Laboratory digital twinning system and method based on unreal engine and self-evolution AI

Abstract

The invention discloses a laboratory digital twinning system and method based on an unreal engine and self-evolution AI. The system comprises a control system, a mechanical arm system, an intelligent logistics scheduling system, a digital twinning platform system and a man-machine interaction system. A local master control server of the control system is integrated with a self-evolution AI decision engine. According to the method, the self-evolution AI decision engine and the material knowledge base are integrated, so that the system can autonomously generate'component-structure-performance 'assumptions and create new material targets; the method comprises the following steps: integrating a physical engine in a high-fidelity digital twin constructed by an unreal engine to carry out ten-thousand-level simulation rehearsal, and establishing a causal intervention channel interacting with a physical entity in real time; distributed intention sharing nodes are established among digital twin bodies of multiple mechanical arms, and cooperative motion planning and collision detection are performed based on a virtual space, so that multi-arm cooperation is upgraded from simple time sequence cooperation to intelligent cooperation with prediction and collision avoidance capabilities.

Description

Technical Field

[0001] This invention relates to the field of intelligent chemical experiment automation technology, and more specifically, to a laboratory digital twin system and method based on Unreal Engine and self-evolving AI. Background Technology

[0002] With the development of intelligent manufacturing and laboratory automation technologies, the field of chemical experiments is gradually evolving from traditional manual operation to automation and intelligence. Currently, the technological paths in this field mainly focus on the following two aspects: Firstly, there are robot-based automated experimental systems. These systems utilize robotic arms to replace manual labor in repetitive operations such as reagent preparation, mixing, and transfer, thereby improving the accuracy and efficiency of experiments. For example, patent document CN117444948A discloses an automated material synthesis system based on artificial intelligence and robotics. This system can generate target materials that meet the set desired performance parameters through an AI model and control robots to perform synthesis experiments. However, the core paradigm of these systems remains a linear process of "human setting goals - machine performing operations." Their artificial intelligence modules mainly play the role of "searcher" or "optimizer," relying on clearly defined performance indicators preset by humans. They lack the ability to autonomously propose scientific hypotheses and create new target materials, failing to achieve the leap from "operational substitution" to "cognitive substitution." Furthermore, the collaboration between multiple robotic arms is mostly a sequential coordination of preset programs, lacking intelligent collaborative mechanisms based on dynamic environments and shared intentions.

[0003] Secondly, there is the application of digital twin technology in the visualization and monitoring of experimental processes. Digital twins, by constructing virtual mappings of physical entities, can be used to simulate and demonstrate experimental procedures. For example, patent document CN116945190A discloses a virtual synchronous control system for a robotic arm based on digital twins, achieving motion synchronization between the virtual and physical robotic arms. However, most existing digital twin systems focus on state monitoring and one-way visualization mapping, with insufficient coupling depth between the virtual model and the physical entity. Specifically, the virtual space mainly serves as a "mirror" of the physical world, lacking high-fidelity process simulation capabilities based on first principles or physics engines; more importantly, existing systems generally lack a real-time, closed-loop "reverse intervention" channel from the virtual space to the physical entity. In other words, when the virtual simulation predicts potential problems or optimization paths, the system cannot directly and automatically apply corrective strategies to the physical experimental equipment to achieve online dynamic adjustment of reaction conditions and operation sequences, thus failing to fully realize the "control" and "optimization" value of digital twins.

[0004] Furthermore, both automated robotic systems and existing digital twin solutions generally face the problem of a broken closed loop of experimental knowledge. The large amount of data generated during experiments is often merely recorded and archived, failing to be effectively fed back into the AI ​​model in real time for online learning and knowledge iteration. This prevents the system from continuously evolving from historical experiments, causing its decision-making capabilities and adaptability to stagnate.

