An unmanned aerial vehicle-mounted insulator intelligent detection system and detection method

Abstract

The application discloses an unmanned aerial vehicle-mounted insulator intelligent detection system and a detection method, the detection system comprising a detection device, and further comprising a visual perception unit, a multi-modal detection unit, an edge computing unit and an environment perception unit arranged on the detection device; through the cooperative control of the visual perception unit and the edge computing unit, the detection device realizes the whole-process autonomous operation from hanging, moving, positioning, detection to recycling. Compared with the prior art, the number of manual interventions required for single insulator string detection is reduced to zero, the operation efficiency is improved by more than 60%, and the human operation delay and the risk of misoperation are significantly reduced. The application further constructs a multi-modal fusion diagnosis architecture of "non-contact pre-screening-contact precision measurement-environment compensation correction", which can significantly reduce the misjudgment caused by poor contact or environmental interference.

Description

Technical Field

[0001] This invention relates to the field of power equipment maintenance technology, specifically to an intelligent insulator detection system and method based on a drone. Background Technology

[0002] Insulators, as critical insulating components in power systems such as transmission lines and substations, directly determine the level of safe and stable operation of the power grid. During long-term operation, insulators are subject to corrosion from natural environments such as high temperature and humidity, dust pollution, ultraviolet radiation, and ice-melt cycles, leading to degradation phenomena such as insulation performance decline, skirt aging, porcelain cracking, and internal defects. The occurrence of zero-value and low-value insulators is a core cause of line flashovers, short-circuit trips, and large-scale power outages. Therefore, conducting regular, accurate, and efficient zero-value and degradation testing of insulators is a necessary technical step in ensuring power supply reliability in the power operation and maintenance field.

[0003] Traditional insulator inspection relies primarily on manual tower climbing and high-altitude operations, which suffers from significant drawbacks such as high labor intensity, low work efficiency, high risks of falls and electric shocks, and poor adaptability to complex terrain. With the integration of drone technology and power inspection equipment, drone-mounted insulator inspection devices are gradually replacing manual labor. By deploying the inspection device onto insulator strings via drones, unmanned high-altitude inspection is achieved, significantly reducing operational risks and improving inspection flexibility and coverage.

[0004] Existing UAV-mounted insulator testing devices mostly employ a mechanical straddle structure (our company's previously developed testing device, patent number 2025118190460), using a flexible belt drive to move along the insulator string and working with contact-type testing probes to perform zero-value insulation resistance testing. While this type of device achieves basic testing functions, it still has significant technical shortcomings in practical engineering applications, making it difficult to meet the demands for intelligent, high-precision, and high-efficiency testing. Firstly, the entire process of the device moving along the insulator string and stopping at the target insulator requires remote manual control of its movement and start / stop by the operator. It cannot autonomously identify the position, spacing, and numbering information of the insulator discs. The detection path and stopping point rely on manual experience for judgment, which not only results in low detection efficiency and long operation time, but is also prone to positioning deviations due to control delays, visual errors, or high-altitude airflow interference. This can cause misalignment and overlap between the detection needle and the insulator electrode, leading to distorted detection data, missed detections, or repeated detections, which seriously affects the reliability of the detection.

[0005] Secondly, existing devices only use contact-type detection probes to obtain the single parameter of insulation resistance. They cannot make non-contact predictions of the insulator's surface condition, such as dirt, discharge marks, and cracks, before contact testing, nor can they conduct non-destructive screening for hidden defects such as looseness and cracks inside the insulator. Single contact-type testing is insufficient to comprehensively characterize the insulator's degradation status throughout its entire life cycle, easily missing early and latent defects, resulting in delayed fault warnings and failing to provide complete and multi-dimensional status information for operation and maintenance decisions.

