High-precision industrial drawing element recognition and intelligent error correction system

Abstract

This invention discloses a high-precision industrial drawing element recognition and intelligent error correction system. The system includes: a drawing input preprocessing unit for forming standardized drawing data and outputting page area boundaries; a drawing element joint recognition unit for generating a set of basic element relationships; a multi-view correspondence construction unit for establishing cross-view mapping of the same structural unit; a suspected conflict screening unit and a local configuration inversion unit for generating intermediate structural expressions of candidate areas; and a back-projection consistency verification unit for outputting view consistency deviations and deviation details. This invention relates to the field of industrial drawing processing technology. This high-precision industrial drawing element recognition and intelligent error correction system achieves cross-view structural mapping through projection direction, centerline alignment, dimension inheritance, and section mapping relationships; it performs local configuration inversion on candidate areas to generate intermediate structural expressions, thereby improving the ability to detect and correct local errors.

Description

Technical Field

[0001] This invention relates to the field of industrial drawing processing, and more specifically, to a high-precision industrial drawing element recognition and intelligent error correction system. Background Technology

[0002] In traditional industrial manufacturing, design drawings are an important basis for product manufacturing and assembly. As product structure becomes more complex, industrial drawings typically contain multiple views, multiple scales, and multiple types of elements, such as outlines, dimension lines, symbols, and text annotations, and may originate from various formats such as scanned paper drawings, portable document formats like PDF, or vector graphics.

[0003] In existing technologies, the identification and correction of industrial drawing elements mainly suffer from the following problems: The diverse sources of drawings, with significant differences in orientation, scale, resolution, and line width, make it difficult for traditional recognition algorithms to adapt to diverse inputs. Furthermore, the difficulty in associating information across multiple views is a concern; inconsistencies in dimensions, projections, and annotations may exist in the front view, top view, side view, and enlarged details of the same structure, making it impossible for existing methods to efficiently establish cross-view mappings and associations. Conflict detection is inaccurate; traditional methods often only consider single-type constraints and cannot comprehensively analyze conflicts by considering semantic, geometric, and technological constraints, leading to inaccurate error location. Finally, the reliance on manual error correction is high; for complex structures and multi-view drawings, manual inspection and correction are time-consuming and error-prone, impacting design and production efficiency.

[0004] Therefore, there is an urgent need for a system capable of high-precision feature recognition and intelligent error correction for multi-view, multi-constraint industrial drawings, in order to improve the accuracy of drawing analysis, reduce manual intervention, and enhance the efficiency of industrial design and manufacturing. Summary of the Invention

[0005] The purpose of this invention is to provide a high-precision industrial drawing element recognition and intelligent error correction system. This system solves the problems of existing industrial drawing element recognition and error correction systems, which suffer from diverse drawing sources, difficulty in associating multi-view information, inaccurate conflict detection, and the fact that traditional methods often only consider single-type constraints and cannot comprehensively analyze conflicts by considering semantic, geometric, and process constraints, resulting in inaccurate error location. Furthermore, the system relies heavily on manual error correction, which is time-consuming and error-prone for complex structures and multi-view drawings, affecting design and production efficiency and failing to meet usage requirements.

[0006] This invention achieves the above objectives through the following technical solution: a high-precision industrial drawing element recognition and intelligent error correction system, the system comprising: The system includes a drawing input preprocessing unit, a drawing element joint identification unit, a multi-view correspondence construction unit, a suspected conflict screening unit, a local configuration inversion unit, a back projection consistency verification unit, a cross-domain constraint map construction unit, a joint conflict analysis unit, an error source localization unit, and a candidate error correction generation unit. The drawing input preprocessing unit is used to generate standardized drawing data and output the page area boundaries; The drawing element joint identification unit is used to generate a set of basic element relationships; Multi-view corresponding building units are used to establish cross-view mappings for the same structural unit; Suspected conflict screening units and local configuration inversion units are used to generate intermediate structural representations of candidate regions; The back projection consistency verification unit is used for output view Figure 1 Consistency deviations and deviation details; Cross-domain constraint graph construction unit is used to construct a unified constraint graph; The joint conflict analysis unit is used to combine the view Figure 1 Consistency deviations and conflicting results in the unified constraint map identify error regions; The error source localization unit is used to output the error location, error type, error confidence score, and impact range; The candidate error correction generation unit is used to output the error correction results after consistency verification and constraint satisfaction verification.

[0007] Furthermore, the drawing input preprocessing unit includes: Drawing access subunit, page layout positioning subunit, and standardized transformation subunit; The drawing access subunit is used to access scanned drawings, portable document format drawings, and vector drawings; The layout positioning subunit is used to locate the view area, title bar area, detail bar area and technical requirements area, and output the boundary coordinates and area number of each area; The standardized transformation subunit is used to perform tilt correction, scaling normalization, noise reduction and enhancement, edge enhancement and line width merging to reduce the differences in orientation, scale, resolution and line element thickness of drawings from different sources, and write the processed drawing data into a unified data cache.

[0008] Furthermore, the drawing element joint identification unit includes the following steps: The system uniformly identifies contour lines, dimension lines, center lines, section lines, leader lines, geometric tolerance symbols, datum symbols, welding symbols, surface roughness symbols, dimension figures, title block text, and technical requirement text. Based on the endpoints of the graphic elements, text directions, symbol attachment positions, and local adjacency relationships, it generates a basic set of element relationships containing element categories, coordinate positions, anchor point relationships, adjacency relationships, and direction relationships. This set serves as the association basis for subsequent view association, conflict propagation, and error correction verification, and outputs the corresponding attachment target number and association direction.

