A method and apparatus for determining blade machining parameters

Abstract

The application provides a blade machining parameter determination method and device. The method provided by the application integrates various types of data related to complex curved surface machining, obtains preprocessed data through preprocessing, constructs a basic knowledge graph and a causal relationship graph based on the preprocessed data, trains a large model using the preprocessed data and the knowledge graph, and obtains a trained large model. The trained large model is used to extract features of each piece of data in the machining preparation stage and actual machining data, obtain corresponding feature vectors, and construct a feature vector library based on attribute information and the feature vectors of each piece of data. When determining a machining track of a blade to be machined, a target feature vector of the blade is extracted using the large model, a matching feature vector matching the target feature vector is searched from the feature vector library, attribute information of the matching feature vector is analyzed, corresponding target machining parameters are extracted from a target knowledge graph, and the target machining parameters and the matching feature vector are input into the large model for optimization to obtain optimized machining parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blade processing, and in particular to a blade processing parameter determination method and device. BACKGROUND

[0002] In the industrial field, especially in aviation, energy and mechanical engineering, blades have very wide applications. When processing blades, titanium alloy or high-temperature alloy is usually used to manufacture blades. Such materials are known for their excellent mechanical properties, but also bring challenges in processing. For example, in the process of cutting blades, large cutting force and rapid tool wear are generated, which easily causes processing deformation and vibration, thereby affecting the dimensional accuracy and surface quality of the parts.

[0003] In order to solve this problem, in the traditional blade processing method, a phase change material such as a low-melting-point alloy is used to fill and assist support. Although this helps to improve some processing problems, it also has problems such as decreased positioning accuracy, complex production preparation, residual on the surface of the workpiece and difficult cleaning, and the environmental conditions during pouring and melting are relatively harsh, resulting in low precision of the generated blade. Therefore, there is an urgent need for an optimization method for blade processing parameters to ensure that the produced blade meets the requirements of high precision and high quality. SUMMARY

[0004] Therefore, the present application provides a blade processing parameter determination method and device to accurately determine the processing parameters of the blade to improve the processing precision and quality of the blade.

[0005] Specifically, the present application is realized by the following technical solutions:

[0006] The first aspect of the present application provides a blade processing parameter determination method, which comprises:

[0007] Obtaining various types of data related to complex surface processing; wherein the various types of data reflect the surface data, feature data and machining data of the blade;

[0008] Preprocessing the various types of data to obtain preprocessed data, and constructing a basic knowledge graph related to the field of blade processing according to the preprocessed data; the basic knowledge graph takes the blade, material, tool, quality control, processing technology and processing trajectory as nodes, and the inherent characteristics of the blade, material, tool, quality control, processing technology and processing trajectory as node attributes;

[0009] According to the preprocessed data, a causal relationship graph related to the field of blade processing is constructed, and the basic knowledge graph is updated using the causal relationship graph to obtain a target knowledge graph;

[0010] train a large model by using the preprocessed data and the target knowledge graph, so that the large model learns knowledge in the blade machining field, and obtain a trained large model;

[0011] For each piece of data in the machining preparation stage data and the actual machining data, the trained large model is used to extract features of the piece of data, to obtain a feature vector of the piece of data, and a feature vector library is constructed based on the attribute information and the feature vector of each piece of data;

[0012] When a machining trajectory of a to-be-machined blade needs to be determined, a target feature vector of the to-be-machined blade is extracted by using the trained large model, and a matching feature vector matching the target feature vector is searched from the feature vector library;

[0013] According to the attribute information of the matching feature vector, a target machining parameter corresponding to the attribute information is searched from the target knowledge graph, and the matching feature vector and the target machining parameter are input into the trained large model, so that the target machining parameter is optimized by the trained large model, to obtain an optimized machining parameter.

[0014] The second aspect of the application provides a blade machining parameter determination device, the device comprising an acquisition module, a processing module, a construction module, a training module and an optimization module, wherein,

[0015] The acquisition module is configured to acquire various types of data related to complex curved surface machining; wherein the various types of data reflect curved surface data, feature data and machining data of a blade;

[0016] The processing module is configured to preprocess the various types of data to obtain preprocessed data, and construct a basic knowledge graph related to the blade machining field according to the preprocessed data; the basic knowledge graph takes a blade, a material, a tool, quality control, a machining process and a machining trajectory as nodes, and takes inherent characteristics of the blade, the material, the tool, the quality control, the machining process and the machining trajectory as node attributes;

[0017] The construction module is configured to construct a causal relationship graph related to the blade machining field according to the preprocessed data, and update the basic knowledge graph by using the causal relationship graph, to obtain a target knowledge graph;

[0018] The training module is configured to train a large model by using the preprocessed data and the target knowledge graph, so that the large model learns knowledge in the blade machining field, and obtain a trained large model;

[0019] The construction module is configured to, for each piece of data in the machining preparation stage data and the actual machining data, extract features of the piece of data by using the trained large model to obtain a feature vector of the piece of data, and construct a feature vector library based on attribute information and the feature vector of each piece of data.

[0020] The acquisition module is configured to, when it is necessary to determine a machining trajectory of a to-be-machined blade, extract a target feature vector of the to-be-machined blade by using the trained large model, and search for a matching feature vector matching the target feature vector from the feature vector library.

[0021] The optimization module is configured to search for a target machining parameter corresponding to the attribute information from the target knowledge graph according to the attribute information of the matching feature vector, and input the matching feature vector and the target machining parameter into the trained large model to optimize the target machining parameter by the trained large model to obtain an optimized machining parameter.

[0022] The blade machining parameter determination method and device provided in the application integrate various types of data related to complex curved surface machining, pre-process the various types of data to obtain pre-processed data, construct a basic knowledge graph and a causal relationship graph according to the pre-processed data, train a large model by using the pre-processed data and a target knowledge graph to obtain a trained large model, then extract features of each piece of data in the machining preparation stage and the actual machining data by the trained large model to obtain corresponding feature vectors, construct a feature vector library based on attribute information and the feature vectors of each piece of data, further extract a target feature vector of a blade by the large model when it is necessary to determine a machining trajectory of a to-be-machined blade, search for a matching feature vector from the feature vector library, extract a corresponding target machining parameter from the target knowledge graph by analyzing attribute information of the matching feature vector, input the matching feature vector and the target machining parameter into the large model for further optimization to obtain an optimized machining parameter. In this way, firstly, the knowledge graph of blade machining is constructed, and the large model is trained in combination with the causal relationship, so that the large model not only remembers knowledge, but also reasons the relationship between the machining parameter and the machining effect, improves the explainability of the knowledge, further uses the trained large model to extract features of the machining preparation stage and the actual machining data, and constructs a feature vector library, which enables the model to perform in-depth analysis before and after machining. In addition, the matching of the knowledge graph and the feature vector library realizes the extraction of effective machining parameters from similar historical data, and further optimization in the large model ensures that the output parameter is more suitable for the current machining requirement, and the machining parameter can be accurately determined. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1A flow chart of the blade machining parameter determination method embodiment one provided in the present application;

[0024] Figure 2 A flow chart of the blade machining parameter determination method embodiment two provided in the present application;

[0025] Figure 3 A schematic diagram of the sub-body model of the blade shown in an exemplary embodiment of the present application;

[0026] Figure 4 A schematic diagram of the sub-body model of the material shown in an exemplary embodiment of the present application;

[0027] Figure 5 A schematic diagram of the sub-body model of the tool shown in an exemplary embodiment of the present application;

[0028] Figure 6 A schematic diagram of the sub-body model of the quality control shown in an exemplary embodiment of the present application;

[0029] Figure 7 A schematic diagram of the sub-body model of the machining process shown in an exemplary embodiment of the present application;

[0030] Figure 8 A schematic diagram of the sub-body model of the machining track shown in an exemplary embodiment of the present application;

[0031] Figure 9 A structural schematic diagram of the blade machining parameter determination device embodiment one shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0032] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The following description is made with reference to the accompanying drawings in which like reference numerals refer to like elements in the several figures. The following exemplary embodiments described in the following description do not represent all of the implementations consistent with the present application.

