A method and system for mapping unstructured process data process knowledge

Abstract

The application discloses a kind of unstructured processing data process knowledge mapping method and system, according to enterprise process decision-making process definition multilayer process knowledge model, and definition process decision-making key element;According to the actual production situation of current enterprise, the content that each kind of key element should include is selected to build high-quality data set;Small sample data set is marked using BIEOS marking method;LERT model fine-tuning, the unstructured data saved in the production process of enterprise is input into fine-tuned LERT model and CRF model, and process entity in input text is obtained;Entity is stored in Neo4J graph database according to the relationship framework that is preset in advance.The application realizes the conversion of mechanical processing process unstructured data to process knowledge, can automatically convert the long-term storage of enterprise unstructured data into knowledge, reduce the burden of process personnel, in addition, storage in the form of knowledge graph can further realize process parameter intelligent decision-making, improve the efficiency and quality of process planning.

Description

Technical Field

[0001] This invention belongs to the field of process design technology, specifically relating to a method and system for mapping process knowledge from unstructured processing data. Background Technology

[0002] In recent years, the demand for customized products in industries such as aerospace and automotive parts has surged, many of which are complex parts with intricate structures, high processing quality requirements, and harsh operating environments. These parts are characterized by small batch production and customization, and their increasingly complex structures and ever-increasing processing quality requirements pose significant challenges to process planning. Process planning is a crucial aspect of the technical preparation work in mechanical manufacturing, serving as a link between product design and manufacturing. It is a highly experience-based decision-making process that is highly adaptable to changing environments. However, most manufacturing enterprises still employ traditional process planning methods, relying heavily on manual experience to formulate process solutions. This often requires repeated trial production and testing, sacrificing efficiency and cost to ensure the processing quality of complex parts, leading to low processing efficiency and uncontrollable processing cycles. Furthermore, there is a lack of effective coordination between the processing of complex parts and process design, making iterative upgrades of process solutions difficult and time-consuming. Therefore, there is an urgent need to innovate process planning methods to achieve scientific decision-making for process solutions.

[0003] Automated or intelligent process design has always been the goal of Computer-Aided Process Design (CAPP) systems. Achieving intelligent process design requires empowering process design software with capabilities such as understanding design problems, generating design results, and learning design knowledge. The core of these capabilities is process design knowledge, encompassing how knowledge is acquired, expressed, updated, and applied under specific conditions. As a crucial component of modern intelligent process decision-making systems, the process knowledge base serves as a bridge between product design and manufacturing in modern manufacturing. By integrating expert knowledge and experience, it directly impacts production efficiency, product costs, and the company's production cycle. In an intelligent manufacturing environment, building a high-quality process knowledge base in a short time is critical for achieving short-cycle development and process design of complex products under modern manufacturing models. The acquisition of process knowledge, i.e., the construction of the process knowledge base, remains a key issue.

[0004] Existing process knowledge bases are mostly built manually. In order to obtain supporting information from different sources, engineers spend about 40% of their time searching and reorganizing raw data. A lot of time is spent manually creating knowledge bases during the product preparation phase, which greatly increases the product manufacturing cycle. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of the prior art by providing a mapping method and system for unstructured processing data and process knowledge. This method addresses the technical problems of current process planning relying heavily on human experience, having a low level of intelligence, lacking a knowledge base to support process decisions, and the fact that existing methods for constructing process knowledge bases are mostly manual, with long preparation cycles and quality heavily dependent on the experience of domain experts.

[0006] The present invention adopts the following technical solution:

[0007] A method for mapping process knowledge into unstructured processing data includes the following steps:

[0008] S1. Design a multi-layered process knowledge model based on the actual processing process and the process decision-making process;

[0009] S2. Based on the multi-layer process knowledge model constructed in step S1, design the content framework required for the dataset and collect unstructured data to construct a small sample dataset.

[0010] S3. Use the BIEOS annotation method to annotate the small sample dataset constructed in step S2;

[0011] S4. Based on the task objective, the problem is defined as the named entity recognition task. The BERT-based optimized model LERT is selected, and the pre-trained model LERT is fine-tuned for downstream sub-tasks using the small sample dataset labeled in step S3.

[0012] S5. Process the unstructured data from the enterprise's production process and input it into the fine-tuned pre-trained model LERT+CRF. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge.

