A robot assembly process simulation method based on digital twinning
By using digital twins and phase-invariant synchronized-time Petri nets, the problem of insufficient expression of assembly phase changes during robot assembly was solved, achieving high phase consistency and accurate anomaly correction, thereby improving simulation reliability and the consistency of the assembly process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI CHANGSHENGBANG INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2026-05-26
- Publication Date
- 2026-07-24
AI Technical Summary
Existing simulation methods for robot assembly processes are insufficient to represent the physical phase changes between moving and stationary parts during assembly, resulting in inadequate phase consistency and anomaly correction capabilities.
By employing digital twins and phase-invariant synchronized-time Petri nets, assembly phase sequences are generated by extracting assembly features, and a phase-invariant synchronized-time Petri net is constructed to drive robot assembly simulation. When the actual assembly state is inconsistent with the virtual assembly state, phase boundary correction motion is generated.
It improves the physical consistency and process interpretability of the robot assembly process, has high phase consistency and accurate anomaly correction capabilities, reduces the deviation between virtual assembly and real assembly, and improves the credibility of simulation and the accuracy of anomaly localization.
Smart Images
Figure CN122452165A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robot assembly technology, and in particular to a simulation method for robot assembly process based on digital twins. Background Technology
[0002] Existing robot assembly process simulation methods typically construct virtual assembly scenarios based on digital twin models, mapping robots, end effectors, fixtures, moving parts, and stationary parts into a virtual space. They then combine robot trajectory planning, 3D model-driven approaches, sensor feedback, or production cycle time to simulate assembly actions. Some methods also introduce Petri nets to describe the logical relationships between assembly processes, using storage locations, transitions, and tokens to express the sequence of process actions such as loading, clamping, gripping, moving, inserting, and detecting, thereby achieving visualized deduction and synchronized virtual-real display of the assembly process.
[0003] However, existing methods often rely on process completion signals, fixed time intervals, or preset robot trajectories as the basis for simulation progression. This makes it difficult to represent the physical phase changes in the assembly process between moving and stationary parts, from approach, introduction, contact, guidance, advancement to locking. When the actual assembly state differs from the virtual simulation state, existing methods typically only perform position error comparisons, status alarms, or overall replanning. They cannot locate failed assembly phase boundaries, nor can they generate targeted corrective movements based on these boundaries. This results in insufficient phase consistency and anomaly correction capabilities in the robot assembly process simulation.
[0004] Therefore, how to provide a simulation method for robot assembly process based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose a simulation method for robot assembly process based on digital twins. This invention utilizes digital twins and phase invariant synchronized-time Petri nets to achieve phase-driven simulation of robot assembly, which has the advantages of high phase consistency, accurate anomaly correction, and high simulation reliability.
[0006] A robot assembly process simulation method based on digital twin according to an embodiment of the present invention includes the following steps: The assembly datum of moving and stationary parts is read from a pre-built digital twin model of robot assembly, and the assembly feature pairs between moving and stationary parts are extracted. Based on the assembly features, an assembly phase sequence is generated, and phase invariants are constructed. The assembly phase sequence is converted into a phase place, the phase boundary between adjacent assembly phases is converted into a phase boundary transition, the phase invariants are configured as the ignition rules of the phase boundary transition, and a phase invariant synchronization time Petri net is constructed. The current assembly state is generated based on the initial pose of the robot's assembly digital twin model, and then converted into a phase token and written to the corresponding phase library. The robot assembly simulation is driven by the phase invariant of the next assembly phase. When the phase invariant of the next assembly phase is established, the phase boundary transition is triggered, so that the phase token enters the next phase library and a virtual assembly phase is obtained. The system collects execution feedback when the physical robot drives moving parts to perform assembly, forms the real assembly state, and identifies it as the real assembly phase. When the real assembly phase is inconsistent with the virtual assembly phase, it generates phase splitting results, generates phase boundary correction motion, and re-drives the robot assembly simulation until an assembly phase evidence chain is formed.
[0007] Optionally, the generation of the assembly feature pairs specifically includes: In the robot assembly digital twin model, the parts that are driven by the robot end effector and change pose relative to the workstation are identified as moving parts, and the parts that restrict the moving parts from entering the target position during the assembly process are identified as fixed parts. Extract the motion assembly datum that moves with the robot end effector from the moving parts, extract the fixed assembly datum that constrains the motion assembly datum to enter the target position from the fixed parts, and determine the direction vector and datum point of the motion assembly datum and the direction vector and datum point of the fixed assembly datum. Transform the motion assembly datum and the fixed assembly datum to the same coordinate system of the robot assembly digital twin model, and calculate the directional deviation and datum spacing of the motion assembly datum relative to the fixed assembly datum. The assembly allowable directional deviation and assembly allowable spacing are obtained from the assembly process parameters of the robot assembly digital twin model. When the directional deviation is not greater than the assembly allowable directional deviation and the datum spacing is not greater than the assembly allowable spacing, the corresponding moving assembly datum and fixed assembly datum are determined as the assembly feature pair, and the assembly main direction of the assembly feature pair is determined.
[0008] Optionally, the generation of the phase invariant specifically includes: Based on the assembly main direction of the assembly feature pair, calculate the axial approach relationship and lateral deviation relationship of the moving assembly datum relative to the fixed assembly datum, and determine the allowable contact direction, termination positioning position, import margin and constraint range. Based on the axial proximity relationship, lateral deviation relationship, allowable contact direction, termination positioning position, imported allowance and constraint range, generate the assembly phase sequence; Construct phase invariants for each assembly phase in the assembly phase sequence.
