Six-degree-of-freedom three-dimensional constraint collaborative solving method and system based on digital twinning

Abstract

The application discloses a six-degree-of-freedom three-dimensional constraint collaborative solving method and system based on digital twinning, relates to the technical field of digital twinning modeling and simulation, and comprises the following steps: classifying and degree-of-freedom coding an assembly, defining a hyperedge and constructing a constraint hypergraph, and building a constraint solver containing a double partition after completing a geometric lightweight twin. Degree-of-freedom initialization is completed, tree belief propagation iteration solving is adopted, and constraint force distribution and pose correction are calculated. The results are mapped to a spatial field strength, and when the operation enters the constraint tolerance range, the pose is automatically adsorbed, and operation guidance is pushed to the operation end. The application solves the technical problem that it is difficult to efficiently and conflict-free determine the complex position and attitude constraint relationship between components in the multi-component assembly process, achieves the technical effects that the complex constraint scene of the complex assembly body is adapted, the best position of each component is calculated in real time, accurate docking is automatically completed, and intuitive guidance information is provided to the operation end.

Description

Technical Field

[0001] This invention relates to the field of digital twin modeling and simulation technology, and in particular to a six-degree-of-freedom three-dimensional constraint collaborative solution method and system based on digital twins. Background Technology

[0002] Collaborative solving of six-DOF 3D constraints in complex assemblies based on digital twins is crucial for the assembly accuracy, collaborative operation efficiency, and stable operation of high-end equipment, with constraint data processing being a core component. Existing technologies mostly employ traditional 3D constraint solving and conventional data processing methods, relying on ordinary graph modeling and simple iterative algorithms. These methods can achieve basic solutions in closed assembly scenarios with simple structures and single constraints. However, when faced with complex assemblies involving multiple coupled entities and six-DOF linkages in a digital twin environment, traditional methods have significant limitations. Their data processing capabilities are insufficient, making it impossible to accurately characterize and efficiently solve multi-region constraint relationships. This results in incomplete constraint data and inaccurate pose calculations, failing to meet the demands of high-precision assembly collaboration and integrated virtual-physical management driven by digital twins. Summary of the Invention

[0003] This application provides a six-degree-of-freedom three-dimensional constraint collaborative solution method and system based on digital twins, which solves the technical problem of determining the complex position and attitude constraint relationships between components in a high-efficiency and conflict-free manner during the assembly of multiple components.

[0004] The first aspect of this application provides a six-DOF three-dimensional constraint collaborative solution method based on digital twins. The method includes: performing geometric entity classification and motion degree-of-freedom encoding on an assembly; defining hyperedges based on geometric constraint relationships and reconstructing a constraint hypergraph; performing geometrically lightweight twinning on the assembly; constructing a constraint solver containing a degree-of-freedom solution partition and a constraint conflict detection partition based on the constraint hypergraph; performing degree-of-freedom initialization; iteratively updating the degree-of-freedom mask until convergence using tree-like belief propagation; calculating constraint force allocation and pose correction amounts for each hyperedge based on the current degrees of freedom of its two endpoints; wherein constraint conflict detection is interspersed during the iteration process; mapping the constraint force allocation and pose correction amounts to the spatial field strength distribution in the first physical twin layer of the constraint solver; and pushing operation guidance to the operator; wherein when the operation of the operator enters the constraint tolerance range, automatic pose snapping is performed to generate operation guidance information.

[0005] The second aspect of this application provides a six-DOF three-dimensional constraint collaborative solving system based on digital twins. The system includes: a constraint hypergraph reconstruction module, used to perform geometric entity classification and motion degree-of-freedom encoding on the assembly, define hyperedges according to geometric constraint relationships, and reconstruct a constraint hypergraph; a constraint solver construction module, used to perform geometrically lightweight twinning on the assembly, construct a constraint solver containing a degree-of-freedom solving partition and a constraint conflict detection partition based on the constraint hypergraph, perform degree-of-freedom initialization, iteratively update the degree-of-freedom mask using tree-like belief propagation until convergence, and calculate constraint force allocation and pose correction amounts for each hyperedge based on the current degrees of freedom of its two endpoints, wherein constraint conflict detection is interspersed during the iteration process; and an operation guidance execution module, used to map the constraint force allocation and pose correction amounts to the spatial field strength distribution in the first physical twin layer of the constraint solver, and push operation guidance to the operation end, wherein when the operation of the operation end enters the constraint tolerance range, automatic pose snapping is performed to generate operation guidance information.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application classifies the geometric entities of the assembly and constructs a constraint association model. Through lightweight twin modeling and iterative update processing, it obtains the degrees of freedom and constraint data, calculates the constraint force distribution and pose correction information, and optimizes the results of constraint conflict detection. This allows for accurate solution of multi-degree-of-freedom constraints in complex assemblies, making the collaborative solution of assembly constraints in a digital twin environment more accurate and efficient. It achieves the technical effect of adapting to complex constraint scenarios of complex assemblies, calculating the optimal position of each component in real time and automatically completing precise docking, while providing intuitive guidance information to the operator. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins provided in this application embodiment.

[0009] Figure 2 This is a schematic diagram of the structure of the six-degree-of-freedom three-dimensional constraint collaborative solution system based on digital twin provided in the embodiments of this application.

[0010] Explanation of reference numerals in the attached diagram: Constraint hypergraph reconstruction module 1, constraint solver construction module 2, operation guidance execution module 3. Detailed Implementation

[0011] This application provides a six-degree-of-freedom three-dimensional constraint collaborative solution method and system based on digital twins, which solves the technical problem of determining the complex position and attitude constraint relationships between components in a high-efficiency and conflict-free manner during the assembly of multiple components.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, a six-DOF three-dimensional constraint collaborative solution method based on digital twins is provided, wherein the method includes: Perform geometric entity classification and motion degree of freedom encoding on the assembly, define hyperedges according to geometric constraints, and reconstruct it into a constrained hypergraph.

[0015] In this embodiment, the assembly is an industrial product assembled from multiple components, such as an engine or a robotic arm assembly. A hyperedge is an associative unit connecting multiple components and representing the geometric constraints between them, such as the fit or alignment between parts. The constraint hypergraph is a holistic associative model integrating all components and their mutual geometric constraints, intuitively mapping the structure and constraints of the assembly.

[0016] Specifically, the assembly is divided into point, line, surface, and volume nodes according to its geometric type. The motion degrees of freedom of each node are encoded as six-dimensional spinor vectors. Hyperedges connecting multiple nodes are defined based on geometric constraints, and then a constraint hypergraph is constructed. The specific implementation method of this step will be described in detail later.

[0017] The assembly is geometrically lightweight twinned. A constraint solver containing a degree-of-freedom solution partition and a constraint conflict detection partition is constructed based on the constraint hypergraph. By performing degree-of-freedom initialization, the degree-of-freedom mask is iteratively updated until convergence using tree-like belief propagation. Each hyperedge calculates the constraint force allocation and pose correction based on the current degree of freedom of the two end nodes. Constraint conflict detection is interspersed during the iteration process.

[0018] In this embodiment, the constraint solver is a software algorithm engine embedded in the digital twin system, used to automatically calculate the optimal position and orientation of each component in the assembly while satisfying all geometric constraints.

