Drone-Based Overhead Crane Inspection with LiDAR & Thermal Imaging
Core Challenge: How Can a Multi-Sensor Drone Automate High-Altitude Structural Inspections of Overhead Cranes?
A three-module sensing system—20MP visible-light camera, 32-line LiDAR, and 640×480 thermal imaging—is deployed in a three-tier flight plan (overview, fine scan, runway survey). YOLOv8x-seg performs AI-based identification of weld seam cracks, loose bolts, and coating defects, while ICP point-cloud registration enables main girder deflection measurement with ±3 mm accuracy. A single overhead crane inspection takes just 12–18 minutes, improving efficiency by 85%.
Per the periodic inspection and maintenance requirements for metal structures outlined in the ISO 4301 Crane Design Standard, drone-based multi-sensor inspection offers a fresh approach to this long-standing challenge. This article details the complete technical workflow—sensor configuration, flight path planning, image acquisition, and AI-driven defect identification—for using drones to inspect overhead crane structures.
Sensor Configuration and Selection for Drone Inspection Systems
To achieve comprehensive detection of high-altitude crane structures, a drone inspection system must carry multiple sensor types. The core configuration includes four categories: high-resolution visible-light cameras, LiDAR, infrared thermal imagers, and gas sensors. The visible-light camera is used for visual detection of surface defects and should deliver a resolution of at least 20 megapixels, paired with a 20× or higher optical zoom lens to capture crisp detail images of the main girder cover plates, web plates, and weld seams from a safe standoff distance.
LiDAR is employed for structural deformation measurement and 3D reconstruction. A 32-line or higher mechanical LiDAR—or a solid-state unit with equivalent performance—is recommended, offering ranging accuracy of ±2 cm and a scanning range of 360°×30° or greater. This captures geometric parameters such as main girder deflection, rail straightness, and overall structural deformation.
The infrared thermal imager detects temperature anomalies at structural connections (see the thermal imaging AI inspection solution in this series). A high-performance thermal module with 640×480 resolution and NETD ≤0.04°C is recommended. For inspections inside enclosed factory buildings, additional lighting and an RTK high-precision positioning module are required to maintain positioning accuracy in GPS-denied environments.
The drone platform's own specifications directly influence inspection quality. For indoor overhead crane inspections, a six-rotor industrial drone with obstacle avoidance sensors is recommended—600 mm or larger wheelbase, maximum payload of at least 3 kg, and a flight time of 25 minutes or more with payload. It should feature forward, downward, and side-facing binocular vision obstacle avoidance systems, enabling autonomous positioning and hovering via visual SLAM in GPS-denied factory interiors.
For outdoor gantry cranes and port quay cranes, an industrial drone with an IP43 or higher weatherproof rating is recommended, capable of withstanding wind levels up to 6, flying at altitudes above 100 m, and sustaining flight for 40 minutes or more.
Across multiple projects involving Kelude Heavy Industry, the combination of a DJI M300 RTK or M350 RTK equipped with a Zenmuse H20T (20MP visible light + 12MP wide-angle + laser distance measurement + 640×480 thermal imaging) and an L1 LiDAR module has proven to be the most cost-effective and practical configuration.
| Sensor | Detection Target | Key Technology Indicator | detection accuracy |
|---|---|---|---|
| Visible Light Camera | Weld Seam Crack, Coating Peeling, Bolt Loosening, Corrosion Pit | 2000Megapixels,20Optical Zoom | 0.5mm Crack Resolution(10mDistance) |
| Li DARLi DAR | Main Girder Deflection, Crane Rail Straightness, Structure Deformation | 32Line, Ranging Accuracy±2cm, Scanning360°×30° | Deflection Measurement Accuracy±3mm |
| Infraredthermal imager | Connection Bolt Friction Heating from Loosening, Bearing Overheating | 640×480, NETD≤0.04℃ | Temperature Differencedetection accuracy±2℃ |
| Supplementary Lighting | Factory building Auxiliary Lighting for Low-Illumination Areas | 3000lm Color Temperature5000KAdjustable Angle | Illumination Distance Up to30m |
Drone Inspection Route Planning and Autonomous Flight Strategy
Drone inspection routes are planned using a zone-based, multi-tier strategy tailored to the structural characteristics of overhead cranes. The first tier is a general overview route, where the drone ascends to a height of 5–8 meters above the crane structure, centering on the overhead crane itself. It then flies along the main girder axis, capturing visible-light and thermal imaging overview data across the full girder length at a controlled speed of 2–3 m/s.
