Material knowledge base management method and system based on autonomous memory evolution and medium

Abstract

The invention provides a material knowledge base management method and system based on autonomous memory evolution and a storage medium, and the method comprises the steps: an atomic note construction step: responding to a received material data unit, and generating a structured atomic note containing a context description field and a semantic embedding vector field; an intelligent link generation step: responding to the generation of a new atomic note, retrieving a candidate note based on a semantic embedding vector, and analyzing whether scientific association exists or not by using a large language model to establish a link; and a knowledge evolution step: in response to establishment of a new link, triggering an evolution operation on a linked historical atomic note, and judging and updating a context description field, a keyword field or a tag field by utilizing a large language model. By constructing a link-evolution closed loop, normal form transformation from static storage to dynamic evolution is realized, so that the knowledge base can be autonomously learned and self-perfected, and the data value is deeply mined.

Description

Technical Field

[0001] This application relates to the fields of materials informatics and artificial intelligence, and in particular to a method, system and storage medium for managing materials knowledge bases based on autonomous memory evolution. Background Technology

[0002] In the field of materials science, accumulating and utilizing massive amounts of experimental data and literature knowledge is key to driving research and innovation. Existing knowledge management technologies mainly employ relational databases or knowledge graph systems that combine artificial intelligence technologies.

[0003] Relational databases store data through pre-defined table structures, which are relatively rigid and difficult to adapt to the complex and multimodal relationships between data related to composition, processes, microstructure, and properties in materials science. More importantly, the knowledge in these databases is static; once the data is stored, it is fixed and cannot be updated or evolved based on subsequent discoveries, making it difficult to effectively uncover the deep scientific connections between the data.

[0004] In recent years, some methods have emerged that utilize large language models to assist in building knowledge bases. For example, knowledge triples are extracted from unstructured documents or reports to construct knowledge graphs, which are then optimized or supplemented with new knowledge. However, the knowledge bases built using these methods are essentially static. Their optimization and supplementation operations are usually aimed at the completeness and accuracy of the knowledge base itself, such as fixing errors and filling in missing links, but they do not establish an inherent, dynamic evolutionary mechanism that triggers the re-cognition and enrichment of old knowledge content by the addition of new knowledge. In other words, when new experimental data is associated with historical data, existing technologies lack a mechanism to reflect on and update the contextual understanding of the historical data itself. For example, an old experimental conclusion may prove to have limitations under new experimental conditions, and existing systems cannot automatically supplement this new cognitive context for the old conclusion. Therefore, existing technologies generally suffer from static and isolated knowledge systems, unable to simulate the cognitive process of experts learning and gaining insights through the collision of new and old knowledge, resulting in the knowledge base's inability to achieve organic growth and continuous evolution, and its deep value cannot be fully released. Summary of the Invention

[0005] The purpose of this application is to provide a material knowledge base management method, system and storage medium based on autonomous memory evolution, so as to solve the technical problem that the material knowledge base in the prior art is static and the knowledge system cannot achieve autonomous learning and evolution through the interaction of new and old knowledge.

[0006] To achieve the above objectives, this application provides a material knowledge base management method based on autonomous memory evolution, comprising the following steps: an atomized note construction step: in response to receiving material data units, generating corresponding structured atom notes, wherein the atom notes include at least an original data field, a context description field generated by a large language model, a keyword field, a tag field, a semantic embedding vector field, and a link set field; an intelligent link generation step: in response to the generation of a new atom note, retrieving a candidate note set from historical atom notes based on its semantic embedding vector, and using a large language model to analyze whether there is a preset type of scientific association between the new atom note and the notes in the candidate note set, establishing links between related atom notes based on the analysis results, and updating the respective link set fields; a knowledge evolution step: in response to the establishment of new links in the intelligent link generation step, triggering an evolution operation on the linked historical atom notes, wherein the large language model is used to determine whether the context description field, keyword field, or tag field of the historical atom notes needs to be updated, and performing corresponding update operations according to the determination results.

[0007] Optionally, in the atomized note construction step, generating the context description field using the large language model includes: instructing the large language model to analyze the composition, process, structure, and performance data in the material data unit, and summarizing its inherent causal mechanisms and relationships to generate the context description.

