An indoor substation intelligent inspection method and a UAV system

Abstract

This invention provides an intelligent inspection method and unmanned aerial vehicle (UAV) system for indoor substations, belonging to the field of intelligent inspection technology for power equipment. The method includes: after the UAV completes initialization and takes off to a preset inspection altitude, constructing a path optimization objective function to generate an inspection path covering the equipment to be inspected; controlling the UAV to fly along the path and simultaneously collecting visible light image data and infrared thermal imaging data; fusing the two types of data and using an improved deep learning model to jointly identify equipment appearance anomalies and temperature anomalies, generating alarm information when an anomaly is detected; during flight, real-time monitoring of the UAV's operating status and the external environment, dividing the flight process into multiple states according to a preset state management mechanism and dynamically switching them based on priority rules. This invention improves the safety, reliability, and fault detection accuracy of indoor substation inspections.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for power equipment, specifically to an intelligent inspection method for indoor substations and an unmanned aerial vehicle (UAV) system. Background Technology

[0002] As the scale of power systems continues to expand and the number of substation equipment increases, the difficulty of real-time monitoring and regular inspection of equipment operating status also rises. Traditional substation inspections mainly rely on manual inspections, with maintenance personnel conducting on-site checks of equipment appearance, temperature status, and operating conditions. However, manual inspections suffer from low efficiency, high workload, and accuracy issues due to human factors. Especially in complex equipment environments and environments with high electromagnetic interference, inspection personnel also face certain safety risks. Furthermore, the indoor spaces of substations are enclosed during working hours, making it impossible for personnel to enter. Ensuring safe operation and maintenance around the clock is also a challenge for indoor substation inspection work.

[0003] In recent years, drone technology has been widely used in power line inspection, but primarily for outdoor transmission line inspections. Drones, equipped with visible light cameras and infrared thermal imagers, enable remote inspection and data collection. However, indoor substations present challenges due to their confined spaces, dense equipment, complex inspection paths, and increased task difficulty. When drones experience command loss, delays, or insufficient battery life during indoor inspections, they must autonomously switch states to prevent flight accidents. Ensuring the safety of drone flights has become a major obstacle to indoor drone applications. Furthermore, most existing indoor inspection systems rely on preset routes or manual remote control, lacking intelligent autonomous planning capabilities. Additionally, in equipment fault detection, most systems still depend on manual review of video images, with limited automatic identification capabilities, making it difficult to detect equipment anomalies promptly.

[0004] The prior art discloses an automatic inspection method for unmanned aerial vehicles (UAVs). This method obtains the offline point cloud map of the scene to be inspected, the inspection route, and the waypoint task by receiving and parsing the inspection command. It determines the real-time pose of the UAV by using real-time image information of the scene around the UAV and the offline point cloud map, and performs path planning based on the real-time pose and the inspection route to generate flight control commands to control the UAV to perform the inspection task.

[0005] However, the aforementioned existing technologies primarily focus on the positioning and path planning of UAVs in complex environments. Their core lies in generating and executing inspection paths through environmental perception and path planning. But for the special application scenario of indoor substations, which presents high security and constraints, relying solely on path planning and positioning technologies is insufficient to meet the safety and task controllability requirements of UAV inspection processes. Specifically, these existing technologies have at least the following shortcomings: They lack a unified state management mechanism for the entire inspection task process, failing to systematically constrain the behavior of UAVs at different task stages; they do not address the characteristics of dense equipment and high safety requirements in indoor substation environments, failing to implement hierarchical management and safety control of abnormal states during UAV flight, making it difficult to effectively reduce operational risks; and during inspection execution, they lack a collaborative decision-making mechanism based on task status and environmental changes, making it difficult to achieve effective linkage between path execution, task switching, and anomaly handling.

[0006] In summary, for the special enclosed task scenario of indoor substations, an intelligent inspection drone system has been developed that can achieve autonomous inspection path planning and flight status management, intelligent identification of equipment operating status, and abnormal alarm in the indoor substation environment. This system liberates people from dangerous working environments, improves inspection efficiency and accuracy, and enhances the guarantee for the stable operation of indoor substation equipment. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an intelligent inspection method for indoor substations and an unmanned aerial vehicle (UAV) system.

[0008] The present invention adopts the following technical solution.

[0009] The first aspect of the present invention provides an intelligent inspection method for indoor substations, comprising: After the drone completes initialization and takes off to the preset inspection altitude, an inspection path is generated based on preset path optimization constraints, starting from the current position and covering the equipment to be inspected. The drone is controlled to fly along the inspection path and collect visible light image data and infrared thermal imaging data of the equipment to be inspected. The collected visible light image data and infrared thermal imaging data are fused together, and the abnormal status of the equipment to be inspected is identified based on the fused multimodal data. When an abnormality is detected, corresponding alarm information is generated and sent. During flight, the operating status and external environment of the UAV are monitored in real time, and the flight status of the UAV is switched according to a preset status management mechanism in response to different monitored events.

[0010] Optionally, the inspection path generated based on preset path optimization constraints, starting from the current position and covering the equipment to be inspected, includes: A path optimization objective function is constructed, which is composed of a weighted sum of path length cost, path turning angle change cost, obstacle avoidance cost, and equipment safety distance constraint cost. The weight coefficient of each cost term is a preset value and satisfies the normalization constraint. The equipment safety distance constraint cost is calculated based on the distance between the path point and different types of inspection equipment, the preset minimum safety distance for different types of equipment, and the preset weight representing the risk level of the equipment. The inspection path is generated using the path optimization objective function.

[0011] Optionally, identifying abnormal states of the equipment to be inspected based on the fused multimodal data includes: inputting the visible light image data into a deep learning backbone network to extract visual feature maps; The infrared thermal imaging data is input into a separate feature extraction branch to extract temperature feature maps; In the feature pyramid network layer, the visual feature map and the temperature feature map are concatenated at the channel level to generate a fused feature map; Target detection is performed based on the fused feature map to identify abnormalities in the appearance and temperature of the device.

