A virtual machine tool simulation data processing system and method for numerical control machining teaching

Abstract

This application discloses a virtual machine tool simulation data processing system and method for CNC machining teaching. The system includes: a teaching management module for distributing machining tasks and standard 3D mesh models; a simulation data acquisition module for acquiring machining data generated by students during simulation and the completed 3D mesh model data in real time; an automatic scoring module for identifying manufacturing features based on the 3D mesh model, sequentially performing ICP spatial registration, multi-level comparison, and multi-level scoring, weighted fusion, and outputting the final score; receiving various parameters input by students, calling a process expert system to perform parameter comparison analysis and cutting force safety assessment to obtain rationality analysis results; and a report generation module for automatically generating a structured teaching feedback report based on the scoring results, a list of matching failure features, parameter deviation items, and process rationality analysis results. This application can optimize the teaching process for CNC machining and improve teaching quality.

Description

Technical Field

[0001] This application relates to the field of teaching data processing technology, specifically to a virtual machine tool simulation data processing system and method for CNC machining teaching. Background Technology

[0002] With the rapid development of the manufacturing industry, CNC machining technology is playing an increasingly important role in industrial production. CNC machining education, as a crucial component in cultivating professional talent, has also received widespread attention. The emergence of virtual machine tool simulation software has provided a new means of CNC machining education, allowing students to practice CNC programming and simulated machining in a virtual environment, improving their operational skills and practical abilities. This teaching method not only saves on the cost of raw materials and equipment required for actual machining but also allows students to practice repeatedly in a safe environment. At the same time, virtual machine tool simulation software also provides teachers with a more convenient teaching tool, enabling a more intuitive demonstration of the machining process and results.

[0003] In traditional CNC machining instruction, teachers primarily distribute drawings and tasks to students via USB drives. This method is inefficient and prone to file loss or corruption. For students, after completing the simulation machining, the lack of effective recording methods makes the machining process like a "black box," making it difficult to trace back and pinpoint the specific error if problems arise. Furthermore, when grading students' work, teachers often rely on visual observation of the parts' appearance and manual measurement. This method is not only inefficient but also prone to significant differences in subjective judgment among teachers, leading to inconsistent grading standards. Students receive only a general score, unable to pinpoint specific operational shortcomings or make targeted improvements.

[0004] The shortcomings of these traditional teaching methods make it difficult to improve teaching efficiency and greatly reduce students' learning outcomes, failing to meet the requirements of modern CNC machining teaching for high quality and high efficiency. Summary of the Invention

[0005] To meet the high-quality and high-efficiency requirements of modern CNC machining teaching, optimize the CNC machining teaching process, and improve teaching quality, this application provides a virtual machine tool simulation data processing system and method for CNC machining teaching.

[0006] In a first aspect, this application provides a virtual machine tool simulation data processing system for CNC machining teaching, comprising: The teaching management module, simulation data acquisition module, automatic scoring module, and report generation module are interconnected. The teaching management module is used to distribute processing tasks and standard 3D mesh models to student terminals; The simulation data acquisition module is used to acquire, in real time, the machining path data, tool trajectory data, material removal sequence data, and the final 3D mesh model data generated by the student during the simulation process by calling the open interface of the virtual machine tool software. The automatic scoring module includes a manufacturing feature extraction unit, a model comparison unit, a comprehensive scoring unit, and a process rationality analysis unit. The manufacturing feature extraction unit identifies manufacturing features based on a 3D mesh model and outputs the geometric primitive parameters and semantic types of each feature. The model comparison unit sequentially performs ICP spatial registration, multi-level comparison, and multi-level scoring on the standard 3D mesh model and the student 3D mesh model. The multi-level comparison includes global geometric statistics comparison, feature-level semantic comparison, and voxel distance field comparison. The comprehensive scoring unit performs weighted fusion of the multi-level scores and outputs the final score. The process rationality analysis unit receives machining condition parameters input by the student and tool trajectory parameters actually used in the simulation. It then calls a process expert system containing a material database, rule reasoning engine, parameter calculation engine, and cutting force analysis module to perform parameter comparison analysis and cutting force safety assessment to obtain the rationality analysis results. The report generation module is used to automatically generate a structured teaching feedback report based on the scoring results, the list of matching failure features, parameter deviation items, and the results of process rationality analysis.

[0007] By adopting the above scheme, the design of the teaching management module efficiently distributes processing tasks and standard 3D mesh models to student terminals, improving the efficiency of the distribution method; the design of the simulation data acquisition module acquires various data in real time during the student simulation process, facilitating subsequent analysis; the design of the automatic scoring module identifies manufacturing characteristics and outputs parameters and types, and achieves objective and accurate evaluation of student processing results through multi-level comparison and scoring; the design of the process rationality analysis unit calls upon the process expert system to conduct parameter comparison analysis and cutting force safety assessment, obtaining rationality analysis results; the design of the report generation module automatically generates structured teaching feedback reports based on relevant results, thus making the entire teaching process verifiable and data-driven, enabling teachers to quickly understand student situations and provide targeted feedback, improving teaching efficiency and quality.

[0008] Preferably, the manufacturing feature extraction unit is used to perform RANSAC robust fitting, normal clustering segmentation and curvature feature description sequentially based on a triangular mesh model to identify manufacturing features including holes, slots, bosses, surfaces, steps, cavities and composite features.

[0009] By adopting the above scheme, a triangular mesh model is designed, and operations such as RANSAC robust fitting, normal clustering segmentation and curvature feature description are used to effectively identify multiple types of manufacturing features, improve the accuracy and comprehensiveness of manufacturing feature identification, and provide a more reliable foundation for subsequent model comparison and comprehensive scoring.

[0010] Preferably, the process rationality analysis unit is also used to receive the machining condition parameters input by the student, the actual tool trajectory parameters used in the simulation, and the defect features of mismatch or deviation exceeding the limit identified in the model comparison, and call the process expert system including the material database, rule reasoning engine, parameter calculation engine, cutting force analysis module and correlation diagnosis module to perform parameter comparison analysis, cutting force safety assessment and feature-process correlation diagnosis to obtain the process rationality analysis results of the student simulation process; The process expert system comprises several components. The rule-based reasoning engine in the system infers adjustment factors by reasoning step-by-step based on material category fundamental factor rules, material physical property perception rules, material-specific rule sets, and general process rules. The parameter calculation engine calculates theoretical cutting parameters based on the adjustment factors and machine tool capability constraints, providing theoretical cutting parameter references for students during machining simulations and comparing them with actual simulation parameters to generate deviation indicators. The cutting force analysis module provides a cutting force calculation model, calculates the machine tool's cutting force based on user-defined parameters, and compares it with the machine tool's maximum thrust, maximum torque, and rated power to generate a safety level. The feature-process correlation diagnosis uses a neural network algorithm to correlate and analyze defect features identified in model comparisons with the actual tool trajectory parameters and machining condition parameters associated with the corresponding machining program segment, thus pinpointing problems in student simulation machining processes and tool usage.