[0009] Optionally, the generation of the phase invariant synchronization time Petri net specifically includes: According to the order of the assembly phase sequence, each assembly phase is converted into a corresponding phase repository, which is used to receive phase tokens representing the same assembly phase. Based on the sequential relationship between adjacent assembly phases in the assembly phase sequence, the phase boundary between adjacent assembly phases is converted into a phase boundary transition. Configure the phase invariant of the next assembly phase as the ignition rule for the corresponding phase boundary transition to limit the triggering of the phase boundary transition; A synchronization time constraint is configured for each phase boundary transition, which, together with the ignition rule, serves as the triggering condition for the phase boundary transition. The phase storage, phase boundary transition, ignition rules, and synchronization time constraints are connected into a phase invariant synchronization time Petri net, so that the phase tokens flow in the phase invariant synchronization time Petri net according to the assembly phase sequence.
[0010] Optionally, the phase invariant synchronization time Petri net includes the phase place obtained from the assembly phase transformation, the phase boundary transition obtained from the phase boundary transformation, the ignition rule formed by the phase invariant of the subsequent assembly phase, and the synchronization time constraint used to limit the triggering timing of the phase boundary transition.
[0011] Optionally, the generation of the phase token specifically includes: Based on the initial pose of the robot assembly digital twin model, the motion assembly reference and the fixed assembly reference are placed in the same coordinate system to generate the current assembly state; According to the order of the assembly phase sequence, the current assembly state is matched with the phase invariant of each assembly phase in turn to determine the current assembly phase corresponding to the current assembly state. Convert the current assembly phase into a phase token to indicate the current assembly phase of the robot assembly simulation; In the phase invariant synchronization time Petri net, find the phase location corresponding to the current assembly phase, write the phase token into the corresponding phase location, and use it as the starting state for the phase invariant driven robot assembly simulation of the next assembly phase.
[0012] Optionally, the generation of the virtual assembly phase specifically includes: Based on the phase library where the phase token is located, determine the current assembly phase, and determine the next assembly phase adjacent to the current assembly phase according to the assembly phase sequence; Based on the phase invariant of the next assembly phase, the simulated propulsion action of the robot end effector is generated. The robot end effector drives the moving parts to update their pose in the robot assembly digital twin model, thus obtaining the simulated assembly state. The simulation post-assembly state is matched with the phase invariant of the next assembly phase, and the phase boundary transition is determined by the synchronization time constraint of the corresponding phase boundary transition. After the phase boundary transition is triggered, the phase token is moved from the phase library corresponding to the current assembly phase to the phase library corresponding to the next assembly phase, and the next assembly phase is determined as the virtual assembly phase.
[0013] Optionally, the generation of the assembly phase evidence chain specifically includes: Collect execution feedback when the physical robot drives moving parts to perform assembly, and convert the execution feedback to the same coordinate system as the robot assembly digital twin model to form the real assembly state; The actual assembly state is matched sequentially with the phase invariants of each assembly phase in the assembly phase sequence, and the assembly phase that matches is determined as the actual assembly phase. Compare the actual assembly phase with the virtual assembly phase. When the actual assembly phase and the virtual assembly phase are inconsistent, generate a phase splitting result. Calculate the phase residual of the actual assembly state relative to the next assembly phase based on the phase invariant of the next assembly phase corresponding to the phase boundary to be corrected. The phase boundary correction motion is generated based on the phase residual, and the robot assembly simulation is re-driven using the phase boundary correction motion. After the phase boundary correction motion re-drives the robot assembly simulation, if the real assembly phase is consistent with the virtual assembly phase, the corresponding phase boundary transition and the corresponding assembly phase are written into the assembly phase evidence chain. If the real assembly phase is still inconsistent with the virtual assembly phase, the phase boundary correction motion is generated again based on the phase splitting result.
[0014] Optionally, the phase splitting result uses the real assembly phase as the actual dwell phase and the virtual assembly phase as the simulated arrival phase. Based on the sequential relationship between the actual dwell phase and the simulated arrival phase in the assembly phase sequence, the phase boundaries that are not passed by the real assembly state are determined, and the phase boundaries that are not passed by the real assembly state are determined as the phase boundaries to be corrected.
[0015] The beneficial effects of this invention are: This invention extracts assembly feature pairs between moving and fixed parts from a digital twin model of robot assembly, and generates an assembly phase sequence based on the main assembly direction, axial approach relationship, lateral deviation relationship, permissible contact direction, and termination positioning position. This allows the robot assembly process to progress step-by-step from simple grasping, moving, inserting, and detecting actions to a series of physical phases, including free approach phase, introduction and capture phase, unilateral contact phase, constraint guidance phase, depth advancement phase, and bottom locking phase. Therefore, the digital twin simulation can more accurately represent the actual interaction and evolution process between the moving and fixed assembly datums, improving the physical consistency and interpretability of the robot assembly process simulation.
[0016] This invention converts assembly phases into phase places, phase boundaries into phase boundary transitions, and configures the phase invariants of the subsequent assembly phase as the ignition rules for phase boundary transitions. This ensures that phase tokens only enter the next phase place when the phase invariants of the subsequent assembly phase are true and the synchronization time constraint is satisfied. Compared to assembly Petri nets triggered by process action completion signals or fixed beats, this invention enables robot assembly simulation to flow according to assembly phase boundaries, avoiding the problem of virtual robots only completing actions according to preset trajectories without the moving parts actually entering the corresponding assembly phase, thereby improving the accuracy of virtual assembly phase generation.
