High-precision self-adaptive assembling and adjusting system for automobile door cover based on physical field digital twinning

Abstract

The invention discloses an automobile door cover high-precision self-adaptive assembling and adjusting system based on physical field digital twinning, and the system comprises a multi-mode flexible sensing system which is used for collecting the geometric and mechanical data of a door cover and a door frame in real time; the digital twinning and real-time simulation prediction module is used for constructing and synchronizing a high-fidelity virtual model and predicting a gap and a surface difference after assembly based on a physical information neural network; the reinforcement learning intelligent decision model is used for outputting an optimal adjustment instruction according to the prediction deviation; and the automatic guiding and executing module is used for executing the adjusting instruction and completing assembling and tightening of the door cover. According to the automobile door cover high-precision self-adaptive assembling and adjusting system based on physical field digital twinning, high automation of door cover assembling is achieved through intelligent sensing, real-time optimization and self-adaptive control.

Description

Technical Field

[0001] This invention belongs to the field of automotive intelligent manufacturing technology, specifically a high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins. Background Technology

[0002] In the automotive manufacturing industry, the assembly and adjustment of doors and hoods (including four doors and two hoods) is a crucial step in the overall vehicle manufacturing process. The accuracy of door and hood assembly not only affects the consistency of the vehicle's appearance but also directly impacts sealing performance, safety performance, and service life. Currently, there are two main operating modes for door and hood installation in the industry: manual installation assisted by a power-assisted arm and automated installation using industrial robots. In the manual installation mode, door and hood assembly typically requires multiple people working together, resulting in high labor intensity. Furthermore, the assembly quality largely depends on the operator's experience and skill level, easily leading to inconsistencies. While the industrial robot installation mode achieves partial automation, it suffers from limitations such as insufficient visual perception accuracy, a single rigid positioning method, and poor adaptability to dynamic working conditions, resulting in a low first-pass yield rate. This necessitates heavy reliance on subsequent manual adjustments and refinements to meet assembly requirements.

[0003] As the automotive industry shifts towards multi-model, small-batch, and flexible production, the traditional assembly and adjustment model relying on human experience and rigid robot teaching is no longer sufficient to meet the comprehensive requirements of high precision, high efficiency, and consistency. Therefore, the industry urgently needs a revolutionary technological solution that can overcome existing limitations. This solution must be able to construct a real-time digital twin and intelligent agent that closely approximates physical reality, endowing automated systems with the ability to perceive and think. This would enable the system to adapt to the individual differences of each workpiece in real time, like an experienced expert, achieving intelligent assembly and adjustment with high precision, high efficiency, and high consistency. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins, which achieves a high degree of automation in door cover assembly through intelligent sensing, real-time optimization and adaptive control.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins includes: A multimodal flexible sensing system is used to collect geometric and mechanical data of the door cover and door frame in real time; The digital twin and real-time simulation prediction module is used to build and synchronize a high-fidelity virtual model and predict the gaps and surface differences after assembly based on physical information neural networks. A reinforcement learning intelligent decision-making model is used to output the optimal adjustment instructions based on the deviation between the predicted gap and the surface difference; An automated guidance and execution module is used to execute the adjustment instructions and complete the assembly and tightening of the door cover.

[0006] Furthermore, the multimodal flexible sensing system includes: A 3D optical scanner is used to acquire geometric data, including high-density 3D point cloud data of the door cover, door frame and surrounding area, as well as gap and surface difference data of hundreds of key measuring points distributed along the edge of the door cover. A six-dimensional force / torque sensor is used to collect mechanical data, including contact forces and torques during assembly.

[0007] Furthermore, the digital twin and real-time simulation prediction module includes: A high-fidelity twin model, including geometric and physical property models of the door cover, door frame, hinges, sealing strips, and bolts; A physical information neural network model is used to predict the final gap and surface difference distribution of the door cover after elastic deformation under the action of gravity, assembly stress and sealing strip extrusion force; A real-time state synchronization engine is used to perform millisecond-level data synchronization and state mapping between the real-time data collected by the multimodal flexible sensing system and the high-fidelity twin model.

[0008] Furthermore, the loss function of the physical information neural network model includes a data-driven term and a physical law constraint term, expressed as: in: It is a hyperparameter that balances the two; It is the mean square error between the model's predicted value and the actual measured value; It is the residual based on the Navier-Cauchy equations of elasticity; for the displacement field Its residual Defined as: in: and These are Lamé parameters; It is gravity; Let L2 norm be the integral of the residual over the solution domain.

