Power plant inspection robot and method based on 5G network

Abstract

The invention relates to the field of power systems, in particular to a power plant inspection robot and method based on a 5G network. The robot is provided with the movable base to achieve movement in a power plant, and the camera shooting assembly and the sensing device are integrated on the equipment base and used for monitoring the environment and automatically detecting the position of workers, the equipment temperature and obstacles correspondingly. And the 5G communication module is connected with a base station to realize real-time transmission of data. The control module is internally provided with a path planning algorithm and a deep learning model so as to intelligently plan an inspection path and identify personnel in the inspection path. The method comprises the steps of robot driving, environment monitoring, data sensing, data transmission, path planning, personnel identification and path updating based on an identification result. Through combination of a path planning algorithm and a deep learning model, the autonomous navigation and intelligent identification capabilities of the robot in a complex environment are enhanced, and the inspection efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a 5G network-based power plant inspection robot and method. Background Art

[0002] In the field of power systems, especially the safety inspection work of thermal power plants, it is crucial to ensure the stable operation of power generation facilities and the safety of personnel. Traditionally, this task relies on manual inspections, but this method has problems such as low efficiency, high cost, and safety risks. With the development of automation technology, automated inspection robot systems have begun to be introduced to improve the efficiency and safety of inspection work. However, existing automated inspection robot systems, such as the inspection control method, device and storage medium of an inspection robot described in CN115237131A, and an outdoor autonomous inspection robot for a substation based on collaborative optimization described in CN118051054A, still have some limitations.

[0003] The inspection robot described in CN115237131A mainly focuses on receiving the original route map and scheduled inspection points of the power plant, planning the scheduled inspection path, and recording the inspection data. Although this method improves the automation of inspection, it still needs to be improved in terms of real-time performance, accuracy and intelligence, especially in complex environments, where the robot's path is easily out of control, communication is delayed, image transmission is unclear, and it is unable to intelligently identify and process inspection data.

[0004] The inspection robot in CN118051054A emphasizes collaborative optimization and improvement of inspection capabilities under multi-tasking, but still lacks in-depth discussion on 5G network applications, which is crucial for achieving real-time and intelligent requirements in the context of big data, cloud computing and the Internet of Things.

[0005] Thermal power plants have concentrated equipment, widely distributed pipelines, small spaces, and a relatively complex environment. Manual inspections are costly and inefficient, and there are certain safety risks for workers. With the widespread promotion of the concept of "smart grid", the construction of smart power plants has become an inevitable trend in the development of power grid technology. Using intelligent robots to replace manual labor can effectively avoid the shortcomings of manual inspections, such as high labor intensity, low work efficiency, and low controllability. It has the characteristics of stable working conditions, high detection accuracy, and intelligent analysis of detection information. This technology will greatly improve the reliability and safety of thermal power plant power generation, strengthen the intelligent construction of thermal power plant inspection systems, and improve the "smart grid" system.

[0006] In recent years, with the development of science and technology, some robots have been used to participate in the daily inspection of thermal power plants. The robots record videos on the moving path, and then the staff watch the videos in the background to realize the daily inspection of thermal power plants. However, these robots generally have the disadvantages of easy loss of control of the inspection path, communication delay, unclear image transmission, and inability to intelligently identify and process data, which limits their application capabilities in complex environments. Therefore, there is an urgent need for a power plant inspection robot and method based on a 5G network. Summary of the invention

[0007] The purpose of the present invention is to provide a power plant inspection robot and method based on a 5G network to solve the problems of path loss of control, communication delay, and inability to intelligently identify and process inspection data in complex environments.

[0008] To achieve the above purpose, the following technical solution is adopted.

[0009] A power plant inspection robot based on 5G network, including:

[0010] Mobile base for movement within the power plant;

[0011] The equipment base is arranged on the mobile base, and includes a camera assembly and a sensing device, wherein the camera assembly is used to monitor the environment around the inspection path, and the sensing device is used to automatically sense the position of the staff, the temperature data of the equipment and obstacles in the inspection path;

[0012] The 5G communication module installed on the equipment base is interconnected with the 5G communication base station;

[0013] A control module includes a path planning algorithm for planning the robot's inspection path when the environment in the power plant changes; and also includes a deep learning model and computer vision technology for identifying people in the inspection path and performing data processing.

[0014] A power plant inspection method based on 5G network includes the following steps:

[0015] The inspection robot is driven to move by the mobile base of the inspection robot;

[0016] The camera assembly disposed in the equipment base on the mobile base is used to monitor the environment around the inspection path to obtain detection data;

[0017] Use sensing devices to automatically sense the position of staff, temperature data of equipment, and obstacles in the inspection path, and sense data;

[0018] The 5G communication module installed on the equipment base is interconnected with the 5G communication base station to send the monitoring data and sensing data to the control terminal;

[0019] Run the path planning algorithm in the control module to plan the robot's inspection path when the environment in the power plant changes;

[0020] In the control module, deep learning models and computer vision technology are used to identify people in the inspection path and perform data processing;

[0021] Based on the recognition results, the inspection route is updated to avoid the identified staff locations until the inspection is completed.

