A knowledge graph construction method for intelligent design of an electric arc additive process

Abstract

The application discloses a kind of electric arc additive process intelligent design knowledge graph construction method, comprising the following steps: S1, based on electric arc additive process design file includes process parameters, with the characteristic of experiential knowledge summarized by process design personnel, determine the electric arc additive knowledge hierarchy of the ontology layer of electric arc additive process intelligent design knowledge graph, construct ontology layer;S2, generate electric arc additive process knowledge entity identification corpus set;S3, construct the word vector space of electric arc additive;S4, construct the electric arc additive process knowledge entity extraction model of multiple neural network model, realize the extraction to unstructured electric arc additive process knowledge;S5, the knowledge extracted in step S4 is fused with the ontology layer constructed in step S1 according to corresponding process parameter entity, and the knowledge graph of electric arc additive process intelligent design is formed;S6, the formed knowledge graph is stored in graph database, realizes the knowledge service to electric arc additive process intelligent design.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electric arc additive manufacturing, and particularly relates to a knowledge graph construction method for intelligent design of an electric arc additive process. BACKGROUND

[0002] The electric arc additive technology is a method for rapid near-net shaping of metal parts based on a digital three-dimensional model under the control of a program, by using an electric arc generated by a welding machine of a shielded metal arc welding as a heat source, synchronously adding a metal wire at the same time. The method has advantages such as high material utilization, high deposition efficiency, and being applicable to forming of complex parts, and is currently widely used in manufacturing and repairing of parts. The process parameter design of the electric arc additive directly affects the forming effect and quality of the parts. Due to the multiple types of process parameters involved in the additive process, the influence relationship between the parameter design and the forming effect is relatively fuzzy, and the forming effect of the parts cannot be accurately predicted. In order to use the electric arc additive process for part processing and manufacturing, a large number of process experiments are carried out before formal processing to obtain process parameters with good forming effect for part forming. In addition, process experiments need to be carried out again to obtain optimal process parameters for each new part forming, which greatly increases the cost of part manufacturing, and ignores the process knowledge generated by previous process experiments. Therefore, in order to realize the reuse of process knowledge, help the electric arc additive process to realize intelligent design of parameters, improve the manufacturing quality of parts and reduce the production cost, it is necessary to construct an electric arc additive knowledge base based on a field knowledge graph.

[0003] The addition of the electric arc additive process design knowledge graph can utilize the electric arc additive process design knowledge stored in the graph to help process designers in this field to quickly complete the process parameter design of parts, and quickly search for similar process design cases to assist designers in rapid design. Based on the case knowledge in the graph, the problem of repeated process experiments is avoided, and the cost consumed by part manufacturing is greatly reduced.

[0004] However, there is no knowledge graph or knowledge base applied to the electric arc additive part process design field, so it is currently very difficult to realize knowledge-driven intelligent design of the electric arc additive process. In the process of constructing the electric arc additive process design knowledge graph, the named entity recognition task is a key link in the whole process. Through this task, the key process knowledge is extracted from the input text knowledge, and the knowledge graph data collection work is completed, so the quality of the electric arc additive process design knowledge in the knowledge graph is directly determined by the completion of this task.

[0005] However, the main problems existing in the named entity recognition of the electric arc additive process knowledge data at present are as follows:

[0006] In order to make the computer recognize the input text data, it is necessary to use a representation model to vectorize the text data, and currently a pre-trained model BERT is usually used to represent the input text data, although such a model is pre-trained with a large amount of data and has excellent vector representation capability, but for electric arc additive such a specific field, there is no effective model training set, and the internal features of the process vocabulary in the field are complex, so that the pre-trained model cannot effectively reflect the semantic features of the input text, resulting in low entity recognition accuracy.

[0007] In the named entity recognition task, after the vectorization of the input text is completed, the key entities in the text need to be labeled, and the sequence labeling model of BiLSTM+CRF or BIGRU+CRF is commonly used, which is prone to feature extraction failure when labeling long sequence text.