The second tier is a detailed scanning route. The drone descends to a level parallel with the main girder web plates (at a horizontal distance of 3–5 meters from the web), flying slowly along each side of the girder at 0.3–0.5 m/s. This pass captures high-resolution imagery of critical areas including weld seams, stiffening plate connections, and end carriage connecting bolts.
The third tier is a crane rail inspection route. The drone flies 1–2 meters above the crane rail along its full length, with LiDAR continuously scanning to collect rail cross-section point cloud data for straightness and rail gauge deviation analysis. For a QD Type double-girder bridge crane with a 22.5-meter span, this route planning strategy yields a total inspection path of approximately 260 meters, with a single inspection flight lasting 12–18 minutes and covering 160–200 detection points.
Autonomous flight in indoor GPS-denied environments is a key technical challenge for crane drone inspections. In engineering practice, a multi-sensor fusion approach delivers precise positioning: 4–6 UWB positioning base stations are pre-installed throughout the factory building as a spatial reference framework, while the drone carries a UWB tag for sub-meter indoor positioning (accuracy ±30 cm).
Additionally, a downward-facing visual sensor on the drone supports positioning through texture matching. QR code markers are placed on the factory floor every 10 meters as visual waypoints in well-lit areas. Combined with onboard IMU inertial navigation data, an Extended Kalman Filter fuses these multi-source positioning inputs to achieve indoor positioning accuracy better than ±10 cm. During route execution, the system detects flight deviations in real time and automatically corrects them; when positioning confidence falls below a set threshold, the drone automatically hovers and triggers a re-localization procedure.
AI-Powered Image and Point Cloud Processing for Defect Detection
Raw data collected by the drone falls into three categories—high-definition image sequences, LiDAR point cloud data, and thermal infrared image sequences—each processed through a dedicated AI pipeline.
Image data is analyzed using a YOLOv8 instance segmentation model for surface defect detection. The training dataset covers six typical defect categories: weld seam cracks (longitudinal weld bead cracking and transverse weld toe cracks), coating blistering and peeling (area >1 cm²), missing or loose bolts (bolt head displacement >2 mm classified as loose), corrosion pitting (corrosion pits with depth >0.5 mm), structural deformation (local web plate buckling, flange plate waviness), and fatigue cracking (occurring in weld heat-affected zones or stress concentration areas).
The model input resolution is 1920×1080, and the output provides defect class, bounding box, segmentation mask, and confidence score for each detection. On field data from Kelude Heavy Industry, the YOLOv8x-seg model achieves 91.3% mAP@0.5 for weld crack detection and 88.7% mAP@0.5 for loose bolt detection, with a single-image inference time of 45 ms on an NVIDIA RTX 4090.
LiDAR point cloud processing focuses on structural deformation analysis. The captured point cloud data is first registered against the crane's BIM design model or historical inspection point clouds using the ICP (Iterative Closest Point) algorithm, generating a three-dimensional deviation map of the structure.
Key deviation indicators analyzed include: main girder mid-span deflection (the difference between design camber and actual measurement, with deflection deviation required to be <L/700 per FEM 1.001 for general purpose bridge cranes), rail straightness deviation (<±2 mm per 10-meter measuring segment), end carriage diagonal deviation (difference between the two diagonal lengthsths of the end carriages <5 mm), and main girder web plate flatness (deviation <3 mm within any 1-meter range).