[0022] Optionally, the path planning algorithm includes a global path planning algorithm, a local path planning algorithm and a fusion navigation algorithm.

[0023] Optionally, the global path planning algorithm includes the following steps:

[0024] Prepare preset map module data in the power plant area;

[0025] Set the starting and ending target positions of the robot inspection on the map module;

[0026] Enter the map data of the area, manually control the robot to move along the required inspection path, and mark the stop position and stop time during the movement to form a planned path;

[0027] Compare and match the coordinate points of the moving path with the path on the map to determine the location of the path on the map;

[0028] Use the map matching algorithm to generate the global optimal path from the starting point to the end point based on the comparison and matching results;

[0029] The global optimal path is used as the initial value of local path planning and as a reference for local path planning.

[0030] Optionally, the specific steps of the local path planning algorithm include:

[0031] Build a grid map of the environment and specify the robot's starting and target locations to provide basic data for path planning;

[0032] Based on the constructed grid graph, a safe collision-free path is generated from the starting point to the target location;

[0033] When the robot is performing inspections, its moving path is monitored. If an obstacle is encountered that causes the global planning path to be interrupted, it will automatically switch to local path planning.

[0034] Within the local path planning range, the TEB algorithm is used to continuously adjust the robot's posture and direction according to the current robot position and the target position;

[0035] Combine static obstacle costmap and dynamic obstacle costmap to evaluate the robot's motion performance in the environment;

[0036] When a dynamic obstacle is detected, the old robot posture is deleted and a new robot posture is added based on the evaluation results, and the robot position is updated in real time;

[0037] Through continuous iteration, a new obstacle avoidance path is generated and optimized so that the robot can bypass obstacles smoothly;

[0038] The generated optimized path is used as the actual moving path of the robot, so that the robot can update and adjust the route in real time during the inspection process.

[0039] Optionally, the specific steps of the fusion navigation algorithm include:

[0040] Receive an initial path plan provided by a global path planning algorithm, where the initial path plan defines an approximate path from the robot's current position to the target position;

[0041] While executing global path planning, the robot's surrounding environment is monitored in real time, and sensor data is used to identify static and dynamic obstacles to obtain obstacle information;

[0042] The local path planning algorithm generates a collision-free local path based on the identified obstacle information and determines the robot's immediate movement instructions;

[0043] The local path generated by the local path planning algorithm is integrated with the initial path provided by the global path planning algorithm to adjust and optimize the actual travel route of the robot;

[0044] During the fusion process, the adjustment of the local path is made not to deviate from the predetermined direction of the global path, and the robot is made to avoid all identified obstacles;

[0045] Dynamically update the global path planning to adapt to environmental changes and local path adjustments, so that the global path and local path remain consistent and continuous;

[0046] Through an iterative process, the path is continuously optimized, allowing the robot to flexibly respond to local environmental changes while following the global path.

[0047] Optionally, the steps of using deep learning models and computer vision technology in the control module to identify personnel in the inspection path and perform data processing include:

[0048] Use cameras or other visual sensors to capture images or videos in the area and obtain image data of the environment where workers may be present;

[0049] Preprocess the acquired image data using image processing techniques, including denoising, enhancement, and standardization;

[0050] Use computer vision technology to analyze the pre-processed images and identify the characteristics of the staff;

[0051] Based on the identified features, a pre-trained deep learning algorithm is used to train the model to locate the identified workers and determine their exact position and posture in the image data;

[0052] The results are combined with the robot’s current position and path planning data to update the robot’s inspection path to avoid the identified worker locations.

[0053] Optionally, the pre-training step of the pre-trained deep learning algorithm training model includes:

[0054] Collect facial image data and work clothes image data of all workers in the factory and build a data set for visual detection of workers;

[0055] Divide the prepared data set into a training set and a test set, where the training set is used for subsequent training and optimization of the deep learning algorithm training model, and the test set is used to evaluate the performance of the deep learning algorithm training model;

[0056] Use deep learning techniques to build convolutional neural network models;

[0057] The constructed convolutional neural network model is used to extract features from the image of the suspected staff member, and the target object is identified through the classifier to obtain the identification result;

[0058] While identifying the target object, a regressor is used to determine the position of the target object, which is completed through bounding box positioning to obtain the position result;

[0059] Combine the recognition results and positions to build a complete convolutional neural network model;

[0060] The constructed convolutional neural network model is evaluated using the divided test set, and the convolutional neural network model is optimized and adjusted according to the evaluation results to obtain an optimized convolutional neural network model.

[0061] Optionally, use the deep learning framework of TensorFlow or PyTorch to train and learn the convolutional neural network model.

