Method for establishing laboratory safety knowledge base

Abstract

The invention relates to the technical field of laboratory knowledge base systems, in particular to a method for establishing a laboratory safety knowledge base. A file type database, a directory type database, an entity and entity relation library, a data processing module, an application module and a user side interface are included. And the file type database is used for storing standard specifications, management methods, rules and regulations, operation regulations, case analysis, equipment and facility files and official directory files. According to the method, a cross-laboratory type knowledge association network is established, the risk source directory database solves the problem of cross-domain knowledge integration fragmentation through parameter standardization and a security risk level-regulation term mapping rule, and the systematic management efficiency of multi-type laboratory data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of laboratory knowledge base systems, and in particular to a method for establishing a laboratory safety knowledge base. Background Art

[0002] In recent years, with the rapid development of scientific research activities, the complexity and importance of laboratory safety management have become increasingly prominent. Laboratory safety data covers multiple types of content, including regulatory standards, equipment operating specifications, risk source lists (such as chemicals, pathogenic microorganisms), laboratory operating specifications, accident cases, etc., involving multimodal data such as text, tables, pictures, and videos.

[0003] With the popularization of information technology, some institutions have tried to build a safety knowledge base through electronic databases or knowledge graph technology. However, existing databases usually use single-field structured storage, which is difficult to be compatible with the differentiated formats of different sources (such as national standards, local documents, and laboratory internal records). It is impossible to achieve unified association and rapid retrieval, making the integration of multi-source heterogeneous data inefficient. Therefore, it is urgent to design a method to establish a laboratory safety knowledge base to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for establishing a laboratory safety knowledge base to solve the problem proposed in the above background technology that the existing database usually adopts a single field structured storage, is difficult to be compatible with the differentiated formats of different sources, cannot achieve unified association and rapid retrieval, and makes the integration efficiency of multi-source heterogeneous data low.

[0005] To achieve the above object, the present invention provides the following technical solutions: a method for establishing a laboratory safety knowledge base: comprising a file type database, a directory type database, an entity and entity relationship database, a data processing module, an application module and a user terminal interface;

[0006] The file type database is used to store standards and specifications, management methods, rules and regulations, operating procedures, case studies, equipment and facility documents, and official directory documents. It supports PDF / Word / Excel / video / audio multimodal data, and each file type is set with structured fields.

[0007] The list type database includes a list of hazardous chemicals, a list of pathogenic microorganisms, and a list of radioactive substances, each of which is associated with an official document source and defines an entity relationship;

[0008] The entity and entity relationship library automatically generates entity association relationships based on the AI big model to build a knowledge graph;

[0009] The data processing module integrates BERT+TF-IDF text analysis, FFmpeg frame hashing+OpenCV multimedia feature extraction technology to perform data cleaning, illegal content judgment and entity disambiguation operations;

[0010] The application module supports laboratory filing review guidance, intelligent risk source identification, knowledge graph query, and file viewing and downloading functions;

[0011] The client interface supports multiple clients to upload / update data and triggers automatic prompts to generate entity relationship data.

[0012] Each type of document structured field in the document type database includes a document number, an effective date, a revision, and an associated laboratory type code;

[0013] The illegal content judgment function of the data processing module is realized through a pre-trained classification model, and the model integrates the knowledge graph entity relationship to perform multi-dimensional verification;

[0014] The AI model of the entity and entity relationship library adopts a graph neural network architecture to dynamically update the causal chain and risk transmission path between entities;

[0015] The risk source intelligent identification function of the application module is configured with a threshold trigger mechanism, which generates an early warning map when it is detected that the risk correlation of multi-source heterogeneous data exceeds the set threshold.

[0016] Preferably, the equipment and facility files and case analysis files in the file type database support the storage of video / audio attachments, and the association index between multimedia content and text description is realized through VIT feature extraction technology.

[0017] Preferably, the hazardous chemicals list field of the list type database includes 16 structured parameters of the MSDS document, and the pathogenic microorganism list field includes virus culture laboratory level and transport packaging requirements, and is automatically linked to the original terms of the corresponding regulatory documents.