The point cloud processing pipeline also automatically extracts critical dimensional features—main girder flange plate width, web plate height, stiffening plate spacing, and rail center distance—and compares them against design values to generate a deviation report. By comparing point clouds from successive inspections, the system can also generate deformation rate trend charts to predict structural degradation over time.
| Defect Type | Detection Method | AIModel | detection accuracy | Determination Standard |
|---|---|---|---|---|
| Weld Seam Crack | Visible Light Image+AISegmentation | YOLOv8x-seg | m AP@0.5=91.3% | Crack Length>10mm Requires Re-inspection |
| Bolt Loosening | Visible Light Image+Angle Detection | YOLOv8x-seg | m AP@0.5=88.7% | Offset>2mm Loosening Determination |
| Coating Peeling | Visible Light Image+Semantic Segmentation | Deep Lab V3+ | m Io U=85.6% | Area>1cm²Marking |
| Structure Deformation | Li DARPoint Cloud+ICPRegistration | Point Net++ | Deflection±3mm | Deviation> L/700Requires Evaluation |
Structured Inspection Reports & Maintenance Decision Support
Once the drone inspection is complete, the system automatically generates a structured, 3D-visualized inspection report. Using the overhead crane's BIM model or a 3D point cloud model as the base map, all detected defects are marked with icons at their corresponding locations on the 3D model. Red icons indicate critical defects requiring immediate attention, yellow icons denote defects scheduled for planned maintenance, and green icons represent normal conditions.
Clicking any defect icon expands to reveal detailed information: defect type, dimensions, positional coordinates, on-site photos, AI confidence scores, historical comparison data for the same location, and maintenance recommendations.
The report also includes an equipment health scorecard. In accordance with the structural safety evaluation requirements outlined in ISO 4301 (Crane Design Standard) and FEM 1.001 (General Purpose Bridge Cranes), a comprehensive score is calculated across five dimensions: structural deformation, weld seam condition, coating integrity, connector status, and crane rail geometry. The weighting for each dimension is as follows: structural deformation 30%, weld seam condition 25%, coating integrity 15%, connector status 20%, and crane rail geometry 10%.
When the comprehensive score falls below 70, the system automatically issues a maintenance alert and recommends specific repair plans and timeline suggestions.
The long-term value of accumulated drone inspection data is substantial. By performing time-series analysis on multiple inspection datasets from the same overhead crane, trend curves for various inspection indicators can be generated, enabling proactive Predictive Maintenance alerts before thresholds are reached. For instance, the annual rate of change in main girder mid-span deflection can indicate the degradation trend of structural Stiffness. When the annual deflection growth rate exceeds 2%, the system recommends a detailed structural safety assessment.
Similarly, the annual growth rate of coating delamination area reflects the aging of the corrosion protection system. An annual increase exceeding 5% triggers a recommendation for full-machine Anti-Corrosion Treatment. Trends in crack count and length are used directly to determine whether operations should be halted for necessary repairs.
In its drone inspection service practice across multiple factories, Kelude has inspected over 200 overhead crane structures. Through comparative historical data analysis, it has successfully provided early warnings for 12 potential structural safety hazards, including 3 cases of local buckling in the main girder web plate and 9 cases of batch bolt loosening on end carriages.
Drone Inspection Workflow & Safety Protocols
Conducting drone inspections inside industrial factory buildings requires managing safety across three key areas: personnel, equipment, and airspace. For personnel safety, temporary warning tape or signs must be set up around the inspection zone to keep non-essential personnel out of the flight area. Operators must hold a valid CAAC drone pilot license or an equivalent AOPA certificate.
Equipment safety involves a pre-flight checklist confirmation: battery level (minimum 30% reserve), propeller integrity, sensor lens cleanliness, RTK/visual positioning status, and obstacle avoidance system self-check results. For airspace safety, the flight path must be clear of obstacles such as hoisting wire ropes, cable trays, and process piping. Before the first flight in any workshop, an experienced pilot should conduct a manual reconnaissance flight to establish a 3D obstacle map.
Safety strategies for crash back scenarios include: automatic switching to visual positioning mode upon GPS signal loss, automatic return-to-home or hover-in-place upon communication interruption, and automatic calculation of the shortest return path when battery levels are low.
