Radiation field reconstruction method, computer device and readable storage medium

Abstract

The application relates to a radiation field reconstruction method, a computer device and a readable storage medium. The method comprises the following steps: constructing a kernel data digital twin environment according to three-dimensional basic data, and generating virtual scene data; performing radiation field calculation on a three-dimensional space of the virtual scene data based on radiation source parameters and shielding material parameters, and generating radiation dose distribution data; deploying virtual nodes corresponding to virtual robots and physical robots; planning an inspection path of the virtual robots and the virtual nodes; controlling the physical robots to perform an inspection task according to the planned inspection path of the virtual nodes; acquiring real-time simulation data of the virtual nodes and receiving real-time feedback data of the physical robots; and verifying and updating the radiation dose distribution data based on the real-time simulation data and the real-time feedback data. The method can solve the problem of large deviation between static simulation and a real environment through virtual-real closed-loop correction, continuously improve the radiation field model precision, and provide a reliable basis for subsequent inspection path optimization.

Description

Technical Field

[0001] This application relates to the field of intelligent inspection technology for nuclear power plants, and in particular to a radiation field reconstruction method, computer equipment, and computer-readable storage medium. Background Technology

[0002] During the decommissioning and routine maintenance of nuclear facilities, operators frequently need to enter high-radiation areas to inspect equipment and assess operating conditions. The annual cumulative dose approaches regulatory limits, resulting in significant radiation exposure risks. Although robotic systems are considered an alternative, they require thorough testing and verification before deployment. However, frequent testing is prohibited at nuclear facilities, and there is a lack of representative sites containing real radiation sources. This leads to insufficient verification of core algorithms such as radiation-sensing navigation and dose-optimized path planning, resulting in low system reliability and a high failure rate after on-site deployment.

[0003] Traditional robot simulation tools (such as Gazebo, V-REP, and Webots) can accurately simulate traditional sensors like LiDAR and cameras, but they completely lack physical modeling of radiation fields, treating radiation as an invisible factor and failing to simulate key processes such as radiation attenuation, shielding effects, and detector response. Developers can only visualize radiation distribution through color overlays or numerical annotations, lacking physical realism. This leads to problems such as inappropriate radiation response, path planning failures, and excessive cumulative doses when simulation-based algorithms are deployed in real-world scenarios. Furthermore, traditional simulation environments differ significantly from real nuclear facilities in terms of environmental complexity, shielding effects, and detector response characteristics, resulting in low algorithm migration success rates. Extensive on-site data recalibration is required, leading to long cycles, high labor costs, and the debugging process forcing operators and robots to frequently enter high-radiation areas, contradicting the initial goal of reducing radiation exposure. Summary of the Invention

[0004] Therefore, it is necessary to provide a radiation field reconstruction method, computer equipment, and computer-readable storage medium to address the aforementioned technical problems. This method can solve the problem of large deviations between static simulation and the real environment through virtual-real closed-loop correction, continuously improve the accuracy of the radiation field model, and provide a reliable basis for subsequent inspection path optimization.

[0005] In a first aspect, this application provides a radiation field reconstruction method, including:

[0006] A digital twin environment for the nuclear facility is constructed based on 3D baseline data, generating virtual scene data;

[0007] Receive radiation source parameters and shielding material parameters;

[0008] Based on radiation source parameters and shielding material parameters, radiation field calculations are performed on the three-dimensional space of virtual scene data to generate radiation dose distribution data.

[0009] Deploy virtual robots and corresponding virtual nodes for physical robots;

[0010] Based on virtual scene data and radiation dose distribution data, the inspection paths of virtual robots and virtual nodes are planned.

[0011] Control the physical robot to perform inspection tasks according to the planned inspection path of the virtual nodes;

[0012] Acquire real-time simulation data of virtual nodes and receive real-time feedback data from the physical robot;

[0013] The radiation dose distribution data is verified and updated based on real-time simulation data and real-time feedback data.

[0014] In one embodiment, radiation field calculations are performed on the three-dimensional space of the virtual scene data based on radiation source parameters and shielding material parameters to generate radiation dose distribution data, including:

[0015] Based on radiation source parameters, the virtual scene data is decomposed into radiation units, and based on shielding material parameters, material properties are set for each radiation unit.

[0016] Deploy at least one detector in the three-dimensional space of the virtual scene data;

[0017] For each detector, based on the radiation source parameters, the current detector is used as the observation point, and a virtual ray is emitted towards the radiation source;

[0018] Based on a preset parallel computing strategy, the virtual ray traverses all target radiation units it passes through from the current detector to the radiation source, and records the material type of each target radiation unit and the path length of the virtual ray within the target radiation unit.

[0019] Calculate the total attenuation factor for each of the multiple energy levels based on the material type and path length of all target radiating elements.

[0020] For each energy level, the radiation intensity corresponding to the energy level is calculated based on the initial source intensity of the energy level, the total attenuation factor corresponding to the energy level, the equivalent radius of the current detector, and the distance from the current detector to the radiation source.

[0021] Based on the weight information corresponding to multiple energy levels, the radiation intensity corresponding to multiple energy levels is weighted and summed to obtain the radiation dose rate of the current detector.

[0022] The radiation dose distribution data is determined based on the radiation dose rate of at least one detector.

[0023] In one embodiment, based on the weight information corresponding to multiple energy levels, the radiation intensity corresponding to multiple energy levels is weighted and summed to obtain the radiation dose rate of the current detector, including:

[0024] Based on the weight information corresponding to multiple energy levels, the radiation intensity corresponding to multiple energy levels is weighted and summed to obtain the initial dose rate of the current detector.

[0025] The radiation dose rate of the current detector is obtained by superimposing a scattering correction term on the initial dose rate.

[0026] In one embodiment, the virtual scene data is decomposed into radiation elements based on radiation source parameters, including:

[0027] Build an octree mesh for the virtual scene data;

[0028] Based on radiation source parameters, the octree mesh is adaptively divided according to the preset radiation source location and the estimated dose gradient to determine multiple voxels.

[0029] Based on a pre-defined voxel clustering algorithm, multiple voxels are clustered and merged to obtain several radiative units.

[0030] In one embodiment, after determining radiation dose distribution data based on the radiation dose rate of at least one detector, the method further includes:

[0031] Based on the radiation dose distribution data, the dose contribution of each radiation unit to other radiation units within a preset range is determined, and a radiation contribution lookup table is constructed.

[0032] In one embodiment, the physical robot carries multiple different types of target detectors; it acquires real-time simulation data of virtual nodes, including:

[0033] Based on a pre-built detector type library, the sensitivity function corresponding to each target detector is determined;

[0034] Based on radiation dose distribution data, the initial dose rate at the location of the virtual node is determined;

[0035] Based on the azimuth angle of the source relative to the detector coordinate system and the sensitivity function corresponding to each target detector, the initial dose rate is corrected to obtain the effective dose rate corresponding to each target detector.

[0036] Statistical fluctuation simulations were performed based on the effective dose rate corresponding to each target detector to obtain the simulated dose rate;

[0037] The simulated dose rates corresponding to each target detector are fused to obtain real-time simulation data of the virtual node.

[0038] In one embodiment, each radiation unit maintains a probability distribution of radiation dose rate; based on virtual scene data and radiation dose distribution data, the inspection paths for the virtual robot and virtual nodes are planned, including:

[0039] The information entropy of each radiation unit is determined based on the probability distribution of the radiation dose rate of each radiation unit.

[0040] Construct a multi-robot joint observation gain function that includes a penalty for overlapping robot detection ranges;

[0041] The Hungarian algorithm is used to solve for the optimal task allocation, which maximizes the total information gain corresponding to the multi-robot joint observation gain function.

[0042] Based on optimal task allocation, the inspection paths for virtual robots and virtual nodes are planned.

[0043] In one embodiment, the method further includes:

[0044] Physical degradation models, radiation source evolution models, and shielding failure simulation models corresponding to virtual scene data are constructed respectively.

[0045] The radiation dose distribution data is dynamically updated based on the outputs of the physical degradation model, the radiation source evolution model, and the shielding failure simulation model.

[0046] Secondly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect above.

[0047] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0048] The aforementioned radiation field reconstruction method, computer equipment, and computer-readable storage medium construct a digital twin environment for nuclear facilities based on 3D baseline data, generating virtual scene data; receive radiation source parameters and shielding material parameters; calculate the radiation field in the 3D space of the virtual scene data based on the radiation source parameters and shielding material parameters, generating radiation dose distribution data; deploy virtual robots and corresponding virtual nodes of physical robots; plan the inspection paths of virtual robots and virtual nodes based on the virtual scene data and radiation dose distribution data; control physical robots to perform inspection tasks according to the planned inspection paths of virtual nodes; acquire real-time simulation data of virtual nodes and receive real-time feedback data from physical robots; verify and update the radiation dose distribution data based on the real-time simulation data and real-time feedback data. Through this method, a digital twin environment for nuclear facilities is constructed based on 3D baseline data, eliminating the need for frequent equipment testing in real high-radiation sites, saving on-site testing costs and reducing personnel radiation exposure risks. The radiation field calculation based on radiation source parameters and shielding material parameters generates dose distribution data, solving the problem that traditional simulations cannot quantitatively describe the spatial distribution of radiation. Virtual nodes corresponding to physical robots are deployed, and path planning is performed based on virtual scenes and dose data. The physical robots are then controlled to execute tasks according to the virtual node paths, avoiding direct and blind deployment and improving inspection safety and task success rates. Simulation data from virtual nodes is acquired, and real-time feedback data from the physical robots is received. Radiation dose distribution data is verified and updated, and optimization is performed using a small number of real measurement points, significantly reducing the amount of real data required. A virtual-real closed-loop correction addresses the problem of large deviations between static simulation and the real environment, continuously improving the accuracy of the radiation field model and providing a reliable basis for subsequent inspection path optimization. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a diagram illustrating the application environment of the radiation field reconstruction method in one embodiment;

[0051] Figure 2 This is a flowchart illustrating a radiation field reconstruction method in one embodiment;

[0052] Figure 3 This is a schematic diagram of the platform architecture in one embodiment;

[0053] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0056] The radiation field reconstruction method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, robot 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed in the cloud or on other network servers. Server 104 constructs a digital twin environment of the nuclear facility based on 3D basic data, generating virtual scene data; receives radiation source parameters and shielding material parameters; calculates the radiation field in the 3D space of the virtual scene data based on the radiation source parameters and shielding material parameters, generating radiation dose distribution data; deploys virtual robots and corresponding virtual nodes of physical robot 102; plans the inspection paths of virtual robots and virtual nodes based on the virtual scene data and radiation dose distribution data; controls physical robot 102 to perform inspection tasks according to the planned inspection paths of virtual nodes; acquires real-time simulation data of virtual nodes and receives real-time feedback data from physical robot 102; and verifies and updates the radiation dose distribution data based on the real-time simulation data and real-time feedback data. Robot 102 can be, but is not limited to, various mobile robots, including wheeled robots, tracked robots, legged robots, drones, etc. Server 104 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0057] In one exemplary embodiment, such as Figure 2 As shown, a radiation field reconstruction method is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes:

[0058] Step 202: Construct a digital twin environment for the nuclear facility based on the 3D basic data and generate virtual scene data.

