An artificial intelligence-based steel bridge welding process evaluation test method and system

Abstract

This invention discloses an artificial intelligence-based testing method and system for steel bridge welding process evaluation, involving the intersection of bridge manufacturing and artificial intelligence. It solves the problems of low efficiency and low utilization of historical data in traditional evaluation processes. The method includes steps such as intelligent extraction and verification of project information, recommendation of process parameters, planning of evaluation projects, substitution retrieval, automatic generation of schemes and reports, and collection and storage of test data and reports. It constructs a four-stage closed-loop architecture of "cognition-decision-verification-evolution." The system integrates modules such as multimodal analysis, intelligent recommendation, and substitution retrieval, as well as a closed-loop knowledge evolution mechanism. It combines computer vision, knowledge graph, and other technologies to achieve full-process intelligence, realizing traceable automation from design drawing analysis to evaluation report generation. Simultaneously, it completes knowledge accumulation and dynamic evolution, improving the efficiency, accuracy, and standardization of evaluation work, avoiding repeated experiments, and reducing evaluation preparation time by more than 70%.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of bridge manufacturing, welding engineering, and artificial intelligence, specifically to an artificial intelligence-based test method and system for evaluating the welding process of steel bridges. Background Technology

[0002] As a vital transportation infrastructure, the structural safety and durability of steel bridges are of paramount importance. Welding is a core process in steel bridge manufacturing, and the quality of welded joints directly determines the overall performance of the bridge. To ensure welding quality, welding procedure qualification tests must be conducted before welding to verify the correctness of the proposed welding procedure specifications.

[0003] The existing steel bridge welding procedure qualification process still relies heavily on manual operation, resulting in serious efficiency and quality bottlenecks, specifically in the following aspects: I. Low efficiency and long cycle: According to the "2024 China Steel Bridge Manufacturing White Paper", the welding process qualification of a single large steel bridge project takes an average of 25-30 working days. About 70% of the time is spent manually extracting drawing information, compiling weld statistics tables and writing evaluation plans, and the error rate of manual processing is as high as 12-15%.

[0004] Second, the evaluation scheme is highly dependent on experience and lacks consistency: the development of the evaluation scheme relies heavily on the engineer's personal experience and understanding of the specifications. Different engineers may have different evaluation schemes for the same project, making it difficult to guarantee the consistency and reliability of the evaluation results.

[0005] Third, low utilization rate of historical data: Enterprises have accumulated a large number of historical welding process qualification reports, test data and engineering data, but these valuable data are mostly in the form of paper or unstructured electronic documents, forming "data silos". When conducting new project qualification, it is difficult to quickly and accurately retrieve and reuse existing qualification results, resulting in a large number of repetitive tests and wasting resources.

[0006] Fourth, insufficient level of intelligence: Although some artificial intelligence models are used in the existing technology for optimizing welding parameters or detecting weld defects, they only focus on the application of intelligent tools in a certain link and have not solved the problems of knowledge reuse in pre-process evaluation and intelligent planning and decision-making throughout the entire process. Summary of the Invention

[0007] To address the shortcomings of the aforementioned technologies, this invention provides a method and system for evaluating the welding process of steel bridges based on artificial intelligence.

[0008] The technical solution adopted by the present invention to achieve the above-mentioned technical effects is as follows: An artificial intelligence-based test method for steel bridge welding process qualification includes the following steps: S1. Intelligent extraction of project information: Using artificial intelligence models to analyze the design drawings and related technical documents of the steel bridge project, extracting information on welds, steel plate materials, and plate thickness, generating a statistical table of project weld information, and extracting the standards and specifications on which the project is based and the weld quality grade requirements. S2. Intelligent statistics and process recommendation of weld information: Based on weld quality requirements and successful historical welding process qualification test cases, the artificial intelligence model is used to recommend welding process parameters for different welds. S3. Intelligent planning of evaluation items: Based on the confirmed weld information statistics table and the preset welding process evaluation coverage principle, it automatically generates a list of test items to be evaluated. S4. Test item substitution search: Based on the historical welding procedure qualification test result database, search for qualified welding procedure qualification reports that can cover the weld to be evaluated. If they exist, generate a list of substitution items for engineers to make decisions. S5. Generation of Qualification Test Plan: Referencing the welding procedure qualification test plan templates of historical projects and combining them with the current project information, automatically generate a "Welding Procedure Qualification Test Plan" that includes a project overview, welding inspection requirements, and a list of qualification test items. S6. Test plate welding and performance testing: Take test plates to perform welding procedure qualification tests, and after completing the test plate welding, carry out mechanical property tests and obtain mechanical property test reports. S7. Generation and storage of evaluation report: After obtaining the mechanical property test results, the "Welding Procedure Qualification Report" is automatically generated and stored in the historical welding procedure qualification test results database by integrating the welding procedure qualification test plan, process documents and mechanical property test results.

