CN117332669A - Digital twin mechanical arm design method and system based on graph convolution neural network - Google Patents
Digital twin mechanical arm design method and system based on graph convolution neural network Download PDFInfo
- Publication number
- CN117332669A CN117332669A CN202210800445.2A CN202210800445A CN117332669A CN 117332669 A CN117332669 A CN 117332669A CN 202210800445 A CN202210800445 A CN 202210800445A CN 117332669 A CN117332669 A CN 117332669A
- Authority
- CN
- China
- Prior art keywords
- graph
- matrix
- mechanical arm
- neural network
- digital twin
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 49
- 238000013461 design Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 37
- 239000011159 matrix material Substances 0.000 claims abstract description 156
- 238000005096 rolling process Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 11
- 238000003860 storage Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 abstract description 14
- 238000004519 manufacturing process Methods 0.000 abstract description 12
- 238000013473 artificial intelligence Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a digital twin mechanical arm design method and a system based on a graph rolling neural network, wherein the method comprises the steps of establishing a feature matrix and an adjacent matrix for the mechanical arm, obtaining a learning matrix through the graph rolling neural network, constructing a learning matrix with n multiplied by n, preprocessing the learning matrix to obtain a graph adjacent matrix and obtaining a threshold value; and binarizing elements in the preprocessed graph adjacent matrix based on a threshold value, traversing by adopting depth-first search to obtain a maximum connected graph, and finding a group of mechanical arm structure diagrams with the largest connecting nodes in the binarized graph adjacent matrix as a design graph of the digital twin mechanical arm. The invention aims to add an artificial intelligent design method into the traditional manufacturing design, and intelligently designs the mechanical arm structure with required requirements, so as to be used for intelligently generating the mechanical arm structure in the digital twin world.
Description
Technical Field
The invention relates to an intelligent design technology of a digital twin model, in particular to a digital twin mechanical arm design method and system based on a graph convolution neural network.
Background
From the development of the mechanical manufacturing industry, it has undergone manual manufacturing, taylor manufacturing, highly automated, flexible automated and integrated manufacturing. Automation of the manufacturing industry is a trend of future development, and intelligent design is also a trend of future development of the manufacturing industry. The conventional design requires a lot of manpower and material resources. The intelligent design can effectively reduce the consumption of resources, and a large number of intelligent design schemes meeting the appearance requirements, the physical requirements and the industrial requirements are designed. Artificial intelligence technology is incorporated into many industries, and the advent of artificial intelligence has led to many tasks not using human resources alone. In the traditional mechanical manufacturing field, a three-dimensional modeling mode and a dynamic model are generally adopted to simulate a mechanical structure, so that a large amount of computing resources and a large amount of space model storage are generally required to be consumed, and the time and the labor are consumed. The artificial intelligence technology is added into the traditional mechanical manufacturing field, and the intelligent design and simulation work can be realized by adopting an artificial intelligence model which is more intelligent and consumes less resources. At present, the field of intelligent design of the mechanical arm is blank by artificial intelligence.
Disclosure of Invention
The invention aims to solve the technical problems: aiming at the problems in the prior art, the invention provides a digital twin mechanical arm design method and a digital twin mechanical arm design system based on a graph convolution neural network, which aims at adding an artificial intelligence method into the traditional manufacturing design to intelligently design a mechanical arm structure with required requirements for intelligently generating the mechanical arm structure in the digital twin world.