[0011] By adopting the above scheme, based on the analysis of process rationality, the design combines the machining condition parameters input by students, the actual tool trajectory parameters, and the defect characteristics in the model comparison, and calls the process expert system to conduct a comprehensive parameter comparison analysis, cutting force safety assessment, and feature-process correlation diagnosis. This allows for the precise identification of problems students encounter in the simulated machining process and tool usage, providing more targeted feedback for teaching.

[0012] Preferably, the model comparison unit is further used to perform multi-level comparisons between the standard 3D mesh model and the student 3D mesh model. The comparison process includes: calling a preset processing type-geometric statistical quantum set mapping table and feature template according to the processing type corresponding to the distributed processing task, and dynamically selecting to perform a complete three-level comparison or a simplified level comparison according to the size of the task queue to be scored; the complete three-level comparison includes: at the global level, selecting an appropriate geometric feature statistical quantum set according to the processing type, calculating the deviation of the two models on the subset to generate a global geometric similarity score; at the feature level, extracting appropriate features from the two models to select the processing type, calculating single feature similarity, and adaptively selecting feature matching based on the Hungarian algorithm or feature matching based on the spatial nearest neighbor of the KD tree according to the number of features, and calculating a feature-level semantic matching score; at the voxel level, converting the two models into voxel space representations, calculating the voxel intersection-union ratio and Chamfer distance to generate a voxel similarity score; The comprehensive scoring unit is also used to obtain a weight matrix of processing skill points associated with the current teaching progress. The weight matrix assigns weights to different comparison levels and different manufacturing features that are adjusted according to the teaching stage. The weight matrix is ​​used to perform weighted fusion of global geometric similarity score, feature-level semantic matching score and voxel similarity score to calculate a comprehensive score that is suitable for the current teaching progress and processing type.

[0013] By adopting the above scheme, the model comparison unit calls the mapping table and template according to the processing task, selects the comparison method according to the size of the queue of tasks to be scored, and efficiently performs multi-level comparison of the standard and student 3D mesh models, generating corresponding similarity scores at the global layer, feature layer and voxel layer respectively; the comprehensive scoring unit obtains the processing skill point weight matrix associated with the teaching progress, and uses this matrix to calculate the comprehensive score by weighting and fusing the scores at different levels, so that the score is adapted to the current teaching progress and processing type, providing more accurate evaluation results for teaching.

[0014] Preferably, the automatic scoring module further includes: a behavior sequence mining and analysis unit connected to the process rationality analysis unit; The behavior sequence mining and analysis unit is used to extract the student's operation event flow on the virtual machine based on the processing path data generated by the student during the simulation, construct the operation behavior sequence and extract operation fluency and risk behavior indicators; associate the operation behavior sequence with the defect feature-process after feature-process correlation diagnosis and align them by time to form a defect feature-process-behavior correlation evidence chain; and call the causal diagnosis model in the process expert system according to the evidence chain to obtain the causal analysis results of the student's simulation process defects. The causal diagnosis model uses a neural network algorithm to perform causal analysis on machining task problems with a comprehensive score lower than a preset comprehensive score based on the defect features identified in the model comparison, the actual tool trajectory parameters associated with the machining program segments corresponding to the defect features, and the operation behavior sequence data, and to determine the causal result of the machining task problem. The report generation module is also used to automatically generate a structured teaching feedback report based on the scoring results, the list of matching failure features, parameter deviation items, the results of process rationality analysis, and the causal results of processing task problems.

[0015] By adopting the above scheme, the behavior sequence mining and analysis unit is designed to extract student operation event flow, construct operation behavior sequence and extract relevant indicators, form a defect feature-process-behavior correlation evidence chain, and call the causal diagnosis model to obtain the causal analysis results of student simulation process defects, which helps to accurately locate the cause of student processing task problems; then the report generation module generates a structured teaching feedback report based on more results, making the teaching feedback more comprehensive and accurate.

[0016] Preferably, the report generation module is further used to summarize the structured teaching feedback reports of each member in the student group specified by the teaching management module, extract the common matching failure feature types and parameter deviation items in the reports, and generate a group teaching diagnosis report when the frequency of the same feature type or parameter deviation item in the student group exceeds a preset frequency threshold; the group teaching diagnosis report includes common error types, processing skill points and corresponding teaching improvement suggestions.

[0017] By adopting the above scheme, the report generation module summarizes the structured teaching feedback reports of each member in the student group, extracts the common matching failure feature types and parameter deviation items, and generates a group teaching diagnostic report when the frequency of occurrence exceeds the preset frequency threshold. This can help teachers discover common problems in the student group, identify the processing skill points involved, and then put forward targeted suggestions for improving the teaching process.

[0018] Preferably, the manufacturing feature extraction unit is further configured to perform multi-level collaborative manufacturing feature extraction and enhancement, which includes: before identifying manufacturing features, calculating the higher-order geometric moments of the triangular mesh model and extracting global features of manufacturing orientation, including the size distribution of the main processing direction and the theoretical material removal rate; when identifying feature geometric primitives, firstly using a graph neural network or an attention-based model to perform preliminary feature classification of the mesh patches, and then using the preliminary feature classification to adapt to the corresponding geometric rules for boundary correction and segmentation, and outputting feature regions with semantic labels.

[0019] By adopting the above scheme, the higher-order geometric moments of the triangular mesh model and the global features of manufacturing orientation are calculated before identifying manufacturing features, providing more comprehensive reference information for subsequent feature identification; a graph neural network or an attention-based model is used to initially classify the mesh patches and combine geometric rules for boundary correction and segmentation, thereby improving the accuracy of feature region identification.

[0020] Secondly, this application provides a virtual machine tool simulation data processing method for CNC machining teaching, which utilizes the aforementioned system and includes: Distribute processing tasks and standard 3D mesh models to student terminals; By calling the open interface of the virtual machine tool software, the machining path data, tool trajectory data, material removal sequence data and the final 3D mesh model data generated by the student during the simulation can be obtained in real time. Based on the 3D mesh model, manufacturing features are identified and the geometric primitive parameters and semantic types of each feature are output. ICP spatial registration, multi-level comparison, and multi-level scoring are performed sequentially on the standard 3D mesh model and the student 3D mesh model. The multi-level scores are weighted and fused to output the final score. The system receives machining condition parameters and actual tool trajectory parameters input by students, and calls a process expert system containing a material database, rule reasoning engine, parameter calculation engine and cutting force analysis module to perform parameter comparison analysis and cutting force safety assessment to obtain rationality analysis results. Based on the scoring results, the list of matching failure features, parameter deviation items, and the results of process rationality analysis, a structured teaching feedback report is automatically generated.

