CNC machining defect tracing system based on knowledge graph
By using a knowledge graph-based approach, tool wear and workpiece clamping strain areas in CNC machining are identified, solving the problem of inaccurate defect tracing in traditional methods and achieving efficient defect tracing and improved machining quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN YAYUAN PRECISION MANUFACTURING CO LTD
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-24
AI Technical Summary
Traditional CNC machining defect tracing methods struggle to effectively correlate and analyze complex and variable machining conditions such as tool wear and workpiece clamping deformation, resulting in inaccurate defect location, low tracing efficiency, and an inability to effectively improve machining quality and efficiency.
A knowledge graph-based approach is adopted. The deviation detection module extracts trajectory deviation feature nodes, and the wear screening module and strain marking module are combined. Real-time data is mapped and calculated with the pre-built knowledge graph to identify local tool wear and workpiece clamping strain concentration areas. The spatial topology structure is integrated to accurately extract defect-sensitive nodes.
It enables efficient and accurate tracing of processing defects, improves processing quality and production efficiency, and ensures the reliability and data support for processing technology optimization.
Smart Images
Figure CN122442437A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of CNC machining technology, specifically a CNC machining defect tracing system based on knowledge graphs. Background Technology
[0002] In the field of modern intelligent manufacturing, CNC machining technology has become an important means of determining product processing quality and efficiency. However, due to the combined influence of multiple factors involved in the machining process, traditional machining defect tracing methods are often limited to a single data source or static analysis, making it difficult to effectively correlate and analyze complex and ever-changing machining states such as tool wear and workpiece clamping deformation. This results in inaccurate defect location and low tracing efficiency, which seriously restricts the improvement of machining quality and production efficiency.
[0003] Currently, although some solutions have attempted to incorporate multi-source data analysis technology, they generally suffer from insufficient mining of correlation features among different factors and neglect of spatial topological relationships. This makes accurate identification of defect-sensitive areas difficult and hinders effective guidance for subsequent machining optimization. Therefore, a more efficient and accurate defect tracing system is urgently needed to achieve rapid location and precise tracing of machining defects influenced by multiple factors such as tool condition and clamping deformation, thereby improving the overall stability of CNC machining and product quality. Summary of the Invention
[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a CNC machining defect tracing system based on knowledge graph.
[0005] To achieve the above objectives, this invention provides a CNC machining defect tracing system based on a knowledge graph, comprising: The deviation detection module is used to extract trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory after obtaining theoretical and actual tool tip trajectory data. The wear screening module is used to map the local wear area of the tool through a pre-built tool wear knowledge graph based on real-time tool compensation data, and to screen tool wear sensitive nodes in the trajectory deviation feature nodes. The strain marking module is used to match the workpiece strain concentration area with the pre-built clamping strain drift knowledge graph using real-time workpiece clamping strain data, and mark the clamping strain sensitive nodes in the trajectory deviation feature nodes. The defect identification module is used to extract machining defect sensitive trajectory nodes for trajectory deviation feature nodes that are jointly covered by tool wear sensitive nodes and clamping strain sensitive nodes.
[0006] Furthermore, the extraction of trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory includes: The local curvature distribution of the trajectory nodes is determined based on theoretical blade tip trajectory data; A scale-adaptive curvature filter is used to perform multi-scale spatial filtering on the local curvature distribution to obtain a multi-scale curvature response; Trajectory deviation feature nodes are identified based on the gradient abrupt change locations of the multi-scale curvature response.
[0007] Furthermore, the multi-scale spatial filtering process performed on the local curvature distribution using a scale-adaptive curvature filter includes: The spatial scale factor of each node is determined based on the statistical characteristics of the spatial gradient of the local curvature distribution. The local curvature distribution is adjusted node by node using a spatial scale factor to complete multi-scale spatial filtering and output a multi-scale curvature response.
[0008] Furthermore, the mapping of local tool wear regions through a pre-constructed tool wear knowledge graph includes: The local pressure distribution on the tool edge is determined based on real-time tool compensation data; Node topology features are extracted from the tool wear knowledge graph through a graph node embedding-enhanced graph convolutional network. The local wear area of the tool is located by mapping the local pressure distribution with the topological characteristics of the node.