In order to solve the technical problems, the invention adopts the following technical scheme:
a digital twin mechanical arm design method based on a graph convolution neural network comprises the following steps:
s1, establishing a feature matrix X and an adjacent matrix A for the mechanical arm, wherein the feature matrix X comprises attributes of all rods in the mechanical arm, and the adjacent matrix A comprises connection among all rods in the mechanical arm;
s2, the feature matrix X and the adjacent matrix A are rolled through a graph to obtain a learning matrix Z;
s3, multiplying the learning matrix Z by a transpose Z of the learning matrix Z T Obtaining a matrix with the size of n multiplied by n, normalizing by using a sigmoid function to obtain a normalized learning matrix M 0 ；
S4, normalizing the learning matrix M 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 ；
S5, adjoining matrix M from the preprocessed graph 1 Respectively taking the maximum value of each row to construct a set m i And from set m i Taking the minimum value as a threshold value m for node risk avoidance;
s6, preprocessing the graph adjacent matrix M based on the threshold value M 1 Binarizing the elements in the matrix to obtain a binarized graph adjacent matrix M 2 ；
S7, aiming at the binarized graph adjacent matrix M 2 And traversing by adopting depth-first search to obtain a maximum connected graph, and taking the found maximum connected graph as a design graph R of the digital twin mechanical arm.
Optionally, in step S1, the number of rows of the feature matrix X is the number of rods in the mechanical arm, the number of columns is the number of attributes of the rods, and the number of rows and columns of the adjacent matrix a are both the number of rods.
Optionally, the attribute of each rod in step S1 includes a left end kinematic pair, a type of the rod, a size of the rod, a movement mode, and a right end kinematic pair, where the left end kinematic pair and the right end kinematic pair represent kinematic pairs of each end of the rod respectively; the kinematic pair comprises a fixed connection, an R connection which is connected in a hinged manner by adopting a revolute pair, an S connection which is connected in a hinged manner by adopting a ball pair and a U connection which is connected in a Hooke hinge; the types of the rods comprise straight rods, folding rods and telescopic rods; the dimensions of the rod include a variety of dimensional size classes; the movement modes include active and passive.
Optionally, in step S2, the function expression of the learning matrix Z obtained by using the feature matrix X and the adjacent matrix a through the graph convolution neural network is:
Z＝G(G(X,A),A)
in the above formula, G represents a graph roll-up neural network.
Optionally, the graph roll-up neural network is a graph neural network GCN or a graph annotation force network GAT.
Optionally, the normalized learning matrix M is used in step S4 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 The functional expression of (2) is:
in the above, a ij Representing a graph adjacency matrix M 1 The ith row and jth column element, a ij,0 Representing normalized learning matrix M 0 The ith row and jth column element, a ji,0 Representing normalized learning matrix M 0 The j-th row and i-th column elements of (c).
Optionally, the matrix M is adjoined from the preprocessed graph in step S5 1 Respectively taking the maximum value of each row to construct a set m i And from set m i The function expression taking the minimum value as the threshold value m of node risk avoidance is as follows:
m i ＝max(a i1 ,a i2 ,...,a in )
m＝min(m 1 ,m 2 ,...,m n )
in the above, a i1 ～a in Respectively represent the graph adjacent matrix M 1 The (i) th row 1 to n column element, m 1 ～m n The maximum values of rows 1 to n are shown, respectively.
Optionally, the preprocessed graph adjacency matrix M is based on a threshold M in step S6 1 Binary values of the elements in (a)Obtaining a binarized graph adjacent matrix M 2 The functional expression of (2) is:
in the above formula, x represents the graph adjacency matrix M after pretreatment 1 M represents the result of x binarization.
In addition, the invention also provides a digital twin mechanical arm design system based on the graph rolling neural network, which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the steps of the digital twin mechanical arm design method based on the graph rolling neural network.
Furthermore, the present invention provides a computer readable storage medium having a computer program stored therein, the computer program being for being programmed or configured by a microprocessor to perform the steps of the graph roll-up neural network based digital twin mechanical arm design method.
Compared with the prior art, the invention has the following advantages: the digital twin mechanical arm design method based on the graph rolling neural network comprises the steps of establishing a feature matrix and an adjacent matrix for the mechanical arm, obtaining a learning matrix through the graph rolling neural network, constructing a learning matrix with the size of n multiplied by n, preprocessing the learning matrix to obtain the graph adjacent matrix and obtaining a threshold value; and binarizing elements in the preprocessed graph adjacent matrix based on a threshold value, traversing by adopting depth-first search to obtain a maximum connected graph, and taking the maximum connected graph with the most connection nodes as a design graph of the digital twin mechanical arm. The invention adds an artificial intelligent design method into the traditional manufacturing design, thereby intelligently designing the mechanical arm structure with required requirements, being used for intelligently generating the mechanical arm structure in the digital twin world, being capable of intelligently generating a brand-new mechanical structure diagram according to the existing mechanical structure.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the invention.