[0013] S6. Further process the entities identified in step S5, and automatically import them into the graph database Neo4J according to the set knowledge architecture to complete the automated construction and visualization of the process knowledge base.

[0014] Specifically, the multi-layered process knowledge model (Process_model) is as follows:

[0015] Process_model={Design_layer,Process_layer,manufacturing_resource_layer}

[0016] Among them, Design_layer is the design layer, Process_layer is the process layer, and manufacturing_resource_layer is the manufacturing resource layer.

[0017] Furthermore, the design layer stores key design requirements for part design drawings, including part type, part material, machining characteristics, accuracy, and roughness; the process layer stores process examples of actual machining processes, including cutting parameters and machining procedures; and the manufacturing resource layer stores recommended equipment for machining processes, including machine tool, cutting fluid, fixture, tool type, and tool material.

[0018] Specifically, in step S2, the small sample dataset includes:

[0019] Part types: Shaft parts, sleeve parts, disc parts, plate parts, cylindrical gear parts, and box-type parts;

[0020] Parts materials: cast iron, mold steel, high manganese steel, stainless steel, titanium alloy, high temperature alloy, copper alloy;

[0021] Machining features: holes, keyways, outer cylindrical surfaces, annular grooves, cones, inner cylindrical surfaces, planes, curved surfaces, threads, splines, fillets;

[0022] Tool materials: high-speed steel, cemented carbide, ceramics, diamond, and cubic boron nitride;

[0023] Cutting fluids: dry cutting fluids, aqueous solutions, emulsions, extreme pressure emulsions, rust-preventive emulsions, mineral oils, animal and vegetable oils, compound oils, machine oils, kerosene, water-based cutting fluids, and oil-based cutting fluids;

[0024] Precision and roughness are represented by the fixed symbols IT and Ra;

[0025] Machining processes: turning, milling, drilling, reaming, boring, heat treatment;

[0026] Cutting parameters: Cutting parameters for a machining process.

[0027] Specifically, in step S3, each dataset sample is labeled with BIOES, where B represents the beginning of an entity, I represents the inside of an entity, O represents the outside of an entity, E represents the end of an entity, and S represents a single character entity.

[0028] Specifically, step S5 is as follows:

[0029] S501. The input text is processed by embedding representation and adding position encoding to form a vector sequence. After being input into LERT, different information is first extracted through a multi-head attention mechanism, then normalized and connected to the residual network, and then input into the fully connected layer. Then normalization and connection to the residual network are performed again. The above steps are repeated multiple times to finally obtain the output vector.

[0030] S502. Input the vector sequence obtained in step S501 into the optimization model LERT, extract features from the input sequence through a multi-head self-attention mechanism, and output a vector sequence containing contextual information.

[0031] S503. Input the output vector of the optimization model LERT into the fully connected layer, calculate the label probability of each word to obtain a state score, then input the state score into the CRF layer, use the state score as a state feature, and obtain the optimal route through the transition score matrix.

[0032] Furthermore, in step S502, multiple attention methods are learned through multi-head attention, with each head learning different attention weights. The calculation process is as follows:

[0033] q∈R q Let k be the input query vector, and v be m key-value pairs, (k1, v1), ..., (k...). m ,v m ), k∈R k ,v∈R v The attention convergence function is expressed as:

[0034]

[0035] The two vectors are mapped to scalars using the attention scoring function 'a', and then the attention weights for query 'q' and key 'k1' are obtained through softmax operation.

[0036]

[0037] The scoring function for scaled dot product attention is:

[0038]

[0039] The output vector containing contextual information is obtained.

[0040] Furthermore, in step S503, the probability of the tag sequence appearing for each word is:

[0041]

[0042] Among them, S real The score represents the actual path, and the text length is N.

[0043] Specifically, in step S6, the knowledge is automatically imported into the graph database Neo4J according to the pre-defined knowledge structure.

[0044] Knowledge_relation={Has_attribute,Has_relation}

[0045] Here, Knowledge_relation represents knowledge relationships, Has_attribute represents attributes, and Has_relation represents relationships.

[0046] Secondly, embodiments of the present invention provide a mapping system for unstructured processing data and process knowledge, including:

[0047] The design module designs a multi-layered process knowledge model based on the actual processing process and the process decision-making process.

[0048] The collection module designs the required content framework for the dataset based on the multi-layered process knowledge model constructed by the design module, and collects unstructured data to construct a small sample dataset.