[0009] Furthermore, node topology features are extracted from the tool wear knowledge graph, including: The topology sensitivity weights of nodes are determined by analyzing the path diversity of nodes. The graph convolutional network is trained using the topology-sensitive weights to extract node topology features.
[0010] Furthermore, the matching of workpiece strain concentration regions through a pre-constructed clamping strain drift knowledge graph includes: Establish the local strain spatial distribution on the workpiece surface using real-time workpiece clamping strain data; The topological patterns of nodes are extracted from the clamping strain drift knowledge graph using a graph structure-sensitive enhanced attention network. Based on the spatial distribution of local strain on the workpiece surface and the topological pattern of the nodes, a pattern matching function is used to identify the strain concentration area of the workpiece.
[0011] Furthermore, the topological structure pattern of the nodes is extracted from the clamping strain drift knowledge graph, including: Extract the topological sensitivity factor of a node based on its local adjacency distribution; The node focusing strategy of the attention network is optimized using a topology sensitivity factor to obtain the topological structure pattern of the nodes.
[0012] Furthermore, the extraction of processing defect-sensitive trajectory nodes includes: A spatial topology correlation fusion algorithm is used to fuse the spatial topology of the tool local wear region and the workpiece strain concentration region. Calculate the sensitivity distribution based on the fused spatial topology; Based on the sensitivity distribution, the trajectory deviation feature nodes are filtered for sensitivity, and the sensitive trajectory nodes for processing defects are extracted.
[0013] Furthermore, the spatial topology of the tool local wear region and the workpiece strain concentration region is fused using a spatial topology correlation fusion algorithm, including: Extract the local topological structure information of the tool wear area and the workpiece strain concentration area respectively; The fusion computation is performed by utilizing the spatial neighborhood associations between local topologies, and the fused spatial topology is output.
[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention extracts trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory, which can effectively identify the difference between the theoretical trajectory and the actual trajectory, accurately locate feature areas that are prone to processing deviations, solve the problem of inaccurate identification of trajectory abnormality sensitive areas by traditional methods, and improve the reliability of defect tracing.
[0015] This invention constructs a knowledge graph of tool wear and clamping strain drift, and uses real-time data and the topological structure pattern of the knowledge graph for mapping calculations to achieve joint identification of local wear areas of the tool and concentrated clamping strain areas of the workpiece. This overcomes the problem that existing technologies cannot effectively correlate the influence of tool state and clamping state on machining defects, and greatly improves the accuracy of tracing the source of machining defects.
[0016] This invention fuses the spatial topology of the tool's local wear area and the workpiece's strain concentration area using a spatial topology correlation fusion algorithm, accurately extracting sensitive trajectory nodes of machining defects. This makes the location of sensitive areas clearer and more specific, thus providing more efficient and reliable data support for machining process optimization and effectively ensuring machining quality and production efficiency. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1This is a block diagram of the system of the present invention. Detailed Implementation
[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] Please see Figure 1 This embodiment provides a CNC machining defect tracing system based on knowledge graphs, including: The deviation detection module 101 is used to extract trajectory deviation feature nodes based on the location of local curvature gradient change in the trajectory after obtaining theoretical and actual tool tip trajectory data. Specifically, the extraction of trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory includes: The local curvature distribution of the trajectory nodes is determined based on theoretical blade tip trajectory data; It should be understood that theoretical tool tip trajectory data refers to the tool path planned by the CNC equipment before machining, which is usually represented as a sequence of discrete coordinate points. Local curvature distribution is used to characterize the degree of geometric change near each node on the tool tip trajectory and is an important basis for judging trajectory characteristics.
[0021] Specifically, the curvature value at each node is calculated using the spatial coordinates of adjacent nodes on the trajectory, thereby forming a local curvature sequence distributed along the entire trajectory.
[0022] For example, the spatial coordinates of three consecutive nodes on the trajectory are as follows:
[0023] It is understandable that intermediate nodes can be determined using known formulas for calculating the curvature of three points in space. Local curvature The above calculations traverse all nodes on the trajectory to obtain complete local curvature distribution data.