[0017] This invention further identifies the actual assembly state formed by the feedback from the physical robot as the actual assembly phase and compares it with the virtual assembly phase. When the two are inconsistent, a phase splitting result is generated, the phase boundary to be corrected is determined, and the phase residual is calculated based on the phase invariants of the subsequent assembly phase. This generates a phase boundary correction motion and re-drives the robot assembly simulation. Thus, the system can locate the asynchrony between the virtual and real systems to a specific assembly phase boundary, rather than simply outputting position errors or status alarms. This enables the robot assembly simulation to have a closed-loop correction capability for phase deviations and forms an assembly phase evidence chain, improving the reliability of assembly anomaly localization, simulation verification, and process optimization. Attached Figure Description
[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a robot assembly process simulation method based on digital twin proposed in this invention; Figure 2 This is a flowchart of the phase Petri net construction process for a robot assembly process simulation method based on digital twin proposed in this invention. Figure 3This is a flowchart illustrating the phase splitting correction process of a robot assembly process simulation method based on digital twins proposed in this invention. Detailed Implementation
[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0020] refer to Figures 1-3 A simulation method for robot assembly process based on digital twins includes the following steps: The assembly datum of moving and stationary parts is read from a pre-built digital twin model of robot assembly, and the assembly feature pairs between moving and stationary parts are extracted. Based on the assembly features, an assembly phase sequence is generated, and phase invariants are constructed. The assembly phase sequence is converted into a phase place, the phase boundary between adjacent assembly phases is converted into a phase boundary transition, the phase invariants are configured as the ignition rules of the phase boundary transition, and a phase invariant synchronization time Petri net is constructed. The current assembly state is generated based on the initial pose of the robot's assembly digital twin model, and then converted into a phase token and written to the corresponding phase library. The robot assembly simulation is driven by the phase invariant of the next assembly phase. When the phase invariant of the next assembly phase is established, the phase boundary transition is triggered, so that the phase token enters the next phase library and a virtual assembly phase is obtained. The system collects execution feedback when the physical robot drives moving parts to perform assembly, forms the real assembly state, and identifies it as the real assembly phase. When the real assembly phase is inconsistent with the virtual assembly phase, it generates phase splitting results, generates phase boundary correction motion, and re-drives the robot assembly simulation until an assembly phase evidence chain is formed.
[0021] In this embodiment, the generation of assembly feature pairs specifically includes: In the robot assembly digital twin model, the parts that are driven by the robot end effector and change pose relative to the workstation are identified as moving parts, and the parts that restrict the moving parts from entering the target position during the assembly process are identified as fixed parts. Extract the motion assembly datum that moves with the robot end effector from the moving parts, extract the fixed assembly datum that constrains the motion assembly datum to enter the target position from the fixed parts, and determine the direction vector and datum point of the motion assembly datum and the direction vector and datum point of the fixed assembly datum. Among them, the direction vector of the axis-type assembly datum is the direction of the central axis, and the direction vector of the planar assembly datum is the direction of the normal to the positioning surface; Transform the motion assembly datum and the fixed assembly datum to the same coordinate system of the robot assembly digital twin model, and calculate the directional deviation and datum spacing of the motion assembly datum relative to the fixed assembly datum. The directional deviation is calculated by reading the directional vectors of the moving assembly datum and the fixed assembly datum, and the angle between the two directional vectors is used as the directional deviation. The datum spacing is calculated by reading the datum points on the moving assembly datum and the datum points on the fixed assembly datum, and the displacement between the two datum points is used as the length of the component of the displacement perpendicular to the fixed assembly datum. The assembly allowable directional deviation and assembly allowable spacing are obtained from the assembly process parameters of the robot assembly digital twin model. When the directional deviation is not greater than the assembly allowable directional deviation and the datum spacing is not greater than the assembly allowable spacing, the corresponding moving assembly datum and fixed assembly datum are determined as the assembly feature pair, and the assembly main direction of the assembly feature pair is determined. The assembly principal direction is obtained by normalizing the direction vector of the fixed assembly datum, and the normalized direction vector is used as the assembly principal direction of the assembly feature pair.
