High-voltage power distribution room robot state monitoring system and method based on digital twinning

Abstract

The invention discloses a high-voltage power distribution room robot state monitoring system and method based on digital twinning, and relates to the technical field of intelligent operation and maintenance of power equipment, and the system comprises a state monitoring module which comprises a multi-source data collection layer, an edge data processing layer, a digital twinning engine, an intelligent decision-making layer and a reverse control channel. The state monitoring module obtains state information of the high-voltage power distribution room robot and regulates and controls the high-voltage power distribution room robot. The high-voltage power distribution room robot comprises a short-axis mechanical arm, a long-axis mechanical arm, a flange plate, an aviation plug platform and an AGV unmanned transport vehicle. By installing the multi-source data acquisition layer, the multi-modal data synchronous acquisition function is achieved, the environment where the robot is located is rapidly obtained, the state of the high-voltage power distribution room robot is rapidly and accurately known, the single-source misjudgment rate is reduced, the working efficiency is improved, and manual troubleshooting is facilitated.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent operation and maintenance of power equipment, and particularly relates to a high-voltage distribution room robot state monitoring system and method based on digital twinning. BACKGROUND

[0002] Safe and stable replacement of equipment accessories of power plants, and real-time monitoring of the operation state of power plants are prerequisites for ensuring the stable operation of a power system. A high-voltage distribution robot has the advantages of improved safety, continuous operation capability, high accuracy, fast response speed and strong environmental adaptability in the operation and maintenance management of a power system. The robot can be remotely controlled, an operator can operate in a safe area, and the robot can also cooperate with other systems to improve the intelligent level of operation and maintenance.

[0003] However, due to the variable and complex state of the robot in the power operation process, it is difficult for the robot to independently complete the distribution task, and errors caused by errors need to be manually checked. There are certain problems in manual checking. The operator needs to understand the current state of the robot, and it is difficult for the operator to directly observe the running state of the robot through pure data. The error of the data measured by the sensor during manual checking may bring poor information to the operator, thereby causing the operator to misjudge. Manual checking has a certain lag, and long-time checking may cause the robot to stop, thereby causing the power system components to fail and causing more serious problems.

[0004] Patent document CN115464661B discloses a robot control system based on digital twinning. The above patent realizes the improvement of the efficiency of industrial production. The first attitude data of the industrial robot is obtained through the sensor unit, the attitude image of the industrial robot is obtained through the camera unit, the second attitude data of the industrial robot is obtained through the verification unit, and whether the first attitude data is excessively large error data is verified, thereby reducing the misjudgment of the staff.

[0005] The above patent obtains the first attitude data and the second attitude data of the industrial robot, judges whether the first attitude data is excessively large error data based on the obtained second attitude data, and controls the industrial robot based on the judgment result of the staff, which can effectively reduce the probability that the staff incorrectly stops the operation of the industrial robot due to the first attitude data being excessively large error data. However, the problem of quickly and accurately understanding the state of the robot has not been solved.

[0006] Therefore, the application provides a high-voltage distribution room robot state monitoring system and method based on digital twinning, which can quickly and accurately understand the state of the robot. SUMMARY

[0007] The application aims to provide a high-voltage power distribution room robot state monitoring system and method based on digital twinning to solve the technical problem of being difficult to quickly and accurately understand the state of the high-voltage power distribution room robot in the background art.

[0008] To achieve the above-mentioned purpose, the application provides the following technical scheme: a high-voltage power distribution room robot state monitoring system based on digital twinning, characterized in that: comprising a state monitoring module, the state monitoring module comprising a multi-source data acquisition layer, an edge data processing layer, a digital twinning engine, an intelligent decision-making layer and a reverse control channel, the edge data processing layer and the digital twinning engine performing data processing and digital simulation after receiving the data collected by the multi-source data acquisition layer, the intelligent decision-making layer generating operation decisions according to the results of data processing and digital simulation transmitted by the edge data processing layer and the digital twinning engine, the reverse control channel delivering the operation decisions generated by the intelligent decision-making layer to an execution terminal after review and verification, and the state monitoring module acquiring the state information of the high-voltage power distribution room robot and regulating and controlling the same.

[0009] The high-voltage power distribution room robot comprises a short-shaft mechanical arm, a long-shaft mechanical arm, a flange, a navigation plug-in and plug-out platform and an AGV unmanned transport vehicle, the short-shaft mechanical arm and the long-shaft mechanical arm are composed of a mechanical arm joint connecting rod and are provided with an end effector, the short-shaft mechanical arm and the long-shaft mechanical arm are rigidly connected with the navigation plug-in and plug-out platform by taking the flange as a mounting base, the end effector is provided with an image acquisition suite composed of a camera and a light source, and the navigation plug-in and plug-out platform is mounted on the AGV unmanned transport vehicle.

[0010] Preferably, the multi-source data acquisition layer comprises a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is connected with the visual acquisition unit and is used for collecting the pose and environmental data of the high-voltage power distribution room robot in real time.

[0011] The multi-modal sensor array comprises an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit, the angle sensor is arranged at the mechanical arm joint, an absolute photoelectric encoder is adopted to feed back data in real time to monitor the motion state of the mechanical arm, the transient ground voltage sensor with a bandwidth of 100 kHz-3 GHz and the infrared thermal imager with a resolution of 0.03℃ are mounted on the surface of the power distribution cabinet to capture the partial discharge signal and the temperature field distribution, and the environmental monitoring unit integrates a temperature and humidity sensor with an accuracy of ±1% RH to monitor the environmental parameters of the robot in real time.

[0012] The visual acquisition unit comprises a binocular stereo vision module, a laser profiler and an adaptive light compensation module; the binocular stereo vision module with a baseline distance of 80 mm±5% in combination with the laser profiler constructs a three-dimensional point cloud model of the device from physical scanning to point cloud registration to twin mapping, the point density is greater than 200 points / cm2, the color temperature of the adaptive light compensation module is 5000 K±300 K, the illuminance is adjustable between 50 lux and 1000 lux, and the imaging quality under complex lighting conditions is ensured.

