A cloud CAD-based three-dimensional modeling complexity evaluation and end-to-cloud resource collaborative scheduling method and system

Abstract

This invention discloses a method and system for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD, belonging to the field of computer-aided design technology. The method constructs a multi-layered associated object containing sketches, feature objects, geometric topology, and assembly relationships, and assesses the overall complexity based on this multi-layered associated object. It generates operation semantic signatures based on the operations to be executed, extends along feature dependencies, topological adjacency relationships, and assembly relationships to form an operation influence subgraph, and forms an incremental reconstruction work package to determine the effective complexity of the operation and the resource requirement profile. Combining terminal performance, network status, cloud load, task queue, model cache status, and result submission risks, it predicts the completion time of different execution paths and selects an execution plan. Semantic checkpoints support task migration, and after verifying the consistency of execution results, the prediction parameters are updated based on the actual time consumption feedback. This invention can improve the efficiency of modeling task scheduling and the consistency of edge-cloud execution results.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, specifically to a method and system for assessing the complexity of 3D modeling and coordinating the scheduling of edge-cloud resources based on cloud CAD. Background Technology

[0002] Cloud-based 3D CAD typically consists of a browser-based or lightweight client, cloud-based geometric computation services, model data services, and collaboration services. Compared to traditional desktop CAD, cloud-based 3D CAD allows different terminals to share model data and computational power. However, the response time of modeling operations is simultaneously affected by model content, the scope of operation, terminal performance, network status, cloud load, and data synchronization methods. Fixedly placing all geometric computations on the terminal or entirely in the cloud makes it difficult to consistently achieve good response times across different models and operating states.

[0003] Existing model size descriptions often approximate complexity using a single metric such as file size, number of faces, number of triangles, or number of features. However, these methods fail to account for computational differences arising from variations in sketch contour complexity, constraint coupling strength, feature dependency depth, topology pruning range, and the number of assembly matings, even with the same number of objects. For example, two sketches may have the same number of primitives, but significantly different closed contours, constraint numbers, and associated edge numbers; two parts may have similar face counts, but different feature tree dependency depths and local reconstruction propagation ranges. Existing complexity evaluation methods based on a single geometric or feature number have limitations in describing the overall structural burden of a 3D model.

[0004] Furthermore, the overall model complexity differs from the operational complexity of a single user action. Overall model complexity describes the background burden of the current document, while single-action complexity depends on the type of operation, the direct input object, and the scope of reconstruction triggered by dependencies. Treating both as a single value can easily lead to over-measurement of the entire model for local lightweight operations or underestimate operations such as Boolean operations, rounded corners, sweeping, and assembly updates that trigger large-scale impacts.

[0005] Existing edge-to-cloud offloading strategies often employ fixed rules based on device type, network thresholds, or task type, or simply compare the computing speed of the terminal and server. They fail to uniformly map the internal structure of the 3D model, the scope of operational impact, the scale of data transmission, and the cloud queuing status to the user-perceived completion time. Such strategies may fail to accurately determine the appropriate execution location when terminal performance is limited, network conditions fluctuate, the amount of data transmitted back is large, or the cloud load is high, leading to an increase in actual completion time. Existing edge-to-cloud scheduling methods primarily target general computing tasks or AI inference tasks, lacking mechanisms for describing the complexity of parametric 3D CAD model structures and the modeling process.

[0006] Therefore, there is a need for a method that can form a unified, interpretable, and rapidly updatable overall complexity from multiple layers of content such as sketches, features, topology, and assembly, and can locate the affected scope for the current operation, predict the completion time of different execution paths, and perform edge-cloud resource scheduling accordingly. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is to provide a method and system for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD. This method describes the overall structural burden of the cloud 3D CAD model and the incremental computational burden caused by a single modeling operation. It also predicts the completion time of multiple candidate execution paths by considering the execution environment, model cache status, and result submission risks, and selects the optimal execution plan accordingly. During execution, semantic checkpoints are used to achieve cross-node task migration, and consistency checks and actual time feedback are used to update prediction parameters, thus forming a closed loop of complexity assessment, path prediction, scheduling execution, and online calibration.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for evaluating the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD includes the following steps: Step S1: Obtain sketch objects, feature objects, boundary representation geometric topology objects, assembly objects, and the constraints, dependencies, adjacencies, inclusions, and mating relationships between objects in the cloud 3D CAD model; construct multi-layered associated objects with cross-layer relationships; and obtain the model baseline and command flow associated with the cloud 3D CAD model; the command flow records the operation type, operation object, parameter changes, model version, and execution status. Step S2: Extract sketch structure quantities, feature dependency quantities, geometric topology quantities, and assembly relation quantities from the multi-layered associated objects, and map each type of structural quantity into complexity components; generate a model state summary based on the model version, cross-layer relation state, and topology relation state. Step S3: Perform non-negative weight aggregation and bottleneck correction on the complexity components to obtain the overall complexity of the cloud 3D CAD model, and generate a complexity assessment confidence level based on the data field completeness rate, relationship establishment coverage rate, and contour or topological evidence status. Write the overall complexity, complexity assessment confidence level, and model status summary into the task description information. Step S4: Receive the 3D modeling operation to be executed, and generate an initial operation semantic signature based on the operation type, input object, reference object, parameter changes, and source model version; determine the directly acting object from the multi-layered associated objects, and extend it along at least one of feature dependency relationship, topological adjacency relationship, and assembly association relationship to form an operation influence subgraph; supplement the initial operation semantic signature based on the set of directly acting objects and the cross-layer relationship boundary information of the operation influence subgraph to obtain the operation semantic signature; Step S5: Generate an incremental reconstruction work package according to the dependency order of the operation influence subgraph, and calculate the effective complexity of the 3D modeling operation and the task resource requirement profile based on the operation influence subgraph and the incremental reconstruction work package; Step S6: Obtain terminal performance, network status, candidate computing node load, candidate computing node queuing status, model cache status, and the uplink and downlink data volume corresponding to the 3D modeling operation; Step S7: Input the overall complexity, effective operational complexity, task resource requirement profile, and running status information obtained in step S6 into the incremental reconstruction completion time prediction model to obtain the predicted completion time and result submission risk of different execution paths. Select the target execution plan based on the predicted completion time, result submission risk, resource constraints, and load-aware offloading margin. Step S8: Execute the corresponding 3D modeling operation according to the target execution plan; record semantic checkpoints at the end of the separable reconstruction phase, and when the migration conditions are met, migrate the unfinished phase to other execution locations according to the command flow record and semantic checkpoints; Step S9: Obtain the final result generated by the target execution plan, and perform consistency verification based on the model state summary, operation semantic signature, parameter change record, feature order and topology change summary carried by it; after the verification is passed, submit the final result, write the execution information into the command stream, record the expected time and actual time of each execution stage, and update the calibration parameters of the incremental reconstruction completion time prediction model according to the execution path, model cache status and topology change category.

[0009] Further, in step S1, the multi-layered associated objects include a sketch layer, a feature layer, a geometric topology layer, and an assembly layer; the sketch layer includes geometric primitive nodes, constraint nodes, contour object nodes, and edges interacting with primitives and constraints; the feature layer includes modeling feature nodes and parent-child dependency edges; the geometric topology layer includes volume, shell, face, loop, edge, and vertex nodes, as well as adjacency relationships; the assembly layer includes component instance nodes, hierarchical containment edges, and mating constraint edges; different layers are associated through generation relationships, reference relationships, or attribution relationships.

[0010] Further, in step S3, let the normalized complexity components corresponding to the sketch structure, feature dependencies, geometric topology, and assembly relationships be C1, C2, C3, and C4; the non-negative weights are determined according to the actual full model reconstruction time in the historical task records using the following formula, and equal weights are used as initial values ​​when historical samples are insufficient:

[0011] in, Indicates the optimal weight. This represents the k-th complexity component of the nth historical model. This represents the actual full model reconstruction time after normalization of the nth historical model; The overall complexity of the model is determined by the following formula:

[0012] in, Indicates the overall complexity of the model. This is the bottleneck correction factor. This is the bottleneck trigger threshold.