[0033] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the present application, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0034] It should be understood that, although the terms first, second, third, etc. can be employed in this application to describe various information, the information should not be limited to these terms. These terms are only used to distinguish one type of information from another type of information. For example, a first information can also be termed a second information, similarly, a second information can also be termed a first information without departing from the scope of the present application. Depending on the context, the word "if' as used herein can be interpreted to mean "when" or "in response to determining".

[0035] The following specific embodiments are given to introduce the technical solutions of the present application in detail.

[0036] Figure 1 The flow chart of the blade machining parameter determination method embodiment one provided in the present application is shown in the following figure. Figure 1 The blade machining parameter determination method provided in the embodiment includes:

[0037] S101, acquiring various types of data related to complex curved surface machining; wherein the various types of data reflect curved surface data, feature data and machining data of the blade.

[0038] Specifically, the various types of data include blade design data, simulation data, machining preparation stage data, actual machining data, quality control data and maintenance improvement data.

[0039] In an embodiment, the various types of data related to complex curved surface machining can be acquired according to the following steps:

[0040] Step 1: determine the data collection type as structured text, semi-structured text, unstructured text, table, literature, picture, audio and / or video and / or webpage;

[0041] Step 2: determine the data sources including CAD models and design drawings in the product design stage, blade geometric parameters, material properties, CFD analysis results in the simulation simulation stage, FEA stress and strain prediction, thermodynamic performance evaluation, tool selection parameters in the machining accuracy stage, machining tool specifications, NC programming information, real-time machining parameters, online measurement data, temperature monitoring data in the machining process, final product quality detection report, defect record, customer feedback in the quality control stage, fault history record, maintenance log, improvement measures document in the maintenance and improvement stage;

[0042] Step 3: determine the data collection method including direct collection, system docking, manual import and / or input;

[0043] Step 4: collect the various types of data indicated by the data collection type from the data sources according to the data collection method.

[0044] In summary, through the above steps, the collection type, data source and collection method of various data related to complex curved surface machining can be determined, and various data required in the complex curved surface machining process can be obtained from multiple links (such as design, simulation, machining and quality control, etc.), so that the subsequent machining process of the blade according to the obtained various data can be optimized, and strong data support can be provided for selecting suitable machining parameters to obtain the required blade.

[0045] It should be noted that various data related to complex curved surface machining include curved surface data, feature data and machining data of the blade, and the curved surface data, feature data and machining data respectively represent technical information in different aspects in the machining process, and each type of data can provide support and decision basis in different stages and links of machining.

[0046] Specifically, the curved surface data includes information related to the collective shape, surface feature and curved surface structure of the blade, wherein the collective shape of the blade includes the blade root, blade tip, blade profile, film hole, cooling channel, damping hole, damping platform, tenon slot, tenon, leading edge, trailing edge, blade basin and blade back, etc. of the blade; further, the surface feature of the blade includes surface feature information of each collective shape part of the blade, for example, whether the blade tip is sharp or smooth, the length and width information of the blade tip; further, the curved surface structure of the blade includes the specific structure at each part of the blade, for example, the connection mode of the blade root is through inlay milling connection, key connection or bolt connection, and the curvature radius of the blade back, etc.

[0047] Further, the feature data of the blade refers to data describing the functional, structural or operational characteristics of the blade, wherein the feature data of the blade includes material properties and thermodynamic performance, etc. of the blade. For example, the manufacturing material of the blade can be steel, cast iron or non-ferrous metal, and further, the steel material, cast iron material and non-ferrous metal material for manufacturing the blade can also select specific refined materials according to different requirements. Among them, the steel material is divided into tool steel, alloy steel, medium alloy steel and high-quality steel, etc.; the cast iron material includes forgeable cast iron, wear-resistant cast iron and ordinary gray cast iron, etc.; the non-ferrous metal material includes hard metal, single crystal high-temperature metal and conformal material, etc.

[0048] The machining data of the blade relates to information related to the actual machining of the blade, including selection parameters of a tool for cutting the blade, a cutting speed, specifications and settings of a machining machine tool, and actual machining parameters. The selection parameters of the tool include a material classification, a use classification, and a structure classification of the tool. For example, the material classification of the tool can include high-speed steel, ceramic, and hard metal. According to the use classification of the tool, the tool can be classified into a turning tool, a hole machining tool, a broach, a milling cutter, and a gear tool. According to the structure classification of the tool, the tool can be classified into a whole tool, a machine clamp tool, a composite tool, and an insert tool. It should be noted that, in the specific blade machining process, appropriate blade machining data can be set according to the curved surface data, the feature data, and the blade requirements of the blade, so as to obtain a blade meeting the preset requirements. In this application, the blade machining data is not limited.

[0049] S102, pre-processing the various types of data to obtain pre-processed data, and constructing a basic knowledge graph related to the blade machining field according to the pre-processed data; the basic knowledge graph takes a blade, a material, a tool, quality control, a machining process, and a machining trajectory as nodes, and takes inherent characteristics of the blade, the material, the tool, the quality control, the machining process, and the machining trajectory as node attributes.

[0050] It should be noted that the pre-processing method of the various types of data can be selected according to actual needs, and the application does not limit the pre-processing method. For example, in an embodiment, the pre-processing process can include obtaining key structured information from unstructured text, cleaning noise and interference information in text data, removing garbled code and illegal characters in structured data, repairing syntax errors and / or logically incoherent sentences, and the like.

[0051] Figure 2 For the flowchart of the blade machining parameter determination method embodiment two provided in the application, please refer to Figure 2 Based on the embodiment one, the method provided in the embodiment can include the following process in the process of constructing a basic knowledge graph related to the blade machining field according to the pre-processed data:

[0052] S201, identifying entities in the blade machining field and attributes corresponding to each entity based on the pre-processed data.

[0053] In this step, after pre-processing the various types of data, the entities in the blade machining field and the attributes corresponding to each entity are identified.

[0054] In a specific implementation, the key entities in the blade processing field can be extracted from the preprocessed data based on rule matching, machine learning-based NER models, or pre-trained language models (such as BERT), and further, the attributes of each entity can be identified based on statistical analysis combined with TF-IDF, word embedding, and other technologies.

[0055] In this embodiment, the entities in the blade processing field include blade, material, tool, quality control, processing technology, and processing trajectory, which are important components in the blade processing process. Further, each entity has multiple attributes to describe the properties of the entity. For example, the attributes of the blade include geometric shape, blade type, etc., and the attributes of the tool include tool type and tool material.

[0056] For example, in an embodiment, the attributes of the blade include Gaussian curvature, average curvature, principal curvature, torsion, slope, normal direction, tangent direction, area, volume, connectivity, topological type, continuity, fairness, boundary, characteristic line, characteristic point, shape description, and machining accuracy, where the topological type includes closed surface, open surface, simply connected surface, and multiply connected surface. The attributes of the processing trajectory can include linear trajectory, curve trajectory, spiral trajectory, random trajectory, and hybrid trajectory. The attributes of the tool can include tool name, tool number, tool angle, tool diameter, tool durability, and / or tool life, etc.

[0057] In S202, based on the preprocessed data, the first hierarchical relationship between entities and the second hierarchical relationship between attributes of different entities are analyzed and determined, and for each entity, the hierarchical relationship between its attributes is analyzed and determined, obtaining a sub-ontology model corresponding to each entity.

