A substation high voltage switchgear switching operation robot system

Abstract

The application provides a transformer substation high-voltage switch cabinet switching operation robot system and belongs to the technical field of robots.In the process of executing the switching operation of the switch cabinet, the dispatching instruction of the main control room is transmitted remotely based on the WebSocket communication protocol by integrating ChatGPT, the mobile robot system moves to the specified switch cabinet through a path planning algorithm, the state of the switch cabinet is identified through an improved YOLOv8 algorithm, and the vehicle-mounted unmanned aerial vehicle is used to verify that the grounding knife switch has been completely disconnected and the switch cabinet has met the switching conditions.The center pose of the operation hole is positioned based on a feature constraint ellipse detection algorithm, and a six-degree-of-freedom mechanical arm is controlled by using a cubic B-spline combined with an improved sparrow search algorithm to execute the switching operation in the optimal trajectory in terms of time, and the robot system is used to replace manual operation to execute the potentially dangerous task according to the standardized switching operation process of the transformer substation, so that the safety risk of the staff is reduced.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology, specifically relating to a robot system for switching operations of high-voltage switchgear in substations. Background Technology

[0002] With the continuous development of society, people's demand for a stable power supply is increasing, and the foundation for ensuring a continuous and stable power supply is the stable operation of the power supply system. As a crucial link in the power supply system, a failure in the transmission and transformation equipment of a substation will cause the entire power supply system to malfunction. Therefore, regular maintenance of the substation's transmission and transformation equipment can effectively ensure the normal operation of power equipment. Substations are usually staffed with dedicated personnel who conduct regular inspections, record instrument readings, and check various power equipment for any abnormalities. Power inspection personnel need to perform repetitive, mechanical work every day, consuming a large amount of manpower and resulting in low inspection efficiency. Furthermore, in daily life, adjustments to the power grid's operating mode require power inspection personnel to perform a series of switching operations. Therefore, identifying the status of high-voltage switchgear in advance and using intelligent robotic arms to perform switching operations will help ensure the safety of equipment, personnel, and the power grid.

[0003] In power systems, substations play a crucial role in power transmission, and high-voltage switchgear switching operations are an important task in the daily maintenance and emergency response processes of substations. Traditionally, following standardized substation switching operation procedures, workers first check the switch positions and meter readings from the front of the high-voltage switchgear, then observe and verify from the back of the cabinet that the grounding switch has been fully opened before performing the high-voltage switchgear switching operation. After confirming that the cabinet is in normal condition, workers need to manually complete the switching operation, which involves contact with high-voltage equipment and poses certain safety risks. In addition, the efficiency and accuracy of manual operation are limited by the operator's skills and experience. Existing mobile switching operation robots on the market cannot simultaneously detect the status information of the front and back of the cabinet, and the narrow passages and other terrain factors at the rear of the cabinet prevent the mobile robots from moving smoothly behind the cabinet. The hardware of this new switching operation robot system combines a mobile unmanned vehicle equipped with a six-degree-of-freedom robotic arm and a quadcopter drone, while the software integrates ChatGPT and deep learning image processing. This system can reduce the safety risks for workers by remotely controlling the system according to the standardized substation switching operation procedures. Summary of the Invention

[0004] To achieve switching operations of high-voltage switchgear, this invention proposes a robot system for switching operations of high-voltage switchgear in substations, including a human-machine interaction subsystem, a mobile chassis navigation and positioning subsystem, a deep learning image processing subsystem, a UAV target detection subsystem, and a robotic arm switching execution subsystem, such as... Figure 1 and Figure 2 As shown. The human-computer interaction subsystem integrates the ChatGPT natural language processing module to reduce manual training costs and improve operational efficiency and safety, including language processing, interactive interface, and decision support. The mobile chassis navigation and positioning subsystem autonomously navigates to a designated target switchgear based on a known global map and adjusts its own posture to the working mode, including a lidar ranging module, omnidirectional wheels, and an odometer reading module. The deep learning image processing subsystem identifies switchgear indicator lights, switch positions, and meter readings in real time to determine the switchgear's working status, including an image acquisition module, an omnidirectional gimbal, an image processing module, and a signal transmission module. The UAV target detection subsystem detects the status of the grounding switch. Due to the limitations of traditional manual inspection and the narrow working space, this system utilizes a vehicle-mounted UAV to autonomously navigate to the window of the lower cabinet on the back of the target switchgear to detect whether the grounding switch is fully open, including an image acquisition module and an image processing module. The robotic arm switching execution subsystem replaces human operators in performing switching operations, avoiding personal safety accidents caused by electrical leakage, including an operating hole positioning and ranging module, a trajectory planning module, and an inverse kinematics solution module.