[0022] In this embodiment, the generation of phase invariants specifically includes: Based on the assembly main direction of the assembly feature pair, calculate the axial approach relationship and lateral deviation relationship of the moving assembly datum relative to the fixed assembly datum, and determine the allowable contact direction, termination positioning position, import margin and constraint range. The axial approach relationship is determined along the main assembly direction, judging the sequential relationship between the entry position, constraint range, and target position of the moving assembly datum relative to the fixed assembly datum. The states of the moving assembly datum not reaching the entry position, entering the constraint range, continuing to approach the target position, or reaching the target position are defined as the axial approach relationship. The lateral deviation relationship is calculated in a plane perpendicular to the main assembly direction, determining the deviation direction and deviation distance of the moving assembly datum relative to the fixed assembly datum. The magnitude of the deviation distance relative to the introduced allowance is defined as the lateral deviation relationship. The introduced allowance is obtained from the assembly process parameters of the robot assembly digital twin model. The constraint range is limited by the entry position, target position, and introduced allowance of the fixed assembly datum. The permissible contact direction is determined by the normal direction of the contact surface in the fixed assembly datum used to guide the moving assembly datum into the target position. The termination positioning position is determined by the position where the moving assembly datum moves along the main assembly direction and completes contact with the positioning surface of the fixed assembly datum. Based on the axial proximity relationship, lateral deviation relationship, allowable contact direction, termination positioning position, imported allowance and constraint range, generate the assembly phase sequence; The assembly phase sequence is generated as follows: a free approach phase is generated when the moving assembly datum does not enter the constraint range; an import capture phase is generated when the moving assembly datum enters the constraint range and the deviation distance is not greater than the import margin; a single-sided contact phase is generated when the normal direction of the contact surface between the moving assembly datum and the fixed assembly datum matches only one allowed contact direction; a constraint guidance phase is generated when the moving assembly datum advances along the main assembly direction and the current deviation distance is not greater than the deviation distance of the previous simulation step; a depth advancement phase is generated when the moving assembly datum continues to advance along the main assembly direction and the assembly depth increases; and a bottoming lock phase is generated when the moving assembly datum reaches the termination positioning position. Construct phase invariants for each assembly phase in the assembly phase sequence; The phase invariant is the assembly state judgment condition defined by the assembly features when the corresponding assembly phase is established. The phase invariant of the free approach phase is constructed as the motion assembly datum has not entered the constraint range. The phase invariant of the import capture phase is constructed as the motion assembly datum enters the constraint range and the deviation distance is not greater than the import margin. The phase invariant of the single-sided contact phase is constructed as the normal direction of the contact surface between the motion assembly datum and the fixed assembly datum matches only one allowed contact direction. The phase invariant of the constraint guidance phase is constructed as the motion assembly datum advances along the main assembly direction and the current deviation distance is not greater than the deviation distance of the previous simulation step. The phase invariant of the depth advancement phase is constructed as the motion assembly datum continues to advance along the main assembly direction and the assembly depth increases. The phase invariant of the bottom locking phase is constructed as the motion assembly datum reaches the termination positioning position.
[0023] In this embodiment, the generation of the phase invariant synchronization time Petri net specifically includes: According to the order of the assembly phase sequence, each assembly phase is converted into a corresponding phase repository, which is used to receive phase tokens representing the same assembly phase. Based on the sequential relationship between adjacent assembly phases in the assembly phase sequence, the phase boundary between adjacent assembly phases is converted into a phase boundary transition. The phase boundary transition uses the phase library corresponding to the previous assembly phase as the input library and the phase library corresponding to the next assembly phase as the output library. A phase boundary transition is set between the input library and the output library so that the phase token can only enter the output library from the input library after the phase boundary transition is triggered. Configure the phase invariant of the next assembly phase as the ignition rule for the corresponding phase boundary transition to limit the triggering of the phase boundary transition; The ignition rule configuration involves writing the phase invariant of the next assembly phase into the corresponding phase boundary transition. During the robot assembly simulation, it is determined whether the phase invariant of the next assembly phase is valid. The phase boundary transition is only triggered when the phase invariant of the next assembly phase is valid. A synchronization time constraint is configured for each phase boundary transition, which, together with the ignition rule, serves as the triggering condition for the phase boundary transition. The synchronization time constraint determines the earliest and latest trigger times of phase boundary transitions based on the allowed entry timing between adjacent assembly phases in the assembly phase sequence. When the robot assembly simulation time has not reached the earliest trigger time, the phase boundary transition is prevented from being triggered. When the robot assembly simulation time has exceeded the latest trigger time and the phase invariant of the next assembly phase has not yet been established, the phase token is kept in the phase library corresponding to the previous assembly phase. The phase storage, phase boundary transition, ignition rules, and synchronization time constraints are connected into a phase invariant synchronization time Petri net, so that the phase tokens flow in the phase invariant synchronization time Petri net according to the assembly phase sequence.
[0024] In this embodiment, the phase invariant synchronization time Petri net includes the phase place obtained from the assembly phase transformation, the phase boundary transition obtained from the phase boundary transformation, the ignition rule formed by the phase invariant of the subsequent assembly phase, and the synchronization time constraint used to limit the triggering timing of the phase boundary transition. Among them, the phase library carries the assembly phase in which the robot assembly simulation is located. The phase boundary transition control phase token moves from the previous assembly phase to the next assembly phase. The ignition rule uses whether the phase invariant of the next assembly phase is valid as the triggering basis for the phase boundary transition. The synchronization time constraint restricts the phase boundary transition to be triggered within the allowed entry time corresponding to the assembly phase sequence. This makes the phase invariant synchronization time Petri net different from the assembly Petri net that uses the process action completion signal or fixed beat as the triggering basis. This allows the robot assembly simulation to advance according to the assembly phase evolution process between the moving assembly reference and the fixed assembly reference.
[0025] In this embodiment, the generation of the phase token specifically includes: Based on the initial pose of the robot assembly digital twin model, the motion assembly reference and the fixed assembly reference are placed in the same coordinate system to generate the current assembly state; According to the order of the assembly phase sequence, the current assembly state is matched with the phase invariant of each assembly phase in turn to determine the current assembly phase corresponding to the current assembly state. The matching process first determines whether the current assembly state satisfies the phase invariant of the free approach phase. If it does, the free approach phase is determined as the current assembly phase. If it does not, it continues to determine whether the current assembly state satisfies the phase invariant of the import capture phase. If it does, the import capture phase is determined as the current assembly phase. The process continues to determine the phase invariants of the single-sided contact phase, constraint guidance phase, depth propulsion phase, and bottoming lock phase in sequence until the current assembly phase corresponding to the current assembly state is determined. Convert the current assembly phase into a phase token to indicate the current assembly phase of the robot assembly simulation; In the phase invariant synchronization time Petri net, find the phase location corresponding to the current assembly phase, write the phase token into the corresponding phase location, and use it as the starting state for the phase invariant driven robot assembly simulation of the next assembly phase.
