An intelligent casting process whole-flow optimization system and method

Abstract

The application relates to the technical field of casting, and discloses an intelligent casting process whole-flow optimization system and method, which comprises the following steps: digitizing and analyzing casting drawings and process flow documents to generate a structured process table; based on the structured process table, using a DeepSeek-NLP model to perform multi-round question answering, outputting process structure analysis results and initial process parameter suggestions; based on the process structure analysis results and the initial process parameter suggestions, constructing a three-dimensional process model, simulating a casting process based on a simulation tool, and generating simulation results; testing the simulation results; if there are defects that do not meet quality standards or process parameters that do not match, feeding back correction requirements to an AI intelligent analysis module, triggering iterative optimization of process parameters, and completing closed-loop optimization of the process flow. The application can significantly shorten the parameter optimization cycle, improve the intelligent level of production and the product yield.

Description

Technical Field

[0001] This application relates to the field of casting technology, and in particular to an intelligent casting process optimization system and method. Background Technology

[0002] In traditional casting processes, process design often relies on manual experience and a step-by-step trial-and-error method. This not only results in a long optimization cycle for process parameters but also, due to a lack of effective data support, easily leads to design deviations and casting defects such as porosity and shrinkage cavities. Especially when facing complex casting designs and the application of new materials, traditional methods cannot effectively adapt to new process requirements, increasing the risk of casting defects. At the same time, existing casting process optimization processes typically rely on manually drawn casting drawings and process flow documents, which results in low information transmission efficiency and insufficient accuracy in the conversion between diagrams and actual production processes. This often leads to a gap between the implemented process and the design, thus affecting production efficiency and product quality.

[0003] In this context, the development of intelligent casting process optimization methods has become particularly important. This application provides a full-process optimization method for casting processes that combines deep learning and 3D modeling. By utilizing artificial intelligence models (DeepSeek-NLP models), especially DeepSeek-NLP models, multiple rounds of question-and-answer analysis are conducted during the process design process to progressively optimize process parameters such as parting line position and gating system configuration, avoiding the dependence on and blindness of traditional methods in process parameter design. This method significantly improves the efficiency and accuracy of process optimization through data digitization, automated analysis, and feedback mechanisms. It can promptly identify and correct design defects in complex casting processes, ensuring that the final casting process meets quality standards.

[0004] Furthermore, the application of simulation technology allows for the verification and optimization of various stages of the casting process, such as filling, solidification, and cooling, during the design phase, reducing waste and costs in the experimental process. Through the close integration of the DeepSeek-NLP model and simulation tools, design schemes can be dynamically adjusted, improving production efficiency and reducing the defect rate. Summary of the Invention

[0005] This application provides a smart casting process whole-process optimization system and method to improve the intelligence level and production efficiency of the casting process.

[0006] Firstly, this application provides a method for optimizing the entire intelligent casting process, the method comprising: Step S1: Digitally analyze the casting drawings and process flow documents to generate structured process tables; Step S2: Based on the structured process table, use the DeepSeek-NLP model to perform multiple rounds of question and answer, and output the process structure analysis results and preliminary process parameter suggestions; Step S3: Based on the process structure analysis results and the preliminary process parameter suggestions, construct a three-dimensional process model, and simulate the casting process using simulation tools to generate simulation results; Step S4: Verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the AI ​​intelligent analysis module, and iterative optimization of the process parameters will be triggered to complete the closed-loop optimization of the process flow.

[0007] Secondly, this application provides an intelligent casting process optimization system, the system comprising: The digital analysis module is used to digitally analyze casting drawings and process flow documents to generate structured process tables. The AI ​​intelligent analysis module is used to perform multi-round question and answer based on the structured process table using the DeepSeek-NLP model, and output process structure analysis results and preliminary process parameter suggestions. The 3D modeling and simulation module is used to construct a 3D process model based on the process structure analysis results and the preliminary process parameter suggestions, and to simulate the casting process using simulation tools to generate simulation results. The feedback optimization module is used to verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the AI ​​intelligent analysis module, triggering iterative optimization of the process parameters to complete the closed-loop optimization of the process flow.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: The technical solution provided in this application significantly improves the design efficiency and accuracy of the casting process through an intelligent, end-to-end optimization method. In traditional processes, design often relies on manual experience and trial-and-error, resulting in long optimization cycles and inherent biases. However, this application, by combining the DeepSeek-NLP model with deep learning technology, can automatically analyze and optimize process parameters in a shorter time, reducing the impact of human intervention and improving the scientific rigor and rationality of the design.

[0009] Through structured data processing and multi-round question-and-answer analysis, this application can accurately extract and analyze key information in casting drawings and process flow documents. Based on structured process tables and deep learning using the DeepSeek-NLP model, it automatically identifies and optimizes various process parameters, ensuring the accuracy of parameter design and avoiding the problems of information loss or inaccurate transmission in traditional design methods. Furthermore, by combining with simulation tools, it can simulate each stage of the casting process in advance during the design phase, predict and evaluate casting defects in real time, provide data support for optimized design, and ensure the feasibility of the process design scheme.

[0010] This application also enhances the adaptability of the DeepSeek-NLP model to new materials, new technologies, or special working conditions through reinforcement learning and incremental learning mechanisms, enabling it to adjust and optimize in a timely manner based on actual production data, avoiding the problems of outdated data and poor adaptability that may occur in traditional training methods. In addition, through a closed-loop optimization mechanism, the model can continuously adjust process parameters based on simulation results and production feedback, ensuring continuous improvement of the process flow and further enhancing the quality and production efficiency of casting products.

[0011] This application realizes intelligent design and optimization from casting drawings to final process solutions, which not only improves the efficiency and precision of casting processes, but also greatly reduces trial and error costs and production costs, improves the stability of the casting process and the consistency of products, and ultimately enables the casting industry to achieve significant technological progress in process design and optimization. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the steps of a smart casting process optimization method in an embodiment of this application. Figure 2 This is a schematic diagram of the casting process drawing for the body cover in an embodiment of this application; Figure 3 This is a schematic diagram of the structured process flow table for castings in the embodiments of this application; Figure 4 This is a structural diagram of a smart casting process optimization system according to an embodiment of this application. Detailed Implementation

[0014] This application provides an intelligent casting process optimization system and method. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Example 1: Traditional casting processes rely on experience for process parameter design, lacking systematicity and precision, resulting in low process optimization efficiency and difficulty in accurately predicting defects during casting. Furthermore, existing simulation tools often rely on idealized models for prediction, failing to fully consider the influence of variables such as temperature and flow rate in actual production, leading to significant deviations between simulation results and actual production, and hindering effective closed-loop optimization of process parameters. In casting process design, mismatches between process parameters and quality requirements frequently occur, easily leading to defects such as porosity and shrinkage cavities, affecting casting quality. Simultaneously, traditional methods are insufficiently responsive to dynamic changes in the production environment, making accurate optimization of casting processes for complex castings or new materials difficult. Therefore, this application combines artificial intelligence technology and 3D simulation analysis, employing a multi-round question-and-answer optimization method based on the DeepSeek-NLP model. By utilizing real-time data feedback and production environment simulation, it improves process design accuracy, achieving a closed loop from process structure analysis to process parameter optimization, ultimately improving casting quality and production efficiency.

[0016] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 The intelligent casting process optimization method in this application includes: Step S1: Digitally analyze the casting drawings and process flow documents to generate structured process tables.