[0012] Optionally, the state management mechanism divides the UAV's flight process into multiple states, including: initialization state, takeoff state, hovering state, inspection mission execution state, path adjustment state, return to home state, and manual takeover state. The initialization state is used to complete device self-testing, sensor initialization, and environmental map loading after the system is powered on. The takeoff status is used to control the drone to take off automatically and reach the preset inspection altitude; The hovering state is used for path planning and task preparation. The inspection task execution status is used to inspect substation equipment according to the planned route; The path adjustment status is used for local path replanning when the path is blocked or the environment changes; The manual takeover status is used to ensure that the drone can be manually controlled at any time; The return-to-home status is used to perform autonomous return and landing when the mission is completed or safety conditions are triggered.

[0013] Optionally, the state management mechanism defines the priority of different states: Emergency landing status has the highest priority. When emergency landing status is triggered, regardless of the current state of the drone, the response of emergency landing status takes precedence over the current state, causing the drone to switch to emergency landing status. The return-to-home status has a higher priority than hovering and normal inspection status, but a lower priority than emergency landing status. When the return-to-home status is triggered and the drone is currently in hovering or normal inspection status, the response of the return-to-home status takes precedence over the current status, causing the drone to switch to the return-to-home status. The hovering state has a higher priority than the normal inspection state. When the hovering state is triggered and the drone is currently in the normal inspection state, the hovering state response takes precedence over the normal inspection state, causing the drone to switch to the hovering state. Normal inspection status has the lowest priority. When an emergency landing status, return-to-base status, or hovering status is triggered, the response to the triggered status takes precedence over the normal inspection status.

[0014] Optionally, the flight status of the UAV can be dynamically switched according to a preset state management mechanism, including: When the UAV detects that its flight path is blocked by an obstacle while performing an inspection mission, it switches to path adjustment mode. By updating the obstacle avoidance cost and equipment safety distance constraint cost in the path optimization objective function, the current inspection path is replanned to achieve dynamic adjustment of the inspection path.

[0015] Optionally, the flight status of the UAV can be dynamically switched according to a preset state management mechanism, including: The system monitors the drone's power system and flight attitude data in real time. When the drone's angular velocity exceeds the preset safe angular velocity threshold, or the attitude angle deviates from the command value by more than the preset safe envelope threshold, it determines that an abnormal state of flight attitude instability has occurred, triggers the highest priority emergency landing mode, controls the drone to perform a controlled landing operation, and simultaneously sends fault information containing fault codes, abnormality types, and current position coordinates to the ground station.

[0016] Optionally, the flight status of the UAV can be dynamically switched according to a preset state management mechanism, including: Real-time monitoring of the drone's communication quality, positioning system status, and temperature measurement equipment operating status; When at least one sensing and communication anomaly is detected, the drone is controlled to switch to pure inertial navigation mode and enter a fixed-point hovering state, while a self-check process is initiated. Abnormal situations in perception and communication include: data packet loss rate exceeding the first preset threshold, communication delay exceeding the second preset threshold, trace of the multi-source fusion covariance matrix of the positioning system exceeding the third preset threshold, or abnormal jumps in the data of the temperature measuring device that prevent normal collection of temperature data of the device to be inspected. The self-test process includes: attempting to restart the sensor driver of the temperature measuring device to restore its normal operation, and reinitializing the real-time localization and mapping algorithm; If all abnormal indicators return to normal within the preset stable time, the self-test is deemed passed, the drone is controlled to exit the hovering state and continue to perform the inspection task; otherwise, the self-test is deemed to have failed, the emergency landing mode is triggered, the coordinates of the faulty equipment are marked simultaneously, and a manual inspection command is uploaded to the ground station.

[0017] Optionally, the flight status of the UAV can be dynamically switched according to a preset state management mechanism, including: An energy decision evaluation function is constructed based on the drone's current remaining battery power, estimated return path distance, and battery health status. The energy decision evaluation function is composed of a weighted sum of the power safety coefficient, the distance feasibility coefficient, and the battery health coefficient, with the weight of each coefficient being a preset value. When the output value of the energy decision evaluation function is higher than the first preset decision threshold, the drone is controlled to continue to perform the current inspection task. When the output value of the energy decision evaluation function is not higher than the first preset decision threshold and not lower than the second preset decision threshold, control the drone to switch to return-to-home mode. When the output value of the energy decision evaluation function is lower than the second preset decision threshold, the drone is controlled to perform an emergency landing on the spot.

[0018] A second aspect of the present invention provides an intelligent inspection drone system for indoor substations, used to implement the intelligent inspection method for indoor substations described in the first aspect of the present invention, comprising: Unmanned aerial vehicles (UAVs), intelligent status management and planning subsystem, intelligent equipment fault detection subsystem, alarm subsystem, and ground station control terminal; The drone is used to perform inspection flight missions and collect visible light image data and infrared thermal imaging data of the equipment to be inspected. The intelligent state management planning subsystem is used to generate inspection paths based on preset path optimization constraints and control the UAV to fly along the paths, as well as to dynamically switch the UAV's flight state based on the intelligent state management mechanism. The intelligent equipment fault detection subsystem is used to fuse visible light image data and infrared thermal imaging data, and to identify abnormal conditions of the equipment to be inspected based on the fused multimodal data. The alarm subsystem is used to generate and send alarm information when an equipment malfunction is detected; The ground station control terminal is used for patrol task configuration, patrol process monitoring, data management, alarm information reception and display, and to achieve data interaction with the drone through a wireless communication network.

[0019] Compared with the prior art, the beneficial effects of the present invention include at least the following: 1. This invention introduces an intelligent state management planning subsystem to uniformly manage the state of UAVs during the execution of inspection tasks, and coordinates path planning and intelligent state switching to achieve linkage control between inspection execution, path adjustment and abnormal state handling. This solves the problem of the lack of a full-process state management mechanism in the existing technology, effectively ensures the safe and stable flight of UAVs in indoor substation environments, and reduces the risk of safety accidents caused by UAV failures.