[0019] Optionally, first perform geometric lightweight twinning on the assembly, open the original CAD assembly model, export the triangular mesh data of all geometric entities, extract the 3D coordinates of the vertices and the triangular facet index information of each entity, and at the same time extract the overall hierarchical structure of the assembly, record the relationship between the parent entity and the child entity of each entity, and complete the extraction of the basic data of the original model.

[0020] After extracting the raw data, the dihedral angle thresholding method is used to identify the local feature boundaries of all geometric entities. All edges of each triangular facet are traversed, and the angle between the normal vectors of adjacent triangular faces is calculated. When the angle is greater than a preset threshold of 30 degrees, the edge is determined to be a feature edge. The vertices of all feature edges are collected and marked as non-mergeable protected vertices. Special emphasis is placed on marking vertices of critical geometric boundaries directly related to subsequent constraint solving, such as mating surface boundaries, datum axes, and positioning hole edges, to ensure that constraint-related geometric features are not destroyed during the lightweighting process.

[0021] After feature boundary marking is completed, a vertex clustering algorithm is first executed to perform initial compression of the unprotected region. Specifically, the 3D space is divided into a uniform cubic mesh with a preset side length. For example, the mesh side length can be set to one-hundredth of the diagonal length of the model's bounding box, or between 1 mm and 5 mm depending on the lightweight precision requirements. All unprotected vertices are traversed, and vertices falling within the same cubic mesh are clustered into a new vertex. The coordinates of the new vertex are taken as the arithmetic mean of the coordinates of all original vertices within that mesh. After vertex clustering, an edge folding algorithm is executed for further compression. The folding cost of each unprotected edge is calculated. The folding cost is obtained by weighting the edge length with the average curvature of the two endpoints of the edge, prioritizing the folding of edges with the lowest folding cost. Each time an edge is folded, the two endpoints of the edge are merged into a new vertex, and the vertex indices of all triangles containing that edge are updated. This process is repeated until the total number of triangles is compressed to less than ten percent of the original model. During the compression process, all edges containing protected vertices are always skipped, preserving the complete geometry of the local feature boundaries.

[0022] After mesh compression, a globally unique identifier is generated for each lightweight entity node using the UUID algorithm. A one-to-one correspondence index mapping table between the original CAD entity ID and the lightweight entity's unique identifier is established. At the same time, the extracted assembly hierarchy relationship is bound to the corresponding lightweight entity, and the parent and child entity association information of each entity is saved to ensure the structural consistency between the lightweight model and the original model.

[0023] The processed lightweight triangular mesh data is then imported into the real-time rendering engine, and a unique identifier is bound to each imported entity node according to the index mapping table. The initial world space pose coordinates of each entity are extracted from the original CAD model and assigned to the transformation component of the corresponding lightweight entity node. At the same time, each entity node is given translational and rotational motion properties, thus completing the construction of the geometric lightweight twin model.

[0024] Next, a constraint solver is constructed based on the constraint hypergraph, which includes a first physical twin layer and a second constraint solver layer containing degree-of-freedom solutions and constraint conflict detection partitions. After initializing the degree of freedom of the entity nodes and iteratively updating using tree-like belief propagation, the constraint force allocation and pose correction are obtained. The specific implementation method of this step will be described in detail later.

[0025] The above steps achieve efficient and lightweight assembly geometric modeling, significantly reducing the computational load of the model while fully preserving the key geometric features and assembly structure required for constraint solving, providing a foundation for real-time constraint solving and virtual-real synchronization in the subsequent digital twin environment.

[0026] The constraint force allocation and pose correction amount are mapped to the spatial field intensity distribution in the first physical twin layer of the constraint solver, and operation guidance is pushed to the operator. When the operation of the operator enters the constraint tolerance range, the pose is automatically snapped and operation guidance information is generated.

[0027] In one embodiment of this application, the constraint force allocation and pose correction amount are mapped to the spatial field strength distribution of the first physical twin layer of the constraint solver and synchronized. When the operation of the assembly by the operator enters the constraint tolerance range based on the field strength distribution, the automatic pose adsorption processing is performed according to the field strength distribution to obtain theoretical data. Then, based on the theoretical data, operation guidance is generated and pushed on the operation interface. The specific implementation method of this step will be described in detail later.

[0028] Furthermore, the method provided in this application embodiment includes: The geometric entities in the assembly are divided into multiple types of entity nodes, including points, lines, surfaces, and volumes. The multiple types of entity nodes are traversed, and the degrees of freedom of the entity motion are encoded as six-dimensional spinor vectors, where the degrees of freedom are defined by three translational degrees of freedom and three rotational degrees of freedom. Hyperedges are defined based on geometric constraints, where the hyperedge types include concurrent, collinear, coplanar, parallel, perpendicular, fixed, and distance-based. Each hyperedge can connect multiple entity nodes. A constraint hypergraph is constructed based on the multiple types of entity nodes, the six-dimensional degree-of-freedom vectors, and the hyperedges.

[0029] Specifically, in 3D constraint modeling technology, the geometric entities of the assembly are usually simply divided, and ordinary graph structures are often used to carry the constraint relationships between entities. Ordinary graphs can only realize the association of two entities, which is difficult to adapt to complex constraint scenarios with multiple entities coupled. Moreover, the motion state of entities is often described in a low-dimensional simplification, which cannot fully represent the six degrees of freedom motion characteristics. This results in insufficient comprehensiveness and accuracy of constraint modeling, making it difficult to support the needs of high-precision constraint solving.

[0030] Therefore, this step first involves traversing and identifying all geometric entities in the assembly. Using a geometric dimension attribute classification method, based on the spatial geometric morphology of the entities, they are divided into four basic entity nodes: points, lines, surfaces, and volumes. Point features (such as vertices and corners) are classified as point nodes; straight edges and axes are classified as line nodes; planes and curved surfaces are classified as surface nodes; and solid blocks are classified as volume nodes. By clarifying the geometric attribute boundaries of each type of node, the standardized classification of entity nodes is completed, laying the foundation for subsequent constraint relationship analysis.

[0031] After classifying the entity nodes, a screw motion encoding method is used to traverse each of the classified point, line, surface, and volume entity nodes one by one. The spatial motion state of each entity is decomposed into three mutually orthogonal translational motion dimensions and three mutually orthogonal rotational motion dimensions. Then, the state parameters of the six motion dimensions are integrated and encoded to generate the corresponding six-dimensional screw vector. Specifically, the three translational degrees of freedom are encoded as vectors (dx, dy, dz), and the three rotational degrees of freedom are encoded as vectors (rx, ry, rz), which are then combined into a six-dimensional screw [dx, dy, dz, rx, ry, rz]. This completely and accurately represents the six-degree-of-freedom motion information of each entity node, avoiding the simplification bias in the motion state description.

[0032] After the degrees of freedom of the entity nodes are encoded, the multi-entity constraint association definition method is adopted to sort out the geometric constraint relationships between the entity nodes of the assembly one by one. For the seven common constraint types in the assembly scenario, namely, common point, collinear, coplanar, parallel, perpendicular, fixed and distance, corresponding constraint hyperedges are defined. Unlike the bilateral connection of ordinary graphs, each hyperedge can simultaneously associate two or more entity nodes with corresponding constraint relationships, realizing the direct expression of multi-entity coupling constraints.

