A Knowledge Graph-based Sewing Process Knowledge Fusion and Intelligent Recommendation Method

Abstract

The present invention discloses a method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph. Aiming at the characteristics of long processing flow links and decentralized information in the sewing process, it integrates and manages multi-link information of processing from a global perspective, mainly including the following steps: designing a sewing knowledge ontology model in combination with the characteristics of the sewing industry knowledge structure and service system requirements; extracting process knowledge entities based on a bidirectional long short-term memory network model with conditional random fields; extracting process knowledge entity relationships based on a stacked pointer network model with a pre-trained language model; realizing knowledge storage based on the Neo4j graph database; establishing a sewing parameter recommendation method based on the mechanical properties of fabrics, and realizing intelligent applications in combination with intelligent search and intelligent recommendation functions. Based on knowledge graph technology, the present invention realizes the rapid call and recommendation of sewing process knowledge for the massive process information generated in the sewing process, and improves the independent decision-making ability of sewing production.

Description

Technical Field

[0001] The present invention relates to a method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph, belonging to the fields of sewing machinery, artificial intelligence, and information technology. Background Art

[0002] The clothing sewing process is the process link with the longest process flow, the most types of processes, and the most personnel and equipment involved in clothing production. Its processing technology is complex, production factors are diverse, and processing information is decentralized and isolated. It is difficult to integrate and manage the multi-link information of processing from a global perspective, resulting in information blockage in solving problems in a certain link. Currently, the recording and expression of sewing process knowledge mainly rely on traditional methods such as text, tables, graphics, and personnel experience. The equipment selection, process parameter adjustment, and fault repair means in the existing sewing production process are difficult to meet the production needs of multiple batches and personalization. Formulating a scientific and reasonable process route and sewing process parameters, reducing quality problems, and improving production efficiency have become the key to enhancing the competitiveness of clothing production enterprises. Currently, there is a lack of effective technical means for the fusion of sewing process knowledge in actual factory production, for the digital storage management, process reuse, and process intelligent optimization of process knowledge, to achieve rapid and accurate formulation of a reasonable sewing process route.

[0003] The knowledge graph has the ability to standardize the storage of unstructured information such as expert knowledge and use it in actual production, and can realize the utilization and recommendation of scattered knowledge. It has been widely used in recent years, and its forms mainly include search, question answering, reasoning, and recommendation, etc. Drawing on the extensive application of the knowledge graph in fields such as medicine and finance, domestic and foreign experts have introduced the knowledge graph technology into the manufacturing industry, aiming to achieve the acquisition, organization, and utilization of a large amount of process knowledge in the manufacturing field. However, currently, the application of the knowledge graph in the field of sewing machinery is still rare. Summary of the Invention

[0004] The object of the present invention is to establish a sewing process knowledge graph model and a knowledge recommendation method to realize the intelligent recommendation of sewing equipment knowledge and fabric mechanical property parameters.

[0005] In order to achieve the above object, the technical solution of the present invention provides a method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph, which is characterized by including the following steps:

[0006] Step 1, design a sewing knowledge ontology model:

[0007] According to the situation of sewing process corpus and the requirements of the knowledge graph, combined with the characteristics of the knowledge structure in the sewing industry and the functions that the service system needs to achieve, a knowledge graph ontology model is designed, including an equipment utilization layer and a resource utilization layer. Among them, the equipment utilization layer is used to store sewing process information and the categories of sewing production equipment, and the resource utilization layer is used to store product design knowledge, sewing electrical knowledge, sewing equipment knowledge, and sewing equipment operation and maintenance knowledge that can be called by the equipment utilization layer;

[0008] Step 2: Implement entity extraction of the corpus based on a bidirectional long short-term memory network with conditional random fields;

[0009] Step 3: Implement relation extraction of the corpus based on a stacked pointer network entity relation extraction model of a pre-trained language model;

[0010] Step 4: Knowledge storage based on the Neo4j graph database:

[0011] Fill the entities and their relations extracted in Step 2 and Step 3 into the sewing knowledge ontology model constructed in Step 1, and use the Neo4j graph database as a data storage tool to construct a sewing process knowledge graph;

[0012] Step 5: Sewing parameter recommendation based on the mechanical properties of fabrics, including the following steps:

[0013] Step 5.1: Establish a knowledge system for the basic properties of fabrics, measure the basic specifications of various fabrics, and conduct a preliminary classification of fabrics by level as input data that can be selected by users later;

[0014] Step 5.2: Establish a knowledge system for the mechanical properties of fabrics, conduct an analysis of the mechanical property parameters of fabrics, measure the mechanical properties of fabrics, conduct factor analysis on each performance index using the principal component analysis method, and extract the main factors;

[0015] Step 5.3: Establish a knowledge system for the sewing performance of fabrics, conduct sewing experiments on the fabrics described in Step 5.1, and conduct an objective evaluation in combination with the clothing sewing flatness standard. Use the evaluation data obtained in the sewing experiment to establish a stepwise multiple regression model, with the independent variable being the main factors extracted in Step 5.2 and the dependent variable being the sewing appearance flatness grade in the warp and weft directions of the fabric;

[0016] Step 5.4: Establish a fabric sewing shrinkage theoretical model based on the data obtained in Step 5.2 and Step 5.3, establish a relationship model between the mechanical properties of fabrics and the maximum sewing shrinkage rate, and realize the prediction of the fabric appearance flatness and the maximum fabric sewing shrinkage rate after sewing;

[0017] Step 5.5: Based on the fabric sewing shrinkage theoretical model established in Step 5.4 and the obtained experimental parameters, establish a sewing product design knowledge graph. After the user inputs the basic parameters of the fabric, the system identifies the fabric category and related process parameters, and calculates and outputs the processing and production performance of the fabric, including recommendations for the maximum sewing shrinkage rate of the fabric, bonding parameters, and processing technology descriptions.

[0018] Step 6: Implement intelligent search and intelligent recommendation functions based on the intelligent application of the knowledge graph. Among them, the intelligent search function includes entity recognition and entity and relationship query functions, and the intelligent recommendation function realizes the automatic output of recommended fabric sewing parameters based on the basic fabric information input by the user.

[0019] Preferably, in Step 1, the entity types available for designing the sewing knowledge ontology model include process documents, sewing equipment operation manuals, sewing equipment fault repair documents, and sewing machine enterprise website page information.

[0020] Preferably, Step 2 includes the following steps:

[0021] Step 2.1: Manually perform entity annotation on the original corpus based on the brat software, then use the BIO three-way annotation method to generate the annotated file, and use the annotated document as the dataset.

[0022] Step 2.2: Divide the dataset into a training set and a test set in a ratio of 7:3, generate cache files for the vocabulary and label tables according to the training set, and perform equal-length segmentation on the text, with each group consisting of 50 characters.

[0023] Step 2.3: Perform data preprocessing, select the top 5000 words with the highest word frequencies, and use <unk> \ <num> \ <pad>Instead;

[0024] Step 2.4: Obtain the output feature vectors of each word in the dataset through BiLSTM network training, and then determine the optimal output label sequence of the sentence through the CRF layer combined with Viterbi decoding, so as to determine the named entities;

[0025] Step 2.5: Use the trained model to extract labels from the test set corpus, and compare and test them with the manually labeled labels. The measurement criteria are precision and recall;

[0026] Step 2.6: Use the trained model to extract entities from the remaining corpus.

[0027] Preferably, the said step 3 includes the following steps:

[0028] Step 3.1: On the basis of the annotation in step 2.1, perform manual relation annotation, and divide the annotated document into a training set and a test set according to a ratio of 7:3;

[0029] Step 3.2: Use the main body, object and sentence of the corpus in the training set as inputs, and the relation as the label to train the entity relation extraction model of the stacked pointer network based on the pre-trained language model. When training:

[0030] The decoding layer of the entity relation extraction model of the stacked pointer network based on the pre-trained language model divides the corpus into words, obtains the context semantic information, and represents the words / characters;

[0031] Through the main body recognition layer, directly decode the encoding layer to recognize all possible main bodies;

[0032] Perform joint recognition of relations and objects, check all possible relations for each possible main body to determine whether there is a relation that can connect the object in the sentence with it;

[0033] Output triples in the form of corresponding main body, relation, and object;

[0034] Step 3.3: Use the trained model to extract labels from the test set corpus, and compare and test them with the manually labeled labels. The measurement criteria are precision and recall;

[0035] Step 3.4: Use the trained model to extract relations from the remaining corpus.

