System for Identifying the Condition of Hoisting Crane Runways

Abstract

Introduction . Inspecting the overhead crane runways in production workshops and warehouses is a challenging and dangerous task. The risks for specialists are associated with the high altitude at which the runways are located and the lack of walkways along them. At the same time, standard visual and dimensional inspection is characterized by a low inspection speed. The scientific literature contains numerous studies on the potential of artificial intelligence (AI) to ensure occupational safety: methods for monitoring occupational risks and preventing accidents, the relationship between accident rates and the competencies of crane operators, the detection of defects in removable load‑handling devices using computer vision tools, as well as remote monitoring of crane safety based on video data from IP cameras. However, these solutions do not address the inspection of crane runways, which have specific features, primarily their considerable length (often up to 200 meters or more). The application of AI requires video-analytical monitoring along the entire length of the runway and training a neural network to recognize local defects and geometric deviations, which existing methods do not provide. Today, crane runway inspection is conducted through visual and dimensional control methods, including direct inspection and surveying using total stations and theodolites. While these methods are accurate, they are labor-intensive and time-consuming. Therefore, there is a need for the use of AI to improve safety and speed while maintaining accuracy in identifying faults in crane runways. The aim of this research is to develop a method for remote inspection of the overhead crane runways located at height in industrial facilities. This will help minimize the exposure of workers to hazardous and harmful working conditions while maintaining accuracy, speed, and reliability of the inspection results. Materials and Methods . Data on crane runway defects collected during inspections at industrial facilities was used as the basis for the study. The methodology for identifying defects in crane runways was based on GOST R 56944–2016 . Computer vision neural networks were trained using open libraries for Python language. A modernized pre-trained YOLOv8 neural network was used to detect defects. Results. A method for remote detection of defects on overhead crane runways was developed using an unmanned aerial vehicle designed by the authors (a quadcopter with a protective frame, equipped with a Livox MID‑40 lidar, an Orbbec Gemini 2 depth camera, a 4K RGB camera, and a DWM1000 positioning system supporting TWR and TDOA). Based on survey data collected in 2024 from 352 overhead cranes with a total runway length of approximately 14 kilometers, a modernized YOLOv8 neural network for computer vision was trained. This resulted in the creation of a complete three-dimensional point cloud that covered the runways, crane beams and supports, as well as reference to column lines and centers. The three-dimensional model allowed for the automatic identification and classification of local defects, as well as the estimation of their sizes with accuracy of one millimeter. Automated geometry assessment showed that the deviations in the runway markings in the model in question did not exceed the permitted values according to GOST R 56944–2016  (40 millimeters in one section and 10 millimeters on adjacent columns), which confirmed the effectiveness of the method. Discussion. The results obtained indicate that the authors’ goal has been achieved — the development of a method for remote inspection of overhead crane runways. This was made possible by conducting a significant number of surveys, which provided a diverse range of defects for training neural networks. A comparison with previous studies has shown the uniqueness of the proposed approach to inspecting overhead crane runways. Methods based on artificial intelligence and unmanned aerial vehicles (UAVs) have previously been used to monitor personnel, assess removable lifting attachments, and inspect tower cranes outdoors. However, these methods were not suitable for detecting local defects in overhead crane runways inside production facilities. The main limitation of the developed method was the flight time of the UAV (no more than 20 minutes), which was due to the low battery capacity. This capacity could not be increased without increasing the maximum size of the device (0.5 meters) in the confined conditions of enclosed spaces. The new method's results were positive, as it enabled inspection of tracks along their entire length from a close distance. By building a 3D model using photogrammetry, it was possible to assess the size of defects and reduce labor intensity and duration of the survey by at least half. These benefits made its further development and practical implementation worthwhile. Conclusion . The main outcome of the study was the development of a method for remote inspection of overhead crane runways located at height in industrial facilities. During the research, neural networks were trained to detect defects, and an algorithm was created for inspection. This involved creating a three-dimensional model that allowed for automated assessment of geometric deviations in the runways in both longitudinal and transverse planes, as well as the identification of local defects. The key benefit of this method was that it eliminated the need for experts to climb to heights, ensuring their safety. Additionally, it allowed for the inspection of hard-to-reach areas and the detection of previously invisible defects, reducing the likelihood of future emergency situations. Further research in this field will focus on improving the system's ability to automatically identify the condition of load-bearing metal structures and possible defects in cranes.