[0013] Preferably, the edge data processing layer is connected with the multi-modal sensor array and the visual collection unit through time-triggered Ethernet (TTEthernet), the edge data processing layer acquires data collected by the multi-source data acquisition layer through Ethernet and analyzes and processes the data, and the edge data processing layer is internally provided with a dynamic data cleaning module and a feature fusion module;

[0014] The dynamic data cleaning module adopts a sliding window with a window size T=1 s±0.3 s to adaptively filter and an improved DBSCAN clustering algorithm with parameters ε=0.3 and MinPts=5 to eliminate electromagnetic interference noise, and the data cleaning efficiency is improved by 40%, and the feature fusion module constructs a space-time alignment algorithm, and through an EPOCH synchronizer with a precision of ±10 ns, vibration frequency spectrum, temperature gradient and partial discharge pulses with a frequency of 0-5 kHz are fused into a multi-dimensional tensor;

[0015] The sensor data is transmitted to the edge node through time-triggered Ethernet (TTEthernet), the clock synchronization accuracy is ±10 ns, the time sequence consistency of vibration, temperature and partial discharge data is ensured, the edge node performs data cleaning and feature extraction, and the data amount is compressed to 30% of the original value.

[0016] Preferably, the digital twin engine receives data collected by the multi-element data acquisition layer and processed by the edge data processing layer, performs digital simulation, and the digital twin engine includes a model construction and virtual-real synchronization mechanism, drives a simulation model through a virtual-real synchronization protocol, and the digital twin engine constructs a millimeter-level precision model based on a Unity platform, including a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module, the robot kinematics model includes 6 degrees of freedom joint parameters, the joint angle synchronization error is less than 0.5°, the electromagnetic-thermal coupling simulation module calculates the core eddy current loss by discretely solving Maxwell equation, the error is less than 3%, the digital model is constructed through physical data, and the stress nephogram and electric field distribution are updated once every 100 ms; the physical device is controlled through a reverse control instruction, the action instruction is issued after pre-rehearsal verification, the torsion deviation is less than 5%, and the virtual-real bidirectional synchronization is realized.

[0017] The cloud digital twin engine receives feature data, drives a simulation model to update, and generates a decision instruction, and the response delay is less than 150 ms.

[0018] Preferably, the intelligent decision layer integrates an operation order parser, a dynamic path planner, and a risk prediction matrix, and is configured with a task decomposition logic tree, a visual terminal, a three-dimensional heat map production module, and a fault prediction cloud display interface.

[0019] The operation order parser uses a 32-dimensional vector BERT model to extract operation item features to eliminate natural language ambiguity, and the parsing accuracy reaches 98.7%; the dynamic path planner integrates a DDPG reinforcement learning model to optimize the action sequence, and the reward function contains three elements of safety distance, energy consumption, and timeliness; the risk prediction matrix defines 12 types of high-risk operation combinations, such as closing with live voltage, to block real-time operation instructions with high risk and avoid existing safety hazards.

[0020] Preferably, the task decomposition logic tree of the intelligent decision layer includes an atomic operation unit library, an operation dependency graph, and a resource conflict detector, the atomic operation unit library contains multiple types of basic mechanical actions such as tightening, plugging, and detection, the operation dependency graph is modeled using Petri nets, and the resource conflict detector is implemented based on the banker's algorithm; the visual terminal of the intelligent decision layer generates three-dimensional heat maps and fault probability cloud maps; the fault prediction module of the intelligent decision layer integrates a wavelet packet-convolutional neural network WP-CNN model to realize collaborative diagnosis of mechanical transmission faults and electrical faults.

[0021] Preferably, the reverse control channel integrates a CRC32 check module, an operation logic review unit, and a physical feedback verification module.

[0022] After the control instructions are checked by CRC32 and reviewed by logic rules, they are issued to the execution terminal; after every two atomic actions are completed, the physical feedback and the simulation results are compared in real time by the physical feedback verification module, and the error tolerance is less than 3%; if the error tolerance exceeds the limit, a self-correction process is triggered.

[0023] Preferably, the method comprises the following steps:

[0024] S1: receiving an operation order task issued by the system, which contains at least 3 associated sub-tasks;

[0025] S2: acquiring pose data of the target related equipment through a multi-modal sensor array and a visual collection unit;

[0026] S3: preforming the operation process in the digital twin environment to identify potential mechanical interference areas;

[0027] S4: decomposing the operation into an atomic action sequence, such as positioning → clamping → rotating 120° → torque verification;

[0028] S5: issuing atomic action instructions to the execution terminal and synchronously updating the twin model state.

[0029] Preferably, the mechanical interference recognition of S3 adopts Unity physics engine collision detection to dynamically calculate the following parameters:

[0030] The distance between the mechanical arm and the power distribution cabinet, the overlap rate of the dual mechanical arm actuator workspace, and whether the joint acceleration exceeds the limited range.

[0031] Preferably, the atomic action sequence generation of S4 is specifically:

[0032] S41: The operation ticket parser extracts operation item key parameters;

[0033] S42: The DDPG model generates an initial action path;

[0034] S43: The risk prediction matrix intercepts high-risk operation combinations;

[0035] S44: Output atomic action sequence.