[0013] Furthermore, in step S4, the set of objects directly affected is... Starting from the operation type, select the set of relation edges that are allowed to propagate. And expand the influence of the subgraph layer by layer according to the following formula:

[0014] in, , The first sequence The collection of objects obtained after the second expansion Indicates from object To object Relationship edge; The expansion terminates when the sets of objects obtained from two adjacent expansions are the same, or when a feature branch that does not require reconstruction, does not depend on the current operation object, a stable topological boundary, or an assembly boundary that does not transfer degrees of freedom is reached; when the proportion of affected objects exceeds a preset threshold or an unresolved circular dependency occurs, global reconstruction is switched; size modification is extended along the constrained connected region and downstream features, fillets or chamfers are extended along adjacent faces and tangent continuous edge chains, Boolean operations are extended along the participating entities and their downstream references, and assembly updates are extended along the mate constraints that transfer pose or degrees of freedom.

[0015] Further, in step S4, the operation semantic signature includes at least the operation type, source model version, input object identifier, reference object identifier, parameter change summary, set of directly affected objects, and cross-layer relationship boundary information; the cross-layer relationship boundary information is used to represent the generation relationship, reference relationship, or assembly association relationship between the affected feature, topology object, or assembly object and the unaffected object.

[0016] Furthermore, in step S5, the incremental reconstruction work package is organized according to one or more reconstruction stages, including constraint solving, feature reconstruction, topology update, assembly solving, and result encapsulation; each reconstruction stage is associated with at least its input object, output object, preceding stage, expected resource requirements, and corresponding verification items, so that unaffected model objects do not participate in the repeated reconstruction of the current modeling operation. The effective complexity of the operation is determined based on the difficulty of the operation type, the weighted sum of nodes in the operation-affected subgraph, the weighted sum of relation edges, the depth of dependency propagation, the number of reconstruction stages to be executed, the expected topology change category, and the expected result increment.

[0017] Further, in step S6, the model cache state is determined based on whether there is a model baseline in the candidate computing nodes that matches the model state summary and geometric modeling kernel configuration, and whether the cache object set associated with the model baseline covers the input objects corresponding to the operation influence subgraph; the operation influence subgraph summary includes at least the set of affected object identifiers, relation edge types, dependency propagation boundaries, source model version, and identifier of the reconstruction stage to be executed; when the model cache state is cache hit, the prediction completion time of the cloud execution path is not included in the complete preparation time and state recovery time of the corresponding model baseline, and when the model cache state is cache miss, it is included in the model baseline preparation time and state recovery time.

[0018] Further, in step S7, the predicted completion time of the local execution path is calculated using the following formula:

[0019] in, The predicted completion time for the local execution path. The local side is used to predict the computation time for incremental reconstruction. The synchronization time for submitting local execution results to the model baseline; The predicted completion times for cache-hit and cache-miss execution paths are calculated using the following formula:

[0020] in, For the first The predicted total completion time for each cloud execution path. The cache value is 0 when the model cache is hit and 1 when it is not hit. For queuing time, For uplink transmission time, For the first The computation time for predictive incremental reconstruction of each candidate cloud node. For downlink transmission time, For the result submission verification time, Preparation time for model baseline For model state recovery time; The predicted completion time of the edge-cloud collaborative execution path is the sum of local preview time, queuing time, uplink transmission time, cloud node prediction incremental reconstruction calculation time, downlink transmission time, result submission verification time, and final result replacement time.

[0021] Further, in step S7, the risk of submitting the result is determined by the following formula:

[0022] in, The four risk components are: the difference between the source model version and the current model version, the change in the relationship at the boundary of the operation influence subgraph, the overlap of the influence domains of concurrent operations, and the prediction uncertainty. For the first The normalized weighting coefficients corresponding to each risk component; Load-aware offload margin Determine by the following formula:

[0023] in, , and These represent the candidate node queue load, network volatility, and prediction time error, respectively. , , , These are the weighting coefficients; The optimal cloud node is selected using the following formula:

[0024] Prioritize selecting the cache hit cloud execution path that meets resource constraints and has the shortest predicted completion time. If no cache hit path meets the above conditions, select the cloud node only if the predicted completion time of the cloud node with the shortest predicted completion time is greater than the predicted time saved by the local end than the load-aware offloading margin of the node, and the result submission risk of the node does not exceed the preset risk threshold. Otherwise, keep the execution on the local end.

[0025] Furthermore, when multiple users submit 3D modeling operations concurrently, conflict relationships in the influence domain are generated based on the operation semantic signature and operation influence subgraph of each operation. Parallel scheduling is allowed when the direct objects, affected objects, or cross-layer relationship boundaries of different operations do not overlap. When there is overlap, serialization, recalculation, or rollback processing is performed according to the source model version, operation sequence number, and pre-reconstruction relationship.

[0026] This invention also provides a cloud-based CAD-based 3D modeling complexity assessment and edge-cloud resource collaborative scheduling system for implementing the method. The system includes: The model object adaptation module is used to obtain sketches, feature objects, geometric topologies and assembly objects in cloud 3D CAD models, construct multi-layer related objects with cross-layer relationships and generate model state summaries; The complexity assessment module is used to extract structural information corresponding to sketch structure, feature dependencies, geometric topology and assembly relationships, and to calculate the overall complexity of the model, complexity components and complexity assessment confidence. The influence domain analysis and work package generation module is used to receive 3D modeling operations, generate operation semantic signatures, determine operation influence subgraphs, and generate incremental reconstruction work packages, operation effective complexity, and task resource requirement profiles according to the CAD parametric reconstruction order. The state awareness and cache recognition module is used to obtain terminal device performance, network status, cloud computing node utilization, task queue length, available resources and model cache status. The time consumption prediction and risk assessment module is used to predict the incremental reconstruction completion time and result submission risk of local, cache hit cloud, cache miss cloud and end-to-cloud collaborative execution paths based on the overall complexity of the model, the effective complexity of the operation, the task resource requirement profile and execution environment information. The scheduling decision module is used to select the target execution plan and target computing node based on the prediction completion time, result submission risk, model cache status, resource constraints and task priority, and to trigger task migration or node switching when the migration conditions are met. The local execution and preview module is used to execute modeling operations assigned to the local end on the terminal device, or to generate replaceable local preview results based on the directly affected object and the subgraph of operation influence. The cloud-based incremental reconstruction module is used to execute one or more reconstruction stages in the incremental reconstruction work package on candidate computing nodes in the cloud, and generate transferable semantic checkpoints at the end of the stage. The authoritative result submission module is used to synchronize the geometric results, model change information and operation status generated by the execution on the local end or in the cloud, and submit the final result to the target model version after verifying the consistency based on the model status summary, operation semantic signature, parameter change record and topology change summary. The log labeling module is used to record model cache status, target execution plan, reconstruction stage, estimated completion time, actual completion time, target computing node, resource consumption, and result submission anomalies, and to update the parameters of the incremental reconstruction completion time prediction model according to operation type, execution path, model cache status, and topology change category. The model data storage module is used to store model version, multi-level related objects, model complexity results, and operation influence domain information; The scheduling log storage module is used to store scheduling results, execution records, time prediction results, and parameter update records.

[0027] The beneficial effects of the above technical solution are as follows: First, this invention constructs a multi-layered associated object that includes sketches, modeling features, geometric topology, and assembly relationships, thereby achieving a comprehensive description of the structural complexity of a 3D model. This avoids evaluation based solely on a single geometric quantity or feature quantity, and improves the completeness and reliability of the complexity assessment results.

[0028] Second, this invention describes the actual impact range of modeling operations on the model by using operation semantic signatures, operation influence subgraphs, and incremental reconstruction work packages, distinguishing between the overall complexity of the model and the complexity of a single operation. This allows for the determination of affected objects, reconstruction stages, and resource requirements for specific modeling tasks, reducing unnecessary computation and data transmission.

[0029] Third, this invention considers execution paths such as local computation, cache hit in the cloud, cache miss in the cloud, and end-to-end cloud collaboration, and comprehensively considers computation time, transmission time, queuing time, result return time, and consistency verification time to select an execution plan with a better predicted completion time and that meets resource constraints, thereby improving the accuracy of end-to-end cloud scheduling.

[0030] Fourth, this invention establishes multi-user task conflict relationships based on the overlapping relationships between the influence domains of model operations, performs parallel scheduling for tasks without influence domain conflicts, and performs serial scheduling for tasks with conflicts according to version relationships, operation order, and prerequisite dependencies, thereby improving resource utilization and result consistency in multi-user collaborative modeling scenarios.