[0005] To address the aforementioned issues, how to deeply integrate high-fidelity simulation, self-evolving AI decision-making, and real-time closed-loop control to achieve autonomous cognition and operation of the entire laboratory process has become a pressing technical problem to be solved in this field. Summary of the Invention

[0006] In view of the above-mentioned technical problems in related technologies, the present invention proposes a laboratory digital twin system and method based on Unreal Engine and self-evolving AI, which can overcome the above-mentioned shortcomings of the prior art.

[0007] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: A laboratory digital twin system based on Unreal Engine and self-evolving AI includes a control system, a robotic arm system, an intelligent logistics scheduling system, a digital twin platform system, and a human-computer interaction system. The control system includes a PLC controller as the core main control unit and a local central control server, which are used to coordinate the system operation logic, store data and process high-level logic operations. The local central control server integrates a self-evolving AI decision engine. The robotic arm system includes at least one collaborative robot, which is connected to the control system and the digital twin platform system via a first communication protocol; The intelligent logistics scheduling system is used to schedule automated transportation equipment for the transfer of experimental materials, and maintains real-time two-way communication with the control system and digital twin platform system through the second communication protocol. The digital twin platform system is built on Unreal Engine, forming a 1:1 high-fidelity virtual mapping with the physical experimental environment, and integrates a physics engine; the digital twin platform system interacts with the robotic arm system and the intelligent logistics scheduling system through the first communication protocol and the second communication protocol respectively, and has a causal intervention channel for real-time bidirectional data exchange with the control system. The human-machine interaction system is used for system monitoring and command input, and obtains underlying operating data from the PLC controller of the control system through a data acquisition protocol; The self-evolving AI decision engine actively generates experimental objectives and hypotheses based on a knowledge base, and performs virtual simulation verification in the digital twin platform system to generate the optimal experimental strategy. The digital twin platform system sends correction instructions to the control system through the causal intervention channel based on the differences between the virtual simulation and the physical entity operation, thereby intervening in the physical experiment process in reverse. The end-to-end data generated by the experiment is fed back to the self-evolving AI decision engine in real time for online incremental learning.

[0008] Furthermore, the self-evolving AI decision engine includes a generative algorithm module, which is used to proactively generate scientific hypotheses and formulation parameters about the "composition-structure-performance" of materials based on materials genome databases and literature knowledge, thereby achieving AI self-creation of experimental objectives.

[0009] Furthermore, the self-evolving AI decision engine also includes a model coupled with graph neural networks and deep reinforcement learning, used to deeply analyze the structure-property relationship of materials and dynamically optimize experimental strategies; the local central control server collects the underlying operating data of the PLC controller through the OPCUA protocol and feeds it back to the self-evolving AI decision model for online incremental learning.

[0010] Furthermore, after receiving the experimental strategy, the digital twin platform system uses a physics engine to perform high-speed simulation and logical verification of the robotic arm's motion trajectory and logistics transportation path in a virtual environment before issuing physical instructions, thereby selecting the optimal reaction path and operation sequence.

[0011] Furthermore, the causal intervention channel specifically refers to the following: when the physical sensor data collected through the human-computer interaction system does not match the virtual prediction result in the digital twin platform system, the digital twin platform system sends a correction command to the PLC controller of the control system through a real-time bidirectional communication interface to adjust the operating parameters of the robotic arm or related experimental equipment in real time.

[0012] Furthermore, the robotic arm system includes multiple collaborative robots; the digital twin platform system creates a digital twin for each collaborative robot in virtual space, and realizes intention sharing and motion coordination pre-play among multiple robots through a distributed communication mechanism to avoid collisions.

[0013] Furthermore, the first communication protocol is the TCP / IP protocol, and the second communication protocol is the WebSocket protocol.

[0014] Furthermore, the PLC controller is a Siemens PLC controller, the collaborative robot is a JAKA collaborative robot, and the collaborative robot integrates a position sensor, a speed sensor, and a force control feedback module.

[0015] Furthermore, the system adopts a distributed hardware-in-the-loop debugging and verification method. First, the device twin and the virtual controller are used to perform full simulation logic joint debugging in the digital twin platform system. After the verification is passed, the physical devices are gradually connected to replace the corresponding virtual modules.