[0049] The annotation module uses the BIEOS annotation method to annotate the small sample dataset constructed by the collection module;

[0050] The fine-tuning module defines the problem as a named entity recognition task based on the task objective, selects the BERT-based optimized model LERT, and uses the small sample dataset labeled by the annotation module to fine-tune the pre-trained model LERT for downstream sub-tasks.

[0051] The conversion module processes unstructured data from the enterprise's production process and inputs it into the pre-trained model LERT+CRF, which has been fine-tuned by the fine-tuning module. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge.

[0052] The output module further processes the entities identified by the conversion module and automatically imports them into the graph database Neo4J according to the predefined knowledge architecture, thus completing the automated construction and visualization of the process knowledge base.

[0053] Compared with the prior art, the present invention has at least the following beneficial effects:

[0054] A mapping method for process knowledge in unstructured machining data is proposed. This method fully considers various elements in the machining process, constructs a layered process knowledge model, and represents the information needed for process parameter decision-making through a design layer, a process layer, a manufacturing resource layer, and 12 key elements within these three layers. This model is suitable for subsequent intelligent decision-making tasks related to process parameters. For each of the 12 defined key elements, the specific content to be included is defined based on the actual processing situation of the enterprise, enhancing the robustness of the method and making it applicable to different processing and manufacturing processes. The BIOES annotation method is used to annotate the dataset, improving the algorithm's recognition efficiency and enabling better differentiation of entity boundaries. A new BERT-optimized model, LERT, is adopted, which is more suitable for Chinese name recognition tasks. A CRF layer is connected after the LERT model, and the rationality of the output label sequence is further improved through state feature functions and transition feature functions. Finally, the identified process knowledge entities are stored in Neo4J according to a predefined knowledge framework, stored in the form of a knowledge graph, such as... Figure 6 As shown, this invention abandons the traditional method of manually building knowledge bases, realizes the mapping between unstructured data of machining processes and process knowledge, and achieves automated construction and storage of process knowledge bases, which greatly improves the efficiency of knowledge base construction.

[0055] Furthermore, by constructing a layered process knowledge model, the various process knowledge required for process parameter decisions can be more clearly represented, conforming to the overall process decision-making process. The specific requirements of the task to be processed are obtained through the design layer, and then linked to the content of the process layer to provide process parameter decisions. Furthermore, the manufacturing resource layer is linked to provide the equipment required for the processing, thereby improving the rationality of the process knowledge base.

[0056] Furthermore, by selecting the content of the 12 key elements, some outdated process knowledge has been discarded, making it more in line with the current processing realities of enterprises. In addition, the content is comprehensive, applicable to various processing tasks, robust, and can be generalized to different processing tasks.

[0057] Furthermore, current deep learning models often require massive datasets to achieve good results, and the heavy data annotation work will greatly increase the cost of building a knowledge base. Therefore, this patent selects the pre-trained model LERT and builds a small sample dataset in the field for fine-tuning on this basis, so as to achieve good named entity recognition results without the need for a large amount of data annotation.

[0058] Furthermore, unlike the more widely used BIO annotation method, the BIOES annotation method can better distinguish entity boundaries, which helps improve recognition accuracy. In addition, the annotation results of the BIEOS annotation method can be easily converted into continuous entity text, so the recognition results can be easily visualized and analyzed, with good readability and visualization.

[0059] Furthermore, when using the LERT model alone for named entity recognition tasks, its output labels do not consider the information of adjacent labels, which may result in incorrect BIB sequences (the correct sequence should be BIE). By further processing the LERT input using CRF and employing the transition feature matrix, the contextual information of the output labels can be fully considered, avoiding the output of label sequences that do not conform to reality, and further improving the accuracy of entity recognition.

[0060] Furthermore, through the attention mechanism, the model can automatically learn the feature extraction method. The input text sequence often contains semantic relationships. When performing named entity recognition, it is necessary to consider the content of the text that is close to each other as well as the content of the text that is far away. One attention mechanism learns one type of feature. Through multi-head attention, different information can be extracted, which can obtain both short-distance relationships and long-distance relationships.

[0061] Furthermore, CRF uses the transition score matrix to find the path with the highest probability and best performance among all possible paths. This label sequence is the model's output. When the text length is n and the number of labels is m, there are a total of m... n One possibility. Using S i Let represent the score of the i-th path. Construct a loss function with increasing its probability as the objective function, and iterative optimization will yield the optimal label sequence.

