A mold defect recognition method and system based on artificial intelligence

Abstract

The present application relates to the technical field of defect identification, in particular to a mold defect identification method and system based on artificial intelligence, comprising the following steps: analyzing the change of adjacent measurement points based on three-dimensional height data, screening abnormal points, extracting defect edge track, comparing direction vector and included angle to identify same-direction edge section, integrating parameters combined with jump points and regional characteristics, sorting and dividing response levels, and obtaining defect part response sequence. The present application refines the spatial positioning and partitioning of surface abnormal points by fusing spatial height information and boundary trend characteristics, progressively extracts the boundary continuity and jump point distribution, realizes hierarchical sorting of defect division, strengthens the capture ability of small structure mutations and complex morphology regions of the mold, promotes the transition of defect response sorting from single feature judgment to multi-feature collaborative decision making, realizes efficient and systematic multi-defect hierarchical scheduling in automatic detection, and enhances the depth of defect risk identification and detection rationality in batch process.

Description

Technical Field

[0001] This invention relates to the field of defect recognition technology, and in particular to a mold defect recognition method and system based on artificial intelligence. Background Technology

[0002] The field of defect identification involves the detection and classification of product surfaces or structures using image acquisition and analysis methods. This technical field encompasses the construction of detection systems, the acquisition and preprocessing of defect images, the extraction and analysis of feature parameters, and the application of pattern recognition methods in various product quality inspection scenarios. It is widely used in automated quality control and inspection processes in industrial production. Traditional mold defect identification refers to the process of detecting defects such as cracks, pores, and pits by manually inspecting or using image processing methods relying on fixed thresholds during mold manufacturing or use, processing two-dimensional grayscale or color images acquired from the mold surface, and then combining this with feature comparison.

[0003] Existing technologies, which rely primarily on single grayscale or color parameters, have limitations in processing three-dimensional information such as height and distribution for defect characterization. Parameter identification methods mainly depend on fixed standards or human experience, resulting in limited accuracy in identifying the spatial distribution and structural anomalies of abnormal points. In practice, when the surface structure of the mold is complex or defects manifest as minor undulations and abrupt changes in distribution, there are often delays in response judgment and confusion in priorities. The accuracy of defect identification is unstable, resource scheduling is unbalanced, and it is difficult to provide timely warnings of product quality risks in complex production processes. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an artificial intelligence-based method and system for identifying mold defects.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a mold defect identification method based on artificial intelligence, comprising the following steps:

[0006] S1: Based on the three-dimensional height data of the mold surface, analyze the height changes of adjacent measurement points, determine the height difference trend, compare the spatial distribution of jump points, screen out abnormal points under the resolution accuracy of the equipment, optimize the spatial matching of measurement points and abnormal points, and obtain an abnormal structure distribution layer.

[0007] S2: Based on the abnormal structure distribution layer, analyze the contour edge of the defect area in the layer, calculate the coordinate difference between adjacent edge points, determine the direction change, optimize the direction vector arrangement, filter the true edge direction trajectory, adjust the edge point order, and obtain the edge segment direction trajectory sequence.

[0008] S3: Based on the edge segment directional trajectory sequence, compare the changes in directional vectors of each segment, analyze the trend of the defect boundary, calculate the angle between adjacent directional vectors, determine whether the angle is within the reference bandwidth range, identify the same-direction aligned segments, and obtain the boundary contour overlap ratio index.

[0009] S4: Based on the boundary contour overlap ratio index, analyze the distribution of jump points and defect areas, calculate the ratio of jump points to area within the region, compare the slope changes of jump points, determine the relationship between slope and area ratio, and obtain the structural jump amplitude parameter.

[0010] The present invention is improved in that the abnormal structure distribution layer includes surface abnormal distribution data, abnormal location identifiers, and feature marker set; the edge segment direction trajectory sequence includes edge direction information, trajectory segment index, and continuous contour set; the boundary contour overlap ratio index includes overlap ratio value, overlap segment index, and overlap trend parameter; and the structure jump ratio parameter includes density distribution parameter and slope change parameter.

[0011] The present invention is improved in that the specific steps for obtaining the abnormal structure distribution layer are as follows:

[0012] S111: Based on the three-dimensional height data of the mold surface, collect the three-dimensional coordinates and height data of the measurement points in the detection area, and extract the height information between adjacent measurement points in sequence to construct the height difference sequence between adjacent measurement points and obtain the height difference trend sequence.

[0013] S112: Based on the height difference trend sequence, compare the height differences of adjacent data within the sequence, analyze the difference fluctuations using a continuous sliding window method, screen for jump points based on the abnormal jump benchmark, statistically analyze the distribution pattern of jump points in the detection area, and then extract the spatial density, arrangement orientation and concentration of jump points in each local segment to obtain the jump point distribution index.

[0014] S113: Based on the jump point distribution index, sequentially match the spatial coordinates of the jump points and the measurement points, optimize the matching path and eliminate duplicate matching items, integrate the density parameters, position offset parameters and spatial correspondence characteristics of the jump points, retrieve the spatial mapping region of the jump points in the three-dimensional structure, and obtain the abnormal structure distribution layer.

[0015] The present invention is improved in that the step of obtaining the edge segment direction trajectory sequence is specifically as follows:

[0016] S211: Based on the abnormal structure distribution layer, select adjacent edge points, calculate the horizontal coordinate difference and vertical coordinate difference of each group of edge points, generate direction vectors, determine the spatial direction change trend of each direction vector, count the node positions where continuous direction changes occur, and obtain the direction change node index sequence.

[0017] S212: Based on the direction change node index sequence, extract continuous direction vector segments. According to the spatial direction distribution of each vector segment, measure the spatial angle of each segment. Compare the angle of each segment with the boundary continuity index, and select vector segments with spatial angles close to the boundary continuity standard to obtain a sequence of segments with consistent spatial angles.

[0018] S213: Based on the spatial angle consistent segment sequence, extract all associated edge points, adjust the arrangement order according to the spatial coordinate relationship between each point, calculate the spatial continuity change during the arrangement process point by point, find the optimal point sequence combination of arrangement perturbation, and obtain the edge segment direction trajectory sequence.

[0019] The present invention is improved in that the step of obtaining the boundary contour overlap ratio index is specifically as follows:

[0020] S311: Based on the edge segment direction trajectory sequence, compare the spatial angles of each pair of adjacent direction vectors, determine the correspondence between each angle and the reference bandwidth interval of the boundary angle, filter the segment pairs whose spatial angles satisfy the reference bandwidth interval, and calculate the proportion of the filtered segment pairs to all adjacent segment pairs to obtain the ratio of the same direction segment pairs.

[0021] S312: Based on the ratio of the same-direction segments, calculate the total length of the same-direction segments on the boundary line, compare the total length of the same-direction segments with the length of the boundary line, analyze the proportional relationship between the two, and determine the distribution characteristics of the same-direction segments in the boundary contour to obtain the ratio of the same-direction lengths.