[0024] A scale-adaptive curvature filter is used to perform multi-scale spatial filtering on the local curvature distribution to obtain a multi-scale curvature response; It should be noted that the scale-adaptive curvature filter is used to perform scale-selective filtering on the original curvature distribution to adapt to the differences in geometric features at different scales in the trajectory and reflect the multi-scale morphological information of the trajectory.
[0025] Specifically, the scale-adaptive curvature filter comprises filter kernels at multiple scale levels. Each trajectory node determines the corresponding filter kernel size through a spatial scale factor before performing filtering operations. This method yields the filtered response of each node at different spatial scales, forming a complete multi-scale curvature response distribution.
[0026] The step of performing multi-scale spatial filtering on the local curvature distribution using a scale-adaptive curvature filter includes: The spatial scale factor of each node is determined based on the statistical characteristics of the spatial gradient of the local curvature distribution. Specifically, the spatial scale factor determines the size of the filter kernel at each node. Its value directly depends on the spatial gradient variation characteristics of the local curvature near the trajectory node, which is manifested as an inverse correlation between the gradient magnitude and the spatial scale factor.
[0027] It should be noted that the spatial scale factor of a node is obtained by calculating the curvature gradient values at the node and its neighborhood and performing statistical analysis. Nodes with larger curvature gradient magnitudes have smaller spatial scale factors; conversely, nodes with gentler gradient changes have relatively larger spatial scale factors.
[0028] For example, for a specific node on the axis motion trajectory of a CNC machine, if the calculated local curvature gradient value of the node is 0.4, which is much larger than the gradient values of the surrounding nodes (e.g., an average of 0.1), then the spatial scale factor corresponding to the node is set to a smaller value (e.g., 0.3 mm) to more sensitively capture its trajectory details.
[0029] The local curvature distribution is adjusted node by node using a spatial scale factor to complete multi-scale spatial filtering and output a multi-scale curvature response. It should be understood that once the spatial scale factor is determined, the corresponding filter kernel size is selected for each node based on this factor, thereby realizing the dynamic adjustment of the filter scale; Specifically, the scale-adaptive curvature filter uses a one-dimensional discrete Gaussian filter kernel to perform spatial convolution on the local curvature distribution, and its kernel function expression is as follows: In the formula, This represents the spatial displacement of a neighboring node relative to the central node to be processed. That is, the spatial scale factor corresponding to the node to be processed.
[0030] Specifically, based on the spatial scale factor determined by the node, a matching Gaussian filter kernel is selected to perform filtering operations on the local curvature distribution data, thereby obtaining the curvature response values of each node under different scale conditions.
[0031] For example, assuming the spatial scale factor of a specific node on a certain processing path is 0.3 mm, then the Gaussian filter kernel parameter is selected. Spatial filtering is applied to the local curvature data around the node to obtain the filtered multi-scale curvature response value of that node. After performing similar processing on all nodes, complete multi-scale curvature response data can be obtained.
[0032] Trajectory deviation feature nodes are identified based on the gradient abrupt change locations of the multi-scale curvature response; It should be noted that the identification of trajectory deviation feature nodes is based on the locations of drastic gradient changes shown in the multi-scale curvature response data. These locations usually correspond to feature regions where the theoretical trajectory differs significantly from the actual trajectory, and are prone to processing deviations.
[0033] Specifically, after calculating the gradient values of the multi-scale curvature response data along the trajectory direction, the distribution pattern of the gradient values of all nodes is statistically analyzed. Based on the distribution results, a threshold for judging gradient abrupt changes is determined, thereby accurately identifying nodes whose gradients exceed the threshold as trajectory deviation feature nodes.
[0034] For example, a multi-scale curvature response analysis is performed on a segment of the robot arm's trajectory. The gradient threshold for the multi-scale curvature response is set to 0.2. When the gradient calculation result of a node is 0.25, that node is identified as a trajectory deviation feature node. All nodes on the entire trajectory that meet this condition are marked as trajectory deviation feature nodes, thus completing the complete node identification.
