WO2022007753A1 - Digital twin modeling method oriented to mobile robot milling processing - Google Patents

Abstract

A digital twin modeling method oriented to mobile robot milling processing, comprising the steps of: (1) a unit division performed with respect to a mobile robot milling processing system; (2) a multilevel division of a mobile robot processing process oriented to digital twin modeling; (3) a key feature extraction on the basis of the multilevel division; (4) the acquisition of measured data on the basis a physical entity; (5) noise reduction and normalization processing of the data; (6), a multilevel reverse modeling method; and (7) the dynamic reconstruction and matching of a digital twin model. As such, the accuracy in the construction of a digital twin model is increased, updates are done in real-time, the effective merging of a virtual environment and a physical environment is ensured, and a foundation is laid for subsequent processes of predicting, controlling, and optimizing a twin data-driven mobile robot processing process.

Description

A digital twin modeling method for mobile robot milling

This application claims the priority of the Chinese patent application with the application number 202010642150.8 and the application title "A Digital Twin Modeling Method for Mobile Robot Milling", which was filed with the China Patent Office on July 6, 2020, the entire content of which is approved by Reference is incorporated in this application.

technical field

The invention relates to a digital twin modeling method for mobile robot milling and belongs to the technical field of mobile robot application and digital twin.

Background technique

Mobile robot milling is an effective way to solve the problem of high-efficiency and high-precision manufacturing of large and complex components. Large and complex components, such as spacecraft cabins, large wind turbine blades, and high-speed rail car body structural components, have the characteristics of large size, complex structure, weak rigidity, and high processing accuracy requirements. The manufacture of such components has exceeded the processing capacity of existing machine tools. . Based on the perspective of patent bibliometric analysis, the research and development of industrial robots has gradually shifted to mobile robots and multi-robot systems. Aviation Manufacturing Network has included the precision machining of mobile robots in the top 20 trends in foreign defense manufacturing technology in 2019. For example, the mobile robot developed by the Fraunhofer Society in Germany is used for milling of aircraft wings. Existing research shows that mobile processing robots have important research significance and broad application prospects.

With the gradual maturity of model digital expression technologies such as model lightweighting, MBD, and physics-based modeling, the rapid popularization and application of new-generation information and communication technologies such as big data, Internet of Things, and cloud computing, as well as the rapid popularization and application of new-generation information and communication technologies such as machine learning and deep learning The continuous emergence of optimization algorithms has enriched the forms and concepts of digital twins. Digital twins have been used by many famous international companies, such as Siemens, Dassault, GM, NASA, Boeing, etc., in product design, manufacturing and service to ensure the final quality of products. In particular, the combination of manufacturing mode and digital twin can not only solve specific manufacturing closed-loop optimization decisions and changes in product development mode, but also ensure the application of digital twin. For example, Guo Feiyan and others conducted research on the optimization-feedback-improvement loop mechanism of aviation product assembly process driven by digital twin model by means of "integration of virtual and real, and virtual control of real" based on the coordinated transmission of all digital quantities in the manufacturing process. Digital twin technology is very suitable for application in the field of large and complex component mobile robot milling. Real-time monitoring and control are carried out, and twin data-driven system optimization and improvement are realized from the system level. To effectively combine the digital twin technology with the mobile robot processing technology, the first premise to realize the "virtual-real integration and virtual control of the real" of the digital twin is to carry out digital twin modeling for the milling processing of mobile robots.

SUMMARY OF THE INVENTION

The technical solution of the present invention is to overcome the deficiencies of the prior art and provide a digital twin modeling method for mobile robot processing.

The technical solution of the present invention is:

A digital twin modeling method for mobile robot processing, the steps are as follows:

(1) Unit division of the mobile robot milling processing system;

The mobile robot milling processing system is divided into units, which are specifically divided into: laser tracker measurement unit, mobile robot unit, milling actuator unit, visual measurement unit, and large and complex component unit;

(2) Based on the units divided in step (1), a multi-level relationship is divided for the mobile robot milling processing system;

(3) Based on the multi-level relationship divided in step (2), it is determined that key features are to be extracted;

(4) carry out physical entity measurement to the key features determined to be extracted in step (3);

(5) performing noise reduction and normalization processing on the data obtained by the entity measurement in step (4);

(6) carry out reverse modeling of key features according to the processed data in step (5);

(7) Dynamically reconstruct the digital twin model to complete the digital twin modeling for mobile robot processing.