[0008] Optionally, the intelligent link generation step includes: calculating vector similarity in historical atomic notes based on the semantic embedding vector of the new atomic note to retrieve a preset number of candidate notes, thereby forming the candidate note set; inputting the new atomic note and the notes in the candidate note set into the large language model to perform in-depth analysis on whether there is a preset type of scientific association between the new atomic note and the candidate notes.

[0009] Optionally, the preset type of scientific association includes at least one of the following: component similarity association, process comparison association, tissue commonality association, performance mechanism association, and causal chain association.

[0010] Optionally, the knowledge evolution step includes: taking the new atomic note and one or more historical atomic notes that are newly linked to the new atomic note as input and sending them to the large language model for analysis to determine whether the content of the historical atomic notes needs to be updated.

[0011] Optionally, the update operation includes at least one of the following: supplementing or correcting the text in the context description field of the historical atomic note; adding keywords or tags to the historical atomic note.

[0012] Optionally, the method further includes a topic index generation step: when the number of atomic notes related to a specific topic reaches a preset threshold, a topic index note is automatically created, wherein the topic index note is used to link all atomic notes related to the specific topic.

[0013] This application also provides a materials knowledge base management system based on autonomous memory evolution, comprising: one or more processors; a memory storing computer program instructions; wherein, when the computer program instructions are executed by the one or more processors, the system implements: an atomic note construction function, used to generate corresponding structured atomic notes in response to receiving material data units, wherein the atomic notes include at least an original data field, a context description field generated by a large language model, a keyword field, a tag field, a semantic embedding vector field, and a link set field; an intelligent link generation function, used to retrieve a candidate note set from historical atomic notes based on its semantic embedding vector in response to the generation of a new atomic note, and use a large language model to analyze whether there is a preset type of scientific association between the new atomic note and the notes in the candidate note set, establish links between related atomic notes based on the analysis results, and update the respective link set fields; and a knowledge evolution function, used to trigger an evolution operation on the linked historical atomic notes in response to the establishment of new links in the intelligent link generation function, wherein a large language model is used to determine whether the context description field, keyword field, or tag field of the historical atomic notes needs to be updated, and the corresponding update operation is performed according to the determination result.

[0014] Optionally, the system further includes: a vector retrieval engine, used to efficiently retrieve the candidate note set based on semantic embedding vectors when performing the intelligent link generation function; and a large language model interface, used to communicate with the large language model service to invoke the large language model to perform scientific association analysis and knowledge evolution judgment.

[0015] This application also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the method as described in any of the preceding claims.

[0016] Compared with the prior art, this application has the following beneficial effects: 1. This application achieves a paradigm shift from "static storage" to "dynamic evolution." Through a self-learning closed-loop mechanism of "construction-linking-evolution," the knowledge base is no longer a passive data archive, but an active cognitive system capable of self-learning and self-improvement. The addition of new knowledge is not merely an increase in quantity, but a catalyst that triggers qualitative optimization of existing knowledge, achieving the organic growth of knowledge.

[0017] 2. It deeply mines the value of data, enabling accelerated R&D and innovation. Through intelligent linking and knowledge evolution, the system can proactively reveal the deep scientific connections hidden behind the data, such as process optimization paths and cross-system mechanism analogies, providing researchers with unexpected insights and inspiration, thereby effectively shortening the R&D cycle and reducing trial-and-error costs.

[0018] 3. It has solidified and revitalized the organization's knowledge assets. This application transforms the tacit knowledge scattered in the minds of experts and in experimental reports into an interactive, evolving, and perpetual explicit knowledge network, solving the problem of knowledge transfer and building a core knowledge barrier that is difficult for organizations to replicate.

[0019] 4. It empowers complex scientific discovery tasks. The solution provided in this application can help scientists discover universal patterns and cross-scale correlations in massive, multi-dimensional data, transcending the role of traditional data retrieval tools and becoming a new type of scientific discovery engine. Attached Figure Description

[0020] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A flowchart illustrating a material knowledge base management method based on autonomous memory evolution, provided for an embodiment of this application; Figure 2 A schematic diagram of a materials knowledge base management system based on autonomous memory evolution is provided for an embodiment of this application; Figure 3 This is a signaling interaction timing diagram of a complete knowledge evolution process in an embodiment of this application; Figure 4 This is a schematic diagram showing the state comparison of the knowledge network before and after one evolution in an embodiment of this application. Detailed Implementation

[0021] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0022] Example 1 This application provides a material knowledge base management method based on autonomous memory evolution. To better understand the technical solution of this application, the following will be discussed in conjunction with the appendix. Figure 1 , Figure 3 and Figure 4 Taking a specific scenario of "discovering new paths for process optimization" as an example, this paper illustrates the complete process of the method. This method simulates the cognitive process of materials experts through a closed loop of "construction-linking-evolution," thereby achieving dynamic growth and self-improvement of the knowledge base and transforming new experimental data into opportunities to deepen the understanding of historical knowledge.