[0009] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process evaluation test method, in step S1, the artificial intelligence model integrates computer vision, document understanding and natural language processing technologies to perform end-to-end joint analysis of the design drawings, thereby realizing automatic positioning of the weld and correlation matching of component materials and plate thickness.

[0010] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process evaluation test method, in step S2, a "quality-process-performance" knowledge graph is constructed, and a graph neural network is used to perform semantic matching and contextual reasoning on historical successful cases to generate a process parameter recommendation result with confidence.

[0011] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process qualification test method, in step S3, by quantifying the compliance similarity between welds in terms of thickness, bevel, position, welding method and welding material, a coverage relationship matrix is ​​constructed and the minimum set coverage problem is solved to determine the minimum number of representative qualification test items.

[0012] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process qualification test method, in step S4, the compliance filtering and confidence ranking of historical qualification projects are performed by integrating welding specification constraints and process element similarity, and at the same time, the reliability of existing qualification projects for the substitution of new welds is quantified.

[0013] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process qualification test method, in steps S5 and S8, based on the embedded specification clauses and historical report templates, a constraint-aware generative artificial intelligence model is used to integrate multi-source structured data extracted from design drawings and technical documents, automatically assemble the contents of each chapter, and generate a logically rigorous, compliant and auditable welding process qualification scheme or qualification report draft.

[0014] Preferably, in the above-mentioned artificial intelligence-based steel bridge welding process qualification test method, the welding process parameters include at least the groove type, backing type, welding method, and welding materials.

[0015] An artificial intelligence-based steel bridge welding process qualification test system for implementing the above method includes: Data input and processing module: used to receive electronic project files and convert them into a format that can be recognized by artificial intelligence; Drawing parsing and key information extraction module: used to parse drawings and related technical documents, automatically locate weld seams, associate them with the material and plate thickness of the components to which they belong, output a complete "Project Weld Seam Information Statistics Table", and extract key information on applicable standards and quality levels from it; The intelligent welding process recommendation module performs semantic matching and contextual reasoning based on weld quality requirements and successful historical welding process qualification test cases to generate preliminary process parameter combinations that meet specifications and have been verified by engineering for new welds, and outputs recommendation results with confidence for engineers to verify. Intelligent Welding Procedure Qualification Report Planning Module: Based on the confirmed weld information statistics table and the preset welding procedure qualification coverage principle, it automatically generates a list of test items to be evaluated; Intelligent Substitute Retrieval Module: By integrating welding specification constraints and process element similarity, the module performs compliance filtering and confidence ranking on historical evaluation projects, and intelligently retrieves existing welding procedure qualification projects that can replace the new weld. Document generation and output module: Integrates multi-source structured data extracted from design drawings and technical documents to generate welding procedure qualification test plans and initial drafts of qualification reports; Human-computer interaction module: used to display the content of files generated by the artificial intelligence model and provide interfaces for human verification, confirmation, modification and decision-making; Historical database: Used to store and retrieve weld information, welding procedure qualification test plans, qualification reports and all related process data from historical projects.

[0016] Preferably, in the above-mentioned AI-based steel bridge welding process qualification test system, the information stored in the historical database includes joint type, representative weld, welding specification parameters, mechanical property test results, cross-sectional macroscopic metallographic photographs, non-destructive testing reports, and test process data.

[0017] Preferably, the above-mentioned artificial intelligence-based steel bridge welding process evaluation test system also includes a closed-loop knowledge evolution mechanism. The closed-loop knowledge evolution mechanism stores the new evaluation results in a structured manner into the process knowledge graph through a knowledge channel to expand the knowledge base. At the same time, it records the modification behavior of engineers to the output of the artificial intelligence model through a behavior channel, and dynamically optimizes the model parameters and reasoning logic by using comparative learning and rule mining.

[0018] The beneficial effects of this invention are as follows: This invention deeply integrates multimodal technologies such as computer vision, natural language processing, knowledge graphs, graph neural networks, and generative artificial intelligence to construct an end-to-end intelligent evaluation framework from unstructured design drawings to structured process decisions. It can realize knowledge accumulation, dynamic reasoning, risk pre-control, and closed-loop evolution in steel bridge welding process evaluation. It has artificial intelligence models and artificial intelligence-driven intelligent decision-making and knowledge transformation of welding processes, which greatly improves the efficiency, accuracy, and standardization of steel bridge welding process evaluation. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the method described in this invention; Figure 2 This is a block diagram of the module structure of the system described in this invention. Detailed Implementation

[0020] To provide a further understanding of the present invention, the invention will be further described below with reference to the accompanying drawings and specific embodiments: In the description of this invention, it should be noted that the terms "vertical," "upper," "lower," and "horizontal," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0021] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or a connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] An embodiment of the present invention proposes an artificial intelligence-based test method for the welding process qualification of steel bridges, comprising the following steps: S1. Intelligent extraction of project information: Utilizes artificial intelligence models to analyze the design drawings and related technical documents of the steel bridge project, identify and extract information on component type, weld location, steel plate material, and plate thickness combination, and generate a statistical table of project weld information; At the same time, according to instructions, extracts project overview, standards and specifications on which the project is based, and weld quality grade requirements from the design drawings and related technical documents of the steel bridge project.