[0005] The robotic system integrates ChatGPT natural language processing and deep learning image processing to reduce human training costs and improve operational efficiency and safety. This robot can not only perform basic switching operations but also identify abnormal conditions through drone inspections.

[0006] A shell is installed on a mobile chassis, a six-degree-of-freedom robotic arm is installed on the shell, and an omnidirectional gimbal is added to the shell to install an RGB camera. Laser ranging radars are installed on the front and rear sides inside the mobile chassis, and four omnidirectional wheels are installed at the bottom.

[0007] The six-degree-of-freedom robotic arm base is mounted on a mobile chassis shell, with an end effector at one end and a binocular depth camera mounted at the end of the robotic arm.

[0008] The quadcopter drone is parked at the far end of the mobile chassis shell, without interfering with the robotic arm, and has an RGB camera underneath. Attached Figure Description

[0009] Figure 1 This is a system workflow diagram of an embodiment of this application;

[0010] Figure 2 This is a schematic diagram of the system operation according to an embodiment of this application;

[0011] Figure 3 This is a schematic diagram of system communication according to an embodiment of this application. Detailed Implementation

[0012] This invention proposes a substation high-voltage switchgear switching operation robot system, comprising a human-machine interaction subsystem, a mobile chassis navigation and positioning subsystem, a deep learning image processing subsystem, a UAV target detection subsystem, and a robotic arm switching execution subsystem, such as... Figure 1 and Figure 2 As shown.

[0013] The human-machine interaction subsystem integrates the ChatGPT natural language processing module to reduce manual training costs and improve operational efficiency and safety, including language processing, interactive interface, and decision support. The mobile chassis navigation and positioning subsystem autonomously navigates to a designated target switchgear based on a known global map and adjusts its own posture to the working mode, including a lidar ranging module and an odometer reading module. The deep learning image processing subsystem identifies switchgear indicator lights, switch positions, and meter readings in real time to determine the switchgear's working status, including an image acquisition module, an image processing module, and a signal transmission module. The UAV target detection subsystem detects the status of the grounding switch. Due to the limitations of traditional manual inspection methods and the narrow working space, this system utilizes a vehicle-mounted UAV to autonomously navigate to the window of the lower cabinet on the back of the target switchgear to detect whether the grounding switch is fully open, including an image acquisition module and an image processing module. The robotic arm switching execution subsystem replaces human operators in performing switching operations, avoiding personal safety accidents caused by electrical leakage, including an operating hole positioning and ranging module, a trajectory planning module, and an inverse kinematics solution module.

[0014] This invention proposes a robot system for switching operations of high-voltage switchgear in substations, the specific working process of which is as follows:

[0015] Step 1: The main control room issues the instruction "Designate 10KV high-voltage switchgear to switch from maintenance to operation". Traditionally, this requires two staff members to enter the substation to inspect and perform switching operations. Using the UI of the PC host / tablet of this system and the WebSocket network communication protocol, staff members can remotely control the switchgear by entering the number of the designated switchgear or by entering "Perform switching operation on xx switchgear" on the PC host / tablet. This avoids safety accidents caused by equipment leakage or improper operation by personnel.

[0016] Step 1.1, as follows Figure 3 To explain the communication method between the client and server of this system, the PC host / tablet, the unmanned vehicle and the drone are connected to the same wireless local area network (Wi-Fi) and the correct IP address and subnet mask are set to ensure that each part can access each other;

[0017] Step 1.2: In order to remotely control the vehicle through the UI interface of the PC host / tablet, the WebSocket network communication protocol based on TCP / IP is used for data transmission. Specifically, the PC host / tablet and the unmanned vehicle transmit full-duplex, real-time data by publishing and subscribing to ROS topics.

[0018] Step 1.3: Drones and unmanned vehicles under the same Wi-Fi network communicate with each other in the ROS system through message passing between nodes. The ROS Master runs on the unmanned vehicle, and the drone connects to this Master as a node. The unmanned vehicle and the drone also communicate by publishing and subscribing to ROS topics.