Fig. 2 is a diagram illustrating an exemplary structure of a mechanical arm according to an embodiment of the present invention.
Fig. 3 is a topology diagram of a robotic arm in the form of nodes and edges in an embodiment of the invention.
Fig. 4 is a structural example of a kinematic pair in the embodiment of the present invention.
Fig. 5 is a combined morphological structure example of the type and movement pattern of the lever in the embodiment of the present invention.
FIG. 6 is an example of average pooling in an embodiment of the invention.
Fig. 7 is a threshold selection example in an embodiment of the present invention.
Fig. 8 is an example of a graph adjacency matrix before generating a maximum connectivity graph in an embodiment of the invention.
Fig. 9 is a maximum connectivity diagram generated in an embodiment of the present invention.
Detailed Description
As shown in fig. 1, the digital twin mechanical arm design method based on the graph roll-up neural network in this embodiment includes:
s1, providing a data set for a graph convolution neural network: establishing a characteristic matrix X and an adjacent matrix A for the mechanical arm, wherein the characteristic matrix X comprises the attribute of each rod in the mechanical arm, and the adjacent matrix A comprises the connection between each rod in the mechanical arm;
s2, realizing feature learning by adopting a graph neural network: the feature matrix X and the adjacent matrix A are rolled through a graph to obtain a learning matrix Z;
s3, normalizing the learned matrix to conveniently obtain a matrix capable of carrying out probability operation: multiplying the learning matrix Z by the transpose Z of the learning matrix Z T Obtaining a matrix with the size of n multiplied by n, normalizing by using a sigmoid function to obtain a normalized learning matrix M 0 ；
S4, willThe learned matrix is further normalized: the normalized learning matrix M 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 ；
S5, selecting a threshold value, and providing data support for a method for screening out effective values in a matrix: from the preprocessed graph adjacency matrix M 1 Respectively taking the maximum value of each row to construct a set m i And from set m i Taking the minimum value as a threshold value m for node risk avoidance;
s6, screening out effective data in the matrix through a threshold value: graph adjacency matrix M after preprocessing based on threshold M 1 Binarizing the elements in the matrix to obtain a binarized graph adjacent matrix M 2 ；
S7, obtaining a matrix carrying effective values (a binarized graph adjacent matrix M 2 ) Reversely reducing into a graph structure of the mechanical arm: for the binarized graph adjacency matrix M 2 And traversing by adopting depth-first search to obtain a maximum connected graph, and taking the found maximum connected graph as a design graph R of the digital twin mechanical arm.
The digital twin mechanical arm design method based on the graph convolution neural network in the embodiment has the following basic principle: the complex mechanical arm structure is converted into a simple graph structure, the probability of connection between nodes in the graph is learned through a deep learning algorithm (graph neural network), and according to the threshold value selection, whether the connection relationship exists between the parts of the mechanical arm or not is judged, and compared with a traditional exhaustion method, the speed and the usability are improved to a great extent.
FIG. 2 is a diagram illustrating an example structure of a robot arm in the present embodiment, wherein the number identified is the number of a lever (node); fig. 3 shows a structure of a mechanical arm expressed in the form of nodes and edges, wherein the numbers of the nodes are the serial numbers of the nodes, the rod of the mechanical arm is used as the node in the mechanical arm structure, and the connection of the mechanical arm is used as the edge in the mechanical arm structure, so that the complete structure of the mechanical arm is formed. In this embodiment, the number of rows of the feature matrix X in step S1 is the number of rods in the mechanical arm, the number of columns is the number of attributes of the rods, and the number of rows and columns of the adjacent matrix a are both the number of rods.