[0059] The 3D foundational data includes, but is not limited to, computer-aided design (CAD) models of nuclear facilities, LiDAR point cloud data, Building Information Modeling (BIM) data, photogrammetric mesh models, and material and texture information. After acquiring the above 3D foundational data, the client or simulation server uses a 3D rendering engine (such as Unity3D or Unreal Engine) or a robot simulation platform (such as Gazebo or Webots) to construct a high-fidelity digital twin scene.

[0060] Optionally, during the construction process, coordinate alignment and fusion are performed on the multi-source 3D basic data: the structural geometric information in the CAD model, the actual spatial contours in the point cloud data, and the semantic information of equipment and pipelines in the BIM data are unified into the same world coordinate system; the scene is meshed and material is assigned to generate static virtual scene data containing objects such as walls, equipment, pipes, shielding bodies, and radiation source containers. For movable objects (such as temporary shielding plates, the robot itself) or dynamically changing areas, the virtual scene data also includes the pose and state parameters of the dynamic objects, supporting runtime updates.

[0061] In one optional implementation, the virtual scene data supports incremental updates and version management. When the layout of the real nuclear facility changes (e.g., adding equipment or removing temporary shielding), operators can use visual editing tools or import new 3D base data to locally reconstruct only the changed areas and generate updated virtual scene data. The system automatically records the timestamp and change log for each update, facilitating the traceability and reproduction of subsequent simulation results.

[0062] Step 204: Receive radiation source parameters and shielding material parameters.

[0063] The client or simulation server acquires radiation source parameters and shielding material parameters through a user input interface, configuration file, or real-time data stream. Optionally, it receives user-input radiation source parameters (including location, nuclide, activity, etc.) and shielding material parameters (including density, attenuation coefficient, etc.). Optionally, it acquires radiation source parameters from real-time feedback data from the physical robot and shielding material parameters from configuration information.

[0064] In one alternative implementation, before receiving radiation source parameters and shielding material parameters, the virtual scene data is preprocessed to identify the types of labeled objects in the scene and automatically associate the corresponding default shielding parameters from the built-in material database. Users can then fine-tune these default parameters, such as modifying the actual thickness of concrete walls or the density of lead plates. The received parameters are stored in memory as key-value pairs or structured data tables, and an index mapping is established with the corresponding objects in the virtual scene for quick lookup during radiation field calculations.

[0065] Step 206: Based on the radiation source parameters and shielding material parameters, perform radiation field calculations on the three-dimensional space of the virtual scene data to generate radiation dose distribution data.

[0066] The Adaptive Voxel Radiation Field (AVRF) model is used for real-time calculations. A bounding box is created based on the spatial extent of the virtual scene data, and this bounding box is divided into a coarse-resolution initial voxel mesh (e.g., 1m × 1m × 1m). The distance from the center of each voxel to each radiation source is calculated, and the radiation gradient of the region containing that voxel is evaluated. For regions with large radiation gradients (e.g., near strong radiation sources, shielding boundaries, or through-wall openings), the voxels are automatically subdivided into finer meshes (e.g., 0.1m × 0.1m × 0.1m), while for regions far from sources or completely blocked by thick shielding, the coarse mesh is maintained or further merged. This hierarchical voxelization strategy significantly reduces computational cost while maintaining accuracy.

[0067] In one alternative implementation, the core of the radiation field calculation is to solve for the gamma-ray dose rate at each voxel. For each radiation source, several rays are emitted from the source point in all directions in three-dimensional space (e.g., using uniform solid angle sampling or importance sampling). As each ray passes through the voxel grid, the energy attenuation is cumulatively calculated based on the material properties of each voxel or geometric object along the ray path. For multi-spectral radiation sources, the attenuation and energy deposition for each energy level are calculated separately, and then weighted and summed according to the dose conversion factor to obtain the total dose rate. Optionally, for the contribution of Compton scattering, a simplified scattering correction model is used, adding an isotropic scattering component from surrounding voxels to the ray tracing results.

[0068] In one alternative implementation, radiation dose distribution data is generated as a three-dimensional scalar field. When radiation source parameters or shielding material parameters change (e.g., the user adjusts source activity or moves a temporary shield), incremental radiation field recalculation is automatically triggered, updating only the voxels in the affected area, rather than recalculating the entire scene.

[0069] In one optional implementation, after generating the radiation dose distribution data, it is also displayed in the visualization interface as a pseudo-color overlay. Users can choose to display two-dimensional slices (horizontal, vertical, or arbitrary directional profiles) or three-dimensional isodose surfaces (e.g., 1 mSv / h, 10 mSv / h, 100 mSv / h isosurfaces). During navigation, the virtual robot (or a virtual node of a real robot) can query the dose rate value at its current pose in real time for subsequent dose-optimized path planning.

[0070] Step 208: Deploy the virtual robot and the corresponding virtual node of the physical robot.

[0071] In this system, the client or simulation server creates several virtual robot instances within the constructed digital twin environment of the nuclear facility, based on task requirements and robot type, and generates a corresponding virtual node for each physical robot connected to the system. A virtual robot is a robot model that runs entirely within the simulation environment, possessing a complete kinematic model, sensor model (LiDAR, camera, IMU, radiation detector, etc.), and collision detection capabilities. Its behavior is directly controlled by the planning algorithm or reinforcement learning strategy within the simulation platform, independent of real hardware. The virtual node corresponding to the physical robot is an abstract proxy entity that corresponds one-to-one with the real physical robot. This node runs on the simulation server, receiving real-time feedback data from the physical robot, such as pose, sensor readings, and operating status, and injecting this data into the virtual environment. This ensures that the virtual node's state in the digital twin space remains synchronized with the actual state of the real physical robot. Simultaneously, the virtual node can also receive control commands or planned paths from the simulation platform, which, after security verification, are forwarded to the real physical robot for execution.

[0072] In one alternative implementation, before deploying the virtual robot, a robot description file is selected or imported from a robot model library. The robot description file defines the robot's geometry, joint structure, mass attributes, sensor configuration, and the type and installation location of radiation detectors. The simulation server parses this file, instantiates the virtual robot in the virtual scene, and initializes its kinematic states (initial position, orientation, joint angles). For different types of robots (such as wheeled inspection robots, tracked robots, legged robots, or drones), the system automatically adapts the corresponding motion control interface and collision model.

[0073] In one optional implementation, deploying virtual nodes corresponding to the physical robot includes: creating a unique virtual node identifier for each physical robot to be connected in the simulation server and establishing a network connection for communication with that physical robot; extracting the same kinematics and sensor configuration from the physical robot's description file to ensure that the virtual node and the virtual robot have consistent physical properties; setting up a state synchronization module for the virtual node, which receives real-time data reported by the physical robot at a fixed frequency, including: pose, joint state, raw measurements of each sensor (especially radiation detector readings), fault codes, etc.; applying the received data to the virtual node to update its position, attitude, and sensor outputs in the digital twin environment. Simultaneously, the radiation detector model on the virtual node generates simulated readings based on the radiation dose distribution data at the current location of the virtual node (provided by the radiation field calculation module). These simulated readings can be compared with the actual detector readings of the physical robot for subsequent parameter calibration and model verification.

[0074] In one alternative implementation, virtual robots and their corresponding virtual nodes in the physical robot can coexist and interact within the same simulation scene. For example, one physical robot corresponds to one virtual node, and three virtual robots are deployed simultaneously. The simulation server maintains a unified scene graph that includes both virtual robot entities and virtual node entities. Virtual robots can perceive the pose and sensor data of virtual nodes, and vice versa. Thus, during the planning algorithm's execution, virtual nodes can be viewed as virtual avatars of the real robot, enabling virtual robots to collaboratively plan paths with the real robot. For instance, a virtual robot can explore a large high-risk area, while the physical robot (through its virtual node) performs verification measurements in a relatively safe area. Virtual nodes can also receive planned paths from the simulation platform, and after safety distance detection and radiation dose estimation, issue feasible control commands to the physical robot for execution.

[0075] Step 210: Based on virtual scene data and radiation dose distribution data, plan the inspection path for the virtual robot and virtual nodes.

[0076] The client or simulation server takes virtual scene data (including geometry, traversable areas, obstacle locations, equipment layout, etc.) and radiation dose distribution data (a three-dimensional scalar field, with each voxel corresponding to a dose rate or cumulative dose) as input to calculate an inspection path that meets the task requirements and minimizes radiation exposure risk for each virtual robot and the corresponding virtual node of each physical robot. The inspection path refers to the spatial trajectory starting from the starting point, passing through several predetermined inspection points (e.g., instrument reading locations, equipment status checkpoints, radiation hotspot verification points), and finally reaching the destination (e.g., returning to the charging station or evacuation area). It is typically represented as a series of discrete waypoints (position and orientation) or a continuous time-parameterized curve.

[0077] In one alternative implementation, multi-robot collaborative planning is supported when planning the inspection paths for virtual robots and virtual nodes. For example, the system maintains a global task list and assigns sub-tasks to each robot (including virtual robots and virtual nodes). The collaborative planning algorithm employs a distributed consensus algorithm to avoid multiple robots competing for the same narrow passage or simultaneously approaching the same high-dose hotspot. When a virtual robot's planned path needs to pass through an area currently occupied by a virtual node, the virtual node broadcasts its real-time pose to all virtual robots. The virtual robots treat this area as a dynamic obstacle and invoke the dynamic path replanning module to avoid it in real time.

[0078] In one optional implementation, the path planning results are visualized in a 3D simulation interface. Different robot paths are represented by lines of different colors, and the predicted dose rates along the paths are marked with color bands or numerical labels. Operators can manually drag and drop waypoints to modify them, and the system automatically replans local paths and updates cumulative dose estimates in real time. After planning is complete, users can save the path as a task file for subsequent batch execution or playback analysis.

[0079] Step 212: Control the physical robot to perform the inspection task according to the planned inspection path of the virtual node.

[0080] The simulation server obtains the pre-planned inspection path (typically represented as a series of timestamped waypoints or continuous trajectory curves) for the virtual nodes corresponding to the physical robot. Through the communication link established between the virtual nodes and the physical robot, the server sends the path instructions to the physical robot in the real nuclear facility. Following the received instructions, the physical robot moves sequentially to each waypoint and performs pre-defined inspection actions at each waypoint (e.g., taking pictures of instrument readings, measuring dose rate with a radiation detector, and checking equipment temperature with a thermal imager). During execution, the physical robot continuously feeds back its real-time pose, sensor readings, status information, and task progress to the simulation server, where the virtual nodes synchronize and record the data.

[0081] In one optional implementation, the command issuance employs a segmented confirmation mechanism, rather than issuing the entire path at once. The specific steps are as follows: The simulation server divides the complete inspection path of the virtual node into several sub-segments, each containing 2-5 consecutive waypoints and an accompanying safety summary (maximum dose rate, estimated time, cumulative dose); the path command for the first sub-segment is sent to the physical robot; the physical robot executes the sub-segment and reports the execution result to the simulation server upon completion (including the actual poses of each waypoint reached, measured dose rate, and time taken); the simulation server compares the execution result with the simulation prediction for that sub-segment in the virtual node. If the deviation is within acceptable limits (e.g., pose error <0.1m, dose rate deviation <20%), the next sub-segment is sent; if the deviation exceeds the limits, subsequent command issuance is paused, and local online replanning is initiated or operator intervention is requested; the above steps are repeated until all sub-segments are executed or the task is interrupted.