Further, the laser tracker measuring unit is used to construct different coordinate systems, and is realized by a laser tracker;

The vision measuring unit is used to measure the milling plane and target point;

The milling actuator unit is used for the execution of milling processing, and the vision measurement unit is installed on the milling actuator unit;

Mobile robot unit: including an omnidirectional mobile platform and a robot installed on the mobile platform, the milling actuator unit is installed on the mobile robot, and the mobile robot adjusts the spatial pose of the milling actuator unit;

Large-scale complex component unit: refers to the cabin component, the bracket is installed on the periphery of the cabin, and the upper surface of the bracket is the area to be milled.

Further, the multi-level relationship division of the mobile robot milling processing system refers to dividing the existing relationship between the units into a geometric matching layer, a pose alignment layer and a deformation compensation layer;

Geometric matching layer: The geometric matching layer is used to assign the actual accuracy information of each unit to the virtual model, so that the virtual model and the physical entity are mapped one by one at the geometric level; the actual accuracy information includes tolerance type, tolerance value, accuracy grade and surface roughness;

Pose alignment layer: There is actually a relative positional relationship between each unit itself and between the units, and with the milling of the mobile robot, the relative positional relationship changes in real time. The pose alignment layer is based on the coordinate system established by each unit. The relative positional relationship between the two is described by the pose, so that the virtual model and the physical entity are mapped one by one at the pose level;

Deformation compensation layer: Each unit is deformed due to gravity and the generated cutting force during the processing. The deformation compensation layer assigns the physical deformation to the virtual model, so that the virtual model and the physical entity are mapped one by one in the deformation compensation layer.

Further, the step (3) determines to extract key features, specifically:

In the geometric matching layer, for the area to be milled, the upper surface of the bracket of the large and complex component unit is measured by the visual measurement unit, and the plane of the data fitting is obtained as the key feature;

In the pose constraint layer, according to the mutual position relationship of each unit, ≥3 key measurement points are selected in each unit coordinate system, and the pose of the data fitting is obtained as the key feature;

Each unit coordinate system includes the laser tracker unit design coordinate system, the visual measurement unit design coordinate system, the milling actuator unit design coordinate system, the mobile robot unit base design coordinate system, and the large-scale complex component unit design coordinate system;

In the deformation compensation layer, the mobile robot mills the upper surface of the bracket, converts the deformation amount to compensate the key measurement points, and obtains the pose of the data fitting as the key feature; at the same time, the contact between the milling actuator unit and the large and complex component unit is passed. The pose of the mobile robot unit obtained by force calculation is used as the key feature.

Further, described step (4) carries out physical entity measurement to determine the key features to be extracted, specifically:

In the geometric matching layer, the upper surface of the support of the large and complex component unit is measured by the visual measurement unit, and the measured surface data is obtained, and the measured surface data is discrete point cloud data;

In the pose constraint layer, the laser tracker unit is used to measure ≥3 key measurement points selected in each unit coordinate system, and the measured data of the key measurement points are obtained;

In the deformation compensation layer, the offset weights of key measurement points are calculated by mechanical calculation or finite element method, and the above key measurement points are measured by the laser tracker unit to obtain the measured data; the milling execution installed on the mobile robot unit is measured by the six-dimensional force sensor. The force of the device unit is obtained, and the measured data of the contact force is obtained.

Further, the step (6) performs reverse modeling of key features, including reverse modeling of key features of the geometric matching layer, reverse modeling of key features of the pose constraint layer, and reverse modeling of key features of the deformation compensation layer;

The reverse modeling of the key features of the geometric matching layer is carried out, specifically: based on the normalized data, an error curve is used to describe the topography of the upper surface of the bracket. The fluctuation is described as the extension of the error curve along the y direction. The least squares method is used to perform polynomial fitting on the discrete data of the error curve, so as to reconstruct the continuous curve function close to the surface shape of the actual part, and realize the inverse modeling of the upper surface of the bracket, that is, geometric matching. Reverse modeling of layer key features.