[0036] Preferably, in step 3.2, when recognizing all possible main bodies: first judge whether each character is the start or end of a certain main body, and then use the nearest matching principle to pair the recognized start and end to obtain a set.

[0037] Preferably, in step 6, the entity recognition function adopts methods of Chinese word segmentation, part-of-speech tagging, and entity naming recognition, and realizes the system's automatic recognition of sewing process information by traversing the dictionary of the established sewing process information corpus for the user's input query corpus.

[0038] The entity and its relationship query function means that based on the Cypher query language of Neo4j, when a user inputs a certain entity, the relevant entity information and the relationships between them can be displayed.

[0039] Preferably, in step 6, when implementing the intelligent recommendation function, the basic fabric information input by the user is selected from the fabric basic performance knowledge system established in step 5.1, and the fabric sewing parameters output are selected from the sewing product design knowledge graph established in step 5.5, including the maximum seam shrinkage rate, bonding parameters, and processing technology descriptions of the fabric.

[0040] The present invention can achieve the precise automatic construction of the sewing process knowledge graph and the intelligent recommendation of sewing equipment knowledge and fabric mechanical property parameters. Compared with the prior art, the present invention has the following advantages:

[0041] The present invention faces the characteristics of a long sewing process chain and isolated process information islands, integrates and manages the multi-link information of processing from a global perspective, avoids the blocking of problem-solving information in a certain link, and proposes a knowledge graph construction technology and application process for sewing process information fusion. For a large amount of unstructured text data such as design data, operation and use data, and maintenance support data generated during the sewing process, the knowledge graph technology is used to extract knowledge from the above unstructured text, achieving the effect of automatic unified standardization expression of unstructured knowledge, with the advantages of clear data organization and easy expansion. At the same time, based on the constructed knowledge graph, the intelligent recommendation of process parameters is realized by using the sewing parameter recommendation method based on fabric mechanical properties, improving the autonomous decision-making ability of the knowledge-based sewing process route. Description of the Drawings

[0042] Figure 1 It is a knowledge graph construction and application flow chart for sewing process information fusion provided by an embodiment of the present invention;

[0043] Figure 2 It is a structure diagram of the sewing process knowledge ontology model provided by an embodiment of the present invention;

[0044] Figure 3 It is a schematic structural diagram of a bidirectional long short-term memory network model based on conditional random fields provided by an embodiment of the present invention;

[0045] Figure 4 It is a schematic structural diagram of a stacked pointer network model based on a pre-trained language model provided by an embodiment of the present invention;

[0046] Figure 5 It is the flowchart of the recommended application of sewing parameters provided by the embodiments of the present invention;

[0047] Figure 6 It is the method for classifying and labeling fabric layers provided by the embodiments of the present invention;

[0048] Figure 7 It is the functional architecture diagram of the sewing process knowledge fusion and intelligent recommendation system provided by the embodiments of the present invention. Detailed implementation manners

[0049] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application.

[0050] Combined with Figure 1 , a sewing process knowledge fusion and intelligent recommendation method based on a knowledge graph provided by the embodiments of the present invention includes:

[0051] Step 1: Design a sewing knowledge ontology model. The structure of the sewing process knowledge ontology model designed in this embodiment is as Figure 2 shown.

[0052] The sewing knowledge ontology model disclosed in this embodiment is divided into an equipment utilization layer and a resource utilization layer. Among them, the equipment utilization layer mainly stores sewing process information and the categories of sewing production equipment, and the resource utilization layer stores product design knowledge, sewing electrical knowledge, sewing equipment knowledge, and sewing equipment operation and maintenance knowledge that can be called by the equipment utilization layer. Figure 2 The solid lines in [[ ]] represent the "classes" in the sewing knowledge ontology model and the relationships between them, the dashed lines represent the "entities" included in the "classes" and the relationships between them, and the dotted lines represent the attributes included in the "entities" and the relationships between them.

[0053] The equipment utilization layer further includes a process layer and an equipment layer. The sewing process categories included in the process layer include but are not limited to flat sewing, bartacking, buttonholing, overlock sewing, and button sewing processes. The equipment layer includes the production enterprises, equipment origins, and equipment categories of sewing equipment. The equipment categories further include but are not limited to flat sewing machines, bartacking machines, buttonholing machines, overlock sewing machines, button sewing machines, overlock sewing machines, double needle machines, and overlock sewing machines. Each equipment category also includes different series categories, and each series category also includes a series of branch models, and each branch model is used as an "entity". Figure 2 Shows some categories included in the flat sewing machine. For example, the computerized flat sewing machine includes the H93S series, and the H93S model includes a series of branch models. The branch models are regarded as "entities" and stored in the H93S series category.