[0023] Specifically, an artificial intelligence model that integrates computer vision, document understanding and natural language processing technologies is used to perform end-to-end joint analysis of drawings and their associated technical documents, automatically locate the weld seam, associate it with the material and plate thickness combination information of the component to which it belongs, output a complete "Project Weld Seam Information Statistics Table", and extract key information on applicable standards and quality levels from it.

[0024] Before extracting project information, the design drawings and related technical documents of the steel bridge project are manually input as the data source. For example, the engineer uploads the design drawings (DWG / PDF format), bidding documents (PDF format), and applicable manufacturing specifications (such as the "Technical Specification for Construction of Highway Bridges and Culverts" JTG / T3650) of a steel truss bridge. Then, using an existing mature artificial intelligence model, the system automatically identifies the list of components, material list, and details in the drawings, extracts all weld locations, steel plate materials (such as Q345qD, Q420qE), and thicknesses (such as 16mm, 32mm, 50mm), and initially generates a "Weld Information Statistics Table." At the same time, the system identifies welding process requirements from the bidding documents, such as "steel structure manufacturing shall be carried out in accordance with the 'Technical Specification for Construction of Highway Bridges and Culverts' JTG / T3650" and "all butt welds of main truss members must undergo 100% ultrasonic testing, with a quality level of I." Then, the verification engineer quickly browsed and verified the table, correcting the information misidentified by the artificial intelligence model. For example, a piece of incorrectly identified steel plate thickness information was manually corrected after being manually verified and identified by the verification engineer.

[0025] S2. Intelligent statistics and process recommendation of weld information: Based on weld quality requirements and successful historical welding process qualification test cases, the system uses an artificial intelligence model to recommend preliminary process parameters such as groove type, backing type, welding method, and welding material for different welds, and adds them to the weld information statistics table for manual verification and confirmation.

[0026] Specifically, by constructing a "quality-process-performance" knowledge graph, its graph neural network is used to perform semantic matching and contextual reasoning on historical successful cases. This generates preliminary process parameter combinations that conform to specifications and have been verified by engineering for new welds, and outputs recommended results with confidence levels for engineers to verify. The recommended results mainly include welding methods, welding materials, shielding gas, groove type, preheating temperature, and welding specification parameters. Simultaneously, a warning is given when a certain mechanical index of the process is close to the standard value based on historical data. Successful historical welding procedure qualification test cases are stored in a historical database and used for process recommendations to adapt to new weld construction.

[0027] In embodiments of this invention, the construction of the "quality-process-performance" knowledge graph is based on the collection of previous welding procedure qualification and welding process knowledge, such as welding procedure qualification reports and welding procedure specifications. The large model guides the extraction of entities and relationships from the data in these documents according to categories such as quality (including steel plate strength grade, quality grade, joint type, welding position, etc.), process (including welding method, welding materials, welding equipment, groove type, welding specification parameters, etc.), and performance (including strength, elongation, impact toughness, etc.). The extracted triples and relational data are then stored in a graph database. During this process, it is necessary to align and merge the same entity from different sources (such as "CO2 gas shielded welding" and "MAG") to eliminate ambiguity and form a unified knowledge representation. Ultimately, a network knowledge structure is constructed with "weld quality requirements," "welding process," and "welding performance" as core entities, interconnected through rich relationships. Furthermore, a feedback loop is established. When engineers adopt or modify the process parameters recommended by the system, new success cases and corresponding "quality-process-performance" relationships will be added back to the database and the knowledge graph will be updated, enabling continuous accumulation and iterative optimization of knowledge.

[0028] Regarding semantic matching and contextual reasoning: For new welds, the system extracts their known feature vectors, such as feature vectors related to quality requirements like steel plate strength grade, quality grade, joint type, and welding position. These feature vectors are then fused and converted into a vector representation identical to historical cases. The vector representation of the new weld is then compared with the vectors of all historical successful cases in the knowledge graph (a score is assigned based on the degree of similarity). The historical case node with the highest similarity is the semantically best-matching "neighbor." During the matching process, an attention mechanism can be introduced, enabling the model to automatically learn and assign different weights to different historical cases or different features within cases, based on the emphasis of different weld quality requirements. For example, when a new weld has extremely high requirements for low-temperature impact toughness, the model will automatically increase the matching weight of historical cases that meet similar high impact requirements.