[0006] In summary, existing UAV-mounted insulator inspection devices suffer from problems such as lack of autonomous positioning capabilities, limited detection modes, and insufficient intelligence, which restrict the automation level, detection accuracy, and comprehensiveness of defect identification in insulator inspection. They also fail to meet the technical requirements of smart grids for power transmission equipment status awareness and precise operation and maintenance. Therefore, developing intelligent UAV-mounted insulator inspection technology with autonomous positioning and navigation and multi-modal fusion detection has become an urgent technical challenge to be solved in the power operation and maintenance field. Summary of the Invention

[0007] The purpose of this invention is to address the problems of lack of autonomous positioning capability, single detection mode, and insufficient intelligence level in existing UAV-mounted insulator detection devices, and to provide an intelligent insulator detection system and method based on UAV mounting.

[0008] To address the shortcomings of the aforementioned technical problems, the present invention adopts the following technical solution: an intelligent insulator detection system based on a drone, comprising a detection device, and further comprising a visual perception unit, a multimodal detection unit, an edge computing unit, and an environmental perception unit disposed on the detection device; the visual perception unit is used to acquire real-time images of the insulator string; the multimodal detection unit includes an ultraviolet imaging sensor and an infrared thermal imaging sensor, as well as a contact force sensor and a contact resistance monitoring module; the edge computing unit includes an embedded AI chip, a memory, a wireless communication module, and a positioning module; the environmental perception unit includes a temperature and humidity sensor, an air pressure sensor, and an electric field strength sensor.

[0009] As a further optimization of the intelligent insulator detection system based on UAV mounting described in this invention, the visual perception unit is an industrial camera, which is installed at both ends of the mounting bracket along the extension direction of the insulator string, and there are at least two of them.

[0010] As a further optimization of the intelligent insulator detection system based on UAV as described in this invention, the visual perception unit is equipped with a supplementary light that automatically provides supplementary lighting in low-light environments.

[0011] As a further optimization of the intelligent insulator detection system based on UAV as described in this invention, the edge computing unit runs a lightweight deep learning model and a SLAM localization algorithm.

[0012] As a further optimization of the intelligent insulator detection system based on UAV as described in this invention, the ultraviolet imaging sensor and infrared thermal imaging sensor in the multimodal detection unit are integrated on the outer frame of the detection device, and the contact force sensor and contact resistance monitoring module are set on the detection needle; the edge computing unit is set inside the center plate or protective cover; and the environmental sensing unit is installed on the outside of the center plate.

[0013] A detection method for an insulator intelligent detection system mounted on a drone includes the following steps: S1: Autonomous Attachment and Initial Positioning The drone hoisted the detection system above the insulator string. The visual perception unit collected images in real time, the edge computing unit identified the end features of the insulator string, and automatically controlled the drone to fine-tune its position so that the inverted U-shaped structure of the detection device's mounting bracket was aligned with the insulator string, completing the autonomous unhooking and placement. S2: Multimodal Pre-detection and Path Planning The multimodal detection unit first scans the insulators before and after the current station. The edge computing unit combines the visual image to identify the spatial position of each insulator, establishes an initial "position-state" mapping table for the insulator string, and plans the movement path. Then, the edge computing unit stores the planned path in the memory. S3: Closed-loop movement and autonomous positioning The edge computing unit triggers the drive unit of the detection device, which drives the detection system to move along the insulator string. The vision perception unit collects images of the insulator in front in real time, estimates the relative displacement of the device, and combines the feedback from the encoder of the drive motor to achieve closed-loop position control. When the system device is detected to have reached the preset detection position of the target insulator piece, the edge computing unit sends a stop command to the drive unit. S4: Adaptive Contact Detection Upon reaching the target station, the flipping motor drives the detection probe to flip downwards. The contact force sensor monitors the contact pressure in real time. When the pressure reaches a preset threshold, the flipping stops to ensure reliable contact without damaging the insulator surface. The zero-value detector performs insulation resistance measurement, while the contact resistance monitoring module records the contact resistance value between the probe and the insulator for data correction. S5: Multimodal Data Fusion and Real-time Diagnosis The edge computing unit performs spatiotemporal alignment on the collected data of the current insulator, then runs the fusion diagnostic model, outputs diagnostic conclusions, and provides confidence levels; the diagnostic results and related data are uploaded to the cloud or ground station in real time. S6: Automatic Loop and Termination Detection After the inspection is completed, the detection probe is reset, and the drive unit continues to move to the next station. When the vision perception unit recognizes the end of the insulator string, it determines that the inspection is complete, the device automatically stops moving, and sends a retrieval signal to the drone via wireless communication.

