A laboratory intelligent management system based on an active virtual digital human

Abstract

The application relates to the technical field of intelligent management, in particular to a laboratory intelligent management system based on an active virtual digital person, which comprises a server; the server comprises a perception layer, a digital twin layer, a cognition layer and an application layer, and forms an intelligent management system of perception, twin model, intelligent decision and execution; wherein the perception layer collects laboratory environment, equipment, personnel and access control state data through Internet of Things sensors and other equipment; the digital twin layer constructs and updates a virtual twin model of a physical laboratory in real time, and realizes two-way data binding; the cognition layer comprises a multi-agent module, and intelligently decides in combination with a laboratory safety knowledge base and a large language model; the virtual digital person of the application layer serves as an explicit manager, autonomously executes active inspection, risk early warning and emergency response tasks, and communicates with users through multi-modal interaction.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology, specifically to a laboratory intelligent management system based on an active virtual digital human. Background Technology

[0002] University laboratories are important venues for teaching and research activities. Their management is complex and cumbersome, involving equipment maintenance, consumable management, safety monitoring, personnel access control, and operational procedures. Therefore, current laboratory management faces the following challenges: 1. Passive management model, reliant on manpower: Traditional laboratory management heavily relies on the responsibility and experience of management personnel. Inspections, recording, and supervision consume significant manpower and are difficult to maintain uninterrupted, 24 / 7 coverage. Management actions are mostly reactive, lacking foresight regarding potential risks; 2. Isolated information systems, ineffective data utilization: Although many laboratories have introduced various information technology systems… Systems (such as equipment management systems and access control systems) exist, but these systems are often independent "data silos," failing to achieve interconnection and deep integration analysis of information. In addition, a large amount of sensor data is only used for simple threshold alarms, and its deeper value has not been explored. 3. There are blind spots in safety supervision and training methods are monotonous: Laboratory safety is of paramount importance, but traditional monitoring methods (such as cameras) require real-time manual viewing, which is inefficient and prone to omissions. There is a delay in response to violations or dangerous situations. At the same time, safety training is mostly based on lectures or written materials, which is monotonous and makes it difficult for students to obtain an immersive and contextualized learning experience.

[0003] In recent years, some new technologies have been attempted to be applied to laboratory management. For example, IoT-based monitoring systems can remotely monitor environmental parameters (such as temperature, humidity, and concentration of harmful gases), but they usually only reach the level of data collection and simple alarms, lacking intelligent decision-making capabilities. Digital twin technology has also been introduced to visualize monitoring by building virtual models of the laboratory, but in most applications, the twin model mainly serves as a passive mapping of the physical world, and its predictive and simulation capabilities have not been fully utilized in proactive management. In addition, although virtual digital human technology has been developed, in existing applications, virtual digital humans are mostly used as passive question-and-answer assistants or virtual customer service representatives, and their potential to play an active and core role in professional management scenarios has not yet been explored.

[0004] In summary, there is an urgent need for a technical solution that can organically combine the intuitive interaction of virtual digital humans, the global insight of digital twins, and the real-time sensing capabilities of the Internet of Things to achieve proactive, forward-looking, efficient, and intelligent management of laboratories. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a laboratory intelligent management system based on an active virtual digital human. By shaping the virtual digital human into a routine and intuitively interactive manager in the laboratory, it achieves comprehensive, proactive, and forward-looking intelligent management of the laboratory.