[0029] In the "Quality-Process-Performance" knowledge graph, "weld quality requirements" are achieved through the implementation of "welding processes" and verified using mechanical properties. The neural network can perform multi-hop reasoning along these relational paths, automatically discovering the causal chain from "quality requirements" to "process parameters." When a new weld node lacks a perfectly matching successful case, the model can analyze the contextual information of relevant materials, similar requirements, and successful cases, combining this with engineering experience from the entire knowledge graph to recommend a suitable process. In this process, rule and logical constraint mechanisms can be used to embed industry standards and empirical knowledge of welding metallurgy principles as hard or soft constraints into the training and reasoning process of the graph neural network, improving its accuracy.

[0030] For example, the system retrieves weld information from historical databases of "similar span steel truss bridge projects," combines this information with the current project's structural and weld requirements, and performs intelligent welding process recommendations to generate the "XX Bridge Project Weld Information Statistics Table." This table lists the codes (e.g., WELD-TRUSS-001), locations (e.g., "upper chord H1-H2 butt joint"), welding positions (e.g., PA / 1G), and plate thickness combinations (e.g., 32mm+32mm) for all welds on the entire bridge. Based on the requirements for "Class I welds" and historical data, it recommends solutions such as "double-sided V-groove + ceramic backing," "using gas metal arc welding (GMAW)," and "using ER50-6 solid welding wire" for butt welds, and fills these recommendations into the weld information statistics table. After engineer confirmation, this weld information statistics table serves as the basis for subsequent welding process parameters.

[0031] S3. Intelligent planning for evaluation projects: Based on the confirmed weld information statistics table and the preset welding procedure qualification coverage principle, it automatically analyzes and lists all test items that must be evaluated for welding procedure qualification.

[0032] Specifically, by quantifying the compliance similarity between welds in key elements such as steel plate thickness, groove shape, welding position, welding method and welding materials, a coverage relationship matrix is ​​constructed, and the minimum set coverage problem is solved, thereby automatically selecting the minimum number of representative welding procedure qualification test items to achieve specification coverage of all welds and optimal cost.

[0033] In embodiments of the present invention, the welding procedure qualification coverage principle is to compare the current weld to be qualified with existing weld procedure qualification items. Specifically, for butt welds, full penetration fillet welds, and partial penetration fillet welds, coverage is possible when the steel plate thickness δ of the new weld meets the following relationship with the steel plate thickness t of the already qualified weld. The specific relationships are: when t≤16, 0.5t≤δ≤1.5t; when 16<t≤25, 0.75t≤δ≤1.5t; when 25<t≤80, 0.75t≤δ≤1.3t.

[0034] Among the situations where coverage is not possible are: When the steel grade, welding material type, welding method or welding position, gasket material, or current type and polarity are changed, it cannot be covered and needs to be re-evaluated.

[0035] If the welding current, welding voltage, and welding speed change by more than ±10%; the groove shape and size change (groove angle decreases by more than 10°, blunt edge of penetration weld increases by more than 2mm; root gap changes by more than 2mm without backing, root gap changes by more than -2mm or +6mm with backing); or the preheating temperature is 20°C lower than the specified lower limit temperature, it cannot be covered and needs to be re-evaluated.

[0036] Compared with existing welding procedure qualification items, if post-weld heat treatment is added or removed, or filler metal is added or removed, it cannot be covered and re-qualification is required.

[0037] In an embodiment of the present invention, the process of solving the minimum set covering problem is as follows: Feature vectors are extracted from all welds and those evaluated in the historical database and converted into data. Based on standards (such as steel plate thickness, welding position, welding material, etc.), it is determined which welds each evaluated weld can cover, generating a "coverage relation matrix" (i.e., which welds can cover which). A set-based covering algorithm, such as a greedy algorithm, is used for automatic filtering. The strategy prioritizes welds with the "widest coverage and strongest versatility," using the minimum set to encompass all welds. If there are any remaining welds that cannot be covered, they are treated as welds to be evaluated and the above process is repeated along with all remaining welds until the minimum number of new welds to be evaluated is determined.

[0038] Exemplarily, a set of joints with steel plate material of Q345qD, steel plate thickness of 20 mm, welding method of automatic submerged arc welding, and welding materials of H10Mn2 + SJ101q are selected for evaluation so that it can cover all butt joint welds of the whole bridge. If other butt joint welds are evaluated, at least two sets are required to cover all of them.