[0036] Compared with the prior art, the beneficial effects of the present application are:

[0037] 1. The present application realizes the function of multi-modal data synchronous acquisition by installing a multi-source data acquisition layer, quickly acquires the environment of the robot, quickly and accurately understands the state of the robot, reduces the single-source misjudgment rate, improves the work efficiency, reduces the labor intensity, and is convenient for manual troubleshooting;

[0038] 2. The present application realizes the function of millisecond-level virtual-real bidirectional synchronization by installing a digital twin engine, virtually and virtually simulates operation instructions, improves the accuracy of mechanical interference recognition, avoids the risk of mechanical arm collision, shortens the abnormal recognition response time, and reduces the fault false alarm rate;

[0039] 3. The present application realizes the function of anti-interference feature extraction by installing an edge data processing layer, reduces the influence of the electromagnetic environment on feature extraction, compresses the volume of original data, lightens data to speed up transmission speed, reduces transmission time, and improves effective information density;

[0040] 4. The present application realizes the function of atomic action level optimization by installing an intelligent decision-making layer, improves the task execution success rate of the robot, intercepts high-risk operation combinations, effectively avoids high-risk operations, reduces the rate of human error operation accidents, and ensures the safety of power distribution. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a front view structural diagram of the high-voltage power distribution robot of the present application;

[0042] Figure 2 is a digital twin algorithm flowchart of the present application;

[0043] Figure 3A mechanical arm structure schematic diagram of the present application;

[0044] Figure 4 An intelligent decision-making process schematic diagram of the present application.

[0045] In the figure: 1, short-axis mechanical arm; 2, long-axis mechanical arm; 3, end effector; 4, mechanical arm joint; 5, connecting rod; 6, photoelectric absolute encoder; 7, flange; 8, sailing plug-in and plug-out platform; 9, AGV unmanned transport vehicle. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0047] In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first" and "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0048] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "provided with", "connection" and the like should be broadly understood, for example, "connection" can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can also be electrical connection; can be directly connected, can also be indirectly connected through an intermediate medium; can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0049] Embodiment 1, please refer to Figure 1 、 Figure 2 、 Figure 3 and Figure 4 , a high-voltage distribution room robot state monitoring system based on digital twinning, comprising a state monitoring module, the state monitoring module comprises a multi-source data acquisition layer, an edge data processing layer, a digital twinning engine, an intelligent decision-making layer and a reverse control channel, the state monitoring module acquires the state information of the high-voltage distribution room robot and controls it;

[0050] The multi-source data acquisition layer includes a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is signal connected with the visual collection unit, the multi-modal sensor array includes: an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit; the visual acquisition unit includes: a binocular stereo vision module, a laser profiler and an adaptive light supplement module;

[0051] The edge data processing layer is connected with the multi-modal sensor array and the visual collection unit through time-triggered Ethernet, the edge data processing layer is built-in with a dynamic data cleaning module and a feature fusion module, the dynamic data cleaning module adopts a sliding window adaptive filter and an improved DBSCAN clustering algorithm, and the feature fusion module constructs a space-time alignment algorithm;

[0052] The digital twin engine includes a model construction and a virtual-real synchronization mechanism, a simulation model is driven through a virtual-real synchronization protocol, the digital twin engine constructs a millimeter-level precision model based on a Unity platform, including a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module;

[0053] The intelligent decision layer integrates an operation ticket parser, a dynamic path planner and a risk prediction matrix, is configured with a task disintegration logic tree, a visual terminal, a three-dimensional heat map production module and a fault prediction module, the operation ticket parser extracts operation item features by using a BERT model, the dynamic path planning integrates a DDPG reinforcement learning model, and the risk prediction matrix defines 12 types of high-risk operation combinations;

[0054] The reverse control channel integrates a CRC32 check module, an operation logic review unit and a physical feedback verification module;

[0055] Further, the whole process is deployed and implemented as follows:

[0056] Step 1: hardware system construction:

[0057] Mobile robot platform deployment: a collaborative robot chassis is adopted, which is equipped with a UWB positioning module and is configured with an electromagnetic interference shielding cover;

[0058] Industrial arm joint sensing system integration: four photoelectric absolute encoders are installed on each mechanical arm joint, which can monitor and return joint angle data to the upper computer in real time; a high-definition image acquisition kit is installed at the end effector of the short-shaft mechanical arm and the long-shaft mechanical arm, which can clearly capture the image details of the robot's visual angle, including but not limited to the state of the aircraft plug, the state information displayed by the electric cabinet, the opening and closing of the electric cabinet door, etc., and the data is uploaded in real time through the EtherCAT protocol with a transmission period of 1 ms;

[0059] Step 2: multi-source data acquisition and transmission:

[0060] Sensor array initialization: The photoelectric absolute encoder 6 determines the absolute initial position at the moment of power-on, and the reference point of the mechanical arm joint 4 position needs to be calibrated before power-on; the binocular vision module performs stereo calibration, the chessboard calibration board is 7x9, the square size is 30 mm±0.01 mm, and the re-projection error is less than 0.15 pixels.

[0061] Real-time data upload mechanism: Joint angle data is packaged as EtherCAT messages and uploaded to the edge node through the switch every 50 ms, with a time delay jitter of less than 5μs. A data buffer queue with a depth of 100 frames is established to deal with network burst congestion.

[0062] Step 3: Digital twin engine construction:

[0063] Three-dimensional model construction: Use SolidWorks to export the robot STEP assembly file, import it into the Unity 2021.3 engine; set the physical engine parameters: gravitational acceleration 9.80665 m / s², friction coefficient dynamically adjusted within 0.05-0.15; import the power distribution cabinet CAD model, and subdivide the mesh to 5 mm accuracy.

[0064] Virtual-real synchronization protocol configuration: Physical to virtual data mapping: establish Topic subscription through ROS Bridge; reverse control channel: use Protobuf protocol to package control instructions; synchronization verification: compare the virtual model with the actual joint angle after completing two atomic actions such as clamping and rotating, error threshold: rotation axis ±0.5°, linear axis ±0.2 mm.

[0065] Step 4: Intelligent decision-making module deployment:

[0066] Operation ticket parsing algorithm training: Based on the accumulated 10,000 historical operation tickets, a database is constructed, and natural language processing technology is used to parse the key step text in each operation ticket. Finally, structured information describing the complete operation process is extracted and output, where each operation is represented as one or more operation item entities containing core attributes such as "action type".

