camera and a six-dimensional torque sensor; the industrial camera is used to collect contour point cloud data of automotive seals, the six-dimensional torque sensor is used to collect torque data of actuators, and the industrial data acquisition terminal also acquires the action status of actuators through a communication interface to achieve synchronous acquisition of multi-source industrial data.

[0010] Preferably, the actuator is a six-axis robotic arm, and the system also includes an industrial control computer, which is used to discretize the assembly trajectory of the actuator into energy balance check points, so as to perform segmented analysis and control of energy changes during the assembly process of the seal in conjunction with the assembly trajectory.

[0011] Compared with the prior art, the present invention has the following beneficial effects: This invention utilizes the collaborative processing of industrial data acquisition terminals, edge computing servers, and programmable logic controllers to comprehensively analyze contour point cloud data, torque data, and motion states. It combines a standard assembly reference model, a preset viscoelastic coefficient matrix, a virtual tension-strain tensor field, accumulated stress, visual distortion observations, and residual internal stress gradients to generate feedforward compensation control commands. This elevates the control object in automotive sealing component assembly from existing external positional deviations to the early identification and intervention of internal stress states and their propagation trends. This allows for positioning adjustments to be completed before significant warping, bulging, springback offset, or insufficient pressing occurs in the assembly area. Compared to methods relying solely on visual detection or positional deviation feedback correction, this approach offers stronger foresight, real-time performance, and assembly stability. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1This is a schematic diagram of the structure of an automotive seal positioning and adjustment system based on industrial data processing, provided in an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] An automotive seal positioning and adjustment system based on industrial data processing is used for assembling and positioning automotive seals, including: an industrial data acquisition terminal, an edge computing server, and a programmable logic controller; Among them, the industrial data acquisition terminal is used to acquire torque data and action status from external actuators, and to acquire contour point cloud data of automotive seals, and send the contour point cloud data, torque data and action status to the edge computing server; Edge computing servers are used to acquire 3D models of car doors as standard assembly reference models; Based on a pre-configured viscoelastic coefficient matrix characterizing the physical properties of automotive seals, the pre-configured viscoelastic coefficient matrix includes the relaxation time constant, shear modulus and elastic recovery coefficient of the corresponding discrete segment of the automotive seal; and combined with contour point cloud data and torque data, a virtual tension-strain tensor field is constructed. Based on the action state and the virtual tension-strain tensor field, the dynamic stress transmission characteristics in the virtual tension-strain tensor field are extracted as deformation stress waves, and the stress accumulation of the deformation stress waves is calculated. Spatial registration and alignment are performed between the contour point cloud data and the standard assembly datum model. Based on the registration and alignment results, the spatial deviation of the point cloud coordinates is extracted from the contour point cloud data as the visual distortion observation value. Residual analysis is performed on the stress accumulation and the visual distortion observation value to extract the residual internal stress gradient and send the residual internal stress gradient to the programmable logic controller. The programmable logic controller (PLC) is connected to the actuator to generate feedforward compensation control commands based on the standard assembly reference model and residual internal stress gradient, and then sends the feedforward compensation control commands to the actuator. This embodiment provides a positioning adjustment mechanism for the automatic assembly of automotive door sealing strips, such as... Figure 1 As shown; specifically, the system is deployed at the door sub-assembly station of the vehicle assembly line to press the continuous flexible rubber sealing strip into the door sheet metal flange groove; at this station, the door is fed into the positioning fixture by the conveyor line, and the actuator runs along the outer contour of the door to complete the gripping, pulling, bonding and pressing of the sealing strip. Since the sealing strip is a viscoelastic material, the stress state is significantly different in the straight section, the rounded transition section and the upper corner area of ​​the door frame. If only the current visual position error is used for hysteresis correction, delayed rebound, local bulging and insufficient pressing are likely to occur at the corner. Therefore, this embodiment realizes the prediction and feedforward compensation of the internal stress state of the sealing strip through the collaboration between the industrial data acquisition terminal, the edge computing server and the programmable logic controller. Specifically, the industrial data acquisition terminal simultaneously collects three types of information during the assembly process. The first type is the contour point cloud data of the outer surface of the sealing strip, which reflects the spatial shape of the sealing strip relative to the door groove at the current moment. The second type is the torque data applied to the sealing strip by the end of the actuator, which reflects the current mechanical loads such as tension, torsion, and compression. The third type is the action status of the actuator, including the trajectory segment number, current pose, running speed, direction of attitude change, and pressure stage indicator. Among them, the pressure stage indicator specifically includes discrete state codes that represent whether the actuator is currently in the initial contact stage, stable pressure stage, or pressure holding stage. The edge computing server pre-stores a 3D model of the car door for this model and uses it as a standard assembly reference model. This reference model is not just a simple geometric reference, but is used to define the normal direction, curvature change and allowable assembly window of each slot, so that subsequent analysis can distinguish between normal fitting curvature change and torsion caused by abnormal force. After receiving the above data, the edge computing server does not directly treat the sealing strip as a rigid profile, but abstracts it as a flexible body that continuously transmits stress along its length. A preset viscoelastic coefficient matrix is ​​used to characterize the response characteristics of this batch of sealing strip materials under tensile recovery, shear hysteresis, and local torsion. This matrix can be understood as a set of calibrated material response relationships, which reflect that when the same external force is applied to a straight segment and a curved segment, the rate of energy accumulation and release inside the sealing strip is different. Specifically, the preset viscoelastic coefficient matrix includes the relaxation time constant, shear modulus, and elastic recovery coefficient corresponding to each discrete segment. These parameters are obtained by conducting standard tensile and torsional mechanical calibration tests on the same batch of sealing strip samples beforehand. A virtual tension-strain tensor field, constructed based on point cloud data and torque data, is used to represent the current stress and deformation distribution state of each area of ​​the sealing strip. Combined with the action state of the actuator, the edge computing server calculates the cumulative trend of deformation stress waves excited by gripping, pulling, and pressing actions inside the sealing strip. The amount of stress accumulation mentioned here actually corresponds to how much deformation energy that may cause rebound or displacement has been stored inside a certain area to be assembled before it is pressed into the groove. On the visual side, the system extracts the point cloud coordinate space deviation from the contour point cloud data as the visual distortion observation value. This observation value is not limited to the overall translation error, but focuses on local warping, twisting, wavy undulations, and deviations from the relative groove normal. These distortions are often external manifestations of internal stress that has not yet been released. The edge computing server performs residual analysis on the stress accumulation and the visual distortion observation value. If the stress result predicted by the model is consistent with the actual surface distortion, it means that the current material state can be explained by known physical parameters. If there is a significant residual between the two, it means that there is still residual internal stress inside the sealing strip that has not been fully reflected by the vision. The system extracts the residual internal stress gradient and sends it to the programmable logic controller (PLC). The PLC then combines the standard assembly reference model to generate feedforward compensation control instructions, enabling the actuator to adjust its motion posture and rhythm in advance before the error actually spreads to the area to be pressed in front. To illustrate the organization of the data flow, a simplified process can be used. Suppose that a certain sealing strip is currently divided into four continuous regions S1, S2, S3, and S4 along its length, where S1 is close to the clamping end and S4 is close to the area to be assembled that is about to enter the upper corner of the door. The industrial camera detected a slight distortion on the outer surface of S3, while the six-dimensional torque data indicated a continuous change in torsional load at the end within the current preset time window. Based on this, the edge computing server determined that although S4 had not yet entered the pressing point, the stress transmission from S2 and S3 had already moved towards S4. Therefore, the control did not wait for S4 to show a significant deviation, but instead output the compensation amount for S4 to the programmable logic controller in advance. This avoids correcting the issue after a significant warping occurs at the corner. In abnormal or boundary situations, the system adopts a degradation strategy: if the point cloud is partially missing due to surface reflection or occlusion, the edge computing server retains the continuous morphological information of the most recent stable period and increases the weight of torque data and action status; if the torque sensor experiences a short-term noise surge while the visual morphology remains stable, instead of a large-scale feedforward compensation action, the system confirms whether a continuous stress accumulation has formed through multiple consecutive control cycles; if the binding information between the standard assembly reference model and the current vehicle model workstation is inconsistent, the automatic issuance of feedforward compensation control commands is prohibited, and the actuator is only allowed to maintain a safe trajectory and wait for manual confirmation to avoid erroneous actions caused by incorrect vehicle model calls. In the assembly station of the right front door sealing strip of a new energy vehicle, a six-axis robotic arm starts from the lower edge of the door frame, moves up along the A-pillar, and turns into the roof beam section at the upper corner. When the robotic arm passes the upper rounded corner of the A-pillar at a high pace, the industrial data acquisition terminal detects fine ripples on the outer surface of the sealing strip, and the six-dimensional torque data simultaneously shows that the torque around the end normal direction continues to rise. The edge computing server determines that it is about to enter a high curvature corner area based on the three-dimensional model of the door, and then constructs a virtual tension-strain tensor field at that moment, and finds that there is a clear stress accumulation trend in the unpressed front area. Therefore, the programmable logic controller issues a feedforward compensation control command before the robotic arm enters the corner, so that the end effector deflects according to the preset angle compensation amount and is adjusted according to the preset correction parameters; as a result, when the sealing strip enters the corner groove, it can form a coordinated fit with the external pressing action through its own rebound, thereby improving the local fit quality of the sealing strip. Furthermore, to avoid ambiguity in the terminology used in subsequent implementations, this specification adopts the following unified conventions for related names: whenever an actuator is mentioned, it refers to the end effector system that is connected to the programmable logic controller and actually performs gripping, traction, bonding, and pressing actions; when a robotic arm is mentioned in subsequent implementations in conjunction with specific workstation examples, it only indicates that the actuator preferably adopts the specific implementation form of a six-axis robotic arm, without changing the overarching concept of the actuator; When referring to an industrial camera, it refers to the visual acquisition component within an industrial data acquisition terminal used to output contour point cloud data; camera descriptions appearing in subsequent paragraphs have the same meaning as industrial camera; when referring to a controller, unless otherwise specified, it refers to a programmable logic controller, and not an edge computing server or industrial control computer; when referring to a 3D model of a car door as an assembly reference, the standard assembly reference model is the sole designation, and subsequent descriptions of the reference model and the 3D model of the car door shall be based on this same technical object; This ensures that the data flow and control relationship between the industrial data acquisition terminal, edge computing server, programmable logic controller, and actuator remain consistent throughout the text. Specifically, the industrial data acquisition terminal is responsible for collecting and uploading data, the edge computing server is responsible for modeling, prediction, and residual extraction, the programmable logic controller is responsible for generating and issuing feedforward compensation control instructions, and the actuator is responsible for performing assembly actions according to the feedforward compensation control instructions. The purpose of this step is to change the control object in the assembly of flexible sealing strips from static position deviation to dynamic stress state and future deformation trend; thereby achieving early interception of internal stress propagation in high-speed assembly scenarios and reducing the probability of corner misalignment, local bulging and springback offset.