[0023] Reference Figure 1 This illustrates the overall flow of a material knowledge base management method based on autonomous memory evolution provided by an embodiment of this application. Specifically, the method may include: receiving material data units (S100), constructing atomized notes (S110), generating smart links (S120), determining whether to establish a new link (S130), performing knowledge evolution (S140), and updating the database (S150).

[0024] In a specific application scenario, we can assume that the materials knowledge base already stores a large number of atomic notes about different materials. Initially, the knowledge base contains a historical atomic note A, which records an experiment on the preparation of ceramic material X. For example... Figure 4 As shown in the left half, the knowledge network already contains nodes such as Atomic Notebooks A, C, and D, as well as existing links L1 between them. The specific content of Atomic Notebook A can be structured and stored in the material database 20, for example, in JSON format. { "note_id": "A", "creation_date": "2023-01-15", "raw_data": { "composition": "Ceramic Material X", "process": { "type": "Traditional sintering", "temperature": "1500°C", "duration": "2h" }, "structure": { "grain_size": "5μm }, "performance": { "density": "90%", "strength": "200MPa" } }, "context_description": "Ceramic material X was prepared using traditional sintering process A and held at 1500°C for 2 hours, achieving a density of 90% and a flexural strength of 200 MPa." Keyword: ["Ceramic Material X", "Traditional Sintering", "Density", "Mechanical Strength"], "tags": ["sintering process", "zirconia ceramics"], "embedding_vector": [0.12, -0.34, ..., 0.56], / / A high-dimensional floating-point array "links": ["note_id_C"] / / Assuming there is an existing link L1 to note C } Understandably, in the above data structure, note_id is the unique identifier of the note; the raw_data field is used to store objective, raw experimental data; the context_description field is a natural language summary of the experiment generated by the large language model; the keywords and tags fields are keywords and tags used for retrieval and classification, respectively; the embedding_vector field is a semantic embedding vector calculated by a text encoding model based on the core text information of the note (such as context_description and keywords); and the links field is a collection of links used to store the identifiers of other atomic notes associated with this note.

[0025] First, step S100 of receiving material data units is performed. For example, when a researcher completes a new experiment and enters its data into the system, the system receives a new material data unit B. The content of this data unit B may be: Ceramic material X with a similar composition to that in note A was prepared using a spark plasma sintering process, achieving a density of 99% and a strength of 400 MPa.

[0026] Subsequently, step S110, constructing the atomic note, is executed. In response to the received new data unit B, the system generates a completely new structured atomic note B for it. Specifically, this step can be performed by… Figure 2The atomic note-building module 11 in the intelligent management system 10 is executed. This module 11 packages the raw data (e.g., composition, process, performance, etc.) of the new data unit B and sends a request to the large language model service 50 through the large language model interface 40. The request instruction can be designed as: "Please analyze the following material experimental data, generate a contextual description containing scientific mechanism analysis, and extract core keywords and classification labels. Data: {…}". After understanding the data, the large language model service 50 returns the generated text. Accordingly, the atomic note-building module 11 combines the raw data, the returned text, and the semantic embedding vector calculated through the text encoding model into a complete atomic note B and stores it in the material database 20. The structure of the newly generated atomic note B can be exemplified as follows: { "note_id": "B", "creation_date": "2024-05-20", "raw_data": { "composition": "Ceramic Material X", "process": { "type": "Discharge Plasma Sintering", "temperature": "1300°C", "duration": "5min }, "structure": { "grain_size": "0.5μm" }, "performance": { "density": "99%", "strength": "400MPa" } }, "context_description": "Using advanced spark plasma sintering process B, through rapid heating and pressurization, grain growth is effectively suppressed, thereby significantly improving the density and mechanical strength of ceramic material X." Keyword: ["Ceramic Materials X", "Spark Plasma Sintering", "High Density", "High Strength"] Tags: ["Advanced sintering process", "Zirconium oxide ceramics", "Performance optimization"] "embedding_vector": [0.21, -0.11, ..., 0.78], "links": [] / / Initially empty } After Atomic Note B is successfully created and stored in the database, step S120, which generates smart links, will be automatically triggered. This step can be executed by the smart link generation module 12. In this step, the smart link generation module 12 first extracts the semantic embedding vector of Atomic Note B and submits it as a query vector to the vector retrieval engine 30. Accordingly, the vector retrieval engine 30 performs an efficient similarity retrieval (e.g., based on cosine similarity or Euclidean distance) in the index that stores all historical note vectors and returns the identifiers of the top-K most semantically relevant candidate notes (where K can be 10). It is understandable that since both Atomic Note A and B involve "ceramic material X" and similar performance indicators, Atomic Note A is highly likely to appear in the returned set of candidate notes.