[0022] S313: Based on the aforementioned length ratio in the same direction, the length ratios of the segments in the same direction are accumulated using a weighted superposition method, and combined with the total boundary length ratio data, the boundary contour overlap ratio index is obtained.

[0023] The present invention is improved in that the step of obtaining the structural jump ratio parameter is specifically as follows:

[0024] S411: Based on the boundary contour overlap ratio index, analyze the transition points and spatial range within each region, calculate the ratio between the number of transition points and the corresponding spatial area in each region, determine the density of transition point distribution, and obtain the transition point area ratio index.

[0025] S412: Compare the area ratio of the jump points in each region, analyze the slope changes of the jump points on the contour boundary, optimize the calculation order of the slope difference between each jump point, and obtain the set of slope differences of the jump points.

[0026] S413: Based on the set of slope differences at the transition points, filter the numbering information of each region, calculate the joint characteristics of the area ratio of the transition points and the slope differences, and obtain the structural transition amplitude parameter.

[0027] The present invention is improved in that the steps further include:

[0028] S5: Based on the structural jump amplitude parameter, determine the order of parameters in the defect area, compare and sort the parameter performance of each area, filter the preceding number to be classified as high response level, and classify the rest as low response level, and analyze the overlap ratio and number mapping to obtain the response sequence of the defect location.

[0029] The defect location response sequence includes a high response number sequence, a low response number sequence, and a hierarchical sorting list.

[0030] The present invention is improved in that the step of obtaining the response sequence of the defective part is specifically as follows:

[0031] S511: Based on the structural jump amplitude parameters, compare the density distribution, slope change and amplitude characteristics of each defect region, calculate the arrangement order of the parameters in the task queue, determine the relative position of each region in the sequence, and obtain the jump parameter sequence arrangement index.

[0032] S512: Based on the jump parameter sequence arrangement index, filter priority regions, analyze the sequence distribution of the corresponding numbers of the regions, optimize the classification method of high response and low response levels, adjust the level distribution, and obtain the response level classification sequence structure.

[0033] S513: Based on the response hierarchy classification sequence structure, determine the corresponding number group, analyze the combination distribution of the number under the boundary contour overlap ratio index, compare the correspondence between the number and the contour distribution, optimize the number sorting, and obtain the response sequence of the defect location.

[0034] An artificial intelligence-based mold defect identification system, the system comprising:

[0035] The abnormal structure extraction module analyzes the height changes between adjacent measurement points based on the three-dimensional height data of the mold surface, determines the trend range of the continuous height difference sequence, compares the spatial distribution pattern of jump points, filters out height anomalies that match the resolution accuracy of the matching equipment, optimizes the spatial matching relationship between the coordinates of each measurement point and the anomaly points, and obtains the abnormal structure distribution layer.

[0036] The edge trajectory construction module analyzes the shape of the defect area contour edge based on the abnormal structure distribution layer, calculates the difference between the horizontal and vertical coordinates of each group of adjacent edge points, judges the directional changes between edge points, optimizes the arrangement and combination order of the directional vectors, filters the directional trajectory segments that are consistent with the real edge state, and then adjusts the edge point wrapping order to obtain the edge segment directional trajectory sequence.

[0037] The contour overlap recognition module compares the changes in the direction vectors of each segment based on the edge segment direction trajectory sequence, analyzes the trend of the direction after the mold defect boundary segment is normalized, calculates the angle between adjacent direction vectors, determines whether the angle is within the reference bandwidth range of the mold detection boundary angle, identifies edge segments that are aligned in the same direction, and obtains the boundary contour overlap ratio index.

[0038] Based on the boundary contour overlap ratio index, the structural feature integration module analyzes the distribution relationship between the jump points and each defect area, calculates the proportion of the space area occupied by the jump points in each area, compares the contour slope change parameters corresponding to all jump points, judges the relationship between the slope change and the area ratio of the area, optimizes the data integration method of each jump point parameter, and obtains the structural jump amplitude parameter.

[0039] The response sorting and grading module determines the order of each parameter in the defect area of ​​the mold inspection task queue based on the structural jump ratio parameter, compares the parameter performance between regions and sorts the parameters, filters the top-ranked numbers as high response levels, and classifies the remaining numbers as low response levels. It also analyzes the corresponding mapping between the boundary contour overlap ratio index and the region number to obtain the response sequence of the defect location.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0041] In this invention, by fusing spatial height information with boundary orientation features, the spatial positioning and partitioning of surface anomalies are refined. Based on multi-parameter linkage screening, the responsiveness to local structural changes is improved. The progressive extraction of boundary continuity and abrupt point distribution enables hierarchical sorting of defects, strengthens the ability to capture micro-structural mutations and complex morphological regions of molds, promotes the defect response sorting from single feature judgment to multi-feature collaborative decision-making, improves the discrimination adaptability to complex defect scenarios, realizes efficient and systematic multi-defect hierarchical scheduling in automated detection, and significantly enhances the depth of defect risk identification and detection rationality in batch processes. Attached Figure Description

[0042] Figure 1 This is a flowchart of the main steps of the present invention;

[0043] Figure 2 This is a flowchart illustrating the process of obtaining the abnormal structure distribution layer in this invention;

[0044] Figure 3 This is a flowchart illustrating the process of obtaining the edge segment directional trajectory sequence in this invention.

[0045] Figure 4 This is a flowchart illustrating the process of obtaining the boundary contour overlap ratio index in this invention.

[0046] Figure 5This is a flowchart illustrating the process of obtaining the structural jump ratio parameter in this invention.

[0047] Figure 6 This is a flowchart of the process for obtaining the response sequence of the defect location in this invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0049] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0050] Example: Please refer to Figure 1 This invention provides a technical solution: a mold defect identification method based on artificial intelligence, comprising the following steps:

[0051] S1: Based on the three-dimensional height data of the mold surface, analyze the height changes between adjacent measurement points, determine the trend range of the continuous height difference sequence, compare the spatial distribution pattern of jump points, screen out height anomalies that match the resolution accuracy of the matching equipment, optimize the spatial matching relationship between the coordinates of each measurement point and the anomalies, and obtain the anomaly structure distribution layer.

[0052] S2: Based on the abnormal structure distribution layer, analyze the shape of the defect area contour edge, calculate the difference between the horizontal and vertical coordinates of each group of adjacent edge points, determine the directional changes between edge points, optimize the arrangement and combination order of direction vectors, filter the direction trajectory segments that are consistent with the real edge state, and then adjust the edge point wrapping order to optimize the continuity of the area contour and obtain the edge segment direction trajectory sequence.

[0053] S3: Based on the edge segment directional trajectory sequence, compare the changes of directional vectors of each segment, analyze the trend of the direction after the normalization of the mold defect boundary segment, calculate the angle between adjacent directional vectors, determine whether the angle is within the reference bandwidth range of the mold detection boundary angle, identify edge segments that are aligned in the same direction, and obtain the boundary contour overlap ratio index.