[0019] Step 2, the Human-Computer Interaction (HCI) subsystem, is designed to understand natural language commands and interact with operators. The language processing module uses ChatGPT to parse operator instructions and convert them into commands the robot can understand. The user interface module allows operators to interact with the system via text or voice through a UI. The decision support module provides decision support, such as suggestions or safety tips for uncertain operations, offering operational guidance while reducing training costs.

[0020] Step 3: After receiving the execution command from the designated switch cabinet, the Raspberry Pi main controller plans a path to the designated location of the target switch cabinet using the RRT-Connected path planning algorithm and publishes topic information including running speed and X and Y attitude. The X and Y attitude information ensures that the robot system is facing the switch cabinet and in working condition. The odometry module can calculate and report the position of the mobile chassis on the known global cost map. The lidar ranging module can enable the mobile chassis to avoid obstacles in real time by setting the minimum and maximum collision distances.

[0021] Step 4, the deep learning image processing subsystem, uses an RGB camera to identify the status of indicator lights, switches, meters and grounding wire switching in the switch cabinet in real time to determine the working status of the switch cabinet.

[0022] Step 4.1: First, create a dataset of high-voltage switchgear switching statuses, and use LabelMe annotation software to label and classify the data. Due to the small sample size of the dataset, image enhancement techniques such as flipping, translating, scaling, mirroring, and adding noise are used to expand the dataset.

[0023] Step 4.2: The high-voltage switchgear dataset is trained using the improved YOLOv8 algorithm. mAP is used as the evaluation metric to compare the object detection experimental results and select the optimal training weights. Specifically, all C2f modules in the YOLOv8 neck network are replaced with C3Ghost modules based on lightweight Ghost convolutions, reducing computational and parameter counts. Simultaneously, spatial and channel attention mechanisms are added to the three network paths before reaching the feature pyramid, enhancing the network's ability to learn features at different scales and improving accuracy. The improved network structure significantly reduces inference latency, model parameter count, and computational load without sacrificing much accuracy, thus enabling real-time detection of high-voltage switchgear switching states.

[0024] Step 4.2.1, the design concept of C3Ghost is to combine the computational efficiency of Ghost convolution with the feature representation capability of the C3 module to improve the computational efficiency and accuracy of the model. By combining the advantages of Ghost convolution and the C3 module, C3Ghost can improve feature representation capability while maintaining low computational cost, thus achieving better results in object detection tasks.

[0025] Step 4.2.2: Ghost convolution is used to reduce computation by using a small number of convolution kernels (compared to standard convolution) to generate the original feature map. Then, the original feature map is processed by depthwise separable convolution to generate the Ghost feature map. Finally, the original feature map and the Ghost feature map are concatenated to form the final feature output.

[0026] Step 4.2.3, the C3 module is used to improve feature representation capabilities. The C3 module is a lightweight residual connection structure consisting of three convolutional layers. It extracts features using 1×1 and 3×3 convolutional kernels and uses cross-layer residual connections and linear activation functions to improve the model's accuracy and robustness. The C3 module can improve model performance with relatively few computational resources. Its input and output are both feature maps. In its specific implementation, the C3 module first uses a 1×1 convolutional layer to reduce the dimensionality of the input feature map, thus reducing the amount of computation. It also reduces the number of channels in the feature map to half of its original value. Then, it uses a 3×3 convolutional layer to perform convolution operations on the dimensionality-reduced feature map, increasing the receptive field size and extracting more global features. It then uses another 1×1 convolutional layer to double the number of channels in the feature map, increasing the dimensionality of the feature map. At the same time, it uses residual connections to add the input and output feature maps, preserving the high-level features of the input feature map and the low-level features of the output feature map. Finally, it performs BatchNorm and linear activation operations on the concatenated feature map to obtain the final feature map output.

[0027] Step 4.3: Convert the optimal training weights file (best.pt) from Step 4.2 into an ONNX file and deploy it on a Raspberry Pi for real-time monitoring of the switch cabinet's status. Specifically, use PyTorch's `torch.onnx.export` function to convert the .pt file to .onnx format.