[0062] Furthermore, the output entities are processed further, pre-defined relationships are added, and the data is stored in a Neo4J graph database in the form of a knowledge graph. The data and knowledge storage structure of the knowledge graph is a directed graph structure, which allows the knowledge graph to effectively store the relationships between data and knowledge. The knowledge graph can achieve efficient data and knowledge access through graph matching algorithms. It can automatically and intelligently discover and reason about multi-faceted implicit knowledge from existing knowledge. This completes the mapping from unstructured data of machining processes to process knowledge.

[0063] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0064] In summary, this invention defines a model of process knowledge, applies LERT and CRF models to the field of manufacturing process knowledge identification, and, unlike traditional relational databases, presents knowledge in the form of a knowledge graph. This enables the automated construction of a process knowledge base, effectively helping process engineers build a high-quality knowledge base, thereby improving the efficiency and quality of process parameter decisions and reducing the burden on process engineers.

[0065] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0066] Figure 1 This is a flowchart of the present invention;

[0067] Figure 2 This is a multi-layer process knowledge model diagram of the present invention;

[0068] Figure 3 This is a schematic diagram of the dataset structure of the present invention;

[0069] Figure 4 This is a schematic diagram of the BIEOS dataset annotation method of the present invention;

[0070] Figure 5 This is a flowchart illustrating the specific process of the LERT+CRF algorithm of this invention.

[0071] Figure 6 This is a framework diagram of the knowledge graph storage of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0074] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0075] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0076] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0077] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0078] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0079] This invention provides a method for mapping process knowledge from unstructured machining data. It defines a multi-layered process knowledge model based on the enterprise's process decision-making process and defines 12 key elements for process decisions. Then, based on the current actual production situation of the enterprise, it selects the content that each key element should include to construct a high-quality dataset. The dataset is then further processed and labeled using the BIEOS annotation method. A small sample dataset is then used to fine-tune the LERT model, making it more suitable for mechanical processing entity recognition tasks. Unstructured data stored during the enterprise's production process is then input into the fine-tuned LERT model and CRF model to obtain process entities in the input text. Finally, the entities are stored in a Neo4J graph database according to a pre-set relational framework. This invention realizes the transformation of unstructured mechanical processing data into process knowledge. Compared with traditional methods, this invention can automatically transform long-term stored unstructured data into knowledge, reducing the burden on process personnel. Furthermore, storing it in the form of a knowledge graph can further realize intelligent decision-making on process parameters, improving the efficiency and quality of process planning.

[0080] Please see Figure 1 The present invention discloses a method for mapping process knowledge of unstructured processing data, comprising the following steps:

[0081] S1. Based on the actual processing process and the process decision-making process, design a multi-layer process knowledge model, which includes 12 key elements of process decision-making, thereby fully representing the process knowledge required for process decision-making.

[0082] Please see Figure 2 The multi-layered process knowledge model (Process_model) is divided into a design layer, a process layer, and a manufacturing resource layer, specifically:

[0083] Process_model={Design_layer,Process_layer,manufacturing_resource_layer}

[0084] Among them, Design_layer is the design layer, Process_layer is the process layer, and manufacturing_resource_layer is the manufacturing resource layer.

[0085] The design layer stores the key design requirements of the part design drawings, including part type (Part_type), part material (Part_material), machining characteristics (Machining_characteristics), accuracy, and roughness.

[0086] The process layer stores process examples of the actual machining process, including cutting parameters Machine_parameter and machining procedures Processing_procedure.

[0087] The manufacturing resource layer stores recommended equipment for the processing, including machine tool, cutting fluid, fixture, tool type, and tool material.

[0088] S2. Based on the multi-layer process knowledge model constructed in step S1, design the content framework required for the dataset, and further collect unstructured data from databases, process books, and electronic documents to construct a high-quality small sample dataset.

[0089] Based on the 12 key factors for process decision-making in step S1, a fine-grained Chinese named entity recognition dataset is constructed. When selecting unstructured data, dataset sentences are selected according to enterprise surveys and literature searches.

[0090] Part types include shaft parts, sleeve parts, disc parts, plate parts, cylindrical gear parts, box-type parts, etc.