[0035] The wear screening module 102 is used to map the local wear area of the tool through a pre-constructed tool wear knowledge graph based on real-time tool compensation data, and to screen tool wear sensitive nodes in the trajectory deviation feature nodes. Specifically, the mapping of local tool wear regions through a pre-constructed tool wear knowledge graph includes: The local pressure distribution on the tool edge is determined based on real-time tool compensation data; It should be noted that real-time tool compensation data specifically refers to the compensation values that the machining equipment adjusts the machining trajectory in real time based on the actual wear of the tool. By mapping the compensation values to the spatial position of the tool cutting edge, the pressure distribution characteristics corresponding to the tool cutting edge area can be obtained, which reflects the true situation of local tool wear.
[0036] Specifically, based on real-time tool compensation data and combined with tool geometry, the finite element method is used to perform force analysis on the tool cutting edge to determine the pressure distribution at various locations during workpiece machining, thus forming complete local pressure distribution data of the tool cutting edge.
[0037] For example, taking a tool used in CNC machining as an example, assuming that the real-time tool compensation data shows that the compensation amount at the tool tip reaches 0.05mm, then through finite element modeling analysis, it is found that there is a concentrated area of pressure distribution in the local area of the tool cutting edge, with a pressure value of about 350MPa, thus identifying the highly sensitive area of tool wear.
[0038] Node topology features are extracted from the tool wear knowledge graph through a graph node embedding-enhanced graph convolutional network. It should be understood that the tool wear knowledge graph is a graph-structured knowledge base pre-built based on historical data. Nodes represent different local regions of the tool cutting edge, and edges represent the wear state relationships between these regions. By using graph node embedding-enhanced graph convolutional networks, topological features can be efficiently extracted from the graph for accurate identification of local tool wear locations.
[0039] Specifically, the graph node embedding-enhanced graph convolutional network first performs high-dimensional embedding representation of nodes in the knowledge graph, then aggregates the association information between nodes layer by layer through graph convolution operations, and finally extracts the topological feature vector of each node.
[0040] Specifically, the graph node embedding process involves the initial feature vector of the nodes. The construction of the node; the initial feature vector of the node is composed of the physical service attributes of the local area of the tool, and its components include: the three-dimensional spatial coordinates corresponding to the node. The data includes the local geometric curvature value of the cutting edge, the wear resistance rating of the material in that area, and the average pressure load recorded in historical machining tasks. By performing feature extraction and vectorization on the above multidimensional attributes, a high-dimensional semantic vector that can characterize the local state of the tool is generated.
[0041] The extraction of node topology features from the tool wear knowledge graph includes: The topology sensitivity weights of nodes are determined by analyzing the path diversity of nodes. Specifically, node path diversity analysis refers to calculating the number and length distribution of paths between each node and other nodes in a knowledge graph, thereby obtaining the topological sensitivity weight of the node. The richer the path diversity of a node, the higher its topological importance in the graph, and the larger its topological sensitivity weight.
[0042] For example, if the statistical analysis of the number of paths between a certain node and other nodes in the knowledge graph of CNC machining tool wear is 35, which is significantly more than the average number of paths of other nodes in the graph (e.g., 15), then the node is given a higher topology sensitivity weight, for example, a weight of 0.8, to highlight the importance of the node.
[0043] The graph convolutional network is trained based on the topology-sensitive weights to extract node topology features; It should be noted that the training input of the graph convolutional network includes the initial feature vectors corresponding to each node in the tool wear knowledge graph and the determined topology-sensitive weights. The network parameters are trained through supervised learning, enabling the network to adaptively learn the topological relationships between nodes and finally output representative node topological structure features.
[0044] Specifically, the training process adopts a supervised learning approach, using historical wear data to mark typical patterns of tool wear, and using the node topology sensitivity weights as weighting factors of the loss function for backpropagation optimization, so that the graph convolutional network pays full attention to nodes with high topology sensitivity, thereby obtaining accurate topological structure feature representations of each node.
[0045] For example, in the knowledge graph of CNC part machining tools, if nodes with larger topology sensitivity weights are found to appear more frequently in areas of severe historical wear, the network can learn the topological characteristics of these nodes through training, such as the connection density between nodes or the local network structure, and can effectively reflect the true state of local tool wear.