[0020] 2. This invention improves the EGO intelligent planning algorithm by adding equipment safety distance constraints for indoor substation scenarios, enabling path planning to automatically maintain safe distances based on different equipment types. This solves the problem that existing path planning methods do not consider equipment safety constraints and improves the flight safety of UAVs in densely populated equipment environments.

[0021] 3. This invention improves the YOLO-based deep learning model by adding a high-resolution detection layer and fusing visible light and infrared multimodal features, which solves the problem of low defect recognition rate of existing detection methods for small-sized equipment and improves the accuracy of automatic identification of substation equipment anomalies.

[0022] 4. This invention replaces manual inspection with automatic inspection by drones, solving the problems of low efficiency, high intensity and safety risks of traditional manual inspection. It realizes intelligent monitoring of the operating status of indoor substation equipment at all times, improving the safety and efficiency of substation operation and maintenance. Attached Figure Description

[0023] Figure 1 Flowchart of intelligent inspection method for indoor substations; Figure 2 Block diagram of an intelligent inspection drone system for indoor substations; Figure 3 This is a block diagram of the intelligent status management module. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0025] This invention provides an intelligent inspection method for indoor substations in Embodiment 1, applicable to typical indoor substation environments such as flexible DC converter stations, distribution rooms, and enclosed power equipment rooms. These environments typically feature dense equipment, limited space, complex navigation environments, and unavailability of global satellite positioning signals. Figure 1As shown, the method includes the following steps: Step 1: After the UAV completes initialization and takes off to the preset inspection altitude, an inspection path is generated based on the preset path optimization constraints, starting from the current position and covering the equipment to be inspected.

[0026] Preferably, the step of generating the inspection path based on the preset path optimization constraints includes: constructing a path optimization objective function and generating the inspection path using the path optimization objective function; the path optimization objective function is composed of a weighted sum of path length cost, path turning angle change cost, obstacle avoidance cost, and equipment safety distance constraint cost, and the weight coefficient of each cost term is a preset value and satisfies the normalization constraint.

[0027] More preferably, the obstacle avoidance cost is calculated based on the distance between the path point and the obstacle and the preset expansion range of the obstacle, and is used to constrain the UAV to maintain a safe distance from the obstacle.

[0028] More preferably, the equipment safety distance constraint cost is calculated based on the distance between the waypoint and different types of inspection equipment, as well as the minimum safety distance and risk weight coefficient preset for different types of equipment; wherein, different types of equipment correspond to different minimum safety distances and risk weight coefficients, so that the UAV automatically maintains a safety distance that matches the risk level of the equipment when flying close to different types of equipment.

[0029] Specifically, this step employs an improved EGO intelligent planning algorithm with added constraints from the indoor substation environment for path generation. First, the indoor substation environment map information and the location distribution information of the inspection equipment are loaded. After the drone reaches the preset inspection altitude and enters a hovering state, the path planning process is triggered. The expression for the path optimization objective function is:

[0030] in, The path length cost is determined by the three-dimensional coordinates of the i-th path point in space. ,in Represents the coordinates of the i-th path point; Indicates the cost of changes in path turning angles; Indicates the cost of obstacle avoidance. ,in This represents the distance from path point i to the j-th obstacle. This indicates the extent of the obstacle's expansion, and can be set to 0.5m. The cost of the equipment safety distance constraint is represented as follows:

[0031] in, This represents the distance from path point i to the k-th inspection device, obtained by binocular ranging from a vision camera. This represents the minimum safe distance for the k-th device. Taking high-voltage switchgear, transformers, and ordinary structural components as examples, this value is set to 0.7m, 0.5m, and 0.3m respectively. The risk weights for equipment are determined based on the type of equipment. For example, high-voltage switchgear has the highest risk, followed by transformers, and then ordinary structural components have the lowest risk. It is a small constant used to prevent the denominator from being zero.

[0032] By introducing a safe distance constraint for the device, the UAV automatically maintains a safe distance when flying close to the detection device along the path result of the path planning algorithm, thus avoiding collisions and electromagnetic interference. , , , The weights are set with initial values ​​based on experience, and then normalized. The sum of the four weights is 1.

[0033] By unifying the modeling of path length cost, corner smoothing cost, obstacle avoidance cost, and equipment safety distance constraints, and combining it with an adaptive adjustment mechanism for weight coefficients, path planning can significantly improve the flight safety and mission feasibility of UAVs in the complex environment of indoor substations while ensuring path efficiency. Step 2: Control the UAV to fly along the inspection path and collect visible light image data and infrared thermal imaging data of the equipment to be inspected.

[0034] Preferably, during flight, visible light image data and infrared thermal imaging data are acquired simultaneously, and the two types of data are aligned using timestamp synchronization and spatial calibration methods to form unified multimodal detection input data.

[0035] Specifically, the drone is equipped with a high-definition visible light camera and an infrared thermal imaging sensor. While inspecting various devices along a planned path, the two sensors are triggered synchronously at a preset acquisition frequency. The visible light images are used for device appearance integrity detection, while the infrared thermal imaging data is used for temperature distribution anomaly detection. The acquired raw data undergoes timestamp alignment in the onboard computing module, and the infrared temperature data is mapped to the visible light image coordinate system using pre-calibrated camera intrinsic and extrinsic parameters, completing spatial alignment and providing registered input data pairs for subsequent multimodal fusion detection.

[0036] Step 3: The collected visible light image data and infrared thermal imaging data are fused together, and the abnormal status of the equipment to be inspected is identified based on the fused multimodal data. When an abnormality is detected, corresponding alarm information is generated and sent.

[0037] Preferably, the abnormal state identification of the equipment to be inspected based on the fused multimodal data includes: The visible light image data is input into a deep learning backbone network to extract visual feature maps; The infrared thermal imaging data is input into a separate feature extraction branch to extract temperature feature maps; The visual feature map and the temperature feature map are concatenated at the channel level in the feature pyramid network layer to generate a fused feature map; target detection is performed based on the fused feature map to identify device appearance anomalies and temperature anomalies.

[0038] More preferably, the detection head used for target detection adds a high-resolution detection layer for detecting small-sized targets on the basis of the original detection layer.