Further, perform reverse modeling of key features of the pose constraint layer, specifically: based on the coordinate system of each unit, measure key measurement points through the laser tracker unit, obtain the measured data in the laser tracker coordinate system, and combine the location of the key measurement points. Theoretical data in different coordinate systems are used to solve the pose of each unit in the laser tracker coordinate system, and then determine the mutual position of each unit to realize the inverse modeling of the pose.

Further, solving the pose of each unit in the laser tracker coordinate system specifically includes the following steps:

①Theoretical data and measurement data given the coordinates of the key measurement point set;

②The three-point method is used for rough registration of the point set, and the initial transformation matrix T0 is obtained;

③ Establish the objective function, take the initial transformation matrix T0 as the initial value, use the LM algorithm to iteratively optimize, and obtain the transformation matrix Tk; the LM algorithm refers to the Levenberg-Marquardt algorithm;

(4) Transform the theoretical data of point set coordinates based on the transformation matrix Tk obtained by the LM algorithm to obtain new theoretical coordinate data;

⑤Using the singular value decomposition method to register the measured data and the new coordinate theory data to obtain the accurately registered transformation matrix TSVD;

⑥ Calculate the transformation matrix Tc=TSVD·Tk between the theoretical data of the key measurement point set coordinates and the measurement data, and Tc is the desired pose.

Further, perform reverse modeling of key features of the deformation compensation layer, including reverse modeling of weighted pose and reverse modeling of pose caused by contact force;

Perform weighted pose inverse modeling, including:

①The 3D model of the measured cabin support is imported through finite element analysis, and the relevant parameters are set according to the actual milling environment, and the force and deformation of the cabin support are analyzed to obtain the offset vector of each key measurement point;

② Calculate the distance between the offset position and the original position according to the offset vector of each key measurement point, and then distribute the weight of each key measurement point according to the distance to obtain its weight factor;

3. Introduce a weight factor into the objective function, and use the improved objective function to iteratively solve the pose, so as to realize the inverse modeling of the weighted pose caused by deformation compensation;

Perform inverse modeling of pose caused by contact force, including:

① Determine the mapping relationship between the contact force on the milling actuator unit installed on the mobile robot unit and the robot pose as F=K·X; where F is the 6-dimensional generalized force vector measured by the 6-dimensional force sensor; X is the 6-dimensional generalized deformation vector of the robot pose; K is a 6×6 Cartesian stiffness matrix, and has:

K=J ^-T K _q J ^-1

Among them, J is the robot Jacobian matrix, K _q is the joint stiffness matrix,

(2) Then the 6-dimensional generalized deformation vector X of the robot pose is obtained as:

X=J ^-1 K _q J ^-T F

The robot pose is corrected by X, and then the reverse modeling of the pose caused by the contact force is realized.

Further, perform dynamic reconstruction of the digital twin model, specifically:

①Combined with the division of each unit, import the virtual model of each unit;

②According to the reverse modeling results based on key features, correct the information of the virtual model from the geometric matching layer, the pose constraint layer, and the deformation compensation layer to obtain the digital twin model;

③ With the dynamic changes of the milling process of the mobile robot unit, the digital twin model is continuously updated, thereby realizing the dynamic reconstruction of the digital twin model.

The beneficial effects of the present invention compared with the prior art are:

(1) The present invention improves the accuracy of the digital twin model construction, updates in real time, ensures the effective integration of the virtual environment and the physical environment, and lays a foundation for the subsequent twin data-driven mobile robot processing process prediction, regulation, optimization and other processes .

(2) The present invention combines the actual process of milling with mobile robots, and based on the existing theoretical methods of machining analysis, establishes a multi-dimensional, multi-space-time scale, and multi-physical dynamic virtual model of the physical entity in a digital manner to simulate and describe the physical entity in the real world. The attributes and behaviors in the environment not only lay the foundation for optimal decision-making, but also ensure the application of digital twin technology.