[0046] Based on the mapping calculation between the local pressure distribution and the node topology characteristics, the local wear area of the tool is located; It should be understood that by mapping the real-time determined local pressure distribution on the tool edge with the node topology features extracted from the knowledge graph, local areas on the tool prone to wear can be accurately identified. This mapping process specifically involves calculating the degree of matching between the local pressure distribution data and the node topology features. A higher degree of matching indicates a greater likelihood of tool wear in that area, thus achieving accurate localization of tool wear regions.
[0047] It should be noted that before performing the mapping calculation, the spatial geometric model of the tool cutting edge and the tool wear knowledge graph are aligned in spatial coordinate system to establish the spatial mapping relationship between the local pressure distribution sampling points and the graph nodes.
[0048] Specifically, the real-time local pressure distribution of the tool is represented as a feature vector. The node topology features output by the graph convolutional network are represented as feature vectors. ,in and Let represent the pressure value and topological feature value corresponding to the i-th node, respectively. By calculating the similarity measure between the two, such as cosine similarity or Euclidean distance, the matching between the pressure feature and the topological feature is determined, thereby locating the wear area.
[0049] For example, in the tool analysis of CNC machining equipment, when the cosine similarity between the local pressure distribution vector and the topological feature vector reaches 0.92, it indicates that the pressure distribution and wear characteristics of the corresponding node area are highly matched. This area is located as the local wear-sensitive area of the tool, thereby achieving accurate screening of tool wear-sensitive nodes.
[0050] The strain marking module 103 is used to match the workpiece strain concentration area with the pre-built clamping strain drift knowledge graph by using real-time workpiece clamping strain data, and mark the clamping strain sensitive nodes in the trajectory deviation feature nodes. Specifically, the matching of workpiece strain concentration regions through a pre-constructed clamping strain drift knowledge graph includes: Establish the local strain spatial distribution on the workpiece surface using real-time workpiece clamping strain data; It should be understood that real-time workpiece clamping strain data is obtained through strain measurement sensors positioned at key locations on the workpiece. Specifically, based on the real-time measured strain values, a local strain spatial distribution corresponding to the geometric position on the workpiece surface is established. This spatial distribution is represented by a dataset containing position coordinates and corresponding strain values, reflecting the spatial distribution characteristics of the workpiece caused by clamping.
[0051] For example, during the machining of key CNC components, real-time strain measurement sensors are arranged at several key measurement points on the surface of the component. After obtaining the strain measurement values corresponding to each measurement point, the discrete strain measurement values are transformed into a continuous spatial distribution of strain on the workpiece surface through interpolation or fitting methods. This spatial distribution can intuitively reflect the changing trend of clamping strain in a local area of the workpiece, providing the necessary input basis for further identification of strain concentration areas.
[0052] The topological patterns of nodes are extracted from the clamping strain drift knowledge graph using a graph structure-sensitive enhanced attention network. It should be noted that the clamping strain drift knowledge graph is pre-constructed based on strain data from a large number of historical clamping conditions. Its nodes represent the workpiece clamping area, and the edges between nodes represent the strain interactions under clamping conditions. A graph-structure-sensitive enhanced attention network extracts features from this graph to capture the topological pattern characteristics of different nodes.
[0053] The topological structure pattern of nodes extracted from the clamping strain drift knowledge graph includes: Extract the topological sensitivity factor of a node based on its local adjacency distribution; Specifically, the local adjacency distribution of a node refers to the local connectivity state of a node in a knowledge graph. By statistically analyzing the number and strength of connections in a node's neighborhood, topological sensitivity factors that reflect the importance and sensitivity of a node in the graph can be extracted.
[0054] Preferably, for example, when a node is closely connected to multiple strain-sensitive regions, its topology sensitivity factor is relatively large. This means that the local topology of the node plays a key role in the formation of strain concentration regions, and the attention network will give priority to such nodes.
[0055] The node focusing strategy of the attention network is optimized using a topology-sensitive factor to obtain the topological structure pattern of the nodes; It should be understood that when using topology-sensitive factors for attention network optimization, the node topology-sensitive factors are specifically used as weighting parameters in the attention mechanism to affect the weight allocation of node features, thereby enhancing the network's attention to topology-sensitive nodes and making the extracted node topology patterns more reflective of clamping strain concentration features.