[0031] Fifth, by setting semantic checkpoints and execution result consistency verification, this invention supports task migration between different execution nodes, and updates the prediction model parameters based on actual execution time, queuing time, transmission time and result submission status, enabling the scheduling process to have continuous calibration and adaptive capabilities. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall process of the three-dimensional modeling complexity assessment and edge-cloud resource collaborative scheduling method of the present invention; Figure 2This is a schematic diagram of the structure of the multi-layered associated objects of the present invention; Figure 3 This is a schematic diagram illustrating the hierarchical calculation and aggregation of the overall complexity of this invention; Figure 4 This is a schematic diagram illustrating the modeling operation influence subgraph and the incremental reconstruction work package generation process of this invention. Figure 5 This is a schematic diagram illustrating the prediction of completion time for different cloud execution paths according to the present invention; Figure 6 This is a schematic diagram of the execution location selection and task migration process of the present invention; Figure 7 This is a schematic diagram of the module structure of the 3D modeling complexity assessment and edge-cloud resource collaborative scheduling system of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of this invention. All equivalent substitutions, improvements, and modifications made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0034] It should be noted that, unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0035] To facilitate understanding of this invention, the following definitions apply to several terms used in this application: Model baseline: refers to the state of complete steady-state model data with a defined model version number, including all sketch objects, feature objects, geometric topology objects, assembly objects and their cross-layer relationships under that version.

[0036] Model version: refers to the version number or version identifier string that uniquely identifies the status of the model data, and is updated with each successful submission of a modeling operation.

[0037] Current model status: Model data loaded and running in real time in the terminal, compute node memory, and session, which may include temporary edits that have not yet been submitted to generate a new model baseline.

[0038] Example 1

[0039] This embodiment proposes a method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD, applied to a cloud 3D CAD platform. This method converts the current model state and pending operations into verifiable model state summaries and operation semantic signatures, respectively. Then, it organizes incremental reconstruction work packages based on the operation impact subgraph. The system compares not only the computation speed of local and cloud environments but also the model cache state, result submission risks, and migration costs at each reconstruction stage to select the local, cache-hit-cloud, cache-miss-cloud, or edge-cloud coordinating execution plan.

[0040] In a preferred embodiment, the present invention is deployed in a cloud-based 3D CAD platform. The platform may include a user interaction environment, a model data management environment, a complexity assessment environment, a task scheduling environment, a cloud computing environment, and a result synchronization environment. These environments collaborate through interface calls, data exchange, or messaging mechanisms to achieve 3D model complexity assessment, modeling task time prediction, edge-cloud resource allocation, task execution, and result feedback.

[0041] The following is combined with Figures 1 to 6 The overall process of the method of the present invention will be further described in detail.

[0042] Step S1: Construct multi-level related objects.

[0043] like Figure 1 and Figure 2 As shown, the system obtains the 3D CAD model created or opened by the user in the cloud 3D CAD platform, and parses the model to obtain sketch objects, geometric feature objects, boundary representation geometric topology objects, assembly objects, parameter objects, constraint objects, and feature history relationships.

[0044] The boundary represents a geometric topology object used to describe the body, shell, face, ring, edge, and vertex of a 3D CAD entity, as well as their compositional relationships, adjacency relationships, trimming relationships, and containment relationships. The assembly object includes component instances, assembly levels, mating constraints, assembly connection areas, and external reference relationships. The system organizes the above objects and their relationships into sketch layers, feature layers, geometric topology layers, and assembly layers, forming cross-layer associations through generation relationships, reference relationships, or attribution relationships; and acquires the model baseline and command stream data associated with the cloud 3D CAD model.

[0045] The command stream records the operation type, operation object, parameter changes, model version, and execution status in the order in which the modeling operations occur, and is used for model state recovery, task migration recovery, and eventual consistency verification.

[0046] In a preferred embodiment, the system maintains the multi-layered associated objects incrementally. When only local parameters of the model are modified, only the nodes and relationships related to the modified object are updated; when the model is imported, the file is reopened, or the entire system is rebuilt, the global object relationships are reconstructed and the model version identifier is recorded.

[0047] Specifically, the sketch layer includes geometric primitive nodes, constraint nodes, contour object nodes, and edges connecting primitives and constraints. The feature layer includes modeling feature nodes and parent-child dependency edges, specifically including one or more of the following: extrusion features, rotation features, sweep features, loft features, hole features, fillet features, chamfer features, shell features, array features, mirror features, Boolean features, and direct editing features. It may also include other parametric modeling or direct modeling operations, as well as parent-child relationships and dependencies between features. The geometric topology layer includes solids, faces, loops, edges, vertices, surface types, topological adjacency relationships, trim boundaries, and display data. The assembly layer includes component instances, assembly levels, mating constraints, assembly connection areas, flexible subassemblies, and external reference relationships. Different layers are associated through generation relationships, reference relationships, or attribution relationships.

[0048] Step S2: Extract complexity evaluation information and calculate complexity components.

[0049] like Figure 3 As shown, the system extracts sketch structure quantities, feature dependencies, geometric topology quantities, and assembly relation quantities from the multi-layered associated objects, and maps each type of structural quantity to complexity components. Simultaneously, it generates a model state summary for subsequent validation based on the model version, cross-layer relation status, and topology relation status. The model state summary includes at least the model version, feature order, key topology, parameter summary, and assembly status. Further, the model state summary is jointly generated from the source model version, cross-layer relation status, key topology relation status, and object status related to the current task, and is used for consistency verification during migration and final result submission.

[0050] Extract the number of valid primitives, primitive types, number of closed contours, number of inner contours, contour nesting levels, number of constraints, constraint types, number of constraint-related primitives, and degree of relationship coupling in the sketch to determine the structural complexity of the sketch; Extract the number of modeling features, feature types, parent-child dependencies, feature dependency depth, maximum dependency region, feature reconstruction propagation range, and external reference relationships to determine the feature dependency complexity. Extract the number of entities, faces, cycles, edges, vertices, surface types, topological adjacency counts, trimmed boundary counts, and displayed triangle counts to determine the geometric topological complexity; Extract the number of component instances, the number of unique components, the assembly tree depth, the number of mate constraints, the assembly connected regions, the number of flexible subassemblies, and the number of external references to determine the complexity of the assembly relationship; The complexity of the sketch structure, feature dependency complexity, geometric topology complexity, and assembly relationship complexity are uniformly scored and weighted to obtain the overall complexity of the model, and the complexity level, main sources of complexity, and complexity assessment confidence level are output.

[0051] In one feasible method for calculating complexity components, the system sets a corresponding reference scale for each type of structural quantity. As the structural quantity increases, a standardized complexity component is generated based on the growth rate of the structural quantity relative to the reference scale. This ensures that the complexity component increases monotonically, and the rate of increase gradually decreases as the structural quantity continues to increase, thereby preventing a small number of abnormally large objects from distorting the overall complexity.

[0052] The system assigns non-negative weights to components of different complexity and performs weighted summation according to preset weights. When a component is significantly higher than other components, the influence of that component on the overall complexity is increased, forming a bottleneck correction to reflect the limiting effect on modeling and computation when feature dependencies are too deep, assembly constraints are too dense, or topology pruning is too complex.

[0053] Step S3: Aggregate to form the overall complexity and generate the complexity assessment confidence level.

[0054] The system forms the overall model complexity based on the complexity components obtained in step S2, and divides the overall complexity into low, medium, and high complexity, or into multiple consecutive levels. The source information of the overall complexity includes at least the sketches, features, topology, or assembly components that have the main influence.

[0055] In one specific implementation, the sketch structure complexity, feature dependency complexity, geometric topology complexity, and assembly relation complexity are denoted as C1, C2, C3, and C4, respectively, and each complexity component is normalized to a value between 0 and 1. The actual full model reconstruction time in historical task records is normalized to the same scale and used as... Non-negative weights are obtained according to formula (1). When there are insufficient historical valid samples, the weights of all four categories are taken as 0.25 as the initial value and are re-determined after new valid records are accumulated.