[0009] Furthermore, the multi-view corresponding construction unit includes the following steps: Based on the projection direction relationship, centerline alignment relationship, datum boundary correspondence relationship, dimension inheritance relationship and section mark mapping relationship, the same structural units in different views are associated to form cross-view structural mapping results between the main view and the top view, the main view and the side view, the main view and the section view, as well as the local enlarged view and the original view area. It also outputs the view number, structure identifier, mapping confidence information and constraint category label corresponding to each mapping result, which are used to support subsequent candidate region selection and local configuration inversion.

[0010] Furthermore, the suspected conflict screening unit is used to identify the corresponding region as a candidate region for inversion when the following conditions are met: The projected outline of the same structure is inconsistent in different views; The dimension values ​​of the same local feature are inconsistent in different views; The cross-section representation in the sectional view does not match the outline of the external view; The geometry of the enlarged view does not correspond to the selected area in the original view; The semantic meaning of the text annotation is inconsistent with the positional relationship of the graphic elements; The identification results show a break in the relationship with adjacent structures; It also outputs the candidate region coordinates, candidate region number, initial conflict type, conflict source view, and filtering priority for each candidate region to be inverted.

[0011] Furthermore, the local configuration inversion unit includes the following steps: Local structure restoration is performed on the candidate region to be inverted, generating intermediate structure expression units including local geometric skeleton, boundary topological relationship, key dimension parameters, symmetry parameters, hole axis relationship parameters, slot boundary relationship parameters, local contour closure state, local dimension chain and cross-sectional contour relationship after sectioning. The intermediate structure expression units are used as unified input objects for cross-view verification, constraint solving and candidate error correction generation.

[0012] Furthermore, the back-projection consistency verification unit includes the following steps: The intermediate structural representation units are projected back to the main view, top view, side view, sectional view, and enlarged view respectively to obtain the contour projection deviation, dimension chain deviation, symbol anchor point deviation, projection missing deviation, and projection redundancy deviation. And based on the weighted results of various deviations, candidate regions are generated. Figure 1 Consistency deviation score and view deviation details are recorded, and the source view of the deviation is also recorded. The weight of each deviation is determined based on the contribution of the corresponding deviation in the historical annotated drawings to the error judgment result, and the weight of each deviation satisfies the normalization constraint.

[0013] Furthermore, the cross-domain constraint graph construction unit includes the following steps: Semantic constraints, geometric constraints, and process constraints are mapped into a unified constraint network, and drawing elements and their relationships are mapped into constraint nodes and constraint edges to support the propagation analysis of constraint conflicts and the search for satisfyable solutions. Among them, semantic constraints include the orientation relationship between dimension values ​​and dimension lines, the correspondence between geometric tolerances and constrained features, the connection relationship between datum symbols and target surfaces or target axes, and the association relationship between sectional markings and sectional results. Geometric constraints include collinearity, perpendicularity, parallelism, closure, symmetry, concentricity, and closed-loop dimensional chain. Process constraints include fit feasibility, accessibility of hole and groove machining, process compatibility of chamfers and fillets, assembly rationality of datum selection, and the matching relationship between tolerance allocation and machining capability.

[0014] Furthermore, the joint conflict analysis unit includes the following steps: Combined with vision Figure 1 The consistency deviation score, semantic conflict score, geometric conflict score and process conflict score are used to generate a comprehensive anomaly score. The comprehensive anomaly score is compared with a preset anomaly judgment threshold. When the comprehensive anomaly score reaches the anomaly judgment threshold, the error region is determined and the conflict type, conflict intensity and conflict propagation path are output. The error source localization unit is used to determine the error source from primitive nodes, dimension nodes, symbol nodes and section expression nodes based on the conflict propagation path in the unified constraint network, and outputs the view to which the error source belongs, its location coordinates, error type, error confidence score and scope of influence.

[0015] Furthermore, the candidate error correction generation unit includes the following steps: Based on the multi-view back projection results, the set of basic element relationships, the satisfyable solutions of the unified constraint network, and the stable expression patterns of similar structures in historical drawings, the corrected dimension values, the completed outline, the corrected cutting direction, the repositioned tolerance symbol, the corrected datum attachment position, and the corrected local enlarged area identifier are generated. The candidate error correction results are then ranked according to the consistency review results, constraint satisfaction review results, and comprehensive anomaly scores. Output the candidate error correction results that are ranked first and have passed the review as the final error correction results, while retaining the candidate results that have not passed the review and the reasons for their failure.

[0016] The beneficial effects of this invention are as follows: 1. By performing tilt correction, scaling normalization, noise reduction and enhancement, edge enhancement and line width merging, we achieve unified and standardized processing of drawings from different sources, thereby improving the accuracy of subsequent recognition.

[0017] 2. Jointly identify multiple types of elements such as outlines, dimension lines, center lines, symbols, and text, and establish a basic set of element relationships based on element categories, coordinate positions, anchor points, and adjacency relationships, providing a reliable foundation for cross-view association, conflict analysis, and error correction.

[0018] 3. Cross-view structure mapping is achieved through projection direction, centerline alignment, size inheritance, and section mapping relationships; local configuration inversion is performed on candidate regions to generate intermediate structure representations, thereby improving the ability to detect and correct local errors.

[0019] 4. Combine visual Figure 1 Consistency deviations, semantic, geometric, and process constraints are used to generate a comprehensive anomaly score, enabling multi-dimensional conflict analysis; and the source, type, and scope of error are accurately located based on the constraint propagation path.

