Modular robot digital twin simulation building and virtual debugging system

Abstract

The application relates to the technical field of industrial control software, and discloses a modular robot digital twin simulation building and virtual debugging system, which comprises a topological atlas analysis module, a local state solving module and a dynamic constraint arbitration module.The topological atlas analysis module is used for translating the logical connection relationship of a topological unit into a global topological sparse tensor, so that virtual assembly topological data is abstracted into a discrete topological graph.The local state solving module is used for extracting a diagonal block matrix in the global topological sparse tensor, and decoupling and splitting a system overall control equation set into a plurality of independent sub-state equations.The dynamic constraint arbitration module is used for triggering a local neighborhood solving operator to start relaxation iteration when a transient boundary deviation value is greater than a discrete convergence threshold value, so as to correct a constraint boundary variable and output a state update instruction.The application breaks the inertia of maintaining a global continuous dynamics control matrix solution, avoids repeated inverse operation of a full matrix caused by frequent changes of a local topological graph, and reduces data flow delay.

Description

Technical Field

[0001] This invention belongs to the field of industrial control software technology, and in particular relates to a modular robot digital twin simulation construction and virtual debugging system. Background Technology

[0002] Current multi-joint component motion simulation systems form the core support for virtual debugging. They typically employ a globally centralized rigid body dynamics solver to establish a single dynamic control matrix that encompasses the system's geometry and physical parameters. The state variables are iterated alternately within continuous time steps to simulate motion trajectories and predict physical behavior. The core step in solving the dynamic equations with this type of solver is to calculate the inverse of the multidimensional control matrix or solve the full matrix triangular decomposition. The overall computational overhead diverges cubically with increasing system degrees of freedom. When facing the construction and virtual debugging of modular robot digital twin simulation systems with frequently changing topologies, the centralized architecture has transmission defects. Frequent changes in physical configuration cause the simulation software to frequently reconstruct the global control matrix during motion intervals or dynamic interaction nodes, resulting in computational delays in data flow paths and causing the simulation clock to become out of sync with the physical clock.

[0003] To reconcile the trade-off between computational latency and fidelity, conventional improvement strategies attempt to employ multi-core parallel acceleration or pre-set static rigid body model templates. These linear solutions fail to overcome the computational cost losses caused by configuration changes. Simply increasing computational power or adding hardware resources, when faced with real-time changes in the topology network structure, still leads to frequent disruptions of the inherent sparse matrix arrangement, causing the system to reallocate memory space and rebuild the global elimination tree, resulting in escalating computational costs and hindering the provision of continuous and stable state evolution. Multi-joint components, due to the inherent limitations of hardware assembly physical rigidity or roller shape, restrict the overall machine's flexible tuning. The underlying control methods also have shortcomings, for example... Chinese invention patent CN119596709B discloses an adaptive control method for the topology of a homogeneous modular robot. This scheme relies on a globally unified continuous dynamic control equation and completes motion control by recursively calculating higher-order partial derivatives and gradient matrices. However, when faced with the frequently changing configuration topology in a digital twin simulation environment, the implicit basis for this control scheme is mismatched. Real-time topology reconstruction causes the entire state space control matrix and higher-order partial derivatives to be globally recalculated, leading to frequent inversions of multidimensional coupling matrices, resulting in computational blockages and timing delays. This makes it impossible to meet the high-precision real-time alignment requirements between the simulation clock and the physical clock.

[0004] Therefore, the technical problem to be solved by this invention is how to achieve the decoupling and asynchronous scheduling of the dynamic control matrix through sparse tensor topology decomposition and independent mounting mechanism of local state equations, so as to reduce the matrix calculation overhead during topology changes without losing the stress fidelity of the components. Summary of the Invention

[0005] This invention aims to solve the problem of simulation calculation delay divergence caused by centralized solution of the global control matrix and frequent topology changes.

[0006] In this technical solution, a modular robot digital twin simulation construction and virtual debugging system includes:

[0007] The topology graph parsing module is used to translate the logical connection relationship of the topological unit of the modular robot into a global topological sparse tensor by acquiring the topological architecture data that represents the topological transformation of the rigid body model data of the modular robot. This abstracts the virtual assembly topology data into a discrete topology graph composed of logical nodes and constraint keys.

[0008] The local state solution module is connected to the topology graph analysis module. The local state solution module includes multiple parallel discrete solution kernels. The local state solution module is used to extract the diagonal block matrix corresponding to each logical node in the global topology sparse tensor by relying on the block sparsity characteristics in the global topology sparse tensor. The system's overall control equations are decoupled and decomposed into multiple independent sub-state equations mounted on each logical node. Each sub-state equation contains only the intrinsic state parameters of the corresponding logical node and the numerical transmission load of the adjacent constraint keys.

[0009] The dynamic constraint arbitration module is mounted on the data flow interface of each sub-state equation. The dynamic constraint arbitration module is used to monitor the transient boundary deviation value of two adjacent logical nodes during the simulation operation. When the transient boundary deviation value is greater than the given discrete convergence threshold, it only triggers the local neighborhood solver operator that has a causal dependence on the corresponding connection boundary to start relaxation iteration. The correction matrix parameter obtained by the iteration is reversed to cover the constraint boundary variable in the corresponding sub-state equation, and the state update instruction is output to the virtual debugging driver module to update the twin scene.