[0018] Preferably, the entity and relationship library automatically prompts to generate entity association relationships across data sources through the AI big model, including three types of relationship networks: "file-directory entity association", "file-file entity association", and "directory-risk source hierarchical relationship".

[0019] Preferably, the data processing module adopts a multimodal violation detection strategy:

[0020] The text content is judged as sensitive information through the BERT violation detection model;

[0021] The video / audio content is processed by FFmpeg to extract key frames, and OpenCV feature matching is used to identify illegal operation scenarios.

[0022] Preferably, the laboratory filing and review guidance function of the application module automatically matches standard specifications, management system templates and necessary equipment lists according to the laboratory type (biology / chemistry / radiation) to generate a compliance checklist.

[0023] Preferably, the user interface sets hierarchical permissions:

[0024] Ordinary user terminals only support file viewing / downloading and knowledge graph query;

[0025] The operational user end supports file upload / update, triggers the AI big model to automatically generate entity relationship data, and synchronously updates the graph.

[0026] Preferably, the data processing module adopts dynamic entity relationship weight calculation, and for repeatedly uploaded content, based on Jaccard similarity and Siamese network double verification, gives priority to retaining high timestamp or high confidence version data.

[0027] Preferably, the knowledge graph query supports multi-hop reasoning, inputs a natural language description, and returns the associated original text of rules and regulations, equipment operation videos, and emergency plan cases.

[0028] Preferably, the user-end interface provides a visual interactive interface, integrates a force-directed graph to display the laboratory-risk source-control measure association network, and supports clicking on a node to directly download a PDF / video attachment or jump to the original text of the regulation.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. This system realizes automatic classification and storage of multi-source regulatory documents through preset laboratory type codes (biological / chemical / radiation, etc.) and structured field systems (document number + effective date + revision version). The document type database adopts laboratory type labeling index, intelligently matches national standards, local specifications and internal documents according to scenarios such as biosafety cabinet operation and hazardous chemical storage, and establishes a knowledge association network across laboratory types. The risk source catalog database solves the problem of fragmented cross-domain knowledge integration through parameter standardization and safety risk level-regulatory clause mapping rules, thereby improving the systematic management efficiency of multi-type laboratory data.

[0031] 2. The system uses a BERT + TF-IDF fusion model to analyze text semantics, combining FFmpeg frame segmentation with OpenCV feature matching for audio and video processing, enabling automated feature extraction from unstructured data. A dual deduplication mechanism, using Jaccard similarity screening and Siamese network semantic verification, dynamically calculates entity relationship weights and prioritizes high-confidence data versions, improving cleaning efficiency by 80% compared to traditional manual review. Multimodal data is aligned across multiple modalities using VIT + CLIP to generate a unified semantic index, addressing the integration bottleneck of heterogeneous PDF, video, and text data and reducing data management labor costs.

[0032] 3. The system builds a dynamically evolving knowledge graph, mining causal chains and risk transmission pathways between entities using graph neural networks. When users enter natural language queries, BERT entity recognition and a three-hop inference engine aggregate relevant regulatory texts, operational videos, and historical cases to form a scenario-based knowledge package. The filing and review function automatically associates a list of required documents (e.g., mandatory radioactive operation guidelines for radiation laboratories) based on the laboratory type code, providing comprehensive knowledge from risk identification to compliance, ensuring that practitioners can access systematic solutions with a single search.