[0006] The basic solution provided by this invention is a laboratory intelligent management system based on an active virtual digital human, comprising a server, wherein the server includes: The perception layer is used to collect laboratory environmental data, equipment status data, personnel activity data, and access control status data in real time through IoT sensors, visual acquisition devices, and RFID readers deployed in the laboratory setting. The digital twin layer is used to construct and update a digital twin model in real time based on the static three-dimensional model of the physical laboratory and the laboratory environment data, equipment status data, personnel activity data and access control status data collected by the perception layer. It achieves two-way data binding between the physical and virtual worlds through a unified data model and synchronization protocol. The cognitive layer, including the multi-agent module, is used to receive and analyze laboratory environment data, equipment status data, personnel activity data, access control status data, and the status of the digital twin layer model constructed by the digital twin layer collected by the perception layer, to make intelligent decisions and generate corresponding management instructions. The application layer includes virtual digital humans playing an explicit management role, driven by a multi-agent module. These are used to visualize the laboratory status, issue management instructions, and interact with users in a multimodal manner, autonomously performing tasks such as proactive inspections, risk warnings, and emergency response management.

[0007] The principle of this invention lies in achieving proactive laboratory management through the construction of an intelligent management system. The perception layer, acting as the data acquisition entry point, comprehensively captures real-time status data of the laboratory environment, equipment, personnel, and access control through various acquisition devices deployed in key locations within the laboratory, providing complete data support for subsequent management. The digital twin layer combines the static structure of the physical laboratory with the dynamic data from the perception layer to construct a digital twin model, achieving a precise mapping between the physical and virtual worlds and providing a visual platform for global monitoring and decision analysis. The multi-agent module of the cognition layer, acting as the "brain" of the system's core analysis and reasoning, comprehensively analyzes the collected multi-source data and the twin model's status, completing risk assessment and task planning through intelligent reasoning, and generating targeted management instructions. The virtual digital human of the application layer, acting as an explicit management platform, transforms abstract management instructions into visualized and interactive management behaviors, autonomously executing tasks such as inspections and early warnings, while simultaneously achieving efficient communication with users through multimodal interaction.

[0008] The beneficial effects of this invention are as follows: Through real-time data collection at the perception layer and intelligent decision-making at the cognition layer, the system can proactively identify risks and plan management tasks, transforming the management model from post-event response to pre-event prevention. It eliminates the need for continuous manual monitoring, achieving 24 / 7 uninterrupted proactive management. The digital twin layer, as the core of data integration, links and binds dispersed environmental, equipment, and personnel data, breaking down "data silos" and providing a complete data foundation for in-depth analysis and intelligent decision-making, fully unlocking data value. The perception layer comprehensively covers key laboratory data collection, the digital twin layer achieves global visual monitoring, and the virtual digital human proactively performs inspection tasks. The three work together to eliminate omissions and delays in manual monitoring, improving the comprehensiveness and timeliness of safety supervision. The virtual digital human, as an explicit interactive carrier, changes the traditional single-text or lecture training model, providing users with contextualized guidance through intuitive visualization and multimodal interaction, improving operational standardization and training effectiveness.

[0009] Furthermore, the unified data model includes an identification field, a time field, a data type field, and a business data field. The identification field integrates a unique device identifier, and the business data field contains business data key-value pairs. The unique device identifier adopts a hierarchical naming rule of "region-category-number". The synchronization protocol adopts the MQTT protocol. Through the unified data model and the MQTT protocol, real-time data interaction between the physical entity and the digital twin model is realized, and two-way data binding is completed.

[0010] A unified data model clarifies the structural specifications of various types of collected data, ensuring that the data formats of different types of sensors and devices are consistent, reducing the difficulty of data integration and analysis, and improving data processing efficiency. Among them, the hierarchical naming rules give each collected object (device, sensor) a unique identifier, which facilitates quick location of data source and enables accurate tracking and management of device status. As a lightweight IoT communication protocol, the MQTT protocol is suitable for low-latency transmission of multi-source data. Its synchronization mechanism not only ensures the stable synchronization of regular data, but also meets the real-time data requirements of management scenarios, ensuring the synchronization consistency between the physical and virtual worlds.