[0056] (1) in, Indicates the optimal weight. This represents the k-th complexity component of the nth historical model. This represents the actual full model reconstruction time after normalization of the nth historical model; The overall complexity of the model is calculated according to equation (2): (2) in, This is the bottleneck correction factor. The bottleneck trigger threshold is determined based on the historical task verification error, so as to minimize the prediction error between the overall model complexity and the actual full model reconstruction time.

[0057] The overall complexity is used to describe the background structural burden generated by the current model editing, and is used to correct the background time consumption generated by model state management, cache access, version synchronization and global data maintenance. It does not participate in the geometric computation calculation corresponding to the incremental reconstruction work package; the operational effective complexity is used to describe the incremental burden corresponding to the subgraph affected by the current operation.

[0058] The system simultaneously assesses confidence based on structural data field completeness, cross-layer relationship coverage, contour information completeness, and the validity of topological adjacency evidence. Missing fields are marked as missing and are not replaced with zero values; when critical topological or assembly data is missing, the system lowers the confidence level and outputs corresponding prompts in the task description information.

[0059] In practice, the complexity of the sketch structure, feature dependency complexity, geometric topology complexity, and assembly relationship complexity are uniformly scored and weighted to obtain the overall complexity of the model, and the complexity level, main sources of complexity, and complexity assessment confidence level are output.

[0060] In one numerical implementation, after a model is standardized, the four complexity components—sketching, features, topology, and assembly—are at medium, medium-high, medium-high, and low levels, respectively. The system performs weighted summation based on preset weights and then performs bottleneck correction based on the highest component to obtain a medium-high level overall complexity. If the structural field completeness rate is high, the relation coverage is complete, and the contour and topological evidence are valid, a higher confidence level is generated; otherwise, the overall complexity level is retained and its confidence level is reduced.

[0061] The system writes the overall complexity, complexity level, complexity components, main sources of complexity, and complexity assessment confidence into the task description information of the current cloud CAD model, which is used for subsequent effective complexity analysis and completion time prediction.

[0062] Step S4: Generate operation semantic signatures and form an operation influence subgraph.

[0063] like Figure 4As shown, the system receives a 3D modeling operation to be executed, obtains the operation type, input object, reference object, and operation parameters, and determines the direct object from the multi-layered associated objects. For dimension modification operations, the direct object is the dimension being modified and its attached sketch or feature; for fillet, chamfer, and Boolean operations, the direct object is the edge, face, or solid being operated on; for assembly update operations, the direct object is the modified component or mating constraint.

[0064] The system generates an initial operation semantic signature based on the operation type, input object, reference object, parameter change, and source model version. After the operation influence subgraph is formed, it supplements the set of directly affected objects and cross-layer relationship boundary information. In addition to identifying the directly affected objects, the operation semantic signature also records the cross-layer relationship boundary between the affected and unaffected ranges and is used to determine the expansion start point and stopping boundary of the following operation influence subgraph.

[0065] Starting from the directly affected object, the system extends along at least one of the following: feature parent-child dependency relationship, geometric topological adjacency relationship, and assembly association relationship. Objects that may be resolved, reconstructed, updated, or synchronized are added to the operation influence subgraph, and the relationship type and propagation layer number traversed during the extension are recorded.

[0066] The operation influence subgraph refers to a local association subgraph formed by starting from the direct object of the current modeling operation and extending along feature dependencies, topological adjacency relationships, and assembly association relationships.

[0067] For local parameter editing, the system only extends to the affected downstream features and adjacent geometric objects; for model import, file reopening, global reconstruction, or large-scale assembly updates, the system treats the corresponding global objects or global dependency closures as the affected scope.

[0068] Different methods for determining the operation influence subgraph are used for different types of 3D modeling operations: For size modification operations, the affected objects are determined along the features to which the modified size is attached and the downstream dependent features; For stretch, revolve, sweep, and loft operations, the affected objects are determined based on the input sketch, generated features, and subsequent dependent features. For fillet, chamfer, and Boolean operations, the affected objects are determined based on the edges, faces, entities being operated on, and their topological adjacency relationships. For assembly update operations, the affected objects are determined based on the modified parts, mating constraints, and assembly connectivity areas. For model import, file reopening, and global reconstruction operations, the entire or a large range of objects in the current model are identified as the affected objects.

[0069] In one specific implementation, the system uses the direct target as... Select the set of relation edges based on the current operation type. And expand the influence subgraph layer by layer according to equation (3); (3) Objects already included in the operation influence subgraph will not be expanded again. For boundary objects whose impact is uncertain, conservative rules will be applied to include them in the operation influence subgraph.

[0070] The expansion terminates when the set of objects obtained from two consecutive expansions no longer changes, or when the expansion reaches objects that do not need to be rebuilt, independent feature branches, stable topological boundaries, or assembly boundaries that do not transfer degrees of freedom; when the number of affected objects exceeds a preset proportion of the total number of model objects or when unresolvable circular dependencies occur, the current task is converted into a global reconstruction task.

[0071] The set of relational edges for size modification operations includes constraint connectivity, feature generation relationships, and downstream dependencies; the set of relational edges for fillet or chamfer operations includes adjacent face relationships, tangent continuous edge chains, and topological reference relationships; the set of relational edges for Boolean operations includes participating entities, Boolean results, and downstream reference relationships; and the set of relational edges for assembly update operations includes mating constraints that can transmit pose or degrees of freedom.

[0072] The operation semantic signature includes at least the operation type, source model version, input object identifier, reference object identifier, parameter change summary, set of directly affected objects, and cross-layer relationship boundary information. The cross-layer relationship boundary information is used to represent the generation relationship, reference relationship, or assembly association relationship between the affected feature, topology object, or assembly object and the unaffected object.

[0073] Step S5: Generate the incremental reconstruction work package and calculate the effective complexity of the operation.

[0074] The system generates an incremental reconstruction work package based on the dependency order of the operation influence subgraph, and calculates the effective complexity of the 3D modeling operation and the task resource requirement profile based on the operation influence subgraph and the incremental reconstruction work package.

[0075] Specifically, based on the operation influence subgraph, the system first organizes the task into one or more incremental reconstruction stages according to the pre-dependencies: constraint solving, feature reconstruction, topology updating, assembly solving, and result encapsulation. Each reconstruction stage records the input objects, output objects, expected resource requirements, expected result increments, and verification items, forming an incremental reconstruction work package. The incremental reconstruction work package is organized according to the cloud CAD parametric reconstruction process. Subsequently, the system counts the number of nodes, the number of relation edges, the types of nodes and relation edges, the dependency propagation depth, the number of reconstruction stages to be executed, and the expected topology change category, and determines the effective complexity of the operation by combining the difficulty coefficient of the current operation type.

[0076] Furthermore, the system breaks down the current modeling task into one or more incremental reconstruction stages based on the pre-dependencies in the operation impact subgraph, including constraint solving, feature reconstruction, topology updating, assembly solving, and result encapsulation, forming an incremental reconstruction work package. Each reconstruction stage is associated with at least its input objects, output objects, pre-stages, expected resource requirements, and corresponding validation items to ensure that unaffected model objects do not participate in the repeated reconstruction of the current modeling operation. The effective complexity of the operation is determined based on the difficulty of the operation type, the weighted statistics of affected nodes and relation edges, the depth of dependency propagation, the number of reconstruction stages to be executed, the expected topology change category, and the expected result increment.

[0077] The task resource requirement profile includes at least processor computing requirements, memory requirements, graphics processing requirements, storage read / write requirements, interaction response level, and result submission deadline. The processor computing requirements, memory requirements, graphics processing requirements, and storage read / write requirements are generated based on the projected resource requirements for each reconstruction stage in the incremental reconstruction work package. The interaction response level and result submission deadline are determined based on the operation type, user interaction status, or preset service level.

[0078] In one possible implementation, the system assigns non-negative weights to sketch nodes, feature nodes, topology nodes, assembly nodes, and different types of dependency edges. It then performs weighted statistics on the nodes and relational edges in the operation-affected subgraph, and further adjusts the propagation depth, the number of reconstruction stages to be executed, the type of topology change, and the expected result increment. The effective complexity of the operation only describes the incremental solution and reconstruction burden caused by the current operation and is not fully included in the overall model complexity again.

[0079] Step S6: Obtain execution environment information and model cache status.

[0080] like Figure 5 As shown, terminal performance, network status, candidate computing node load, candidate computing node queuing status, model cache status, and the uplink and downlink data volume corresponding to the 3D modeling operation are obtained.

