A Smart Recommendation System and Method for Welding Processes Based on Clustering and Large Models
By constructing a hierarchical and progressive process recommendation architecture, combined with rule constraints, data mining, and knowledge completion, the problem of existing welding process design relying on experience has been solved, achieving accurate and complete generation of welding process parameters, and improving welding quality and production efficiency.
Patent Information
- Application Number
- CN202610505220.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-16
- Publication Date
- 2026-06-02
Smart Images

Figure CN122132451A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of welding process parameter design and intelligent manufacturing technology, and in particular to an intelligent recommendation system and method for welding processes based on clustering and large models. Background Technology
[0002] The rational design of welding process parameters is a crucial step in ensuring welding quality and production efficiency. Current welding process design mainly relies on manual experience and standard lookup tables, which suffers from low quantification, complex parameter combinations, and strong dependence on engineer experience. With the development of data-driven methods, process parameter optimization methods based on historical data have gradually emerged; however, relying solely on data mining can easily produce results that do not conform to process specifications. Meanwhile, while general-purpose large language models possess strong language generation and reasoning capabilities, they lack specialized knowledge in the welding field, and directly applying them to welding process design can easily produce unreliable "illusion" results. Therefore, there is an urgent need for an intelligent recommendation method that can simultaneously ensure process compliance, parameter accuracy, and process integrity. Summary of the Invention
[0003] The purpose of this invention is to provide a data-driven intelligent recommendation method for welding processes. By constructing a hierarchical and progressive process recommendation architecture, it organically combines rule constraints, data mining, and knowledge completion to achieve safe screening, accurate optimization, and complete generation of welding process parameters. This overcomes the problems of existing technologies, such as process design relying on experience, inaccurate recommendation results, and insufficient completeness. To achieve the above objective, this invention adopts the following technical solution:
[0004] Step 1: Obtain the basic process information for the welding task. The basic process information includes at least the plate thickness and joint type.
[0005] Step 2: In the basic constraint layer, based on the pre-built welding process rules and national standard database, the basic process information is matched according to rules, and the initial feasible parameter range of welding current and welding voltage that meets the process specification requirements is output, which is used to construct the feasible domain of welding process parameters.
[0006] Step 3: In the core decision-making layer, based on historical high-quality welding condition data, the welding current and welding voltage parameters located in the feasible domain are clustered using a Gaussian mixture model to obtain several stable welding condition clusters.
[0007] Step 4: Within each stable welding condition cluster, the K-nearest neighbor density estimation method is introduced to determine the welding current and welding voltage parameter points with the highest local data density, which are then used as the optimal core process parameters for the current welding task.
[0008] Step 5: In the knowledge completion layer, using the optimal core process parameters as an index, the matching auxiliary process parameters are retrieved from the private welding process knowledge base through retrieval enhancement generation technology.
[0009] Step 6: When matching private process data is retrieved, a complete welding process specification is generated using a direct mapping method; when the search results are insufficient, the large language model is called to infer and complete the missing auxiliary process parameters under the conditions of rules and physical constraints, thereby generating a complete welding process specification. Attached Figure Description
[0010] The accompanying drawings are used to further illustrate the technical solutions of the present invention and do not constitute a limitation on the scope of protection of the present invention.
[0011] Figure 1 The diagram shows the hierarchical structure of the intelligent recommendation method for welding processes described in this invention, which includes a basic constraint layer, a core decision layer, and a knowledge completion layer.
[0012] Figure 2 : A schematic diagram of the overall system architecture of the data-driven intelligent recommendation method for welding processes described in this invention.
[0013] Figure 3 Information Collection Module Flowchart
[0014] Figure 4 Data Flow Diagram of Large Model Inference Module
[0015] Figure 5 Automatic Flowchart Generation in WPS Documents
[0016] Figure 6 : Process Recommendation System Page Illustration Detailed Implementation
[0017] The technical solutions and embodiments of the present invention will be further described below with reference to the accompanying drawings, but these descriptions do not limit the scope of protection of the present invention. Equivalent modifications or substitutions made to these embodiments by those skilled in the art without departing from the spirit and substance of the present invention should all be covered within the scope of protection of the present invention.
[0018] Example 1
[0019] This embodiment provides a specific implementation process for a data-driven intelligent recommendation method for welding processes, such as... Figure 1 As shown, the method includes the following steps.
[0020] Step 1: In this embodiment, the basic input information required for welding process development is obtained through a web-based front-end interactive interface. This basic input information includes, but is not limited to, parameters such as base material and grouping, welding method, joint type, weld size, and welding wire type. The front-end interface supports inter-field linkage validation and intelligent prompts. For example, when the user selects base material one, the system automatically matches and recommends base material two based on a preset material database, thereby reducing human input errors. The core of this step is to ensure the structured and accurate nature of the input data, providing reliable input for subsequent data-driven reasoning.
