Overhead Crane 4-in-1 Fusion Diagnosis: Infrared, Drone & Acoustic
The Kelude Quad-Sensor Fusion Diagnostic Platform for overhead cranes integrates infrared thermography, drone-based high-altitude inspection, and acoustic-vibration-oil analysis into a unified monitoring system. By leveraging multi-source signal correlation, it enables early fault warning and precise fault localization.
In line with the equipment health management requirements outlined in ISO 4301 (Crane Design Standard) and GB/T 28264 Safety Monitoring and Management System, the failure modes of critical crane components are diverse and often interrelated. For instance, reducer gear wear generates abnormal vibration and metallic particles in the oil; motor winding aging leads to localized temperature rise; and main girder structural cracks are difficult to detect visually at height. No single detection method can cover all these failure types. Kelude's KL-DIAG-4000 Quad-Sensor Fusion Diagnostic Platform addresses this by integrating three subsystems—infrared thermography, multi-sensor drone inspection, and acoustic-vibration-oil analysis—into a unified data platform. This enables spatiotemporal alignment and collaborative warning based on multi-source signals, significantly improving diagnostic accuracy and coverage.
Quad-Sensor Fusion System Architecture
The KL-DIAG-4000 system employs a three-tier architecture. The perception layer comprises an infrared thermal imager (resolution 640×480, NETD <30mK), a multi-sensor drone payload (4K visible light + 640×512 thermal imaging + LiDAR), vibration accelerometers (IEPE type, frequency response 0.5Hz~10kHz), and online oil sensors (combined particle count, viscosity, and moisture). The data fusion layer is deployed on an edge computing gateway (NXP i.MX 8M Plus, 4-core Cortex-A53 + 2-core Cortex-M7), running unified time-stamp synchronization and feature-level fusion algorithms. The application layer uploads data to the cloud platform via 4G/5G, providing dashboard visualization, alert push notifications, and trend analysis.
Data from each subsystem is sampled at rates between 10Hz and 1kHz. The edge gateway uses the IEEE 1588v2 Precision Time Protocol (PTP) to achieve sub-microsecond synchronization, ensuring consistent time stamps between vibration phase data and infrared images. All data is stored in the InfluxDB time-series database. Each crane generates approximately 2.8GB of raw data daily, which is compressed at the edge to about 120MB per day.
After feature extraction at the edge, only feature vectors and alarm events are uploaded, significantly reducing bandwidth pressure on the workshop's industrial network. For example, in a steel plant where 12 overhead cranes in the steelmaking bay are connected to the KL-DIAG-4000 platform, the total daily upload from edge gateways is approximately 1.4GB (compared to ~33.6GB/day of raw data). Each gateway consumes only 12W of power (passive cooling, fanless) and is housed in an IP65-rated enclosure, designed to withstand the extreme temperature variations and vibration shocks of the crane environment. The platform supports Modbus TCP, OPC UA, and MQTT industrial protocols, and is compatible with major PLC brands (Siemens S7-1200/1500, Rockwell ControlLogix, Mitsubishi FX5U), facilitating integration with existing MES and SCADA systems.
Infrared Thermography for Temperature Rise Diagnosis
The infrared thermography module consists of a fixed thermal imager (mounted above the operator cabin, field of view 67°×52°) and a handheld inspection thermal imager. It automatically scans the crane's main electrical and mechanical components every 4 hours. The system uses an AI semantic segmentation model (based on U-Net architecture) to automatically extract 38 key temperature measurement regions (ROIs), including motor junction boxes, VFD radiators, brake friction linings, and cable joints. Temperature thresholds for each ROI are automatically matched according to the equipment's technical specification.
When the temperature rise rate of a specific ROI exceeds 5°C/hour, or the absolute temperature surpasses the threshold (e.g., motor winding ≥130°C, referencing the Class F insulation limit per GB/T 755-2019), the system triggers an alert and saves the thermal imaging video segment for that period. Field data shows that this module can provide 48 to 72 hours of advance warning for motor bearing preheating issues. The infrared data also participates in multi-source fusion decisions, cross-validating with vibration features to reduce false alarm rates.
The thermal imager is equipped with a motorized focus lens (focal length 25mm, field of view 24°×18°), providing a single-pixel spatial resolution of approximately 0.6mm at a working distance of 15m. This is sufficient to distinguish terminal overheating and localized hot spots on the bearing outer ring. Each inspection generates a thermal image sequence that is automatically archived to the equipment's digital profile, supporting historical trend comparison. Maintenance personnel can retrieve temperature curves for the same ROI over the past 30, 90, or 180 days from the platform interface. These curves are overlaid with an ambient temperature correction line (based on workshop ambient temperature sensor data) to identify abnormal temperature rises caused by faults rather than seasonal variations. The built-in temperature difference alarm algorithm is based on the two-level thresholds recommended by ISO 18434-1: Δt≥5°C (compared to a symmetrical point) and Δt≥8°C (compared to the historical baseline).