[0056] Specifically, the optimization strategy includes incorporating topological sensitivity factors as significant weighting factors into the calculation process of attention scores, and increasing the aggregation weight of neighboring nodes with high sensitivity features, so that the feature vector output by the network can accurately map the spatial drift law of clamping strain.
[0057] Specifically, after optimization of the node focusing strategy of the attention network, each node outputs a feature vector representing its topological structure characteristics in the graph. This feature vector can reflect the importance of the node and the characteristics of its local structural patterns in the strain drift graph.
[0058] For example, in the clamping strain analysis of CNC parts machining, through the optimized attention network, the feature vectors output by nodes with higher topology sensitivity factors can clearly represent the features related to the clamping strain concentration area. These feature vectors are then used to match the spatial distribution of real-time strain data, thereby accurately identifying the workpiece strain concentration area.
[0059] Based on the spatial distribution of local strain on the workpiece surface and the topological structure pattern of the nodes, a pattern matching of the spectrum is performed to identify the strain concentration area of the workpiece. Specifically, the method for map pattern matching is as follows: It should be noted that the real-time obtained spatial distribution of local strain on the workpiece surface is transformed into a spatial feature vector with the same form as the node features of the knowledge graph, and the matching degree between the spatial feature vector and the feature vector of the topological structure pattern of the graph nodes is calculated. During this process, it is ensured that the spatial coordinate system of the workpiece surface strain sampling is consistent with the node position mapping of the clamping strain drift knowledge graph.
[0060] Specifically, the cosine similarity method is preferred for calculating the matching degree, which gives the degree of matching between two vectors. The calculation formula is as follows: In the formula: X represents the real-time local strain spatial feature vector of the workpiece, and Y represents the topological structure pattern feature vector of the node. The Euclidean norm of a vector is denoted by .
[0061] Understandably, after the matching degree calculation is completed, a screening process is performed based on a pre-set matching threshold. The regions corresponding to nodes whose matching degree values reach the preset threshold are identified as areas of concentrated workpiece strain. These regions are highly sensitive to machining deviations caused by clamping strain.
[0062] For example, in the machining of key CNC components, when the matching degree between the strain feature vector of a certain local area and the topological feature vector of a certain node in the knowledge graph reaches or approaches a threshold (e.g., 0.85), the area is marked as a clamping strain concentration area and identified as a sensitive location that may cause machining errors.
[0063] Furthermore, after determining the strain concentration area, the corresponding trajectory deviation feature nodes are located and marked as clamping strain sensitive nodes, based on the information of the aforementioned trajectory deviation feature nodes, for the precise extraction of machining defect sensitive trajectory nodes in the next stage.
[0064] The defect identification module 104 is used to extract machining defect sensitive trajectory nodes for trajectory deviation feature nodes that are jointly covered by tool wear sensitive nodes and clamping strain sensitive nodes. Specifically, the extraction of processing defect-sensitive trajectory nodes includes: A spatial topology correlation fusion algorithm is used to fuse the spatial topology of the tool local wear region and the workpiece strain concentration region. It should be noted that the role of the spatial topology correlation fusion algorithm is to identify and fuse the trajectory regions affected by both tool wear and workpiece clamping strain, so as to more accurately identify the sensitive nodes of machining defects.
[0065] The method of fusing the spatial topology of the tool local wear region and the workpiece strain concentration region using a spatial topology correlation fusion algorithm includes: Extract the local topological structure information of the tool wear area and the workpiece strain concentration area respectively; In some specific embodiments, the local topological structure information of the tool wear area and the workpiece strain concentration area can be represented by the feature node set obtained in the previous steps, wherein each node includes specific spatial location coordinates and related feature information.
[0066] For example, the topological information of the local wear region of the tool is represented as a set of nodes:
[0067] The topological information of the strain concentration region of the workpiece is represented as follows:
[0068] In the formula: Spatial nodes representing the tool wear area and the workpiece strain concentration area, respectively. This represents the spatial coordinates of the node, where M and N are the total number of nodes in the two regions, respectively.