[0033] 4. The system constructs a three-dimensional interactive network based on a force-directed graph algorithm, using node size to identify risk levels and edge color to distinguish relationship types (red / blue corresponds to risk / compliance). Users can click on a risk source node to drill down to view MSDS documents, emergency videos, and historical accident chains. Path highlighting dynamically displays the trajectory of risk transmission. The visual interface supports zooming and focusing, synchronized with real-time data, and transforms complex entity relationships into actionable topological maps, helping managers quickly locate risk nodes and formulate prevention and control strategies, thereby improving safety decision-making efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A platform block diagram of the entity and entity relationship library of the present invention;

[0035] Figure 2 A platform block diagram created for the data of the present invention;

[0036] Figure 3 This is a platform block diagram of the database application of the present invention. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] See also Figure 1-3 , an embodiment provided by the present invention:

[0039] A method for establishing a laboratory safety knowledge base: comprising a file type database, a directory type database, an entity and entity relationship database, a data processing module, an application module and a user terminal interface;

[0040] The file type database is used to store standards and specifications, management methods, rules and regulations, operating procedures, case studies, equipment and facility documents, and official directory files. It supports PDF / Word / Excel / video / audio multimodal data, and each file type has structured fields.

[0041] The list type database includes the list of hazardous chemicals, the list of pathogenic microorganisms, and the list of radioactive substances. Each list is associated with the official document source and defines the entity relationship;

[0042] The entity and entity relationship database automatically generates entity association relationships based on the AI big model to build a knowledge graph;

[0043] The data processing module integrates BERT+TF-IDF text analysis, FFmpeg frame hashing+OpenCV multimedia feature extraction technology to perform data cleaning, illegal content judgment and entity disambiguation operations;

[0044] The application module supports laboratory filing review guidelines, intelligent risk source identification, knowledge graph query, and file viewing and downloading functions;

[0045] The client interface supports multiple clients to upload / update data and triggers automatic prompts to generate entity relationship data.

[0046] The structured fields for each type of document in the document type database include the document number, effective date, revision, and associated laboratory type code;

[0047] The illegal content judgment function of the data processing module is implemented through a pre-trained classification model, and the model integrates the knowledge graph entity relationship to perform multi-dimensional verification;

[0048] The AI model of the entity and entity relationship database uses a graph neural network architecture to dynamically update the causal chain and risk transmission path between entities;

[0049] The risk source intelligent identification function of the application module is equipped with a threshold trigger mechanism. When it is detected that the risk correlation of multi-source heterogeneous data exceeds the set threshold, an early warning map is generated. The modules of the system form a closed-loop collaborative mechanism through data flow and knowledge chain: after the user-end interface receives multi-source heterogeneous data, the data processing module uses BERT+TF-IDF to perform text semantic analysis, and combines FFmpeg+OpenCV to extract audio and video features to complete structured cleaning and violation detection; the purified data is stored in the file type database (with laboratory type coding) and the directory type database (bound to official traceability information) respectively, to achieve real-time data protection. The graph neural network of the entity and entity relationship library parses multimodal data in real time, generates entity relationship triples with causal chains and updates the knowledge graph; the application module synchronously calls the graph data, calculates the correlation of multi-source data through risk transmission path analysis, and when the threshold trigger mechanism is activated, it links the file database to retrieve related cases to generate an early warning graph; when the user records are reviewed, the system automatically matches the standard documents corresponding to the laboratory type code, forming a full-chain closed loop of knowledge retrieval-risk warning-decision support. Each module realizes data feedback based on a unified entity relationship framework, and improves the real-time and predictability of safety management through dynamic knowledge evolution.

[0050] Furthermore, the equipment and facility files and case analysis files in the file type database support the storage of video / audio attachments, and the association index of multimedia content and text descriptions is realized through VIT feature extraction technology. The (VIT) model is used to extract features of video key frames to generate 512-dimensional visual semantic vectors; the text description is simultaneously encoded through the CLIP model to generate matching text semantic vectors. The two are cross-modally associated through cosine similarity calculation, and support the positioning of video clips by text queries. When a new video / audio attachment is added, the system automatically triggers the VIT feature extraction process and updates the multimedia-text association matrix to ensure the real-time and consistency of the knowledge base retrieval results.