[0010] Preferably, the system also includes a discrete state event asynchronous scheduling module; the discrete state event asynchronous scheduling module is deployed at the input end of the local state solution module. The discrete state event asynchronous scheduling module is used to receive the change signals generated by the concurrent collision of multiple logical nodes of the modular robot, convert the collision conflict into discrete state requests, extract the local kinetic energy change rate fed back by each component node, allocate multi-threaded computing resource weights through an asymmetric thread pool scheduling mechanism, and establish an asynchronous queuing processing sequence for conflict data. This limits the thread blocking caused by the intertwining of solution clues under sudden multi-point contact states, and enables the convergence of the underlying state flow timing of the simulation solver under the scenarios of multi-configuration component replacement and concurrent load mutation.

[0011] Preferably, the system further includes a control equation damping coefficient adaptive adjustment module; the control equation damping coefficient adaptive adjustment module is connected in the feedback loop of the dynamic constraint arbitration module, and is used to: capture the fluctuation data value generated when the transient boundary deviation value approaches the system stability critical point when the dynamic constraint arbitration module monitors the transient boundary deviation value; adjust the virtual damping coefficient of adjacent constraint keys to adjust the local constraint flexibility, balance the boundary sudden load by means of the adaptive convergence characteristics of the local variables of each sub-state equation, avoid the geometric truncation error caused by the use of hard-coded static model, and maintain the continuity of force transmission data in multi-configuration assembly scenarios.

[0012] Preferably, the dynamic constraint arbitration module includes: real-time monitoring of the motion evolution and changes between adjacent component logic nodes, capturing discrete jump data of boundary feature parameters; triggering adjacent local neighborhood solvers to start iterative relaxation calculations, and reverse-overwriting the constraint boundary variables inside the sub-state equations to balance the boundary abrupt loads.

[0013] Preferably, the discrete state event asynchronous scheduling module includes: guiding concurrent computation data flow to thread nodes with high response weights through an asymmetric thread pool scheduling mechanism; and establishing an asynchronous queuing sequence for conflicting data to eliminate thread blockage caused by the intertwining of computation threads under sudden multi-point contact conditions.

[0014] Preferably, the topology parsing module includes: converting each independent rigid body data unit in the three-dimensional virtual assembly model of the modular robot into a corresponding logical node; and converting the hinge constraint data and contact surface contact constraint data between the independent rigid body data units into corresponding constraint keys.

[0015] Preferably, the local state solution module includes: extracting the diagonal block matrix corresponding to each logical node in the global topological sparse tensor; separating the rigid body dynamics matrix belonging to global coupling in the overall system control equations; and importing the separated rigid body dynamics matrix as the block solution boundary of the sub-state equations into the corresponding discrete solution kernel.

[0016] Preferably, the dynamic constraint arbitration module further includes: comparing the transient boundary deviation value with the given convergence tolerance range in real time; and sending a trigger signal to the control equation damping coefficient adaptive adjustment module to adjust the virtual damping coefficient when the transient boundary deviation value exceeds the upper limit of the convergence tolerance range for three consecutive control cycles.

[0017] Preferably, the system also includes a virtual debugging driver module; the virtual debugging driver module is connected to the data flow interface of the dynamic constraint arbitration module, and the virtual debugging driver module is used to receive the correction state command output by the dynamic constraint arbitration module to correct the virtual motion pose of each rigid body data unit in the twin scene in real time, so that the data flow path of the virtual debugging simulation clock and the physical side clock signal is kept synchronized.

[0018] Compared with existing technologies, the modular robot digital twin simulation construction and virtual debugging system of the present invention has the following advantages:

[0019] 1. In the construction of robot digital twin simulation, the topology map analysis unit transforms the spatial assembly relationship into a sparse tensor topology structure. The local state solver partitions and separates the global control matrix of the system, extracts the intrinsic parameters of physical components and adjacent constraint loads, and decomposes the solution of the overall algebraic equation system into independent calculations of the sub-state equations of each component node. This mechanism breaks the inertia of maintaining the solution of the global continuous dynamic control matrix, cuts off the strong coupling link in the solution process of the motion state of each component, avoids repeated matrix inversion operations caused by frequent changes in local topology, blocks the spread of redundant computing power loss to the far-end non-changing region, greatly reduces data flow delay, and makes the computational cost of the simulation system evolve linearly with the component scale.

[0020] 2. The dynamic constraint arbitration closed loop connects the data interaction interfaces of each sub-state equation, monitors the motion evolution and changes between adjacent component logic nodes in real time, captures the boundary stress transition waveform, triggers the adjacent local neighborhood solver to start iterative relaxation calculation, reverses the unknown constraint parameters inside the sub-state equation, and outputs the corrected state command to the virtual debugging drive component; this interaction loop establishes a local incremental feedback compensation channel, without relying on external hardware data, relies on the adaptive convergence characteristics of local variables to balance the boundary sudden load, avoids the geometric truncation error caused by the use of hard-coded static models, and maintains high fidelity and physical compactness of local force transmission in multi-configuration assembly scenarios.