[0014] As a further optimization of the detection method of the intelligent insulator detection system based on UAV of the present invention, when the multi-modal detection unit scans the front and rear insulators in step S2, the ultraviolet imaging sensor is activated to collect the corona discharge intensity and discharge pulse frequency around each insulator, and the infrared thermal imaging sensor acquires the surface temperature distribution data of the insulator skirt, steel cap and iron foot area.

[0015] As a further optimization of the detection method of the intelligent insulator detection system based on UAV mounting of the present invention, the pressure threshold of the contact force sensor in step S4 for real-time monitoring of contact pressure is 0.5-2.0N.

[0016] As a further optimization of the detection method of the intelligent insulator detection system based on UAV mounting of the present invention, the contact resistance monitoring module in step S4 can adopt the four-wire method or the AC impedance measurement method.

[0017] Compared with the prior art, the present invention has the following beneficial effects: I. This invention achieves fully autonomous operation of the detection device from attachment, movement, positioning, detection, and retrieval through the collaborative control of a visual perception unit and an edge computing unit. The edge computing unit runs a lightweight deep learning model to automatically identify the end features of the insulator string and the center of the steel cap of each insulator. Combined with visual SLAM and motor encoder feedback, it performs millimeter-level closed-loop positioning, completing path planning, precise docking, and detection actions without human intervention. Compared with existing technologies, this invention reduces the number of manual interventions required for a single insulator string detection to zero, improves work efficiency by more than 60%, and significantly reduces human operation delays and the risk of misoperation.

[0018] II. This invention constructs a multimodal fusion diagnostic architecture of "non-contact pre-screening - contact-based precision measurement - environmental compensation correction". The multimodal detection unit collects ultraviolet discharge signals and infrared thermograms before the detection needle makes contact, the contact force sensor controls the detection needle pressure in a closed loop to ensure stable contact, and the environmental sensing unit performs real-time compensation correction for temperature, humidity and electric field strength. The edge computing unit inputs the above multi-source data into the fusion diagnostic model and outputs the insulator status category and confidence level, which can significantly reduce misjudgments caused by poor contact or environmental interference. Attached Figure Description

[0019] Figure 1 This is a structural block diagram of the present invention; Figure 2 This is a structural block diagram of the multimodal detection unit in this invention; Figure 3 This is a structural block diagram of the edge computing unit in this invention; Figure 4 This is a structural block diagram of the environmental sensing unit in this invention. Detailed Implementation

[0020] To better understand the present invention, the following embodiments further illustrate the content of the present invention, but the content of the present invention is not limited to the following embodiments.

[0021] like Figure 1-4 As shown, an intelligent insulator detection system based on a drone includes a detection device, and further includes a visual perception unit, a multimodal detection unit, an edge computing unit, and an environmental perception unit mounted on the detection device. The visual perception unit is used to acquire real-time images of the insulator string. The multimodal detection unit includes an ultraviolet imaging sensor and an infrared thermal imaging sensor, as well as a contact force sensor and a contact resistance monitoring module. The edge computing unit includes an embedded AI chip, a memory, a wireless communication module, and a positioning module. The environmental perception unit includes a temperature and humidity sensor, a barometric pressure sensor, and an electric field strength sensor.

[0022] For better implementation results, the visual perception unit is an industrial camera, which is installed at both ends of the mounting bracket along the extension direction of the insulator string, and there are at least two of them.

[0023] Furthermore, the visual perception unit is equipped with a supplementary light, such as an LED array, that automatically provides supplementary lighting in low-light environments.

[0024] To make the intelligent detection system of the present invention more intelligent and efficient, the edge computing unit runs a lightweight deep learning model and a SLAM localization algorithm.