[0021] Step 2: After the user completes the input, the front-end sends the basic process information to the back-end service module in JSON format. The back-end performs integrity verification, format standardization, and parameter mapping on the received data, uniformly mapping fields from different user input habits to an internal standard parameter model. For example, the base material name is mapped to the corresponding material group number, and the welding method is mapped to a predefined process parameter template. This step eliminates input discrepancies, ensuring the stability and consistency of the subsequent intelligent recommendation process.
[0022] Step 3: After parameter standardization, the system uses the processed process input as model input, calls a large language model or data-driven reasoning model with domain knowledge reasoning capabilities to comprehensively analyze the welding process parameters and generate core process parameters such as welding current and voltage that conform to the process specifications. This step uses historical welding data and process rules as constraints to achieve intelligent recommendation of core parameters.
[0023] Step 4: Based on the recommended process parameters and the joint type information input by the user, the system automatically retrieves the pre-built welding process image database and matches the corresponding joint diagram and welding sequence diagram. For example, when the joint type is a butt joint, the corresponding butt welding diagram is returned; when the joint type is a T-joint, the corresponding T-joint welding sequence diagram is returned. The image data is returned to the front end in the form of a file path or encoding for subsequent process specification display.
[0024] Step 5: After receiving the recommended process parameters and matching process images, the front end automatically fills them into a preset welding process specification template. The template includes a process parameter table, joint diagram, welding sequence diagram, process key point description area, and multiple sets of process description fields. The template engine dynamically renders the content to generate a complete and uniformly formatted welding process specification page.
[0025] Step 6: After the welding process specification page is generated, the system calls the document conversion module to perform format verification, style rendering, image embedding and table layout adjustment on the page content, and automatically converts the welding process specification into a standardized PDF file, realizing one-click export function, which is convenient for archiving and on-site use.
[0026] Step 7: The method described in this embodiment is deployed in a welding process management platform. During actual production, users only need to input basic process information to automatically obtain recommended welding process parameters and corresponding welding process specification documents. Verification through multiple production applications shows that this method can significantly improve the efficiency of welding process preparation, enhance the consistency of process parameters among different engineers, and reduce the error rate caused by manual table lookups and experience-based judgments, thus verifying the effectiveness and practicality of the method of this invention.
[0027] This embodiment verifies the effectiveness and practicality of the present invention.
[0028] It should be noted that all contents not described in detail in this embodiment are existing technologies known to those skilled in the art; the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention.
Claims
1. A welding process intelligent recommendation system and method based on clustering and large models, characterized in that, Includes the following steps: Step 1: Obtain the basic process information for the welding task, which includes at least the plate thickness and joint type; Step 2: In the basic constraint layer, the pre-built national welding process standard database is called according to the basic process information. Through rule matching and constraint reasoning, feasible parameter ranges for welding current and welding voltage that meet the standard requirements are output. Step 3: In the core decision-making layer, based on historical high-quality welding condition data, the welding current and welding voltage are clustered using a Gaussian mixture model to obtain several stable condition clusters. Within each stable condition cluster, the core parameter point with the largest local density is determined using the K-nearest neighbor density estimation method, which serves as the optimal welding current and voltage. Step 4: In the knowledge completion layer, using the optimal welding current and voltage as the retrieval index, the matching auxiliary process parameters are retrieved from the private welding process knowledge base using retrieval enhancement generation technology; Step 5: When matching data exists in the private welding process knowledge base, the complete welding process specification is obtained directly by mapping; when no matching data exists in the private welding process knowledge base, the large language model is called to infer and complete the missing auxiliary process parameters based on physical constraints and process logic, and a complete welding process specification is generated.
2. The method according to claim 1, characterized in that, The feasible parameter ranges of welding current and welding voltage output by the basic constraint layer constitute the feasible domain of welding process parameters, which are used to eliminate parameter combinations that violate process specifications or common sense physics.
3. The method according to claim 1, characterized in that, The historical high-quality welding condition data includes welding current, welding voltage, and corresponding welding quality evaluation results, and only data whose welding quality meets the preset threshold are selected as clustering input.
4. The method according to claim 1, characterized in that, The Gaussian mixture model is used to identify stable welding condition clusters under equipment or process conditions to characterize the stability features of the welding arc.
5. The method according to claim 1, characterized in that, The private welding process knowledge base includes the company's welding process specifications, experimental data, and historical production data, and a retrieval index is established through feature vectorization.
6. The method according to claim 1, characterized in that, When performing auxiliary process parameter completion, the large language model introduces prompting engineering constraints to ensure that the generated results satisfy the welding physical mechanism and process experience logic.