[0027] Next, the intelligent link generation module 12 enters the deep judgment stage of scientific association. In this stage, the module pairs the content of the new atomic note B with the content of each note in the candidate note set (including atomic note A) and requests the large language model service 50 for analysis again through the large language model interface 40. At this time, the request instruction is more specific, aiming to determine whether there is a pre-defined type of scientific association. For example, for the combination of notes A and B, the instruction can be designed as: "Analyze the following two experimental notes on ceramic material X. Note A uses conventional sintering, and note B uses spark plasma sintering. Is there a 'process comparison association' between them? Please answer 'yes' or 'no' and provide reasons." After analysis, the large language model service 50 may return the following response: "Yes. The two notes describe different preparation processes for the same material system, and the performance results are significantly different, constituting a typical process comparison relationship, which can be used to evaluate the advantages and disadvantages of different processes." In step S130, the system determines whether a positive association judgment has been received from the large language model. In this embodiment, since the received judgment is positive, the intelligent link generation module 12 will perform a link establishment operation. Specifically, this module adds the identifier "B" of Atomic Note B to the links field of Atomic Note A, and simultaneously adds the identifier "A" of Atomic Note A to the links field of Atomic Note B, thereby establishing a bidirectional new link L_new between Note A and Note B, as shown below. Figure 4 As shown in the right half. Next, step S150, updating the database, will persist the updates to these two notes.

[0028] When a new link is successfully established in the system, a key triggering condition is met, and step S140 of knowledge evolution will be immediately triggered. This step can be executed by the knowledge evolution module 13, which aims to reflect on and enrich old knowledge (note A) using new knowledge (note B). As an optional implementation, the knowledge evolution module 13 takes the entire contents of the new atomic note B and the entire contents of the historical atomic note A with which it has newly established a link as input, and sends a third request to the large language model service 50 through the large language model interface 40. The instructions in this request are used to obtain evolutionary decisions, such as: "It is known that note A is a historical record of traditional sintering processes. A note B about advanced sintering processes has been found to have a 'process comparison association' with A. Based on the new information provided by note B, how should the understanding of note A be updated or supplemented to make its knowledge more comprehensive and insightful? Please provide specific update suggestions, including supplementary text for the 'context_description' field and 'tags' that need to be added." After understanding the above context, the Big Language Model Service 50 will generate evolutionary suggestions, such as: "1. Add the following to the end of the context_description field of Note A: 'It is worth noting that subsequent research (see Note B) shows that the use of spark plasma sintering can achieve higher density (99%) and strength (400MPa) in this material system, which is an effective optimization direction for this process.' 2. Add a new tag to Note A: 'Optimizable - Sintering Process'." After receiving and parsing the above suggestions, the knowledge evolution module 13 performs a corresponding update operation on atomic notebook A in the materials database 20. The updated atomic notebook A (in...) Figure 4 The content of the evolved atomic notebook A* can be exemplified as follows: { "note_id": "A", "version": 2, / / Optional version number ... "context_description": "Ceramic material X was prepared using conventional sintering process A and held at 1500°C for 2 hours, achieving a density of 90% and a flexural strength of 200 MPa. Notably, subsequent research (see Note B) indicates that spark plasma sintering can achieve even higher density (99%) and strength (400 MPa) in this material system, representing an effective direction for optimizing this process." ... Tags: ["Sintering process", "Zirconium oxide ceramics", "Optimizable sintering process"] ... "links": ["note_id_C", "B"] } Finally, the system executes step S150 again to update the database and persistently store the evolved atomic note A*.