[0011] Furthermore, the multi-agent module includes: The sensing agent is used to clean and convert the laboratory environment data, equipment status data, personnel activity data, and access control status data collected by the sensing layer. The decision-making intelligent agent, with a built-in DeepSeek series of large-scale language models, combined with retrieval enhancement generation technology and a laboratory safety knowledge base, is used for logical reasoning, risk assessment, task planning and simulation, and generates management instructions. An execution agent is used to parse the management instructions generated by the decision-making agent into equipment control protocols and schedule them to the intelligent actuators in the laboratory. Interactive intelligent agents are used to drive the rendering of virtual digital human images, the generation of actions, the synthesis and recognition of speech, process multimodal interactive information, extract user intent, and transmit it to the decision-making intelligent agent.

[0012] The perceptual agent is responsible for data preprocessing, forming data in a unified format for subsequent analysis and processing. The decision-making agent integrates a professional large-scale language model and a laboratory safety knowledge base, combined with retrieval-enhanced generation technology, to accurately integrate professional knowledge with real-time data, enabling logical reasoning and risk assessment in complex scenarios, and ensuring the scientific nature and relevance of management instructions. The execution agent realizes the precise conversion from abstract decisions to equipment control, ensuring the effective implementation of management instructions. The interactive agent drives virtual digital humans to achieve natural and smooth multimodal interaction, reducing the user's operating threshold and improving the efficiency of management instruction transmission.

[0013] Furthermore, computational fluid dynamics is employed in the simulation and deduction of the decision-making agent. - A turbulence model is used to simulate gas diffusion, a CFAST regional model is used to simulate fire spread, and a multi-agent model based on social forces is used to simulate personnel evacuation and predict risk evolution trends.

[0014] For the three core risk scenarios commonly encountered in laboratories—gas leaks, fires, and personnel evacuation—we employ industry-mature specialized simulation algorithms to ensure the accuracy of risk evolution trend predictions, providing a reliable basis for emergency decision-making. Through simulation and deduction, we can predict the risk diffusion path, impact range, and response challenges in advance, enabling the decision-making agent to generate forward-looking management strategies, avoiding passive responses, and minimizing risk losses.

[0015] Furthermore, the laboratory environment data collected by the sensing layer includes temperature, relative humidity, pressure difference, carbon dioxide concentration, and oxygen concentration. The equipment status data includes the airflow speed of the biosafety cabinet, the internal temperature of the ultra-low temperature freezer, the pressure of the autoclave cavity, and the vibration parameters of the centrifuge. Each parameter is set with a warning threshold and an alarm threshold.

[0016] The collection of these data covers key indicators of the laboratory environment and the operating status of core equipment, ensuring the comprehensiveness of data collection and providing complete data support for safety monitoring and equipment management. By setting early warning thresholds and alarm thresholds, risk-based responses can be achieved, enabling early warning of potential risks and timely triggering of emergency measures when risks escalate, preventing the risks from spreading and improving the accuracy and effectiveness of safety monitoring. By collecting key operating parameters and setting thresholds for core experimental equipment, abnormal equipment operation can be detected in a timely manner, providing data basis for equipment maintenance and fault prevention.

[0017] Furthermore, the virtual digital human's proactive inspection tasks are executed based on dynamic path rules: the shortest basic path covering key areas is preset and executed according to a preset cycle; the inspection frequency is adjusted according to the equipment risk level, historical failure rate and current usage status; when abnormal data collected by the perception layer or abnormal access control status occurs, a temporary inspection path directly to the target area is generated.

[0018] Furthermore, the cognitive layer adopts a three-dimensional anomaly judgment standard in the intelligent decision-making process: after a single data exceeds the standard, the status is continuously observed for a preset time, and if it does not recover, an early warning command is sent to the application layer; hidden anomalies are identified through multi-dimensional sensor data correlation analysis; and personnel behavior sequences are analyzed through visual recognition and location tracking to determine whether they comply with standard operating procedures.

[0019] The basic inspection path covers key areas of the laboratory, ensuring that routine inspections do not miss core management points and providing a fundamental guarantee for laboratory safety. The inspection frequency is dynamically adjusted according to the equipment risk level, failure rate, and usage status, concentrating inspection resources on high-risk and high-demand areas, improving inspection efficiency and avoiding resource waste. The event-triggered temporary inspection path can quickly focus on the target area when an anomaly occurs, enabling rapid verification and handling of anomalies, shortening response time, and reducing the impact of anomalies.