[0067] Dynamic robot motion planning: a learning framework is constructed to deeply integrate environmental perception and robot state. The state space is composed of three parts: 6D joint state of the robot: real-time feedback of joint angles, accurately describing the robot configuration; 6D end-effector pose: capturing the position and orientation of the end-effector in the form of position coordinates (x, y, z) and Euler angles; 128D environmental perception: extracting the geometric feature vector of the scene through a point cloud encoding network such as PointNet, and analyzing the spatial relationship between obstacles and the work target in real time. The neural network is trained based on the 140D state vector to generate a motion path that takes into account both time optimization and operational safety. The planning process needs to optimize two objectives simultaneously: timeliness: minimize the execution time of the trajectory under the premise of no collision, dynamically compressing the idle time of the movement process; safety: hard-embed collision avoidance mechanisms in the motion sequence, and calculate the Euclidean distance field SDF between the robot and the obstacle in real time. When the predicted distance is below the safety threshold, automatically trigger trajectory correction. The safety layer checks each path point through a fast distance query algorithm (such as BVH tree), and constrains the joint acceleration within the physical limit, such as ±180° / s², to prevent torque overload and hardware damage.

[0068] Implementation path: pre-train the policy network in a simulated environment using reinforcement learning, and the loss function combines a time penalty term (proportional to path length) and a collision risk term (inversely proportional to the minimum obstacle distance). When migrating to the physical robot, add an online adaptive module ADA to fine-tune the network parameters using real point cloud data. The final output is a smooth sequence of joint angle increments or an end-effector pose trajectory, ensuring that the robot can complete high-dynamic tasks with millimeter-level precision and second-level response speed.

[0069] Step 5: Control the flow of execution:

[0070] Operation rehearsal and verification: collect all environmental data obtained by sensors and current robot state data for data cleaning and reconstruction, and synchronize to the digital twin engine; read the operation ticket task and automatically generate a robot motion sequence (such as: move → lift → clamp → rotate 120° → release) through the intelligent decision-making module; execute the generated motion sequence step by step through the physics engine, if safety problems are detected, such as collision between two robot arms, collision between robot arm path and external environment, regenerate the robot motion sequence; if the digital twin simulation is successful, issue the action instruction to the entity. During the execution of the action instruction by the entity, the twin synchronizes the state of the physical entity and monitors the possibility of collision in real time.

[0071] Example 2, see Figure 1 、 Figure 2 、 Figure 3 and Figure 4The application discloses a high-voltage distribution room robot state monitoring system based on digital twinning, which comprises a state monitoring module, wherein the state monitoring module comprises a multi-source data acquisition layer, an edge data processing layer, a digital twinning engine, an intelligent decision-making layer and a reverse control channel, the state monitoring module acquires state information of the high-voltage distribution room robot and regulates and controls the state information.

[0072] The multi-source data acquisition layer comprises a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is signal-connected with the visual acquisition unit, the multi-modal sensor array comprises an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit, and the visual acquisition unit comprises a binocular stereo vision module, a laser profiler and an adaptive light supplement module.

[0073] The edge data processing layer is connected with the multi-modal sensor array and the visual acquisition unit through a time-triggered Ethernet, the edge data processing layer is internally provided with a dynamic data cleaning module and a feature fusion module, the dynamic data cleaning module adopts a sliding window self-adaptive filter and an improved DBSCAN clustering algorithm, and the feature fusion module constructs a space-time alignment algorithm.

[0074] The digital twinning engine comprises a model construction and a virtual-real synchronization mechanism, a simulation model is driven through a virtual-real synchronization protocol, the digital twinning engine constructs a millimeter-level precision model based on a Unity platform, and the millimeter-level precision model comprises a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module.

[0075] The intelligent decision-making layer is integrated with an operation ticket parser, a dynamic path planner and a risk prediction matrix, is configured with a task disintegration logic tree, a visual terminal, a three-dimensional heat map production module and a fault prediction module, the operation ticket parser extracts operation item features by adopting a BERT model, the dynamic path planner is integrated with a DDPG reinforcement learning model, and the risk prediction matrix defines 12 types of high-risk operation combinations.

[0076] The reverse control channel is integrated with a CRC32 check module, an operation logic review unit and a physical feedback verification module.

[0077] Further, in a 10 kV distribution room of a 500 kV hub substation, the system receives an operation ticket task: "replace a standby plug-in module of a 3# distribution cabinet, model HXJ-35 kV", the multi-source data acquisition layer immediately starts:

[0078] A photoelectric absolute encoder 6 with a resolution of 17 bits installed at four joints of the mechanical arm feeds back 6-degree-of-freedom joint angles in real time, data is uploaded through an EtherCAT protocol at a period of 1 ms, a host computer calculates real-time poses of a short-shaft mechanical arm 1 and a long-shaft mechanical arm 2, and the positioning accuracy is ±0.05 mm.

[0079] The transient voltage sensor with bandwidth of 100 kHz-3 GHz is deployed on the surface of the power distribution cabinet to capture partial discharge pulses with amplitude of 35 mV and rise time of 3 ns. The external thermal imager scans the temperature of the connector interface synchronously, and the NETD is 0.03℃. The highest temperature is 72.5℃, and the ambient temperature is 28℃.

[0080] The binocular vision module with a baseline of 82 mm is combined with a laser profiler with a joint accuracy of ±0.1 mm to perform three-dimensional reconstruction on the connector insertion and extraction platform 8, generating a point cloud model with a point density of up to 230 points / cm². The offset angle of the connector pin is identified to be 1.7°.