[0069] It is understandable that the extracted topological information, in addition to spatial coordinates, may also include feature information such as the connection relationship between nodes and the curvature change, in order to more completely express the spatial topological structure.
[0070] The fusion computation is performed by utilizing the spatial neighborhood associations between local topologies, and the fused spatial topology is output. Specifically, during the fusion calculation, the spatial neighborhood correlation strength between the node sets of the tool local wear area and the workpiece strain concentration area is calculated to determine the nodes that need to be fused.
[0071] For example, the spatial neighborhood association strength between nodes is preferably calculated using a distance-weighted neighborhood algorithm, such as the following Gaussian neighborhood weight function: In the formula, Let the neighborhood association weights be those between node i and node j. Let be the Euclidean distance between the two nodes. The preset spatial dimension parameter is, for example, 0.5-1.5mm. The specific value can be adjusted according to the accuracy requirements of the actual processing equipment.
[0072] Furthermore, when the neighborhood association strength between nodes is greater than a preset fusion threshold, the corresponding node is marked as a node to be fused, thus completing the topology fusion and forming a fused spatial topology node set. ; It should be noted that the fusion spatial topology set obtained through the above fusion calculation can effectively reveal the sensitive areas of the interaction between tool wear and workpiece clamping strain, which is helpful for the next stage of sensitivity distribution calculation and sensitive trajectory node selection.
[0073] Calculate the sensitivity distribution based on the fused spatial topology; Specifically, the processing defect sensitivity distribution represents the processing defect sensitivity of each node in the fused topology node, which facilitates the determination of subsequent processing defect sensitive trajectory nodes.
[0074] It should be understood that the specific process for calculating the sensitivity to processing defects is as follows: First, regarding the fused spatial topology node set The sensitivity index of each node is calculated separately. Preferably, the sensitivity index calculation takes into account both the degree of tool wear and the degree of workpiece strain.
[0075] For example, let the set of nodes in the merged topology be: ; Specifically, the sensitivity index of each node The calculation method is as follows:
[0076] In the formula, For nodes Sensitivity index; This represents a normalized index indicating the local wear degree of the tool at the corresponding node location. This represents a normalized index indicating the degree of workpiece clamping strain at the corresponding node position; coefficient Let be the weight coefficient, and satisfy... The specific value can be set according to the weight requirements of the actual processing task. For example, take... .
[0077] Furthermore, obtain sensitivity indicators Then, based on the set of sensitivity indicators for all nodes Normalization is performed, preferably using the Min-Max normalization method: ; Specifically, the execution process of the Min-Max normalization is as follows: Step 1, traverse the set of sensitivity indicators. The maximum sensitivity index in the set is retrieved through numerical comparison. With minimum sensitivity index Step 2: Perform linear mapping calculation on each node according to the above normalization formula; Step 3: If the above occurs... and If the values are equal, then the normalized scores of all nodes in the set are uniformly applied. Marked as 1, this achieves data unit standardization. In the formula, These represent the maximum and minimum values in the set of sensitivity indicators, respectively. This is the normalized sensitivity distribution value, ranging from 0 to 1, which makes it easier to define thresholds and screen sensitive nodes later.
[0078] It should be noted that the purpose of sensitivity distribution calculation is to clarify the relative degree of sensitivity of the fused nodes to processing defects, which is beneficial for classifying and identifying trajectory nodes.
[0079] Based on the sensitivity distribution, the trajectory deviation feature nodes are filtered for sensitivity, and the sensitive trajectory nodes for processing defects are extracted. Specifically, the method for extracting the processing defect sensitive trajectory nodes includes: Preferably, a sensitivity threshold is first set to distinguish between sensitive and non-sensitive nodes. For example, the threshold can be a value in the range of 0.75 to 0.85, and the specific value is determined according to the accuracy and process requirements of the actual processing equipment.
[0080] In practical implementation, when the node's normalization sensitivity... A node can be marked as a machining defect sensitive trajectory node when the following conditions are met: In the formula, This is the sensitivity threshold.
[0081] For example, if a threshold is set Then when node When the normalization sensitivity reaches or exceeds 0.80, the node is marked as a processing defect sensitive trajectory node, which is used to clearly indicate the specific trajectory location that is prone to causing processing defects.