[0009] Furthermore, the reinforcement learning intelligent decision-making model includes a reinforcement learning-based agent, which uses the high-fidelity twin model as its simulation environment for interactive training, and: The state of the agent is defined as a vector consisting of the deviations between the current gap measurement values ​​and surface difference measurement values ​​of all key measuring points and the ideal target value, expressed as: in: It is a state vector; and These are the clearance deviation item and the surface difference deviation item, respectively. , Let be the number of key measuring points, and: in: and They are respectively in For the first moment The gap value and surface difference value obtained from the measurement at each measuring point; and These are the ideal target values ​​for gap and surface difference, respectively; The action of the intelligent agent is defined as: a six-degree-of-freedom fine-tuning command vector of three-dimensional spatial translation and rotation issued to the door cover or hinge adjustment actuator, expressed as: in: For action vectors; , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command; The reward function of the intelligent agent is used to guide the system to converge quickly to the optimal assembly quality. Its reward value is inversely proportional to the root mean square error of the adjusted gap and surface difference, and a negative penalty is applied to excessive adjustment actions to ensure process smoothness, expressed as: in: It is the root mean square error of all gap deviations and surface differences under the new condition; It is the reward coefficient. It is a small constant to prevent the denominator from being zero; It is the penalty coefficient.

[0010] Furthermore, the automated boot and execution module includes: The door cover grasping and positioning robot is used to grasp the door cover to be assembled and perform coarse positioning according to the initial instructions; The high-precision adjustment robot, as the adjustment execution unit, performs micron-level precise pose adjustment based on the optimal adjustment instructions output by the reinforcement learning intelligent decision-making model. The automatic tightening robot is equipped with a high-precision servo tightening shaft and an automatic bolt feeding system, which is used to automatically complete the bolt tightening operation after the door cover is adjusted into place.

[0011] Furthermore, the end effector of the high-precision adjustment robot is equipped with a sensor kit of the multimodal flexible sensing system.

[0012] Furthermore, the method for high-precision adjustment robot to perform micron-level precise pose adjustment includes the following steps: S1: Transfer the action vector Transform into a homogeneous transformation matrix : in: It is a 3x3 rotation matrix, composed of motion vectors. Rotation command in Generate; and: in: , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command; S2: Solve the target pose of the robot's end effector : in: Given the current pose and a 4x4 homogeneous transformation matrix, it describes the current position and orientation of the robot's end effector relative to the robot base. The current pose is obtained using the DH parameter method for a robot with n joints: The DH parameters for each joint are: ; The joint angle readings are: ; For the Establish transformation matrix for each joint for: We obtain this through chain multiplication: in: The total transformation from the base coordinate system to the end effector coordinate system is equal to the product of all individual joint transformation matrices from beginning to end; Indicates the length of the link; Indicates the linkage torsion angle; Indicates the distance between the links; Indicates joint angle; Number of joints; S3: Solve the robot's inverse kinematics to calculate the realization The target angles of each joint are determined, and the motion trajectory is generated to drive the robot to perform adjustment actions; S4: Repeat steps S1 to S3 until... The quality is less than the preset quality threshold.

[0013] Furthermore, it also includes a cloud-edge collaborative computing platform, comprising cloud computing servers and edge computing servers; The cloud computing server is equipped with the digital twin and real-time simulation prediction module and the reinforcement learning intelligent decision-making model. It is used to store historical production data and high-fidelity twin models, perform offline training and iterative optimization of the physical information neural network and the reinforcement learning intelligent decision-making model, and deploy the optimized model to the edge server. The edge computing server is deployed next to the production line to receive real-time data streams collected by the multimodal flexible sensing system, run the pre-trained physical information neural network model and reinforcement learning intelligent decision-making model, perform real-time state synchronization, deviation prediction and calculation of optimal adjustment strategy, and send the optimal adjustment command to the automation guidance and execution module.