[0033] Finally, a graph structure construction process is adopted, using the classified point, line, surface, and volume entity nodes as vertices of the graph, and the defined constraint hyperedges as the connection relationships between vertices. At the same time, the six-dimensional spinor vectors corresponding to each entity node are bound to the corresponding vertices. By integrating all vertices, connection relationships, and bound degrees of freedom information, the overall construction of the constraint hypergraph is completed, realizing the integrated structured expression of the geometric entity and constraint relationship of the assembly.

[0034] Based on this, the constraint hypergraph constructed by the above steps can be further used for subsequent constraint strength quantification and degree-of-freedom mask representation. The specific implementation methods will be detailed in the following examples regarding the extended properties of the constraint hypergraph.

[0035] By sequentially classifying the geometric entities of the assembly, encoding the degrees of freedom of the entity motion, defining the constraint hyperedge, and constructing the constraint hypergraph, a complete and accurate model of the complex multi-entity constraints of the assembly is achieved, providing a standardized and reliable basic model for subsequent collaborative solution of six-degree-of-freedom constraints.

[0036] Furthermore, the method provided in this application embodiment includes: In the definition of a hyperedge, point constraints eliminate three translations, line constraints eliminate translations along the axis and rotations around the axis, and surface constraints eliminate translations in the normal direction and rotations around the normal. A degree-of-freedom mask based on the labeled constraints is added to the constrained hypergraph, and a constraint strength weight is added to each hyperedge.

[0037] Optionally, after completing the basic construction of the constraint hypergraph in the aforementioned steps, the effect of the degree of freedom constraint on each type of hyperedge should be clearly quantified and defined to provide accurate rules for subsequent constraint solving.

[0038] For point constraints, the three mutually orthogonal translational degrees of freedom of the entity nodes are explicitly eliminated, preventing the entity from moving in any direction in space. For line constraints, the translational degrees of freedom of the entity nodes along the constraint axis and the rotational degrees of freedom around the constraint axis are explicitly eliminated, allowing the entity to move only along a plane perpendicular to the axis and rotate in two directions perpendicular to the axis. For surface constraints, covering typical assembly constraint types such as surface mating and surface parallelism, the translational degrees of freedom of the entity nodes along the surface normal and the rotational degrees of freedom around the normal are explicitly eliminated. That is, the entity cannot move along the normal or rotate around the normal, but can translate in the surface along two orthogonal directions and tilt around two orthogonal axes within the surface, retaining two in-plane translational degrees of freedom and two in-plane rotational degrees of freedom. For other types of hyperedges such as parallel, perpendicular, fixed, and distance, the specific restriction rules for each degree of freedom of the entity nodes are also explicitly defined. By defining the degree-of-freedom restriction rules corresponding to each type of hyperedge, the clear effects of various constraints on the motion of the entity are obtained.

[0039] After defining the effects of hyperedge constraints, binary masking is used to add a degree-of-freedom mask based on the labeled constraints to each entity node in the constraint hypergraph. The degree-of-freedom mask is a six-dimensional binary vector, with its six dimensions corresponding one-to-one with the six-dimensional spinor vector of the entity node, representing three translational degrees of freedom and three rotational degrees of freedom, respectively. A binary value of 0 indicates that the corresponding degree of freedom is constrained, and a binary value of 1 indicates that the corresponding degree of freedom remains free. The value of the degree-of-freedom mask is determined dimension by dimension based on the constraint effects of all hyperedges associated with the entity node. If a degree of freedom is constrained by any associated hyperedge, the mask value for that dimension is set to 0; otherwise, it is set to 1. Through this encoding process, the initial degree-of-freedom state representation of each entity node is obtained.

[0040] While adding a degree-of-freedom mask, constraint strength weights are added to each hyperedge in the constraint hypergraph. Constraint strength weights are represented by values ​​between 0 and 1, where 0 indicates no constraint effect on the hyperedge, 1 indicates a completely rigid constraint, and intermediate values ​​represent the degree of rigidity. The constraint strength weights for each hyperedge are determined based on the constraint type and assembly process requirements. For example, the weight of fixed constraints is uniformly set to 1; the weights of rigid geometric constraints such as concurrent points, collinearities, and coplanarities are set to values ​​between 0.9 and 1; and the weights of flexible constraints such as distance and parallelism can be set to values ​​between 0.5 and 0.9 according to assembly tolerance requirements, and can be flexibly adjusted according to actual assembly tolerances. By assigning corresponding constraint strength weights to each hyperedge, differentiated representations of the strength of different constraints are achieved.

[0041] By clarifying the degree-of-freedom constraint effects of various types of hyperedges, adding degree-of-freedom masks to entity nodes, and adding constraint strength weights to hyperedges, the quantitative enhancement of constrained hypergraphs was achieved, providing accurate quantitative basis for subsequent six-degree-of-freedom constraint collaborative solution and effectively improving the accuracy and reliability of constraint solution.

[0042] Furthermore, the method provided in this application embodiment includes: Based on the constraint hypergraph, a constraint solver for the assembly is constructed, wherein the constraint solver includes a first physical twin layer and a second constraint solving layer, and the second constraint solving layer includes a degree-of-freedom solving partition and a constraint conflict detection partition; according to the constraint solver, the degree of freedom of the entity nodes of the constraint hypergraph is initialized, and the hypergraph topology is iteratively updated using a tree-like belief propagation method to obtain the constraint force allocation and pose correction amount.

[0043] Specifically, based on the completed constraint hypergraph, a modular software architecture is adopted to build a two-layer constraint solver that adapts to the constraint solving requirements of the assembly. The constraint solver is divided into two independent functional layers: the first physical twin layer and the second constraint solving layer, and the functional boundaries and data interaction rules of each layer are clearly defined.

[0044] After completing the overall architecture partitioning, the first physical twin layer is constructed. The lightweight geometric twin model of the assembly is imported, establishing a one-to-one mapping between each entity node in the constraint hypergraph and its corresponding geometric element in the lightweight twin model. Spatial coordinates and real-time pose attributes are bound to each entity node. Simultaneously, a spatial field intensity rendering module is constructed, pre-setting mapping rules between field intensity values ​​and color / transparency. This is used to transform the subsequently calculated constraint force allocation and pose correction quantities into a visualized spatial field intensity distribution, achieving an intuitive presentation of the solution results.

[0045] Next, a second constraint solving layer is constructed, employing a partitioned parallel design approach. This second constraint solving layer is further divided into a degree-of-freedom solving partition and a constraint conflict detection partition. The two partitions establish a real-time data interaction channel through shared memory. The degree-of-freedom solving partition undertakes the core constraint solving calculation tasks, while the constraint conflict detection partition undertakes the parallel constraint conflict detection tasks. The two partitions run independently with real-time data synchronization, avoiding efficiency bottlenecks caused by serial execution.

[0046] When constructing the degree-of-freedom (DOF) solution partition, the vertex and hyperedge data of the constraint hypergraph are imported into the graph computation engine to establish a computational topology consistent with the constraint hypergraph topology. Then, message passing rules for tree-like belief propagation are defined, setting computation functions for hyperedges to transmit constraint beliefs to connected nodes, and for nodes to transmit degree-of-freedom states to connected hyperedges. An iterative convergence criterion is also set: the change in the node degree-of-freedom mask between two adjacent iterations must be less than a preset threshold. Finally, a constraint force and pose correction calculation module is constructed to calculate the corresponding constraint force allocation and pose correction based on the converged node degree-of-freedom states.