Description of drawings

Fig. 1 is the flow chart of the method of the present invention;

Fig. 2 is the error curve schematic diagram of the present invention;

FIG. 3 is a schematic diagram of the mobile robot processing system of the present invention.

detailed description

The present invention relates to a digital twin modeling method oriented to mobile robot processing, comprising the following steps: (1) cell division of the mobile robot milling processing system; (2) multi-level division of the mobile robot processing process oriented to digital twin modeling; ( 3) Key feature extraction based on multi-level division; (4) Measurement data acquisition based on physical entities; (5) Data noise reduction and normalization processing; (6) Multi-level inverse modeling method; (7) Digital Dynamic reconstruction and matching of twin models. The invention improves the accuracy of the digital twin body model construction, updates in real time, ensures the effective integration of the virtual environment and the physical environment, and lays a foundation for the subsequent twin data-driven mobile robot processing process prediction, regulation, optimization and other processes.

Specifically, as shown in FIG. 1 , the present invention proposes a digital twin modeling method for mobile robot processing. The steps are as follows:

(1) Divide the mobile robot milling processing system into units, as shown in Figure 3;

Laser tracker measurement unit: including the laser tracker body, measurement data processing software and supporting aids, etc., used to construct different coordinate systems;

Vision measurement unit: including cameras, lenses, light sources and other hardware, data processing software and supporting aids, etc., used to measure the milling plane and target points;

Milling actuator unit: including spindle, tool, sensor and supporting auxiliary, etc., used for milling process execution, on which the vision measurement unit is installed;

Mobile robot unit: including omnidirectional mobile platform, robot (installed on the mobile platform), control cabinet and other hardware, numerical control software system and supporting auxiliary, etc., used to install and fix the milling actuator unit, move the robot and adjust the milling actuator unit to adjust the spatial pose;

Large-scale complex component unit: It includes large-scale cabin-type components made of aluminum alloys, and brackets are installed on the periphery of the cabin. The upper surface of the bracket is the area to be milled.

(2) Based on the units divided in step (1), the multi-level relationship division of the mobile robot milling processing system; for digital twin modeling, the multi-level relationship division of the mobile robot milling processing system refers to dividing the unit itself and the unit with the The relationship between units is divided into geometric matching layer, pose alignment layer and deformation compensation layer;

Because there is a pose relationship between each unit, there is also a pose relationship between the mobile robot unit itself; each unit has its own geometric dimensions; the deformation includes the deformation of large and complex component units, the deformation of mobile robot units, and the milling execution unit. Only key feature extraction is performed in the subsequent modeling, ignoring some pose relationships, geometric dimensions, deformations, etc., which simplifies the modeling process.

Geometric matching layer: The design model of each unit in the virtual environment is a theoretical model, but there are geometric errors in the actual unit manufacturing process, and the accuracy information is determined by parameters such as tolerance type, tolerance value, accuracy grade, and surface roughness. The geometric matching layer is used to assign the actual accuracy information of each unit to the virtual model, so that the virtual model and the physical entity are mapped one by one at the geometric level; the actual accuracy information includes tolerance type, tolerance value, accuracy grade and surface roughness;

Pose alignment layer: There is actually a relative positional relationship between each unit itself and between the units, and with the milling of the mobile robot, the relative positional relationship changes in real time. The pose alignment layer is based on the coordinate system established by each unit. The relative positional relationship between the two is described by the pose, so that the virtual model and the physical entity are mapped one by one at the pose level;

Deformation compensation layer: The material properties of each element are composed of material properties, mechanical properties, and heat treatment methods. Each unit is deformed due to gravity and the generated cutting force during the machining process. The deformation compensation layer assigns the physical deformation to the virtual model, so that the virtual model and the physical entity are mapped one by one in the deformation compensation layer.

(3) Based on the multi-level relationship divided in step (2), it is determined that key features are to be extracted;

Considering the lightweight of the model in the virtual environment, reducing the resource occupation and operation time, and locally correcting the key feature information based on the design model, the mapping between the virtual model and the physical entity is realized.