[0039] S4. Substitute item retrieval for test items. Based on the historical welding procedure qualification test result database, automatically retrieve whether there is a qualified welding procedure qualification report that can cover the currently to-be-evaluated weld. If such a welding procedure qualification report is retrieved, generate a list of substitute items for the engineer to make a decision.

[0040] Specifically, the list of substitute items includes the item name, number, and relevant report materials. This list of substitute items can be used by the engineer to decide whether to directly引用 it to avoid repeated tests. Specifically, by integrating welding specification constraints and process element similarity, conduct compliance filtering and confidence ranking on historical evaluation items, intelligently retrieve existing welding procedure qualification items that can replace new welds, and quantify their substitution reliability.

[0041] Exemplarily, when retrieving the historical database, a qualification report numbered "PQR - 2023 - 088" is found. Its evaluated steel plate thickness is 20 mm, steel plate material is Q345qD, welding method is automatic submerged arc welding, welding materials are H10Mn2 + SJ101q, and its mechanical properties are qualified. After retrieval and comparison, this welding procedure qualification report can cover the currently to-be-evaluated weld, so that it can be directly引用 to the weld test item with a current plate thickness of 20 mm. Then, a prompt appears: "Detected substitute qualification PQR - 2023 - 088. Do you want to directly引用 it?" After the engineer's evaluation, it is decided to引用 it, and this qualification report is automatically associated with the corresponding weld.

[0042] S5. Generation of the qualification test plan. Refer to the welding procedure qualification test plan template of historical projects and combine with the current project information to automatically generate the "Welding Procedure Qualification Test Plan" including project introduction, welding inspection requirements, and a list of qualification test items.

[0043] Specifically, the "Welding Procedure Qualification Test Plan" includes the following contents: Project introduction: Automatically extract information such as project overview and structural features from design drawings and bidding documents to generate.

[0044] Scope of application and compilation basis: Automatically generate according to project information and standard specifications extracted from the drawings.

[0045] Main joint forms and steel plates for qualification: Extract and organize from the weld information statistical table.

[0046] Welding methods and welding materials: extracted from the evaluation test items determined in step S3.

[0047] The assessment report's content framework is generated based on standards and historical templates.

[0048] List of welding procedure qualification test items: directly referencing the results generated in step S3.

[0049] That is, step S5 integrates multi-source structured data extracted from design drawings, technical documents and other modules, and automatically assembles the content of each chapter based on the embedded specification clauses and historical report templates using a constraint-aware generative artificial intelligence model to generate a logically rigorous and compliant auditable welding process qualification scheme.

[0050] The entire process is as follows: First, a standard evaluation scheme template is retrieved from the historical database; then, the project overview is extracted from the project documents to generate a "Project Introduction"; next, the welding methods and materials are determined from the list of projects to be evaluated; then, the weld inspection requirements are generated according to the Class I weld requirements in the specification clauses; next, other information is extracted and summarized from relevant documents; finally, the contents of each chapter are automatically assembled to generate a complete "Welding Procedure Qualification Test Scheme".

[0051] S6. Test plate welding and performance testing: Take test plates to perform welding procedure qualification tests, and after completing the test plate welding, carry out mechanical property tests and obtain mechanical property test reports. Specifically, before performing the welding procedure qualification test, there is also an intermediate step S5-6: test process management and data collection. That is, after generating the "Welding Procedure Qualification Test Plan" in step S5, it is necessary to list the test plates, including information such as plate thickness, material, size and bevel size, based on the generated qualification plan and the determined test plate size rules, and send the test plate list to the production workshop for the cutting and processing of the test plates.

[0052] Specifically, in step S6, before and during the test, all process documents, including the steel plate quality certificate, steel plate re-inspection report, welding material quality certificate, welding material re-inspection report, welding test record, and welding procedure qualification guide, are electronically collected and entered. The collected and entered documents are then automatically OCR-recognized, structurally archived, and associated with the corresponding qualification items for storage. After welding is completed, the welded test plate is sent to a third-party testing institution for mechanical property testing.

[0053] S7. Generation and Storage of the Qualification Report: After obtaining the mechanical performance test report, it is entered into the system. The system references the format of historically qualified process qualification reports and integrates all relevant data (including test plans, process documents, mechanical performance results, etc.) to automatically generate a complete draft of the "Welding Procedure Qualification Report." This draft, after manual review and confirmation, is officially stored in the historical welding procedure qualification test results database to provide data support for subsequent projects.

[0054] Specifically, in step S7, based on the embedded normative clauses and historical report templates, a constraint-aware generative artificial intelligence model is used to integrate multi-source structured data extracted from design drawings and technical documents, automatically assemble the content of each chapter, and generate a logically rigorous and compliant draft of the evaluation report.