[0008] In summary, the problems and defects of the knowledge graph construction technology in the application field of electric arc additive are: there is no knowledge graph for electric arc additive process design; there is no named entity recognition method and data set for electric arc additive process knowledge data; the existing word vector representation model cannot accurately represent the text data in the electric arc additive field, and the existing model has low accuracy in process entity recognition. SUMMARY

[0009] The purpose of the present application is to solve the above problems, and provide an electric arc additive process intelligent design knowledge graph construction method which uses multi-class deep learning model to accurately and efficiently extract input process knowledge text, and lays a data foundation for knowledge-driven electric arc additive process intelligent design.

[0010] To solve the above technical problems, the technical scheme of the present application is: a knowledge graph construction method for electric arc additive process intelligent design, comprising the following steps:

[0011] S1, based on the characteristics of the process parameters contained in the electric arc additive process design file and the experiential knowledge summarized by the process design personnel, determine the electric arc additive knowledge hierarchy of the ontology layer of the electric arc additive process intelligent design, and construct the ontology layer;

[0012] S2, collect electric arc additive process design files and experiential design knowledge, preprocess the data, and generate an electric arc additive process knowledge entity recognition corpus set;

[0013] S3, collect electric arc additive field vocabulary, and construct an electric arc additive word vector space;

[0014] S4, construct an electric arc additive process knowledge entity extraction model composed of multiple neural network models, and realize the extraction of unstructured electric arc additive process knowledge;

[0015] S5, the knowledge extracted in step S4 is fused with the ontology layer constructed in step S1 according to corresponding process parameter entities, and a knowledge graph of the electric arc additive process intelligent design is formed;

[0016] S6, the formed knowledge graph is stored in a graph database, and knowledge service for the electric arc additive process intelligent design is realized.

[0017] Further, in step S1, the characteristic parameters of the electric arc additive process design knowledge include but are not limited to product model characteristics, equipment parameter characteristics, process parameter characteristics, defect characteristics, path planning experience, and process parameter design experience, and the construction of the ontology framework model of the knowledge graph is completed.

[0018] Further, in step S2, the data preprocessing is the manual cleaning of the obtained process knowledge, including but not limited to deleting duplicate data, correcting wrong characters, supplementing unclearly described knowledge, and based on the characteristic parameters of the electric arc additive process design knowledge, data labeling is performed on the cleaned knowledge to obtain an electric arc additive process knowledge entity recognition corpus set.

[0019] Further, the field vocabulary in step S3 includes but is not limited to materials: 2319 aluminum alloy, 5A06 aluminum alloy, nickel-titanium metal; path planning: contour swing, linear scan filling; heat treatment: solid solution aging, vacuum annealing, stress relief annealing; product entity: battery box, oil tank shell; finally, a field dictionary set for electric arc additive process intelligent design is generated, the corpus set is segmented based on the introduction of the dictionary set, the segmentation result is trained based on the Skip-gran skip word model, and a word vector space special for electric arc additive process intelligent design is obtained.

[0020] Further, step S4 includes the following steps:

[0021] S41, based on the input unstructured electric arc additive process design knowledge, a text word vector representation layer composed of RoBERTa and the word vector space of the electric arc additive process intelligent design is input, and the representation of the input text vectorization is completed;

[0022] S42, the word vector representations output by RoBERTa and the word vector space for the input text are fused to generate a new word vector representation, and the weight of the fusion is dynamically modified;

[0023] S43, for the input text, the RoBERTa model is used to represent the text as a feature vector to complete the classification of the process knowledge;

[0024] S44, using BiGRU-MHA-CRF to perform arc additive knowledge named entity recognition task on the word vector representation obtained in step S42, completing labeling according to the labeling strategy of BIOES, and completing extraction of triple knowledge.