[0020] Furthermore, the multimodal interaction methods of virtual digital humans include: responding to voice or text commands through natural language processing technology; recognizing user identity, gestures, and body posture through computer vision technology; providing immersive operation guidance and training in conjunction with AR or VR devices; and providing personalized services in conjunction with user identity, location, and current task during the interaction process.

[0021] It supports multiple interaction methods such as voice, text, gestures, and body language, adapting to different usage scenarios and user habits, and lowering the barrier to interaction between users and the system; it provides customized services based on user identity, location, and current task, such as providing targeted operation guidance for experimenters and global management data for administrators, improving the accuracy and practicality of services; AR / VR immersive guidance can simulate real experimental scenarios, allowing users to have a contextualized learning experience, deepening their understanding of operating procedures and safety knowledge, and improving training quality and operational proficiency. Attached Figure Description

[0022] Figure 1 This is a system block diagram of an embodiment of an intelligent laboratory management system based on an active virtual digital human according to the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the collaboration between the multi-agent module and external entities in an embodiment of the present invention. Detailed Implementation

[0024] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A laboratory intelligent management system based on an active virtual digital human, including a server, the server comprising: The perception layer is used to collect laboratory environmental data, equipment status data, personnel activity data, and access control status data in real time through IoT sensors, visual acquisition devices, and RFID readers deployed in the laboratory setting. The digital twin layer is used to construct and update a digital twin model in real time based on the static three-dimensional model of the physical laboratory and the laboratory environment data, equipment status data, personnel activity data and access control status data collected by the perception layer. It achieves two-way data binding between the physical and virtual worlds through a unified data model and synchronization protocol. The cognitive layer, including the multi-agent module, is used to receive and analyze laboratory environment data, equipment status data, personnel activity data, access control status data, and the status of the digital twin layer model constructed by the digital twin layer collected by the perception layer, to make intelligent decisions and generate corresponding management instructions. The application layer includes virtual digital humans playing an explicit management role, driven by a multi-agent module. These are used to visualize the laboratory status, issue management instructions, and interact with users in a multimodal manner, autonomously performing tasks such as proactive inspections, risk warnings, and emergency response management.

[0025] In this embodiment, the server adopts a high-performance industrial server, which has the capabilities of data storage, parallel computing and multi-module collaborative scheduling. It can meet the computing power requirements of real-time processing of multi-source data, dynamic updating of digital twin models and intelligent decision-making, and provide hardware support for the stable operation of the entire system.

[0026] Furthermore, the unified data model includes an identifier field, a time field, a data type field, and a business data field. The identifier field integrates a unique device identifier, and the business data field contains business data key-value pairs. The unique device identifier adopts a hierarchical naming rule of "region-category-number". The synchronization protocol adopts the MQTT protocol. Through the unified data model and the MQTT protocol, real-time data interaction between the physical entity and the digital twin model is realized, and two-way data binding is completed. The unified data model is defined using JSON Schema, including two parts: a header and a payload. The header integrates the unique device identifier, timestamp, and data type, and the payload contains business data key-value pairs.

[0027] In this embodiment, a specific example of the device's unique identifier is "P2-ENV-TEMP-01", which represents temperature sensor number 1 in the P2 experimental area's environmental category, and "P2-BSC-02" represents biosafety cabinet number 2 in the P2 experimental area. The preset data synchronization period is set to 10 seconds to ensure timely data updates while avoiding excessive transmission pressure. The MQTT protocol's Topic path corresponds to the device's unique identifier, with the format "lab_id / asset_id / data". For example, the status data of biosafety cabinet number 2 in the P2 experimental area is published through the path "lab01 / P2-BSC-02 / data", which facilitates accurate subscription and parsing by the sensing agent.