[0054] In the resource utilization layer, in this embodiment, the principle of resource invocation in the equipment utilization layer of the sewing knowledge ontology model is introduced with an example in the equipment operation and maintenance knowledge system. The fault operation and maintenance part stores the fault codes of each device. For example, "E605" is a fault code of the H93S model device and is stored as an "entity" in the "category" of H93S fault codes. The relationship between the H93S series in the equipment utilization layer and the fault code _H93S is "has_fault_content". At the same time, "E605", as a fault code entity, also has three attributes: "fault_content", "possible_cause_of_fault", and "fault_processing". The specific knowledge included in the attributes has been listed in the figure.

[0055] After completing the above work, it is necessary to further refine the knowledge information in the sewing knowledge ontology model, such as Figure 1 shown, perform step 2 entity extraction and step 3 relationship extraction.

[0056] Step 2: Perform entity extraction using a bidirectional long short-term memory network based on conditional random fields. Figure 3 It is a schematic diagram of the entity extraction model structure provided by the embodiment of the present invention.

[0057] The original corpus is entity-annotated based on the brat software, and then the BIO three-way annotation method is used to generate the annotated file. BIO annotation is an entity annotation method. The first character of a named entity is represented by "B", the last character is represented by "I", and non-named entities are represented by "O". For example Figure 3 In the sentence "The flat sewing machine has needle skipping during operation" in the input layer shown in, it is: flat / B seam / I machine / I work / O during / O operation / O appears / O needle skipping / B, where "flat" is the first character "B" of the named entity in the "equipment type" category, "machine" is the last character of the named entity in the "equipment type" category, and the subsequent "work", "operation", etc. are all non-entity components "O".

[0058] The annotated document is used as a dataset, which is divided into a training set and a test set in a ratio of 7:3. Cache files of the vocabulary table and the label table are generated according to the training set, and the text is segmented into equal lengths, with each group consisting of 50 characters. Subsequently, data preprocessing is performed. The first 5000 characters with the highest word frequencies are taken, and the remaining irrelevant vocabulary, numerals, and words used for extended alignment are respectively <unk> \ <num> \ <pad>Instead.

[0059] Subsequently, the output feature vectors of each word in the dataset are obtained through the training of the BiLSTM network. The BiLSTM network consists of a forward and a backward neural network. It takes the word vectors from the previous layer as input, breaks down the forward and backward output states to obtain a complete state sequence, and maps it to the corresponding labels.

[0060] Then, through the CRF layer combined with Viterbi decoding, the optimal output label sequence of the sentence is determined, thereby determining the named entities. The role of the CRF layer is to solve the problem that the label sequence output from the BiLSTM model may be invalid by learning the constraints between labels. Finally, the trained model is used to extract labels from the test set corpus and compare them with the manually annotated labels. The measurement criteria are precision and recall. The trained model is used to extract entities from the remaining corpus.

[0061] Step 3: Entity relation extraction based on the stacked pointer network of the pre-trained language model (RoBERTa-wwm-CasRel), Figure 4 This is the schematic diagram of the relation extraction model structure provided by the embodiment of the present invention.

[0062] First, based on the annotation in Step 2, relation annotation is carried out, and the annotated document is divided into a training set and a test set according to a ratio of 7:3. The main entities, object entities, and sentences in the corpus of the training set are used as input, and the relations are used as labels. After passing through the RoBERTa-wwm decoding layer, the corpus is segmented word by word to obtain context semantic information and represent the words / characters.

[0063] Subsequently, through the "main entity" recognition layer, the encoding layer is directly decoded to identify all possible "main entities". First, it is judged whether each character is the "start" or "end" of a certain "main entity", and then the identified "start" and "end" are paired using the nearest matching principle to obtain a set. For example, in Figure 4 the input corpus "EO8 represents a broken needle, please replace the sewing needle", it can be considered that "E08", "broken needle", and "replace the sewing needle" are 3 possible "main entities". Taking the broken needle as an example, the character "broken" is the "start", and the character "needle" is the "end".

