CN117021118A - Dynamic compensation method for digital twin track error of parallel robot - Google Patents

Abstract

The invention belongs to the technical field of robot running track compensation, and solves the problem of poor virtual-real interaction performance of parallel robots. The method comprises the steps of generating a digital twin model of the parallel robot and constructing a parallel robot track sample data set; training and iterating the digital twin model by a dynamic coordination reinforcement learning method to obtain an optimal motion track under optimal displacement and speed; compensating the speed error between the physical entity of the parallel robot and the optimized digital twin model by a symmetrical error compensation method; and integrating the digital twin frames to realize real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model. The invention is beneficial to improving the reconstruction capability of the digital twin model and the error control performance of the parallel robot, can realize the communication synchronization between the physical entity and the digital twin model and improve the virtual-real interaction rate of the physical entity and the digital twin model.

Description

A dynamic compensation method for trajectory error of digital twins of parallel robots

Technical field

The invention belongs to the technical field of robot running trajectory compensation, and specifically relates to a dynamic compensation method for trajectory errors of parallel robot digital twins.

Background technique

At present, production operations such as intelligent boxing and handling in the field of parallel robots combined with digital twin technology have become a hot topic in intelligent manufacturing. However, for parallel robots, high-precision and high-speed operation are the key. Most digital twin system frameworks cannot meet the high-precision and high-speed requirements of parallel robots.

The current digital twin technology has matured the application of virtual and real synchronous interaction. However, due to the rigidity error and assembly error of the parallel robot body, the displacement of the parallel robot is not synchronized during movement; in digital twin technology, there is a gap between the physical entity and the digital twin model. Communication delay and communication time will cause speed errors in the synchronization of parallel robots, resulting in a decrease in the virtual and real interaction rate between the two.

Contents of the invention

In order to solve at least one of the above technical problems existing in the prior art, the present invention provides a parallel robot digital twin trajectory error dynamic compensation method.

The present invention adopts the following technical solution to realize: a parallel robot digital twin trajectory error dynamic compensation method, which includes the following steps: S1: Construct a virtual model of the parallel robot, perform rod constraints and space constraints on the virtual model, and generate a parallel robot. Digital twin model; S2: Perform dynamic analysis on the digital twin model, collect the motor torque and kinetic energy parameters of the parallel robot, convert them into trajectory displacement and speed parameters, and construct a parallel robot trajectory sample data set; S3: Strengthen through dynamic coordination The learning method performs training iterations on the digital twin model to obtain the optimal motion trajectory under optimal displacement and speed; S4: Compensate the speed error between the physical entity of the parallel robot and the optimized digital twin model through the symmetric error compensation method ; S5: Integrate the digital twin framework to achieve real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model.

Preferably, in step S2, a parallel robot dynamics set is constructed to obtain the trajectory displacement of the parallel robot. and speed/> ; Establish a mathematical model of trajectory displacement and speed of the digital twin model, collect parallel robot trajectory sample data, and collect a total of N sample data point sets as a parallel robot trajectory sample data set.

Preferably, step S3 specifically includes the following steps: S31: Construct a trajectory displacement DQN subnetwork and a velocity DQN subnetwork: an optimal displacement model and an optimal velocity model; S32: Divide the trajectory sample data set into trajectory displacement and velocity sample data. Set, conduct training of DQN subnetworks, introduce dynamic coordination coefficients, change trajectory strategies, and dynamically coordinate the displacement Q values and velocity Q values of the two DQN subnetworks through the improved optimal trajectory strategy; S33: Construction for obtaining optimal motion For the DQN total network of the trajectory, the displacement Q value and the velocity Q value are input for training and iteration, and finally the optimal motion trajectory under the optimal displacement and velocity is obtained.

Preferably, in step S32, the improved optimal strategy The formula is:

In the formula, with/> is the dynamic coordination coefficient under this strategy;/> is the initial trajectory strategy;/> is the displacement error;/> is the speed error;/> Solve the function for the minimum value;/> To execute the initial trajectory strategy/> Q value.

Preferably, in step S32, the digital twin model and the physical entity are Displacement error at time/> for:

In the formula, Indicates that the digital twin model is in/> The position coordinates of the end effector running at the moment; Indicates that the physical entity is in/> The position coordinates of the end effector running at the moment;

Digital twin model and physical entity Time speed error/> for:

In the formula, Represents the mathematical twin model of the parallel robot in/> The velocity of the end effector at the moment, /> Indicates that the physical entity of the parallel robot is in/> The speed of the end effector at the moment.