[0047] When constructing the constraint conflict detection partition, parallel computing is used to build constraint conflict detection, over-constraint determination, and under-constraint determination modules respectively. The constraint conflict detection module receives the node degree-of-freedom (DOF) state data output from each iteration of the DDF solution partition in real time, checking for situations where the same DDF is simultaneously constrained by multiple hyperedges with contradictory constraint rules. The over-constraint determination module constructs constraint force balance equations based on solving a system of linear equations, determining over-constraint by judging the existence and uniqueness of the equation solutions. The under-constraint determination module traverses the DDF masks of all entity nodes, locating the DDF type with a value of 1 that is not constrained by any hyperedge. Finally, a result feedback module is constructed to feed back the detected over-constraint and under-constraint results to the DDF solution partition in real time for subsequent DDF fusion processing.

[0048] After completing the construction of each layer and partition, a standardized data interface specification is adopted to define the pose data interface and field strength data interface between the first physical twin layer and the second constraint solution layer, and to define the degree of freedom state data interface and conflict result data interface between the degree of freedom solution partition and the constraint conflict detection partition, so as to ensure that the data transmission format between each module is unified and the interaction is smooth.

[0049] Next, when the constraint solver initializes the degrees of freedom of entity nodes in the constraint hypergraph, the constraints are divided into explicit constraints and implicit constraints. Explicit constraints include labeled constraints and unlabeled constraints. In unlabeled constraints, the degrees of freedom of fixed entities are all zero and the degrees of freedom of floating entities are all one. Implicit constraints are obtained through constraint propagation mining based on graph isomorphic networks. The specific implementation method of this step will be described in detail later.

[0050] Finally, the constraint solver iteratively updates the hypergraph topology using a tree-like belief propagation method. The constraint beliefs of the degrees of freedom from the source entity node to the target entity node are calculated by the hyperedge according to the constraint type. The beliefs of each entity node are collected and merged with the initial degrees of freedom to update the degree of freedom mask. Multiple iterations are performed until the change of the degree of freedom mask of the node in adjacent iterations is less than a preset threshold to determine the target degree of freedom state. Then, based on this state, the constraint force allocation and pose correction amount are calculated by each hyperedge. The specific implementation method of this step will also be described in detail later.

[0051] By constructing a constraint solver using a hierarchical and partitioned modular architecture, parallel processing of constraint solving and conflict detection, as well as the integration of digital twin virtual-real mapping, are achieved.

[0052] Furthermore, the method provided in this application embodiment includes: The initialization of the degrees of freedom of entity nodes in the constrained hypergraph includes explicit constraints and implicit constraints; among them, the explicit constraints include labeled constraints and unlabeled constraints, and the degrees of freedom of fixed entities in the unlabeled constraints are all zero, while the degrees of freedom of floating entities are all one; the implicit constraints are obtained by constraint propagation mining based on graph isomorphic networks.

[0053] In one embodiment, in the degree-of-freedom solving partition of the constraint solver, the initialization of explicit constraints is completed first, laying the foundation for subsequent implicit constraint mining.

[0054] Traverse all entity nodes in the constraint hypergraph and, based on the fixed marker information in the assembly process document, classify the entity nodes into two categories: fixed entities and floating entities. For entity nodes corresponding to labeled constraints, directly initialize them using the degree-of-freedom mask rules defined in the previous steps. For fixed entities without labeled constraints, uniformly set all dimensions of their six-dimensional degree-of-freedom mask to 0, locking all six degrees of freedom. For floating entities without labeled constraints, uniformly set all dimensions of their six-dimensional degree-of-freedom mask to 1, retaining all six degrees of freedom. Through the above operations, complete the initialization of the degrees of freedom corresponding to all explicit constraints.

[0055] Next, the constraint embedding vectors of entity nodes and hyperedges are extracted from the input graph isomorphic network of the constraint hypergraph. The constraint propagation mode and the embedding vectors are combined with the pre-learned constraint propagation mode to predict the constraint type and constraint strength of entity node pairs. A candidate implicit constraint list is obtained based on the confidence level. The list is then cross-validated with the labeled constraints. If there is a constraint contradiction, the corresponding labeled constraint is marked. If there is no contradiction, the candidate content is included in the implicit constraint. The specific implementation method of this step will be described in detail later.

[0056] By combining explicit constraint initialization with graph isomorphic network mining of implicit constraints, a complete modeling of assembly constraint relationships was achieved, effectively making up for the shortcomings of manually labeled constraints and significantly improving the accuracy and reliability of subsequent six-degree-of-freedom constraint collaborative solution.

[0057] Furthermore, the method provided in this application embodiment includes: The constrained hypergraph is input into a graph isomorphic network, and the constraint embedding vectors of each entity node and hyperedge are extracted. Based on the pre-learned constraint propagation mode and the constraint embedding vectors, the constraint type and constraint strength between entity node pairs are predicted. Based on the confidence level, a candidate implicit constraint list is output. The candidate implicit constraint list is cross-validated with the labeled constraints. If there is a constraint contradiction, the mapped labeled constraint entries are marked as over-constrained. If there is no contradiction, they are added to the implicit constraints.

[0058] Optionally, after completing the explicit constraint initialization, an implicit constraint mining model based on a graph isomorphic network is constructed. The input to this model is the structured data of the constraint hypergraph, specifically including the geometric type attributes and initial degree-of-freedom mask attributes of each entity node, the constraint type attributes and constraint strength attributes of each hyperedge, and the topological connections between entity nodes and hyperedges. The geometric type attributes of entity nodes and the constraint type attributes of hyperedges are one-hot encoded and converted into corresponding numerical vectors. These vectors are then concatenated with the initial degree-of-freedom mask, constraint strength, and other numerical features to serve as the input features for the graph isomorphic network. The output of this model is a 128-dimensional constraint embedding vector corresponding to each entity node and a 128-dimensional constraint embedding vector corresponding to each hyperedge.

[0059] When constructing the graph isomorphic network, a three-layer graph convolutional layer architecture is adopted. The input dimension of the first graph convolutional layer is set to match the total length of the input feature vector, for example, 32, and the output dimension is 64. The input dimension of the second graph convolutional layer is 64, and the output dimension is 128. The input dimension of the third graph convolutional layer is 128, and the output dimension is 128. Each graph convolutional layer uses a summation aggregation function as the feature aggregation method, performing bidirectional aggregation in the order of node to hyperedge, and then hyperedge to node: first, it aggregates the features of all neighboring nodes of each hyperedge, updating the feature vector of the hyperedge; then it aggregates the features of all neighboring hyperedges of each node, updating the feature vector of the node. The initial input dimension of the hyperedge features is 16, consistent with the node feature dimension. Each graph convolutional layer is followed by a batch normalization layer and a ReLU activation function to accelerate model convergence and prevent overfitting. Finally, the model sets two independent output heads, outputting the entity node constraint embedding vector and the hyperedge constraint embedding vector, both with a dimension of 128. After the entity node constraint embedding vector output head, two independent fully connected layers are connected, serving as a constraint type classifier and a constraint strength regressor, respectively. The classifier's output dimension is 7, corresponding to 7 common assembly constraint types, and it uses the Softmax activation function to output the probability of each category. The regressor's output dimension is 1, and it uses the Sigmoid activation function to output the constraint strength value in the range of 0 to 1.