[0064] Then, the joint recognition of "relation" and "object" is carried out. For each possible "main entity", all possible relations are checked to determine whether there is a relation that can connect the "object" in the sentence with it. For example, "broken needle" is the "object" of "E08", and the relation between them is "fault_content", and "replace the sewing needle" is the "object" of "E08", and the relation between them is "fault_processing".

[0065] Finally, output triples in the form of corresponding "subject", "relationship", and "object".

[0066] Train a stacked pointer network entity relation extraction model based on a pre-trained language model. Use the trained model to extract labels from the test set corpus and compare them with the manually annotated labels. The measurement criteria are precision and recall. Use the trained model to extract the relationships between entities in the remaining corpus.

[0067] After completing the above steps, perform Step 4: Knowledge storage based on the Neo4j graph database. Fill the entities and their relationships extracted in Step 2 and Step 3 into the sewing knowledge ontology model constructed in Step 1, and use the Neo4j graph database as a data storage tool to construct a sewing process knowledge graph.

[0068] Step 5: Sewing parameter recommendation based on the mechanical properties of the fabric. For ease of understanding, please refer to Figure 5 which is the application flow chart of the sewing parameter recommendation provided by the embodiment of the present invention.

[0069] Data layer. First, select 95 common fabrics in the flat seam process as experimental materials, test the basic specification parameters of the selected fabrics, use the KES experiment to test the mechanical properties of the materials, and construct a basic performance knowledge system for the fabrics; subsequently, perform a correlation analysis on the mechanical property data of the fabrics, and combine expert evaluations to achieve a correlation analysis of the fabric mechanical properties with sewing flatness and maximum seam shrinkage rate.

[0070] Service layer. First, combine the previous experimental data to establish a basic performance knowledge system for the fabrics as input data that can be selected by users later, Figure 6 which is the fabric hierarchical classification labeling method provided by the embodiment of the present invention. Subsequently, establish a fabric mechanical property knowledge system based on the fabric mechanical property parameters and fabric basic specification parameters. Through mathematical model calculations, through the fabric mechanical property knowledge system established based on sewing parameters, bonding parameters, and processing technology descriptions, predict the appearance flatness of the fabric and the maximum seam shrinkage rate after sewing, and provide intelligent recommendations for bonding parameters and processing technology descriptions.

[0071] Application layer. According to the established mathematical model and the obtained experimental parameters, establish a sewing product design knowledge graph. The user inputs the basic parameters of the fabric, and the system identifies the fabric category and relevant process parameters, and calculates and outputs the processing and production performance of the fabric.

[0072] Its implementation includes the following steps:

[0073] Step 5.1: Establish a knowledge system for the basic properties of fabrics. Select 95 common fabrics in the flat-seam process as experimental materials. Measure the basic specifications of the experimental materials, such as thickness, mass per unit area, warp density, and weft density, and preliminarily classify the fabrics by level as input data that can be selected by users later.

[0074] Step 5.2: Establish a knowledge system for the mechanical properties of fabrics, analyze the mechanical property parameters of fabrics. Use the KES fabric style tester system to measure the mechanical properties of fabrics, conduct a correlation analysis on the mechanical property data of fabrics, perform factor analysis on each performance index using the principal component analysis method, and extract the main factors to construct a basic performance knowledge system for fabrics. Combine expert evaluation to achieve a correlation analysis between the mechanical properties of fabrics and sewing flatness and the maximum seam shrinkage rate.

[0075] Step 5.3: Establish a knowledge system for the sewing performance of fabrics. Conduct sewing experiments on the above-mentioned fabrics and conduct an objective evaluation in combination with the AATCC-88B clothing sewing flatness standard. Use the evaluation data obtained in the sewing experiment to establish a stepwise multiple regression model, with the independent variable being the main factor extracted in Step 5.2 and the dependent variable being the sewing appearance flatness grade in the warp and weft directions of the fabric.

[0076] Step 5.4: Establish a fabric seam shrinkage theoretical model based on the data described in Steps 5.2 and 5.3, establish a relationship model between the mechanical properties of fabrics and the maximum seam shrinkage rate, and realize the prediction of the fabric appearance flatness and the maximum seam shrinkage rate after sewing.