[0062] Optionally, it also includes identifying the number, name and work status of the staff through deep learning models and computer vision technology, and sending the identification results to the control terminal.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] The present invention provides a power plant inspection robot integrated with a 5G communication module, which can achieve fast, high-bandwidth data transmission, improve the real-time and reliability of inspections. Through the combination of path planning algorithms and deep learning models, the robot's autonomous navigation and intelligent recognition capabilities in complex environments are enhanced, improving inspection efficiency and safety.

[0065] The present invention provides a power plant inspection method based on 5G network, which reduces manual intervention, labor intensity and safety risks through automated inspection process. Real-time data transmission realized by 5G network ensures accurate control and immediate response of inspection robot by control terminal. By combining global path planning algorithm, local path planning algorithm and fusion navigation algorithm, the path adaptability and obstacle avoidance ability of inspection robot in complex environment are improved. The response speed and processing ability of robot to environmental changes are enhanced, ensuring the optimization and safety of inspection path.

[0066] The global path planning algorithm improves the accuracy and reliability of path planning, allowing the robot to cover the predetermined inspection area more effectively.

[0067] The map matching algorithm improves the accuracy of robot positioning and provides a solid foundation for the generation of the global optimal path. The local path planning algorithm enhances the robot's ability to cope with sudden obstacles and ensures the continuity and safety of the inspection process. The dynamic path adjustment optimizes the robot's route, improves the inspection efficiency and the robot's operational flexibility.

[0068] The steps of the fusion navigation algorithm provide an effective mechanism to combine global path planning with local path planning, improving the robot's navigation performance in complex environments. By dynamically updating the path planning, it ensures that the robot can flexibly respond to local environmental changes while following the global path.

[0069] Through the application of deep learning models and computer vision technology, the accuracy of identifying people in the inspection path is improved, and the data processing capability is enhanced. The robot's ability to avoid staff is optimized, reducing potential safety risks and improving the safety of inspections.

[0070] The pre-trained deep learning algorithm training model improves the accuracy and robustness of staff identification. By building and evaluating convolutional neural network models, the performance of target detection is improved, providing strong technical support for the robot's intelligent decision-making. The use of TensorFlow or PyTorch deep learning frameworks improves the efficiency and flexibility of model training, allowing the model to quickly adapt to new data and environmental changes. By using advanced deep learning frameworks, the performance and accuracy of the model are improved, providing strong computing support for the robot's intelligent identification. The number, name and working status of staff are identified through deep learning models and computer vision technology, which improves the richness and availability of inspection information. The identification results are sent to the control terminal, which enhances the monitoring ability of the on-site situation and provides real-time data support for safety management and emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a module schematic diagram of an embodiment of a power plant inspection robot based on a 5G network according to the present invention.

[0072] Figure 2 It is a schematic diagram of the step flow of an embodiment of a power plant inspection method based on a 5G network according to the present invention.

[0073] Figure 3 It is a grid schematic diagram of a local path planning algorithm according to an embodiment of a power plant inspection method based on a 5G network of the present invention.

[0074] Figure 4 It is a flowchart of the steps of a deep learning algorithm training model according to an embodiment of a power plant inspection method based on a 5G network of the present invention. DETAILED DESCRIPTION

[0075] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0076] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.

[0077] Example 1

[0078] like Figure 1As shown, a power plant inspection robot based on a 5G network includes a mobile base designed to move smoothly within a thermal power plant, capable of carrying the main components of the robot and maintaining stability in a complex power plant environment. An equipment base is mounted on the mobile base, on which a camera assembly and a sensing device are integrated. The camera assembly includes a high-definition camera for real-time monitoring of the environment around the inspection path and capturing image or video data for subsequent analysis and processing. The sensing device includes, but is not limited to, infrared sensors, thermal imagers, and lidars, which are used to automatically sense the position of staff in the inspection path, the temperature data of the equipment, and obstacle information to ensure that the robot can respond to environmental changes in a timely manner.

[0079] The equipment base is also equipped with a 5G communication module, which can establish a high-speed, low-latency wireless connection with a 5G communication base station to achieve real-time data transmission. Through the 5G network, the robot can quickly send the collected environmental images, sensing data and processing results to the remote control terminal and receive control instructions at the same time.

[0080] The control module is the robot’s brain and contains a path planning algorithm that can dynamically plan and adjust the robot’s inspection path when the plant environment changes, such as when equipment moves or obstacles appear. The module also includes deep learning models and computer vision technology to process image data captured by the camera component, identify people in the inspection path, and perform related data processing. The deep learning model is pre-trained to recognize specific features of workers, such as human shape, color, and movement patterns, while computer vision technology is used to analyze image content, extract features, and perform target recognition and tracking.

[0081] In actual applications, the robot first collects environmental data through camera components and sensing devices, and then transmits the data to the control module through the 5G communication module. The path planning algorithm in the control module calculates the optimal inspection path in real time based on the received data, and guides the robot's mobile base to move along the path. At the same time, the deep learning model and computer vision technology analyze the collected images, identify the staff in the inspection path, and feed back their position information to the control module so that the inspection path can be adjusted in time to ensure the safety of personnel.