[0081] Terminal device performance is determined at least based on one or more of the following: processor performance, available memory, logical concurrency, graphics processing capability, and historical modeling operation time. Network status is determined at least based on average round-trip latency, latency fluctuation, quantile latency, effective uplink throughput, and effective downlink throughput. The status of candidate computing nodes in the cloud is determined at least based on processor utilization, available memory, number of active tasks, task queue length, current concurrency, and node response status.

[0082] Specifically, the system estimates the uplink model baseline data volume, incremental work package data volume, downlink geometric result data volume, and state synchronization data volume based on the model state summary, operation impact subgraph summary, and current operation. It also checks whether candidate cloud nodes have cached matching model baselines, associated objects, and intermediate reconstruction results. Available memory in the terminal browser is used to determine if the local end meets the execution conditions, and available memory in the cloud nodes is used to determine if the cloud nodes meet the carrying conditions; both are recorded separately.

[0083] Furthermore, the model cache state is determined based on whether there is a model baseline in the candidate computing nodes that matches the model state summary and geometric modeling kernel configuration, and whether the cached object set associated with the model baseline covers the input objects corresponding to the operation influence subgraph. The operation influence subgraph summary includes at least the set of affected object identifiers, relation edge types, dependency propagation boundaries, source model versions, and identifiers of the reconstruction stage to be executed.

[0084] Step S7: Predict the path completion time, result submission risk, and select an execution plan.

[0085] The incremental reconstruction completion time prediction model is established based on historical modeling task execution records, and prediction parameters are maintained according to operation type, model cache status, topology change category, and execution path. The system inputs the overall complexity, effective operation complexity, task resource requirement profile, and terminal, network, and cloud node status into the prediction model to predict the completion time of local terminal, cache hit cloud, cache miss cloud, and terminal-cloud collaborative execution path, and assesses the risk of result submission; the target execution plan is selected by combining the predicted completion time, result submission risk, resource constraints, and load-aware offloading margin.

[0086] The risk of submitting results is generated based on one or more of the following: differences between the source model version and the current model version, changes in relationships at the boundaries of the subgraph affected by operations, overlap of the influence domains of concurrent operations, and prediction uncertainty.

[0087] The incremental reconstruction completion time prediction model adopts a non-negative linear regression structure and predicts the computation time for different execution paths according to equation (4):

[0088] Where r represents the execution path, to These represent the overall model complexity, effective operational complexity, proportion of affected objects, number of reconstruction stages to be executed, processor requirements, memory requirements, time consumed per unit of terminal workload, terminal load, candidate node load, candidate node queue length, estimated uplink transmission time, and estimated downlink transmission time, respectively. Model parameters are determined through training on historical samples and validated using the mean absolute percentage error and high quantile absolute error of independent validation samples.

[0089] In one specific implementation, each historical modeling task execution record that passes the final result consistency check constitutes a training sample. The sample format is {task identifier, operation type, model cache state, topology change category, execution path, feature vector x, actual completion time t, sample state}, where the feature vector x is a 12-dimensional real number vector, and the actual completion time t is the total time taken from receiving the final result from the task to passing the consistency check. The operation type, model cache state, topology change category, and execution path serve as the sample grouping conditions.

[0090] Before training, all input features are normalized, and the samples are divided into training, validation, and test sets according to the order of completion time. Regression parameters are determined using the training set, regularization parameters are selected using the validation set, and the model's prediction accuracy is verified using the mean absolute percentage error and the 90th percentile absolute error of the test set, ensuring that the model's prediction error is within an acceptable range. If a group has insufficient samples, shared parameters from the same execution path and model cache state are used.

[0091] The predicted completion times for the local execution path and the cloud execution path are calculated according to Equation (5) and Equation (6), respectively.

[0092] (5) (6) in, The predicted completion time for the local execution path. The local side is used to predict the computation time for incremental reconstruction. The synchronization time for submitting local execution results to the model baseline; For the first The predicted total completion time for each cloud execution path. The cache value is 0 when the model cache is hit and 1 when it is not hit. For queuing time, For uplink transmission time, For the first The computation time for predictive incremental reconstruction of each candidate cloud node. For downlink transmission time, For the result submission verification time, Preparation time for model baseline For model state recovery time; When the model cache status is hit, the prediction completion time of the cloud execution path is not included in the corresponding model baseline preparation time and state recovery time. When the model cache status is miss, it is included in the model baseline preparation time and state recovery time.

[0093] The predicted completion time of the edge-cloud collaborative execution path is the sum of local preview time, queuing time, uplink transmission time, cloud node prediction incremental reconstruction calculation time, downlink transmission time, result submission verification time, and final result replacement time.

[0094] Risk of submitting results Determine according to formula (7): (7) in, The four risk components are: the difference between the source model version and the current model version, the change in the relationship at the boundary of the operation influence subgraph, the overlap of the influence domains of concurrent operations, and the prediction uncertainty. For the first The normalized weight coefficients corresponding to each risk component; each risk component is normalized to 0 to 1, and the weights are... Determined by historical consistency verification records.

[0095] Load-aware offload margin Determine according to formula (8): (8) in, , and These represent the candidate node queue load, network volatility, and prediction time error, respectively. , , , The weighting coefficient is determined by the historical task verification error; when the node queuing increases, the network fluctuation increases, or the prediction error increases, the unloading margin increases accordingly.

[0096] The optimal cloud node that satisfies the resource constraints is selected according to equation (9): (9) When the 3D modeling operation is a sketch, attribute query, selection highlight, short preview or other preset lightweight operation, it is preferred to execute it on the local end; when the network risk reaches the preset level and the expected amount of returned data exceeds the preset data amount threshold, it is preferred to execute it on the local end.

[0097] For tasks that need to be executed in the cloud, the cache hit cloud execution path that meets resource constraints and has the shortest predicted completion time is selected first. If there is no cache hit path that meets the above conditions, the cloud node is selected only if the predicted time saving of the cloud node with the shortest predicted completion time relative to the local end is greater than the load-aware offloading margin of the node, and the result submission risk of the node does not exceed the preset risk threshold. Otherwise, the execution is kept on the local end.

[0098] When multiple candidate cloud computing nodes exist, the target cloud computing node is selected based on the model cache status, prediction completion time, current load, task queue, available processor resources, available memory resources, and task priority. Under the premise of meeting resource constraints, for nodes that have cache hits and model state summaries, even if their processor utilization is slightly higher, as long as the prediction completion time is short and the risk of result submission is acceptable, they can still be used as target nodes.

[0099] During task execution, when the estimated remaining time on the local end increases significantly, the load on the cloud node exceeds the preset range, the network condition deteriorates, or the risk of result submission increases significantly, the system re-predicts the remaining completion time and migrates the unfinished computation stages to other execution locations according to the migration conditions; the migration carries a model state summary, operation semantic signature, completed stages, and intermediate result summaries.

[0100] When multiple users submit 3D modeling operations concurrently, the conflict relationship of the influence domain is generated based on the operation semantic signature and operation influence subgraph of each operation; when the direct objects, affected objects or cross-layer relationship boundaries of different operations do not overlap, parallel scheduling is allowed; when there is overlap, serialization, recalculation or rollback processing is performed according to the source model version, operation sequence number and reconstruction stage prerequisite relationship.

[0101] Steps S8 to S9: Execution, submission of authoritative results, and online calibration.

[0102] Step S8: The system performs 3D modeling operations at the target execution location according to the incremental reconstruction work package, and records semantic checkpoints at the end of each reconstruction stage with complete semantic boundaries and independent division. When migration conditions are met, the incomplete stages are migrated to other execution locations based on the command flow records and semantic checkpoints. The semantic checkpoints include at least a model state summary, operation semantic signatures, completed stages, and intermediate result summaries.

[0103] In one specific implementation, the semantic checkpoint includes task identifier, model identifier, source model version, target model version, operation sequence number, current stage identifier, operation semantic signature, set of completed stages, set of stages to be executed, topology mapping, intermediate result reference, generation time and status flag.