[0039] More preferably, when an equipment malfunction is detected, an alarm message is automatically generated, which includes the location of the malfunctioning equipment, the type of malfunction, the corresponding inspection image, and the time when the malfunction occurred.

[0040] Specifically, equipment fault detection is achieved using a deep learning model based on an improved YOLO. First, a large amount of substation equipment image data is collected using drones, and equipment faults in the images are labeled, such as damaged areas and missing component locations, thus constructing an equipment fault dataset. The trained equipment fault identification model then performs real-time analysis of equipment images during inspections.

[0041] Specifically, the equipment fault identification model employs a feature layer fusion approach to achieve multimodal information fusion. Visible light images are input into the backbone network to extract visual features; infrared thermal images are processed through an independent lightweight feature extraction branch to extract temperature features; and channel-level stitching and fusion are performed at the Feature Pyramid Network (FPN) layer.

[0042] in, This is a visible light feature map. This is an infrared feature map.

[0043] The fused features are used as input for subsequent target detection heads, thereby enabling joint identification of equipment appearance anomalies and temperature anomalies.

[0044] It should be noted that the YOLO model has been improved to address the characteristics of indoor substation equipment, which include "large differences in target size, numerous small target defects, and uneven category distribution." To enhance the detection capability of small targets, a high-resolution detection branch (P2 layer) has been added to the original YOLO detection head. The original detection layers were P3, P4, and P5; the improved layers are P2, P3, P4, and P5. The P2 layer is used to detect tiny defects smaller than 32×32 pixels (such as cracks and loose parts), thereby improving the detection rate of small-sized faults.

[0045] Step 4: During flight, the operating status and external environment of the UAV are monitored in real time, and the flight status of the UAV is switched according to the preset status management mechanism to respond to different monitored events.

[0046] Preferably, the state management mechanism divides the flight process of the UAV into multiple states, which include at least: initialization state, takeoff state, hovering state, inspection task execution state, path adjustment state, return state, and manual takeover state.

[0047] The initialization state is used to complete the device self-test, sensor initialization, and environmental map loading after power-on. The takeoff status is used to control the drone to take off automatically and reach the preset inspection altitude; The hovering state is used for path planning and task preparation. The inspection task execution status is used to inspect substation equipment according to the planned path; The path adjustment state is used for local path replanning when the path is blocked or the environment changes; The manual takeover status is used to ensure that the drone can be manually controlled at any time to ensure flight safety; The return-to-home status is used to perform autonomous return and landing when the mission is completed or safety conditions are triggered.

[0048] The state management mechanism triggers the UAV to dynamically switch between different states based on received task instructions, environmental perception information, or operating status.

[0049] More preferably, the state management mechanism defines the priority of different states: Emergency landing status has the highest priority. When emergency landing status is triggered, regardless of the current state of the drone, the response of emergency landing status takes precedence over the current state, causing the drone to switch to emergency landing status. The return-to-home status has a higher priority than hovering and normal inspection status, but a lower priority than emergency landing status. When the return-to-home status is triggered and the drone is currently in hovering or normal inspection status, the response of the return-to-home status takes precedence over the current status, causing the drone to switch to the return-to-home status. The hovering state has a higher priority than the normal inspection state. When the hovering state is triggered and the drone is currently in the normal inspection state, the hovering state response takes precedence over the normal inspection state, causing the drone to switch to the hovering state. Normal inspection status has the lowest priority. When an emergency landing status, return-to-base status, or hovering status is triggered, the response to the triggered status takes precedence over the normal inspection status.

[0050] Specifically, this step utilizes a state-event driven control mechanism to achieve dynamic switching between states. The definitions and switching rules for each state are shown in Table 1. Table 1. Definitions and switching rules for each state

[0051] All status events are categorized into four levels, from highest to lowest: highest level, high level, medium level, and low level. The preemption rules for each level are clearly defined to ensure the rationality of system decision-making priorities. Through this arbitration mechanism, the system can quickly respond to various anomalies in complex scenarios, rationally allocate system resources, ensure the continuity and security of inspection tasks, and achieve efficient collaborative operation across multiple modules and states.

[0052] Preferably, the flight state of the UAV is dynamically switched according to a preset state management mechanism, including: When the UAV detects that its flight path is blocked by an obstacle while performing an inspection mission, it switches to path adjustment mode. By updating the obstacle avoidance cost and equipment safety distance constraint cost in the path optimization objective function, the current inspection path is replanned to achieve dynamic adjustment of the inspection path.

[0053] Preferably, the flight state of the UAV is dynamically switched according to a preset state management mechanism, including: The system monitors the drone's power system and flight attitude data in real time. When the drone's angular velocity exceeds the preset safe angular velocity threshold, or the attitude angle deviates from the command value by more than the preset safe envelope threshold, it determines that an abnormal state of flight attitude instability has occurred, triggers the highest priority emergency landing mode, controls the drone to perform a controlled landing operation, and simultaneously sends fault information containing fault codes, abnormality types, and current position coordinates to the ground station.

[0054] Specifically, when the UAV detects the following abnormal states through the accelerometer and gyroscope, it immediately triggers the controlled emergency landing mode and simultaneously triggers the onboard alarm device to send the fault code, abnormality type and the current position coordinates of the UAV to the ground station, so as to facilitate the ground station operators to quickly troubleshoot and handle the fault. The abnormal state is defined as follows: the gyroscope detects an angular velocity exceeding the safe range, i.e. , A preset safe angular velocity threshold is set; if the attitude angle deviates from the command value by more than the safe envelope, i.e. or , The pitch angle deviates from the commanded value. This is the roll angle deviation from the command value.