[0082] In addition, the control module can also perform fault prediction and status monitoring based on the equipment temperature data collected by the sensing device, and discover potential safety hazards in advance. The operation of the robot can be monitored and intervened through the remote control terminal to ensure the smooth completion of the inspection task.

[0083] Through the above implementation, the power plant inspection robot based on the 5G network can effectively improve the efficiency and safety of thermal power plant inspections, while reducing the risks and costs of manual inspections.

[0084] Example 2

[0085] like Figure 2 As shown, a power plant inspection method based on a 5G network uses a power plant inspection robot to perform the following steps:

[0086] The mobile base of the inspection robot is activated to move along the predetermined starting path in the power plant. The mobile base is equipped with a power system that can smoothly travel on different ground conditions and can carry the robot body and the equipment it carries.

[0087] The equipment base on the robot's mobile base is equipped with a camera component to monitor the environment around the inspection path. The camera component includes a high-definition camera that can capture high-resolution images or videos for subsequent environmental analysis and personnel identification.

[0088] At the same time, the sensing device automatically senses relevant information in the inspection path, including the location of the staff, the temperature data of the equipment, and potential obstacles. The sensing device may include infrared sensors, thermal imagers, and lidar, which can collect data on the environment and obstacles in real time.

[0089] The 5G communication module on the robot establishes a connection with the 5G communication base station, and transmits the detection data captured by the camera component and the sensing data collected by the sensing device to the remote control terminal at high speed and low latency through the 5G network.

[0090] In the remote control terminal or the robot's control module, a path planning algorithm is run. This algorithm can dynamically plan and adjust the robot's inspection path based on real-time data and changes in the power plant environment to ensure inspection efficiency and safety.

[0091] The deep learning model and computer vision technology in the control module start working, analyzing the transmitted image data and identifying the personnel in the inspection path. The deep learning model can recognize the specific characteristics of the staff, such as human shape, color and movement pattern, while computer vision technology is used to analyze the image content, extract features, and perform target recognition and tracking.

[0092] Based on the recognition results of deep learning models and computer vision technology, the control module updates the inspection path to avoid the identified positions of staff and other obstacles, ensuring that the robot's inspection path is safe and efficient.

[0093] The robot continues to perform inspection tasks according to the updated path until it completes all the scheduled inspection areas. During the entire inspection process, the robot can respond to environmental changes in real time to ensure the smooth completion of the inspection task.

[0094] Through the above implementation methods, the power plant inspection method based on 5G network can realize efficient and safe inspection of thermal power plants, reduce the risks and costs of manual inspections, and improve the intelligence level of inspections.

[0095] As a specific example of the above example, a power plant inspection method based on a 5G network specifically includes:

[0096] Global path planning algorithm: This method first uses laser SLAM technology to scan the power plant environment using a two-dimensional laser radar to build a high-precision map of the power plant. On this basis, the A algorithm is used for global path planning to improve navigation efficiency and help the robot quickly plan the best path. The A algorithm combines heuristic search and breadth-first algorithm. It selects the search direction through the cost function and expands around the starting point, repeating until it reaches the end point, generating a path from the starting point to the end point.

[0097] Local Path Planning Algorithm: Real-time obstacles may appear while the robot is moving along the global path. To avoid collisions, a technique for local path planning with time elasticity, the Timed Elastic Band (TEB) algorithm, is used to achieve real-time obstacle avoidance based on global path planning. The TEB algorithm is an optimization algorithm that follows the path generated by the global path planner and generates a local trajectory consisting of a series of continuous time and posture sequences.

[0098] Fusion navigation algorithm: The fusion navigation algorithm combines the results of global path planning and local path planning to achieve optimal global path planning and real-time obstacle avoidance. The algorithm uses lidar as the core sensor, integrates IMU, GNSS, OD and other sensors to increase data observation sources, combines pre-collected scene maps, and uses data source reliability adaptive analysis, weighted fusion, and extended Kalman filtering to perform data fusion processing, thereby improving the system's fault tolerance and adaptability.

[0099] Application of 5G communication module: In terms of data transmission, this method uses 5G communication for data transmission, and utilizes the technical characteristics of the 5G network such as large bandwidth, low latency, high reliability, and massive connections to provide strong guarantees for the robot to receive instructions from the monitoring center and transmit the collected information back to the monitoring center.

[0100] Deep learning and computer vision technology: In the control module, deep learning models and computer vision technology are used to identify people in the inspection path and perform data processing. The deep learning model is pre-trained to recognize specific characteristics of workers, such as human shape, color, and movement patterns, while computer vision technology is used to analyze image content, extract features, and perform target recognition and tracking.

[0101] As a specific example of the above example, the detailed steps of the global path planning algorithm include:

[0102] First, map module data for the power plant area is prepared. This data includes a detailed layout of the power plant, which clearly identifies the location of each device, pipeline, building, and other obstacles. The map module data can be two-dimensional or three-dimensional and is stored in a digital form so that the robotic system can read and process it.

