Coal preparation electromechanical equipment drawing multi-modal semantic recognition system

Abstract

The present application relates to the technical field of graphic data reading and electric digital data processing, and discloses a multi-modal semantic recognition system for coal preparation machine electric equipment drawings. The system obtains key parameters recognized from the drawings, binds two-dimensional source coordinates and recognition channel attributes for the key parameters, and generates a set of identified key parameters. According to the equipment template, the system maps the key parameters into modeling instructions and executes to generate a three-dimensional model object. The two-dimensional source coordinates, recognition channel attributes, key parameters and instruction identifiers are written into traceability metadata. After the model is generated, the bounding box size, interface coordinates or interference volume are read reversely, and deviation checking is performed in combination with the identified key parameter set and the modeling instruction constraints. When an abnormality occurs, the source drawing position, recognition channel and affected modeling instruction set are located, and then local reconstruction and weight rewriting are performed. The present application can form a traceable closed loop between the drawing recognition result and the three-dimensional model abnormality, and reduce overall troubleshooting and overall reconstruction.

Description

Technical Field

[0001] This invention relates to the field of graphic data reading and electronic digital data processing technology, and in particular to a multimodal semantic recognition system for drawings of coal preparation equipment. Background Technology

[0002] In the digital design process of electromechanical engineering in coal preparation plants, components such as vibrating screens, filter presses, thickeners, crushers, feeders, chutes, hoppers, supports, equipment foundations, and pipeline interface connections are typically sourced from two-dimensional engineering drawings. To utilize these drawings for subsequent design deliverables, it is necessary to read the graphics, dimensions, text tables, detailed drawings, and equipment parameters from the drawings and convert them into data capable of driving CAD modeling.

[0003] Existing technologies typically output key parameters through image analysis, OCR recognition, dimension annotation recognition, multimodal semantic understanding, or rule verification, and then generate a 3D model by calling a CAD interface based on equipment templates or parametric modeling rules. Some solutions also establish a positive mapping relationship between 2D drawing elements and 3D model objects to facilitate model generation, parameter management, or 2D / 3D consistency verification.

[0004] However, the same key parameter in coal preparation equipment drawings may appear simultaneously in the main view, detailed drawings, equipment list, and technical specifications, leading to inconsistencies between different sources. Even after the front-end identifies the parameters to drive the CAD model generation, interface misalignment, incorrect hole counts, bounding box overflows, or spatial interference may still occur. Existing solutions typically lack a fine-grained traceability chain that reverse-engineers abnormal 3D model objects to the source 2D parameters, identification channels, and modeling instructions. This makes it difficult to determine the location of the source drawing that caused the anomaly and the scope of the affected modeling instructions after an anomaly occurs, requiring manual global investigation or complete reconstruction.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] This invention provides a multimodal semantic recognition system for coal preparation equipment drawings to solve the problem in the prior art that there is no reversible tracing chain between drawing recognition parameters, CAD modeling instructions and 3D model objects, which makes it difficult to locate the source 2D parameters, recognition channels and affected modeling instructions of 3D model anomalies.

[0007] The technical solution of this invention is: a multimodal semantic recognition system for coal preparation equipment drawings, comprising the following modules: an identifier parameter acquisition module: used to acquire key parameters for drawing recognition, bind two-dimensional source coordinates and recognition channel attributes to the key parameters, and generate an identifier key parameter set; a template and instruction assembly module: used to map the identifier key parameter set to modeling instructions according to the equipment template, establish the association between parameters and instructions, and generate a modeling instruction sequence; a model generation and traceability injection module: used to execute the modeling instruction sequence to generate a three-dimensional model object, and write the two-dimensional source coordinates, the recognition channel attributes, the key parameters, and the instruction identifier into the traceability metadata of the three-dimensional model object; and a reverse spatial feature reading module. The module is used to read the three-dimensional spatial physical features of the generated three-dimensional model object in reverse, including bounding box size, interface coordinates, or interference volume; the cross-checking and anomaly tracing module is used to check the deviation between the three-dimensional spatial physical features and the identified key parameter set and modeling instruction constraints, and to parse the traceability metadata when an anomaly occurs, locate the two-dimensional source coordinates, the identification channel attributes, and the set of affected modeling instructions that caused the anomaly; the local reconstruction and weight write-back module is used to obtain correction parameters and freeze non-associated objects, perform local reconstruction only on the set of affected modeling instructions based on the correction parameters, and adjust the weight of the identification channel in subsequent parameter fusion according to the tracing results.

[0008] Optionally, the identifier parameter acquisition module is further configured to: encapsulate the key parameters obtained from drawing identification into a parameter object, wherein the parameter object includes at least parameter identifier, parameter name, parameter value, unit, two-dimensional source coordinates, source view, identification channel attribute, confidence level, verification status, and modeling field mapping identifier; wherein the two-dimensional source coordinates are used to represent the page number, frame position, or drawing coordinate range of the key parameter in the source drawing, and the identification channel attribute is used to represent the extraction source of the key parameter.

[0009] Optionally, the identifier parameter acquisition module is further configured to: when there are candidate values ​​for the same key parameter from multiple identification channels, record the source view, identification channel attribute and confidence level corresponding to each candidate value; when the candidate value does not meet the consistency condition, mark the corresponding key parameter as a conflicting parameter; mark the conflicting parameter or the key parameter whose confidence level does not meet the input condition as a parameter to be reviewed, and prevent the parameter to be reviewed from directly entering the template and instruction assembly module.

[0010] Optionally, the template and instruction assembly module is further configured to: match the corresponding equipment template according to the equipment type and parameter integrity; map the parameter values ​​in the identified key parameter set to CAD modeling parameters; generate a modeling instruction sequence according to the equipment template and the CAD modeling parameters; and establish a correspondence between parameter identifiers, CAD modeling parameters, modeling instruction identifiers and three-dimensional model object identifiers, so that the modeling instruction sequence can retain the traceability basis from parameters to model objects.