[0060] A supervised training approach was used to train the graph isomorphic network. First, a training dataset was constructed, containing 10,000 annotated assembly constraint hypergraph samples. Each sample included complete explicit constraint annotations and corresponding implicit constraint labels. The training dataset was randomly divided into a training set and a validation set in an 8:2 ratio. The training set was used for model parameter updates, while the validation set was used to monitor the model's generalization ability during training. The model's total loss function was defined as a weighted sum of constraint type classification loss and constraint strength regression loss. Constraint type classification used the cross-entropy loss function, calculated for seven common assembly constraint types, while constraint strength regression used the mean squared error loss function. The weights for both were set to 0.7 and 0.3, respectively. The Adam optimizer was used for model training, with an initial learning rate of 0.001 and a batch size of 32. During training, after each full training iteration, the constraint prediction accuracy of the model was calculated on the validation set. When the improvement in validation set accuracy was less than 0.1% for 10 consecutive iterations, the model was considered converged, training was stopped, and the model weights were saved.

[0061] Next, the constrained hypergraph with explicit constraints will be used as input to the trained graph isomorphic network. During training, this graph isomorphic network is constructed using a combination of supervised learning and rule-based constraint propagation. The rule-based constraint propagation is used to generate auxiliary training samples. For example, based on the transitivity of fixed and coaxial constraints, as well as parallel and coplanar constraints, the coaxiality of A and C is derived from the fact that A and B are fixed and B and C are coaxial, thus expanding the implicit constraint labels in the training data. After feature extraction and aggregation through three layers of graph convolutional layers, the constraint embedding vector corresponding to each entity node and each hyperedge is obtained.

[0062] Based on the constraint propagation pattern learned by the model pre-learning, the constraint correlation degree between any two entity nodes is calculated. This pre-learned constraint propagation pattern can capture implicit constraint relationships such as transitive constraints. For example, when A and B are fixed and B and C are coaxial, it can be deduced that A and C are coaxial. The constraint correlation degree is calculated using the cosine similarity of the constraint embedding vectors of the two entity nodes. When the constraint correlation degree is greater than a preset threshold of 0.8, it is determined that the entity node pair has a potential implicit constraint.

[0063] For entity node pairs with potential implicit constraints, a classifier predicts the corresponding constraint type, and a regressor predicts the corresponding constraint strength. The classifier outputs the probabilities of seven common assembly constraint types, and the regressor outputs the constraint strength value in the range of 0 to 1, along with the confidence score of each prediction. All predictions with a confidence score greater than 0.9 are collected to generate a candidate implicit constraint list. The candidate implicit constraint list is then cross-validated with the labeled explicit constraints one by one to check for contradictory constraint types for the same entity node pair. If a constraint contradiction exists, the mapped labeled constraint entries are marked as constraints, and the marking results are fed back to the constraint conflict detection partition. If no constraint contradiction exists, the candidate implicit constraint is added to the constraint hypergraph and participates as an implicit constraint in subsequent degree-of-freedom updates.

[0064] By extracting constraint embedding vectors and predicting constraint types and strengths through a constraint hypergraph input graph isomorphic network, and then verifying them with labeled constraints, the automatic discovery and supplementation of implicit geometric constraints in assemblies is realized, thereby improving the completeness of constraint modeling.

[0065] Furthermore, the method provided in this application embodiment includes: Using a tree-like belief propagation method, each hyperedge calculates the degree-of-freedom constraint belief from the source entity node to the target entity node based on the constraint type. The source entity node and the target entity node are determined by the directed connection end of the hyperedge. Each entity node collects the degree-of-freedom constraint beliefs of the incoming edges and merges them with the initial degree of freedom to update the degree-of-freedom mask. By performing multiple rounds of degree-of-freedom mask update iterations until convergence, the target degree-of-freedom state is determined. The convergence condition is that the change in the degree-of-freedom mask corresponding to the node between the previous iteration and the next iteration is less than a preset threshold. For the constraint hypergraph based on the target degree-of-freedom state, each hyperedge calculates the constraint force allocation and pose correction amount based on the current degree-of-freedom state of the nodes at both ends.

[0066] Specifically, in the degree-of-freedom solution partition of the constraint solver, a tree-like belief propagation algorithm is used to iteratively update the constraint hypergraph topology, achieving efficient solution of constraint force allocation and pose correction. The specific steps are as follows: First, a directed graph transformation technique from graph theory is employed to direct all undirected hyperedges in the constrained hypergraph. Based on the assembly process sequence and the fixed state of the entities, the directed connection endpoints of each hyperedge are determined. Entities upstream in the assembly sequence or already fixed are designated as source nodes, while those downstream or unfixed are designated as target nodes. For hyperedges connecting multiple entities, directed connections are generated from each source node to each target node, forming a complete set of directed hyperedges. This provides a clear message transmission direction for subsequent belief propagation.

[0067] After the hyperedges are directed, all directed hyperedges are traversed one by one. Based on the constraint type corresponding to the hyperedge, the degree-of-freedom constraint belief from the source entity node to the target entity node is calculated. The degree-of-freedom constraint belief is a six-dimensional vector, with its six dimensions corresponding one-to-one with the six-dimensional degree-of-freedom mask of the entity node. Each dimension in the vector ranges from 0 to 1, where a value of 0 indicates that the corresponding degree of freedom is not restricted by the constraint, and a value of 1 indicates that the corresponding degree of freedom is completely restricted by the constraint. For example, for a coaxial constraint hyperedge, according to the constraint rules defined in the previous steps, the generated degree-of-freedom constraint belief vector has a translation dimension along the axis and a rotation dimension around the axis of 0, while the other four dimensions have a value of 1, indicating that the target entity can only translate along the axis and rotate around the axis. Through the above calculations, the degree-of-freedom constraint belief corresponding to each directed hyperedge is obtained.

[0068] Each entity node collects all the degree-of-freedom (DOF) constraint beliefs pointing to itself from all directed hyperedges, and then fuses all the collected constraint belief vectors with its own initial DDF mask. The fusion process uses a bitwise logical AND operation: for each dimension of the DDF mask, if any constraint belief vector has a value of 1 in that dimension, the DDF mask value for that dimension is updated to 0; otherwise, the original value remains unchanged. In fact, the new value of the DDF mask is the result of a logical AND operation between the original mask value and the bitwise inversion of each constraint belief vector. For example, if an entity node's initial DDF mask is all 1s, and it receives a belief vector for a coaxial constraint and a belief vector for a coplanar constraint, after a bitwise logical AND operation, the updated DDF mask is obtained, accurately representing the current remaining DDF state of the entity node.

[0069] The process proceeds in multiple rounds, updating the degree-of-freedom (DOF) mask by propagating outwards layer by layer along the directed hyperedge, starting from a fixed entity node. After each full iteration, the Hamming distance between the current DDF mask of all entity nodes and the previous DDF mask is calculated. The Hamming distance is the total number of different corresponding bit values ​​in two binary vectors. When the sum of the Hamming distances of all entity nodes is less than a preset threshold of 1, the iteration is considered converged, the iteration process is stopped, and the current DDF mask is taken as the target DDF state of the entity node. If the iteration count exceeds the preset maximum of 50 iterations without convergence, a constraint conflict is identified, and the conflict information is fed back to the constraint conflict detection partition for processing.