[0103] Set the starting and ending target locations of the robot inspection on the map module. These locations can be key equipment points or inspection areas in the power plant to ensure that the robot can cover all the predetermined inspection points.

[0104] The map data of the area is entered into the robot's control system. Subsequently, the robot is manually controlled to move along the required inspection path. During this process, the system will mark the robot's stop position and stop time, thus forming a planned path. This data will be used for subsequent path optimization and automatic path planning.

[0105] Compare and match the coordinate points of the robot's moving path with the path on the map to determine the robot's actual position on the map. This step is achieved through technologies such as GPS positioning, lidar scanning or visual odometer to ensure the accuracy of the path.

[0106] Using the map matching algorithm, a global optimal path from the starting point to the end point is generated based on the comparison and matching results. The algorithm considers factors such as path length, obstacle avoidance, and energy consumption minimization to determine the most efficient inspection route.

[0107] The generated global optimal path is used as the initial value of local path planning to provide a reference for local path planning. Local path planning will take into account real-time environmental changes and dynamic obstacles, and fine-tune the global path to ensure the safe and efficient operation of the robot in the actual environment.

[0108] As a specific example of the above example, the detailed steps of the local path planning algorithm include:

[0109] First, the internal environment data of the power plant is acquired in real time through sensors such as lidar and cameras to build a grid map of the environment. This step includes dividing the environment into several grid cells and specifying the robot's starting and target positions to provide basic data for path planning.

[0110] Based on the constructed grid graph, a path search algorithm such as A* algorithm or Dijkstra algorithm is used to generate a safe and collision-free path from the starting point to the target location. This path takes into account the location of static and dynamic obstacles to ensure the safety and feasibility of the path.

[0111] When the robot performs inspection tasks, it continuously monitors its moving path. If it encounters obstacles that interrupt the global planning path, such as when a worker suddenly enters the inspection area or equipment failure creates new obstacles, the system automatically switches to local path planning to avoid these obstacles.

[0112] Within the local path planning range, the robot’s posture and orientation are continuously adjusted based on the current robot position and the target position using the TEB (Time-Elastic Band) algorithm. The TEB algorithm takes into account the robot’s kinematic constraints, such as velocity and acceleration limits, to achieve smooth path tracking.

[0113] Combine static obstacle costmap and dynamic obstacle costmap to evaluate the robot's motion performance in the environment. The static obstacle costmap is generated based on the environment layout, while the dynamic obstacle costmap is updated in real time to reflect the dynamically changing obstacle information in the environment.

[0114] When dynamic obstacles are detected, such as moving workers or equipment, the old robot posture is deleted and a new robot posture is added based on the evaluation results, and the robot's position is updated in real time to ensure the accuracy and safety of path planning.

[0115] Through continuous iteration, new obstacle avoidance paths are generated and optimized to ensure that the robot can bypass obstacles smoothly. This step involves continuous adjustment and optimization of the path to cope with real-time changes in the environment.

[0116] The generated optimized path is used as the actual moving path of the robot, so that the robot can update and adjust the route in real time during the inspection process. This step ensures that the robot can flexibly cope with the complex power plant environment and improve the efficiency and safety of the inspection.

[0117] Specifically, Figure 3 As shown in the figure, on the basis of global path planning, during the movement of the robot, if an obstacle blocks the global planning path and causes the path to be interrupted, the grid map is used to automatically plan the optimal detour path, and the end point of the detour path is the coordinate point on the global planning path that is closest to the interruption position.

[0118] Global path planning takes a static obstacle cost map as input and does not consider the robot's mechanical properties and kinematic constraints when planning the path. It uses a graph algorithm to plan the optimal path from the robot's current position to the target position and provides initial values ​​for local path planning.

[0119] Local path planning collects path nodes on the global optimal path and optimizes the global path subset between the robot's current node and the collected path nodes. It combines the static obstacle cost map and the dynamic obstacle cost map, and uses the TEB algorithm to continuously adjust the robot's posture and direction within the local path planning range, taking into account its shape, dynamic model, and motion performance. When encountering a dynamic obstacle, it deletes the old robot posture and adds a new robot posture, so that each iteration can generate a new path and obtain the optimized path through continuous iterations.

[0120] By integrating navigation algorithms, optimal global path planning and real-time obstacle avoidance are achieved during mobile robot navigation.

[0121] As a specific example of the above example, the detailed steps of the fusion navigation algorithm include:

[0122] The method starts by receiving an initial path plan provided by a global path planning algorithm. This initial path plan is based on the map data of the power plant and defines the approximate path from the robot's current position to the target position, providing a basic navigation framework for the robot's inspection task.

[0123] While executing global path planning, the robot monitors its surrounding environment in real time through its onboard sensor systems, such as lidar, cameras, infrared sensors, etc. The data collected by these sensors are used to identify static and dynamic obstacles and obtain obstacle information, including the location, size, and movement speed of the obstacles.

