Additive manufacturing industry information collection method and system based on artificial intelligence

Abstract

The invention belongs to the field of data collection, and provides an additive manufacturing industry information collection method and system based on artificial intelligence, and the method comprises the steps: obtaining original data related to the additive manufacturing industry from a multi-source heterogeneous data channel; the original data is preprocessed; performing representation learning on entities in the map to obtain an analogy score relationship among the entities for reasoning a potential technical evolution direction; fusing the generated new terms and the inferred potential technical evolution direction with originally extracted industry knowledge entities and relationships, and constructing a dynamic semantic knowledge graph containing dominant entity relationships and new concept inference edges; periodic updating and reasoning calculation are conducted on the dynamic semantic knowledge graph, and structured industry intelligence information is output and comprises new terms, high-potential technology paths, cross-domain technology fusion trends and concept evolution trajectories. The method can adapt to additive manufacturing industry information collection with rapid information change.

Description

Technical Field

[0001] The present invention belongs to the field of data acquisition, and more particularly relates to a method and system for collecting information on the additive manufacturing industry based on artificial intelligence. Background Art

[0002] With the continuous progress of artificial intelligence technology and natural language processing capabilities, methods for automatically obtaining industry-related information based on intelligent systems have been widely used in fields such as policy research, industrial analysis, and technology forecasting. Existing industry information collection methods mainly use technical means such as keyword retrieval, rule extraction, and co-occurrence statistics, combined with traditional information scraping and tagging methods, to structurally organize text resources such as industry news, patent documents, and academic papers. However, the additive manufacturing industry is developing rapidly, with new technologies and new terms emerging continuously. The existing technologies have obvious deficiencies in the following aspects: Existing methods rely on manually constructed keyword vocabularies or domain entity libraries, and it is difficult to identify newly emerged terms, expressions, or unnamed emerging concepts in the additive manufacturing industry, and cannot adapt to the dynamic characteristics of the continuously evolving term system. For example, with the emergence of new processes such as multi-material hybrid printing and laser energy adaptive control, a large number of professional terms have not been standardized, and traditional methods are difficult to identify or classify.

[0003] Existing methods mostly focus on the extraction and statistics of explicit knowledge, lacking the ability to understand the implicit semantic relationships between terms in the additive manufacturing industry, and cannot simulate the reasoning, induction, and concept evolution judgment in the cognitive process of domain experts, and cannot identify the functional analogy relationships, structural similarities, or technology migration paths between terms.

[0004] Existing systems are difficult to construct an updatable, inferable, and evolvable knowledge structure. Most of the results obtained are static knowledge tables or collections of entries, lacking a time-series update mechanism and semantic graph support, and cannot model and predict the evolution trajectory of concepts, trend intersection points, or cross-domain integration behaviors.

[0005] Therefore, there is an urgent need for an artificial intelligence-based method for collecting information on the additive manufacturing industry that can integrate deep language modeling and analogical reasoning capabilities, has the functions of new term discovery and dynamic trend analysis, and supports knowledge structuring and graph evolution and update, so as to break through the semantic bottleneck and structural limitations of existing technologies and improve the intelligent perception and deduction capabilities of the system in the complex industry background of the rapid development of emerging technologies. Summary of the Invention

[0006] To solve the problems in the existing technology, the present invention provides an artificial intelligence-based method for collecting information on the additive manufacturing industry, including the following steps: Obtain raw data related to the additive manufacturing industry from multi-source heterogeneous data channels according to a preset industry theme; Preprocess the original data, and use a deep language representation model to encode the original data to generate a high-dimensional semantic vector representation; Construct a term space based on the high-dimensional semantic vector, and combine and / or transform and / or expand terms or technical descriptions in the semantic neighborhood through a concept combination function to generate new terms that do not directly appear in the original data but are semantically derivable; Based on the identified industry entities, construct an analogical relationship graph with structural, functional, or evolutionary path similarities, and perform representation learning on the entities in the graph through a graph neural network model to obtain the analogical score relationship between entities for inferring potential technology evolution directions; Fuse the generated new terms and the inferred potential technology evolution directions with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge graph containing explicit entity relationships and new concept inference edges; Periodically update and perform inference calculations on the dynamic semantic knowledge graph, and output structured industry intelligence information, where the industry intelligence information includes new terms, high-potential technology paths, cross-domain technology integration trends, and concept evolution trajectories.

[0007] Further, the steps of generating new terms that do not directly appear in the original data through a concept combination function include: Based on the term space, define multiple types of concept combination functions, and the combination functions include: term combination function, term transformation function, and term expansion function; Among them, the term combination function is used to perform structural-level combination of two or more semantically related or complementary terms to form new terms expressing more complex or refined concepts; the term transformation function is used to perform semantic replacement, expression transformation, or grammatical rewriting on existing terms to form variant terms that are semantically connected or evolved; the term expansion function is used to add context modification components, applicable conditions, or technical constraints to terms to form semantically enhanced terms.

[0008] Further, the steps of constructing an analogical relationship graph with structural, functional, or evolutionary path similarities and performing representation learning on the entities in the graph through a graph neural network model include: Construct an industry entity set and an entity attribute tensor, and construct structural analogy edges, functional analogy edges, and evolutionary analogy edges based on the structural similarity, functional relevance, or evolutionary chain between entities in terms of control logic, module composition, and process constraints; Based on the constructed graph structure, use a graph neural network to perform multiple rounds of updates on the node vectors, and aggregate adjacent node information based on edge type weights in each iteration; Calculate the semantic similarity score between nodes for determining the analogy strength and potential evolution direction between technology entities.

[0009] Further, the steps of constructing a dynamic semantic knowledge graph including explicit entity relationships and new concept inference edges include: Integrate the new terms generated in step S30 into the original industry knowledge graph, construct new term nodes and attach term source type, generation path, first occurrence time, and semantic cluster label attribute information; Establish semantic relationship edges such as term extension edges, combined derivative edges, and function enhancement edges according to the term generation path and semantic similarity; Integrate the potentially technological evolution paths inferred in step S40 into the graph in the form of analogy edges or trend edges. Terms that have not been verified can be set as placeholder nodes and the "to be confirmed" status can be set; Record the snapshot difference of the graph and mark the node update time and update frequency for subsequent graph evolution path tracking and structure maintenance.

