AGV / AMR digital twin operation and maintenance system fusing dynamic modeling

Abstract

This invention belongs to the field of intelligent logistics and industrial digital twin technology, specifically disclosing an AGV / AMR digital twin operation and maintenance system that integrates dynamic modeling. The system includes a physical entity perception unit, a visual semantic parsing unit, a multibody dynamics solution unit, a twin model synchronization unit, and a parameter evolution control unit. It uses visual semantic parsing to inversely solve for load distribution and structural deformation as dynamic boundary conditions, and combines this with a recursive Newton-Euler algorithm to calculate forces and deformations in real time, driving the synchronous update of the twin's geometric mesh. Simultaneously, it identifies and corrects mass, friction, and stiffness parameters online based on multi-source observation data, and introduces an aging factor to simulate long-term degradation behavior. Through the above technical solutions, this invention achieves high-fidelity, adaptive, and full-lifecycle digital twin construction, improving predictive control accuracy and operational reliability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent logistics and industrial digital twin technology, specifically involving an AGV / AMR digital twin operation and maintenance system that integrates dynamic modeling. Background Technology

[0002] With the rapid transformation of intelligent manufacturing and smart logistics systems, Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs), as core execution units of flexible production, require precise monitoring and predictive maintenance of their operational status to ensure the stability of automated production lines. Digital twin technology, by constructing a virtual mapping of physical entities in cyberspace, enables real-time synchronization of mobile robot motion trajectories, operating environments, and system states, providing data support and a simulation platform for cluster scheduling and full lifecycle operation and maintenance in complex industrial scenarios.

[0003] High-fidelity dynamic modeling and parameter evolution mechanisms are the core foundation for realizing the leap from geometric representation to physical deduction in digital twin systems. In actual operation, mobile robots need to combine multi-source sensor fusion perception technology to accurately characterize their own kinematic characteristics, load distribution, and mechanical structure losses, ensuring that the virtual model can realistically reflect the dynamic response process of the physical entity, thereby providing an accurate physical benchmark for control decision optimization and equipment fault early warning.

[0004] Traditional digital twin operation and maintenance systems often employ pre-set static physical parameter modeling, making it difficult to address performance drift issues in mobile robots during long-term operation. Due to frequent load fluctuations, tire wear, and weight shifts caused by battery consumption during task execution, fixed-parameter models frequently suffer from simulation mismatch, leading to predictive control failures or increased false alarm rates. Existing perception solutions typically treat computer vision and dynamic analysis in isolation, lacking a deep coupling mechanism to convert visual semantic information into dynamic boundary conditions in real time, and failing to effectively solve for nonlinear changes in physical properties through visual observation. The system lacks the ability to update model parameters dynamically, making it difficult to capture the real-time evolution of structural deformation and inertia matrices, resulting in logical discrepancies between the twin model and the real physical entity over time. Summary of the Invention

[0005] The purpose of this invention is to provide an AGV / AMR digital twin operation and maintenance system that integrates dynamic modeling, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides an AGV / AMR digital twin operation and maintenance system integrating dynamic modeling, comprising:

[0007] The physical entity sensing unit is configured to collect multi-source sensor data in real time during the operation of automated guided vehicles or autonomous mobile robots in order to obtain raw observation data on their motion state and internal load changes.

[0008] The visual semantic parsing unit is configured to receive depth images and color image sequences, perform point cloud reconstruction and visual synchronous localization and mapping, extract load distribution features and vehicle body structural deformation features, and convert the load distribution features and vehicle body structural deformation features into physical attribute parameters.

[0009] The multibody dynamics solving unit is configured to build a high-fidelity dynamics model, receive the physical property parameters as boundary conditions, and calculate the force distribution, rotational inertia matrix and tire contact physical quantities of each component of the vehicle body in real time.

[0010] The twin model synchronization unit is configured to maintain a digital twin corresponding to the physical entity in the virtual space, and drive the digital twin to perform attitude updates and geometric deformations based on the dynamic response results output by the multibody dynamics solving unit.

[0011] The parameter evolution control unit is configured to perform online identification and adaptive correction of the model parameters in the multibody dynamics solution unit, and realize the dynamic update of the model parameters based on the multi-dimensional observation data provided by the physical entity perception unit and the visual semantic parsing unit.

[0012] Preferably, the physical entity sensing unit specifically includes:

[0013] The sensor acquisition subunit integrates a high-precision six-axis inertial measurement unit, a drive motor controller interface, and a battery management system interface. It is configured to acquire the angular velocity vector and linear acceleration vector of the vehicle body in three-dimensional space, read the phase current feedback signal of each wheel drive motor, encoder pulse signal and output torque estimation value, and obtain the real-time state of charge, cell voltage distribution and operating temperature of the battery pack.

[0014] The signal preprocessing subunit is connected to the sensor acquisition subunit and is configured to perform recursive least squares filtering or low-pass filtering on the acquired raw signal to remove noise signals introduced by mechanical vibration or electromagnetic interference.

[0015] The data synchronization encapsulation subunit is configured to assign a unified timestamp to the preprocessed inertial data, motion data, and energy data, and encapsulate multi-source heterogeneous data into standardized telemetry data frames according to a preset communication protocol.

[0016] Preferably, the visual semantic parsing unit specifically includes:

[0017] The point cloud reconstruction module is configured to receive point cloud streams from a color depth camera and uses an iterative nearest-point algorithm to incrementally construct visual odometry, generating local point cloud maps of the operating environment and the automated guided vehicle itself.

[0018] The semantic segmentation module, which embeds a deep convolutional neural network, is configured to classify objects in the point cloud map and identify the cargo outline, pallet posture, and key structural components of the vehicle body on the cargo platform.

[0019] The physical property mapping module is configured to inversely calculate the equivalent mass and centroid coordinates of the cargo based on the identified cargo geometry and a preset density dictionary.

[0020] When the semantic segmentation module detects that the cargo is stacked tilted or unbalanced, the physical attribute mapping module calculates the deviation moment of the cargo from the centerline of the loading platform using a voxelization method and outputs a rotational inertia correction term.

[0021] The visual semantic parsing unit is further configured to identify the stacking tilt angle and center of mass offset of goods on the cargo platform, and use the stacking tilt angle and center of mass offset as key inputs for the correction of the rotational inertia matrix, so that the dynamic model can dynamically reflect the changes in inertial characteristics under different load conditions.

[0022] Preferably, the multibody dynamics solution unit specifically includes:

[0023] The topology description module is configured to construct a topology description matrix that includes rigid bodies, elastic bodies, and constraint pairs.

