Insulator chain maintenance robot placing method based on AI

Abstract

The invention discloses an AI-based insulator chain maintenance robot placement method, and relates to the technical field of a chain inspection robot, and the method comprises the steps: collecting the parameter information of an insulator chain and a tower where the insulator chain is located, storing the data in a database, training an AI model through a learning model, and carrying out the training of the AI model; the insulator string placement method has the advantages that the insulator string and the surrounding environment thereof can be accurately sensed and analyzed by utilizing an advanced sensor technology and an intelligent algorithm based on the A I placement method, the three-dimensional coordinate, the shape, the posture and the detailed information of surrounding obstacles of the insulator string can be obtained through fusion of the laser radar and the visual sensor, and the insulator string placement method is suitable for popularization and application. The placement precision of the robot is controlled at a centimeter level or even a higher level, the situation that maintenance operation is difficult or failed due to the fact that the placement position is inaccurate is reduced, it is ensured that the robot can accurately approach an insulator chain, and various maintenance tasks such as detection, cleaning and repairing are smoothly carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of string inspection robots, and in particular to an AI-based insulator string maintenance robot placement method. Background Art

[0002] Insulator strings play an indispensable role in the power transmission system. They can effectively isolate transmission lines and towers to ensure safe and stable power transmission. However, due to long-term exposure to the natural environment, insulator strings face threats such as pollution deposition, lightning strikes, mechanical damage and other factors. This makes regular and efficient maintenance work a key link in ensuring the reliable operation of the power system. With the continuous advancement of science and technology, the methods and technologies of insulator string maintenance are also evolving, gradually moving from traditional manual maintenance to intelligent and automated maintenance modes.

[0003] The current level of intelligent decision-making in the robot placement process is low. The determination of the placement position mainly relies on pre-set fixed rules or manual experience judgment, and lacks the ability to deeply analyze and learn a large amount of actual operation data. To this end, we propose an AI-based insulator string maintenance robot placement method. Summary of the invention

[0004] The purpose of the present invention is to provide an AI-based insulator string maintenance robot placement method.

[0005] In order to solve the problems raised in the above background technology, the present invention provides the following technical solutions: an AI-based insulator string maintenance robot placement method, comprising collecting parameter information of the insulator string and the tower where it is located, storing the data in a database, and training the AI ​​model using a learning model. The specific operation steps of the placement method are as follows:

[0006] Step 1: Use a drone to photograph the shape and height parameter information of the insulator string;

[0007] Step 2: Collect data on the type, height and topographic environment of the tower where the insulator string is located;

[0008] Step 3: Store the collected data in a database and record the performance parameters of the maintenance robot itself;

[0009] Step 4: Use the convolutional neural network learning model in deep learning to train the selected AI model using the collected data;

[0010] Step 5: Optimize the joint accuracy of the robot arm and adjust the robot's motion parameters using the model predictive control algorithm;

[0011] Step 6: Scan the on-site environment in real time, and the AI ​​system plans the robot's movement path based on the on-site environment and the initial placement position.

[0012] As a further solution of the present invention: In step 1, a drone equipped with a high-definition camera is used to take clear images of the insulator string from different angles and distances, and the average value method formula is used:

[0013]

[0014] Among them, R represents the pixel value of the red channel, G represents the pixel value of the green channel, B represents the pixel value of the blue channel, and 3 represents the number of color channels calculated. The original color image can be converted into a grayscale image. During the image acquisition process, due to atmospheric jitter and lighting changes, the image contains noise. The bilateral filtering algorithm is used to remove the noise in the image. The spatial distance weight formula is:

[0015]

[0016] Among them, ω d represents the spatial distance weight, σ d represents the standard deviation of spatial distance, P represents the current pixel, Q represents the neighboring pixel, d(P,Q) represents the spatial distance between the current pixel and the neighboring pixel, and the grayscale difference weight formula is:

[0017]

[0018] Among them, ω r represents the grayscale difference weight, σ r Grayscale difference standard deviation, I(P) represents the grayscale value of the current pixel, I(Q) represents the grayscale value of the neighboring pixel, and the final weight ω=ω d ×ω r , multiply the weights of all pixels in the neighborhood by the pixel value and then sum them, and then divide them by the sum of all weights to get the filtered pixel value, which can remove noise while retaining the edge of the image. Then use the Canny operator edge detection technology to extract the contour information of the insulator string, and use the contour tracking algorithm to connect the edge pixels to form a complete insulator string contour curve. The insulator string contour curve in the processed image is annotated, and the annotation information is saved in JSON format. The annotated data can be used as the actual situation of the target for model learning comparison in subsequent training.