[0020] 5. Based on the multi-view back projection results, basic element relationships and constraint solutions, the system automatically generates corrected dimensions, outlines, cutting directions, symbol positions and local magnified area identifiers, achieving automated and intelligent error correction, and can retain candidate results that fail the review and the reasons. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is the overall system flowchart of the present invention; Figure 2 This is a flowchart of the drawing input preprocessing unit of the present invention; Figure 3 This is a flowchart of the joint conflict analysis and error source localization process of the present invention. Detailed Implementation

[0022] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0023] Example 1: Please see Figure 1-3 This invention provides a technical solution: a high-precision industrial drawing element recognition and intelligent error correction system, comprising: The drawing input and preprocessing unit is used to acquire industrial drawings to be processed, and to perform tilt correction, scale normalization, noise reduction and enhancement, line width normalization, page partitioning, and regional positioning of the view area, title block area, detail block area and technical requirements area on the industrial drawings to form standardized drawing data. Among them, unprocessed industrial drawings, which are graphic files containing information about industrial products, may contain various factors that are detrimental to subsequent identification and analysis, such as tilt, inconsistent scale, and noise. Tilt correction adjusts industrial drawings with tilt angles to restore them to standard orientations such as horizontal or vertical for accurate subsequent processing. Scale normalization unifies industrial drawings drawn at different scales to the same standard, eliminating errors in size and shape analysis caused by scale differences. Noise reduction and enhancement remove noise interference in the drawings, such as speckles and unclear lines, while enhancing the clarity and contrast of useful information to improve drawing quality. Line width normalization unifies lines of different thicknesses in the drawings to a standard line width, eliminating the impact of line width differences on subsequent identification and analysis. Layout partitioning divides the industrial drawings into different areas for targeted processing and analysis of each area. View area, title block area, detail block area, and technical requirements area are common specific functional areas in industrial drawings. The view area displays various views of the product; the title bar area contains information such as the name, number, and scale of the drawing; the parts list area lists the product's components; the technical requirements area describes the product's manufacturing, inspection, and other technical requirements; and the standardized drawing data, after being processed such as tilt correction, scale normalization, noise reduction and enhancement, line width normalization, page partitioning, and area positioning, forms drawing data that conforms to a unified standard and is easy to identify and analyze later. The drawing element joint identification unit is used to jointly identify contour lines, dimension lines, center lines, section lines, leader lines, text annotations, tolerance symbols, datum symbols, welding symbols, surface roughness symbols, as well as local elements corresponding to holes, slots, fillets and chamfers in standardized drawing data, and outputs element category, coordinate position, anchor point information, adjacency relationship and initial identification result; Among them, the outline lines describe the shape and contour of an object, used to define its boundaries; dimension lines are used to indicate the dimensions of an object, usually used in conjunction with dimension figures to clarify the size of each part; center lines represent the center of symmetry or axis of an object, generally thin dashed lines, used to assist in positioning and express symmetry; section lines indicate the sectioning position, usually composed of thick solid lines and arrows, indicating the direction of the sectioning plane; leader lines extend from the drawing elements, used to connect text labels or symbols, serving an indicative and explanatory function; text labels provide textual descriptions of drawing elements, dimensions, technical requirements, etc.; tolerance symbols indicate the allowable range of variation in the size, shape, and position of a part, used to control the machining accuracy of the part; and datum symbols determine the geometric tolerance zone of the part. The symbols used for identification are as follows: reference symbols, typically composed of a base letter and a short horizontal line; welding symbols, which indicate welding information such as welding method, weld type, and weld size; surface roughness symbols, used to indicate the smoothness of the surface finish of a part, reflecting the microscopic geometric errors of the part's surface; local elements, graphic elements in the drawing representing local features such as holes, slots, fillets, and chamfers; element categories, classifying the identified drawing elements, such as outlines and dimension lines; coordinate positions, the specific position information of the drawing elements in the drawing coordinate system; anchor point information, key position point information related to the symbols, used to accurately locate the symbols; adjacency relationships, the spatial relationships such as adjacency and connection between different drawing elements; and initial identification results, the results obtained after preliminary identification of the drawing elements, which may contain certain errors and uncertainties. The multi-view correspondence building unit is used to identify the same structural unit from the front view, top view, side view, sectional view and partial enlarged view, and to establish cross-view correspondence; Among them, the front view, top view, and side view are views obtained by projecting an object from different directions. The front view usually reflects the main shape features of the object, while the top view and side view show the shape of the object from the top and side, respectively. The sectional view is a view obtained by cutting the object, removing part of the material, and then projecting the remaining part onto the projection plane. It is used to show the internal structure of the object. The enlarged view is a view drawn by enlarging a part of the object's structure according to a certain scale. It is used to express local features in detail. The same structural unit is a structural element that represents the same part of the object in different views. Cross-view correspondence establishes the relationship between the same structural unit in different views so as to carry out consistency analysis and processing of multiple views. The suspected conflict screening unit is used to screen candidate areas to be inverted based on inconsistencies in projected contours, conflicts in dimension annotations, mismatches in sectional expressions, inconsistencies between the selected areas of the local magnified view and the original view, abnormal relationships between the semantics of text annotations and the position of graphic elements, and abnormal recognition results. Among these issues are: inconsistent projected outlines (mismatched projected outlines of objects in different views, potentially indicating drawing errors or inaccurate representation); conflicting dimensioning (contradictions or inconsistencies exist between dimensioning in different views or within the same view); mismatched sectional views (inconsistencies exist between the representation in the sectional view and the actual structure of the object or in other views); inconsistent selection areas between the enlarged partial view and the original view (the area shown in the enlarged partial view does not correspond to the selected area in the original view); abnormal semantic relationships between text annotations and graphic element positions (the content of text annotations does not match the position and meaning of the corresponding graphic element); abnormal recognition results (the initial recognition results output by the joint recognition unit for drawing elements are unreasonable or erroneous); and candidate areas to be inverted (drawing areas that may contain errors, selected based on the above-mentioned conflicts, require further analysis and verification). The local configuration inversion unit is used to perform local structural inversion on the candidate region to be inverted, forming an intermediate structural expression unit that includes local geometric skeleton, boundary topological relationship, key size parameters, symmetry parameters, hole axis or slot boundary relationship parameters, and cross-sectional contour relationship after sectioning. The process includes: local structure inversion, which involves reverse analysis of candidate regions to infer their original geometric structure and shape features; local geometric skeleton, which represents the basic geometric shape and topological relationships of the object's local structure; boundary topological relationships, which describe the connections and adjacency relationships between local boundary elements of the object; key dimensional parameters, which are important dimensional information that determines the shape and size of the object's local structure; symmetry parameters, which reflect the symmetry characteristics of the object's local structure, such as symmetry axis and symmetry plane; hole-axis or slot boundary relationship parameters, which describe the boundary relationships between holes and axes, slots and adjacent structures, such as fit clearance and positional relationships; cross-sectional profile relationships after sectioning, which are the shape of the cross-sectional profile and its relative positional relationship with other parts after the object is sectioned; and intermediate structural expression units, which are intermediate representations of various key information of the local structure obtained after local configuration inversion, used for subsequent verification and analysis. The back-projection consistency verification unit is used to project the intermediate structure representation unit back to the front view, top view, side view, sectional view, and enlarged partial view, respectively, to obtain multi-view results. Figure 1 Consistency deviation; This involves projecting back to the main view, top view, side view, sectional view, and enlarged partial view, reprojecting the intermediate structural units onto the corresponding views according to the projection rules of different views; multi-view Figure 1 Consistency bias refers to the differences and deviations between the intermediate structural representation units projected back to each view and the original view, which are used to evaluate the accuracy and consistency of local configuration inversion. The cross-domain constraint graph construction unit is used to construct a unified constraint graph composed of semantic constraints, geometric constraints, and technological constraints; Among them, semantic constraints, based on information such as text annotations and symbol meanings in the drawings, constrain the relationships and meanings between drawing elements; geometric constraints, based on the geometric shape, size relationship, positional relationship, etc. of objects, constrain the geometric characteristics between drawing elements; process constraints, considering the actual production requirements such as the manufacturing process and assembly process of parts, constrain the drawing elements; and the unified constraint map integrates semantic constraints, geometric constraints, and process constraints to form a comprehensive and unified constraint system, which is used to guide and verify the correctness of drawings. Joint conflict analysis unit is used for multi-view Figure 1 The consistency deviation and the conflict in the unified constraint map are solved jointly to identify the error region; Among them, multi-view Figure 1 Consistency deviations and conflicts in the unified constraint map will lead to multiple views obtained from the back-projection consistency verification unit. Figure 1 A comparative analysis of consistency deviations and unified constraint graphs established by cross-domain constraint graph construction units is conducted to identify existing conflicts and contradictions; joint solutions are then implemented, taking into account multiple perspectives. Figure 1 To identify erroneous regions by employing appropriate methods and algorithms to resolve consistency deviations and conflicts in the unified constraint map; The error source location unit is used to locate the error to a specific graphic element, specific dimension annotation, specific symbol anchor point, or specific section expression location; Specifically, the error is located in the specific drawing elements, such as lines and graphics in the drawing; the error is related to specific dimensions, such as incorrect dimension numbers or incorrect position of dimension lines; the error is located in the specific symbol anchor points, such as incorrect position of the datum letter in the datum symbol; and the error is related to the sectioning position, sectioning direction, and other aspects of the section view. The candidate error correction generation unit is used to generate candidate error correction results based on the error source and output the final error correction result; Among them, the candidate error correction results generate multiple possible error correction schemes and results based on the error source; the final error correction result selects the optimal and most reasonable error correction scheme from the candidate error correction results and outputs it as the final error correction result.