[0029] Reference Figure 3 This diagram, a sequence diagram of signaling interaction in this embodiment, fully illustrates the series of interaction processes described above. From the user submitting new data (message 1), to the three core interactions between the intelligent management system S and the large language model service L for creating notes (message 2), determining associations (message 5), and requesting evolution (message 8), as well as the retrieval interaction with the vector retrieval engine V (messages 3 and 4), this diagram clearly depicts the data flow and control logic in the entire closed-loop process.

[0030] Through the complete process described in this embodiment, the originally static and isolated historical note A is given new context and meaning with the addition of the new note B. When researchers subsequently search for or browse note A about "traditional sintering process A," the system will not only display its original information but also proactively suggest the existence of a superior process B and provide a direct link. Understandably, this mechanism can significantly enhance the application value of the knowledge base, transforming it from a static data set into a dynamic knowledge system that proactively provides relevant insights, thereby helping to accelerate R&D innovation.

[0031] Example 2 This embodiment will further illustrate the ability of the large language model to make scientific association depth judgments in the intelligent link generation step S120, especially its application in establishing deep mechanism associations across different material systems. As a preferred implementation, the preset scientific association types may include "performance mechanism associations".

[0032] It should be noted that in the field of materials science, many underlying physical or chemical principles are universal, but their specific applications are scattered across different material systems (e.g., steel, aluminum alloys, high-temperature alloys, ceramics, etc.). Researchers often find it difficult to achieve cross-disciplinary knowledge application due to the limitations of their own fields. The technical solution provided in this application can effectively break down such knowledge barriers.

[0033] Suppose that the following two historical atomic notes already exist in the materials database 20: Atomic Note C, which contains a literature abstract on "precipitation strengthening in low-carbon steel by forming nanoscale NbC precipitates to hinder dislocation movement," and whose keywords may include "low-carbon steel," "microalloying," "NbC," and "precipitation strengthening"; and Atomic Note D, which contains experimental data on "excellent high-temperature creep resistance in nickel-based superalloys by forming a large number of ordered γ' phases as strengthening phases," and whose keywords may include "nickel-based alloys," "γ' phase," and "precipitation strengthening."

[0034] Clearly, notes C and D belong to completely different material systems (i.e., iron-based alloys and nickel-based alloys), and their specific strengthening phases, compositions, and application scenarios are also quite different. In traditional databases, it is almost impossible to establish a connection between the two through simple keyword matching.

[0035] When a researcher enters a recent paper on “a novel high-entropy alloy exhibits significantly improved yield strength by precipitating L12-structured nanoprecipitates through heat treatment,” the system will execute step S110 to construct an atomized note in order to create a new atomized note E.

[0036] Subsequently, in the step S120 of generating intelligent links, the system first finds a set of candidate notes including notes C and D through vector retrieval (because they are all related to macroscopic concepts such as "alloy," "strengthening," and "precipitation"). In the next deep judgment stage, the intelligent link generation module 12 combines notes E with notes C and notes E with notes D respectively, and sends a judgment request to the large language model service 50.

[0037] For example, for a combination of notes E and C, the instruction could be designed as: "Analyze notes E (regarding the L12 precipitate in high-entropy alloys) and C (regarding the NbC precipitate in steel). Despite the different material systems and precipitate types, do the material strengthening phenomena they describe originate from the same fundamental physical metallurgical principle? If so, identify the principle and confirm the 'performance mechanism correlation' between them." Based on its knowledge gained from training on massive amounts of scientific literature, the Large Language Model Service 50 can identify that both improve material strength by impeding dislocation movement through second-phase particles dispersed in the matrix. Therefore, the service returns an affirmative judgment, confirming a "performance mechanism correlation" based on the "precipitation strengthening" principle between the two. Similarly, for the combination of notes E and D, the Large Language Model can also identify the underlying common "precipitation strengthening" mechanism.

[0038] Accordingly, the intelligent link generation module 12 will establish "performance mechanism association" type links between Atomic Notebook E and C, and between E and D, and update their respective link set fields.