[0017] In this process, computer vision technology is used to digitally analyze the geometric contours, dimensions, and tolerance requirements of casting drawings, transforming them into a set of structured geometric parameters that can be recognized by computers. The rationality of the dimensions is then verified through geometric constraints.

[0018] Among them, natural language processing technology is used to extract the process execution steps, parameter ranges and quality control standards from the process flow documents, transform them into a standardized set of process parameters and rules, and verify the matching of process steps and parameters through semantic consistency.

[0019] Based on the set of structured geometric parameters and the set of standardized process parameters and rules, after data mapping and format conversion, they are integrated into a computer-recognizable structured process table containing process execution steps, parameter ranges and quality control standards, forming a digital database of the original casting process.

[0020] Specifically, the technical solution of this application aims to solve the problem of information acquisition and processing in the process of intelligent casting process optimization, especially how to extract accurate process parameters and structural information from casting drawings and process flow documents, and convert them into computer-readable data to support subsequent process analysis and optimization. In traditional casting processes, the information in process drawings and documents is usually difficult to be directly processed by computer systems. Therefore, this application uses digital parsing technology to convert this information into structured process tables, providing basic data support for process optimization.

[0021] In a specific embodiment, firstly, for casting drawings, computer vision technology is used to digitally analyze the drawings, including pixel-level scanning to identify key information such as geometric contours, dimensions, and tolerance requirements. By using image processing algorithms, the graphic information is transformed into a set of structured geometric parameters. For example, when processing pipe drawings, the pipe's diameter, length, and angle are automatically identified and converted into computer-recognizable vector data. At the same time, the rationality of the dimensions is verified based on geometric constraints to ensure that the dimensions and annotations in the drawings meet the actual engineering requirements. The key technology in this process lies in the application of graphic recognition and geometric constraints, ensuring that the converted data can be directly used in subsequent process simulations.

[0022] For process flow documents, natural language processing (NLP) technology is used to extract process logic, parameter values, and quality control standards from the text. Each process step, process parameter, and quality standard in the document is transformed into a standardized set of process parameters and rules after word segmentation, semantic analysis, and entity extraction. For example, from the casting process document, parameters such as the execution order, corresponding temperature range, pouring time, and pouring volume of each process are extracted and transformed into structured data. In this process, semantic consistency verification is a key step, which ensures the matching between process steps and parameters and avoids data inconsistency or errors.

[0023] After completing the two extraction processes above, the obtained set of structured geometric parameters and the set of process parameters and rules are integrated into a unified structured process table through data mapping and format conversion. This table includes fields such as the execution steps, parameter ranges, and quality control standards for each process, ensuring that the computer or DeepSeek-NLP model can directly read and process this information. The core technology of this process is data formatting and mapping, which transforms the multiple extracted data sets into a standardized process table, facilitating subsequent database storage and retrieval. Finally, the generated structured process table forms a digital database of the original casting process, providing data support for subsequent process analysis, simulation, and optimization.

[0024] Throughout the digital analysis process, a multi-dimensional verification mechanism is employed to ensure the accuracy and completeness of information extraction. For example, geometric information is verified through geometric constraints to ensure the consistency and rationality of dimension annotations. Document information is verified through semantic consistency to ensure the matching of process steps and parameters. After these verifications, the information is transformed into structured data, which can greatly improve the reliability of subsequent fault prediction or process optimization.

[0025] The technical solution in this embodiment solves the problems of low information processing efficiency and low accuracy in traditional casting processes. By digitizing the information in drawings and documents and converting it into structured data, it not only improves the efficiency of data processing, but also provides accurate basic data support for subsequent intelligent analysis, process simulation and fault prediction.

[0026] The basic digital database used in the specific embodiment is based on more than forty sets of casting process drawings, full-process process documents and supporting technical data of cast steel / aluminum accumulated by our unit over a long period of time.

[0027] In this embodiment, for Figure 2 The following steps are taken to digitize and analyze a casting process drawing of a fuselage cover, and generate a structured process table: S11. Data Source Preparation: First, collect relevant technical documents for the fuselage cover, including casting drawings, process flow documents, and part structure and material information. Casting drawings include the three-dimensional outline of the fuselage cover, geometric dimension markings of blades and ribs, spatial structure diagram of the three-way oil passage, and dimensional sketches of the gating system such as the sprue, runner, and ingate. The process flow documents record each step in the casting process, including melting, degassing, sand mold preparation, pouring, pressure holding, cooling, demolding, and post-processing. Each step corresponds to specific process parameters and quality requirements, such as melting temperature range of 680-720℃, pouring speed range of 1.0-1.5m / s, pressure holding time of 30-45s, and quality requirements including no porosity or shrinkage cavities in the oil passage area and surface roughness Ra≤12.5μm. The part structure and material information details the external dimensions of the fuselage cover, the minimum wall thickness of blades and ribs, and the fluidity and thermal expansion coefficient of the alloy.

[0028] S12. Application of Multimodal Analysis Technology: For casting drawings, computer vision technology is used to digitally analyze the drawings. Pixel-level scanning of the casting drawings is performed using the Python programming language combined with the OpenCV library (version 4.5). The Canny edge detection algorithm is used to identify the vector coordinates of the outer contour of the body cover, blades, and ribs. For example, in the image coordinate system, the contour coordinate range of the top blade is (x: 50-120, y: 30-80). This coordinate is converted into a set of geometric parameters, including 28 data items such as the outer contour, blade and rib dimensions, and gating system sketch dimensions. For dimension annotation information, the PaddleOCR tool is used to extract text, such as the minimum wall thickness of 3mm and the blade length of 40mm, and convert it into numerical data. At the same time, the geometric constraint verification module verifies the rationality of the connection angle between the blade and the rib (design requirement is 30°±2°) and the perpendicularity of each branch of the T-junction oil passage (perpendicularity tolerance ≤0.05mm), ultimately forming a set of geometric parameters to ensure that all geometric data are accurate and meet design requirements.

[0029] For process flow documents, Natural Language Processing (NLP) technology is used to segment the documents, and key process information is extracted using a BERT-based entity recognition model. Specific operations include extracting process keywords such as smelting, sand mold preparation, and casting; process parameter values ​​such as smelting temperature 680-720℃ and casting speed 1.0-1.5m / s; quality control standards such as no porosity or shrinkage in the T-junction oil passage and Ra≤12.5μm; and material properties such as EN AC-42100 solidus 548℃. A semantic consistency verification module ensures the matching of process steps and parameters. For example, it verifies the consistency between the holding time (30-45s) and the cooling start time (≥5s after holding time), ensuring the logical rationality of the process flow.

[0030] S13. Generation of Structured Process Tables: After integrating the geometric parameter set and parameter and rule set generated in the previous step, this is converted into a structured process table using the pandas library in Python. This table includes fields such as process execution steps, corresponding process parameter values ​​for each process, and quality control standards. Figure 3 The structured process flow chart shown covers all the key parameters for casting aluminum body cover, such as the geometric dimensions of blades and ribs, melting and pouring process parameters, and quality acceptance standards. This structured process chart transforms the original design drawings and process flow documents into a computer-readable data format, ensuring efficient data storage and retrieval. It provides accurate input data for subsequent process optimization and simulation analysis. Through this digital analysis and structured transformation, the precision and efficiency of the casting process can be greatly improved, while reducing manual intervention and improving data consistency and reliability.