[0054] S4: Based on the boundary contour overlap ratio index, analyze the distribution relationship between the jump points and each defect area, calculate the proportion of the space area occupied by the jump points in each area, compare the contour slope change parameters corresponding to all jump points, determine the relationship between the slope change and the area ratio of the area, optimize the data integration method of each jump point parameter, and obtain the structural jump amplitude parameter.

[0055] S5: Based on the structural jump amplitude parameter, determine the order of each parameter in the defect area of ​​the mold inspection task queue, compare the parameter performance between regions and sort the parameters, filter the number of the first sorted number to classify as high response level, classify the remaining number as low response level, analyze the corresponding mapping between the boundary contour overlap ratio index and the region number, optimize the region response sorting, and obtain the response sequence of the defect location.

[0056] The abnormal structure distribution layer includes surface abnormal distribution data, abnormal location identifiers, and feature marker sets. The edge segment direction trajectory sequence includes edge direction information, trajectory segment index, and continuous contour set. The boundary contour overlap ratio index includes overlap ratio value, overlap segment index, and overlap trend parameter. The structural jump amplitude parameter includes density distribution parameter, slope change parameter, and comprehensive amplitude index. The defect location response sequence includes high response number sequence, low response number sequence, and hierarchical sorting list.

[0057] In S1, a measurement point refers to each specific spatial coordinate position acquired on the mold surface by a 3D scanning device or height sensor, representing a specific surface point; the height difference sequence refers to the set of height change data obtained by point-by-point difference of the height values ​​of adjacent measurement points, used to analyze the trend of surface height change; a jump point refers to a point in the height difference sequence where the change amplitude suddenly and drastically changes, reflecting the spatial location of small protrusions, depressions, or defects on the surface; the distribution pattern refers to the spatial distribution of jump points on the entire mold surface or within a specified detection area, including statistical characteristics such as dense areas, sparse areas, and arrangement trends; the equipment resolution accuracy refers to the minimum height change range that the 3D scanning device or height sensor can distinguish, i.e., the detection accuracy standard used to screen valid anomalies; a height anomaly point refers to a measurement point that is identified as exceeding the normal change range and exhibiting abnormal fluctuations when analyzing height data, and the point usually corresponds to an actual surface defect; the spatial matching relationship refers to the one-to-one correspondence between the detected height anomaly points and the actual mold surface coordinates, ensuring that subsequent defect analysis can be accurately located in three-dimensional space.

[0058] In S2, the difference between the lateral and longitudinal coordinates of adjacent edge points refers to the difference in coordinates along the X-axis (lateral) and Y-axis (longitudinal) between two adjacent edge points on the boundary line of the defect area, reflecting the specific direction of the boundary; directional change refers to the change in the direction of the boundary line between points, that is, the change in the orientation of the line segment from one point to the next, used to reflect the curvature or straightness of the boundary shape; the arrangement and combination order of direction vectors refers to arranging the direction vectors formed between all adjacent edge points in the order of the boundary line encircling, forming a complete and continuous description of the boundary direction; the true edge state refers to the boundary shape and spatial position of the actual physical surface when the mold defect actually exists, used as the standard for selecting direction trajectories; the direction trajectory segment refers to the continuous direction vector sequence that is highly consistent with the true defect boundary shape after screening among all boundary direction vectors, reflecting the local true contour; the encircling order refers to arranging the points on the boundary line in a clockwise or counterclockwise order according to a specific starting point, used to describe the overall coherence of the boundary.

[0059] In S3, the change of direction vector in each segment refers to the difference in orientation of the direction vector between different segments along the boundary of the defect area, which is used to depict the trend of boundary contour change; the trend after direction normalization refers to the standardization of each segment's direction vector (such as unit length) to compare the direction differences and overall shape between different segments; the included angle reference bandwidth range refers to the numerical range used to judge whether the included angle between boundary segments meets a specific standard, representing the normal or expected alignment tolerance range; the edge segments aligned in the same direction refer to the boundary segments whose included angle is within the reference bandwidth range and whose direction height is consistent, indicating that the spatial orientation of the segments tends to be consistent.

[0060] In S4, the distribution relationship refers to the spatial arrangement characteristics of the jump points in each defect region, including their distribution density and positional relationship within the defect region; the contour slope change parameter refers to the slope (i.e., elevation change slope) of each jump point on the region boundary, used to reflect the abrupt changes in the defect boundary in space; the relationship with the area ratio refers to the ratio between the number of all jump points within the region and the total area of ​​the defect region, used to measure the complexity or degree of abrupt change of the defect structure; the data integration method refers to the unified processing of parameters such as jump points, slope changes, and area ratios through statistical, weighted, and normalized operations to obtain structural characteristic parameters.

[0061] In S5, the mold inspection task queue refers to the set of all defect areas that the system needs to process sequentially during the automated mold defect inspection process. It is the object of defect identification, sorting, and scheduling. The sorting order refers to the sequential ranking of all defect areas based on feature parameters, usually used for priority determination or subsequent inspection resource allocation. Parameter sorting refers to determining the priority order of each area by comparing the size of the feature parameters of each defect area, which facilitates subsequent hierarchical processing. High response level refers to the group of defect areas that are identified as requiring higher priority or processing after parameter sorting, usually corresponding to areas with high risk or prominent features. Low response level refers to the group of defect areas that are identified as requiring subsequent processing or with lower risk after parameter sorting, usually with inconspicuous features. The mapping to the area number refers to the one-to-one correspondence between the parameter results obtained from the analysis and the number of each defect area in the system, which facilitates result management and tracking.

[0062] Please see Figure 2 The specific steps for obtaining the abnormal structure distribution layer are as follows:

[0063] S111: Based on the three-dimensional height data of the mold surface, collect the three-dimensional coordinates and height data of the measurement points in the detection area, and extract the height information between adjacent measurement points in sequence to construct the height difference sequence between adjacent measurement points and obtain the height difference trend sequence.

[0064] A total of 10,000 measurement points were acquired using a 3D scanning device at 0.5mm intervals. Each measurement point recorded its two-dimensional position in the x and y directions and its corresponding z-axis height value. The measurement point data were arranged sequentially in the horizontal direction, and the height difference between any two adjacent points was extracted sequentially. Points with a height difference exceeding 0.12mm were marked as potential jump data points. This difference was calculated by multiplying the difference by a redundancy factor of 1.2 within a ±0.05mm allowable error range. After processing all the horizontal data, the same height difference calculation was performed in the vertical direction, resulting in two height difference sequences in each direction. A sliding window operation was then used, selecting five consecutive difference data points at a time to calculate their fluctuation range. When the fluctuation range exceeded 0.12mm, the position corresponding to the center of the window was identified as an outlier. After completing the above operations, all jump points were extracted. The detection area is positioned on the original x, y coordinate plane and divided into 100 grid cells based on a fixed 5mm × 5mm grid. The jump points are then assigned to the corresponding grid cells according to their spatial coordinates. The number of jump points in each grid cell is counted. If a grid cell contains 4 or more jump points, it is marked as a dense jump area because the density of 4 points in a 25mm² area is 0.16 points / mm². The position and orientation of the jump points in the dense area are then analyzed. For example, the orientation angle is obtained by measuring the average direction of the line connecting all jump points relative to the geometric center of the grid. The distance from each jump point to the main direction line is used as a concentration reference value. If the average distance of multiple points is less than 0.3mm, the points are considered to be closely arranged and have a consistent distribution direction. Finally, the jump density, orientation angle, and concentration level in each grid cell are output as the jump distribution index dataset for that area.