[0028] Furthermore, the multi-agent module includes: The sensing agent is used to clean and convert the laboratory environment data, equipment status data, personnel activity data, and access control status data collected by the sensing layer. The decision-making intelligent agent, with a built-in DeepSeek series of large-scale language models, combined with retrieval enhancement generation technology and a laboratory safety knowledge base, is used for logical reasoning, risk assessment, task planning and simulation, and generates management instructions. An execution agent is used to parse the management instructions generated by the decision-making agent into equipment control protocols and schedule them to the intelligent actuators in the laboratory. Interactive intelligent agents are used to drive the rendering of virtual digital human images, the generation of actions, the synthesis and recognition of speech, process multimodal interactive information, extract user intent, and transmit it to the decision-making intelligent agent.

[0029] In this embodiment, as shown in the appendix Figure 2 As shown, after receiving laboratory environment data, equipment status data, personnel activity data and access control status data transmitted from the perception layer, the sensing agent first performs data cleaning (removing invalid and duplicate data) and format conversion (unifying to JSON format), and then extracts abnormal data features through simple pattern recognition algorithms (such as threshold preliminary screening and data fluctuation identification). The preprocessed data is then transmitted to the decision-making agent. The decision-making agent incorporates the DeepSeek-V3 general logic processing engine, the DeepSeek-VL2 visual language model, and the DeepSeek-R1 reinforcement learning inference model. The laboratory safety knowledge base includes the "General Requirements for Laboratory Biosafety," equipment operation manuals (SOPs), chemical safety data sheets (MSDS), and emergency plans. During decision-making, it extracts knowledge fragments relevant to the current scenario through retrieval enhancement generation technology and performs logical reasoning by combining preprocessed data. The executing agent parses the management commands generated by the decision-making agent, such as "start forced ventilation" and "cut off power", into control protocols such as Modbus and TCP / IP that can be recognized by the laboratory intelligent actuators, and sends them to the corresponding actuators through the server interface; The interactive intelligent agent integrates text-to-speech (TTS), automatic speech recognition (ASR), and image rendering engines to drive a virtual digital human to present a professional butler image. It can process users' voice commands, text queries, and gestures in real time, and convert users' intentions into data signals to transmit to the decision-making intelligent agent.

[0030] Furthermore, computational fluid dynamics is employed in the simulation and deduction of the decision-making agent. - A turbulence model is used to simulate gas diffusion, a CFAST regional model is used to simulate fire spread, and a multi-agent model based on social forces is used to simulate personnel evacuation and predict risk evolution trends.

[0031] In this embodiment, when the sensing layer detects that the concentration of toxic gas exceeds the standard, the decision-making agent calls... - The turbulence model, combined with the laboratory space structure in the digital twin model (such as the location of doors and windows, and the distribution of ventilation openings), can simulate the gas diffusion path, concentration distribution, and time to reach key areas within 30 seconds, providing a basis for evacuation route planning. When a fire alarm is triggered, the CFAST regional model divides the laboratory into an upper hot smoke layer and a lower cold air layer. By solving the mass and energy conservation equations, it quickly predicts the fire spread rate, smoke layer height and temperature changes, providing decision support for fire-fighting resource allocation and personnel evacuation. In personnel evacuation simulation, each experimenter is simulated as an independent intelligent agent whose behavior is driven by the goal of "reaching the exit", the environmental perception of "avoiding the fire source" and the interpersonal interaction of "avoiding collision". It can realistically simulate the flow of people and evacuation bottlenecks in emergency situations and optimize the design of evacuation routes.

[0032] Furthermore, the laboratory environment data collected by the sensing layer includes temperature, relative humidity, pressure difference, carbon dioxide concentration, and oxygen concentration. The equipment status data includes the airflow speed of the biosafety cabinet, the internal temperature of the ultra-low temperature freezer, the pressure of the autoclave cavity, and the vibration parameters of the centrifuge. Each parameter is set with a warning threshold and an alarm threshold.