Preferably, As the output of the optimal displacement model and fed back into the optimal displacement model, the smallest displacement error is finally output/> As the displacement Q value;/> As the output of the optimal speed model and fed back into the optimal speed model, the minimum speed error is finally output. As the speed Q value.

Preferably, in the DQN total network, the minimum trajectory comprehensive error is defined as:

in the formula is the overall dynamic balance coefficient.

Preferably, there is a time delay between the physical entity of the parallel robot and the digital twin model. In order to ensure the consistency of the running speeds of the two, the speed error needs to be compensated based on the concept of speed symmetry, which specifically includes the following steps:

S41: Synchronize the digital twin model operation cycle with the time period during the movement of the physical entity. The digital twin model operation cycle includes 5 links, which are: startup acceleration link, uniform acceleration link, uniform acceleration to uniform deceleration link, and uniform deceleration link. and the final braking deceleration link; divide the physical entity of the parallel robot into the same operating cycle; S42: Make the physical entity perform periodic motion according to the optimal motion trajectory of the digital twin model, and set the state node to correspond to the door-shaped trajectory motion; S43 : The door-shaped motion trajectory is a symmetrical shape, and the symmetrical error compensation method is used to divide the motion with consistent transformation rules within the cycle; the uniform deceleration link and the final braking deceleration link are simplified into a speed symmetric model of the uniform acceleration link and the start-up acceleration link, Taking the center line as the benchmark, the left side is the acceleration trend speed model, and the right side is the deceleration trend speed model. At the same time, the speed strategy of the digital twin model and the physical entity is changed to make the movement trends of the digital twin model and the physical entity consistent; S44: In In the start-up acceleration link, uniform acceleration to uniform deceleration link, and final braking deceleration link, the error compensation is linear error compensation; in the uniform acceleration link and uniform deceleration link, the error compensation is arc error compensation.

Preferably, in step S5, the real-time synchronous interaction system between the parallel robot physical entity and the digital twin model includes: physical entity layer, information interaction layer, error feedback layer, data accumulation layer, virtual twin layer, data processing layer, dynamic coordination layer, Virtual-real interaction module and dynamic coordination reinforcement learning module;

The physical entity layer includes the parallel robot body and communication ports; the information interaction layer is used to control the communication linkage between the running software and the digital twin model; the error feedback layer eliminates synchronization errors in the optimal trajectory displacement and speed of the physical entity layer and the virtual twin layer; Data accumulation establishes database information, saving the training parameters of the DQN network and the compensation parameters during the movement of the parallel robot; the virtual twin layer includes the parallel robot digital twin model, software platform interface, and mobile control system; the dynamic coordination layer controls the current trajectory error and speed error Combined with the optimal strategy in dynamic coordination reinforcement learning for coordination, it is responsible for the feedback coordination between the physical entity layer and the virtual twin layer to form an overall closed-loop interactive system.

Preferably, the process of real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model is:

Start the physical entity layer to make the parallel robot make periodic movements and enter the information interaction layer through the communication interface; start the virtual twin layer to dynamically coordinate the reinforcement learning module for optimal trajectory displacement, and eliminate redundant data information in the data processing layer;

The information interaction layer analyzes the kinematic parameters in the physical entity layer, converts them into digital information and transmits them to the error feedback layer; at the same time, it dynamically coordinates the reinforcement learning module to calculate and accumulate reward values, and transmits the current data to the data accumulation layer and error feedback layer;

The error feedback layer analyzes the synchronization results, achieves trajectory velocity synchronization compensation based on trajectory displacement synchronization, obtains smaller velocity and displacement errors, and achieves synchronization control effects between high-precision digital twin models and object entities;

The historical information in the data accumulation layer is dynamically coordinated with the information in the current error feedback layer, and finally fed back to the physical entity layer and the virtual twin layer in a closed-loop manner to complete the virtual-real interaction process.