[0055] Preferably, the flight state of the UAV is dynamically switched according to a preset state management mechanism, including: Real-time monitoring of the drone's communication quality, positioning system status, and temperature measurement equipment operating status; When at least one sensing and communication anomaly is detected, the drone is controlled to switch to pure inertial navigation mode and enter a fixed-point hovering state, while a self-check process is initiated. Abnormal situations in perception and communication include: data packet loss rate exceeding the first preset threshold, communication delay exceeding the second preset threshold, trace of the multi-source fusion covariance matrix of the positioning system exceeding the third preset threshold, or abnormal jumps in the data of the temperature measuring device that prevent normal collection of temperature data of the device to be inspected. The self-test process includes: attempting to restart the sensor driver of the temperature measuring device to restore its normal operation, and reinitializing the real-time positioning and mapping algorithm to ensure positioning and mapping accuracy; If all abnormal indicators return to normal within the preset stable time, the self-test is deemed passed, the drone is controlled to exit the hovering state and continue to perform the inspection task; otherwise, the self-test is deemed to have failed, the emergency landing mode is triggered, the coordinates of the faulty equipment are marked simultaneously, and a manual inspection command is uploaded to the ground station.

[0056] Specifically, perception and communication anomalies include: communication anomalies, i.e., data packet loss rate > 10% or latency > 0.5s; and localization anomalies, i.e., the trace of the covariance matrix of the visual / LiDAR multi-source localization results. The temperature measuring equipment is malfunctioning, meaning that the infrared temperature measuring equipment is experiencing abnormal fluctuations in data and is unable to collect temperature data from the equipment to be inspected.

[0057] The self-test process is as follows: Restart the driver program for the abnormal sensor to restore normal sensor operation; reinitialize the SLAM algorithm to ensure positioning and mapping accuracy. If all abnormal indicators return to normal and remain stable for 2 seconds, the self-test is considered passed. The UAV exits hovering mode and continues to perform the inspection mission; if any abnormal indicator remains abnormal, the self-test is considered failed, the emergency landing mode is immediately triggered, the coordinates of the faulty equipment are simultaneously marked, and a manual troubleshooting instruction is uploaded to the ground station.

[0058] Preferably, the flight state of the UAV is dynamically switched according to a preset state management mechanism, including: An energy decision evaluation function is constructed based on the drone's current remaining battery power, estimated return path distance, and battery health status. The energy decision evaluation function is composed of a weighted sum of the power safety coefficient, the distance feasibility coefficient, and the battery health coefficient, with the weight of each coefficient being a preset value. When the output value of the energy decision evaluation function is higher than the first preset decision threshold, the drone is controlled to continue to perform the current inspection task. When the output value of the energy decision evaluation function is not higher than the first preset decision threshold and not lower than the second preset decision threshold, control the drone to switch to return-to-home mode. When the output value of the energy decision evaluation function is lower than the second preset decision threshold, the drone is controlled to perform an emergency landing on the spot.

[0059] Specifically, the energy decision evaluation function is constructed as follows: , Among them, the weighting coefficient 0.5, 0.2, 0.3, For power safety factor, Battery charge (%) At 30%, it is 1; at 20%, it is 1. 30% linear decay to 0.5, 20% is 0. The distance feasibility coefficient is the estimated remaining range coverage based on the current battery level. The highest value of the coefficient is 1. The return distance is the sum of the distances between discrete points in the 3D path planning.

[0060] The battery health coefficient changes from 1 to 0 when the voltage drop rate is >0.1V / s, indicating an abnormal battery health condition; otherwise, the coefficient remains at 1. This represents the voltage sag rate.

[0061] The decision rule is: when At 0.7, continue executing the task; at 0.4... At 0.7, switch to return-to-home mode. At 0.4, immediately make an emergency landing on the spot.

[0062] This method is based on a UAV platform. It uses an intelligent state management planning subsystem to uniformly schedule and control the state of the UAV inspection process. The improved EGO path planning algorithm enables autonomous path generation and dynamic optimization. Furthermore, it improves the automatic identification and alarm of equipment anomalies based on the YOLO deep learning model.

[0063] In Embodiment 1, this invention provides an intelligent inspection drone system for indoor substations, used to implement the intelligent inspection method for indoor substations described in Embodiment 1, such as... Figure 2 As shown, it includes: Unmanned aerial vehicles (UAVs), intelligent status management and planning subsystem, intelligent equipment fault detection subsystem, alarm subsystem, and ground station control terminal; The drone is used to perform inspection flight missions and collect visible light image data and infrared thermal imaging data of the equipment to be inspected. The intelligent state management planning subsystem is used to generate inspection paths based on preset path optimization constraints and control the UAV to fly along the paths, as well as to dynamically switch the UAV's flight state according to a preset state management mechanism. The intelligent equipment fault detection subsystem is used to fuse visible light image data and infrared thermal imaging data, and to identify abnormal conditions of the equipment to be inspected based on the fused multimodal data. The alarm subsystem is used to generate and send alarm information when an equipment malfunction is detected; The ground station control terminal is used for patrol task configuration, patrol process monitoring, data management, alarm information reception and display, and to achieve data interaction with the drone through a wireless communication network.

[0064] Preferably, the UAV employs a multi-rotor UAV platform for performing inspection flight missions and collecting visible light image data and infrared thermal imaging data of the equipment to be inspected. The UAV includes a flight control module, an onboard computing module, a communication module, a power supply module, and an equipment status perception module.

[0065] More preferably, the flight control module is used to realize the attitude control and flight stability control of the UAV. This module internally includes an inertial measurement unit (IMU), an attitude calculation module, and a flight control algorithm module, wherein: The inertial measurement unit (IMU) is used to measure the angular velocity and acceleration information of the UAV in real time. An attitude calculation algorithm then calculates the UAV's current attitude angles, including roll, pitch, and yaw. The flight control algorithm generates control commands based on the attitude error, thereby driving the motors to achieve stable flight.

[0066] In this embodiment, the drone can maintain a stable hover and fly smoothly along a preset path.

[0067] More preferably, the airborne computing module is used to execute algorithmic tasks related to the autonomous flight of the UAV, including path planning algorithms, intelligent state management and control logic, and equipment detection algorithms.

[0068] More preferably, the communication module is used to realize data communication between the UAV and the ground station control terminal. The communication method can be wireless local area network communication or dedicated industrial wireless communication network. The communication module can realize bidirectional data transmission, including UAV flight status information transmission, inspection image data transmission, and ground station control command reception.

