A method for constructing a numerical control machining whole-process multi-modal data space-time correlation model

Abstract

The application discloses a kind of construction methods of numerical control processing whole-process multi-modal data space-time correlation model, it is related to numerical control processing whole-process processing data correlation modeling field, comprising: firstly, part geometric model is discretized into point cloud, and multi-level granularity is divided according to point, feature, overall geometry;Then, feature level granularity point cloud entity is constructed based on tool position point sequence;Next, process data is based on machining feature, monitoring data is aligned based on time, and based on space mapping, detection data is based on spatial positioning, respectively associated to the point cloud entity of corresponding granularity;Finally, through the mutual reference of the unique identification and attribute of entity, the fusion of multi-modal, multi-scale, multi-granularity data such as geometry, process, monitoring, detection in unified space-time coordinates is realized;The application solves the problem that multi-source heterogeneous processing data is difficult to express and correlate uniformly, and provides technical support for constructing high-quality processing data set and industrial big model application.

Description

Technical Field

[0001] This invention relates to the field of CNC machining process data association modeling, and in particular to a method for constructing a spatiotemporal association model of multimodal data for the entire CNC machining process. Background Technology

[0002] With the deepening application of artificial intelligence in the manufacturing field, large-scale industrial models for manufacturing have become an important direction for promoting the development of intelligent manufacturing. Constructing high-quality, generalizable datasets is a prerequisite for obtaining reliable and accurate prediction and decision-making models. However, in actual manufacturing processes, parts have complex geometries, long processing cycles, and numerous process chains, involving heterogeneous data from multiple sources, including geometry, process, monitoring, and inspection. This data generally exhibits characteristics such as multimodality, multi-granularity, and multi-scale, with low levels of structure and weak semantic correlation, making it difficult to achieve unified expression and knowledge reasoning, thus limiting the effective training and application of large-scale industrial models. Therefore, the structured expression and correlation fusion of manufacturing data has become a key challenge for the development of intelligent manufacturing.

[0003] Existing research mainly focuses on manufacturing information modeling at the geometric feature level of parts: ontology-based methods can unify semantic and structured information, but have limited capabilities in processing multi-granularity, multi-scale, and multi-modal unstructured data and geometric data, and lack dynamism, making them difficult to adapt to complex manufacturing scenarios; knowledge graph-based manufacturing information construction frameworks use part geometric features as the smallest unit of analysis, achieving semantic association, but lack a modeling perspective for multi-granularity and multi-scale collaborative fusion, resulting in unclear data hierarchy; the STEP-NC standard uses part geometric features as the smallest programming unit, defines machining tasks through process step sequences, and uses an object-oriented data model to separate geometric and process information, but fails to cover the modeling and fusion of monitoring and inspection data; dynamic machining feature models use machining features as the smallest unit of analysis, integrating machining, monitoring, and online inspection data to support the optimized control of complex parts, but do not fully consider the multi-granularity characteristics of geometric features and their correlation with geometric depth. Currently, geometry-process-monitoring information association methods mostly use tool positions as a medium to associate static geometric information with dynamic process and monitoring information to theoretical tool positions, achieving synchronous association of multi-source data and providing a data foundation for intelligent manufacturing. However, this type of method still uses processing features as the smallest unit of analysis and has not yet incorporated detection information into a unified association framework, making it difficult to achieve a unified expression of processing data throughout the entire CNC machining process.

[0004] In the manufacturing process, processing data comes from diverse sources, encompassing information on geometry, technology, monitoring, and inspection. This data exhibits multi-granularity, multi-scale, and multi-modal characteristics, leading to challenges such as inconsistent primary keys, spatiotemporal asynchrony, and semantic misalignment during the data construction process. Current technologies struggle to achieve a unified representation of manufacturing data, thus limiting the full potential of data-driven decision-making in intelligent manufacturing. Summary of the Invention

[0005] To address the above issues, this invention proposes a method for constructing a spatiotemporal correlation model of multimodal data throughout the entire CNC machining process. Using part geometric information as a unified spatial carrier, a point cloud space is constructed through part geometry to form a spatial index, thereby realizing the spatiotemporal mapping and correlation of multi-source data such as geometry, process, monitoring and detection during the machining process.