Compared with the prior art, the beneficial effects of the present invention are:

This invention builds a virtual model of a parallel robot and generates a digital twin model of the parallel robot; collects the torque and kinetic energy parameters of the parallel robot and converts them into trajectory displacement and speed parameters; and trains the digital twin model through a dynamic coordination reinforcement learning method. Iterate to obtain the minimum displacement error and speed error under the optimal motion trajectory; use the symmetrical error compensation method to compensate for the speed error between the physical entity of the parallel robot and the optimized digital twin model; integrate the digital twin framework to achieve parallel connection Real-time synchronous interaction between the robot's physical entity and the digital twin model. It is conducive to improving the digital twin model reconstruction capability and the error control performance of parallel robots, enabling communication synchronization between physical entities and digital twin models, and improving the virtual and real interaction rate between the two.

Description of the drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the drawings of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a schematic diagram of the overall process of the present invention;

Figure 2 is a flow chart of the implementation of the CD-DQN network of the present invention;

Figure 3 is the implementation flow chart of the DQN network before improvement;

Figure 4 is a comparison diagram of the speed error between the CD-DQN network of the present invention and the DQN network before improvement;

Figure 5 is a schematic diagram of the trajectory displacement and velocity mathematical model of the present invention;

Figure 6 is a schematic diagram of the symmetric error compensation structure of the present invention;

Figure 7 is a structural block diagram of the dynamic virtual and real interaction method of the present invention;

Figure 8 is a structural block diagram of the virtual-real interaction method before improvement;

Figure 9 is a comparison diagram of the interaction synchronization rate between the dynamic virtual and real interaction method of the present invention and the improved virtual and real interaction method.

Detailed ways

The technical solutions in the embodiments of the present invention are clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementations obtained by those of ordinary skill in the art without any creative work fall within the scope of protection of the present invention.

It should be noted that the structures, proportions, sizes, etc. shown in the drawings of this specification are only used to coordinate with the content disclosed in the specification for the understanding and reading of those familiar with this technology, and are not used to limit the conditions for the implementation of the present invention. , so it has no technical substantive significance. Any structural modifications, changes in proportions or adjustments in size shall fall within the scope of what is disclosed in the present invention as long as it does not affect the effects that the present invention can produce and the purposes that can be achieved. To the extent that the technical content can be covered, it should be noted that in this specification, relational terms such as first and second are only used to distinguish one entity from several other entities, and do not necessarily require or imply There is no actual relationship or ordering between these entities.

The present invention provides an embodiment:

As shown in Figure 1, a dynamic compensation method for trajectory error of a parallel robot digital twin includes the following steps:

S1: Construct a virtual model of the parallel robot, perform rod constraints and space constraints on the virtual model, and generate a digital twin model of the parallel robot;

S2: Perform dynamic analysis on the digital twin model, collect the motor torque and kinetic energy parameters of the parallel robot, convert them into trajectory displacement and speed parameters, and collect the parallel robot trajectory sample data set;

S3: Use the dynamic coordination reinforcement learning method to conduct training iterations on the digital twin model to obtain the optimal motion trajectory under optimal displacement and speed;

S4: Compensate the speed error between the physical entity of the parallel robot and the optimized digital twin model through the symmetrical error compensation method;

S5: Integrate the digital twin framework to achieve real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model.

In step S1, a three-dimensional model of the parallel robot is established and divided into modules, including a static platform, a moving platform, driving rods, springs, and screw parts. The main components of the three-dimensional model are assembled to build a virtual model of the parallel robot. Use the physics engine to fix and constrain the parallel robot virtual model, give the parallel robot's work space, set the parallel robot's spatial working range, and establish a parallel robot digital twin model.

The specific steps are: use the Unity3D physics engine to fix and constrain the parallel robot, align the static platform coordinate system in the parallel robot with the actual world coordinate system; fix the static platform components of the parallel robot, and perform spherical hinge constraints on the driving rods. , carry out parallel constraints on the moving platform components of the parallel robot; constrain the structural relationship between the driving rod and the moving platform to complete the fixing and constraining operations; give the moving platform components equal stresses in different directions, and the stresses are parallel to the spatial coordinate system where the static platform is located , given the normal working space of the parallel robot, set a reasonable working range; add the structural mapping relationship of the parallel robot to generate a digital twin model of the parallel robot.