[0015] Furthermore, the edge computing server is also used to: define discrete segments of automotive seals as multiple virtual nodes, and map contour point cloud data and torque data to the initial energy state of each virtual node; based on the initial energy state and a preset viscoelastic coefficient matrix, construct a spring-damped directed graph composed of virtual nodes as a virtual tension-strain tensor field.

[0016] This embodiment provides a modeling mechanism for discretizing flexible sealing strips. Specifically, in the previous embodiment, although it was possible to determine the internal stress propagation of the sealing strip based on point cloud, torque, and motion state, if the overall calculation is performed directly on the continuous physical body of the entire sealing strip, the local state is easily masked by the overall mean when facing different curvature segments, different material response zones, and different clamping states. Especially in the sealing area at the upper corner or lower end of the door, tension, compression, and torsional coupling may exist simultaneously within a short distance. If fine-grained state representation is lacking, feedforward control will be difficult to accurately fall on the effective segment. Therefore, this embodiment further introduces the expression method of virtual nodes and spring-damped directed graphs. Specifically, the edge computing server first maps the real-time contour point cloud to several continuous segments along the centerline of the sealing strip, with each segment corresponding to a virtual node. Each node is not equivalent to a single point cloud point, but represents the comprehensive state of a small segment of the sealing strip at the current moment. The point cloud data mainly provides the geometric orientation, bending trend, and deviation relative to the groove of the segment. The torque data is used to describe the degree of effect of the traction, torsion, and pressing loads transmitted along the length of the segment. The system constructs an energy functional equation that incorporates the deformation energy caused by local curvature changes and the work done by external loads, mapping this information to the initial energy state of the nodes. Specifically, the energy functional equation is set as follows: the initial energy state of a node is equal to the local curvature space integral value of the corresponding discrete segment multiplied by the elastic recovery coefficient, and superimposed with the work integral values ​​of the torque data along the tangential and normal directions of the target assembly trajectory; this can be understood as how much mechanical energy that segment of the sealing strip currently stores that may cause subsequent deformation; among them, nodes with high energy usually mean that the segment is in a stress concentration area, and is more likely to experience release rebound or torsional diffusion in the future; Between nodes, the edge computing server establishes directional connection edges based on the preset viscoelastic coefficient relationship of the material; a spring-damped directed graph is used here because the stress transmission inside the sealing strip has both elastic recovery and viscous dissipation characteristics, and the propagation direction is related to the force direction of the actuator, the position of the clamping point, and the continuous topology of the unpressed area; Springs on the connecting edge are used to reflect the deformation and entanglement between adjacent sections; damping on the connecting edge is used to reflect the gradual attenuation of stress waves due to material internal friction during propagation; directionality is used to indicate which direction the current stress is more likely to continue propagating, such as from the pressed section to the unpressed section, or from the clamped end to the free end. A simplified process is described below: Assume that at a certain moment, the sealing strip is divided into node N1, which is close to the stable section that has been pressed into the groove; node N2, which is located in the gripping transition section; node N3, which is located in the bend entrance section; and node N4, which is located in the leading section that has not yet been pressed in. If the point cloud shows that N3 has a local outward turning trend, and the torque data shows that there is an increased torsional load at the end recently, then the system can set the initial energy state of N3 to be higher than that of N2 and N4. Based on the viscoelastic response of the material, the connections between N2 and N3, and between N3 and N4 are set as stronger connections, while the connection between N1 and N2 is set as weaker connections. This is because N1 is constrained by the slot, and its free propagation ability is reduced. The resulting diagram structure can more realistically express the propagation characteristics of the current stress mainly continuing to be pushed from N3 to N4, rather than simply assuming that the entire sealing strip is uniformly stressed. In abnormal or boundary situations, if the number of local points is insufficient when partitioning the point cloud, resulting in unstable geometric representation of a certain node, the edge computing server can merge the node with neighboring nodes to avoid constructing false high-energy regions due to sparse point clouds. If a section of the sealing strip is temporarily blocked by a clamp, the system can prioritize compensation modeling based on the continuity of neighboring nodes and the most recent historical cycle state, but will not output the blocked section as a high-confidence node. If a change in material batch causes the viscoelastic response to be significantly inconsistent with the preset value, the parameter file of the corresponding batch can be reloaded, or a conservative mode can be entered, retaining only the node-level trend judgment while reducing the compensation magnitude. In the same door assembly station, when the robotic arm turns from the A-pillar to the roof beam, the edge computing server divides the sealing strip near the corner into five virtual nodes. Due to the combined effects of traction and torsion on the corner entrance section, this node is identified as a high initial energy node and establishes a strong propagation connection with the node to be assembled in front. The constructed spring-damped directed graph shows that the stress does not remain at the currently visible torsion point, but is transmitted forward to the unassembled area. Based on this, the control system begins to prepare for compensation before the node in front shows a significant positional deviation. The purpose of this step is to compress the complex stress state of a continuous flexible body into a node network structure that can be processed in real time, thereby enabling rapid identification of local high-risk sections and a structured expression of the direction of subsequent stress propagation.

[0017] Furthermore, the edge computing server is also used to: apply a wave propagation dissipation operator containing a damping attenuation coefficient to a virtual tension-strain tensor field; calculate the propagation speed and propagation phase of the deformation stress wave generated by the actuator's action along the automotive seal based on the wave propagation dissipation operator; and predict the amount of stress accumulation in the assembly area of ​​the automotive seal on the assembly trajectory in the future based on the propagation speed and propagation phase.