[0024] It should be noted that during use, the drawing input and preprocessing units unify drawing standards, laying the foundation for subsequent processing and improving recognition accuracy. The drawing element joint recognition unit comprehensively identifies various elements and outputs detailed information, facilitating accurate understanding of the drawing content. The multi-view correspondence construction unit can accurately associate different views, ensuring the consistency of multi-view analysis. The suspected conflict screening unit can quickly locate potentially erroneous areas, improving error correction efficiency. The local configuration inversion and back projection consistency verification units work together to deeply analyze local structures and verify accuracy. The cross-domain constraint map construction unit provides comprehensive constraints, ensuring that error correction meets multiple requirements. The joint conflict analysis, error source location, and candidate error correction generation units work in a progressive manner to accurately locate errors and provide reasonable error correction solutions, effectively improving the quality of industrial drawings and reducing the cost and error of manual error correction.

[0025] In this embodiment, the industrial drawings to be processed can be mechanical part drawings, assembly drawings, sectional views, or enlarged partial views. The drawing sources include scanned drawings, PDF drawings, and vector drawings. For scanned drawings, it is preferred to use... Resolution input system; for PDF drawings and vector graphics, it can convert to equivalent resolution. The bitmap data is then processed uniformly to ensure a relatively stable spatial resolution during line element extraction, character recognition, and symbol detection. This resolution is determined by comparing... , , and Under certain conditions, the fidelity of fine lines, character distinguishability, and overall computational burden are considered. It can achieve a better balance between the accuracy of dimension numbers, geometric tolerance symbols and fine line structure recognition and computational efficiency.

[0026] I. Drawing Input and Preprocessing: First, acquire the industrial drawings to be processed and perform standardized preprocessing. The preprocessing process includes tilt correction, scale normalization, noise reduction and enhancement, line width normalization, page layout partitioning, and regional positioning of the view area, title block area, detail block area, and technical requirements area. This step generates standardized drawing data required for subsequent identification.