[0028] Step 5, the UAV target detection subsystem, checks if the grounding switch is fully open. Previously, this required personnel to physically go to the lower cabinet window at the rear of the designated switchgear for inspection. This system automates this process using UAV autonomous navigation and detection technology, improving efficiency and safety. Personnel can remotely check the status of the grounding switch via PC / tablet in the main control room and issue corresponding operating commands accordingly.

[0029] Step 5.1: After confirming that there are no abnormal values ​​in the high-voltage switchgear and it can be switched from maintenance to operation, the unmanned vehicle runs roscore as the Master, and the drone's Raspberry Pi starts the path planning node. The drone uses the RRT* path planning algorithm to plan a 3D path to the lower cabinet window behind the designated switchgear, and then transmits data to the drone controller through the serial port, thereby guiding the drone to the designated working position.

[0030] Step 5.2: Deploy the improved YOLOv8 algorithm from Step 4.2 onto the drone's Raspberry Pi, and use an RGB camera to detect the status of the grounding switch. This deployment ensures the drone can effectively identify the status of the grounding switch, providing reliable visual information for subsequent operations.

[0031] Step 5.3: The data transmission module sends the current detection results back to the PC / tablet in the main control room via the WebSocket network communication protocol. The staff then reviews the results and issues a switching command. The switching command is transmitted to the drone and the unmanned vehicle. After the drone automatically returns to the stop position, the unmanned vehicle runs the switching operation node.

[0032] Step 6: The robotic arm switching execution subsystem is designed to replace human workers in performing switching operations, thus preventing personal safety accidents caused by electrical leakage.

[0033] Step 6.1 First, the center pose of the switching operation hole is located using an ellipse detection algorithm with a binocular depth camera. Specifically, the width, height and center point coordinates of the predicted bounding box detected by the YOLOv8 target detection algorithm are obtained. Within the predicted bounding box, the ellipse detection algorithm based on coordinate constraints, quadrant constraints and feature constraints is used to identify the switching operation hole, and the X and Y coordinates and depth value of the center of the operation hole are output.

[0034] Step 6.2: Further, the position information of the operating holes obtained by the binocular depth camera will undergo camera calibration and hand-eye calibration processes. The purpose of this step is to convert the position information in the object's pixel coordinate system into position information in the robot's coordinate system, and then transmit the converted coordinates from the Raspberry Pi to the host computer (PC / tablet) via WebSocket communication.

[0035] Step 6.3: Further, use the MoveIt! software package on a PC to build a simulation environment for the robotic arm's operation within the ROS environment, and input the converted position information to achieve simulated grasping. MoveIt! is an open-source ROS software package designed to provide robotic arms with advanced operation functions such as motion planning, obstacle avoidance, and grasping. MoveIt! provides a set of advanced motion planning interfaces and algorithms; given the position of the robotic arm's gate, it can automatically plan its motion trajectory. MoveIt! can also visualize the robot model and planned path in RViz, facilitating users to debug and optimize the robot's movement and behavior.

[0036] Step 6.4 uses a combination of an improved sparrow search algorithm and a cubic B-spline interpolation algorithm to effectively find the optimal or near-optimal motion trajectory, reducing energy consumption and operation time while ensuring operational efficiency and stability. It also smoothly interpolates the robotic arm's motion trajectory, ensuring the continuity and smoothness of movements, thereby improving operational accuracy and reliability. Integrate this algorithm plugin into the MoveIt! configuration by modifying the MoveIt! configuration file ompl_planning.yaml to use this trajectory planning algorithm.

[0037] The basic steps of time-optimal robotic arm trajectory planning combining cubic B-spline interpolation and an improved sparrow search algorithm are as follows:

[0038] Step 6.4.1, Improved Circle Chaotic Mapping: Make the initial individual (N=100 in this system) uniformly distributed throughout the space.

[0039]

[0040] Step 6.4.2, Define the fitness function: Time optimality usually means finding a path from the starting point to the ending point that is feasible under the dynamic constraints of the robotic arm. The fitness function will primarily focus on path length and smoothness.

[0041] The fitness function is f(θ) = w1·L(θ) + w2·S(θ).

[0042] The path length is

[0043] Smoothness is In the formula, w1 and w2 are weighting factors, and θ j,i is the joint angle of the robotic arm, j is a positive integer in [1,6] representing the joints of the robotic arm, i∈[1,n], and n represents the number of control points on the path.