[0018] This embodiment provides a stress propagation prediction mechanism for future moments; specifically, node-based representation alone is insufficient to solve the time delay problem in high-speed assembly, because high node energy only indicates where stress concentration exists but cannot answer when this stress will propagate to the area to be assembled ahead. In the assembly of automotive sealing strips, many misalignments do not occur at the current stress point, but rather at the leading pressure point after a preset time delay. If the control system waits until the leading point has warped before taking action, it is easy to conflict with the high-speed operation rhythm of the robotic arm. Therefore, this embodiment further introduces a wave propagation dissipation operator to describe the propagation and attenuation law of deformation stress inside the flexible sealing strip. Specifically, the edge computing server applies the wave propagation dissipation operator to the aforementioned spring-damped directed graph; the operator here can be understood as a set of propagation rules: mathematically, the wave propagation dissipation operator is composed of a state transition matrix containing the damping attenuation coefficient and the equivalent length between nodes, which is used to perform iterative multiplication and addition operations on the node energy state vector of the previous control cycle, thereby realizing the propagation rules; First, the stretching or torsion introduced by the current action of the robotic arm will not spread throughout the entire sealing strip instantly, but will propagate segment by segment along the length direction; second, during the propagation process, it will gradually decrease due to material internal friction, groove contact constraints and local friction; third, the curvature and boundary constraints of different sections will change the propagation speed, making the arrival time of the wave crest uneven. Based on these rules, the edge computing server extracts the propagation speed and propagation phase. The propagation speed is used to characterize how fast the stress influence moves forward, and the propagation phase is used to characterize whether a certain assembly area is in the stage of approaching, reaching, or having passed the peak. In this way, the system can convert the currently visible local stress into a prediction of the stress state of the assembly area in the future within a short period of time. Furthermore, in order to avoid the propagation speed and propagation phase remaining in abstract description, the edge computing server updates the energy transfer between adjacent virtual nodes sequentially according to the control cycle. Specifically, the following rules apply: First, based on the current node energy, edge connection strength, damping attenuation coefficient, and velocity change direction in the action state, determine the share of propagable energy from the upstream node to the downstream node within a certain period; then, based on this share and the equivalent length between nodes, calculate how many node distances the stress influence has moved forward within the current period, thereby obtaining the propagation speed. At the same time, record how many periods remain before the node to be assembled reaches the peak energy of this round. If the remaining period is greater than 1, it is recorded as the peak is approaching; if it is approximately equal to 1 control period, it is recorded as the peak has arrived. If the peak value has been exceeded and the decay has begun, it is recorded as the peak value has been exceeded. In other words, the propagation phase in this embodiment is preferably expressed using discrete phase quantities, rather than requiring a highly complex solution for the continuous waveform, so as to be compatible with the real-time computing capabilities of the edge computing server. Furthermore, the cumulative stress over future time is not directly given by the instantaneous value of a single node, but is jointly determined by the cumulative transferred energy received by the area to be assembled within the prediction window, the initial energy that the area itself has not yet released, and the dissipation caused by local contact constraints. For nodes that have been stabilized by the slot, their transfer weight should be reduced; for nodes that are close to the free end or corner entrance and have not yet been pressed in, their forward receiving weight should be increased. This allows the prediction results to distinguish between two different risk states: energy is still in the process of propagation and energy is about to accumulate in front, thus avoiding classifying all forward nodes as high-risk. A simplified process is described below: Along the sealing strip, there is an upstream constraint node N1 near the already press-fitted stable section, a transition node N2 currently mainly affected by gripping traction, a main excitation node N3 currently under significant stress and transmitting energy forward, a leading risk node N4 near the corner entrance to be assembled, and a further forward free node N5 downstream of N4 that has not yet entered the main wave crest action zone. The current action of the robotic arm mainly falls between N2 and N3. If the material damping is small and the curvature of this section is gentle, the stress effect from N3 will be transmitted to N4 more quickly. If N4 is about to enter a high curvature corner, although the propagation continues to move forward, local accumulation will occur at the corner entrance, which means that the future stress accumulation of N4 is higher than that of N5. Conversely, if the same force occurs near the stable section that has been pressed into the groove, the wave energy will be dissipated faster because the groove provides constraint on the sealing strip, and the impact on the leading area will be weaker. It can be seen that this embodiment does not simply predict geometric deviation, but predicts when the area to be assembled will become unstable due to the arrival of stress that is not yet visible in front. The aforementioned N1 to N5 are only used to illustrate the propagation order and relative positional relationship, and do not change the technical meaning of the virtual nodes mentioned above. In abnormal or boundary situations, if the actuator suddenly slows down or stops, causing the original propagation trend to be interrupted, the edge computing server should update the propagation speed and propagation phase again with the current action state as the boundary to avoid using outdated prediction results. If it is detected that a certain section of the sealing strip has been actually pressed into the groove and formed a contact lock, the propagation path after that node should reduce the free propagation weight to avoid the constrained energy being mistakenly passed forward. If the action status message arrives late in a certain cycle, the system can use the action trend of the previous stable cycle for short-term extrapolation, but it is only effective within a limited cycle. After that, the future prediction is paused and conservative control is switched to. During the pressing stage at the upper corner of the right front door, the robotic arm moves quickly along the door contour. The current gripping point is still located on the upper section of the A-pillar, while the assembly area is already close to the starting section of the upper crossbeam. Based on the damping attenuation characteristics of the material near the corner, the edge computing server judges that the torsional stress will be transmitted to the pressing area in front in a short time and form a local peak at the upper corner transition point. Although the industrial camera has not yet seen obvious warping at the front, the system has predicted that high stress accumulation will occur in this area during the upcoming control cycle, so it prepares compensation actions in advance. The purpose of this step is to transform the current stress into a future predicted deviation trend, thereby enabling early intervention in the assembly area and reducing feedback catch-up and oscillation control caused by viscoelastic hysteresis.