[0006] To achieve the above objectives, this invention provides a method for constructing a spatiotemporal correlation model of multimodal data throughout the entire CNC machining process, comprising the following steps:

[0007] S1. Using the geometric point cloud of the part as a unified spatial representation carrier, and adopting the entity-identifier-attribute organizational structure as a unified data association paradigm; establishing multi-granularity hierarchical identification for the part geometry, including single-point level granular identification, feature level granular identification, and overall geometric level granular identification.

[0008] S2. Based on the tool position sequence during the part machining process, construct feature-level granularity point cloud entities in the point cloud space. The feature-level granularity point cloud entities are composed of point cloud points corresponding to the tool sweep area. Point cloud points not covered by any machining feature corresponding sweep area belong to the overall geometric-level granularity point cloud entity. Each single point cloud after the geometric model is discretized belongs to the single point-level granularity point cloud entity.

[0009] S3. Map the processing technology data to the corresponding feature-level granularity point cloud entities constructed in S2 based on their corresponding processing steps and processing features.

[0010] S4. Perform spatiotemporal alignment processing on the monitoring data of the processing process, and then associate the aligned monitoring data with the corresponding feature-level granularity point cloud entities constructed in S2 based on the spatial mapping relationship.

[0011] S5. Map the part inspection data to the corresponding feature-level granularity point cloud entity, overall geometric-level granularity point cloud entity, and single-point granularity point cloud entity based on the spatial coordinates of the inspection points and the spatial proximity relationship of the point cloud.

[0012] S6. By defining entities containing unique identifiers and attributes for each data object, and using the identifier of one entity as the attribute of another entity, the association relationship between data entities of different modalities, scales and granularities in S2-S5 is established, and finally a spatiotemporal association model of multimodal data of the entire CNC machining process is formed.

[0013] Preferably, the entity-identifier-attribute organizational structure is represented by constructing entity, identifier, and attribute triples for each type of data object:

[0014] ;

[0015] in Represents an entity, A unique identifier that corresponds one-to-one with an entity. It represents a set of attributes for an entity; it defines data entities of different modalities, scales, and granularities, defines a unique identifier for each entity, and identifies an entity by... logo Define as another entity Attributes Establish the association between two entities in the following ways:

[0016] ;

[0017] Establish entity With entity The relationships between entities are defined; each entity defines multiple different types of attributes, including attributes of the entity itself and attributes associated with other entities. By selecting a specified attribute of an entity, the corresponding modality, scale, or granularity data associated with that entity can be obtained, thereby realizing the unified expression and association of machining data throughout the entire CNC machining process.

[0018] Preferably, single-point level granularity identifiers are used to identify individual points in a point cloud;

[0019] Feature-level granularity identifiers are used to identify processing feature entities composed of multiple point-level entities;

[0020] The overall geometric level granularity identifier is used to identify the overall geometry of a part, and its corresponding point cloud is the overall point cloud of the part.

[0021] Preferably, S2 contains the following: The overall geometric point cloud is obtained based on the discretization of the part's geometric model, and the theoretical tool position sequence of different machining steps is mapped to the point cloud space; based on the tool's geometric parameters and the theoretical tool position sequence, the spatial region swept by the tool during machining is determined, and the point cloud points located within the swept region are divided and identified as feature-level granularity point cloud entities corresponding to the same machining feature; point cloud points not covered by any swept region corresponding to a machining feature are assigned to the overall geometric-level granularity point cloud entity; through spatial mapping and region division based on the tool position sequence, multi-granularity expression of the part geometry is realized at the point level, feature level, and overall geometric level, providing a spatial basis for the subsequent mapping and association of multimodal machining data in the point cloud space.