[0021] By adopting the above scheme, the distribution of machining tasks and standard 3D mesh models can be completed, various data in the student simulation process can be acquired in real time, manufacturing features can be identified and model comparison and scoring can be completed, the rationality of the process can be analyzed through the process expert system, and finally a structured teaching feedback report can be generated. This solves the problems of low task distribution efficiency, no recording of machining process, difficulty in scoring and inconsistent standards, and lack of targeted feedback for students in traditional CNC machining teaching, making the entire teaching process verifiable.

[0022] Thirdly, this application provides a computer-readable storage medium including a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to perform the method described above.

[0023] Fourthly, this application provides a computer device, the computer device including a memory, a processor and a program stored in the memory and executable thereon, the program being executed by the processor to implement the steps of the method described above.

[0024] In summary, this application has the following beneficial effects: 1. The teaching management module, simulation data acquisition module, automatic scoring module, and report generation module work together. The teaching management module can distribute processing tasks and standard 3D mesh models to student terminals, solving the problem of low efficiency in traditional distribution methods; the simulation data acquisition module acquires student simulation data in real time; the manufacturing feature extraction unit of the automatic scoring module identifies manufacturing features, performs multi-level comparison and scoring, and outputs the final score, solving the problems of time-consuming and inconsistent standards in traditional scoring; the process rationality analysis unit calls the process expert system to evaluate the rationality of parameters and assists the report generation module in generating feedback reports to provide targeted feedback to students. 2. The model comparison unit calls the preset mapping table and feature template according to the processing task, and dynamically selects the complete three-layer comparison or the simplified layer comparison. The complete three-layer comparison performs similarity scoring from the global layer, feature layer and voxel layer. The comprehensive scoring unit obtains the weight matrix associated with the teaching progress, and weights and merges the scores of each layer to calculate the comprehensive score that is suitable for the current teaching progress and processing type. This allows for a more accurate evaluation of students' processing results based on the actual teaching situation, and improves the accuracy and adaptability of the scoring. 3. The behavior sequence mining and analysis unit in the automatic scoring module can construct operation behavior sequences based on student simulation processing data, extract relevant indicators, associate them with defect features and processes to form an evidence chain, call the causal diagnosis model to perform defect causal analysis, determine the causal results of processing task problems, and the report generation module generates a structured teaching feedback report based on multiple results, which can deeply analyze the root causes of students' problems in the simulation process and improve teaching quality. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the virtual machine tool simulation data processing system for CNC machining teaching described in a specific embodiment; Figure 2 This is a schematic diagram of the structure of a virtual machine tool simulation data processing system for CNC machining teaching described in a specific embodiment, showing how a student performs CNC machining simulation operations based on parameter recommendations. Figure 3 This is a flowchart of the virtual machine tool simulation data processing method for CNC machining teaching described in a specific embodiment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] like Figure 1As shown in the illustration, this application discloses a virtual machine tool simulation data processing system for CNC machining teaching. Specifically, it includes a teaching management module 100, a simulation data acquisition module 200, an automatic scoring module 300, and a report generation module 400. These modules are interconnected and work together to efficiently distribute various tasks in CNC machining teaching, acquire and analyze student simulation data, and improve teaching efficiency and the quality of machining tasks and standards. Each specific module will be described in detail below.

[0028] First, the teaching management module 100 is used to distribute processing tasks and standard 3D mesh models to student terminals.

[0029] Specifically, the teaching management module 100 includes a task distribution unit 110 and a model distribution unit 120. The task distribution unit 110 is used by teachers to set up processing tasks and distribute them to various student terminals. It adopts a lightweight RESTful service architecture based on the HTTP / 2 protocol, with its core being a task queue manager with a transaction rollback mechanism. It supports the import of processing task templates with JSON Schema validation, embedding processing type identifiers, skill point codes, standard model hash values, and version control fields in the templates.

[0030] Model distribution unit 120 is used to distribute standard 3D mesh models. It adopts a dual-caching strategy. The main cache is deployed on a local NAS server to store the standard 3D mesh model after format conversion and normal consistency verification. The secondary cache is deployed in the teacher's terminal memory to preload the vertex index tree and bounding box hierarchy (BVH) of the model required for the current lesson. The two are synchronized with zero copy through memory-mapped files.

[0031] Secondly, the simulation data acquisition module 200 includes an interface calling unit 210 and a data acquisition unit 220.

[0032] Specifically, the interface calling unit 210 is used to call the open interface of the virtual machine bed software. In this embodiment, the virtual machine bed software adopts the Siemens SINUMERIK ONE / 828D system and uses the OPCUA protocol to exchange data with the SINUMERIK ONE / 828D system, capturing data from the system in real time for intelligent teaching.

[0033] The data acquisition unit 220 is used to collect machining data generated by students during the simulation in real time, including machining path data, tool trajectory data, material removal sequence data, and the final 3D mesh model data. The data acquisition unit adopts a circular buffer structure. Its writing end is driven by the event callback function of the interface calling unit, and its reading end is periodically polled by the feature extraction unit of the automatic scoring module. The buffer depth is set to 2048 frames, and each frame stores a data packet for a single event. The header of the data packet contains an event type marker, timestamp, data length, and CRC32 checksum, while the tail contains the raw binary data.

[0034] Then, the automatic scoring module 300 includes a manufacturing feature extraction unit 310, a model comparison unit 320, a comprehensive scoring unit 330, and a process rationality analysis unit 340.

[0035] The manufacturing feature extraction unit 310 is used to identify manufacturing features based on a three-dimensional mesh model and output the geometric primitive parameters and semantic types of each feature. In this embodiment, the manufacturing feature extraction unit 310 is used to sequentially perform RANSAC robust fitting, normal clustering segmentation, and curvature feature description based on a triangular mesh model to identify manufacturing features including holes, slots, bosses, curved surfaces, steps, cavities, and composite features (such as fillets / chamfers, keyways, conical surfaces, spherical surfaces, countersunk holes, stiffeners, and flanges).

[0036] Among them, the triangular mesh model , For vertex set, Let D be a set of triangular facets; define the diagonal length D = , , For the model bounding box focus points, all distance thresholds are defined using relative diagonal lengths. Planarity determination: For a point set P, construct a centered covariance matrix: If its smallest eigenvalue If P is coplanar, then it is considered that P is coplanar. Normal and area, surface normal. The area is obtained by normalizing the cross product of the triangle vertices; the area is calculated by summing the triangulations. The distance from a point to the triangle is solved using the central coordinate method.