[0082] In one alternative implementation, the physical robot synchronizes its status to virtual nodes in real time while performing inspection tasks. The virtual nodes receive the latest pose of the physical robot and radiation detector readings at fixed time intervals. The simulation server uses this real-time data to dynamically correct the radiation dose distribution model, thereby providing a more accurate cost estimate for subsequent path segments. Simultaneously, the virtual nodes can compare the simulated dose rate with the measured dose rate at the physical robot's current location. If the difference exceeds a preset threshold (e.g., 30%), the parameter calibration module is automatically triggered to optimize the radiation field calculation parameters online.

[0083] In one alternative implementation, during the physical robot's inspection task, the operator can monitor it in real time through a human-machine interface. The interface simultaneously displays virtual nodes in the virtual scene (whose poses are synchronized with the actual robot) and video streams transmitted from the physical robot's cameras. The operator can pause the task at any time, manually take over control, or modify waypoints on the remaining path. The modified path is then replanned by the simulation server and sent to the physical robot with a new sequence of instructions for continued execution.

[0084] Step 214: Obtain real-time simulation data of the virtual node and receive real-time feedback data from the physical robot.

[0085] During operation, the simulation server extracts real-time simulation data from virtual nodes based on its own maintained simulation status, and receives real-time feedback data reported by physical robots in the real nuclear facility via a network communication interface. The real-time simulation data of the virtual nodes refers to various simulation output values ​​calculated by the virtual nodes (i.e., the abstract agent entities corresponding to the physical robots) in the digital twin environment based on radiation dose distribution data, virtual scene geometry data, kinematic models, and control commands at the current simulation moment. This data includes simulation outputs from sensor models (such as lidar point clouds, camera images, IMU acceleration, and especially simulated dose rates or count rates of radiation detectors). The real-time feedback data from the physical robots refers to the measured data periodically transmitted by the physical robots deployed in the real nuclear facility via wireless communication links, including but not limited to: the actual pose calculated by the physical robot itself through the positioning system, the actual measurement values ​​of various physical sensors (especially the measured dose rates or count rates of radiation detectors), motion execution status, fault codes, and the completion status of task execution phases.

[0086] In one optional implementation, the real-time simulation data of the virtual node and the real-time feedback data of the physical robot need to be aligned in time. Since there may be discrepancies between the simulated virtual time and the physical real time, and network transmission has variable latency, the simulation server employs a timestamp alignment and interpolation strategy. For example, each frame of simulation data carries a virtual simulation timestamp, and each frame of feedback data carries the physical robot's local real timestamp. When it is necessary to compare the simulated value and the measured value at the same moment, the system searches for the simulation frame closest to the real time in the historical buffer of the simulation data based on the real timestamp, or obtains the simulation data for that real moment through linear interpolation. Conversely, when the feedback data needs to be used to update the simulation model, the system aligns the timestamp of the feedback data to the simulation virtual timeline and applies it sequentially.

[0087] Step 216: Verify and update the radiation dose distribution data based on real-time simulation data and real-time feedback data.

[0088] The simulation server acquires real-time simulation data from virtual nodes (including simulated dose rate at the current location of the virtual node, radiation gradient of surrounding voxels, etc.) and real-time feedback data from physical robots (including radiation dose rate and detector count rate measured by the physical robots at the same or adjacent locations in the real nuclear facility, etc.). It then compares and analyzes the two data to assess the accuracy of the current radiation dose distribution data. Based on the comparison results, it corrects and optimizes the dose distribution data, gradually bringing the simulation model closer to the spatial distribution of the real radiation field. Verification involves comparing the differences between simulated and measured values ​​to test the reliability of existing radiation dose distribution data at a specific spatial location and quantifying the prediction error. Updating involves using measured data as observations and employing data assimilation or parameter optimization methods to adjust scene data, radiation source parameters, attenuation coefficients, scattering coefficients, etc., to recalculate or locally correct the radiation dose distribution data, thereby reducing the deviation between simulation and reality.

[0089] In one alternative implementation, the update process employs different strategies based on the verification results. If the overall error is small, only the radiation dose distribution data is fine-tuned using Kalman filtering or Bayesian update methods. Each new measurement point is treated as an observation, and the dose rate estimates and uncertainties of neighboring voxels are updated. For example, for each measured point, the system updates the dose rate estimates of all voxels within a sphere centered at that point and with a radius of R.

[0090] In one optional implementation, if the verification results show a large overall error or systematic bias (e.g., simulated values ​​are generally lower than measured values), a parameter-level update is triggered. The system initiates an online calibration module, using radiation source parameters (such as source activity and energy spectrum distribution), shielding material parameters (such as linear attenuation coefficient and thickness), and scattering correction coefficients as variables to be optimized. The goal is to reduce the difference between simulated and measured values, solving the parameter optimization problem. After optimization, the system recalculates the radiation field using the updated parameters, generating new radiation dose distribution data.

[0091] In one alternative implementation, the verification and update process supports closed-loop iteration. After each physical robot completes an inspection task, the system automatically updates the data using all the measured feedback data accumulated during that task, generating optimized radiation dose distribution data. The updated data is then used when planning the next inspection path, making dose predictions for subsequent tasks more accurate and further reducing the risk of radiation exposure during actual execution. Through this closed-loop mechanism, the accuracy of the radiation dose distribution data continuously improves with the number of task executions, ultimately making the simulation environment highly approximate the radiation field distribution of a real nuclear facility.

[0092] For example, after acquiring real-time simulation data from virtual nodes and real-time feedback data from the physical robot, the simulation server uses Bayesian inference and particle filtering algorithms to dynamically correct radiation source parameters (location, intensity) and radiation dose distribution data, using the actual measured values ​​as observation data. Simultaneously, an incremental map update strategy is employed, utilizing methods such as sliding windows, anomaly detection, and RANSAC removal to enable the radiation field model to adapt to environmental changes (such as the addition of new radiation sources, source movement, and source attenuation), continuously improving model accuracy.

[0093] In one alternative implementation, as the physical robot moves in a real-world environment and measures the dose rate, the system receives real-time feedback data, including the spatial coordinates of the measurement point and the measured dose rate. This data is based on the robot's position... Measured dose rate Construct the likelihood function (considering measurement noise):

[0094] ;

[0095] in, Given the locations and intensities of K radiation sources, For source Position Dosage contribution, To measure the variance of noise.

[0096] Using Bayesian inference to update the posterior distribution of radiation source parameters:

[0097] ;

[0098] in, The prior distribution is assumed to be uniformly distributed in terms of source location and intensity following a log-normal distribution.

[0099] The particle filter algorithm is used to approximate the posterior distribution, maintaining N=1000 particles. Each time a new measurement arrives, the particle weights are updated. Resample low-weighted particles; calculate the expected dose rate and uncertainty for each voxel:

[0100] ;

[0101] ;

[0102] in, For particles The theoretical total dose rate at the current voxel center location under the source parameters represented; This represents the expected dose rate at the current voxel. This represents the posterior variance of the dose rate at the current voxel, reflecting uncertainty.

[0103] These statistics can be used for subsequent information entropy calculations and path planning.

[0104] In one optional implementation, for incremental map updates, the system employs a sliding window strategy and a change detection mechanism to adapt to dynamic changes in the radiation field. The specific steps are as follows:

[0105] 1. Streaming Data Processing. As the physical robot moves continuously, it generates new measurement data (pose + dose rate). The system stores this data in chronological order into a fixed-length queue (e.g., retaining the most recent 100 measurements). Each new data point is added to the queue; when the queue length exceeds the limit, the oldest data is discarded. This sliding window strategy limits the computational burden while allowing the model to discard outdated information, thus tracking changes in the radiation field.

[0106] 2. Change Detection. The system periodically (e.g., after every 10 measurements) compares the current particle filter-predicted dose rate (calculated based on the latest particle set) with the new measured value. Residuals are defined. If the absolute value of the residual exceeds a preset threshold (e.g., 3 times the historical residual standard deviation), a significant change is considered to have occurred at that location. Possible causes of change include: the addition of a radiation source, the relocation of an existing radiation source, attenuation or enhancement of radiation source activity, and changes in the shielding structure.

[0107] 3. Triggering Source Parameter Reestimation. When a significant change is detected, the system initiates a change response mechanism. Specifically, the system adds new particle hypotheses near the changed region (e.g., generating a set of candidate source locations around the measurement point) and assigns them certain initial weights. Simultaneously, for the existing particle set, its weights are adjusted according to the direction of the change (e.g., if the measured value is significantly higher than the prediction, the weights of particles hypothesized to have additional sources or higher source intensities near the measurement point are increased). After several measurements, the particle set gradually converges to the new radiation source configuration.

[0108] 4. Outlier Removal. Due to potential detector malfunctions (such as electronic noise spikes) or momentary environmental obstructions (such as temporary blockage of the radiation path caused by personnel passing by), some measurements may deviate significantly from the true radiation field. For robustness, the system employs the RANSAC (Random Sample Consensus) algorithm to remove outliers.

[0109] In one alternative implementation, the real-time simulation data of the virtual nodes plays a dual role in this process: firstly, the initial particle set can be generated based on the simulation data (i.e., assuming the simulation model is a priori approximation of the real radiation field), thereby reducing the number of actual measurements required; secondly, during the update process, the system can compare the measured values ​​of the physical robot with the simulated values ​​of the virtual nodes. If the deviation between the two remains large, the parameter calibration module is triggered to adjust the radiation source intensity or shielding coefficient, gradually aligning the simulation model with the real environment. Specifically, the system constructs an auxiliary optimization objective: minimizing the root mean square error between the measured and simulated values, and updating the simulation model by adjusting simulation parameters (such as radiation source activity and material attenuation coefficient).

[0110] For example, parameters such as radiation source intensity, shielding material parameters, and detector efficiency in the simulated environment are uncertain. To address this issue, a real robot measures the radiation dose rate at a known location to obtain a calibration dataset.

[0111] ;

[0112] Reproduce the same location in a virtual environment and calculate the simulated dose rate. ;

[0113] Construct the objective function for parameter optimization:

[0114] ;

[0115] in, For real robots in The dose rate measured at the site, The vector of parameters to be optimized (including radiation source intensity, attenuation coefficient, and scattering coefficient). As a priori value, The regularization coefficient is used.

[0116] The optimal parameters are solved using the L-BFGS-B algorithm, and the simulation model is updated accordingly.

[0117] Iterative calibration involves collecting 10-20 new measurement points with the real robot each time, re-optimizing the parameters, and gradually bringing the simulation accuracy closer to the real environment.