Specifically, the attributes of the respective levers in step S1 include a left-end kinematic pair, a type of lever, a size of lever, a manner of movement, and a right-end kinematic pair, wherein the left-end kinematic pair and the right-end kinematic pair represent kinematic pairs of the respective ends of the lever, respectively (even if named as an upper-end kinematic pair and a lower-end kinematic pair, they are essentially the same as in the present embodiment); the kinematic pair comprises a fixed connection, an R connection which is hinged by adopting a revolute pair, an S connection which is hinged by adopting a ball pair and a U connection which is hinged by adopting a Hooke hinge, as shown in the figure 4, wherein (a) is the fixed connection, (b) is the R connection, (c) is the S connection, and (d) is the U connection; the types of the rods comprise straight rods, folding rods and telescopic rods; the dimensions of the rod include a variety of dimensional size classes; the movement modes include active and passive. The type of the rod and the movement mode can be combined with each other to form various composite forms, for example, in fig. 5, (a) is a straight rod driving mode, (b) is a straight rod driven mode, (c) is a folding rod driving mode, (d) is a folding rod driven mode, (e) is a telescopic rod driving mode, and (f) is a telescopic rod driven mode. In this embodiment, the attribute codes of the rods are shown in table 1.
Table 1: the properties of the rods encode tables.
In this embodiment, in step S2, the feature matrix X and the adjacent matrix a are rolled through the graph to obtain the learning matrix Z, which is specifically performed by two convolution operations. Experiments prove that the effect of multiple convolution operations is poor, the probability in the matrix is relatively close, the results obtained by two convolution operations have obvious probability distinction, and the effect of generating a mechanical structure is better. In the step S2, the function expression of the learning matrix Z obtained by the characteristic matrix X and the adjacent matrix A through the graph convolution neural network is as follows:
Z＝G(G(X,A),A)
in the above formula, G represents a graph roll-up neural network.
Referring to fig. 1, the graph roll-up neural network is either graph neural network GCN (Graph Convolutional Network) or graph meaning network GAT (Graph Attention Network). By using a graph neural network (GCN, graph Convolutional Network) or a graph attention network (GAT, graph Attention Network), the connection rules of the mechanical arms can be learned, and the mechanical structures can be intelligently generated so as to further optimize the designed mechanical arms according to the constraint between each mechanical structure. It should be noted that, the graph neural network (GCN, graph Convolutional Network) or the graph attention network (GAT, graph Attention Network) is an existing neural network, and the present embodiment only relates to an application of the graph neural network (GCN, graph Convolutional Network) or the graph attention network (GAT, graph Attention Network), and does not relate to an improvement of the graph neural network (GCN, graph Convolutional Network) or the graph attention network (GAT, graph Attention Network).
Multiplying the learning matrix Z by the transpose Z of the learning matrix Z in step S3 T Obtaining a matrix ZZ with the size of n multiplied by n T Normalization is performed by using a sigmoid function, and a normalized learning matrix M0 is obtained and can be expressed as follows:
M 0 ＝sigmoid(ZZ T )
in the above formula, sigmoid represents normalization using a sigmoid function.
In this embodiment, the normalized learning matrix M is used in step S4 0 And carrying out average pooling pretreatment, ensuring that the learned matrixes contain information among each other according to an average pooling thought, normalizing matrix values by a sigmoid method to be probabilities, and carrying out subsequent operation as the probability of connection between mechanical structures learned by deep learning after information fusion of the two. Specifically, the normalized learning matrix M is used in step S4 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 The functional expression of (2) is:
in the above, a ij Representing a graph adjacency matrix M 1 The ith row and jth column element, a ij,0 Representing normalized learning matrix M 0 The ith row and jth column element, a ji,0 Representing normalized learning matrix M 0 The j-th row and i-th column elements of (c). The principle is shown in FIG. 6, wherein the left side of the figure is a normalized learning matrix M 0 Right-hand graph adjacency matrix M 1 The average pooling pretreatment is to normalize the learning matrix M 0 The element on the diagonal is 0, the element addition with symmetrical diagonal is averaged to be used as the graph adjacent matrix M 1 . The output data is changed into a symmetrical matrix by adopting the idea of average pooling, the diagonal is 0, the element values symmetrical with the diagonal are added to obtain an average value, the learned information is more fully utilized after the data fusion, and the information loss is reduced.