[0037] The system can be configured with RANSAC fitting sub-units, normal clustering sub-units, and curvature description sub-units (vertex-level geometric feature pre-extraction). The curvature description sub-unit is used to pre-extract vertex-level geometric features. It takes the original triangular mesh vertices, triangular facet topology, and adjacency relationships as input, determines the surface type based on curvature classification rules, and assigns a geometric attribute label to each vertex, serving as a multi-dimensional feature vector for subsequent clustering and segmentation. The normal clustering sub-unit is used to group facet regions. It can construct multi-dimensional feature vectors based on curvature, normal vectors, and 3D coordinates to perform coarse mesh region division, splitting the entire mesh into several continuous homogeneous facets. The RANSAC fitting sub-unit is used to perform accurate surface modeling. For each homogeneous facet obtained from clustering, it performs robust RANSAC fitting, solves the analytical geometric equations, determines the precise surface type, and outputs manufacturing feature parameters, achieving precise geometric parameter solving at the facet level.

[0038] The specific steps for extracting basic features include: First, for hole features, a dual-path complementary strategy is adopted: one path obtains the axis and radius parameters through RANSAC cylindrical fitting, and the other path constructs an undirected graph by extracting grid boundary edges and searching for closed loops, calculating the loop center and equivalent radius; the hole features obtained from the two paths are merged and deduplicated based on the center distance and radius difference, retaining the one with higher confidence. Second, for groove features, PCA planarity verification is performed on the face set to segment planar regions, and pairs of parallel planes with approximately opposite normals are searched as the upper and lower surfaces of the groove. Third, for boss features, regions are segmented by clustering along the face normal, a convex hull is constructed for the region point set, and convexity is determined based on the ratio of the region area to the convex hull area; then, geometric primitives are identified for the convex region through RANSAC planar fitting or cylindrical fitting. Fourth, for curved surface features. For each triangular facet, a neighborhood normal set is constructed, and the approximate values ​​of mean curvature and Gaussian curvature are calculated. MeanShift clustering is used to divide the surface region based on the 2D curvature features. Then, RANSAC plane fitting and cylindrical fitting are performed on each region sequentially. Unfitted regions are marked as freeform surfaces. Fifth, for step features, PCA is performed on the vertex set to obtain three principal axis directions. Vertices are projected along each principal axis direction, and MeanShift clustering is used to separate different height layers. Sixth, for cavity features, the distance from the center of each facet to the model's convex hull is calculated as concavity. Concave regions are clustered using DBSCAN, and the center, depth, and area are extracted. Simultaneously, boundary closed loops are extracted, and the projected area and normal within the loops are calculated.

[0039] For composite feature extraction, such as rounded / beveled features, the adjacent faces and their included angles are provided by `mesh.face_adjacency`. The included angles are then filtered. For edge features, small transition surfaces with an area less than a threshold between two adjacent faces are identified. RANSAC screen fitting is performed on these transition surfaces to obtain chamfers and fillets. For keyway features: RANSAC cylinder detection is used to identify shaft / hole structures. Radial distance anomalies are searched on the cylinder surface to calculate external shaft keyways and internal hole keyways. PCA is performed on the candidate point set, aspect ratios are calculated, and the convex hull area ratio is used to distinguish rectangular, single-arc, double-arc, and through-slot keyways. For conical features: RANSAC cone fitting is used; 5 points are randomly sampled, and PCA is used to estimate the initial axis and vertex. The half-cone angle is estimated using the median. For spherical features: RANSAC spherical fitting is used, selecting points in each iteration and solving a system of linear equations to find the sphere center. For other composite features: such as countersunk holes: coaxial with a radius difference... The two holes are paired; the algorithms are combined to automatically extract common machining features from the triangular mesh model and output numerical descriptions.

[0040] The model comparison unit 320 is used to sequentially perform ICP spatial registration, multi-level comparison, and multi-level scoring on the standard 3D mesh model and the student 3D mesh model. In this embodiment, model spatial registration is achieved in advance using the Iterative Closest Point (ICP) algorithm; then, similarity is quantified at three levels: global geometric statistics, local manufacturing feature matching, and voxel / distance field; finally, a weighted fusion is used to give a comprehensive score.

[0041] This includes the implementation of ICP space registration.

[0042] For a given standard 3D mesh model and The point-to-point iterative nearest point algorithm is used to... Registration to Coordinate system. N=1000 points are uniformly sampled from each of the two models to obtain point clouds. and The following steps are executed iteratively: KD-tree nearest neighbor matching: ; Calculate the centroid and decentralize it: ; ; ; Construct the covariance matrix and perform SVD decomposition: ; ; Calculate the rotation matrix: ,like Then, the last column of V is inverted to ensure that the determinant of the rotation matrix is ​​positive.

[0043] Calculate the translation vector: Construct the transformation matrix: ; and act on Point cloud, up to average matching distance Or it may reach the maximum number of iterations.

[0044] This embodiment addresses multi-level alignment and multi-level scoring. Multi-level alignment includes global geometric statistics alignment, feature-level semantic alignment, and voxel distance field alignment. The following sections elaborate on the alignment process and scoring for each alignment level.

[0045] For global geometric statistics comparison, a simplified geometric comparison method is adopted. Specifically, this includes comparing the basic geometric statistics of the two models at a global scale, including vertex count, face count, volume, surface area, centroid, and bounding box. Let the index values ​​of the standard model and the student model be s and t, respectively, and the difference is measured using relative bias (centroid is Euclidean distance): ; ; The relative deviations between the number of vertices and the number of faces are calculated similarly. The simplified comparison score definition uses a weighted calculation of each geometric deviation; for example: For feature-level semantic comparison, based on the preceding feature extraction module (extracting 13 types of manufacturing features such as holes, slots, bosses, curved surfaces, steps, cavities, fillets / chamfers, stiffeners, keyways, countersunk holes, conical surfaces, spherical surfaces, and flanges), the feature sets of the two models are grouped by type, and optimal matching is performed for each type and similarity is calculated.

[0046] First, single-feature similarity calculation: Each feature type corresponds to a FeatureComparator, which internally defines a set of parameter differences. and weight Overall similarity: The normalized differences among the parameters are unified as follows: positional difference, size difference, orientation difference, and type matching. Taking the hole feature as an example, its weights are set to {0.25, 0.3, 0.2, 0.15, 0.1}, corresponding to position, radius, depth, axis, and hole type, respectively. The weighted similarity of the hole features is then obtained. Other feature types each define their own parameter sets and weights.

[0047] Then, feature matching and the Hungarian algorithm are used to calculate: for each feature type t, a similarity matrix is ​​constructed: ,in, Transforming the matching problem into maximum weight matching in a bipartite graph is equivalent to solving the cost matrix. Minimize allocation problem: The optimal matching pair (i,j) is obtained using the Hungarian algorithm. If... If the match is successful, then it is considered a successful match. Type-related matching thresholds: basic features (holes, slots, bosses, cavities, steps): Curved surfaces, stiffening plates, countersunk holes, flanges: Rounded corners / bevels, conical surfaces, spherical surfaces: 0.3, etc. Unmatched standard features are counted as missing, and unmatched student features are counted as redundant.