[0025] The detection device involved in this invention includes a mounting frame, a drive unit, a detection unit, a stabilizing unit, and an auxiliary unit. The mounting frame has an inverted U-shaped structure and is used to straddle an insulator string. The drive unit is disposed on the mounting frame and is used to drive the mounting frame to move along the insulator string. The detection unit includes a zero-value detector and a detection probe driven by a flip motor. The specific structure of this detection device can be found in the prior invention patent application of this company, patent number 2020118190460. The structural name involved in this invention is directly referenced from the structural name in that patent.

[0026] To achieve better detection results, the ultraviolet imaging sensor and infrared thermal imaging sensor in the multimodal detection unit are integrated on the epitaxial frame, and the contact force sensor and contact resistance monitoring module are mounted on the detection needle; the edge computing unit is located inside the center plate or protective cover, and the AI ​​chip (such as NPU), wireless communication module (such as 4G / 5G / WiFi) and positioning module (such as GPS / BeiDou positioning module) are mounted on the outside of the center plate.

[0027] A detection method for an insulator intelligent detection system based on UAV mounting includes the following steps: S1: Autonomous mounting and initial positioning The drone hoisted the detection system above the insulator string. The visual perception unit collected images in real time, and the edge computing unit identified the end features of the insulator string (such as the equalizing ring and the first insulator steel cap) through the target detection algorithm. It also automatically controlled the drone to fine-tune its position so that the inverted U-shaped structure of the detection device's mounting bracket was aligned with the insulator string, thus completing the autonomous unhooking and placement. S2: Multimodal Pre-detection and Path Planning The multimodal detection unit first scans the insulators before and after the current station. The edge computing unit combines the visual image to identify the spatial position of each insulator, establishes an initial "position-state" mapping table for the insulator string, and plans the movement path (such as detecting each insulator piece by piece from the conductor side to the crossarm side). Then, the edge computing unit stores the planned path in the memory. Specifically: After the drone places the detection device on the insulator string, the drive unit does not start temporarily, and the edge computing unit automatically triggers the multimodal pre-detection mode.

[0028] First, the multimodal detection unit activates the ultraviolet imaging sensor and the infrared thermal imaging sensor. Using the current position of the mounting frame as the reference position, it scans 2 to 3 insulators forward and backward along the extension direction of the insulator string. The ultraviolet imaging sensor collects the corona discharge intensity and discharge pulse frequency around each insulator, while the infrared thermal imaging sensor acquires the surface temperature distribution data of the insulator skirts, steel caps, and iron feet. The edge computing unit stores the collected ultraviolet signals and infrared thermal images in real time and correlates them with the spatial position of the corresponding insulator.

[0029] Simultaneously, the visual perception unit continuously acquires global and local detail images of the insulator string. The edge computing unit runs a lightweight deep learning model to perform target detection on each insulator in the image, outputting the bounding box coordinates, steel cap center point coordinates, and insulator type. Based on the detection results, the edge computing unit further identifies the conductor-side end features (such as equipotential rings) and crossarm-side end features (such as ball-end rings) of the insulator string, thereby determining the extension direction of the insulator string and sequentially numbering the insulator pieces along that direction.

[0030] Subsequently, the edge computing unit spatially registers the aforementioned non-contact detection data (ultraviolet discharge intensity, infrared temperature distribution) with the visual recognition results (insulator number, steel cap center coordinates) to establish an initial "position-state" mapping table indexed by the insulator number. This mapping table includes at least the number of each insulator, the three-dimensional coordinates of the steel cap center, the ultraviolet discharge intensity level, the infrared temperature characteristic value, and the detection timestamp.

[0031] Based on the initial mapping table, the edge computing unit automatically plans the movement path of the inspection device. The default path planning strategy is to inspect insulators piece by piece from the conductor side to the crossarm side, that is, starting from the end closest to the conductor and moving sequentially towards the crossarm. When the conductor-side end feature is identified, the edge computing unit numbers the nearest insulator at the current workstation as the 1st piece and sets the movement direction to gradually advance from the 1st piece to the Nth piece (crossarm side). If the inspection operation needs to start from the crossarm side, the operator can remotely switch the planning direction through the ground station.