[0010] Further, the steps of outputting structured industry intelligence information include: Set up a graph update scheduling mechanism to perform incremental updates on the dynamic semantic knowledge graph regularly, including node attribute updates, edge weight adjustments, node active status evaluations, and version snapshot generations; Perform path reasoning, graph representation learning, hot spot clustering analysis, and multi-hop logical chain tracking on the updated graph to obtain term semantic change trends, fusion nodes, and technological paths; Organize the analysis results into structured intelligence for output. The output content includes: new term intelligence reports, high-potential technological path lists, cross-domain fusion trend graphs, and concept evolution trajectory graphs, and supports export in the form of structured data files or a combination of text and graphics.

[0011] On the other hand, the present invention also provides an additive manufacturing industry information collection system based on artificial intelligence, including the following modules: A data collection module for obtaining raw data related to the additive manufacturing industry from multi-source heterogeneous data channels according to a preset industry theme; A semantic encoding module for preprocessing the raw data and encoding the raw data using a deep language representation model to generate a high-dimensional semantic vector representation; A term generation module for constructing a term space based on the high-dimensional semantic vectors, and generating new terms that do not directly appear in the raw data but are semantically derivable by combining and / or transforming and / or expanding terms or technical descriptions in the semantic neighborhood through a concept combination function; The analogical reasoning module is used to construct an analogical relationship graph with similarity in structure, function, or evolutionary path based on the identified industry entities, and perform representation learning on the entities in the graph through a graph neural network model to obtain the analogical score relationship between the entities for inferring potential technological evolution directions; The graph construction module is used to fuse the generated new terms and the inferred potential technological evolution directions with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge graph containing explicit entity relationships and new concept inference edges; The intelligence output module is used to perform periodic updates and inference calculations on the dynamic semantic knowledge graph and output structured industry intelligence information, where the industry intelligence information includes new terms, high-potential technology paths, cross-domain technology integration trends, and concept evolution trajectories.

[0012] Furthermore, the term generation module includes: The concept combination function library is used to define term combination functions, term transformation functions, and term extension functions; The combination control unit is used to select semantic neighborhood term pairs in the term space, call the concept combination function to generate term semantic vectors, and input the generated term semantic vectors into the term decoder to generate a new term text that can be read in language; The term scoring unit is used to perform semantic consistency scoring and knowledge matching verification on the generated new terms, and mark the terms with a scoring result higher than the preset threshold as semantic emergence terms.

[0013] Furthermore, the analogical reasoning module includes: The entity attribute construction unit is used to construct an industry entity set and its attribute tensors; The analogical relationship construction unit is used to construct structural analogical edges, functional analogical edges, and evolutionary analogical edges based on features such as control logic, module composition, and process constraints; The graph neural network learning unit is used to update the vectors of entity nodes through multiple rounds of aggregation of adjacent node information in the graph structure; The analogical scoring unit is used to calculate the semantic similarity based on node vectors and output the analogical scores between entities for subsequent evolution path inference.

[0014] Furthermore, the graph construction module also includes: The node access unit is used to access the new terms in the form of nodes into the original industry graph, attaching term sources, generation paths, first appearance times, and semantic cluster labels; The relationship generation unit is used to establish term extension edges, combination derivative edges, and functional enhancement edges in the graph based on the term generation path and semantic similarity; The reasoning edge access unit is used to add potential technology evolution paths to the knowledge graph in the form of analogy edges or trend edges, set the unvalidated term nodes as placeholder nodes and set the "to be confirmed" status; The evolution management unit is used to record the snapshot differences of the knowledge graph and maintain the node update time, active status and evolution tracking information.

[0015] Furthermore, the intelligence output module includes: The update scheduling unit is used to periodically trigger the incremental update operation of the knowledge graph, and update the node attributes, edge weights and knowledge graph version snapshots; The reasoning and analysis unit is used to perform path reasoning, graph representation learning, hot term clustering and multi-hop logical chain analysis on the updated knowledge graph; The intelligence organization unit is used to extract new terms, high-potential technology paths, cross-domain trend nodes and evolution trajectory information, and organize them into a structured intelligence report; The output interface unit is used to export the industry intelligence information in the form of structured data format or graphic report.

[0016] Compared with the existing industry information collection technologies, the present invention significantly improves the system's capabilities in term recognition, trend insight and knowledge structure organization by introducing a number of artificial intelligence technologies such as term space construction, concept combination function, semantic emergence reasoning mechanism, analogy relationship graph and graph neural network evolution modeling, and has the following beneficial effects: The present invention no longer relies on a fixed term vocabulary or manual rules, but constructs a term semantic vector space, and uses combination and / or transformation and / or extension mechanisms to automatically generate new terms that are not explicitly present in the original data but are semantically reasonable, breaking through the limitation of traditional methods that can only extract existing terms, and significantly enhancing the ability to recognize emerging concepts.

[0017] By constructing a multi-dimensional relationship graph of structural analogy, functional analogy and evolutionary analogy, and using graph neural network to deeply model the semantic commonalities between entities, potential migration paths and fusion directions between different technologies can be identified, with the ability to simulate "heterogeneous analogy" in expert cognition, and improving the system's cross-domain association and trend judgment level.

[0018] The present invention can integrate explicit terms, newly generated terms and analogy paths into a unified dynamic semantic knowledge graph, support periodic incremental updates of nodes and edges, semantic activity evaluation and time evolution trajectory tracking, and provide structured support for trend evolution analysis.

[0019] By performing periodic reasoning calculations on the knowledge graph, the system can automatically output structured industry intelligence information, including a list of new terms, high-potential technology paths, cross-domain integration trends, and concept evolution chains, providing an intelligent decision-making basis for policy-making, enterprise R & D, technology investment, etc.

[0020] In summary, the present invention breaks through many limitations of traditional industry information collection technologies in terms of term recognition scope, semantic association depth, trend evolution modeling ability, and structural organization flexibility, and can be widely applied in industries with rapid term evolution and highly cross-cutting technologies such as additive manufacturing, electronics manufacturing, synthetic biology, and intelligent transportation, having significant technical value and practical prospects. Brief Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 is a flowchart of the method of the present invention; Figure 2 is a system block diagram of the present invention. Detailed Embodiments

[0023] The following will make a preferred description of the invention in combination with the drawings and specific embodiments.

[0024] This embodiment solves the above problems through the following steps: In one embodiment, referring to Figure 1 , the present invention provides a method for collecting information on the additive manufacturing industry based on artificial intelligence, which automatically mines new concepts, technical contexts, and industrial trends related to the target field by simulating the interpretation and deduction behaviors of domain experts on additive manufacturing industry data, and is applicable to multiple scenarios such as policy research, market prediction, and technology reconnaissance.