Drone-Based Multi-Sensor Inspection Module
The drone inspection module is designed for high-altitude components that are difficult to access, such as main girders, end carriages, and crane rails. The inspection drone features a 4K visible light camera, a 640×512 thermal imager, and a LiDAR sensor for 3D mapping and dimensional measurements. It can operate for up to 45 minutes on a single charge and is equipped with obstacle avoidance radar for safe operation in complex industrial environments.
During an automated inspection flight, the drone follows a pre-programmed path to capture high-resolution images and thermal data of critical structural areas. The AI-based image recognition algorithm automatically identifies surface cracks, corrosion, loose bolts, and other anomalies on the main girder and structural welds. The LiDAR data is used to generate a 3D point cloud model of the crane, which can be compared against the original CAD model to detect structural deformation or misalignment. All inspection data is transmitted to the edge gateway in real-time via 5GHz Wi-Fi for immediate processing and analysis.
The drone is also equipped with a thermal imager that can detect temperature anomalies on high-altitude electrical lines and busbars, which are often overlooked by fixed thermal cameras. The system's intelligent flight control allows it to hover at a safe distance (typically 3-5 meters) from the target component to capture precise thermal images without risking collision. The entire inspection process, from takeoff to landing, is fully automated and can be scheduled remotely, reducing the need for manual inspections and improving overall safety.
Acoustic-Vibration-Oil Fusion Diagnosis Module
This module combines three complementary sensing technologies to provide a comprehensive health assessment of rotating machinery. Vibration analysis uses IEPE accelerometers mounted on motor bearings, reducer housings, and drum bearings to detect imbalance, misalignment, bearing wear, and gear mesh issues. Acoustic emission sensors capture high-frequency stress waves generated by friction, impacts, and crack propagation, providing early detection of incipient faults. Oil analysis monitors the condition of lubricating oil in reducers, detecting metal particles, water ingress, and viscosity changes that indicate internal wear.
The fusion algorithm correlates features from all three data sources to identify fault patterns with high confidence. For example, an increase in vibration at gear mesh frequency combined with the presence of iron particles in the oil and specific acoustic emission patterns strongly indicates reducer gear wear. The system automatically generates diagnostic reports with recommended maintenance actions, helping to transition from reactive to predictive maintenance strategies.
The acoustic-vibration-oil fusion module is particularly effective for detecting faults in reducers and gearboxes, which are critical components in overhead cranes. By integrating data from these three sensing modalities, the system can distinguish between different fault types (e.g., gear wear vs. bearing pitting) and estimate the remaining useful life of components. This allows maintenance teams to schedule repairs during planned downtime, minimizing production disruptions.
Edge Computing and Data Management
The edge computing gateway is the central processing unit of the KL-DIAG-4000 platform. It performs real-time data acquisition, signal processing, feature extraction, and local decision-making. The gateway runs a Linux-based operating system and supports containerized applications for flexible deployment and updates. It is equipped with multiple communication interfaces, including Gigabit Ethernet, 4G/5G, and Wi-Fi, ensuring reliable data transmission in various industrial settings.
Data management is a key aspect of the platform. All raw and processed data is stored in a centralized cloud platform, providing a single source of truth for all crane health information. The platform offers a user-friendly dashboard that displays real-time status, historical trends, and alarm notifications. It also supports role-based access control, allowing different stakeholders (e.g., maintenance technicians, plant managers, safety officers) to view relevant information according to their needs.
The system is designed to be scalable and interoperable. It supports standard industrial protocols (Modbus TCP, OPC UA, MQTT) and can be easily integrated with existing plant systems such as MES and SCADA. The cloud platform provides RESTful APIs for data export and integration with third-party analytics tools. This open architecture ensures that the KL-DIAG-4000 platform can adapt to evolving industrial IoT ecosystems and support future enhancements.
Frequently Asked Questions
Q: What is the main advantage of the Quad-Sensor Fusion Diagnostic Platform compared to traditional single-sensor monitoring systems?
A: The primary advantage is its ability to detect a wider range of fault types with higher accuracy. By integrating infrared thermography, drone inspection, and acoustic-vibration-oil analysis, the platform provides a more comprehensive view of equipment health. Multi-source data fusion reduces false alarms and enables earlier detection of incipient faults, leading to more effective predictive maintenance.