[0022] Preferably, the machining process data is obtained by extracting process information from different steps in the machining process. Its process attribute information includes at least machine tool features, tool information, machining method, step sequence and process parameters. The machining process data is mapped to the corresponding feature-level granularity point cloud entities based on the steps and machining features, realizing the spatiotemporal correlation between the machining process data and the geometric point cloud of the part.

[0023] Preferably, the machining process monitoring data includes multi-source monitoring attribute information such as machine tool operating status, tool motion information, and machining process monitoring information.

[0024] Preferably, the spatiotemporal alignment process in S4 is as follows: matching the actual tool position point and the theoretical tool position point obtained during the processing, and then realizing the spatial matching and alignment of multimodal data according to the sweep space region determined in S2; setting a unified standard time window according to the time characteristics of the processing process, and mapping monitoring data with different sampling frequencies to the standard time window for alignment, thereby realizing the alignment of multi-monitoring channel data in the time dimension.

[0025] Preferably, the part inspection data includes multi-scale inspection attribute information from macro to micro, reflecting the internal stress state of the material, material properties, geometric deformation and surface quality.

[0026] Preferably, the part inspection data is spatially located using the coordinates of the inspection points obtained from on-machine inspection or external inspection equipment: the spatial coordinates of the inspection points are obtained from on-machine inspection or external inspection equipment, and the inspection points and the part geometric point cloud are placed in the same spatial coordinate system. The mapping relationship between the inspection points and the corresponding geometric points or geometric regions in the point cloud is determined by the spatial proximity relationship, and the inspection attribute information is associated with the corresponding feature-level granularity point cloud entity, the overall geometric-level granularity point cloud entity, and the single-point-level granularity point cloud entity.

[0027] Therefore, the present invention employs the above-mentioned method for constructing a multimodal data spatiotemporal correlation model for the entire CNC machining process, which has the following beneficial effects:

[0028] (1) This invention constructs a high-quality spatiotemporal correlation dataset by unifying the modeling and association of multimodal data such as geometry, process, monitoring and detection, providing a solid data foundation for industrial large model training and intelligent decision-making in the manufacturing field;

[0029] (2) This invention uses point cloud as a unified spatiotemporal coordinate carrier to realize the association and spatial mapping of multimodal data such as geometry, process, materials, monitoring and quality, and establishes a unified representation framework for multimodal, multiscale and multigranular data, thereby improving the integrability and applicability of CNC machining process data.

[0030] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0031] Figure 1 This is a multimodal, multi-scale, and multi-granularity machining data representation diagram of the entire CNC machining process according to an embodiment of the present invention;

[0032] Figure 2 This is a framework diagram of the spatiotemporal correlation model of multimodal, multi-scale, and multi-granularity machining data for the entire CNC machining process according to an embodiment of the present invention.

[0033] Figure 3 This is a diagram showing the results of the analytical study of the geometric feature data of the parts according to an embodiment of the present invention;

[0034] Figure 4 This is a mapping diagram of the theoretical and actual cutting tool positions in an embodiment of the present invention, wherein (a) is a schematic diagram of the theoretical cutting tool position, (b) is a schematic diagram of the actual cutting tool position, and (c) is a matching diagram of the theoretical and actual cutting tool positions;

[0035] Figure 5 This is a diagram illustrating the multimodal data association between geometry, process, monitoring, and detection in an embodiment of the present invention. Detailed Implementation

[0036] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0037] Please see Figures 1-5 A method for constructing a spatiotemporal correlation model of multimodal data throughout the entire CNC machining process includes the following steps:

[0038] S1. Using the part's geometric point cloud as a unified spatial representation carrier and employing an entity-identifier-attribute organizational structure as a unified data association paradigm, multi-granularity hierarchical identifiers are established for the part geometry, including single-point level identifiers, feature level identifiers, and overall geometric level identifiers. Single-point level identifiers are used to identify individual points in the point cloud; feature level identifiers are used to identify machining feature entities composed of multiple point-level entities; and overall geometric level identifiers are used to identify the overall geometry of the part, with the corresponding point cloud being the overall point cloud of the part. This multi-granularity hierarchical division of geometry enables hierarchical expression of the same geometry, thereby supporting the unified expression of multi-layered machining semantics in the point cloud space. It should be noted that the part geometric point cloud is first processed by discretizing the part's geometric model, extracting spatial coordinate information, and generating point cloud data. Based on this, the overall geometric point cloud is divided into granularities by theoretical discrete tool positions. Multi-granularity representation of geometric information is realized according to point level, feature level and overall part level. The point cloud region is divided into theoretical geometric domain, machining space domain and detection space domain according to the multimodal data association region, providing a unified geometric basis for the spatial mapping of subsequent multimodal data.

[0039] The entity-identifier-attribute organizational structure is represented by constructing a triplet of entity, identifier, and attribute for each type of data object:

[0040] ;

[0041] in Represents an entity, A unique identifier that corresponds one-to-one with an entity. It represents a set of attributes for an entity; it defines data entities of different modalities, scales, and granularities, defines a unique identifier for each entity, and identifies an entity by... logo Define as another entity Attributes Establish the association between two entities in the following ways:

[0042] ;

[0043] Establish entity With entity The relationships between entities are defined; each entity defines multiple different types of attributes, including attributes of the entity itself and attributes associated with other entities. By selecting a specified attribute of an entity, the corresponding modality, scale, or granularity data associated with that entity can be obtained, thereby realizing the unified expression and association of machining data throughout the entire CNC machining process.

[0044] S2. Based on the tool position sequence during part machining, construct feature-level granular point cloud entities in the point cloud space. Feature-level granular point cloud entities consist of point cloud points corresponding to the tool sweep area. Point cloud points not covered by any sweep area corresponding to a machining feature belong to the overall geometric-level granular point cloud entity. Each single point cloud after discretization of the geometric model belongs to a single-point granular point cloud entity. The content is as follows: Obtain the overall geometric point cloud based on the discretization of the part's geometric model, and map the theoretical tool position sequence of different machining steps to the point cloud space. Based on the tool's geometric parameters and the theoretical tool position sequence, determine the spatial region swept by the tool during machining, and divide and identify the point cloud points within the swept area as feature-level granular point cloud entities corresponding to the same machining feature. Point cloud points not covered by any sweep area corresponding to a machining feature belong to the overall geometric-level granular point cloud entity. Through spatial mapping and region division based on the tool position sequence, realize multi-granularity expression of part geometry at the point, feature, and overall geometric levels, providing a spatial basis for the subsequent mapping and association of multimodal machining data in the point cloud space.

[0045] S3. Map the machining process data to the corresponding feature-level granularity point cloud entities constructed in S2 based on its corresponding machining steps and machining features. The machining process data is obtained by extracting the process information of different machining steps in the machining process. Its process attribute information includes at least machine tool features, tool information, machining method, machining step sequence and process parameters. The machining process data is mapped to the corresponding feature-level granularity point cloud entities based on the machining steps and machining features to realize the spatiotemporal correlation between the machining process data and the geometric point cloud of the part.

[0046] It should be noted that a mapping relationship is established between machining process data and feature-level point clouds in the machining spatial domain. Each point cloud point corresponds to a specific location on the surface or volume of the part, and the machining process information generated at that location during the machining process is mapped to the corresponding point cloud point, thus forming a one-to-one correspondence in the spatial domain. The point cloud points in the machining spatial domain not only have spatial coordinates but also additional multimodal process attributes. This representation method allows the machining process information to be uniformly presented in the point cloud space, forming a geometrically-process unified data representation. Through this representation structure, the machining process can be queried and analyzed at the point cloud level.