[0025] Specifically, the method includes the following steps: Step S10, obtaining raw data related to the additive manufacturing industry from multi-source heterogeneous data channels according to a preset industry theme.

[0026] Specifically, the "industry theme" can be input by the user, preset by the task system, or automatically recommended according to industry hotspots. The theme is used to limit the content scope and semantic boundary of information collection, such as "energy control technology for metal additive manufacturing equipment", "application of powder bed fusion (PBF) process in aviation parts", "market trend of new printing material Ti6Al4V", etc.

[0027] The multi-source heterogeneous data channels refer to an information collection containing multiple data formats, structures, and sources, including but not limited to: Structured data sources: such as industry standard databases, technology regulatory websites, manufacturing cost structures in corporate financial reports, equipment sales quantities, etc.

[0028] Semi-structured data sources: such as patent documents, academic papers, technical white papers, etc., which contain structured field information and free text.

[0029] Unstructured data sources: such as news reports, corporate website information, industry interviews, social media technology topics, forum comments, etc.

[0030] Multi-modal data sources: such as physical pictures of printing devices, videos of additive manufacturing processes uploaded by users, spectral images of powder materials, conference recordings, etc.

[0031] Taking the additive manufacturing industry as an example, after the system receives the industry theme of "development trends of metal additive manufacturing in the aerospace field", the following data collection process is automatically triggered: Retrieve the full text and metadata of papers containing keywords such as "laser powder bed fusion", "additive manufacturing in the aerospace field", "L-PBF Ti6Al4V", etc. from academic websites.

[0032] Obtain the public patent texts and claim contents containing keywords such as "aerospace parts", "additive manufacturing", "printing path optimization", etc. from public patent databases.

[0033] Collect industry dynamic news about new equipment, new processes, and new materials in the past six months from 3D printing industry portals.

[0034] Extract the technical specifications, video demonstration links, and user feedback documents of new model printers released by the corporate website.

[0035] Download the conference schedule, abstract collection, and expert speech PPT documents from industry conference websites.

[0036] Retrieve explanatory videos containing "additive manufacturing of aerospace engine blades" from platforms such as video platforms, and automatically identify the voice content therein and transcribe it into text.

[0037] After the data is collected, meta-properties such as its source, time, language, modality, etc. will be marked, and it will be uniformly transmitted to the subsequent preprocessing and semantic modeling steps.

[0038] Step S20, preprocess the original data, and use a deep language representation model to encode the original data to generate a high-dimensional semantic vector representation.

[0039] The original data may come from various sources, such as web texts, social media comments, news reports, etc., which may contain a large amount of noise information and unregulated expressions. The purpose of preprocessing is to improve the data quality and make it more suitable for subsequent model processing. Preprocessing mainly includes the following sub-steps: Data cleaning: The original data may contain noise information such as HTML tags, special symbols, redundant spaces and line breaks, etc. These information will interfere with the subsequent processing process, so they need to be removed. For example, if the original data is news content crawled from a web page, there may be such as 、 HTML tags and special characters such as spaces need to be filtered out using regular expressions.

[0040] Remove stop words: Stop words refer to words that appear frequently in the text but contribute little to the semantic expression of the text, such as "的", "是", "在", "和", etc. Removing stop words can reduce the dimension of the data and improve processing efficiency. You can use a pre-defined stop word list to filter out stop words in the text.

[0041] Word segmentation: For Chinese text, continuous text needs to be segmented into individual words. Word segmentation is a basic step in Chinese natural language processing. Commonly used word segmentation tools include Jieba word segmentation and HanLP. For example, the sentence "I love natural language processing" is segmented into "I", "love", and "natural language processing".

[0042] Part-of-speech tagging and stemming (optional): Part-of-speech tagging is to mark each word with its part of speech, such as noun, verb, adjective, etc. Stemming is to restore words to their stem form, such as restoring "learning" and "learner" to "learning". These two steps can further improve the quality of data and the consistency of features, but they are not required for all tasks.

[0043] After completing data preprocessing, you need to use a deep language representation model to convert text data into a high-dimensional semantic vector representation. The deep language representation model can learn the semantic information of the text and map it to a high-dimensional vector space, so that texts with similar semantics are close in the vector space. Here we take the BERT model as an example: Loading pre-trained models: BERT is a pre-trained language model based on the Transformer architecture. It performs unsupervised learning on large-scale text data and learns rich language knowledge. You can use Hugging Face's Transformers library to load the pre-trained BERT model and the corresponding tokenizer.

[0044] Input processing: The preprocessed text data is input into the tokenizer to convert it into an input format acceptable to the model. It is usually necessary to add special tags, such as [CLS] and [SEP], to indicate the beginning and end of a sentence, respectively. At the same time, the words are converted into corresponding word indices, and padding and truncation operations are performed to make the length of all input sequences consistent.

[0045] Model inference: The processed input sequence is input into the BERT model for inference, and the model outputs the contextual representation of each word. The output vector corresponding to the [CLS] tag is usually taken as the semantic representation of the entire sentence.

[0046] After being processed by the BERT model, the output vector corresponding to the [CLS] token is the high-dimensional semantic vector representation of the sentence. This vector usually has a relatively high dimension (such as 768 dimensions or 1024 dimensions) and contains the semantic information of the sentence. This vector can be used for subsequent tasks such as text classification, sentiment analysis, information retrieval, etc.

[0047] Step S30: Based on the high-dimensional semantic vector, construct a term space, and through a concept combination function, combine and / or transform and / or expand terms or technical descriptions in the semantic neighborhood to generate new terms that do not directly appear in the original data but are semantically derivable.

[0048] This step is to further implement a term generation mechanism based on semantic space logical reasoning on the basis of completing the acquisition and semantic modeling of original industry data. Through this step, the system can not only identify existing terms, but also predictively generate new terms that are semantically related but not explicitly expressed, expand the industry semantic boundary, and enhance the trend prediction ability.

[0049] Industry terms often have the characteristic of periodic update. With the continuous emergence of new technologies, new materials, and new processes, their term systems are also evolving rapidly. For example, in the field of additive manufacturing in recent years, a series of sub-concepts such as "metal additive manufacturing", "multi-material composite printing", and "selective laser sintering" have emerged from "3D printing".

[0050] Traditional methods can only extract terms based on existing word lists or entity recognition models and are difficult to handle the following scenarios: A new process that has not been named begins to be described in papers but has not been uniformly expressed; The combination of terms from multiple different fields forms a new technology path; Some expression forms are in a fuzzy state but have a clear semantic reference.