In step S2, the kinematics set constructed from the physical information of the parallel robot is , collect various parameters of the parallel robot and convert them into kinematic information of the parallel robot, through the radius of the static platform/> , moving platform radius/> , active rod length/> , driven rod length/> Obtain the end pose relationship of the parallel robot; obtain the trajectory displacement of the parallel robot through the constructed kinematic model/> with speed ; Use the ML-agent plug-in in Unity3D to make the digital twin model of the parallel robot move in the virtual space, and collect the parallel robot trajectory sample data. A total of 50,000 sample data point sets are collected as the parallel robot trajectory sample data set.

As shown in Figure 5, the digital twin model includes 5 links when running, which is defined as a mathematical model of trajectory displacement and velocity:

Corresponding operation cycle ;

,/> and/> Respectively expressed in the interval/> Momentary acceleration, velocity and displacement;/> is the maximum acceleration during the operating cycle.

definition The interval corresponds to the startup acceleration link, which corresponds to link 1 in Figure 5. Within this interval:

,/> ,

;

definition The interval corresponds to the uniform acceleration link, which corresponds to link 2 in Figure 5. Within this interval:

,/> ,

;

definition The interval corresponds to the period from uniform acceleration to uniform deceleration, which corresponds to link 3 in Figure 5. Within this interval:

,

,

definition The interval corresponds to the uniform deceleration link, which corresponds to link 4 in Figure 5. Within this interval:

,/> ,

definition The interval corresponds to the final braking deceleration link, which corresponds to link 5 in Figure 5. Within this interval:

,/> ,

;

In step S3, the digital twin model is iteratively trained using the sample data set of the parallel robot trajectory, and two DQN subnetworks are constructed for solving the optimal displacement and optimal speed. The sample data set of the trajectory is divided into Sample data set of displacements and velocities. Carry out model training for two DQN sub-networks respectively. During the training process, the trajectory displacement and velocity are dynamically changed, and the displacement and velocity errors in the digital twin model under the optimal trajectory are solved. When the error is minimum, the displacement Q value and velocity Q value are obtained.

Then, continuous trial and error is carried out in the DQN total network. The error states include operating chaos and strange postures of the robot's end effector. Finally, while obtaining the optimal displacement and speed, the optimal door-shaped operating trajectory is obtained. Two DQN sub-networks and a DQN total network form a CD-DQN network. The characteristic is that the two sub-networks are biased towards exploring the possibility of trajectory operation, while the total network reduces exploration and is biased towards the execution activities of the optimal trajectory, combining exploration and execution. The work is carried out in series, paying attention to the coordination relationship between multiple parameters (velocity, displacement).

As shown in Figures 2 to 4, step S3 specifically includes the following steps:

S31: Construct two DQN sub-networks: the optimal displacement model A1 and the optimal speed model B1; define various parameters of the coordinated dynamic network, and the agent is an intelligent execution body used to execute the convergence target strategy.

Specifically, S311: Define two different learning environments respectively: Definition for/> The overall environmental state corresponding to the moment,/> ,/> respectively/> The environmental state of the optimal displacement model and optimal velocity model corresponding to each moment;

S312: Define separately ,/> for/> Actions that can be performed in the optimal displacement model and optimal velocity model at all times;

S313: Define reward mechanism ,/> Indicates the total reward value; in/> Total environmental status at all times/> with action/> ,/> The total reward value obtained is/> ,/> is the displacement reward value, is the speed bonus value;

S314: Digital twin model in The current position of the end effector running at the moment is , the physical entity of the parallel robot is in/> The current position of the end effector running at the moment is ;Digital twin model and physical entity in/> The displacement error at time is:

In the formula, Indicates that the digital twin model is in/> The position coordinates of the end effector running at the moment; Indicates that the physical entity is in/> The position coordinates of the end effector running at the moment;/> represents the displacement error.

Mathematical twin model of parallel robot in The speed of the end effector at time is/> , the physical entity of the parallel robot is in/> Moment of end effector speed/> , the digital twin model and the physical entity are in/> The speed error at time is:

In the formula, Indicates speed error.

S315: The initial trajectory policy function is , formula in/> To execute the initial trajectory strategy/> Q value,/> is the minimum value function.