[0091] The materials used for the parts include cast iron, mold steel (hardened steel), high manganese steel, stainless steel, titanium alloy, high temperature alloy, copper alloy and other commonly used materials.

[0092] Machining features include holes, keyways, outer cylindrical surfaces, annular grooves (grooves), cones, inner cylindrical surfaces, planes (end faces), curved surfaces, threads, splines, fillets, and other features.

[0093] The main materials used in cutting tools include high-speed steel, cemented carbide, ceramics, diamond (natural and synthetic) and cubic boron nitride.

[0094] Cutting fluids include dry cutting (without added cutting fluid), aqueous solutions, emulsions, extreme pressure emulsions, rust-preventive emulsions, mineral oils, animal and vegetable oils, compound oils, engine oils, kerosene, water-based cutting fluids, oil-based cutting fluids, and other commonly used cutting fluids.

[0095] Precision and roughness are represented by the fixed symbols IT and Ra.

[0096] The machining process includes specific steps such as turning, milling, drilling, reaming, boring, and heat treatment.

[0097] The cutting parameters include the cutting parameters for the above machining processes.

[0098] Machine tools, fixtures, and cutting tools are mainly the various processing equipment required for the above-mentioned processes.

[0099] Please see Figure 3 Each data entry should contain the input text:

[0100] When machining difficult-to-machine materials with high-speed steel tools, 10% to 15% extreme pressure emulsion or extreme pressure cutting oil should be selected.

[0101] The location of the entity in the text:

[0102] "start":1,"end":4

[0103] The entity's text content:

[0104] "text":"high-speed steel"

[0105] Entity tag type:

[0106] "labels":"Tool material".

[0107] S3. Use the BIEOS annotation method to annotate the dataset constructed in step S2 so that it is suitable for the fine-tuning task of the pre-trained model in step S4.

[0108] Each dataset sample is labeled with BIOES: where B represents the beginning of an entity, I represents the inside of an entity, O represents the outside of an entity, E represents the end of an entity, and S represents a single-character entity.

[0109] Please see Figure 4 The input sample sequence is:

[0110] Input sample: {When cutting difficult-to-machine materials with high-speed steel tools, the appropriate tool should be selected}

[0111] The labeled sequence is:

[0112] Label sequence: {OB 刀具材料 I 刀具材料 E 刀具材料 OOOOOOOOOOOOOO}.

[0113] S4. Based on the task objective, the problem is defined as the named entity recognition task. The BERT-based optimized model LERT is selected, and the pre-trained model LERT is fine-tuned for downstream sub-tasks using the dataset labeled in step S3, so that it is more suitable for the subsequent process knowledge extraction task.

[0114] The labeled datasets are input in batches into the pre-trained model LERT, and its content parameters are continuously optimized to make it more suitable for downstream mechanical processing entity recognition tasks.

[0115] S5. Process unstructured data such as natural language in the enterprise's production process and input it into the fine-tuned LERT+CRF model. The CRF layer further ensures the rationality of the labeling through the transition score matrix, completes the named entity recognition task, and realizes the conversion of unstructured data into process knowledge.

[0116] Please see Figure 5 The details are as follows:

[0117] S501. Process the input text, perform embedding representation and add position encoding to form a vector sequence;

[0118] Form a vector sequence:

[0119] {E <cls>< / cls> E <用> E <高> E <速> E <钢> ,...,E <用>}

[0120] S502. The input sequence is feature extracted through a multi-head self-attention mechanism, and a vector sequence containing contextual information is output.

[0121] By capturing long-distance dependencies in the input sequence through self-attention, the global dependencies of the sequence can be captured more effectively.

[0122] The calculation of the self-attention mechanism is as follows:

[0123] y i =f(x) i ,(x1,x1),...,(x n ,x n ))∈R d

[0124] Where, x i ,...,x n Given an input sequence consisting of word units, any x i ∈R d (1≤i≤n), where d is the feature dimension.

[0125]

[0126] Here, f(x) is a general attention pooling formula.

[0127] Multi-head attention learns various attention patterns, thereby capturing different semantic information in the input sequence. Each head learns different attention weights, enabling the model to observe the input sequence from different perspectives and thus better capture the information in the sequence. The calculation process is as follows:

[0128] q∈R q Let k be the input query vector, and v be m key-value pairs, (k1, v1), ..., (k...). m ,v m ), where k∈R k ,v∈R v The attention convergence function is expressed as:

[0129]

[0130] The attention weights for query q and key k1 are obtained by mapping the two vectors into scalars through the attention scoring function a, and then performing a softmax operation.