[0026] In this embodiment, the generation of the virtual assembly phase specifically includes: Based on the phase library where the phase token is located, determine the current assembly phase, and determine the next assembly phase adjacent to the current assembly phase according to the assembly phase sequence; Based on the phase invariant of the next assembly phase, the simulated propulsion action of the robot end effector is generated. The robot end effector drives the moving parts to update their pose in the robot assembly digital twin model, thus obtaining the simulated assembly state. The simulated propulsion actions are generated as follows: When the next assembly phase is the introductory capture phase, the robot end effector drives the motion assembly datum to approach the entry position of the fixed assembly datum along the main assembly direction; when the next assembly phase is the single-sided contact phase, the robot end effector drives the motion assembly datum into the constraint range of the fixed assembly datum along the direction of reducing lateral deviation; when the next assembly phase is the constraint guiding phase, the robot end effector drives the motion assembly datum to advance along the main assembly direction within the contact surface defined by the allowed contact direction; when the next assembly phase is the depth advancement phase, the robot end effector keeps the motion assembly datum advancing along the main assembly direction; when the next assembly phase is the bottom locking phase, the robot end effector drives the motion assembly datum to move to the termination positioning position. The simulation post-assembly state is matched with the phase invariant of the next assembly phase, and the phase boundary transition is determined by the synchronization time constraint of the corresponding phase boundary transition. The triggering of phase boundary transitions is determined as follows: when the post-simulation assembly state satisfies the phase invariant of the next assembly phase, and the robot assembly simulation time is between the earliest and latest triggering times of the corresponding phase boundary transition, the corresponding phase boundary transition is triggered. When the post-simulation assembly state does not satisfy the phase invariant of the next assembly phase, or the robot assembly simulation time is not between the earliest and latest triggering times of the corresponding phase boundary transition, the phase token is kept in the phase library corresponding to the current assembly phase, and the robot assembly simulation continues to be driven according to the phase invariant of the next assembly phase. After the phase boundary transition is triggered, the phase token is moved from the phase library corresponding to the current assembly phase to the phase library corresponding to the next assembly phase, and the next assembly phase is determined as the virtual assembly phase.
[0027] In this embodiment, the generation of the assembly phase evidence chain specifically includes: Collect execution feedback when the physical robot drives moving parts to perform assembly, and convert the execution feedback to the same coordinate system as the robot assembly digital twin model to form the real assembly state; The actual assembly state is matched sequentially with the phase invariants of each assembly phase in the assembly phase sequence, and the assembly phase that matches is determined as the actual assembly phase. The matching process determines whether the moving assembly reference satisfies the phase invariants of the free approach phase, the introduction and capture phase, the one-sided contact phase, the constraint and guidance phase, the deep advance phase, and the bottom locking phase relative to the fixed assembly reference based on the actual assembly state. When the phase invariant of any of the aforementioned assembly phases is met, the assembly phase is determined as the actual assembly phase. Compare the actual assembly phase with the virtual assembly phase. When the actual assembly phase and the virtual assembly phase are inconsistent, generate a phase splitting result. Calculate the phase residual of the actual assembly state relative to the next assembly phase based on the phase invariant of the next assembly phase corresponding to the phase boundary to be corrected. The phase residual is calculated as follows: When the subsequent assembly phase is the introductory capture phase, the approach difference required for the moving assembly datum to enter the constraint range of the fixed assembly datum in the actual assembly state is calculated; when the subsequent assembly phase is the single-sided contact phase, the deviation difference of the contact direction relative to the allowable contact direction in the actual assembly state is calculated; when the subsequent assembly phase is the constraint guiding phase, the deviation difference required for the lateral deviation relationship in the actual assembly state to change from an increasing state to a non-increasing state is calculated; when the subsequent assembly phase is the depth advancement phase, the depth difference required for the assembly depth in the actual assembly state to change from a non-increasing state to an increasing state is calculated; when the subsequent assembly phase is the bottom locking phase, the positioning difference required for the moving assembly datum to reach the termination positioning position in the actual assembly state is calculated. The phase boundary correction motion is generated based on the phase residual, and the robot assembly simulation is re-driven using the phase boundary correction motion. The phase boundary correction motion generates an approach correction motion along the main assembly direction based on the approach difference, a contact correction motion tending towards the allowable contact direction based on the deviation difference, a guiding correction motion reducing the lateral deviation relationship based on the deviation difference, a propulsion correction motion along the main assembly direction based on the depth difference, and a locking correction motion towards the termination positioning position based on the positioning difference. The generated correction motion is used as a simulated propulsion action for the robot end effector to drive the moving parts back into the phase boundary to be corrected. After the phase boundary correction motion re-drives the robot assembly simulation, if the real assembly phase is consistent with the virtual assembly phase, the corresponding phase boundary transition and the corresponding assembly phase are written into the assembly phase evidence chain. If the real assembly phase is still inconsistent with the virtual assembly phase, the phase boundary correction motion is generated again based on the phase splitting result.
[0028] In this embodiment, the phase splitting result uses the real assembly phase as the actual dwell phase and the virtual assembly phase as the simulated arrival phase. Based on the sequential relationship between the actual dwell phase and the simulated arrival phase in the assembly phase sequence, the phase boundaries that are not passed by the real assembly state are determined, and the phase boundaries that are not passed by the real assembly state are determined as the phase boundaries to be corrected.