[0047] S4. Perform spatiotemporal alignment processing on the machining process monitoring data, and then associate the aligned monitoring data with the corresponding feature-level granularity point cloud entities constructed in S2 based on spatial mapping relationships. The machining process monitoring data includes multi-source monitoring attribute information of machine tool operating status, tool motion information, and machining process monitoring information. The monitoring data comes from different acquisition systems with different sampling frequencies and time resolutions. The spatiotemporal alignment processing is specifically as follows: match the actual tool position points and theoretical tool position points acquired during the machining process, and then achieve multi-modal data spatial matching and alignment based on the swept space region determined in S2; set a unified standard time window according to the time characteristics of the machining process, and map monitoring data with different sampling frequencies into the unified time window to achieve alignment of multi-monitoring channel data in the time dimension; after completing the temporal alignment, associate the aligned monitoring data with the corresponding identified feature-level granularity point cloud entities based on spatial mapping relationships.

[0048] It should be noted that, to eliminate the timing differences caused by different sampling frequencies, the multi-source monitoring signals are first subjected to unified spatiotemporal alignment processing in the time domain, with high-frequency signals sampled based on standard time intervals. Specifically, statistical features, including mean, root mean square, maximum, and minimum values, are extracted from the high-frequency monitoring signals within each standard time window to effectively characterize the transient changes in the signals. Low-frequency signals are directly mapped to their corresponding time windows, thus completing the fusion of high and low frequency signals in the time domain. This ensures that monitoring data from different frequencies maintain their dynamic characteristics while possessing a unified time index.

[0049] Building upon this foundation, actual sampling tool positions are introduced as the spatiotemporal correlation medium for multi-source monitoring data, enabling the mapping transition of monitoring information from the time domain to the spatial domain. Specifically, by combining time series matching and spatial coordinate matching, actual sampling tool positions are mapped one-to-one with theoretical discrete tool positions in the machining program. In the time dimension, timestamp alignment ensures that positions corresponding to the same machining moment are close to each other. In the spatial dimension, Euclidean distance or trajectory projection ensures the consistency of the two types of tool positions in three-dimensional space, thus establishing a mapping relationship between actual and theoretical tool positions. Subsequently, the multi-source monitoring data attached to the actual tool positions are associated with the corresponding theoretical discrete tool positions. The theoretical discrete tool positions are generated by discretizing the tool trajectory and are used to describe the spatial motion sequence of the tool during machining. When the tool passes through a theoretical tool position, it will perform material removal behavior on the corresponding area of ​​the part's geometric model. Since the part has been resolved into a point cloud, a spatial neighborhood relationship is established between the theoretical tool position point and the point cloud set covered by its swept body. This determines the subset of point clouds to be cut, and associates this subset with the monitoring data of the corresponding tool position point, forming a mapping link between tool position point, monitoring data, and point cloud points. Given that the same spatial point cloud may be repeatedly cut by multiple adjacent tool positions, the corresponding monitoring data exhibits multi-valued characteristics. Therefore, a statistical fusion mechanism is introduced for the multi-source monitoring data associated with a single point cloud. For example, methods such as mean, weighted mean, or time window aggregation are used to comprehensively process the data, ultimately constructing a unique monitoring attribute vector for each feature point cloud, realizing the expression of monitoring data at the point cloud level. This results in a multi-dimensional attribute structure based on point clouds. Each point cloud point not only contains its spatial coordinate information but also includes its corresponding multi-source monitoring feature parameters, such as cutting force characteristics, vibration characteristics, and spindle load change rate. This structure achieves a unified expression of geometric information and monitoring information on the same spatial carrier.

[0050] S5. Map the part inspection data to the corresponding feature-level granular point cloud entities, overall geometric-level granular point cloud entities, and single-point-level granular point cloud entities based on the spatial coordinates of the inspection points and the spatial proximity relationship of the point cloud. The part inspection data includes multi-scale inspection attribute information from macro to micro that reflects the internal stress state of the material, material properties, geometric deformation, and surface quality.