[0031] The above embodiments, by combining multimodal analysis technology and using computer vision and natural language processing technology, accurately extract the core information from casting drawings and process flow documents, and transform this information into machine-readable structured process tables. This solves the data processing problem in casting process design, provides basic data support for intelligent casting process whole-process optimization methods, not only improves the efficiency of process design, but also provides reliable support for subsequent fault prediction, process optimization and intelligent decision-making.

[0032] Step S2: Based on the structured process table, use the DeepSeek-NLP model to conduct multiple rounds of question and answer, and output the process structure analysis results and preliminary process parameter suggestions.

[0033] The DeepSeek-NLP model, based on a structured process table, analyzes the process structure and parameters through multiple rounds of question-and-answer sessions. It gradually optimizes the parting line position, gating system configuration, and preliminary process parameter suggestions, and outputs the final process design scheme. During the multiple rounds of question-and-answer sessions, the DeepSeek-NLP model combines process logic, parameter association rules, and quality constraints to gradually adjust the output process parameters, ensuring that the final design scheme meets the casting quality standards.

[0034] Specifically, this embodiment describes in detail how to use the DeepSeek-NLP model, based on a structured process table, to analyze the process structure and parameters through multiple rounds of question and answer, gradually optimize the parting line position, gating system configuration, and preliminary process parameter suggestions, and finally output a process design scheme that meets quality standards.

[0035] In this embodiment, firstly, the structured process table is converted into JSON format, including fields such as part attributes, material attributes, process logic, process parameters, and quality requirements. This data is then input into the DeepSeek-NLP model as a training dataset. The model is built on the TensorFlow 2.10 framework and trained using batch_size=8, epoch=50, and a learning rate of 1e-4. The model encodes the data using a BERT word segmenter to construct word vector representations. The Transformer encoder is then used to learn related knowledge such as "body cover structure (blade / rib / tee oil passage) → process parameter selection" and "EN AC-42100 material properties → process adaptation rules," generating a casting process knowledge graph that covers 200+ "structure-parameter" and "material-process" association rules (such as "minimum wall thickness 3mm → recommended pouring speed 1.2-1.4m / s" and "EN AC-42100 has good fluidity → melting temperature can approach the liquidus").

[0036] Next, based on the structured process table, a multi-round interactive process of precise questioning and DeepSeek-NLP model responses was initiated to gradually optimize the process design scheme. The first round of questions focused on the parting line location design. The question was: "Based on the geometric features of the part in the structured process table—'the outer contour of the engine cover is 210mm × 221mm × 49mm, containing 5 blades and 11 ribs (minimum wall thickness 3mm), with internal three-way oil passages requiring sealing'—combined with the material properties of 'EN AC-42100 solidus 548℃, liquidus 613℃,' and considering the process logic recorded in the table for similar cylinder block parts that 'the parting line must avoid the sealing area and facilitate sand mold release,' where should the parting line of the engine cover be located? What constraints related to the part structure must this location meet?"

[0037] The DeepSeek-NLP model's response is: "The parting line is recommended to be set at a horizontal section 25mm from the top flange of the body cover (spatial coordinate Z=25mm). The structural constraints to be met are: ① Avoid the sealing area of ​​the internal three-way oil passage (this area is located in the Z=10-35mm segment, and the parting line Z=25mm is in the middle of the area, requiring the auxiliary sand core to avoid the sealing surface); ② There is no risk of rigid interference when the blades and ribs are formed in the upper and lower sand molds (simulation shows that the draft angle of the blades in the lower sand mold at this parting line is ≥3°, which meets the draft requirements); ③ The ratio of the projected area of ​​the upper and lower sand molds is 1.1 (≤ the threshold of 1.2 for the projected area ratio of conventional drafting in casting process); ④ The wall thickness of the body cover structure at the parting line is ≥5mm (the actual wall thickness at this section is 6mm, which meets the pressure resistance requirements when the sand mold is closed)."

[0038] The second round of questions asked: "Based on the 'Z=25mm parting surface' determined in the previous round, and combined with the quality requirement in the structured process table that 'the three-way oil passage area is not allowed to have pores / shrinkage cavities with a diameter ≥2mm,' how should the current casting position of the body cover be designed? During the design process, how should the wall thickness distribution of the blades (minimum wall thickness 3mm) and ribs be correlated to ensure uniform filling?"

[0039] The DeepSeek-NLP model's response: "Bottom-in pouring is recommended, with the ingate located on both sides of the bottom of the machine cover (sidewalls at Z=0mm from the parting line). The logic for ensuring uniform filling based on wall thickness distribution is as follows: ① For the thick area from the bottom to the middle (wall thickness ≥6mm), the ingate cross-sectional area is designed to be 200mm²." 2 ① Control the molten metal flow rate to 1.3 m / s to ensure rapid filling and sufficient feeding; ② For areas with a minimum wall thickness of 3 mm, such as the top blades and ribs, reduce the cross-sectional area of ​​the ingate from 200 mm at the bottom. 2 The flow rate gradually transitions to 120mm² at the top, keeping the molten metal flow rate within the range of 1.2-1.3m / s. This matches the flowability of EN AC-42100 at this flow rate, preventing air entrapment due to excessively high flow rates or incomplete filling due to excessively low flow rates.

[0040] The DeepSeek-NLP model's response: "Bottom-in pouring is recommended, with the ingate located on both sides of the bottom of the machine cover (sidewalls at Z=0mm from the parting line). The logic for ensuring uniform filling based on wall thickness distribution is as follows: ① For the thick area from the bottom to the middle (wall thickness ≥6mm), the ingate cross-sectional area is designed to be 200mm²." 2 ① Control the molten metal flow rate to 1.3 m / s to ensure rapid filling and sufficient feeding; ② For areas with a minimum wall thickness of 3 mm, such as the top blades and ribs, reduce the cross-sectional area of ​​the ingate from 200 mm at the bottom. 2 Gradient to 120mm at the top corresponding area 2 This ensures that the molten metal flow rate is within the range of 1.2-1.3 m / s, matching the flowability of ENAC-42100 at this flow rate, thus avoiding air entrapment due to excessively high flow rates or incomplete filling due to excessively low flow rates.

[0041] The three rounds of questions are as follows: "Based on the established 'Z=25mm parting surface' and 'bottom-pouring casting position,' referring to the process parameter range of 'melting temperature 680-720℃, holding time 30-45s' in the structured process table, and the quality requirement of 'T-connector oil passage surface roughness Ra≤12.5μm,' how should the dimensions of the current body cover's gating system (straight runner diameter, horizontal runner cross-sectional dimensions) be set? What should the machining allowances for each surface of the casting (such as the top flange face, bottom mounting face, and oil pipe connection face) be set? How should the 'parameter constraints of the melting and casting process' (such as the relationship between melting temperature and alloy fluidity) recorded in the table be matched during the setting process?"