[0044] Step 6.4.3, Sparrow foraging behavior with adaptive weights:

[0045]

[0046] In the formula, iter max Let R2 be a random number in the range (0,1) representing the maximum number of iterations.

[0047] [0,1]) and ST (ST∈[0.5,1.0]) represent the warning value and the safety value, respectively. Q is a random number that follows a normal distribution in the range [0,1].

[0048] Step 6.4.4, Improved Sparrow Alertness and Avoidance Behaviors:

[0049]

[0050] Among them, X best Let f represent the globally optimal position, β be the step size adjustment coefficient, and k be a uniformly distributed random number within the range [-1, 1] that is normally distributed with a mean of 0 and a variance of 1. i This is the current fitness value of the sparrow. g This is the current globally optimal fitness value.

[0051] Step 6.4.5, Update Sparrow Locations: Based on foraging, alertness, and escape behaviors, update the location of each sparrow. Each new location represents a new possible solution.

[0052] Step 6.4.6, Applying the Cubic B-Spline Algorithm: For each sparrow's position, the cubic B-spline algorithm is used to generate a smooth path. Assuming a set of control points P(0), P(1), ..., P(n), the cubic B-spline curve B(t) can be represented as:

[0053]

[0054] Among them, B i,3 (t) is a basis function of a cubic B-spline, where t is a parameterized variable (usually between 0 and 1). The basis function B... i,3 (t) is defined by the following recursive formula:

[0055]

[0056]

[0057] Where k is the degree of the spline (for a cubic B-spline, k = 3), t i It is the i-th element in the node vector, which determines the shape and parameterization of the spline curve.

[0058] Step 6.4.7, Evaluate and select the best solution: Use the fitness function to evaluate each generated path and select the path with the highest fitness as the current best solution.

[0059] Step 6.4.8, Iterative Optimization: Repeat Steps 6.4.3 to 6.4.7 until the maximum number of iterations is reached. In this system, iter max =1000.

[0060] Step 6.4.9, Output the optimal path: Output the optimal path found during the iteration process as the final solution.

[0061] Step 6.5: Visualize the path using RViz and, after ensuring it meets expectations, execute the trajectory planning, i.e., embed the end effector of the 6-DOF robotic arm into the switching operation hole. After the operator confirms the correctness via the image transmitted back from the RGB camera, issue the switching command, and the robotic arm rotates the end effector according to the preset values.

[0062] Step 7: To ensure the switching operation is complete, a deep learning image processing subsystem is used to identify the indicator lights and switch status of the switchgear. The detection results are then transmitted back to a PC / tablet via topic communication. Once the work team confirms that the switchgear is in the closed and energized state, they issue a "return home" command. The robotic arm returns to its starting and ending positions, and the mobile chassis returns to the starting point.