[0058] In this step, based on the preprocessed data, the first hierarchical relationship between entities and the second hierarchical relationship between attributes of different entities are analyzed and determined. The first hierarchical relationship is the hierarchical relationship or association between entities, for example, there is a first hierarchical relationship between processing technology and tool selection, where processing technology is one of the core entities, tool selection is closely related to processing technology, and there is a clear dependency relationship between them. In a certain processing technology, a specific type of tool is required, therefore, there is a first hierarchical relationship between processing technology and tool selection. Further, the second hierarchical relationship between attributes of different entities is the interaction relationship between attributes of different entities, for example, tool material is an attribute of the entity tool, and cutting speed is an attribute of the entity processing technology. Between them, tool material directly affects the applicable cutting speed range, that is, different tool materials will affect the selection of cutting speed in the processing technology, and there is a second hierarchical relationship of dependency between them.

[0059] In a specific implementation, the first-level relationship of the entity and the second-level relationship between the attributes of different entities can be extracted based on a rule-based method (e.g., dependency syntax analysis), a machine learning-based method, or a knowledge reasoning-based method.

[0060] Further, in a possible implementation, the step of analyzing and determining the hierarchical relationship between the attributes of each entity to obtain the sub-ontology model corresponding to each entity can include:

[0061] S2021, for the blade, according to the geometric shape, physical property, processing method, optimization target and application field of the blade, the characteristics of the blade are classified in detail, and the relationship between the attributes in each category is determined, the hierarchical relationship is established according to the classification result and the attribute relationship, and the reasoning rule is formulated according to the above classification and attribute relationship, so as to obtain the sub-ontology model of the blade.

[0062] In this step, the blade is classified according to its characteristics, wherein the geometric shape of the blade includes the size and shape of the blade, and the physical property of the blade includes the strength and density of the blade, and the processing method of the blade is determined by how the blade is processed. Then, by analyzing the interaction between these characteristics, the dependency relationship between each attribute is determined, and these dependency relationships are connected through certain reasoning rules, for example, the strength of the blade material will affect the design of the blade geometric shape. In this way, the attributes and mutual relationships of the blade can be clearly presented through the hierarchical structure between them, thereby forming the sub-ontology model of the blade and providing effective data support for subsequent decision-making and optimization.

[0063] Figure 3 A schematic diagram of the sub-ontology model of the blade is shown for an exemplary embodiment of the present application. Please refer to Figure 3 As can be seen from Figure 3 , in the sub-ontology model of the blade, the attributes under the geometric shape of the blade include shape and size, the attributes under the physical property of the blade include strength and density, the attributes under the processing method of the blade include milling and laser cutting, the attributes under the optimization target of the blade include strength and corrosion resistance, and the attributes under the application field of the blade include aviation and wind power generation; further, in the sub-ontology model, the geometric shape, physical property, processing method, optimization target and application field of the blade are taken as top-level nodes, and they form different hierarchical sub-nodes through dependency relationship, reflecting the interaction and dependency in the blade design and optimization process, for example, the geometric shape (such as size and shape) of the blade will affect the selection of its physical properties (such as strength and density), and at the same time, these physical properties also determine the processing method of the blade.

[0064] S2022, for the material, classify the tool material and the workpiece material according to the category, and classify each material after the category, and determine the relationship between each attribute under each category after the manufacturing process classification, and finally detect the hierarchical relationship according to the classification result and the attribute relationship to obtain the sub ontology model of the material.

[0065] It should be noted that the category of tool material and workpiece material includes metal material and alloy material, and each material also has a corresponding specific manufacturing process, such as casting, forging and welding, etc., and then for each manufacturing process, by analyzing the physical and chemical properties of the tool material, the change of the specific properties of the material under different processes can be determined. Further, the relationship between the properties and attributes of the material under different manufacturing processes is analyzed to form a hierarchical structure, ensuring that each material and its related attributes are clearly linked to ultimately obtain a sub ontology model that can reflect the relationship between material properties and processes. For example, when using aluminum alloy as the manufacturing material, the manufacturing processes that can be used during manufacturing include casting, forging and welding. Different manufacturing processes will have different effects on the properties of aluminum alloy products, including strength, hardness, plasticity and corrosion resistance. Specifically, when using casting process to process aluminum alloy material, the strength property of the obtained aluminum alloy product is lower, the hardness property is lower, and the plasticity property is higher. That is, there is no significant difference between the strength property and the hardness property of the aluminum alloy product obtained by casting process and the strength property and the hardness property of the aluminum alloy material before casting process. However, the plasticity of the aluminum alloy product after casting process is higher than that of the aluminum alloy material before casting process. Similarly, when using forging process to process aluminum alloy material, the strength property of the obtained aluminum alloy product is higher, the hardness property is higher, and the plasticity property is moderate. That is, the strength of the aluminum alloy product obtained by forging process is greatly enhanced compared with the strength property and the hardness property of the aluminum alloy material before forging process. However, the plasticity of the aluminum alloy product after forging process is improved to a certain extent compared with the plasticity of the aluminum alloy material before forging process. When using welding process to process aluminum alloy material, the strength property of the welding area of the obtained aluminum alloy product is higher, the hardness property is moderate, and the corrosion resistance property is lower. That is, the welding area of the aluminum alloy product obtained by welding process has lower strength, a certain degree of improved hardness and basically the same corrosion resistance compared with the aluminum alloy material before welding process. In this way, the ontology model of aluminum alloy material is obtained, which can clearly understand the effect of different processes on the performance of aluminum alloy, so as to select a more scientific manufacturing strategy in actual application.

[0066] Figure 4For the schematic diagram of the sub ontology model of the material shown in an exemplary embodiment of the present application, please refer to Figure 4 , Figure 4 In the sub ontology model in the material category of aluminum alloy material, after selecting aluminum alloy material as the tool and the workpiece material, manufacturing processes such as casting, forging and welding can be selected for processing. When the casting process is selected for processing, the material properties of the aluminum alloy material include strength properties, hardness properties and corrosion resistance properties, which specifically make the tool and the process obtained by manufacturing have low strength, low hardness and low corrosion resistance.

[0067] S2023, for the tool, the numerical control machining tool is classified according to the processing technology, and each tool after classification is functionally classified, and the relationship between each attribute under each category after functional classification is determined, and finally the hierarchical relationship is established according to the classification result and the attribute relationship to obtain the sub ontology model of the tool.

[0068] Specifically, the numerical control machining tool can be classified into milling cutter, turning tool and drill bit according to the processing technology. Taking the milling cutter as an example, the milling cutter can be divided into cutting milling cutter and feeding milling cutter according to the functional classification of the milling cutter under different conditions. Further, the attributes of the cutting milling cutter include cutting force, surface quality, tool life and cutting depth, and the attributes of the feeding milling cutter include cutting force, surface quality, tool life and feeding speed. In this way, the hierarchical relationship can be established according to the classification result and the attribute relationship of the milling cutter, and the sub ontology model of the milling cutter can be obtained, thereby providing a basis for the selection of the tool.

[0069] Figure 5 For the schematic diagram of the sub ontology model of the tool shown in an exemplary embodiment of the present application, please refer to Figure 5 , Figure 5 Taking the milling cutter in the numerical control machining tool as an example, the milling cutter is classified into cutting milling cutter and feeding milling cutter according to the processing technology, and the cutting force, surface quality, tool life and cutting depth are defined under the cutting milling cutter classification, and the cutting force, surface quality, tool life and feeding speed are defined under the feeding milling cutter classification, and the hierarchical structure relationship between the classification result of the tool and the attributes under each tool definition is clear.