[0051] This step precisely proposes an overall mechanism of term space construction + semantic combination function + new term generation + multi-dimensional verification to solve the above problems.

[0052] Specifically, step S30 includes the following sub-steps: Step S301: Semantic neighborhood modeling and mining of term aggregation relationships Based on the set of term semantic vectors obtained in the previous step , use a vector clustering method based on semantic distance (such as K-Means, hierarchical clustering, or spectral clustering) to construct a neighbor network between terms and extract a set of semantic neighborhoods , and each term is mapped to the cluster structure of its semantic context.

[0053] Adjust the association strength between terms according to the term frequency, context relevance, and neighborhood semantic density, and finally construct a term relationship network with the following structure: Strong semantic adjacency edges: such as "laser sintering" and "laser focusing control"; Potential replacement edges: such as "superalloy" and "high-entropy alloy"; Heterogeneous analogy edges: such as "metal powder bed melting" and "ceramic wire synchronous deposition".

[0054] Term space graph The edge set 𝐸 in the records different types of combination possibilities.

[0055] Step S302: Combination function design and term generation vector construction Based on the term space, the system defines multiple types of concept combination functions for different types of semantic reasoning tasks. The combination functions are divided into the following according to different combination methods: 1. Term combination Structurally combine two or more semantically related or complementary terms to form a new term that expresses a more complex or refined concept. The combination relationship can be language structures such as coordination, modification, subject-predicate, noun-adjective, etc.

[0056] Design template structures, such as "[material] + [process]", "[functional word] + [equipment name]", and embed the term text for combination.

[0057] Exemplarily: Term pairs: "Multi-laser scanning" "Titanium alloy powder" After combination, generate: "Multi-laser scanning printing of titanium alloy powder" "Multi-laser titanium-based powder deposition process" 2. Term transformation Perform semantic replacement, expression transformation, and grammar rewriting on existing terms to form variant terms with different expressions but semantically equivalent or evolved, which is applicable to industry backgrounds with inconsistent term standards and coexistence of multiple description methods.

[0058] The implementation methods include: Semantic replacement: Replace a certain component in the term, such as replacing "laser" with "electron beam".

[0059] Structure inversion: Change "control energy density" to "energy density control".

[0060] Stylistic style transformation: For example, change "selective sintering technology" to "selective sintering method".

[0061] Vector difference operation: Simulate "laser printing": "metal" = X: "ceramic" ⇒ X = "ceramic printing" Exemplarily: Original term: "High-entropy alloy printing path optimization" Transformed term: "High-entropy printing path adjustment" "Complex alloy printing parameter control" "Multi-component alloy path adaptive adjustment" 3. Term expansion Add context modifiers, specific context limitations, applicable object descriptions, or technical condition supplements to the original term to expand new term expressions and strengthen its semantic dimension and description clarity.

[0062] Implementation methods include: Superordinate and subordinate limitation expansion: Add specific limiting components before / after the original term, such as "space-grade", "microstructure", "sub-millimeter scale".

[0063] Process environment / constraint condition expansion: For example, add "under variable air pressure environment" to "fused deposition".

[0064] Semantic reference expansion: Expand "laser sintering" to "laser sintering technology applicable to multi-material hybrid forming".

[0065] Nested context vector fusion: Weight and fuse the term vector and the context paragraph vector to generate a semantic enhanced term.

[0066] Exemplarily: Original term: "Powder bed melting" Expansion term candidates: "Powder bed melting process under low air pressure environment" "Powder bed melting technology applicable to refractory metals" "Special powder bed forming method for aerospace engine components" The above three types of operations can not only be used independently, but the system also allows them to be used in combination, for example: First combine the terms and then transform them; Add extended descriptions to the combined terms; Perform restrictive expansion on the basis of transformation.

[0067] Exemplarily: Term 1: "Laser energy modulation" Term 2: "Ceramic matrix composite" Combination: → "Ceramic composite laser modulation" Transformation: → "Composite ceramic laser multi-zone power adjustment" Expansion: → "Multi-zone Power Laser Composite Ceramic Printing Strategy for Heterogeneous Interface Control" Step S303: Term Candidate Generation and Multi-candidate Hierarchical Decoding For each newly constructed vector , the following strategy is adopted to generate term candidates: Top-k Beam Search: Generate multiple term candidates with different syntactic structures but close semantics; Semantic Segmentation Decoding: First generate the basic stem, and then generate the modification structure; Language Structure Evaluator Scoring: Eliminate verbose or illogical expressions.

[0068] Exemplarily: Combined Input: Term A: "High-energy density scanning" Term B: "Composite ceramic material" Generate term candidates: "High-energy density composite ceramic melting" "Laser scanning of composite ceramic materials" "Ceramic composite high-energy sintering technology" Each candidate term retains its semantic generation path: starting term pair, combination method, vector distance, language model probability.

[0069] Step S304: Term Confidence Learning and Semantic Consensus Verification To determine whether the term candidates are acceptable in the industry and semantically valid, the system introduces a confidence scoring model , and the scoring metrics include: Semantic Consistency Score : The average semantic proximity to the input terms; Language Naturalness Score : The language model's judgment on the smoothness of the generated expression; Industry Consensus Score : The matching frequency of the candidate terms in the known term graph or external database; Substitutability Score : Whether it can naturally replace some existing expressions.

[0070] Total Score: , where are configurable weight coefficients. Terms with scores higher than the threshold (e.g., 0.75) are marked as "semantically emerging terms" and enter the term library.

[0071] Step S305: Construction of Term Tracing Path and Preparation for Updating the Evolution Map The system constructs a "term generation path" for all newly verified terms, including: Source term list; Combination function number used; Semantic neighborhood to which it belongs; Decoding model version; Timestamp; Whether the external corpus matches.

[0072] All candidates marked as semantically emerging terms are constructed into the nodes of the term evolution graph to support trend modeling and knowledge fusion in the next step.

[0073] Step S40: Based on the identified industry entities, construct an analogical relationship graph with similarities in structure, function, or evolution path, and use a graph neural network model to perform representation learning on the entities in the graph to obtain the analogical score relationship between entities for inferring potential technology evolution directions.

[0074] The purpose of this step is to construct an analogical relationship graph based on entity content such as industry terms, technology names, material types, and equipment parameters extracted in the previous steps, and use a graph neural network model to learn the representations of the entities in the graph, thereby obtaining the analogical score relationship between entities for inferring technology evolution paths, cross-domain migration paths, or process fusion trends that may occur semantically but have not been clearly proposed.