[0131]

[0132] Scaled dot-product attention is used to improve computational efficiency, and its scoring function is:

[0133]

[0134] We obtain the output vector containing contextual information:

[0135] {C,T 用 ,T 高 ,T 速 ,T 钢 ,...,T 用}

[0136] S503. Input the output vector of LERT into the fully connected layer to calculate the label probability of each word, thereby obtaining a state score. Then input the score into the CRF layer, use the state score as the state feature, and obtain the optimal route through the transition score matrix.

[0137] The formula for conditional probability is as follows:

[0138] (y|x)=P(y0,y1,...,y n |x0,x1,...,x n )

[0139] The probability of a label sequence occurring is:

[0140]

[0141] Among them, S real The score represents the actual path, and the text length is N.

[0142] Its loss function is defined as:

[0143]

[0144] Finally, a set of label sequences {O, B} is obtained. 刀具材料用 ,I 刀具材料 E 刀具材料 The values ​​of , ..., O} are assigned to the input to complete the entity recognition task.

[0145] S6. Further process the entities identified in step S5, and automatically import them into the graph database Neo4J according to the set knowledge architecture to complete the automated construction and visualization of the process knowledge base.

[0146] The output entities from step S5 are processed and imported into the Neo4J graph database. Relationships are then added according to the pre-defined process knowledge framework, specifically as follows:

[0147] Knowledge_relation={Has_attribute,Has_relation}

[0148] The specific attributes are:

[0149]

[0150] The relationship is as follows:

[0151] Has_Relation={HasMachiningCharacteristic,HasProcess,HasCase,HasResources}

[0152] Part types are connected to part materials via HasMaterial, and to machining features via HasMachiningCharacteristics; machining features are connected to surface roughness via HasRoughness, to accuracy via HasIT, and to machining operations via HasProcess; machining operations are connected to case studies via HasCase; case studies are connected to cutting parameters via HasResourses. Part materials are connected via HasMaterials; cutting parameters are connected to manufacturing resources via HasResourses; manufacturing resources are connected to cutting fluid via HasCuttingFlude, to fixtures via HasFixture, to tool types via HasToolType, to machine tools via HasMachine, and to tool materials via HasToolMaterial; ultimately, this achieves the mapping from unstructured machining process data to process knowledge.

[0153] In another embodiment of the present invention, a mapping system for unstructured processing data process knowledge is provided. This system can be used to implement the above-mentioned mapping method for unstructured processing data process knowledge. Specifically, the mapping system for unstructured processing data process knowledge includes a design module, a collection module, an annotation module, a fine-tuning module, a conversion module, and an output module.

[0154] The design module designs a multi-layered process knowledge model based on the actual processing and process decision-making process.

[0155] The collection module designs the required content framework for the dataset based on the multi-layered process knowledge model constructed by the design module, and collects unstructured data to construct a small sample dataset.

[0156] The annotation module uses the BIEOS annotation method to annotate the small sample dataset constructed by the collection module;

[0157] The fine-tuning module defines the problem as a named entity recognition task based on the task objective, selects the BERT-based optimized model LERT, and uses the small sample dataset labeled by the annotation module to fine-tune the pre-trained model LERT for downstream sub-tasks.

[0158] The conversion module processes unstructured data from the enterprise's production process and inputs it into the pre-trained model LERT+CRF, which has been fine-tuned by the fine-tuning module. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge.

[0159] The output module further processes the entities identified by the conversion module and automatically imports them into the graph database Neo4J according to the predefined knowledge architecture, thus completing the automated construction and visualization of the process knowledge base.

[0160] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a mapping method for unstructured processing data technology knowledge, including:

[0161] A multi-layered process knowledge model is designed based on the actual processing procedure and the process decision-making process. The required content framework for the dataset is designed based on the multi-layered process knowledge model, and unstructured data is collected to construct a small sample dataset. The small sample dataset is labeled using the BIEOS annotation method. Based on the task objective, the problem is defined as a named entity recognition task. The BERT-based optimized model LERT is selected, and the pre-trained model LERT is fine-tuned for downstream subtasks using the labeled small sample dataset. Unstructured data from the enterprise's production process is processed and input into the fine-tuned pre-trained model LERT+CRF. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge. The identified entities are further processed and automatically imported into the graph database Neo4J according to the set knowledge architecture, completing the automated construction and visualization of the process knowledge base.