[0118] In the aforementioned radiation field reconstruction method, a digital twin environment of the nuclear facility is constructed based on three-dimensional basic data to generate virtual scene data; radiation source parameters and shielding material parameters are received; based on the radiation source parameters and shielding material parameters, radiation field calculations are performed on the three-dimensional space of the virtual scene data to generate radiation dose distribution data; virtual robots are deployed, along with corresponding virtual nodes of physical robots; based on the virtual scene data and radiation dose distribution data, inspection paths for the virtual robots and virtual nodes are planned; the physical robots are controlled to perform inspection tasks according to the planned inspection paths of the virtual nodes; real-time simulation data of the virtual nodes is acquired, and real-time feedback data from the physical robots is received; based on the real-time simulation data and real-time feedback data, the radiation dose distribution data is verified and updated. Through this method, a digital twin environment of the nuclear facility is constructed based on three-dimensional basic data, eliminating the need for frequent equipment testing in real high-radiation sites, thus saving on-site testing costs and reducing personnel radiation exposure risks. The radiation field calculations based on radiation source parameters and shielding material parameters to generate dose distribution data solve the problem that traditional simulations cannot quantitatively describe the spatial distribution of radiation. Virtual nodes corresponding to physical robots are deployed, and path planning is performed based on virtual scenes and dose data. The physical robots are then controlled to execute tasks according to the virtual node paths, avoiding direct and blind deployment and improving inspection safety and task success rates. Simulation data from virtual nodes is acquired, and real-time feedback data from the physical robots is received. Radiation dose distribution data is verified and updated, and optimization is performed using a small number of real measurement points, significantly reducing the amount of real data required. A virtual-real closed-loop correction addresses the problem of large deviations between static simulation and the real environment, continuously improving the accuracy of the radiation field model and providing a reliable basis for subsequent inspection path optimization.

[0119] In an exemplary embodiment, step 206 includes: decomposing the virtual scene data into radiation units based on radiation source parameters, and setting material properties for each radiation unit based on shielding material parameters; deploying at least one detector in the three-dimensional space of the virtual scene data; for each detector, emitting a virtual ray towards the radiation source using the current detector as the observation point according to the radiation source parameters; traversing all target radiation units traversed by the virtual ray from the current detector to the radiation source based on a preset parallel computing strategy, recording the material type of each target radiation unit and the path length of the virtual ray within the target radiation unit; calculating the total attenuation factor corresponding to multiple energy levels based on the material type and path length of all target radiation units; for each energy level, calculating the radiation intensity corresponding to the energy level based on the initial source intensity of the energy level, the total attenuation factor corresponding to the energy level, the equivalent radius of the current detector, and the distance from the current detector to the radiation source; weighting and summing the radiation intensities corresponding to multiple energy levels based on the weight information corresponding to multiple energy levels to obtain the radiation dose rate of the current detector; and determining the radiation dose distribution data based on the radiation dose rate of at least one detector.

[0120] In this process, the three-dimensional space of the nuclear facility is voxelized. A radiation unit can be a single voxel or an aggregate of multiple voxels. Each radiation unit stores material properties, such as material density, linear attenuation (which can be stored separately for different energy levels), and scattering cross section.

[0121] In one optional implementation, at least one detector is deployed in the three-dimensional space of the virtual scene data. The detector can be a radiation detector model integrated into a virtual robot or virtual node, or it can be an independently set virtual detection point (e.g., a user-specified spatial location). The parameters for each detector include: detector type (omnidirectional, collimated, Compton camera, etc.), equivalent radius (default value 1 cm), energy response function, and spatial pose (position coordinates and orientation). The system supports the simultaneous deployment of multiple detectors, calculating the radiation dose rate at each point separately, and finally combining them to form radiation dose distribution data.

[0122] In one alternative implementation, for each detector, based on the radiation source parameters, a virtual ray is emitted from the current detector as the observation point towards the radiation source. Unlike traditional Monte Carlo methods that emit a large number of particles from the source to the detector, this method employs reverse ray tracing: starting from the detector's location, one (or a small number) virtual rays are emitted towards each known radiation source. Since the number of radiation sources is typically much smaller than the number of detectors or spatial points, reverse tracing significantly reduces the total number of rays. Each virtual ray is considered as the reverse of the energy transfer path from the source to the detector.

[0123] In one optional implementation, based on a preset parallel computing strategy, all target radiating cells traversed by the virtual ray from the current detector to the radiation source are iterated. The system employs the GPU-accelerated DDA (Digital Differential Analyzer) 3D ray tracing algorithm, executing the traversal of multiple rays in parallel on the massive threads of the graphics processor. The specific steps are as follows: An independent thread block or thread is allocated to each virtual ray on the GPU; the DDA algorithm is used to calculate the intersection points of the ray with the 3D radiating cell mesh; the DDA algorithm determines each radiating cell traversed by the ray sequentially through integer increments, which, compared to traditional geometric intersection methods, has a smaller computational load and is suitable for parallelization; for each target radiating cell traversed by the ray, the material type of the cell is recorded, and the path length of the ray within that cell (i.e., the distance from the ray's entry point to its exit point) is calculated; all recorded material types and path lengths are stored in a thread-local cache for subsequent attenuation calculations.

[0124] Since GPUs can process thousands of rays simultaneously, the computational complexity is significantly reduced compared to traditional serial methods. Reduced to ,in The total number of rays (equal to the number of radiation sources × the number of detectors). This represents the number of parallel threads on the GPU.

[0125] In one alternative implementation, the total attenuation factor corresponding to multiple energy levels is calculated based on the material type and path length of all target radiating elements. Considering the diversity of the gamma-ray energy spectrum (e.g., the cesium-137 main peak at 662 keV, and the cobalt-60 bimodal peaks at 1173 keV and 1332 keV), the system discretizes the energy spectrum of the radiation source into K energy levels (e.g., K=3 or K=5), assigning a weight to each energy level. and the linear decay coefficient at this energy For the k-th energy level, the total attenuation factor is... The calculation formula is:

[0126] ;

[0127] Where i represents all target radiation elements traversed by the traversing ray. Let be the linear decay coefficient of the material in the i-th unit at the k-th energy level. The path length of the ray within the radiation unit.

[0128] In one optional implementation, for each energy level, the radiation intensity corresponding to that energy level is calculated based on the initial source intensity, total attenuation factor, equivalent radius of the current detector, and distance from the detector to the radiation source. Specifically, an improved solid angle model is used to consider geometric attenuation. The radiation source is considered as a point source or surface source, emitting rays isotropically into space. The detector has a certain effective cross-sectional area, and the radiation intensity it receives is proportional to the solid angle subtended by the detector. The formula for calculating the radiation intensity of the k-th energy level at the detector location is:

[0129] ;

[0130] in, d is the equivalent radius of the detector (default value is 1cm), and d is the distance from the source to the detector. The initial source intensity for the kth energy level (i.e., the reference dose rate at a distance of 1m from the source without attenuation). This is the total attenuation factor.

[0131] In one optional implementation, the radiation dose rate of the current detector is obtained by weighting the radiation intensities corresponding to multiple energy levels based on their respective weight information. The weight information reflects the proportion of each energy level in the radiation source spectrum. The final dose rate calculation formula is:

[0132] ;

[0133] in, The total dose rate at the detector location. The weight of the k-th energy level. Let be the detector's response efficiency to the k-th energy level.

[0134] In one alternative implementation, radiation dose distribution data is determined based on the radiation dose rate of at least one detector. The system can deploy multiple virtual detectors throughout the three-dimensional space (e.g., at the center point of a voxel grid or user-specified key locations), repeating the aforementioned reverse ray tracing process to obtain the dose rate at each virtual detector location. The dose rates of these discrete points are then interpolated (e.g., trilinear interpolation or radial basis function interpolation) to fill the entire space, forming a continuous three-dimensional scalar field. This scalar field represents the radiation dose distribution data and can be output as a voxelized grid. For regions with the same or similar radiation characteristics (e.g., within the same radiation cell), where the dose rate changes gradually, it is unnecessary to deploy a detector at each voxel. The dose rate can be calculated only at the representative point of the radiation cell and then assigned to all voxels within the cell, further saving computational resources.

[0135] In one alternative implementation, for multi-source scenarios (i.e., virtual scenarios with multiple radiation sources), the system calculates the contribution of each radiation source for each detector separately, and then sums the contributions of all sources. Since the radiation field has linear superposition, the total dose rate equals the sum of the individual contributions of each source. The system can simultaneously calculate the rays from all sources to all detectors in parallel and perform the reduction and summation on the GPU.

[0136] In this embodiment, virtual scene data is decomposed into radiation units based on radiation source parameters and material properties are set. This efficiently and physically realistically calculates the radiation dose rate at any detection point, thereby generating radiation dose distribution data across the entire space. It balances real-time computation with physical accuracy (considering material attenuation, geometric attenuation, and energy spectrum grading), providing a high-fidelity radiation field foundation for subsequent path planning and virtual-real fusion verification.

[0137] In an exemplary embodiment, the radiation dose rate of the current detector is obtained by weighted summation of the radiation intensities corresponding to multiple energy levels based on the weight information corresponding to each energy level. This includes: weighted summation of the radiation intensities corresponding to multiple energy levels based on the weight information corresponding to each energy level to obtain the initial dose rate of the current detector; and superimposing a scattering correction term on the initial dose rate to obtain the radiation dose rate of the current detector.

[0138] After completing ray tracing and attenuation calculations, the simulation server calculates the direct dose rate without scattering correction (i.e., only considering the dose rate contributed by γ photons that arrive directly at the detector without collision) by weighted summation according to energy levels. Taking into account the Compton scattering phenomenon that occurs when γ photons penetrate shielding materials and media, the contribution of scattered photons is superimposed on the initial dose rate as a correction term, and finally the total dose rate of the detector that is closer to the real physical process is obtained.

[0139] In one alternative implementation, the Compton scattering probability is calculated for each target radiating element (voxel or radiating element) along the ray path. The formula for calculating the scattering probability is:

[0140] ;

[0141] in, The Compton scattering cross section, This refers to the material density (unit: g / cm³). This represents the path length of the ray within the radiation unit. The Compton scattering cross section is related to the atomic number of the material and the energy of the incident photon; the system can pre-build a lookup table of scattering cross sections for different materials based on the energy level.

[0142] Because calculating multiple scattering precisely involves a huge computational burden, an empirically-based simplified model is used, approximating the scattering contribution as 10%–30% of the direct intensity. The specific proportionality can be pre-calibrated based on the actual operating conditions of the nuclear facility or calibrated using offline MCNP simulation results. For example, for high-density, high-atomic-number materials (such as lead), the scattering contribution is relatively small (approximately 10%); for low-density, low-atomic-number materials (such as water and concrete), the scattering contribution is relatively large (up to 30%). Simultaneously, the energy of the scattered photons is typically reduced to 50%–70% of their original energy. This energy reduction affects the detector's response efficiency (because the detector's energy response curve usually varies with energy). When superimposing the scattering terms, the system re-queries the corresponding detector response efficiency based on the scattered energy.

[0143] The final radiation dose rate is obtained by adding a scattering correction term to the initial dose rate. The calculation formula is:

[0144] ;

[0145] in, This is the initial dose rate (i.e., the direct dose rate). is the scattering correction coefficient (ranging from 0.1 to 0.3, with different coefficient values ​​available for different materials or regions), and is the effective dose rate after the scattered photons pass through the detector's energy response weighted.

[0146] In this embodiment, the initial dose rate is obtained by weighted summation based on energy levels, and then a scattering correction term based on Compton scattering probability and material density estimation is superimposed to finally obtain the radiation dose rate of the current detector. A good balance is achieved between computational efficiency and physical accuracy, avoiding complex multiple scattering Monte Carlo simulations and significantly improving the systematic underestimation problem when only direct radiation is considered, providing a more reliable physical basis for the generation of subsequent radiation dose distribution data.