[0019] As a further solution of the present invention: in step 2, the staff observes the shape of the tower on site to determine its type, and uses a total station to accurately measure the tower height, and the calculation formula is:

[0020] H=S×sinα+h2-h1

[0021] Among them, H represents the height of the tower, S represents the slant moment from the center of the total station to the top of the tower, α represents the inclination angle of the telescope in the total station, h1 represents the height difference between the center of the total station and the bottom of the tower, and h2 represents the height of the total station itself. Aerial remote sensing is used to obtain high-resolution images of the area where the tower is located. The band range of remote sensing images ranges from visible light to infrared. Different band combinations can be used to identify different terrain and geomorphic features. In the visible light band, the color and texture of the surface can be observed, and the vegetation type and water body can be distinguished. In the infrared band, the health status and coverage of vegetation can be judged according to the infrared reflection characteristics of vegetation. The remote sensing data is integrated and analyzed with the existing GIS terrain and geomorphic data, and the tower location information is superimposed with the terrain and geomorphic data to intuitively obtain the terrain and geomorphic environmental information around the tower.

[0022] As a further solution of the present invention: in the step three, a relational database is used to store the collected data, a data table named insulator string is created in the relational database, including the serial number of the insulator string, the location of the insulator string, including the specific latitude and longitude coordinates and the number of the pole tower where it is located, the shape, height information, manufacturing material and installation date of the insulator string, a data table named pole tower is created in the relational database, including the type, height, location and construction date of the pole tower, a data table named robot performance parameters is created in the relational database, including the robot model, weight, size and power source, the working range of the robot, including the maximum extension length of the robot arm and the angle range that can be covered, for the insulator string and pole tower data collected by the drone, a corresponding data entry interface program is developed, and the parsed data is inserted into the corresponding data table according to the designed database structure, and the robot performance parameter data is entered into the database through the data entry interface by exporting relevant parameter information from the robot control system after the robot is manufactured.

[0023] As a further solution of the present invention: in the step 4, a convolutional neural network learning model for processing insulator string related data is constructed, and the convolutional neural network learning model includes a convolution layer, a pooling layer and a fully connected layer. The convolution layer can extract key feature information of different levels from the insulator string image data, from shallow texture, edge to deep structure, and shape through the convolution kernel structure and parameter setting. The pooling layer can reduce the data dimension, reduce the amount of calculation and enhance the translation invariance of the feature. After the data is processed by the convolution and pooling layers in turn, the extracted feature map is expanded and connected to the fully connected layer. The number of neurons in the fully connected layer shows a decreasing trend according to the network depth, and the data can be feature integrated and classified and regressed. The last layer of the fully connected layer can be flexibly set according to specific task requirements, and can not only accurately output the coordinate information of the placement position area to provide accurate placement position guidance for the insulator string maintenance robot, but also output probability values ​​representing the possibility of different placement areas to assist the robot in comprehensive judgment and selection of the optimal placement area.

[0024] As a further solution of the present invention: in the step 4, the insulator string shape and height data, tower type and height data, and terrain and geomorphic environment data obtained from drone photography are used in combination with the robot's own performance parameter data to train the convolutional neural network learning model. The training process adopts a back propagation algorithm, and the connection weights between network layers are adjusted according to the difference between the data label and the model prediction result, so that the convolutional neural network learning model can accurately identify the characteristics of the insulator string, judge its operating status, and plan the robot's movement path and operation actions according to the robot's capabilities and environmental conditions, and gradually learn to select a more suitable placement position according to factors such as the position of the insulator string and the structure of the tower.