Claims

1. A robot system for switching operations of high-voltage switchgear in substations, characterized in that: Step 1: The main control room issues the instruction "Select a 10KV high-voltage switchgear to switch from maintenance to operation". Using the UI operation interface of the PC host / tablet of this system, and using the WebSocket network communication protocol, the staff can enter the number of the selected switchgear or enter "Execute switching operation on the selected 10KV switchgear" on the PC host / tablet to remotely control it, so as to avoid safety accidents caused by equipment leakage or improper operation by personnel. Step 1.1: First, connect the PC host / tablet, unmanned vehicle, and drone to the same Wi-Fi network and set the correct IP address and subnet mask to ensure that each part can access each other; Step 1.2: In order to remotely control the vehicle through the UI interface of the PC host / tablet, the WebSocket network communication protocol based on TCP / IP is used for data transmission. Specifically, the PC host / tablet and the unmanned vehicle transmit full-duplex, real-time data by publishing and subscribing to ROS topics. Step 1.3: Drones and unmanned vehicles under the same Wi-Fi network communicate with each other in the ROS system through message passing between nodes. The ROS Master runs on the unmanned vehicle, and the drone connects to this Master as a node. The unmanned vehicle and the drone also communicate by publishing and subscribing to ROS topics. Step 2: The human-computer interaction subsystem is capable of understanding natural language instructions and interacting with operators. The language processing module uses ChatGPT for natural language processing, parses the operator's instructions, and converts them into commands that the robot can understand. The interaction interface module allows operators to interact with the system via text or voice through a UI interface. The decision support module provides decision support. Step 3: After receiving the execution command for the designated switch cabinet, the Raspberry Pi main controller plans the path to the designated location of the target switch cabinet using the RRT-Connected path planning algorithm and publishes topic information including running speed and X and Y attitude. The X and Y attitude information ensures that the robot system is facing the switch cabinet and in working state. The odometry module calculates and feeds back the position of the mobile chassis on the known global cost map; the lidar ranging module enables the mobile chassis to avoid obstacles in real time by setting the minimum and maximum collision distances. Step 4: The function of the deep learning image processing subsystem is to use an RGB camera to identify the status of indicator lights, switches, meters and ground wire switching in the switch cabinet in real time to determine the working status of the switch cabinet. Step 4.1: First, create a high-voltage switchgear switching status dataset, and use labelme annotation software to annotate and classify the data; because the dataset has a small number of samples, image enhancement and noise addition are used to expand the dataset. Step 4.2: The high-voltage switchgear dataset is trained using the improved YOLOv8 algorithm. mAP is used as the evaluation metric to compare the target detection experimental results and select the optimal training weights. Specifically, all C2f modules in the YOLOv8 neck network are replaced with C3Ghost modules based on lightweight Ghost convolutions to reduce computational cost and parameter count. At the same time, spatial attention and channel attention mechanisms are added to the three networks before the network reaches the feature pyramid, thereby completing the real-time detection of the switching state of the high-voltage switchgear. Step 4.2.1, the design idea of ​​C3Ghost is to combine the computational efficiency of Ghost convolution with the feature representation capability of C3 module to improve the computational efficiency and accuracy of the model; Step 4.2.2: Ghost convolution is used to reduce computation by using a small number of convolution kernels to generate the original feature map. Then, depthwise separable convolution is used to process the original feature map to generate the Ghost feature map. Finally, the original feature map and the Ghost feature map are concatenated to form the final feature output. Step 4.2.3, the C3 module is used to improve feature representation capabilities. The C3 module is a lightweight residual connection structure consisting of three convolutional layers. It extracts features by using 1×1 and 3×3 convolutional kernels and uses cross-layer residual connections and linear activation functions to improve the accuracy and robustness of the model. The computation process of the C3 module first uses a 1×1 convolutional layer to reduce the dimensionality of the input feature map, reducing the amount of computation; and reduces the number of channels of the feature map to half of the original number. Then, a 3×3 convolutional layer is used to perform convolution operations on the dimensionality-reduced feature map, increasing the receptive field size and extracting more global features. The number of channels in the feature map is doubled again using a 1×1 convolutional layer, increasing the dimension of the feature map. At the same time, residual connections are used to add the input and output feature maps, preserving the high-level features of the input feature map and the low-level features of the output feature map. BatchNorm and linear activation operations are then performed on the concatenated feature map to obtain the final feature map output. Step 4.3: Convert the best training weights file (best.pt) from Step 4.2 into an ONNX file and deploy it on a Raspberry Pi to monitor the status of the switch cabinet in real time. Specifically, use PyTorch's torch.onnx.export function to convert the .pt file to .onnx format. Step 5: The function of the UAV target detection subsystem is to check that the grounding switch is fully open; staff in the main control room can remotely confirm the status of the grounding switch via PC / tablet and issue corresponding operation instructions accordingly. Step 5.1 After confirming that there are no abnormal values ​​in the high-voltage switchgear