[0039] The beneficial effect of this embodiment is that when a researcher focusing on high-entropy alloys browses their newly created note E, the system will display links to classic research notes (C and D) on precipitation strengthening in steel and nickel-based superalloys. This provides researchers with important reference information; for example, they can draw upon well-established control strategies in the field of steel and superalloys regarding the size, shape, volume fraction, and coherence relationships of precipitates to optimize their heat treatment processes for high-entropy alloys. This cross-disciplinary automatic knowledge linking and recommendation function is difficult to achieve effectively in traditional knowledge management systems, and it can stimulate innovative thinking, thereby helping to accelerate the process of scientific discovery. This embodiment fully demonstrates the effective support of this application for the "performance mechanism association" among the preset scientific association types (including but not limited to composition similarity association, process comparison association, common microstructure association, performance mechanism association, and causal chain association).

[0040] Example 3 This embodiment describes a preferred implementation method based on the basic process, namely a knowledge aggregation function, specifically manifested in the generation of topic indexes. This function aims to automatically abstract and summarize knowledge when the system accumulates a predetermined number of notes on a specific scientific connection or topic, in order to create a higher-level knowledge entry point.

[0041] As one implementation, this method can add a step to generate a topic index. This step can be continuously monitored by a separate background service, or implemented as an extension function of knowledge evolution step S140. Its core logic lies in monitoring "knowledge clusters" in the knowledge network.

[0042] Based on the aforementioned Example 2, it can be assumed that over time, more atomic notes related to "precipitation strengthening" will be added to the knowledge base, such as notes about precipitation strengthening in aluminum or magnesium alloys. At this time, in the knowledge network of the materials database 20, a dense subgraph will be formed, consisting of atomic notes C, D, E and other related notes, interconnected by links of the "performance mechanism association - precipitation strengthening" type.

[0043] At this point, the triggering condition for generating the topic index is met. This condition can be set as: "when the number of links with the same type and subtype (e.g., 'performance mechanism association - extraction enhancement') exceeds a preset threshold N (e.g., N=5), or when the number of independent notes connected by such links exceeds a preset threshold M (e.g., M=3).

[0044] Once the conditions are triggered, the system will perform the following operations: 1. Identification and Aggregation: The system first queries the database to find a list of identifiers (e.g., {C, D, E, G, H}) of all atomic notes that participate in constituting the "extracted reinforcement" knowledge cluster. 2. Content Abstraction and Generation: The system packages the core information of these notes (e.g., note_id, summary of context_description, keywords) and sends it to the large language model service 50 through the large language model interface 40. The instruction at this point can be designed as a generative summarization and naming task, for example: "The following is a set of notes on the 'precipitation strengthening' mechanism in different material systems. Please complete the following tasks: a) Create a concise and comprehensive title for this topic; b) Write a review description of about 200 words introducing the basic principles of precipitation strengthening and mentioning its application characteristics in these different material systems (e.g., steel, nickel-based alloys, high-entropy alloys, etc.); c) Extract common keywords for this topic." 3. Create Topic Index Notes: The Large Language Model Service 50 may return the following results: * Title: "Application and Comparison of Precipitation Strengthening Mechanisms in Multimetallic Systems" * Review: "Precipitation strengthening is a key material strengthening mechanism, the core of which lies in effectively hindering dislocation movement by forming dispersed second-phase particles in the matrix… This topic compiles specific application cases of this mechanism in various advanced metallic materials such as low-carbon steel (through the NbC phase), nickel-based superalloys (through the γ' phase), and high-entropy alloys (through the L12 phase)…" * Keywords: ["precipitation strengthening", "second-phase strengthening", [Dislocation Interaction, Material Design Principles] Based on the returned content, the system creates a new, special type of atomic note called "Topic Index Note F". 4. Link Aggregation: The links field of Topic Index Note F is populated with links to all related atomic notes (C, D, E, G, H). Alternatively, as an optional implementation, a link back to Topic Index Note F can be added to each linked atomic note (C, D, E, G, H) and marked as "Belongs to Topic: Extraction Enhancement".

[0045] Through the functions described in this embodiment, the system is no longer limited to point-to-point knowledge connections, but can spontaneously construct a hierarchical structure of knowledge. Researchers can not only conduct "bottom-up" exploration (i.e., discovering another related note from a specific note), but also "top-down" learning. For example, a beginner in materials science can directly search for "strengthening mechanism" to find the topic index note F of "precipitation strengthening," and then systematically understand the application examples of this principle in various materials in a one-stop manner, which greatly improves the efficiency and systematicness of knowledge acquisition. Understandably, this function enables the knowledge base to evolve from a flat network graph into a dynamic knowledge system with hierarchical directories and structured content.