[0082] It should be understood that after the above-mentioned sensitive node extraction and marking process, the obtained processing defect sensitive trajectory nodes can directly guide subsequent defect tracing and correction measures, thereby effectively improving processing quality.
[0083] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. 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 be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A CNC machining defect tracing system based on knowledge graphs, characterized in that, include: The deviation detection module is used to extract trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory after obtaining theoretical and actual tool tip trajectory data. The wear screening module is used to map the local wear area of the tool through a pre-built tool wear knowledge graph based on real-time tool compensation data, and to screen tool wear sensitive nodes in the trajectory deviation feature nodes. The strain marking module is used to match the workpiece strain concentration area with the pre-built clamping strain drift knowledge graph using real-time workpiece clamping strain data, and mark the clamping strain sensitive nodes in the trajectory deviation feature nodes. The defect identification module is used to extract machining defect sensitive trajectory nodes for trajectory deviation feature nodes that are jointly covered by tool wear sensitive nodes and clamping strain sensitive nodes.
2. The system according to claim 1, characterized in that, The extraction of trajectory deviation feature nodes based on the location of local curvature gradient changes in the trajectory includes: The local curvature distribution of the trajectory nodes is determined based on theoretical blade tip trajectory data; A scale-adaptive curvature filter is used to perform multi-scale spatial filtering on the local curvature distribution to obtain a multi-scale curvature response; Trajectory deviation feature nodes are identified based on the gradient abrupt change locations of the multi-scale curvature response.
3. The system according to claim 2, characterized in that, The process of performing multi-scale spatial filtering on the local curvature distribution using a scale-adaptive curvature filter includes: The spatial scale factor of each node is determined based on the statistical characteristics of the spatial gradient of the local curvature distribution. The local curvature distribution is adjusted node by node using a spatial scale factor to complete multi-scale spatial filtering and output a multi-scale curvature response.
4. The system according to claim 3, characterized in that, The mapping of local tool wear regions using a pre-constructed tool wear knowledge graph includes: The local pressure distribution on the tool edge is determined based on real-time tool compensation data; Node topology features are extracted from the tool wear knowledge graph through a graph node embedding-enhanced graph convolutional network. The local wear area of the tool is located by mapping the local pressure distribution with the topological characteristics of the node.
5. The system according to claim 4, characterized in that, Extracting node topology features from the tool wear knowledge graph includes: The topology sensitivity weights of nodes are determined by analyzing the path diversity of nodes. The graph convolutional network is trained using the topology-sensitive weights to extract node topology features.
6. The system according to claim 5, characterized in that, The matching of workpiece strain concentration regions through a pre-constructed clamping strain drift knowledge graph includes: Establish the local strain spatial distribution on the workpiece surface using real-time workpiece clamping strain data; The topological patterns of nodes are extracted from the clamping strain drift knowledge graph using a graph structure-sensitive enhanced attention network. Based on the spatial distribution of local strain on the workpiece surface and the topological pattern of the nodes, a pattern matching function is used to identify the strain concentration area of the workpiece.
7. The system according to claim 6, characterized in that, Extracting the topological pattern of nodes from the clamping strain drift knowledge graph includes: Extract the topological sensitivity factor of a node based on its local adjacency distribution; The node focusing strategy of the attention network is optimized using a topology sensitivity factor to obtain the topological structure pattern of the nodes.
8. The system according to claim 7, characterized in that, The extraction of processing defect-sensitive trajectory nodes includes: A spatial topology correlation fusion algorithm is used to fuse the spatial topology of the tool local wear region and the workpiece strain concentration region. Calculate the sensitivity distribution based on the fused spatial topology; Based on the sensitivity distribution, the trajectory deviation feature nodes are filtered for sensitivity, and the sensitive trajectory nodes for processing defects are extracted.
9. The system according to claim 8, characterized in that, The spatial topology structure of the tool local wear region and the workpiece strain concentration region is fused using a spatial topology correlation fusion algorithm, including: Extract the local topological structure information of the tool wear area and the workpiece strain concentration area respectively; The fusion computation is performed by utilizing the spatial neighborhood associations between local topologies, and the fused spatial topology is output.