and the switchgear is switched from maintenance to operation, the unmanned vehicle runs roscore as the Master, and the drone's Raspberry Pi starts the path planning node. The drone uses the RRT* path planning algorithm to plan a 3D path to the lower cabinet window on the back of the designated switchgear, and then transmits data to the drone controller through the serial port to guide the drone to the designated working position. Step 5.2: Deploy the improved YOLOv8 algorithm onto the Raspberry Pi of the drone and use an RGB camera to detect the status of the grounding switch; Step 5.3: The data transmission module sends the current detection results back to the PC / tablet in the main control room via the WebSocket network communication protocol. The staff will then review the data and issue a switching command. The switching command is transmitted to the drone and the unmanned vehicle. After the drone automatically returns to the stop position, the unmanned vehicle will run the switching operation node. Step 6: The function of the robotic arm switching execution subsystem is to replace the workers in performing switching operations and avoid personal safety accidents caused by electrical leakage. Step 6.1 First, the center pose of the switching operation hole is located using an ellipse detection algorithm with a binocular depth camera. Specifically, the width, height and center point coordinates of the predicted bounding box detected by the YOLOv8 target detection algorithm are obtained. Within the predicted bounding box, the ellipse detection algorithm based on coordinate constraints, quadrant constraints and feature constraints is used to identify the switching operation hole, and the X and Y coordinates and depth value of the center of the operation hole are output. Step 6.2, further, the operation hole position information obtained by the binocular depth camera will undergo camera calibration and hand-eye calibration processes; the purpose of this step is to convert the position information in the object pixel coordinate system into the position information in the robot coordinate system, and transmit the converted coordinates from the Raspberry Pi to the host computer through WebSocket communication; Step 6.3: Use the MoveIt! software package on the host computer to build a simulation environment for the operation of the robotic arm in the ROS environment, and input the converted position information to realize the simulation grasping; Step 6.4: Use a combination of an improved sparrow search algorithm and a cubic B-spline interpolation algorithm; integrate the algorithm plugin into the MoveIt! configuration and modify the MoveIt! configuration file ompl_planning.yaml to use the trajectory planning algorithm; The basic steps of time-optimal robotic arm trajectory planning combining cubic B-spline interpolation and an improved sparrow search algorithm are as follows: Step 6.4.1, Improved Circle Chaos Mapping: Makes the initial individual uniformly distributed throughout the space; Step 6.4.2, Define the fitness function: Time optimal means finding a path from the starting point to the ending point that is feasible under the dynamic constraints of the robotic arm; focus the fitness function on path length and smoothness; The fitness function is , The path length is , Smoothness is In the formula and It is a weighting factor, θ j,i is the joint angle of the robotic arm, j is a positive integer in [1,6] representing the joints of the robotic arm, i∈[1,n], and n represents the number of control points on the path; Step 6.4.3, Sparrow foraging behavior with adaptive weights: , In the formula, iter max R2 represents the maximum number of iterations, α is a random number in the range (0,1), R2 and ST represent the warning value and the safety value respectively, R2∈[0,1], ST∈[0.5,1.0]; Q is a random number that follows a normal distribution in [0,1]. Step 6.4.4, Improved Sparrow Alertness and Avoidance Behaviors: Among them, X best Let f represent the globally optimal position, β be the step size adjustment coefficient, and k be a normally distributed random number with a mean of 0 and a variance of 1; f be a uniform random number in the range [-1, 1]. i This is the current fitness value of the sparrow; f g This is the current globally optimal fitness value; Step 6.4.5, Update Sparrow Locations: Based on foraging, alertness, and escape behaviors, update the location of each sparrow; each new location represents a new possible solution; Step 6.4.6, Applying the cubic B-spline algorithm: For each sparrow's position, use the cubic B-spline algorithm to generate a smooth path; assuming a set of control points P(0), P(1), …, P(n), the cubic B-spline curve B(t) is represented as: in, These are the basis functions of a cubic B-spline, where t is a parameterized variable t∈[0,1]; basis functions Defined by the following recursive formula: Where k is the degree of the spline, t i It is the i-th element in the node vector, which determines the shape and parameterization of the spline curve; Step 6.4.7, Evaluate and select the best solution: Use the fitness function to evaluate each generated path and select the path with the highest fitness as the current best solution; Step 6.4.8, Iterative optimization: Repeat Step 6.4.3 to Step 6.4.7 until the maximum number of iterations is reached; Step 6.4.9, Output the optimal path: Output the optimal path found during the iteration process as the final solution; Step 6.5: Use RViz for visualization and ensure that the path meets expectations before performing trajectory planning. Embed the end effector of the 6-DOF robotic arm into the switching operation hole. After the staff confirms that there are no errors by sending back the image through the RGB camera, issue the switching command. The robotic arm will rotate the end effector according to the preset value. Step 7: To ensure that the switching has been completed, the deep learning image processing subsystem is used to identify the indicator lights and switch status of the switch cabinet, and the detection results are sent back to the PC / tablet via topic communication. After the work team confirms that the switch cabinet is in the closed and energized state, the command to return home is issued; the robotic arm returns to the start and stop positions, and the mobile chassis returns to the starting point.

Images