[0147] In an exemplary embodiment, virtual scene data is decomposed into radiation units based on radiation source parameters, including: establishing an octree mesh for the virtual scene data; adaptively dividing the octree mesh according to the radiation source parameters, based on the preset radiation source location and the estimated dose gradient, to determine multiple voxels; and clustering and merging the multiple voxels based on a preset voxel clustering algorithm to obtain several radiation units.

[0148] After acquiring virtual scene data from the digital twin environment of the nuclear facility, the simulation server first constructs an octree mesh structure covering the entire three-dimensional space. An octree is a hierarchical tree-like data structure where each node represents a cubic spatial region, the root node corresponds to the entire bounding box of the nuclear facility, and each non-leaf node can be recursively divided into eight equally sized child nodes (sub-cubes). Through this hierarchical organization, the system can dynamically adjust the spatial resolution in subsequent steps based on the radiation field distribution characteristics, achieving a balance between computational accuracy and efficiency.

[0149] In one optional implementation, when building the octree mesh, the system first determines the size of the root node based on the geometric extent of the virtual scene data (e.g., the length, width, and height are taken as the side lengths of the 3D bounding box of the core facility). The root node is initially divided into eight child nodes, and each child node is further subdivided as needed. The depth of the octree (i.e., the maximum number of subdivision levels) can be set by the user or automatically determined based on the minimum voxel size (e.g., when the depth is 8 levels, the minimum voxel side length is approximately the total side length / 256). The system stores the following information for each octree node: the spatial bounding box in which the node is located, the node level, child node pointers (if any), and the mean or gradient value of the radiation dose rate calculated subsequently.

[0150] In one optional implementation, based on the radiation source parameters, the octree mesh is adaptively divided according to the preset radiation source location and the estimated dose gradient to determine multiple voxels. Specifically, the system estimates the dose gradient distribution throughout the space based on the source location (e.g., spatial coordinates of a point source), source activity, and the initially estimated dose rate decay trend with distance in the radiation source parameters. In regions close to the radiation source, the dose rate changes drastically with distance (large gradient), requiring a finer voxel resolution to capture rapid changes in the radiation field; in regions far from the radiation source or through thick shielding, the dose rate changes gradually (small gradient), allowing for a coarser voxel resolution.

[0151] The adaptive partitioning strategy is as follows: For each octagonal leaf node (currently considered a candidate voxel), the system calculates the distance from the node's center to the radiation source based on the radiation source location, and estimates the dose rate at the node and the dose rate difference between adjacent nodes based on the shielding material along the path. If one of the following conditions is met, the node is further subdivided (i.e., its eight child nodes are activated as new leaf nodes): ① The estimated dose rate is higher than a high dose threshold (e.g., 10 mSv / h); ② The relative difference in estimated dose rates between the node's adjacent sibling nodes exceeds a preset gradient threshold (e.g., 20%); ③ The node contains the radiation source itself or known shielding boundaries, holes, or other geometric features. The subdivision process is recursively performed until a preset maximum depth or minimum voxel size (e.g., 0.1 m) is reached. After subdivision, all leaf nodes are the finally determined voxels. Through this adaptive partitioning, the system generates fine voxels with a size of 0.1m in the high-dose region, voxels with a size of 0.5m in the medium-dose region (1~10mSv / h), and voxels with a size of 2m in the low-dose region (<1mSv / h), thereby significantly reducing the total number of voxels while ensuring accuracy.

[0152] In one alternative implementation, the number of voxels may still be large (e.g., in the millions) after the system completes adaptive voxel partitioning. To further reduce the computational load of subsequent ray tracing and radiation field calculations, a voxel clustering algorithm is introduced to merge adjacent voxels with similar radiation characteristics (i.e., similar material properties, dose rate values, or attenuation coefficients) into larger radiation units. The specific steps are as follows: Initialize each voxel as an independent cluster; traverse all voxels, and for each voxel, check its adjacent voxels in six directions (or twenty-six neighborhoods); if the adjacent voxels have the same material type as the current voxel (e.g., both are concrete), and the estimated dose rate ratio is within a preset range (e.g., 0.8~1.2), and do not cross the boundary of the radiation source or shielding layer, then merge the clusters containing the two voxels; repeat the above steps until no new merging occurs; define each cluster formed in the end as a radiation unit. A radiation unit can be a single voxel (if the voxel cannot merge with any adjacent voxels, such as at the center of a high-dose hotspot or at a gap in the shielding layer), or it can be an aggregate of multiple voxels (for example, in a low-dose homogeneous region far from the radiation source, hundreds of voxels may merge into a single radiation unit).

[0153] For example, each radiative element stores the following attributes: the element's spatial bounding box, a list of voxels within the element, the element's representative location (e.g., centroid), material density, linear attenuation coefficient (stored separately for each energy level), and scattering cross section. Through clustering, the number of radiative elements is typically smaller than the original number of voxels, significantly reducing the number of basic elements that need to be traversed and computed during ray tracing, while preserving the fineness at the individual voxel level in high gradient regions.

[0154] In this embodiment, an octree mesh is established for the virtual scene data. Adaptive voxel partitioning is performed based on radiation source parameters and estimated dose gradients. Then, a voxel clustering algorithm is used to merge similar voxels to form radiation units, achieving intelligent optimization of spatial resolution in radiation field calculation. This ensures the fine modeling accuracy of high-dose, high-gradient regions while significantly reducing computational redundancy in low-dose, low-gradient regions, laying an efficient data structure foundation for subsequent real-time ray tracing and dose rate calculation.

[0155] In an exemplary embodiment, after determining radiation dose distribution data based on the radiation dose rate of at least one detector, the method further includes: determining the dose contribution of each radiation unit to other radiation units within a preset range based on the radiation dose distribution data, and constructing a radiation contribution lookup table.

[0156] After dividing the radiation elements and setting the material properties, the simulation server uses existing radiation dose distribution data (which can be the preliminary dose field obtained through ray tracing and attenuation calculations, or the dose field generated in the previous iteration) to calculate the dose contribution value of each radiation element as a virtual source point to all other radiation elements within a certain spatial range (e.g., within 100m). These contribution values ​​are stored in a look-up table (LUT) for subsequent rapid querying and incremental updates. In this way, when it is necessary to query the dose contribution between the same pair of radiation elements multiple times (e.g., repeatedly evaluating the dose rate at different locations in path planning or multi-robot collaborative simulation), the results can be obtained directly from the table without repeatedly performing ray tracing calculations, thus significantly improving the efficiency of real-time simulation.

[0157] In one optional implementation, after constructing a radiation contribution lookup table, the system uses this table to quickly calculate the dose rate at any location during actual operation. For example, when it is necessary to query the dose rate at the current position P of the virtual robot, the system first determines the radiation element j to which P belongs (or the nearest radiation element to P), then iterates through all source elements i (or only iterates through the source elements in the lookup table that contribute to j), obtains the dose rate from the lookup table, and accumulates it to obtain the total dose rate. If the location to be queried is not a representative point of a radiation element, a more accurate result can be obtained through spatial interpolation of the contribution values ​​of neighboring elements. The computational complexity of this query process is much lower than that of real-time ray tracing, thus supporting high frame rate real-time simulation.

[0158] In this embodiment, the dose contribution of each radiation unit to other radiation units within a preset range is determined based on radiation dose distribution data, and a radiation contribution lookup table is constructed, realizing the pre-storage and rapid reuse of radiation field calculation results. This provides efficient data support for large-scale multi-robot collaborative simulation, path planning iteration, and real-time verification of virtual-real fusion that require frequent dose rate queries.

[0159] In one exemplary embodiment, the physical robot carries multiple different types of target detectors; acquiring real-time simulation data of the virtual node includes: determining the sensitivity function corresponding to each target detector based on a pre-built detector type library; determining the initial dose rate at the location of the virtual node based on radiation dose distribution data; correcting the initial dose rate based on the azimuth angle of the source relative to the detector coordinate system and the sensitivity function corresponding to each target detector to obtain the effective dose rate corresponding to each target detector; performing statistical fluctuation simulation based on the effective dose rate corresponding to each target detector to obtain the simulated dose rate; and fusing the simulated dose rates corresponding to each target detector to obtain the real-time simulation data of the virtual node.

[0160] The simulation server pre-builds a detector type library containing various common radiation detector types and their parameterized response models. This library is used to simulate the directionality, energy response, statistical fluctuations, and dead time characteristics of different detectors in a virtual environment. When a specific type of radiation detector is configured on a virtual node (or virtual robot), the system loads the corresponding sensitivity function and parameters from the detector type library. Combining this with the radiation dose distribution data at the current location of the virtual node and the spatial geometric relationship between the radiation source and the detector, the system calculates a physically accurate simulated dose rate output. It also supports multi-detector data fusion to improve the accuracy of radiation source localization and dose estimation.

[0161] In one alternative implementation, the pre-built detector type library includes at least:

[0162] Type 1 (Omnidirectional Detector): For example, a NaI scintillator detector, whose sensitivity function is an isotropic constant, i.e. This indicates that the detector's response is the same in all directions and does not change with the incident direction.

[0163] Type 2 (Collimating Detector): For example, the LaBr3 detector with a lead collimator, whose sensitivity function is related to the incident direction and the collimator field of view. The expression is:

[0164] ;

[0165] in, Half of the field of view The angle between the incident direction and the detector axis.

[0166] Type 3 (Compton camera): For example, a dual-layer CZT (cadmium zinc telluride) detector, whose sensitivity function takes into account the Compton scattering angle, and is expressed as:

[0167] ;

[0168] in, The differential cross section for Compton scattering. This refers to the intrinsic efficiency of the detector. The energy of the incident photon. , The linear attenuation coefficient of the detector material. The thickness of the detector's sensitive layer.

[0169] Type 4 (Encoded Aperture Imaging Detector): Its sensitivity function is the convolutional response of a two-dimensional encoded mask.

[0170] The system stores the parameters of the above detector types (such as field of view half angle, intrinsic efficiency, energy response range, angular resolution, dead time constant, etc.) in the form of configuration files (JSON / YAML) or database tables. Users can select the required detector type and specify parameter values ​​in the configuration of virtual robots or virtual nodes.

[0171] In one alternative implementation, the simulation server obtains the dose rate value at a given location by querying or interpolating from radiation dose distribution data (a three-dimensional scalar field) based on the virtual node's current spatial coordinates within the digital twin environment (e.g., synchronized from real-time feedback data from the physical robot). This value is denoted as... The initial dose rate can be understood as the theoretical dose rate of an ideal omnidirectional detector (non-directional, non-statistical fluctuation, and non-dead time) at its current location. The system first needs to determine the spatial geometric relationship between the radiation source and the virtual node (detector). Since there may be multiple radiation sources in a nuclear facility scenario, the system calculates the azimuth angle (i.e., the angle between the incident direction and the detector axis) of each radiation source relative to the detector, and then calculates the directional weight of each radiation source relative to the detector based on the detector's sensitivity function. For example, the pose information of the virtual node (including position and orientation) and the local installation parameters of the detector mounted on the virtual node are obtained. For each radiation source in the scenario (whose position is known), the azimuth angle of the source relative to the detector coordinate system is calculated. Based on the detector type, the corresponding sensitivity function is queried, and directional weights are applied to obtain the effective dose rate: .