[0081] The edge data processing layer receives raw data through time-triggered Ethernet (TTEthernet) and performs key processing: a sliding window adaptive filter with a window size of T=0.9 s is used to eliminate 50 Hz power frequency interference in the power distribution room. The DBSCAN clustering algorithm is improved with ε=0.3 and MinPts=5 to eliminate electromagnetic noise, and the data cleaning efficiency is improved by 43%. The EPOCH synchronizer with an accuracy of ±10 ns aligns the time bases of each sensor. The vibration frequency spectrum, peak frequency of 2.3 kHz, temperature gradient ΔT / Δt=4.2℃ / s, and partial discharge pulse characteristics are fused into a 128-dimensional tensor, and the data volume is compressed to 28% of the original value.

[0082] The digital twin engine starts dynamic simulation on the Unity platform: import the robot CAD model in STEP format, set the joint friction coefficient to 0.08, and detect mechanical interference risks during the replacement process: the distance between the long-axis robot arm 2 motion trajectory and the power distribution cabinet door is only 6.8 cm, which is lower than the safety threshold of 10 cm, and automatically triggers trajectory correction. The electromagnetic-thermal coupling module solves Maxwell's equation, and the simulation shows that poor contact between the connector and the power distribution cabinet leads to concentrated eddy current loss, with a local temperature rise simulation value of 71.8℃.

[0083] The intelligent decision-making layer analyzes the operation ticket and generates atomic action sequences: the AGV unmanned transport vehicle 9 moves to the coordinates (X=3.215m, Y=1.503m) based on UWB positioning, with a positioning error of less than 2 mm; the short-axis robot arm 1 clamps the connector with a torque of 85 N·m±2% (verified by a torque sensor in real time); the long-axis robot arm 2 performs obstacle avoidance path: first rotate 42° to avoid the cabinet door, then translate 15 cm; measure the contact resistance after inserting the connector;

[0084] The risk prediction matrix intercepts the "live operation" instruction and forces the insertion of the "remote power-off" step. The reverse control channel issues instructions after CRC32 verification, and physical feedback verification is performed after each completion of 2 atomic actions. The joint angle deviation is 0.28°, which is less than the threshold of 0.5°. Compared with traditional manual operation, the efficiency is improved, and no collision warning is triggered.

[0085] Embodiment 3, please refer to Figure 1 , Figure 2 and Figure 4 , a high-voltage distribution room robot state monitoring system based on digital twinning, comprising a state monitoring module, the state monitoring module comprising a multi-source data acquisition layer, an edge data processing layer, a digital twinning engine, an intelligent decision-making layer and a reverse control channel, the state monitoring module acquires the state information of the high-voltage distribution room robot and regulates it;

[0086] The multi-source data acquisition layer comprises a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is connected with the visual acquisition unit, the multi-modal sensor array comprises: an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit; the visual acquisition unit comprises: a binocular stereo vision module, a laser profiler and an adaptive light supplement module;

[0087] The edge data processing layer is connected with the multi-modal sensor array and the visual acquisition unit through time-triggered Ethernet, the edge data processing layer is built-in dynamic data cleaning module and feature fusion module, the dynamic data cleaning module adopts sliding window adaptive filter and improved DBSCAN clustering algorithm, the feature fusion module constructs a spatio-temporal alignment algorithm;

[0088] The digital twinning engine contains model construction and virtual-real synchronization mechanism, and drives the simulation model through the virtual-real synchronization protocol, the digital twinning engine constructs a millimeter-level precision model based on the Unity platform, including a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module;

[0089] The intelligent decision-making layer integrates an operation ticket parser, a dynamic path planner and a risk prediction matrix, is configured with a task decomposition logic tree, a visual terminal, a three-dimensional heat map production module and a fault prediction module, the operation ticket parser extracts operation item features using a BERT model, the dynamic path planning integrates a DDPG reinforcement learning model, and the risk prediction matrix defines 12 high-risk operation combinations;

[0090] The reverse control channel integrates a CRC32 check module, an operation logic review unit and a physical feedback verification module;

[0091] Further, the system monitors that the main transformer core temperature of the model SFSZ-120000 / 220 exceeds the threshold alarm, and the threshold is set to 95 DEG C;

[0092] The multi-source data acquisition layer starts the emergency diagnosis mode: the infrared thermal imager scans to generate a temperature field distribution map, identifies the B-phase winding hot spot, the maximum temperature is 98.7 DEG C, and the transient ground voltage sensor synchronously captures a 3.2 MHz discharge pulse with a repetition rate of 120 times / s;

[0093] The vibration sensor array detects abnormal core vibration at a fundamental frequency of 100 Hz with an amplitude of 0.8 g; the environmental temperature and humidity sensor records the environmental parameters: temperature 42℃, humidity 85% RH;

[0094] The edge node performs multi-modal feature extraction: the improved DBSCAN algorithm separates electromagnetic noise from effective discharge pulses, and the sliding window filter extracts the temperature change trend; after the EPOCH synchronizer aligns the time scale, the 0-5 kHz vibration spectrum, temperature gradient field, and partial discharge pulse sequence are fused into a 256-dimensional tensor, which is compressed and uploaded to the cloud through a 5G private network, with a transmission delay of 18 ms;

[0095] The digital twin engine performs deep simulation analysis: the electromagnetic-thermal coupling module discretely solves Maxwell's equations to reconstruct the core eddy current loss cloud map, with a simulation error of 2.1%, and locks the fault point as the 15th turn insulation damage of the B-phase winding; the kinematics model simulates the approaching path of the mechanical arm, predicts that the minimum distance between the cooling fan maintenance opening and the mechanical arm is 8 cm, which has a mechanical interference risk, and adjusts the motion path of the mechanical arm in real time to avoid mechanical interference;

[0096] The intelligent decision layer activates the core algorithm: the fault prediction model integrates the wavelet packet decomposition of the vibration signal into 8 frequency bands, inputs the convolutional neural network, and the structure is Conv3D-32→MaxPool→Conv3D-64→FC-128. The output probability of mechanical fault bearing wear is 89%, and the probability of electrical fault insulation carbonization is 94%. The visualization terminal generates a three-dimensional heat map with a confidence level greater than 96% superimposed on the fault probability cloud map, and recommends the operation ticket: "start the standby fan, isolate the B-phase, and apply for maintenance";

[0097] The reverse control channel executes the treatment scheme: issues the fan start command, the fan speed is 1500 rpm, and the digital twin body synchronously updates the temperature field model. Every 120 seconds, the actual temperature drop rate is compared with the simulation temperature drop rate, and the deviation is less than 3% threshold;

[0098] The risk prediction matrix intercepts the "isolating with load" instruction and adds the "transferring load" step. After treatment, the hotspot temperature drops to 61.4℃, and the fault probability cloud map shows that the risk level decreases from "danger" to "attention".