[0065] S112: Based on the height difference trend sequence, the height difference between adjacent data in the sequence is compared, and the difference fluctuation is analyzed by using a continuous sliding window method. Jump points are screened according to the abnormal jump benchmark, and the distribution pattern of jump points in the detection area is statistically analyzed. Then, the spatial density, arrangement orientation and concentration of jump points in each local segment are extracted to obtain the jump point distribution index.

[0066] Based on the labeled jump distribution index data, each jump point is matched in the three-dimensional measurement point set. First, according to the jump point's number index in the data, the corresponding measurement point in spatial coordinates is found. If the point falls within 3μm of a measurement point, the two are confirmed to be in the same position and the match is completed. In the case of multiple jump points close to the same measurement point, the jump density value with higher value will be retained first, and the rest will be discarded as redundancy to prevent multiple outliers from pointing to the same measurement point. If the spatial distance between jump points is less than 3mm, their arrangement order will be reordered so that the close points are arranged first, thereby obtaining a reasonable spatial path streamline. Then, three feature parameters are extracted for each matched jump point: the first is the jump density value of its grid, for example, between 0.16 and 0.25 jump points / mm². The second is the spatial offset value between the measurement point and the measurement point in the x, y, and z directions. If this value is within 0.05 mm, it indicates that the spatial consistency is tight. The third is the Euclidean distance between the two spatial points, which is used as a reliability reference for the spatial correspondence. If this distance is within 0.08 mm, it is considered a good matching point. Subsequently, the matching relationship is constructed into a point dataset required for layer drawing. Each point will be represented by a different color to indicate its jump density level (for example, red represents a density higher than 0.25 points / mm², yellow represents a density between 0.16 and 0.25 points / mm², and blue represents a density lower than 0.16 points / mm²), and its arrangement direction is marked with an arrow. The radius of the dot indicates the concentration of jump points. Finally, the abnormal structure distribution layer with graphic markers is output in the detection interface for defect visualization processing in three-dimensional space.

[0067] S113: Based on the distribution index of jump points, sequentially match the spatial coordinates of jump points and measurement points, optimize the matching path and eliminate duplicate matching items, integrate the density parameters, position offset parameters and spatial correspondence characteristics of jump points, retrieve the spatial mapping area of ​​jump points in the three-dimensional structure, and obtain the abnormal structure distribution layer.

[0068] After reading all point data from the abnormal structure distribution layer, the jump point number, its 3D coordinates, grid number, and arrangement direction angle are extracted one by one. Based on the arrangement direction, the direction difference between any two adjacent jump points is extracted. If the direction difference is less than 10 degrees, the two points are considered to belong to the same direction segment. Multiple points with similar continuous directions are then organized into a segment, forming multiple consecutive point segments. Within each segment, the point sequence is arranged according to its actual spatial position. If the distance between points exceeds 3mm, it is moved to the next segment. Then, by scanning the point cloud within each segment, its smallest rectangular bounding box in the 2D plane is constructed, and the jump points within the rectangle are counted. The number of points is counted and their regional density is calculated. For example, if a paragraph contains 8 jump points and occupies a rectangular area of ​​32 mm², the density is 0.25 points / mm², which exceeds the average density benchmark value of 0.2 points / mm² set in the historical analysis samples. Therefore, it can be classified as a high-density jump segment. At the same time, the structural continuity is judged based on the arrangement order and direction consistency of the jump points in the paragraph. If there are no jump points or large jumps in the arrangement order of all points, the paragraph structure remains intact. Finally, the paragraphs that are judged to be high-density, consistent in direction, and continuous in arrangement are extracted from the layer as complete spatial structural anomaly areas. Combined with the spatial location, arrangement direction, and aggregation attribute data of the corresponding jump points, a new layer data structure is generated.

[0069] Please see Figure 3 The specific steps for obtaining the edge segment directional trajectory sequence are as follows:

[0070] S211: Based on the abnormal structure distribution layer, select adjacent edge points, calculate the horizontal and vertical coordinate differences of each group of edge points, generate direction vectors, determine the spatial direction change trend of each direction vector, count the node positions where continuous direction changes occur, and obtain the direction change node index sequence.

[0071] Select pairs of edge points sequentially based on their adjacency in spatial coordinates. The difference in the lateral coordinates of each pair is obtained by subtracting the X-axis coordinates of the two points from each other, and the difference in the longitudinal coordinates is obtained by subtracting the Y-axis coordinates of the two points from each other. Record the numerical range of the lateral and longitudinal differences in millimeters; for example, the lateral difference is between -2.0 mm and 2.0 mm, and the longitudinal difference is between -1.5 mm and 1.5 mm. Generate a direction vector by combining the lateral and longitudinal differences, and record the spatial orientation angle of this vector, which is between 0° and 360°. The values ​​are taken within the range, and the spatial trend of each directional vector is judged in turn. That is, the absolute value of the angle difference between two adjacent directional vectors is taken and compared with the directional change reference interval. The directional change reference interval is set with reference to the physical continuity of the boundary contour. For example, the angle is less than 15° and it is determined that the direction is basically the same. The angle is greater than 45° and it is determined that the direction has changed significantly. When the directional vector change falls into the significant change interval, the position is marked as the directional change node and its index position in the edge point sequence is recorded. The node numbers of all nodes that have changed significantly are counted to obtain the directional change node index sequence.

[0072] S212: Based on the direction change node index sequence, extract continuous direction vector segments. According to the spatial direction distribution of each vector segment, measure the spatial angle of each segment. Compare the angle of each segment with the boundary continuity index, and select vector segments with spatial angles close to the boundary continuity standard to obtain a sequence of segments with consistent spatial angles.