[0172] In one alternative implementation, the readings of the real radiation detector are random, primarily due to statistical fluctuations in radioactivity (following a Poisson distribution) and electronic noise (Gaussian distribution). Furthermore, the detector suffers from a dead-time effect, where some events are lost due to pulse accumulation at high count rates. The steps for simulating these effects are as follows:

[0173] 1. Continuous dose rate Convert to count rate The relationship between dose rate (in Sv / h) and count rate (in cps) depends on the detector's calibration factor. (For example, obtained through offline calibration), and the user can specify the sampling time. (e.g., 1 second).

[0174] 2. Simulate Poisson statistical fluctuations. Sample a random integer from the Poisson distribution as the actual observation count: .

[0175] 3. Add electronic noise. Electronic noise typically manifests as Gaussian white noise superimposed on the count, and its standard deviation is correlated with the count level or is a fixed value, such as the standard deviation. .

[0176] 4. Simulate detector dead time effect, when the count rate exceeds During CPS, the actual count ,in. It is the dead time constant (typical value 1-10 μs).

[0177] In one alternative implementation, when a virtual node (or physical robot) carries multiple detectors of different types (e.g., one omnidirectional detector for background monitoring and four collimating detectors forming a detector array for orientation localization), the system calculates the simulated dose rate for each detector separately and combines these readings into an observation vector. For example, the readings are weighted according to the confidence level of each detector to obtain a comprehensive dose estimate. This observation vector serves as part of the virtual node's real-time simulation data for subsequent radiation source localization, dose field reconstruction, or path planning.

[0178] For example, the location and intensity of the radiation source are inferred by using readings from multiple detectors and response models for each detector. Readings are calculated independently for each detector. Construct observation vectors The location and intensity of the inverted radiation source were estimated using maximum likelihood estimation.

[0179] ;

[0180] in, The likelihood function is calculated based on the detector response model. The sum of factors that maximize the likelihood function is obtained through optimization algorithms (such as gradient descent or grid search), leading to a more accurate estimate of the radiation source parameters.

[0181] In one alternative implementation, the multi-detector fusion results can also be packaged together with other real-time simulation data of the virtual node (such as pose, lidar point cloud, camera images, etc.) and sent to the path planning module or parameter calibration module through the communication interface to guide the next action of the physical robot.

[0182] In this embodiment, based on a detector type library, sensitivity function, directionality correction, statistical fluctuation simulation, and multi-detector data fusion, the system can provide virtual nodes with highly realistic radiation measurement simulation data that possesses statistical characteristics and supports complex detector types. This simulation data can not only be used for algorithm training and verification in the virtual environment, but also compared with real feedback data from physical robots, supporting online calibration and virtual-real fusion calibration, significantly improving the physical reliability of radiation sensing simulations.

[0183] In an exemplary embodiment, each radiation unit maintains a probability distribution of radiation dose rate; step 210 includes: determining the information entropy of each radiation unit based on the probability distribution of radiation dose rate of each radiation unit; constructing a multi-robot joint observation gain function including robot detection range overlap penalty; using the Hungarian algorithm to solve for the optimal task allocation to maximize the total information gain corresponding to the multi-robot joint observation gain function; and planning the inspection path of the virtual robot and virtual node based on the optimal task allocation.

[0184] After the simulation server generates the initial radiation dose distribution data, the initial simulation model may contain uncertainties (e.g., the prior values ​​of radiation source parameters and shielding material parameters are not accurate enough). Therefore, it is necessary to collect more real or simulated measurement data through multi-robot collaborative exploration to reduce the uncertainty of the radiation field model. To this end, the system introduces an information entropy-driven task allocation mechanism, treating radiation units in three-dimensional space as targets to be observed. The radiation dose rate of each radiation unit is modeled as a random variable and its probability distribution is maintained. By calculating the information entropy, the uncertainty of the current perception of the dose rate of the unit is quantified, and then the detection paths of multiple robots (including virtual robots and virtual nodes corresponding to physical robots) are planned to minimize the uncertainty of the overall radiation field with limited movement and measurement resources.

[0185] In one alternative implementation, the three-dimensional space is divided into a voxel mesh, and the system maintains a probability distribution of the radiation dose rate at each radiation element. (Initially a uniform distribution). For each radiating unit, calculate the information entropy based on its current probability distribution:

[0186] ;

[0187] Information entropy The larger the value, the more uncertain the understanding of the dose rate of that unit, meaning that the unit has higher exploratory value. High-entropy regions typically appear far from existing measurement points, near radiation source boundaries, or in areas with complex shielding structures that cause drastic dose rate changes.

[0188] In one alternative implementation, the system needs to decide which candidate locations to assign multiple robots (virtual robots and / or virtual nodes) to for measurement to maximize overall information gain. Directly selecting the maximum entropy location independently for each robot may result in multiple robots heading to the same area, wasting measurement resources. Therefore, the joint observation gain function introduces an overlap penalty term to encourage robots to explore in a dispersed manner. The multi-robot joint observation gain function is defined as follows:

[0189] ;

[0190] Where M is the total number of robots, Let be the candidate position (or candidate path endpoint) of the i-th robot. The center of voxel v, The information entropy of voxel v, For spatial attenuation scale, The robot detection range overlap penalty is used to measure the degree of overlap between the detection ranges of multiple robots.

[0191] In one alternative implementation, the Hungarian algorithm is used to solve for the optimal task allocation. To maximize the overall information gain, after obtaining the optimal detection position for each robot, the system needs to plan an inspection path from the current position to the target position for each robot. The path planning process, for example, uses a preset algorithm (such as RRT or DWA) based on virtual scene data (obstacle map) and radiation dose distribution data (cost map) to generate a path that safely and efficiently reaches the target waypoint from the starting point, while minimizing the cumulative radiation dose along the path. For virtual robots, path planning is performed entirely in the simulation environment, and the generated path is directly used to control the movement of the virtual robot; for virtual nodes (representing physical robots), the planned path, after safety verification, is sent to the real physical robot for execution via a communication link. During movement, the robot travels along the predetermined path, and upon reaching the target position, performs radiation measurement (for virtual robots, the simulated dose rate at that position is obtained by querying the radiation dose distribution data or ray tracing; for physical robots, the measured dose rate is obtained through their real detectors). The measurement results are used to update the probability distribution of the corresponding radiation unit, thereby reducing information entropy. Subsequently, the system can recalculate the information entropy distribution and enter the next round of task allocation and path planning iterations, achieving continuous refinement of the radiation field model.

[0192] In this embodiment, a multi-robot joint observation gain function with overlap penalty is constructed based on the information entropy of each radiation unit. The optimal task allocation is solved using the Hungarian algorithm, and inspection paths are planned for virtual robots and virtual nodes based on the allocation results. This achieves multi-robot collaborative active exploration and information-driven task allocation. It can intelligently guide robots to the most information-valuable areas, avoiding redundant detection and significantly improving the efficiency and accuracy of radiation field reconstruction, providing an efficient exploration strategy for intelligent inspection of nuclear facilities.

[0193] In an exemplary embodiment, the method further includes: constructing a physical degradation model, a radiation source evolution model, and a shielding failure simulation model corresponding to the virtual scene data; and dynamically updating the radiation dose distribution data based on the outputs of the physical degradation model, the radiation source evolution model, and the shielding failure simulation model.

[0194] The simulation server takes into account the physical degradation phenomena that nuclear facilities undergo during long-term operation, such as equipment corrosion, pipeline leaks, and aging of shielding materials, as well as the evolutionary processes of radioactive decay of the radiation source itself, including the generation of decay chain particles and leakage diffusion. Static virtual scene data and radiation dose distribution data cannot reflect these real-world conditions that change over time. To address this, the system constructs a parameterized environmental evolution tool, including a physical degradation model, a radiation source evolution model, and a shielding failure simulation model. This tool dynamically updates the virtual scene data and radiation dose distribution data in a time-driven or event-triggered manner, enabling the digital twin environment to simulate gradual changes and sudden accidents throughout the entire lifecycle of a nuclear facility. This provides a testing and verification basis for the long-term adaptability and anomaly response capabilities of the robotic inspection algorithm.

[0195] In one optional implementation, a physical degradation model corresponding to the virtual scene data is constructed. The physical degradation model is used to simulate the geometric changes and structural performance degradation of metal equipment, pipes, containers, and other structures in nuclear facilities caused by mechanisms such as corrosion, deformation, and cracking. Specifically, it includes the following sub-models:

[0196] Corrosion effect: A rust texture is generated using Perlin noise, and the color of the metal surface is randomly modified (RGB shifts towards reddish-brown). The corrosion ratio increases linearly over time. ; For corrosion ratio, This is the initial corrosion ratio. Here is the corrosion rate constant;

[0197] Deformation effect: Applying internal pressure to the storage tank causes deformation at the apex. Simulates bulges and dents; Let be the vertex displacement vector. The original position of the vertex. The coordinates of the tank center are: The bulge amplitude coefficient, For the spatial dimensions of the bulge;

[0198] Crack effect: A crack network is generated using a Voronoi diagram. The crack width and depth increase over time, and a leakage event is triggered when a crack penetrates the network.

[0199] In one optional implementation, a radiation source evolution model corresponding to the virtual scene data is constructed. The radiation source evolution model is used to simulate the decay of the radionuclide itself, the generation of decay chain progeny, and the spatial distribution changes of radioactive material due to container leakage. Specifically, it includes the following sub-models:

[0200] Radioactive attenuation: ,in, For activity, Initial activity, The decay constant;

[0201] Secondary source generation: Considering decay chains (e.g., U-238→Th-234→Pa-234), the activity of daughter nuclides increases:

[0202] ;

[0203] in, For the activity of the offspring, Initial maternal activity The parent decay constant, The decay constant of the daughter body;

[0204] Leakage and diffusion: When a container leaks, radioactive material diffuses into the surrounding voxels, using the diffusion equation:

[0205] ;

[0206] in, This refers to the concentration of radioactivity. is the diffusion coefficient.

[0207] In one optional implementation, a shielding failure simulation model corresponding to the virtual scene data is constructed. The shielding failure simulation model is used to simulate the decrease or even complete failure of shielding capabilities of shielding materials (such as concrete walls, lead plates, boron-containing polyethylene boards, etc.) due to aging, radiation damage, mechanical impact, etc. The specific steps are as follows: Define the health status of the shielding material. When health is below the threshold ( The shielding capability decreases, and the linear attenuation coefficient is corrected to... In extreme cases, the shielding material may detach. The area becomes air, and radiation penetrates it directly. For initial health, The degradation rate constant is For effective linear attenuation coefficient, This is the initial linear decay coefficient.