[0075] Different from traditional co-occurrence graphs, the focus of this step is to mine the structural analogy, functional similarity, and evolution relevance between entities, emphasizing the ability of analogical reasoning rather than explicit co-occurrence relationships.

[0076] Specific implementation steps Step S401: Construct an industry entity set and entity attribute tensors The system first structures all industry terms, material names, equipment names, process methods, control strategies, etc. extracted in the previous steps to form an entity set: , Each entity is equipped with an attribute tensor , including but not limited to: Technical attributes (such as whether it is laser-related, whether it involves melting, whether it is powder / wire input); Functional attributes (such as whether it controls energy, whether it supports multi-channel output); Application scenarios (such as whether it is used in aviation, medical, microstructural control); Temporal attributes (time of first appearance, active period, trend curve, etc.); Module structure (such as print head structure, composition of control modules, etc.).

[0077] These attributes will be used to judge and score the structure - function analogy edges in the subsequent graph construction process.

[0078] Step S402: Establishing analogy relation edges and constructing graph structure Taking the entity set E as nodes, the system constructs three types of analogy edges based on the structural similarity, functional relevance, and historical evolution path between entity attribute tensors: Structure analogy edge For example: Although the entities "laser cladding" and "electron beam deposition" use different energy sources, their control logics and module architectures are similar.

[0079] Function analogy edge For example: "Laser focusing control" and "current density control" can form an equivalent analogy in the energy field regulation logic.

[0080] Evolution analogy edge For example: "Titanium alloy powder bed printing" evolves into "Titanium aluminum alloy multi - laser printing" → mapping out that "High - entropy alloy powder bed melting" may become a trend evolution node.

[0081] The graph is denoted as: , where it contains the above three types of edge relations and their edge attributes (similarity score, edge type, edge direction, etc.).

[0082] Step S403: Analogy graph embedding and graph neural network modeling The system encodes the analogy graph using a graph neural network. The representation vector of each entity node is updated through the following iterative formula: , where: represents the set of adjacent entities of entity ; represents the relation weight corresponding to the edge type 𝑟; represents the activation function; represents the normalization factor; represents the initial vector projected from the entity attribute tensor.

[0083] After the graph embedding learning is completed, the system obtains the context representation of each entity in the analogy graph, which can be used to calculate the analogy score.

[0084] Step S404: Analogy score calculation and evolution trend reasoning For any two entities , Its analogy score function is defined as: , The commonly used similarity calculation method is cosine similarity: , The system sorts all potential relationships according to the analogy scores and selects entity pairs with scores higher than a set threshold (such as 0.85) as candidate paths with potential technological evolution.

[0085] For example: Structural Analogy Reasoning Entity 1: Laser cladding printing system (including optical module, scanning path control, material injection system) Entity 2: Plasma spray deposition (including high temperature energy source, path control, powder feeding) The system extracts the structural modules of the two and matches them, determining them as structural analogy edges and assigning them an analogy score of 0.82.

[0086] → Reasoning conclusion: If the plasma dual energy field control technology is introduced into laser cladding, it is possible to form a new "hybrid energy field deposition path control technology".

[0087] Functional Analogy Reasoning Entity 1: Multi-area scan path adjustment Entity 2: Multi-point pressure sensing feedback mechanism Although the two technologies are different, one is printing path control and the other is closed-loop control, they share the same functional semantics of "regional-level adaptive regulation".

[0088] → Score 0.89, system prompt: The concept of "multi-area sensing-path integrated linkage printing system" may appear.

[0089] Evolutionary Analogy The entity evolution path is known: "Single laser titanium alloy printing" → "Multi-laser titanium alloy printing" → "Synchronous multi-laser printing of high entropy alloys" → The system analogy is: "Multi-material micro-powder composite printing" has a multi-component structure in terms of materials, or it can draw on the evolution path of "high entropy alloy".

[0090] The system generates a future candidate term: "high-component particle multi-laser partitioning forming".

[0091] In this step, by constructing an analogy relationship map that includes the similarity of structure, function, and evolution path, and introducing a graph neural network for entity representation learning and analogy score calculation, the deep semantic association between industry entities is modeled, and the evolution trend is inferred. It can effectively identify potential technology transfer paths and integration directions, break through the limitations of traditional co-occurrence statistics or rule matching methods, significantly improve the forward-looking judgment ability and cross-domain association ability of the information collection system in complex technical fields, and provide more intelligent and semantic support for new term discovery, technology prediction, and strategic judgment.

[0092] Step S50: Integrate the generated new terms and the inferred potential technology evolution directions with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge graph that includes explicit entity relationships and new concept inference edges.

[0093] This step aims to integrate the semantic emergence terms generated in step S30 and the potential technology evolution directions inferred in step S40 with the industry terms, technology entities, material entities, relationship peer information, etc. explicitly extracted from multi-source data in the previous steps, and construct a dynamic semantic knowledge graph with the ability of incremental update, clear structural semantics, and support for evolutionary reasoning, which is used to support subsequent industry analysis, trend monitoring, and strategy decision-making.

[0094] The specific implementation process includes: Step S501: Preparation of the initial structure of the graph The system first loads or constructs the original industry semantic knowledge graph , where: represents the set of original entities, including technical terms, material types, equipment structures, control strategies, application fields, etc.; represents the set of original relationships, including common domain semantic relationships such as "applicable to", "contains", "based on", "belongs to", "improved from", "replaced by", etc.

[0095] The original graph usually comes from the previous information extraction module (step S20) or existing standard knowledge bases (such as CMeKG, patent co-occurrence graphs, etc.).

[0096] Step S502: Access of new term nodes and edge construction For the new terms generated and verified in step S30 , the system accesses them to the graph one by one and constructs the following information: Create a new node for each term , and attach the following attributes: The type of term source (such as "combination", "transformation", "expansion"); The original term combination path; The timestamp of the first generation; The semantic cluster label it belongs to; The external verification result (whether it has appeared in the literature or database); The system automatically matches the semantic proximity relationship between the new term and the existing nodes, constructs relationship edges such as "evolved from", "belongs to", "function enhanced", "term extended from", etc., and writes them into .