[0070] S2024, for quality control, taking quality control as the core entity, taking detection method, quality standard, error range and detection frequency as the attributes of the entity, and taking detection method including detection means, detection tool, detection program and detection environment, quality standard including international standard, national standard, industry standard, enterprise standard and technical specification, error range including absolute error, relative error, quality uncertainty, tolerance range, process error and detection error, and detection frequency including time frequency, sampling ratio, dynamic adjustment and key point detection, a sub ontology model with quality control as the core entity is established.

[0071] In this step, the specific sub-attributes of each attribute for entity quality control can be selected according to actual needs, and this application does not limit them. For example, in the testing method, the testing means include electrical testing and functional testing, the testing tools include a multimeter and an oscilloscope, the testing procedure includes performing electrical testing first, followed by functional testing, and the testing environment is under the conditions of room temperature (between 20° and 25°) and humidity less than 70%.

[0072] Figure 6 This is a schematic diagram of a sub-ontology model for quality control shown in an exemplary embodiment of this application. Please refer to... Figure 6 , Figure 6 Taking the detection method attribute in the core entity as an example, the detection method includes detection means, detection tools, detection procedures and detection environment. By refining these contents and through the hierarchical relationship between the attribute detection method and sub-attributes, a systematic and specific detection method can be provided for actual quality control work to determine the detection quality.

[0073] S2025. For machining processes, a sub-ontology model with machining process as the core entity is established, and the machining process is divided into the material to be processed, process, step, cutting speed, depth of cut and feed rate as the attributes of the entity. The type of material to be processed is used as the attribute of the material to be processed.

[0074] In this step, the materials to be processed include metallic materials, non-metallic materials, composite materials, and special materials; the processes include roughing, finishing, and ultra-finishing; the steps include cutting, grinding, and electrical discharge machining; the cutting speeds include low, medium, and high cutting speeds; the depth of cut includes small, medium, and large depths of cut; and the feed rate includes low, medium, and high feed rates. Assuming a milling operation is being performed on an aluminum alloy, the above properties can be used to construct a sub-body model for this machining process. For example, the material to be processed in the machining process is an aluminum alloy, the process is roughing, the step is cutting, the cutting speed is medium (100 m / min to 150 m / min), the depth of cut is medium (2 mm to 4 mm), and the feed rate is medium (0.1 mm / rev to 0.2 mm / rev).

[0075] It should be noted that a feed rate of 0.1 mm / rev means that for every revolution of the tool, the tool will advance 0.1 mm along the cutting direction of the workpiece, and a feed rate of 0.2 mm / rev means that for every revolution of the tool, the tool will advance 0.2 mm along the cutting direction of the workpiece.

[0076] Figure 7For a schematic diagram of the sub ontology model of the processing process shown in an exemplary embodiment of the present application, please refer to Figure 7 , Figure 7 From the core entity processing process, layer by layer to each specific processing link, through this hierarchical structure, the relationship between different processing processes and their attributes can be clearly understood, and the relationship between each level structure can be intuitively displayed.

[0077] S2026, for the processing trajectory, the processing trajectory is classified according to the processing trajectory type, the hierarchical relationship of the processing trajectory is obtained, and for each type of processing trajectory, the scene suitable for the processing trajectory and the tool suitable for the processing trajectory are taken as the attributes of the sub ontology model with the processing trajectory as the core entity.

[0078] Specifically, according to the classification of the processing trajectory, the processing trajectory can be classified into straight line trajectory, circular arc trajectory, spiral trajectory and arbitrary curve trajectory, wherein the scene suitable for the straight line trajectory is processing plane, groove and hole, and the tool suitable for the straight line trajectory is milling cutter and turning tool; the scene suitable for the circular arc trajectory is processing curved surface and inner and outer circle, and the tool suitable for the circular arc trajectory is milling cutter, turning tool and tool radius compensation tool; the scene suitable for the spiral trajectory is processing thread and hole spiral structure, and the tool suitable for the spiral trajectory is milling cutter, thread tool and drill bit; the scene suitable for the arbitrary curve trajectory is processing free curve, complex ink mold and special surface, and the tool suitable for the arbitrary curve trajectory is 5-axis milling cutter, electrochemical machining tool and special tool.

[0079] Figure 8 For a schematic diagram of the sub ontology model of the processing trajectory shown in an exemplary embodiment of the present application, please refer to Figure 8 From Figure 8 It can be seen that the core entity processing trajectory is classified into straight line trajectory, circular arc trajectory, spiral trajectory and arbitrary curve trajectory, and each processing trajectory attribute includes sub attribute suitable scene and suitable tool, Figure 8 In the straight line trajectory, the scene suitable for the straight line trajectory includes processing plane, groove and hole, and the tool suitable for the straight line trajectory is milling cutter and turning tool, so that different processing trajectories can be adopted according to different scenes and tools, the hierarchical structure is clear, and the appropriate processing trajectory can be selected according to the sub ontology model of the processing trajectory.

[0080] S203, according to the first hierarchical relationship, the sub ontology model corresponding to each entity and the second hierarchical relationship, the ontology model of the blade processing field is obtained; the ontology model is used to represent the hierarchical relationship and the mutual relationship between the entities and the attributes.

[0081] In this step, the first hierarchical relationship, the sub ontology model corresponding to each entity and the second hierarchical relationship can be fused to obtain the ontology model of the blade processing field, and the ontology model can accurately represent the hierarchical structure of the blade processing field.

[0082] S204, identify entity, attribute and value triple data from the pre-processed data based on the ontology model, and construct a basic knowledge graph of the blade processing field by using the identified triple data; wherein the basic knowledge graph represents the blade field knowledge through entities and the relationship between entities, and forms a fine-grained relationship and attribute description through triple data, which can accurately represent various entities, attributes and their relationships in the blade processing field.

[0083] Specifically, in the ontology model, the hierarchical relationship and interaction between each entity and attribute are explicitly represented. Then, by extracting entities, attributes and values from the pre-processed data, specific triple data is formed, wherein the entities include processing technology, tool selection, tool material, etc., the attributes include cutting speed, feed rate, tool material and hardness, etc., and the values include cutting speed 150 m / min, feed rate 150 m / rev, etc. The formed triple can be (processing technology, cutting speed, 150 m / min), (tool selection, tool material, hard alloy), (processing technology, feed rate, 0.05 mm / rev), etc.

[0084] Further, after obtaining the triple data, the basic knowledge graph of the blade processing field can be constructed by the triple data, and a fine-grained relationship and attribute description is formed by the triple data, so that various entities, attributes and their relationships in the blade processing field can be accurately represented.

[0085] It should be noted that based on the knowledge graph, the processing parameters can be determined, for example, when selecting a suitable processing trajectory for a workpiece that needs to be processed on a complex surface, the workpiece may include a complex free-form surface and needs to be precisely three-dimensionally cut. Therefore, an arbitrary curve trajectory is selected as the processing trajectory, and a 5-axis milling cutter is selected as the tool due to the complexity of the trajectory.

[0086] It should be noted that by constructing the ontology model and then obtaining the basic knowledge graph of the blade processing field based on the ontology model, the hierarchical relationship and interaction between each entity and its attribute in the blade processing process can be effectively represented and organized, wherein the basic knowledge graph can clearly describe the dependency relationship between entities such as processing technology, tool selection, material properties, etc. through fine-grained triple data, and provide more accurate information for subsequent processing parameter determination.

[0087] S103, according to the pre-processed data, construct a causal relationship graph related to the blade processing field, and update the basic knowledge graph by using the causal relationship graph to obtain a target knowledge graph.