[0051] Part inspection data is spatially located using the coordinates of inspection points obtained from on-machine inspection or external inspection equipment: The spatial coordinates of inspection points are obtained from on-machine inspection or external inspection equipment, and the inspection points and the geometric point cloud of the part are placed in the same spatial coordinate system. The mapping relationship between the inspection points and the corresponding geometric points or geometric regions in the point cloud is determined by the spatial proximity relationship, and the inspection attribute information is associated with the corresponding feature-level granularity point cloud entity, the overall geometric-level granularity point cloud entity, and the single-point-level granularity point cloud entity.

[0052] It should be noted that the part inspection data contains multi-scale information from macroscopic to microscopic levels, such as residual stress, geometric deformation, material properties, and surface quality. To achieve an effective correlation between inspection information and part geometry, a spatial association mechanism between inspection data and point clouds needs to be constructed, enabling the inspection results to be mapped from the measurement coordinate system to the part point cloud space for unified representation. First, the inspection data acquired by the inspection equipment is spatially located, and the coordinates of the measurement points or the measurement area in the workpiece coordinate system are extracted. Through coordinate system transformation and calibration parameters, the inspection data is transformed from the inspection coordinate system to the unified spatial coordinate system used by the point cloud, and a spatial neighborhood relationship is established between the inspection points and point cloud points. Using methods such as nearest neighbor matching and interpolation mapping, the inspection results are mapped to the corresponding point cloud points or point cloud regions, thus forming a correlation between inspection data, spatial location, and point cloud points. To address the different scales of inspection information, a multi-scale association strategy is introduced. Macroscopic inspection data (such as residual stress distribution) is assigned to the point cloud region through field mapping to achieve region-level attribute association. Microscopic inspection data (such as surface roughness) is directly attached to the corresponding point cloud points using point-level matching to ensure accurate representation of microscopic information. Each point cloud point is expanded into a feature point carrying multidimensional inspection attributes. This structure supports spatial traceability analysis of part machining quality.

[0053] S6. By defining entities containing unique identifiers and attributes for each data object, and using the identifier of one entity as the attribute of another entity, the association relationship between data entities of different modalities, scales and granularities in S2-S5 is established, and finally a spatiotemporal association model of multimodal data of the entire CNC machining process is formed.

[0054] Example

[0055] The following section uses a typical aerospace structural component as an example to provide a detailed explanation of the construction of a multimodal, multi-scale, and multi-granular data spatiotemporal correlation model for the entire CNC machining process. Figure 1 As shown, the specific steps include:

[0056] Step 1. Taking the machining deformation control dataset of an aerospace structural component as an example. A typical aluminum alloy structural component has dimensions of 640mm × 240mm × 24mm, and the material is 7075 aluminum alloy. The residual stress field inside the material is uniform in the XY plane, but complex in the Z direction. Therefore, the residual stress field of the component is discretized into 12 regions in the Z direction, each region being a 2mm deep region in the XY plane. The residual stress is located in both the X and Y directions, i.e. and 24 residual stress values ​​can be obtained through residual stress measurement / inference methods.

[0057] Step 2. This structural component has 7 groove features with a depth of 20mm, a web thickness of 4mm, a rib thickness of 4mm, and a cutting depth of 2mm for each machining operation. The eleventh and twelfth layers are the remaining material. Therefore, a total of 70 machining operations were performed on the 7 grooves, establishing a process operation of 10 processes and 70 steps. The machining operations were analyzed, and the process information was extracted, including machine tool type, clamping method, tool information, milling method, machining process parameters, and other machining process information.

[0058] Step 3. Acquire monitoring data such as spindle current, spindle power, real-time coordinates, cutting force, and vibration during the machining process through the machine tool OPC protocol, force measuring tool holder, and vibration sensor.

[0059] Step 4. Perform numerical model analysis on the part blank model. Discretize the part blank model to extract node and element information. Represent the part as a point cloud by reading the numerical model information. Combine the theoretical tool position point with the tool shape to form the theoretically removed envelope, creating different levels of geometry, and then use the point cloud for unified representation. Figure 3 As shown.