[0033] In this embodiment, the deployment scheme of the perception layer is as follows: Environmental data acquisition: Temperature and humidity sensors, differential pressure sensors, carbon dioxide sensors, and oxygen sensors are evenly deployed throughout the laboratory. Differential pressure sensors are primarily deployed at the connection points between the core work area and the outside environment to ensure negative pressure monitoring. Equipment status data acquisition: Airflow velocity sensors are installed at the air outlets of biosafety cabinets, temperature sensors are embedded inside ultra-low temperature freezers, pressure sensors are installed in the autoclave cavity, and vibration sensors are installed on the centrifuge base to capture key equipment operating parameters in real time. Threshold settings are based on national standards and equipment specifications. For example, the temperature warning threshold is below 18℃ or above 26℃, and the alarm threshold is below 16℃ or above 28℃. The biosafety cabinet inflow airflow velocity warning threshold is below 0.5m / s, and the alarm threshold is below 0.4m / s, ensuring the scientific validity and safety of the threshold settings.

[0034] Furthermore, the virtual digital human's proactive inspection tasks are executed based on dynamic path rules: the shortest basic path covering key areas is preset and executed according to a preset cycle; the inspection frequency is adjusted according to the equipment risk level, historical failure rate and current usage status; when abnormal data collected by the perception layer or abnormal access control status occurs, a temporary inspection path directly to the target area is generated.

[0035] In this embodiment, the basic path is generated by a path planning algorithm, covering key areas such as hazardous chemical cabinets, biosafety cabinets, high-pressure equipment areas, and fire exits. The preset cycle is a routine inspection every 2 hours. The inspection frequency adjustment rules are as follows: for high-risk equipment areas such as operating autoclaves and centrifuges, the inspection frequency is increased to 15 minutes / time; for equipment with a historical failure rate of more than 30%, the inspection frequency is increased to 30 minutes / time; for ordinary equipment in standby mode, the inspection frequency is reduced to 4 hours / time. When the sensing layer detects that the temperature sensor data in a certain area continues to exceed the standard, or the access control is abnormally opened during non-working hours, the system immediately generates a temporary inspection path. The virtual digital human in the digital twin model directly reaches the area, retrieves real-time monitoring images and relevant sensor data, and performs anomaly verification.

[0036] Furthermore, the cognitive layer adopts a three-dimensional anomaly judgment standard in the intelligent decision-making process: after a single data exceeds the standard, the status is continuously observed for a preset time, and if it does not recover, an early warning command is sent to the application layer; hidden anomalies are identified through multi-dimensional sensor data correlation analysis; and personnel behavior sequences are analyzed through visual recognition and location tracking to determine whether they comply with standard operating procedures.

[0037] In this embodiment, the preset duration for continuous state observation is 60 seconds. For example, when the refrigerator door is opened, causing a brief temperature rise, the system starts a 60-second timer. If the temperature does not drop after 60 seconds, it is determined to be an early warning event. Example of multidimensional data correlation analysis: The vision system detects personnel activity in the laboratory, but the carbon dioxide concentration does not rise significantly within 10 minutes. The system determines that the air quality sensor may be faulty and triggers a sensor verification command. Behavioral sequence analysis tracks personnel actions through visual acquisition devices. For example, if the system detects the behavioral sequence of "opening the ultra-low temperature refrigerator, taking out the bacterial strain, not closing the refrigerator door, and walking towards the clean bench", if the "not closing the refrigerator door" state lasts for more than 30 seconds, it is determined to be an unsafe behavior, and an early warning command is sent to the application layer.

[0038] Furthermore, the multimodal interaction methods of virtual digital humans include: responding to voice or text commands through natural language processing technology; recognizing user identity, gestures, and body posture through computer vision technology; providing immersive operation guidance and training in conjunction with AR or VR devices; and providing personalized services in conjunction with user identity, location, and current task during the interaction process.