[0025] Further, in step S41, the input text uses the method of simultaneously calculating word vector representation by RoBERTa and arc additive word vector space, and the RoBERTa model optimizes and iterates parameters based on attention mechanism according to the advantages and disadvantages of the recognition result; in step S42, the word vector representations output by RoBERTa and arc additive word vector space respectively are dynamically optimized in the weight distribution of vector fusion based on GA optimization algorithm in the vector fusion process; in step S43, the RoBERTa model outputs the classification effect of the input process knowledge text based on the sentence vector representation of the input text, which is used to distinguish which arc additive case the knowledge belongs to, such as the case name of "battery box" or "cabin body"; in step S44, the BiGRU-MHA-CRF performs arc additive knowledge named entity recognition task, connects the output generated by RoBERTa and word vector space fusion with the input of BiGRU, uses multi-head attention MHA to improve the recognition ability of BiGRU, and outputs the score of each label, and finally constraints the predicted label based on the CRF layer to ensure the effectiveness of the labeled entity.

[0026] Further, in step S5, according to the text classification result of step S43, the obtained triple knowledge entity category is mapped into the word vector space in step S3, similarity calculation and knowledge fusion are performed, and finally the construction of the knowledge graph of the arc additive process intelligent design field is completed.

[0027] Further, in step S5, according to the corresponding process parameter entity, the knowledge graph of arc additive process intelligent design is formed, and in the process of process entity fusion, the similarity calculation is completed in the arc additive vector space, and the formula is as follows:

[0028]

[0029] Based on the size of the entity similarity, the fusion and alignment of the data layer entity and the ontology layer entity are completed, and finally the knowledge graph for arc additive process intelligent design is completed.

[0030] Furthermore, in step S6, knowledge services for intelligent design of electric arc additive manufacturing processes are implemented. For problems encountered in the design process of electric arc additive manufacturing processes, a knowledge question-and-answer format based on knowledge graphs is adopted. Process designers raise questions to the question-and-answer system, and semantic parsing is used to understand the system. The questions are then converted into query statements of graph data and entered into the electric arc additive manufacturing knowledge base to complete the answers to the corresponding process knowledge questions.

[0031] The beneficial effects of this invention are as follows: This invention provides a knowledge graph construction method for intelligent design of arc additive manufacturing processes. It also constructs a word vector space specifically for the arc additive manufacturing field, proposes a method to enhance the domain knowledge of pre-trained word vector representation models using a professional domain word vector space, thereby improving the model's ability to recognize domain knowledge. Furthermore, it dynamically optimizes the weights of the vector fusion process based on a genetic algorithm, enhancing the model's ability to represent the semantic features of arc additive manufacturing process knowledge. Additionally, the entity extraction model provided by this invention improves the accuracy of extracting process knowledge entities. On the other hand, the intelligent design knowledge graph for arc additive manufacturing processes constructed using this invention enables the reuse of design knowledge from the arc additive manufacturing domain knowledge graph based on a knowledge question-and-answer method. This helps designers in the arc additive manufacturing field improve the efficiency of process parameters, while simultaneously improving the manufacturing quality of parts and reducing production costs. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the knowledge graph construction process for intelligent design of electric arc additive manufacturing process according to the present invention.

[0033] Figure 2 This is a structural diagram of the intelligent design knowledge model for arc additive manufacturing proposed in this invention;

[0034] Figure 3 This invention proposes an extraction model for intelligent design knowledge applicable to electric arc additive manufacturing processes.

[0035] Figure 4 This is a diagram illustrating the effect of the invention's construction of a domain dictionary set to help the model better identify relevant terms in the electric arc additive manufacturing process. Detailed Implementation

[0036] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0037] like Figures 1 to 4 As shown, the present invention provides a knowledge graph construction method for intelligent design of arc additive manufacturing processes, comprising the following steps:

[0038] S1, the feature parameters of the arc additive process design file include process parameters, and the feature parameters of the experience knowledge summarized by the process designers, determine the arc additive knowledge hierarchy of the knowledge graph ontology layer of the arc additive process intelligent design, and construct the ontology layer.