[0060] Step 5. Read the NC-CODE code of the post-processing program generated from the part's process document, using it as the theoretically discrete tool position point. Simultaneously, interpolate the theoretical tool position points according to a fixed discretization accuracy, ensuring the interpolation accuracy is not lower than the discretization accuracy of the point cloud, ultimately forming a theoretical tool position point set. The actual tool position points during machining are obtained through the real-time coordinates of the machine tool's OPC. Both are searched sequentially forward through the machining process time sequence to find points on the theoretical tool position points that meet the matching conditions, and these points are determined as the corresponding theoretical tool position points for the current actual sampled tool position points, such as... Figure 4 As shown in the figure. In this way, a precise correspondence between the theoretical and actual tool positions is achieved in time and space, providing a foundation for multi-source data correlation modeling.

[0061] Step 6. Using a point cloud-based multi-granularity geometric model, the overall geometry is divided into a theoretical spatial domain, a machining spatial domain, and a detection spatial domain. Tool points are aligned with monitoring data along the spatial domain and machining time axis, and detection coordinate points are matched and aligned with monitoring data along the spatial domain and detection coordinate axis. This achieves the matching and association of geometry-process-monitoring-detection data. Multimodal data are associated under a unified point cloud coordinate and framework, achieving a unified representation of multi-scale, multi-granularity data. Figure 2 As shown.

[0062] Step 7. Based on point cloud, realize the spatial correlation representation of multimodal data, achieve intuitive visualization of processing parameters, monitoring information, and detection information, and combine with interactive queries to realize flexible analysis of local processing areas and global processes, such as... Figure 5 As shown.

[0063] Therefore, this invention employs the aforementioned method for constructing a multimodal data spatiotemporal correlation model for the entire CNC machining process. First, a point cloud is generated by discretizing the part model, and then divided into multi-level granular spatial entities such as points, features, and the entire system using the machining tool position sequence, establishing a data correlation framework. Next, process data is mapped to feature-level granular point cloud entities according to machining features; monitoring data with different step sampling frequencies are aligned along the time axis and then associated with corresponding feature-level granular point cloud entities based on spatial mapping relationships; detection data is matched to space through coordinate transformation and associated with entities of different granularities according to its scale. Finally, all data is attached to the corresponding spatial entities in the form of attributes, and a complete and traceable spatiotemporal correlation network of "geometry-process-monitoring-detection" is formed through mutual referencing of entity identifiers.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a spatiotemporal correlation model of multi-modal data in a whole process of numerical control machining, characterized in that, Includes the following steps: S1. Using the geometric point cloud of the part as a unified spatial representation carrier, and adopting the entity-identifier-attribute organizational structure as a unified data association paradigm; establishing multi-granularity hierarchical identification for the part geometry, including single-point level granular identification, feature level granular identification, and overall geometric level granular identification. S2. Based on the tool position sequence during the part machining process, construct feature-level granularity point cloud entities in the point cloud space. The feature-level granularity point cloud entities are composed of point cloud points corresponding to the tool sweep area. Point cloud points not covered by any machining feature corresponding sweep area belong to the overall geometric-level granularity point cloud entity. Each single point cloud after the geometric model is discretized belongs to the single point-level granularity point cloud entity. S3. Map the processing technology data to the corresponding feature-level granularity point cloud entities constructed in S2 based on their corresponding processing steps and processing features. S4. Perform spatiotemporal alignment processing on the monitoring data of the processing process, and then associate the aligned monitoring data with the corresponding feature-level granularity point cloud entities constructed in S2 based on the spatial mapping relationship. S5. Map the part inspection data to the corresponding feature-level granularity point cloud entity, overall geometric-level granularity point cloud entity, and single-point granularity point cloud entity based on the spatial coordinates of the inspection points and the spatial proximity relationship of the point cloud. S6. By defining entities containing unique identifiers and attributes for each data object, and using the identifier of one entity as the attribute of another entity, the association relationship between data entities of different modalities, scales and granularities in S2-S5 is established, and finally a spatiotemporal association model of multimodal data of the entire CNC machining process is formed. The content of S2 is as follows: Based on the discretization processing of the part's geometric model, the overall geometric point cloud is obtained, and the theoretical tool position sequence of different steps in the machining process is mapped to the point cloud space; according to the tool's geometric parameters and the theoretical tool position sequence, the spatial region swept by the tool during the machining process is determined, and the point cloud points located in the swept region are divided into and identified as feature-level granularity point cloud entities corresponding to the same machining feature; point cloud points not covered by the swept region corresponding to any machining feature are classified as overall geometric-level granularity point cloud entities; The spatiotemporal alignment process in S4 is as follows: the actual tool position point and the theoretical tool position point are matched according to the actual tool position point collected during the machining process, and then the multimodal data spatial matching and alignment is achieved according to the sweep space region determined in S2. A unified standard time window is set according to the time characteristics of the processing, and monitoring data with different sampling frequencies are mapped to the standard time window for alignment, so as to realize the alignment of data from multiple monitoring channels in the time dimension.