[0051] Furthermore, the hazardous chemicals list field of the list type database contains 16 structured parameters of the MSDS document, and the pathogenic microorganism list field contains the virus culture laboratory level and transportation packaging requirements, and is automatically linked to the original clauses of the corresponding regulatory documents. Through the rule engine "List of Pathogenic Microorganisms Transmitted by Humans", the clause number is matched, and the semantic similarity between the Latin name of the virus and the regulatory text is calculated to achieve accurate mapping of the list items and the original text of the regulations. The 16 structured parameters of the MSDS document of the hazardous chemicals list include: chemical name (PRODUCT), chemical composition information (COMPOSITION), hazard information (HAZARDS IDENTIFICATION), first aid measures (FIRST-AID MEASURES), fire fighting measures (FIRE-FIGHTING MEASURES), accidental release measures (ACCIDENTAL RELEASE MEASURES), exposure control and personal protection measures (EXPOSURE CONTROLS / PERSONAL PROTECTION), physical and chemical properties (PHYSICAL AND CHEMICAL PROPERTIES), stability and reactivity (STABILITY AND REACTIVITY), TOXICOLOGICAL INFORMATION, HANDLING AND STORAGE, DISPOSAL CONSIDERATION, TRANSPORT INFORMATION, REGULATION INFORMATION, OTHER INFORMATION, NATIONAL EMERGENCY TELEPHONY NUMBER.

[0052] Furthermore, the entity and relationship library automatically prompts to generate entity association relationships across data sources through the AI big model, and adopts a multi-task learning framework that combines a pre-trained language model with a graph neural network (GNN), including three types of relationship networks: "file-directory entity association", "file-file entity association", and "directory-risk source hierarchical relationship", to perform cross-data source associations, and realize cross-modal retrieval of text and video through VIT+CLIP. The AI big model is combined with the rule engine to achieve high-precision entity association.

[0053] Furthermore, the data processing module adopts a multimodal violation detection strategy:

[0054] The text content is judged as sensitive information through the BERT violation detection model;

[0055] Key frames are extracted from video / audio content through FFmpeg frame processing, and OpenCV feature matching is used to identify illegal operation scenarios. The BERT+graph neural network model is used to automatically generate a "file-directory-risk source" entity relationship network across data sources through prompts, dynamically updating the causal chain and risk transmission path. The text content is identified through the BERT model to identify sensitive information and integrate knowledge graph verification; key frames are extracted from video / audio through FFmpeg, and illegal operations are identified through YOLOv5 behavior detection and OpenCV feature matching.

[0056] Furthermore, the laboratory filing and review guidance function of the application module automatically matches standard specifications, management system templates and necessary equipment lists according to the laboratory type (biology / chemistry / radiation) to generate a compliance checklist.

[0057] Furthermore, the user interface sets hierarchical permissions:

[0058] Ordinary user terminals only support file viewing / downloading and knowledge graph query;

[0059] The operational user interface supports file uploads and updates, triggering the AI model to automatically generate entity-relationship data and synchronously update the graph. The application module's laboratory filing and review guidance function allows real-time queries of laboratory type (biology / chemistry / radiation) nodes based on the knowledge graph. The rules engine matches associated standards, management system templates, and equipment lists to generate a dynamic compliance checklist. The user interface utilizes an RBAC permissions model, limiting ordinary users to querying and downloading only. After the operational user uploads data, the AI model generates entity-relationship triplets. The graph is incrementally updated via the Neo4j API, ensuring real-time data and traceability.

[0060] Furthermore, the data processing module uses dynamic entity relationship weight calculation. For duplicate uploads, it prioritizes retaining data with high timestamps or high-confidence versions based on Jaccard similarity and Siamese network double verification. The data processing module uses a dynamic entity relationship weight calculation strategy to perform double verification on duplicate uploads: initially screening text similarity based on Jaccard similarity (threshold > 0.85), and then confirming duplication by extracting entity semantic vectors (cosine similarity > 0.9) through the Siamese network. Conflicting data is automatically retained based on version priority rules (timestamp in reverse order + model confidence sorting), and the knowledge graph node weights are updated, and data lineage relationships are simultaneously recorded to ensure traceability.