[0021] 3. The discrete state event asynchronous scheduling queue is deployed at the input end of the local state solver partition. It receives the change signals generated by concurrent collisions of multiple nodes, transforms the collision conflicts into discrete state requests, extracts the local kinetic energy change rate of each component node, and dynamically allocates the weight of multi-threaded computing resources. This structure guides the concurrent computing flow to the high dynamic response region through asymmetric thread pool scheduling, establishes an asynchronous queuing sequence for conflict data processing, suppresses the risk of thread deadlock caused by the intertwining of solution threads under sudden multi-point contact states, and ensures the convergence stability and solution continuity of the underlying state flow of the simulation solver under scenarios of frequent replacement of multi-configuration components and sudden changes in concurrent loads. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the structure of the digital twin virtual debugging system of the present invention;

[0023] Figure 2 This is the state transition diagram of the digital twin virtual debugging system of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0025] A modular robot digital twin simulation construction and virtual debugging system includes:

[0026] The topology graph parsing module is used to translate the logical connection relationship of the topological unit of the modular robot into a global topological sparse tensor by acquiring the topological architecture data that represents the topological transformation of the rigid body model data of the modular robot. This abstracts the virtual assembly topology data into a discrete topology graph composed of logical nodes and constraint keys.

[0027] The local state solution module is connected to the topology graph analysis module. The local state solution module includes multiple parallel discrete solution kernels. The local state solution module is used to extract the diagonal block matrix corresponding to each logical node in the global topology sparse tensor by relying on the block sparsity characteristics in the global topology sparse tensor. The system's overall control equations are decoupled and decomposed into multiple independent sub-state equations mounted on each logical node. Each sub-state equation contains only the intrinsic state parameters of the corresponding logical node and the numerical transmission load of the adjacent constraint keys.

[0028] The dynamic constraint arbitration module is mounted on the data flow interface of each sub-state equation. The dynamic constraint arbitration module is used to monitor the transient boundary deviation value of two adjacent logical nodes during the simulation operation. When the transient boundary deviation value is greater than the given discrete convergence threshold, it only triggers the local neighborhood solver operator that has a causal dependence on the corresponding connection boundary to start relaxation iteration. The correction matrix parameter obtained by the iteration is reversed to cover the constraint boundary variable in the corresponding sub-state equation, and the state update instruction is output to the virtual debugging driver module to update the twin scene.

[0029] Preferably, the system also includes a discrete state event asynchronous scheduling module; the discrete state event asynchronous scheduling module is deployed at the input end of the local state solution module. The discrete state event asynchronous scheduling module is used to receive the change signals generated by the concurrent collision of multiple logical nodes of the modular robot, convert the collision conflict into discrete state requests, extract the local kinetic energy change rate fed back by each component node, allocate multi-threaded computing resource weights through an asymmetric thread pool scheduling mechanism, and establish an asynchronous queuing processing sequence for conflict data. This limits the thread blocking caused by the intertwining of solution clues under sudden multi-point contact states, and enables the convergence of the underlying state flow timing of the simulation solver under the scenarios of multi-configuration component replacement and concurrent load mutation.

[0030] Preferably, the system further includes a control equation damping coefficient adaptive adjustment module; the control equation damping coefficient adaptive adjustment module is connected in the feedback loop of the dynamic constraint arbitration module, and is used to: capture the fluctuation data value generated when the transient boundary deviation value approaches the system stability critical point when the dynamic constraint arbitration module monitors the transient boundary deviation value; adjust the virtual damping coefficient of adjacent constraint keys to adjust the local constraint flexibility, balance the boundary sudden load by means of the adaptive convergence characteristics of the local variables of each sub-state equation, avoid the geometric truncation error caused by the use of hard-coded static model, and maintain the continuity of force transmission data in multi-configuration assembly scenarios.

[0031] Preferably, the dynamic constraint arbitration module includes: real-time monitoring of the motion evolution and changes between adjacent component logic nodes, capturing discrete jump data of boundary feature parameters; triggering adjacent local neighborhood solvers to start iterative relaxation calculations, and reverse-overwriting the constraint boundary variables inside the sub-state equations to balance the boundary abrupt loads.

[0032] Preferably, the discrete state event asynchronous scheduling module includes: guiding concurrent computation data flow to thread nodes with high response weights through an asymmetric thread pool scheduling mechanism; and establishing an asynchronous queuing sequence for conflicting data to eliminate thread blockage caused by the intertwining of computation threads under sudden multi-point contact conditions.

[0033] Preferably, the topology parsing module includes: converting each independent rigid body data unit in the three-dimensional virtual assembly model of the modular robot into a corresponding logical node; and converting the hinge constraint data and contact surface contact constraint data between the independent rigid body data units into corresponding constraint keys.

[0034] Preferably, the local state solution module includes: extracting the diagonal block matrix corresponding to each logical node in the global topological sparse tensor; separating the rigid body dynamics matrix belonging to global coupling in the overall system control equations; and importing the separated rigid body dynamics matrix as the block solution boundary of the sub-state equations into the corresponding discrete solution kernel.

[0035] Preferably, the dynamic constraint arbitration module further includes: comparing the transient boundary deviation value with the given convergence tolerance range in real time; and sending a trigger signal to the control equation damping coefficient adaptive adjustment module to adjust the virtual damping coefficient when the transient boundary deviation value exceeds the upper limit of the convergence tolerance range for three consecutive control cycles.