[0032] After path planning is completed, the edge computing unit stores the planned path in memory and sends the number of the first target insulator and the center coordinates of its steel cap to the drive unit, triggering the closed-loop movement control in step S3. Throughout the pre-detection and path planning process, the drive unit remains stationary to ensure that the data collected by the multi-modal detection unit is not disturbed by vibrations or displacements caused by device movement.

[0033] S3: Closed-loop movement and autonomous positioning The edge computing unit triggers the drive unit of the detection device, which drives the detection system to move along the insulator string. The vision perception unit collects images of the insulator in front in real time, estimates the relative displacement of the device, and combines the feedback from the encoder of the drive motor to achieve closed-loop position control. When the system device is detected to have reached the preset detection position of the target insulator piece, the edge computing unit sends a stop command to the drive unit. Specifically: The edge computing unit incorporates an embedded AI chip and runs a lightweight deep learning model for target detection and number recognition of insulator string images acquired in real time by the visual perception unit. This lightweight deep learning model employs an INT8-quantized convolutional neural network architecture, including YOLO-Nano or MobileNet-SSD networks, with fewer than 5M parameters and an inference speed of at least 30 frames per second on the embedded platform. The model input is the RGB image acquired by the visual perception unit, and the output includes the bounding box coordinates of the insulator segments, category labels, steel cap center point coordinates, and detection confidence scores.

[0034] During the numbering and identification process, the edge computing unit first identifies the end features at both ends of the insulator string using the model, including the equalizing ring on the conductor side and the ball-end ring on the crossarm side, and determines the extension direction of the insulator string based on these end features. Subsequently, the edge computing unit sequentially detects the steel cap area of ​​each insulator along the extension direction and runs a sequential target tracking algorithm to correlate the detection results of consecutive frames in the video stream, assigning a unique sequential number to each insulator. As the detection device moves along the insulator string, the edge computing unit updates the insulator numbers within its field of view in real time and smooths the correspondence between the numbers and spatial positions using Kalman filtering to prevent number jumps caused by occlusion or rapid movement.

[0035] To achieve precise alignment of the detection pins, the edge computing unit matches the coordinates of the steel cap center point output by the model with the preset detection station. When the steel cap center point is located in the center region of the image or the fixed reference position of the mounting bracket, the edge computing unit determines that the device has reached the detection station of the target insulator and sends a stop command to the drive unit. Simultaneously, it transmits the current insulator number, steel cap center coordinates, and detection station information to the detection unit to guide the precise alignment of the detection pins. The lightweight deep learning model operates using a dual-thread architecture. The inference thread continuously receives image frames and outputs detection results, while the control thread combines the detection results with encoder feedback from the drive unit to achieve vision-encoder fusion positioning, ensuring decoupling between the control cycle and the inference cycle. Furthermore, the edge computing unit uploads high-confidence detection results to the cloud training platform via a wireless communication module for incremental training and remote iterative updates of the model.

[0036] S4: Adaptive Contact Detection Upon reaching the target station, the flipping motor drives the detection probe to flip downwards. The contact force sensor monitors the contact pressure in real time. When the pressure reaches a preset threshold, the flipping stops to ensure reliable contact without damaging the insulator surface. The zero-value detector performs insulation resistance measurement, while the contact resistance monitoring module records the contact resistance value between the probe and the insulator for data correction. Specifically: Once the mounting bracket is precisely positioned at the detection station of the target insulator under the control of the edge computing unit, the edge computing unit sends a flipping command to the flipping motor, driving the detection pin to flip downwards from its initial raised position and move towards the steel cap and iron foot of the target insulator.