[0025] As a further solution of the present invention: in the step five, a high-precision harmonic reducer is used. The harmonic reducer can accurately convert the high-speed rotation of the servo motor into the low-speed and high-torque movement required by the robot arm joint, and can provide an accurate transmission ratio and a smaller tooth gap, thereby reducing the error transmission during joint movement. A laser interferometer is used to detect the motion error of the robot arm joint, and software compensation is used to optimize the robot arm joint accuracy based on the detected error. The corresponding error compensation amount is calculated based on the current working state of the robot arm, including the load weight and movement speed of the robot arm, and then the target angle or displacement of the joint is corrected in the control signal to reduce the actual motion error.

[0026] As a further solution of the present invention: in the step five, a dynamic model is established for the robot based on the Lagrange equation to describe the relationship between the position, velocity and acceleration state variables of each joint of the robot mechanical arm and the motor torque control input, the current state parameters of the robot mechanical arm, including the angle, angular velocity and angular acceleration of each joint, are obtained through sensors, and the system state at the future moment is predicted using a model predictive control (MPC) algorithm. The MPC algorithm predicts the motion trajectory in the prediction time domain based on the current system state using the established prediction model, and based on the set optimization goals, including minimizing joint motion errors, reducing energy consumption and constraints, such as joint angle range limitations and motor torque limitations, the optimal control sequence in the control time domain is solved by the optimization algorithm, the first control input is taken out from the solved optimal control sequence and applied to the system, after the control input is implemented, the new state information of the system is obtained again through sensors, the system state variables are updated as the initial state for the next control cycle prediction, and the above prediction, optimization and implementation processes are repeated, and the cycle is rolled in this way, so that the system can continuously adapt to internal and external changes and always run towards the optimization goal.

[0027] As a further solution of the present invention: in the step six, a laser radar is used to scan the pole tower, insulator string and obstacles in the surrounding environment, and a visual sensor is used to obtain image information of the on-site environment. The high-definition camera can clearly capture the appearance of the insulator string, the structural characteristics of the pole tower and the overall appearance of the surrounding environment. The binocular vision camera can also obtain the depth information of the object by calculating the parallax, which complements the data of the laser radar to further improve the three-dimensional perception of the on-site environment. The AI ​​system uses the point cloud data of the laser radar and the image information of the visual sensor to construct a detailed on-site environment map, and relies on the path planning algorithm to quickly determine the optimal path from the starting point to the target point. In the process of path planning, the robot's own performance characteristics and actual operating conditions are always closely combined to ensure the scientificity and feasibility of the planned path. When the robot moves along the planned path, the AI ​​system continuously monitors the on-site environment through sensors, and promptly detects changes in the environment, such as the appearance of new obstacles and the movement of the original obstacle position, and immediately replans and partially adjusts the path.

[0028] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are:

[0029] 1. The AI-based placement method of the present invention utilizes advanced sensor technology and intelligent algorithms to accurately perceive and analyze the insulator string and its surrounding environment. Through the fusion of laser radar and visual sensors, it can obtain detailed information on the three-dimensional coordinates, shape, posture and surrounding obstacles of the insulator string, and control the placement accuracy of the robot to the centimeter level or even higher, reducing the difficulty or failure of maintenance operations caused by inaccurate placement positions, ensuring that the robot can accurately approach the insulator string, smoothly carry out various maintenance tasks such as detection, cleaning and repair, and effectively improve the quality and reliability of maintenance work;

[0030] 2. The AI-based robot placement method of the present invention has excellent dynamic adjustment capabilities. Once the insulator string or the surrounding environment changes, such as a slight displacement of the insulator string due to external force, or the addition of temporary obstacles in the surrounding area, the robot can instantly re-plan the placement position and route of action through continuous environmental monitoring and intelligent analysis. By learning and analyzing a large amount of historical data, real-time monitoring data, and simulation data, the robot can automatically generate the optimal placement plan based on factors such as different insulator string types, aging degrees, fault characteristics, and environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the placement steps of the AI-based insulator string maintenance robot in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] The present invention provides an AI-based insulator string maintenance robot placement method, which includes collecting parameter information of the insulator string and the pole tower where it is located, storing the data in a database, and training the AI ​​model using a learning model. The specific operation steps of the placement method are as follows:

[0034] Step 1: Use a drone to photograph the shape and height parameter information of the insulator string;

[0035] Step 2: Collect data on the type, height and topographic environment of the tower where the insulator string is located;

[0036] Step 3: Store the collected data in a database and record the performance parameters of the maintenance robot itself;

[0037] Step 4: Use the convolutional neural network learning model in deep learning to train the selected AI model using the collected data;

[0038] Step 5: Optimize the joint accuracy of the robot arm and adjust the robot's motion parameters using the model predictive control algorithm;

[0039] Step 6: Scan the on-site environment in real time, and the AI ​​system plans the robot's movement path based on the on-site environment and the initial placement position.

[0040] In one embodiment of the present invention: In step 1, a drone equipped with a high-definition camera is used to capture clear images of the insulator string from different angles and distances, and the average method formula is used:

[0041]

[0042] Among them, R represents the pixel value of the red channel, G represents the pixel value of the green channel, B represents the pixel value of the blue channel, and 3 represents the number of color channels calculated. The original color image can be converted into a grayscale image. During the image acquisition process, due to atmospheric jitter and lighting changes, the image contains noise. The bilateral filtering algorithm is used to remove the noise in the image. The spatial distance weight formula is:

[0043]

[0044] Among them, ω d represents the spatial distance weight, σ d represents the standard deviation of spatial distance, P represents the current pixel, Q represents the neighboring pixel, d(P,Q) represents the spatial distance between the current pixel and the neighboring pixel, and the grayscale difference weight formula is:

[0045]

[0046] Among them, ω r represents the grayscale difference weight, σ r Grayscale difference standard deviation, I(P) represents the grayscale value of the current pixel, I(Q) represents the grayscale value of the neighboring pixel, and the final weight ω=ω d ×ω r , multiply the weights of all pixels in the neighborhood by the pixel value and then sum them, and then divide them by the sum of all weights to get the filtered pixel value, which can remove noise while retaining the edge of the image. Then use the Canny operator edge detection technology to extract the contour information of the insulator string, and use the contour tracking algorithm to connect the edge pixels to form a complete insulator string contour curve. The insulator string contour curve in the processed image is annotated, and the annotation information is saved in JSON format. The annotated data can be used as the actual situation of the target for model learning comparison in subsequent training.

[0047] In one embodiment of the present invention: In step 2, the staff observes the shape of the tower on site to determine its type, and uses a total station to accurately measure the tower height, and the calculation formula is:

[0048] H=S×sinα+h2-h1

[0049] Among them, H represents the height of the tower, S represents the slant moment from the center of the total station to the top of the tower, α represents the inclination angle of the telescope in the total station, h1 represents the height difference between the center of the total station and the bottom of the tower, and h2 represents the height of the total station itself. Aerial remote sensing is used to obtain high-resolution images of the area where the tower is located. The band range of remote sensing images ranges from visible light to infrared. Different band combinations can be used to identify different terrain and geomorphic features. In the visible light band, the color and texture of the surface can be observed, and the vegetation type and water body can be distinguished. In the infrared band, the health status and coverage of vegetation can be judged according to the infrared reflection characteristics of vegetation. The remote sensing data is integrated and analyzed with the existing GIS terrain and geomorphic data, and the tower location information is superimposed with the terrain and geomorphic data to intuitively obtain the terrain and geomorphic environmental information around the tower.