[0070] After iterative convergence to obtain the target degree of freedom state, all hyperedges in the constraint hypergraph are traversed one by one. Based on the current target degree of freedom state and initial pose of the entity nodes at both ends of the hyperedge, the corresponding constraint force allocation and pose correction are calculated. The constraint balance equation is constructed using the least squares method. Combined with the constraint strength weights assigned to each hyperedge in the aforementioned constraint hypergraph, the magnitude and direction of the constraint force corresponding to each hyperedge are solved. The pose correction is the difference between the pose corresponding to the target degree of freedom state of the entity node and the current actual pose, including three translation corrections and three rotation corrections. Through the above calculations, the constraint force allocation results of all hyperedges and the pose correction of all entity nodes are obtained.

[0071] By employing a tree-like belief propagation method for iterative updates on the constraint hypergraph topology, efficient parallel solutions for multi-entity coupled constraints are achieved, significantly improving the convergence speed and accuracy of constraint force allocation and pose correction calculation.

[0072] Furthermore, the method provided in this application embodiment includes: In the degree-of-freedom solution partition, a round of propagation processing is performed. Each entity node collects the degree-of-freedom constraint beliefs of the incoming edges and imports them into the constraint conflict detection partition for constraint contradiction detection. If a constraint contradiction exists, it is determined whether it constitutes an over-constraint by using the constraint force balance equation, and an over-constraint judgment result is generated. The degree-of-freedom type that is not constrained by any hyperedge is located as the under-constraint judgment result. Based on the over-constraint judgment result and the under-constraint judgment result, degree-of-freedom fusion processing is performed.

[0073] Specifically, the constraint solver adopts a dual-branch architecture that combines solving and verification. The degree-of-freedom solving partition and the constraint conflict detection partition establish a real-time data synchronization channel through shared memory. After each round of tree-like belief propagation processing is completed, the degree-of-freedom solving partition immediately synchronizes the current degree-of-freedom constraint beliefs and the updated degree-of-freedom mask of all entity nodes to the constraint conflict detection partition. The two partitions run independently at the same time, realizing the parallel execution of the solving process and conflict verification.

[0074] After receiving the synchronized degrees of freedom data, the constraint conflict detection partition first performs a preliminary constraint conflict detection. It iterates through all six degrees of freedom of all entity nodes, collecting constraint belief values ​​transmitted by all associated hyperedges for each dimension. It checks whether there are constraint beliefs with values ​​of 0 and 1 simultaneously within the same dimension—that is, one hyperedge requires the degree of freedom to be locked while the other allows it to be free. If so, a constraint conflict is determined for that degree of freedom. The entity node numbers, degree of freedom dimension numbers, and corresponding conflicting hyperedge information for all conflicting entities are recorded, generating a preliminary constraint conflict list.

[0075] For all conflicting items in the initial constraint contradiction list, a constraint force balance equation is constructed to determine over-constraint. The constraint strength weight of each hyperedge is used as the equation coefficients, and the difference between the target pose and the current pose of the entity node is used as the right-hand side of the equation, constructing a system of linear equations. Gaussian elimination is used to calculate the rank of the coefficient matrix and the rank of the augmented matrix. If the rank of the coefficient matrix is ​​less than the rank of the augmented matrix, the constraint contradiction is determined to be over-constrained. All over-constrained entity nodes, conflicting hyperedges, and their corresponding constraint strength weights are recorded to generate the over-constraint judgment result.

[0076] While performing over-constraint determination, under-constraint detection is also performed. The degree-of-freedom masks of all entity nodes are traversed one by one, extracting all degrees-of-freedom dimensions with a value of 1. For each degree-of-freedom dimension with a value of 1, it is checked whether there exists at least one associated hyperedge with a constraint belief value of 1. If no associated hyperedge imposes a constraint on this dimension, it is determined that the degree-of-freedom dimension is under-constrained. The entity node numbers and corresponding degree-of-freedom dimensions of all under-constrained entities are recorded, generating the under-constraint determination result.

[0077] Finally, based on the generated over-constraint and under-constraint judgment results, degree-of-freedom fusion processing is performed in the degree-of-freedom solution partition. For over-constraint problems, a cooperative constraint weakening allocation method based on constraint strength weight is adopted, retaining the constraint belief of the hyperedge with the highest constraint strength weight, and proportionally reducing the constraint strength weight of the remaining conflicting hyperedges until the constraint contradiction is eliminated. For under-constraint problems, based on the assembly type and existing constraint relationships, the supplementary constraint type that best fits the assembly logic is matched from the preset assembly constraint type library, a constraint recommendation list is generated, and fed back to the operation terminal. The processed constraint beliefs are then re-fused with the original initial degrees of freedom, the degree-of-freedom mask of the entity nodes is updated, and subsequent tree-like belief propagation iterations continue.

[0078] By introducing parallel constraint conflict detection and real-time degree-of-freedom fusion processing during the tree-like belief propagation iteration process, early detection and early handling of constraint contradictions are achieved, effectively reducing the number of invalid iterations and significantly improving the convergence speed and stability of constraint solution.

[0079] Furthermore, the method provided in this application embodiment includes: The constraint force allocation and pose correction amount are mapped to the field strength distribution in space and synchronized to the first physical twin layer, wherein the first physical twin layer establishes information interaction with the operation terminal; when the operation terminal performs an operation on the assembly, when the operation enters the constraint tolerance range based on the field strength distribution, the pose automatic adsorption processing is performed according to the field strength distribution to obtain theoretical data; based on the theoretical data, operation guidance is provided for the operation interface.

[0080] In one embodiment, the constraint tolerance range is first calculated to provide a basis for subsequent automatic pose snapping. Specifically: the design tolerance requirements corresponding to each assembly constraint are extracted from the assembly process document, including positional and angular tolerances. Combining the constraint strength weights assigned to each hyperedge in the previous steps, the constraint tolerance range corresponding to each constraint is calculated. The constraint tolerance range is a six-dimensional spatial region centered on the target pose of the entity node. The tolerance threshold for each translational degree of freedom is the product of the corresponding design positional tolerance and the constraint strength weight, and the tolerance threshold for each rotational degree of freedom is the product of the corresponding design angular tolerance and the constraint strength weight. A higher constraint strength weight corresponds to a smaller tolerance threshold, and more stringent triggering conditions for automatic snapping.

[0081] After calculating the constraint tolerance range, the constraint force distribution and pose correction obtained from the constraint solution are mapped to a three-dimensional spatial field intensity distribution. The magnitude of the pose correction corresponds to the field intensity value, the direction of the pose correction corresponds to the gradient direction of the field intensity, and the magnitude of the constraint force is superimposed on the field intensity corresponding to the pose correction, forming a comprehensive constraint field intensity distribution. For example, the field intensity value can be set as a weighted sum of the magnitude of the pose correction and the magnitude of the constraint force, and the field intensity gradient direction is taken as the unit vector direction of the pose correction. The calculated spatial field intensity distribution data is synchronized to the first physical twin layer of the constraint solver, aligning it with the spatial coordinate system of the geometrically lightweight twin model, so that each spatial location corresponds to a unique field intensity value and direction.

[0082] The first physical twin layer establishes a real-time two-way information exchange channel with the operator. The operator collects the spatial pose data of the current object in real time and synchronizes it to the first physical twin layer. The first physical twin layer calculates the difference between the current operation pose and the target pose of the entity node for each degree of freedom, and compares the difference of each degree of freedom with the corresponding constraint tolerance threshold. When the difference of all degrees of freedom is less than or equal to the corresponding tolerance threshold, the operation is determined to have entered the constraint tolerance range.