[0088] Optionally, in one possible implementation, the process of constructing a causal relationship graph related to the blade processing field based on the preprocessed data includes:

[0089] (1) Based on the causal discovery algorithm, causal relationships are mined from the preprocessed data to obtain the first causal relationship set.

[0090] It should be noted that the first causal relationship set contains the potential causal relationships in the preprocessed data. These causal relationships can reveal the direct influence between different processing parameters. Furthermore, unsupervised causal algorithms include PC algorithm, LiNGAM algorithm, and GES algorithm. Taking the PC algorithm as an example, the principle of this unsupervised causal algorithm is to infer causal relationships by constructing a conditional independence graph. First, a fully connected undirected graph is constructed. Then, based on the conditional independence test in the preprocessed data, edges that do not conform to the causal assumption are gradually deleted, and finally, the causal relationship is represented by a directed acyclic graph, which is the first causal relationship graph.

[0091] (2) Based on statistical methods, identify the relevant relationships from the preprocessed data to obtain the second causal relationship set.

[0092] Specifically, first, select an appropriate statistical method based on the data type of the preprocessed data. Taking the Pearson correlation coefficient as an example, first, it is necessary to ensure that the preprocessed data is continuous and meets the linear relationship assumption. Then, calculate the Pearson correlation coefficient for each pair of variables, and then perform hypothesis testing. Set the null hypothesis as "no correlation" (H0: r=0), and use the t-test to calculate the p-value to determine whether the correlation is significant. For example, if the p-value is less than the significance level (e.g., 0.05), then a significant linear correlation is considered to exist; if the p-value is greater than the significance level (e.g., 0.05), then a significant linear correlation is considered not to exist.

[0093] Furthermore, after calculating the correlation coefficients between variables in the preprocessed data, related variable pairs can be identified based on these coefficients, and the set of related variable pairs can be defined as a second causal relationship set. This second causal relationship set identifies potential correlations within the data using statistical methods; these relationships can provide statistical association information between different entities and attributes.

[0094] (3) Merge the first causal relationship set and the second causal relationship set, and use prior causal relationships and intervention learning to verify whether each causal relationship in the fused causal relationship base is valid, so as to find the valid target causal relationship from the causal relationship set.

[0095] In a specific implementation, the first set of causal relationships and the second causal relationship can be combined (union or intersection) to obtain fused causal relationships.

[0096] It should be noted that the prior causal relationship is a causal relationship between entities and attributes that is set in advance based on knowledge, experience or known theoretical models in the field before data analysis; and the intervention learning is a causal reasoning method based on "intervention" or "operation". In intervention learning, the causal effect of a variable on other variables is inferred by artificially intervening in the variable (for example, adjusting the value of a parameter in a certain process step of a processing process) and observing the reaction of the entire system, and the causal relationship is obtained.

[0097] In a specific implementation, when verifying the causal relationship based on intervention learning, the causal relationship can be verified based on A / B test (control variable method), causal inference model, data-driven intervention experiment (using reinforcement learning RL to adjust the processing parameters in a simulation environment and observing the results), etc.

[0098] In this step, the target causal relationship that is established is selected again by fusing the first set of causal relationships and the second set of causal relationships, combining the prior causal relationship and the verification method of intervention learning, so that the effectiveness and feasibility of the causal relationship in the causal relationship set can be ensured.

[0099] (4) According to the target causal relationship, a causal relationship diagram related to the blade processing field is constructed.

[0100] It should be noted that in the embodiment, the unsupervised causal discovery algorithm (such as PC algorithm) and the statistical method (such as Pearson correlation coefficient) are used to extract potential causal relationships from data. These methods can reveal the true relevance in the data. Secondly, the preprocessed data is strictly preprocessed to remove noise and missing values, ensuring the quality of the analysis results. Further, the causal relationship is verified by combining the prior knowledge of the field and intervention learning, ensuring that the identified causal relationship not only holds in data but also is verified in actual operation. In this way, through multi-level verification and fusion, the causal relationship diagram obtained finally can accurately reflect the causal mechanism in the blade processing field. Therefore, the causal relationship diagram obtained finally not only has a scientific basis but also can provide effective decision support in actual application.

[0101] Further, after obtaining the causal relationship, the causal relationship diagram is used to update the basic knowledge graph to obtain a target knowledge graph.

[0102] Specifically, by mapping the entities and relationships in the causal relationship graph to the nodes and edges of the basic knowledge graph, the existing graph content is expanded, and the effectiveness of the new relationship is verified through domain knowledge; at the same time, based on the statistical support and experimental verification in the causal relationship graph, the weights and credibility of the relationships in the basic knowledge graph are updated to obtain the target knowledge graph. In this way, the target knowledge graph obtained will be more comprehensive and accurate, and can reflect the causal mechanism in the blade processing field in real time, providing more effective support for decision-making and optimization.

[0103] It should be noted that the target knowledge graph can also clearly show the causal action chain between entities such as tool materials, processing technology, and tool selection. In this way, it can provide more comprehensive data support and decision-making basis for subsequent processing optimization and parameter adjustment.

[0104] S104, training a large model using the preprocessed data and the target knowledge graph, so that the large model learns the knowledge in the blade processing field to obtain a trained large model.

[0105] The goal of training the large model is to enable the model to fully learn the knowledge in the blade processing field and be used for feature extraction, processing parameter prediction, etc. Specifically, the causal reasoning ability of the model can be enhanced in combination with the knowledge graph. In specific implementation, knowledge can be effectively captured from the preprocessed data and the target knowledge graph as pre-training large model data for data labeling and training; further, knowledge injection can be performed to enable the large model to also obtain associated information of entities when learning word vectors; at the same time, the triple relationship in the target knowledge graph is introduced to effectively make up for the knowledge gap of the large model; in addition, auxiliary tasks such as edge prediction or triple prediction can be added to enable the large model to better understand the entities and relationships in the blade processing field; further, in the pre-training process, the large model learns the blade feature attributes and the statistical characteristics and patterns of blade processing data, extracts features related to blade quality, and finally, in the fine-tuning and verification, the pre-training model is fine-tuned according to specific tasks. In the fine-tuning training data, introduce the annotations in the target knowledge graph to provide additional supervision information; and according to different downstream tasks, formulate clear evaluation indicators.

[0106] Optionally, in a possible implementation, the specific implementation process of this step can include:

[0107] (1) constructing a training sample based on the preprocessed data to train the large model using the training sample.

[0108] In this step, the pre-processed data includes key parameters and related information in the blade processing field, which can provide necessary data support for the large model. According to the pre-processed data, training samples for training the large model are constructed, so that the large model can obtain rich and representative data in the training process.

[0109] (2) Extract triples from the target knowledge graph, and introduce the triples as additional context information into the word vector training when training the large model.

[0110] In this step, triples such as (processing technology, cutting speed, 150 m / min) are extracted from the target knowledge graph, and the triples are introduced as additional context information when training the word vector of the large model. In this way, the large model can not only consider the context of words when learning word vectors, but also understand the interaction and background information between words according to the relationship in the knowledge graph, thereby enhancing the large model's understanding of the knowledge in the blade processing field.

[0111] (3) Add edge prediction and triple prediction as auxiliary tasks to the training process, and through joint optimization, the large model can not only learn word vectors, but also understand and infer the relationship between domain entities.

[0112] Specifically, edge prediction refers to predicting the potential relationship between entities in the knowledge graph, for example, for two entities "tool type" and "processing technology", the model needs to predict whether there is some kind of mutual influence between them; triple prediction involves predicting the integrity or existence of triples, for example, given a triple including entity 1, entity 2 and attribute, given the relationship between entity 1 and attribute, predict another entity 2.