S32: Introduce a dynamic coordination coefficient and change the trajectory strategy; in the optimal trajectory state, dynamically coordinate the target Q values of the two DQN subnetworks through the improved optimal strategy, and dynamically coordinate the relationship between the two; the initial trajectory is S315 The trajectory strategy moves according to the correct door-shaped trajectory during the running trajectory. The trajectory strategy function changes at the next moment and becomes , begins to emphasize the changes in speed and displacement errors, and the optimal speed and error are minimized.

Optimal strategy in dynamic coordination process The formula is:

In the formula, with/> is the dynamic coordination coefficient under this strategy;/> is the initial trajectory strategy;/> is the displacement error;/> is the speed error;/> To execute the initial trajectory strategy/> Q value.

When the speed error of the digital twin model is greater than 0.03m/s and the displacement error is less than 0.2mm, focus on minimizing the displacement error; when the speed error of the digital twin model is less than 0.03m/s and the displacement error is greater than 0.2mm, focus on minimizing the speed error. ization; the advantage is that kinematic models with larger errors are eliminated and better network evaluation results can be obtained.

If both meet the basic error, the initial trajectory strategy remains unchanged; if the running speed and trajectory error exceed the fixed value, the strategy value is set to 0 and the accumulation and calculation of reward values are not performed to reduce the amount of network calculations.

Training subnetwork: train the CD-DQN network, As the output of the optimal displacement model and fed back into the optimal displacement model, the smallest displacement error is finally output/> As the displacement Q value;/> As the output of the optimal speed model and fed back into the optimal speed model, the minimum speed error is finally output. As the velocity Q value; the original 50,000 sample points are integrated into 15,000 better trajectory samples, and the displacement Q value and velocity Q value are used as the input Q value of the overall DQN network.

S33: Construct a DQN total network for obtaining the optimal motion trajectory, input the displacement Q value and the velocity Q value for training iterations; the DQN total network training process obtains the running trajectory with the smallest error;

Specifically, S331: First perform DQN total network training on the data sample set, offline strategy Complete exploration and learning, and conduct multiple batch sample samplings; define the minimum trajectory comprehensive error as:

, formula in/> is the overall dynamic balance coefficient;

The total reward value is , formula in/> is the minimum trajectory comprehensive error,/> is the maximum value function.

S332: The optimal trajectory obtained from the displacement error and speed error needs to undergo periodic motion when the parallel robot is running; the parallel robot digital twin model will run periodically further divided into/> status nodes, the time interval is 20ms, here/> ,/> Each status node is further divided into 5 links. The number of simplified samples of the subnetwork is 15,000, and 5,000 iterative calculations are started;

S333: Solve the loss function of the DQN total network;

In the formula, Mathematical expectations for empirical replay;/> is the total weight,/> are the weights in the displacement and velocity subnetworks respectively;/> ,/> , To approximate the input Q value function,/> for/> The overall environmental state of the moment,/> For the initial trajectory strategy/> Q-value function.

S334: Carry out training of the overall DQN network. During the weight alternation process, first set the target Q value is fixed, and then the updated total weights in the network are evaluated after every 10 iterations of training/> Give/> , perform preliminary iterations, and train the overall network 5,000 times. The calculation formula of the target Q value is:

In the formula, for execution/> The current optimal Q value of the strategy,/> is the current optimal Q-value function close to the optimal strategy,/> for mathematical expectations.

Specifically, the optimal displacement and optimal speed parameters are obtained in two DQN sub-networks, and the two are substituted into the total network as input Q values to obtain the optimal trajectory of the robot. The displacement sample parameters and velocity sample parameters in the trajectory sample data set obtained in S2 are divided into two DQN subnetworks: the optimal displacement model and the optimal velocity model. The number of samples changes from 50,000 to 15,000. Set the optimal running trajectory as the total DQN network (target Q value), the speed Q value and displacement Q value obtained from the DQN subnetwork are used as the input Q value of the total network; under the optimal strategy, the DQN total network can predict the optimal Q value in real time, and use the optimal Q value to with/> Compare so that the loss function in the iteration process/> Convergence; during the training iteration process, the trajectory sample set data is simplified. The end effector of the robot digital twin model will predict a variety of trajectories during the movement and simplify the trajectory samples; the input will be updated every 10 times of training in the network iteration Q value, enter the Q value to change the weight/> The target Q value is assigned and the target Q value is updated; new input Q values and target Q values are continuously formed during network training iterations to achieve dynamic trajectory error compensation.