[0019] Further, residual analysis is performed on the accumulated stress and the observed visual distortion to extract the residual internal stress gradient, including: converting the accumulated stress into a stress level score and converting the observed visual distortion into a distortion score; calculating the deviation between the stress level score and the distortion score as the fitting residual; if the fitting residual is greater than a preset residual threshold, the fitting residual is used as the residual internal stress gradient; otherwise, the residual internal stress gradient is determined to be zero.

[0020] This embodiment provides a residual extraction mechanism for identifying latent internal stress. Specifically, in the aforementioned scheme, the edge computing server is already able to predict future stress accumulation based on material parameters and stress state. However, there is still a practical problem in industrial settings: the visible distortion of the sealing strip surface is not always completely synchronized with the internal energy. Sometimes the surface deformation is within the preset tolerance range, but the material has already accumulated enough stress to be released in the next stage. Sometimes slight ripples may appear visually, but this is a normal transitional deformation and will not affect the final pressing quality. If the model prediction and actual observation are not uniformly verified, the control system may overcompensate or undercompensate. Therefore, this embodiment performs residual analysis on the predicted stress results and visual distortion observations to extract the residual internal stress gradient that is truly worth intervening in. Specifically, the edge computing server views the accumulated stress as the deformation trend that the material should exhibit; it views the visual distortion observation as the actual shape that the sealing strip surface has already exhibited, reflecting the deviation between the two, i.e. the fitting residual, which can reflect whether the internal stress is still retained by the material's viscoelasticity; if the prediction shows that a certain section has accumulated a large amount of stress, but the visual distortion is only very small, it means that the section may be in a critical unstable state that is about to be released but has not yet been released. If the prediction is basically consistent with the vision, it means that the material response is consistent with the model. The current additional compensation intervention system that exceeds the conventional threshold will identify the deviation that is greater than the preset residual threshold as the residual internal stress gradient, which is used to prompt the controller to perform feedforward compensation. For deviations that do not exceed the threshold, they are regarded as normal modeling errors, minor texture disturbances or imaging noise, and are not used as the basis for compensation. Furthermore, to avoid direct comparison between stress accumulation and visual distortion observations due to their different dimensions, the edge computing server maps both to the same risk dimension before calculating the fitting residual. Specifically, the stress accumulation can be converted into a stress level score according to the historical safety range corresponding to the current vehicle model, material batch, and trajectory segment, and the visual distortion observations can be converted into a distortion score according to the relative deviation of the slot normal, the local warping height, the cross-sectional torsion amplitude, and the degree of ripple undulation. Specifically, a normalization function and preset weighting coefficients are used to map the accumulated stress to a stress level score within a preset standard range. At the same time, a multi-dimensional weighted summation model is used to sum the visual distortion observations, which include relative deviation from the slot normal, local warping height, cross-sectional torsion amplitude, and ripple undulation degree, according to the preset weights of each dimension, and then normalize them to a distortion score of the corresponding magnitude. The two are then calculated after falling into the same scoring range. In this way, the fitting residual represents the difference between the predicted risk and the actual appearance risk, rather than simply subtracting heterogeneous data such as Newtons, millimeters, or angles, thus avoiding distortion in the comparison. Furthermore, the residual internal stress gradient is preferably extracted along the length of the sealing strip or along the direction of the virtual node sequence, rather than outputting only a single-point difference. That is, the edge computing server can compare the changing trends of the fitting residuals of adjacent nodes. If the residual of a continuous section gradually increases from upstream to downstream, it indicates that the implicit energy is accumulating in the leading assembly area, and this growth trend can be regarded as the residual internal stress gradient. If only a few discrete nodes have short-term high residuals and the nodes before and after are discontinuous, they are preferentially regarded as local observation noise or occasional disturbances and are not directly used as control inputs. This can keep the term gradient consistent with the segmental compensation in subsequent control and avoid mistaking a single point anomaly as the entire stress shift. To illustrate with a simplified implementation scenario: Suppose that for the same node segment, the model predicts high stress accumulation, while the visual distortion observation is only medium to low; then the residual is judged to be significant, indicating that there is still unreleased deformation driving force within the segment, which should be extracted as residual internal stress gradient; conversely, if another segment is predicted to be medium, and the visual observation is also medium; then the residual is small, and no additional hidden risk is considered; furthermore, if the visual observation is high, but the prediction is low; then the system will further check whether the segment is affected by external occlusion, local contamination, point cloud stitching errors, or sudden collisions, to prevent non-material internal stress factors from being mistakenly taken as control targets; As an anomaly protection mechanism, the preset residual threshold should be calibrated in combination with vehicle model, material batch and camera resolution. The preset residual threshold is specifically the maximum allowable deviation fluctuation constant calculated by comparing historical normal assembly samples on visual distortion score and stress level score. When there is a large light interference or obvious reflection on the surface of the sealing strip, the visual distortion observation may be amplified. The system can improve the continuous period requirement for the residual to be established, that is, only when multiple consecutive periods exceed the threshold will the residual internal stress gradient be output. If the model prediction deviates from the overall direction due to abnormal material parameter loading, the residual results can be prevented from directly entering the control loop through calibration state verification, and should be manually confirmed or rolled back to a conservative trajectory. If the residual remains close to zero for a long time, but the final assembly contour comparison results show a systematic deviation, it indicates that the threshold is too large or the model is too blunt, and it should be recalibrated. During the assembly process at the upper corner of the car door, the edge computing server predicted that a high level of stress accumulation had formed in the leading area, but the industrial camera only observed a slight outward bulge, which had not yet reached a significant mismatch level. Based on this, the system believed that there was still unmanifested residual internal stress in the area and extracted the residual as an internal stress gradient that gradually increased along the length of the sealing strip, and sent it to the programmable logic controller. Conversely, in the straight section at the bottom of the car door, since the sealing strip was subjected to uniform stress and the visual distortion was basically consistent with the model prediction, the residual did not exceed the threshold. The system regarded this section as a normal assembly process and did not trigger additional compensation. The purpose of this step is to distinguish between visible deformation and unmanifested internal stress, thereby identifying hidden risks and avoiding delayed or excessive control decisions based solely on visual appearances.