[0055] The historical welding procedure qualification test results database stores complete qualification information, including but not limited to: joint type, representative weld, welding method and welding materials, welding specification parameters, mechanical property test results, cross-sectional macroscopic metallographic photographs, non-destructive testing reports, and other relevant process data. All data is stored in a structured format and can be retrieved and reused for future projects. This concludes the entire qualification process.

[0056] Furthermore, in a preferred embodiment of the present invention, a closed-loop knowledge evolution mechanism is also included. This mechanism expands the knowledge base by structuring and storing the new evaluation results into the process knowledge graph through a "knowledge channel" and recording the modification behavior of engineers on the output of the artificial intelligence model through a "behavior channel". By using comparative learning and rule mining, the model parameters and reasoning logic are dynamically optimized to achieve continuous evolution of the system's cognitive ability.

[0057] With this, the entire welding process qualification test for the steel truss bridge has been completed, and subsequent similar steel bridge projects can directly retrieve and reuse the evaluation results from the historical database.

[0058] On the other hand, the present invention also proposes an artificial intelligence-based steel bridge welding process qualification test system for implementing the above method, the system comprising: Data input and processing module: used to receive electronic project documents and convert them into a format that can be recognized by artificial intelligence; specifically, the aforementioned electronic project documents include design drawings, technical specifications, tender documents and process documents; for example, an engineer uploads the design drawings (DWG / PDF format), tender documents (PDF format) and applicable manufacturing specifications (such as the "Technical Specification for Construction of Highway Bridges and Culverts" JTG / T3650) of a steel truss bridge to the system's data input module.

[0059] The drawing parsing and key information extraction module is used to parse drawings and related technical documents, automatically locate weld seams, associate them with the material and plate thickness combinations of their respective components, output a complete "Project Weld Information Statistics Table," and extract key information on applicable standards and quality levels from it. Specifically, by integrating computer vision, document understanding, and natural language processing technologies, it performs end-to-end joint parsing of drawings and their related technical documents, automatically locates weld seams, and associates them with the material and plate thickness combinations of their respective components.

[0060] After engineers upload the electronic project files to the data input and processing module, the system's drawing parsing and key information extraction module is activated. It automatically identifies the component lists, material lists, and details in the drawings, extracting all weld locations, steel plate materials (e.g., Q345qD, Q420qE), and thicknesses (e.g., 16mm, 32mm, 50mm), generating a preliminary "Weld Information Statistics Table." Simultaneously, it identifies requirements from the tender documents such as "Steel structure manufacturing shall be carried out in accordance with the 'Technical Specifications for Highway Bridge and Culvert Construction' (JTG / T3650)" and "All main truss butt welds must undergo 100% ultrasonic testing, with a quality level of I." Engineers quickly browse and verify the table in the human-computer interaction module, correcting a misidentification of steel plate thickness by the artificial intelligence model.

[0061] The intelligent welding process recommendation module performs semantic matching and contextual reasoning based on weld quality requirements and successful historical welding process qualification test cases. It generates preliminary process parameter combinations that meet specifications and have been verified by engineering for new welds, and outputs recommendation results with confidence for engineers to verify. Specifically, the intelligent welding process recommendation module constructs a "quality-process-performance" knowledge graph, which uses graph neural networks to perform semantic matching and contextual reasoning on historical successful cases.

[0062] After completing the drawing analysis and key information extraction, the system retrieves weld information from the historical database of "similar span steel truss bridge projects." Combining this with the current project's structural and weld requirements, the system activates the intelligent welding process recommendation module to generate the "XX Bridge Project Weld Information Statistics Table." This table lists the codes (e.g., WELD-TRUSS-001), locations (e.g., "upper chord H1-H2 butt joint"), welding positions (e.g., PA / 1G), and plate thickness combinations (e.g., 32mm+32mm) for all welds on the entire bridge. Based on the requirements for "Class I welds" and historical data, the system recommends solutions such as "double-sided V-groove + ceramic backing," "using gas metal arc welding (GMAW)," and "using ER50-6 solid welding wire" for butt welds, and fills these recommendations into the weld information statistics table. After engineer confirmation, this table becomes the basis for subsequent work.

[0063] The intelligent welding procedure qualification report planning module automatically generates a list of test items to be evaluated based on the confirmed weld information statistics table and preset welding procedure qualification coverage principles. Specifically, by quantifying the compliance similarity between welds in key elements such as thickness, bevel, location, method, and material, a coverage relationship matrix is ​​constructed, and the minimum set coverage problem is solved. This automatically selects the minimum number of representative welding procedure qualification test items, achieving optimal coverage of all welds at the cost-effectiveness level. For example, evaluating a set of joints with Q345qD material, 20mm plate thickness, submerged arc welding method, and H10Mn2+SJ101q welding material can cover all butt welds of the entire bridge. If other butt welds are to be evaluated, at least two sets are needed to achieve full coverage.