2. The method of claim 1, wherein the method is characterized by: The entity-identifier-attribute organizational structure represents each type of data object by constructing a triplet of entity, identifier, and attribute: ； in Represents an entity, A unique identifier that corresponds one-to-one with an entity. It represents a set of attributes for an entity; it defines data entities of different modalities, scales, and granularities, defines a unique identifier for each entity, and identifies an entity by... logo Define as another entity Attributes Establish the association between two entities in the following ways: ； Establish entity With entity The relationships between entities are defined; each entity defines multiple different types of attributes, including attributes of the entity itself and attributes associated with other entities. By selecting a specified attribute of an entity, the corresponding modality, scale, or granularity data associated with that entity can be obtained, thereby realizing the unified expression and association of machining data throughout the entire CNC machining process.

3. The method for constructing a spatiotemporal correlation model of multimodal data for the entire CNC machining process according to claim 2, characterized in that: Single-point level granularity identifiers are used to identify individual points in a point cloud; Feature-level granularity identifiers are used to identify processing feature entities composed of multiple point-level entities; The overall geometric level granularity identifier is used to identify the overall geometry of a part, and its corresponding point cloud is the overall point cloud of the part.

4. The method of claim 3, wherein the method comprises: Machining process data is obtained by extracting process information from different steps in the machining process. Its process attribute information includes at least machine tool features, tool information, machining method, step sequence and process parameters. Machining process data is mapped to the corresponding feature-level granularity point cloud entities based on the steps and machining features, realizing the spatiotemporal correlation between machining process data and part geometric point cloud.

5. The method of claim 4, wherein the method comprises: The machining process monitoring data includes multi-source monitoring attribute information such as machine tool operating status, tool movement information, and machining process monitoring information.

6. The method of claim 5, wherein the method comprises: Part inspection data includes multi-scale inspection attribute information from macro to micro, reflecting the internal stress state of the material, material properties, geometric deformation, and surface quality.

7. The method for constructing a spatiotemporal correlation model of multimodal data for the entire CNC machining process according to claim 6, characterized in that: Part inspection data is spatially located using the coordinates of inspection points obtained from on-machine inspection or external inspection equipment: The spatial coordinates of inspection points are obtained from on-machine inspection or external inspection equipment, and the inspection points and the geometric point cloud of the part are placed in the same spatial coordinate system. The mapping relationship between the inspection points and the corresponding geometric points or geometric regions in the point cloud is determined by the spatial proximity relationship, and the inspection attribute information is associated with the corresponding feature-level granularity point cloud entity, the overall geometric-level granularity point cloud entity, and the single-point-level granularity point cloud entity.

Images