[0097] Exemplarily: New term: "Multi-region energy density adaptive melting strategy" The system determines that it is derived from the combination of "energy density control" and "multi-region scanning" → establish the following edges: "Multi-region energy density adaptive melting strategy" → "energy density control": Relationship = "Combination derivative" "Multi-region energy density adaptive melting strategy" → "multi-region scanning": Relationship = "Semantic fusion" Step S503: Evolution path edge fusion For the potential evolution path inferred in step S40 , the system processes it according to the following rules: If the target node is a new term, directly connect the "evolution trend edge"; If does not exist yet, the system can temporarily set a "potential term placeholder node" with a "to be confirmed" flag for future verification; If the confidence level of the inference relationship is relatively high (such as the score > 0.9), it is marked as "strong evolution connection"; if it is between 0.7 and 0.9, it is "possible trend connection".

[0098] The system embeds the evolution inference result and the term generation path into the knowledge graph through this mechanism, forming a complete semantic link from concept generation → analogical reasoning → trend evolution → structural organization.

[0099] Step S504: Semantic graph structure update and evolution management After the fusion is completed, the system constructs the latest state of the graph: , At the same time, record the graph snapshot difference , which is used to support subsequent evolution path backtracking and trend evaluation. Each node or edge is added with a time attribute, an update frequency mark, and a verification status (such as "original term", "graph model prediction", "externally confirmed").

[0100] To control the complexity of the graph structure, the system can set the following mechanism: Edge relation sparsification threshold: Delete weakly related edges with low correlation; Node activity assessment: If a term has not appeared in subsequent data for a long time, set the "cold node" flag; Term merging mechanism: For new and old terms with highly similar semantics, the system avoids graph redundancy by merging nodes or setting an "alias" relationship.

[0101] Exemplarily: New terms: "Ceramic composite microwave synchronous sintering" "High-frequency multimode energy adaptive printing" Inference edge: "Laser sintering" → "Multimode energy deposition": Evolution edge (score 0.91) "Electron beam printing" → "Microwave printing": Structural analogy edge (score 0.88) Create a new node: "Ceramic composite microwave synchronous sintering" Attributes: "Combination + extended generation"; Semantic source terms: "Ceramic composite powder" + "Microwave energy synchronous control" Edge update: "Ceramic composite microwave synchronous sintering" → "Microwave energy synchronous control": Edge type = "Term extended from" "Electron beam printing" → "Microwave printing": Edge type = "Analogy migration path" "Microwave printing" → "Ceramic composite microwave synchronous sintering": Edge type = "Potential evolution edge"; Score = 0.84; Status = "To be confirmed" In step S50, the system realizes the expansion of the industry knowledge graph system from explicit term knowledge to emerging concepts, from co-occurrence graph to analogy graph, and from static structure to dynamic evolution, and has the following technical advantages: New terms and trend edges can be naturally embedded in the original knowledge graph; It is convenient to display the path of "term birth - development - integration"; It can accommodate different source information from data, language models, and reasoning models; It provides structural support for trend prediction, hot spot discovery, and technology evolution modeling.

[0102] In step S60, perform periodic update and inference calculation on the dynamic semantic knowledge graph, and output structured industry intelligence information, where the industry intelligence information includes new terms, high-potential technology paths, cross-domain technology integration trends, and concept evolution trajectories.

[0103] The purpose of this step is to, based on the dynamic semantic knowledge graph constructed in step S50, extract and output structured industry intelligence information with strategic value, foresight, and interpretability by setting a periodic update mechanism and an inference calculation model, thereby realizing the closed-loop logic chain of "information → knowledge → trend → decision".

[0104] The industry intelligence information not only includes the statistical results of explicit data, but also includes the potential knowledge deduced by the artificial intelligence model based on the graph structure, such as the dissemination path of new terms in different fields, the possible direction of technology evolution, the phased law of the evolution of industry concepts, etc.

[0105] The specific implementation steps include: Step S601: Graph update scheduling mechanism The system sets a regular trigger mechanism, such as once a day, once a week or once a month, to update and iterate the knowledge graph. The update content includes: Integrate new term nodes and relationship edges; Delete, downgrade or freeze low-active nodes and edges (such as no citation for 6 months); Update node attributes (such as the first appearance time, current activity, appearance frequency, etc.); Generate a snapshot of the graph version to form a time-series graph structure .

[0106] The graph update is based on the graph incremental calculation strategy, and gives priority to processing new / modified content to reduce the computational complexity.

[0107] Step S602: Inference calculation module For the knowledge graph that has been updated and completed , a variety of graph inference algorithms are used for deductive analysis, including: Path inference Identify high-confidence paths between terms - technologies - materials - products, such as: "High-entropy alloy" → "Multi-laser sintering" → "Aero-engine nozzle" → Output possible technology implementation routes; Graph representation learning Use graph neural networks (such as GAT, RGCN) to learn the vector representation of entities in the graph, and then calculate: The semantic change trend between terms; The evolution trajectory of entity vectors over time; Hot spot clustering and semantic diffusion analysis Judge the diffusion degree of new terms in the graph, whether they quickly form multi-edge connections and participate in multiple trend paths, as an indicator to judge "concept popularity".

[0108] Multi-hop logical chain tracking Identify the multi-hop logical chain of concept A → technology B → material C → application D for complex trend modeling.

[0109] Step S603: Industry intelligence information output module After the reasoning and calculation are completed, the system automatically organizes the results into a structured intelligence report. The output information includes but is not limited to: New term intelligence report New term name Source path (combined terms / analogical reasoning) Current graph activity index Whether it has been verified in an external database List of high-potential technology paths Path starting point (usually basic technology or material) Intermediate control strategy node Potential output structure or application scenario Path confidence score and recommendation level Cross-domain technology fusion trend chart In the past two rounds of graphs, the significant growth path of the connection relationship between terms in different fields Form cross-border fusion nodes (such as "microwave-assisted metal sintering" connecting "electronic communication modulation technology" and "metallurgical laser forming") Concept evolution trajectory chart Describe the node expansion trajectory, relationship density change, and evolution path of a typical concept (such as "energy density control") in multiple time graphs Visualize as a time series flow chart or a vein tree chart structure The output format can be PDF data, or generate an "Intelligent Industry Intelligence Weekly / Monthly Report" in the form of charts + text summary.

[0110] Exemplarily, as shown in Table 1: Table 1 New term intelligence output New Term Source Path Activity Whether to Verify Multi-channel Laser Scanning Path Control "Multi-laser System" + "Path Adjustment Strategy" High No Microwave-assisted Powder Bed Sintering "Powder Bed Melting" + Analogy to "Microwave Heating" Medium Yes (Appeared in arXiv) High-potential technology path Starting point: "High-entropy alloy powder" Path: "High-entropy alloy powder" → "Zone temperature control sintering" → "Interlayer microstructure regulation" → "Directional printing of high-temperature fatigue parts" Path confidence: 0.89 Application potential: Aeroengine support frame Cross-domain fusion trend "Laser galvanometer synchronous scanning" ↔ "Optical communication beam splitting regulation" → The system identifies it as an analogical fusion point of "multi-channel optical modulation scheme", which may give birth to a new field of "tunable frequency laser sintering".