[0208] In one alternative implementation, the system iteratively executes the following update process at discrete time steps (e.g., daily, weekly, or monthly): Based on the current simulation time, it updates the corrosion ratio, deformation displacement, crack width, and depth in the physical degradation model. If a crack penetration triggers a leakage event, the leakage information is passed to the radiation source evolution model. Based on the radiation source evolution model, the activity (decay and secondary source generation) of each radiation source is updated. If a leakage event occurs, the diffusion equation is solved, the radioactivity concentration of each voxel is updated, and a new voxel source term is added. Based on the shielding failure simulation model, the health and linear decay coefficient of each shielding material are updated. If a detachment event occurs, the material properties of the corresponding radiation element are modified. Based on the updated radiation source parameters (activity, location, energy spectrum), shielding material parameters (decay coefficient), and virtual scene geometry (deformed mesh, newly added voxels), the radiation dose distribution data is recalculated. To improve efficiency, the system can employ an incremental update strategy: performing local ray tracing or table lookup corrections only on affected areas (e.g., near the leakage point, shielding failure area, voxels with significant changes in radiation source activity), instead of recalculating the entire scene. The recalculated radiation dose distribution data overwrites the old data and updates all modules that depend on it (such as path planning and virtual sensor simulation). The timestamp and summary of each update are recorded and stored in a database to support subsequent time-lapse and trend analysis.

[0209] In this embodiment, a physical degradation model, a radiation source evolution model, and a shielding failure simulation model are constructed respectively. Radiation dose distribution data is dynamically updated based on the outputs of these models, enabling the virtual environment to simulate the gradual changes and sudden events of nuclear facilities over medium- to long-term timescales. This method overcomes the limitations of static simulation environments and provides a powerful tool for verifying the adaptability, safety, and reliability of robotic algorithms throughout the entire lifecycle of nuclear facilities.

[0210] In one exemplary embodiment, refer to Figure 3 This application's embodiments employ a platform architecture comprising a virtual twin layer, a radiation reconstruction layer, and a collaborative verification layer. The virtual twin layer, based on 3D rendering engines such as Unity3D / Unreal, imports multi-source data including model data of the core facility, LiDAR point cloud data, and BIM information. The imported data undergoes coordinate alignment, meshing, and material assignment to generate a high-fidelity 3D scene (virtual scene data) containing objects such as walls, equipment, pipes, and shielding structures.

[0211] The radiation reconstruction layer employs an Adaptive Voxel Radiation Field Model (AVRF), which hierarchically voxels the 3D space based on the radiation gradient (fine mesh in high-dose regions, coarse mesh in low-dose regions), and merges similar voxels into radiative units using a voxel clustering algorithm. A GPU-accelerated fast ray tracing algorithm is used to emit virtual rays from the detector location towards the radiation source, traversing the radiative units and calculating the geometric and material attenuation at each energy level. Combined with the detector response function (supporting omnidirectional, collimated, Compton camera, and other types), directional response, statistical fluctuations (Poisson distribution), and dead-time effects are simulated to generate real-time radiation dose rate distribution data.

[0212] The collaborative verification layer deploys a multi-robot collaborative simulation platform, integrating the ROS2 interface to support mixed formation of virtual and physical robot virtual nodes. Based on virtual scene data and radiation dose distribution data, it performs information entropy-driven multi-robot task allocation and inspection path planning. Through a virtual-real closed loop: virtual nodes send the planned path to the physical robots for execution, and the real-time feedback data from the physical robots is returned to the simulation server for online calibration (parameter optimization) and verification and updating of radiation dose distribution data, enabling rapid migration and reliability verification of the algorithm from simulation to the field.

[0213] In an exemplary embodiment, during the offline training phase, model data of the nuclear facility, lidar point cloud data, BIM information, etc., are imported to construct a virtual environment; radiation source parameters (location, nuclide, activity) and shielding material parameters (density, attenuation coefficient) are configured; the AVRF model is run to generate an initial radiation field distribution and visualize a three-dimensional heat map; a virtual robot is deployed to train radiation avoidance strategies (e.g., deep reinforcement learning, PPO algorithm) and cooperative exploration algorithms; multiple task simulations are performed in the virtual environment to statistically analyze the success rate, cumulative dose, and completion time.

[0214] For example, a radiation avoidance strategy is trained in a virtual environment (using a deep reinforcement learning algorithm, such as the PPO algorithm), where the agent learns path planning to minimize the cumulative dose; a small amount of trajectory data (10-20 tracks) is collected in a real environment to construct a domain-adaptive loss function.

[0215] ;

[0216] Fine-tune the strategy network parameters to adapt them to the real radiation field distribution; online adaptive, after each task is performed by the real robot, the experience data is fed back to the simulation environment to continuously update the strategy.

[0217] in, For the real task loss, MMD (Maximum Mean Discrepancy) is a measure of the difference between the simulated and real feature distributions; The weighting coefficient is used to control the contribution of the MMD term to the total loss. This weighting coefficient can be preset according to the experiment or dynamically adjusted according to the adaptive learning results. This example does not limit its specific value.

[0218] In an exemplary embodiment, during the virtual-real fusion verification phase, a real robot enters the nuclear facility and collects radiation data from 10-20 calibration points; online calibration of simulation parameters is performed to match the virtual radiation field with the real distribution; a virtual-real hybrid formation (e.g., 2 real robots + 3 virtual robots) is initiated for collaborative exploration; real robots prioritize exploring high-risk areas, while virtual robots supplement and cover low-risk areas; virtual and real data are fused to reconstruct a complete three-dimensional map of the radiation field.

[0219] For example, a hybrid virtual-real robot formation: 1-2 physical robots (equipped with LiDAR, cameras, and radiation detectors) are deployed in a real nuclear facility, and 3-5 simulated robots are deployed in a virtual environment; the real robots send sensor data (point cloud, images, radiation readings) to the simulation server via 5G / WiFi; the simulation server injects the poses and sensor data of the real robots into the virtual environment, and the virtual robots perceive the real robots and collaboratively plan paths; the planning results of the virtual robots are fed back to the real robots, forming a virtual-real collaborative decision-making closed loop.

[0220] In one exemplary embodiment, during the field deployment phase, the algorithm, which has been verified in both virtual and real environments, is deployed to the field robot; the robot autonomously performs inspection tasks and updates the radiation field map in real time; the robot monitors the cumulative dose and automatically evacuates when the warning threshold is exceeded; the data is transmitted back to the digital twin platform to support remote expert analysis.

[0221] For example, this application can be applied to intelligent inspection systems in nuclear power plants. In critical areas such as reactor buildings, spent fuel storage areas, and radioactive waste treatment facilities, the system provides accurate radiation field prediction support for robot inspection path planning. Maintenance personnel pre-plan inspection routes in a virtual digital twin environment. The system automatically calculates the cumulative radiation dose of each path, selecting the path with the lowest dose to ensure that the robot's radiation exposure is kept to a minimum while completing the inspection task. Repeated practice in the virtual environment allows robot operators to become familiar with complex facility layouts, master abnormal operating condition response strategies, and optimize task execution processes under zero radiation risk conditions. The system's multi-robot collaborative function significantly improves inspection efficiency. High-radiation areas that previously required multiple operators to enter in batches can now be completed autonomously by robot teams, greatly reducing the cumulative dose to personnel. More importantly, the historical inspection data and radiation field evolution records accumulated by the system provide data support for predictive maintenance decisions, helping nuclear power plants to promptly detect equipment anomalies, assess shielding material aging, and predict potential leakage risks, shifting from passive response to proactive prevention, and continuously improving unit availability and operational safety.

[0222] As some early-stage nuclear power units under management gradually enter the decommissioning phase, this application plays a crucial role in decommissioning and dismantling projects. Based on historical operational data and activation calculation results, the system reconstructs the three-dimensional radiation field distribution of the decommissioning facility in a virtual environment, providing accurate dose assessment data for dismantling scheme design. Engineers can simulate different dismantling sequences, evaluate temporary shielding schemes, optimize cutting paths, pre-identify high-dose hotspots and potential hazardous areas, and formulate targeted protective measures on the digital twin platform. The virtual-real fusion verification mechanism ensures that the dismantling robot undergoes thorough testing before on-site deployment, avoiding task interruptions or equipment damage due to algorithmic defects, thus providing dual assurance for the progress and safety of the dismantling project. In the field of emergency response, the system provides highly realistic training scenarios for nuclear power companies' emergency drills, capable of simulating the rapid evolution of the radiation field under various accident conditions such as pipeline rupture, radioactive material leakage, and shielding failure. Emergency commanders and robot operators repeatedly practice accident response procedures in the virtual environment, training key capabilities such as rapid radiation field reconstruction, personnel evacuation route planning, and source term location and tracking. The low cost and high frequency of virtual drills have greatly improved the efficiency of emergency response team capacity building. In the event of a real accident, the response speed is faster, the decision-making is more scientific, and the personnel protection is more in place, thus building a solid technical defense line to ensure the safe and stable operation of nuclear power plants and the safety of the surrounding public.

[0223] This application, by constructing a virtual-real integrated intelligent inspection and verification platform, has a multi-dimensional positive impact on improving the productivity of nuclear power enterprises. First, the algorithm pre-verification mechanism in the virtual environment significantly shortens the on-site commissioning cycle of the robot system. What previously required months of on-site testing and iteration is compressed into weeks, reducing the time spent on unit shutdown maintenance windows and improving the availability of power generation equipment and annual power generation. Second, the multi-robot collaborative inspection function greatly improves the efficiency of radiation monitoring coverage. High-radiation areas that previously required multiple shifts of manual entry can now be quickly completed by robot teams, shortening the inspection cycle from monthly to weekly or even daily. Equipment anomalies are detected more promptly, and unplanned shutdowns are significantly reduced. Third, the virtual training environment reduces operator training costs. New employees and those changing positions can repeatedly practice complex operating procedures on the digital twin platform, significantly shortening the training cycle and making human resource allocation more flexible and efficient. Fourth, the precise reconstruction of the radiation field and dose optimization technology minimizes the time required for personnel to enter high-radiation areas, reducing work interruptions due to radiation protection requirements and improving the efficiency of maintenance and repair tasks. Fifth, the massive amount of simulation data and field verification data accumulated by the system provides strong support for nuclear power companies to formulate technical standards, summarize best practices, and build knowledge bases, promoting the digital and intelligent transformation of nuclear power operation and maintenance, and accelerating the transformation of new technologies from research and development to engineering applications.

[0224] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0225] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores virtual scene data or intermediate processing data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a radiation field reconstruction method.

[0226] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0227] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: constructing a digital twin environment of a nuclear facility based on three-dimensional basic data to generate virtual scene data; receiving radiation source parameters and shielding material parameters; calculating the radiation field in the three-dimensional space of the virtual scene data based on the radiation source parameters and shielding material parameters to generate radiation dose distribution data; deploying a virtual robot and deploying virtual nodes corresponding to the physical robot; planning the inspection paths of the virtual robot and virtual nodes based on the virtual scene data and radiation dose distribution data; controlling the physical robot to perform inspection tasks according to the planned inspection paths of the virtual nodes; acquiring real-time simulation data of the virtual nodes and receiving real-time feedback data from the physical robot; and verifying and updating the radiation dose distribution data based on the real-time simulation data and real-time feedback data.

[0228] In one embodiment, when the processor executes the computer program, it further performs the following steps: decomposing the virtual scene data into radiation units based on radiation source parameters, and setting material properties for each radiation unit based on shielding material parameters; deploying at least one detector in the three-dimensional space of the virtual scene data; for each detector, emitting a virtual ray towards the radiation source using the current detector as the observation point according to the radiation source parameters; traversing all target radiation units through which the virtual ray passes from the current detector to the radiation source based on a preset parallel computing strategy, recording the material type of each target radiation unit and the path length of the virtual ray within the target radiation unit; calculating the total attenuation factor corresponding to multiple energy levels based on the material type and path length of all target radiation units; for each energy level, calculating the radiation intensity corresponding to the energy level based on the initial source intensity of the energy level, the total attenuation factor corresponding to the energy level, the equivalent radius of the current detector, and the distance from the current detector to the radiation source; weighting and summing the radiation intensities corresponding to multiple energy levels based on the weight information corresponding to multiple energy levels to obtain the radiation dose rate of the current detector; and determining the radiation dose distribution data based on the radiation dose rate of at least one detector.