[0021] Furthermore, the programmable logic controller is also used to: calculate the reverse compensation energy required to counteract the deformation stress wave based on the residual internal stress gradient; convert the reverse compensation energy into the fine-tuning speed offset and pose offset of the actuator according to the preset energy-kinematic mapping relationship; and generate feedforward compensation control instructions with timestamps based on the fine-tuning speed offset and pose offset.

[0022] This embodiment provides a feedforward control mechanism that converts stress state into mechanical action; specifically, the aforementioned scheme can identify residual internal stress gradient, but if the gradient is only output as an alarm quantity, it still cannot directly act on the actuator; the control objective in automotive sealing strip assembly is not simply to eliminate the error, but to release or counteract the stress wave that is about to reach the assembly area at a pre-calculated speed and posture at a preset matching time. If the compensation action does not match the energy state inside the sealing strip, two types of problems may occur: first, insufficient compensation, which cannot prevent the corner from lifting up; second, excessive compensation, which may introduce new torsion and local pressure damage. Therefore, this embodiment introduces reverse compensation energy and its mapping relationship to kinematic parameters. Specifically, after receiving the residual internal stress gradient, the programmable logic controller first determines the stress wave intensity level and the area of ​​action that needs to be offset, and then calculates the corresponding reverse compensation energy. Here, "reverse" does not necessarily mean completely opposite to the current direction of action, but rather refers to the part of the control energy that can weaken, disperse, or delay the arrival of the wave peak at the assembly area in terms of physical effect. For example, if the current stress is mainly manifested as torsional load forward transmission, then reverse compensation can be manifested as a combination of small anti-torsion and cycle slowdown. If the current stress is mainly manifested as local tensile energy storage, then reverse compensation can be manifested as a slow-release posture retraction along the tangential direction and a timing adjustment of the normal indentation force. Based on the preset energy-kinematic mapping relationship, the system converts the reverse compensation energy into executable fine-tuning speed offset and posture offset. The speed offset determines whether the robotic arm decelerates, accelerates, or maintains its speed in a local section according to a preset ratio. The posture offset determines the direction of the end effector's small posture correction in space, such as slight twisting around the end effector axis, slight pre-compression along the slot normal, or slight release along the tangential direction. Furthermore, the energy-kinematic mapping relationship is preferably established by combining categorized lookup tables with segment constraints, rather than directly converting abstract energy values ​​into single motion quantities. Specifically, the residual internal stress gradient can be classified into four categories according to the dominant type: torsional dominant, tension dominant, compression dominant, and coupling dominant, and then corresponding to different kinematic adjustment priorities: when torsional dominant, priority is given to increasing the correction ratio around the end axis in the pose offset. When tension is dominant, priority is given to configuring the velocity offset and slight back-off along the trajectory tangential direction; when compression is dominant, priority is given to configuring the normal pressing angle and pressing timing correction; when coupling is dominant, a multi-component limiting combination is used; this ensures that the velocity offset and pose offset output by the controller are consistent with the stress source, avoiding different force mechanisms being incorrectly processed by the same motion template; during mapping conversion, the Joule value of the reverse compensation energy is directly converted into the rotation angle offset and velocity increase / decrease value along the trajectory tangential direction of the corresponding joint at the end of the actuator according to the preset linear or polynomial proportional coefficient; Furthermore, the feedforward compensation control instruction with timestamp should include at least three elements: target trajectory segment identifier, compensation start time, and compensation duration cycle number; the target trajectory segment identifier is used to ensure that the compensation action is performed near the corresponding energy balance check point, preventing the risk of the previous segment from being incorrectly carried over to the next segment; the compensation start time is used to align with the current position and speed of the robotic arm and the target position in the standard assembly reference model; The number of compensation duration cycles is used to limit the duration of the compensation action, preventing local energy release actions from continuing to act after passing through high-risk areas, thereby disrupting the stable assembly of subsequent straight segments; therefore, the timestamp is not just a simple record of the sending time, but execution constraint information bound to the specific trajectory landing point; A simplified implementation scenario is used as follows: Suppose the residual internal stress gradient of a certain section to be assembled is identified as continuously increasing along the length direction and dominated by torsion; then the controller can map it as a speed offset that decreases by a first preset ratio, and an attitude offset that rotates in the opposite direction around the end axis by a second preset offset angle; if another section is identified as dominated by local compression and energy storage, it can be mapped to maintain the speed but slightly adjust the normal pressing angle; so that the sealing strip releases a portion of the local compression energy before entering the slot; the feedforward compensation control command with timestamp is used to prevent these adjustments from causing timing conflicts or being triggered out of order, but instead corresponds one-to-one with the trajectory position that the robotic arm is about to reach, preventing misalignment of the action timing; As an anomaly protection mechanism, if the amplitude of the residual internal stress gradient is lower than the minimum control resolution that the system can stably execute, no new feedforward compensation action will be generated to avoid frequent shaking of the robotic arm under noise drive. If compensation requirements from multiple sections conflict within the same control cycle, such as one section requiring deceleration while another section requires acceleration, the controller will prioritize the high-risk section that is about to enter the pressing window and postpone or weaken the low-priority compensation. If there is a safety interlock state at the current workstation, such as the door clamp not being locked, the robotic arm approaching the anti-collision boundary, or the manual intervention signal being activated, all timestamped feedforward compensation instructions should be issued only after safety adjudication. In the short period before the right front door enters the pressing position, the residual internal stress gradient output by the edge computing server shows that there is a clear trend of the torsional peak moving forward in the leading section of the sealing strip. Based on this, the programmable logic controller determines that a set of compensation actions is needed, which mainly releases the torsion and secondarily reduces the accumulation of the peak. It converts this into a slight reverse attitude offset of the sixth axis at the end and a small speed reduction along the trajectory direction. The compensation instruction is accompanied by a corresponding execution timestamp to ensure that the robotic arm completes the pre-action before reaching the corner entrance, rather than correcting the deviation after it has been pressed in. The purpose of this step is to transform the abstract internal stress risk into a directly executable kinematic correction, thereby achieving a correspondence between the control quantity and the material stress state, and improving the feasibility and stability of feedforward compensation.