[0042] The DeepSeek-NLP model's answer is: "① Gating system dimensions: The sprue diameter is designed to be φ50mm (based on flow rate calculations, meeting the requirement of a molten metal flow rate of 1.8L / s, matching the filling rate of bottom-pouring casting); the runner adopts a rectangular cross-section with dimensions of 30mm (width) × 40mm (height) (flow velocity approximately 1.35m / s, within a stable flow range); ② Machining allowance: A machining allowance of 2mm is set for the top flange face and bottom mounting face (to meet the accuracy requirements of subsequent machining to the design dimensions); a machining allowance of 1.5mm is set for the oil pipe connection face (balancing sealing and machining efficiency); ③ Parameter matching logic: The melting temperature is set to 700℃ (close to the liquidus of EN AC-42100 at 613℃, which can enhance alloy fluidity and compensate for the filling capacity of thin-walled areas such as blades and ribs); the holding time is set to 35s (5s longer than the median of the process parameter range, corresponding to the solidification time requirement of a maximum wall thickness of 6mm for the body cover, reducing the risk of shrinkage cavities)."

[0043] Four rounds of questioning: Verification of consistency between parameters and quality requirements; Question content: "Based on the above output of 'Z=25mm parting surface, bottom pouring position, sprue φ50mm / glide 30mm×40mm, machining allowance 2mm / 1.5mm, melting temperature 700℃, holding time 35s', and comparing it with the quality requirements of 'T-junction oil passage porosity / shrinkage cavity diameter ≤2mm' and 'Ra≤12.5μm' in the structured process table, do the current output process structure and parameters meet the quality standards? If there are potential mismatch risks, how should the direction be adjusted?"

[0044] The DeepSeek-NLP model's response: "Based on the model's knowledge graph-based deduction, the current process structure and parameters basically meet the quality standards. Potential mismatch risk: At the connection between the top blade and the rib (minimum wall thickness 3mm), due to the molten metal flow rate of 1.3m / s, the probability of air entrapment is approximately 8% (potentially forming micropores of φ1.5-2mm, close to the '≤2mm' threshold). Recommended adjustment direction: ① Reduce the cross-sectional area of ​​the ingate in the corresponding top area by 5% (from 120mm)." 2Reduced to 114mm 2 ① This reduces the flow velocity to 1.25 m / s, decreasing the probability of air entrapment; ② It increases the permeability of the sand mold from the conventional 700 cm³ / s. 3 / min increased to 800cm 3 / min, enhancing the gas discharge capacity within the cavity.

[0045] Results Integration: The results obtained from multiple rounds of questioning, including "parting surface location (Z=25mm), pouring location (bottom pouring type, inner gates on both sides of the bottom), gating system dimensions (sprue φ50mm, runner 30mm×40mm), machining allowance (flange face / mounting face 2mm, oil pipe face 1.5mm), process parameters (melting 700℃, holding pressure 35s), and risk adjustment suggestions," were integrated to form a complete casting process structure analysis result and initial process parameter scheme, which were then written into the "geometric feature parameters and process execution parameters" module of the structured process table.

[0046] This embodiment utilizes the multi-round question-and-answer mechanism of the DeepSeek-NLP model to perform step-by-step optimization based on a structured process table. The generated process design scheme meets the quality requirements of the casting. Through continuous adjustments via multiple rounds of interaction, the accuracy and reliability of the design scheme are ensured, guaranteeing that the final output process scheme meets all quality standards in the casting process and providing data support for further process optimization.

[0047] Among them, the DeepSeek-NLP model combines process logic, parameter association rules and quality constraints, and automatically adjusts the output process parameters in each round of question and answer to ensure that the final design scheme meets the casting quality standards. When dealing with new materials, new technologies or special working conditions, it automatically absorbs new data and updates the knowledge graph through an incremental learning mechanism.

[0048] Specifically, although the DeepSeek-NLP model uses multi-turn question-and-answer for reasoning, complex casting processes may involve extremely detailed and complex process parameters and design details that cannot be fully answered through limited multi-turn question-and-answer sessions. This is especially true when dealing with casting processes involving new materials, new technologies, or special working conditions, where the training data of existing models may not fully cover all scenarios. To improve the intelligence and depth of multi-turn question-and-answer, this technical solution further optimizes the DeepSeek-NLP model, enabling it to effectively handle complex casting process scenarios, particularly when dealing with process design under new materials, new technologies, or special working conditions, thus enhancing the DeepSeek-NLP model's adaptability to these scenarios. During implementation, reinforcement learning and incremental learning mechanisms are used to periodically introduce new data and optimize the training process of the DeepSeek-NLP model, enabling it to analyze more accurately and provide reasonable process suggestions when dealing with complex processes. Specifically, the implementation process includes the following key steps: data preprocessing, generation and input of structured process tables, model training and learning process, intelligent analysis of multi-turn question-and-answer, and feedback and optimization of output results.

[0049] First, the data processing begins with the digital analysis of casting drawings and process flow documents: computer vision technology is used to perform pixel-level scanning of casting drawings, extracting information such as geometric features, dimensioning, and tolerance requirements, and converting them into structured data; for process flow documents, natural language processing technology is used for word segmentation, semantic analysis, and entity extraction to identify textual information such as process steps, process parameter ranges, and quality requirements; this processed data is organized into standardized structured process tables, which include the execution sequence of each process, process parameters, and quality standards, ultimately forming data input that can be used for analysis by the DeepSeek-NLP model.

[0050] Next, the DeepSeek-NLP model, by inputting this structured data and combining it with a pre-defined model structure and knowledge graph, can automatically identify the correlations between data during multi-turn question answering. The quality and accuracy of the input data directly affect the model's performance. Therefore, this application incorporates a multi-dimensional data verification mechanism during the data structuring process to ensure data consistency and integrity. For example, the extraction and verification of geometric parameters are verified through geometric constraints, while text data is checked for semantic consistency to ensure the matching of process logic and parameter descriptions. The pre-defined model structure and knowledge graph refer to the knowledge system pre-established in the DeepSeek-NLP model to represent the relationships between process structure, material properties, process parameters, and quality requirements. Through association rules and knowledge graphs, the model can identify and learn the inherent connections between different process conditions, providing support for multi-turn question answering and optimizing process parameter design.

[0051] In the training and learning process of the DeepSeek-NLP model, algorithms based on deep learning and natural language processing (NLP) are used to train the data. During training, the model continuously learns from a large amount of process data to construct a knowledge graph of the casting process. The nodes in the knowledge graph include different casting process steps, parameter association rules, material properties, and quality control standards, and the relationships between nodes are continuously optimized and adjusted through model learning. During training, the model continuously adjusts weights and parameters to enable it to accurately understand and parse the information in the process table and generate process parameter suggestions.

[0052] The core feature of this step is intelligent analysis through multi-round question answering. When conducting multi-round question answering using the DeepSeek-NLP model, the model can perform progressive reasoning based on existing process structures and parameter data. The output of each round of question answering not only depends on the results of the previous round but also incorporates the model's deep understanding of casting processes and parameter design. For example, when analyzing a casting process, the model may first start with the geometric features of the part to determine the location of the parting surface. Then, the model will continue to analyze the relationship between the pouring position and the uniformity of filling through question answering and output appropriate process parameter suggestions based on process requirements, such as pouring temperature and holding time. The output of each round increases the understanding of the overall process design and gradually deepens the optimization of design details.

[0053] Through this process, the feedback and optimization mechanism of the output results plays a key role. The results of each round are fed back into the process design to adjust and correct the existing design scheme. When the simulation finds that the design does not meet the requirements, it is automatically fed back to the model for further optimization. This closed-loop optimization process not only ensures the accuracy of the model analysis, but also ensures the continuous improvement of the design results.