[0037] The detection probe integrates a contact force sensor, which communicates with the edge computing unit in real time. As the probe rotates downwards, the edge computing unit continuously reads the feedback signal from the contact force sensor. When the tip of the probe touches the surface of the insulator's steel cap and iron foot, the pressure value detected by the contact force sensor rises from zero. The edge computing unit compares this real-time pressure value with a preset pressure threshold, which is pre-calibrated based on the insulator surface material (such as glass, ceramic, or composite silicone rubber) and the mechanical properties of the detection probe, typically set between 0.5N and 2.0N.

[0038] Once the real-time pressure value reaches or exceeds the preset threshold, the edge computing unit immediately sends a stop command to the flipping motor, maintaining the current flipping angle of the detection probe. This ensures stable and reliable mechanical contact between the probe and the insulator surface, while preventing damage to the insulator glaze or deformation of the detection probe due to excessive pressure. If the pressure value exceeds the safety limit (e.g., 3.0N) during the flipping process, the edge computing unit controls the flipping motor to make reverse fine adjustments, causing the pressure to drop back to within the preset threshold range, achieving closed-loop adaptive adjustment.

[0039] After contact stabilization, the edge computing unit activates the zero-value detector, applying a detection voltage (typically DC 2500V or 5000V, depending on the insulator's rated voltage) to the insulator via the detection probe. It then measures the leakage current flowing through the insulator and calculates the insulation resistance value. Simultaneously, the contact resistance monitoring module records the contact resistance value between the detection probe and the insulator's metal components (steel cap, iron foot). This contact resistance monitoring module can employ either the four-wire method or AC impedance measurement to eliminate the influence of conductor resistance and contact interface resistance on the measurement results.

[0040] The edge computing unit uses the acquired contact resistance value as a correction parameter to compensate for the insulation resistance value measured by the zero-value detector. Specifically, if the contact resistance value exceeds the preset range (e.g., greater than 100Ω), it indicates that there is dirt, oxide layer, or poor contact between the probe and the insulator. The edge computing unit will correct the insulation resistance value according to the pre-calibrated contact resistance-measurement error mapping curve, or mark the measurement as "abnormal contact" and trigger the automatic cleaning program of the detection probe (e.g., multiple minor vibrations or flipping and resetting followed by re-contact). If the contact resistance value is within the normal range, the measurement result is directly adopted.

[0041] After completing a single insulation resistance measurement and data correction, the edge computing unit stores the current target insulator number, the corrected insulation resistance value, the contact resistance value, the leakage current waveform, and the detection timestamp into the local memory, and prepares to execute the subsequent detection pin reset action.

[0042] S5: Multimodal Data Fusion and Real-time Diagnosis The edge computing unit performs spatiotemporal alignment on the following data collected from the current insulator: location information (piece number, GPS coordinates, relative position), non-contact data (ultraviolet discharge intensity, infrared temperature distribution), contact data (insulation resistance value, contact resistance, leakage current waveform), and environmental data (temperature, humidity, electric field strength). Then, it runs a fusion diagnostic model (such as a decision tree, support vector machine, or lightweight neural network), outputs diagnostic conclusions, and provides confidence levels. The diagnostic results and related data are uploaded to the cloud or ground station in real time. S6: Automatic Loop and Termination Detection After the test is completed, the test needle is reset, and the drive unit continues to move to the next station. When the vision perception unit recognizes the end of the insulator string (such as the crossarm fitting), it determines that the test is completed, the device automatically stops moving, and sends a retrieval signal to the drone via wireless communication.

[0043] This invention achieves fully autonomous operation of the detection device from attachment, movement, positioning, detection to retrieval by integrating a visual perception unit, an edge computing unit and a driving unit through closed-loop collaborative control, which significantly reduces the burden of manual operation and the risk of operational errors.