[0113] In this step, edge prediction and triple prediction are added as auxiliary tasks to the training process, and through joint optimization, the large model not only learns how to represent word vectors, but also understands and infers the relationship and influence mechanism between domain entities.

[0114] Specifically, during joint optimization, a joint loss function can be set based on the training task, and joint optimization is performed based on the joint loss function.

[0115] It should be noted that joint optimization means simultaneously optimizing the tasks of word vector learning, triple prediction, and edge prediction.

[0116] In this embodiment, a large model is trained by combining pre-processed data and a target knowledge graph. The pre-processed data is used to construct training samples to ensure that the large model can receive rich feature information related to the blade processing field during the training process, thereby improving the accuracy of feature extraction and processing parameter prediction. Further, the triples in the target knowledge graph are introduced as context information into word vector training, so that the model can learn the context of words and understand the relationship and background knowledge between different entities, enhancing its causal reasoning ability. Then, edge prediction and triple prediction are used as auxiliary tasks, and the model can infer the potential relationship and interaction between entities in the blade processing field while learning word vectors through joint optimization, thereby improving the understanding of domain knowledge and decision support capability. In this way, the large model can perform more accurate feature extraction, processing parameter prediction, and optimization suggestions based on domain knowledge when optimizing the blade processing process.

[0117] S105, for each piece of data in the processing preparation stage data and the actual processing data, the trained large model is used to extract features of the data to obtain a feature vector of the data, and a feature vector library is constructed based on the attribute information and the feature vector of each piece of data.

[0118] Specifically, the trained large model is used to process each piece of data and convert each piece of data into a feature vector. These data can include records of processing parameters, processing technology, and tool types during the processing preparation stage and the actual processing stage. Then, the trained large model can convert these data into high-dimensional vector representations to capture the potential information in the data.

[0119] Further, based on the attribute information of each piece of data and the generated feature vector, a feature vector library is constructed. The feature vector library stores multiple records, and each record records the correspondence between the attribute information and the feature vector of a piece of data. The attribute information of a piece of data can include the blade type, material, processing state, and other information carried by the data.

[0120] In one possible implementation, after the feature vector library is constructed, the method can further include:

[0121] Step 1: According to the entities in the target knowledge graph, each feature vector in the feature vector library is mapped to the corresponding entity in the target knowledge graph to ensure that each feature vector is correctly associated with the corresponding entity.

[0122] Specifically, in the mapping process, first determine the corresponding item of each entity in the target knowledge graph in the feature vector library, for example, according to the entity name or attribute information in the target knowledge graph Match the corresponding feature vector in the feature vector library, associate each feature vector in the feature vector library with the corresponding entity, wherein if the entity has multiple attributes or relationships, select the most relevant feature vector for mapping; further, if the feature vector cannot be directly matched with the entity, it can be matched with the upper layer concept.

[0123] In this step, according to the entity information in the target knowledge graph, each feature vector in the feature vector library is mapped to the corresponding entity in the knowledge graph, so that each feature vector can be correctly associated with its corresponding domain entity, so that the feature vector not only abstractly represents the data, but also accurately reflects the semantic information of the entity.

[0124] Step 2: For each feature vector, adjust the expression of the feature vector using the relationship and attribute of the entity corresponding to the feature vector in the target knowledge graph.

[0125] In specific implementation, the entity attributes (such as the hardness of the leaf and the material type) in the target knowledge graph can be fused with the feature vector to form an enhanced feature representation. In addition, the relationship and attribute of the entity in the target knowledge graph can be combined into the feature vector through splicing, weighting, etc.

[0126] Optionally, the relationship information (such as the processing relationship between the leaf and the material) in the target knowledge graph can also be used to adjust or enhance the feature vector. For example, through the semantic relationship between entities in the target knowledge graph, the position of the feature vector in the vector space is adjusted, so that the related feature vectors are closer, thereby improving the accuracy of feature matching.

[0127] In this step, for each feature vector, the expression of the feature vector is adjusted according to the relationship and attribute of the entity corresponding to the feature vector in the knowledge graph. It can be understood that the influence relationship and attribute in the knowledge graph provide deep background information of the leaf processing field, which helps to adjust the expression of the feature vector, so that the feature vector can more accurately reflect the interaction and influencing factors between entities, thereby optimizing the representation effect of the feature vector.

[0128] In summary, after the construction of the feature vector library is completed, through the above two steps, it can be ensured that each feature vector in the feature vector library not only has a high-quality semantic representation, but also can make each feature vector closely combined with the entity and the influence relationship in the knowledge graph, which can improve the accuracy of feature matching.

[0129] S106, when it is needed to determine the machining trajectory of a blade to be machined, a target feature vector of the blade to be machined is extracted by using the trained large model, and a matching feature vector matching the target feature vector is searched from the feature vector library.

[0130] In this step, first, the relevant data of the blade to be machined, such as the geometry of the blade and the material of the blade, are feature-extracted by using the trained large model, to obtain a target feature vector containing important feature information in the blade machining process; then, in the feature vector library, a matching feature vector matching the extracted target feature vector is searched.

[0131] It should be noted that searching for a matching feature vector matching the extracted target feature vector in the feature vector library can be based on the similarity measurement of the feature vector, for example, the feature vector most similar to the target feature vector can be searched in the feature vector library according to the cosine similarity or the Euclidean distance similarity, or the feature vector with a similarity greater than a preset threshold can be searched.

[0132] S107, according to the attribute information of the matching feature vector, a target machining parameter corresponding to the attribute information is searched from the target knowledge graph, and the matching feature vector and the target machining parameter are input into the trained large model, so that the target machining parameter is optimized by the trained large model, to obtain an optimized machining parameter.

[0133] In this step, first, according to the attribute information of the matching feature vector, a target machining parameter corresponding to the attribute information of the matching feature vector is searched from the target knowledge graph. For example, the attribute information in the matching feature vector can contain information about the blade material, the tool type and the cutting speed, and from the target knowledge graph, machining parameters related to the above information can be searched, such as the optimal cutting depth and the rotating speed of the tool.

[0134] Further, after the machining parameter is searched, the trained large model can be further used to optimize the machining parameter. In specific implementation, the matching feature vector and the target machining parameter are input into the trained large model, and based on the understanding of the historical machining data and the blade machining field knowledge, the target machining parameter can be further optimized by considering the factors such as the geometry of the current blade, the machining environment and the machining process requirements, and then the optimized machining parameter is obtained.

[0135] Further, in a possible implementation manner, before the matching feature vector and the target machining parameter are input into the trained large model, the method can further include:

[0136] Step 1: Verify the rationality of the matching feature vector according to the target knowledge graph and the attribute information of the blade to be processed, to exclude the matching feature vectors that do not conform to the target knowledge graph, and to screen out the target matching feature vector that conforms to the target knowledge graph.

[0137] Step 2: Input the target matching feature vector and the target processing parameter corresponding to the target matching feature vector into the trained large model.

[0138] In the above steps, by verifying the rationality of the matching feature vector according to the target knowledge graph and the attribute information of the blade to be processed, the matching feature vectors that do not conform to the target knowledge graph are excluded, and the target matching feature vector that conforms to the target knowledge graph is screened out. In this way, it can be ensured that the matching feature vector before inputting the data into the large model meets the technical effectiveness at the same time, and also conforms to the actual processing rules in the blade processing field, thereby improving the accuracy of the large model in optimizing the processing parameters.

[0139] It should be noted that in the present embodiment, by constructing the knowledge graph and the causal relationship graph, and combining the training of the large model, the optimization of the processing parameters is no longer dependent on single experience, but combines data driving, knowledge graph, causal reasoning and large model optimization to optimize the processing parameters, which can improve the scientificity and accuracy of parameter optimization. In addition, through the feature vector matching method, the most suitable historical processing parameter for the current blade processing can be found, and further optimization is performed, so that the processing parameter is more accurate.

