**A Comprehensive Review of Conveyor Belt Deviation (Misalignment) Detection Methods Based on Machine Vision, Deep Learning, and Condition Monitoring**

**A Comprehensive Review of Conveyor Belt Deviation (Misalignment) Detection Methods Based on Machine Vision, Deep Learning, and Condition Monitoring**

sina soleimani1 mahmod eskandari nasab2

1) M.Sc. Student in Mineral Processing Engineering, Shahid Bahonar University of Kerman
2) Ph.D. in Mineral Processing Engineering, Shahid Bahonar University of Kerman

IOI Code : ABEB-ACZGGA
Publication : 4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
    Lateral conveyor belt deviation (belt mistracking/deviation) is one of the most common and costly failures in conveyor systems used in mining, bulk material handling, and production lines. It can lead to longitudinal belt tearing, material spillage, severe wear of rollers and frames, increased drive energy consumption, and, in severe cases, fires as well as human and financial losses. According to reported statistics, production downtime caused by belt deviation may account for 10–30% of the total downtime of a conveying line. This paper presents an analytical review of five representative studies published in reputable international journals—Algorithms (2019), Applied Sciences (2023), Complexity (2021), Scientific Reports (2024), and Sustainability (2021)—in which state-of-the-art methods for conveyor belt deviation detection, monitoring, and mitigation are classified, analyzed, and compared. These approaches include a multiscale feature-fusion network based on depthwise separable convolutions for real-time belt-edge detection; an inspection robot equipped with a lightweight OM-SSD detector, combined with the Hough transform and arc-pattern matching for continuous estimation of the degree of deviation (DD); an improved YOLOv8 algorithm incorporating attention mechanisms (ESE, BiFPN_DoubleAttention, and MHSA) for simultaneous detection of rollers and conveyor belts; and four non-contact monitoring approaches based on bearing vibration measurements, RGB imaging, thermal imaging, and hydraulic tension-force measurements. Furthermore, the effects of roller-bearing design parameters—including bearing type, radial clearance class, and axial clearance—on rolling resistance and the overall energy consumption of conveyor systems are investigated. The results indicate that deep-learning-based methods can achieve mean average precision values exceeding 99% (mAP@0.5) while maintaining real-time processing speeds of 46–67 frames per second (FPS), substantially outperforming traditional mechanical-switch methods and conventional image-processing techniques such as Canny/Hough. In addition, appropriate selection of roller bearings and axial clearance can reduce rolling resistance by more than 60% and the total energy consumption of the conveyor system by approximately 12%. By providing comparative tables and graphical analyses, this paper proposes an integrated framework for selecting an appropriate conveyor belt deviation monitoring solution according to industrial operating conditions.

Keywords : industrial conveyor belt belt deviation detection machine vision deep learning yolov8 inspection robot roller rolling resistance condition monitoring