[0024] The recursive solution module employs a recursive Newton-Louras cooperative algorithm, configured to calculate the kinematic state of each component through forward recursion, and combined with the mass distribution and moment of inertia matrix output by the physical property mapping module, to solve the generalized forces and constraint reactions at the contact points of each joint and wheel train through backward recursion.

[0025] The wheel-ground contact modeling module integrates a nonlinear tire force model, configured to describe the nonlinear mapping relationship between tire longitudinal slip ratio, sideslip angle and ground reaction force, and calculates tire sideslip stiffness drift and tangential adhesion coefficient by combining real-time wheel speed and vehicle trajectory.

[0026] The structural stress calculation module is configured to calculate the small deformation field of the vehicle body structure under non-uniform loads and output flexible deformation data including the compression of the suspension system and the torsional angle of the chassis frame.

[0027] When calculating the tire deformation under stress, the multibody dynamics solver combines the ground reaction force model with the tire material elastic parameters to output the predicted values ​​of tire compression and slip angle.

[0028] Preferably, the twin model synchronization unit is configured with a three-layer geometric representation architecture, specifically including:

[0029] Rigid body skeleton layer, configured to map the kinematic state of the automated guided vehicle's global pose and major mechanical connections;

[0030] The outer shell layer is composed of a high-density triangular mesh. The spatial position of the mesh vertices is driven in real time by the structural deformation field output by the multibody dynamics solving unit. Through radial basis function interpolation or finite element interpolation algorithm, the visualization of tire compression deformation, shock absorber compression and frame micro-twist is realized.

[0031] The load semantic layer is configured to dynamically render the cargo status in the virtual space. Based on the load attributes extracted by the visual semantic parsing unit, it updates the shape, texture, and relative position of the virtual cargo with respect to the vehicle body in real time.

[0032] The twin model synchronization unit supports multi-timescale synchronization strategies. For motion posture updates, a fast channel at the millisecond level is used to achieve following the physical entity, while a slow channel at the second level or task cycle level is used for model parameter updates.

[0033] Preferably, the parameter evolution control unit specifically includes:

[0034] The residual calculation module is configured to store the nominal dynamic model and calculate the mean or variance of the residual sequence between the actual motor current fed back by the physical entity sensing unit and the expected current predicted by the multibody dynamics solving unit under the same working conditions by comparing them in real time.

[0035] An adaptive correction logic module is configured to trigger an online correction process when the residual sequence continuously exceeds a preset tolerance threshold, and to use a gain extended Kalman filter algorithm to correct the equivalent friction coefficient, transmission system damping, and overall mass in the model online.

[0036] The parameter evolution control unit uses a system identification algorithm to analyze the deviation between motor current and wheel speed collected by the physical entity sensing unit. When the residual continuously exceeds the preset threshold, the online calibration process of the model parameters is triggered, and the equivalent mass and damping coefficient in the dynamic model are automatically adjusted.

[0037] Preferably, the parameter evolution control unit also integrates an aging factor accumulation submodule, configured to construct an exponential evolution model of tire wear, motor efficiency decay and suspension stiffness degradation based on the cumulative mileage, frequent start-up times and battery cycle count of the automated guided vehicle, so that the digital twin can simulate the performance degradation process of the physical entity throughout its entire life cycle.

[0038] The aging factor accumulation submodule is specifically configured to execute the following logic:

[0039] Based on operating mileage, number of operation cycles, and battery charge / discharge cycles, a trend model is constructed to investigate the relationship between tire wear and suspension system stiffness reduction.

[0040] Based on the stress cycle sequence of key welding points calculated by the multibody dynamics solution unit, the stress amplitude distribution is statistically analyzed using the rainflow counting method. Combined with the fatigue life curve of the material, the remaining life estimate of each structural component is marked in real time in the digital twin.

[0041] Preferably, a physical feature buffer is provided between the visual semantic parsing unit and the multibody dynamics solving unit, and a bidirectional data channel is established:

[0042] The visual semantic parsing unit writes the inversely solved physical attributes, including the cargo's instantaneous center of mass height, eccentricity, and detected structural elastic deformation, into the physical feature buffer at a first frequency.

[0043] The multibody dynamics solving unit reads data from the physical feature buffer at a second frequency greater than the first frequency, and performs smooth interpolation of the physical features through an internally integrated second-order kinematics interpolator to generate a continuous boundary condition flow.

[0044] The tire normal force calculated by the multibody dynamics solving unit is fed back to the visual semantic parsing unit. The visual semantic parsing unit uses the fed-back force information to constrain and optimize the vertices of the tire-ground contact area in the point cloud, and uses the dynamically predicted structural deformation trend as a priori constraint for visual point cloud registration.

[0045] Preferably, the system adopts an edge-cloud collaborative architecture, specifically including:

[0046] The edge-aware execution layer, deployed on the automated guided vehicle, includes a lightweight front-end of the physical entity perception unit and the visual semantic parsing unit, as well as a real-time kinematics core of the multibody dynamics solving unit. It is configured to perform point cloud feature dimensionality reduction and only uploads the physical feature vectors to the cloud.

[0047] The cloud-based simulation evolution layer, deployed on a server cluster, includes a high-order simulation module of the multibody dynamics solution unit, a parameter evolution control unit, and a twin model synchronization unit. It is configured to run a fully parameterized dynamics model and calculate the stress distribution and long-term dynamic evolution of the vehicle body's internal structure.

[0048] The cloud-based simulation evolution layer also includes a global digital twin resource scheduler, configured to dynamically adjust the computing resource quotas of each twin in the cloud based on the operational intensity and task priority of each automated guided vehicle.

[0049] Preferably, the system also includes:

[0050] The predictive simulation module is configured to enable the multibody dynamics solution unit to be decoupled from the physical entity perception unit in real time. Based on the current parameter state and the preset planning path, it performs advanced time step iterations to simulate the dynamic response in the future time period in order to predict whether there is a risk of instability due to overload or excessive center of gravity.

[0051] A tactile feedback device, connected to the twin model synchronization unit, is configured to apply corresponding resistance feedback to the operator when the operator manually drags and drops goods in the virtual space to simulate adjusting the layout, based on the virtual gravity and friction calculated by the multibody dynamics model.

[0052] The environment mapping feedback link is configured to automatically associate the ground material features identified by the visual semantic parsing unit with a preset friction coefficient reference value, and the parameter evolution control unit performs system identification in the vicinity of the friction coefficient reference value.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention provides an AGV / AMR digital twin operation and maintenance system that integrates dynamic modeling. By constructing a deep coupling mechanism between the visual semantic parsing unit and the multibody dynamics solution unit, it solves the functional separation in the traditional digital twin system where vision is only used for positioning and dynamics is only used for simulation, and realizes the transformation from looking at position to looking at physical attributes.