[0069] More preferably, the power module is used to provide power support for the various modules of the UAV.

[0070] In this embodiment, the power module uses a high-energy-density lithium battery and is equipped with a power monitoring unit. When the system detects that the battery power is lower than a set threshold, the intelligent state management system will trigger a return-to-home mechanism.

[0071] More preferably, the equipment status sensing module is used to collect substation equipment operating status information. This module includes a high-definition visible light camera and an infrared thermal imaging sensor. The high-definition visible light camera is used to collect equipment appearance image data. Through high-resolution images, abnormalities in the equipment appearance can be detected, such as equipment surface damage, missing equipment parts, or equipment surface contamination. Infrared thermal imaging sensors are used to acquire surface temperature distribution maps of equipment. By analyzing the temperature distribution, it can be determined whether the equipment has localized overheating or abnormal temperature rise. During the inspection process, visible light images and infrared images are acquired in a time-synchronized manner and spatially aligned in the onboard computing module to form input data that can be used for multimodal equipment detection.

[0072] Preferably, the intelligent state management planning subsystem is used to generate an inspection path based on preset path optimization constraints and control the UAV to fly along the path, as well as to dynamically switch the flight state of the UAV according to a preset state management mechanism. The intelligent state management planning subsystem includes a path planning module and an intelligent state management module.

[0073] More preferably, the path planning module is used to automatically generate inspection paths based on indoor substation environment map information and inspection equipment location distribution information, and to optimize the paths in response to the characteristics of limited space, dense equipment and high safety constraints in the indoor substation environment.

[0074] When generating an inspection path, the path planning module constructs a path optimization objective function based on path length cost, path turning angle change cost, obstacle avoidance cost, and equipment safety distance constraint cost related to equipment type, and uses the path optimization objective function to generate the inspection path.

[0075] The path length cost is determined based on the three-dimensional coordinates of each path point in space; the path turning angle change cost is determined based on the changes in heading angle and pitch angle between adjacent path segments, used to constrain the smoothness of the flight trajectory; the obstacle avoidance cost is determined based on the distance between the path point and the obstacle and the preset expansion range of the obstacle, used to constrain the UAV to maintain a safe distance from the obstacle; the equipment safety distance constraint cost is determined based on the distance between the path point and different types of inspection equipment, as well as the preset minimum safety distance and risk weight coefficient for different types of equipment.

[0076] Different types of equipment correspond to different minimum safety distances and risk weight coefficients. In this embodiment, high-voltage switchgear has the highest risk weight and the largest minimum safety distance, followed by transformers, and ordinary structural components have the lowest. By introducing the equipment safety distance constraint cost, the UAV automatically maintains a safety distance matching the equipment type when flying along the planned path and approaching the detection equipment, avoiding collisions and electromagnetic interference.

[0077] More preferably, the intelligent state management module is used to uniformly divide and dynamically schedule the entire process of the UAV inspection mission in the indoor substation, and realize the dynamic switching between states based on the state event-driven mechanism. The state switching is jointly triggered by the task command, environmental perception information and the operating status.

[0078] The intelligent state management module divides the flight process of the UAV into multiple states, including at least: initialization state, takeoff state, hovering state, inspection task execution state, path adjustment state, return state, and manual takeover state.

[0079] like Figure 3 As shown, the intelligent state management module includes a normal inspection unit, a power and flight control anomaly emergency handling unit, a perception and communication fault-tolerant unit, and an energy management intelligent decision-making unit. Each unit participates in collaborative decision-making based on the current state and triggering events, and outputs unified control commands.

[0080] The normal inspection unit is used to call the path planning module to generate an inspection path and control the UAV to perform the inspection task according to the path when the inspection task is being executed; when the path is being adjusted, the path planning module is called again to perform local path optimization; when the inspection task is completed, the UAV is controlled to enter the return-to-home state.

[0081] The power and flight control anomaly emergency response unit is used to monitor the UAV's power system and flight attitude in real time, quickly respond to various anomalies, and execute emergency response operations to ensure UAV flight safety. When the UAV detects through the accelerometer and gyroscope that the angular velocity exceeds the preset safe angular velocity threshold, or the attitude angle deviates from the command value exceeding the preset safe envelope threshold, the power and flight control anomaly emergency response unit immediately triggers the controlled emergency landing mode and simultaneously triggers the onboard alarm device to send the fault code, anomaly type, and the UAV's current position coordinates to the ground station.

[0082] The perception and communication fault-tolerant unit is used to establish a hierarchical fault-tolerant mechanism for system anomalies that are prone to occur in indoor substations, so as to ensure the continuity of UAV operations in complex electromagnetic environments. When a perception and communication anomaly occurs, the perception and communication fault-tolerant unit controls the UAV to switch to pure inertial navigation mode and enter a fixed-point hovering state, while initiating a self-test process.

[0083] The abnormal sensing and communication conditions include: data packet loss rate exceeding a first preset threshold, communication delay exceeding a second preset threshold, trace of multi-source fusion positioning covariance matrix exceeding a third preset threshold, or abnormal jumps in infrared temperature measurement equipment data that prevent normal acquisition of equipment temperature data.

[0084] The self-test process includes: restarting the faulty sensor driver to restore normal sensor operation, and reinitializing the real-time localization and mapping algorithm to ensure localization and mapping accuracy.

[0085] If all abnormal indicators return to normal within the preset stable time, the self-test is deemed passed, and the drone exits the hovering state to continue performing the inspection task; if any abnormal indicator remains abnormal, the self-test is deemed to have failed, the emergency landing mode is immediately triggered, the coordinates of the faulty equipment are marked, and a manual inspection command is uploaded to the ground station.

[0086] The intelligent energy management decision-making unit is used to construct a decision model, assess the feasibility of the drone's return to base in real time, and dynamically determine the flight strategy based on the assessment results to avoid the drone losing control due to insufficient energy. The intelligent energy management decision-making unit constructs an energy decision assessment function based on the drone's current remaining battery power, the estimated return path distance, and the battery health status, and determines the flight strategy based on the comparison result of the output value of the energy decision assessment function with a preset threshold.