[0016] According to another aspect of the present invention, a laboratory digital twin method based on Unreal Engine and self-evolving AI is provided, comprising the following steps using the system: S1 is generated by the self-evolving AI decision engine based on the knowledge base, which actively generates experimental objectives and preliminary strategies. S2 constructs a virtual experimental scenario based on the experimental objective in the digital twin platform system, and uses a physics engine to simulate and optimize the preliminary strategy to generate an executable optimal experimental strategy. S3 transforms the optimal experimental strategy into low-level control commands, which are then sent to the robotic arm system and intelligent logistics scheduling system through the control system to execute the physical experiment. During the execution of the physical experiment, S4 compares the virtual simulation data and the physical entity data in real time through the causal intervention channel. If a deviation occurs, the digital twin platform system sends a correction command to the control system to intervene in the physical experiment process in real time. After the S5 experiment is completed, the full-link experimental data is fed back to the self-evolving AI decision engine in real time to drive it to perform online incremental learning and update the knowledge base and model.

[0017] The beneficial effects of this invention are as follows: By integrating a self-evolving AI decision engine and a materials knowledge base, the system can autonomously generate "composition-structure-performance" hypotheses and create new material targets, thereby transforming materials discovery from relying on manual pre-setting to AI-driven autonomous scientific exploration; by integrating a physics engine into a high-fidelity digital twin built with Unreal Engine for tens of thousands of simulations and establishing a causal intervention channel for real-time interaction with physical entities, experimental strategies can be optimized in virtual space and then directly intervened in reverse to correct the physical experimental process; by establishing distributed intent-sharing nodes among multiple robotic arm digital twins and performing collaborative motion planning and collision detection based on virtual space, multi-arm collaboration is upgraded from simple temporal coordination to intelligent collaboration with predictive and collision avoidance capabilities; by collecting full-link experimental data in real time based on the OPCUA protocol and feeding it back to an AI model based on graph neural networks and reinforcement learning for online incremental learning, the system can dynamically update the knowledge base and decision model based on each experimental result, achieving autonomous evolution that becomes smarter with use. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art are within the scope of protection of the present invention.

[0019] According to an embodiment of the present invention, a laboratory digital twin system based on Unreal Engine and self-evolving AI is provided, comprising a control system, a robotic arm system, an intelligent logistics scheduling system, a digital twin platform system, and a human-computer interaction system. The control system includes a PLC controller as the core main control unit and a local central control server, which are used to coordinate the system operation logic, store data and process high-level logic operations. The local central control server integrates a self-evolving AI decision engine. The robotic arm system includes at least one collaborative robot, which is connected to the control system and the digital twin platform system via a first communication protocol; The intelligent logistics scheduling system is used to schedule automated transportation equipment for the transfer of experimental materials, and maintains real-time two-way communication with the control system and digital twin platform system through the second communication protocol. The digital twin platform system is built on Unreal Engine, forming a 1:1 high-fidelity virtual mapping with the physical experimental environment, and integrates a physics engine; the digital twin platform system interacts with the robotic arm system and the intelligent logistics scheduling system through the first communication protocol and the second communication protocol respectively, and has a causal intervention channel for real-time bidirectional data exchange with the control system. The human-machine interaction system is used for system monitoring and command input, and obtains underlying operating data from the PLC controller of the control system through a data acquisition protocol; The self-evolving AI decision engine actively generates experimental objectives and hypotheses based on a knowledge base, and performs virtual simulation verification in the digital twin platform system to generate the optimal experimental strategy. The digital twin platform system sends correction instructions to the control system through the causal intervention channel based on the differences between the virtual simulation and the physical entity operation, thereby intervening in the physical experiment process in reverse. The end-to-end data generated by the experiment is fed back to the self-evolving AI decision engine in real time for online incremental learning.