Q: Can the system be retrofitted to existing overhead cranes?
A: Yes, the KL-DIAG-4000 platform is designed for easy retrofitting. The sensors and edge gateway can be installed on existing cranes without major modifications. The system's flexible mounting options and wireless communication capabilities minimize installation time and disruption to crane operations.
Q: What kind of training is required for maintenance personnel to use this system?
A: The platform is designed with a user-friendly interface that requires minimal training. Basic operation, such as viewing dashboards and acknowledging alarms, can be learned in a few hours. For advanced features like configuring diagnostic rules or analyzing historical data, Kelude provides comprehensive training sessions and detailed documentation.
Drone-Based Multi-Sensor Overhead Crane Inspection
A: A multi-sensor payload mounted on a quadcopter drone (DJI Matrice 350 RTK) combines a 4K visible-light camera, a 640×512 thermal imaging camera, and a Livox Mid-360 LiDAR unit. The drone follows a pre-programmed 3D flight path to automatically inspect elevated crane components, including the top flange plate of the main girder, end carriage connections, trolley rails, conductor rails, and wire ropes. Navigation relies on RTK differential GPS (centimeter-level positioning accuracy) integrated with a pre-built BIM model, enabling fully autonomous flight with no manual intervention.
Visual imagery is processed in real time using the YOLOv8 object detection model, which automatically identifies coating delamination areas, loose bolts (via Hough transform detection of bolt head angular deviation), and broken wire rope strands (using line-scan analysis to detect surface texture anomalies). LiDAR point cloud data is used to measure crane rail straightness deviation (standard requirement ≤2 mm/m per FEM 1.001) and changes in main girder camber. Each inspection generates a structured report that flags defect type, position coordinates (relative to the crane coordinate system), and risk level. The typical inspection cycle is monthly, with each overhead crane inspection taking approximately 15 minutes.
Inspection reports are automatically compared against historical data to highlight new defects and track the progression of existing ones. For coating delamination, the system uses an image segmentation algorithm (DeepLabV3+) to calculate the percentage of paint loss and generate trend curves—when the delaminated area in a given zone grows by more than 20% across two consecutive measurements, the system flags accelerated coating failure and recommends scheduling anti-corrosion treatment ahead of the regular maintenance plan. The main girder camber trend curve also feeds directly into predictive maintenance: per FEM 1.001 requirements, mid-span camber must fall within 0.9/1000 to 1.4/1000 of the span. An alert is triggered when LiDAR-measured camber drops below 0.5/1000 or the rate of decline exceeds 0.1/1000 per year. The KL-DIAG-4000 platform consolidates these structural health indicators with vibration and infrared data into a unified equipment health dashboard.
Acoustic-Vibration-Oil Fusion Diagnosis for Gearbox Health
The acoustic-vibration-oil fusion diagnosis module serves as the system's core diagnostic engine. IEPE acceleration sensors (PCB 352C33, measuring range ±50 g) are mounted on the reducer, motor, and drum bearing housing to capture vibration signals, while an online oil condition sensor (Parker Kittiwake KS10) installed at the reducer base continuously monitors metal particle count, viscosity, and moisture content in the lubricating oil. Vibration analysis applies the four-level vibration severity thresholds of ISO 10816-3:2009 (≤1.8 / 1.8–4.5 / 4.5–11.2 / ≥11.2 mm/s for Class I equipment), while also extracting frequency-domain features (FFT spectral sideband energy ratios at 1× and 2× rotational frequency) and time-domain features (kurtosis factor, impulse factor).
When vibration severity crosses the yellow threshold (4.5 mm/s) and ferromagnetic particle concentration in the oil simultaneously exceeds 2,000 particles/mL, the system diagnoses early-stage gear wear (rather than bearing failure), with a confidence score of ≥0.85. This vibration-oil cross-validation mechanism reduces the false alarm rate from approximately 18% for single-dimensional diagnosis to about 4% for fusion diagnosis. The acoustic-vibration-oil module performs a comprehensive analysis automatically every 12 hours, with each analysis cycle taking roughly 8 minutes (including vibration pulse acquisition, oil sampling, and AI inference).
Multi-Source Collaborative Warning Mechanism and Performance Evaluation
The four-in-one platform's collaborative warning system employs a multi-level evidence fusion decision algorithm. Each sub-module independently calculates a fault probability score (0–1), which is then combined at the decision level using weighted Dempster-Shafer (D-S) evidence theory. Weight coefficients are dynamically adjusted based on each sub-module's diagnostic confidence for a given failure mode—for example, gear wear uses vibration weight 0.45, oil weight 0.40, infrared weight 0.10, and drone weight 0.05.