For the graph adjacent matrix M after pretreatment 1 And node risk avoidance is carried out, a value with higher probability in the matrix is screened out by adding a threshold value, the representative mechanical structures can be connected, the node in the matrix is screened out and reserved by setting a threshold value for the overall situation, and unnecessary edges and nodes are deleted to ensure that a mechanical arm structure is generated and is called node risk avoidance. And selecting a matrix, putting the maximum value of each row of the generated graph adjacent matrix into a set, taking the minimum value in the set as a node risk avoiding threshold value, ensuring that each row has an effective value, and connecting at least one side of each node with other nodes. In this embodiment, the matrix M is adjoined from the preprocessed graph in step S5 1 Respectively taking the maximum value of each row to construct a set m i And from set m i The function expression taking the minimum value as the threshold value m of node risk avoidance is as follows:
m i ＝max(a i1 ,a i2 ,...,a in )
m＝min(m 1 ,m 2 ,...,m n )
in the above, a i1 ～a in Respectively represent the graph adjacent matrix M 1 The (i) th row 1 to n column element, m 1 ～m n The maximum values of the 1 st to n th rows are shown in fig. 7. The maximum value of each row in the learning matrix is adopted as the set through threshold selection, and then one minimum value is selected from the maximum value set, so that a large number of loss of connection points can be reduced, and node loss with a large area can not occur.
In this embodiment, the preprocessed graph adjacency matrix M is based on the threshold M in step S6 1 Binarizing elements in (a), i.e. the graph-adjacency matrix M after pretreatment 1 Setting the value smaller than the threshold value as 0 and the rest as 1 to obtain a binarized graph adjacent matrix M 2 The functional expression of (2) is:
in the above formula, x represents the graph adjacency matrix M after pretreatment 1 M represents the result of x binarization.
In this embodiment, step S7 is directed to the binarized graph adjacency matrix M 2 The depth-first search is adopted to traverse and obtain the maximum connected graph, and the found maximum connected graph (the connected graph with the most connecting nodes) is used as a design graph R of the digital twin mechanical arm, and can be expressed as follows:
R＝DFS(M)
in the above formula, DFS represents depth-first search, which is an existing well-known search method, and details of implementation thereof are not described herein. For example, FIG. 8 shows a matrix carrying effective values before binarization (a binarized graph adjacent matrix M 2 ) Fig. 9 is a diagram of an example based on the matrix carrying significant values shown in fig. 8 (binarized graph adjacency matrix M 2 ) And generating a maximum connected graph. Finally, defining a connection mode according to an adjacent matrix of the design drawing R of the mechanical arm, constructing an intelligent generation mechanical arm structure in the digital twin world, and displaying the intelligent generation mechanical arm structure.
In summary, in the method of this embodiment, the conversion between the mechanical arm structure and the graph is set regularly, and the rod of the mechanical arm structure is used as a point in the graph; the joints of the mechanical arm structures are used as edges in the drawing to construct transformation rules. And constructing a feature matrix X according to the attribute of the mechanical arm structure rod, and constructing an adjacent matrix A according to the graph. Putting the two into a graph neural network for learning to obtain a learning matrix, performing matrix pretreatment, node risk avoidance and maximum connected graph search on the learning matrix to obtain an intelligently generated mechanical arm structure diagram, wherein the rods of the mechanical arm structure are used as points in the graph; the connection part of the mechanical arm structure is used as an edge in the drawing, a transformation rule is constructed, and an intelligent generation mechanical arm structure is constructed and displayed in the digital twin world. The method of the embodiment can meet the appearance requirement, the physical requirement and the industrial requirement by intelligently designing basic mechanical parts.
In addition, the embodiment also provides a digital twin mechanical arm design system based on the graph rolling neural network, which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the steps of the digital twin mechanical arm design method based on the graph rolling neural network.