[0039] In this step S1, the feature parameters of the arc additive process design knowledge include but are not limited to: product model features, equipment parameter features, process parameter features, defect features, path planning experience, and process parameter design experience, and the construction of the ontology framework model of the knowledge graph is completed.

[0040] Since the logical relationship between the specific process parameters included in the arc additive process is complex, and there is currently no knowledge representation model for the arc additive field, in order to clearly represent the process knowledge, a top-down knowledge graph construction method is adopted, the construction of the ontology layer of the arc additive process intelligent design knowledge graph is first completed, the basic framework of the knowledge graph is determined according to the ontology layer, and the arc additive process knowledge can be clearly represented by the knowledge graph.

[0041] The ontology structure of the arc additive process is as shown in Figure 2 According to the requirements of the arc additive process intelligent design, the top layer parameter contents such as but not limited to material type, part feature, processing record, heat treatment feature, forming defect feature, equipment parameter feature, and forming process feature are determined, and the top layer parameter contents are refined, such as that the part feature includes part size, part name, part code, and part model drawing; the equipment parameter feature includes equipment name, forming voltage, forming current, and wire feeding speed; and the forming process feature includes interlayer cooling time, forming mode, and path planning method. The above process parameters are imported into the Protégé ontology modeling software for modeling, and the hierarchy construction of the ontology model is completed.

[0042] S2, collect the arc additive process design file and the experience design knowledge, pre-process the data, and generate the arc additive process knowledge entity recognition corpus set.

[0043] In step S2, the data preprocessing is the manual cleaning of the obtained process knowledge, including but not limited to deleting duplicate data, correcting errors, supplementing unclear knowledge, and based on the feature parameters of the arc additive process design knowledge, the cleaned knowledge is data labeled to obtain the arc additive process knowledge entity recognition corpus set.

[0044] Due to the low degree of intelligent informationization of the current arc additive process design, and no special knowledge base, this step collects the process design file and experiential design knowledge in the arc additive. Since the collected process knowledge may contain some characters or symbols that cannot be recognized by the computer, in order to improve the quality of the process knowledge, the process knowledge is preprocessed and the arc additive process knowledge entity recognition corpus set is generated.

[0045] S3, collect the vocabulary in the field of arc additive, and construct the word vector space of arc additive.

[0046] The field vocabulary in step S3 includes but is not limited to material type: 2319 aluminum alloy, 5A06 aluminum alloy, nickel-titanium metal; path planning type: contour swing, linear scan filling; heat treatment type: solid solution aging, vacuum annealing, stress relief annealing; product entity type: battery box, oil tank shell; finally generate a dictionary set for the field of arc additive process intelligent design, introduce the dictionary set to segment the corpus set, train the segmentation results based on Skip-gran skip word model, and obtain the word vector space special for arc additive process intelligent design.

[0047] Since the vectorization representation of the text data for a specific field ignores the field specialty at present, the entity recognition accuracy is low, the pre-training model fusion specific field word vector space method is proposed for the text vectorization representation part to solve the existing problems.

[0048] According to the hierarchical model constructed in step S1, for the corresponding process knowledge entity, the professional vocabulary of arc additive is crawled based on python, and after the professional vocabulary is obtained through artificial verification, the arc additive process intelligent design dictionary set is formed by classification. The features of the arc additive field vocabulary include but are not limited to material type: "2319 aluminum alloy, 5A06 aluminum alloy, nickel-titanium metal", path planning type: "contour swing, linear scan filling", heat treatment type: "solid solution aging, vacuum annealing, stress relief annealing", product entity type: "battery box, oil tank shell", etc. The purpose of training the dictionary set is to prevent the generation of key process vocabulary in the segmentation process, which causes the field word vector space to be unable to accurately describe the semantic relationship between each process vocabulary. The specific effect can be shown in Figure 4 .