[0011] Optionally, the model generation and traceability injection module is further configured to: execute the modeling instruction sequence to generate the three-dimensional model object; generate traceability metadata based on the correspondence, wherein the traceability metadata includes at least parameter identifiers, two-dimensional source coordinates, identification channel attributes, modeling instruction identifiers, template identifiers, and three-dimensional model object identifiers; and write the traceability metadata into the extended attributes, additional data dictionary, or associated data record of the three-dimensional model object in the form of serialized data, key-value mapping data, or object association data.

[0012] Optionally, the model generation and traceability injection module is further configured to: verify whether the traceability metadata has been written; re-execute the writing if the writing fails; and if the re-writing is still not completed, mark the corresponding 3D model object as a non-traceability model object and suspend it as an object to be reviewed, so as to prevent the non-traceability model object from being directly output as a verification passed object.

[0013] Optionally, the reverse spatial feature reading module and the cross-verification and anomaly tracing module are further used to: reversely read the bounding box size, interface coordinates, number of holes, number of components, attribute fields, or interference volume of the three-dimensional model object; generate a bounding box deviation based on the bounding box size and the theoretical bounding box size determined by the identified key parameter set; generate an interface position deviation based on the interface coordinates and the theoretical interface coordinates; and generate a spatial interference verification result based on the interference volume; when the bounding box deviation, the interface position deviation, the spatial interference verification result, the number of holes, the number of components, or the attribute fields do not meet the set conditions, determine the abnormal model object and generate an anomaly type.

[0014] Optionally, the cross-verification and anomaly tracing module is further configured to: when the geometric data of the three-dimensional model object does not meet the interference volume determination conditions or the reading fails, downgrade to using the bounding box size for boundary verification; generate a geometric reading anomaly identifier based on the boundary verification result, and associate the geometric reading anomaly identifier with the corresponding three-dimensional model object identifier, so that subsequent anomaly tracing does not depend on the missing interference volume data.

[0015] Optionally, the cross-verification and anomaly tracing module is further configured to: parse the corresponding traceability metadata based on the three-dimensional model object identifier of the anomaly model object; extract the parameter identifier that caused the anomaly, the source two-dimensional source coordinates, the identification channel attribute, and the set of affected modeling instructions; generate source drawing positioning information based on the source two-dimensional source coordinates, and output the set of affected modeling instructions to the local reconstruction and weight write-back module.

[0016] Optionally, the local reconstruction and weight write-back module is further configured to: after obtaining the correction parameters, freeze non-associated objects that have no dependency relationship with the parameter identifier, and perform local reconstruction only on the affected set of modeling instructions; when the same type of anomaly points to the same identification channel attribute multiple times within a set evaluation period, reduce the adoption weight of the identification channel attribute in subsequent parameter fusion according to the number of anomalies or the deviation value, and retain the lowest adoption weight; when spatial interference occurs again after local reconstruction, roll back to the model state before local reconstruction, and mark the corresponding associated object as an object to be manually reviewed.

[0017] This invention binds key parameters obtained from drawing recognition to two-dimensional source coordinates and recognition channel attributes, ensuring that these parameters have a traceable source before entering CAD modeling. It establishes a link between parameters and modeling instructions through a template and instruction assembly module, and writes key parameters, two-dimensional source coordinates, recognition channel attributes, and instruction identifiers into the traceability metadata of the three-dimensional model object during model generation, creating a reversible and resolvable correspondence between the three-dimensional model object and the source two-dimensional parameters and modeling instructions. By reverse-reading bounding box dimensions, interface coordinates, or interference volumes and performing deviation checks against the identifiable key parameter set and modeling instruction constraints, interface misalignment, bounding box out-of-bounds, or spatial interference can be identified as traceable anomalies. By parsing the traceability metadata when anomalies occur, the source two-dimensional source coordinates, recognition channel attributes, and the affected set of modeling instructions are located. Based on corrected parameters, only the affected set of modeling instructions is partially reconstructed, reducing the need for overall reconstruction of unrelated model objects and improving the problems of data fragmentation and difficulty in anomaly traceability between existing drawing recognition, CAD modeling, and model verification processes. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the logical architecture of a multimodal semantic recognition system for coal preparation equipment drawings provided in an embodiment of the present invention;

[0019] Figure 2 This is a schematic diagram of the generation of key identification parameters and the interception of conflicts in multiple identification channels provided in one embodiment of the present invention;

[0020] Figure 3This is a schematic diagram of the device template matching, modeling instruction assembly, and parameter-instruction-model correspondence provided in one embodiment of the present invention;

[0021] Figure 4 This is a schematic diagram of writing traceability metadata to a 3D model object and the write failure suspension provided in one embodiment of the present invention;

[0022] Figure 5 This is a schematic diagram of reverse spatial feature reading, deviation verification, and anomaly tracing of a three-dimensional model object provided in one embodiment of the present invention;

[0023] Figure 6 This is a schematic diagram of the timing of local reconstruction, channel weight write-back, and secondary verification rollback provided in one embodiment of the present invention. Detailed Implementation

[0024] 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 only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0025] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0026] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] As mentioned earlier, existing solutions for drawing recognition and CAD modeling of coal preparation equipment typically focus on front-end parameter extraction or forward model generation, lacking a fine-grained traceability chain that reverse-engineers abnormal 3D model objects back to the source 2D parameters, recognition channels, and modeling instructions. When the 3D model generated by the recognition parameters exhibits interface misalignment, bounding box out-of-bounds errors, or spatial interference, the system struggles to determine the source of the anomaly and the affected local instruction range, easily relying on manual global investigation or overall reconstruction.