[0036] Preferably, the system also includes a virtual debugging driver module; the virtual debugging driver module is connected to the data flow interface of the dynamic constraint arbitration module, and the virtual debugging driver module is used to receive the correction state command output by the dynamic constraint arbitration module to correct the virtual motion pose of each rigid body data unit in the twin scene in real time, so that the data flow path of the virtual debugging simulation clock and the physical side clock signal is kept synchronized.

[0037] Example 1: When the system faces the situation of frequent topological changes of modular robots to adapt to multi-task operations, the global centralized rigid body dynamics solver iterates state variables alternately within continuous time steps to solve the inverse matrix of the multidimensional control matrix. This centralized architecture causes the computational overhead caused by local configuration changes to spread to the non-changing region, frequently breaking the inherent sparse matrix arrangement state and causing the system to reallocate memory space and rebuild the global elimination tree. The increased computational cost hinders the system from providing continuous and stable state evolution. The system obtains the topological architecture data representing the topological transformation of the rigid body model data of the modular robot through the topological graph analysis module, and translates the logical connection relationship of the topological unit of the modular robot into a global topological sparse tensor. Thus, the virtual assembly topological data is abstracted into a discrete topological graph composed of logical nodes and constraint keys, providing a topological structure basis for subsequent matrix block isolation. The local state solving module connected to the topological graph analysis module uses multiple internally integrated parallel discrete solving kernels to extract the diagonal block matrix corresponding to each logical node in the global topological sparse tensor based on the block sparsity characteristics in the global topological sparse tensor.

[0038] The rigid body dynamics matrix belonging to global coupling in the overall system control equations is separated and imported into the corresponding discrete solution kernel as the block solution boundary. It is decoupled and split into multiple independent sub-state equations mounted on each logic node. Each sub-state equation contains only the intrinsic state parameters of the corresponding logic node and the numerical transmission load of the adjacent constraint keys. Specifically, when extracting the diagonal block matrix, this invention partitions the non-zero elements in the global topological sparse tensor according to the degrees of freedom of the rigid body component. Each independent rigid body component corresponds to a 6-row by 6-column diagonal block matrix to characterize the component's mass, moment of inertia and intrinsic force state.

[0039] For strongly coupled elements at off-diagonal positions, i.e., hinge constraints and contact constraints across components, this invention does not directly ignore them, but converts them into external boundary constraint loads, and extracts the corresponding constraint force values ​​from the solution results of the previous step, and attaches them as known external force terms to the data flow interface of the corresponding sub-state equations; by converting the off-diagonal coupling terms in the global matrix into known load boundaries of each sub-state equation, complete decoupling at the algebraic level is achieved, enabling each discrete solution kernel to solve the dynamic response of a single rigid body component in parallel and independently, avoiding the global full matrix inversion operation of the multidimensional control matrix.

[0040] The dynamic constraint arbitration module, mounted on the data flow interface of each sub-state equation, monitors the transient boundary deviation value of two adjacent logical nodes in real time during the simulation operation. When the transient boundary deviation value is greater than the given discrete convergence threshold, it triggers the local neighborhood solver operator that has a causal dependency with the corresponding connection boundary to start relaxation iteration. The correction matrix parameters obtained by the iterative solution are then used to inversely cover the constraint boundary variables in the corresponding sub-state equation. The state update command is then output to the virtual debugging drive module to update the twin scene. This local node adaptive convergence driven by discrete connection events relies on the independent sub-state equation obtained by the local state solver module from the global dynamic matrix as the basis. The calculation range is limited to the affected neighborhood subgraph, preventing the computational overhead from spreading to other nodes outside the change area of ​​the entire network. This prevents the system computational overhead from increasing cubically with the expansion of the overall joint size of the robot. While maintaining the fidelity of mechanical stress simulation, it reduces the matrix solution delay when the topology changes frequently.

[0041] When the system encounters a multi-node concurrent physical collision generating a change signal, the discrete state event asynchronous scheduling module deployed at the input of the local state solution module receives the change signal, converts the collision conflict into a discrete state request, extracts the local kinetic energy change rate fed back by each component node, and allocates multi-threaded computing resource weights through an asymmetric thread pool scheduling mechanism. An asynchronous queuing sequence for conflict data is established to limit thread blocking caused by the intertwining of solution threads under sudden multi-point contact conditions, guiding concurrent computing data flow to thread nodes with high response weights. This ensures the convergence of the underlying state transition timing of the simulation solver under multi-configuration component replacement and concurrent load mutation conditions. In the actual allocation of multi-threaded computing resource weights, this invention uses virtual sensors set on each component node to obtain its local kinetic energy change rate value in real time and compares this value with a preset collision energy benchmark value. When the local kinetic energy change rate generated by a concurrent collision of a certain group of logical nodes reaches more than twice the benchmark value, the asymmetric thread pool scheduling mechanism assigns the node to the appropriate thread pool scheduling mechanism. Discrete state requests are marked as high priority and allocated core computing threads accounting for 60% of the total thread pool resources; if the kinetic energy change rate is between 1 and 2 times the baseline value, 30% of the thread resources are allocated; the remaining low dynamic response regions share the remaining 10% of the thread resources. This quantization mapping rule, which uses a step-by-step weight allocation based on the magnitude of the kinetic energy change rate, ensures that computing resources can be targeted and concentrated in the micro-regions with the most intense physical conflicts, effectively eliminating thread deadlocks caused by the intertwining of decomposition threads. At the same time, the control equation damping coefficient adaptive adjustment module connected to the feedback loop of the dynamic constraint arbitration module receives the adjustment trigger signal when the dynamic constraint arbitration module compares the transient boundary deviation value with the given convergence tolerance range in real time, and the transient boundary deviation value exceeds the upper limit of the convergence tolerance range for three consecutive control cycles. It captures the fluctuation data value generated when the transient boundary deviation value approaches the system stability critical point, and adjusts the virtual damping coefficient of adjacent constraint keys according to the ratio of the transient boundary deviation value to the total system energy.