[0023] Furthermore, the programmable logic controller is also used to: extract the position of the current action state of the actuator on the preset target assembly trajectory based on the timestamp as the assembly progress, and determine whether the remaining time span between the current assembly progress of the actuator and the preset key trajectory point in the standard assembly reference model is less than or equal to the preset time advance; if it is less than or equal to the preset time advance, control the actuator to perform a reverse micro-torsion action to release internal stress; otherwise, maintain the current motion state of the actuator. This embodiment provides an action release mechanism triggered by critical moments. Specifically, in the previous embodiment, the controller was already able to generate feedforward compensation control commands with timestamps. However, if all compensation is executed immediately upon completion, it often disrupts the currently stable bonding state. For flexible sealing strips, the timing of releasing internal stress is just as important as the magnitude of the action. If executed too early, the release effect will accumulate again before truly entering the high-risk area; if executed too late, it will be impossible to offset the stress before the peak arrives. Therefore, this embodiment further introduces a judgment logic for a preset time lead, so that the compensation action is implemented within the optimal window relative to the standard assembly reference model. Specifically, the programmable logic controller determines whether the actuator has reached a certain key trajectory position by a preset lead time based on the timestamp and the current assembly progress. This lead time is not simply a fixed time, but a control window that is adapted to the vehicle trajectory length, current running speed, corner curvature, and material release time constant. Specifically, the preset time advance is determined by calculating the ratio of the expected high curvature trajectory distance where the peaks are expected to accumulate to the current operating speed of the actuator to obtain the base time, and then adding the compensation margin corresponding to the material release time constant; if it is determined that the window has been reached, the actuator is controlled to perform a reverse micro-torsion action to release the torsional internal stress before the sealing strip actually enters the high curvature groove. If the window has not yet been reached, the current motion state is maintained to prevent premature disturbance of the sealing strip in an inappropriate position; the reverse micro-twist here is a small-angle, short-duration attitude movement performed against the identified dominant torsional direction, the purpose of which is to release the torsional energy stored in the sealing strip cross section, rather than to change the overall assembly trajectory. A simplified implementation scenario is used to illustrate this: Assume that the key position K is the entrance at the upper corner of the car door. According to historical calibration, the system knows that under the current material and speed conditions, the reverse micro-twist should be triggered within a short window before reaching K to be effective. If the controller determines that the robotic arm is still in a position beyond the distance threshold corresponding to the preset time advance before K, it will maintain the original motion and will not apply attitude disturbance in advance. Once the progress enters this window, the controller triggers one or several micro-torsion actions to partially release the torsional load on the high-energy node before it enters the slot; thus, when the sealing strip reaches K, it is closer to the initial assembly position with no residual internal stress. As an anomaly protection mechanism, if the current speed changes abruptly, the previously set time advance may lose its corresponding relationship. In this case, the controller should recalculate the trigger window based on the real-time speed. If the robotic arm waits for a short time due to fluctuations in the upstream station's cycle time, all compensation commands that have not yet arrived should be rescheduled to prevent accidental triggering of reverse micro-torsion while waiting. If it is detected that the sealing strip has formed a strong contact constraint with the groove, the release-type micro-torsion will no longer be executed to avoid pulling out or wearing the pressed section. Instead, the current motion state will be maintained or a gentler speed compensation will be adopted. During the assembly process at the upper corner of the same right front door, the controller had received an attitude compensation command for the corner entrance in advance, but the robotic arm was still located in the upper part of the A-pillar at this time. The system judged that it had not yet entered the preset advance window, so it maintained the current pressing progress. As the robotic arm approached the upper corner entrance, the assembly progress met the triggering conditions, and the controller immediately executed a short-term reverse micro-torsion action to release the torsional stress of the sealing strip section before it entered the groove. The robotic arm continued to press into the upper crossbeam section according to the predetermined trajectory, reducing outward turning and edge rebound at the corner. The purpose of this step is to ensure that feedforward compensation acts on the most sensitive trajectory position at the most effective time, thereby achieving precise timing control of the internal stress release action and avoiding premature release failure and delayed release failure.

[0024] Furthermore, the industrial data acquisition terminal includes an industrial camera and a six-dimensional torque sensor; wherein, the industrial camera is used to acquire contour point cloud data of automotive seals; the six-dimensional torque sensor is used to acquire torque data of the actuator; the industrial data acquisition terminal is also used to acquire the action status of the actuator through a communication interface.