[0044] The visual perception unit acquires images of the insulator string in real time, while the edge computing unit runs a lightweight deep learning model to automatically identify the end features of the insulator string (such as equalizing rings and ball-head hanging rings) and the center position of the steel cap of each insulator, eliminating the need for manual pre-setting of the number or number of insulators. Based on this, the edge computing unit combines visual SLAM positioning with encoder feedback from the drive motor to perform millimeter-level closed-loop control of the mounting frame's position, autonomously planning its movement path and precisely stopping at the inspection station of the target insulator. During inspection, the edge computing unit automatically controls the flipping angle, contact pressure, and reset action of the inspection pins, and automatically triggers a retrieval signal after completing the inspection of the entire string by visually recognizing the end features of the insulator string. The entire operation process requires no remote human intervention, upgrading the traditional operation mode that relies on operator visual judgment and manual control to a machine vision-guided autonomous operation mode. This significantly reduces the risk of inspection failures caused by human control delays, limited viewing angles, or misoperations, improving the consistency and reliability of the inspection operation.

[0045] Furthermore, the present invention employs a dual-thread architecture to decouple visual reasoning from control commands, ensuring that the detection device can continuously and stably perform target detection and position updates during movement, thereby further guaranteeing the smoothness and automation of the entire process.

[0046] Compared with existing technologies, this invention reduces the number of manual interventions required for a single insulator string inspection from more than ten to zero, improving work efficiency by more than 60%, and is particularly suitable for large-scale, repetitive insulator inspection tasks in transmission lines.

[0047] This invention significantly reduces the false positive and false negative rates of insulator condition detection by constructing a multimodal fusion diagnostic architecture of "non-contact pre-screening - contact-based precision measurement - environmental compensation correction" and combining contact force closed-loop control with environmental adaptive correction. Before the detection needle contacts the insulator, the multimodal detection unit acquires the discharge signal intensity of the insulator using an ultraviolet imaging sensor and obtains the temperature distribution characteristics of the insulator surface and steel cap area using an infrared thermal imaging sensor. This non-contact pre-detection can quickly identify insulators with abnormal phenomena such as surface discharge and localized overheating without damaging the insulator surface, providing key screening criteria for subsequent contact-based detection. The edge computing unit spatially aligns the ultraviolet and infrared features with the insulator positions and numbers identified by the visual perception unit to form a multimodal data association matrix.

[0048] In the contact testing stage, a contact force sensor mounted on the probe monitors the contact pressure between the probe and the insulator's steel cap and iron foot in real time. The edge computing unit uses the force feedback signal to control the rotation angle of the flipping motor in a closed loop, ensuring that the contact pressure is always maintained within the preset optimal threshold range. This design avoids the problems of increased contact resistance and low measured values ​​caused by insufficient contact pressure, while also preventing mechanical damage to the glaze layer on the insulator surface or the probe caused by excessive pressure, thus ensuring the repeatability and accuracy of insulation resistance measurements.

[0049] The environmental sensing unit collects real-time data on temperature, humidity, and electric field strength at the work site, while the edge computing unit corrects the insulation resistance measurements based on a built-in environmental compensation model. For example, in high humidity environments, a water film may form on the insulator surface, leading to increased leakage current. The environmental compensation model normalizes the measured values ​​using a pre-calibrated humidity-resistance correction curve, eliminating the interference of environmental factors on the test results and ensuring the comparability of test data obtained under different seasons and climatic conditions.

[0050] The edge computing unit inputs non-contact detection data, contact measurement data, and environmental parameters into a pre-trained fusion diagnostic model (such as a decision tree or lightweight neural network), and outputs the insulator status category (good / attentive / deteriorated / zero value) and confidence level. This fusion diagnostic strategy overcomes the shortcomings of single detection modes (relying solely on insulation resistance values) which are easily affected by contact status and environmental factors, and significantly reduces the probability of false alarms and false negatives through cross-validation of multi-source data.

[0051] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. An intelligent insulator detection system based on UAV mounting, comprising a detection device, characterized in that: It also includes a visual perception unit, a multimodal detection unit, an edge computing unit, and an environmental perception unit installed on the detection device; the visual perception unit is used to acquire real-time images of the insulator string; the multimodal detection unit includes an ultraviolet imaging sensor and an infrared thermal imaging sensor, as well as a contact force sensor and a contact resistance monitoring module. The edge computing unit includes an embedded AI chip, a memory, a wireless communication module, and a positioning module; the environmental sensing unit includes a temperature and humidity sensor, a barometric pressure sensor, and an electric field strength sensor.