[0027] In a specific implementation, the tilt correction angle is preferably controlled at [value]. Within this range; noise reduction and enhancement employ a combination of median filtering and Gaussian filtering, where the Gaussian filter parameters are... Preferred selection The target linewidth after linewidth normalization is preferably... By locating and cropping the view area, title block area, detail area, and technical requirements area in the drawing separately, non-target areas can be avoided from interfering with subsequent view mapping and conflict analysis. The basis for this determination is: after statistically analyzing the page deflection angle, noise level, and line width distribution in historical samples, it was found that tilt correction is controlled within... The parameters σ of the inner and Gaussian filters are taken as follows: Normalized target line width When the pixel count is 1, it can better balance noise reduction effect, line element continuity and subsequent recognition stability.

[0028] II. Joint Identification of Drawing Elements: After preprocessing, the standardized drawing data is subjected to joint recognition of drawing elements. The recognition objects include outlines, dimension lines, center lines, section lines, leader lines, text annotations, tolerance symbols, datum symbols, welding symbols, surface roughness symbols, and local elements corresponding to holes, slots, fillets, and chamfers. For different elements, the system outputs the element category, coordinate position, anchor point information, adjacency relationship, and initial recognition results.

[0029] In this embodiment, contour lines, dimension lines, center lines, section lines, and leader lines are identified by a line element detection model; text annotations, tolerance symbols, datum symbols, welding symbols, and surface roughness symbols are identified by a text symbol joint recognition model; and local elements corresponding to holes, slots, fillets, and chamfers are extracted through contour analysis and local template matching. To improve the stability of subsequent multi-view matching, the system also establishes anchor point relationships, adjacency relationships, and pointing relationships between drawing elements, forming a basic element relationship set.

[0030] In one specific implementation, the recognition confidence threshold for a single drawing element can be set to 0.78. When the recognition confidence of a drawing element is lower than 0.78, the drawing element is marked as a low-confidence element. If the adjacency relationship between the low-confidence element and its adjacent structures is unstable, then the area where the low-confidence element is located is taken as the priority analysis area. The basis for this determination is that after traversing tests on different confidence thresholds on the validation samples, 0.78 can achieve a better balance between recognition accuracy and recall, and effectively reduce the interference of low-quality false recognition results on subsequent structural mapping.

[0031] III. Establishing the correspondence between multiple views: After the drawing elements are identified, the front view, top view, side view, sectional view, and enlarged partial view are identified from the drawing, and a cross-view correspondence matrix is ​​established. The correspondence matrix is ​​constructed based on projection direction relationship, centerline alignment relationship, datum boundary correspondence relationship, dimension inheritance relationship, similarity of contour features of the same structure in different views, and mapping relationship between section marks and section results, thereby associating the same structural unit scattered in different views.

[0032] In a specific implementation, a comprehensive matching score can be calculated for structural units between the front view and top view, the front view and side view, and the front view and sectional view. When the comprehensive matching score is not lower than 0.72, it is determined to be a corresponding representation of the same structural unit in different views. When the comprehensive matching score is lower than 0.72, the structural unit is marked as a suspected inconsistency region to be further verified. The determination is based on the fact that after statistically analyzing the distribution of comprehensive matching scores between true corresponding structures and non-corresponding structures, a score around 0.72 can effectively distinguish between valid cross-view correspondences and pseudo-correspondences, thus balancing matching recall and matching accuracy.

[0033] IV. Screening of Suspected Conflict Areas: After establishing the cross-view correspondence, the correspondence results are initially screened for consistency. If any of the following situations occur, the corresponding area is defined as a candidate area to be inverted: the projection contour of the same structure is inconsistent in different views; the dimension annotation of the same local feature is numerically conflicting in different views; the cross-sectional expression in the sectional view does not match the contour of the external view; the geometric structure of the enlarged local view is inconsistent with the selected area of ​​the original view; the semantic relationship between the text annotation and the position of the graphic element is abnormal; the initial identification confidence is lower than the set threshold and the relationship between neighboring features is unstable.

[0034] In this embodiment, it is preferred to use A local window of a pixel is used as the candidate region analysis unit. When a contour projection deviation greater than 0.15 or a dimension value deviation exceeding the annotation value occurs within a local window... A local window can be designated as a candidate region for inversion if either the symbol anchor point distance deviation is greater than 8 pixels or the overlap between the boundary of the local magnified image and the selected area of ​​the original view is less than 0.80. The determination is based on verification using a combination of local feature size distribution and abnormal region coverage data from the sample. The pixel window can cover most holes, slots, rounded corners, sections, and locally magnified areas; the contour projection deviation is 0.15, and the dimensional deviation is... Anchor point deviation of 8 pixels and boundary overlap of 0.80 correspond to the effective boundaries of normal and abnormal regions in statistical distribution, respectively.

[0035] V. Local configuration inversion: For candidate regions to be inverted, the system performs local configuration inversion to obtain intermediate structural representation units. These intermediate structural representation units include at least the local geometric skeleton, boundary topological relationships, key dimensional parameters, symmetry parameters, hole-axis or slot boundary relationship parameters, and the cross-sectional profile relationships after sectioning. Local configuration inversion does not directly recover the complete 3D model, but rather recovers the model that satisfies multiple views. Figure 1 Consistent local intermediate structure models are used for subsequent comparison and error correction.

[0036] For example, for holes, slots, fillets, and chamfers in a mechanical part drawing, the system prioritizes restoring intermediate structural information such as hole axes, slot boundary parallelism, fillet radii, chamfer angles, and section profiles. In a specific implementation, the deviation of the hole axis direction is preferably no greater than [value missing]. The parallelism deviation of the groove boundary is preferably no greater than 0.05, and the fillet radius recovery error is preferably no greater than the design radius. The basis for this determination is as follows: After statistically analyzing the fitting errors of typical holes, slots, and fillets in mechanical part drawings, it was found that when the deviation of the hole axis direction is no greater than 2°, the deviation of the slot boundary parallelism is no greater than 0.05, and the fillet radius recovery error is no greater than the design radius... When the inversion results are stable, they can represent the original local configuration relatively well.