[0077] Step 5.5: According to the mathematical model described in Step 5.4 and the obtained experimental parameters, establish a sewing product design knowledge graph. When the user inputs the basic parameters of the fabric, the system identifies the fabric category and related process parameters, and calculates and outputs the processing and production performance of the fabric, including recommendations for the maximum seam shrinkage rate, bonding parameters, and processing process descriptions of the fabric.

[0078] After completing the above steps, integrate the above functions and proceed to Step 6. Figure 7 This is the functional architecture diagram of the sewing process knowledge integration and intelligent recommendation system provided by the embodiment of the present invention.

[0079] Establish a sewing process knowledge integration and intelligent recommendation system based on the Web side to realize intelligent search and intelligent recommendation functions based on the knowledge graph. The intelligent search function includes entity recognition, entity and its relationship query functions, and the intelligent recommendation function is the sewing parameter recommendation based on the mechanical properties of fabrics described in Step 5.

[0080] The entity recognition function is implemented using a dictionary-based method. By adopting Chinese word segmentation, part-of-speech tagging, and entity naming recognition methods, and traversing the established sewing process information corpus with a dictionary, the named entity recognition in the sewing field is realized. The entity recognition function can enable the system to automatically recognize sewing process information when the user inputs a query corpus.

[0081] The entity query function is built based on the Cypher query language of Neo4j and can search for entities and relationships related to a certain entity. For example, when inputting "Juki Group" (sewing machine enterprise), the corresponding sewing machine models will be displayed. The relationship query is to query the triple relationship entity1-[relation]->entity2. For example, by specifying the query results of the first entity entity1 and the second entity entity2, that is, specifying "China Standard" (sewing machine enterprise) and "GC6180MT3" (sewing machine model), a corresponding relationship visualization interface can be output.

[0082] The intelligent recommendation function can enable the system to automatically output recommended fabric sewing parameters when the user inputs the basic fabric information. The basic fabric information is selected from the fabric basic performance knowledge system established in step 5.1, and the fabric sewing parameters are selected from the sewing product design knowledge graph established in step 5.5, including the maximum sewing shrinkage rate, bonding parameters, and processing technology instructions of the fabric.< / pad> < / num> < / unk> < / pad> < / num> < / unk>

Claims

1. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph, characterized in that, it includes the following steps: Step 1: Design a sewing knowledge ontology model: According to the sewing process corpus situation and graph requirements, combined with the characteristics of the sewing industry knowledge structure and the functions to be realized by the service system, design a knowledge graph ontology model, including an equipment utilization layer and a resource utilization layer. Among them, the equipment utilization layer is used to store sewing process information and the categories of sewing production equipment, and the resource utilization layer is used to store product design knowledge, sewing electrical knowledge, sewing equipment knowledge, and sewing equipment operation and maintenance knowledge that can be called by the equipment utilization layer; Step 2: Implement entity extraction of the corpus based on a bidirectional long short-term memory network with conditional random fields; Step 3: Implement relationship extraction of the corpus based on a stacked pointer network entity relationship extraction model with a pre-trained language model; Step 4: Knowledge storage based on the Neo4j graph database: Fill the entities and their relationships extracted in Step 2 and Step 3 into the sewing knowledge ontology model constructed in Step 1, and use the Neo4j graph database as a data storage tool to construct a sewing process knowledge graph; Step 5: Sewing parameter recommendation based on the mechanical properties of fabrics, including the following steps: Step 5.1: Establish a fabric basic performance knowledge system, measure the basic specifications of various fabrics, and conduct a preliminary classification of fabrics by level as input data that can be selected by users later; Step 5.2: Establish a fabric mechanical property knowledge system, conduct fabric mechanical property parameter analysis, measure the mechanical properties of fabrics, perform factor analysis on each performance index using the principal component analysis method, and extract the main factors; Step 5.3: Establish a fabric sewing performance knowledge system, conduct sewing experiments on the fabrics described in Step 5.1, and conduct an objective evaluation in combination with the clothing sewing flatness standard. Use the evaluation data obtained in the sewing experiments to establish a stepwise multiple regression model, with the independent variable being the main factors extracted in Step 5.2 and the dependent variable being the sewing appearance flatness grade in the warp and weft directions of the fabric; Step 5.4: Establish a fabric shrinkage theory model based on the data obtained in Step 5.2 and Step 5.3, establish a relationship model between the fabric mechanical properties and the maximum shrinkage rate, and realize the prediction of the fabric appearance flatness and the maximum shrinkage rate of the fabric after sewing; Step 5.5: According to the fabric shrinkage theory model established in Step 5.4 and the obtained experimental parameters, establish a sewing product design knowledge graph. After the user inputs the basic parameters of the fabric, the system identifies the fabric category and related process parameters, and calculates and outputs the processing and production performance of the fabric, including recommendations for the maximum shrinkage rate of the fabric, bonding parameters, and processing process descriptions; Step 6: Intelligent applications based on the knowledge graph, realizing intelligent search functions and intelligent recommendation functions. Among them, the intelligent search function includes entity recognition functions and entity and their relationship query functions, and the intelligent recommendation function realizes the automatic output of recommended fabric sewing parameters based on the basic information of the fabric input by the user.