[0042] By increasing the virtual damping coefficient in a stepwise manner to adjust the flexibility of local constraints, and leveraging the adaptive convergence characteristics of local variables in each sub-state equation, the system balances abrupt boundary load changes and dissipates the oscillating energy generated by numerical calculation truncation. This avoids geometric truncation errors caused by hard-coded static models and maintains the continuity and temporal alignment of force transmission data in multi-configuration assembly scenarios. To prevent numerical explosions or geometric position penetrations within the transition blind zone of three consecutive control cycles, this invention deploys a temporary state-holding mechanism in the dynamic constraint arbitration module. During these three buffer cycles that exceed the convergence tolerance but have not yet triggered damping coefficient adjustment, the system does not directly use divergent transient positions for rendering. Instead, it locks the current motion constraint boundary and estimates a temporary motion pose using an extrapolation prediction algorithm, while limiting the maximum displacement variable within a single simulation step to no more than 0.5 mm. This fallback-style temporal flow control effectively dissipates abnormal abrupt energy within the blind zone, facilitating smooth transitions in subsequent adaptive adjustment modules. Increasing the virtual damping coefficient ensures a stable transition clock cycle, preventing system collapse during sudden state changes in the digital twin scenario. In the topology reconstruction test of a robotic arm with 48 modular joints, when high-frequency topology switching occurs, the local state solution module extracts the diagonal block matrix corresponding to each logical node from the global topology sparse tensor output by the topology graph analysis module. By eliminating the continuous full matrix inversion operation of the global control matrix, the single-step update delay of the system's virtual debugging is stably controlled from 45.2ms to 3.4ms to 3.8ms. This decreasing trend confirms the downward evolution of the system's computational complexity from cubic to linear order. In this state, the virtual debugging drive module receives the correction state command output by the dynamic constraint arbitration module, and corrects the virtual motion pose of each rigid body data unit in the twin scenario in real time. The residual convergence accuracy of the structural stress simulation reaches within 0.0001, ensuring that the virtual debugging simulation clock and the physical clock signal data flow path remain synchronized.

[0043] Example 2: When the system performs virtual debugging data verification on a modular robot containing 48 rigid body components in a multi-body rigid body dynamics simulation platform built based on Newton's laws of motion and the law of conservation of momentum, the original physical input data used in this experiment comes from a multi-axis series load actuator under electromagnetic interference environment on the physical side. Gaussian white noise with a signal-to-noise ratio of 20dB and a 50Hz power frequency harmonic interference signal are actively superimposed to reproduce the measurement noise of the industrial site. Furthermore, this experiment calibrates the thread scheduling cycle in the discrete state event asynchronous scheduling module, which is a control parameter. The setting of the thread scheduling cycle depends on the burst density of concurrent physical collision events and the processing... The processor's multi-core parallel computing resources are used to balance the response time of multi-threaded parallel computing with the thread synchronization overhead caused by lock contention. The quantitative judgment rule is as follows: when the frequency of concurrent physical collision events within a single discrete simulation step is greater than 1000 times per second, the thread scheduling period is set to the upper limit of the corresponding range to reduce the context switching overhead caused by thread wake-up and suspension. When the frequency is less than 100 times per second, the thread scheduling period is set to the lower limit of the corresponding range to improve the transient refresh rate of the virtual debugging scenario. Under the assembly modification condition of this experiment, the engineering instance value of the thread scheduling period calculated according to the aforementioned quantitative judgment rule is 5ms.

[0044] After the aforementioned assembly modification was initiated, this experiment simultaneously ran a control group using a global centralized rigid body dynamics solver and an experimental group using this invention's technical solution. When faced with the dynamic signals generated by 12 concurrent joint physical collisions, the control group directly received the raw motion load data containing Gaussian white noise. Because the overall geometric information was forcibly bound to the physical parameters, the dimension of its global control matrix increased to 288 rows by 288 columns, causing the elimination tree to be destroyed and reconstructed within consecutive time steps. In contrast, the experimental group of this invention used a topology graph analysis module to translate the disturbing logical connections into a global topological sparse tensor, which was then used by the local state solution module. Based on the block sparsity characteristics, the internally integrated multiple parallel discrete solution kernels directly extract 48 independent diagonal block matrices with dimensions of 6 rows by 6 columns. Through this matrix block isolation action, the number of non-zero elements that were originally coupled in the control group was reduced from 82,944 to 1,728. The numerical change of the intermediate matrix feature data presents the discretization decomposition process of the system control equations. The discrete state event asynchronous scheduling module mounted on the sub-state equation data flow interface performs asymmetric allocation of multi-threaded computing resources according to the asynchronous queuing processing sequence of conflict data, so that thread nodes with high response weights can preferentially absorb the collision kinetic energy change rate.