[0039] In this embodiment, natural language interaction allows users to ask the virtual android questions such as "current temperature and humidity of the P2 laboratory" and "operating status of the biosafety cabinet" via the laboratory's large screen or mobile app. The virtual android responds in real time using speech synthesis technology and displays relevant data on the interface. Visual recognition interaction allows the virtual android to automatically match the user's experimental task with the user's experimental task and push corresponding operational guidelines. Users can trigger the virtual android to display inspection reports, operation instructions, and other content by using preset gestures such as "waving" or "nodding." AR / VR immersive interaction allows new employees to see operation steps overlaid by the virtual android next to real experimental equipment while wearing AR glasses. For example, the operation process of an autoclave is presented in the form of a 3D animation. During emergency response training, VR equipment simulates chemical spill scenarios, and the virtual android guides users to complete the correct evacuation and disposal operations.

[0040] The following example illustrates the workflow of an emergency response to a hazardous chemical leak: Data Acquisition: Gas sensors deployed near the chemical storage tank detect excessive concentrations of toxic gases and immediately upload the data to the server's perception layer via the MQTT protocol; Model Update: After receiving the data, the digital twin layer renders the area near the leak point as a red highlight in the 3D model, simulating gas diffusion trends with particle effects, and synchronously updates all associated terminals; Intelligent Decision-Making: The perception agent preprocesses the gas concentration data, identifies it as a high-risk leak event, and transmits it to the decision-making agent; The decision-making agent calls the gas diffusion simulation algorithm, combines it with the Material Safety Data Sheet (MSDS) in the knowledge base, and generates emergency decision instructions: activate the forced ventilation system, switch... The system cuts off power to non-explosion-proof equipment in the leak area, triggers audible and visual alarms, and plans evacuation routes. Equipment linkage: The executing intelligent agent parses decision commands into control protocols and sends them to actuators such as the laboratory's exhaust controllers and smart sockets, causing physical devices to respond and execute immediately. Interactive early warning: The interactive intelligent agent drives a virtual digital human to pop up on all laboratory displays and user mobile apps, broadcasting in voice and text: "Warning! A hazardous gas leak has occurred near medicine cabinet number three. Please evacuate immediately along the green route in an orderly manner. Do not use the elevator!" It also displays the optimal evacuation route. Emergency response: After rescue personnel wearing AR glasses arrive at the scene, the virtual digital human guides them to the precise location of the leak point within the AR field of view, providing real-time updates on chemical handling procedures and guiding rescue personnel in safe handling.

[0041] In this embodiment, the significance of the virtual digital human lies in achieving a key upgrade from "passive data monitoring" to "proactive anthropomorphic management." First, the dynamic path planning of proactive inspection routes transforms the management logic of prioritizing key areas and managing abnormal areas immediately into visualized inspection behavior. Second, it realizes a shift from "people searching for data" to "system finding people," eliminating the need for manual screening of anomalies from massive amounts of data. The system automatically identifies problems through proactive inspection and anomaly detection, accurately pushing the results to relevant personnel. Finally, the virtual digital human can not only autonomously perform management tasks such as inspection, early warning, and emergency dispatch, but also transform professional decisions into intuitive guidance through multimodal interaction, reducing information interpretation costs and improving the efficiency of management command execution.

[0042] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A laboratory intelligent management system based on an active virtual digital human, characterized in that, Includes a server, the server comprising: The perception layer is used to collect laboratory environmental data, equipment status data, personnel activity data, and access control status data in real time through IoT sensors, visual acquisition devices, and RFID readers deployed in the laboratory setting. The digital twin layer is used to construct and update a digital twin model in real time based on the static three-dimensional model of the physical laboratory and the laboratory environment data, equipment status data, personnel activity data and access control status data collected by the perception layer. It achieves two-way data binding between the physical and virtual worlds through a unified data model and synchronization protocol. The cognitive layer, including the multi-agent module, is used to receive and analyze laboratory environment data, equipment status data, personnel activity data, access control status data, and the status of the digital twin layer model constructed by the digital twin layer collected by the perception layer, to make intelligent decisions and generate corresponding management instructions. The application layer includes virtual digital humans playing an explicit management role, driven by a multi-agent module. These are used to visualize the laboratory status, issue management instructions, and interact with users in a multimodal manner, autonomously performing tasks such as proactive inspections, risk warnings, and emergency response management.