[0104] Checkpoint metadata is stored in the scheduling log storage module, and intermediate results are stored in the model data storage module. At the end of the rebuild phase, the system writes a temporary checkpoint, which is marked as valid after consistency checks on model version, operation semantic signature, phase status, and intermediate result references. During migration recovery, the most recent valid checkpoint is loaded and the pending phase continues execution; if verification fails, it reverts to the previous valid checkpoint; if no valid checkpoint exists, the model baseline is loaded and execution is re-executed according to the command flow.

[0105] Step S9: Obtain the final result generated by the target execution plan, and perform a consistency check based on the model state summary, operation semantic signature, parameter change record, feature order, and topology change summary it carries. After the check passes, submit the final result, write the operation type, object identifier, parameter change, model version, execution location, and execution status into the command stream, record the estimated and actual execution times of each execution stage, and update the calibration parameters of the incremental reconstruction completion time prediction model according to the execution path, model cache status, and topology change category. For the end-cloud collaborative path, the local preview result is replaced by the final result in the cloud after the check passes.

[0106] The execution record in step S9 is associated with the model baseline, incremental reconstruction work package, and semantic checkpoint. It includes at least the operation type, object identifier, model version, model state summary, operation semantic signature, overall model complexity, effective operation complexity, model cache status, target execution plan, estimated and actual time consumption of each reconstruction stage, resource consumption, execution status, and reasons for result submission or migration anomalies. It is used for model state recovery, task migration recovery, and tracing of model change processes.

[0107] Based on the execution records, the time consumption prediction parameters are classified and updated according to the execution path, model cache status, operation type, complexity level, topology change category, terminal device category, network status level, and cloud node load level; the time consumption components such as queuing, data preparation, transmission, calculation, result verification, and preview replacement are calibrated separately to avoid miscounting time consumption anomalies caused by cache misses, network fluctuations, or result conflicts into geometric calculation capabilities.

[0108] The system records model cache status, target execution plan, incremental reconstruction phase, estimated completion time, actual completion time, target execution location, target computing node, processor and memory usage, data transfer duration, recalculation count, task migration status, and reasons for result submission anomalies. When the deviation between actual and predicted time continuously exceeds a preset range, the system calibrates the parameters corresponding to data preparation, transmission, computation, verification, and preview replacement according to execution path, cache status, operation type, complexity range, topology change category, terminal category, network status level, and node load level.

[0109] To verify the feasibility of the above method, a small-scale verification was conducted based on the existing cloud CAD front-end and back-end code, focusing on complexity assessment, operational impact analysis, and edge-cloud scheduling processes. The existing scheduling module can receive model complexity, effective operational complexity, terminal status, network status, cloud node load, and model cache status, calculate the predicted completion time for different candidate execution paths, and output local, cloud, or edge-cloud collaborative execution plans. It also records the execution path, actual completion time, and consistency verification results.

[0110] The complexity assessment selected three sketch samples with progressively increasing structural size. Samples S1, S2, and S3 contain 4, 11, and 68 primitives, respectively, and 2, 10, and 80 constraints, respectively. Each sample was preheated 500 times and then continuously evaluated 5000 times. The resulting complexity scores were 12.474, 41.998, and 88.718, corresponding to simple, medium, and very complex levels, respectively. The 95th percentile of the longest single evaluation time for the longest sample was less than 0.3 ms, indicating that the complexity assessment results can reflect the changes in sketch structural size and constraint relationships, and meet the requirements of online evaluation.

[0111] The operational impact analysis module was validated using 44 regression test cases, covering scenarios such as disjoint operations, read-write dependencies, write conflicts, dangling references, topological invariant conflicts, Boolean operation topology mapping, and reference migration. All 44 test cases passed. Simultaneously, the input-output process of the scheduling module was validated by setting different network states, node loads, and model cache states, checking whether its candidate path prediction, execution plan selection, and state change response conformed to preset scheduling rules. This validation demonstrates that the scheduling process can actually run, and the predicted time output by the scheduling module is not directly used as the actual performance test result for the edge-cloud connection.

[0112] Further performance verification can be conducted under the same terminal, network, cloud nodes, and modeling task set, comparing the full local execution strategy, fixed threshold offloading strategy, and the scheduling strategy described in this invention, respectively. Actual completion time, predicted completion time, execution path, cache status, node load, network status, and consistency check results should be recorded. Average completion time, high-quantile completion time, proportion of timed-out tasks, prediction error, and path selection accuracy should also be compared. Relevant performance results are based on the actual operation logs of the cloud CAD platform and should not be replaced by simulated task time or prediction results.

[0113] Example 2

[0114] Combination Figure 1 , Figure 4 and Figure 5This section describes the modification of local parameters in a single part model. The system obtains the current model version, the operation to be executed, and the operation object information, determines the feature objects, associated geometric entities, and subsequent dependent objects involved in the operation, and generates the corresponding operation semantic signature and operation influence subgraph.

[0115] Since this operation did not cause any changes to the assembly relationships and its impact was concentrated on local features and their associated surfaces, the system generated a local incremental reconstruction work package and predicted the completion time of different execution paths based on terminal computing power, network status, and cloud load. When the local execution path meets the time constraints, the calculation is completed locally; otherwise, the cached cloud execution path is selected.

[0116] After the task is completed, the system performs consistency checks on the model version, parameter changes, and local topology results, and records the execution results and actual time consumption for subsequent prediction parameter updates.

[0117] Example 3

[0118] like Figure 5 As shown in the figure, this embodiment illustrates the edge-cloud collaborative execution process of a medium-complexity model.

[0119] In this embodiment, the user performs operations to add hole features and modify the array quantity on a part model containing multiple feature objects. The system first obtains the number of features, feature tree level, parameter constraint relationships, number of hole features, and the range of geometric entities affected by the array features in the model, and determines that the model has medium complexity based on this.

[0120] The system identifies that adding hole features affects the topological relationships of local faces, edges, and volumes, and modifying the array quantity will cause multiple identical features to be regenerated, but this operation does not affect the overall assembly model. Based on historical records of similar operations, the system predicts that this task will involve a relatively short local interaction processing time and a relatively long background feature reconstruction time.

[0121] To ensure users can see operation feedback promptly, the system first generates a preview of the hole features locally, while simultaneously submitting the complete feature reconstruction, array calculation, and topology update tasks to the cloud computing node for execution. After the cloud completes the calculations, it returns the updated feature tree, geometric entities, and parameter status to the local machine.

[0122] After receiving the results from the cloud, the local end performs a consistency check on the parameters, feature order, and geometric topology of the returned model. If the check passes, the local end replaces the preview results with the cloud results and writes the execution process, estimated time, and actual time of this end-to-cloud collaborative execution into the scheduling log.

[0123] Example 4

[0124] like Figure 6As shown in the figure, this embodiment illustrates the cloud resource scheduling process for a highly complex assembly model.

[0125] In this embodiment, the user opens a 3D CAD assembly model containing multiple components, complex assembly constraints, and multi-level feature history relationships, and modifies the dimensional parameters of one key component. The system obtains the number of assembly components, assembly level, number of mating relationships, feature history depth, and the range of affected components, determining that the overall complexity of the model is high.

[0126] The system further identified that this parameter modification would not only affect the feature reconstruction of the current component, but also the positional relationships of adjacent components, assembly constraint states, and local interference check results. Based on historical task records, the system predicted that this task would require a long model reconstruction time and significant computational resources.

[0127] The system acquires local processing capacity, cloud node load, task queue length, network status, and current user task priority, and selects a suitable target node from multiple cloud computing nodes. When the target node's cache is hit, only parameter change data and incremental reconstruction work packages are sent; when the cache is missed, the model baseline and the data required to restore the current model state are sent or loaded simultaneously.

[0128] After completing component reconstruction, assembly constraint updates, and interference relationship calculations, the cloud node returns the model update results to the local node. If the target node experiences excessive load or abnormal response during execution, the system will transfer the unfinished portion of the task to other available cloud nodes for continued execution, based on the completed calculation stages of the task.

[0129] After receiving the final results locally, the system performs consistency checks on the model version, component locations, assembly constraints, and interference check results. If the checks pass, the system updates the local model and saves the cloud scheduling process, resource node information, task migration information, and actual time consumed.

[0130] Example 5

[0131] Combination Figure 7 This paper describes task scheduling in a multi-user collaborative modeling scenario. The system obtains the model version, operation semantic signature, operation impact subgraph, prediction completion time and resource requirements submitted by each user, and obtains the processing capacity, model cache status, task queue and concurrency usage of candidate execution nodes.