[0050] In one embodiment of the present invention: in step three, a relational database is used to store the collected data, a data table named insulator string is created in the relational database, including the serial number of the insulator string, the location of the insulator string, including the specific latitude and longitude coordinates and the number of the pole tower where it is located, the shape, height information, manufacturing material and installation date of the insulator string, a data table named pole tower is created in the relational database, including the type, height, location and construction date of the pole tower, a data table named robot performance parameters is created in the relational database, including the robot model, weight, size and power source, the working range of the robot, including the maximum extension length of the robot arm and the angle range that can be covered, for the insulator string and pole tower data collected by the drone, a corresponding data entry interface program is developed, and the parsed data is inserted into the corresponding data table according to the designed database structure, and the robot performance parameter data is entered into the database through the data entry interface by exporting relevant parameter information from the robot control system after the robot is manufactured.

[0051] In one embodiment of the present invention: in step 4, a convolutional neural network learning model for processing insulator string related data is constructed, and the convolutional neural network learning model includes a convolutional layer, a pooling layer and a fully connected layer. The convolutional layer can extract key feature information of different levels from the insulator string image data, from shallow texture, edge to deep structure, and shape through the convolution kernel structure and parameter setting. The pooling layer can reduce the data dimension, reduce the amount of calculation and enhance the translation invariance of the feature. After the data is processed by the convolution and pooling layers in turn, the extracted feature map is expanded and connected to the fully connected layer. The number of neurons in the fully connected layer decreases according to the network depth, and the data can be feature integrated and classified and regressed. The last layer of the fully connected layer can be flexibly set according to specific task requirements, and can accurately output the coordinate information of the placement position area to provide accurate placement position guidance for the insulator string maintenance robot, and can also output probability values ​​representing the possibility of different placement areas to assist the robot in comprehensive judgment and selection of the optimal placement area.

[0052] In one embodiment of the present invention: In step 4, the insulator string shape and height data, tower type and height data, and terrain and geomorphic environment data obtained from drone photography are used in combination with the robot's own performance parameter data to train the convolutional neural network learning model. The training process uses a back propagation algorithm to adjust the connection weights between network layers based on the difference between the data label and the model prediction result, so that the convolutional neural network learning model can accurately identify the characteristics of the insulator string, determine its operating status, and plan the robot's movement path and operation actions according to the robot's capabilities and environmental conditions, and gradually learn to select a more suitable placement position based on factors such as the position of the insulator string and the structure of the tower.

[0053] In one embodiment of the present invention: In step five, a high-precision harmonic reducer is used. The harmonic reducer can accurately convert the high-speed rotation of the servo motor into the low-speed and high-torque movement required by the robot arm joint, and can provide an accurate transmission ratio and a smaller tooth gap, thereby reducing the error transmission during joint movement. A laser interferometer is used to detect the motion error of the robot arm joint, and software compensation is used to optimize the robot arm joint accuracy based on the detected error. The corresponding error compensation amount is calculated based on the current working state of the robot arm, including the load weight and movement speed of the robot arm, and then the target angle or displacement of the joint is corrected in the control signal to reduce the actual motion error.

[0054] In one embodiment of the present invention: in step five, a dynamic model is established for the robot based on the Lagrange equation to describe the relationship between the position, velocity and acceleration state variables of each joint of the robot mechanical arm and the motor torque control input, the current state parameters of the robot mechanical arm are obtained through sensors, including the angle, angular velocity and angular acceleration of each joint, and the model predictive control (MPC) algorithm is used to predict the system state at the future moment. The MPC algorithm predicts the motion trajectory in the prediction time domain based on the current system state using the established prediction model, and based on the set optimization goals, including minimizing joint motion errors, reducing energy consumption and constraints, such as joint angle range limitations and motor torque limitations, the optimal control sequence in the control time domain is solved by the optimization algorithm, the first control input is taken from the solved optimal control sequence and applied to the system, after the control input is implemented, the new state information of the system is obtained again through sensors, the system state variables are updated as the initial state for the next control cycle prediction, and the above prediction, optimization and implementation processes are repeated, and the cycle is rolled so that the system can continuously adapt to internal and external changes and always run towards the optimization goal.