[0083] Once the operation is determined to be within the constraint tolerance range, the first physical twin layer performs automatic pose snapping based on the field strength distribution at the current spatial location. Guided by the field strength gradient direction, the pose of the operation object is automatically adjusted to the target pose position with the strongest field strength, obtaining theoretically aligned pose data. The adjustment speed of automatic snapping is positively correlated with the field strength at the current position; the stronger the field strength, the faster the adjustment speed, ensuring that the operation object can be quickly and accurately aligned to the target pose.

[0084] Finally, based on the obtained theoretical alignment pose data, corresponding operation guidance information is generated and pushed to the operation interface. When the operation pose is outside the constraint tolerance range, the operation interface displays a directional arrow pointing to the target pose and a distance indicator, guiding the operator to move the object in the correct direction. When the operation enters the constraint tolerance range, the operation interface displays an automatic snap-in prompt. After automatic snap-in is complete, the operation interface displays a confirmation message indicating that the constraint alignment is complete and highlights the completed constraint relationships.

[0085] By mapping constraint force allocation and pose correction amount to an intuitive spatial field strength distribution, combined with constraint tolerance-triggered automatic pose snapping and visual operation guidance, the accuracy and efficiency of assembly operations are effectively improved, while reducing the skill requirements and workload of operators.

[0086] In summary, the six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins provided in this application has the following technical effects: This application maps constraint force allocation and pose correction to spatial field strength distribution and synchronizes it to the first physical twin layer. Relying on the information interaction between this layer and the operating terminal, after the assembly operation enters the constraint tolerance range, the pose is automatically snapped in according to the field strength distribution and theoretical data is obtained. Then, the interface operation guidance is generated in combination with the theoretical data, realizing real-time auxiliary guidance for assembly operations. This effectively improves the accuracy and execution efficiency of assembly operations, and achieves the technical effect of adapting to complex constraint scenarios of complex assemblies, calculating the optimal position of each component in real time and automatically completing precise docking, while providing intuitive guidance information to the operating terminal.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a six-degree-of-freedom three-dimensional constraint collaborative solution system based on digital twins, the system comprising: The constraint hypergraph reconstruction module is used to perform geometric entity classification and motion degree of freedom encoding on the assembly, define hyperedges according to geometric constraint relationships, and reconstruct a constraint hypergraph.

[0088] The constraint solver construction module is used to perform geometric lightweight twinning on the assembly. It constructs a constraint solver containing a degree-of-freedom solving partition and a constraint conflict detection partition based on the constraint hypergraph. By performing degree-of-freedom initialization, the degree-of-freedom mask is iteratively updated until convergence using tree-like belief propagation. Each hyperedge calculates the constraint force allocation and pose correction based on the current degree of freedom of the two end nodes. Constraint conflict detection is interspersed during the iteration process.

[0089] The operation guidance execution module is used to map the constraint force allocation and pose correction amount to the spatial field strength distribution in the first physical twin layer of the constraint solver, and push operation guidance to the operation terminal. When the operation of the operation terminal enters the constraint tolerance range, the pose is automatically snapped and operation guidance information is generated.

[0090] Furthermore, the constrained hypergraph reconstruction module is used to perform the following steps: The geometric entities in the assembly are divided into multiple types of entity nodes, including points, lines, surfaces, and volumes. The multiple types of entity nodes are traversed, and the degrees of freedom of the entity motion are encoded as six-dimensional spinor vectors, where the degrees of freedom are defined by three translational degrees of freedom and three rotational degrees of freedom. Hyperedges are defined based on geometric constraints, where the hyperedge types include concurrent, collinear, coplanar, parallel, perpendicular, fixed, and distance-based. Each hyperedge can connect multiple entity nodes. A constraint hypergraph is constructed based on the multiple types of entity nodes, the six-dimensional degree-of-freedom vectors, and the hyperedges.

[0091] Furthermore, the constrained hypergraph reconstruction module is used to perform the following steps: In the definition of a hyperedge, point constraints eliminate three translations, line constraints eliminate translations along the axis and rotations around the axis, and surface constraints eliminate translations in the normal direction and rotations around the normal. A degree-of-freedom mask based on the labeled constraints is added to the constrained hypergraph, and a constraint strength weight is added to each hyperedge.

[0092] Furthermore, the constraint solver building module is used to perform the following steps: Based on the constraint hypergraph, a constraint solver for the assembly is constructed, wherein the constraint solver includes a first physical twin layer and a second constraint solving layer, and the second constraint solving layer includes a degree-of-freedom solving partition and a constraint conflict detection partition; according to the constraint solver, the degree of freedom of the entity nodes of the constraint hypergraph is initialized, and the hypergraph topology is iteratively updated using a tree-like belief propagation method to obtain the constraint force allocation and pose correction amount.

[0093] Furthermore, the constraint solver building module is used to perform the following steps: Using a tree-like belief propagation method, each hyperedge calculates the degree-of-freedom constraint belief from the source entity node to the target entity node based on the constraint type. The source entity node and the target entity node are determined by the directed connection end of the hyperedge. Each entity node collects the degree-of-freedom constraint beliefs of the incoming edges and merges them with the initial degree of freedom to update the degree-of-freedom mask. By performing multiple rounds of degree-of-freedom mask update iterations until convergence, the target degree-of-freedom state is determined. The convergence condition is that the change in the degree-of-freedom mask corresponding to the node between the previous iteration and the next iteration is less than a preset threshold. For the constraint hypergraph based on the target degree-of-freedom state, each hyperedge calculates the constraint force allocation and pose correction amount based on the current degree-of-freedom state of the nodes at both ends.

[0094] Furthermore, the constraint solver building module is used to perform the following steps: In the degree-of-freedom solution partition, a round of propagation processing is performed. Each entity node collects the degree-of-freedom constraint beliefs of the incoming edges and imports them into the constraint conflict detection partition for constraint contradiction detection. If a constraint contradiction exists, it is determined whether it constitutes an over-constraint by using the constraint force balance equation, and an over-constraint judgment result is generated. The degree-of-freedom type that is not constrained by any hyperedge is located as the under-constraint judgment result. Based on the over-constraint judgment result and the under-constraint judgment result, degree-of-freedom fusion processing is performed.

[0095] Furthermore, the constraint solver building module is used to perform the following steps: The initialization of the degrees of freedom of entity nodes in the constrained hypergraph includes explicit constraints and implicit constraints; among them, the explicit constraints include labeled constraints and unlabeled constraints, and the degrees of freedom of fixed entities in the unlabeled constraints are all zero, while the degrees of freedom of floating entities are all one; the implicit constraints are obtained by constraint propagation mining based on graph isomorphic networks.

[0096] Furthermore, the constraint solver building module is used to perform the following steps: The constrained hypergraph is input into a graph isomorphic network, and the constraint embedding vectors of each entity node and hyperedge are extracted. Based on the pre-learned constraint propagation mode and the constraint embedding vectors, the constraint type and constraint strength between entity node pairs are predicted. Based on the confidence level, a candidate implicit constraint list is output. The candidate implicit constraint list is cross-validated with the labeled constraints. If there is a constraint contradiction, the mapped labeled constraint entries are marked as over-constrained. If there is no contradiction, they are added to the implicit constraints.