Furthermore, the present embodiment also provides a computer readable storage medium having a computer program stored therein, the computer program being configured or programmed by a microprocessor to perform the steps of the aforementioned digital twin mechanical arm design method based on a graph roll-up neural network.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the present invention may occur to one skilled in the art without departing from the principles of the present invention and are intended to be within the scope of the present invention.
Claims (10)
1. A digital twin mechanical arm design method based on a graph convolution neural network is characterized by comprising the following steps:
s1, establishing a feature matrix X and an adjacent matrix A for the mechanical arm, wherein the feature matrix X comprises attributes of all rods in the mechanical arm, and the adjacent matrix A comprises connection among all rods in the mechanical arm;
s2, the feature matrix X and the adjacent matrix A are rolled through a graph to obtain a learning matrix Z;
s3, multiplying the learning matrix Z by a transpose Z of the learning matrix Z T Obtaining a matrix with the size of n multiplied by n, normalizing by using a sigmoid function to obtain a normalized learning matrix M 0 ；
S4, normalizing the learning matrix M 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 ；
S5, adjoining matrix M from the preprocessed graph 1 Respectively taking the maximum value of each row to construct a set m i And from set m i Taking the minimum value as a threshold value m for node risk avoidance;
s6, preprocessing the graph adjacent matrix M based on the threshold value M 1 Binarizing the elements in the matrix to obtain a binarized graph adjacent matrix M 2 ；
S7, aiming at the binarized graph adjacent matrix M 2 And traversing by adopting depth-first search to obtain a maximum connected graph, and taking the found maximum connected graph as a design graph R of the digital twin mechanical arm.
2. The method for designing a digital twin mechanical arm based on a graph roll-up neural network according to claim 1, wherein the number of rows of the feature matrix X in the step S1 is the number of rods in the mechanical arm, the number of columns is the number of attributes of the rods, and the number of rows and the number of columns of the adjacent matrix a are both the number of rods.
3. The method for designing a digital twin mechanical arm based on a graph roll-up neural network according to claim 1, wherein the attribute of each rod in the step S1 includes a left end kinematic pair, a type of the rod, a size of the rod, a movement mode, and a right end kinematic pair, wherein the left end kinematic pair and the right end kinematic pair respectively represent kinematic pairs of each end of the rod; the kinematic pair comprises a fixed connection, an R connection which is connected in a hinged manner by adopting a revolute pair, an S connection which is connected in a hinged manner by adopting a ball pair and a U connection which is connected in a Hooke hinge; the types of the rods comprise straight rods, folding rods and telescopic rods; the dimensions of the rod include a variety of dimensional size classes; the movement modes include active and passive.
4. The digital twin mechanical arm design method based on the graph rolling neural network according to claim 1, wherein in the step S2, a function expression for obtaining a learning matrix Z by passing the feature matrix X and the adjacent matrix a through the graph rolling neural network is:
Z＝G(G(X,A),A)
in the above formula, G represents a graph roll-up neural network.
5. The digital twin mechanical arm design method based on a graph roll-up neural network according to claim 4, wherein the graph roll-up neural network is a graph neural network GCN or a graph annotation force network GAT.
6. The method for designing a digital twin mechanical arm based on a graph roll-up neural network according to claim 4, wherein the normalized learning matrix M is used in step S4 0 The average pooling pretreatment is carried out to obtain a graph adjacency matrix M 1 The functional expression of (2) is:
in the above, a ij Representing a graph adjacency matrix M 1 The ith row and jth column element, a ij,0 Representing normalized learning matrix M 0 The ith row and jth column element, a ji,0 Representing normalized learning matrix M 0 The j-th row and i-th column elements of (c).
7. The method for designing a digital twin mechanical arm based on a graph roll-up neural network according to claim 4, wherein the graph adjacency matrix M after preprocessing is used in step S5 1 Respectively taking the maximum value of each row to construct a set m i And from set m i The function expression taking the minimum value as the threshold value m of node risk avoidance is as follows:
m i ＝max(a i1 ,a i2 ,...,a in )
m＝min(m 1 ,m 2 ,...,m n )
in the above, a i1 ～a in Respectively represent the graph adjacent matrix M 1 The (i) th row 1 to n column element, m 1 ～m n Respectively represent the maximum of the 1 st to n th rowsValues.