[0020] Preferably, the self-evolving AI decision engine includes a generative algorithm module, which is used to actively generate scientific hypotheses and formulation parameters about the "composition-structure-performance" of materials based on materials genome databases and literature knowledge, so as to achieve the AI ​​self-creation of experimental objectives.

[0021] Preferably, the self-evolving AI decision engine further includes a model coupled with graph neural networks and deep reinforcement learning, used for deep analysis of material structure-property relationships and dynamic optimization of experimental strategies; the local central control server collects the underlying operating data of the PLC controller through the OPCUA protocol and feeds it back to the self-evolving AI decision model for online incremental learning.

[0022] Preferably, after receiving the experimental strategy, the digital twin platform system uses a physics engine to perform high-speed simulation and logical verification of the robotic arm's motion trajectory and logistics transportation path in a virtual environment before issuing physical instructions, and selects the optimal reaction path and operation sequence.

[0023] Preferably, the causal intervention channel is specifically configured such that when the physical sensor data collected by the human-computer interaction system does not match the virtual prediction result in the digital twin platform system, the digital twin platform system sends a correction command to the PLC controller of the control system through a real-time bidirectional communication interface to adjust the operating parameters of the robotic arm or related experimental equipment in real time.

[0024] Preferably, the robotic arm system includes multiple collaborative robots; the digital twin platform system creates a digital twin for each collaborative robot in virtual space, and realizes intention sharing and motion coordination pre-play among multiple robots through a distributed communication mechanism to avoid collisions.

[0025] Preferably, the first communication protocol is the TCP / IP protocol, and the second communication protocol is the WebSocket protocol.

[0026] Preferably, the PLC controller is a Siemens PLC controller, the collaborative robot is a JAKA collaborative robot, and the collaborative robot integrates a position sensor, a speed sensor, and a force control feedback module.

[0027] Preferably, the system adopts a distributed hardware-in-the-loop debugging and verification method. First, the device twin and the virtual controller are used to perform full simulation logic joint debugging in the digital twin platform system. After the verification is passed, the physical devices are gradually connected to replace the corresponding virtual modules.

[0028] According to another aspect of the present invention, a laboratory digital twin method based on Unreal Engine and self-evolving AI is provided, comprising the following steps using the system: S1 is generated by the self-evolving AI decision engine based on the knowledge base, which actively generates experimental objectives and preliminary strategies. S2 constructs a virtual experimental scenario based on the experimental objective in the digital twin platform system, and uses a physics engine to simulate and optimize the preliminary strategy to generate an executable optimal experimental strategy. S3 transforms the optimal experimental strategy into low-level control commands, which are then sent to the robotic arm system and intelligent logistics scheduling system through the control system to execute the physical experiment. During the execution of the physical experiment, S4 compares the virtual simulation data and the physical entity data in real time through the causal intervention channel. If a deviation occurs, the digital twin platform system sends a correction command to the control system to intervene in the physical experiment process in real time. After the S5 experiment is completed, the full-link experimental data is fed back to the self-evolving AI decision engine in real time to drive it to perform online incremental learning and update the knowledge base and model.

[0029] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention will be provided through specific usage methods.

[0030] In practical application, the laboratory digital twin system based on Unreal Engine and self-evolving AI, as described in this invention, mainly includes a control system, a robotic arm system, an intelligent logistics scheduling system (RDS), a digital twin platform system, and a human-computer interaction system. These systems are deeply coupled through specific communication protocols and logical architectures, jointly forming an intelligent experimental platform with autonomous perception, decision-making, execution, and evolutionary capabilities.

[0031] The control system uses a Siemens PLC controller as the core control unit, responsible for coordinating the operation logic of the robotic arm system, intelligent logistics scheduling system, and other automated equipment. The PLC constructs the underlying control network via industrial Ethernet, enabling high-speed command issuance and status feedback. It is equipped with a local central control server for storing production data and processing high-level logic operations. This local central control server not only stores end-to-end production data but also integrates a self-evolving AI decision engine for handling high-level logic operations and scientific decision-making, such as material discovery and path planning.