[0073] The original direction vector sequence is segmented according to the index position. The vector set between every two direction change nodes is extracted as a continuous direction vector segment. For each vector in the segment, its spatial direction angle is read sequentially from the start point to the end point, and the direction angle difference between adjacent vectors is compared. The difference is compared with the boundary continuity index. The value of the boundary continuity index is set with reference to the continuity of the actual mold defect edge. For example, the direction angle difference is less than 10° and is considered highly continuous. The direction angle difference is between 10° and 20° and is considered acceptablely continuous. The difference is greater than 20° and is considered discontinuous. For the angle difference of all adjacent vectors in each segment, the proportion of those falling into the highly continuous range is counted. If the proportion is not less than 80%, the segment is selected as a segment with a spatial angle close to the boundary continuity standard. Otherwise, the segment is removed. During this selection process, the segment numbers that meet the conditions are recorded sequentially to form a segment sequence with consistent spatial angles. Each element in this sequence contains the segment's start edge point number, end edge point number, and a description of the overall direction angle distribution of the segment.

[0074] S213: Based on the spatial angle consistent segment sequence, extract all associated edge points, adjust the arrangement order according to the spatial coordinate relationship between each point, calculate the spatial coherence change during the arrangement process point by point, find the optimal point sequence combination of arrangement perturbation, and obtain the edge segment direction trajectory sequence.

[0075] Extract the set of edge points involved in all segments, and sort the edge points according to their spatial positional relationship in the X, Y, and Z directions. First, sort them by X-direction coordinate value from smallest to largest. If there are identical X-values, sort them by Y-direction coordinate value, and finally sort them by Z-direction. Check the spatial continuity change of adjacent edge points point by point. The change is calculated by subtracting the displacement of adjacent points in the X, Y, and Z directions one by one, and comparing the absolute value of the difference with the continuity threshold. The continuity threshold is set with reference to the maximum allowable edge point spacing on the mold surface. For example, the threshold is 0.5 mm. When the displacement of adjacent points in any direction exceeds 0.5 mm, it is marked as a continuity interruption point. During the adjustment process, different combinations of adjacent point sequences in the segment are tested. The number of continuity interruption points under each sequence is counted. The sequence with the fewest interruption points is the optimal sequence, and the sequential number of the edge points is output. The continuous set of edge points formed by the sequential numbers is the edge segment directional trajectory sequence.

[0076] Please see Figure 4 The specific steps for obtaining the boundary contour overlap ratio index are as follows:

[0077] S311: Based on the edge segment direction trajectory sequence, compare the spatial angle between each pair of adjacent direction vectors, determine the correspondence between each angle and the reference bandwidth interval of the boundary angle, filter the segment pairs whose spatial angles satisfy the reference bandwidth interval, and calculate the proportion of the filtered segment pairs to all adjacent segment pairs to obtain the ratio of the same direction segment pairs.

[0078] The spatial orientation angle data of each direction vector in the sequence is read, and adjacent vectors are paired sequentially to form segment pairs. For each pair of adjacent vectors, their orientation angle values ​​are extracted, and the angle difference between them is calculated. The angle difference is calculated by directly subtracting the two vectors and taking the absolute value. For example, when the first vector's orientation angle is 35° and the second vector's orientation angle is 40°, the angle difference is 5°. Then, the obtained angle difference is compared one by one with a pre-set boundary angle reference bandwidth interval. The value of this interval is based on the continuity statistics of the mold defect boundary. The bandwidth range is set, for example, 0° to 10° for scenarios with high continuity requirements, 10° to 20° for medium continuity, and is considered discontinuous if it exceeds 20°. For each segment pair, if the difference in their included angles falls within the reference bandwidth range, the segment pair is marked as a co-directional segment pair; otherwise, it is marked as a non-co-directional segment pair. After comparing all segment pairs, the number of all co-directional segment pairs is counted, and the ratio of this number to the total number of segment pairs is calculated. For example, if there are 42 co-directional segment pairs and 50 total segment pairs, the ratio is 84%, which is the co-directional segment pair ratio.

[0079] S312: Based on the ratio of the same-direction segments, calculate the total length of the same-direction segments on the boundary line, compare the total length of the same-direction segments with the length of the boundary line, analyze the proportional relationship between the two, and determine the distribution characteristics of the same-direction segments in the boundary contour to obtain the ratio of the same-direction lengths.

[0080] For all boundary segments marked as pairs of segments running in the same direction, their length data on the boundary line is obtained one by one. The length is calculated by statistically analyzing the distance between the start and end points of the segment in three-dimensional space, in millimeters. The lengths of all segments running in the same direction are summed to obtain the total length. For example, if there are 42 pairs of segments running in the same direction, and each segment has an average length of 2.8 mm, the total length is 117.6 mm. Then, the total length of the entire boundary line is obtained. For example, if the total length of the boundary line is 150 mm, the ratio of the total length of segments running in the same direction to the total length of the boundary line is calculated. Here, it is 117.6 / 150 = 78.4%. This refers to the proportion of lengths in the same direction. At the same time, it determines the distribution characteristics of segments in the same direction on the boundary contour. Specifically, the boundary line is divided into segments of equal length and numbered, for example, each segment is 10mm, for a total of 15 segments. Each segment is checked to see if there are pairs of segments in the same direction. The proportion of segments covered by segments in the same direction to the total number of segments is counted. The classification is based on whether the number of continuous segments covered exceeds the set continuous distribution threshold. For example, if the continuous distribution threshold is 3 segments, if the continuous coverage of segments in the same direction on the boundary reaches or exceeds 3 segments, it is marked as continuous distribution; otherwise, it is marked as discrete distribution. The proportion of lengths in the same direction is then output.

[0081] S313: Based on the length ratio in the same direction, a weighted summation method is used to accumulate the length ratios of segments in the same direction. Combined with the total boundary length ratio data, the formula is as follows:

[0082] ;

[0083] Obtain the boundary contour overlap ratio index ,in, Representing the The length of each segment in the same direction, Representing the The directional change range of each segment in the same direction, Representing the The position coefficients of each segment in the same direction within the boundary line Representing the The length of each boundary segment Represents the total number of boundary segments. Represents the total number of segments moving in the same direction;

[0084] The boundary contour overlap ratio index is a numerical representation that reflects the degree of spatial overlap and consistency of the segments in the same direction in the overall boundary contour by weighting and summing parameters such as the length, direction variation range, and spatial position of all segments in the same direction and comparing them with the overall length of the boundary. It is used to quantitatively reflect the continuity, directionality, and spatial distribution overlap characteristics of the boundary morphology. The higher the index, the more obvious the directionality consistency and contour continuity of the defect boundary.