[0064] The intelligent substitution retrieval module integrates welding specification constraints and process element similarity to filter and rank historical qualification projects based on compliance and confidence levels. It intelligently retrieves existing welding procedure qualification projects that can replace the new weld and quantifies their substitution reliability. Specifically, after generating the list of test projects to be evaluated, the system activates the intelligent substitution retrieval module. It finds a qualification report numbered "PQR-2023-088," which evaluates a 20mm thick plate made of Q345qD material, using submerged arc welding (SAW) and H10Mn2+SJ101q welding material. The report meets mechanical property requirements. The system determines that this report can replace the current 20mm plate test project and displays a prompt: "Substitute qualification report PQR-2023-088 detected. Directly reference it?". After evaluation by the engineer, the system automatically associates the qualification report with the corresponding weld.

[0065] Document generation and output module: Integrates multi-source structured data extracted from design drawings, technical documents and other modules, and automatically assembles the content of each chapter based on embedded specification clauses and historical report templates using a constraint-aware generative artificial intelligence model to generate a logically rigorous, compliant and auditable welding process qualification scheme or qualification report draft. Human-computer interaction module: used to display the content of files generated by the artificial intelligence model and provide interfaces for human verification, confirmation, modification and decision-making; Historical database: Used to store and retrieve weld information, welding procedure qualification test plans, qualification reports and all related process data from historical projects.

[0066] Furthermore, in a preferred embodiment of the present invention, the information stored in the historical database includes joint type, representative weld, welding specification parameters, mechanical property test results, cross-sectional macroscopic metallographic photographs, non-destructive testing reports, and test process data.

[0067] As a preferred implementation, the system also includes a closed-loop knowledge evolution mechanism. This mechanism expands the knowledge base by structurally storing new evaluation results into the process knowledge graph through a knowledge channel, and simultaneously records engineers' modifications to the AI ​​model's output through a behavior channel. It dynamically optimizes model parameters and inference logic using comparative learning and rule mining. By recording engineer feedback, expanding the knowledge base, and optimizing model parameters through this closed-loop knowledge evolution mechanism, the system's capabilities in process recommendation and project planning continuously improve with actual application, adapting to more steel bridge welding scenarios.

[0068] The aforementioned modules—data input and processing, drawing parsing and key information extraction, intelligent welding process recommendation, intelligent welding process qualification report planning, intelligent substitute retrieval, document generation and output, historical database, human-computer interaction, and closed-loop knowledge evolution mechanism—are connected via a network and work collaboratively to form a complete intelligent welding process qualification management platform.

[0069] The steel bridge welding procedure qualification test system of this invention is deployed on a local enterprise server and includes a data input terminal, an image processing server, an AI inference server, a historical database server, and a human-computer interaction terminal. Each module achieves data communication and collaborative work through the enterprise intranet. The historical database pre-stores structured information such as steel bridge welding procedure qualification reports, test data, and engineering documents from the past 10 years. The AI ​​inference server is equipped with existing mature AI artificial intelligence models. It achieves full automation from design drawing analysis to qualification report generation, while simultaneously completing knowledge accumulation and dynamic evolution, improving the efficiency, accuracy, and standardization of the qualification work.

[0070] In summary, this invention, through an artificial intelligence model, can automatically complete tedious tasks such as information extraction, statistics, and scheme preparation, freeing engineers from repetitive labor and reducing the preparation time for process qualification by more than 70%, significantly improving work efficiency. Intelligent recommendation of welding processes based on historical successful cases and specification requirements reduces the uncertainty caused by differences in human experience, making the qualification scheme more scientific, standardized, and unified, thus improving the quality and standardization level of the qualification. Through test item substitution retrieval and intelligent generation of qualification test schemes, the coverage and substitution of qualification results can be maximized, significantly reducing unnecessary welding and mechanical performance tests and saving material and testing costs. Furthermore, by integrating scattered and unstructured historical qualification data into a structured and searchable database, effective accumulation and reuse of enterprise knowledge are achieved. From project input to report storage, this invention forms a complete digital and intelligent closed loop, ensuring the traceability of all qualification processes and providing strong support for quality management and engineering auditing.