[0099] Example 4, please refer to Figure 1 、 Figure 2 and Figure 4 A high-voltage distribution room robot state monitoring system based on digital twinning, comprising a state monitoring module, the state monitoring module comprising a multi-source data acquisition layer, an edge data processing layer, a digital twin engine, an intelligent decision layer, and a reverse control channel, the state monitoring module acquires state information of the high-voltage distribution room robot and regulates it;

[0100] The multi-source data acquisition layer includes a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is connected with the visual acquisition unit, the multi-modal sensor array includes an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit; the visual acquisition unit includes a binocular stereo vision module, a laser profiler and an adaptive light compensation module;

[0101] The edge data processing layer is connected with the multi-modal sensor array and the visual acquisition unit through a time-triggered Ethernet, the edge data processing layer is internally provided with a dynamic data cleaning module and a feature fusion module, the dynamic data cleaning module adopts a sliding window adaptive filter and an improved DBSCAN clustering algorithm, and the feature fusion module constructs a space-time alignment algorithm;

[0102] The digital twin engine includes a model construction and a virtual-real synchronization mechanism, a simulation model is driven through a virtual-real synchronization protocol, the digital twin engine constructs a millimeter-level precision model based on a Unity platform, and the millimeter-level precision model includes a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module;

[0103] The intelligent decision layer integrates an operation ticket parser, a dynamic path planner and a risk prediction matrix, is configured with a task disintegration logic tree, a visual terminal, a three-dimensional heat map production module and a fault prediction module, the operation ticket parser extracts operation item features by using a BERT model, the dynamic path planner integrates a DDPG reinforcement learning model, and the risk prediction matrix defines 12 types of high-risk operation combinations;

[0104] The reverse control channel is integrated with a CRC32 check module, an operation logic review unit and a physical feedback verification module;

[0105] Further, a 220 kV substation distribution room is subjected to continuous rainstorm invasion during the passage of a typhoon, a water level sensor triggers an alarm, the system starts an emergency inspection task at night, at this time, the ambient illuminance is lower than 5 lux, the rainstorm intensity reaches 50 mm / h, and a traditional monitoring system has failed;

[0106] The multi-source data acquisition layer immediately activates an anti-interference mode: the binocular stereo vision module has a baseline distance accurately controlled at 83 mm, emits a 905 nm near-infrared laser beam to penetrate a rain curtain, the laser profiler scans ground deformation, the adaptive light compensation module stabilizes the color temperature in a range of 5000 K±50 K, and the illuminance is dynamically improved to 1000 lux to suppress raindrop reflection. An ultrasonic water level gauge installed on the chassis of an AGV unmanned transport vehicle 9 monitors the water depth in real time, and a six-axis IMU sensor continuously feeds back the pitch and roll angle data of the AGV;

[0107] The edge data processing layer receives the original data stream through time-triggered Ethernet (TTEthernet), adopts a sliding window adaptive filtering algorithm with a window size of T = 1.2 s, and combines an improved DBSCAN clustering algorithm with parameters of ε = 0.25 and MinPts = 6 to separate effective data points from raindrop noise. The EPOCH synchronizer unifies the time bases of various sensors, fuses a 5 cm x 5 cm resolution water level grid map, joint angles, and instrument readings extracted by OCR into a 128-dimensional environment tensor, compresses the data volume to 32% of the original value, and uploads the data through a 5G private network;

[0108] The digital twin engine constructs a flood evolution model based on a finite element method, updates a flooded area prediction map every 200 ms, and dynamically calculates the distance between the mechanical arm and floating obstacles by using a Unity physical engine. The intelligent decision-making layer calls a DDPG model to plan a path, a state space includes three-dimensional coordinates of the AGV, a 32-dimensional water depth grid, and coordinates of a target cabinet, and when the system identifies that the water depth in front of the No. 3 power distribution cabinet reaches 18 cm, a bypass path is automatically re-planned. A risk prediction matrix intercepts a "wading to detect a live terminal" instruction, and forcibly switches the long-axis mechanical arm 2 to use an insulating clamp. During the execution process, the binocular vision module successfully reads voltage and current values of 32 instruments, the OCR accuracy is 99%, physical feedback verification is triggered every 3 atomic actions, trajectory correction is triggered 3 times throughout the process, and no equipment wading accident occurs.