[0061] Furthermore, knowledge graph query supports multi-hop reasoning. It inputs natural language descriptions and returns the associated original text of regulations, equipment operation videos and emergency plan cases. The user inputs natural language, the BERT model parses the key entities, traverses the related nodes within 3 hops of the knowledge graph, and returns the associated original text of regulations, equipment operation video links and emergency plan case summaries.

[0062] Furthermore, the user-side interface provides a visual interactive interface, integrating a force-directed graph to display the laboratory-risk source-control measure association network, and supports clicking on nodes to directly download PDF / video attachments or jump to the original regulatory terms. The user-side interface constructs a three-dimensional laboratory-risk source-control measure association network based on the force-directed graph algorithm. Clicking on a node can trigger the download of an attachment (PDF / video) or jump to the regulatory terms, and supports zooming and highlighting paths to track risk transmission links.

[0063] Working principle: By constructing a multimodal database (the file library presets structured fields according to laboratory types such as biology / chemistry, and is compatible with national standards, local documents and internal records; the catalog library standardizes MSDS parameters for hazardous chemicals and pathogenic microorganism operation levels), it supports multi-format storage and dynamic association such as PDF / video; combined with AI knowledge graph technology (BERT+graph neural network), it parses text and video semantics and automatically associates cross-source data (such as local regulations and national standard clauses) to generate a unified semantic index; through a cross-modal cleaning process (BERT verifies text contradictions, OpenCV extracts video key frames and aligns them with text time sequence), it solves the problem of unstructured data integration; using the Jaccard+Siamese network dual verification mechanism, it dynamically selects high-efficiency, high-confidence data versions and updates the graph node weights, realizing multi-source data association retrieval in seconds, breaking through the rigid bottleneck of traditional database formats, and improving safety management accuracy and response efficiency.

[0064] Automated cleaning is achieved through multimodal feature fusion and dynamic weight calculation. Semantic features are extracted from text data using BERT+TF-IDF, and combined with knowledge graph entity relationship verification (e.g., verifying whether "ethanol storage conditions" meet regulatory requirements) to filter out erroneous or contradictory content. Video / audio data is segmented into key frames using FFmpeg, and violation scenarios (e.g., not wearing protective equipment) are identified using OpenCV and YOLOv5, and then associated with text descriptions. For duplicate data, after initial Jaccard similarity screening, the Siamese network calculates semantic vector similarity, automatically retaining the optimal version based on timestamp and confidence priority, reducing manual intervention.

[0065] A systematic knowledge service is built based on multi-hop reasoning and intelligent retrieval of the knowledge graph. When a user enters natural language (e.g., "BSL-2 laboratory processes Ebola virus inactivated materials"), the system uses BERT to parse key entities, traverses the associated nodes within three hops of the knowledge graph, and aggregates the original text of regulations, equipment operation videos, and emergency response case studies to produce a one-stop query result. The application module's filing and review function automatically associates a list of required documents based on the laboratory type code (e.g., radiation laboratories must match the "Radioactive Material Operation Guide"), ensuring the integrity of knowledge acquisition and its adaptability to specific scenarios.

[0066] Relationship visualization is achieved through dynamic force-directed graphs and interactive node operations. The user-side interface builds a three-dimensional association network of laboratory-risk source-control measures based on D3.js. The node size indicates the risk level, and the edge color distinguishes the relationship type (red for risk transmission, blue for compliance requirements). Clicking on the risk source node allows drilling down to view the associated MSDS documents, leakage handling videos, and historical accident chains. Path highlighting and zoom focusing are supported, and visualized data is synchronized with the knowledge graph in real time to ensure the intuitive presentation and dynamic update of complex relationships.