[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a robot hole-shaft insertion station in an electronic component assembly workshop. In this station, an industrial robot drives the shaft-like moving parts into the housing hole, while the fixed parts are positioned by a fixture. During assembly, issues such as hole not being captured, one-sided friction, skewed insertion, and incomplete bottoming can easily occur. Existing simulation methods mainly rely on the robot's preset trajectory and motion completion signals to advance the virtual assembly process. While the robot may have completed the insertion action in the virtual image, in actual assembly, it may still be stuck in a state of one-sided contact or insufficient guidance. This forces subsequent debugging personnel to rely on alarm logs and manual observation to determine the source of the anomaly.
[0030] In this scenario, the outer cylindrical surface of the shaft and the inner cylindrical surface of the housing hole are first read from the robot assembly digital twin model. The outer cylindrical surface of the shaft is determined as the motion assembly datum, and the inner cylindrical surface of the housing hole is determined as the fixed assembly datum. The main assembly direction is determined based on the central axis direction of the fixed assembly datum. Subsequently, based on the axial approach relationship, lateral deviation relationship, allowable contact direction, and termination positioning of the motion assembly datum relative to the fixed assembly datum, a free approach phase, an introduction capture phase, a one-sided contact phase, a constraint guiding phase, a depth propulsion phase, and a bottoming-locking phase are generated, and a phase invariant is constructed for each assembly phase. The assembly phase is converted into a phase place, the phase boundary is converted into a phase boundary transition, and the phase invariant of the subsequent assembly phase is configured as an ignition rule, thereby forming a phase invariant synchronization time Petri net.
[0031] When the robot simulation begins, the initial pose of the digital twin model is converted into the current assembly state. The current assembly state is matched with the phase invariants of each assembly phase to generate a phase token, which is then written to the corresponding phase place. The simulation no longer simply follows the complete insertion trajectory; instead, it generates the simulated propulsion action of the robot's end effector based on the phase invariants of the next assembly phase. For example, when the next assembly phase is the introductory capture phase, the robot's end effector moves the motion assembly reference close to the fixed assembly reference's entry position along the main assembly direction; when the next assembly phase is the constraint guiding phase, the robot's end effector moves the motion assembly reference along the main assembly direction within the contact surface defined by the allowed contact direction. Only when the post-simulation assembly state satisfies the phase invariants of the next assembly phase, and the synchronization time constraint allows triggering, is the phase boundary transition triggered, and the phase token enters the next phase place, forming a virtual assembly phase.
[0032] When a physical robot performs assembly, the robot controller feedback, visual detection results, and changes in end-effector force collectively form the true assembly state, which is then identified as the true assembly phase. Continuous workshop operation records show that, after adopting this invention, the deviation between the virtual assembly phase and the true assembly phase can be located to a specific phase boundary, rather than merely manifesting as pose error or action timeout warnings. In cases where the housing opening is not fully introduced, the true assembly phase remains in the single-sided contact phase, while the virtual assembly phase has entered the depth-advancing phase. The algorithm generates a phase splitting result and determines the phase boundary where the single-sided contact phase enters the constraint-guided phase as the phase boundary to be corrected. Then, based on the phase invariants of the subsequent assembly phase, a phase boundary correction motion is generated, causing the robot's end effector to perform a corrective action that reduces lateral deviation while maintaining the main assembly direction of advancement.
[0033] In assembly logs, phase token transfer records, and manual review records from multiple production shifts, after adopting this invention, assembly anomalies no longer appear only as end-position deviations or abnormal force peaks, but are instead marked as failed phase boundaries. Debugging personnel can directly see which assembly phase the assembly process is currently at, and whether the phase boundary correction movement causes the real assembly phase to follow the virtual assembly phase again. The complete assembly phase evidence chain preserves the phase boundary transitions, virtual assembly phase, real assembly phase, and corrected phase entry results, providing continuous evidence for the simulation advancement, anomaly localization, and correction review of the robot assembly process. This improves the credibility of digital twin simulation, the consistency of the assembly process, and the targeted nature of anomaly correction.
[0034] Table 1 Comparison of Simulation Performance of Assembly Phase Drive
[0035] The data in Table 1 comes from the continuous assembly verification process of the same robot hole-shaft pressing station. All three algorithms use the same robot assembly digital twin model, the same moving and stationary part models, the same physical robot execution feedback, and the same assembly process parameters. The preset trajectory-driven digital twin simulation algorithm uses the robot end-effector trajectory as the basis for virtual process advancement; the time-Petri net clockwise simulation algorithm uses the assembly action clockwise and action completion state as the basis for flow; and the phase-invariant synchronized time-Petri net algorithm uses the assembly phase sequence, phase invariants, and phase boundary transitions as the basis for simulation advancement. As shown in the table, the phase-invariant synchronized time-Petri net algorithm achieves a virtual-real phase consistency determination accuracy of 94.1%, which is 11.5 percentage points higher than the preset trajectory-driven digital twin simulation algorithm and 7.2 percentage points higher than the time-Petri net clockwise simulation algorithm. This indicates that converting both the real and virtual assembly states into assembly phases reduces misjudgments caused by solely relying on pose differences, action completion signals, or clockwise states.