[0045] By designing comparative tests spanning multiple scale gradients to record the nonlinear effects of the system and the turning points of data change trends, the test data shows that when the number of robot joints increases progressively from 6, 12, and 24 to the set boundaries of 48 and 96, the simulation update delay of the control group exhibits a cubic order nonlinear increase, with delay times recorded as 4.2ms, 12.5ms, 31.8ms, 156.4ms, and 1124.5ms, respectively. Delay divergence occurs after exceeding 48 joints. In contrast, the proposed sample group, under the same scale gradient, exhibits stable single-step solution update delays of 1.2ms, 1.6ms, 2.4ms, 3.6ms, and 6.2ms, respectively, showing a linear evolution trend. However, further data indicates that when the discrete convergence threshold is set below 0.00001, due to nonlinear saturation caused by approaching the hardware precision boundary of computer floating-point operations and thread synchronization overhead, the residual convergence time of the proposed sample group no longer continues to shorten but tends to flatten, even reaching... At 0.000001, local micro-oscillations caused by numerical shearing error led to a delay rebound to 14.8ms. This turning point in the trend of this data shows that the discrete convergence threshold range defined in this invention is a working range that balances numerical fidelity and computational timing convergence. The multi-dimensional quantitative comparison results of this experiment show that the modular robot digital twin simulation construction and virtual debugging system, which adopts global topological sparse tensor decomposition and independent mounting mechanism of sub-state equations, transforms the global full matrix inversion required by the control matrix into independent adaptive relaxation iteration of local neighborhood solvers under the conditions of topological reconstruction and multi-point collision concurrency. Under electromagnetic interference, the single-step update delay is controlled within 3.8ms and the residual convergence accuracy reaches within 0.0001. It suppresses the computational delay divergence and twin clock signal loss caused by centralized solution of the control matrix. When the virtual debugging drive module receives the correction state command output by the dynamic constraint arbitration module, it synchronizes the motion stress data stream to the virtual debugging simulation clock signal in situ.

[0046] Example 3: This example combines Figures 1 to 2 This section describes the modular robot digital twin simulation construction and virtual debugging system, such as... Figure 1As shown, the topology data representing the rigid body model data topology transformation is acquired and input into the topology graph parsing module. The topology graph parsing module translates the logical connection relationship of the topology unit and extracts and abstracts it into a discrete topology graph, outputting the global topology sparse tensor and the discrete topology graph to the local state solution module. At the same time, the changing signal and the concurrent collision conflict of multiple nodes trigger the discrete state event asynchronous scheduling module. The discrete state event asynchronous scheduling module extracts the local kinetic energy change rate and assigns multi-threaded calculation resource weights to output discrete state requests to the local state solution module. The local state solution module extracts the diagonal block matrix and decouples and splits the overall system control equations. It separates the matrix boundary through the discrete solution kernel and outputs it to the logical node, which includes the intrinsic state parameters and the transmitted... The independent sub-state equations of the guided load, the adaptive adjustment module of the control equation damping coefficient captures critical point fluctuation data and dynamically adjusts the virtual damping coefficient to regulate adjacent constraint keys and act on the independent sub-state equations. The independent sub-state equations output boundary parameters to the dynamic constraint arbitration module through the data transfer interface. The dynamic constraint arbitration module monitors transient boundary deviation values ​​and triggers relaxation iteration of the local neighborhood solver operator. Its output correction matrix parameters are reverse-overwritten to the independent sub-state equations. When continuous deviation exceeds the limit, it outputs a continuous deviation exceedance trigger signal to the adaptive adjustment module of the control equation damping coefficient. Finally, the dynamic constraint arbitration module outputs a state update command to the virtual debugging driver module. The virtual debugging driver module receives the correction state command and updates the twin scene in real time.

[0047] like Figure 2 As shown, the system transitions from the topology analysis state to the logic node decoupling state via the translated topology unit. The logic node decoupling state enters the constraint deviation monitoring state by triggering decoupling. When the boundary deviation exceeds the limit, the system transitions from the constraint deviation monitoring state to the relaxation iterative solution state and enters the state update instruction output state by correcting the boundary variables. When the iteration converges, the system directly transitions from the constraint deviation monitoring state to the state update instruction output state.

[0048] Example 4: When the system runs in a multi-core processor computing node environment with a main frequency of not less than 2.5GHz, a cache capacity of not less than 32MB, and a memory bandwidth of not less than 50GB / s, and the multi-core processor computing node faces the topology architecture of a modular robot continuously changing 48 rigid body components under alternating stress conditions, high-frequency stress oscillations and numerical truncation errors are generated at the local boundary during the instantaneous configuration change. The damping coefficient adaptive adjustment module of the system control equation is adjusted according to parameter calibration to control the divergence of simulation calculation. During the assembly and transformation process, the transient energy non-equilibrium transfer is caused by the discontinuous force on the topological boundary nodes. The system introduces the topological deformation perturbation potential index. Based on the state space analysis principle, the topological deformation perturbation potential index calculates the cumulative influence trend of the local kinetic energy change rate on the block sparse matrix density in the global topological sparse tensor within a sliding sampling window of 10 consecutive simulation time steps, converting the boundary topological evolution into continuously observed scalar physical data, thereby establishing the physical eigenmap path of the topological deformation perturbation potential.