[0109] Embodiment 5, please refer to Figure 1 、 Figure 2 and Figure 4 , a high-voltage power distribution room robot state monitoring system based on digital twinning, comprising a state monitoring module, the state monitoring module comprising a multi-source data acquisition layer, an edge data processing layer, a digital twin engine, an intelligent decision-making layer, and a reverse control channel, the state monitoring module acquires state information of the high-voltage power distribution room robot and regulates the state information;

[0110] The multi-source data acquisition layer comprises a multi-modal sensor array and a vision acquisition unit, the multi-modal sensor array is connected with the vision acquisition unit, the multi-modal sensor array comprises an angle sensor, a transient ground voltage sensor, an infrared thermal imager, and an environment monitoring unit, and the vision acquisition unit comprises a binocular stereo vision module, a laser profiler, and an adaptive light supplement module;

[0111] The edge data processing layer is connected with the multi-modal sensor array and the vision acquisition unit through time-triggered Ethernet, the edge data processing layer is internally provided with a dynamic data cleaning module and a feature fusion module, the dynamic data cleaning module adopts a sliding window adaptive filtering algorithm and an improved DBSCAN clustering algorithm, and the feature fusion module constructs a space-time alignment algorithm;

[0112] The digital twin engine includes a model construction and virtual-real synchronization mechanism, drives a simulation model through a virtual-real synchronization protocol, and is based on a Unity platform to construct a millimeter-level precision model, including a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module;

[0113] The intelligent decision layer integrates an operation ticket parser, a dynamic path planner, and a risk prediction matrix, is configured with a task decomposition logic tree, a visual terminal, a three-dimensional heat map production module, and a fault prediction module, the operation ticket parser extracts operation item features using a BERT model, the dynamic path planner integrates a DDPG reinforcement learning model, and the risk prediction matrix defines 12 high-risk operation combinations;

[0114] The reverse control channel integrates a CRC32 check module, an operation logic review unit, and a physical feedback verification module.

[0115] Further, the system initiates preventive maintenance based on historical operation data, the operation ticket parser receives a natural language instruction: "Annual maintenance: tighten busbar bolt, detect insulation resistance". The BERT model outputs a 32-dimensional feature vector, which is deconstructed into atomic operation items: "action type: torque control, target: busbar bolt, parameter: 65 N·m" and "action type: measurement, target: insulation resistance, device: megohmmeter", with an analysis accuracy of 98.7%. The fault prediction module activates a wavelet packet-convolutional neural network WP-CNN: the input layer receives temperature curves in the past 30 days, vibration frequency spectra with a bandwidth of 0-5 kHz, and partial discharge pulse data; the wavelet packet decomposition layer disassembles the vibration signal into 8 frequency bands; the convolution module uses a Conv1D-32→MaxPool→Conv1D-64 structure to extract features; the fully connected layer outputs the fault probability: the probability of fatigue fracture of the cabinet door hinge is 89%, and the probability of cable joint aging is 94%. The digital twin engine calls the electromagnetic-thermal coupling simulation module, discretely solves Maxwell's equation, and shows that the contact resistance rises to 82 μΩ due to bolt loosening, and predicts a local temperature rise of 7.2℃;

[0116] The intelligent decision layer generates an atomic sequence: the manipulator A actuator 1 scans the hinge area with an infrared thermal imager, generating a temperature field three-dimensional heat map with a maximum temperature of 68℃; the bolt is tightened with a target torque of 65 N·m, and the actual output of 69.5 N·m triggers a deviation alarm; after the self-correction process issues a -4% torque compensation instruction, it meets the standard;

[0117] The risk prediction matrix in the maintenance process intercepts the request for "non-power-off operation" in real time, and forcibly inserts the remote power-off step; the retest of the partial discharge sensor shows that the pulse count decreases from 120 times / s to 5 times / s. The verification data after maintenance: the maximum temperature decreases to 63.5°C, the vibration RMS value decreases from 0.15 g to 0.08 g, and the WP-CNN model re-evaluates the failure probability to 4%. The whole process realizes a response delay of 150 ms through TTEthernet network, the data cleaning efficiency of the edge node is improved by 40%, the visual terminal outputs the maintenance report and the temperature field distribution comparison graph, which greatly reduces the time-consuming of preventive maintenance compared with the traditional method, and successfully avoids potential short-circuit accidents.

[0118] Working principle: After the system is started, the multi-source data acquisition layer is activated synchronously, the photoelectric encoder 6 feeds back the angle of the mechanical arm in real time, the binocular vision module generates a point cloud model, the transient voltage sensor captures the partial discharge signal, the original data is transmitted to the edge node through time-triggered Ethernet synchronization, the sliding window adaptive filtering and improved DBSCAN algorithm are used to clean electromagnetic noise, the vibration frequency spectrum, temperature gradient and visual data are fused into a multi-dimensional tensor through the EPOCH synchronizer, the data volume is compressed to 30% of the original value before uploading to the cloud, the digital twin engine drives the millimeter-level precision model on the Unity platform, the kinematic model detects the interference risk of the power distribution cabinet when simulating the trajectory of the mechanical arm, and the electromagnetic-thermal coupling module solves Maxwell's equation to generate an eddy current loss cloud map. The intelligent decision layer uses the BERT model to analyze the operation ticket, the DDPG reinforcement learning model optimizes the action sequence, and the risk prediction matrix intercepts 12 types of high-risk operations in real time. The atomic action sequence is verified by CRC32 before being sent to the execution end, and physical feedback verification is triggered after each two atomic actions are completed: comparing the actual joint angle with the simulation value, and starting the self-correction process if it is out of limit. The virtual-real synchronization protocol is used to realize closed-loop control throughout the process, and the visual terminal renders the three-dimensional thermal diagram and the failure probability cloud diagram in real time.

[0119] It is apparent to those skilled in the art that the application is not limited to the details of the foregoing exemplary embodiments, and that the application can be implemented in other concrete forms without departing from the spirit or essential characteristics of the application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, the scope of the application being defined by the appended claims rather than the foregoing description, and it is intended to encompass all changes and modifications that fall within the meaning and scope of the equivalent elements of the claims. Any reference signs in the claims should not be considered as limiting the claims involved.

Claims

1. A digital-twin-based robot condition monitoring system for high-voltage switchgear, characterized in that: The state monitoring module comprises a multi-source data acquisition layer, an edge data processing layer, a digital twin engine, an intelligent decision layer and a reverse control channel, the edge data processing layer and the digital twin engine receive data collected by the multi-source data acquisition layer and perform data processing and digital simulation, the intelligent decision layer generates operation decisions according to the results of data processing and digital simulation, and the reverse control channel delivers the operation decisions to an execution end after examination; the state monitoring module acquires state information of the high-voltage power distribution room robot and regulates and controls the same; The high-voltage power distribution room robot comprises a short-shaft mechanical arm (1), a long-shaft mechanical arm (2), a flange (7), a plug-in and plug-out platform (8) and an AGV unmanned transport vehicle (9), the short-shaft mechanical arm (1) and the long-shaft mechanical arm (2) are connected by a mechanical arm joint (4) and a connecting rod (5) and are provided with an end effector (3), the short-shaft mechanical arm (1) and the long-shaft mechanical arm (2) are rigidly connected with the plug-in and plug-out platform (8) through the flange (7), and the plug-in and plug-out platform (8) is carried on the AGV unmanned transport vehicle (9).

2. The high-voltage power distribution room robot state monitoring system based on digital twinning of claim 1, wherein: The multi-source data acquisition layer comprises a multi-modal sensor array and a visual acquisition unit, the multi-modal sensor array is signal-connected with the visual acquisition unit, and the pose and environmental data of the high-voltage power distribution room robot are collected; The multi-modal sensor array comprises an angle sensor, a transient ground voltage sensor, an infrared thermal imager and an environmental monitoring unit, the angle sensor is arranged on the mechanical arm joint (4), an optical absolute encoder (6) is adopted, the optical absolute encoder (6) feeds back data in real time to monitor the motion state of the mechanical arm, the transient ground voltage sensor and the infrared thermal imager are mounted on the surface of the power distribution cabinet, and local discharge signals and temperature field distribution are captured, and the environmental monitoring unit integrates a temperature and humidity sensor; The visual acquisition unit comprises a binocular stereo vision module, a laser profiler and an adaptive light compensation module, the binocular stereo vision module is combined with the laser profiler to construct a three-dimensional point cloud model of the device, and the adaptive light compensation module ensures the imaging quality under complex lighting conditions.

3. The high-voltage power distribution room robot condition monitoring system based on digital twinning of claim 1, wherein: The edge data processing layer is connected with the multi-modal sensor array and the visual acquisition unit through time-triggered Ethernet, the edge data processing layer acquires the data collected by the multi-source data acquisition layer through Ethernet and analyzes and processes the same, and the edge data processing layer is internally provided with a dynamic data cleaning module and a feature fusion module; The dynamic data cleaning module adopts a sliding window adaptive filter and an improved DBSCAN clustering algorithm, and the feature fusion module synchronizes the time bases of various sensors through an EPOCH synchronizer, and fuses vibration spectrum, temperature gradient and local discharge pulses into a multi-dimensional tensor.

4. The high-voltage power distribution room robot state monitoring system based on digital twinning of claim 1, wherein: The digital twin engine receives the data collected by the multi-source data acquisition layer and processed by the edge data processing layer, performs digital simulation, and comprises model construction and virtual-real synchronization mechanisms, and drives a simulation model through a virtual-real synchronization protocol. The digital twin engine is built based on the Unity platform to construct a millimeter-level precision model, including a robot kinematics model and a transformer electromagnetic-thermal coupling simulation module.

5. The high-voltage power distribution room robot condition monitoring system based on digital twinning of claim 1, wherein: The intelligent decision layer integrates an operation ticket parser, a dynamic path planner, and a risk prediction matrix, and is configured with a task decomposition logic tree, a visual terminal, a three-dimensional heat map production module, and a fault prediction module. The operation ticket parser extracts operation item features using a BERT model, and the dynamic path planner integrates a DDPG reinforcement learning model to optimize the action sequence and simultaneously optimize timeliness and safety.

6. The high-voltage switch room robot condition monitoring system based on digital twinning of claim 5, wherein: The task decomposition logic tree is composed of an atomic operation unit library, an operation dependency graph, and a resource conflict detector, the atomic operation unit library contains various basic mechanical actions, the operation dependency graph is modeled using Petri nets, and the resource conflict detector is based on the banker's algorithm.

7. The high-voltage switch room robot condition monitoring system based on digital twinning of claim 1, wherein: The visual terminal generates three-dimensional heat maps and fault probability cloud maps, and the fault prediction module integrates a wavelet packet-convolutional neural network (WP-CNN) model to collaboratively diagnose mechanical and electrical faults. The reverse control channel integrates a CRC32 check module, an operation logic review unit, and a physical feedback verification module.

8. A method for robot condition monitoring in a high-voltage power distribution room based on digital twinning, applicable to a robot condition monitoring system in a high-voltage power distribution room based on digital twinning according to any one of claims 1-7, characterized in that: After the control instructions are verified by CRC32 and logical rule review, they are sent to the execution terminal. The method includes the following steps: S1: receiving the operation ticket task issued; S2: obtaining the pose data of the target unit through a multi-modal sensor array and a visual collection unit; S3: preforming the operation process in the digital twin environment and identifying potential mechanical interference areas; S4: decomposing the operation into an atomic action sequence; 9. The method for robot condition monitoring of high-voltage power distribution rooms based on digital twinning according to claim 8, characterized in that: S5: issuing atomic action instructions to the execution terminal and synchronously updating the twin model state. The mechanical interference identification of S3 uses Unity physics engine collision detection to dynamically calculate the following parameters:

10. The method for robot condition monitoring of high-voltage power distribution rooms based on digital twinning according to claim 8, characterized in that: The distance between the robot arm and the power distribution cabinet, the overlap rate of the dual-robot arm actuator workspace, and whether the joint acceleration exceeds the limited range. The generation of the atomic action sequence of S4 is as follows: S41: the operation ticket parser extracts operation item key parameters; S42: the DDPG model generates an initial action path; S43: the risk prediction matrix intercepts high-risk operation combinations; S44: output the atomic action sequence.

Images