The original kinematic parameters are converted into trajectory coordinate parameters through the digital twin model; the ML-agent plug-in in the Unity3D engine is used to adopt the experience playback principle, that is, the agent continuously iterates the interactive data to form a parameter set , right/> The total reward value at the moment/> Accumulation, during training, environment/> Make changes to/> , the minimum comprehensive error is obtained, and the final synchronization mapping obtains the synchronization between the optimal path and the end effector of the robot digital twin model; after speed and trajectory displacement compensation adjustment, the operation in the digital twin model is synchronized with the optimal trajectory, and the minimum comprehensive error is obtained. The smaller the error, the higher the reward value. After synchronization is completed, // ,/> Return to the experience replay for the next training iteration.

After the training is completed, the digital twin model data is updated in real time to obtain virtual and real synchronization of the optimal motion trajectory with the twin model's motion trajectory. Test the quality of network training and compare the speed error of the original DQN and the improved method CD-DQN. As shown in Figure 5, 60 seconds is selected as the motion period and the trajectory is repeatedly executed. The speed error of the CD-DQN method decreases faster with time, which is significantly better than the original DQN method.

As shown in Figure 5 and Figure 6, after completing the dynamic reinforcement learning process, the digital twin model of the parallel robot completes the operation of optimal trajectory displacement and speed, and performs periodic motion in the virtual space. Due to the time delay between the physical entity and the digital twin model, , it is necessary to carry out actual error compensation between the physical entity and the digital twin model of the parallel robot, and propose the concept of speed symmetry to make the movement trend of the digital twin model consistent with the running speed of the physical entity.

Specifically, S41: Synchronize the running cycle of the digital twin model with the time cycle during the movement of the physical entity. The running cycle of the digital twin model is , divide the physical entity of the parallel robot into the same operating cycle;

S42: Make the physical entity move periodically according to the optimal displacement trajectory of the digital twin model, and set the state node to move corresponding to the gate-shaped trajectory;

S43: The door-shaped motion trajectory is a symmetrical shape, and the symmetrical error compensation method is used to divide the motion with consistent transformation rules within the cycle; the uniform deceleration link and the final braking deceleration link are simplified into a speed symmetric model of the uniform acceleration link and the start-up acceleration link. (The speed model of link 4 and link 5 is simplified, link 4 becomes link 2', and link 5 becomes 1'. Link 1' and link 2' are the speed symmetry models of links 1 and 2), based on the center line, The left side is the acceleration trend speed model, and the right side is the deceleration trend speed model. At the same time, the speed strategy in the digital twin model and the physical entity is changed to make the motion trends of the digital twin model and the physical entity consistent;

S44: The specific compensation methods are: ,/> ,/> When , the error compensation is mainly linear error compensation;/> ,/> When , the error compensation is mainly arc error compensation.

As shown in Figures 7 to 9, based on the optimal trajectory executed by the parallel robot trained in step S3 and the feedback error compensation obtained in step S4, the digital twin framework is integrated, and the physical entity layer, virtual twin layer, and information interaction are set. layer, error feedback layer, data processing layer, data accumulation layer, and dynamic coordination layer to realize real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model.

The real-time synchronous interaction system includes: physical entity layer, information interaction layer, error feedback layer, data accumulation layer, virtual twin layer, data processing layer, dynamic coordination layer, virtual and real interaction module and dynamic coordination reinforcement learning module; the physical entity layer includes parallel robots Ontology, communication port; the information interaction layer is used to control the communication linkage between the running software and the digital twin model; the error feedback layer is used to eliminate the synchronization error of the optimal trajectory displacement and speed of the physical entity layer and the virtual twin layer; the data accumulation layer establishes database information , saves the training parameters of the DQN network and the compensation parameters during the movement of the parallel robot; the virtual twin layer includes the parallel robot digital twin model, software platform interface, and mobile control system; the dynamic coordination layer combines dynamic coordination reinforcement learning with the current trajectory error and speed error. It coordinates the optimal strategy in the system and is responsible for the feedback coordination between the physical entity layer and the virtual twin layer to form an overall closed-loop interactive system.