Specifically:

In the geometric matching layer, for the area to be milled, the upper surface of the bracket of the large and complex component unit is measured by the visual measurement unit, and the plane of the data fitting is obtained as the key feature;

In the pose constraint layer, according to the mutual positional relationship of each unit, ≥3 key measurement points are selected in each unit coordinate system. The theoretical coordinates of the measurement points are known, and they have the characteristics of high stiffness and measurability. The coordinate system of each unit includes the laser tracker unit design coordinate system, the visual measurement unit design coordinate system, the milling actuator unit design coordinate system, the mobile robot unit base design coordinate system, and the large complex component unit design coordinate system. Tie;

In the deformation compensation layer, the mobile robot mills the upper surface of the bracket, considering the deformation of the aluminum alloy cabin bracket, which leads to the deviation of the key measurement points, converts the deformation amount to compensate the key measurement points, and obtains the pose of the data fitting as the key feature; At the same time, considering the force on the end of the robot, which leads to the deformation of the weakly rigid robot structure, the pose of the mobile robot unit obtained by the calculation of the contact force between the milling actuator unit and the large and complex component unit is used as the key feature.

(4) Perform physical entity measurement on the key features determined to be extracted in step (3); specifically:

In the geometric matching layer, the upper surface of the support of the large and complex component unit is measured by the visual measurement unit, and the measured surface data is obtained, and the measured surface data is discrete point cloud data;

In the pose constraint layer, the laser tracker unit is used to measure ≥3 key measurement points selected in each unit coordinate system, and the measured data of the key measurement points are obtained;

In the deformation compensation layer, the offset weights of key measurement points are calculated by mechanical calculation or finite element method, and the above key measurement points are measured by the laser tracker unit to obtain the measured data; the milling execution installed on the mobile robot unit is measured by the six-dimensional force sensor. The force of the device unit is obtained, and the measured data of the contact force is obtained.

(5) performing noise reduction and normalization processing on the data obtained by the entity measurement in step (4);

Obtain measurement data of different precision and magnitude through different measurement equipment. In order to facilitate calculation and unify the precision, the measurement data is denoised and normalized. The processing steps include: point cloud alignment, error point elimination, filtering, streamlining, etc. , to correct the measured accuracy information of key features.

(6) carry out reverse modeling of key features according to the processed data in step (5);

Perform reverse modeling of key features, including reverse modeling of key features of geometric matching layer, reverse modeling of key features of pose constraint layer, and reverse modeling of key features of deformation compensation layer;

The reverse modeling of the key features of the geometric matching layer is carried out, specifically: based on the normalized data, an error curve is used to describe the topography of the upper surface of the bracket. The fluctuation is described as the extension of the error curve along the y direction. As shown in Figure 2, the least squares method is used to perform polynomial fitting on the discrete data of the error curve, so as to reconstruct a continuous curve function close to the surface shape of the actual part, and realize the inversion of the upper surface of the bracket. Modeling, that is, the reverse modeling of the key features of the geometric matching layer.

Denote the discrete point cloud data set as {Q _i (x _i , z _i )}, where i=1,2,...,n, use the two norm of the difference between the fitted continuous function f(x) and all discrete point values Count e ² to evaluate the degree of fit, as follows:

If f (x) is a polynomial, i.e., _{f (x) = a 0 +} a 1 x + ... + a m x m, when e ² takes a minimum value, called f (x) is the least squares fit of discrete points clouds polynomial. Denote the squared value of the fitted curve and the two-norm of the discrete points as e, that is,

Since e is the polynomial coefficients _{{a 0, a 1, ...} , a m} polyvalent function, find the minimum value e ² by the partial derivative of a _k, i.e.,

k is a positive integer from 0 to m, the above formula can be expressed in matrix form, and the polynomial order m is set according to the design requirements of key features, the unique solution of the combination of polynomial coefficients can be calculated, and then the geometric shape closest to the real line feature can be obtained.

Carry out the reverse modeling of the key features of the pose constraint layer, specifically: based on the coordinate system of each unit, measure the key measurement points through the laser tracker unit, obtain the measured data under the laser tracker coordinate system, and combine the different coordinate systems where the key measurement points are located. According to the theoretical data below, the pose of each unit in the laser tracker coordinate system is solved, and then the mutual position of each unit is determined to realize the inverse modeling of the pose.