[0054] In terms of technical implementation, the combination of reinforcement learning and incremental learning mechanisms further improves the model's adaptability to complex process scenarios. Reinforcement learning enables the model to continuously adjust and optimize decisions based on new data during actual production, while incremental learning helps the model quickly adapt and continuously update its knowledge base when dealing with new materials, new technologies, or special working conditions. For example, when encountering new aluminum alloy materials or complex casting shapes, the model can automatically absorb new data and update its training parameters through incremental learning, avoiding the problems of outdated data and poor adaptability that may occur in traditional training methods. The reinforcement learning and incremental learning mechanisms... This refers to a learning method that self-optimizes through interaction with the environment. Reinforcement learning continuously adjusts decision-making strategies through reward mechanisms, while incremental learning enables the model to dynamically update its knowledge base when processing new data, avoiding problems such as obsolescence and poor adaptability. In this embodiment, reinforcement learning and incremental learning mechanisms refer to the model's feedback learning through interaction with the environment. Reinforcement learning helps the model optimize decisions based on new data in the actual production process, while incremental learning enables the model to quickly adapt and update its knowledge base when encountering new materials, new technologies, or special working conditions, ensuring that it maintains high accuracy and adaptability in the constantly changing process environment.

[0055] Furthermore, the progressive nature of multi-round question-and-answer sessions is improved by optimizing the order of questions and the setting of questions. Based on the needs of actual process design, the question setting is not limited to a single-dimensional process analysis, but can cover multiple dimensions and levels, enabling the model to more comprehensively consider the interaction between design parameters and generate optimization solutions that better meet actual needs. For example, the model can focus on the general framework of the process structure in the early stages of a process, and then gradually delve into the details of specific process parameter design, quality control standards, etc., thereby avoiding the problem of neglecting a key factor that may occur in traditional models.

[0056] In practical applications of the technology presented in this application, the DeepSeek-NLP model can provide optimized design solutions in the early stages of casting process design. This not only effectively reduces the workload of manual design but also greatly improves design efficiency and quality. For example, when designing a process for a new type of aluminum alloy casting, the model can quickly generate appropriate parameters such as gating system design, parting line position, and holding time based on existing knowledge graphs, avoiding the inefficient practice of relying on experience and repeated trials in traditional methods. At the same time, based on simulation feedback, the model can predict potential defects in the casting before production and make timely adjustments and optimizations, ensuring the consistency and stability of product quality.

[0057] Through this deep optimization, the DeepSeek-NLP model enables a shift from experience-driven to data-driven casting process design, improving design accuracy and efficiency. Secondly, through reinforcement learning and incremental learning mechanisms, the model can continuously adapt to the requirements of new processes and materials, enhancing its ability to handle complex process scenarios. Finally, the combination of closed-loop optimization and multi-round question answering ensures the continuous improvement of process design solutions, thereby effectively reducing trial-and-error costs and production costs, and enhancing the company's competitiveness and product quality.

[0058] Step S3: Based on the process structure analysis results and preliminary process parameter suggestions, construct a three-dimensional process model, and simulate the casting process using simulation tools to generate simulation results; Based on the process structure analysis results and preliminary process parameter suggestions, a three-dimensional process model is constructed using three-dimensional modeling software, and physical field simulation is performed using simulation tools to generate simulation results such as filling animation, defect cloud map, flow field distribution cloud map, and temperature field cloud map.

[0059] The three-dimensional process model includes the gating system design, parting line location, and process parameters. The process parameters output by the DeepSeek-NLP model are mapped into the simulation model. The simulation calculation is performed by combining material properties, geometric features, and physical quantities to optimize the casting process design scheme to adapt to the actual production environment.

[0060] Specifically, to address the issues of constructing a three-dimensional process model and simulating physical fields in casting process optimization, this step constructs a three-dimensional process model based on the output results of the DeepSeek-NLP model, and uses simulation tools to simulate the casting process, generating simulation results of physical fields such as flow field and temperature field, thereby further optimizing the casting process design and improving process accuracy and production efficiency.

[0061] In the specific implementation process, firstly, a three-dimensional process model is constructed based on the optimized design results of the DeepSeek-NLP model. The output of the DeepSeek-NLP model typically includes parameters such as the parting line position, gating system design, and machining allowance settings. These design schemes are transformed into three-dimensional geometric shapes using 3D modeling software such as SolidWorks and CATIA, forming an executable process model. In this process, the 3D model not only represents the geometric shape but also needs to accurately set physical properties, such as thermal conductivity, fluidity, and expansion coefficient. For example, in the gating system design, the DeepSeek-NLP model outputs reasonable pouring points, gating gates, and runner dimensions based on the geometric characteristics of the casting, which are visualized through 3D modeling to ensure the rationality of the model design.

[0062] After the three-dimensional process model is completed, the simulation stage begins. In this stage, computational fluid dynamics (CFD) and thermodynamic analysis tools, such as ANSYS, CASTsoft, and Flow3D, are used to simulate the physical processes of molten metal flow, cooling, and deformation during casting. The simulation tools help designers assess the filling status of the casting by calculating the flow path, velocity, and temperature distribution of the molten metal. The simulation results can reveal potential defect areas, such as gas entrapment caused by excessively high flow rates or incomplete pouring caused by excessively low flow rates, thus enabling targeted design optimization.

[0063] Temperature field simulation simulates the entire process from pouring to cooling, analyzing the temperature distribution inside and on the surface of the casting. The uniformity of the temperature field is crucial to the quality of the casting; uneven cooling may lead to cracks or deformation. During temperature field simulation, by optimizing the cooling system design, setting appropriate cooling channels, or adjusting the pouring temperature, the temperature gradient can be reduced, avoiding uneven stress in the casting during cooling. Through joint simulation of the flow field and temperature field, the casting process can be comprehensively evaluated, ensuring the effectiveness of the optimization scheme.

[0064] To further improve the accuracy of simulation results and ensure their consistency with the actual production environment, this application introduces a real-time data feedback and production environment simulation mechanism. Real-time data feedback utilizes industrial internet technology to collect real-time data from the production line, such as mold temperature, melting temperature, and flow rate, and compares this data with the parameters of the simulation model. This mechanism allows for dynamic adjustment of relevant parameters in the simulation model to ensure that the simulation results better reflect actual production conditions. For example, mold temperature and melting temperature directly affect the fluidity and cooling rate of the molten metal; therefore, during the simulation, real-time measurement data from the production line is used as input to dynamically correct model parameters and improve the accuracy of the simulation results.

[0065] Furthermore, the dynamic optimization of the simulation model and the actual production environment can be achieved through intelligent optimization algorithms. Real-time acquired production data provides input for the intelligent optimization algorithm, which automatically adjusts the simulation parameters based on the production data to ensure the model's adaptability to different working conditions. After each simulation, the model output is compared with the actual production data, and if there is a deviation, the parameters in the simulation model will be adjusted accordingly. Through this closed-loop feedback mechanism, the simulation model can be optimized in real time according to changes in actual production, thereby improving simulation accuracy and reducing the difference between idealized assumptions and actual production.

[0066] Machine learning and incremental learning techniques played a crucial role in optimizing simulation results. By training on historical production data, the DeepSeek-NLP model continuously learns from actual production, identifying patterns under different process conditions. Based on these patterns, the simulation tool can more accurately predict future casting results. Especially when facing new materials or special castings, it can automatically infer appropriate process parameters from existing experience, making each simulation result more accurate.