The process of real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model is:

Start the physical entity layer to make the robot make periodic movements and enter the information interaction layer through the communication interface; start the virtual twin layer to dynamically coordinate the reinforcement learning module for optimal trajectory displacement, and eliminate redundant data information in the data processing layer;

The information interaction layer analyzes the dynamic parameters in the physical entity layer, converts them into digital information and transmits them to the error feedback layer; at the same time, it dynamically coordinates the reinforcement learning module to calculate and accumulate reward values, and transmits the current data to the data accumulation layer and error feedback layer;

The error feedback layer analyzes the synchronization results, achieves trajectory velocity synchronization compensation based on trajectory displacement synchronization, obtains smaller velocity and displacement errors, and achieves synchronization control effects between high-precision digital twin models and object entities;

The historical information in the data accumulation layer is dynamically coordinated with the information in the current error feedback layer, and finally fed back to the physical entity layer and the virtual twin layer in a closed-loop manner to complete the virtual-real interaction process.

The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or modifications within the technical scope disclosed in the present invention. All substitutions should be within the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (10)

1. The dynamic compensation method for the digital twin track error of the parallel robot is characterized by comprising the following steps of:

s1: constructing a virtual model of the parallel robot, and performing rod member constraint and space constraint on the virtual model to generate a digital twin model of the parallel robot;

s2: carrying out dynamic analysis on the digital twin model, collecting motor torque and kinetic energy parameters of the parallel robot, converting the motor torque and kinetic energy parameters into track displacement and speed parameters, and constructing a track sample data set of the parallel robot;

s3: training and iterating the digital twin model by a dynamic coordination reinforcement learning method to obtain an optimal motion track under optimal displacement and speed;

s4: compensating the speed error between the physical entity of the parallel robot and the optimized digital twin model by a symmetrical error compensation method;

s5: and integrating the digital twin frames to realize real-time synchronous interaction between the physical entity of the parallel robot and the digital twin model.

2. The method for dynamically compensating digital twin track errors of parallel robot according to claim 1, wherein the method comprises the following steps: in step S2, a parallel robot dynamics set is constructed to obtain the track displacement of the parallel robotAnd speed->The method comprises the steps of carrying out a first treatment on the surface of the And establishing a track displacement and speed mathematical model of the digital twin model, collecting parallel robot track sample data, and collecting N sample data point sets as a parallel robot track sample data set.

3. The method for dynamically compensating digital twin track errors of parallel robot according to claim 2, wherein the method comprises the following steps: the step S3 specifically comprises the following steps:

s31: constructing a track displacement DQN sub-network and a speed DQN sub-network: an optimal displacement model and an optimal velocity model;

s32: dividing a track sample data set into track displacement and speed sample data sets, training the DQN sub-networks, introducing dynamic coordination coefficients, changing track strategies, and dynamically coordinating the displacement Q values and the speed Q values of the two DQN sub-networks through the improved optimal track strategies;

s33: and constructing a DQN total network for acquiring the optimal motion trail, inputting a displacement Q value and a speed Q value, and performing training iteration on the DQN total network to finally obtain the optimal motion trail under the optimal displacement and the speed.

4. A method for dynamically compensating digital twin track error of a parallel robot according to claim 3, wherein: in step S32, the improved optimal strategyThe formula of (2) is:

in the method, in the process of the invention,and->Dynamic coordination coefficients under the strategy; />Is an initial trajectory strategy; />Is displacement error; />Is a speed error; />Solving a function for a minimum value; />To execute the initial trajectory strategy->Is a Q value of (C).

5. The method for dynamically compensating digital twin track errors of parallel robot according to claim 4, wherein the method comprises the following steps: in step S32, the digital twin model and the physical entity are inTime displacement error->The method comprises the following steps:

in the method, in the process of the invention,representing a digital twin model at +.>Position coordinates of the end effector operation at the moment;representing physical entity in->Position coordinates of the end effector operation at the moment;

digital twin model and physical entitySpeed error +.>The method comprises the following steps:

in the method, in the process of the invention,representing that the mathematical twin model of the parallel robot is +.>Speed of end effector at time, +.>Representing that the physical entity of the parallel robot is +.>The speed of the end effector at the moment in time.

6. The method for dynamically compensating digital twin track errors of parallel robot according to claim 5, wherein the method comprises the following steps:as an output of the optimal displacement model and feedback acting on the optimal displacement model, the displacement error is finally output to be the smallest>As a displacement Q value; />As an output of the optimal speed model and feedback is applied to the optimal speed model, the final output is the smallest speed error +.>As the speed Q value.