[0087] When the output value is higher than the first preset decision threshold, control the drone to continue executing the current task; When the output value is between the first preset decision threshold and the second preset decision threshold, control the drone to switch to return-to-home mode. When the output value is lower than the second preset decision threshold, the drone is controlled to immediately perform an emergency landing on the spot.

[0088] Furthermore, the intelligent state management module defines different priorities for each state. Wherein: Emergency landing status has the highest priority. When emergency landing status is triggered, regardless of the current state of the drone, the response of emergency landing status takes precedence over the current state, causing the drone to switch to emergency landing status. The return-to-home status has a higher priority than the hovering status and the normal inspection status, but a lower priority than the emergency landing status. When the return-to-home status is triggered and the UAV is currently in the hovering status or the normal inspection status, the response of the return-to-home status takes precedence over the current status, causing the UAV to switch to the return-to-home status. The hovering state has a higher priority than the normal inspection state. When the hovering state is triggered and the drone is currently in the normal inspection state, the response of the hovering state takes precedence over the normal inspection state, causing the drone to switch to the hovering state. Normal inspection status has the lowest priority. When an emergency landing status, return-to-base status, or hovering status is triggered, the response to the triggered status takes precedence over the normal inspection status.

[0089] This priority mechanism enables rapid response to various anomalies in complex scenarios, reasonable allocation of system resources, ensuring the continuity and security of inspection tasks, and achieving efficient collaborative operation of multiple units and multiple states.

[0090] Preferably, the intelligent equipment fault detection subsystem is used to fuse visible light image data and infrared thermal imaging data collected by the UAV, and to identify abnormal states of the equipment to be inspected based on the fused multimodal data.

[0091] The intelligent equipment fault detection subsystem uses a deep learning backbone network to extract visual feature maps from visible light images and an independent lightweight feature extraction branch to extract temperature feature maps from infrared thermal images. The visual feature maps and temperature feature maps are then concatenated at the channel level in the feature pyramid network layer to generate a fused feature map. Based on the fused feature map, a target detection head is used to jointly identify equipment appearance anomalies and temperature anomalies.

[0092] Furthermore, the target detection head of the intelligent equipment fault detection subsystem adds a high-resolution detection layer for detecting small targets, based on the original detection layer. The original detection layer is used to detect medium and large targets, while the high-resolution detection layer is used to detect minute defects smaller than a preset pixel threshold, such as surface cracks and loose components. This improvement effectively increases the detection rate of small-sized faults in indoor substation equipment.

[0093] The intelligent equipment fault detection subsystem is trained using a pre-built equipment status image dataset. This dataset is obtained by collecting a large amount of substation equipment image data and labeling equipment faults in the images. The labeling includes information such as damaged areas, missing component locations, surface contamination areas, and areas of abnormal temperature. After training, the intelligent equipment fault detection subsystem is deployed in the onboard computing module of the UAV, performing real-time analysis of equipment images during inspections.

[0094] Preferably, the alarm subsystem is used to generate and send alarm information when the intelligent equipment fault detection subsystem detects an equipment abnormality.

[0095] When the intelligent equipment fault detection subsystem detects an equipment anomaly, the alarm subsystem automatically generates structured alarm information. The alarm information includes: the location information of the abnormal equipment, the anomaly type, the corresponding inspection image information, and the timestamp of the anomaly occurrence.

[0096] The alarm subsystem transmits the alarm information to the ground station control terminal via the UAV's communication module. Upon receiving the alarm information, the ground station control terminal displays the abnormal device location and related images on the monitoring interface and alerts maintenance personnel to handle the situation promptly.

[0097] Preferably, the ground station control terminal is used for patrol task configuration, patrol process monitoring, data management, and alarm information reception and display. The ground station control terminal interacts with the UAV through a wireless communication network. Maintenance personnel can use the ground station control terminal to issue patrol task commands to the UAV, monitor the UAV's flight status and patrol footage in real time, receive and process alarm information, and take over control of the UAV when necessary, switching the UAV to manual control mode.

[0098] Through the coordinated operation of the above components, the indoor substation intelligent inspection drone system provided in this embodiment can realize autonomous inspection path planning, intelligent flight status management, multimodal equipment anomaly detection and hierarchical safety response in the indoor substation environment. It effectively replaces manual inspection methods, improves inspection efficiency and accuracy, and provides strong support for the stable operation of indoor substation equipment.

[0099] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for intelligent inspection of indoor substations, characterized in that, include: After the drone completes initialization and takes off to the preset inspection altitude, an inspection path is generated based on preset path optimization constraints, starting from the current position and covering the equipment to be inspected. The drone is controlled to fly along the inspection path and collect visible light image data and infrared thermal imaging data of the equipment to be inspected. The collected visible light image data and infrared thermal imaging data are fused together, and the abnormal status of the equipment to be inspected is identified based on the fused multimodal data. When an abnormality is detected, corresponding alarm information is generated and sent. During flight, the operating status and external environment of the UAV are monitored in real time, and the flight status of the UAV is switched according to a preset status management mechanism to respond to different monitored events.

2. The intelligent inspection method for indoor substations according to claim 1, characterized in that: Based on preset path optimization constraints, an inspection path is generated starting from the current position and covering the equipment to be inspected, including: A path optimization objective function is constructed, which is composed of a weighted sum of path length cost, path turning angle change cost, obstacle avoidance cost, and equipment safety distance constraint cost. The weight coefficient of each cost term is a preset value and satisfies the normalization constraint. The equipment safety distance constraint cost is calculated based on the distance between the path point and different types of inspection equipment, the preset minimum safety distance for different types of equipment, and the preset weight representing the risk level of the equipment. The inspection path is generated using the path optimization objective function.