[0067] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A method for establishing a laboratory safety knowledge base, characterized by: It includes document type database, directory type database, entity and entity relationship database, data processing module, application module and user terminal interface; The file type database is used to store standards and specifications, management methods, rules and regulations, operating procedures, case studies, equipment and facility documents, and official directory documents. It supports PDF / Word / Excel / video / audio multimodal data, and each file type is set with structured fields. The list type database includes lists of hazardous chemicals, pathogenic microorganisms, radioactive substances, precursor chemicals, explosives, psychotropic substances, and narcotic drugs, etc. Each list is associated with an official document source and defines an entity relationship. The entity and entity relationship library automatically generates entity association relationships based on the AI big model to build a knowledge graph; The data processing module integrates BERT+TF-IDF text analysis, FFmpeg frame hashing+OpenCV multimedia feature extraction technology to perform data cleaning, illegal content judgment and entity disambiguation operations; The application module supports laboratory filing review guidance, intelligent risk source identification, knowledge graph query, and file viewing and downloading functions; The client interface supports multiple clients to upload / update data and triggers automatic prompts to generate entity relationship data. Each type of document structured field in the document type database includes a document number, an effective date, a revision, and an associated laboratory type code; The illegal content judgment function of the data processing module is realized through a pre-trained classification model, and the model integrates the knowledge graph entity relationship to perform multi-dimensional verification; The AI model of the entity and entity relationship library adopts a graph neural network architecture to dynamically update the causal chain and risk transmission path between entities; The risk source intelligent identification function of the application module is configured with a threshold trigger mechanism, which generates an early warning map when it is detected that the risk correlation of multi-source heterogeneous data exceeds the set threshold.

2. The method for establishing a laboratory safety knowledge base according to claim 1, wherein: The equipment and facility files and case analysis files in the file type database support the storage of video / audio attachments, and realize the association index of multimedia content and text description through VIT feature extraction technology.

3. The method for establishing a laboratory safety knowledge base according to claim 1, wherein: The hazardous chemicals list field of the list type database includes 16 structured parameters of the MSDS document, and the pathogenic microorganism list field includes the virus culture laboratory level and transportation packaging requirements, and is automatically linked to the original clauses of the corresponding regulatory documents.

4. The method for establishing a laboratory safety knowledge base according to claim 3, wherein: The entity and relationship library automatically prompts to generate entity association relationships across data sources through the AI big model, including three types of relationship networks: "file-directory entity association", "file-file entity association", and "directory-risk source hierarchical relationship".

5. The method for establishing a laboratory safety knowledge base according to claim 4, characterized in that: The data processing module adopts a multimodal violation detection strategy: The text content is judged as sensitive information through the BERT violation detection model; The video / audio content is processed by FFmpeg to extract key frames, and OpenCV feature matching is used to identify illegal operation scenarios.

6. A method for establishing a laboratory safety knowledge base according to claim 5, characterized in that: The laboratory filing and review guidance function of the application module automatically matches standard specifications, management system templates and necessary equipment lists according to the laboratory type (biology / chemistry / radiation) to generate a compliance checklist.

7. The method for establishing a laboratory safety knowledge base according to claim 6, characterized in that: The user interface sets hierarchical permissions: Ordinary user terminals only support file viewing / downloading and knowledge graph query; The operational user end supports file upload / update, triggers the AI big model to automatically generate entity relationship data, and synchronously updates the graph.

8. The method for establishing a laboratory safety knowledge base according to claim 1, characterized in that: The data processing module adopts dynamic entity relationship weight calculation, and for repeatedly uploaded content, prioritizes retaining data with high timestamps or high confidence versions based on Jaccard similarity and Siamese network dual verification.

9. The method for establishing a laboratory safety knowledge base according to claim 2, wherein: The knowledge graph query supports multi-hop reasoning, inputs a natural language description, and returns the associated original text of regulations, equipment operation videos, and emergency plan cases.

10. The method for establishing a laboratory safety knowledge base according to claim 9, characterized in that: The user-end interface provides a visual interactive interface, integrates a force-directed graph to display the laboratory-risk source-control measure association network, and supports clicking on nodes to directly download PDF / video attachments or jump to the original text of the regulations.

Images