[0049] In this computing node environment, when the measured local kinetic energy change rate read via the data bus is 15.4 joules per second and the scalar ratio of the block sparse matrix density is 0.12, the processor divides these two values ​​to obtain the current topological alteration perturbation potential of 128.3 joules per second; when specifically calculating the cumulative impact trend, The result value of 128.3 joules per second is used as the transient perturbation feature value of the current 10th simulation time step. The complete sliding sampling window includes the transient perturbation feature values ​​independently calculated within the current step and the aforementioned 9 consecutive simulation time steps. The processor calculates a moving average of the 10 transient perturbation feature values ​​within the sliding sampling window to eliminate numerical glitches caused by high-frequency random noise. The final moving average is used as the final scalar physical data reflecting the cumulative influence trend. When the local kinetic energy change rate changes abruptly or the block sparse matrix density undergoes discrete jumps due to topology switching, this moving average can smoothly capture the non-equilibrium state of energy transfer, thus providing stable time-series data support for the adaptive update of the convergence threshold. The discrete convergence threshold, which matches the topological alteration perturbation potential, is adjusted through a parameter adaptive update loop. The control equation damping coefficient adaptive adjustment module receives the topological alteration perturbation potential and calculates the discrete gradient optimization. During the detection phase, five different discrete convergence threshold gradients are set sequentially. The discrete convergence threshold gradients cover the numerical boundary decreasing from 0.01 to 0.00001. The processor computing nodes synchronously monitor the thread lock-up waiting time data during parallel elimination solutions, capturing the efficiency change inflection point caused by thread blocking due to excessively tight residuals. The coordinate point with the minimum sum of data processing latency and the squared sum of boundary synchronization residuals is determined as the solution boundary, thus forming a technical control criterion derived from the data chain. The method of obtaining the squared sum of boundary synchronization residuals is to calculate the difference between the transient boundary deviation value of two adjacent logical nodes in the current control cycle and the given upper limit of convergence tolerance under the current discrete convergence threshold gradient, and then sum the squared differences of all neighboring nodes with causal dependencies. The data processing latency is directly calculated by the processor to determine the hardware clock time consumed by parallel elimination solutions in a single control cycle. The parameter adaptive update loop adds the above two physical quantities with different dimensions after normalization to construct a comprehensive evaluation benchmark, and iterates and compares them in 5 set threshold gradients. The gradient coordinate point with the minimum absolute value of the sum of the two is determined as the most suitable discrete convergence threshold, thereby reducing the thread blocking overhead of the underlying computation while ensuring numerical fidelity.

[0050] Within the discrete state update cycle triggered by assembly transformation, the discrete state event asynchronous scheduling module extracts the motion physical quantity inputs fed back by each component node and allocates multi-threaded computational resource weights through an asymmetric thread pool scheduling mechanism. At this time, the control equation damping coefficient adaptive adjustment module is connected to the computation flow and adjusts the virtual damping coefficient of adjacent constraint keys in real time according to the ratio of the topological deformation disturbance potential to the baseline stable energy. The virtual damping coefficient is increased by stepwise adjustment, so that the oscillation energy generated by the numerical integration truncation when multiple nodes have concurrent physical collisions is adjusted and smoothed in the local variable convergence loop of the corresponding sub-state equation, avoiding the geometric position penetration defect caused by hard-coded hard constraints in the static model. In the continuous deformation virtual debugging test that runs for 45 minutes, the single-step simulation calculation delay of the system is controlled within 3.6ms from 45.2ms, and the residual convergence accuracy of the structural force simulation reaches within 0.0001. The virtual debugging simulation clock and the physical side clock signal data flow path are kept synchronized.

[0051] Example 5: When the system faces a situation where the feature data of the modular robot deviates from the design model due to wear and tear of on-site components, the original pose flow collected by the sensor has an initial state error. If there is no pre-program, the equation calculation of the logic node within the initial step size will drift. To address this, the system uses a calibration program to correct the error before scene setup. This calibration program is applied to the robot body, which consists of motion axes and links. A spatial reference is established using a ranging component with a sampling frequency of 100Hz and a resolution of 0.01mm. The motion axes are controlled to drive the robot components to perform swaying movements in 5 directions under no-load conditions. The ranging component collects the three-dimensional coordinates of the endpoints and calculates the difference between the three-dimensional coordinates and the theoretical values ​​to convert them into a compensation operator, correcting the initial boundary parameters in the global topological sparse tensor.

[0052] After the calibration program outputs the compensation operator, the local state solution module substitutes the operator into the sub-state equation. By multiplying the compensation operator with the inherent rigid body characteristic matrix, the initial load difference caused by noise is canceled out before the first simulation step starts. In this state, the discrete state event asynchronous scheduling module receives the change signal and allocates multi-threaded resources to keep the residual accuracy between the simulated and measured values ​​of the angular velocity under no-load motion axis within 0.0001. As the robot enters the load condition, the single-step update delay of the system is stably maintained within 3.6ms. The virtual debugging drive module receives the instructions output by the dynamic constraint arbitration module and corrects the virtual pose of the unit in the twin scene. The simulation clock signal and the physical clock signal achieve timing alignment.