[0097] Furthermore, the operation guidance execution module is used to perform the following steps: The constraint force allocation and pose correction amount are mapped to the field strength distribution in space and synchronized to the first physical twin layer, wherein the first physical twin layer establishes information interaction with the operation terminal; when the operation terminal performs an operation on the assembly, when the operation enters the constraint tolerance range based on the field strength distribution, the pose automatic adsorption processing is performed according to the field strength distribution to obtain theoretical data; based on the theoretical data, operation guidance is provided for the operation interface.

[0098] The six-degree-of-freedom three-dimensional constraint collaborative solution system based on digital twin provided in this invention can execute the six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twin provided in any embodiment of this invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0099] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0100] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins, characterized in that, The method includes: Perform geometric entity classification and motion degree of freedom encoding on the assembly, define hyperedges according to geometric constraints, and reconstruct it into a constraint hypergraph; The assembly is geometrically lightweight twinned. A constraint solver containing a degree-of-freedom solution partition and a constraint conflict detection partition is constructed based on the constraint hypergraph. By performing degree-of-freedom initialization, the degree-of-freedom mask is iteratively updated until convergence using tree-like belief propagation. Each hyperedge calculates the constraint force distribution and pose correction based on the current degree of freedom of the two end nodes. Constraint conflict detection is interspersed during the iteration process. The constraint force allocation and pose correction amount are mapped to the spatial field intensity distribution in the first physical twin layer of the constraint solver, and operation guidance is pushed to the operation terminal. When the operation of the operation terminal enters the constraint tolerance range, the pose is automatically snapped and operation guidance information is generated. The calculation of constraint force allocation and pose correction includes: Based on the constraint hypergraph, a constraint solver for the assembly is constructed, wherein the constraint solver includes a first physical twin layer and a second constraint solving layer, and the second constraint solving layer includes a degree-of-freedom solving partition and a constraint conflict detection partition; According to the constraint solver, the degrees of freedom of the entity nodes of the constraint hypergraph are initialized, and the topology of the hypergraph is iteratively updated using a tree-like belief propagation method to obtain the constraint force allocation and pose correction amount. The initialization of the degrees of freedom of entity nodes in the constrained hypergraph includes both explicit and implicit constraints. The explicit constraints include labeled constraints and unlabeled constraints. In the unlabeled constraints, the degrees of freedom of fixed entities are all zero, and the degrees of freedom of floating entities are all one. The implicit constraints are obtained by using constraint propagation mining based on graph isomorphic networks; The implicit constraints are obtained by using constraint propagation mining based on graph isomorphic networks, including: The constraint hypergraph is input into a graph isomorphic network, and the constraint embedding vectors of each entity node and hyperedge are extracted. Based on the pre-learned constraint propagation pattern and the constraint embedding vector, the constraint type and constraint strength between entity node pairs are predicted, and a candidate implicit constraint list is output based on the confidence level. The candidate implicit constraint list is cross-validated with the labeled constraints. If there is a constraint contradiction, the mapped labeled constraint entries are marked as having been constrained; if there is no contradiction, they are added to the implicit constraints.

2. The six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins as described in claim 1, characterized in that, Reconstructing into a constrained hypergraph includes: The geometric entities in the assembly are divided into multiple types of entity nodes, which include points, lines, surfaces, and volumes; Traverse the various types of entity nodes and encode the entity's motion degrees of freedom as a six-dimensional spinor vector, wherein the motion degrees of freedom are defined by three translational degrees of freedom and three rotational degrees of freedom; Hyperedges are defined based on geometric constraints. Hyperedge types include concurrent, collinear, coplanar, parallel, perpendicular, fixed, and distance-based. Each hyperedge can connect multiple entity nodes. Based on the various types of entity nodes, the six-dimensional degree-of-freedom vector, and the hyperedges, a constrained hypergraph is constructed.

3. The six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins as described in claim 2, characterized in that, In the definition of a hyperedge, point constraints eliminate three translations, line constraints eliminate translations along the axis and rotations about the axis, and face constraints eliminate translations in the normal direction and rotations about the normal. Add a degree-of-freedom mask based on the labeled constraints to the constrained hypergraph, and add a constraint strength weight to each hyperedge.

4. The six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twin as described in claim 1, characterized in that, A tree-like belief propagation method is used to iteratively update the hypergraph topology, obtaining constraint assignments and pose corrections, including: Using a tree-like belief propagation method, each hyperedge calculates the degree-of-freedom constraint belief from the source entity node to the target entity node according to the constraint type, where the source entity node and the target entity node are determined according to the directed connection end of the hyperedge; Each entity node collects the freedom constraint beliefs of the incoming edges, merges them with the initial freedom, and updates the freedom mask; By performing multiple rounds of freedom mask update iterations until convergence, the target freedom state is determined. The convergence condition is that the change in the freedom mask corresponding to the node between the previous iteration and the next iteration is less than a preset threshold. For a constrained hypergraph based on the target degree of freedom state, each hyperedge calculates the constraint force allocation and pose correction amount according to the current degree of freedom state of the nodes at both ends.

5. The six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins as described in claim 1, characterized in that, After collecting the freedom constraint beliefs of the incoming edges for each entity node, the following are included: A round of propagation processing is performed in the degree-of-freedom solution partition, and each entity node collects the degree-of-freedom constraint beliefs of the incoming edges, which are then imported into the constraint conflict detection partition for constraint contradiction detection. If there is a constraint contradiction, determine whether it constitutes an over-constraint by using the constraint force balance equation, and generate an over-constraint judgment result. Identify the type of degree of freedom that is not constrained by any hyperedge and use it as the result of the underconstraint judgment; Based on the over-constraint judgment result and the under-constraint judgment result, a degree of freedom fusion process is performed.

6. The six-degree-of-freedom three-dimensional constraint collaborative solution method based on digital twins as described in claim 1, characterized in that, The constraint force allocation and pose correction are mapped to the spatial field intensity distribution in the first physical twin layer of the constraint solver, and operation instructions are pushed to the operator, including: The constraint force allocation and pose correction amount are mapped to the field strength distribution in space and synchronized to the first physical twin layer, wherein the first physical twin layer establishes information interaction with the operation terminal; When the operating terminal performs an operation on the assembly, and the operation enters the constraint tolerance range based on the field strength distribution, automatic pose adsorption processing is performed according to the field strength distribution to obtain theoretical data; Based on the theoretical data, provide operation guidance for the user interface.

7. A six-degree-of-freedom three-dimensional constraint collaborative solution system based on digital twins, characterized in that, The system is used to implement the six-DOF three-dimensional constraint collaborative solution method based on digital twins as described in any one of claims 1-6, the system comprising: The constraint hypergraph reconstruction module is used to perform geometric entity classification and motion degree of freedom encoding on the assembly, define hyperedges according to geometric constraint relationships, and reconstruct a constraint hypergraph. The constraint solver construction module is used to perform geometric lightweight twinning on the assembly. It constructs a constraint solver containing a degree-of-freedom solving partition and a constraint conflict detection partition based on the constraint hypergraph. By performing degree-of-freedom initialization, the degree-of-freedom mask is iteratively updated until convergence using tree-like belief propagation. Each hyperedge calculates the constraint force distribution and pose correction based on the current degree of freedom of the two end nodes. Constraint conflict detection is interspersed during the iteration process. The operation guidance execution module is used to map the constraint force allocation and pose correction amount to the spatial field strength distribution in the first physical twin layer of the constraint solver, and push operation guidance to the operation terminal. When the operation of the operation terminal enters the constraint tolerance range, the pose is automatically snapped and operation guidance information is generated.

Images