[0124] Based on the identified obstacle information, the local path planning algorithm is triggered to generate a collision-free local path and determine the robot's immediate movement instructions. The algorithm can quickly respond to environmental changes and ensure that the robot can safely avoid obstacles when encountering them.

[0125] The local path generated by the local path planning algorithm is merged with the initial path provided by the global path planning algorithm. This step involves adjusting and optimizing the robot's actual path to ensure that the adjustment of the local path does not deviate from the intended direction of the global path, while enabling the robot to effectively avoid all identified obstacles.

[0126] The global path planning algorithm is dynamically updated based on environmental changes and adjustments to the local path. This step ensures that the global and local paths remain consistent and continuous, allowing the robot to flexibly respond to local environmental changes while following the global path.

[0127] Through an iterative process, the above steps are repeated to continuously optimize the path so that the robot can flexibly respond to local environmental changes while following the global path. This step involves continuous evaluation and adjustment of the path to achieve optimal inspection efficiency and safety.

[0128] The fusion navigation algorithm can effectively combine the advantages of global path planning and local path planning, and improve the robot's navigation ability and adaptability in complex power plant environments. This method not only improves the automation and intelligence level of inspection, but also reduces the risk and cost of manual inspection, while improving the coverage and accuracy of inspection.

[0129] As a specific example of the above example, the steps of using the deep learning model and computer vision technology in the control module to identify people in the inspection path and perform data processing include:

[0130] Use cameras or other visual sensors to capture images or videos in the power plant area to obtain environmental image data where workers may be present. These image data include but are not limited to visible light images and infrared images, which are used for subsequent personnel identification and safety monitoring.

[0131] The acquired image data is preprocessed by image processing technology to improve image quality and prepare for the subsequent recognition step. The preprocessing steps include denoising to reduce random noise in the image, enhancement to highlight the characteristics of the staff, and standardization to unify the image data format.

[0132] Computer vision technology is used to analyze the pre-processed images to identify the characteristics of the staff. These characteristics may include human shape, color, movement pattern, etc., providing input data for the deep learning model.

[0133] Based on the identified features, a pre-trained deep learning algorithm is used to train a model to locate the worker. The model is able to identify and locate the worker’s exact position and posture from the image data, and identify the target object through a classifier.

[0134] The recognition results of the deep learning model are combined with the robot’s current position and path planning data. Based on the identified worker locations, the robot’s inspection path is dynamically updated to avoid these locations, ensuring the safety and efficiency of the inspection process.

[0135] During the inspection process, the image data and recognition results are continuously monitored, and the robot's inspection path is adjusted in real time. This step ensures that the robot can flexibly respond to changes in the internal environment of the power plant, such as the movement of staff or other emergencies.

[0136] This method can effectively improve the automation and intelligence level of power plant inspections, reduce the risks and costs of manual inspections, and improve the coverage and accuracy of inspections. In addition, through the high-speed data transmission of the 5G network, the cloud server can quickly process image data and provide real-time feedback on abnormal situations, enhancing the safety monitoring capabilities of power plants.

[0137] like Figure 4 As shown, as a specific example of the above example, the pre-training steps of the pre-trained deep learning algorithm training model include:

[0138] Collect facial image data and work clothes image data of all workers in the factory to build a dataset for worker visual detection. This step involves using high-definition cameras to capture images under different lighting conditions and backgrounds to ensure the diversity and comprehensiveness of the dataset to cover a variety of possible working environments and conditions.

[0139] The training set and the test set are divided from the prepared data set. The training set is used for subsequent training and optimization of the deep learning algorithm training model, while the test set is used to evaluate the performance of the model. Usually, the data set can be divided into 80% for training and 20% for testing, but the specific ratio can be adjusted according to actual needs.

[0140] Use deep learning techniques to build a convolutional neural network (CNN) model. This model contains multiple convolutional layers, pooling layers, and fully connected layers to automatically learn feature representations from images. Model building also includes choosing a suitable activation function, such as ReLU, and setting an optimization algorithm, such as Adam or SGD.

[0141] The constructed convolutional neural network model is used to extract features from the image of the suspected staff member, and the target object is identified through the classifier to obtain the identification result. This step involves inputting the preprocessed image into the CNN model, and the model automatically extracts image features and classifies them.

[0142] While identifying the target object, a regressor is used to determine the location of the target object, which is completed through bounding box positioning to obtain the location result. This step involves integrating or running a regression network such as YOLO or SSD in parallel with the CNN model to determine the precise location of the worker.

[0143] Combining the recognition results and the position, a complete convolutional neural network model is constructed. This model can not only recognize the worker, but also determine their exact position in the image, providing precise data for the robot's obstacle avoidance and path planning.

[0144] Use the divided test set to evaluate the constructed convolutional neural network model, and optimize and adjust the convolutional neural network model based on the evaluation results. Evaluation indicators may include accuracy, recall rate, and F1 score. According to the evaluation results, adjust the model parameters or network structure to improve the performance and generalization ability of the model.