[0025] This embodiment provides a hardware configuration mechanism for data acquisition at the workstation. Specifically, in the control link described above, if the acquisition source is too singular, it is impossible to simultaneously perceive the external shape and the internal stress trend. For example, relying solely on a camera can only show the deviation that has already appeared on the surface of the sealing strip, and it is difficult to reflect the internal stress that has not yet manifested. Relying solely on a torque sensor makes it difficult to distinguish whether the force comes from normal press-fit contact or from local torsion accumulation. Therefore, this embodiment adopts a combined acquisition architecture of industrial camera, six-dimensional torque sensor and communication interface, so that the edge computing server can obtain complementary information. Specifically, an industrial camera is installed at a vantage point that covers the end-effector's working area and the leading assembly area to collect point cloud data of the sealing strip's local contour. Point clouds are used instead of simple two-dimensional images because anomalies in the sealing strip are manifested not only in changes in edge position but also in three-dimensional deviations in cross-sectional orientation, outward flaring height, and relative groove normal. A six-dimensional torque sensor is positioned between the end of the actuator and the clamping device to acquire triaxial force and torque information, reflecting the mechanical interaction state between the sensor and the sealing strip during gripping, traction, pressing, and torsion processes. The communication interface reads the motion status from the actuator control bus, including the current pose, joint speed, trajectory segment switching information and process stage signals. After combining these three, the visual morphological features, mechanical force status and actuator motion parameters can be uniformly sent to the edge computing server. To illustrate with a simplified implementation scenario: If, at a certain moment, the industrial camera detects that the edge of the sealing strip's leading section has lifted, and the six-dimensional torque sensor simultaneously detects an increase in the end-axis torque, and the communication interface indicates that the actuator is in the acceleration phase before entering a corner, the system can determine that the anomaly is due to torsional stress accumulation caused by high-curvature trajectory motion, consistent with the preset confidence level. If the camera observes the edge lifting, but the torque data is stable and the motion status shows the robotic arm is stationary, it is more likely that the apparent anomaly is caused by external obstruction, lighting errors, or micro-vibration of the fixture. Thus, it is evident that multi-source acquisition plays a fundamental supporting role in subsequent residual analysis and feedforward compensation. As an anomaly protection mechanism, if the industrial camera experiences a short-term defocusing or excessive surface reflection, the system can activate a transition mode that combines the most recent stable point cloud frame with the current torque data, but the duration should be limited; if the six-dimensional torque sensor experiences zero-point drift, the edge computing server can automatically correct the baseline during the no-load calibration phase, and if the drift exceeds the limit, the channel will be disabled and the control level will be reduced; if the communication interface data update is interrupted, long-term prediction should not continue in an outdated action state, but the system should be reverted to a safe mode that only performs local feedback monitoring. During the continuous cycle operation of the car door assembly line, an industrial camera is fixedly installed above the outer contour of the car door, covering the transition area from the A-pillar to the upper crossbeam; a six-dimensional torque sensor is located before the pressing head at the end of the robotic arm; the communication interface is connected to the robotic arm controller in real time; when the assembly reaches the upper corner of the car door, the three types of data simultaneously reflect the increase in local distortion, the increase in torque, and the trajectory entering the high curvature section. Based on this, the edge computing server quickly establishes a reliable stress propagation judgment and outputs effective compensation basis to the downstream controller. Furthermore, to maintain complete consistency between the terminology used in this embodiment and the aforementioned embodiments, the camera in this embodiment refers to the same visual acquisition component as the industrial camera in the embodiments, and the data type output by the industrial camera is uniformly contour point cloud data; the six-dimensional torque sensor in this embodiment is the only corresponding hardware source for acquiring torque data of the actuator in the embodiments, and its output is uniformly classified as torque data; In this embodiment, the communication interface only undertakes the function of acquiring and transmitting action status, and does not replace the computing function of the edge computing server and the control function of the programmable logic controller. As a result, each hardware and data item in the industrial data acquisition terminal forms a one-to-one correspondence, that is, the industrial camera corresponds to the contour point cloud data, the six-dimensional torque sensor corresponds to the stress torque data, and the communication interface corresponds to the action status, thus avoiding the confusion of the same data object being referred to by different names in the whole text. The purpose of this step is to provide a three-dimensional data foundation of form, force, and motion for feedforward control, thereby achieving a comprehensive perception of the flexible assembly state and improving the reliability of subsequent modeling and compensation.

[0026] Furthermore, the actuator is a six-axis robotic arm; the system also includes an industrial computer, which is used to discretize the assembly trajectory of the actuator into energy balance check points.

[0027] This embodiment provides an execution and discrete planning mechanism adapted to complex spatial trajectories. Specifically, in the assembly of door sealing strips, the trajectory includes not only long straight segments but also multiple rounded corners and transition segments with significant curvature changes. If the control system only treats the entire trajectory as a continuous smooth curve without distinguishing which positions are more prone to energy accumulation and release, the feedforward compensation will lack a landing point, resulting in overly averaged control. On the other hand, the sealing strip not only needs to move along the door frame boundary in three-dimensional space, but also needs to enter the slot under different normal and tangential relationships. This determines that the actuator should be a six-axis robotic arm with multi-posture adjustment capability. Based on this, this embodiment uses a six-axis robotic arm as the actuator, and the industrial control computer discretizes the assembly trajectory into multiple energy balance check points. Specifically, before the assembly starts, the industrial control computer loads the three-dimensional model of the door, the standard trajectory and the process parameters, and divides the entire assembly trajectory into multiple checkpoints according to geometric features and process risks. The so-called energy balance checkpoints refer to the locations where the system pays special attention to whether the sealing strip will change from a stable fit state to a stress accumulation state, or from a stress accumulation state to a release state. Common checkpoints include: the initial gripping entry point, the transition point from the straight section to the corner, the point where the upper corner peak is likely to form, the speed recovery point in the middle section of the long side, and the end taper point; the six-axis robotic arm executes different posture strategies based on these checkpoints, because the six-axis structure can simultaneously complete composite actions such as trajectory advancement, micro-torsion around the axis, normal pressing, and local posture repositioning, which is more suitable for the spatial fitting requirements of flexible sealing strips; A simplified implementation scenario is used to illustrate this: The industrial control computer can discretize a trajectory that starts from the lower edge of the car door, rises through the A-pillar, turns into the upper crossbeam, and then descends to the B-pillar into five types of inspection points: the initial grab entry point P1, the transition point from the straight section to the corner P2, the point where the upper corner peak is likely to form P3, the speed recovery point in the middle section of the long side P4, and the end closing point P5. If the edge computing server identifies a stress forward shift trend near the transition point P2 from the straight section to the corner, the controller can focus on feedforward compensation at the point P3 where the upper corner peak is easily formed; if the stress has significantly decreased at the velocity recovery point P4 in the middle of the long side, the control strategy can resume normal propulsion; P1 to P5 are only used as checkpoint markers after trajectory discretization to establish a one-to-one correspondence between subsequent checkpoints, predictions, and compensation. As an anomaly protection mechanism, if the actual vehicle model changes and the industrial control computer fails to switch the corresponding trajectory template in time, the old checkpoint configuration must not be used directly in production; if a checkpoint loses its original physical meaning due to tooling interference, fixture modification or workstation adjustment, it should be recalibrated before use; if a joint of the six-axis robotic arm enters the limit protection, resulting in the inability to complete the planned posture compensation, the control system can pre-allocate some compensation actions to the adjacent checkpoint, or reduce the cycle time when necessary, to ensure that no mechanical interference or assembly damage occurs. In the assembly line of new energy vehicle doors, the industrial control computer discretizes the pressing trajectory of the right front door sealing strip into multiple inspection points and marks the upper corner entry point as a high-risk energy balance inspection point. Before approaching the inspection point, the six-axis robotic arm continuously makes predictions based on the stress state output by the edge computing server. Once it is determined that there is a trend of the peak moving forward, the controller combines the spatial attitude constraints of the inspection point and issues micro-torsion and speed offset commands in advance. If the upper corner is successfully passed, the subsequent straight section resumes high-frequency operation. Furthermore, to ensure that the hardware relationships remain consistent with the aforementioned implementation methods, the six-axis robotic arm in this implementation is a specific definition of the actuator. Therefore, the robotic arms appearing in subsequent examples should all be understood as actuators, rather than new modules parallel to actuators. The industrial control computer in this implementation is only responsible for discretizing the assembly trajectory into energy balance checkpoints and providing trajectory checkpoint indexes to relevant modules. It does not replace the edge computing server's function of calculating the virtual tension-strain tensor field, stress accumulation, and residual internal stress gradient, nor does it replace the function of the programmable logic controller in generating and issuing feedforward compensation control commands. Accordingly, the functional division of labor in the entire text is unified as follows: the industrial control computer is responsible for trajectory discretization and checkpoint organization, the industrial data acquisition terminal is responsible for data acquisition, the edge computing server is responsible for stress modeling and prediction, the programmable logic controller is responsible for control decision-making, and the six-axis robotic arm is responsible for performing assembly actions. Through this unified definition, the robotic arm controller, industrial control computer, and programmable logic controller can be avoided from being mistakenly regarded as interchangeable in the description, thereby ensuring the consistency of hardware connection relationships and algorithm flow. The purpose of this step is to transform continuous and complex trajectories into key state nodes suitable for real-time control, and to leverage the multi-degree-of-freedom attitude capabilities of a six-axis robotic arm to achieve the plannable, predictable, and compensable assembly of flexible sealing strips under complex door contours.