[0055] In one embodiment of the present invention: in step six, a laser radar is used to scan the pole tower, insulator string and obstacles in the surrounding environment, and a visual sensor is used to obtain image information of the on-site environment. The high-definition camera can clearly capture the appearance of the insulator string, the structural characteristics of the pole tower and the overall appearance of the surrounding environment. The binocular visual camera can also obtain the depth information of the object by calculating the parallax, which complements the data of the laser radar to further improve the three-dimensional perception of the on-site environment. The AI ​​system uses the point cloud data of the laser radar and the image information of the visual sensor to build a detailed on-site environment map, and relies on the path planning algorithm to quickly determine the optimal path from the starting point to the target point. In the process of path planning, the robot's own performance characteristics and actual operating conditions are always closely combined to ensure the scientificity and feasibility of the planned path. When the robot moves along the planned path, the AI ​​system continuously monitors the on-site environment through sensors, and promptly detects changes in the environment, such as the appearance of new obstacles and the movement of the original obstacle position, and immediately replans and partially adjusts the path.

[0056] Embodiment 1: After configuring the relevant parameters, run the training script to start the model training process. The YOLOv5 model will be based on the architecture of the convolutional neural network (CNN). By continuously performing forward propagation (predicting the position and category of the insulator string based on the input image) and back propagation (adjusting the network parameters of the model according to the difference between the predicted result and the actual situation of the annotation) on the training set image, the model is gradually optimized to enable it to more accurately identify and locate the insulator string. The trained model is applied to the test set image. The model will output the predicted position (presented in the form of bounding box coordinates) and category probability of the insulator string in each image. By comparing with the actual annotation of the test set image, various accuracy indicators of the model on the test set can be calculated, such as accuracy (the proportion of the number of correctly identified insulator strings to the total predicted number) and recall rate (the proportion of actually existing insulator strings that are correctly identified), so as to comprehensively evaluate the recognition and positioning effect of the model.

[0057] Embodiment 2: The AI ​​system continuously detects changes in the position and status of objects in the map based on data collected by sensors in real time, and modifies the map in a timely manner. If an obstacle that did not exist originally appears on the robot's scheduled path, such as a tree branch blown down by the wind onto the path that the robot is about to pass, after the sensor detects this change, the AI ​​system will quickly calculate a new path segment that bypasses the branch, and reasonably connect it with the original path to ensure that the robot can continue to move forward safely and smoothly. The map will immediately update the information of the area and mark the new obstacle location so that the robot can replan the path to avoid it, ensuring that the path always adapts to the actual on-site environment.

[0058] Example 3: Construct an accurate dynamic model to describe the relationship between the robot's joint angle, angular velocity, angular acceleration and other state variables and the motor torque and other control inputs, and determine the MPC parameters, including the prediction time domain N p and control time domain N c The prediction horizon determines how long in the future the system state is predicted, and the control horizon is the length of the control input sequence obtained by each optimization calculation, usually N c <N p , whose size affects the flexibility and computational complexity of the controller. Various sensors installed on the robot (such as joint angle encoders, angular velocity gyroscopes, accelerometers, etc.) collect the current state information of the system, and according to the established system dynamics model and the current system state, predict the future N p The system state at each moment is predicted and the optimization objective function is set to determine the control target that the MPC algorithm wants to achieve.

[0059] As attached Figure 1As shown, by collecting various information about the insulator string and the pole tower where it is located, the learning model uses information parameters to train the AI, so that the AI ​​system can plan the robot's movement path according to the site environment and the initial placement position.

[0060] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

[0061] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0062] The above contents are merely examples and explanations of the present invention. Various modifications or additions to the specific embodiments described or replacements in similar ways by technicians in the technical field shall fall within the protection scope of the present invention as long as they do not deviate from the invention or exceed the scope defined by the claims.

[0063] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An AI-based insulator string maintenance robot placement method, comprising collecting parameter information of the insulator string and the tower where it is located, storing the data in a database, and training the AI ​​model using a learning model, characterized in that: The specific steps of the placement method are as follows: Step 1: Use a drone to photograph the shape and height parameter information of the insulator string; Step 2: Collect data on the type, height and topographic environment of the tower where the insulator string is located; Step 3: Store the collected data in a database and record the performance parameters of the maintenance robot itself; Step 4: Use the convolutional neural network learning model in deep learning to train the selected AI model using the collected data; Step 5: Optimize the joint accuracy of the robot arm and adjust the robot's motion parameters using the model predictive control algorithm; Step 6: Scan the on-site environment in real time, and the AI ​​system plans the robot's movement path based on the on-site environment and the initial placement position.

2. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In step 1, a drone equipped with a high-definition camera is used to capture clear images of the insulator string from different angles and distances, and the average method formula is used: Among them, R represents the pixel value of the red channel, G represents the pixel value of the green channel, B represents the pixel value of the blue channel, and 3 represents the number of color channels calculated. The original color image can be converted into a grayscale image. During the image acquisition process, due to atmospheric jitter and lighting changes, the image contains noise. The bilateral filtering algorithm is used to remove the noise in the image. The spatial distance weight formula is: Among them, ω d represents the spatial distance weight, σ d represents the standard deviation of spatial distance, P represents the current pixel, Q represents the neighboring pixel, d(P,Q) represents the spatial distance between the current pixel and the neighboring pixel, and the grayscale difference weight formula is: Among them, ω r represents the grayscale difference weight, σ r Grayscale difference standard deviation, I(P) represents the grayscale value of the current pixel, I(Q) represents the grayscale value of the neighboring pixel, and the final weight ω=ω d ×ω r , multiply the weights of all pixels in the neighborhood by the pixel value and then sum them, and then divide them by the sum of all weights to get the filtered pixel value, which can remove noise while retaining the edge of the image. Then use the Canny operator edge detection technology to extract the contour information of the insulator string, and use the contour tracking algorithm to connect the edge pixels to form a complete insulator string contour curve. The insulator string contour curve in the processed image is annotated, and the annotation information is saved in JSON format. The annotated data can be used as the actual situation of the target for model learning comparison in subsequent training.

3. The Al-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In step 2, the staff observes the shape of the tower on site to determine its type, and uses a total station to accurately measure the tower height. The calculation formula is: H=S×sinα+h2-h1 Among them, H represents the height of the tower, S represents the slant moment from the center of the total station to the top of the tower, α represents the inclination angle of the telescope in the total station, h1 represents the height difference between the center of the total station and the bottom of the tower, and h2 represents the height of the total station itself. Aerial remote sensing is used to obtain high-resolution images of the area where the tower is located. The band range of remote sensing images ranges from visible light to infrared. Different band combinations can be used to identify different terrain and geomorphic features. In the visible light band, the color and texture of the surface can be observed, and the vegetation type and water body can be distinguished. In the infrared band, the health status and coverage of vegetation can be judged according to the infrared reflection characteristics of vegetation. The remote sensing data is integrated and analyzed with the existing GIS terrain and geomorphic data, and the tower location information is superimposed with the terrain and geomorphic data to intuitively obtain the terrain and geomorphic environmental information around the tower.

4. The Al-based insulator string maintenance robot placement method according to claim 3 is characterized in that: In the step three, a relational database is used to store the collected data, and a data table named insulator string is created in the relational database, which includes the serial number of the insulator string, the location of the insulator string, including the specific latitude and longitude coordinates and the number of the pole tower where it is located, the shape, height information, manufacturing material and installation date of the insulator string. A data table named pole tower is created in the relational database, including the type, height, location and construction date of the pole tower. A data table named robot performance parameters is created in the relational database, including the robot model, weight, size and power source, and the working range of the robot, including the maximum length of the robot arm and the angle range that can be covered. For the insulator string and pole tower data collected by the drone, a corresponding data entry interface program is developed, and the parsed data is inserted into the corresponding data table according to the designed database structure. After the robot is manufactured, the robot performance parameter data is exported from the robot control system. The relevant parameter information is then entered into the database through the data entry interface.

5. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In the step 4, a convolutional neural network learning model for processing insulator string related data is constructed. The convolutional neural network learning model includes a convolutional layer, a pooling layer and a fully connected layer. The convolutional layer can extract key feature information of different levels from the insulator string image data, from shallow texture, edge to deep structure, and shape through the convolution kernel structure and parameter setting. The pooling layer can reduce the data dimension, reduce the amount of calculation and enhance the translation invariance of the feature. After the data is processed by the convolution and pooling layers in turn, the extracted feature map is expanded and connected to the fully connected layer. The number of neurons in the fully connected layer decreases according to the network depth, and the data can be feature integrated and classified and regressed. The last layer of the fully connected layer can be flexibly set according to specific task requirements, and can accurately output the coordinate information of the placement position area to provide accurate placement position guidance for the insulator string maintenance robot, and can also output probability values ​​representing the possibility of different placement areas to assist the robot in comprehensive judgment and selection of the optimal placement area.

6. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In the step 4, the insulator string shape and height data, tower type and height data, and terrain and geomorphic environment data obtained from drone photography are used to train the convolutional neural network learning model in combination with the robot's own performance parameter data. The training process adopts a back-propagation algorithm, and the connection weights between network layers are adjusted according to the difference between the data label and the model prediction result, so that the convolutional neural network learning model can accurately identify the characteristics of the insulator string, judge its operating status, and plan the robot's movement path and operation actions according to the robot's capabilities and environmental conditions, and gradually learn to select a more suitable placement position according to factors such as the position of the insulator string and the structure of the tower.

7. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In the step five, a high-precision harmonic reducer is used. The harmonic reducer can accurately convert the high-speed rotation of the servo motor into the low-speed and high-torque movement required by the robot arm joint, and can provide an accurate transmission ratio and a small tooth gap, thereby reducing the error transmission during joint movement. A laser interferometer is used to detect the motion error of the robot arm joint. According to the detected error, software compensation is used to optimize the robot arm joint accuracy. According to the current working state of the robot arm, including the load weight and movement speed of the robot arm, the corresponding error compensation amount is calculated, and then the target angle or displacement of the joint is corrected in the control signal to reduce the actual motion error.

8. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In the step five, a dynamic model is established for the robot based on the Lagrange equation to describe the relationship between the position, velocity and acceleration state variables of each joint of the robot mechanical arm and the motor torque control input. The current state parameters of the robot mechanical arm, including the angle, angular velocity and angular acceleration of each joint, are obtained through sensors. The system state at the future moment is predicted using a model predictive control (MPC) algorithm. The MPC algorithm predicts the motion trajectory in the prediction time domain based on the current system state using the established prediction model. Based on the set optimization goals, including minimizing joint motion errors, reducing energy consumption and constraints, such as joint angle range limitations and motor torque limitations, the optimal control sequence in the control time domain is solved by the optimization algorithm. The first control input is taken from the solved optimal control sequence and applied to the system. After the control input is implemented, the new state information of the system is obtained again through sensors, and the system state variables are updated as the initial state for the next control cycle prediction. The above prediction, optimization and implementation processes are repeated, and the cycle is repeated so that the system can continuously adapt to internal and external changes and always run towards the optimization goal.

9. The AI-based insulator string maintenance robot placement method according to claim 1 is characterized in that: In the step six, a laser radar is used to scan the pole tower, insulator string and obstacles in the surrounding environment, and a visual sensor is used to obtain image information of the on-site environment. The high-definition camera can clearly capture the appearance of the insulator string, the structural characteristics of the pole tower and the overall appearance of the surrounding environment. The binocular vision camera can also obtain the depth information of the object by calculating the parallax, which complements the data of the laser radar to further improve the three-dimensional perception of the on-site environment. The AI ​​system uses the point cloud data of the laser radar and the image information of the visual sensor to build a detailed on-site environment map, and relies on the path planning algorithm to quickly determine the optimal path from the starting point to the target point. In the process of path planning, the robot's own performance characteristics and actual operating conditions are always closely combined to ensure the scientificity and feasibility of the planned path. When the robot moves along the planned path, the AI ​​system continuously monitors the on-site environment through sensors, and promptly detects changes in the environment, such as the appearance of new obstacles and the movement of the original obstacle position, and immediately replans and partially adjusts the path.

Images