[0046] Example 4 This embodiment will combine Figure 2 and Figure 3 The hardware and software architecture of the intelligent management system 10 that implements the above method, as well as the collaborative workflow between its components, are described in detail.

[0047] Reference Figure 2 This is a schematic diagram of a materials knowledge base management system based on autonomous memory evolution. In one embodiment of this application, the intelligent management system 10 can be deployed on one or more servers. Each server includes basic hardware components, such as one or more processors (e.g., high-performance multi-core central processing units) and memory (e.g., including high-speed random access memory and solid-state drives for persistent storage). The memory stores computer program instructions, which, when executed by the processor, realize the various functions described in this application.

[0048] At the software level, the intelligent management system 10 can be instantiated from the following core functional modules: Atomic Note Building Module 11: This module is responsible for handling all new knowledge entry points and can be implemented as a microservice that provides an application programming interface (API). When a user or other system submits a new material data unit through this interface (e.g., ... Figure 3 (Message 1) This module will validate and format the data, and construct a request based on a preset template, communicating with the large language model service 50 through the large language model interface 40. Figure 3 The module retrieves the generated context description, keywords, and tags from message 2. Simultaneously, it calls a text encoding service (which can be integrated locally or via an interface) to generate semantic embedding vectors. Finally, the module assembles all the information into a complete atomic note object, writes it to the material database 20, and writes the note's ID and vector data to the vector retrieval engine 30.

[0049] Intelligent Link Generation Module 12: This module is the core for realizing knowledge networking. Its execution can be activated by triggers in the material database 20 (e.g., when a new note is inserted). After activation, module 12 obtains the ID and semantic embedding vector of the new note from the triggering event and uses this vector to initiate a query through the client library of the vector retrieval engine 30. Figure 3 Message 3 in the middle requests the return of the top-K similar neighbor note IDs. The vector retrieval engine 30 (e.g., using an open-source vector database) utilizes its efficient indexing structure to return a list of candidate notes in milliseconds. Figure 3 Message 4 in the document). Subsequently, module 12 retrieves the complete content of these candidate notes in batches from the material database 20 based on the returned ID list, and pairs the new notes with each candidate note to construct a request for association judgment, which is then sent to the large language model service 50 through the large language model interface 40. Figure 3 Message 5 in the middle). After receiving all judgment results ( Figure 3 (Message 6 in the document), module 12 filters out all note pairs determined to have scientific relevance and performs an update operation in the materials database 20, updating the links field of these note pairs bidirectionally. Figure 3 Action 7 in the middle.

[0050] Knowledge Evolution Module 13: This module is crucial for realizing knowledge evolution. Its execution can be activated by another trigger in the material database 20 (e.g., when the links field of a note is updated). Upon activation, module 13 identifies the note pair whose links have changed (i.e., the new note and the linked historical note) from the triggering event and retrieves the complete content of these two (or more) notes from the database to construct a specific request for requesting evolutionary recommendations. This request is sent to the large language model service 50 via the large language model interface 40. Figure 3 Message 8 in the middle. After receiving a response containing specific update suggestions ( Figure 3 (Message 9 in the document), module 13 is responsible for parsing the response and performing precise update operations on the historical notes in the material database 20, such as modifying the context_description field or adding new elements to the tags array. Figure 3 Action 10 in the middle).

[0051] In addition to the core modules mentioned above, the system also relies on several key underlying services: Material Database 20: A document database can be used, whose flexible schema characteristics are well-suited for storing atomic notes whose structure may evolve over time. This database is responsible for persistent knowledge storage and precise retrieval. Vector Retrieval Engine 30: Responsible for storing, indexing, and efficiently retrieving large-scale vector data, it is crucial for implementing the "vector initial screening" step and ensuring system response speed and scalability. Large Language Model Interface 40: As an abstract communication layer, it encapsulates all the details of interacting with the Large Language Model Service 50, including interface authentication, request serialization, response deserialization, and error handling. This allows upper-layer modules to easily invoke the intelligence of the large language model without needing to concern themselves with its specific deployment form. Large Language Model Service 50: As the engine providing core intelligence, it can be a commercial cloud service (such as various publicly available large model application programming interfaces) or an open-source or self-developed large language model deployed locally or in a private cloud.