[0049] Relying on the constructed field dictionary set, the preprocessed corpus in step S2 is segmented, and the word vector space of arc additive is trained based on the Skip-gram skip word model.

[0050] S4, construct an arc additive process knowledge entity extraction model composed of multiple neural network models, and realize the extraction of unstructured arc additive process knowledge. As Figure 3 shown.

[0051] The step S4 comprises the following sub-steps:

[0052] S41, based on the input unstructured electric arc additive process design knowledge, inputting a text word vector representation layer composed of RoBERTa and electric arc additive process intelligent design word vector space, completing the representation of input text vectorization.

[0053] The method of simultaneously calculating word vector representation of input text by RoBERTa and electric arc additive word vector space solves the problem that the specific field vocabulary is not considered at present and the BERT model is often used for representation. RoBERTa is an optimized model based on the pre-training model BERT, which uses a larger amount of data for training and is superior to the BERT model in text representation ability. In the representation process, the RoBERTa is optimized and iterated in parameters according to the advantages and disadvantages of the identification results through the multi-head attention mechanism in the Transformer, and the input text is calculated in two different word vectors in RoBERTa and word vector space.

[0054] S42, for the word vector representation output by RoBERTa and word vector space respectively for the input text, the two are fused to generate new word vector representation, and the weight of fusion is dynamically modified.

[0055] In the fusion process of vectors, the two vectors obtained are calculated by using the weight method, and the weight distribution of vector fusion is dynamically modified based on the GA optimization algorithm according to the calculation results of the verification set, and the optimal weight value is finally obtained. The two vectors are fused to generate new word vector representation.

[0056] S43, for the input text, the RoBERTa model performs feature vector representation on the text to complete the classification of process knowledge.

[0057] For the input text knowledge, the RoBERTa model outputs the classification effect of the input process knowledge text based on the sentence vector representation of the input text, which is used to distinguish which kind of electric arc additive case the knowledge belongs to, such as the case name of "battery box" or "cabin", which is used to distinguish the specific data of which kind of process case and which kind of process knowledge, prevent the wrong classification of electric arc additive process knowledge and experience, and thus reduce the knowledge service ability of the electric arc additive process design knowledge graph.

[0058] S44, for the word vector representation obtained in step S42, BiGRU-MHA-CRF is used to recognize electric arc additive knowledge named entity recognition task, the annotation is completed according to the annotation strategy of BIOES, and the extraction of triple knowledge is completed.

[0059] BiGRU-MHA-CRF is used to perform the named entity recognition task of the arc additive process knowledge, and the knowledge recognition and extraction are performed on the arc additive process knowledge corpus constructed above. First, the text data is input into the representation layer composed of RoBERTa and word vector space to represent the text data. Second, the output generated by the fusion of RoBERTa and word vector space is connected with the input of BiGRU, and the multi-head attention MHA is used to improve the recognition ability of BiGRU, and the corresponding score of each label category is output. The final constraint of the predicted label is performed based on the CRF layer to ensure the effectiveness of the labeled entity, and the labeling is completed according to the labeling strategy of BIOES, and the extraction of the triple knowledge is completed. The results of different entity recognition models are shown in the following table.

[0060]

[0061] In the above steps S41 to S44, in step S41, the input text uses the method of simultaneously calculating the word vector representation of RoBERTa and the arc additive word vector space, and the RoBERTa model optimizes and iterates the parameters based on the advantages and disadvantages of the recognition result according to the attention mechanism; in step S42, the word vector representation output by RoBERTa and the arc additive word vector space respectively is dynamically corrected in the vector fusion process based on the GA optimization algorithm; in step S43, the RoBERTa model outputs the classification effect of the input process knowledge text based on the sentence vector representation of the input text, which is used to distinguish which arc additive case the knowledge belongs to, such as the case name of “battery box” or “cabin body”; in step S44, the BiGRU-MHA-CRF performs the named entity recognition task of the arc additive knowledge, connects the output generated by the fusion of RoBERTa and word vector space with the input of BiGRU, and uses the multi-head attention MHA to improve the recognition ability of BiGRU, and outputs the score of each label. The final constraint of the predicted label is performed based on the CRF layer to ensure the effectiveness of the labeled entity.