[0067] This technical solution overcomes the problem of mismatch between simulation results and actual conditions in traditional casting process optimization by combining real-time data feedback, production environment simulation, intelligent optimization algorithms, 3D modeling, and physical simulation technology. This significantly improves simulation accuracy and process optimization efficiency. Through this technology, designers can more accurately predict potential defects in castings and optimize design schemes in a timely manner, thereby improving the quality and production efficiency of cast products. Furthermore, the introduction of machine learning and incremental learning allows for continuous accumulation of experience from historical data and optimization of process design, further enhancing simulation accuracy and production efficiency, and ensuring the feasibility and stability of the process optimization scheme in actual production.

[0068] In this embodiment, for the casting process of the fuselage cover, based on the process structure analysis results and preliminary process parameter suggestions output by the DeepSeek-NLP model, a 3D process model is constructed using the 3D modeling software SolidWorks 2023. Then, a physical field simulation is performed using simulation tools (such as CASTsoft), ultimately generating a filling animation, defect cloud map, flow field distribution cloud map, and temperature field cloud map to optimize the casting process design. The specific implementation steps are as follows: S31. 3D Process Model Construction (SolidWorks 2023): Parting Surface Creation: Open the original 3D model of the fuselage cover in SolidWorks and create the parting surface at the position of "Z=25mm" as output by the DeepSeek-NLP model. Divide the model into an upper sand mold (including top blades and some ribs) and a lower sand mold (including bottom structure and T-junction oil passage sand core mating surface); Gating System Modeling: Create the sprue and gating system using the "Rotate" and "Extrude" features according to the dimensions of "sprue φ50mm, gating system 30mm×40mm"; then create the bottom two side ingates with a cross-sectional area of ​​200mm². 2 Gradient to top 120mm 2To meet the requirements, the following steps were taken: First, a gradient section feature for the ingate was created. Second, machining allowance was added: the "shell" feature was used on the top flange face and bottom mounting face, with an offset distance of 2mm. Third, an offset distance of 1.5mm was set on the oil pipe connection face, generating a 3D model of the casting with machining allowance. Fourth, assembly and integration: the gating system model was assembled with the body cover model with machining allowance, ensuring that the ingate was tangent to the bottom sidewall of the body cover, and that all parts of the gating system had good connectivity, thus ensuring the integrity and rationality of the entire casting 3D process model.

[0069] S32. Simulation System Mapping (CASTsoft): Process Parameter Mapping: The "pouring temperature 1575℃" output by the DeepSeek-NLP model is mapped to the "initial molten metal temperature" parameter in the CASTsoft simulation model, and the "holding time 2.5 hours" is mapped to the "holding stage duration" parameter. This allows the preliminary process parameters given by the DeepSeek-NLP model to be directly applied in the simulation, ensuring the accuracy of the simulation process. Material Property Mapping: The material "EN AC-42100" is selected from the CASTsoft material library. Its solidus temperature of 548℃ and liquidus temperature of 613℃ (specific heat capacity, thermal conductivity, etc.) are automatically loaded, along with its thermal properties such as specific heat capacity and thermal conductivity. This mapping ensures that the material properties in the simulation model match the thermal performance of the actual material, improving the accuracy of the simulation results. Geometric Feature and Physical Quantity Correlation: The "blade (3mm wall thickness), rib (gradual wall thickness), and three-way oil passage (uniform wall thickness area)" in the 3D process model are mapped. The geometric features of the "( )" are respectively associated with the "thermal conductivity gradient" in the simulation (the thermal conductivity coefficient is set to 0.15W / (m·K) in the thin region and 0.12W / (m·K) in the uniform region); the dimensions of "straight runner φ50mm, horizontal runner 30mm×40mm" are associated with the "fluid velocity calculation model" to ensure the accuracy of fluid dynamics calculations in the simulation; sand mold and environmental parameter settings: "furan resin self-hardening sand" is selected as the sand mold material, and its density is set to 1600kg / m³. 3 The specific heat capacity is 800 J / (kg・K); the ambient temperature is set to 25℃ to simulate the workshop production environment. This setting ensures that the influence of materials and environment on the casting process is fully considered in the simulation.

[0070] Through the mapping of the aforementioned 3D process model and simulation system, simulation results including filling animation, defect cloud map, flow field distribution cloud map, and temperature field cloud map were generated. The simulation results provide detailed predictions of the casting process, helping to determine whether the flow rate is uniform and whether there are defects such as gas entrapment or incomplete pouring. At the same time, the temperature field simulation results provide detailed temperature distribution for the cooling process of the casting, guiding the optimization of the cooling system design and avoiding cracks or deformation. Through these simulation results, designers can make necessary adjustments to the process, optimize process parameters such as the gating system, parting surface position, and machining allowance, thereby improving the quality and production efficiency of the casting process. In addition, through real-time data feedback and production environment simulation, this application further improves the accuracy of the simulation model, ensuring the feasibility of the optimization scheme in actual production.

[0071] This embodiment combines DeepSeek-NLP models with 3D modeling software to generate accurate 3D process models, and further verifies and optimizes the casting process through physical field simulation. This method can effectively predict potential defects and optimize the design based on simulation results, significantly improving the accuracy and production efficiency of the casting process.

[0072] Step S4: Verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the DeepSeek-NLP model analysis module, and iterative optimization of the process parameters will be triggered to complete the closed-loop optimization of the process flow.

[0073] Based on three core dimensions—defect characteristics, flow field distribution characteristics, and process parameter matching characteristics—the defect size in the defect cloud map is extracted and compared with the quality standard; the flow field cloud map is analyzed to check for abnormal flow velocity areas to ensure compliance with filling uniformity requirements; the actual physical quantities in the simulation are compared one by one with the parameters output by the DeepSeek-NLP model to identify deviations, and the results are fed back to the DeepSeek-NLP model to trigger iterative optimization of process parameters, thus completing the closed-loop optimization of the process flow.

[0074] The requirements for mold filling uniformity include: during the casting process, the flow rate of the molten metal should remain uniform, not less than the first threshold and not greater than the second threshold, with the first threshold being less than the second threshold; the absolute value of the filling time difference between different regions should not exceed the third threshold; during the mold filling process, all regions should be uniformly filled with molten metal, and dead zones or bubble zones are not allowed; the flow channels and gate positions of the gating system should be rationally designed so that the molten metal can be evenly distributed to all parts of the casting; the first threshold, second threshold, and third threshold are numerical ranges dynamically set based on the geometric characteristics, material properties, and quality standards of the specific casting, through simulation data and process experience, to ensure mold filling uniformity and process parameter matching.

[0075] Specifically, to address the mismatch between simulation results and actual production in the casting process, particularly in defect prediction and process parameter optimization, the generated simulation visualization results are verified to ensure that the process parameters and simulation results meet quality standards. An iterative optimization process is then implemented through a feedback mechanism, forming a closed-loop optimization process. Specifically, based on three core dimensions—defect characteristics, flow field distribution characteristics, and process parameter matching characteristics—the simulation results are thoroughly verified, triggering adjustments to the process parameters to further improve the accuracy and production efficiency of the casting process.

[0076] During implementation, the defect features in the defect cloud map are first extracted from the simulation results, and the volume and size of the defects are measured using simulation tools. For example, the volume and size of potential defects such as shrinkage cavities and porosity are extracted from the filling animation and defect annotation layers. These defects are then analyzed in detail using simulation tools and compared with the defect control thresholds set in the generated table. In this way, it can be determined whether the defects exceed the quality standards, and the casting process can be adjusted based on the inspection results. If the defect size is found to be out of standard or does not meet the quality requirements, the cause needs to be analyzed, and the correction requirements are fed back to the table to further trigger iterative optimization of process parameters.