[0140] The blade machining parameter determination method provided by the embodiment integrates various types of data related to complex curved surface machining, pre-processes the various types of data to obtain pre-processed data, constructs a basic knowledge graph and a causal relationship graph according to the pre-processed data, trains a large model using the pre-processed data and the target knowledge graph to obtain a trained large model, then extracts features of each piece of data in the machining preparation stage and actual machining data by using the trained large model to obtain corresponding feature vectors, constructs a feature vector library based on the attribute information and the feature vectors of each piece of data, further, when the machining trajectory of a blade to be machined needs to be determined, the target feature vector of the blade is extracted by using the large model, and a matching feature vector is searched from the feature vector library, the corresponding target machining parameter is extracted from the target knowledge graph by analyzing the attribute information of the matching feature vector, and the matching feature vector and the target machining parameter are input into the large model for further optimization to obtain optimized machining parameters. In this way, first, the knowledge graph of blade machining is constructed, and the large model is trained in combination with the causal relationship, so that the large model not only remembers the knowledge, but also reasons the relationship between the machining parameters and the machining effect, improves the explainability of the knowledge, further, the trained large model is used to extract features of the machining preparation stage and the actual machining data, and a feature vector library is constructed, which enables the model to perform in-depth analysis before and after machining. In addition, in combination with the matching of the knowledge graph and the feature vector library, effective machining parameters are extracted from similar historical data, and are further optimized in the large model to ensure that the output parameters are more suitable for the current machining requirements, and the machining parameters can be accurately determined.

[0141] Optionally, in a possible implementation, the method further includes:

[0142] The trained large model is used to analyze the processed data according to the machining process, realize the extraction of the causal relationship knowledge between the blade, the tool, the machining process, the machining trajectory, the quality problem and the influencing factor, and correct the causal relationship graph based on the extracted causal relationship indication.

[0143] In specific implementation, the pre-processed data is input into the trained large model to enable the large model to identify the causal relationship, and further, the causal relationship graph can be corrected based on the causal relationship.

[0144] Specifically, in the method, the pre-processed data is analyzed layer by layer according to the machining process (such as design, machining, quality detection, etc.), the causal relationship knowledge between entities is extracted, so that the large model can understand the influence of each stage and step on the final machining result, then the causal relationship graph is corrected based on the extracted causal relationship indication, and problems that may occur in the machining process can be identified and prevented before the machining process starts, thereby improving the machining efficiency and the machining quality.

[0145] Optionally, in another possible implementation, the natural language and multi-modal understanding capability of the trained large model is used to identify entities and relationships related to blade processing quality from the preprocessed data, and the identified entities and relationships are used to supplement or correct the target knowledge graph.

[0146] Specifically, in the method, the trained large model is used to identify entities and relationships related to blade processing quality from the preprocessed data, and the target knowledge graph is supplemented and corrected according to the identified entities and relationships. In this way, the target knowledge graph is continuously iteratively updated, so that the data and logic in the large model can be continuously updated and improved, and the intelligence and decision accuracy of the large model are improved.

[0147] Corresponding to the foregoing embodiment of the blade processing parameter determination method, the present application also provides an embodiment of a blade processing parameter determination device.

[0148] Figure 9 For the structure diagram of the first embodiment of the blade processing parameter determination device of the exemplary embodiment of the present application, please refer to Figure 9 The device comprises an acquisition module 5910, a processing module 920, a construction module 930, a training module 940 and an optimization module 950, wherein,

[0149] The acquisition module 910 is configured to acquire various types of data related to complex curved surface processing; wherein the various types of data reflect curved surface data, feature data and machining data of the blade;

[0150] The processing module 920 is configured to pre-process the various types of data to obtain preprocessed data, and construct a basic knowledge graph related to the blade processing field according to the preprocessed data; the basic knowledge graph takes the blade, material, tool, quality control, processing technology and machining trajectory as nodes, and the inherent characteristics of the blade, material, tool, quality control, processing technology and machining trajectory as node attributes;

[0151] The construction module 930 is configured to construct a causal relationship graph related to the blade processing field according to the preprocessed data, and update the basic knowledge graph using the causal relationship graph to obtain a target knowledge graph;

[0152] The training module 940 is configured to train a large model using the preprocessed data and the target knowledge graph, so that the large model learns the knowledge of the blade processing field to obtain a trained large model;

[0153] The construction module 930 is configured to, for each piece of data in the machining preparation stage data and the actual machining data in the types of data, perform feature extraction on the piece of data by using the trained large model to obtain a feature vector of the piece of data, and construct a feature vector library based on attribute information and the feature vector of each piece of data.

[0154] The acquisition module 910 is configured to, when it is necessary to determine a machining trajectory of a to-be-machined blade, extract a target feature vector of the to-be-machined blade by using the trained large model, and search for a matching feature vector matching the target feature vector from the feature vector library.

[0155] The optimization module 950 is configured to search for a target machining parameter corresponding to attribute information of the matching feature vector from the target knowledge graph according to the attribute information, and input the matching feature vector and the target machining parameter to the trained large model to optimize the target machining parameter by using the trained large model to obtain an optimized machining parameter.

[0156] The device of the embodiment can be used to execute the steps of the method embodiment, and the specific implementation principle and implementation process are similar, and will not be described here. Figure 1 The steps of the method embodiment are similar to the specific implementation principle and implementation process, and will not be described here.

[0157] For the device embodiment, since it basically corresponds to the method embodiment, the related parts are described in the part of the method embodiment. The device embodiment described above is only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the scheme of the present application. Those skilled in the art can understand and implement without creative labor.

[0158] The above is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method of determining a blade machining parameter, characterized by, The blade machining parameter determination method comprises: Obtaining various types of data related to complex surface machining; wherein the various types of data reflect the surface data, feature data and machining data of the blade; Pretreating the various types of data to obtain pretreated data, and constructing a basic knowledge graph related to the blade machining field according to the pretreated data; the basic knowledge graph takes the blade, material, tool, quality control, machining process and machining trajectory as nodes, and takes the inherent characteristics of the blade, material, tool, quality control, machining process and machining trajectory as node attributes; According to the pretreated data, a causal relationship graph related to the blade machining field is constructed, and the basic knowledge graph is updated by using the causal relationship graph to obtain a target knowledge graph; Training a large model by using the pretreated data and the target knowledge graph, so that the large model learns the knowledge in the blade machining field to obtain a trained large model; For each piece of data in the machining preparation stage data and the actual machining data in the various types of data, the trained large model is used to extract the feature vector of the data, and a feature vector library is constructed based on the attribute information and the feature vector of each piece of data; When the machining trajectory of a to-be-machined blade needs to be determined, the target feature vector of the to-be-machined blade is extracted by using the trained large model, and a matching feature vector matching the target feature vector is searched from the feature vector library; According to the attribute information of the matching feature vector, a target machining parameter corresponding to the attribute information is searched from the target knowledge graph, and the matching feature vector and the target machining parameter are input into the trained large model to optimize the target machining parameter by the trained large model to obtain an optimized machining parameter; The method further comprises: Analyzing the pretreated data according to the machining process by using the trained large model, realizing the extraction of the causal association knowledge among the blade, tool, machining process, machining trajectory, quality problem and influencing factor, and correcting the causal relationship graph based on the extracted causal association indication; Or, using the natural language and multi-modal understanding ability of the trained large model, identifying entities and relationships related to blade machining quality from the pretreated data, and supplementing or correcting the target knowledge graph by using the identified entities and relationships.