Solve the pose of each unit in the laser tracker coordinate system, which includes the following steps:

①Theoretical data and measurement data given the coordinates of the key measurement point set;

②The three-point method is used for rough registration of the point set, and the initial transformation matrix T0 is obtained;

③ Establish the objective function, take the initial transformation matrix T0 as the initial value, use the LM algorithm to iteratively optimize, and obtain the transformation matrix Tk; the LM algorithm refers to the Levenberg-Marquardt algorithm;

(4) Transform the theoretical data of point set coordinates based on the transformation matrix Tk obtained by the LM algorithm to obtain new theoretical coordinate data;

⑤Using the singular value decomposition method to register the measured data and the new coordinate theory data to obtain the accurately registered transformation matrix TSVD;

⑥ Calculate the transformation matrix Tc=TSVD·Tk between the theoretical data of the key measurement point set coordinates and the measurement data, and Tc is the desired pose.

Perform reverse modeling of key features of the deformation compensation layer, including reverse modeling of weighted pose and reverse modeling of pose caused by contact force;

Considering the deformation of the aluminum alloy cabin bracket, which leads to the deviation of key measurement points, the deformation analysis of the aluminum alloy cabin bracket model is carried out by means of the finite element analysis method. On this basis, an offset vector weight factor is introduced into the theoretical coordinates of each key measurement point. , combined with the measured data of the laser tracker, the weighted pose fitting algorithm is used to solve the pose. The weighted pose inverse modeling is performed, and the steps are as follows:

①Through the 3D model of the measured cabin support imported into the finite element analysis software, set relevant parameters according to the actual milling environment, analyze the force and deformation of the cabin support, and obtain the offset vector of each key measurement point;

② Calculate the distance between the offset position and the original position according to the offset vector of each key measurement point, and then distribute the weight of each key measurement point according to the distance to obtain its weight factor;

3. Introduce a weight factor into the objective function, and use the improved objective function to iteratively solve the pose, so as to realize the inverse modeling of the weighted pose caused by deformation compensation;

Perform inverse modeling of pose caused by contact force, including:

Considering the structural deformation of the robot's end, the force on the end of the robot is measured by a six-dimensional force/torque sensor to obtain the measured contact force data. The amount of pose change caused by the force and deformation of the robot, in the coordinate system of the milling actuator unit, satisfies the following Hooke's law:

F=K·X

Among them, F is the 6-dimensional generalized force vector measured by the 6-dimensional force/torque sensor; X is the 6-dimensional broad deformation vector; K is the 6×6 Cartesian stiffness matrix. Since the mapping relationship between the end stiffness matrix K and the joint stiffness matrix K _{q is}

K=J ^-T K _q J ^-1

Among them, J is the Jacobian matrix of the robot, and the joint stiffness matrix K _{q is} known after calibration. Therefore, the satisfied Hooke's law can be changed to:

X=J ^-1 K _q J ^-T F

The 6-dimensional wide deformation vector X can be solved, and then the robot pose can be corrected, so as to realize the reverse modeling of the pose caused by the deformation compensation.

(7) Dynamically reconstruct the digital twin model to complete the digital twin modeling for mobile robot processing. In the virtual environment, the digital twin model is dynamically reconstructed to support the functions of high-fidelity simulation prediction, system optimization, and decision-making control of the milling process of the mobile robot. The steps are as follows:

①Combined with the division of each unit, import the virtual model of each unit;

②According to the reverse modeling results based on key features, correct the information of the virtual model from the geometric matching layer, the pose constraint layer, and the deformation compensation layer to obtain the digital twin model;

③ With the dynamic changes of the milling process of the mobile robot unit, the digital twin model is continuously updated, thereby realizing the dynamic reconstruction of the digital twin model.

On the basis of considering model lightweight, reducing resource occupation and computing time, the invention establishes a multi-dimensional, multi-space-time scale, and multi-physical dynamic virtual model of the physical entity in a digital manner to simulate and describe the properties of the physical entity in the real environment, Behaviors, etc., lay the foundation for the prediction, regulation, optimization, and decision-making of the twin data-driven mobile robot milling process.