[0085] Call the parameters of all filtered segments in the same direction, and extract their corresponding segment lengths. , directional change range and position coefficient Let the original lengths of segments 1 to 3 be 3.74, 2.45, and 2.24 respectively, and the total length of the boundary segment be 12.6. To eliminate the difference in dimensions, we will... When all values ​​are normalized to the boundary line scale, the normalized results are 0.297, 0.194, and 0.178, respectively. Meanwhile, the magnitude of the directional change... The original values ​​were 0.2, 0.15, and 0.1, which remained unchanged after normalization because they were dimensionless angular proportions and position coefficients. Reflecting the relative distribution of paragraphs within the boundary line, their normalized values ​​are 0.64, 0.36, and 0.49, respectively. Substituting these values ​​into the formula, the numerator expands to:

[0086] The first item is ;

[0087] Item 2 is ;

[0088] Item 3 is ;

[0089] Summing the above three terms, we get:

[0090] ;

[0091] because (Since the total boundary length is 1 after normalization), the calculation result is:

[0092] ;

[0093] The result shows that the calculated boundary contour coincidence ratio index If the overlap rate falls within the preset overlap rate judgment interval [0.5, 0.75], this interval is defined as the "moderate overlap segment" judgment interval. This indicates that there are a considerable number of edge segments with consistent direction in the mold boundary line, but there are still some directional turning points or spatial offset points distributed in the contour direction. The boundary morphology is a mixed state of local continuity and local discontinuity. The specific interval division criteria are as follows:

[0094] Low overlap intervals indicate significant inconsistencies in boundary directions and strong morphological fragmentation.

[0095] The second-lowest overlapping segment contains some segments with the same direction, but they are spatially discontinuous.

[0096] The intervals with moderate overlap show obvious but unstable boundary continuity.

[0097] : The highly overlapping segments have boundaries that are highly consistent with their spatial location.

[0098] Please see Figure 5 The specific steps for obtaining the structural jump ratio parameter are as follows:

[0099] S411: Based on the boundary contour overlap ratio index, analyze the transition points and spatial range within each region, calculate the ratio between the number of transition points and the corresponding spatial area in each region, determine the density of transition point distribution, and obtain the transition point area ratio index.

[0100] According to the area number in the defect detection task, the coordinate data of the transition points in each area are read one by one, and the spatial boundary of each area is extracted. The spatial boundary is determined by the smallest bounding rectangle that encloses all the transition points in the area. Its length and width are recorded in millimeters, and the area is obtained by multiplying the length and width. For example, if the bounding rectangle of a certain area is 12.4 mm long and 8.6 mm wide, then the area is 106.64 mm². Then, the total number of transition points in the area is counted, for example, 28 points. The total number of transition points is then multiplied by the spatial area of ​​the area. The ratio is calculated to be 0.262 individuals / mm². Then, it is judged according to the preset distribution density range, which can be divided into sparse (ratio < 0.15 individuals / mm²), medium (ratio between 0.15 and 0.25 individuals / mm²), and dense (ratio > 0.25 individuals / mm²). In the example above, the ratio of 0.262 individuals / mm² is judged as a dense distribution. This process is repeated for all areas to perform the same ratio calculation and density judgment. The ratio value of each area is recorded as the area proportion index of the jump point.

[0101] S412: Compare the area ratio of transition points in each region, analyze the slope changes of transition points on the contour boundary, optimize the calculation order of slope differences between transition points, and obtain the set of slope differences of transition points.

[0102] The indicator values ​​of each region are compared pairwise according to region number, and the regions with larger differences are recorded. The difference is calculated as the absolute difference between the area ratio indicators of the two regions. For example, region A has 0.262 points / mm², region B has 0.180 points / mm², and the difference is 0.082 points / mm². Then, for the transition points in each region, the height values ​​of the adjacent two points are extracted in order of their arrangement on the boundary line. The height difference between the adjacent points is calculated as the local slope change. This process is performed sequentially according to the sequence of transition points in the region to obtain a slope change list. To avoid the calculation order from interfering with the overall difference set, the adjacent point pairs in the original sequence are reordered according to the spatial distance from smallest to largest, and the slope difference calculation is performed again to obtain an optimized slope difference list. Finally, the absolute values ​​of the slope difference between adjacent points in this list are recorded in order to form a transition point slope difference set. This set is stored in units of each region and contains both the slope difference sequence between each transition point pair and the information on the order of the differences.

[0103] S413: Based on the set of slope differences at transition points, filter the numbering information of each region, calculate the joint characteristics of the area ratio of transition points and slope differences, using the formula:

[0104] ;

[0105] The structural jump ratio parameter is obtained, where, Indicates the region The structural jump ratio parameter, Indicates the region Inner The area percentage of each transition point Indicates the region Inner The slope change at each transition point Indicates the region The average slope change at all transition points within the area. Indicates the region The total number of internal jump points;

[0106] The structural jump amplitude parameter is a parameter used to comprehensively reflect the integrated characteristics of jump points in spatial distribution and slope changes within a specific detection area. It describes the intensity and complexity of the changes in local height fluctuations and distribution density of jump points in the area. It can quantify the change amplitude, distribution clustering, and abnormality of defect points in the surface morphology within the detection area. The larger the parameter, the denser the distribution of jump points, the greater the slope fluctuation, and the more drastic the change trend in the area. It is used to characterize the complexity, severity, or priority of the defect structure.

[0107] Set up a region Number of jump points in The normalized area percentage of each jump point is: The normalized slope change at each jump point is as follows: The average slope change is The structural jump ratio parameter of this region is calculated as follows:

[0108] ;

[0109] ;

[0110] ;

[0111] The formula combines two key detection parameters with different dimensions into the same dimension by weighted fusion of area ratio and slope deviation, making the expression of structural complexity more complete and comparable.

[0112] Structural jump amplitude parameter The effective recognition interval range is:

[0113] Low intensity range: This indicates that the slope of the transition point fluctuates relatively little and is spatially dispersed, usually corresponding to slight processing marks or areas with slow edge changes;

[0114] Moderate intensity range: This indicates that the jump points are clustered in a certain area, and the slope fluctuation has become directional or continuous, which usually corresponds to suspected defect areas that need to be manually re-inspected.

[0115] High-intensity zone: This indicates a dense distribution of abrupt change points and drastic slope changes, commonly found in areas of structural abrupt change such as mold erosion, edge damage, or obvious gaps.

[0116] According to the calculation results This value is within the preset range. Within this range, which falls into the medium intensity zone, this result indicates that the region... The corresponding structural transition characteristics show a clear joint fluctuation trend in both the slope change and the density of transition points, indicating that there is a certain abnormal concentration in this area. It should be listed as a priority for review in the subsequent defect response ranking to support the quantitative assessment of the severity of the defect area.

[0117] Please see Figure 6 The specific steps for obtaining the response sequence of the defective part are as follows:

[0118] S511: Based on the structural jump amplitude parameter, compare the density distribution, slope change and amplitude characteristics of each defect region, calculate the arrangement order of the parameters in the task queue, determine the relative position of each region in the sequence, and obtain the jump parameter sequence arrangement index.