[0014] The beneficial effects of this invention are as follows: This invention relates to a high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins. It utilizes a multimodal flexible sensing system to collect high-precision geometric and mechanical data in real time. Leveraging digital twins and real-time simulation prediction modules, it achieves accurate forward prediction of assembly results. Furthermore, a reinforcement learning intelligent decision-making module outputs the optimal adjustment strategy, and a high-precision robotic actuator performs micron-level adjustments and automatic tightening. This invention overcomes the problems of existing technologies, such as reliance on human experience, insufficient visual perception accuracy, and poor adaptability to rigid positioning. It achieves stable assembly accuracy control at the 0.1mm level, improving consistency and quality. Simultaneously, the system possesses strong flexibility and self-learning capabilities, enabling rapid adaptation to the production needs of multiple vehicle models and batches. It completely replaces traditional manual adjustment processes, significantly reducing manufacturing costs and providing key technological support for intelligent manufacturing and "lights-out factories." Attached Figure Description

[0015] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of an embodiment of the high-precision adaptive assembly and adjustment system for automobile door covers based on physical field digital twins of the present invention. Figure 2 This is a schematic diagram of a physical information neural network. Figure 3A schematic diagram of the principle of a reinforcement learning intelligent decision-making model; Figure 4 A flowchart for adaptive assembly; Figure 5 This is a schematic diagram illustrating the principles of real-time simulation prediction and intelligent decision-making. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0017] like Figure 1 As shown, this embodiment presents a high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins. This system includes a multimodal flexible sensing system, a digital twin and real-time simulation prediction module, a reinforcement learning intelligent decision-making model, and an automated guidance and execution module. Specifically, the multimodal flexible sensing system is used to collect geometric and mechanical data of the door cover and door frame in real time; the digital twin and real-time simulation prediction module is used to construct and synchronize a high-fidelity virtual model, and predict the gaps and surface differences after assembly based on a physical information neural network; the reinforcement learning intelligent decision-making model is used to output optimal adjustment commands based on the deviations in the predicted gaps and surface differences; and the automated guidance and execution module is used to execute the adjustment commands and complete the assembly and tightening of the door cover.

[0018] (1) Multimodal flexible sensing system.

[0019] The multimodal flexible sensing system comprises a 3D optical scanner and a six-dimensional force / torque sensor. Specifically, the 3D optical scanner acquires geometric data, including high-density 3D point cloud data of the door cover, door frame, and surrounding areas, as well as gap and surface difference data of hundreds of key measuring points distributed along the edge of the door cover; the six-dimensional force / torque sensor acquires mechanical data, including contact forces and torques during assembly. The multimodal flexible sensing system is responsible for providing a high-dimensional photometric and dynamic input data stream for the real-time evolution of the digital twin.

[0020] (2) Digital twin and real-time simulation prediction module.

[0021] The digital twin and real-time simulation prediction module serves as a bridge connecting physical entities and virtual models, running on cloud or edge computing platforms. In this embodiment, the digital twin and real-time simulation prediction module includes a high-fidelity twin model, a physical information neural network model, and a real-time state synchronization engine.

[0022] The high-fidelity twin model includes geometric and physical property models of the door cover, door frame, hinges, sealing strips, and bolts. Specifically, the geometric model includes accurate CAD geometric models of the door cover, door frame, hinges, sealing strips, and bolts; the physical property model defines material properties such as elastic modulus, Poisson's ratio, and density.

[0023] A physical information neural network model is used to predict the final gap and surface difference distribution of a door cover after elastic deformation under the action of gravity, assembly stress, and sealing strip compression. For example... Figure 2 As shown, in this embodiment, the physical information neural network model uses the six-degree-of-freedom pose of the hinge and the preload of the bolt as inputs. It can predict in real-time and with high accuracy the final gap and surface difference distribution after the door cover undergoes elastic deformation under the combined effects of gravity, assembly stress, and sealing strip pressure. Specifically, in this embodiment, the loss function of the physical information neural network model includes a data-driven term and a physical constraint term, expressed as: in: It is a hyperparameter that balances the two; It is the mean square error between the model's predicted value and the actual measured value; It is a residual based on the Navier-Cauchy equations of elasticity.

[0024] For displacement field Its residual Defined as: in: and These are Lamé parameters; It is gravity; Let L2 norm be the integral of the residual over the solution domain.

[0025] The real-time state synchronization engine is used to perform millisecond-level data synchronization and state mapping between the real-time data collected by the multimodal flexible sensing system and the high-fidelity twin model, ensuring that the model state in the virtual space is completely consistent with the state of the physical entity.

[0026] By minimizing the total loss function, the trained physical information neural network model can accurately predict pose adjustments. and tightening force The final deformation under the action, and thus the predicted next state. .