[0145] The pre-trained deep learning algorithm training model can accurately identify and locate personnel in the power plant inspection path, providing strong technical support for the power plant inspection method based on 5G network. This method improves the automation and intelligence level of inspection and ensures the safety and efficiency of the power plant.

[0146] As a preferred example of the above example, the convolutional neural network model is trained and learned using the deep learning framework of TensorFlow or PyTorch:

[0147] Choose TensorFlow or PyTorch as your deep learning framework. TensorFlow is known for its scalability, versatile API, and TensorBoard visualization tool, which is suitable for large-scale training and production deployment. PyTorch is favored for its dynamic computational graphs, Python nature, and rich libraries and tools, and is particularly suitable for research and development.

[0148] Collect facial image data and work clothes image data of all workers in the factory and build a dataset for worker visual detection. The dataset needs to contain diverse images to ensure the generalization ability of the model.

[0149] The constructed dataset is divided into training set and test set, usually in a certain ratio (such as 80% training set and 20% test set) to evaluate the model performance.

[0150] Construct a convolutional neural network (CNN) model in the selected deep learning framework. The model contains an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer.

[0151] Use the training set to train the CNN model. In TensorFlow, you can train the model by defining the model structure, loss function, and optimizer. In PyTorch, you can define the model class, data loader, loss function, and optimizer, and perform iterative training.

[0152] The trained CNN model is used to extract features from images of suspected staff members, and the target object is identified through the classifier to obtain the identification result.

[0153] While identifying the target object, a regressor is used to determine the position of the target object, which is completed through bounding box positioning to obtain the position result.

[0154] Use the test set to evaluate the constructed CNN model, and optimize and adjust the model based on the evaluation results. Evaluation indicators may include accuracy, recall, etc.

[0155] The optimized convolutional neural network model is deployed into the robot’s control module to identify and locate personnel in the inspection path in real time.

[0156] Example 3

[0157] A power plant inspection method based on a 5G network also includes the steps of identifying the number, name and working status of staff members by using a deep learning model and computer vision technology, and sending the identification results to a control terminal:

[0158] First, use high-definition cameras or other visual sensors to capture images or videos of workers in the power plant area and obtain image data of the environment where the workers may be present. This image data will be used for subsequent personnel identification and status monitoring.

[0159] The acquired image data is preprocessed by image processing techniques, including denoising, enhancement, and standardization, to improve image quality and prepare for subsequent recognition steps. The preprocessing step may include calculating the 0th and 1st order moments of the image, the neighborhood centroid of the feature points, and the main direction of the feature points, thereby extracting the key features in the image.

[0160] The convolutional neural network model is trained and learned using the deep learning framework of TensorFlow or PyTorch. During the model training process, the training set divided from the data set is used to train the model to recognize the characteristics of the staff through forward propagation and optimization of the loss function.

[0161] The trained deep learning model is used to extract features from the preprocessed image, and the target object is identified through the classifier to obtain the recognition result. At the same time, the regressor is used to determine the position of the target object, which is completed through bounding box positioning to obtain the position result.

[0162] After identifying the staff, the deep learning model is further used to identify the staff’s name and work status. This may involve facial recognition, work clothes identification, and behavioral pattern analysis to determine the staff’s identity and current work status.

[0163] The recognition results, including the number, name and working status of the workers, are combined with the robot's current location and path planning data to update the robot's inspection path to avoid the identified workers. At the same time, these recognition results are sent to the control terminal in real time via the 5G network for further analysis and response by the monitoring center.

[0164] After the control terminal receives the identification results, the operator at the monitoring center can monitor the status of the workers in the power plant in real time and intervene when necessary to ensure the safe operation of the power plant and the safety of the workers.

[0165] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A power plant inspection robot based on 5G network, characterized in that: include, Mobile base for movement within the power plant; The equipment base is arranged on the mobile base, and includes a camera assembly and a sensing device, wherein the camera assembly is used to monitor the environment around the inspection path, and the sensing device is used to automatically sense the position of the staff, the temperature data of the equipment and obstacles in the inspection path; The 5G communication module installed on the equipment base is interconnected with the 5G communication base station; A control module includes a path planning algorithm for planning the robot's inspection path when the environment in the power plant changes; and also includes a deep learning model and computer vision technology for identifying people in the inspection path and performing data processing.

2. A 5G network-based power plant inspection method, based on the 5G network-based power plant inspection robot according to claim 1, characterized in that: The following steps are included: The inspection robot is driven to move by the mobile base of the inspection robot; The camera assembly disposed in the equipment base on the mobile base is used to monitor the environment around the inspection path to obtain detection data; Use sensing devices to automatically sense the position of staff, temperature data of equipment, and obstacles in the inspection path, and sense data; The 5G communication module installed on the equipment base is interconnected with the 5G communication base station to send the monitoring data and sensing data to the control terminal; Run the path planning algorithm in the control module to plan the robot's inspection path when the environment in the power plant changes; In the control module, deep learning models and computer vision technology are used to identify people in the inspection path and perform data processing; Based on the recognition results, the inspection route is updated to avoid the identified staff locations until the inspection is completed.