[0032] The robotic arm system utilizes JAKA series collaborative robots, enabling real-time data interaction with the control system and digital twin platform via TCP / IP communication protocol. The robotic arm unit integrates position sensors, speed sensors, and force control feedback modules, supporting precise trajectory planning and collision detection to ensure the flexibility and safety of experimental operations.

[0033] The intelligent logistics scheduling system employs the RDS (Robot Dispatch System) AGV (Automated Guided Vehicle) scheduling system to manage and schedule the automated transport of experimental materials. This system maintains real-time bidirectional communication with the upper-level control terminal and digital twin platform via the WebSocket protocol, achieving millisecond-level synchronization of AGV path planning, task allocation, and location status, ensuring efficient and accurate logistics transportation.

[0034] The digital twin platform system uses Unreal Engine 5 (UE5) as its core development environment, constructing a virtual experimental scene that maps to the physical reality at a 1:1 scale through a high-fidelity 3D visualization interface. The system integrates the PhysX physics engine to enhance the realism of the virtual-real mapping and features multi-protocol communication interfaces: it can receive joint data from the JAKA robotic arm via TCP / IP and monitor the real-time trajectory of the RDS vehicle via WebSocket. It also has a causal intervention channel for interconnection with the control system. The platform supports a sensor-triggered mechanism, allowing for logical verification of the simulation model's actions via virtual ladder diagrams or Python scripts.

[0035] The human-machine interface system includes an industrial touchscreen and a PC workstation for operators to monitor and input commands. The system features a data visualization module that collects low-level operational data from Siemens PLCs via the industry-standard OPCUA protocol. Combined with real-time rendering in UE5, this data presents an immersive production monitoring interface on the PC, enabling managers to intuitively understand the current production status and make informed decisions.

[0036] To achieve a shift in materials discovery from "human-defined goals" to "AI-created goals": By integrating an AI decision engine into the local central control server in the control system, the system can automatically retrieve materials genome databases and literature knowledge using its high-level logical computing capabilities. Based on generative algorithms, it can proactively propose scientific hypotheses of "composition-structure-performance" and automatically generate formulation parameters for target materials, thereby replacing the traditional mode of manually setting single performance indicators.

[0037] The experimental strategy is transformed from "human-written process" to "AI self-evolving process": using a digital twin platform system (based on UE5 and PhysX physics engine) as a virtual trial and error field, the motion trajectory of the robotic arm system and the transportation path of the intelligent logistics scheduling system are simulated and logically verified tens of thousands of times in the virtual environment (virtual ladder diagram) before issuing physical commands. The optimal response path and operation sequence are automatically selected, thereby replacing the inefficient mode of manually writing fixed linear processes.

[0038] This transforms the digital twin system from a "visual mapping" to a "causal intervention engine": It leverages a real-time bidirectional communication interface (TCP / IP and WebSocket) established between the Siemens PLC controller and the digital twin platform to construct a closed loop of "data-physical-semantic" information. When the physical sensor data collected by the human-machine interface system via the OPCUA protocol does not match the virtual prediction, the digital twin platform can send corrective commands in reverse through the causal intervention channel, adjusting the temperature control, stirring, or feeding parameters of the robotic arm in real time. This enables the system to reverse-engineer the physical experiment process.

[0039] To realize the transformation of experimental knowledge from "offline recording" to "online evolutionary ontology": Based on the data feedback mechanism of the control system, the full-link time-series data (such as yield and by-product information) generated in physical experiments is fed back to the AI ​​decision model in real time for online incremental learning. This enables the system to dynamically revise its prediction model and knowledge base based on the results of each experiment, thereby making the system smarter with use and establishing a self-evolving chemical experiment knowledge ontology.

[0040] Example 1: Autonomous Discovery of High-Entropy Oxide Anode Materials Based on AI-Created Targets Scenario Description: A new energy battery laboratory needs to find lithium-ion battery anode materials with a cycle life of >5000 cycles and a capacity of >800 mAh g⁻¹. Faced with the predicament of human scientists exhaustively exploring transition metal combinations for three consecutive months without a breakthrough, this system intervenes through a cognitive substitution mode.