[0027] (3) Reinforcement learning intelligent decision-making model.

[0028] like Figure 3 As shown, the reinforcement learning intelligent decision-making model includes a reinforcement learning-based agent, which uses a high-fidelity twin model as its simulation environment for interactive training.

[0029] The state of the agent is defined as a vector consisting of the deviations between the current gap and surface difference measurements of all key measuring points and the ideal target value, expressed as: in: It is a state vector; and These are the clearance deviation item and the surface difference deviation item, respectively. , Let be the number of key measuring points, and: in: and They are respectively in For the first moment The gap value and surface difference value obtained from the measurement at each measuring point; and These are the ideal target values ​​for gap and surface difference, respectively.

[0030] The action of an intelligent agent is defined as: a six-degree-of-freedom fine-tuning command vector of three-dimensional spatial translation and rotation issued to the door cover or hinge adjustment actuator, expressed as: in: For action vectors; , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command.

[0031] The goal of this strategy is to maximize the expected cumulative reward. The policy network is trained by interacting with a digital twin environment at each time step. : in: These are the parameters of the policy network.

[0032] In performing the action Afterwards, the system transitions to a new state. The intelligent agent receives a reward. The agent's reward function guides the system to converge quickly to the optimal assembly quality. Its reward value is inversely proportional to the root mean square error of the adjusted gap and surface difference, and a negative penalty is applied to excessively large adjustment actions to ensure process smoothness. This is expressed as: in: It is the root mean square error of all gap deviations and surface differences under the new condition; It is the reward coefficient. The first term is a small constant to prevent the denominator from being zero; the second term is a penalty term for adjusting the magnitude of the action. It is a penalty coefficient, designed to encourage agents to achieve their goals with smaller adjustments.

[0033] Through millions of training sessions in a digital twin environment, the agent is able to learn a complex nonlinear mapping from arbitrary biased states to optimal adjustment actions, i.e., the optimal adjustment strategy.

[0034] (4) Automated boot and execution module.

[0035] In this embodiment, the automated guidance and execution module includes a door cover gripping and positioning robot, a high-precision adjustment robot, and an automatic tightening robot.

[0036] The door cover gripping and positioning robot is used to accurately grip the door cover to be assembled from the material rack and perform coarse positioning according to the initial instructions.

[0037] The high-precision adjustment robot is used to grip door covers or hinges. As an adjustment execution unit, it performs micron-level precise pose adjustments based on the optimal adjustment instructions output by a reinforcement learning intelligent decision-making model. In this embodiment, the high-precision adjustment robot's end effector is equipped with a sensor suite of a multimodal flexible sensing system.

[0038] In this embodiment, the method steps for high-precision adjustment robot to perform micron-level precise pose adjustment are as follows.

[0039] S1: Action vector output by the decision module The motion vectors need to be converted into high-precision joint movement commands for adjusting the robot. Transform into a homogeneous transformation matrix : in: It is a 3x3 rotation matrix, composed of motion vectors. Rotation command in These values ​​are generated, and they are usually very small, close to zero.

[0040] and: in: , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command.

[0041] S2: Solve the target pose of the robot's end effector Target pose of the robot's end effector From the current pose With homogeneous transformation matrix Multiplying them together gives: in: Given the current pose and a 4x4 homogeneous transformation matrix, it describes the current position and orientation of the robot's end effector relative to the robot base.

[0042] The current pose is obtained using the DH parameter method, for a given... A robot with 1 joint: The DH parameters for each joint are: ; The joint angle readings are: ; For the first Establish transformation matrix for each joint for: We obtain this through chain multiplication: in: The total transformation from the base coordinate system to the end effector coordinate system is equal to the product of all individual joint transformation matrices from beginning to end; Indicates the length of the link; Indicates the linkage torsion angle; Indicates the distance between the links; Indicates joint angle; This refers to the number of joints.

[0043] S3: Solve the robot's inverse kinematics to calculate the realization The target angles of each joint are determined, and the motion trajectory is generated to drive the robot to perform adjustment actions.

[0044] S4: Repeat steps S1 to S3 until... The quality is less than the preset quality threshold.

[0045] The automatic tightening robot is equipped with a high-precision servo tightening shaft and an automatic bolt feeding system. After the door cover is adjusted to the correct position, it automatically tightens all connecting bolts according to the required torque and angle. In this embodiment, the automatic tightening robot is equipped with a clutch-controlled servo tightening gun and an automatic bolt feeding mechanism at its end, used to automatically complete the bolt tightening operation after the door cover is adjusted to the correct position.

[0046] (5) Cloud-edge collaborative computing platform.

[0047] The cloud-edge collaborative computing platform in this embodiment includes a cloud computing server and an edge computing server.

[0048] Specifically, the cloud computing server is equipped with a digital twin and real-time simulation prediction module and a reinforcement learning intelligent decision-making model. It is used to store historical production data and high-fidelity twin models, conduct offline training and iterative optimization of the physical information neural network and reinforcement learning intelligent decision-making model, and deploy the optimized model to the edge server.

[0049] The edge computing server is deployed next to the production line to receive real-time data streams collected by the multimodal flexible sensing system, run the pre-trained physical information neural network model and reinforcement learning intelligent decision-making model, perform real-time state synchronization, deviation prediction and calculation of optimal adjustment strategy, and send the optimal adjustment instructions to the automation guidance and execution module.

[0050] The following section uses the left front door assembly station in the automobile assembly workshop as an example to further elaborate on the specific implementation of the high-precision adaptive assembly and adjustment system for automobile door covers based on physical field digital twins of the present invention.

[0051] Specifically, the left front door assembly station is mainly composed of a multi-joint robot collaborative system and equipped with an edge computing server located next to the production line.

[0052] 1. System hardware deployment.

[0053] (1) The physical entities of the automated guidance and execution module include the door cover gripping and positioning robot, the high-precision adjustment robot and the automatic tightening robot.

[0054] Door and Cover Gripping and Positioning Robot: A heavy-duty six-axis industrial robot with a specially designed flexible door and cover gripper at its end effector, equipped with pneumatic suction cups and positioning pins. This robot is responsible for gripping the left front door from a dedicated door and cover trolley and transporting it to a predetermined position next to the body-in-white.

[0055] High-precision adjustment robot: A high-precision, high-rigidity six-axis industrial robot whose end flange is directly or via a transition device connected to a door / cover gripper, or directly clamps the hinge body. This robot is the primary executor of fine-tuning actions, achieving micron-level repeatability.

[0056] Automatic Tightening Robot: A lightly loaded six-axis robot equipped with a high-precision servo tightening gun with clutch control and an automatic bolt feeding mechanism at its end. It is responsible for automatically tightening four connecting bolts on the upper and lower hinges after the door cover is adjusted into position.

[0057] (2) The physical entity of the multimodal flexible sensing system is an integrated sensor suite mounted on the end effector of the high-precision adjustment robot, specifically including a three-dimensional optical scanner and a six-dimensional force / torque sensor.

[0058] 3D Optical Scanner: A lightweight blue light structured light scanner, mounted on the robot's wrist, can quickly acquire high-density 3D point clouds of millions of pixels in areas such as door flange edges and body pillars.

[0059] Six-dimensional force / torque sensor: Installed between the robot's wrist and end effector, it is used to monitor the changes in triaxial force and triaxial torque in real time during the contact between the door cover and the body hinge, as well as during the compression process with the sealing strip.

[0060] 2. Cloud-edge collaborative computing platform.

[0061] The software algorithms of the digital twin and real-time simulation prediction module and the reinforcement learning intelligent decision-making module are deployed on the cloud-edge collaborative platform, which includes edge computing servers and cloud computing platforms.

[0062] Edge computing servers, deployed alongside the production line, receive real-time data streams from the sensing system, run pre-trained physical information neural network models and reinforcement learning intelligent decision-making models, perform real-time state synchronization, deviation prediction, and calculation of optimal adjustment strategies, and then send control commands to the robot controller. This deployment method ensures low latency in decision-making, meeting production cycle requirements.

[0063] Cloud computing platform: Responsible for handling non-real-time, computationally intensive tasks. This includes: storing massive amounts of historical production data and twin models; and conducting offline training and iterative optimization of physical information neural networks and reinforcement learning agents. Once the cloud model is updated, the optimized model is then deployed to edge servers.

[0064] 3. Adaptive assembly and adjustment implementation process.

[0065] like Figure 4 As shown, after a body-in-white enters the left front door assembly station via the automated conveyor line and stops, the assembly and adjustment process is as follows: (1) Preparation and identification: The label on the body-in-white is read, and the system retrieves the standard CAD model, process parameters and initial assembly program of the vehicle model from the database. At the same time, the door and cover grabbing and positioning robot grabs the corresponding left front door from the material cart.

[0066] (2) Initial scanning and twin instantiation: The high-precision adjustment robot drives the three-dimensional optical scanner at its end to quickly scan the edges of the door openings of the body-in-white and the four edges of the left front door to be installed. The collected real-time point cloud data is sent to the edge server and fused with the standard CAD model to generate a high-fidelity digital twin instance that reflects the unique tolerances of the current pair of doors and the body.

[0067] (3) Coarse positioning and contact perception: The door cover grasping robot moves the left front door to the installation preparation position of the body hinge and performs preliminary attachment. During this process, the six-dimensional force / torque sensor monitors the weight of the left front door itself and the changes in force and torque when in contact with the hinge and positioning pin in real time, and the data is updated synchronously to the digital twin model.

[0068] (4) Real-time simulation prediction and intelligent decision-making: The digital twin and simulation prediction module in the edge server is activated. The physical information neural network model predicts the gap and surface difference distribution of all key measuring points after the door cover is completely fixed within milliseconds, based on the initial pose and contact force of the current left front door, and taking into account the door's own weight and the virtual squeezing force of the sealing strip. The reinforcement learning intelligent decision-making module receives the deviation state vector of this prediction, and its internal agent immediately outputs an optimal six-degree-of-freedom fine-tuning instruction vector. This instruction is the optimal strategy learned by the agent through millions of virtual trial and error training with the twin model in the cloud, aiming to minimize all deviations in one or several steps. Its principle diagram is as follows. Figure 5 As shown.

[0069] (5) Precise adjustment and execution: After receiving this fine-tuning instruction, the high-precision adjustment robot uses high-precision interpolation motion to drive the left front door or hinge to perform a composite fine-tuning of translation and rotation. At the same time, it performs iterative cycles (generally iterating in steps 3-5) according to the actual door pose state until the required accuracy is achieved.

[0070] (6) Tightening and fixing: Once the adjustment is in place, the system sends a signal and the automatic tightening robot starts immediately, tightening all the hinge bolts in sequence according to the preset tightening strategy.

[0071] (7) Quality Verification and Data Feedback: After tightening, the 3D optical scanner performs a final comprehensive gap and surface difference scan on the assembled left front door, generating a digital quality report for this assembly. This measured result is used for final quality confirmation and is compared with the predicted value in step 4. The difference will be uploaded to the cloud as learning data for continuous optimization and iteration of the twin model and decision model, enabling the system to have self-learning and adaptive capabilities.

[0072] Through the above methods, this system uses a brand-new model of "one-time prediction - intelligent decision-making - precise execution" to achieve automation, high precision and intelligence in the assembly of automotive door panels.

[0073] The embodiments described above are merely preferred embodiments for fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins, characterized in that: include: A multimodal flexible sensing system is used to collect geometric and mechanical data of the door cover and door frame in real time; The digital twin and real-time simulation prediction module is used to build and synchronize a high-fidelity virtual model and predict the gaps and surface differences after assembly based on physical information neural networks. A reinforcement learning intelligent decision-making model is used to output the optimal adjustment instructions based on the deviation between the predicted gap and the surface difference; An automated guidance and execution module is used to execute the adjustment instructions and complete the assembly and tightening of the door cover.

2. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins as described in claim 1, characterized in that: The multimodal flexible sensing system includes: A 3D optical scanner is used to acquire geometric data, including high-density 3D point cloud data of the door cover, door frame and surrounding area, as well as gap and surface difference data of hundreds of key measuring points distributed along the edge of the door cover. A six-dimensional force / torque sensor is used to collect mechanical data, including contact forces and torques during assembly.

3. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 1, characterized in that: The digital twin and real-time simulation prediction module includes: A high-fidelity twin model, including geometric and physical property models of the door cover, door frame, hinges, sealing strips, and bolts; A physical information neural network model is used to predict the final gap and surface difference distribution of the door cover after elastic deformation under the action of gravity, assembly stress and sealing strip extrusion force; A real-time state synchronization engine is used to perform millisecond-level data synchronization and state mapping between the real-time data collected by the multimodal flexible sensing system and the high-fidelity twin model.

4. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 3, characterized in that: The loss function of the physical information neural network model includes a data-driven term and a physical law constraint term, expressed as: in: It is a hyperparameter that balances the two; It is the mean square error between the model's predicted value and the actual measured value; It is the residual based on the Navier-Cauchy equations of elasticity; for the displacement field Its residual Defined as: in: and These are Lamé parameters; It is gravity; Let L2 norm be the integral of the residual over the solution domain.

5. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 1, characterized in that: The reinforcement learning intelligent decision-making model includes a reinforcement learning-based agent, which uses the high-fidelity twin model as its simulation environment for interactive training, and: The state of the agent is defined as a vector consisting of the deviations between the current gap measurement values ​​and surface difference measurement values ​​of all key measuring points and the ideal target value, expressed as: in: It is a state vector; and These are the clearance deviation item and the surface difference deviation item, respectively. , Let be the number of key measuring points, and: in: and They are respectively in For the first moment The gap value and surface difference value obtained from the measurement at each measuring point; and These are the ideal target values ​​for gap and surface difference, respectively; The action of the intelligent agent is defined as: a six-degree-of-freedom fine-tuning command vector of three-dimensional spatial translation and rotation issued to the door cover or hinge adjustment actuator, expressed as: in: For action vectors; , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command; The reward function of the intelligent agent is used to guide the system to converge quickly to the optimal assembly quality. Its reward value is inversely proportional to the root mean square error of the adjusted gap and surface difference, and a negative penalty is applied to excessive adjustment actions to ensure process smoothness, expressed as: in: It is the root mean square error of all gap deviations and surface differences under the new condition; It is the reward coefficient. It is a small constant to prevent the denominator from being zero; It is the penalty coefficient.

6. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 1, characterized in that: The automated boot and execution module includes: The door cover grasping and positioning robot is used to grasp the door cover to be assembled and perform coarse positioning according to the initial instructions; The high-precision adjustment robot, as the adjustment execution unit, performs micron-level precise pose adjustment based on the optimal adjustment instructions output by the reinforcement learning intelligent decision-making model. The automatic tightening robot is equipped with a high-precision servo tightening shaft and an automatic bolt feeding system, which is used to automatically complete the bolt tightening operation after the door cover is adjusted into place.

7. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 6, characterized in that: The high-precision adjustment robot is equipped with a sensor kit of the multimodal flexible sensing system at its end.

8. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 6, characterized in that: The method for high-precision adjustment robot to perform micron-level precise pose adjustment includes the following steps: S1: Transfer the action vector Transform into a homogeneous transformation matrix : in: It is a 3x3 rotation matrix, composed of motion vectors. Rotation command in Generate; and: in: , and This is a three-dimensional spatial translation fine-tuning command; , and This is a three-dimensional rotation fine-tuning command; S2: Solve for the target pose of the robot's end effector. : in: Given the current pose and a 4x4 homogeneous transformation matrix, it describes the current position and orientation of the robot's end effector relative to the robot base. The current pose is obtained using the DH parameter method, for a given... A robot with 1 joint: The DH parameters for each joint are: ; The joint angle readings are: ; For the Establish transformation matrix for each joint for: We obtain this through chain multiplication: in: The total transformation from the base coordinate system to the end effector coordinate system is equal to the product of all individual joint transformation matrices from beginning to end; Indicates the length of the link; Indicates the linkage torsion angle; Indicates the distance between the links; Indicates joint angle; Number of joints; S3: Solve the robot's inverse kinematics to calculate the realization The target angles of each joint are determined, and the motion trajectory is generated to drive the robot to perform adjustment actions; S4: Repeat steps S1 to S3 until... The quality is less than the preset quality threshold.

9. The high-precision adaptive assembly and adjustment system for automotive door covers based on physical field digital twins according to claim 1, characterized in that: It also includes a cloud-edge collaborative computing platform, comprising cloud computing servers and edge computing servers; The cloud computing server is equipped with the digital twin and real-time simulation prediction module and the reinforcement learning intelligent decision-making model. It is used to store historical production data and high-fidelity twin models, perform offline training and iterative optimization of the physical information neural network and the reinforcement learning intelligent decision-making model, and deploy the optimized model to the edge server. The edge computing server is deployed next to the production line to receive real-time data streams collected by the multimodal flexible sensing system, run the pre-trained physical information neural network model and reinforcement learning intelligent decision-making model, perform real-time state synchronization, deviation prediction and calculation of optimal adjustment strategy, and send the optimal adjustment command to the automation guidance and execution module.

Images