In a pilot deployment on six 32 t casting cranes at a steel mill, the system generated 42 warnings over a six-month operating period. Of these, 35 were verified during shutdown inspections as genuine fault precursors (83.3% accuracy), while 7 were false alarms (primarily infrared false positives triggered by sudden ambient temperature changes). Compared with traditional single-dimensional diagnosis, the fusion diagnosis extended the average early warning lead time from 7.2 days (vibration-only) to 12.5 days—a 73.6% improvement. Kelude Heavy Industry continuously refines its algorithm models based on diagnostic data from the platform and delivers quarterly platform upgrade packages to customers.
| Diagnosis Scenario | Single-Dimension Diagnosis Accuracy | Fusion Diagnosis Accuracy | Advance Warning Time |
|---|---|---|---|
| Reducer Gear Wear | 71.3% | 89.7% | 14~21Day |
| motor bearing Preheating | 65.8% | 86.2% | 48~72Hour |
| Wire Rope Wire Break | 78.4% | 91.5% | Real-Time |
| Crane Rail Straightness Out-of-Tolerance | 82.1% | 94.3% | Real-Time(Inspection) |
| Main Girder Crack | 58.6% | 82.9% | 7~14Day |
| Average | 71.2% | 88.9% | — |
Frequently Asked Questions
Q: What advantages does the four-in-one integrated diagnostic platform offer over purchasing three separate systems?
A: The core advantage lies in data synergy. Three independent systems each have their own data acquisition units, upload channels, and management interfaces, making cross-system time synchronization and feature correlation impossible—when vibration signals indicate an anomaly, the inspection drone may still be charging. With KL-DIAG-4000's unified time-base synchronization, when the vibration sensor captures an abnormal impact, the system automatically retrieves the infrared thermal image and oil particle data from the same moment, generating a complete fault evidence chain. Additionally, the unified platform reduces deployment costs by approximately 35% compared to three separate systems (hardware sharing of the edge gateway, power supply, and communication links).
Q: How is flight reliability and safety ensured for drones in high-dust, high-temperature overhead crane workshop environments?
A: We use the DJI Matrice 350 RTK industrial-grade drone with an IP55 protection rating and an operating temperature range of −20 to +50°C. Before takeoff, environmental conditions are automatically checked—takeoff is refused if wind speed exceeds 8 m/s, dust concentration exceeds 5 mg/m³, or temperature falls outside the operating range. Throughout the flight, forward and rear obstacle-avoidance radar (horizontal detection range 0.5–20 m, vertical 0.5–15 m) remains active, automatically decelerating when the drone approaches within 2 m of the overhead crane steel structure. If RTK signal is lost or the remote control link is interrupted for more than 5 seconds, the drone automatically returns to its launch point (RTH function). In actual steel mill deployments, over 1,200 sorties have been completed with zero safety incidents.
Q: Do the oil condition sensors used in the vibration-acoustic-oil fusion diagnosis require regular calibration and maintenance?
A: The Parker Kittiwake KS10 online oil condition sensor has a factory calibration interval of 12 months. The sensor features built-in self-diagnostics that perform a sensitivity check automatically once per day. Metal particle counting uses the laser light-blocking principle, requiring accuracy calibration every 3 months using standard particle dispersion fluid (the calibration kit provides standard samples at 1,000 particles/mL and 5,000 particles/mL concentrations). Viscosity measurement employs the quartz crystal microbalance principle and requires calibration every 6 months. Under normal oil conditions (AS 1638 Hydraulic fluid contamination standard Grade ≤8), the sensor demonstrates reliable continuous operation with a mean time between failures (MTBF) of ≥18,000 hours.
Q: What retrofits are required on existing overhead cranes to deploy the four-in-one platform?
A: The main retrofit work includes: installing an edge computing gateway and 4G/5G communication module in the overhead crane electrical cabinet (occupying 2U of rack space); drilling and tapping mounting points on the motor bearing housing, reducer input/output ends, and drum bearing housing to install IEPE vibration sensors (standard configuration of 8–12 measurement points per crane); adding an online oil condition sensor circulation loop at the reducer drain port; installing a fixed mounting bracket for the thermal imager above the operator cab along with power and communication cabling; and importing drone flight routes—no modification to the crane structure itself is required. These retrofits can be completed within a normal scheduled maintenance shutdown window (approximately 2 days), with crane downtime limited to ≤4 hours (electrical control system wiring).