2. The intelligent insulator detection system based on UAV mounting as described in claim 1, characterized in that: The visual perception unit is an industrial camera, installed at both ends of the mounting bracket along the extension direction of the insulator string, and there are at least two of them.

3. The intelligent insulator detection system based on UAV mounting as described in claim 1, characterized in that: The visual perception unit is equipped with a supplementary light that automatically provides supplementary lighting in low-light environments.

4. The intelligent insulator detection system based on UAV mounting as described in claim 1, characterized in that: The edge computing unit runs a lightweight deep learning model and SLAM localization algorithm.

5. The intelligent insulator detection system based on UAV mounting as described in claim 1, characterized in that: The ultraviolet imaging sensor and infrared thermal imaging sensor in the multimodal detection unit are integrated on the outer frame of the detection device, and the contact force sensor and contact resistance monitoring module are set on the detection needle; the edge computing unit is set inside the center plate or protective cover; the environmental sensing unit is installed on the outside of the center plate.

6. A detection method for an intelligent insulator detection system based on a drone, characterized in that: Includes the following steps: S1: Autonomous Attachment and Initial Positioning The drone hoisted the detection system above the insulator string. The visual perception unit collected images in real time, the edge computing unit identified the end features of the insulator string, and automatically controlled the drone to fine-tune its position so that the inverted U-shaped structure of the detection device's mounting bracket was aligned with the insulator string, completing the autonomous unhooking and placement. S2: Multimodal Pre-detection and Path Planning The multimodal detection unit first scans the insulators before and after the current station. The edge computing unit combines the visual image to identify the spatial position of each insulator, establishes an initial "position-state" mapping table for the insulator string, and plans the movement path. Then, the edge computing unit stores the planned path in the memory. S3: Closed-loop movement and autonomous positioning The edge computing unit triggers the drive unit of the detection device, which drives the detection system to move along the insulator string. The vision perception unit collects images of the insulator in front in real time, estimates the relative displacement of the device, and combines the feedback from the encoder of the drive motor to achieve closed-loop position control. When the system device is detected to have reached the preset detection position of the target insulator piece, the edge computing unit sends a stop command to the drive unit. S4: Adaptive Contact Detection Upon reaching the target station, the flipping motor drives the detection probe to flip downwards. The contact force sensor monitors the contact pressure in real time. When the pressure reaches a preset threshold, the flipping stops to ensure reliable contact without damaging the insulator surface. The zero-value detector performs insulation resistance measurement, while the contact resistance monitoring module records the contact resistance value between the probe and the insulator for data correction. S5: Multimodal Data Fusion and Real-time Diagnosis The edge computing unit performs spatiotemporal alignment on the collected data of the current insulator, then runs the fusion diagnostic model, outputs diagnostic conclusions, and provides confidence levels; the diagnostic results and related data are uploaded to the cloud or ground station in real time. S6: Automatic Loop and Termination Detection After the test is completed, the test needle is reset, and the drive unit continues to move to the next station. When the vision perception unit recognizes the end of the insulator string, it determines that the test is complete, the device automatically stops moving, and sends a retrieval signal to the drone via wireless communication.

7. The detection method of the insulator intelligent detection system based on UAV mounting as described in claim 6, characterized in that: In step S2, when the multi-modal detection unit scans the front and rear insulators, it activates the ultraviolet imaging sensor to collect the corona discharge intensity and discharge pulse frequency around each insulator, and the infrared thermal imaging sensor to acquire the surface temperature distribution data of the insulator skirt, steel cap and iron foot area.

8. The detection method of the intelligent insulator detection system based on UAV mounting as described in claim 6, characterized in that: The pressure threshold for real-time monitoring of contact pressure by the contact force sensor in step S4 is 0.5-2.0N.

9. The detection method of the intelligent insulator detection system based on UAV mounting as described in claim 6, characterized in that: The contact resistance monitoring module mentioned in step S4 can use the four-wire method or the AC impedance measurement method.

Images