[0048] Then, when the number of features of a certain type is large ( When performing a KD-tree spatial nearest neighbor matching algorithm, a fast matching algorithm is used instead of the Hungarian algorithm. A KD-tree is constructed for the locations of standard features. The nearest student feature is then queried, and if the distance is... (Basic characteristics) or (Complex features), then directly Pair them up and calculate their similarity.

[0049] Next, feature level scoring is performed. Let the standard number of features for type t be... The number of successful matches is The average similarity of the matching features is ,definition: ; Feature matching score: In the formula, These are the corresponding coefficient weights.

[0050] For voxel distance field comparison (mesh-level comparison), to further capture global shape differences, voxel intersection-over-union (IoU) and Chamfer distance are used for mesh-level comparison. Mesh normalization: The two models are scaled to the same unit cube region, with the common bounding box center at c and the maximum side length at s. Voxel IoU: Based on a preset resolution (e.g., ... The sampling points on the normalized grid surface are voxelized to obtain binary voxels. , ,calculate: Symmetrical average Chamfer distance: Each sample consists of N=5000 surface point clouds. , Calculate the bidirectional average nearest point distance: Distance rating: Grid-level comprehensive score: In the formula, These are the corresponding coefficient weights.

[0051] The comprehensive scoring unit 330 is used to weight and fuse the multi-level scores and output the final score. The final comparison score is obtained by weighting and fusing the scores from the three levels mentioned above: In the formula, The score corresponds to each level of comparison. The full three-level comparison or the simplified level comparison is dynamically selected based on the size of the task queue to be scored. For example, if the size of the task queue to be scored is greater than the preset size, the comparison without voxel distance field (comparison without grid level) is used; otherwise, the comparison with voxel distance field (comparison with grid level) is used.

[0052] The model comparison passed the condition. (Final score preset threshold).

[0053] Based on the above multi-level 3D model comparison framework, it is possible to quickly and objectively complete the scoring of works and the initial identification of problems. However, in order to further improve the teaching quality, for the non-conforming items found in the test, the rationality of the processing process should be analyzed in depth. It is possible to accurately locate whether it is a programming error, improper parameter selection, incorrect tool selection or process route problem. A process rationality analysis unit 340 is set up to receive the processing condition parameters input by students and the tool trajectory parameters actually used in the simulation. It calls the process expert system to perform parameter comparison analysis and cutting force safety assessment to obtain the rationality analysis results.

[0054] Specifically, the process expert system integrates a material database, a rule-based reasoning engine, a parameter calculation engine, and a cutting force analysis module. Based on user-input machining conditions (material, tool, machine tool, operation type, etc.), it can automatically calculate and recommend optimized cutting parameters, such as... Figure 2 As shown, whether the teacher inputs relevant assignments and their standard models, or the student uploads working condition data related to the actual machining simulation process, the system utilizes its internal material database, rule reasoning engine, parameter calculation engine, and cutting force analysis module to calculate and recommend parameters to assist students in setting actual process parameters for machining simulations. Simultaneously, this system also provides multi-dimensional cutting force analysis and safety assessment. The following sections detail the various modules of the process expert system.

[0055] Firstly, the data loading module. It reads the following six types of data from the database: Material Library: Covering plastics (nylon, glass fiber reinforced nylon, etc.), aluminum alloys (6061, 7075), steel (steel, Q235, 42CrMo), cast iron, stainless steel, titanium alloys (Ti6Al4V), copper alloys, etc. For each material, it records the basic cutting speed, feed per tooth, melting point, thermal conductivity, elastic modulus, and other physical properties at different machining stages. Tool Library: Defines the cutting speed factors and feed factors for tool materials such as cemented carbide, coated cemented carbide, high-speed steel, and CBN. Machine Tool Library: Includes three-axis / five-axis machining centers, CNC lathes, high-speed machining centers, etc., recording capability parameters such as maximum speed, power, torque, and kinematic chain structure. Machining Operation Library: Defines the speed and feed correction coefficients for each operation type. Material Category Restriction Rules: Defines hard and soft restrictions on cutting parameters according to material category, such as the upper limit of Vc for plastics and the cutting width ratio for titanium alloys.

[0056] Second, the rule reasoning engine. Rule reasoning is the core of intelligent decision-making and realizes the generative rule reasoning mechanism. The rule base contains about 40 rules, which are divided into the following levels according to the descending priority: (1) Material category basic factor rule (highest priority, level 110): The depth of cut factor and width of cut factor are set according to the material category, which reflects the essential differences of different materials in terms of cutting resistance, thermal conductivity, elastic modulus, etc. For example: the fixed depth of cut factor of aluminum alloy (small cutting force, large depth of cut) is 1 and the width of cut factor is 0.7; the fixed depth of cut factor of titanium alloy (extremely poor thermal conductivity, strictly limited) is 0.4 and the width of cut factor is 0.2. (2) Material physical property perception rule (priority level 90-100) Based on the thermophysical properties of the material, parameter constraints are made: Low melting point material thermal constraint: For thermoplastic materials with a melting point of less than 300 degrees Celsius, the cutting speed is strictly limited to prevent the material from softening and melting due to frictional heat. Low thermal conductivity material constraint: For materials with extremely low thermal conductivity, the cutting speed and feed are greatly reduced to prevent the tool tip from overheating and burning. Large depth of cut warning: When the axial depth of cut is greater than the preset tool diameter, the lateral force warning is automatically triggered, and the safe cutting width ratio is recommended according to the material category. Built-up edge prevention: In view of the easy sticking characteristics of aluminum alloy, it is suggested to use a large rake angle polishing tool and sufficient emulsion cooling. (3) Material-specific rule set (priority level 80-95) has written a special rule chain for each material category, for example: Stainless steel rule: limit Forced minimum cutting depth 0.15min, requiring high-pressure internal cooling and climb milling. (4) General process rules (priority level 25-60), including general experience rules such as speed reduction for high hardness, parameter reduction for thin-walled parts, feed increase for coated tools, feed reduction for curved surface finishing, speed reduction for deep holes, and speed limit for tapping.

[0057] The core method of this inference engine, infer(), evaluates each rule condition one by one. When the condition is met, the corresponding action is executed (updating coefficient factors or generating warning information), and finally outputs four key adjustment factors: such as cutting speed adjustment factor, feed per tooth adjustment factor, axial depth of cut adjustment factor, radial width of cut adjustment factor, and multi-dimensional process warnings.

