A drawing intelligent management method and device and a storage medium

Abstract

The application relates to an intelligent management method and device for drawings and a storage medium, applied to the technical field of engineering drawing digital management, and comprising the following steps: adopting a CNN+Transformer hybrid model, extracting drawing visual features through a convolutional neural network, combining a Transformer architecture to analyze semantic association, realizing three-level classification of drawings, improving accuracy, innovatively designing a multi-modal retrieval engine, supporting text, sketches and three-dimensional space to jointly realize real-time conflict detection, adopting a dynamic detection algorithm based on operation log hashing, calculating an operation sequence hash value through a distributed node, realizing millisecond-level conflict identification, adopting an incremental rendering strategy, enabling a designer to compare and modify differences in real time, and improving conflict solving efficiency by 5 times; a double-chain blockchain architecture is designed, permission operation logs are stored in a private chain to ensure efficient query, and permission change records are stored in a public chain to realize non-tampering, so that user privacy is protected and supervision and traceability requirements are met, and the calculation overhead is reduced by 30% compared with a traditional digital signature scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering drawing digital management, and particularly relates to a drawing intelligent management method and device and a storage medium. BACKGROUND

[0002] Construction drawing examination refers to the technical examination of construction design drawings of engineering construction projects by government regulatory departments. Previously, this work was completed by government construction regulatory departments themselves by organizing relevant technical personnel. Now, it is mainly completed by third-party professional examination institutions licensed by the government. After the technical examination, examination reports, qualification certificates and record tables and other documents are issued.

[0003] The existing drawing management technology mainly includes the following two schemes:

[0004] One is a traditional management system based on a rule engine: file format analysis and metadata matching technology are used to realize drawing classification, and artificial pre-set rules are relied on. The classification dimension is single and the content semantics of the drawing cannot be identified. The classification accuracy of mixed drawings in multiple fields is insufficient. The operation sequence timestamp comparison is used for collaborative editing conflict detection. When multiple users modify the same drawing area at the same time, the server needs to process the operation request in sequence, which leads to a conflict recognition delay of more than 500 ms, seriously affecting the collaboration efficiency. Moreover, the retrieval function is limited to text keyword matching, and lacks support for unstructured queries such as sketch contour and three-dimensional spatial relationship, with low precision.

[0005] The other is a cloud collaborative platform: it relies on a centralized storage architecture, and the drawing version management adopts linear history records without introducing semantic version coding, which leads to manual comparison of differences when merging versions, with an average time consumption of more than 30 minutes per time. The permission management is based on a role-based access control model, and the permission change records are stored in the central database, which has a risk of data tampering. Moreover, it lacks an intelligent compliance verification module, and problems such as missing drawing labels and size tolerance conflicts rely on manual checking, with a high error and omission rate. SUMMARY

[0006] Therefore, the present application aims to provide a drawing intelligent management method and device and storage medium to solve the problems in the prior art that the drawing classification is realized by using file format analysis and metadata matching technology, relies on artificial preset rules, the classification dimension is single and the content semantics of the drawing cannot be recognized, the classification accuracy for mixed drawings in multiple fields is insufficient, the operation sequence timestamp comparison is used for conflict detection in collaborative editing, when multiple users modify the same drawing area at the same time, the server needs to process the operation request in sequence, which leads to a conflict recognition delay of more than 500 ms, seriously affecting the collaboration efficiency, and the search function is limited to text keyword matching and lacks support for unstructured queries such as sketch contour and three-dimensional spatial relationship, resulting in low precision, and the role-based access control model is used for permission management, and the permission change records are stored in the central database, which has a data tampering risk, and there is no intelligent compliance verification module, and the problems of missing drawing labels and size tolerance conflicts depend on manual checking, and the error and omission rate is high.

[0007] According to a first aspect of an embodiment of the present application, a drawing intelligent management method is provided, which comprises:

[0008] Receiving a drawing uploaded by a user terminal, using a pre-trained CNN-Transformer hybrid model to extract features of the uploaded drawing, and outputting a three-dimensional classification label containing an engineering type, a design stage and a version state;

[0009] Monitoring the three-dimensional classification label by an operation log hash synchronizer deployed on an edge node, recording the editing operation of the user in real time and generating a SHA-3 hash sequence with a time and space stamp, detecting whether there is a concurrent modification of the same area of the drawing by multiple users according to the SHA-3 hash sequence with the time and space stamp, if there is, using a conflict positioning algorithm based on the hash value timestamp to obtain the conflict area and generate a conflict difference heat map, extracting the image features of the conflict area and converting them into search input features;

[0010] Building a hybrid search engine by using a pre-trained text conflict detection model, a sketch conflict detection model and a three-dimensional spatial relationship conflict detection model;

[0011] Inputting the search input features into the hybrid search engine, analyzing the text semantics by the text conflict detection model, extracting the sketch contour features by the Fourier descriptor of the sketch conflict detection model, and matching the three-dimensional spatial relationship by the BIM topological subgraph of the three-dimensional spatial relationship conflict detection model, the hybrid search engine outputs the conflict area as one or more combinations of text conflict, sketch conflict or three-dimensional spatial relationship conflict;

[0012] The permission allocation verification is performed on the conflict area, permission change records are stored in a public chain, permission change operation logs are stored in a private chain, and it is ensured that the permission allocation is traceable and tamper-proof.

[0013] A multi-modal query vectorization association reconstruction labeling network is used on the conflict area to identify an unclosed chain and locate a missing node, a tolerance transfer matrix model is established to predict assembly interference risks, and when the conflict area is not modified for a continuous preset period and passes the assembly interference risk check, cold storage archiving based on ErasureCoding is triggered to complete intelligent management of drawings.

[0014] Preferably,

[0015] The training of the CNN-Transformer hybrid model includes:

[0016] At least N cross-domain drawings are collected to construct a heterogeneous data set, and a three-level label of engineering type, design stage, and version state is labeled for each cross-domain drawing;

[0017] A loss function is set, the labeled cross-domain drawings are input into a CNN architecture and a Transformer architecture respectively, high-dimensional feature maps output by the CNN architecture and semantic vectors extracted by the Transformer architecture are subjected to attention weighted fusion, the fusion weight is dynamically adjusted through a learnable parameter matrix, and the training is iterated until the loss of the CNN architecture and the Transformer architecture no longer decreases, thereby obtaining a trained CNN-Transformer hybrid model;

[0018] When the misclassification rate of the newly added drawings exceeds a preset threshold, online fine-tuning feature incremental learning is started, the full connection layer parameters of the CNN-Transformer hybrid model are updated, and a double-channel dynamic fusion architecture and a restricted incremental learning strategy are used.

[0019] Preferably,

[0020] The SHA-3 hash sequence with time and space stamps is used to detect whether there is a multi-user concurrent modification of the same area of the drawings, and if so, a conflict positioning algorithm based on hash value time stamps is used to obtain a conflict area and generate a conflict difference heat map, which includes:

[0021] The user editing operation is decomposed into the smallest operation unit according to the SHA-3 hash sequence with time and space stamps, and each unit is attached with a position hash value;

[0022] A conflict judgment matrix based on overlap rate and operation type is established, the horizontal axis is the operation type, and the vertical axis is the influence area overlap rate, the influence area overlap rate is obtained through the smallest operation unit, and when the overlap rate is greater than a preset ratio and the operation type conflicts, it is marked as a high-risk conflict.

[0023] The WebGL is used to accelerate rendering of the difference region, and transparency gradient superposition display is realized.

[0024] Preferably,

[0025] The text semantic is analyzed by the text conflict detection model, the sketch contour feature is extracted by the Fourier descriptor of the sketch conflict detection model, and the three-dimensional space relationship is matched by the BIM topological subgraph of the three-dimensional space relationship conflict detection model.

[0026] The text conflict detection adopts a BERT-diagram knowledge graph joint embedding model to map the query statement and the preset diagram metadata to the same vector space, and then performs similarity matching in the same vector space, and selects the diagram corresponding to the diagram metadata with the highest similarity and greater than the preset similarity threshold as the text conflict detection result.

[0027] The sketch conflict detection converts the hand-drawn contour into a frequency domain feature by using an improved Fourier descriptor combined with a dynamic time warping algorithm, performs similarity calculation and sorting on the frequency domain feature and the frequency domain feature of the preset sketch, and selects the preset sketch with the highest similarity and greater than the preset similarity threshold as the sketch conflict detection result.

[0028] The three-dimensional space relationship conflict detection analyzes the bounding box hierarchy structure of the three-dimensional diagram to be queried by point cloud topological subgraph matching, and performs isomorphic comparison with the preset BIM database, and selects the three-dimensional diagram model corresponding to the bounding box hierarchy structure with the highest similarity and greater than the preset similarity threshold as the three-dimensional space relationship conflict detection result.

[0029] Preferably,

[0030] The bounding box hierarchy structure of the three-dimensional diagram to be queried is analyzed by point cloud topological subgraph matching, and includes:

[0031] The point cloud density of the three-dimensional diagram to be queried is standardized to 500 points per cubic meter by using a voxel grid downsampling algorithm, and the point cloud preprocessing is completed;

[0032] The hole structure feature of the three-dimensional diagram to be queried is extracted by an Alpha Shape algorithm, a multi-level bounding box tree is generated, and the topological feature extraction is completed;

[0033] The isomorphic comparison with the preset BIM database includes:

[0034] The VF2 algorithm is used for subgraph isomorphic matching, a rotation invariance constraint condition is introduced, and subgraph matching optimization is realized.

[0035] Preferably,

[0036] The permission allocation verification on the conflict area comprises:

[0037] The deployment of the Chainlink node realizes the data bidirectional verification between the public chain and the private chain, ensures that the verification delay of the data bidirectional verification between the public chain and the private chain is within a preset threshold time;

[0038] A plurality of permission templates are defined, the permission templates are used to limit editable and non-exportable, and support one-key application of the templates to a project group;

[0039] When the user operates the signature, 5 nodes are randomly selected to form a temporary ring, and it is ensured that the signature verification time consumption is within a preset threshold time.

[0040] Preferably,

[0041] The multi-modal query vectorization association reconstruction labeling network for the conflict area comprises:

[0042] The conflict area is dimensionally labeled, the dimensional label is converted into a weighted directed edge, a node represents a geometric element center point, a depth-first search is used to traverse an unclosed path, and a missing labeling position is located;

[0043] The missing labeling position is completed according to an adjacent dimensional tolerance.

[0044] Preferably,

[0045] The completion of the missing labeling position according to the adjacent dimensional tolerance comprises:

[0046] According to an upper and lower deviation range of adjacent dimensions, a maximum material condition is taken as a constraint, it is ensured that an assembly gap after accumulation of all dimensional tolerances still meets a limit requirement, a theoretical upper limit of a theoretically closed dimension is calculated according to a maximum positive deviation, a limit tolerance range of the missing dimension is inversely deduced, and the missing labeling position is completed through the limit tolerance range of the missing dimension;

[0047] Or,

[0048] A root mean square statistical model is used to confirm that the dimensional deviation conforms to a normal distribution, a standard deviation of all dimensions in the chain ring is counted, a theoretical standard deviation of the closed dimension is calculated, and finally a bilateral symmetric tolerance is generated, and the missing labeling position is completed through the bilateral symmetric tolerance;

[0049] Or,

[0050] The total amount of unclosed tolerances is evenly distributed to related dimensions according to a dimensional sensitivity weight, and is distributed to each ring according to a weight proportion, and the missing labeling position is completed according to the distribution result.

[0051] According to a second aspect of the embodiment of the present application, a smart drawing management device is provided, and the device comprises:

[0052] The classification label generation module is configured to receive drawings uploaded by a user terminal, extract features of the uploaded drawings by using a pre-trained CNN-Transformer hybrid model, and output three-dimensional classification labels containing an engineering type, a design stage, and a version state.

[0053] The conflict detection module is configured to monitor the three-dimensional classification labels by an operation log hash synchronizer deployed on an edge node, record editing operations of a user in real time and generate a SHA-3 hash sequence with a time and space stamp, detect whether there is a concurrent modification of the same area of the drawing by multiple users according to the SHA-3 hash sequence with the time and space stamp, if there is, acquire a conflict area and generate a conflict difference heat map by using a conflict positioning algorithm based on a hash value time stamp, extract image features of the conflict area and convert the image features into retrieval input features.

[0054] The retrieval engine building module is configured to build a hybrid retrieval engine by using a pre-trained text conflict detection model, a sketch conflict detection model, and a three-dimensional spatial relationship conflict detection model.

[0055] The conflict relationship analysis module is configured to input the retrieval input features into the hybrid retrieval engine, analyze text semantics by using the text conflict detection model, extract sketch contour features by using a Fourier descriptor of the sketch conflict detection model, match three-dimensional spatial relationships by using a BIM topological subgraph of the three-dimensional spatial relationship conflict detection model, and output the conflict area as one or more combinations of a text conflict, a sketch conflict, or a three-dimensional spatial relationship conflict.

[0056] The permission locking module is configured to perform permission allocation verification on the conflict area, store permission change records in a public chain, store permission change operation logs in a private chain, and ensure that the permission allocation is traceable and tamper-proof.

[0057] The archive management module is configured to use a multi-modal query vectorization correlation reconstruction annotation network to identify an unclosed chain and locate a missing node, establish a tolerance transfer matrix model to predict assembly interference risks, trigger cold storage archiving based on ErasureCoding to complete smart drawing management when the conflict area is not modified for a continuous preset period and passes the assembly interference risk check.

[0058] According to a third aspect of the embodiment of the present application, a storage medium is provided, and the storage medium stores a computer program, and the computer program is executed by a host computer to implement each step in the above method.

[0059] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0060] The application adopts a CNN+Transformer hybrid model, extracts visual features of drawings through a convolutional neural network, analyzes semantic association in combination with a Transformer architecture, realizes three-level classification of drawings, improves accuracy, innovatively designs a multi-modal retrieval engine, supports text, sketch and three-dimensional space to jointly realize real-time conflict detection, realizes millisecond-level conflict identification through a distributed node to calculate a hash value of an operation sequence based on a dynamic detection algorithm of an operation log hash, adopts an incremental rendering strategy, enables designers to compare and modify differences in real time, and improves conflict resolution efficiency by 5 times, designs a double-chain blockchain architecture, stores permission operation logs in a private chain to ensure efficient query, stores permission change records in a public chain to realize non-tamperability, protects user privacy and meets regulatory traceability requirements, and reduces 30% of the computing overhead compared with a traditional digital signature scheme, and a cold and hot data hierarchical storage strategy in combination with Erasure Coding coding technology reduces the storage cost of abandoned drawings, while greatly improving data recoverability.

[0061] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0062] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0063] Figure 1 is a flowchart of a drawing intelligent management method according to an exemplary embodiment;

[0064] Figure 2 is a flowchart of a CNN+Transformer hybrid model training process according to another exemplary embodiment;

[0065] Figure 3 is a system diagram of a drawing intelligent management device according to another exemplary embodiment;

[0066] In the drawings: 1-classification label generation module, 2-conflict detection module, 3-retrieval engine building module, 4-conflict relationship analysis module, 5-permission locking module, 6-archiving management module. DETAILED DESCRIPTION

[0067] The exemplary embodiments will be described in detail below with reference to the drawings. The following description is presented in connection with the accompanying drawings in which the same numbers are used in different drawings to refer to the same or similar elements. The implementations described in the following exemplary embodiments are not meant to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0068] Embodiment One

[0069] Figure 1 is a flowchart of a drawing intelligent management method according to an exemplary embodiment, as shown in Figure 1 , the method comprises:

[0070] S1, receiving the drawings uploaded by the user terminal, using a pre-trained CNN-Transformer hybrid model to extract features of the uploaded drawings, and outputting three-dimensional classification labels containing engineering type, design stage and version state;

[0071] S2, monitoring the three-dimensional classification labels by deploying an operation log hash synchronizer on the edge node, recording the user's editing operation in real time and generating a SHA-3 hash sequence with a time and space stamp, detecting whether there is a multi-user concurrent modification of the same area of the drawing according to the SHA-3 hash sequence with a time and space stamp, if there is, using a conflict positioning algorithm based on hash value timestamp to obtain the conflict area and generate a conflict difference heat map, extracting the image features of the conflict area and converting them into retrieval input features;

[0072] S3, building a hybrid retrieval engine through a pre-trained text conflict detection model, a sketch conflict detection model and a three-dimensional spatial relationship conflict detection model;

[0073] S4, inputting the retrieval input features into the hybrid retrieval engine, analyzing the text semantics through the text conflict detection model, extracting the sketch contour features through the Fourier descriptor of the sketch conflict detection model, and matching the three-dimensional spatial relationship through the BIM topological subgraph of the three-dimensional spatial relationship conflict detection model, the hybrid retrieval engine outputs the conflict area as one or more combinations of text conflict, sketch conflict or three-dimensional spatial relationship conflict;

[0074] S5, verifying the permission allocation of the conflict area, storing the permission change record in the public chain and storing the permission change operation log in the private chain, ensuring that the permission allocation is traceable and tamper-proof;

[0075] S6, adopting a multi-modal query vectorization association reconstruction labeling network on the conflict region, identifying an open chain and locating a missing node, establishing a tolerance transfer matrix model to predict an assembly interference risk, when the conflict region is not modified within a continuous preset period and passes the assembly interference risk check, triggering cold storage archiving based on Erasure Coding to complete intelligent management of drawings;

[0076] It can be understood that,

[0077] S1, receiving drawings uploaded by a user terminal, extracting features of the uploaded drawings by using a pre-trained CNN-Transformer hybrid model, and outputting three-dimensional classification labels containing an engineering type, a design stage, and a version state, wherein training of the CNN-Transformer hybrid model is as shown in the accompanying drawings, and includes: Figure 2

[0078] S101, collecting at least 500,000 cross-domain drawings to construct a heterogeneous data set, and labeling three-level labels of an engineering type, a design stage, and a version state for each cross-domain drawing;

[0079] S102, setting a loss function, inputting the labeled cross-domain drawings into a CNN architecture and a Transformer architecture respectively, performing attention weighted fusion on high-dimensional feature maps output by the CNN architecture and semantic vectors extracted by the Transformer architecture, dynamically adjusting a fusion weight through a learnable parameter matrix, and obtaining a trained CNN-Transformer hybrid model through iterative training until the loss of the CNN architecture and the Transformer architecture no longer decreases;

[0080] S103, when the misclassification rate of newly added drawings exceeds 15%, starting online fine-tuning feature incremental learning, updating full connection layer parameters, and using a double-channel dynamic fusion architecture and a restricted incremental learning strategy;

[0081] S2, monitoring the three-dimensional classification labels by using an operation log hash synchronizer deployed on an edge node, recording editing operations of a user in real time and generating a SHA-3 hash sequence with a time and space stamp, detecting whether there is concurrent modification of the same region of drawings by multiple users according to the SHA-3 hash sequence with the time and space stamp, if there is, using a conflict positioning algorithm based on a hash value timestamp to obtain a conflict region and generate a conflict difference heat map, extracting image features of the conflict region and converting them into retrieval input features, wherein detecting whether there is concurrent modification of the same region of drawings by multiple users according to the SHA-3 hash sequence with the time and space stamp, if there is, using a conflict positioning algorithm based on a hash value timestamp to obtain a conflict region and generate a conflict difference heat map includes:

[0082] ​The SHA-3 hash sequence with time and space stamps is used to detect whether there is multi-user concurrent modification of the same area of the drawing, and if so, a conflict positioning algorithm based on hash value time stamps is used to obtain the conflict area and generate a conflict difference heat map.

[0083] The SHA-3 hash sequence with time and space stamps is used to detect whether there is multi-user concurrent modification of the same area of the drawing, and if so, a conflict positioning algorithm based on hash value time stamps is used to obtain the conflict area and generate a conflict difference heat map.

[0084] A conflict determination matrix based on overlap rate and operation type is established, with the operation type as the horizontal axis and the influence area overlap rate as the vertical axis, the influence area overlap rate being obtained by the minimum operation unit; when the overlap rate is greater than a preset ratio and the operation type conflicts, it is marked as a high-risk conflict.

[0085] The difference area is rendered using WebGL to achieve transparent gradient overlay display.

[0086] S3, a hybrid retrieval engine is built by pre-trained text conflict detection model, sketch conflict detection model and three-dimensional space relationship conflict detection model.

[0087] S4, the retrieval input features are input into the hybrid retrieval engine, the text semantic is analyzed by the text conflict detection model, the sketch contour features are extracted by the Fourier descriptor of the sketch conflict detection model, and the three-dimensional space relationship is matched by the BIM topological subgraph of the three-dimensional space relationship conflict detection model. The hybrid retrieval engine outputs the conflict area as one or more combinations of text conflict, sketch conflict or three-dimensional space relationship conflict, wherein the text semantic is analyzed by the text conflict detection model, the sketch contour features are extracted by the Fourier descriptor of the sketch conflict detection model, and the three-dimensional space relationship is matched by the BIM topological subgraph of the three-dimensional space relationship conflict detection model.

[0088] The text conflict detection uses a BERT-drawing knowledge graph joint embedding model to map the query statement and the preset drawing metadata to the same vector space, and then performs similarity matching in the same vector space, and selects the drawing corresponding to the drawing metadata with the highest similarity and greater than the preset similarity threshold as the text conflict detection result.

[0089] The sketch conflict retrieval converts the hand-drawn contour into frequency domain features by using an improved Fourier descriptor combined with a dynamic time warping algorithm, calculates the similarity of the frequency domain features and the frequency domain features of the preset sketch, and sorts them, and selects the preset sketch with the highest similarity and greater than the preset similarity threshold as the sketch conflict detection result.

[0090] The three-dimensional space relationship conflict detection analyzes the bounding box hierarchy of the three-dimensional drawing to be queried through point cloud topology subgraph matching, and performs isomorphic comparison with the preset BIM database, selects the three-dimensional drawing model corresponding to the bounding box hierarchy with the highest similarity and greater than the preset similarity threshold as the three-dimensional space relationship conflict detection result, and specifically includes:

[0091] The voxel grid downsampling algorithm is used to normalize the point cloud density to 500 points per cubic meter, and the point cloud preprocessing is completed;

[0092] The Alpha Shape algorithm is used to extract the hole structure features of the model, generate a multi-level bounding box tree, and complete the topology feature extraction;

[0093] The VF2 algorithm is used for subgraph isomorphic matching, and the rotation invariance constraint condition is introduced to complete the subgraph matching optimization;

[0094] S5, the permission allocation verification is performed on the conflict area, the permission change record is stored in the public chain, and the permission change operation log is stored in the private chain, so that the permission allocation is traceable and tamper-proof, and specifically includes:

[0095] Deploying a Chainlink node realizes the data bidirectional verification of the public chain and the private chain, and ensures that the data bidirectional verification verification delay of the public chain and the private chain is within the preset threshold time;

[0096] Define multiple permission templates, the permission template is used to limit editable and non-exportable, and supports one-key application of the template to the project group;

[0097] When signing, 5 nodes are randomly selected to form a temporary ring, and the signature verification time is ensured to be within the preset threshold time;

[0098] S6, the multi-modal query vectorization association reconstruction labeling network is used for the conflict area, the open chain is identified and the missing node is located, the tolerance transfer matrix model is established to predict the assembly interference risk, when the conflict area is not modified for a continuous preset period and passes the assembly interference risk check, trigger the cold storage archiving based on ErasureCoding to complete the intelligent management of drawings, wherein the multi-modal query vectorization association reconstruction labeling network is used for the conflict area, the open chain is identified and the missing node is located, including:

[0099] The size labeling is performed on the conflict area, the size labeling is converted into a weighted directed edge, the node represents the center point of the geometric element, the depth first search is used to traverse the open path, and the missing labeling position is located;

[0100] The missing labeling position is completed according to the adjacent size tolerance;

[0101] The embodiment also provides three missing labeling completion schemes, including:

[0102] According to the upper and lower deviation range of the adjacent size, with the maximum material condition as the constraint, ensure that the assembly gap after the accumulation of all size tolerances still meets the ultimate requirement, and calculate the theoretical upper limit of the closed size according to the maximum positive deviation, inversely deduce the limit tolerance range of the missing size, and complete the missing label position through the limit tolerance range of the missing size;

[0103] Or,

[0104] Adopting the root mean square statistical model, confirming that the size deviation conforms to the normal distribution, calculating the theoretical standard deviation of the closed size by statistically calculating the standard deviation of all sizes in the chain, and finally generating the bilateral symmetric tolerance to complete the missing label position through the bilateral symmetric tolerance;

[0105] Or,

[0106] The total amount of unclosed tolerance is evenly distributed to the related size according to the size sensitivity weight, and is distributed to each ring according to the weight proportion, and the missing label position is completed according to the distribution result.

[0107] Embodiment two

[0108] Figure 3 It is a system schematic diagram of a drawing intelligent management device according to another exemplary embodiment, which comprises:

[0109] The classification label generation module 1 is used for receiving the drawings uploaded by the user terminal, extracting the features of the uploaded drawings by using the pre-trained CNN-Transformer hybrid model, and outputting three-dimensional classification labels containing the engineering type, design stage and version state;

[0110] The conflict detection module 2 is used for monitoring the three-dimensional classification labels by deploying the operation log hash synchronizer on the edge node, recording the editing operation of the user in real time and generating the SHA-3 hash sequence with time and space stamp, detecting whether there is concurrent modification of the same area of the drawing by multiple users according to the SHA-3 hash sequence with time and space stamp, if there is, using the conflict positioning algorithm based on the hash value timestamp to obtain the conflict area and generate the conflict difference heat map, extracting the image features of the conflict area and converting them into retrieval input features;

[0111] The retrieval engine building module 3 is used for building a hybrid retrieval engine by pre-training the text conflict detection model, sketch conflict detection model and three-dimensional spatial relationship conflict detection model;

[0112] The conflict relationship analysis module 4 is configured to input the search input features into the hybrid search engine, analyze the text semantics through the text conflict detection model, extract the sketch contour features through the Fourier descriptor of the sketch conflict detection model, and match the three-dimensional space relationship through the BIM topology subgraph of the three-dimensional space relationship conflict detection model. The hybrid search engine outputs the conflict area as one or a combination of text conflict, sketch conflict, or three-dimensional space relationship conflict.

[0113] The permission lock module 5 is configured to verify the permission allocation of the conflict area, store the permission change record in the public chain, store the permission change operation log in the private chain, and ensure that the permission allocation is traceable and tamper-proof.

[0114] The archive management module 6 is configured to use a multi-modal query vectorization associated reconstruction labeling network to identify an unclosed chain and locate a missing node, establish a tolerance transfer matrix model to predict an assembly interference risk, and trigger cold storage archiving based on ErasureCoding when the conflict area is not modified for a continuous preset period and passes the assembly interference risk check, thereby completing intelligent management of drawings.

[0115] It can be understood that the drawing uploaded by the user terminal is received by the classification label generation module 1, the pre-trained CNN-Transformer hybrid model is used for feature extraction of the uploaded drawing, and three-dimensional classification labels containing engineering type, design stage and version state are output; the conflict detection module 2 uses the operation log hash synchronizer deployed in the edge node to monitor the three-dimensional classification labels, records the user's editing operation in real time and generates a SHA-3 hash sequence with a space-time stamp, detects whether there is a concurrent modification of the same area of the drawing by multiple users according to the SHA-3 hash sequence with a space-time stamp, if there is, uses a conflict positioning algorithm based on hash value timestamp to obtain the conflict area and generate a conflict difference heat map, extracts the image features of the conflict area and converts them into retrieval input features; the retrieval engine building module 3 uses the pre-trained text conflict detection model, sketch conflict detection model and three-dimensional spatial relationship conflict detection model to build a hybrid retrieval engine; the conflict relationship analysis module 4 inputs the retrieval input features into the hybrid retrieval engine, analyzes the text semantics through the text conflict detection model, extracts the sketch contour features through the Fourier descriptor of the sketch conflict detection model, and matches the three-dimensional spatial relationship through the BIM topological subgraph of the three-dimensional spatial relationship conflict detection model. The hybrid retrieval engine outputs the conflict area as one or more combinations of text conflict, sketch conflict or three-dimensional spatial relationship conflict; the permission locking module 5 verifies the permission allocation of the conflict area, stores the permission change record in the public chain, and stores the permission change operation log in the private chain, ensuring that the permission allocation is traceable and tamper-proof; the archiving management module 6 uses a multi-modal query vectorization association reconstruction labeling network to identify unclosed chains and locate missing nodes, establishes a tolerance transfer matrix model to predict assembly interference risks, and when there is no modification in the conflict area for a continuous preset period and the assembly interference risk check is passed, triggers cold storage archiving based on ErasureCoding to complete intelligent management of the drawing;

[0116] The embodiment is implemented by deploying a CNN-Transformer hybrid model processor to perform three-level classification of the drawing, output structured metadata containing feature vectors and permission identifiers; through an edge computing node cluster, real-time capture of user operations and generation of SHA-3 hash chains with time and space stamps; based on the hash branch comparison algorithm, the conflict area is located within 50 ms and the heat map is generated; the integration of the BERT text semantic parser, the Fourier descriptor sketch feature extractor and the BIM topology matcher supports cross-modal joint retrieval; the size chain closed loop verification adopts the graph algorithm to traverse the annotation network, and the specification version adapter dynamically loads the national standard library according to the design stage; the private chain stores operation logs and version snapshots, the public chain records permission change smart contracts and notarization hashes, and the cross-chain oracle realizes bidirectional verification of double-chain data; the NVMe SSD is used to store the active version, supporting μs-level response, and the archived version is implemented by sharding coding, the multi-modal retrieval engine is innovatively designed to support text, sketch and three-dimensional model joint query, the sketch retrieval uses the improved Fourier descriptor algorithm to improve the matching accuracy of the hand-drawn contour and the drawing library, and the dynamic detection algorithm based on operation log hash calculates the operation sequence hash value through distributed node calculation to realize millisecond-level conflict identification.

[0117] In the embodiment, the text conflict retrieval is based on the BERT-drawing knowledge graph joint embedding model, the query text is mapped to the same vector space as the drawing title and the annotated text, and the semantic similarity extended retrieval is supported; the sketch conflict retrieval extracts the frequency domain features of the hand-drawn contour through the Fourier descriptor algorithm, and calculates the contour similarity by using the dynamic time warping algorithm; the voxel grid is down-sampled to standardize the point cloud density value, and the Alpha Shape algorithm is used to extract the hole features, the VF2 subgraph isomorphism algorithm is used to realize the rotation invariance matching, the cold and hot data are stored in layers, and the Erasure Coding coding technology is used to reduce the storage cost of the abandoned drawing, while greatly improving the data recoverability, the double-chain blockchain architecture is innovatively designed, the operation log is stored in the private chain to ensure efficient query, the permission change record is stored in the public chain to realize tamper-proofing, which protects user privacy and meets the regulatory traceability requirements, and reduces the computing overhead by 30% compared with the traditional digital signature scheme.

[0118] The double-chain blockchain architecture is innovatively designed, the operation log is stored in the private chain to ensure efficient query, the permission change record is stored in the public chain to realize tamper-proofing, which protects user privacy and meets the regulatory traceability requirements, and reduces the computing overhead by 30% compared with the traditional digital signature scheme.

[0119] Embodiment three:

[0120] The embodiment provides a storage medium, the storage medium stores a computer program, when the computer program is executed by a host, each step in the above method is realized.

[0121] It can be understood that the storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0122] It can be understood that the same or similar parts in the above embodiments can be mutually referred to, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0123] It should be noted that, in the description of the present application, the terms "first", "second", etc. are only for the purpose of description, and cannot be understood as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise specified, the meaning of "a plurality of" is at least two.

[0124] Any process or method descriptions in flow charts or described herein otherwise can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for performing specific logic functions or steps in the process, and that the various embodiments of the application include the additional implementation that the functions can be performed in different orders, in different ways, or in an essentially simultaneous manner, as can be useful in the implementation depending on the functionality involved, as would be understood by those skilled in the art.

[0125] It should be understood that the parts of the present application can be realized by hardware, software, firmware or their combination. In the above embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or their combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA), etc.

[0126] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by program instructions to the relevant hardware, and the program can be stored in a computer readable storage medium, which includes one or a combination of steps of the method embodiments when executed.

[0127] In addition, each function unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware, or in the form of a software function module. When the integrated module is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0128] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0129] In the description of the present specification, the description referring to the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0130] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for intelligent management of drawings, characterized in that, The method includes: The system receives drawings uploaded by user terminals, uses a pre-trained CNN-Transformer hybrid model to extract features from the uploaded drawings, and outputs three-dimensional classification labels that include project type, design stage, and version status. The operation log hash synchronizer deployed on the edge node monitors the three-dimensional classification label, records the user's editing operations in real time and generates a SHA-3 hash sequence with a time stamp. Based on the SHA-3 hash sequence with the time stamp, it detects whether there are multiple users concurrently modifying the same area of ​​the drawing. If so, a conflict location algorithm based on hash value timestamp is used to obtain the conflict area and generate a conflict difference heat map. The image features of the conflict area are extracted and converted into retrieval input features. A hybrid retrieval engine is built using pre-trained text conflict detection models, sketch conflict detection models, and 3D spatial relationship conflict detection models. The search input features are input into the hybrid search engine. The text semantics are parsed by the text conflict detection model, the sketch outline features are extracted by the Fourier descriptor of the sketch conflict detection model, and the three-dimensional spatial relationships are matched by the BIM topology subgraph of the three-dimensional spatial relationship conflict detection model. The hybrid search engine outputs that the conflict area is one or more combinations of text conflict, sketch conflict, or three-dimensional spatial relationship conflict. The permission allocation verification is performed on the conflict area, the permission change record is stored on the public chain, and the permission change operation log is stored on the private chain to ensure that the permission allocation is traceable and tamper-proof. The conflict area is reconstructed using a multimodal query vectorized association to identify unclosed chains and locate missing nodes. A tolerance transfer matrix model is established to predict assembly interference risks. When the conflict area remains unmodified for a preset period and passes the assembly interference risk check, cold storage archiving based on ErasureCoding is triggered to complete intelligent drawing management.

2. The method according to claim 1, characterized in that, The training of the CNN-Transformer hybrid model includes: Collect at least N cross-domain drawings to construct a heterogeneous dataset, and label each cross-domain drawing with three-level labels: project type, design stage, and version status; Set a loss function, input the labeled cross-domain map into the CNN architecture and the Transformer architecture respectively, perform attention-weighted fusion of the high-dimensional feature map output by the CNN architecture and the semantic vector extracted by the Transformer architecture, and dynamically adjust the fusion weight through a learnable parameter matrix. Iterate the training until the loss of the CNN architecture and the Transformer architecture no longer decreases, and obtain the trained CNN-Transformer hybrid model. When the classification error rate of newly added drawings exceeds the preset threshold, online fine-tuning of feature incremental learning is initiated, the fully connected layer parameters of the CNN-Transformer hybrid model are updated, and a dual-channel dynamic fusion architecture and a restricted incremental learning strategy are used.

3. The method according to claim 2, characterized in that, The step of detecting whether multiple users concurrently modify the same area of ​​the drawing based on the SHA-3 hash sequence with spatiotemporal stamp, and if so, using a conflict location algorithm based on hash value timestamps to obtain the conflict area and generate a conflict difference heatmap includes: The user editing operation is decomposed into the smallest operation unit based on the SHA-3 hash sequence with spacetime stamp, and each unit is accompanied by a location hash value; A conflict determination matrix based on overlap rate and operation type is established, with the horizontal axis representing the operation type and the vertical axis representing the overlap rate of the affected area. The overlap rate of the affected area is obtained through the smallest operation unit. When the overlap rate is greater than a preset ratio and the operation types conflict, it is marked as a high-risk conflict. WebGL is used to accelerate the rendering of different areas, enabling the display of transparency gradient overlay.

4. The method according to claim 3, characterized in that, The steps of parsing text semantics using the text conflict detection model, extracting sketch outline features using the Fourier descriptor of the sketch conflict detection model, and matching three-dimensional spatial relationships using the BIM topology subgraph of the three-dimensional spatial relationship conflict detection model include: Text conflict detection uses a BERT-drawing knowledge graph joint embedding model to map the query statement and the preset drawing metadata to the same vector space, and then performs similarity matching in the same vector space. The drawing corresponding to the drawing metadata with the highest similarity and greater than the preset similarity threshold is selected as the text conflict detection result. Sketch conflict retrieval uses an improved Fourier descriptor combined with a dynamic time warping algorithm to convert hand-drawn outlines into frequency domain features. The similarity between the frequency domain features and the frequency domain features of preset sketches is calculated and sorted. The preset sketch with the highest similarity and greater than the preset similarity threshold is selected as the sketch conflict detection result. The 3D spatial relationship conflict detection analyzes the bounding box hierarchy of the 3D drawing to be queried by point cloud topology subgraph matching, and performs isomorphic comparison with the preset BIM database. The 3D drawing model corresponding to the bounding box hierarchy with the highest similarity and greater than the preset similarity threshold is selected as the 3D spatial relationship conflict detection result.

5. The method according to claim 4, characterized in that, The method of parsing the bounding box hierarchy of the queried 3D drawing through point cloud topology subgraph matching includes: The point cloud density of the 3D drawing to be queried is standardized to 500 points / cubic meter using a voxel grid downsampling algorithm to complete the point cloud preprocessing; The Alpha Shape algorithm is used to extract the hole structure features of the 3D drawing to be queried, and a multi-level bounding box tree is generated to complete the topological feature extraction. The isomorphic comparison with the preset BIM database includes: The VF2 algorithm is used for subgraph isomorphic matching, and rotation invariance constraints are introduced to optimize subgraph matching.

6. The method according to claim 5, characterized in that, The permission allocation verification for the conflict area includes: Deploy Chainlink nodes to achieve two-way data verification between the public and private chains, ensuring that the verification delay between the public and private chains is within a preset threshold time. Define multiple permission templates, which are used to restrict permissions to editable but not exportable, and support one-click application of the templates to the project group; When a user performs a signature operation, five nodes are randomly selected to form a temporary ring, and the signature verification time is ensured to be within a preset threshold time.

7. The method according to claim 6, characterized in that, The step of using multimodal query vectorization to reconstruct the annotation network for the conflict region, identifying unclosed chains and locating missing nodes includes: The conflict area is dimensioned and converted into weighted directed edges. Nodes represent the center points of geometric elements. Depth-first search is used to traverse unclosed paths to locate the missing dimensioning positions. Missing annotations are filled in based on adjacent dimensional tolerances.

8. The method according to claim 7, characterized in that, The step of filling in missing annotation positions based on adjacent dimensional tolerances includes: Based on the upper and lower deviation ranges of adjacent dimensions, and constrained by the maximum material conditions, ensure that the assembly clearance after the accumulation of all dimensional tolerances still meets the limit requirements. Calculate the theoretical upper limit of the theoretical closed dimension based on the maximum positive deviation, and deduce the limit tolerance range of the missing dimension in reverse. Then, fill in the missing marking position using the limit tolerance range of the missing dimension. or, The root mean square statistical model is used to confirm that the dimensional deviation conforms to the normal distribution. The standard deviation of all dimensions in the chain is statistically analyzed, the theoretical standard deviation of the closed dimension is calculated, and finally, a bilateral symmetric tolerance is generated. The missing annotation positions are filled by the bilateral symmetric tolerance. or, The total amount of unclosed tolerances is evenly distributed to the relevant dimensions according to the size sensitivity weight, and then distributed to each ring according to the weight ratio. The missing annotation positions are filled in according to the distribution results.

9. A drawing intelligent management device, characterized in that, The device includes: Classification label generation module: It is used to receive drawings uploaded by user terminals, use a pre-trained CNN-Transformer hybrid model to extract features from the uploaded drawings, and output three-dimensional classification labels containing project type, design stage and version status. The conflict detection module is used to monitor the 3D classification labels through the operation log hash synchronizer deployed on the edge node, record the user's editing operations in real time and generate a SHA-3 hash sequence with a time stamp. Based on the SHA-3 hash sequence with the time stamp, it detects whether there are multiple users concurrently modifying the same area of ​​the drawing. If so, it uses a conflict location algorithm based on hash value timestamps to obtain the conflict area and generate a conflict difference heat map, extracts the image features of the conflict area and converts them into retrieval input features. Search engine building module: used to build a hybrid search engine using pre-trained text conflict detection model, sketch conflict detection model and 3D spatial relationship conflict detection model; Conflict Relationship Analysis Module: Used to input the retrieval input features into the hybrid retrieval engine, parse the text semantics through the text conflict detection model, extract the sketch outline features through the Fourier descriptor of the sketch conflict detection model, match the three-dimensional spatial relationship through the BIM topology subgraph of the three-dimensional spatial relationship conflict detection model, and the hybrid retrieval engine outputs that the conflict area is one or more combinations of text conflict, sketch conflict or three-dimensional spatial relationship conflict. The permission locking module is used to verify the permission allocation of the conflict area, store the permission change record on the public chain, and store the permission change operation log on the private chain to ensure that the permission allocation is traceable and tamper-proof. The archive management module is used to reconstruct the annotation network of the conflict area using multimodal query vectorization association, identify unclosed chains and locate missing nodes, establish a tolerance transfer matrix model to predict assembly interference risks, and trigger cold storage archiving based on ErasureCoding to complete intelligent management of drawings when the conflict area has no modifications within a consecutive preset period and passes the assembly interference risk verification.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by the main controller, implements each step of the intelligent drawing management method as described in any one of claims 1-8.

Images