[0041] Deployment methods and implementation details: 1. Activating AI Questioner Mode (Control System - AI Decision Layer): The local central control server in the control system initiates the self-evolving AI decision engine, embedding knowledge from the materials genome database and 2800 recent articles via a data interface. The AI ​​does not simply search; instead, it constructs 137 "performance-composition-structure" hypotheses based on generative algorithms, automatically generating innovative goals such as "introducing 3at%Sc³⁺ into high-entropy rock-salt phase oxides to suppress lattice oxygen evolution."

[0042] 2. Real-time Digital Twin Generation (UE5 + PhysX): The digital twin platform system renders the corresponding virtual experimental scene within 0.8 seconds using the UE5 engine. The system calls the PhysX physics engine to bind the physical properties of the sol-gel synthesis module, tube furnace, in-situ XRD sample cell, and two JAKA collaborative robotic arms, ensuring that the gravity and collision volume of the virtual space are completely consistent with the physical entities.

[0043] 3. Self-evolutionary experimental strategy (virtual simulation verification): Before issuing instructions, the AI ​​performs 60,000 zero-cost reaction path simulations in a twin space. Using virtual ladder diagram logic verification, four optimal routes with a yield > 85%, thermodynamic feasibility, and total duration < 22 hours are selected. These advanced strategies are then transformed into low-level OPCUA instruction flows recognizable by Siemens PLCs.

[0044] 4. Virtual-real closed-loop verification (multi-protocol collaborative control): Physical execution: The intelligent logistics scheduling system receives instructions via the WebSocket protocol and schedules the RDSAGV trolley to accurately deliver the metal nitrate precursor to the synthesis station; the JAKA robotic arm (Arm-A) receives PLC instructions via the TCP / IP protocol and completes the gradient addition of 9 components with an accuracy of 0.05mL; Arm-B coordinates to complete the transfer operation between the oven and the tube furnace.

[0045] Twin-side intervention: The platform collects data from 52 sensors aggregated by the PLC in real time via OPCUA to perform online soft measurement of particle morphology. When the virtual XRD simulation finds that the peak position shift of Sc doping is 0.12° which does not match the prediction, the system triggers a causal intervention mechanism. The AI ​​immediately decides to lower the annealing temperature by 30°C, and the instruction is sent to the physical furnace controller in milliseconds.

[0046] 5. Knowledge Feedback and Iteration (Online Incremental Learning): After the experiment, the in-situ charge-discharge data (0.1C capacity 823 mAh g⁻¹, 96% retention rate after 1000 cycles) was automatically written back to the local central control server. The graph neural network model completed incremental updates (Online Learning) within 15 minutes, automatically avoiding similar inefficient components containing Sc in the next round of target generation, and shifting to the Ta–Nb–Mo–W–Ti space, thus achieving the system's self-evolution.

[0047] Results: Three new materials were identified in just 11 days, with no human intervention throughout the process, and the repeatability error of the experiment was <1.2%.

[0048] Example 2: Total Synthesis of Natural Products Based on Two-Arm Collaboration and "Intent Sharing" Scenario Description: A pharmaceutical company needs to synthesize 20g of the anticancer natural product (+)-Scholarisine A, involving 12 stringent reaction steps. Addressing the low success rate of traditional two-person operations, the system utilizes multi-arm collaboration and digital twin technology for end-to-end control.

[0049] Deployment methods and implementation details: 1. AI-driven self-evolution process: The control system reads historical data and quantum chemical calculation results, reconstructing the 12-step process into 8 "microreactor-separation" units. The system automatically calculates the optimal temperature control and stirring parameters, generating an automated task sequence list based on a time axis.