2. The method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph according to claim 1, characterized in that, In Step 1, the entity types available for designing the sewing knowledge ontology model include process documents, sewing equipment operation manuals, sewing equipment fault repair documents, and sewing machine enterprise website page information.

3. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph as described in Claim 1, characterized in that, Step 2 includes the following steps: Step 2.1: Manually annotate entities for the original corpus based on brat software, and then use the BIO three - dimensional annotation method to generate the annotated file, and use the annotated document as a data set; Step 2.2: Divide the data set into a training set and a test set in a ratio of 7:3, generate cache files for the word list and label list according to the training set, and perform equal - length segmentation on the text, with every 50 characters as a group; Step 2.3: Perform data preprocessing. Select the top 5000 words with the highest word frequencies, and use <unk> \ <num> \ <pad>Replace; < / pad> < / num> < / unk> to replace the remaining irrelevant words, numerals, and words used for extended alignment respectively. <unk> \ <num> \ <pad>Replace; < / pad> < / num> < / unk> Step 2.4: Obtain the output feature vector of each character in the data set through training with a BiLSTM network, and then determine the optimal output label sequence of the sentence through the CRF layer combined with Viterbi decoding, so as to determine the named entities; Step 2.5: Use the trained model to extract labels from the test set corpus, and compare them with the manually annotated labels. The measurement criteria are precision and recall; Step 2.6: Use the model trained and tested to extract entities from the remaining corpus.

4. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph as described in Claim 3, characterized in that, Step 3 includes the following steps: Step 3.1: On the basis of the annotation in Step 2.1, perform manual relationship annotation, and divide the annotated document into a training set and a test set in a ratio of 7:3; Step 3.2: Use the subject, object, and sentence of the corpus in the training set as inputs, and the relationship as the label to train the entity relationship extraction model of the stacked pointer network based on the pre - trained language model. During training: The decoding layer of the entity relationship extraction model of the stacked pointer network based on the pre - trained language model divides the corpus into words, obtains the context semantic information, and represents the characters / words; Pass through the subject recognition layer, directly decode the encoding layer, and recognize all possible subjects; Perform joint recognition of relationships and objects, check all possible relationships for each possible subject to determine whether there is a relationship that can connect the object in the sentence with it; Output triples in the form of corresponding subjects, relationships, and objects; Step 3.3: Use the trained model to extract labels from the test set corpus, and compare them with the manually annotated labels. The measurement criteria are precision and recall; Step 3.4: Use the model trained and tested to extract relationships from the remaining corpus.

5. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph as described in Claim 4, characterized in that, In Step 3.2, when recognizing all possible subjects: first judge whether each character is the start or end of a certain subject, and then use the nearest matching principle to pair the recognized start and end to obtain a set.

6. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph as described in Claim 1, characterized in that, In step 6, the entity recognition function adopts methods of Chinese word segmentation, part-of-speech tagging, and entity named recognition. By traversing the dictionary of the established sewing process information corpus, the system can automatically recognize the sewing process information when the user inputs a query corpus. The entity and its relationship query function means that based on the Cypher query language of Neo4j, when the user inputs a certain entity, the system can display the entity information related to it and the relationships between them.

7. A method for sewing process knowledge fusion and intelligent recommendation based on a knowledge graph as described in claim 1, characterized in that in step 6, when implementing the intelligent recommendation function, the basic fabric information input by the user is selected from the fabric basic performance knowledge system established in step 5.1, and the fabric sewing parameters output are selected from the sewing product design knowledge graph established in step 5.5, including the maximum seam shrinkage rate, bonding parameters, and processing technology description of the fabric.

Images