Claims (10)

1. A digital twin modeling method for mobile robot processing is characterized in that the steps are as follows:

   (1) Unit division of the mobile robot milling processing system;

   The mobile robot milling processing system is divided into units, which are specifically divided into: laser tracker measurement unit, mobile robot unit, milling actuator unit, visual measurement unit, and large and complex component unit;

   (2) Based on the units divided in step (1), a multi-level relationship is divided for the mobile robot milling processing system;

   (3) Based on the multi-level relationship divided in step (2), it is determined that key features are to be extracted;

   (4) carry out physical entity measurement to the key features determined to be extracted in step (3);

   (5) performing noise reduction and normalization processing on the data obtained by the entity measurement in step (4);

   (6) carry out reverse modeling of key features according to the processed data in step (5);

   (7) Dynamically reconstruct the digital twin model to complete the digital twin modeling for mobile robot processing.
2. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: the laser tracker measuring unit is used to construct different coordinate systems, and is realized by a laser tracker;

   The vision measuring unit is used to measure the milling plane and target point;

   The milling actuator unit is used for the execution of milling processing, and the vision measurement unit is installed on the milling actuator unit;

   Mobile robot unit: including an omnidirectional mobile platform and a mobile robot installed on the mobile platform, the milling actuator unit is installed on the mobile robot, and the mobile robot adjusts the spatial pose of the milling actuator unit;

   Large-scale complex component unit: refers to the cabin component, the bracket is installed on the periphery of the cabin, and the upper surface of the bracket is the area to be milled.
3. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: the multi-level relationship division of the mobile robot milling processing system means that the existing relationship between units is divided into geometric matching layers , pose alignment layer and deformation compensation layer;

   Geometric matching layer: The geometric matching layer is used to assign the actual accuracy information of each unit to the virtual model, so that the virtual model and the physical entity are mapped one by one at the geometric level; the actual accuracy information includes tolerance type, tolerance value, accuracy grade and surface roughness;

   Pose alignment layer: There is actually a relative positional relationship between each unit itself and the units, and with the milling of the mobile robot, the relative positional relationship changes. The pose alignment layer is based on the coordinate system established by each unit. The relative positional relationship between the two is described by the pose, so that the virtual model and the physical entity are mapped one by one at the pose level;

   Deformation compensation layer: Each unit is deformed due to its own weight and the generated cutting force during the processing. The deformation compensation layer assigns the physical deformation to the virtual model, so that the virtual model and the physical entity are mapped one by one in the deformation compensation layer.
4. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: the step (3) determines to extract key features, specifically:

   In the geometric matching layer, for the area to be milled, the upper surface of the bracket of the large and complex component unit is measured by the visual measurement unit, and the plane of the data fitting is obtained as the key feature;

   In the pose constraint layer, according to the mutual position relationship of each unit, ≥3 key measurement points are selected in each unit coordinate system, and the pose of the data fitting is obtained as the key feature;

   Each unit coordinate system includes the laser tracker unit design coordinate system, the visual measurement unit design coordinate system, the milling actuator unit design coordinate system, the mobile robot unit base design coordinate system, and the large and complex component unit design coordinate system;

   In the deformation compensation layer, the mobile robot mills the upper surface of the bracket, converts the deformation amount to compensate the key measurement points, and obtains the pose of the data fitting as the key feature; at the same time, the contact between the milling actuator unit and the large and complex component unit is passed. The pose of the mobile robot unit obtained by force calculation is used as the key feature.
5. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: the step (4) is to perform physical entity measurement on the key features to be extracted, specifically:

   In the geometric matching layer, the upper surface of the support of the large and complex component unit is measured by the visual measurement unit, and the measured surface data is obtained, and the measured surface data is discrete point cloud data;

   In the pose constraint layer, the laser tracker unit is used to measure ≥3 key measurement points selected in each unit coordinate system, and the measured data of the key measurement points are obtained;

   In the deformation compensation layer, the offset weights of key measurement points are calculated by mechanical calculation or finite element method, and the above key measurement points are measured by the laser tracker unit to obtain the measured data; the milling execution installed on the mobile robot unit is measured by the six-dimensional force sensor. The force of the device unit is obtained, and the measured data of the contact force is obtained.
6. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: in the step (6), reverse modeling of key features is performed, including reverse modeling of geometric matching layer key features, pose Reverse modeling of the key features of the constraint layer and reverse modeling of the key features of the deformation compensation layer;