[0077] Secondly, regarding the verification of flow field distribution characteristics, the flow velocity distribution in the flow field cloud map is analyzed first, and areas with abnormal flow velocity are marked. For example, by analyzing the flow velocity in the valve cover area, it is checked whether the flow velocity meets the requirements for uniform filling. If the flow velocity is abnormal, it may lead to gas entrapment or incomplete filling, affecting the quality of the casting. At this stage, the flow velocity is compared with the flow velocity requirements output by the DeepSeek-NLP model to determine whether the flow velocity is within a reasonable range. If it does not meet the requirements, the design of the gating system is adjusted to ensure uniform flow velocity and optimize the casting process.

[0078] Finally, the process parameter matching feature check examines the actual physical quantities in the simulation, such as pouring temperature and holding time, and compares them with the process parameters output by the DeepSeek-NLP model. By accurately comparing these parameters, deviations between process parameters are identified to ensure that the simulation results are consistent with the process design output by the DeepSeek-NLP model. If excessive parameter deviations are found during the comparison process, the corresponding process parameters are adjusted according to the deviation, and the optimization process is restarted.

[0079] After the above tests, when the defects, flow field, or process parameters of the casing cover meet the quantitative anomaly threshold, the structured anomaly data is converted into a format and fed back to the DeepSeek-NLP model. The model uses knowledge graph reasoning to output parameter adjustment suggestions and pre-verify them. After updating the parameters, it is re-simulated and verified, thus completing the closed-loop optimization of the casting process. In this way, the casting process is continuously optimized, thereby improving the quality and production efficiency of the cast products. The DeepSeek-NLP model is part of the AI ​​intelligent analysis module.

[0080] In this embodiment, the feedback mechanism and iterative optimization process linked by the simulation verification module and the DeepSeek-NLP model analysis module achieve closed-loop optimization of the casting process by dynamically adjusting process parameters. Through real-time detection and optimization, the design scheme can be continuously improved, ensuring that the process parameters in each production process match the actual production requirements, thereby improving the stability of the overall production process and product quality. At the same time, through multi-dimensional verification of simulation results, it is ensured that the optimization of each process link has been rigorously verified, minimizing the defects that may occur in the production process and further improving the intelligence level of the casting process.

[0081] This technical solution effectively solves the problem of mismatch between simulation results and actual conditions in existing casting processes, and ensures the efficiency and feasibility of casting process optimization through a multi-level feedback and optimization mechanism.

[0082] In this embodiment, for the optimization of the casting process, the defect characteristics, flow field distribution characteristics and process parameter matching characteristics are checked in multiple dimensions based on the simulation results. The test results are fed back to the DeepSeek-NLP model to trigger iterative optimization of process parameters, so as to ensure closed-loop optimization of the process flow. The specific process is as follows.

[0083] S41. 3D Simulation Execution (CASTsoft): First, submit the casting 3D process model and mapped simulation parameters, run CASTsoft's "Filling + Solidification" simulation, and generate the following simulation graphical results: Filling Animation: Simulates the molten metal flowing from the valve cover and side flange inlet gates, filling the lower sand mold area in 15 seconds, filling the upper sand mold flange and reinforcing rib area in 25 seconds, and completing the entire cavity filling process in 35 seconds; Defect Cloud Map: Predicts a shrinkage cavity with a diameter of φ1.9mm in the hot spot area connecting the valve cover and the inner cavity, located at a distance of Y=300mm from the parting surface, with a size close to the "no shrinkage cavity" quality threshold; Flow Field Distribution Cloud Map: In the valve cover area, the molten metal flow velocity is 1.25m / s, slightly higher than the 1.2m / s suggested by the DeepSeek-NLP model; Temperature Field Cloud Map: The solidification time in the reinforcing rib area is 50 seconds, and the solidification time in the thin-walled area of ​​the inner cavity is 30 seconds. A holding time of 2.5 hours is sufficient to cover the main area's shrinkage compensation.

[0084] S42. Multi-dimensional Result Verification: Defect Feature Verification: By extracting the location of the shrinkage cavity (φ1.9mm) in the hot spot area connecting the valve cover and the inner cavity from the defect cloud map, and comparing it with the "no shrinkage cavity in the body cover" quality requirement in the structured process table, although the defect did not exceed the standard, it was close to the risk edge, indicating that attention is needed; Flow Field Distribution Feature Verification: Analyzing the flow field cloud map, the flow velocity in the valve cover area (1.25m / s) was extracted and compared with the flow velocity of 1.2m / s suggested by the DeepSeek-NLP model. It was found that the flow velocity was too high, and the abnormal flow velocity was the main cause of shrinkage cavity formation, which needs to be optimized and adjusted; Process Parameter Matching Feature Verification: The actual physical quantities in the simulation (such as the casting temperature of 1576℃ and the holding time of 2.5 hours) were retrieved and compared with the parameters of 1575℃ and 2.5 hours output and mapped by S2. The results showed that the deviation was within ±1℃ and 0 hours, the matching was good, and there was no significant deviation in the transmission of process parameters.

[0085] S43. Iterative Optimization and Solution Output: Based on the issue of excessively high flow velocity, the cross-sectional area of ​​the ingate at the valve cover was adjusted. Specifically, the cross-sectional area of ​​the ingate at the valve cover was adjusted from 250mm². 2 Reduced to 237.5mm 2 The adjustment requirement of "(reduced by 5%)" was fed back to the S1 structured process table, and the DeepSeek-NLP model interactive Q&A in S2 was retried. After verification by the DeepSeek-NLP model, it was confirmed that after reducing the ingate cross-sectional area by 5%, the flow velocity could be reduced to 1.2 m / s, reaching a suitable range. Next, the 3D model modification and simulation system mapping in S3 were performed again. After modifying the ingate cross-sectional size to 237.5 mm², the simulation execution and verification in S4 were re-executed. The second simulation results showed: filling animation: filling time extended to 38... The filling process was stable, ensuring uniform filling. Defect cloud map: shrinkage cavities in the hot spot area connecting the valve cover and inner cavity have disappeared, and no defects exceeding the standard were found. Flow field cloud map: the flow velocity in the valve cover area is 1.2 m / s, which conforms to the 1.1-1.2 m / s velocity range recommended by the DeepSeek-NLP model. Final solution output: After repeated optimization and verification, the final verified casting process solution includes the following: parting line position: Y=300mm; center-pour gating system design (cross-sectional area at the inner gate valve cover / side flange: 237.5 mm²). 2 Gating system: 50mm sprue, 40mm top × 50mm bottom × 30mm height of runner; Machining allowance: 4mm for valve cover, side flange, and inner cavity wall; Process parameters: Pouring temperature 1575℃, holding time 2.5 hours, pouring speed 1.1-1.2m / s; Sand mold permeability: 800cm 3 / min.

[0086] Through the optimization process of this embodiment, a closed loop of the entire process from casting drawings to digital models, simulation verification, and optimization design is realized. In this process, the DeepSeek-NLP model learns structured data, accurately outputs process parameters, and combines 3D simulation and iterative optimization to ensure the reliability of casting quality and the efficiency of process design, further improving the intelligence and refinement of the casting process.