3. The intelligent inspection method for indoor substations according to claim 1, characterized in that: The abnormal state identification of the equipment to be inspected based on the fused multimodal data includes: inputting the visible light image data into a deep learning backbone network to extract visual feature maps; The infrared thermal imaging data is input into a separate feature extraction branch to extract temperature feature maps; In the feature pyramid network layer, the visual feature map and the temperature feature map are concatenated at the channel level to generate a fused feature map; Target detection is performed based on the fused feature map to identify abnormalities in the appearance and temperature of the device.

4. The intelligent inspection method for indoor substations according to claim 1, characterized in that: The state management mechanism divides the UAV's flight process into multiple states, including: initialization state, takeoff state, hovering state, inspection mission execution state, path adjustment state, return state, and manual takeover state. The initialization state is used to complete device self-testing, sensor initialization, and environmental map loading after the system is powered on. The takeoff status is used to control the drone to take off automatically and reach the preset inspection altitude; The hovering state is used for path planning and task preparation. The inspection task execution status is used to inspect substation equipment according to the planned route; The path adjustment status is used for local path replanning when the path is blocked or the environment changes; The manual takeover status is used to ensure that the drone can be manually controlled at any time; The return-to-home status is used to perform autonomous return and landing when the mission is completed or safety conditions are triggered.

5. The intelligent inspection method for indoor substations according to claim 4, characterized in that: The state management mechanism defines the priorities of different states: Emergency landing status has the highest priority. When emergency landing status is triggered, regardless of the current state of the drone, the response of emergency landing status takes precedence over the current state, and the drone switches to emergency landing status. The return-to-home status has a higher priority than hovering and normal inspection status, but a lower priority than emergency landing status. When the return-to-home status is triggered and the drone is currently in hovering or normal inspection status, the response of the return-to-home status takes precedence over the current status, causing the drone to switch to the return-to-home status. The hovering state has a higher priority than the normal inspection state. When the hovering state is triggered and the drone is currently in the normal inspection state, the hovering state response takes precedence over the normal inspection state, causing the drone to switch to the hovering state. Normal inspection status has the lowest priority. When an emergency landing status, return-to-base status, or hovering status is triggered, the response to the triggered status takes precedence over the normal inspection status.

6. A method for intelligent inspection of indoor substations according to claim 2 or 5, characterized in that: The drone's flight status is dynamically switched according to a preset state management mechanism, including: When the UAV detects that its flight path is blocked by an obstacle while performing an inspection mission, it switches to path adjustment mode. By updating the obstacle avoidance cost and equipment safety distance constraint cost in the path optimization objective function, the current inspection path is replanned to achieve dynamic adjustment of the inspection path.

7. The intelligent inspection method for indoor substations according to claim 5, characterized in that: The drone's flight status is dynamically switched according to a preset state management mechanism, including: The system monitors the drone's power system and flight attitude data in real time. When the drone's angular velocity exceeds the preset safe angular velocity threshold, or the attitude angle deviates from the command value by more than the preset safe envelope threshold, it determines that an abnormal state of flight attitude instability has occurred, triggers the highest priority emergency landing mode, controls the drone to perform a controlled landing operation, and simultaneously sends fault information containing fault codes, abnormality types, and current position coordinates to the ground station.

8. The intelligent inspection method for indoor substations according to claim 5, characterized in that: The drone's flight status is dynamically switched according to a preset state management mechanism, including: Real-time monitoring of the drone's communication quality, positioning system status, and temperature measurement equipment operating status; When at least one sensing and communication anomaly is detected, the drone is controlled to switch to pure inertial navigation mode and enter a fixed-point hovering state, while a self-check process is initiated. Abnormal situations in perception and communication include: data packet loss rate exceeding the first preset threshold, communication delay exceeding the second preset threshold, trace of the multi-source fusion covariance matrix of the positioning system exceeding the third preset threshold, or abnormal jumps in the data of the temperature measuring device that prevent normal collection of temperature data of the device to be inspected. The self-test process includes: attempting to restart the sensor driver of the temperature measuring device to restore its normal operation, and reinitializing the real-time localization and mapping algorithm; If all abnormal indicators return to normal within the preset stable time, the self-test is deemed successful, the drone is controlled to exit the hovering state and continue to perform the inspection task; otherwise, the self-test is deemed unsuccessful, the emergency landing mode is triggered, the coordinates of the faulty equipment are marked simultaneously, and a manual inspection command is uploaded to the ground station.

9. The intelligent inspection method for indoor substations according to claim 5, characterized in that: The drone's flight status is dynamically switched according to a preset state management mechanism, including: An energy decision evaluation function is constructed based on the drone's current remaining battery power, estimated return path distance, and battery health status. The energy decision evaluation function is composed of a weighted sum of the power safety coefficient, the distance feasibility coefficient, and the battery health coefficient, with the weight of each coefficient being a preset value. When the output value of the energy decision evaluation function is higher than the first preset decision threshold, the drone is controlled to continue to perform the current inspection task. When the output value of the energy decision evaluation function is not higher than the first preset decision threshold and not lower than the second preset decision threshold, control the drone to switch to return-to-home mode. When the output value of the energy decision evaluation function is lower than the second preset decision threshold, the drone is controlled to perform an emergency landing on the spot.

10. An intelligent inspection drone system for indoor substations, used to implement the intelligent inspection method for indoor substations as described in any one of claims 1-9, characterized in that, include: Unmanned aerial vehicles (UAVs), intelligent status management and planning subsystem, intelligent equipment fault detection subsystem, alarm subsystem, and ground station control terminal; The drone is used to perform inspection flight missions and collect visible light image data and infrared thermal imaging data of the equipment to be inspected. The intelligent state management planning subsystem is used to generate inspection paths based on preset path optimization constraints and control the UAV to fly along the paths, as well as to dynamically switch the UAV's flight state based on the intelligent state management mechanism. The intelligent equipment fault detection subsystem is used to fuse visible light image data and infrared thermal imaging data, and to identify abnormal conditions of the equipment to be inspected based on the fused multimodal data. The alarm subsystem is used to generate and send alarm information when an equipment malfunction is detected; The ground station control terminal is used for patrol task configuration, patrol process monitoring, data management, alarm information reception and display, and to achieve data interaction with the drone through a wireless communication network.

Images