3. According to a 5G network-based power plant inspection method according to claim 2, it is characterized in that: The path planning algorithm includes a global path planning algorithm, a local path planning algorithm and a fusion navigation algorithm.

4. According to a 5G network-based power plant inspection method according to claim 3, it is characterized in that: The global path planning algorithm comprises the following steps: Prepare preset map module data in the power plant area; Set the starting and ending target positions of the robot inspection on the map module; Enter the map data of the area, manually control the robot to move along the required inspection path, and mark the stop position and stop time during the movement to form a planned path; Compare and match the coordinate points of the moving path with the path on the map to determine the location of the path on the map; Use the map matching algorithm to generate the global optimal path from the starting point to the end point based on the comparison and matching results; The global optimal path is used as the initial value of local path planning and as a reference for local path planning.

5. According to a 5G network-based power plant inspection method according to claim 3, it is characterized in that: The specific steps of the local path planning algorithm include: Build a grid map of the environment and specify the robot's starting and target locations to provide basic data for path planning; Based on the constructed grid graph, a safe collision-free path is generated from the starting point to the target location; When the robot is performing inspections, its moving path is monitored. If an obstacle is encountered that causes the global planning path to be interrupted, it will automatically switch to local path planning. Within the local path planning range, the TEB algorithm is used to continuously adjust the robot's posture and direction according to the current robot position and the target position; Combine static obstacle costmap and dynamic obstacle costmap to evaluate the robot's motion performance in the environment; When a dynamic obstacle is detected, the old robot posture is deleted and a new robot posture is added based on the evaluation results, and the robot position is updated in real time; Through continuous iteration, a new obstacle avoidance path is generated and optimized so that the robot can bypass obstacles smoothly; The generated optimized path is used as the actual moving path of the robot, so that the robot can update and adjust the route in real time during the inspection process.

6. According to a 5G network-based power plant inspection method according to claim 3, it is characterized in that: The specific steps of the fusion navigation algorithm include: Receive an initial path plan provided by a global path planning algorithm, where the initial path plan defines an approximate path from the robot's current position to the target position; While executing global path planning, the robot's surrounding environment is monitored in real time, and sensor data is used to identify static and dynamic obstacles to obtain obstacle information; The local path planning algorithm generates a collision-free local path based on the identified obstacle information and determines the robot's immediate movement instructions; The local path generated by the local path planning algorithm is integrated with the initial path provided by the global path planning algorithm to adjust and optimize the actual travel route of the robot; During the fusion process, the adjustment of the local path is made not to deviate from the predetermined direction of the global path, and the robot is made to avoid all identified obstacles; Dynamically update the global path planning to adapt to environmental changes and local path adjustments, so that the global path and local path remain consistent and continuous; The above steps are repeated through an iterative process to continuously optimize the path, so that the robot can flexibly respond to local environmental changes while following the global path.

7. According to a 5G network-based power plant inspection method according to claim 2, it is characterized in that: The steps of using deep learning models and computer vision technology in the control module to identify people in the inspection path and perform data processing include: Use cameras or other visual sensors to capture images or videos in the area and obtain image data of the environment where workers may be present; Preprocess the acquired image data using image processing techniques, including denoising, enhancement, and standardization; Use computer vision technology to analyze the pre-processed images and identify the characteristics of the staff; Based on the identified features, a pre-trained deep learning algorithm is used to train the model to locate the identified workers and determine their exact position and posture in the image data; The results are combined with the robot’s current position and path planning data to update the robot’s inspection path to avoid the identified worker locations.

8. A power plant inspection method based on 5G network according to claim 7, characterized in that: The pre-training step of the pre-trained deep learning algorithm training model includes: Collect facial image data and work clothes image data of all workers in the factory and build a data set for visual detection of workers; Divide the prepared data set into a training set and a test set, where the training set is used for subsequent training and optimization of the deep learning algorithm training model, and the test set is used to evaluate the performance of the deep learning algorithm training model; Use deep learning techniques to build convolutional neural network models; The constructed convolutional neural network model is used to extract features from the image of the suspected staff member, and the target object is identified through the classifier to obtain the identification result; While identifying the target object, a regressor is used to determine the position of the target object, which is completed through bounding box positioning to obtain the position result; Combine the recognition results and positions to build a complete convolutional neural network model; The constructed convolutional neural network model is evaluated using the divided test set, and the convolutional neural network model is optimized and adjusted according to the evaluation results to obtain an optimized convolutional neural network model.

9. A power plant inspection method based on 5G network according to claim 8, characterized in that: Use TensorFlow or PyTorch deep learning framework to train and learn convolutional neural network models.

10. A power plant inspection method based on 5G network according to claim 7, characterized in that: It also includes identifying the number, name and work status of staff through deep learning models and computer vision technology, and sending the identification results to the control terminal.

Images