[0162] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0163] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the mapping method for unstructured processing data technology knowledge in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0164] A multi-layered process knowledge model is designed based on the actual processing procedure and the process decision-making process. The required content framework for the dataset is designed based on the multi-layered process knowledge model, and unstructured data is collected to construct a small sample dataset. The small sample dataset is labeled using the BIEOS annotation method. Based on the task objective, the problem is defined as a named entity recognition task. The BERT-based optimized model LERT is selected, and the pre-trained model LERT is fine-tuned for downstream subtasks using the labeled small sample dataset. Unstructured data from the enterprise's production process is processed and input into the fine-tuned pre-trained model LERT+CRF. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge. The identified entities are further processed and automatically imported into the graph database Neo4J according to the set knowledge architecture, completing the automated construction and visualization of the process knowledge base.

[0165] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0166] According to the actual process of this invention, a small sample dataset containing 840 texts was constructed, with ten categories of labels: tool material, tool type, cutting parameters, machining process, machining features, fixture, machine tool, surface roughness, and part material. 700 of these texts were used as the training set, and 140 were used as the test set for training and validation.

[0167] The test set contained 327 entities. Under weighted average evaluation, the precision reached 78.81%, the recall reached 0.8807, and the F1 score reached 0.8296. Testing with actual data shows that the invention performs well on small sample datasets.

[0168] In summary, the present invention provides a method and system for mapping process knowledge from unstructured processing data, which realizes the acquisition and visual storage of process knowledge. Compared with traditional methods, this method can automatically transform unstructured enterprise data into process knowledge, thereby improving the efficiency and quality of process parameter decision-making and reducing the burden on process personnel.

[0169] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0170] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0171] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0172] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0173] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0175] If the integrated module / 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, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0176] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0177] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0178] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0179] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for mapping process knowledge from unstructured processing data, characterized in that, Includes the following steps: S1. Design a multi-layered process knowledge model based on the actual processing process and the process decision-making process; S2. Based on the multi-layer process knowledge model constructed in step S1, design the content framework required for the dataset and collect unstructured data to construct a small sample dataset. S3. Use the BIEOS annotation method to annotate the small sample dataset constructed in step S2; S4. Based on the task objective, the problem is defined as the named entity recognition task. The BERT-based optimized model LERT is selected, and the pre-trained model LERT is fine-tuned for downstream sub-tasks using the small sample dataset labeled in step S3. S5. Process the unstructured data from the enterprise's production process and input it into the fine-tuned pre-trained model LERT+CRF. The CRF layer completes the named entity recognition task through the transition score matrix, realizing the conversion of unstructured data into process knowledge. Specifically: The input text is processed by embedding representation and adding positional encoding to form a vector sequence. After being input into LERT, different information is first extracted through a multi-head attention mechanism, then normalized and connected to a residual network, and then input into a fully connected layer. Then, normalization is performed and the residual network is connected again. The above steps are repeated multiple times to finally obtain the output vector. The obtained vector sequence is input into the optimization model LERT. Features are extracted from the input sequence using a multi-head self-attention mechanism, and a vector sequence containing contextual information is output. Multiple attention methods are learned through multi-head attention, with each head learning different attention weights. The calculation process is as follows: The input query vector, , for key-value pairs , The attention convergence function is expressed as: Through attention scoring function Map the two vectors to scalars, then perform a softmax operation to obtain the query. s and keys Attention weights: The scoring function for scaled dot product attention is: Obtain the output vector containing contextual information; The output vector of the optimization model LERT is input into a fully connected layer, the label probability of each word is calculated to obtain a state score, and the state score is then input into a CRF layer. The state score is used as a state feature, and the optimal route is obtained through the transition score matrix. S6. Further process the entities identified in step S5, and automatically import them into the graph database Neo4J according to the set knowledge architecture to complete the automated construction and visualization of the process knowledge base.

2. The mapping method for unstructured processing data technology knowledge according to claim 1, characterized in that, Multi-layer process knowledge model Specifically: in, For the design layer, For the process layer, To create a resource layer.