8. The digital twin mechanical arm design method based on graph roll-up neural network according to claim 4, wherein the pre-processed graph adjacency matrix M is based on a threshold M in step S6 1 Binarizing the elements in the matrix to obtain a binarized graph adjacent matrix M 2 The functional expression of (2) is:
in the above formula, x represents the graph adjacency matrix M after pretreatment 1 M represents the result of x binarization.
9. A digital twin mechanical arm design system based on a graph roll-up neural network, comprising a microprocessor and a memory connected to each other, characterized in that the microprocessor is programmed or configured to perform the steps of the digital twin mechanical arm design method based on a graph roll-up neural network as claimed in any one of claims 1 to 8.
10. A computer readable storage medium having a computer program stored therein, wherein the computer program is for being programmed or configured by a microprocessor to perform the steps of the graph roll-up neural network based digital twin robot design method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210800445.2A CN117332669A (en) | 2022-07-08 | 2022-07-08 | Digital twin mechanical arm design method and system based on graph convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210800445.2A CN117332669A (en) | 2022-07-08 | 2022-07-08 | Digital twin mechanical arm design method and system based on graph convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117332669A true CN117332669A (en) | 2024-01-02 |
Family
ID=89292139
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210800445.2A Pending CN117332669A (en) | 2022-07-08 | 2022-07-08 | Digital twin mechanical arm design method and system based on graph convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117332669A (en) |
-
2022
- 2022-07-08 CN CN202210800445.2A patent/CN117332669A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Anisotropic convolutional networks for 3d semantic scene completion | |
| JP6504590B2 (en) | System and computer implemented method for semantic segmentation of images and non-transitory computer readable medium | |
| Natali et al. | Graph-based representations of point clouds | |
| CN111709321B (en) | Human behavior recognition method based on graph convolution neural network | |
| CN110181508B (en) | Three-dimensional route planning method and system for underwater robot | |
| CN110297490B (en) | Self-reconstruction planning method of heterogeneous modular robot based on reinforcement learning algorithm | |
| CN105427360B (en) | A kind of controllable CAGE sequence expression algorithm of the error of dynamic grid | |
| CN112308115A (en) | Multi-label image deep learning classification method and equipment | |
| Sudo et al. | Associative memory for online learning in noisy environments using self-organizing incremental neural network | |
| CN106363633A (en) | Robot stable gait planning method and device based on improved particle swarm optimization | |
| Liao | Design of line-type cellular manufacturing systems for minimum operating and material-handling costs | |
| CN109711411B (en) | Image segmentation and identification method based on capsule neurons | |
| Brezocnik et al. | A genetic-based approach to simulation of self-organizing assembly | |
| CN115640842A (en) | Network representation learning method based on graph attention self-encoder | |
| Várady et al. | Improved free-form modelling of scattered data by dynamic neural networks | |
| CN113127697B (en) | Method and system for optimizing graph layout, electronic device and readable storage medium | |
| CN112529010B (en) | Point cloud identification method based on online local feature extraction | |
| CN117332669A (en) | Digital twin mechanical arm design method and system based on graph convolution neural network | |
| CN111428873A (en) | Multi-view attribute network embedding method based on self-encoder | |
| Yue et al. | Study of neural network techniques for computer integrated manufacturing | |
| CN114662204B (en) | Elastic bar system structure system data processing method and device based on graph neural network | |
| Strodthoff et al. | Layered Reeb graphs for three-dimensional manifolds in boundary representation | |
| Allison et al. | Managing variable-dimension structural optimization problems using generative algorithms | |
| CN104504719A (en) | Image edge detection method and equipment | |
| CN113011516A (en) | Three-dimensional mesh model classification method and device based on graph topology and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |