A building construction plan intelligent audit system and audit method

Abstract

The present invention discloses an intelligent audit system for construction plans and an audit method thereof, including: in the submission interface of the same construction plan, an independent upload port is set for each classification; all types of files in the same construction plan are parsed according to a unified parsing format; industry experts define the audit rules and vector similarity comparison audit of each type of plan; establish an audit rule synonym platform; and set up a database. Audit method, S1: classify and upload the uploaded construction technical files; S2: parse all types of files in the same construction plan according to a unified parsing format; S3: import each content in the parsed title, paragraph, picture and table sequence into a rule audit engine; S4: set up an audit rule synonym platform on an interactive terminal; S5: set up a database. The present invention has the advantages of high accuracy of professional terms, standardized audit rules, and high similarity between database stored documents and audit texts.

Description

Technical Field

[0001] The present invention relates to the technical field of building construction plan review, and in particular to an intelligent building construction plan review system and a review method thereof. Background Art

[0002] The construction plan is the implementation plan of the project and is a document material that is written in large quantities by construction company technicians on a daily basis. It includes organizational plan, personnel composition plan, technical plan, safety plan, material supply plan, calculation book, etc. Since the construction plan directly guides the construction, its integrity and accuracy are related to the safety and quality of the project. Usually, after the construction plan is compiled by on-site technicians, it is submitted to the technical management line for proofreading and review. The review work mainly includes verifying the integrity of the implementation plan based on key information, verifying the accuracy of the calculation formula based on key data information, and comparing the key elements of the plan to improve the content of the plan. In the traditional review process, the workload of technicians is large, the repetition is high, and the efficiency is low. Therefore, it is necessary to improve the review speed and review quality, and apply the intelligent system management method to the review of the construction plan to improve the review speed and review quality. However, the software system review has problems such as poor professional term recognition, non-standard review rules, and large deviation between database storage documents and review texts. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, one purpose of the present invention is to propose an intelligent review system for building construction plans and a review method thereof, which solves the problems of poor recognition of professional terms, non-standard review rules, and large deviation between database stored documents and review texts in software system review.

[0004] An intelligent review system for building construction plans proposed in the present invention includes:

[0005] 1. File upload end

[0006] Categorize the uploaded files according to their type, and set up an independent upload port for each category in the same construction plan submission interface;

[0007] The review system will identify and review the various types of construction plans uploaded. If the plan fails the review, it cannot be uploaded and the plan will need to be resubmitted after modification.

[0008] The file types include reinforcement scheme, concrete scheme, high formwork scheme and its calculation sheet, scaffolding scheme and its calculation sheet, formwork engineering scheme and secondary structure scheme;

[0009] 2. Parsing of uploaded files

[0010] All types of files in the same building construction plan are parsed according to a unified parsing format;

[0011] Identify the scheme types of all uploaded files and split the files into title sequence, paragraph sequence, image sequence and table sequence;

[0012] For the identification of title content in the document, a rule list compiled based on regular expressions is used to identify all title content in the document to form a content sequence;

[0013] For paragraph content recognition, a deep learning framework based on the fusion of BiLSTM, CRF and Transformer is used to extract the specified paragraph content from the text data to form a paragraph sequence;

[0014] Identify and parse the image and table contents. Each image is parsed into a JSON field in a specified format based on its position number in the document, its content description, and its category. Tables are parsed into a JSON field in a specified format based on their position number in the document and the corresponding contents of rows and columns.

[0015] 3. Rule Audit Engine

[0016] Industry experts define the audit rules for each solution type, and industry expert R&D personnel build an extensible audit rule corpus based on regular expressions, and the rule base builds a rule audit model;

[0017] By summarizing and collating the manual review opinions of various types of construction plans, we screened out the high-frequency review opinion tables for each type of plan, and then combined them with the compilation specification requirements of each type of construction plan. After discussion and evaluation by industry experts, we produced a list of review rule requirements for each type of construction plan.

[0018] Vector similarity comparison and review, collect and organize architectural professional vocabulary, build word vector pre-training model, and build vector similarity calculation and comparison model;

[0019] Conduct rule review on the structured data to be reviewed, output the review results, select the items that have passed and those that have not passed the rule review, conduct semantic comparison review on the items that have not passed the rule review, set a sliding window, and compare them with the first N characters of each paragraph in the paragraph sequence. If the maximum value of the semantic similarity calculation result is above the threshold, the review is passed, otherwise it fails, and the review results are output;

[0020] The final audit results are returned to the interactive terminal and an audit report is generated;

[0021] 4. Audit Rules Synonym Platform

[0022] The interactive terminal sets up a synonym platform for audit rules, and professionals can add synonym entries to the same audit rule to expand the scope of the audit rule;

[0023] Users can use the audit rule synonym platform to expand the synonym rules for a certain audit type in the audit report;

[0024] 5. Set up the database

[0025] The construction technical documents that have passed the review will be uploaded and saved in the construction technology resource library.

[0026] In some embodiments of the present invention, the image categories set in the parsing of the uploaded file include architectural construction drawings, structural construction drawings, real-life pictures, renderings, and organizational charts.

[0027] In some other embodiments of the present invention, in the parsing of the uploaded file: a target detection model is used to perform position detection and type classification on the target content, and the target detection model adopts a YOLOv8 model; an optical character recognition model is used to recognize the table content, and the optical character recognition model adopts an Abinet model; an image classification model is used to recognize the image type, and the image classification model adopts a Resnet18 model.

[0028] In other embodiments of the present invention, when each content in the title, paragraph, picture and table sequence is imported into the rule review engine, it will be structured with JSON field information in a specified format. The JSON structured data records key information unique to each content, including the key value of the review rule to which the review content belongs, wherein the review rules correspond one-to-one to the review rule key values.

[0029] In other embodiments of the present invention, the vector similarity comparison and review engine is established: 1. 100 cases of steel bar schemes, concrete schemes, high-support formwork schemes, formwork engineering schemes, scaffolding schemes and secondary structure schemes are collected, and the text content is retained as a word vector pre-training data set; 2. Industry experts provide a professional vocabulary in the field of construction, and perform word segmentation on the text content of the pre-training data set; 3. Use a universal stop word list to remove stop words from the pre-training data set after word segmentation; 4. Input the pre-training data set into a word vector pre-training model based on Word2Vec to output a word vector embedding representation result; 5. Construct a sentence vector calculation mode, after the sentence is segmented and stop words are removed, the word vectors of each word in the sentence are summed, and then the ratio is calculated with the total number of words in the sentence, and the result is recorded as the representation vector of this sentence; 6. The vector representation results of the two sentences are subjected to cosine similarity calculation, and this calculation result is the vector similarity comparison result.

[0030] In other embodiments of the present invention, a security database is set up to compare the materials used in documents of multiple file types and the component parameters corresponding to each material with the threshold values ​​of the component parameters for each material in the security database, and warning marks will be given for material component parameters that exceed the threshold values ​​in the security database or will be deemed as invalid designs.

[0031] An intelligent review method for a building construction plan, the specific review steps are as follows:

[0032] S1: Classify the uploaded construction technical documents. On the same construction plan submission interface, find the corresponding submission port for each type of construction plan and submit it. The review system will identify and review the uploaded construction plans of various types separately. If the plan fails the review, it cannot be uploaded and needs to be resubmitted after modification.

[0033] S2: Parse all types of files in the same building construction plan according to a unified parsing format, identify the plan types of all types of uploaded files, and split the files according to title sequence, paragraph sequence, picture sequence, and table sequence;

[0034] S3: Import each piece of content in the parsed title, paragraph, picture and table sequence into the rule review engine, where industry experts define the review rules for each solution type, industry expert R&D personnel build an extensible review rule corpus based on regular expressions, and the rule library builds a rule review model;

[0035] Conduct rule review on the structured data to be reviewed, output the review results, select the items that have passed and those that have not passed the rule review, conduct semantic comparison review on the items that have not passed the rule review, set a sliding window, and compare them with the first N characters of each paragraph in the paragraph sequence. If the maximum value of the semantic similarity calculation result is above the threshold, the review is passed, otherwise it fails, and the review results are output;

[0036] The final audit results are returned to the interactive terminal and an audit report is generated;

[0037] S4: Set up an audit rule synonym platform on the interactive terminal. Professionals can add synonym entries to the same audit rule to expand the scope of the audit rule. The audit rule synonym platform can be used to expand the synonym rules for a certain audit file type in the audit report.

[0038] S5: Set up a database, and the construction technical documents that have passed the review will be uploaded and saved in the construction technical resource library.

[0039] In other embodiments of the present invention, a security database is set up in step S3, and the materials used in documents of multiple file types and the component parameters corresponding to each material are compared with the threshold values ​​of the component parameters for each material in the security database. For material component parameters that exceed the threshold values ​​in the security database, a warning mark will be given or the design will be deemed invalid, and a report of failed audit will be output.

[0040] In other embodiments of the present invention, in the dual-channel audit engine based on rule matching and vector comparison in step S3, the audit rules corresponding to the key element content that failed the audit through rule matching will automatically enter the vector comparison audit engine to traverse and search whether there is content similar to the audit rule. If the maximum value of the similarity calculation result is greater than the threshold, the audit is passed.

[0041] In other embodiments of the present invention, in step S5, the construction technology resource library stores all construction technology documents that have passed the review, which can be used as text corpus content in the construction field and provide training data for natural language processing related tasks.

[0042] The beneficial effects of the present invention are as follows: documents in a building construction plan are classified, different types of file types are parsed using a unified parsing format, the parsed text can be audited using a unified rule audit engine, the audit uses audit rules and vector similarity comparison audits defined by industry experts for each type of plan, and can be audited quickly and accurately. After the quick audit, the audit results are output, and the construction technical files audited each time are uploaded and saved in the construction technology resource library to increase the use of training data for natural language processing related tasks in the database for the next audit. The professional terms are highly accurate, the audit rules are standardized, and the similarity between the database stored documents and the audit text is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0044] Figure 1 This is a schematic diagram of the principles of an intelligent review system for building construction plans proposed by the present invention.

[0045] Figure 2 The present invention is a flowchart of an intelligent review method for a building construction plan proposed by the present invention. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0047] The invention proposes an intelligent review system for building construction plans, also known as an intelligent review robot for building construction plans.

[0048] 1. The intelligent audit device obtains the construction technical documents uploaded to the terminal;

[0049] The construction technical documents here refer to construction plan documents, including reinforcement plans, concrete (general type, large volume type) plans, high formwork plans and their calculations, scaffolding (general type, cantilever type, floor-standing type, flower basket type, attached type) plans and their calculations, formwork engineering plans and secondary structure plans, etc., collectively referred to as construction technical documents.

[0050] The intelligent review device parses the uploaded technical documents and performs key element recognition operations. The element recognition engine first identifies the scheme type of the uploaded file (rebar scheme, concrete scheme, etc.), and then splits the elements in the document into: title sequence, paragraph sequence, image sequence and table sequence.

[0051] The key element recognition method here is implemented using a fusion recognition model based on deep learning and rule matching. Specifically, for the recognition of title content in a document, a rule list compiled based on regular expressions is used to recognize all title content in the document to form a content sequence; for paragraph content recognition, a deep learning framework based on the fusion of BiLSTM, CRF and Transformer is used to extract specified paragraph content from text data to form a paragraph sequence.

[0052] The multimodal large model module identifies and parses the image content and table content in the construction plan document, and returns JSON structured data: each image is parsed into a JSON field of a specified format (the position number in the document, the image content description, and the category location of the image); the table is parsed into a JSON field of a specified format (the position number in the document and the corresponding content of the row and column). The image categories are divided into ① architectural construction drawings, ② structural construction drawings, ③ real-life pictures, ④ renderings, ⑤ organizational charts, etc.

[0053] Specifically, the multimodal large model used here mainly completes tasks such as image content recognition, image classification, complex table content alignment and content recognition in construction plan documents. The large model converts the documents to be recognized into images.

[0054] The target detection model is used to detect the position and classify the type of the target content. The target detection model here adopts the YOLOv8 model; the optical character recognition model is used to recognize the table content. The optical character recognition model here adopts the Abinet model; the image classification model is used to recognize the image type. The image classification model here adopts the Resnet18 model.

[0055] The structured data of the identified title sequence, paragraph sequence, image sequence, and table sequence are input into the dual audit engine of vector comparison and rule matching. Specifically, each content in the title, paragraph, image, and table sequence imported into the audit engine is structured with JSON field information in a specified format. The JSON structured data records the key information unique to each content, including the key value of the audit rule to which the content belongs. The audit rules correspond to the audit rule key values ​​one by one, including the category to which the content belongs and the position of the content in the document.

[0056] The rule audit engine is defined by industry experts for audit rules of various scheme types. R&D personnel build an extensible audit rule corpus based on regular expressions, and the rule audit model is built by the rule base. The audit rules here specifically refer to the audit rules of steel bar schemes, concrete (general type, large volume type) schemes, high-support formwork schemes and their calculation books, formwork engineering, scaffolding (general type, cantilever type, flower basket type, attached type, ground type) schemes and their calculation books, and secondary structure schemes. By summarizing and collating the historical manual audit opinions of various types of construction schemes, the high-frequency audit opinion tables of various scheme types are screened out, and then combined with the compilation specification requirements of each type of construction scheme, after discussion and evaluation by industry experts, a list of audit rule requirements for each type of construction scheme is produced. The list is as follows: complete the audit of completeness and sequential arrangement; audit the completeness and accuracy of key elements (tables, pictures, key expression texts); audit the completeness, compliance, and timeliness of the compilation basis; audit the consistency of context content; ⑤ Audit the accuracy and standardization of key parameters and results of the calculation book.

[0057] Vector similarity comparison and review engine, collects and organizes architectural professional vocabulary, builds word vector (based on Word2Vec) pre-training model, and builds vector similarity calculation and comparison model;

[0058] Establishment of the vector similarity comparison and review engine: ① Collect 100 (at least 100) construction plans for steel bar plan, concrete plan, high-support formwork plan, formwork engineering plan, scaffolding plan and secondary structure plan, totaling 600 construction plans, and retain the text content as a word vector pre-training data set; ② Industry experts provide a professional vocabulary in the construction field, and perform word segmentation on the text content of the pre-training data set; ③ Use a universal stop word list to remove stop words from the pre-training data set after word segmentation; ④ Input the pre-training data set into the word vector pre-training model based on Word2Vec to output the word vector embedding representation result; ⑤ Construct a sentence vector calculation mode, after the sentence is segmented and stop words are removed, the word vector of each word in the sentence is summed, and then the ratio is calculated with the total number of words in the sentence, and the result is recorded as the representation vector of this sentence; ⑥ Perform cosine similarity calculation on the vector representation results of the two sentences, and this calculation result is the vector similarity comparison result.

[0059] Conduct rule review on the structured data to be reviewed, output the review results, and select the items that have passed and those that have not passed the rule review; conduct semantic comparison review on the items that have not passed the rule review, set a sliding window of N=50, and compare them with the first N characters of each paragraph in the paragraph sequence respectively. If the maximum value of the semantic similarity calculation result is above the threshold, the review is passed, otherwise it fails, and the review result is output;

[0060] In the dual-channel audit engine based on rule matching and vector comparison, the key elements that have not passed the audit through rule matching will automatically enter the corresponding audit rules into the vector comparison audit engine. The above audit method will traverse and find whether there is content similar to the audit rule. If the maximum value of the similarity calculation result is greater than 0.65, the audit is passed. The dual-channel audit method can effectively improve the accuracy and fault tolerance of the audit.

[0061] The final audit results are returned to the interactive terminal and an audit report is generated, which can be read and downloaded online. For items that fail, standard examples can be viewed online.

[0062] The audit history tasks are saved in the task management list of the interactive terminal;

[0063] The interactive terminal sets up a synonym platform for audit rules, and professionals can add synonym entries to the same audit rule to expand the scope of the audit rule;

[0064] Users can use the audit rule synonym platform to expand the synonymous rules for a certain audit type (audit point) in the audit report. For example, ① the user selects the audit point entry with the audit type name "Emergency Disposal Measures", and the rule library returns all the audit rule content entries corresponding to the audit type point; ② the user selects the corresponding content of the audit rule entry to be expanded, such as selecting the original audit rule "Emergency Rescue Measures" and then adding the synonymous entry "Rescue Hospital Information", then a new synonymous audit rule will be added to this rule.

[0065] The construction technology documents that have passed the review will be uploaded and saved in the construction technology resource library. The construction technology resource library saves all the construction technology documents that have passed the review and can be used as high-quality text corpus content in the construction field and provide training data for natural language processing related tasks.

[0066] Collect the materials and properties used in construction, and create a safety database for the performance and parameters of building components made of each material. During each audit, the components and component parameters in the text must be compared and analyzed with the safety database. If the comparison and analysis fails the audit, it will be marked and displayed as failed.

[0067] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent review system for building construction plans, characterized by: include:

1. File upload end Categorize the uploaded files according to their type, and set up an independent upload port for each category in the same construction plan submission interface; The review system will identify and review the various types of construction plans uploaded. If the plan fails the review, it cannot be uploaded and the plan will need to be resubmitted after modification. The file types include reinforcement scheme, concrete scheme, high formwork scheme and its calculation sheet, scaffolding scheme and its calculation sheet, formwork engineering scheme and secondary structure scheme; 2. Parsing of uploaded files All types of files in the same building construction plan are parsed according to a unified parsing format; Identify the scheme types of all uploaded files and split the files into title sequence, paragraph sequence, image sequence and table sequence; For the identification of title content in the document, a rule list compiled based on regular expressions is used to identify all title content in the document to form a content sequence; For paragraph content recognition, a deep learning framework based on the fusion of BiLSTM, CRF and Transformer is used to extract the specified paragraph content from the text data to form a paragraph sequence; Identify and parse the image and table contents. Each image is parsed into a JSON field in a specified format based on its position number in the document, its content description, and its category. Tables are parsed into a JSON field in a specified format based on their position number in the document and the corresponding contents of rows and columns.

3. Rule Audit Engine Industry experts define the audit rules for each solution type, and industry expert R&D personnel build an extensible audit rule corpus based on regular expressions, and the rule base builds a rule audit model; By summarizing and collating the manual review opinions of various types of construction plans, we screened out the high-frequency review opinion tables for each type of plan, and then combined them with the compilation specification requirements of each type of construction plan. After discussion and evaluation by industry experts, we produced a list of review rule requirements for each type of construction plan. Vector similarity comparison and review, collect and organize architectural professional vocabulary, build word vector pre-training model, and build vector similarity calculation and comparison model; Conduct rule review on the structured data to be reviewed, output the review results, select the items that have passed and those that have not passed the rule review, conduct semantic comparison review on the items that have not passed the rule review, set a sliding window, and compare them with the first N characters of each paragraph in the paragraph sequence. If the maximum value of the semantic similarity calculation result is above the threshold, the review is passed, otherwise it fails, and the review results are output; The final audit results are returned to the interactive terminal and an audit report is generated; When each content in the title, paragraph, picture and table sequence is imported into the rule review engine, it will be structured with JSON field information in a specified format. The JSON structured data records key information specific to each content, including the key value of the review rule to which the review content belongs, wherein the review rule corresponds to the review rule key value one by one; The establishment of the vector similarity comparison and review engine:

1. Collect 100 cases of steel bar scheme, concrete scheme, high-support formwork scheme, formwork engineering scheme, scaffolding scheme and secondary structure scheme, and retain the text content as a word vector pre-training data set; 2. Industry experts provide a professional vocabulary in the field of construction, and perform word segmentation on the text content of the pre-training data set; 3. Use a universal stop word list to remove stop words from the pre-training data set after word segmentation; 4. Input the pre-training data set into a word vector pre-training model based on Word2Vec to output a word vector embedding representation result; 5. Construct a sentence vector calculation mode, after the sentence is segmented and stop words are removed, sum the word vectors of each word in the sentence, and then calculate the ratio with the total number of words in the sentence, and the result is recorded as the representation vector of this sentence; 6. Perform cosine similarity calculation on the vector representation results of the two sentences, and this calculation result is the vector similarity comparison result; 4. Audit Rules Synonym Platform The interactive terminal sets up a synonym platform for audit rules, and professionals can add synonym entries to the same audit rule to expand the scope of the audit rule; Users can use the audit rule synonym platform to expand the synonym rules for a certain audit type in the audit report; 5. Set up the database The construction technical documents that have passed the review will be uploaded and saved in the construction technology resource library.

2. According to claim 1, the intelligent review system for building construction plans is characterized by: The image categories set in the analysis of the uploaded file include architectural construction drawings, structural construction drawings, real-life pictures, renderings, and organizational charts.

3. The intelligent review system for building construction plans according to claim 1 is characterized by: In the parsing of the uploaded file: The target detection model is used to detect the location and classify the type of the target content. The target detection model adopts the YOLOv8 model; The optical character recognition model is used to recognize the table content, and the optical character recognition model adopts the Abinet model; An image classification model is used to identify image types, and the image classification model uses the Resnet18 model.

4. An intelligent review system for building construction plans according to any one of claims 1 to 3, characterized in that: Set up a safety database to compare the materials used in documents of various file types and the corresponding manufacturing component parameters of each material with the threshold values ​​of the manufacturing component parameters of each material in the safety database. For those materials that exceed the manufacturing component parameter thresholds in the safety database, warning marks will be given or the designs will be deemed invalid.

5. A method for intelligent review of construction plans, characterized in that: The specific review steps are as follows: S1: Classify the uploaded construction technical documents. On the same construction plan submission interface, find the corresponding submission port for each type of construction plan and submit it. The review system will identify and review the uploaded construction plans of various types separately. If the plan fails the review, it cannot be uploaded and needs to be resubmitted after modification. S2: Parse all types of files in the same building construction plan according to a unified parsing format, identify the plan types of all types of uploaded files, and split the files according to title sequence, paragraph sequence, picture sequence, and table sequence; S3: Import each piece of content in the parsed title, paragraph, picture and table sequence into the rule review engine, where industry experts define the review rules for each solution type, industry expert R&D personnel build an extensible review rule corpus based on regular expressions, and the rule library builds a rule review model; Conduct rule review on the structured data to be reviewed, output the review results, select the items that have passed and those that have not passed the rule review, conduct semantic comparison review on the items that have not passed the rule review, set a sliding window, and compare them with the first N characters of each paragraph in the paragraph sequence. If the maximum value of the semantic similarity calculation result is above the threshold, the review is passed, otherwise it fails, and the review results are output; The final audit results are returned to the interactive terminal and an audit report is generated; Set up a security database, compare the materials used in documents of various file types and the corresponding manufacturing component parameters of each material with the thresholds of the manufacturing component parameters of each material in the security database, and issue warning marks or treat them as invalid designs if the manufacturing component parameter thresholds of the materials in the security database are exceeded, and output a failed audit report; In the dual-channel audit engine based on rule matching and vector comparison, the key elements that have not passed the audit through rule matching will automatically enter the corresponding audit rules into the vector comparison audit engine to traverse and search for content similar to the audit rules. If the maximum value of the similarity calculation result is greater than the threshold, the audit is passed; S4: Set up an audit rule synonym platform on the interactive terminal. Professionals can add synonym entries to the same audit rule to expand the scope of the audit rule. The audit rule synonym platform can be used to expand the synonym rules for a certain audit file type in the audit report. S5: Set up a database, and the construction technical documents that have passed the review will be uploaded and saved in the construction technical resource library.

6. The intelligent review method for a construction plan according to claim 5 is characterized by: In step S5, the construction technology resource library stores all construction technology documents that have passed the review, which can be used as text corpus content in the construction field and provide training data for natural language processing related tasks.

Images