[0037] VI. Back-projection consistency verification: After local configuration inversion, the intermediate structural representation units are projected back to the front view, top view, side view, sectional view, and enlarged local view, respectively, to obtain the predicted projection results. Then, the predicted projection results are compared item by item with the corresponding areas of the original drawings, calculating contour deviation values, key point deviation values, dimension chain deviation values, section boundary deviation values, symbol anchor point deviation values, projection missing rate, and projection redundancy rate, thereby forming a multi-view... Figure 1 Consistency deviation vector.

[0038] To facilitate a unified evaluation of different types of deviations, this embodiment adopts a visual approach. Figure 1 Consistency deviation score For the first Each candidate region is quantitatively characterized, and the calculation formula is as follows:

[0039] in, Indicates the deviation in contour projection. Indicates dimensional chain deviation, Indicates the deviation of the symbol anchor point. This represents the combined deviation consisting of projection missing bias and projection redundancy bias. , ,and These are the weighting coefficients.

[0040] In one specific embodiment, it is preferred to first... , , and Normalization was performed to ensure that all values ​​fell within the range of 0 to 1; then the weighting coefficients were set as follows: That is, the weight of the contour projection deviation is 0.35, the weight of the dimension chain deviation is 0.30, the weight of the symbol anchor point deviation is 0.15, and the weight of the combined deviation of projection missing and redundancy is 0.20. When If a candidate region exhibits significant inconsistencies in its multi-view representation, it can be determined and output to the joint conflict analysis module. The determination is based on: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] , , and After normalizing the contributions in the labeled samples, contour structure and dimensional chain have a more significant impact on incorrect judgments, and therefore are given higher weights; when When the threshold is set to 0.42, the detection rate of anomalies in multi-views and the false alarm rate can be well balanced.

[0041] VII. Construction of Cross-Domain Constraint Graph: In gaining multi-view Figure 1 After resolving consistency deviations, the identification results and deviation results are mapped together onto the unified constraint map. The unified constraint map includes at least semantic constraints, geometric constraints, and process constraints.

[0042] Semantic constraints are used to represent the directional relationship between dimension values ​​and dimension lines, the correspondence between geometric tolerances and constrained features, the connection between datum symbols and target surfaces or axes, and the association between section marks and section results. Geometric constraints are used to represent collinearity, perpendicularity, parallelism, closure, symmetry, concentricity, and closed-loop dimensional chains between drawing elements. Technological constraints are used to represent the feasibility of fit relationships, the accessibility of hole and slot machining, the matching of chamfer and fillet processes, the rationality of datum selection for assembly, and the tolerance allocation and machining capability boundaries. The system abstracts various drawing elements and their relationships into constraint nodes and constraint edges to form a unified constraint network that supports conflict propagation analysis.

[0043] In the context of machining drawings, the following rules are preferred to be established at least: Dimension figures must have a unique directional relationship with their corresponding dimension lines; Geometric tolerance symbols must have a clear connection with the constrained surface or shaft; the projections of the axes of the same hole system in different views must be collinear or concentric; and the combination of mating dimensions and tolerances must meet the requirements of machining feasibility. The basis for these rules is that the aforementioned semantic, geometric, and process constraints originate from mechanical drawing specifications, geometric tolerance annotation rules, and machining assembly feasibility requirements, and have been validated through sample playback to reliably support conflict propagation analysis.

[0044] VIII. Joint Conflict Analysis: After the unified constraint graph is established, multi-view Figure 1 By jointly solving for consistency bias and cross-domain constraint conflicts, a comprehensive anomaly score for the candidate region is obtained.

[0045] In this embodiment, the first The combined anomaly score of each candidate region Calculate according to the following formula:

[0046] in, Indicates the first The view of each candidate region Figure 1 Consistency deviation score, Indicates the semantic conflict score. Represents geometric conflict fractions. Indicates the score of process conflict. , , and These are the weighting coefficients.

[0047] In one specific embodiment, the weighting coefficient is preferably set as follows: , i.e. Figure 1 The weighting for consistency deviation score is 0.40, semantic conflict score is 0.20, geometric conflict score is 0.25, and process conflict score is 0.15; anomaly detection threshold. The preferred setting is 0.55. When When this happens, the corresponding candidate region is identified as an erroneous region, and the conflict type, conflict intensity, and conflict propagation path are output. The determination is based on: the view... Figure 1 After statistically analyzing the ability of consistency bias, semantic conflict, geometric conflict, and process conflict to distinguish the final conclusion of error, we can obtain the following results. The preferred combination; when the comprehensive abnormal threshold A value of 0.55 can achieve a better balance between error detection rate and false alarm control.

[0048] In a specific application, the aforementioned weights and thresholds can be obtained through statistical analysis of historical annotated drawing samples. For example, at least 200 annotated mechanical part drawings can be selected as a sample set, with the training sample to validation sample ratio preferably at 8:2. The final weights and thresholds are determined by comparing the distribution differences between correct and incorrect regions in various deviations and conflicts. The basis for this determination is that at least 200 samples can cover the main structural representations in common mechanical part drawings, assembly drawings, and sectional views, and the 8:2 ratio is beneficial for balancing the sufficiency of model training with the stability of validation.

[0049] IX. Error Source Location: For candidate regions identified as error areas, the most likely error source location is determined by combining the conflict propagation path. Error source location objects include at least erroneous text labels, erroneous dimension values, missing outlines, redundant elements, misaligned symbols, erroneous section directions, and mapping errors between enlarged views and the original view area. The system outputs the location coordinates, associated view, error type, error confidence level, and affected area for each error source.