[0036] In terms of anomaly localization at phase boundaries, the phase invariant synchronous time Petri net algorithm achieves an anomaly localization accuracy of 89.4%, significantly higher than the 68.3% of the preset trajectory-driven digital twin simulation algorithm and the 73.5% of the time Petri net beat-driven simulation algorithm. This is because the present invention does not compare the final position deviation only after assembly is complete. Instead, it sets phase boundary transitions between the free approach phase, the introduction and capture phase, the unilateral contact phase, the constraint guidance phase, the depth propulsion phase, and the bottom-locking phase, and configures the phase invariant of the subsequent assembly phase as the ignition rule. When the actual assembly phase is inconsistent with the virtual assembly phase, the phase boundary that has not been traversed by the actual assembly state can be directly identified as the phase boundary to be corrected. Therefore, the anomaly localization object changes from a general robot pose deviation to a specific assembly phase boundary, and the localization result is closer to the actual location where the assembly deviation occurred.
[0037] In terms of average anomaly localization time, the phase invariant synchronization time Petri net algorithm is 9.6s, lower than the preset trajectory-driven digital twin simulation algorithm's 18.7s and the time Petri net beat-driven simulation algorithm's 15.2s. This difference mainly stems from the generation method of the phase splitting results: the preset trajectory-driven digital twin simulation algorithm needs to backtrack the source of deviation throughout the entire trajectory, and the time Petri net beat-driven simulation algorithm needs to find the abnormal link between the action beat and the equipment state. However, this invention, through direct comparison of the real assembly phase and the virtual assembly phase, compresses the anomaly range to the phase boundary between the actual dwell phase and the simulated arrival phase, thereby reducing the backtracking range and manual judgment steps.
[0038] In terms of the pass rate of the corrected simulation, the phase invariant synchronous time Petri net algorithm achieved 87.3%, which is 15.8 percentage points higher than the preset trajectory-driven digital twin simulation algorithm and 11.5 percentage points higher than the time Petri net clockwise simulation algorithm. This improvement comes from the generation mechanism of the phase boundary correction motion: after the phase boundary to be corrected is determined, the algorithm calculates the phase residual based on the phase invariant of the next assembly phase and converts the phase residual into approach correction motion, contact correction motion, guide correction motion, propulsion correction motion, or locking correction motion. This allows the robot's end effector to perform local correction around the failed assembly phase boundary, rather than re-executing the complete trajectory or only adjusting the action clockwise. This method can more effectively handle assembly deviations such as unguided orifices, one-sided contact, insufficient guidance, and incomplete bottoming.
[0039] The table also shows that the single simulation calculation time of the phase invariant synchronization time Petri net algorithm is 56.9 ms, which is higher than the 41.8 ms of the preset trajectory-driven digital twin simulation algorithm and the 47.3 ms of the time Petri net beat-driven simulation algorithm. This increase is because the algorithm needs to continuously perform phase invariant matching, synchronization time constraint judgment, phase boundary transition trigger judgment, and phase residual calculation during the simulation process. Although the single calculation time has increased, the anomaly localization time, the first-time simulation pass rate after correction, and the assembly phase evidence chain integrity rate have all improved, indicating that the computational overhead has been traded for higher simulation consistency, anomaly localization capability, and correction effectiveness.
[0040] In summary, the phase invariant synchronous time Petri net algorithm, through the closed-loop coordination between assembly feature pairs, assembly phase sequences, phase invariants, phase boundary transitions, and phase boundary correction motions, transforms robot assembly process simulation from trajectory-driven or cycle-driven to assembly physical phase-driven. It can more accurately identify the phase deviation between the real assembly state and the virtual assembly state, and generate targeted correction motions after locating the deviation to a specific phase boundary. Therefore, it exhibits better overall performance in terms of virtual-real phase consistency, phase boundary anomaly localization, simulation correction pass rate, and assembly process traceability.
[0041] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A simulation method for robot assembly process based on digital twins, characterized in that, Includes the following steps: The assembly datum of moving and stationary parts is read from a pre-built digital twin model of robot assembly, and the assembly feature pairs between moving and stationary parts are extracted. Based on the assembly features, an assembly phase sequence is generated, and phase invariants are constructed. The assembly phase sequence is converted into a phase place, the phase boundary between adjacent assembly phases is converted into a phase boundary transition, the phase invariants are configured as the ignition rules of the phase boundary transition, and a phase invariant synchronization time Petri net is constructed. The current assembly state is generated based on the initial pose of the robot's assembly digital twin model, and then converted into a phase token and written to the corresponding phase library. The robot assembly simulation is driven by the phase invariant of the next assembly phase. When the phase invariant of the next assembly phase is established, the phase boundary transition is triggered, so that the phase token enters the next phase library and a virtual assembly phase is obtained. The system collects execution feedback when the physical robot drives moving parts to perform assembly, forms the real assembly state, and identifies it as the real assembly phase. When the real assembly phase is inconsistent with the virtual assembly phase, it generates phase splitting results, generates phase boundary correction motion, and re-drives the robot assembly simulation until an assembly phase evidence chain is formed.
2. The simulation method for robot assembly process based on digital twin according to claim 1, characterized in that, The generation of the assembly feature pairs specifically includes: In the robot assembly digital twin model, the parts that are driven by the robot end effector and change pose relative to the workstation are identified as moving parts, and the parts that restrict the moving parts from entering the target position during the assembly process are identified as fixed parts. Extract the motion assembly datum that moves with the robot end effector from the moving parts, extract the fixed assembly datum that constrains the motion assembly datum to enter the target position from the fixed parts, and determine the direction vector and datum point of the motion assembly datum and the direction vector and datum point of the fixed assembly datum. Transform the motion assembly datum and the fixed assembly datum to the same coordinate system of the robot assembly digital twin model, and calculate the directional deviation and datum spacing of the motion assembly datum relative to the fixed assembly datum. The assembly allowable directional deviation and assembly allowable spacing are obtained from the assembly process parameters of the robot assembly digital twin model. When the directional deviation is not greater than the assembly allowable directional deviation and the datum spacing is not greater than the assembly allowable spacing, the corresponding moving assembly datum and fixed assembly datum are determined as the assembly feature pair, and the assembly main direction of the assembly feature pair is determined.