[0028] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A positioning and adjustment system for automotive seals based on industrial data processing, used for assembling and positioning adjustments of automotive seals, characterized in that, include: Industrial data acquisition terminals, edge computing servers, and programmable logic controllers; The industrial data acquisition terminal is used to acquire torque data and action status from external actuators, and to acquire contour point cloud data of the automotive seal, and to send the contour point cloud data, the torque data and the action status to the edge computing server. The edge computing server is used to acquire a 3D model of the car door as a standard assembly reference model; based on a pre-configured preset viscoelastic coefficient matrix characterizing the physical properties of the automotive seal, the preset viscoelastic coefficient matrix includes the relaxation time constant, shear modulus and elastic recovery coefficient corresponding to the discrete segment of the automotive seal, and combined with the contour point cloud data and the torque data, a virtual tension-strain tensor field is constructed. Based on the action state and the virtual tension-strain tensor field, the dynamic stress transmission characteristics in the virtual tension-strain tensor field are extracted as deformation stress waves, and the stress accumulation of the deformation stress waves is calculated; the contour point cloud data is spatially registered and aligned with the standard assembly reference model. Based on the registration and alignment results, the point cloud coordinate space deviation is extracted from the contour point cloud data as the visual distortion observation value. Residual analysis is performed on the stress accumulation and the visual distortion observation value to extract the residual internal stress gradient and send the residual internal stress gradient to the programmable logic controller. The programmable logic controller is communicatively connected to the actuator and is used to generate feedforward compensation control commands based on the standard assembly reference model and the residual internal stress gradient, and send the feedforward compensation control commands to the actuator.

2. The automotive seal positioning and adjustment system based on industrial data processing according to claim 1, characterized in that, The edge computing server is also used for: The discrete segments of the automotive seal are defined as multiple virtual nodes, and the contour point cloud data and the torque data are mapped to the initial energy state of each virtual node. Based on the initial energy state and the preset viscoelastic coefficient matrix, a spring-damped directed graph composed of the virtual nodes is constructed as the virtual tension-strain tensor field.

3. The automotive seal positioning and adjustment system based on industrial data processing according to claim 2, characterized in that, The edge computing server is also used for: The wave propagation dissipation operator, which includes a damping attenuation coefficient, is applied to the virtual tension-strain tensor field. Based on the wave propagation dissipation operator, the propagation speed and propagation phase of the deformation stress wave generated by the actuator's action are calculated along the automotive seal. Based on the propagation speed and the propagation phase, the cumulative stress in the area to be assembled on the assembly trajectory of the automotive seal is predicted over future time.

4. The automotive seal positioning and adjustment system based on industrial data processing according to claim 1, characterized in that, The residual analysis of the accumulated stress and the observed visual distortion to extract the residual internal stress gradient includes: The accumulated stress is converted into a stress level score, and the visual distortion observation is converted into a distortion score; the deviation between the stress level score and the distortion score is calculated as the fitting residual. If the fitting residual is greater than a preset residual threshold, then the fitting residual is used as the residual internal stress gradient; otherwise, the residual internal stress gradient is determined to be zero.

5. The automotive seal positioning and adjustment system based on industrial data processing according to claim 1, characterized in that, The programmable logic controller is also used for: Based on the residual internal stress gradient, calculate the reverse compensation energy required to counteract the deformation stress wave; Based on the preset energy-kinematic mapping relationship, the reverse compensation energy is converted into the fine-tuning speed offset and pose offset of the actuator; Based on the fine-tuning speed offset and the pose offset, the feedforward compensation control command with a timestamp is generated.

6. The automotive seal positioning and adjustment system based on industrial data processing according to claim 5, characterized in that, The programmable logic controller is also used for: Based on the timestamp, the position of the current action state of the actuator on the preset target assembly trajectory is extracted as the assembly progress. It is determined whether the remaining time span between the current assembly progress of the actuator and the preset key trajectory point in the standard assembly reference model is less than or equal to the preset time advance. If it is less than or equal to the preset time advance, the actuator is controlled to perform a reverse micro-torsion action to release internal stress. Otherwise, the current motion state of the actuator is maintained.

7. The automotive seal positioning and adjustment system based on industrial data processing according to claim 1, characterized in that, The industrial data acquisition terminal includes: an industrial camera and a six-dimensional torque sensor; The industrial camera is used to collect the contour point cloud data of the automotive seal; The six-dimensional torque sensor is used to collect the torque data of the actuator; The industrial data acquisition terminal is also used to acquire the action status of the actuator through a communication interface.

8. The automotive seal positioning and adjustment system based on industrial data processing according to claim 1, characterized in that, The actuator is a six-axis robotic arm; The system further includes an industrial control computer, which is used to discretize the assembly trajectory of the actuator into energy balance check points.