[0132] The system establishes task conflict relationships based on the overlap of objects, topological associations, and version dependencies between subgraphs affected by different user operations. For tasks whose affected objects do not overlap and have no version dependencies, they are assigned to different execution nodes for parallel processing; for tasks with overlapping affected objects or version dependencies, they are scheduled serially according to model version, operation order, and prerequisite dependencies.

[0133] When the load on the execution node exceeds resource constraints or the estimated completion time of the task exceeds a set threshold, the system will relocate the task to other cache nodes, local endpoints, or end-to-cloud collaborative paths for execution. After the task is completed, the system will perform version consistency, topology consistency, and operation order consistency checks on the results of each user's model; if the checks pass, the collaborative model status will be updated, and the task allocation, queuing, migration, execution, and check results will be recorded.

[0134] Example 6

[0135] This embodiment provides a cloud-based 3D modeling complexity assessment and edge-cloud resource collaborative scheduling system to implement the method described in Embodiment 1 above.

[0136] like Figure 7 As shown, the system includes a model object adaptation module, a complexity assessment module, an influence domain analysis and work package generation module, a state awareness and cache identification module, a time consumption prediction and risk assessment module, a scheduling decision module, a local execution and preview module, a cloud incremental reconstruction module, an authoritative result submission module, a log calibration module, a model data storage module, and a scheduling log storage module.

[0137] The model object adaptation module acquires sketches, feature objects, geometric topologies, and assembly objects from the cloud 3D CAD model, constructs multi-layered associated objects with cross-layer relationships, and generates a model state summary.

[0138] The complexity assessment module extracts structural information corresponding to sketch structure, feature dependencies, geometric topology and assembly relationship, determines non-negative weights according to equation (1), calculates the overall complexity of the model according to equation (2), and outputs complexity components and complexity assessment confidence.

[0139] The influence domain analysis and work package generation module receives 3D modeling operations, generates operation semantic signatures, determines operation influence subgraphs, and generates incremental reconstruction work packages, effective operation complexity, and task resource requirement profiles according to the CAD parametric reconstruction sequence. Specifically, this module uses the set of directly affected objects... Starting from the operation type, select the set of relation edges that are allowed to propagate. The operation influence subgraph is formed by expanding layer by layer according to equation (3).

[0140] The state awareness and cache identification module acquires terminal device performance, network status, cloud computing node utilization, task queue length, available resources, and model cache status. The model cache status is determined based on whether a model baseline exists among the candidate computing nodes that matches the model state summary and geometric modeling kernel configuration.

[0141] The time consumption prediction and risk assessment module predicts the incremental reconstruction completion time of the local end, cache hit cloud, cache miss cloud and end-to-cloud collaborative execution path according to the overall complexity of the model, the effective complexity of the operation, the task resource requirement profile and the execution environment information, and calculates the risk of each path according to the formula (7).

[0142] The scheduling decision module determines the load-aware offloading margin according to Equation (8) and selects the optimal cloud node according to Equation (9) based on the predicted completion time, result submission risk, model cache status, resource constraints, and task priority, and finally determines the target execution plan and target computing node. When the predicted remaining completion time or the execution node status meets the migration conditions, the scheduling decision module triggers task migration or node switching.

[0143] The local execution and preview module executes the modeling operations assigned to the local end on the terminal device, or generates replaceable local preview results based on the directly affected object and the subgraph of operation influence.

[0144] The cloud-based incremental reconstruction module executes one or more reconstruction stages in the incremental reconstruction work package on candidate computing nodes in the cloud, and generates transferable semantic checkpoints at the end of each stage. The semantic checkpoints include at least the task identifier, model version, operation semantic signature, completed stage, pending stage, topology mapping, intermediate result reference, and status flag.

[0145] The authoritative result submission module synchronizes the geometric results, model change information, and operation status generated by local or cloud execution. It then performs consistency verification on the final result based on the model status summary, operation semantic signature, parameter change record, and topology change summary. Once verification is successful, the final result is submitted to the target model version. For edge-cloud collaborative execution paths, the final result replaces the local preview result after successful verification.

[0146] The log calibration module records the model cache status, target execution plan, reconstruction stage, estimated completion time, actual completion time, target computing node, resource usage, and result submission anomalies, and updates the parameters of the incremental reconstruction completion time prediction model according to operation type, execution path, model cache status, and topology change category.

[0147] The model data storage module stores the model version, multi-level related objects, model complexity results, and operation influence domain information.

[0148] The scheduling log storage module stores scheduling results, execution records, time prediction results, and parameter update records.

[0149] In the context of edge-cloud collaborative execution, the local execution and preview module generates a local preview result based on the directly affected object and the subgraph affected by local operations; the cloud incremental reconstruction module executes the complete geometric reconstruction task and generates semantic checkpoints at the end of the separable reconstruction phase; the authoritative result submission module performs consistency verification on the model version, parameter changes, feature order and topological relationship in the cloud returned result, and replaces the local preview result with the final result after the verification is passed.

[0150] The system described in this embodiment corresponds to the method in Embodiment 1. The collaborative work of each module realizes a complete closed loop of cloud CAD 3D modeling complexity assessment, edge-cloud path prediction, scheduling decision, task migration and online calibration.

[0151] Finally, it should be noted that any parts of this invention not described in detail are prior art. Those skilled in the art will understand that the above descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for evaluating the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD, characterized in that, Includes the following steps: Step S1: Obtain sketch objects, feature objects, boundary representation geometric topology objects, assembly objects, and the constraints, dependencies, adjacencies, inclusions, and mating relationships between objects in the cloud 3D CAD model; construct multi-layered associated objects with cross-layer relationships; and obtain the model baseline and command flow associated with the cloud 3D CAD model; the command flow records the operation type, operation object, parameter changes, model version, and execution status. Step S2: Extract sketch structure quantities, feature dependency quantities, geometric topology quantities, and assembly relation quantities from the multi-layered associated objects, and map each type of structural quantity into complexity components; generate a model state summary based on the model version, cross-layer relation state, and topology relation state. Step S3: Perform non-negative weight aggregation and bottleneck correction on the complexity components to obtain the overall complexity of the cloud 3D CAD model, and generate a complexity assessment confidence level based on the data field completeness rate, relationship establishment coverage rate, and contour or topological evidence status. Write the overall complexity, complexity assessment confidence level, and model status summary into the task description information. Step S4: Receive the 3D modeling operation to be executed, and generate an initial operation semantic signature based on the operation type, input object, reference object, parameter changes, and source model version; determine the directly acting object from the multi-layered associated objects, and extend it along at least one of feature dependency relationship, topological adjacency relationship, and assembly association relationship to form an operation influence subgraph; supplement the initial operation semantic signature based on the set of directly acting objects and the cross-layer relationship boundary information of the operation influence subgraph to obtain the operation semantic signature; Step S5: Generate an incremental reconstruction work package according to the dependency order of the operation influence subgraph, and calculate the effective complexity of the 3D modeling operation and the task resource requirement profile based on the operation influence subgraph and the incremental reconstruction work package; Step S6: Obtain terminal performance, network status, candidate computing node load, candidate computing node queuing status, model cache status, and the uplink and downlink data volume corresponding to the 3D modeling operation; Step S7: Input the overall complexity, effective operational complexity, task resource requirement profile, and running status information obtained in step S6 into the incremental reconstruction completion time prediction model to obtain the predicted completion time and result submission risk of different execution paths. Select the target execution plan based on the predicted completion time, result submission risk, resource constraints, and load-aware offloading margin. Step S8: Execute the corresponding 3D modeling operation according to the target execution plan; record semantic checkpoints at the end of the separable reconstruction phase, and when the migration conditions are met, migrate the unfinished phase to other execution locations according to the command flow record and semantic checkpoints; Step S9: Obtain the final result generated by the target execution plan, and perform consistency verification based on the model state summary, operation semantic signature, parameter change record, feature order and topology change summary carried by it; After verification, the final result is submitted, the execution information is written to the command stream, the estimated time and actual time of each execution stage are recorded, and the calibration parameters of the incremental reconstruction completion time prediction model are updated according to the execution path, model cache status and topology change category.

2. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S1, the multi-layered associated object includes a sketch layer, a feature layer, a geometric topology layer, and an assembly layer; the sketch layer includes geometric primitive nodes, constraint nodes, contour object nodes, and interaction edges between primitives and constraints; The feature layer includes modeling feature nodes and parent-child dependency edges; the geometric topology layer includes volume, shell, face, loop, edge and vertex nodes and adjacency relationships; the assembly layer includes component instance nodes, hierarchical containment edges and mating constraint edges; different layers are associated with each other through generation relationships, reference relationships or attribution relationships.

3. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S3, let the normalized complexity components corresponding to the sketch structure, feature dependencies, geometric topology, and assembly relationships be C1, C2, C3, and C4; the non-negative weights are determined according to the actual full model reconstruction time in the historical task records using the following formula, and equal weights are used as initial values ​​when historical samples are insufficient: in, Indicates the optimal weight. This represents the k-th complexity component of the nth historical model. This represents the actual full model reconstruction time after normalization of the nth historical model; The overall complexity of the model is determined by the following formula: in, Indicates the overall complexity of the model. This is the bottleneck correction factor. This is the bottleneck trigger threshold.

4. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S4, the set of objects directly affected is... Starting from the operation type, select the set of relation edges that are allowed to propagate. And expand the influence of the subgraph layer by layer according to the following formula: in, , The first sequence The collection of objects obtained after the second expansion Indicates from object To object Relationship edge; The expansion terminates when the sets of objects obtained from two adjacent expansions are the same, or when a feature branch that does not require reconstruction, does not depend on the current operation object, a stable topological boundary, or an assembly boundary that does not transfer degrees of freedom is reached; when the proportion of affected objects exceeds a preset threshold or an unresolved circular dependency occurs, global reconstruction is switched; size modification is extended along the constrained connected region and downstream features, fillets or chamfers are extended along adjacent faces and tangent continuous edge chains, Boolean operations are extended along the participating entities and their downstream references, and assembly updates are extended along the mate constraints that transfer pose or degrees of freedom.

5. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S4, the operation semantic signature includes at least the operation type, source model version, input object identifier, reference object identifier, parameter change summary, set of directly affected objects, and cross-layer relationship boundary information; the cross-layer relationship boundary information is used to represent the generation relationship, reference relationship, or assembly association relationship between the affected feature, topology object, or assembly object and the unaffected object.

6. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S5, the incremental reconstruction work package is organized according to one or more reconstruction stages, including constraint solving, feature reconstruction, topology update, assembly solving, and result encapsulation. Each reconstruction stage is associated with at least its input object, output object, preceding stage, expected resource requirements, and corresponding verification items, so that unaffected model objects do not participate in the repeated reconstruction of the current modeling operation. The effective complexity of the operation is determined based on the difficulty of the operation type, the weighted sum of nodes in the operation-affected subgraph, the weighted sum of relation edges, the depth of dependency propagation, the number of reconstruction stages to be executed, the expected topology change category, and the expected result increment.

7. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S6, the model cache state is determined based on whether there is a model baseline in the candidate computing nodes that matches the model state summary and geometric modeling kernel configuration, and whether the cache object set associated with the model baseline covers the input objects corresponding to the operation influence subgraph. The operation influence subgraph summary includes at least the set of affected object identifiers, relation edge types, dependency propagation boundaries, source model version, and identifier of the reconstruction stage to be executed. When the model cache state is cache hit, the prediction completion time of the cloud execution path is not included in the complete preparation time and state recovery time of the corresponding model baseline. When the model cache state is cache miss, the model baseline preparation time and state recovery time are included.

8. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, In step S7, the predicted completion time of the local execution path is calculated using the following formula: in, The predicted completion time for the local execution path. The local side is used to predict the computation time for incremental reconstruction. The synchronization time for submitting local execution results to the model baseline; The predicted completion times for cache-hit and cache-miss execution paths are calculated using the following formula: in, For the first The predicted total completion time for each cloud execution path. The cache value is 0 when the model cache is hit and 1 when it is not hit. For queuing time, For uplink transmission time, For the first The computation time for predictive incremental reconstruction of each candidate cloud node. For downlink transmission time, For the result submission verification time, Preparation time for model baseline For model state recovery time; The predicted completion time of the edge-cloud collaborative execution path is the sum of local preview time, queuing time, uplink transmission time, cloud node prediction incremental reconstruction calculation time, downlink transmission time, result submission verification time, and final result replacement time.

9. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 8, characterized in that, In step S7, the risk of submitting the result is determined by the following formula: in, The four risk components are: the difference between the source model version and the current model version, the change in the relationship at the boundary of the operation influence subgraph, the overlap of the influence domains of concurrent operations, and the prediction uncertainty. For the first The normalized weighting coefficients corresponding to each risk component; Load-aware offload margin Determine by the following formula: in, , and These represent the candidate node queue load, network volatility, and prediction time error, respectively. , , , These are the weighting coefficients; The optimal cloud node is selected using the following formula: Prioritize selecting the cache hit cloud execution path that meets resource constraints and has the shortest predicted completion time. If no cache hit path meets the above conditions, select the cloud node only if the predicted completion time of the cloud node with the shortest predicted completion time is greater than the predicted time saved by the local end than the load-aware offloading margin of the node, and the result submission risk of the node does not exceed the preset risk threshold. Otherwise, keep the execution on the local end.

10. The method for assessing the complexity of 3D modeling and coordinating edge-cloud resource scheduling based on cloud CAD according to claim 1, characterized in that, When multiple users submit 3D modeling operations concurrently, the conflict relationship of the influence domain is generated based on the operation semantic signature and operation influence subgraph of each operation; when the direct objects, affected objects or cross-layer relationship boundaries of different operations do not overlap, parallel scheduling is allowed; when there is overlap, serialization, recalculation or rollback processing is performed according to the source model version, operation sequence number and reconstruction stage prerequisite relationship.

11. A cloud-based CAD-based 3D modeling complexity assessment and edge-cloud resource collaborative scheduling system, used to implement the method described in any one of claims 1 to 10, characterized in that, include: The model object adaptation module is used to obtain sketches, feature objects, geometric topologies and assembly objects in cloud 3D CAD models, construct multi-layer related objects with cross-layer relationships and generate model state summaries; The complexity assessment module is used to extract structural information corresponding to sketch structure, feature dependencies, geometric topology and assembly relationships, and to calculate the overall complexity of the model, complexity components and complexity assessment confidence. The influence domain analysis and work package generation module is used to receive 3D modeling operations, generate operation semantic signatures, determine operation influence subgraphs, and generate incremental reconstruction work packages, operation effective complexity, and task resource requirement profiles according to the CAD parametric reconstruction order. The state awareness and cache recognition module is used to obtain terminal device performance, network status, cloud computing node utilization, task queue length, available resources and model cache status. The time consumption prediction and risk assessment module is used to predict the incremental reconstruction completion time and result submission risk of local, cache hit cloud, cache miss cloud and end-to-cloud collaborative execution paths based on the overall complexity of the model, the effective complexity of the operation, the task resource requirement profile and execution environment information. The scheduling decision module is used to select the target execution plan and target computing node based on the prediction completion time, result submission risk, model cache status, resource constraints and task priority, and to trigger task migration or node switching when the migration conditions are met. The local execution and preview module is used to execute modeling operations assigned to the local end on the terminal device, or to generate replaceable local preview results based on the directly affected object and the subgraph of operation influence. The cloud-based incremental reconstruction module is used to execute one or more reconstruction stages in the incremental reconstruction work package on candidate computing nodes in the cloud, and generate transferable semantic checkpoints at the end of the stage. The authoritative result submission module is used to synchronize the geometric results, model change information and operation status generated by the execution on the local end or in the cloud, and submit the final result to the target model version after verifying the consistency based on the model status summary, operation semantic signature, parameter change record and topology change summary. The log labeling module is used to record model cache status, target execution plan, reconstruction stage, estimated completion time, actual completion time, target computing node, resource consumption, and result submission anomalies, and to update the parameters of the incremental reconstruction completion time prediction model according to operation type, execution path, model cache status, and topology change category. The model data storage module is used to store model version, multi-level related objects, model complexity results, and operation influence domain information; The scheduling log storage module is used to store scheduling results, execution records, time prediction results, and parameter update records.

Images