3. The robot assembly process simulation method based on digital twin according to claim 1, characterized in that, The generation of the phase invariant specifically includes: Based on the assembly main direction of the assembly feature pair, calculate the axial approach relationship and lateral deviation relationship of the moving assembly datum relative to the fixed assembly datum, and determine the allowable contact direction, termination positioning position, import margin and constraint range. Based on the axial proximity relationship, lateral deviation relationship, allowable contact direction, termination positioning position, imported allowance and constraint range, generate the assembly phase sequence; Construct phase invariants for each assembly phase in the assembly phase sequence.
4. The robot assembly process simulation method based on digital twin according to claim 1, characterized in that, The generation of the phase invariant synchronization time Petri net specifically includes: According to the order of the assembly phase sequence, each assembly phase is converted into a corresponding phase repository, which is used to receive phase tokens representing the same assembly phase. Based on the sequential relationship between adjacent assembly phases in the assembly phase sequence, the phase boundary between adjacent assembly phases is converted into a phase boundary transition. Configure the phase invariant of the next assembly phase as the ignition rule for the corresponding phase boundary transition to limit the triggering of the phase boundary transition; A synchronization time constraint is configured for each phase boundary transition, which, together with the ignition rule, serves as the triggering condition for the phase boundary transition. The phase storage, phase boundary transition, ignition rules, and synchronization time constraints are connected into a phase invariant synchronization time Petri net, so that the phase tokens flow in the phase invariant synchronization time Petri net according to the assembly phase sequence.
5. The robot assembly process simulation method based on digital twin according to claim 4, characterized in that, The phase invariant synchronization time Petri net includes the phase location obtained from the assembly phase transformation, the phase boundary transition obtained from the phase boundary transformation, the ignition rule formed by the phase invariant of the subsequent assembly phase, and the synchronization time constraint used to limit the triggering timing of the phase boundary transition.
6. The simulation method for robot assembly process based on digital twin according to claim 1, characterized in that, The generation of the phase token specifically includes: Based on the initial pose of the robot assembly digital twin model, the motion assembly reference and the fixed assembly reference are placed in the same coordinate system to generate the current assembly state; According to the order of the assembly phase sequence, the current assembly state is matched with the phase invariant of each assembly phase in turn to determine the current assembly phase corresponding to the current assembly state. Convert the current assembly phase into a phase token to indicate the current assembly phase of the robot assembly simulation; In the phase invariant synchronization time Petri net, find the phase location corresponding to the current assembly phase, write the phase token into the corresponding phase location, and use it as the starting state for the phase invariant driven robot assembly simulation of the next assembly phase.
7. The simulation method for robot assembly process based on digital twin according to claim 1, characterized in that, The generation of the virtual assembly phase specifically includes: Based on the phase library where the phase token is located, determine the current assembly phase, and determine the next assembly phase adjacent to the current assembly phase according to the assembly phase sequence; Based on the phase invariant of the next assembly phase, the simulated propulsion action of the robot end effector is generated. The robot end effector drives the moving parts to update their pose in the robot assembly digital twin model, thus obtaining the simulated assembly state. The simulation post-assembly state is matched with the phase invariant of the next assembly phase, and the phase boundary transition is determined by the synchronization time constraint of the corresponding phase boundary transition. After the phase boundary transition is triggered, the phase token is moved from the phase library corresponding to the current assembly phase to the phase library corresponding to the next assembly phase, and the next assembly phase is determined as the virtual assembly phase.
8. The simulation method for robot assembly process based on digital twin according to claim 1, characterized in that, The generation of the assembly phase evidence chain specifically includes: Collect execution feedback when the physical robot drives moving parts to perform assembly, and convert the execution feedback to the same coordinate system as the robot assembly digital twin model to form the real assembly state; The actual assembly state is matched sequentially with the phase invariants of each assembly phase in the assembly phase sequence, and the assembly phase that matches is determined as the actual assembly phase. Compare the actual assembly phase with the virtual assembly phase. When the actual assembly phase and the virtual assembly phase are inconsistent, generate a phase splitting result. Calculate the phase residual of the actual assembly state relative to the next assembly phase based on the phase invariant of the next assembly phase corresponding to the phase boundary to be corrected. The phase boundary correction motion is generated based on the phase residual, and the robot assembly simulation is re-driven using the phase boundary correction motion. After the phase boundary correction motion re-drives the robot assembly simulation, if the real assembly phase is consistent with the virtual assembly phase, the corresponding phase boundary transition and the corresponding assembly phase are written into the assembly phase evidence chain. If the real assembly phase is still inconsistent with the virtual assembly phase, the phase boundary correction motion is generated again based on the phase splitting result.
9. A simulation method for robot assembly process based on digital twins according to claim 8, characterized in that, The phase splitting result uses the real assembly phase as the actual dwell phase and the virtual assembly phase as the simulated arrival phase. Based on the sequential relationship between the actual dwell phase and the simulated arrival phase in the assembly phase sequence, it determines the phase boundaries that are not passed by the real assembly state and identifies these phase boundaries as phase boundaries to be corrected.