2. The laboratory intelligent management system based on an active virtual digital human as described in claim 1, characterized in that, The unified data model includes an identification field, a time field, a data type field, and a business data field. The identification field integrates a unique device identifier, and the business data field contains business data key-value pairs. The unique device identifier adopts a hierarchical naming rule of "region-category-number". The synchronization protocol adopts the MQTT protocol. Through the unified data model and the MQTT protocol, real-time data interaction between the physical entity and the digital twin model is realized, and two-way data binding is completed.

3. The laboratory intelligent management system based on an active virtual digital human as described in claim 1, characterized in that, The multi-agent module includes: The sensing agent is used to clean and convert the laboratory environment data, equipment status data, personnel activity data, and access control status data collected by the sensing layer. The decision-making intelligent agent, with a built-in DeepSeek series of large-scale language models, combined with retrieval enhancement generation technology and a laboratory safety knowledge base, is used for logical reasoning, risk assessment, task planning and simulation, and generates management instructions. An execution agent is used to parse the management instructions generated by the decision-making agent into equipment control protocols and schedule them to the intelligent actuators in the laboratory. Interactive intelligent agents are used to drive the rendering of virtual digital human images, the generation of actions, the synthesis and recognition of speech, process multimodal interactive information, extract user intent, and transmit it to the decision-making intelligent agent.

4. The laboratory intelligent management system based on an active virtual digital human as described in claim 3, characterized in that, In the simulation and deduction of decision-making agents, computational fluid dynamics is employed. - A turbulence model is used to simulate gas diffusion, a CFAST regional model is used to simulate fire spread, and a multi-agent model based on social forces is used to simulate personnel evacuation and predict risk evolution trends.

5. A laboratory intelligent management system based on an active virtual digital human as described in claim 1, characterized in that, The laboratory environment data collected by the sensing layer includes temperature, relative humidity, pressure difference, carbon dioxide concentration, and oxygen concentration. The equipment status data includes the airflow velocity of the biosafety cabinet, the internal temperature of the ultra-low temperature freezer, the pressure of the autoclave cavity, and the vibration parameters of the centrifuge. Each parameter is set with a warning threshold and an alarm threshold.

6. The laboratory intelligent management system based on an active virtual digital human according to claim 1, characterized in that, The virtual digital human's proactive inspection tasks are executed based on dynamic path rules: the shortest basic path covering key areas is preset and executed according to a preset cycle; the inspection frequency is adjusted according to the equipment risk level, historical failure rate and current usage status. When abnormal data is collected by the perception layer or when access control status is abnormal, a temporary inspection path is generated that leads directly to the target area.

7. A laboratory intelligent management system based on an active virtual digital human as described in claim 1, characterized in that, The cognitive layer adopts a three-dimensional anomaly judgment standard in the intelligent decision-making process: after a single data exceeds the standard, the state is continuously observed for a preset time, and if it does not recover, an early warning instruction is sent to the application layer; Hidden anomalies can be identified through multi-dimensional sensor data correlation analysis; By analyzing personnel behavior sequences through visual recognition and location tracking, it can be determined whether they comply with standard operating procedures.

8. A laboratory intelligent management system based on an active virtual digital human according to claim 1, characterized in that, The multimodal interaction methods of virtual digital humans include: responding to voice or text commands through natural language processing technology; recognizing user identity, gestures, and body posture through computer vision technology; providing immersive operation guidance and training in conjunction with AR or VR devices; and providing personalized services by combining user identity, location, and current task during the interaction process.

Images