[0062] S5, fuse the knowledge extracted in step S4 with the ontology layer constructed in step S1 according to the corresponding process parameter entity to form a knowledge graph of arc additive process intelligent design.

[0063] The fusion of the ontology layer and the knowledge extraction result is characterized in that the obtained triple knowledge entity category and the ontology layer category are mapped into the arc additive word vector space constructed by the application, and the construction of the knowledge graph of the arc additive process intelligent design field is finally completed based on the similarity calculation and knowledge fusion. The fusion alignment of the data layer entity and the ontology layer entity is completed based on the size of the entity similarity, and the knowledge graph construction for the arc additive process intelligent design is finally completed.

[0064] In step S5, according to the text classification result of step S43, the obtained triad knowledge entity category is mapped into the word vector space in step S3 according to the ontology layer category, similarity calculation and knowledge fusion are performed, and finally the construction of the knowledge graph of the electric arc additive process intelligent design field is completed.

[0065] In step S5, according to the corresponding process parameter entity, the knowledge graph of the electric arc additive process intelligent design is formed. In the process of process entity fusion, the similarity calculation is completed in the electric arc additive vector space, and the formula is as follows:

[0066]

[0067] In the formula, i and j respectively represent the process entity feature vectors output by the BiGRU-MHA-CRF model through feature learning according to the BiGRU-MHA-CRF model, and the similarity between the vectors is reflected according to the size of the cosine value.

[0068] Based on the size of the entity similarity, the fusion of the data layer entity and the ontology layer entity is completed, and finally the knowledge graph for the electric arc additive process intelligent design is completed.

[0069] S6, the formed knowledge graph is stored in the graph database, and the knowledge service for the electric arc additive process intelligent design is realized.

[0070] In step S6, the knowledge service for the electric arc additive process intelligent design is realized. For the problems encountered in the electric arc additive process design process, the knowledge question and answer form based on the knowledge graph is adopted, the process designer puts forward questions to the question and answer system, the understanding of the system is completed through semantic analysis, the question and answer system is converted into a query statement of graph data, enters the electric arc additive knowledge base, and completes the answer to the question. Finally, the intelligent design of the electric arc additive process is realized.

[0071] Those skilled in the art will appreciate that the embodiments described herein are intended to aid the reader in understanding the principles of the present application, and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A knowledge graph construction method for intelligent design of an electric arc additive process, characterized in that, The method comprises the following steps: S1, based on the arc additive process design file, the process parameters are determined together with the characteristic of the experiential knowledge summarized by the process designer, the knowledge hierarchy of the ontology layer of the arc additive intelligent design is determined, and the ontology layer is constructed; S2, collect the arc additive process design file and the experiential design knowledge, preprocess the data, and generate the arc additive process knowledge entity recognition corpus set; S3, collect the arc additive field vocabulary, and construct the word vector space of the arc additive; S4, construct the arc additive process knowledge entity extraction model composed of multiple neural network models, and realize the extraction of unstructured arc additive process knowledge; S5, fuse the knowledge extracted in step S4 and the ontology layer constructed in step S1 according to the corresponding process parameter entity, and form the knowledge graph of the arc additive process intelligent design; S6, store the formed knowledge graph in the graph database, and realize the knowledge service of the arc additive process intelligent design; The step S4 comprises the following steps: S41, based on the input unstructured arc additive process design knowledge, input the text word vector representation layer composed of RoBERTa and the word vector space of the arc additive process intelligent design, and complete the representation of the input text vectorization; S42, for the word vector representation output by RoBERTa and the word vector space respectively, fuse the two to generate a new word vector representation, and the weight of fusion is dynamically optimized; S43, for the input text, use the RoBERTa model to complete the feature vector representation of the text and the classification of the process knowledge; S44, use BiGRU-MHA-CRF to perform arc additive knowledge named entity recognition task on the word vector representation obtained in step S42, complete the annotation according to the annotation strategy of BIOES, and complete the extraction of triple knowledge; In step S41, the input text uses the method of simultaneously calculating the word vector representation of RoBERTa and the arc additive word vector space, and the RoBERTa model optimizes and iterates the parameters based on the attention mechanism according to the advantages and disadvantages of the recognition result; In step S42, the word vector representation output by RoBERTa and the arc additive word vector space is dynamically optimized based on the GA optimization algorithm in the vector fusion process; in step S43, the RoBERTa model outputs the classification effect of the input process knowledge text based on the sentence vector representation of the input text, which is used to distinguish which arc additive case the knowledge belongs to, including the case name of "battery box" or "cabin"; in step S44, the BiGRU-MHA-CRF performs arc additive knowledge named entity recognition task, connects the output generated by the fusion of RoBERTa and the word vector space with the input of BiGRU, improves the recognition ability of BiGRU by using the multi-head attention MHA, and outputs the score of each label, and the CRF layer is used to finally constrain the predicted label to ensure the effectiveness of the annotated entity.