[0119] Three key characteristic data points are read for each defect area: density distribution value, slope variation value, and amplitude characteristic value. The density distribution value is calculated from the ratio of the number of abrupt change points to the area. For example, if area A has 30 abrupt change points and an area of ​​100 mm², the density distribution value is 0.30 points / mm². The slope variation value is obtained by dividing the height difference between adjacent abrupt change points by the horizontal distance to get the average slope variation value. For example, if the height difference is 0.18 mm and the horizontal distance is 2.0 mm, the slope variation value is 0.09 (unit: mm / mm). The amplitude characteristic value is the sum of the slope variation value and the density distribution value. The product of the distribution values ​​is normalized, for example, 0.027 is calculated and normalized to the 0-1 range. Then, the three features of each region are arranged in a fixed order to form a feature vector. Pairwise comparisons are performed between regions in turn. When comparing, the size of the density distribution value is judged first. If the difference is within 0.05, the slope change value is compared. If the difference of the slope change value is within 0.02, the amplitude feature value is compared. After the comparison is completed, the relative position of each region in the sorting is recorded. The positions are arranged from best to worst and numbered to obtain the jump parameter sequence arrangement index of each region in the task queue.

[0120] S512: Based on the jump parameter sequence, arrange the index, filter the priority region, analyze the sequence distribution of the corresponding number of the region, optimize the classification method of high response and low response levels, adjust the level distribution, and obtain the response level classification sequence structure.

[0121] The top 30% of regions are selected as priority regions. For example, in 10 regions, regions with indices 1 to 3 are included in the priority region list, and the remaining regions are included in the non-priority list. Then, the distribution of priority region numbers in the sequence is analyzed. For example, if priority region numbers are 2, 5, and 7, it is determined whether they are concentrated or dispersed in the overall sequence. If they are concentrated, the current grouping is retained. If they are dispersed, the priority and non-priority division boundary is readjusted. For example, the top 40% are changed to enter the priority level. Then, the priority level is marked as the high response level, and the non-priority level is marked as the low response level. The spatial distribution of numbers within each level is further examined. If multiple regions in the high response level are located on the same side of the space and are less than 5 mm apart, their numbers are merged and arranged consecutively. Otherwise, the original sequence order is maintained. After adjustment, a new high response and low response level structure is obtained and output as a response level classification sequence structure.

[0122] S513: Based on the response hierarchy classification sequence structure, determine the corresponding number group, analyze the combination distribution of the number under the boundary contour overlap ratio index, compare the correspondence between the number and the contour distribution, optimize the number sorting, and obtain the response sequence of the defect location.

[0123] The system reads the set of all numbers in the high response level and the low response level, and groups them according to the number order. For example, the high response level is {2, 5, 7} and the low response level is {1, 3, 4, 6, 8, 9, 10}. Then, it calls the boundary contour overlap ratio index value corresponding to each number. The ratio value is expressed as a percentage. For example, the overlap ratio of number 2 is 82%, number 5 is 76%, and number 7 is 80%. Within each level, the system analyzes the combination distribution of the number and its overlap ratio. Specifically, the ratio values ​​are sorted by size and the ratio difference between adjacent numbers is compared. If the difference is greater than 5%, the number position is adjusted, and the number with the larger ratio value is moved forward. Finally, a global sort is performed again in the total sequence of the two levels. The global sorting rule is to first sort by overlap ratio from high to low. When the ratios are the same, they are sorted according to the priority order of the original response level. After this sorting optimization, the response sequence of the defect location is output. The order of the numbers in this sequence can directly reflect the comprehensive priority of each defect region in terms of jump features and boundary contour features.

[0124] An artificial intelligence-based mold defect recognition system, the system comprising:

[0125] The abnormal structure extraction module analyzes the height changes between adjacent measurement points based on the three-dimensional height data of the mold surface, determines the trend range of the continuous height difference sequence, compares the spatial distribution pattern of jump points, filters out height anomalies that match the resolution accuracy of the matching equipment, optimizes the spatial matching relationship between the coordinates of each measurement point and the anomaly points, and obtains the abnormal structure distribution layer.

[0126] The edge trajectory construction module is based on the abnormal structure distribution layer. It analyzes the shape of the contour edge of the defect area, calculates the difference between the horizontal and vertical coordinates of each group of adjacent edge points, judges the directional changes between edge points, optimizes the arrangement and combination order of the directional vectors, filters the directional trajectory segments that are consistent with the real edge state, and then adjusts the wrapping order of the edge points to obtain the edge segment directional trajectory sequence.

[0127] The contour overlap recognition module compares the changes in the direction vectors of each segment based on the edge segment direction trajectory sequence, analyzes the trend of the direction after the mold defect boundary segment is normalized, calculates the angle between adjacent direction vectors, determines whether the angle is within the reference bandwidth range of the mold detection boundary angle, identifies edge segments that are aligned in the same direction, and obtains the boundary contour overlap ratio index.

[0128] The structural feature integration module analyzes the distribution relationship between transition points and each defect area based on the boundary contour overlap ratio index, calculates the proportion of the space area occupied by the transition points in each area, compares the contour slope change parameters corresponding to all transition points, judges the relationship between the slope change and the area ratio of the area, optimizes the data integration method of each transition point parameter, and obtains the structural transition amplitude parameter.

[0129] The response sorting and grading module determines the order of each parameter in the defect area of ​​the mold inspection task queue based on the structural jump amplitude parameter. It compares the parameter performance between regions and sorts the parameters. The top-ranked numbers are classified as high response levels, and the remaining numbers are classified as low response levels. The module also analyzes the correspondence between the boundary contour overlap ratio index and the region number to obtain the response sequence of the defect location.

[0130] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A mold defect identification method based on artificial intelligence, characterized in that, Includes the following steps: S1: Based on the three-dimensional height data of the mold surface, analyze the height changes of adjacent measurement points, determine the height difference trend, compare the spatial distribution of jump points, screen out abnormal points under the resolution accuracy of the equipment, optimize the spatial matching of measurement points and abnormal points, and obtain an abnormal structure distribution layer. S2: Based on the abnormal structure distribution layer, analyze the contour edge of the defect area in the layer, calculate the coordinate difference between adjacent edge points, determine the direction change, optimize the direction vector arrangement, filter the true edge direction trajectory, adjust the edge point order, and obtain the edge segment direction trajectory sequence. S3: Based on the edge segment directional trajectory sequence, compare the changes in directional vectors of each segment, analyze the trend of the defect boundary, calculate the angle between adjacent directional vectors, determine whether the angle is within the reference bandwidth range, identify the same-direction aligned segments, and obtain the boundary contour overlap ratio index. S4: Based on the boundary contour overlap ratio index, analyze the distribution of jump points and defect areas, calculate the ratio of jump points to area within the region, compare the slope changes of jump points, determine the relationship between slope and area ratio, and obtain the structural jump amplitude parameter.

2. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The abnormal structure distribution layer includes surface abnormal distribution data, abnormal location identifiers, and feature marker sets. The edge segment direction trajectory sequence includes edge direction information, trajectory segment index, and continuous contour set. The boundary contour overlap ratio index includes overlap ratio value, overlap segment index, and overlap trend parameter. The structure jump ratio parameter includes density distribution parameter and slope change parameter.

3. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the abnormal structure distribution layer are as follows: S111: Based on the three-dimensional height data of the mold surface, collect the three-dimensional coordinates and height data of the measurement points in the detection area, and extract the height information between adjacent measurement points in sequence to construct the height difference sequence between adjacent measurement points and obtain the height difference trend sequence. S112: Based on the height difference trend sequence, compare the height differences of adjacent data within the sequence, analyze the difference fluctuations using a continuous sliding window method, screen for jump points based on the abnormal jump benchmark, statistically analyze the distribution pattern of jump points in the detection area, and then extract the spatial density, arrangement orientation and concentration of jump points in each local segment to obtain the jump point distribution index. S113: Based on the jump point distribution index, sequentially match the spatial coordinates of the jump points and the measurement points, optimize the matching path and eliminate duplicate matching items, integrate the density parameters, position offset parameters and spatial correspondence characteristics of the jump points, retrieve the spatial mapping region of the jump points in the three-dimensional structure, and obtain the abnormal structure distribution layer.

4. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the edge segment directional trajectory sequence are as follows: S211: Based on the abnormal structure distribution layer, select adjacent edge points, calculate the horizontal coordinate difference and vertical coordinate difference of each group of edge points, generate direction vectors, determine the spatial direction change trend of each direction vector, count the node positions where continuous direction changes occur, and obtain the direction change node index sequence. S212: Based on the direction change node index sequence, extract continuous direction vector segments. According to the spatial direction distribution of each vector segment, measure the spatial angle of each segment. Compare the angle of each segment with the boundary continuity index, and select vector segments with spatial angles close to the boundary continuity standard to obtain a sequence of segments with consistent spatial angles. S213: Based on the spatial angle consistent segment sequence, extract all associated edge points, adjust the arrangement order according to the spatial coordinate relationship between each point, calculate the spatial continuity change during the arrangement process point by point, find the optimal point sequence combination of arrangement perturbation, and obtain the edge segment direction trajectory sequence.

5. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the boundary contour overlap ratio index are as follows: S311: Based on the edge segment direction trajectory sequence, compare the spatial angles of each pair of adjacent direction vectors, determine the correspondence between each angle and the reference bandwidth interval of the boundary angle, filter the segment pairs whose spatial angles satisfy the reference bandwidth interval, and calculate the proportion of the filtered segment pairs to all adjacent segment pairs to obtain the ratio of the same direction segment pairs. S312: Based on the ratio of the same-direction segments, calculate the total length of the same-direction segments on the boundary line, compare the total length of the same-direction segments with the length of the boundary line, analyze the proportional relationship between the two, and determine the distribution characteristics of the same-direction segments in the boundary contour to obtain the ratio of the same-direction lengths. S313: Based on the aforementioned length ratio in the same direction, the length ratios of the segments in the same direction are accumulated using a weighted superposition method, and combined with the total boundary length ratio data, the boundary contour overlap ratio index is obtained.

6. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the structural jump ratio parameter are as follows: S411: Based on the boundary contour overlap ratio index, analyze the transition points and spatial range within each region, calculate the ratio between the number of transition points and the corresponding spatial area in each region, determine the density of transition point distribution, and obtain the transition point area ratio index. S412: Compare the area ratio of the jump points in each region, analyze the slope changes of the jump points on the contour boundary, optimize the calculation order of the slope difference between each jump point, and obtain the set of slope differences of the jump points. S413: Based on the set of slope differences at the transition points, filter the numbering information of each region, calculate the joint characteristics of the area ratio of the transition points and the slope differences, and obtain the structural transition amplitude parameter.

7. The mold defect identification method based on artificial intelligence according to claim 1, characterized in that, The steps also include: S5: Based on the structural jump amplitude parameter, determine the order of parameters in the defect area, compare and sort the parameter performance of each area, filter the preceding number to be classified as high response level, and classify the rest as low response level, and analyze the overlap ratio and number mapping to obtain the response sequence of the defect location. The defect location response sequence includes a high response number sequence, a low response number sequence, and a hierarchical sorting list.

8. The mold defect identification method based on artificial intelligence according to claim 7, characterized in that, The specific steps for obtaining the response sequence of the defective site are as follows: S511: Based on the structural jump amplitude parameters, compare the density distribution, slope change and amplitude characteristics of each defect region, calculate the arrangement order of the parameters in the task queue, determine the relative position of each region in the sequence, and obtain the jump parameter sequence arrangement index. S512: Based on the jump parameter sequence arrangement index, filter priority regions, analyze the sequence distribution of the corresponding numbers of the regions, optimize the classification method of high response and low response levels, adjust the level distribution, and obtain the response level classification sequence structure. S513: Based on the response hierarchy classification sequence structure, determine the corresponding number group, analyze the combination distribution of the number under the boundary contour overlap ratio index, compare the correspondence between the number and the contour distribution, optimize the number sorting, and obtain the response sequence of the defect location.

9. A mold defect recognition system based on artificial intelligence, characterized in that, The system is used to implement the artificial intelligence-based mold defect identification method according to any one of claims 1-8, and the system includes: The abnormal structure extraction module analyzes the height changes between adjacent measurement points based on the three-dimensional height data of the mold surface, determines the trend range of the continuous height difference sequence, compares the spatial distribution pattern of jump points, filters out height anomalies that match the resolution accuracy of the matching equipment, optimizes the spatial matching relationship between the coordinates of each measurement point and the anomaly points, and obtains the abnormal structure distribution layer. The edge trajectory construction module analyzes the shape of the defect area contour edge based on the abnormal structure distribution layer, calculates the difference between the horizontal and vertical coordinates of each group of adjacent edge points, judges the directional changes between edge points, optimizes the arrangement and combination order of the directional vectors, filters the directional trajectory segments that are consistent with the real edge state, and then adjusts the edge point wrapping order to obtain the edge segment directional trajectory sequence. The contour overlap recognition module compares the changes in the direction vectors of each segment based on the edge segment direction trajectory sequence, analyzes the trend of the direction after the mold defect boundary segment is normalized, calculates the angle between adjacent direction vectors, determines whether the angle is within the reference bandwidth range of the mold detection boundary angle, identifies edge segments that are aligned in the same direction, and obtains the boundary contour overlap ratio index. Based on the boundary contour overlap ratio index, the structural feature integration module analyzes the distribution relationship between the jump points and each defect area, calculates the proportion of the space area occupied by the jump points in each area, compares the contour slope change parameters corresponding to all jump points, judges the relationship between the slope change and the area ratio of the area, optimizes the data integration method of each jump point parameter, and obtains the structural jump amplitude parameter. The response sorting and grading module determines the order of each parameter in the defect area of ​​the mold inspection task queue based on the structural jump ratio parameter, compares the parameter performance between regions and sorts the parameters, filters the top-ranked numbers as high response levels, and classifies the remaining numbers as low response levels. It also analyzes the corresponding mapping between the boundary contour overlap ratio index and the region number to obtain the response sequence of the defect location.

Images