2. The method of claim 1, wherein, The training of the large model by using the pretreated data and the target knowledge graph, so that the large model learns the knowledge in the blade machining field to obtain a trained large model, comprises: Constructing a training sample based on the pretreated data to train the large model by using the training sample; Extracting triples from the target knowledge graph, and introducing the triples as additional context information into the word vector training when training the word vector of the large model; Adding edge prediction and triple prediction as auxiliary tasks to the training process, and through joint optimization, the large model can learn the word vector while also understanding and reasoning the relationship between the entities in the blade field.

3. The method of claim 1, wherein, After the feature vector library is constructed, the method further comprises: According to the entities in the target knowledge graph, each feature vector in the feature vector library is mapped to the corresponding entity in the target knowledge graph to ensure that each feature vector is correctly associated with the corresponding entity; For each feature vector, the expression of the feature vector is adjusted by the relationship of the entity corresponding to the feature vector in the target knowledge graph.

4. The method of claim 1, wherein, The construction of the causal relationship graph related to the blade processing field according to the preprocessed data comprises: Based on a causal discovery algorithm, causal relationships are mined from the preprocessed data to obtain a first causal relationship set; Based on a statistical method, relevant relationships are identified from the preprocessed data to obtain a second causal relationship set; The first causal relationship set and the second causal relationship set are fused, and each causal relationship in the fused causal relationship set is verified by using prior causal relationships and intervention learning to find the target causal relationship that is established from the causal relationship set; According to the target causal relationship, a causal relationship graph related to the blade processing field is constructed.

5. The method of claim 1, wherein, The construction of the basic knowledge graph related to the blade processing field according to the preprocessed data comprises: Based on the preprocessed data, entities in the blade processing field and attributes corresponding to each entity are identified; Based on the preprocessed data, first-level relationships between entities and second-level relationships between attributes of different entities are analyzed and determined, and for each entity, the hierarchical relationships between the attributes of the entity are analyzed and determined to obtain a sub-ontology model corresponding to each entity; According to the first-level relationships, the sub-ontology model corresponding to each entity, and the second-level relationships, an ontology model of the blade processing field is obtained; the ontology model is used to represent the hierarchical relationships and mutual relationships between entities and attributes; Based on the ontology model, triple data of entities, attributes, and values are identified from the preprocessed data, and the identified triple data is used to construct a basic knowledge graph of the blade processing field; the basic knowledge graph represents blade field knowledge through entities and relationships between entities, and forms fine-grained relationship and attribute descriptions through triple data, and can accurately represent various entities, attributes, and relationships between them in the blade processing field.

6. The method of claim 5, wherein, The entities include blades, materials, tools, quality control, processing technology, and processing trajectories; the analysis and determination of the hierarchical relationships between the attributes of each entity to obtain a sub-ontology model corresponding to each entity comprises: For blades, according to the geometric shape, physical properties, processing methods, optimization targets, and application fields of blades, the features of blades are classified in detail, the relationships between the attributes in each category are determined, the hierarchical relationships are established according to the classification results and attribute relationships, and reasoning rules are formulated according to the above classification and attribute relationships to obtain a sub-ontology model of blades; For materials, the tool material and the workpiece material are classified according to the type, each type of material is classified according to the manufacturing process, and the relationship between each attribute in each category after the manufacturing process classification is determined, and finally the hierarchical relationship is established according to the classification results and attribute relationships to obtain the sub-ontology model of the material; For tools, the numerical control machining tool is classified according to the machining process, each tool after classification is classified according to the function, and the relationship between each attribute in each category after the function classification is determined, and finally the hierarchical relationship is established according to the classification results and attribute relationships to obtain the sub-ontology model of the tool; For quality control, the quality control is taken as the core entity, the detection method, quality standard, error range and detection frequency are taken as the attributes of the entity, the detection method includes detection means, detection tool, detection program and detection environment, the quality standard includes international standard, national standard, industry standard, enterprise standard and technical specification, the error range includes absolute error, relative error, quality uncertainty, tolerance range, process error and detection error, and the detection frequency includes time frequency, sampling ratio, dynamic adjustment and key point detection. The sub-ontology model with quality control as the core entity is established; For machining process, the machining process is taken as the core entity, and the machining material, process, step, cutting speed, back engagement amount and feed amount are taken as the attributes of the entity, and the type of the machining material is taken as the attribute of the machining material to establish the sub-ontology model with machining process as the core entity; For machining trajectory, the machining trajectory is classified according to the machining trajectory type to obtain the hierarchical relationship of the machining trajectory, and for each type of machining trajectory, the scene suitable for the machining trajectory and the tool suitable for the machining trajectory are taken as the attributes of the machining trajectory to establish the sub-ontology model with machining trajectory as the core entity.

7. The method of claim 1, wherein, Before the matching feature vector and the target machining parameter are input into the trained large model, the method further comprises: According to the target knowledge graph and the attribute information of the blade to be machined, the rationality of the matching feature vector is verified to exclude the matching feature vectors that do not conform to the target knowledge graph, and the target matching feature vector that conforms to the target knowledge graph is selected; The matching feature vector and the target machining parameter are input into the trained large model, comprising: The target matching feature vector and the target machining parameter corresponding to the target matching feature vector are input into the trained large model.

8. The method of claim 1, wherein, The various types of data related to complex curved surface machining are obtained, comprising: The data collection type is determined as structured text, semi-structured text, unstructured text, table, literature, picture, audio and / or video and webpage; The data sources include CAD models and design drawings in the product design stage, blade geometric parameters, material properties, CFD analysis results in the simulation stage, FEA stress and strain prediction, thermodynamic performance evaluation, tool selection parameters in the machining accuracy stage, machining tool specifications, NC programming information, real-time machining parameters, online measurement data, and temperature monitoring data in the machining process, final product quality detection reports, defect records, and customer feedback in the quality control stage, and fault history records, repair logs, and improvement measures documents in the maintenance and improvement stage. The data collection methods include direct collection, system docking, manual import, and / or input. According to the data collection method, each type of data indicated by the data collection type is collected from the data source.

9. A blade machining parameter determination apparatus characterized by comprising: The device includes an acquisition module, a processing module, a construction module, a training module, and an optimization module, wherein, The acquisition module is configured to acquire various types of data related to complex curved surface machining, wherein the various types of data reflect curved surface data, feature data, and machining data of the blade. The processing module is configured to preprocess the various types of data to obtain preprocessed data, and construct a basic knowledge graph related to the blade machining field according to the preprocessed data; the basic knowledge graph takes the blade, material, tool, quality control, machining process, and machining trajectory as nodes, and takes the inherent characteristics of the blade, material, tool, quality control, machining process, and machining trajectory as node attributes. The construction module is configured to construct a causal relationship graph related to the blade machining field according to the preprocessed data, and update the basic knowledge graph using the causal relationship graph to obtain a target knowledge graph. The training module is configured to train a large model using the preprocessed data and the target knowledge graph, so that the large model learns the knowledge of the blade machining field to obtain a trained large model. The construction module is configured to, for each piece of data in the machining preparation stage data and the actual machining data, extract features of the piece of data using the trained large model to obtain a feature vector of the piece of data, and construct a feature vector library based on the attribute information and the feature vector of each piece of data. The acquisition module is configured to, when the machining trajectory of a to-be-machined blade needs to be determined, extract a target feature vector of the to-be-machined blade using the trained large model, and find a matching feature vector matching the target feature vector from the feature vector library. The optimization module is configured to find target machining parameters corresponding to the attribute information from the target knowledge graph according to the attribute information of the matching feature vector, and input the matching feature vector and the target machining parameters into the trained large model to optimize the target machining parameters by the trained large model to obtain optimized machining parameters.

Images