[0052] The entire system, through modular design and event-driven mechanisms, forms a highly efficient and automated data processing pipeline. From data input to knowledge evolution, the entire process can be completed without human intervention, thereby achieving autonomous learning and organic growth of the knowledge base. The system architecture described in this embodiment provides a specific, feasible, and efficient engineering implementation scheme for the methods described in the claims of this application. Furthermore, computer-readable storage media (e.g., server hard drives, solid-state drives, or cloud storage) storing computer program instructions for implementing the functions of the above modules also fall within the scope of protection of this application.

[0053] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0054] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A material knowledge base management method based on autonomous memory evolution, characterized in that, Includes the following steps: Atomic note construction steps: In response to receiving material data units, a corresponding structured atomic note is generated, wherein the atomic note includes at least the original data field, the context description field generated by the large language model, the keyword field, the tag field, the semantic embedding vector field, and the link set field; Intelligent link generation steps: In response to the generation of a new atomic note, a candidate note set is retrieved from the historical atomic notes based on its semantic embedding vector, and a large language model is used to analyze whether there is a preset type of scientific association between the new atomic note and the notes in the candidate note set. Based on the analysis results, links are established between related atomic notes, and the respective link set fields are updated. Knowledge evolution steps: In response to the establishment of a new link in the intelligent link generation step, the evolution operation of the linked historical atomic notes is triggered. In this process, a large language model is used to determine whether the context description field, keyword field, or tag field of the historical atomic notes needs to be updated, and the corresponding update operation is performed based on the determination result.

2. The method according to claim 1, characterized in that, In the atomic note-building step, generating the context description field using the large language model includes: The instruction states that the large language model analyzes the composition, process, structure, and performance data in the material data unit, and summarizes its inherent causal mechanisms and relationships to generate the context description.

3. The method according to claim 1, characterized in that, The intelligent link generation step includes: Based on the semantic embedding vector of the new atomic notebook, vector similarity is calculated in the historical atomic notebooks to retrieve a preset number of candidate notes, thereby forming the candidate note set; The new atomic notes and the notes in the candidate note set are input into the large language model to perform in-depth analysis on whether there is a preset type of scientific association between the new atomic notes and the candidate notes.

4. The method according to claim 3, characterized in that, The preset types of scientific associations include at least one of the following: component similarity association, process comparison association, tissue commonality association, performance mechanism association, and causal chain association.

5. The method according to claim 1, characterized in that, The knowledge evolution steps include: The new atomic note, along with one or more historical atomic notes that are newly linked to the new atomic note, are taken as input and analyzed by the large language model to determine whether the content of the historical atomic notes needs to be updated.

6. The method according to claim 5, characterized in that, The update operation includes at least one of the following: Add or correct the text to the context description field of the historical atomic notes; Add keywords or tags to the historical atomic notes.

7. The method according to claim 1, characterized in that, The method further includes: Topic index generation steps: When the number of atomic notes related to a specific topic reaches a preset threshold, a topic index note is automatically created, wherein the topic index note is used to link all atomic notes related to the specific topic.

8. A materials knowledge base management system based on autonomous memory evolution, characterized in that, include: One or more processors; A memory that stores computer program instructions; When the computer program instructions are executed by the one or more processors, the system enables: The atomic note building function is used to generate corresponding structured atomic notes in response to received material data units. The atomic notes include at least the original data field, the context description field generated by the large language model, the keyword field, the tag field, the semantic embedding vector field, and the link set field. The intelligent link generation function is used to respond to the generation of a new atomic note by retrieving a set of candidate notes from the historical atomic notes based on its semantic embedding vector, and using a large language model to analyze whether there is a preset type of scientific association between the new atomic note and the notes in the candidate note set. Based on the analysis results, links are established between related atomic notes, and the respective link set fields are updated. The knowledge evolution function is used to trigger the evolution operation of the linked historical atomic notes in response to the establishment of a new link in the intelligent link generation function. In this process, a large language model is used to determine whether the context description field, keyword field or tag field of the historical atomic notes needs to be updated, and the corresponding update operation is performed according to the determination result.

9. The system according to claim 8, characterized in that, The system also includes: A vector retrieval engine is used to efficiently retrieve the candidate note set based on semantic embedding vectors when performing the intelligent link generation function; The large language model interface is used to communicate with the large language model service to invoke the large language model to perform scientific association analysis and knowledge evolution judgment.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images