[0028] To address this, the present invention provides a multimodal semantic recognition system for coal preparation equipment drawings. The system uses an identifier parameter acquisition module to bind two-dimensional source coordinates and identification channel attributes to key parameters. A template and instruction assembly module establishes the association between parameters and modeling instructions. A model generation and traceability injection module writes relevant traceability metadata into the three-dimensional model object. Finally, a reverse spatial feature reading module, a cross-checking and anomaly tracing module, and a local reconstruction and weight write-back module perform reverse spatial physical feature verification, anomaly tracing, and local reconstruction on the generated three-dimensional model object.

[0029] The following is combined with Figures 1 to 6 This invention is described in detail.

[0030] Example 1:

[0031] like Figure 1 As shown in the figure, this embodiment of the invention provides a multimodal semantic recognition system for coal preparation equipment drawings, which can run on an engineering design data processing server, a CAD secondary development service environment, a model quality verification service environment, or an enterprise's internal drawing modeling service environment. The system receives drawing files that can be DWG, DXF, DGN, PDF, images, or scanned copies. The target objects corresponding to the drawing files can be vibrating screens, filter presses, thickeners, crushers, feeders, chutes, hoppers, supports, equipment foundations, or pipeline interface components.

[0032] The present invention provides a multimodal semantic recognition system for coal preparation equipment drawings, comprising the following modules: an identifier parameter acquisition module, a template and instruction assembly module, a model generation and traceability injection module, a reverse spatial feature reading module, a cross-checking and anomaly tracing module, and a local reconstruction and weight write-back module.

[0033] The identifier parameter acquisition module is used to acquire key parameters for drawing recognition, bind two-dimensional source coordinates and recognition channel attributes to the key parameters, and generate an identifier key parameter set.

[0034] Specifically, the identifier parameter acquisition module can receive front-end drawing recognition results, which may include graphic geometry recognition results, dimension annotation recognition results, text table recognition results, multimodal semantic recognition results, and professional rule verification results.

[0035] The front-end drawing recognition results can be obtained from the drawing recognition access function in this system or provided by an external drawing recognition system. The improvement of this invention focuses on the identification processing of key parameters in the front-end drawing recognition results, the association of modeling instructions, the writing of traceable metadata, the reverse spatial verification, the anomaly tracing, and the local reconstruction processing.

[0036] The key parameters may include the main dimensions of the equipment, installation elevation, interface coordinates, interface diameter, number of holes, support positions, number of components, material fields, attribute fields, and modeling control fields. The two-dimensional source coordinates are used to record the page number, frame position, or coordinate range of the key parameters in the source drawing, and the identification channel attribute is used to record the source from which the key parameters were extracted.

[0037] The template and instruction assembly module is used to map the set of identified key parameters into modeling instructions based on the device template, establish the association between parameters and instructions, and generate a modeling instruction sequence.

[0038] Specifically, after receiving the set of identifiable key parameters, the template and instruction assembly module can match the corresponding equipment template according to the equipment type and parameter integrity, and convert parameter values ​​such as equipment length, equipment width, equipment height, interface coordinates, number of holes, support position, and attribute fields into CAD modeling parameters. The modeling instruction sequence may include instructions such as creating a reference coordinate system, generating the main geometry, creating the contour, extruding, rotating, sweeping, Boolean merging, Boolean subtraction, arraying, moving, mirroring, placing auxiliary components, setting layers, writing attributes, and saving the model.

[0039] The model generation and traceability injection module is used to execute the modeling instruction sequence to generate a three-dimensional model object, and write the two-dimensional source coordinates, the identification channel attributes, the key parameters and instruction identifiers into the traceability metadata of the three-dimensional model object.

[0040] Specifically, when the model generation and traceability injection module calls the CAD interface to generate a 3D model object, it writes the parameter identifier, 2D source coordinates, identification channel attribute, modeling instruction identifier, template identifier, and 3D model object identifier into the extended attributes, additional data dictionary, or associated data record of the 3D model object, so that the corresponding source 2D parameters and modeling instructions can be retrieved from the 3D model object later.

[0041] The reverse spatial feature reading module is used to read the three-dimensional spatial physical features of the three-dimensional model object after it is generated. The three-dimensional spatial physical features include bounding box size, interface coordinates or interference volume.

[0042] Specifically, the reverse spatial feature reading module can read the bounding box size, interface 3D coordinates, number of holes, number of components, attribute fields, layer fields, and interference volume of the generated 3D model object through the CAD geometric measurement interface, and send the reading results to the cross-checking and anomaly tracing module.

[0043] The cross-verification and anomaly tracing module is used to verify the deviation between the three-dimensional spatial physical features and the identified key parameter set and modeling instruction constraints. When an anomaly occurs, the traceability metadata is parsed to locate the two-dimensional source coordinates, the identification channel attributes, and the set of affected modeling instructions that caused the anomaly.

[0044] Specifically, the cross-verification and anomaly tracing module can compare the bounding box size of the actual generated model with the theoretical bounding box size, compare the actual interface coordinates with the theoretical interface coordinates, and determine the abnormal model object based on the interference volume, the number of holes, the number of components, or the consistency of attribute fields.

[0045] The local reconstruction and weight write-back module is used to obtain correction parameters and freeze unrelated objects. Based on the correction parameters, local reconstruction is performed only on the affected modeling instruction set, and the weight of the identification channel in subsequent parameter fusion is adjusted according to the source tracing results.

[0046] Specifically, after receiving the set of affected modeling instructions output by the cross-checking and anomaly tracing module, the local reconstruction and weight write-back module freezes the 3D model objects that are not dependent on the anomaly parameters, only re-executes the local modeling instructions affected by the corrected parameters, and reduces the weight of the identification channel attribute in subsequent parameter fusion, conflict resolution or confidence evaluation when the same anomaly points to the same identification channel attribute multiple times.