2. The knowledge graph construction method for intelligent design of an electric arc additive process according to claim 1, characterized in that: The feature parameters of the arc additive process design knowledge in the step S1 include product model features, equipment parameter features, process parameter features, defect features, path planning experience, and process parameter design experience, and the construction of the ontology framework model of the knowledge graph is completed.

3. The method of claim 2, wherein the method is characterized by: In the step S2, the data preprocessing is the manual cleaning of the acquired process knowledge, including deleting duplicate data, correcting errors, supplementing unclearly described knowledge, and data labeling of the cleaned knowledge based on the feature parameters of the arc additive process design knowledge, to obtain an arc additive process knowledge entity recognition corpus set.

4. The method of claim 3, wherein the method is characterized by: The domain vocabulary in the step S3 includes materials such as 2319 aluminum alloy, 5A06 aluminum alloy, and nickel-titanium metal, path planning such as contour swing and linear scan filling, heat treatment such as solid solution aging, vacuum annealing, and stress relief annealing, and product entities such as battery boxes and oil tank shells. Finally, a domain dictionary set for arc additive process intelligent design is generated, the corpus set is segmented based on the dictionary set, the segmentation results are trained based on the Skip-gran skip word model, and a word vector space specialized for arc additive process intelligent design is obtained.

5. The knowledge graph construction method for intelligent design of an electric arc additive process according to claim 1, characterized in that: In the step S5, according to the text classification result of step S43, the acquired triple knowledge entity category is mapped into the word vector space in step S3, similarity calculation and knowledge fusion are performed, and finally the construction of the arc additive process intelligent design knowledge graph is completed.

6. The method of claim 1, wherein the method is an intelligent design knowledge graph construction method for an electric arc additive process. In the step S5, the corresponding process parameter entities are fused to form the arc additive process intelligent design knowledge graph. In the process of process entity fusion, the similarity calculation is completed in the arc additive vector space. The formula is as follows: Based on the size of the entity similarity, the data layer entity and the ontology layer entity are fused and aligned, and finally the arc additive process intelligent design knowledge graph is completed.

7. The method of claim 1, wherein the method is an intelligent design knowledge graph construction method for an electric arc additive process. In the step S6, the knowledge service for arc additive process intelligent design is realized. For the problems encountered in the arc additive process design process, the knowledge question and answer form based on the knowledge graph is adopted. The process designer asks questions to the question and answer system, the system is understood through semantic analysis, the query statement of the graph data is converted, the arc additive knowledge base is entered, and the corresponding process knowledge problem is solved.

Images