3. The mapping method for unstructured processing data technology knowledge according to claim 2, characterized in that, The design layer stores key design requirements for part design drawings, including part types. Parts materials Processing characteristics precision roughness The process layer stores process examples of the actual machining process, including cutting parameters. Processing steps Recommended equipment for manufacturing resource layer storage and processing includes machine tools. Cutting fluid Fixture Tool type Cutting tool materials .

4. The mapping method for unstructured processing data technology knowledge according to claim 1, characterized in that, In step S2, the small sample dataset includes: Part types: shaft parts, sleeve parts, disc parts, plate parts, cylindrical gear parts, and box-type parts; Parts materials: cast iron, mold steel, high manganese steel, stainless steel, titanium alloy, high temperature alloy, copper alloy; Machining features: holes, keyways, outer cylindrical surfaces, annular grooves, cones, inner cylindrical surfaces, planes, curved surfaces, threads, splines, fillets; Tool materials: high-speed steel, cemented carbide, ceramics, diamond, and cubic boron nitride; Cutting fluids: dry cutting fluids, aqueous solutions, emulsions, extreme pressure emulsions, rust-preventive emulsions, mineral oils, animal and vegetable oils, compound oils, machine oils, kerosene, water-based cutting fluids, and oil-based cutting fluids; Precision and roughness are represented by the fixed symbols IT and Ra; Machining processes: turning, milling, drilling, reaming, boring, heat treatment; Cutting parameters: Cutting parameters for a machining process.

5. The mapping method for unstructured processing data technology knowledge according to claim 1, characterized in that, In step S3, each dataset sample is labeled with BIOES, where B represents the beginning of an entity, I represents the inside of an entity, O represents the outside of an entity, E represents the end of an entity, and S represents a single character entity.

6. The mapping method for unstructured processing data technology knowledge according to claim 1, characterized in that, In step S503, the probability of each word's tag sequence appearing is: in, The score represents the actual path.

7. The mapping method for unstructured processing data technology knowledge according to claim 1, characterized in that, In step S6, the knowledge is automatically imported into the graph database Neo4J according to the pre-defined knowledge structure. Specifically: in, For knowledge relationships, As an attribute, For the relationship.

8. A mapping system for unstructured processing data and technological knowledge, characterized in that, include: The design module designs a multi-layered process knowledge model based on the actual processing process and the process decision-making process. The collection module designs the required content framework for the dataset based on the multi-layered process knowledge model constructed by the design module, and collects unstructured data to construct a small sample dataset. The annotation module uses the BIEOS annotation method to annotate the small sample dataset constructed by the collection module; The fine-tuning module defines the problem as a named entity recognition task based on the task objective, selects the BERT-based optimized model LERT, and uses the small sample dataset labeled by the annotation module to fine-tune the pre-trained model LERT for downstream sub-tasks. The transformation module processes unstructured data from the enterprise's production process and inputs it into the pre-trained model LERT+CRF, which has been fine-tuned by the fine-tuning module. The CRF layer then uses the transition score matrix to complete the named entity recognition task, achieving the transformation from unstructured data to process knowledge. Specifically: The input text is processed by embedding representation and adding positional encoding to form a vector sequence. After being input into LERT, different information is first extracted through a multi-head attention mechanism, then normalized and connected to a residual network, and then input into a fully connected layer. Then, normalization is performed and the residual network is connected again. The above steps are repeated multiple times to finally obtain the output vector. The obtained vector sequence is input into the optimization model LERT. Features are extracted from the input sequence using a multi-head self-attention mechanism, and a vector sequence containing contextual information is output. Multiple attention methods are learned through multi-head attention, with each head learning different attention weights. The calculation process is as follows: The input query vector, , for key-value pairs , The attention convergence function is expressed as: Through attention scoring function Map the two vectors to scalars, then perform a softmax operation to obtain the query. s and keys Attention weights: The scoring function for scaled dot product attention is: Obtain the output vector containing contextual information; The output vector of the optimization model LERT is input into a fully connected layer, the label probability of each word is calculated to obtain a state score, and the state score is then input into a CRF layer. The state score is used as a state feature, and the optimal route is obtained through the transition score matrix. The output module further processes the entities identified by the conversion module and automatically imports them into the graph database Neo4J according to the predefined knowledge architecture, thus completing the automated construction and visualization of the process knowledge base.

Images