[0055] 2. The system can use visual observation to solve the load distribution and structural deformation in real time, and use it as boundary conditions to drive the high-fidelity dynamic model. At the same time, the dynamic prediction of stress and deformation is fed back to the twin geometry expression, forming a closed-loop two-way collaborative update mechanism.

[0056] 3. The parameter evolution control unit introduces an online system identification and aging factor accumulation strategy, which enables the digital twin model to update dynamically. This solves the problem of time-varying physical parameters caused by load changes, tire wear and battery power consumption, and reduces predictive control failure and false alarm rate caused by model mismatch.

[0057] 4. The multi-level geometric representation and multi-timescale synchronization strategy take into account both real-time performance and long-term evolution, enabling the twin to accurately respond to transient conditions and realistically reflect the performance degradation of the equipment throughout its entire life cycle. This provides a highly reliable physical inference basis for predictive maintenance, cluster scheduling optimization, and human-machine collaborative decision-making in intelligent manufacturing and smart logistics scenarios. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0059] Figure 2 This is a schematic diagram of the core principle framework of the deep coupling between visual semantic parsing and multibody dynamics solution in this invention;

[0060] Figure 3 This is a flowchart of the online parameter evolution control logic based on residual analysis and system identification in this invention.

[0061] Figure 4 This is a schematic diagram of the data flow of the multi-level geometric representation and dynamic response of the digital twin in this invention;

[0062] Figure 5 This is a schematic diagram illustrating the principle of long-term dynamic updating of model parameters in this invention, which incorporates the aging factor accumulation mechanism. Detailed Implementation

[0063] Example 1: Reference Figures 1 to 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0064] A digital twin operation and maintenance system for AGV / AMR that integrates dynamic modeling includes a physical entity perception unit, a visual semantic parsing unit, a multibody dynamics solution unit, a twin model synchronization unit, and a parameter evolution control unit.

[0065] The physical entity sensing unit is installed at the hardware layer of the automated guided vehicle or autonomous mobile robot, and is used to collect multi-source sensor raw data in real time during the execution of logistics operations by the automated guided vehicle or autonomous mobile robot, and construct a bottom-level data stream that reflects the real state of the physical world.

[0066] The visual semantic parsing unit is electrically connected to the physical entity perception unit and the vehicle-mounted visual sensing device. It is used to receive three-dimensional depth image data and extract the external load distribution characteristics and structural strain characteristics of the automated guided vehicle or autonomous mobile robot through inverse geometric calculation.

[0067] The multibody dynamics solving unit receives the physical property parameters output by the visual semantic parsing unit through a data interface, which is used to construct a high-fidelity dynamic model based on physical mechanisms and to calculate the dynamic response results of the vehicle body under dynamic conditions in real time.

[0068] The twin model synchronization unit is deployed on a monitoring terminal or cloud server and is used to drive the geometric deformation, posture evolution and behavior of the digital twin in the virtual dimension according to the output results of the multibody dynamics solving unit.

[0069] The parameter evolution control unit establishes feedback loops with the physical entity sensing unit and the multibody dynamics solving unit, respectively, to realize online identification, drift correction and long-term evolution management of dynamic model parameters based on the equipment life cycle.

[0070] The physical entity sensing unit includes a sensor acquisition subunit, a signal preprocessing subunit, and a data synchronization and encapsulation subunit.

[0071] The sensor acquisition subunit integrates a high-precision six-axis inertial measurement unit (IMU), which is configured to acquire the angular velocity vector and linear acceleration vector of the automated guided vehicle (AGV) in three-dimensional space to capture transient disturbances of the vehicle body during acceleration, turning, or traversing uneven surfaces. The sensor acquisition subunit is connected to the controller area network (CLAN) bus of the drive motors to read the phase current feedback signals, encoder pulse signals, and estimated output torque values ​​of each wheel drive motor in real time. It also includes a battery management system (BMS) interface for acquiring the real-time state of charge (SOC), individual cell voltage distribution, and operating temperature of the battery pack.

[0072] The signal preprocessing subunit is configured to perform recursive least squares filtering or Butterworth low-pass filtering on the acquired high-frequency raw signal to remove high-frequency noise introduced by mechanical vibration or electromagnetic interference, ensuring that the data input to subsequent units has a high signal-to-noise ratio.

[0073] The data synchronization and encapsulation subunit is used to assign a unified high-precision timestamp to multi-source heterogeneous data and encapsulate inertial data, motion data and energy data into standardized telemetry data frames according to a preset communication protocol.

[0074] The visual semantic parsing unit includes a point cloud reconstruction module, a semantic segmentation module, and a physical attribute mapping module.

[0075] The point cloud reconstruction module is configured to receive point cloud streams from an onboard color depth camera, and use an iterative nearest point algorithm or a normal distribution transformation algorithm to incrementally construct visual odometry, generating local point cloud maps of the operating environment and the automated guided vehicle itself.

[0076] The semantic segmentation module is embedded with a deep convolutional neural network and is configured to classify objects in the point cloud map, with a focus on identifying cargo outlines, pallet postures, and key structural components of the vehicle body on the cargo platform.

[0077] The physical attribute mapping module is configured to calculate the equivalent mass and centroid coordinates of the cargo in reverse based on the identified cargo geometry and a preset density dictionary. When the semantic segmentation module detects that the cargo has a stacking tilt or off-center loading phenomenon, the physical attribute mapping module calculates the deviation moment of the cargo from the centerline of the loading platform using a voxelization method and outputs a real-time rotational inertia correction term.

[0078] The multibody dynamics solution unit is based on the theory of multibody system dynamics and constructs a topological description matrix that includes rigid bodies, elastic bodies and constraint pairs.

[0079] The multibody dynamics solution unit uses the recursive Newton-Euler algorithm as the core solver. This recursive Newton-Euler algorithm is configured to calculate the kinematic state of each link or vehicle body component through forward recursion, and combine the mass distribution and moment of inertia matrix input by the visual semantic parsing unit to solve the generalized forces and constraint reactions at the contact points of each joint and wheel system through backward recursion.

[0080] In the modeling of wheel-ground contact characteristics, the multibody dynamics solver unit integrates a nonlinear tire force model, which describes the nonlinear mapping relationship between tire longitudinal slip ratio, sideslip angle and ground reaction force. Combined with the real-time wheel speed and vehicle trajectory obtained by the physical entity sensing unit, the tire's sideslip stiffness drift and tangential adhesion coefficient are calculated.