[0053] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A modular robot digital twin simulation construction and virtual debugging system, characterized in that, include: The topology graph parsing module is used to translate the logical connection relationship of the topological unit of the modular robot into a global topological sparse tensor by acquiring the topological architecture data that represents the topological transformation of the rigid body model data of the modular robot. This abstracts the virtual assembly topology data into a discrete topology graph composed of logical nodes and constraint keys. The local state solution module is connected to the topology graph analysis module. The local state solution module includes multiple parallel discrete solution kernels. The local state solution module is used to extract the diagonal block matrix corresponding to each logical node in the global topology sparse tensor by relying on the block sparsity characteristics in the global topology sparse tensor. The system's overall control equations are decoupled and decomposed into multiple independent sub-state equations mounted on each logical node. Each sub-state equation contains only the intrinsic state parameters of the corresponding logical node and the numerical transmission load of the adjacent constraint keys. The dynamic constraint arbitration module is mounted on the data flow interface of each sub-state equation. The dynamic constraint arbitration module is used to monitor the transient boundary deviation value of two adjacent logical nodes during the simulation operation. When the transient boundary deviation value is greater than the given discrete convergence threshold, it only triggers the local neighborhood solver operator that has a causal dependence on the corresponding connection boundary to start relaxation iteration. The correction matrix parameter obtained by the iteration is reversed to cover the constraint boundary variable in the corresponding sub-state equation, and the state update instruction is output to the virtual debugging driver module to update the twin scene.

2. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The system also includes a discrete state event asynchronous scheduling module. This module is deployed at the input of the local state solution module. It receives change signals generated by concurrent collisions of multiple logical nodes of the modular robot, converts collision conflicts into discrete state requests, extracts the local kinetic energy change rate fed back by each component node, allocates multi-threaded computational resource weights through an asymmetric thread pool scheduling mechanism, and establishes an asynchronous queuing sequence for conflict data. This limits thread blocking caused by the intertwining of solution clues under sudden multi-point contact conditions, and enables the convergence of the underlying state transition timing of the simulation solver under scenarios of multi-configuration component replacement and concurrent load mutation.

3. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The system also includes a control equation damping coefficient adaptive adjustment module; the control equation damping coefficient adaptive adjustment module is connected in the feedback loop of the dynamic constraint arbitration module, and is used to: capture the fluctuation data value generated when the transient boundary deviation value approaches the system stability critical point when the dynamic constraint arbitration module monitors the transient boundary deviation value; adjust the virtual damping coefficient of adjacent constraint keys to adjust the local constraint flexibility, balance the boundary abrupt load by using the adaptive convergence characteristics of the local variables of each sub-state equation, avoid the geometric truncation error caused by the use of hard-coded static model, and maintain the continuity of force transmission data in multi-configuration assembly scenarios.

4. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The dynamic constraint arbitration module includes: real-time monitoring of the motion evolution and changes between adjacent component logic nodes, capturing discrete jump data of boundary feature parameters; triggering adjacent local neighborhood solvers to start iterative relaxation calculations, and reverse-overwriting the constraint boundary variables inside the sub-state equations to balance the boundary abrupt loads.

5. The modular robot digital twin simulation construction and virtual debugging system according to claim 2, characterized in that, The discrete state event asynchronous scheduling module includes: guiding concurrent computation data flow to thread nodes with high response weights through an asymmetric thread pool scheduling mechanism; and establishing an asynchronous queuing sequence for conflicting data to eliminate thread blockages caused by the intertwining of computation threads under sudden multi-point contact conditions.

6. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The topology parsing module includes: converting each independent rigid body data unit in the 3D virtual assembly model of the modular robot into a corresponding logical node; and converting the hinge constraint data and contact surface constraint data between the independent rigid body data units into corresponding constraint keys.

7. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The local state solution module includes: extracting the diagonal block matrix corresponding to each logical node in the global topological sparse tensor; separating the rigid body dynamics matrix belonging to global coupling in the overall system control equations; and importing the separated rigid body dynamics matrix as the block solution boundary of the sub-state equations into the corresponding discrete solution kernel.

8. The modular robot digital twin simulation construction and virtual debugging system according to claim 3, characterized in that, The dynamic constraint arbitration module also includes: comparing the transient boundary deviation value with the given convergence tolerance range in real time; and sending a trigger signal to the control equation damping coefficient adaptive adjustment module to adjust the virtual damping coefficient when the transient boundary deviation value exceeds the upper limit of the convergence tolerance range for three consecutive control cycles.

9. The modular robot digital twin simulation construction and virtual debugging system according to claim 1, characterized in that, The system also includes a virtual debug driver module; The virtual debugging driver module is connected to the data flow interface of the dynamic constraint arbitration module. The virtual debugging driver module is used to receive the correction state command output by the dynamic constraint arbitration module and correct the virtual motion pose of each rigid body data unit in the twin scene in real time, so that the data flow path of the virtual debugging simulation clock and the physical side clock signal is kept synchronized.

Images