[0058] Third, parameter calculation engine. Parameter calculation is the core of the numerical calculation of the system, which encapsulates the complete CNC machining parameter calculation logic and cutting mechanics model. The main methods include: (1) Milling parameter calculation: obtain the basic parameters of material, tool, machine tool and machining mode from the database, call the rule engine to reason to obtain the adjustment factor, calculate the theoretical cutting parameters according to the adjustment factor and machine tool capability constraints, and compare them with the actual simulation parameters of the students to generate deviation indicators; such as: cutting speed Vc calculation: the basic Vc can be selected according to the machining type, multiplied by the tool material factor, machining mode factor, etc., and then the material category hard limit is applied; feed per tooth fz calculation: similar to multi-layer factor superposition, finally constrained by the upper limit of material category fz; depth of cut / width of cut ap / ae correction: limited according to the material category exclusive parameter range. (2) Three cutting force calculation methods are provided and can be selected in the settings, including: a simplified estimation model that calculates the combined cutting force based on the unit cutting force and meshing ratio; the Kienzle model that introduces the cutting force exponential relationship, edge force correction and plowing force component; and the Merchant shear angle model that is theoretically derived based on shear strength, shear angle and friction coefficient. The cutting force analysis module outputs tangential force, radial force, axial force, feed force, combined force, torque, power and specific cutting force. (3) Safety analysis.

[0059] The cutting force is compared with the machine tool's maximum thrust, maximum torque, and rated power to generate a safety level.

[0060] Finally, the report generation module 400 automatically generates a structured teaching feedback report based on the scoring results, the list of matching failure features, parameter deviation items, and the process rationality analysis results. The scoring results are the final comprehensive scoring results, and a threshold range for the comprehensive scoring results can be set to assess each student's level of mastery of the current machining simulation process. The matching failure list stores defect features in the student's 3D mesh model and the standard mesh model that are mismatched or have exceeded the deviation limit. The parameter deviation items represent the deviations of various feature parameters. The process rationality analysis results include parameter comparison analysis results and cutting force safety assessment results.

[0061] In order to further automatically identify potential common teaching problems from a large number of student error reports, and transform passive scoring into proactive teaching quality management. The report generation module 400 is also used to summarize the structured teaching feedback reports of each member in a student group (such as the same class or the same training project group) specified by the teaching management module, and extract the common matching failure feature types and parameter deviation items in the reports, such as "counterhead hole axis deviation", "groove bottom flatness deviation" and parameter deviation items (such as "fz is 18% lower", "coolant pressure not enabled"). When the frequency of the same feature type or parameter deviation item in the student group exceeds the preset threshold (default is 60% of the total number of students in the group), a group teaching diagnosis report is generated. The diagnosis report generation process includes: associating machining skill points by querying the feature type-skill point mapping table, such as mapping "counterhead hole axis deviation" to the "hole machining positioning accuracy control" skill point; locating the teaching link by querying the teaching link ID corresponding to the skill point, such as "G83 deep hole cycle programming and parameter setting"; generating improvement suggestions by calling the teaching knowledge base and outputting suggestions such as "increasing the practical training of G83 parameter Q (drilling depth per bite) and setting typical fault cases (Q value is too small, resulting in poor chip removal) for targeted error correction". The group teaching diagnostic report is generated in Markdown format and includes a heatmap of common error types, a radar chart of the distribution of processing skill points, and a list of suggestions for improving teaching procedures.

[0062] One specific embodiment provides targeted feedback for in-depth diagnosis of the root causes of machining problems. Besides using a process rationality analysis unit to obtain specific students' erroneous parameters and safety issues during simulated machining, it further transforms static geometric deviations into dynamic process decision-making errors, enabling causal tracing of processes through result difference analysis to aid teaching. The system also includes: The process rationality analysis unit 340 is also used to receive the machining condition parameters input by the student, the actual tool trajectory parameters used in the simulation, and the defect features of mismatch or deviation exceeding the limit identified in the model comparison. It calls the process expert system, which includes a material database, rule reasoning engine, parameter calculation engine, cutting force analysis module, and correlation diagnosis module, to perform parameter comparison analysis, cutting force safety assessment, and feature-process correlation diagnosis to obtain the process rationality analysis results of the student simulation process.

[0063] The process expert system invoked also includes feature-process correlation diagnosis. This feature-process correlation diagnosis uses a neural network algorithm to correlate and bind the defect features identified in the model comparison, the corresponding feature parameter deviation sequence, and the actual tool trajectory parameters and machining condition parameters associated with the machining program segment corresponding to the defect features. This helps to locate the problems in the student's simulation machining process and tool usage. Specifically, it analyzes and determines the decision parameter chain of the student's actual application of the current defect features, compares this decision parameter chain with the ideal decision domain, and outputs the incorrect process decision.

[0064] Furthermore, if the focus is on how to deeply integrate the 3D geometric comparison of the final model with the behavioral sequence of students' processing and the rationality of process decisions, restoring static geometric deviations to dynamic process decision errors; and associating isolated feature errors with specific behavioral pattern defects to further assist teaching, the system also includes: The automatic scoring module 300 further includes a behavior sequence mining and analysis unit 350 connected to the process rationality analysis unit. The behavior sequence mining and analysis unit 350 is used to extract the student's operation event flow on the virtual machine tool based on the processing path data generated by the student during the simulation. This includes events such as NC program loading, tool replacement (including tool number and lifespan), coordinate system setting, simulation start / pause / termination, and manual intervention. The unit generates an operation behavior sequence sorted by timestamp, with each event marked with an absolute timestamp and a relative task start time offset. It also calculates operation fluency indicators (such as event density per unit time and program segment switching frequency) and risk behavior indicators (such as the number of manual interventions, coordinate system reset frequency, and unplanned tool replacements). The behavior sequence mining and analysis unit 350 is also used to associate the operation behavior sequence with the defect feature-process after feature-process correlation diagnosis and align them by time. Taking the moment when the defect feature is identified in the model comparison as the anchor point, it traces back the operation events within 5 seconds and the tool trajectory parameter change events within 3 seconds to construct a time window evidence chain of "defect moment - preceding operation - subsequent parameter". Based on the evidence chain, it calls the causal diagnosis model in the process expert system to obtain the causal analysis results of the student simulation process defects.

[0065] Among them, the causal diagnosis model in the process expert system adopts a neural network algorithm (such as LSTM network). The input is the parameter deviation sequence extracted from the evidence chain, defect features and their associated actual tool trajectory parameters, operation event type sequence and time interval. It performs causal analysis on machining task problems with a comprehensive score less than the preset comprehensive score and outputs the causal analysis results of machining task problems. For example, when machining countersunk holes, the chip removal is not smooth due to the incorrect setting of the G83 deep hole cycle parameter (Q value is too small), which causes tool wobble after the third manual intervention and ultimately results in a countersunk depth deviation of +0.15mm.

[0066] The report generation module 400 is also used to automatically generate a structured teaching feedback report based on the scoring results, the list of matching failure features, the parameter deviation items, the results of the process rationality analysis, and the causal results of the processing task problems.

[0067] In a specific embodiment, to make the scoring adaptable to various situations, the evaluation focus is automatically adjusted according to the teaching progress, achieving an upgrade from "geometric comparison" to "assessment of the achievement of teaching objectives." The system can adjust weights to calculate a comprehensive score based on the current teaching progress and processing type, improving the flexibility, accuracy, and adaptability of the scoring, and providing a more precise evaluation basis for teaching. The system also includes: The model comparison unit 320 is also used to perform multi-level comparison between the standard three-dimensional mesh model and the student three-dimensional mesh model. The comparison process includes: calling the preset processing type-geometric statistical quantum set mapping table and feature template according to the processing type corresponding to the distributed processing task, and dynamically selecting to perform a complete three-level comparison or a simplified level comparison according to the size of the queue of tasks to be scored.