[0111] Through this step, the system transforms the static term graph into a dynamic industry knowledge system that can be updated, deduced, and output. Its core technical advantages include: The periodic update mechanism ensures the real-time nature of the industry map and the ability to perceive trends; the structured intelligence output makes the system not only a data collector but also a reasoning and analysis engine with an "expert-like thinking"; it supports various types of analysis requirements, such as new term warning, technology path discovery, cross-domain integration identification, and concept cycle management; significantly improves the semantic understanding and strategic judgment capabilities of the artificial intelligence system for complex technology industries See Figure 2 , in another embodiment, the present invention further provides an information collection system for the additive manufacturing industry based on artificial intelligence, including: A data acquisition module for obtaining raw data related to the additive manufacturing industry from multi-source heterogeneous data channels according to a preset industry theme; A semantic encoding module for preprocessing the raw data and encoding the raw data using a deep language representation model to generate a high-dimensional semantic vector representation; A term generation module for constructing a term space based on the high-dimensional semantic vectors, and generating new terms that do not directly appear in the raw data but are semantically derivable by combining and / or transforming and / or expanding terms or technical descriptions in the semantic neighborhood through a concept combination function; An analogical reasoning module for constructing an analogical relationship map with structural, functional, or evolutionary path similarities based on the identified industry entities, and performing representation learning on the entities in the map through a graph neural network model to obtain the analogical score relationship between entities for inferring potential technology evolution directions; A map construction module for fusing the generated new terms and the inferred potential technology evolution directions with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge map containing explicit entity relationships and new concept inference edges; An intelligence output module for periodically updating and performing reasoning calculations on the dynamic semantic knowledge map, and outputting structured industry intelligence information, where the industry intelligence information includes new terms, high-potential technology paths, cross-domain technology integration trends, and concept evolution trajectories.

[0112] In a further implementation manner, the term generation module includes: A concept combination function library for defining term combination functions, term transformation functions, and term expansion functions; A combination control unit for selecting term pairs in the semantic neighborhood in the term space, calling the concept combination function to generate term semantic vectors, and inputting the generated term semantic vectors into a term decoder to generate a new term text that is readable in language; A term scoring unit for performing semantic consistency scoring and knowledge matching verification on the generated new terms, and marking the terms with a scoring result higher than a preset threshold as semantically emergent terms.

[0113] In a further implementation, the analogical reasoning module includes: An entity attribute construction unit for constructing an industry entity set and its attribute tensors; An analogical relationship construction unit for constructing structural analogical edges, functional analogical edges, and evolutionary analogical edges based on characteristics such as control logic, module composition, and process constraints; A graph neural network learning unit for updating entity node vectors through multi-round aggregation of adjacent node information in the graph structure; An analogical scoring unit for calculating semantic similarity based on node vectors and outputting the analogical scores between entities for subsequent evolutionary path reasoning.

[0114] In a further implementation, the graph construction module further includes: A node access unit for accessing new terms as nodes into the original industry graph, attaching term sources, generation paths, first appearance times, and semantic cluster labels; A relationship generation unit for establishing term extension edges, combination derivative edges, and functional enhancement edges in the graph based on term generation paths and semantic similarity; An inference edge access unit for adding potential technology evolutionary paths to the graph in the form of analogical edges or trend edges, setting unvalidated term nodes as placeholder nodes and setting the "to be confirmed" status; An evolution management unit for recording graph snapshot differences and maintaining node update times, active statuses, and evolutionary tracking information.

[0115] In a further implementation, the intelligence output module includes: An update scheduling unit for periodically triggering graph incremental update operations to update node attributes, edge weights, and graph version snapshots; An inference analysis unit for performing path reasoning, graph representation learning, hot term clustering, and multi-hop logic chain analysis on the updated knowledge graph; An intelligence organization unit for extracting new terms, high-potential technology paths, cross-domain trend nodes, and evolutionary trajectory information and organizing them into a structured intelligence report; An output interface unit for exporting the industry intelligence information in a structured data format or a graphic report form.

[0116] It should be noted that the explanations of the foregoing embodiments of the method for collecting information on the additive manufacturing industry based on artificial intelligence also apply to the device of the embodiments of the present application and will not be elaborated here.

[0117] Those of ordinary skill in the art can realize that the units and algorithm steps described in the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0118] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0119] In several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The foregoing storage media include: USB flash drives, mobile hard disks, read-only memories (hereinafter referred to as ROMs), random access memories (hereinafter referred to as RAMs), magnetic disks, or optical discs, and other media that can store program codes.

[0120] The above is only the specific implementation manner of this application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application and should be covered by the protection scope of this application. For the part of the module structure that is not specifically defined in this invention, the content recorded in the prior art shall prevail. The prior art mentioned in the foregoing background art part and specific embodiment part of this invention can be used as a part of this invention to understand the meaning of some technical features or parameters.

Claims

1. An information collection method for additive manufacturing industry based on artificial intelligence, characterized in that: The method comprises the following steps: According to the preset industry themes, obtain the original data related to the additive manufacturing industry from multi-source heterogeneous data channels; Preprocessing the original data, encoding the original data using a deep language representation model, and generating a high-dimensional semantic vector representation; Constructing a term space based on the high-dimensional semantic vector, combining and / or transforming and / or expanding the terms or technical descriptions in the semantic neighborhood through a concept combination function, and generating new terms that do not directly appear in the original data but are semantically derivable; Based on the identified industry entities, we build analogy relationship graphs with similar structures, functions or evolutionary paths, and learn the representations of the entities in the graph through the graph neural network model to obtain the analogy score relationship between entities, which is used to infer the potential direction of technological evolution. The generated new terms and the inferred potential technology evolution direction are integrated with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge graph containing explicit entity relationships and new concept reasoning edges; The dynamic semantic knowledge graph is periodically updated and inferred to output structured industry intelligence information, which includes new terms, technical paths, cross-domain technology integration trends, and concept evolution trajectories.

2. The method for collecting information in the additive manufacturing industry based on artificial intelligence according to claim 1, characterized in that: The step of generating new terms that do not directly appear in the original data through the concept combination function includes: Based on the term space, multiple types of concept combination functions are defined, including: term combination function, term transformation function, and term expansion function; Among them, the term combination function is used to combine two or more semantically related or complementary terms at the structural level to form new terms that express more complex or more detailed concepts; the term transformation function is used to semantically replace, transform or grammatically rewrite existing terms to form semantically similar or evolved variant terms; the term expansion function is used to add contextual modification elements, applicable conditions or technical constraints to the terms to form semantically enhanced terms.

3. The method for collecting information in the additive manufacturing industry based on artificial intelligence according to claim 1, characterized in that: The steps of constructing an analogy relationship graph with similar structure, function or evolutionary path and performing representation learning on entities in the graph through a graph neural network model include: Construct industry entity sets and entity attribute tensors, and construct structural analogy edges, functional analogy edges, and evolutionary analogy edges based on the structural similarities, functional relevance, or evolutionary chains between entities in terms of control logic, module composition, and process constraints; Based on the constructed graph structure, the graph neural network is used to update the node vectors for multiple rounds. In each iteration, the adjacent node information is aggregated based on the edge type weight. The semantic similarity scores between nodes are calculated to determine the analogy strength and potential evolution direction between technical entities.

4. The method for collecting information in the additive manufacturing industry based on artificial intelligence according to claim 1, characterized in that: The steps of constructing a dynamic semantic knowledge graph including explicit entity relationships and new concept reasoning edges include: Connect the generated new terms to the original industry knowledge graph, construct new term nodes and add term source type, generation path, first appearance time, and semantic cluster label attribute information; According to the term generation path and semantic similarity, the semantic relationship edges of term extension edge, combination derivative edge and function enhancement edge are established; The potential technology evolution path obtained by reasoning is connected to the graph in the form of analogy edges or trend edges. The unverified terms can be set as placeholder nodes and set to the "pending confirmation" status; Record the graph snapshot difference and mark the node update time and update frequency for subsequent graph evolution path tracking and structure maintenance.

5. The method for collecting information in the additive manufacturing industry based on artificial intelligence according to claim 1, characterized in that: The step of outputting structured industry intelligence information includes: Set up a graph update scheduling mechanism to regularly perform incremental updates on the dynamic semantic knowledge graph, including node attribute updates, edge weight adjustments, node activity status assessments, and version snapshot generation; Perform path reasoning, graph representation learning, hotspot clustering analysis, and multi-hop logic chain tracking on the updated graph to obtain term semantic change trends, fusion nodes, and technical paths; The analysis results are organized into structured intelligence outputs, including: new terminology intelligence tables, technology path lists, cross-domain fusion trend maps, and concept evolution trajectory maps. It also supports export in the form of structured data files or a combination of text and images.

6. An artificial intelligence-based additive manufacturing industry information collection system, characterized in that: The system includes the following modules: The data acquisition module is used to obtain raw data related to the additive manufacturing industry from multi-source heterogeneous data channels according to preset industry themes; A semantic encoding module, used to pre-process the original data, encode the original data using a deep language representation model, and generate a high-dimensional semantic vector representation; A term generation module, used for constructing a term space based on the high-dimensional semantic vector, combining and / or transforming and / or expanding terms or technical descriptions in the semantic neighborhood through a concept combination function, and generating new terms that do not directly appear in the original data but are semantically derivable; The analogy reasoning module is used to construct an analogy relationship graph with similar structures, functions or evolutionary paths based on the identified industry entities, and to learn the representation of the entities in the graph through the graph neural network model to obtain the analogy score relationship between entities for inferring the potential direction of technology evolution; A graph construction module is used to fuse the generated new terms and the inferred potential technology evolution direction with the originally extracted industry knowledge entities and relationships to construct a dynamic semantic knowledge graph containing explicit entity relationships and new concept reasoning edges; The intelligence output module is used to periodically update and perform inference calculations on the dynamic semantic knowledge graph, and output structured industry intelligence information, which includes new terms, technical paths, cross-domain technology integration trends, and concept evolution trajectories.

7. The artificial intelligence-based additive manufacturing industry information collection system according to claim 6 is characterized in that: The term generation module includes: Concept combination function library, used to define term combination function, term transformation function and term expansion function; A combination control unit, used for selecting a semantic neighborhood term pair in the term space, calling the concept combination function to generate a term semantic vector, and inputting the generated term semantic vector into a term decoder to generate a language-readable new term text; The term scoring unit is used to perform semantic consistency scoring and knowledge matching verification on the generated new terms, and mark the terms with scoring results higher than the preset threshold as semantically emergent terms.

8. The artificial intelligence-based additive manufacturing industry information collection system according to claim 6, characterized in that: The analogy reasoning module comprises: Entity attribute building unit, used to build industry entity sets and their attribute tensors; An analogy relationship building unit, used to build structural analogy edges, functional analogy edges and evolutionary analogy edges according to control logic, module composition and process constraint characteristics; The graph neural network learning unit is used to update the vectors of entity nodes in the graph structure through multiple rounds of adjacent node information aggregation; The analogy scoring unit is used to calculate the semantic similarity based on the node vectors and output the analogy scores between entities for subsequent evolutionary path reasoning.

9. The artificial intelligence-based additive manufacturing industry information collection system according to claim 6, characterized in that: The graph construction module also includes: Node access unit, used to access new terms into the original industry graph in the form of nodes, and to attach term sources, generation paths, first appearance times, and semantic cluster labels; A relationship generation unit is used to establish term extension edges, combination derivative edges, and function enhancement edges in the graph based on the term generation path and semantic similarity; The reasoning edge access unit is used to add potential technology evolution paths to the graph in the form of analogy edges or trend edges. Unverified term nodes are set as placeholder nodes and set to the "pending confirmation" state; The evolution management unit is used to record graph snapshot differences and maintain node update time, active status, and evolution tracking information.

10. The artificial intelligence-based additive manufacturing industry information collection system according to claim 6, characterized in that: The intelligence output module includes: Update scheduling unit, used to regularly trigger graph incremental update operations, update node attributes, edge weights and graph version snapshots; The reasoning analysis unit is used to perform path reasoning, graph representation learning, hot term clustering, and multi-hop logic chain analysis on the updated knowledge graph; Intelligence organization unit, used to extract new terms, technical paths, cross-domain trend nodes and evolution trajectory information, and organize them into structured intelligence reports; The output interface unit is used to export the industry intelligence information in a structured data format or in a graphic report form.

Images