[0081] The multibody dynamics solving unit is also configured to calculate the small deformation field of the vehicle body structure under non-uniform loads, and output flexible deformation data including the compression of the suspension system and the torsional angle of the chassis frame, as a physical benchmark for driving the visual performance of the twin.

[0082] The twin model synchronization unit adopts a multi-level geometric representation architecture, specifically including a rigid skeleton layer, an outer shell layer, and a load semantic layer. The rigid skeleton layer is used to map the kinematic state of the global pose and main mechanical connections of the automated guided vehicle; the outer shell layer is composed of a high-density triangular mesh, and the spatial position of its vertices is driven in real time by the structural deformation field output by the multibody dynamics solution unit. Through radial basis function interpolation or finite element interpolation algorithms, visualization of phenomena such as tire compression deformation, shock absorber compression, and minute frame torsion is achieved.

[0083] The load semantic layer is used to dynamically render the cargo status in virtual space. Based on the attributes extracted by the visual semantic parsing unit, it updates the shape, texture, and relative position of the virtual cargo to the vehicle body in real time. The twin model synchronization unit also supports multi-timescale synchronization strategies. For high-frequency motion posture updates, a millisecond-level fast channel is used to achieve zero-latency following with the physical entity; for low-frequency model parameter updates, a second-level or task-cycle-level slow channel is used to ensure that the virtual system can accurately reflect the evolution trend of physical characteristics while meeting real-time monitoring requirements.

[0084] The parameter evolution control unit is configured to execute a closed-loop system identification process. The parameter evolution control unit stores a nominal dynamic model and calculates the residual sequence between the actual motor current fed back by the physical entity sensing unit and the expected current predicted by the multibody dynamics solving unit under the same operating conditions by comparing them in real time.

[0085] When the mean or variance of the residual sequence continuously exceeds the preset tolerance threshold, the parameter evolution control unit triggers adaptive correction logic and uses the gain extended Kalman filter algorithm to correct the equivalent friction coefficient, transmission system damping, and overall mass in the model online.

[0086] The parameter evolution control unit also integrates an aging factor accumulation submodule. This aging factor accumulation submodule constructs an exponential evolution model of tire wear, motor efficiency decay, and suspension stiffness degradation based on the cumulative mileage, frequent start-up times, and battery cycle count of the automated guided vehicle. This enables the digital twin to simulate the performance degradation process of the physical entity throughout its entire life cycle, providing data-driven decision-making basis for predictive maintenance.

[0087] In actual operation scenarios, a bidirectional collaborative mechanism is established between the visual semantic parsing unit and the multibody dynamics solving unit. The physical features extracted visually serve as the initial boundary conditions for dynamics solving; the vehicle attitude and structural deformation trends predicted by the multibody dynamics solving unit at the next moment are fed back to the visual synchronous localization and mapping algorithm of the visual semantic parsing unit as prior constraints for point cloud feature matching.

[0088] Through this bidirectional coupling, the system can compensate for the lack of visual perception by using dynamics simulation results or correct the parameter drift of the dynamics model by using real-time visual observation under extreme conditions such as insufficient light or sensor obstruction.

[0089] The distributed wheel hub torque sensors in the physical entity sensing unit are configured to independently monitor the force state of each drive wheel. When the automated guided vehicle (AGV) is traveling on uneven surfaces or slopes, the parameter evolution control unit analyzes the asymmetry of torque distribution among the wheels and, in conjunction with the roll angle sensed by the inertial measurement unit, identifies the specific offset coordinates of the vehicle's center of gravity in the vertical and horizontal directions in real time. This high-precision center of gravity identification result is then input into the multibody dynamics solving unit to reconfigure the rotational inertia tensor, improving the accuracy of the digital twin's assessment of rollover and sideslip risks in complex terrain.

[0090] The parameter evolution control unit is also equipped with an environment mapping feedback link. This link can automatically associate the ground material characteristics (e.g., epoxy flooring, cement flooring, or metal grid) identified by the visual semantic parsing unit with a preset friction coefficient benchmark value, and the system's identification algorithm performs a refined search around the friction coefficient benchmark value. This environment-prior-based search strategy shortens the time required for the dynamic model to converge to the actual physical state, ensuring a smooth transition when the digital twin system switches between different workshop areas.

[0091] The twin model synchronization unit is also configured to have a predictive simulation mode. The multibody dynamics solution unit, independent of the real-time drive of the physical entity perception unit, performs advanced time-step iterations based on the current parameter state and a preset planning path. By simulating the dynamic response over a future period, the system can predict whether there are potential instability risks due to overload or excessively high center of gravity, and present early warning information in the digital twin interface in the form of a heatmap or color-coded alarms, guiding maintenance personnel or the cluster scheduling system to take intervention measures in advance.

[0092] Example 2: Based on Example 1, this example provides an AGV / AMR digital twin operation and maintenance system based on an edge-cloud collaborative architecture that integrates dynamic modeling. It aims to solve the balance problem between computing resource allocation and massive dynamic data processing in large-scale cluster deployment scenarios.

[0093] An AGV / AMR digital twin operation and maintenance system integrating dynamic modeling includes an edge-aware execution layer, a cloud-based simulation evolution layer, and a real-time communication bus.

[0094] The edge-aware execution layer is deployed on each automated guided vehicle (AGV) and includes a lightweight front-end for the physical entity perception unit and visual semantic parsing unit described in Example 1, as well as a real-time kinematics core for the multibody dynamics solving unit. The lightweight front-end is configured to utilize a high-performance embedded processor to perform feature reduction and key semantic extraction of the point cloud, uploading only physical feature vectors related to load, pose, and deformation to the cloud to reduce communication bandwidth pressure. The real-time kinematics core ensures the safety of the robot's underlying real-time control, guaranteeing its response to environmental changes within millimeter-level cycles.

[0095] The cloud-based simulation evolution layer is deployed on a high-performance server cluster and includes a high-order simulation module of the multibody dynamics solver unit described in Example 1, a parameter evolution control unit, and a twin model synchronization unit. The high-order simulation module is configured to utilize the powerful computing resources of the cloud to run a fully parameterized dynamic model with multiple degrees of freedom, calculating the stress distribution, minor fatigue damage, and long-term dynamic evolution of the vehicle's internal structure. The parameter evolution control unit aggregates operational data of the same model of automated guided vehicles under different operating scenarios in the cloud, analyzes the common aging characteristics of this batch of robots using swarm intelligence algorithms, and achieves cross-device parameter optimization and aging model calibration.

[0096] The real-time communication bus adopts industrial wireless communication standards with deterministic latency, such as the ultra-reliable low-latency communication protocol in fifth-generation mobile communication technology, to ensure that the data interaction between the edge-aware execution layer and the cloud simulation evolution layer meets the synchronization accuracy requirements. The real-time communication bus is configured with data integrity verification and disconnection protection mechanisms. When the network link is unstable, the edge-aware execution layer automatically switches to a local degraded operation mode, using a simplified dynamic model to maintain basic digital twin synchronization.

[0097] The cloud-based simulation evolution layer also includes a global digital twin resource scheduler. This scheduler is configured to dynamically adjust the computing resource allocation for each twin in the cloud based on the workload and task priority of each robot. For robots performing precision assembly tasks or operating in high-risk areas, the scheduler allocates higher-frequency dynamics solution accuracy; while for robots idle or traveling on flat, straight paths, the frequency of their parameter evolution updates is reduced accordingly, thus achieving optimal utilization of the overall system's computing power.

[0098] In Example 2, the multibody dynamics solving unit enhances the global identification of nonlinear damping and structural friction terms. Since the cloud has the capability to process long-sequence historical data, the parameter evolution control unit employs a deep reinforcement learning-based identification strategy. By observing performance deviations over hundreds of past work cycles, it learns the evolution characteristics of complex physical properties. For example, regarding bearing wear in an electric drive system, the system can accurately model the evolution trend of the bearing friction coefficient by analyzing the relationship between high-frequency current harmonic components and the motor temperature rise curve, and periodically distribute updated parameter packages to the edge execution layer, achieving a closed-loop follow-up of model parameters.

[0099] The digital twin model synchronization unit supports a collaborative operation and maintenance interface for multiple users in the cloud. Personnel with different functions can access the digital twin system through different views. For example, field maintenance personnel can access it through augmented reality glasses to view in real time the force and heat maps of key components calculated by the multibody dynamics solver unit, which are overlaid on the physical entity; system engineers can view the full lifecycle performance report generated by the parameter evolution control unit through a desktop terminal. This cloud-based shared model ensures consistency in multi-party decision-making and reduces operation and maintenance communication costs.

[0100] Example 3: This example details a special implementation method for heavy-duty AGVs in extreme environments, which enhances the online monitoring of tire model and ground adhesion conditions.

[0101] In this embodiment, the physical entity sensing unit adds a distributed array of tactile sensors, mounted on the tire liner or axle bracket, to directly sense the distribution of reaction forces from the ground. The multibody dynamics solving unit incorporates a tire physics model that considers thermodynamic coupling during calculations. This tire physics model is configured not only to calculate mechanical deformation but also to dynamically adjust the elastic modulus and damping characteristics of the tire rubber by incorporating tire surface temperature data collected by infrared sensors, because under heavy loads, tire temperature rise can cause nonlinear fluctuations in material properties.

[0102] In Embodiment 3, the visual semantic parsing unit is configured to detect ground humidity and oil stains. By extracting the reflectivity of the ground texture using an onboard multispectral camera, the system can roughly estimate the potential adhesion coefficient of the ground. This environmental semantic information is used as a priori probability and input into the Bayesian recognition framework of the parameter evolution control unit.

[0103] When the automated guided vehicle slips or its drive wheels spin freely, the parameter evolution control unit can quickly distinguish whether the abnormality is caused by a sudden load shift or a change in ground adhesion. It can also adjust the friction contact parameters in the twin model in real time to realistically reproduce the robot's drifting or loss of control behavior in virtual space, providing a highly reliable backtracking for subsequent accident analysis.

[0104] To address the torsional stiffness of heavy-duty structures, the multibody dynamics solving unit employs a flexible multibody dynamics modeling method, discretizing the frame into a set of finite element nodes. The twin model synchronization unit receives the displacement components of these nodes, driving the virtual model to generate physically realistic bending and torsional animations. This deep physical coupling ensures that the digital twin is not merely an external imitation, but a mirror image of its intrinsic mechanical properties.

[0105] The parameter evolution control unit also integrates a specific fatigue damage accumulation model. Based on the stress cycle sequence of key weld points calculated by the multibody dynamics solver, the stress amplitude distribution is statistically analyzed using the rainflow counting method. Combined with the material's fatigue life curve, the estimated remaining life of each structural component is marked in real time in the digital twin. When the damage factor of a certain node reaches a preset alarm value, the maintenance system automatically generates a work order to guide maintenance personnel to perform non-destructive testing on the corresponding part of the physical entity.

[0106] Example 4: This example describes in detail the specific textual implementation of the present invention in the online parameter identification logic, strictly adhering to the principle of no formulas.

[0107] Within the parameter evolution control unit, a complete parameter tracking process based on state estimation is executed. This control unit first establishes a state prediction mechanism, configured to predict the desired motion state at the current moment based on the optimal model parameters from the previous time step and the current control input commands, using a multibody dynamics solver. The desired motion state includes the expected center-of-mass acceleration, wheel circumferential velocity, and expected vehicle roll angle.

[0108] The parameter evolution control unit receives the actual observation state from the physical entity sensing unit. The parameter evolution control unit is internally configured with a weighted residual calculation module, which is used to calculate the difference between the desired motion state and the actual observation state. To improve the stability of the identification, the difference is not directly applied to parameter correction, but rather processed through a sliding window time integration to eliminate pseudo-random bias introduced by instantaneous sensor jitter.

[0109] When the cumulative value of the residual within the calculated sliding window exceeds a preset first tolerance threshold but is less than a preset second tolerance threshold, the system determines that the parameter shift is due to load changes or slight environmental fluctuations. The parameter evolution control unit initiates local adaptive correction, using the gradient descent principle to fine-tune the mass distribution matrix in the dynamic model along the direction of reducing residuals in the parameter space.

[0110] If the cumulative value of the residual exceeds the second tolerance threshold, the system determines that a sudden structural change or sensor failure has occurred. The parameter evolution control unit triggers a global reconstruction process, coordinating the visual semantic parsing unit to perform a comprehensive point cloud scan and physical attribute reassessment of the vehicle body and cargo until the new dynamic model prediction results are re-aligned with the physical observation data.

[0111] This hierarchical tolerance-based identification logic ensures that the digital twin system can both sensitively capture subtle evolutions in physical characteristics and possess robustness against interference, preventing frequent oscillations in the model due to noise.

[0112] Example 5: This example describes an optimized communication mechanism of the present invention in the bidirectional coupling of vision and dynamics.

[0113] A physical feature buffer is provided between the visual semantic parsing unit and the multibody dynamics solving unit. The visual semantic parsing unit is configured to write the physical properties obtained from the reverse solution to the physical feature buffer at a first frequency, for example, 15 times per second, including the instantaneous center of mass height of the cargo, the eccentricity moment, and the detected structural elastic deformation.

[0114] The multibody dynamics solver then reads data from the physical feature buffer at a second frequency, for example, 200 times per second. To bridge the time frequency gap between the two units, a second-order kinematic interpolator is integrated within the multibody dynamics solver. This second-order kinematic interpolator is configured to smoothly interpolate the less frequently updated physical features output by the vision unit, generating a continuous flow of boundary conditions.

[0115] The multibody dynamics solving unit is configured to feed back its calculated tire normal forces to the visual semantic parsing unit. Because visual point clouds exhibit discretization noise when processing subtle displacements, the visual semantic parsing unit utilizes this feedback force information to constrain and optimize the vertices of the tire-ground contact area in the point cloud. For example, when dynamic calculations show that the left front wheel bears more than 40% of the vehicle's total mass, the visual algorithm will prioritize searching for compressed point cloud features near that wheel position, thereby improving the sensitivity of the vision system to vehicle deformation capture.

[0116] This frequency decoupling and feature complementarity mechanism improves the overall operating efficiency of the system, enabling complex visual semantic understanding and high-frequency dynamic simulation to work seamlessly together on heterogeneous hardware platforms.

[0117] Example 6: This example describes the specific details of how the parameter evolution control unit handles motor degradation.

[0118] The parameter evolution control unit includes a motor electromechanical coupling identification module. This module is configured to convert the three-phase current feedback signal provided by the physical entity sensing unit into current components in a rotating coordinate system using coordinate transformation technology. By analyzing the voltage equation residuals of the motor at a specific speed, the module can identify the motor's back electromotive force coefficient online.

[0119] Since the back electromotive force coefficient of the motor is proportional to the magnetic flux intensity of the permanent magnet, and the magnetic flux intensity gradually decreases with increasing operating time due to demagnetization, this change is defined by the parameter evolution control unit as a long-term physical evolution process. The system compares the identified real-time back electromotive force coefficient with the nominal value at the time of manufacture to calculate the dimensionless motor performance health index.

[0120] The digital twin synchronization unit adjusts the energy conversion efficiency parameters of the drive system in the virtual model based on the motor's performance health index. This means that when the physical robot experiences a decrease in torque under the same current input due to motor aging, the digital twin model can accurately simulate this power reduction, rather than simply attributing it to increased external resistance. This deep characterization of the underlying failure mechanism enables the maintenance system to accurately distinguish between environmental factors and equipment aging factors, achieving a higher level of fault diagnosis.

[0121] Example 7: This example supplements the system's differentiated modeling method for different individuals in a cluster environment.

[0122] In a system with multiple automated guided vehicles (AGVs), the cloud server maintains a global parameter knowledge base. The parameter evolution control unit is configured to periodically upload the adaptive correction parameter set of each AGV to this global parameter knowledge base.

[0123] The knowledge base employs a clustering analysis algorithm to categorize the dynamic parameters of all robots into different feature clusters. For example, robots operating in low-temperature cold storage environments are classified into the first feature cluster, while robots operating in high-temperature forging workshops are classified into the second feature cluster. Since ambient temperature affects lubricant viscosity, metal expansion coefficients, and battery internal resistance, this environmental feature-based parameter clustering technique allows newly added robots to directly download the most suitable initial dynamic parameter package from the knowledge base based on their deployment environment, significantly shortening the digital twin initialization time for new equipment.

[0124] When the parameter evolution path of a robot deviates from the statistical regularity of its feature cluster, the parameter evolution control unit determines that the device has potential structural damage or abnormal wear. This anomaly detection method based on group statistics has higher sensitivity and a lower false alarm rate compared to isolated individual monitoring.

[0125] Example 8: This example details the specific application scheme of the multi-level geometric representation in the human-computer interaction scenario of the present invention.

[0126] In the digital twin synchronization unit, the load semantic layer is configured to overlay and display mechanical sensing data. When an operator observes the digital twin through an interactive terminal, the system dynamically renders a pressure cloud map at the virtual cargo-vehicle connection point based on the stress field output by the multibody dynamics solver.

[0127] For example, if improper placement of goods causes the stress at a support point to exceed the design load, the corresponding grid area on the digital twin will smoothly transition from green to flashing red, accompanied by text prompts regarding the off-center load value. Simultaneously, the twin model synchronization unit drives the virtual camera's viewpoint to automatically focus on the high-risk area.

[0128] To enhance the realism of the interaction, the twin model synchronization unit is also configured to drive the haptic feedback device. When maintenance personnel manually drag goods in the virtual space to simulate adjusting the layout, the system applies corresponding resistance to the personnel's hands through the haptic feedback device based on the virtual gravity and friction calculated by the multibody dynamics model. This force feedback allows maintenance personnel to intuitively feel the impact of different load layouts on the robot's dynamic stability, enabling them to complete the optimal job configuration before physical operations.

[0129] Example 9: This example details the dynamic modeling and optimization strategy of the system in handling complex working conditions.

[0130] When the automated guided vehicle (AGV) performs high-dynamic tasks such as emergency braking or rapid obstacle avoidance, the sampling frequency in the physical entity sensing unit is temporarily increased to capture the high-frequency vibration characteristics of the vehicle body. The multibody dynamics solution unit switches to transient dynamics mode, increasing the computational depth for structural damping effects and nonlinear damper characteristics.

[0131] To prevent a decrease in real-time performance due to excessive computational load under high dynamic conditions, the parameter evolution control unit is configured to employ a parameter freezing strategy based on sensitivity analysis. This strategy identifies key parameters that contribute most to the dynamic response, such as road adhesion coefficient and braking system pressure, based on the severity of the current operating conditions. Only these key parameters are identified at high frequencies, while parameters with less impact on current efficiency, such as battery mass drift, are temporarily frozen.

[0132] This dynamic resource allocation strategy ensures that the system can still provide accurate digital twin mappings under extreme operating conditions, providing reliable data support for safety assessments and emergency response decisions.

[0133] Example 10: This example details the physical aspects of the present invention for tire wear modeling.

[0134] The parameter evolution control unit integrates a tire wear model, which describes the functional relationship between the reduction in tire diameter and the cumulative number of rolling circumferences, average vertical load, and steering sideslip energy. After each dynamics calculation, the multibody dynamics solution unit adds the tire energy dissipation value generated in that calculation to the wear model.

[0135] As operating time progresses, the effective wheel diameter identified by the parameter evolution control unit gradually decreases. Since the change in wheel diameter directly affects the mapping relationship between motor speed and actual vehicle speed, the multibody dynamics solution unit adjusts the radius term in the dynamic equations based on the latest wheel diameter parameters.

[0136] The twin model synchronization unit, displayed in the visualization interface, not only renders the visual wear and tear on the tire tread but also synchronously updates relevant maintenance metrics, such as the expected replacement mileage. This wear evolution based on physical mechanisms avoids the resource waste or insufficient maintenance caused by traditional scheduled maintenance, achieving true condition-based monitoring and maintenance.

[0137] Example 11: This example details the application of the present invention in structural stiffness identification.

[0138] The parameter evolution control unit is configured to identify fatigue loosening at chassis weld points. When the automated guided vehicle (AGV) traverses a known standard protruding obstacle, the physical entity sensing unit records the vehicle's impact response waveform.

[0139] The multibody dynamics solver unit uses the current structural stiffness parameters to simulate the same obstacle impact process. The parameter evolution control unit compares the transfer function differences between the actual vibration waveform and the simulated waveform in the frequency domain. If the resonant frequency of the system is found to shift towards lower frequencies and the attenuation ratio increases, it is determined that the overall stiffness of the structure has decreased.

[0140] This vibration-based identification method enables digital twin systems to detect internal structural degradation that is invisible to the naked eye. The twin model synchronization unit then marks these hidden damages in augmented reality at corresponding locations on the virtual chassis, guiding maintenance personnel to perform targeted tightening or reinforcement work.

[0141] Example 12: This example details the processing logic of the visual semantic parsing unit for irregular goods in this invention.

[0142] When the automated guided vehicle (AGV) carries non-standard, irregular cargo, the point cloud reconstruction module of the visual semantic parsing unit uses the Poisson reconstruction algorithm to generate a complete 3D surface mesh of the cargo. The physical property mapping module then performs volume integration on the complete 3D surface mesh to calculate the precise geometric center of the cargo.

[0143] Because the internal mass distribution of the cargo may be uneven, the parameter evolution control unit further utilizes the pitch angle changes of the automated guided vehicle during small-amplitude acceleration to inversely calculate the true physical center of gravity height. If the identified physical center of gravity deviates from the visually calculated geometric center, the system determines that the cargo has uneven internal mass.

[0144] This deviation value is fed back to the multibody dynamics solver unit to construct an asymmetric inertia tensor. During rendering, the twin model synchronization unit informs maintenance personnel of the actual center of gravity offset of the cargo by displaying a virtual centroid within the cargo, thus preventing tipping during subsequent high-speed turns.

[0145] Example 13: This example describes the details of the present invention in terms of the coupling of energy systems and dynamics.

[0146] The physical entity sensing unit monitors the voltage drop curve of the battery pack in real time. The parameter evolution control unit establishes an equivalent circuit model of the battery and correlates it with the dynamic model. Since the battery voltage drops as the remaining capacity decreases, this causes the actual power output capability of the motor driver to change at the same duty cycle.

[0147] The multibody dynamics solver is configured to consider this energy-constrained characteristic and calculate in real time the maximum achievable acceleration and maximum climbing angle under the current power level. The twin model synchronization unit displays these limiting performance indicators in the form of a dynamic range envelope in virtual space.

[0148] As battery aging leads to increased internal resistance, the parameter evolution control unit reduces the efficiency coefficient of electrical energy conversion to mechanical energy. This interdisciplinary coupling mechanism allows the digital twin to accurately reflect the evolution of the robot's physical properties as it becomes less efficient with use, providing a physical basis for precise task planning and recharging scheduling.

[0149] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0150] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AGV / AMR digital twin operation and maintenance system integrating dynamic modeling, characterized in that, include: The physical entity sensing unit is configured to collect multi-source sensor data in real time during the operation of automated guided vehicles or autonomous mobile robots in order to obtain raw observation data on their motion state and internal load changes. The visual semantic parsing unit is configured to receive depth images and color image sequences, perform point cloud reconstruction and visual synchronous localization and mapping, extract load distribution features and vehicle body structural deformation features, and convert the load distribution features and vehicle body structural deformation features into physical attribute parameters. The multibody dynamics solving unit is configured to build a high-fidelity dynamics model, receive the physical property parameters as boundary conditions, and calculate the force distribution, rotational inertia matrix and tire contact physical quantities of each component of the vehicle body in real time. The twin model synchronization unit is configured to maintain a digital twin corresponding to the physical entity in the virtual space, and drive the digital twin to perform attitude updates and geometric deformations based on the dynamic response results output by the multibody dynamics solving unit. The parameter evolution control unit is configured to perform online identification and adaptive correction of the model parameters in the multibody dynamics solution unit, and realize the dynamic update of the model parameters based on the multi-dimensional observation data provided by the physical entity perception unit and the visual semantic parsing unit.

2. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 1, characterized in that, The physical entity sensing unit specifically includes: The sensor acquisition subunit integrates a high-precision six-axis inertial measurement unit, a drive motor controller interface, and a battery management system interface. It is configured to acquire the angular velocity vector and linear acceleration vector of the vehicle body in three-dimensional space, read the phase current feedback signal of each wheel drive motor, encoder pulse signal and output torque estimation value, and obtain the real-time state of charge, cell voltage distribution and operating temperature of the battery pack. The signal preprocessing subunit is connected to the sensor acquisition subunit and is configured to perform recursive least squares filtering or low-pass filtering on the acquired raw signal to remove noise signals introduced by mechanical vibration or electromagnetic interference. The data synchronization encapsulation subunit is configured to assign a unified timestamp to the preprocessed inertial data, motion data, and energy data, and encapsulate multi-source heterogeneous data into standardized telemetry data frames according to a preset communication protocol.

3. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 2, characterized in that, The visual semantic parsing unit specifically includes: The point cloud reconstruction module is configured to receive point cloud streams from a color depth camera and uses an iterative nearest-point algorithm to incrementally construct visual odometry, generating local point cloud maps of the operating environment and the automated guided vehicle itself. The semantic segmentation module, which embeds a deep convolutional neural network, is configured to classify objects in the point cloud map and identify the cargo outline, pallet posture, and key structural components of the vehicle body on the cargo platform. The physical property mapping module is configured to inversely calculate the equivalent mass and centroid coordinates of the cargo based on the identified cargo geometry and a preset density dictionary. When the semantic segmentation module detects that the cargo is stacked tilted or unbalanced, the physical attribute mapping module calculates the deviation moment of the cargo from the centerline of the loading platform using a voxelization method and outputs a rotational inertia correction term. The visual semantic parsing unit is further configured to identify the stacking tilt angle and center of mass offset of goods on the cargo platform, and use the stacking tilt angle and center of mass offset as key inputs for the correction of the rotational inertia matrix, so that the dynamic model can dynamically reflect the changes in inertial characteristics under different load conditions.

4. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 3, characterized in that, The multibody dynamics solution unit specifically includes: The topology description module is configured to construct a topology description matrix that includes rigid bodies, elastic bodies, and constraint pairs. The recursive solution module employs a recursive Newton-Louras cooperative algorithm, configured to calculate the kinematic state of each component through forward recursion, and combined with the mass distribution and moment of inertia matrix output by the physical property mapping module, to solve the generalized forces and constraint reactions at the contact points of each joint and wheel train through backward recursion. The wheel-ground contact modeling module integrates a nonlinear tire force model, configured to describe the nonlinear mapping relationship between tire longitudinal slip ratio, sideslip angle and ground reaction force, and calculates tire sideslip stiffness drift and tangential adhesion coefficient by combining real-time wheel speed and vehicle trajectory. The structural stress calculation module is configured to calculate the small deformation field of the vehicle body structure under non-uniform loads and output flexible deformation data including the compression of the suspension system and the torsional angle of the chassis frame. When calculating the tire deformation under stress, the multibody dynamics solver combines the ground reaction force model with the tire material elastic parameters to output the predicted values ​​of tire compression and slip angle.

5. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 4, characterized in that, The twin model synchronization unit is configured with a three-layer geometric representation architecture, specifically including: Rigid body skeleton layer, configured to map the kinematic state of the automated guided vehicle's global pose and major mechanical connections; The outer shell layer is composed of a high-density triangular mesh. The spatial position of the mesh vertices is driven in real time by the structural deformation field output by the multibody dynamics solving unit. Through radial basis function interpolation or finite element interpolation algorithm, the visualization of tire compression deformation, shock absorber compression and frame micro-twist is realized. The load semantic layer is configured to dynamically render the cargo status in the virtual space. Based on the load attributes extracted by the visual semantic parsing unit, it updates the shape, texture, and relative position of the virtual cargo with respect to the vehicle body in real time. The twin model synchronization unit supports multi-timescale synchronization strategies. For motion posture updates, a fast channel at the millisecond level is used to achieve following the physical entity, while a slow channel at the second level or task cycle level is used for model parameter updates.

6. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 5, characterized in that, The parameter evolution control unit specifically includes: The residual calculation module is configured to store the nominal dynamic model and calculate the mean or variance of the residual sequence between the actual motor current fed back by the physical entity sensing unit and the expected current predicted by the multibody dynamics solving unit under the same working conditions by comparing them in real time. An adaptive correction logic module is configured to trigger an online correction process when the residual sequence continuously exceeds a preset tolerance threshold, and to use a gain extended Kalman filter algorithm to correct the equivalent friction coefficient, transmission system damping, and overall mass in the model online. The parameter evolution control unit uses a system identification algorithm to analyze the deviation between motor current and wheel speed collected by the physical entity sensing unit. When the residual continuously exceeds the preset threshold, the online calibration process of the model parameters is triggered, and the equivalent mass and damping coefficient in the dynamic model are automatically adjusted.

7. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 6, characterized in that, The parameter evolution control unit also integrates an aging factor accumulation submodule, which is configured to construct an exponential evolution model of tire wear, motor efficiency decay and suspension stiffness degradation based on the cumulative mileage, frequent start-up times and battery cycle count of the automated guided vehicle, so that the digital twin can simulate the performance degradation process of the physical entity throughout its entire life cycle. The aging factor accumulation submodule is specifically configured to execute the following logic: Based on operating mileage, number of operation cycles, and battery charge / discharge cycles, a trend model is constructed to investigate the relationship between tire wear and suspension system stiffness reduction. Based on the stress cycle sequence of key welding points calculated by the multibody dynamics solution unit, the stress amplitude distribution is statistically analyzed using the rainflow counting method. Combined with the fatigue life curve of the material, the remaining life estimate of each structural component is marked in real time in the digital twin.

8. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 7, characterized in that, A physical feature buffer is set up between the visual semantic parsing unit and the multibody dynamics solving unit, and a bidirectional data channel is established: The visual semantic parsing unit writes the inversely solved physical attributes, including the cargo's instantaneous center of mass height, eccentricity, and detected structural elastic deformation, into the physical feature buffer at a first frequency. The multibody dynamics solving unit reads data from the physical feature buffer at a second frequency greater than the first frequency, and performs smooth interpolation of the physical features through an internally integrated second-order kinematics interpolator to generate a continuous boundary condition flow. The tire normal force calculated by the multibody dynamics solving unit is fed back to the visual semantic parsing unit. The visual semantic parsing unit uses the fed-back force information to constrain and optimize the vertices of the tire-ground contact area in the point cloud, and uses the dynamically predicted structural deformation trend as a priori constraint for visual point cloud registration.

9. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 8, characterized in that, The system adopts an edge and cloud collaborative architecture, specifically including: The edge-aware execution layer, deployed on the automated guided vehicle, includes a lightweight front-end of the physical entity perception unit and the visual semantic parsing unit, as well as a real-time kinematics core of the multibody dynamics solving unit. It is configured to perform point cloud feature dimensionality reduction and only uploads the physical feature vectors to the cloud. The cloud-based simulation evolution layer, deployed on a server cluster, includes a high-order simulation module of the multibody dynamics solution unit, a parameter evolution control unit, and a twin model synchronization unit. It is configured to run a fully parameterized dynamics model and calculate the stress distribution and long-term dynamic evolution of the vehicle body's internal structure. The cloud-based simulation evolution layer also includes a global digital twin resource scheduler, configured to dynamically adjust the computing resource quotas of each twin in the cloud based on the operational intensity and task priority of each automated guided vehicle.

10. The AGV / AMR digital twin operation and maintenance system integrating dynamic modeling according to claim 9, characterized in that, The system also includes: The predictive simulation module is configured to enable the multibody dynamics solution unit to be decoupled from the physical entity perception unit in real time. Based on the current parameter state and the preset planning path, it performs advanced time step iterations to simulate the dynamic response in the future time period in order to predict whether there is a risk of instability due to overload or excessive center of gravity. A tactile feedback device, connected to the twin model synchronization unit, is configured to apply corresponding resistance feedback to the operator when the operator manually drags and drops goods in the virtual space to simulate adjusting the layout, based on the virtual gravity and friction calculated by the multibody dynamics model. The environment mapping feedback link is configured to automatically associate the ground material features identified by the visual semantic parsing unit with a preset friction coefficient reference value, and the parameter evolution control unit performs system identification in the vicinity of the friction coefficient reference value.

Images