[0068] The processing tasks are categorized by processing type, including turning, milling, and multi-axis machining. Correspondingly, a processing type-geometric statistical quantum set mapping table and feature templates are set up. The processing type-geometric statistical quantum set mapping table contains geometric feature statistical quantum sets generated based on fundamental geometric statistics. The processing type-feature templates include features emphasized by different processing types. For example, under turning, the focus is on features of rotating bodies (outer circle, inner hole, thread, tapered surface), with the feature semantic layer tilted towards dimensional accuracy, coaxiality, circular runout, and surface roughness. Under milling, the focus is on planes, cavities, hole systems, and contours, with the feature semantic layer tilted towards position, parallelism, perpendicularity, and overcut / undercut volume.

[0069] The complete three-layer comparison includes: at the global layer, selecting an appropriate set of geometric feature statistics based on the processing type, and calculating the deviation of the two models on the subset to generate a global geometric similarity score.

[0070] At the feature layer, suitable features are extracted from the two models based on processing type. Single feature similarity is calculated, and feature matching based on the number of features is adaptively selected (when the number is not less than 10) or feature matching based on the spatial nearest neighbor of the KD tree (when the number is less than 10), and feature-level semantic matching score is calculated. At the voxel layer, the two models are converted into voxel space representation, and voxel intersection-union ratio and Chamfer distance are calculated to generate voxel similarity score.

[0071] The comprehensive scoring unit 330 is also used to obtain a machining skill point weight matrix associated with the current teaching progress; decompose the machining task into sub-skill points (such as external turning, thread machining, cavity milling, and drilling cycles), and bind them with the teaching progress to generate a teaching progress-skill point weight matrix. The weight matrix assigns weights to different comparison levels and different manufacturing features that are adjusted according to the teaching stage, and uses this weight matrix to perform weighted fusion of global geometric similarity score, feature-level semantic matching score, and voxel similarity score to calculate a comprehensive score that fits the current teaching progress and machining type.

[0072] In one specific embodiment, to improve the accuracy of feature extraction, the system further includes: The manufacturing feature extraction unit 300 is also used to perform multi-level collaborative manufacturing feature extraction and enhancement, specifically including: global feature pre-extraction guided by higher-order geometric moments, initial semantic classification of patches driven by graph neural networks, and directed distance field refinement with adaptive subdivision voxelization.

[0073] Before identifying manufacturing features, higher-order statistical moments based on shape diameter function (SDF) or shape distribution are introduced on the basis of traditional first-order (such as volume, centroid) and second-order (such as moment of inertia) statistics to calculate and obtain the higher-order geometric moments of the triangular mesh model. In addition, manufacturing-oriented global features are constructed to extract manufacturing-oriented global features including the dimensional distribution in the main machining direction (e.g., the thickness distribution features of the model calculated along the machine tool spindle direction (Z-axis) to determine the overall deviation of axial dimensions) and the theoretical material removal rate (by comparing the volume difference between the blank model and the finished product model and comparing it with the theoretical removal amount, the processing efficiency or whether there is overcutting / undercutting).

[0074] When identifying feature geometric primitives, a graph neural network or attention-based model is first used to perform preliminary feature classification of mesh patches (such as planes, cylindrical surfaces, and freeform surfaces). Then, the boundary is corrected and segmented by adapting the preliminary feature classification to the corresponding geometric rules, outputting feature regions with semantic labels, such as the feature regions of straight milled surface 1 and through hole 2. For each identified manufacturing feature, a set of process-related attribute parameters are calculated, such as: hole features: diameter, depth, hole bottom shape (flat bottom, drill tip), axial direction, and position.

[0075] like Figure 3 As shown, this application provides a virtual machine tool simulation data processing method for CNC machining teaching, which utilizes the aforementioned virtual machine tool simulation data processing system for CNC machining teaching, including: S1. Distribute processing tasks and standard 3D mesh models to student terminals.

[0076] S2. By calling the open interface of the virtual machine tool software, the system acquires in real time the machining path data, tool trajectory data, material removal sequence data, and the final 3D mesh model data generated by the student during the simulation. During the simulation, the system calls upon the process expert system to generate recommended parameters in real time based on the currently assigned machining task and the actual machining conditions to guide the student's machining simulation.

[0077] S3. Based on the 3D mesh model, identify manufacturing features and output the geometric primitive parameters and semantic types of each feature; perform ICP spatial registration, multi-level comparison, and multi-level scoring sequentially on the standard 3D mesh model and the student 3D mesh model; perform weighted fusion of the multi-level scores and output the final score.

[0078] S4 receives the machining condition parameters input by the student and the actual tool trajectory parameters used in the simulation. It then calls the process expert system, which includes a material database, rule reasoning engine, parameter calculation engine, and cutting force analysis module, to perform parameter comparison analysis and cutting force safety assessment, and obtain the rationality analysis results.

[0079] S5. Based on the scoring results, the list of matching failure features, parameter deviation items, and the results of process rationality analysis, a structured teaching feedback report is automatically generated.

[0080] This application also discloses a computer-readable storage medium.

[0081] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the virtual machine tool simulation data processing method for CNC machining teaching described above. The computer-readable storage medium includes, for example, various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] This application also discloses a computer device.

[0083] Specifically, the computer device includes a memory and a processor. The memory stores a computer program that can be loaded and executed by the processor to perform the virtual machine tool simulation data processing method for CNC machining teaching.

[0084] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A virtual machine tool simulation data processing system for CNC machining teaching, characterized in that, include: The teaching management module, simulation data acquisition module, automatic scoring module, and report generation module are interconnected and communicate with each other. The teaching management module is used to distribute processing tasks and standard 3D mesh models to student terminals; The simulation data acquisition module is used to acquire, in real time, the machining path data, tool trajectory data, material removal sequence data, and the final 3D mesh model data generated by the student during the simulation process by calling the open interface of the virtual machine tool software. The automatic scoring module includes a manufacturing feature extraction unit, a model comparison unit, a comprehensive scoring unit, and a process rationality analysis unit; The manufacturing feature extraction unit is used to identify manufacturing features based on a 3D mesh model and output the geometric primitive parameters and semantic types of each feature; the model comparison unit is used to sequentially perform ICP spatial registration, multi-level comparison, and multi-level scoring on the standard 3D mesh model and the student 3D mesh model; the multi-level comparison includes global geometric statistics comparison, feature-level semantic comparison, and voxel distance field comparison. The comprehensive scoring unit is used to weight and fuse multi-level scores and output the final score; the process rationality analysis unit is used to receive the machining condition parameters and the actual tool trajectory parameters used in the simulation input by the students, and call the process expert system, which includes a material database, rule reasoning engine, parameter calculation engine and cutting force analysis module, to perform parameter comparison analysis and cutting force safety assessment to obtain the rationality analysis results. The report generation module is used to automatically generate a structured teaching feedback report based on the scoring results, the list of matching failure features, parameter deviation items, and the results of process rationality analysis.

2. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 1, characterized in that, The manufacturing feature extraction unit is used to perform RANSAC robust fitting, normal clustering segmentation and curvature feature description in sequence based on a triangular mesh model to identify manufacturing features including holes, slots, bosses, surfaces, steps, cavities and composite features.

3. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 1, characterized in that, The process rationality analysis unit is also used to receive the machining condition parameters input by the student, the actual tool trajectory parameters used in the simulation, and the defect features of mismatch or deviation exceeding the limit identified in the model comparison. It calls the process expert system, which includes a material database, rule reasoning engine, parameter calculation engine, cutting force analysis module, and correlation diagnosis module, to perform parameter comparison analysis, cutting force safety assessment, and feature-process correlation diagnosis to obtain the process rationality analysis results of the student simulation process. The process expert system comprises several components. The rule-based reasoning engine in the system infers adjustment factors by reasoning step-by-step based on material category fundamental factor rules, material physical property perception rules, material-specific rule sets, and general process rules. The parameter calculation engine calculates theoretical cutting parameters based on the adjustment factors and machine tool capability constraints, providing theoretical cutting parameter references for students during machining simulations and comparing them with actual simulation parameters to generate deviation indicators. The cutting force analysis module provides a cutting force calculation model, calculates the machine tool's cutting force based on user-defined parameters, and compares it with the machine tool's maximum thrust, maximum torque, and rated power to generate a safety level. The feature-process correlation diagnosis uses a neural network algorithm to correlate and analyze defect features identified in model comparisons with the actual tool trajectory parameters and machining condition parameters associated with the corresponding machining program segment, thus pinpointing problems in student simulation machining processes and tool usage.

4. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 1, characterized in that, The model comparison unit is also used to perform multi-level comparisons between the standard 3D mesh model and the student 3D mesh model. The comparison process includes: calling a preset processing type-geometric statistical quantum set mapping table and feature template according to the processing type corresponding to the distributed processing task, and dynamically selecting to perform a complete three-level comparison or a simplified level comparison according to the size of the task queue to be scored; the complete three-level comparison includes: at the global level, selecting an appropriate geometric feature statistical quantum set according to the processing type, calculating the deviation of the two models on the subset to generate a global geometric similarity score; at the feature level, extracting appropriate features from the two models to select the processing type, calculating single feature similarity, and adaptively selecting feature matching based on the Hungarian algorithm or feature matching based on the KD tree spatial nearest neighbor according to the number of features, and calculating a feature-level semantic matching score; at the voxel level, converting the two models into voxel space representations, calculating the voxel intersection-union ratio and Chamfer distance to generate a voxel similarity score; The comprehensive scoring unit is also used to obtain a weight matrix of processing skill points associated with the current teaching progress. The weight matrix assigns weights to different comparison levels and different manufacturing features that are adjusted according to the teaching stage. The weight matrix is ​​used to perform weighted fusion of global geometric similarity score, feature-level semantic matching score and voxel similarity score to calculate a comprehensive score that is suitable for the current teaching progress and processing type.

5. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 3, characterized in that, The automatic scoring module also includes: a behavior sequence mining and analysis unit connected to the process rationality analysis unit; The behavior sequence mining and analysis unit is used to extract the student's operation event flow on the virtual machine based on the processing path data generated by the student during the simulation, construct the operation behavior sequence and extract operation fluency and risk behavior indicators; associate the operation behavior sequence with the defect feature-process after feature-process correlation diagnosis and align them by time to form a defect feature-process-behavior correlation evidence chain; and call the causal diagnosis model in the process expert system according to the evidence chain to obtain the causal analysis results of the student's simulation process defects. The causal diagnosis model uses a neural network algorithm to perform causal analysis on machining task problems with a comprehensive score lower than a preset comprehensive score based on the defect features identified in the model comparison, the actual tool trajectory parameters associated with the machining program segments corresponding to the defect features, and the operation behavior sequence data, and to determine the causal result of the machining task problem. The report generation module is also used to automatically generate a structured teaching feedback report based on the scoring results, the list of matching failure features, parameter deviation items, the results of process rationality analysis, and the causal results of processing task problems.

6. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 1, characterized in that, The report generation module is also used to summarize the structured teaching feedback reports of each member in the student group specified by the teaching management module, extract the common matching failure feature types and parameter deviation items in the reports, and generate a group teaching diagnosis report when the frequency of the same feature type or parameter deviation item in the student group exceeds a preset frequency threshold. The group teaching diagnosis report includes common error types, processing skill points and corresponding teaching improvement suggestions.

7. The virtual machine tool simulation data processing system for CNC machining teaching according to claim 2, characterized in that, The manufacturing feature extraction unit is also used to perform multi-level collaborative manufacturing feature extraction and enhancement, which includes: calculating the higher-order geometric moments of the triangular mesh model before identifying manufacturing features, and extracting manufacturing-oriented global features including the size distribution in the main machining direction and the theoretical material removal rate. When identifying feature geometric primitives, a graph neural network or an attention-based model is first used to perform preliminary feature classification on the mesh patches. Then, the boundary is corrected and segmented by adapting the preliminary feature classification to the corresponding geometric rules, and feature regions with semantic labels are output.

8. A virtual machine tool simulation data processing method for CNC machining teaching, using the system described in any one of claims 1-9, characterized in that, include: Distribute processing tasks and standard 3D mesh models to student terminals; By calling the open interface of the virtual machine tool software, the machining path data, tool trajectory data, material removal sequence data and the final 3D mesh model data generated by the student during the simulation can be obtained in real time. Based on the 3D mesh model, manufacturing features are identified and the geometric primitive parameters and semantic types of each feature are output. ICP spatial registration, multi-level comparison, and multi-level scoring are performed sequentially on the standard 3D mesh model and the student 3D mesh model. The multi-level scores are weighted and fused to output the final score. The system receives machining condition parameters and actual tool trajectory parameters input by students, and calls a process expert system containing a material database, rule reasoning engine, parameter calculation engine and cutting force analysis module to perform parameter comparison analysis and cutting force safety assessment to obtain rationality analysis results. Based on the scoring results, the list of matching failure features, parameter deviation items, and the results of process rationality analysis, a structured teaching feedback report is automatically generated.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the method as described in claim 8.

10. A computer device, characterized in that, The computer device includes a memory, a processor, and a program stored in and executable on the memory, which, when executed by the processor, performs the steps of the method as described in 8.

Images