[0229] In one embodiment, when the processor executes the computer program, it further performs the following steps: based on the weight information corresponding to the multiple energy levels, it performs a weighted summation of the radiation intensities corresponding to the multiple energy levels to obtain the initial dose rate of the current detector; and it superimposes a scattering correction term into the initial dose rate to obtain the radiation dose rate of the current detector.

[0230] In one embodiment, when the processor executes the computer program, it further performs the following steps: establishing an octree mesh for virtual scene data; adaptively dividing the octree mesh based on radiation source parameters, according to the preset radiation source location and the estimated dose gradient, to determine multiple voxels; and clustering and merging the multiple voxels based on a preset voxel clustering algorithm to obtain several radiation units.

[0231] In one embodiment, when the processor executes the computer program, it also performs the following steps: determining the dose contribution of each radiation unit to other radiation units within a preset range based on radiation dose distribution data, and constructing a radiation contribution lookup table.

[0232] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the sensitivity function corresponding to each target detector based on a pre-built detector type library; determining the initial dose rate at the location of the virtual node based on radiation dose distribution data; correcting the initial dose rate based on the azimuth angle of the source relative to the detector coordinate system and the sensitivity function corresponding to each target detector to obtain the effective dose rate corresponding to each target detector; performing statistical fluctuation simulation based on the effective dose rate corresponding to each target detector to obtain the simulated dose rate; and fusing the simulated dose rates corresponding to each target detector to obtain the real-time simulation data of the virtual node.

[0233] In one embodiment, when the processor executes the computer program, it also performs the following steps: determining the information entropy of each radiation unit based on the probability distribution of the radiation dose rate of each radiation unit; constructing a multi-robot joint observation gain function that includes a penalty for overlapping robot detection ranges; using the Hungarian algorithm to solve for the optimal task allocation to maximize the total information gain corresponding to the multi-robot joint observation gain function; and planning the inspection paths for virtual robots and virtual nodes based on the optimal task allocation.

[0234] In one embodiment, when the processor executes the computer program, it also performs the following steps: constructing a physical degradation model, a radiation source evolution model, and a shielding failure simulation model corresponding to the virtual scene data, respectively; and dynamically updating the radiation dose distribution data based on the outputs of the physical degradation model, the radiation source evolution model, and the shielding failure simulation model.

[0235] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program performs the following steps: constructing a digital twin environment of a nuclear facility based on three-dimensional basic data to generate virtual scene data; receiving radiation source parameters and shielding material parameters; calculating the radiation field in the three-dimensional space of the virtual scene data based on the radiation source parameters and shielding material parameters to generate radiation dose distribution data; deploying a virtual robot and deploying virtual nodes corresponding to the physical robot; planning the inspection paths of the virtual robot and virtual nodes based on the virtual scene data and radiation dose distribution data; controlling the physical robot to perform inspection tasks according to the planned inspection paths of the virtual nodes; acquiring real-time simulation data of the virtual nodes and receiving real-time feedback data from the physical robot; and verifying and updating the radiation dose distribution data based on the real-time simulation data and real-time feedback data.

[0236] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: decomposing the virtual scene data into radiation units based on radiation source parameters, and setting material properties for each radiation unit based on shielding material parameters; deploying at least one detector in the three-dimensional space of the virtual scene data; for each detector, emitting a virtual ray towards the radiation source using the current detector as the observation point according to the radiation source parameters; traversing all target radiation units through which the virtual ray passes from the current detector to the radiation source based on a preset parallel computing strategy, recording the material type of each target radiation unit and the path length of the virtual ray within the target radiation unit; calculating the total attenuation factor corresponding to multiple energy levels based on the material type and path length of all target radiation units; for each energy level, calculating the radiation intensity corresponding to the energy level based on the initial source intensity of the energy level, the total attenuation factor corresponding to the energy level, the equivalent radius of the current detector, and the distance from the current detector to the radiation source; performing a weighted summation of the radiation intensities corresponding to multiple energy levels based on the weight information corresponding to each energy level to obtain the radiation dose rate of the current detector; and determining the radiation dose distribution data based on the radiation dose rate of at least one detector.

[0237] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the weight information corresponding to the multiple energy levels, it performs a weighted summation of the radiation intensities corresponding to the multiple energy levels to obtain the initial dose rate of the current detector; and it superimposes a scattering correction term into the initial dose rate to obtain the radiation dose rate of the current detector.

[0238] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: establishing an octree mesh for the virtual scene data; adaptively dividing the octree mesh based on radiation source parameters, according to the preset radiation source location and the estimated dose gradient, to determine multiple voxels; and clustering and merging the multiple voxels based on a preset voxel clustering algorithm to obtain several radiation units.

[0239] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the dose contribution of each radiation unit to other radiation units within a preset range based on radiation dose distribution data, and constructing a radiation contribution lookup table.

[0240] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the sensitivity function corresponding to each target detector based on a pre-built detector type library; determining the initial dose rate at the location of the virtual node based on radiation dose distribution data; correcting the initial dose rate based on the azimuth angle of the source relative to the detector coordinate system and the sensitivity function corresponding to each target detector to obtain the effective dose rate corresponding to each target detector; performing statistical fluctuation simulation based on the effective dose rate corresponding to each target detector to obtain the simulated dose rate; and fusing the simulated dose rates corresponding to each target detector to obtain the real-time simulation data of the virtual node.

[0241] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the information entropy of each radiation unit based on the probability distribution of the radiation dose rate of each radiation unit; constructing a multi-robot joint observation gain function that includes a penalty for overlapping robot detection ranges; using the Hungarian algorithm to solve for the optimal task allocation to maximize the total information gain corresponding to the multi-robot joint observation gain function; and planning the inspection paths for virtual robots and virtual nodes based on the optimal task allocation.

[0242] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: constructing a physical degradation model, a radiation source evolution model, and a shielding failure simulation model corresponding to the virtual scene data, respectively; and dynamically updating the radiation dose distribution data based on the outputs of the physical degradation model, the radiation source evolution model, and the shielding failure simulation model.

[0243] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0244] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0245] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0246] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for reconstructing a radiation field, characterized in that, The method includes: A digital twin environment for the nuclear facility is constructed based on 3D baseline data, generating virtual scene data; Receive radiation source parameters and shielding material parameters; Based on the radiation source parameters and the shielding material parameters, the radiation field of the virtual scene data in three-dimensional space is calculated to generate radiation dose distribution data; Deploy virtual robots and corresponding virtual nodes for physical robots; Based on the virtual scene data and the radiation dose distribution data, the inspection paths of the virtual robot and the virtual node are planned; The physical robot is controlled to perform inspection tasks according to the planned inspection path of the virtual nodes. Acquire real-time simulation data of the virtual node and receive real-time feedback data from the physical robot; The radiation dose distribution data is verified and updated based on the real-time simulation data and the real-time feedback data.

2. The method of claim 1, wherein, The step of calculating the radiation field in the three-dimensional space of the virtual scene data based on the radiation source parameters and the shielding material parameters to generate radiation dose distribution data includes: Based on the radiation source parameters, the virtual scene data is decomposed into radiation units, and based on the shielding material parameters, material properties are set for each radiation unit; Deploy at least one detector in the three-dimensional space of the virtual scene data; For each detector, based on the radiation source parameters, the current detector is used as the observation point, and a virtual ray is emitted towards the radiation source; Based on a preset parallel computing strategy, the virtual ray traverses all target radiation units it passes through from the current detector to the radiation source, and records the material type of each target radiation unit and the path length of the virtual ray within the target radiation unit. Calculate the total attenuation factor for each of the multiple energy levels based on the material type and path length of all target radiating elements. For each energy level, the radiation intensity corresponding to the energy level is calculated based on the initial source intensity of the energy level, the total attenuation factor corresponding to the energy level, the equivalent radius of the current detector, and the distance from the current detector to the radiation source. Based on the weight information corresponding to the multiple energy levels, the radiation intensity corresponding to the multiple energy levels is weighted and summed to obtain the radiation dose rate of the current detector; Radiation dose distribution data are determined based on the radiation dose rate of the at least one detector.

3. The method of claim 2, wherein, The step of weighted summation of the radiation intensities corresponding to the multiple energy levels based on their respective weight information to obtain the radiation dose rate of the current detector includes: Based on the weight information corresponding to the multiple energy levels, the radiation intensity corresponding to the multiple energy levels is weighted and summed to obtain the initial dose rate of the current detector; The radiation dose rate of the current detector is obtained by superimposing a scattering correction term on the initial dose rate.

4. The method of claim 2, wherein, The step of decomposing the virtual scene data into radiation units based on the radiation source parameters includes: An octree mesh is created for the virtual scene data; Based on the radiation source parameters, and according to the preset radiation source location and the estimated dose gradient, the octree mesh is adaptively divided to determine multiple voxels; Based on a preset voxel clustering algorithm, the multiple voxels are clustered and merged to obtain several radiative units.

5. The method of claim 2, wherein, After determining the radiation dose distribution data based on the radiation dose rate of the at least one detector, the method further includes: Based on the radiation dose distribution data, the dose contribution of each radiation unit to other radiation units within a preset range is determined, and a radiation contribution lookup table is constructed.

6. The method according to any one of claims 1 to 5, characterized in that, The physical robot carries multiple different types of target detectors; The process of acquiring the real-time simulation data of the virtual node includes: Based on a pre-built detector type library, the sensitivity function corresponding to each target detector is determined; Based on the radiation dose distribution data, the initial dose rate at the location of the virtual node is determined; Based on the azimuth angle of the source relative to the detector coordinate system and the sensitivity function corresponding to each target detector, the initial dose rate is corrected to obtain the effective dose rate corresponding to each target detector. Statistical fluctuation simulations were performed based on the effective dose rate corresponding to each target detector to obtain the simulated dose rate; The simulated dose rates corresponding to each target detector are fused to obtain the real-time simulation data of the virtual node.

7. The method of claim 2, wherein, Each radiation unit maintains a probability distribution of radiation dose rate; the planning of inspection paths for the virtual robot and the virtual nodes based on the virtual scene data and the radiation dose distribution data includes: The information entropy of each radiation unit is determined based on the probability distribution of the radiation dose rate of each radiation unit. Construct a multi-robot joint observation gain function that includes a penalty for overlapping robot detection ranges; The Hungarian algorithm is used to solve for the optimal task allocation, so as to maximize the total information gain corresponding to the multi-robot joint observation gain function; Based on the optimal task allocation, the inspection paths for the virtual robot and the virtual node are planned.

8. The method according to any one of claims 1 to 5, characterized in that, The method further includes: The physical degradation model, radiation source evolution model, and shielding failure simulation model corresponding to the virtual scene data are constructed respectively. The radiation dose distribution data is dynamically updated based on the outputs of the physical degradation model, the radiation source evolution model, and the shielding failure simulation model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images