[0071] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A test method for evaluating the welding process of steel bridges based on artificial intelligence, characterized in that, Including the following steps: S1. Intelligent extraction of project information: Using artificial intelligence models to analyze the design drawings and related technical documents of the steel bridge project, extracting information on welds, steel plate materials, and plate thickness, generating a statistical table of project weld information, and extracting the standards and specifications on which the project is based and the weld quality grade requirements. S2. Intelligent statistics and process recommendation of weld information: Based on weld quality requirements and successful historical welding process qualification test cases, the artificial intelligence model is used to recommend welding process parameters for different welds. S3. Intelligent planning of evaluation items: Based on the confirmed weld information statistics table and the preset welding process evaluation coverage principle, it automatically generates a list of test items to be evaluated. S4. Test item substitution search: Based on the historical welding procedure qualification test result database, search for qualified welding procedure qualification reports that can cover the weld to be evaluated. If they exist, generate a list of substitution items for engineers to make decisions. S5. Generation of Qualification Test Plan: Referencing the welding procedure qualification test plan templates of historical projects and combining them with the current project information, automatically generate a "Welding Procedure Qualification Test Plan" that includes a project overview, welding inspection requirements, and a list of qualification test items. S6. Test plate welding and performance testing: Take test plates to perform welding procedure qualification tests, and after completing the test plate welding, carry out mechanical property tests and obtain mechanical property test reports. S7. Generation and storage of evaluation report: After obtaining the mechanical property test results, the "Welding Procedure Qualification Report" is automatically generated and stored in the historical welding procedure qualification test results database by integrating the welding procedure qualification test plan, process documents and mechanical property test results.

2. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, In step S1, the artificial intelligence model integrates computer vision, document understanding and natural language processing technologies to perform end-to-end joint analysis of the design drawings, thereby achieving automatic positioning of weld seams and correlation matching of component materials and plate thickness.

3. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, In step S2, a knowledge graph of "quality-process-performance" is constructed, and a graph neural network is used to perform semantic matching and contextual reasoning on historical successful cases to generate process parameter recommendation results with confidence.

4. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, In step S3, by quantifying the compliance similarity between welds in terms of thickness, bevel, location, welding method, and welding material, a coverage relationship matrix is ​​constructed and the minimum set coverage problem is solved to determine the minimum number of representative evaluation test items.

5. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, In step S4, the compliance filtering and confidence ranking of historical evaluation projects are performed by integrating welding specification constraints and process element similarity, while quantifying the reliability of existing evaluation projects as substitutes for new welds.

6. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, In steps S5 and S7, based on the embedded standard clauses and historical report templates, a constraint-aware generative artificial intelligence model is used to integrate multi-source structured data extracted from design drawings and technical documents, automatically assemble the content of each chapter, and generate a logically rigorous and compliant auditable welding process qualification scheme or qualification report draft.

7. The test method for steel bridge welding process qualification based on artificial intelligence according to claim 1, characterized in that, The welding process parameters include at least the bevel type, gasket type, welding method, and welding materials.

8. A steel bridge welding process qualification test system based on artificial intelligence, implementing the method of any one of claims 1 to 7, characterized in that, include: Data input and processing module: used to receive electronic project files and convert them into a format that can be recognized by artificial intelligence; Drawing parsing and key information extraction module: used to parse drawings and related technical documents, automatically locate weld seams, associate them with the material and plate thickness of the components to which they belong, output a complete "Project Weld Seam Information Statistics Table", and extract key information on applicable standards and quality levels from it; The intelligent welding process recommendation module performs semantic matching and contextual reasoning based on weld quality requirements and successful historical welding process qualification test cases to generate preliminary process parameter combinations that meet specifications and have been verified by engineering for new welds, and outputs recommendation results with confidence for engineers to verify. Intelligent Welding Procedure Qualification Report Planning Module: Based on the confirmed weld information statistics table and the preset welding procedure qualification coverage principle, it automatically generates a list of test items to be evaluated; Intelligent Substitute Retrieval Module: By integrating welding specification constraints and process element similarity, the module performs compliance filtering and confidence ranking on historical evaluation projects, and intelligently retrieves existing welding procedure qualification projects that can replace the new weld. Document generation and output module: Integrates multi-source structured data extracted from design drawings and technical documents to generate welding procedure qualification test plans and initial drafts of qualification reports; Human-computer interaction module: used to display the content of files generated by the artificial intelligence model and provide interfaces for human verification, confirmation, modification and decision-making; Historical database: Used to store and retrieve weld information, welding procedure qualification test plans, qualification reports and all related process data from historical projects.

9. The artificial intelligence-based steel bridge welding process qualification test system according to claim 8, characterized in that, The historical database stores information including joint type, representative weld, welding specification parameters, mechanical property test results, cross-sectional macroscopic metallographic photographs, non-destructive testing reports, and test process data.

10. The artificial intelligence-based steel bridge welding process qualification test system according to claim 8, characterized in that, It also includes a closed-loop knowledge evolution mechanism, which stores the new evaluation results in a structured manner into the process knowledge graph through a knowledge channel to expand the knowledge base. At the same time, it records the modification behavior of engineers to the output of the artificial intelligence model through a behavior channel, and dynamically optimizes the model parameters and reasoning logic by using comparative learning and rule mining.

Images