   The reverse modeling of the key features of the geometric matching layer is carried out, specifically: based on the normalized data, an error curve is used to describe the topography of the upper surface of the bracket. The fluctuation is described as the extension of the error curve along the y direction. The least squares method is used to perform polynomial fitting on the discrete data of the error curve, so as to reconstruct the continuous curve function close to the surface shape of the actual part, and realize the inverse modeling of the upper surface of the bracket, that is, geometric matching. Reverse modeling of layer key features.
7. A digital twin modeling method for mobile robot processing according to claim 6, characterized in that: performing reverse modeling of key features of the pose constraint layer, specifically: based on the coordinate system of each unit, through the laser tracker unit Measure the key measurement points, obtain the measured data in the laser tracker coordinate system, and combine the theoretical data in different coordinate systems where the key measurement points are located to solve the pose of each unit in the laser tracker coordinate system, and then determine the relationship between the units. The mutual position of , realizes the inverse modeling of pose.
8. A digital twin modeling method for mobile robot processing according to claim 7, characterized in that: solving the pose of each unit in the laser tracker coordinate system specifically includes the following steps:

   ①Theoretical data and measurement data given the coordinates of the key measurement point set;

   ②The three-point method is used for rough registration of the point set, and the initial transformation matrix T0 is obtained;

   ③ Establish the objective function, take the initial transformation matrix T0 as the initial value, use the LM algorithm to iteratively optimize, and obtain the transformation matrix Tk;

   (4) Transform the theoretical data of point set coordinates based on the transformation matrix Tk obtained by the LM algorithm to obtain new theoretical coordinate data;

   ⑤Using the singular value decomposition method to register the measured data and the new coordinate theory data to obtain the accurately registered transformation matrix TSVD;

   ⑥ Calculate the transformation matrix Tc=TSVD·Tk between the theoretical data of the key measurement point set coordinates and the measurement data, and Tc is the desired pose.
9. A digital twin modeling method for mobile robot processing according to claim 8, characterized in that: performing reverse modeling of key features of the deformation compensation layer, including weighted pose reverse modeling and contact force-induced pose reverse modeling mold;

   Perform weighted pose inverse modeling, including:

   ① The 3D model of the measured cabin support is imported through finite element analysis, and the relevant parameters are set according to the actual milling environment, and the force and deformation of the cabin support are analyzed to obtain the offset vector of each key measurement point;

   ② Calculate the distance between the offset position and the original position according to the offset vector of each key measurement point, and then distribute the weight of each key measurement point according to the distance to obtain its weight factor;

   3. Introduce a weight factor into the objective function, and use the improved objective function to iteratively solve the pose, so as to realize the inverse modeling of the weighted pose caused by deformation compensation;

   Perform inverse modeling of pose caused by contact force, including:

   ① Determine the mapping relationship between the contact force on the milling actuator unit installed on the mobile robot unit and the robot pose as F=K·X; where F is the 6-dimensional generalized force vector measured by the 6-dimensional force sensor; X is the 6-dimensional generalized deformation vector of the robot pose; K is a 6×6 Cartesian stiffness matrix, and has:

   K=J ^-T K _q J ^-1

   Among them, J is the robot Jacobian matrix, K _q is the joint stiffness matrix,

   ②Then the 6-dimensional generalized deformation vector X of the robot pose is obtained as:

   X=J ^-1 K _q J ^-T F

   The robot pose is corrected by X, and then the reverse modeling of the pose caused by the contact force is realized.
10. A digital twin modeling method for mobile robot processing according to claim 1, characterized in that: performing dynamic reconstruction of the digital twin model, specifically:

    ①Combined with the division of each unit, import the virtual model of each unit;

    ②According to the reverse modeling results of key features, the virtual model is corrected from the geometric matching layer, the pose constraint layer, and the deformation compensation layer to obtain the digital twin model;

    ③ With the dynamic changes of the milling process of the mobile robot unit, the digital twin model is continuously updated, thereby realizing the dynamic reconstruction of the digital twin model.

Images