[0047] The identified key parameter set is a structured data object that can be continuously used in subsequent CAD modeling instruction assembly, model object traceability metadata writing, reverse spatial feature verification, and anomaly source analysis. The three-dimensional model object can be the main model, auxiliary components, interfaces, holes, supports, foundations, chutes, funnels, or pipeline interface components generated by CAD modeling instructions.

[0048] Based on the above system, when an anomaly occurs in a 3D model object, the present invention can reverse-parse the abnormal object to the source 2D coordinates, identify the channel attributes and the set of affected modeling instructions, and perform local reconstruction only on the set of affected modeling instructions after obtaining the correction parameters, thereby forming a traceable, verifiable and locally repairable processing closed loop between the drawing recognition results, CAD modeling instructions and 3D model objects.

[0049] Example 2:

[0050] To provide a clearer and more complete explanation of the above embodiments, the present invention also provides an embodiment two. In this embodiment two...

[0051] like Figure 2 As shown, preferably, the identifier parameter acquisition module is further used for:

[0052] The key parameters obtained from drawing recognition are encapsulated into parameter objects. The parameter objects include at least parameter identifier, parameter name, parameter value, unit, two-dimensional source coordinates, source view, recognition channel attribute, confidence level, verification status, and modeling field mapping identifier.

[0053] Wherein, the two-dimensional source coordinates are used to represent the page number, frame position or coordinate range of the key parameter in the source drawing, and the identification channel attribute is used to represent the source from which the key parameter is extracted.

[0054] The source view can be a main view, top view, left view, sectional view, partial detail view, equipment list, material list, technical specification, or title block; the recognition channel attribute can be a graphic geometry channel, dimensioning channel, text table / OCR channel, multimodal semantic channel, or professional rule verification channel; the modeling field mapping identifier is used to point to the corresponding CAD modeling parameter field in the equipment template.

[0055] Specifically, after receiving the drawing recognition results, the identifier parameter acquisition module can first establish the association between the drawing file and the project, discipline, equipment object, and drawing number, and then perform drawing frame recognition, title block parsing, layer parsing, view area recognition, scale recognition, and coordinate unification processing on the drawing. The above processing is only used to form a unified coordinate and field basis for drawings of different formats, and does not change the core technical line of this invention, which is based on traceable metadata injection and reverse verification.

[0056] Before the parameters enter the template and instruction assembly module, the identifier parameter acquisition module can set parameter confidence input conditions (for example, the parameter confidence threshold is 0.65, dimensionless, derived from the preset configuration of field integrity requirements in the equipment template library and the manual review records of similar historical drawings). When the confidence of a certain key parameter is lower than the input condition, the key parameter is not directly entered into CAD modeling as a confirmed parameter, but is marked as a parameter to be reviewed.

[0057] More preferably, the identifier parameter acquisition module is further used for:

[0058] When there are candidate values ​​for the same key parameter from multiple recognition channels, record the source view, recognition channel attribute, and confidence level corresponding to each candidate value.

[0059] When the candidate value does not meet the consistency condition, the corresponding key parameter is marked as a conflict parameter.

[0060] The conflicting parameters or key parameters whose confidence levels do not meet the input conditions are marked as parameters to be reviewed, and the parameters to be reviewed are prevented from directly entering the template and instruction assembly module.

[0061] The consistency conditions can be determined by the equipment template library's preset tolerances, drawing scale, CAD modeling unit conversion rules, and manual review records. The consistency conditions are numerical consistency conditions corresponding to the key parameter field types; for length-type parameters, the consistency conditions at least include the maximum absolute difference condition and the relative difference condition between candidate values; for installation elevation-type parameters, the consistency conditions at least include the elevation difference condition.

[0062] For length parameters in millimeters, the consistency condition can be configured such that the maximum absolute difference between candidate values ​​does not exceed 2mm and the relative difference does not exceed 0.5% (for example, if the same interface diameter is identified as 200mm by the dimension annotation channel and 201mm by the text / table / OCR channel, the absolute difference between the two is 1mm and the relative difference is 0.5%, which can be determined to meet the consistency condition; the 2mm and 0.5% are derived from the preset tolerances for interface parameters and the CAD unit conversion rules in the device template library).

[0063] For installation elevation parameters, the consistency condition can be configured so that the elevation difference does not exceed 5mm (for example, the elevation in the main view is identified as +12.500m, and the elevation in the equipment table is converted to +12.504m, with a difference of 4mm, which can be determined to meet the consistency condition; this 5mm comes from the coal preparation electromechanical assembly rules and the project modeling accuracy level configuration).

[0064] For example, the system identifies the equipment length, width, height, interface center coordinates, support hole spacing, and equipment number from the overall drawing of a vibrating screen. The equipment length is derived from the dimensions in the main view, the interface center coordinates from the detailed view, and the equipment number from the title block. The identifier parameter acquisition module encapsulates these parameters into parameter objects and records the source drawing page number, frame position, identification channel attribute, confidence level, and modeling field mapping identifier for each parameter object. If different candidate values ​​for the interface diameter appear in the detailed view and the equipment table, the system records the source view and identification channel attribute of each candidate value. If the consistency condition is not met, the interface diameter is marked as a conflicting parameter and prevented from directly entering automatic modeling.

[0065] like Figure 3 As shown, preferably, the template and instruction assembly module is further used for:

[0066] Match the corresponding device template based on device type and parameter completeness.

[0067] The parameter values ​​in the identified key parameter set are mapped to CAD modeling parameters.

[0068] A sequence of modeling instructions is generated based on the device template and the CAD modeling parameters.