[0050] 2. Digital Twin Pre-Launch: Collision Detection and Intent Synchronization: A 1:1 model of a microreactor and autoclave is established in the UE5 digital twin platform. Two JAKA robotic arms perform thousands of motion planning simulations in virtual space using the PhysX engine, completely eliminating collision risks. The system establishes an intent-sharing node, enabling Arm-A to feed / transfer and Arm-B to quench / sample in real time via the distributed communication bus of the twin layer to synchronize the next sub-target, rather than simply waiting for a specific time sequence.

[0051] 3. Physical execution and intervention, i.e., real-time virtual-real interaction: Step 4: Lithification reaction: Arm-A injects n-BuLi within 2.3 seconds during the liquid injection operation. Simultaneously, the twin platform calculates the reaction heat flow. When the predicted temperature rise is >1°C, the system sends an intention command to Arm-B via PLC to "start the refrigerant pump" 5 seconds in advance, before the physical sensor alarms, thus achieving prediction-based feedforward control.

[0052] Step 7 High-Pressure Hydrogenation: After Arm-B is fed, the virtual sensor detects an abnormal hydrogen consumption rate, corresponding to catalyst deactivation. The AI ​​decision engine generates a remedial plan within 5 seconds, driving Arm-A to replenish Pd(OH)2 / C, successfully restarting the physical reaction.

[0053] 4. Online quality closed-loop: After each reaction step, Arm-B sends the sample to LC-MS. The human-computer interaction system compares the virtual mass spectrometry data with the measured data via OPCUA, finding that the byproduct at m / z 421 exceeds the limit by 0.7%. The system automatically triggers the feedback loop, dynamically fine-tuning the base equivalent (1.5 -> 1.3 eq) in the next coupling operation, successfully suppressing the side reaction to within 0.2%.

[0054] 5. Batch completion and ontology evolution: 20.4g of target product was produced (yield 56%). The entire set of validated "operation-parameter-result" correlation data was written into the "evolvable ontology library" of the control system, serving as a mature process package for direct use in subsequent processes.

[0055] Results: A single person remotely monitored four production lines, increasing equipment utilization by 41%, and the entire operation process complied with FDA 21 CFR Part 11 electronic record standards.

[0056] In summary, by leveraging the technical solutions described above, the system integrates a self-evolving AI decision engine and a materials knowledge base, enabling it to autonomously generate "composition-structure-performance" hypotheses and create new material targets. This transforms materials discovery from relying on manual pre-setting to AI-driven autonomous scientific exploration. Furthermore, by integrating a physics engine into a high-fidelity digital twin built with Unreal Engine for tens of thousands of simulations and establishing a causal intervention channel for real-time interaction with physical entities, experimental strategies can be optimized in virtual space and then directly intervened in and corrected in the physical experimental process. By establishing distributed intent-sharing nodes among multiple robotic arm digital twins and performing collaborative motion planning and collision detection in virtual space, multi-arm collaboration is upgraded from simple temporal coordination to intelligent collaboration with predictive and collision avoidance capabilities. Finally, by collecting full-link experimental data in real-time based on the OPCUA protocol and feeding it back to an AI model based on graph neural networks and reinforcement learning for online incremental learning, the system can dynamically update its knowledge base and decision model based on each experimental result, achieving autonomous evolution that becomes smarter with use.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A laboratory digital twin system based on Unreal Engine and self-evolving AI, characterized in that, This includes control systems, robotic arm systems, intelligent logistics scheduling systems, digital twin platform systems, and human-computer interaction systems; The control system includes a PLC controller as the core main control unit and a local central control server, which are used to coordinate the system operation logic, store data and process high-level logic operations. The local central control server integrates a self-evolving AI decision engine. The robotic arm system includes at least one collaborative robot, which is connected to the control system and the digital twin platform system via a first communication protocol; The intelligent logistics scheduling system is used to schedule automated transportation equipment for the transfer of experimental materials, and maintains real-time two-way communication with the control system and digital twin platform system through the second communication protocol. The digital twin platform system is built on Unreal Engine, forming a 1:1 high-fidelity virtual mapping with the physical experimental environment, and integrates a physics engine; the digital twin platform system interacts with the robotic arm system and the intelligent logistics scheduling system through the first communication protocol and the second communication protocol respectively, and has a causal intervention channel for real-time bidirectional data exchange with the control system. The human-machine interaction system is used for system monitoring and command input, and obtains underlying operating data from the PLC controller of the control system through a data acquisition protocol; The self-evolving AI decision engine actively generates experimental objectives and hypotheses based on a knowledge base, and performs virtual simulation verification in the digital twin platform system to generate the optimal experimental strategy. The digital twin platform system sends correction instructions to the control system through the causal intervention channel based on the differences between the virtual simulation and the physical entity operation, thereby intervening in the physical experiment process in reverse. The end-to-end data generated by the experiment is fed back to the self-evolving AI decision engine in real time for online incremental learning.

2. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The self-evolving AI decision engine includes a generative algorithm module, which is used to actively generate scientific hypotheses and formulation parameters about the "composition-structure-performance" of materials based on materials genome databases and literature knowledge, so as to achieve the AI ​​self-creation of experimental objectives.

3. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The self-evolving AI decision engine also includes a model coupled with graph neural networks and deep reinforcement learning, used to deeply analyze the structure-property relationship of materials and dynamically optimize experimental strategies; the local central control server collects the underlying operating data of the PLC controller through the OPCUA protocol and feeds it back to the self-evolving AI decision model for online incremental learning.

4. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, After receiving the experimental strategy, the digital twin platform system uses a physics engine to perform high-speed simulation and logical verification of the robotic arm's motion trajectory and logistics transportation path in a virtual environment before issuing physical instructions, thereby selecting the optimal reaction path and operation sequence.

5. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The causal intervention channel specifically refers to the following: when the physical sensor data collected through the human-computer interaction system does not match the virtual prediction result in the digital twin platform system, the digital twin platform system sends a correction command to the PLC controller of the control system through a real-time bidirectional communication interface to adjust the operating parameters of the robotic arm or related experimental equipment in real time.

6. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The robotic arm system includes multiple collaborative robots; the digital twin platform system creates a digital twin for each collaborative robot in virtual space, and realizes intention sharing and motion coordination pre-play among multiple robots through a distributed communication mechanism to avoid collisions.

7. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The first communication protocol is the TCP / IP protocol, and the second communication protocol is the WebSocket protocol.

8. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The PLC controller is a Siemens PLC controller, and the collaborative robot is a JAKA collaborative robot. The collaborative robot integrates a position sensor, a speed sensor, and a force control feedback module.

9. The laboratory digital twin system based on Unreal Engine and self-evolving AI according to claim 1, characterized in that, The system adopts a distributed hardware-in-the-loop debugging and verification method. First, the device twin and the virtual controller are used to perform full simulation logic joint debugging in the digital twin platform system. After the verification is successful, the physical devices are gradually connected to replace the corresponding virtual modules.

10. A laboratory digital twin method based on Unreal Engine and self-evolving AI, using the system described in any one of claims 1-9, characterized in that, Includes the following steps: S1 is generated by the self-evolving AI decision engine based on the knowledge base, which actively generates experimental objectives and preliminary strategies. S2 constructs a virtual experimental scenario based on the experimental objective in the digital twin platform system, and uses a physics engine to simulate and optimize the preliminary strategy to generate an executable optimal experimental strategy. S3 transforms the optimal experimental strategy into low-level control commands, which are then sent to the robotic arm system and intelligent logistics scheduling system through the control system to execute the physical experiment. During the execution of the physical experiment, S4 compares the virtual simulation data and the physical entity data in real time through the causal intervention channel. If a deviation occurs, the digital twin platform system sends a correction command to the control system to intervene in the physical experiment process in real time. After the S5 experiment is completed, the full-link experimental data is fed back to the self-evolving AI decision engine in real time to drive it to perform online incremental learning and update the knowledge base and model.