7. The method for dynamically compensating digital twin track errors of parallel robot according to claim 6, wherein the method comprises the following steps: in the DQN total network, defining the minimum track integrated error as:

in the middle ofIs the overall dynamic balance coefficient.

8. The method for dynamically compensating digital twin track errors of parallel robot according to claim 7, wherein the method comprises the following steps: the parallel robot physical entity and the digital twin model have time delay, and in order to ensure the consistency of the two operation speeds, the speed error is compensated according to the speed symmetry concept, and the method specifically comprises the following steps:

s41: synchronizing the operation period of the digital twin model with the time period in the physical entity movement process, wherein the operation period of the digital twin model comprises 5 links which are sequentially: starting an acceleration link, a uniform acceleration to uniform deceleration link, a uniform deceleration link and a final braking deceleration link; dividing the same operation period by the physical entities of the parallel robot;

s42: enabling the physical entity to perform periodic motion according to the optimal motion trail of the digital twin model, and setting a state node to correspond to the gate-type trail motion;

s43: the gate-type motion trail is in a symmetrical shape, and motion with consistent transformation rule in the period is divided by adopting a symmetrical error compensation method; the uniform deceleration link and the final braking deceleration link are simplified into speed symmetrical models of the uniform acceleration link and the starting acceleration link, the center line is taken as a reference, the left side is an acceleration trend speed model, the right side is a deceleration trend speed model, and meanwhile, the speed strategy of the digital twin model and the physical entity is changed, so that the movement trend of the digital twin model and the physical entity is consistent;

s44: in the starting acceleration link, the uniform acceleration to uniform deceleration link and the final braking deceleration link, the error compensation is linear error compensation; in the uniform acceleration link and the uniform deceleration link, the error compensation is arc error compensation.

9. The method for dynamically compensating digital twin track errors of parallel robot according to claim 8, wherein the method comprises the following steps: in step S5, the real-time synchronous interaction system of the parallel robot physical entity and the digital twin model includes: the system comprises a physical entity layer, an information interaction layer, an error feedback layer, a data accumulation layer, a virtual twin layer, a data processing layer, a dynamic coordination layer, a virtual-real interaction module and a dynamic coordination reinforcement learning module;

the physical entity layer comprises a parallel robot body and a communication port; the information interaction layer is used for controlling communication linkage between the running software and the digital twin model; the error feedback layer is used for eliminating the synchronous error of the optimal track displacement and the optimal track speed of the physical entity layer and the virtual twin layer; data accumulation establishes database information, and saves training parameters of the DQN network and compensation parameters in the parallel robot motion process; the virtual twin layer comprises a parallel robot digital twin model, a software platform interface and a mobile control system; the dynamic coordination layer coordinates the current track error and the speed error in combination with the optimal strategy in dynamic coordination reinforcement learning, and is responsible for feedback coordination of the physical entity layer and the virtual twin layer to form an overall closed-loop interaction system.

10. The method for dynamically compensating digital twin track errors of parallel robot according to claim 9, wherein the method comprises the following steps: the real-time synchronous interaction process of the physical entity of the parallel robot and the digital twin model comprises the following steps:

starting a physical entity layer to enable the parallel robot to do periodic motion, and entering an information interaction layer through a communication interface; starting a virtual twin layer, dynamically coordinating and strengthening the learning module to perform optimal track displacement, and removing redundant data information in a data processing layer;

the information interaction layer analyzes the kinematic parameters in the physical entity layer, converts the kinematic parameters into digital information and transmits the digital information to the error feedback layer; meanwhile, the dynamic coordination reinforcement learning module calculates and accumulates the rewarding value and transmits the current data to the data accumulation layer and the error feedback layer;

the error feedback layer analyzes the synchronous result, achieves synchronous compensation of the track speed on the basis of synchronous track displacement, obtains smaller speed and displacement error, and achieves synchronous control effect of the high-precision digital twin model and the object entity;

and finally, the historical information in the data accumulation layer and the information in the current error feedback layer are dynamically coordinated, and are fed back to the physical entity layer and the virtual twin layer in a closed loop mode to complete the virtual-real interaction process.