[0069] Establish a correspondence between parameter identifiers, CAD modeling parameters, modeling instruction identifiers, and 3D model object identifiers, so that the modeling instruction sequence can retain the traceability basis from parameters to model objects.

[0070] The equipment templates can be set according to the types of coal preparation equipment and auxiliary components, including crushing equipment templates, screening equipment templates, feeding equipment templates, filter press equipment templates, thickening equipment templates, chute and funnel templates, support foundation templates, and pipeline interface component templates.

[0071] The equipment template can pre-store required parameter fields, default fields, modeling steps, CAD modeling formal parameter fields, local component dependencies, and verification rules. The default fields are only used for low-detail placeholders or conservative display before manual review, and are not used as final delivery fields without review.

[0072] Specifically, when the target equipment is a screening device, the template and instruction assembly module can read fields such as equipment length, equipment width, equipment height, tilt angle, inlet position, outlet position, support position and equipment number, and convert them into modeling instructions such as main body geometry creation, tilt positioning, interface creation, support assembly, layer setting and attribute writing.

[0073] When the target component is a non-standard chute or funnel, the template and instruction assembly module can read fields such as upper opening size, lower opening size, wall thickness, connection point coordinates, inclination angle and installation elevation, and convert them into modeling instructions such as contour generation, lofting, sweeping, Boolean processing and attribute writing.

[0074] When the equipment type or parameter completeness does not meet the minimum field requirements of the corresponding equipment template, the template and instruction assembly module can mark the component as a weakly dependent object and output a status pending review. The minimum field requirements can be preset by the equipment template library (for example, a screening equipment template requires at least a length of 5200mm, a width of 1800mm, a height of 2100mm, an inclination angle of 15°, and at least two support position coordinates, derived from the modeling field completeness configuration of this type of template). If the above fields are missing, a deliverable model will not be generated directly.

[0075] Preferably, the model generation and traceability injection module is further used for:

[0076] The three-dimensional model object is generated by executing the modeling instruction sequence.

[0077] Based on the correspondence, traceability metadata is generated, which includes at least parameter identifiers, two-dimensional source coordinates, identification channel attributes, modeling instruction identifiers, template identifiers, and three-dimensional model object identifiers.

[0078] The traceability metadata is written into the extended attributes, additional data dictionary, or associated data record of the 3D model object in the form of serialized data, key-value mapping data, or object association data.

[0079] The traceability metadata can be implemented using serialized strings, key-value mapping data, hash table records, or data records associated with the 3D model object identifier, without being limited to a specific CAD vendor interface. The fields of the traceability metadata may include parameter identifier, parameter name, parameter value, unit, 2D source coordinates, identification channel attribute, modeling instruction identifier, template identifier, 3D model object identifier, and verification status. After the traceability metadata is written, the 3D model object no longer only represents the geometry itself, but also preserves the resolvable relationship between its source parameters and generation instructions.

[0080] like Figure 4 As shown, more preferably, the model generation and traceability injection module is further used for:

[0081] Verify whether the traceability metadata has been written completely.

[0082] Re-execute the write operation if it fails.

[0083] While the rewriting process is still incomplete, the corresponding 3D model object is marked as a non-traceable model object and suspended as an object to be reviewed, so as to prevent the non-traceable model object from being directly output as a verification passed object.

[0084] Specifically, the model generation and traceability injection module can read the extended attributes, additional data dictionary, or associated data records of each 3D model object after its generation, and verify whether they contain parameter identifiers, 2D source coordinates, identification channel attributes, modeling instruction identifiers, template identifiers, and 3D model object identifiers. If the write operation fails, the system re-executes the write operation according to the maximum number of retry attempts (for example, the maximum number of retry attempts is 2, derived from the system task scheduling strategy and CAD interface call stability configuration). "Still not completed after rewriting" means that after the preset retry limit has been reached, the traceability metadata still fails the write completion verification. If the write operation still fails, the system marks the object as a non-traceability model object and suspends it as an object awaiting review.

[0085] For example, the template and instruction assembly module generates main body creation instructions, interface creation instructions, support placement instructions, and attribute writing instructions for a specific filter press. After executing the interface creation instructions, the model generation and traceability injection module writes the interface diameter parameter identifier, interface center 2D source coordinates, dimension annotation channel attributes, interface creation instruction identifier, filter press template identifier, and interface object identifier into the corresponding interface 3D model object. Subsequently, if the interface position deviation exceeds the limit, the system can parse the above traceability metadata based on the interface object identifier without needing to perform a global check on the entire filter press model.

[0086] like Figure 5 As shown, preferably, the reverse spatial feature reading module and the cross-verification and anomaly tracing module are further used for:

[0087] The bounding box size, interface coordinates, number of holes, number of components, attribute fields, or interference volume of the 3D model object are read in reverse.

[0088] A bounding box deviation is generated based on the bounding box size and the theoretical bounding box size determined by the set of identified key parameters. An interface position deviation is generated based on the interface coordinates and the theoretical interface coordinates. A spatial interference verification result is generated based on the interference volume.

[0089] When the bounding box deviation, the interface position deviation, the spatial interference verification result, the number of holes, the number of components, or the attribute field does not meet the set conditions, an abnormal model object is determined and an abnormal type is generated.

[0090] The theoretical bounding box size is determined by the device length, device width, and device height from the identified key parameter set; the interface coordinates are read backwards from the generated 3D model object; the theoretical interface coordinates are determined by the interface positioning field in the device template and the CAD modeling unit conversion rules; the interference volume can be determined by the Boolean intersection or collision detection results of the CAD geometric kernel. The bounding box deviation threshold can be derived from the device template library's preset tolerance, drawing scale, and CAD modeling unit conversion rules (for example, a bounding box deviation threshold of 2%, dimensionless, with example device length values ​​of 5200mm, device width values ​​of 1800mm, and device height values ​​of 2100mm).

[0091] Specifically, the bounding box deviation can be expressed by the following formula:

[0092]

[0093] in, This represents the bounding box deviation rate, expressed in % . This is the actual bounding box length of the model, in mm. This represents the actual bounding box width of the model, in mm. This is the actual bounding box height of the model, in mm. The theoretical bounding box length, determined by the identifiable key parameter set, is in mm; This represents the theoretical bounding box width, in mm. This represents the theoretical bounding box height, in mm. (Calculated...) This is used as input to the cross-checking and anomaly tracing module to determine whether the overall size of the 3D model object exceeds the limits. In this formula, This represents the relative volume deviation based on the theoretical bounding box volume, and is one method of calculating the bounding box deviation.

[0094] The interface position deviation threshold can be derived from the interface assembly tolerance, coal preparation electromechanical assembly rules, and CAD modeling unit conversion rules in the equipment template library (for example, the interface position deviation threshold is 5mm, and the theoretical interface coordinate example value is...). Example values ​​of actual interface coordinates of the model ).

[0095] Specifically, the interface position deviation can be expressed by the following formula:

[0096]

[0097] in, This refers to the interface position deviation, in mm. These are the actual interface coordinates of the model, in mm. These are the theoretical interface coordinates, in mm. (Calculated...) This is used as input to the cross-verification and anomaly tracing module to determine whether there is a spatial misalignment in the interface model object.

[0098] The spatial interference verification result can be determined based on the interference volume and the maximum permissible interference volume limit. The maximum permissible interference volume limit can be derived from the pre-set assembly tolerances in the equipment template library, the assembly rules for coal preparation electromechanical systems, and the CAD geometric verification level (e.g., the maximum permissible interference volume limit is...). This value corresponds to small volume error filtering and is not used as customer test data.

[0099] Specifically, the spatial interferometry verification result can be expressed by the following formula:

[0100]

[0101] in, The results of the spatial interference verification are dimensionless. The interference volume is obtained by Boolean intersection or collision detection, in units of ; The maximum permissible interference volume limit, in units of .when When the cross-checking and anomaly tracing module determines that the corresponding 3D model object has a spatial interference anomaly; when At that time, the three-dimensional model object satisfies the set conditions in the interference judgment item.

[0102] More preferably, the cross-verification and anomaly tracing module is further used for:

[0103] When the geometric data of the three-dimensional model object does not meet the conditions for determining the interference volume or fails to be read, the boundary is checked by using the bounding box size.

[0104] Based on the boundary verification results, a geometric reading anomaly identifier is generated, and the geometric reading anomaly identifier is associated with the corresponding 3D model object identifier, so that subsequent anomaly tracing does not depend on the missing interference volume data.

[0105] Specifically, when the geometric patches of a 3D model object are not closed, Boolean intersection calculations fail, or interference volume reading fails, the cross-checking and anomaly tracing module does not directly determine that the object has passed the check. Instead, it calls the bounding box dimensions output by the reverse spatial feature reading module for boundary check. If the bounding box deviation exceeds the bounding box deviation threshold, the system generates a geometric reading anomaly identifier and stores it in association with the 3D model object identifier, parameter identifier, and modeling instruction identifier.

[0106] Preferably, the cross-verification and anomaly tracing module is further used for:

[0107] The corresponding traceability metadata is parsed based on the three-dimensional model object identifier of the abnormal model object.

[0108] Extract the parameter identifiers that caused the anomaly, the source 2D coordinates, identify the channel attributes, and the set of affected modeling instructions.

[0109] The source drawing positioning information is generated based on the source 2D source coordinates, and the affected modeling instruction set is output to the local reconstruction and weight write-back module.

[0110] The set of affected modeling instructions can be determined based on the correspondence between parameter identifiers and modeling instruction identifiers. For interface creation instructions directly driven by a single parameter, the system includes the interface creation instruction and its subsequent dependent attribute writing instructions in the set of affected modeling instructions. For main body geometry creation instructions driven by multiple parameters, the system only includes the main body geometry creation instruction and its dependent instructions in the set of affected modeling instructions when abnormal parameters are involved in the calculation of main body dimensions or positioning. Non-associated objects that do not have a dependency relationship with abnormal parameters remain in a frozen or read-only state.

[0111] For example, in a non-standard chute model, the system reads back and finds an interference volume between the downstream interface model object and the adjacent pipeline interface, and The cross-verification and anomaly tracing module parses and traces metadata based on the 3D model object identifier of the anomaly model object, extracting the parameter identifier that caused the anomaly as the lower opening width parameter, the source 2D coordinates located in the local detail area of ​​the third drawing, and identifying the channel attribute as a text table. The affected modeling instruction set includes the lower contour generation instruction, the lofting instruction, and the interface attribute writing instruction. Based on this, the system generates source drawing positioning information and sends the affected modeling instruction set to the local reconstruction and weight write-back module.

[0112] like Figure 6 As shown, preferably, the local reconstruction and weight write-back module is further used for:

[0113] After obtaining the corrected parameters, freeze non-associated objects that have no dependency on the parameter identifier, and perform local reconstruction only on the affected set of modeling instructions.

[0114] The correction parameter can be a parameter value confirmed after reviewing the conflicting parameter or the parameter to be reviewed, or it can be a parameter value that has passed the consistency condition check and been confirmed for use among multiple candidate values.

[0115] When the same type of anomaly points to the same identification channel attribute multiple times within a set evaluation period, the adoption weight of that identification channel attribute in subsequent parameter fusion is reduced according to the number of anomalies or the deviation value, while the lowest adoption weight is retained.

[0116] When spatial interference occurs again after partial reconstruction, the model is rolled back to the state before the partial reconstruction, and the corresponding associated objects are marked as objects to be manually reviewed.

[0117] The evaluation cycle can be determined by the system task management strategy, the statistical rules of the device template library, and historical records of similar anomalies (for example, the evaluation cycle is set to 30 modeling task cycles, which comes from the statistical window of the system for continuous processing of similar device templates).

[0118] The minimum adoption weight is used to prevent a certain identification channel from being completely blocked due to a small number of anomalies (for example, the minimum adoption weight is 0.20, dimensionless, derived from the minimum retention configuration of the channel fusion strategy), and the attenuation coefficient is used to control the magnitude of weight reduction (for example, the attenuation coefficient is 0.30, dimensionless, derived from historical statistics of similar anomalies and manual verification results).

[0119] Specifically, the weights of the identified channel attributes in subsequent parameter fusion can be updated using the following formula:

[0120]

[0121] in, The updated adoption weights are dimensionless. The weights used before the update are dimensionless; The minimum weight is used, and it is dimensionless. The attenuation coefficient is dimensionless. Risk factor, dimensionless, in one specific embodiment, risk factor Based on the number of similar anomalies within the assessment period Bounding box deviation rate Interface position deviation Spatial interference verification results and manual verification results Determined; among them, , and Each value is converted into a normalized risk value between 0 and 1 according to its corresponding preset threshold. , and In one specific embodiment, , and By respectively , and The value is determined by dividing the result by the corresponding preset threshold and taking the value no greater than 1; The value is 1 when spatial interference anomalies exist and 0 when no spatial interference anomalies exist; when manual verification confirms that the corresponding identification channel attribute is the source of the anomaly. Set to 1, when manual verification fails to confirm that the corresponding identification channel attribute is the source of the anomaly. Set to 0. The risk factor... It can be expressed by the following formula:

[0122]

[0123] in, , , and The preset influence weight coefficients are defined, and the following conditions are met:

[0124]

[0125] Calculated Used to input the weighted update formula; when hour, The local reconstruction and weight write-back module does not reduce the weight of the identification channel attribute in subsequent parameter fusion.

[0126] Specifically, when the same identification channel attribute is repeatedly identified as the source of the same type of anomaly within the set evaluation period, and manual verification confirms that the corresponding parameters need to be corrected, the local reconstruction and weight write-back module will adjust the calculation based on the number of similar anomalies. Bounding box deviation rate Interface position deviation Spatial interference verification results and manual verification of results Calculate the risk factors When manual verification fails to confirm that the identification channel attribute is an anomaly, Set to 0, so that The value is 0, so as not to reduce the weight of the identification channel attribute in subsequent parameter fusion.

[0127] Calculated The identification parameter acquisition module is used to adjust the adoption weight of the corresponding identification channel attribute in subsequent similar parameter fusion, conflict resolution, or confidence evaluation.

[0128] For example, a text table If a channel is identified as the source of interface coordinate anomalies three consecutive times within 30 modeling task cycles, and manual verification confirms that its corresponding parameters all require correction, then the local reconstruction and weight write-back module can reduce the weight of this identified channel attribute in subsequent interface coordinate parameter fusion while retaining the minimum weight. This weight adjustment only applies to subsequent parameter fusion, conflict resolution, or confidence evaluation, and does not represent an improvement to the internal network structure or training method of the multimodal model.

[0129] After the local reconstruction is completed, the reverse spatial feature reading module reads the bounding box size, interface coordinates, and interference volume of the reconstructed model object again, and the cross-verification and anomaly tracing module performs deviation verification again. If spatial interference still occurs after local reconstruction, the local reconstruction and weight write-back module rolls back to the model state before local reconstruction and marks the corresponding associated objects as objects to be manually reviewed; if the manual review times out, the system suspends the delivery output of the device or associated component, and retains the verification log, anomaly type, source drawing location information, and parameter vector to be reviewed.

[0130] The output results may include verified CAD model files, parameter-instruction-model traceability tables, model verification logs, lists of abnormal model objects, source drawing location information, parameter vectors to be reviewed, local reconstruction logs, channel weight update records, and rollback processing records. For untraceable model objects, geometrically abnormal objects, objects with failed local reconstruction, or objects subject to secondary interference, the system will not output them as verified objects, but will retain them as objects to be reviewed.

[0131] Accordingly, this invention establishes two-dimensional source coordinates and identification channel attributes for key parameters obtained from drawing recognition through an identifier parameter acquisition module, establishes the correspondence between key parameters and modeling instructions through a template and instruction assembly module, and writes the two-dimensional source coordinates, identification channel attributes, key parameters, and instruction identifiers into the traceability metadata of the three-dimensional model object through a model generation and traceability injection module, so that the generated three-dimensional model object can retain the resolvable traceability relationship from the source two-dimensional parameters to the modeling instructions. On this basis, the bounding box size, interface coordinates, or interference volume are read through a reverse spatial feature reading module, and deviation verification and anomaly analysis are performed by a cross-checking and anomaly tracing module, so that the abnormal model object can be located to the corresponding source two-dimensional source coordinates, identification channel attributes, and the set of affected modeling instructions. Furthermore, the local reconstruction and weight write-back module performs local reconstruction only on the set of affected modeling instructions based on correction parameters, thereby forming a traceable, verifiable, and locally repairable processing closed loop between drawing recognition, CAD modeling, and model verification.

[0132] The above description is merely a preferred embodiment of the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multimodal semantic recognition system for coal preparation equipment drawings, characterized in that, Includes the following modules: Identification parameter acquisition module: used to acquire key parameters for drawing recognition, bind two-dimensional source coordinates and recognition channel attributes to the key parameters, and generate an identification key parameter set; Template and instruction assembly module: used to map the identified key parameter set to modeling instructions based on the device template, establish the association between parameters and instructions, and generate a modeling instruction sequence; Model generation and traceability injection module: used to execute the modeling instruction sequence to generate a three-dimensional model object, and write the two-dimensional source coordinates, the identification channel attributes, the key parameters and instruction identifiers into the traceability metadata of the three-dimensional model object; Reverse spatial feature reading module: used to read the three-dimensional spatial physical features of the three-dimensional model object after it is generated. The three-dimensional spatial physical features include bounding box size, interface coordinates or interference volume. Cross-verification and anomaly tracing module: used to verify the deviation between the three-dimensional spatial physical features and the identified key parameter set and modeling instruction constraints; when an anomaly occurs, it parses the traceability metadata to locate the two-dimensional source coordinates, the identification channel attributes and the set of affected modeling instructions that caused the anomaly. Local reconstruction and weight write-back module: used to obtain correction parameters and freeze non-associated objects, perform local reconstruction only on the affected modeling instruction set based on the correction parameters, and adjust the weight of the identification channel in subsequent parameter fusion according to the source tracing results.

2. The system according to claim 1, characterized in that, The identifier parameter acquisition module is further used for: The key parameters obtained from drawing recognition are encapsulated into parameter objects, which include parameter identifier, parameter name, parameter value, unit, two-dimensional source coordinates, source view, recognition channel attribute, confidence level, verification status, and modeling field mapping identifier; Wherein, the two-dimensional source coordinates are used to represent the page number, frame position or coordinate range of the key parameter in the source drawing, and the identification channel attribute is used to represent the source from which the key parameter is extracted.

3. The system according to claim 2, characterized in that, The identifier parameter acquisition module is further used for: When there are candidate values ​​for the same key parameter from multiple recognition channels, record the source view, recognition channel attribute, and confidence level corresponding to each candidate value; When the candidate value does not meet the consistency condition, the corresponding key parameter is marked as a conflict parameter; The conflicting parameters or key parameters whose confidence levels do not meet the input conditions are marked as parameters to be reviewed, and the parameters to be reviewed are prevented from directly entering the template and instruction assembly module.

4. The system according to claim 1, characterized in that, The template and instruction assembly module is further used for: Match the corresponding device template based on device type and parameter completeness; Map the parameter values ​​in the identified key parameter set to CAD modeling parameters; Generate a modeling instruction sequence based on the equipment template and the CAD modeling parameters; Establish a correspondence between parameter identifiers, CAD modeling parameters, modeling instruction identifiers, and 3D model object identifiers, so that the modeling instruction sequence can retain the traceability basis from parameters to model objects.

5. The system according to claim 4, characterized in that, The model generation and traceability injection module is further used for: The three-dimensional model object is generated by executing the modeling instruction sequence; Based on the correspondence, traceability metadata is generated, which includes parameter identifiers, two-dimensional source coordinates, identification channel attributes, modeling instruction identifiers, template identifiers, and three-dimensional model object identifiers. The traceability metadata is written into the extended attributes, additional data dictionary, or associated data record of the 3D model object in the form of serialized data, key-value mapping data, or object association data.

6. The system according to claim 5, characterized in that, The model generation and traceability injection module is further used for: Verify whether the traceability metadata has been completely written; Re-execute the write operation if it fails; While the rewriting process is still incomplete, the corresponding 3D model object is marked as a non-traceable model object and suspended as an object to be reviewed, so as to prevent the non-traceable model object from being directly output as a verification passed object.

7. The system according to claim 1, characterized in that, The reverse spatial feature reading module and the cross-verification and anomaly tracing module are further used for: Reverse read the bounding box size, interface coordinates, number of holes, number of components, attribute fields, or interference volume of the 3D model object; A bounding box deviation is generated based on the bounding box size and the theoretical bounding box size determined by the set of identified key parameters; an interface position deviation is generated based on the interface coordinates and the theoretical interface coordinates; and a spatial interference verification result is generated based on the interference volume. When the bounding box deviation, the interface position deviation, the spatial interference verification result, the number of holes, the number of components, or the attribute field does not meet the set conditions, an abnormal model object is determined and an abnormal type is generated.

8. The system according to claim 7, characterized in that, The cross-checking and anomaly tracing module is further used for: When the geometric data of the three-dimensional model object does not meet the conditions for determining the interference volume or fails to be read, the boundary check is performed by using the bounding box size. Based on the boundary verification results, a geometric reading anomaly identifier is generated, and the geometric reading anomaly identifier is associated with the corresponding 3D model object identifier, so that subsequent anomaly tracing does not depend on the missing interference volume data.

9. The system according to claim 7, characterized in that, The cross-checking and anomaly tracing module is further used for: The corresponding traceability metadata is parsed based on the 3D model object identifier of the abnormal model object. Extract the parameter identifiers that caused the anomaly, the source 2D coordinates, identify the channel attributes, and the set of affected modeling instructions; The source drawing positioning information is generated based on the source 2D source coordinates, and the affected modeling instruction set is output to the local reconstruction and weight write-back module.

10. The system according to claim 9, characterized in that, The local reconstruction and weight write-back module is further used for: After obtaining the corrected parameters, freeze non-associated objects that have no dependency on the parameter identifier, and perform local reconstruction only on the affected set of modeling instructions; When the same type of anomaly points to the same identification channel attribute multiple times within a set evaluation period, the adoption weight of that identification channel attribute in subsequent parameter fusion is reduced according to the number of anomalies or the deviation value, while the lowest adoption weight is retained. When spatial interference occurs again after partial reconstruction, the model is rolled back to the state before the partial reconstruction, and the corresponding associated objects are marked as objects to be manually reviewed.

Images