The entire inspection process follows a standardized five-step procedure: Step one involves a Site Survey and flight path planning, creating a 3D obstacle map of the factory building and planning the optimal route. Step two is benchmark placement and equipment Calibration, setting up UWB positioning base stations at the four corners of the building and calibrating the drone. Step three is the execution of the inspection flight, where the drone follows the autonomous route under the supervision of a pilot who monitors the flight status at all times and is ready to take manual control.
Step four is data export and processing, where raw images and point cloud data are uploaded to the AI processing platform for automated Defect Identification and Deviation analysis. Step five is report generation and review, where the AI system produces a draft report that is then verified and confirmed by professional engineers to ensure the reliability of the Detection results.
Kelude offers comprehensive drone inspection engineering services, including equipment rental, on-site execution, data processing, and report delivery. Services are available on a per-inspection or annual service contract basis, helping clients achieve safe, efficient, and Digitalized high-altitude crane structure inspections.
Frequently Asked Questions
Q: Do I need a license to operate a drone for indoor factory inspections?
A: Yes, a Civil Aviation Administration of China (CAAC) drone pilot license is mandatory for commercial inspection operations. According to regulations, drones used for inspection work (with a maximum takeoff weight of 4 kg or more) must be registered, and operators must hold a valid license. While indoor flights do not require airspace approval, operators still need the appropriate level of certification. All drone inspection team members at Kelude hold valid CAAC Visual Line of Sight (VLOS) or Beyond Visual Line of Sight (BVLOS) pilot licenses.
Q: Can drone inspections completely replace manual high-altitude inspections?
A: Drone inspections can replace over 90% of manual high-altitude inspection work, but not all of it. Drones equipped with visible light and LiDAR sensors are excellent for surface and geometric checks. However, detecting internal defects in hidden weld seams (such as lack of fusion or internal Porosity) and precisely measuring Bolt Preload still require traditional methods like Ultrasonic Testing (UT), Magnetic Particle Testing (MT), or torque wrenches. The recommended approach is a combination: quarterly comprehensive drone inspections for high-altitude areas, supplemented by annual detailed manual inspections of critical components.
Q: How does the drone achieve Precise Positioning indoors where GPS signals are weak?
A: Indoor positioning relies on a multi-source fusion system combining UWB base stations, visual SLAM, and IMU inertial navigation. UWB provides centimeter-level absolute position references (Accuracy ±30 cm), visual SLAM achieves relative positioning by matching ground texture patterns with the downward-facing camera (Accuracy ±5 cm), and the IMU supplies high-frequency attitude data. These three data sources are fused in real-time using an Extended Kalman Filter, achieving an overall indoor positioning Accuracy better than ±10 cm. Typically, 4 to 6 UWB base stations are pre-installed in the factory building, with the Density determined by the building's area and structural complexity.
Q: How long does a drone inspection take for a single overhead crane, and what does it cost?
A: A full inspection flight for a single QD Type double-girder bridge crane with a 22.5m span takes 12–18 minutes. Including on-site setup, equipment calibration, and data export, the total time for one operator is approximately 45–60 minutes per crane. For pricing, a one-time inspection runs about $450–$740 per crane (covering equipment, pilot, data processing, and reporting), while an annual contract (quarterly inspections) costs roughly $1,190–$1,780 per crane per year. Compared to traditional manual high-altitude inspection using scaffolding (which costs about $2,220–$3,700 per crane per visit), drone inspection cuts costs by 60–80% while eliminating the safety risks associated with working at height.
Multi-sensor drone inspection technology is transforming how overhead crane structures are inspected at height. By combining a visible light + LiDAR + thermal imaging three-module sensing system, AI-based automatic defect identification, and a professional structured reporting platform, this approach marks a major leap from "sending people up in the air to look" to "drones inspecting automatically." For detailed solutions and technical parameters on drone-based crane inspection services, feel free to contact the Kelude technical team. Kelude holds a team of CAAC-certified drone pilots and extensive experience in overhead crane structural inspection, delivering safe, efficient, and fully traceable high-altitude crane structure inspections for industrial clients.