[0087] Through the coordination of the above steps, this application improves the product quality and production efficiency of castings.

[0088] In summary, this application provides a method for optimizing the entire intelligent casting process, which effectively solves the problems of low efficiency and high cost in traditional casting process design. By combining the DeepSeek-NLP model with deep learning technology, automation and intelligence are achieved from casting drawings to process parameter design, significantly improving the accuracy and precision of casting process design. Based on data structuring, knowledge graph construction, and multi-round question-answering analysis, the DeepSeek-NLP model can quickly analyze and optimize process parameters, avoiding deviations caused by human intervention. Furthermore, by combining with simulation tools, it provides comprehensive simulation and prediction of the casting process, ensuring the feasibility of the design scheme. In addition, this application improves the model's adaptability to new materials and special working conditions through reinforcement learning and incremental learning mechanisms, ensuring the real-time and continuous nature of process optimization. Through this series of innovative technologies, this application not only optimizes the casting process design process but also improves product quality and production efficiency, reduces trial-and-error costs, and promotes the intelligent transformation of the casting industry, demonstrating significant application value and market prospects.

[0089] Example 2: The above describes a method for optimizing the entire intelligent casting process in an embodiment of this application. The following describes a system for optimizing the entire intelligent casting process in an embodiment of this application. Please refer to [link / reference]. Figure 4 The intelligent casting process optimization system 100 in this application embodiment includes: The digital analysis module 101 is used to digitally analyze casting drawings and process flow documents to generate structured process tables. AI intelligent analysis module 102 is used to perform multi-round question and answer based on structured process tables and DeepSeek-NLP model to output process structure analysis results and preliminary process parameter suggestions. The 3D modeling and simulation module 103 is used to construct a 3D process model based on the process structure analysis results and preliminary process parameter suggestions, and to simulate the casting process based on simulation tools to generate simulation results. The feedback optimization module 104 is used to verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the AI ​​intelligent analysis module, triggering iterative optimization of the process parameters to complete the closed-loop optimization of the process flow.

[0090] Through the synergistic cooperation of the aforementioned components, the product quality and production efficiency of the castings have been further improved.

[0091] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0092] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for optimizing the entire intelligent casting process, characterized in that, The method includes: Step S1: Digitally analyze the casting drawings and process flow documents to generate structured process tables; Step S2: Based on the structured process table, use the DeepSeek-NLP model to perform multiple rounds of question and answer, and output the process structure analysis results and preliminary process parameter suggestions; Step S3: Based on the process structure analysis results and the preliminary process parameter suggestions, construct a three-dimensional process model, and simulate the casting process using simulation tools to generate simulation results; Step S4: Verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the AI ​​intelligent analysis module, and iterative optimization of the process parameters will be triggered to complete the closed-loop optimization of the process flow.

2. The intelligent casting process whole-process optimization method according to claim 1, characterized in that, Step S1 further includes: The geometric contours, dimensions, and tolerance requirements of the casting drawings are digitally analyzed using computer vision technology and transformed into a set of structured geometric parameters that can be recognized by a computer. Natural language processing technology is used to extract the process execution steps, parameter ranges, and quality control standards from the process flow document, and transform them into a standardized set of process parameters and rules.

3. The intelligent casting process whole-process optimization method according to claim 2, characterized in that, Step S1 further includes: Based on the set of structured geometric parameters and the set of standardized process parameters and rules, after data mapping and format conversion, they are integrated into a computer-recognizable structured process table containing process execution steps, parameter ranges and quality control standards, forming a digital database of the original casting process.

4. The intelligent casting process whole-process optimization method according to claim 1, characterized in that, Step S2 further includes: Based on the structured process table, the DeepSeek-NLP model analyzes the process structure and parameters through multiple rounds of question-and-answer sessions, gradually optimizing the parting line position, gating system configuration, and preliminary process parameter suggestions, and outputs the final process design scheme. In the process of multiple rounds of question-and-answer sessions, the DeepSeek-NLP model combines process logic, parameter association rules, and quality constraints to gradually adjust the output process parameters, ensuring that the final design scheme meets the casting quality standards.

5. The intelligent casting process whole-process optimization method according to claim 4, characterized in that, Gradually adjust the output process parameters to ensure that the final design meets the casting quality standards, which also includes: The DeepSeek-NLP model combines process logic, parameter association rules, and quality constraints to automatically adjust the output process parameters during each round of question-and-answer sessions, ensuring that the final design scheme meets the casting quality standards. Furthermore, when dealing with new materials, new technologies, or special working conditions, it automatically absorbs new data and updates the knowledge graph through an incremental learning mechanism.

6. The intelligent casting process whole-process optimization method according to claim 1, characterized in that, Step S3 further includes: Based on the process structure analysis results and the preliminary process parameter recommendations, a three-dimensional process model is constructed using three-dimensional modeling software, and physical field simulation is performed using simulation tools to generate simulation results such as filling animation, defect cloud map, flow field distribution cloud map, and temperature field cloud map.

7. The intelligent casting process whole-process optimization method according to claim 6, characterized in that, The three-dimensional process model includes the gating system design, parting surface location, and process parameters. The process parameters output by the DeepSeek-NLP model are mapped into the simulation model. The simulation calculation is performed by combining material properties, geometric features, and physical quantities to optimize the casting process design scheme to adapt to the actual production environment.

8. The intelligent casting process whole-process optimization method according to claim 1, characterized in that, Step S4 further includes: Based on three core dimensions—defect characteristics, flow field distribution characteristics, and process parameter matching characteristics—the defect size in the defect cloud map is extracted and compared with the quality standard. Analyze the flow field cloud map, check for areas of abnormal flow velocity, and ensure that the filling uniformity requirements are met; The actual physical quantities in the simulation are compared one by one with the parameters output by the DeepSeek-NLP model to identify deviations. The results are then fed back to the DeepSeek-NLP model to trigger iterative optimization of process parameters and complete the closed-loop optimization of the process flow.

9. The intelligent casting process whole-process optimization method according to claim 8, characterized in that, The requirements for filling uniformity include: During the casting process, the flow rate of the molten metal should be kept uniform, and should not be less than the first threshold or greater than the second threshold, wherein the first threshold is less than the second threshold. The absolute value of the filling time difference between each region does not exceed the preset third threshold; During the filling process, all areas should be evenly filled with molten metal, and dead zones or air bubbles are not allowed. By rationally designing the flow channels and gate positions of the gating system, the molten metal can be evenly distributed to all parts of the casting.

10. A smart casting process whole-process optimization system, used to implement the smart casting process whole-process optimization method as described in any one of claims 1-9, characterized in that, The system includes: The digital analysis module is used to digitally analyze casting drawings and process flow documents to generate structured process tables. The AI ​​intelligent analysis module is used to perform multi-round question and answer based on the structured process table using the DeepSeek-NLP model, and output process structure analysis results and preliminary process parameter suggestions. The 3D modeling and simulation module is used to construct a 3D process model based on the process structure analysis results and the preliminary process parameter suggestions, and to simulate the casting process using simulation tools to generate simulation results. The feedback optimization module is used to verify the simulation results. If there are defects that do not meet the quality standards or mismatched process parameters, the correction requirements will be fed back to the AI ​​intelligent analysis module, triggering iterative optimization of the process parameters to complete the closed-loop optimization of the process flow.

Images