[0050] In this embodiment, conflict propagation analysis can be performed preferentially in the order of primitive nodes—dimensional nodes—symbol nodes—section representation nodes. If the cumulative contribution of a node in the conflict propagation path is higher than 0.60, the drawing object corresponding to that node can be identified as the main error source. The basis for this determination is that backtracking analysis of the actual error sources shows that when the cumulative contribution of a single node in the conflict propagation path is higher than 0.60, its correspondence with the actual error source is relatively stable.

[0051] 10. Candidate Error Correction Generation and Output: Once the error source is identified, the system automatically generates candidate error correction results based on the multi-view back projection results, the relationship between elements in the drawing context, the satisfyable solutions of the unified constraint map, and the stable representation patterns of similar structures in historical drawings. Candidate error correction results may include corrected dimension values, completed outlines, corrected section directions, repositioned tolerance symbols, corrected datum attachment positions, and valid local enlargement area identifiers.

[0052] In this embodiment, it is preferable to generate 3 to 5 candidate error correction results for each error source. If the error source is an abnormal size value, the candidate results shall at least include: according to adjacent views Figure 1 The correction methods include dimensional value correction, correction based on the closed-loop relationship of the dimensional chain, and correction based on historical similar structure templates. If the error source is an abnormal cutting direction, the candidate results should include at least: cutting direction reversal, cutting line anchor point relocation, and cutting range adjustment. The determination is based on the following: too few candidates will reduce the coverage of feasible error correction paths, while too many candidates will significantly increase the sorting burden. It has been verified that 3 to 5 candidate results can well cover common error correction scenarios such as dimensional correction, contour completion, cutting adjustment, and symbol relocation.

[0053] Subsequently, consistency and constraint satisfaction checks were performed on each candidate error correction result, taking into account constraint satisfaction and multi-view performance. Figure 1 The system sorts candidate error correction results based on consistency recovery and context fit, and outputs the best-ranked candidate result as the final error correction result; at the same time, it creates a visual error correction mark at the corresponding position on the drawing.

[0054] In a specific implementation, the ranking weights of the candidate error correction results can be set as follows: constraint satisfaction weight 0.45, multi-view... Figure 1Consistency recovery is weighted at 0.35, and context fit is weighted at 0.20. When a candidate error correction result has a comprehensive ranking score of at least 0.70 and is the highest among all candidate results, it is output as the final error correction result. The determination is based on the fact that, among the sample ranking results, constraint satisfaction has the greatest impact on the final error correction usability. Figure 1 Consistency recovery is secondary, while context fit is mainly used for finer distinction. Therefore, 0.45, 0.35, and 0.20 are used as preferred ranking weights. When the comprehensive ranking score is not lower than 0.70, the candidate results can usually meet the requirements of constraint consistency and engineering expression rationality at the same time.

[0055] XI. Optional manual confirmation and closed-loop update: In an optional embodiment, the system can also receive manual review results and correct local configuration inversion parameters, various constraint weights, error type judgment boundaries, and candidate error correction sorting strategies based on the manual review results, thereby enabling the system to continuously improve recognition and error correction accuracy in similar industrial drawing scenarios.

[0056] 12. Standards or basis for determining specific numerical values: In this embodiment, the aforementioned resolution, threshold, weighting coefficient, candidate region size, deviation judgment boundary, and sorting threshold are not arbitrarily set, but are comprehensively determined by combining the drawing specifications of industrial drawings, sample statistical results, validation set evaluation results, and engineering expression stability requirements. Specifically, it is preferable to first construct a sample set of labeled industrial drawings, which includes at least correct drawing samples and drawing samples containing errors, and covers drawing types such as mechanical part drawings, assembly drawings, sectional views, and enlarged partial views; then, each candidate parameter is tested and cross-validated in groups on the sample set, and the recognition accuracy, false alarm rate, false negative rate, multi-view matching stability, error location accuracy, and candidate error correction effectiveness are used as evaluation indicators; finally, the parameter combination that can balance recognition accuracy, positioning accuracy, and operational stability is selected as the preferred parameter in this embodiment.

[0057] In a preferred embodiment, each of the above parameters can be determined based on no fewer than 200 annotated industrial drawing samples, wherein the ratio of training samples to validation samples is preferably 8:2. For different application scenarios, the above parameters can also be recalibrated according to differences in drawing type, symbol density, sectional complexity, assembly relationship complexity, and manufacturing constraints. In other words, the values ​​given in this embodiment are preferred parameters obtained under preferred sample conditions and are not intended to limit the scope of protection of this invention. Those skilled in the art can make equivalent adjustments to the parameters under the same technical concept.

[0058] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A high-precision industrial drawing element recognition and intelligent error correction system, characterized in that, The system includes: The system includes a drawing input preprocessing unit, a drawing element joint identification unit, a multi-view correspondence construction unit, a suspected conflict screening unit, a local configuration inversion unit, a back projection consistency verification unit, a cross-domain constraint map construction unit, a joint conflict analysis unit, an error source localization unit, and a candidate error correction generation unit. The drawing input preprocessing unit is used to generate standardized drawing data and output the page area boundaries; The drawing element joint identification unit is used to generate a set of basic element relationships; Multi-view corresponding building units are used to establish cross-view mappings for the same structural unit; Suspected conflict screening units and local configuration inversion units are used to generate intermediate structural representations of candidate regions; The back-projection consistency verification unit is used to output view consistency deviation and deviation details; Cross-domain constraint graph construction unit is used to construct a unified constraint graph; The joint conflict analysis unit is used to determine the error region by combining the view consistency deviation and the conflict results in the unified constraint map; The error source localization unit is used to output the error location, error type, error confidence score, and impact range; The candidate error correction generation unit is used to output the error correction results after consistency verification and constraint satisfaction verification.

2. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 1, characterized in that, The drawing input preprocessing unit includes: Drawing access subunit, page layout positioning subunit, and standardized transformation subunit; The drawing access subunit is used to access scanned drawings, portable document format drawings, and vector drawings; The layout positioning subunit is used to locate the view area, title bar area, detail bar area and technical requirements area, and output the boundary coordinates and area number of each area; The standardized transformation subunit is used to perform tilt correction, scaling normalization, noise reduction and enhancement, edge enhancement and line width merging to reduce the differences in orientation, scale, resolution and line element thickness of drawings from different sources, and write the processed drawing data into a unified data cache.

3. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 1, characterized in that, The drawing element joint identification unit includes the following steps: The system uniformly identifies contour lines, dimension lines, center lines, section lines, leader lines, geometric tolerance symbols, datum symbols, welding symbols, surface roughness symbols, dimension figures, title block text, and technical requirement text. Based on the endpoints of the graphic elements, text directions, symbol attachment positions, and local adjacency relationships, it generates a basic set of element relationships containing element categories, coordinate positions, anchor point relationships, adjacency relationships, and direction relationships. This set serves as the association basis for subsequent view association, conflict propagation, and error correction verification, and outputs the corresponding attachment target number and association direction.

4. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 2, characterized in that, The multi-view corresponding construction unit includes the following steps: Based on the projection direction relationship, centerline alignment relationship, datum boundary correspondence relationship, dimension inheritance relationship and section mark mapping relationship, the same structural units in different views are associated to form cross-view structural mapping results between the main view and the top view, the main view and the side view, the main view and the section view, as well as the local enlarged view and the original view area. It also outputs the view number, structure identifier, mapping confidence information and constraint category label corresponding to each mapping result, which are used to support subsequent candidate region selection and local configuration inversion.

5. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 3, characterized in that, The suspected conflict screening unit is used to identify the corresponding region as a candidate region for inversion when the following conditions are met: The projected outline of the same structure is inconsistent in different views; The dimension values ​​of the same local feature are inconsistent in different views; The cross-section representation in the sectional view does not match the outline of the external view; The geometry of the enlarged view does not correspond to the selected area in the original view; The semantic meaning of the text annotation is inconsistent with the positional relationship of the graphic elements; The identification results show a break in the relationship with adjacent structures; It also outputs the candidate region coordinates, candidate region number, initial conflict type, conflict source view, and filtering priority for each candidate region to be inverted.

6. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 4, characterized in that, The local configuration inversion unit includes the following steps: Local structure restoration is performed on the candidate region to be inverted, generating intermediate structure expression units including local geometric skeleton, boundary topological relationship, key dimension parameters, symmetry parameters, hole axis relationship parameters, slot boundary relationship parameters, local contour closure state, local dimension chain and cross-sectional contour relationship after sectioning. The intermediate structure expression units are used as unified input objects for cross-view verification, constraint solving and candidate error correction generation.

7. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 5, characterized in that, The projection consistency verification unit includes the following steps: The intermediate structural representation units are projected back to the main view, top view, side view, sectional view, and enlarged view respectively to obtain the contour projection deviation, dimension chain deviation, symbol anchor point deviation, projection missing deviation, and projection redundancy deviation. Based on the weighted results of each deviation, a view consistency deviation score and view deviation details for the candidate region are generated. At the same time, the source view of the deviation is recorded. The weight of each deviation is determined according to the contribution of the corresponding deviation in the historical annotated drawings to the error judgment result, and the weight of each deviation satisfies the normalization constraint.

8. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 6, characterized in that, The cross-domain constraint graph construction unit includes the following steps: Semantic constraints, geometric constraints, and process constraints are mapped into a unified constraint network, and drawing elements and their relationships are mapped into constraint nodes and constraint edges to support the propagation analysis of constraint conflicts and the search for satisfyable solutions. Among them, semantic constraints include the orientation relationship between dimension values ​​and dimension lines, the correspondence between geometric tolerances and constrained features, the connection relationship between datum symbols and target surfaces or target axes, and the association relationship between sectional markings and sectional results. Geometric constraints include collinearity, perpendicularity, parallelism, closure, symmetry, concentricity, and closed-loop dimensional chain. Process constraints include fit feasibility, accessibility of hole and groove machining, process compatibility of chamfers and fillets, assembly rationality of datum selection, and the matching relationship between tolerance allocation and machining capability.

9. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 7, characterized in that, The joint conflict analysis unit includes the following steps: A comprehensive anomaly score is generated by combining the view consistency deviation score, semantic conflict score, geometric conflict score, and process conflict score. The comprehensive anomaly score is then compared with a preset anomaly judgment threshold. When the comprehensive anomaly score reaches the anomaly judgment threshold, the error area is determined, and the conflict type, conflict intensity, and conflict propagation path are output. The error source localization unit is used to determine the error source from primitive nodes, dimension nodes, symbol nodes and section expression nodes based on the conflict propagation path in the unified constraint network, and outputs the view to which the error source belongs, its location coordinates, error type, error confidence score and scope of influence.

10. The high-precision industrial drawing element recognition and intelligent error correction system according to claim 8, characterized in that, The candidate error correction generation unit includes the following steps: Based on the multi-view back projection results, the set of basic element relationships, the satisfyable solutions of the unified constraint network, and the stable expression patterns of similar structures in historical drawings, the corrected dimension values, the completed outline, the corrected cutting direction, the repositioned tolerance symbol, the corrected datum attachment position, and the corrected local enlarged area identifier are generated. The candidate error correction results are then ranked according to the consistency review results, constraint satisfaction review results, and comprehensive anomaly scores. Output the candidate error correction results that are ranked first and have passed the review as the final error correction results, while retaining the candidate results that have not passed the review and the reasons for their failure.

Images