A visual detection method for conveyor belt misalignment based on the improved YOLACT network

Abstract

Conveyor belt misalignment is an operational and maintenance issue for many belt conveyor systems and can result in significant downtime and equipment failure. This article presents a visual detection method for belt misalignment detection using real-time instance segmentation. An improved version of the You Only Look At CoeffcienTs (YOLACT) network is discussed, where the feature backbone of the original YOLACT network is optimized. The improved YOLACT network is used to experimentally detect the belt misalignment with different combinations of bulk solid and operating conditions. The results show that the detection precision of the improved YOLACT network achieves 86.05% mean Average Precision (mAP), in which the greatest improvement on the mAP is the FPN modification with an increase of 1.75% compared to the YOLACT network with the ResNeXt-50. In terms of the detection speed, the improved YOLACT network (15.19 Frames Per Second [FPS]) outperforms the Mask R-CNN (3.42 FPS) and the MS R-CNN (3.41 FPS). Moreover, the averaged detection error from the tests is 0.74 mm for a belt loaded with different bulk solids. It can be concluded that the application of the improved YOLACT network is effective for the real-time detection of conveyor belt misalignment.

Keywords:

• Conveyor belt misalignment
• visual detection
• feature backbone
• improved YOLACT

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work is financially supported by National Natural Science Foundation of China (Grant No.: 52202500) and Hainan Institute at Wuhan University of Technology (Grant No.: 2021KF0028).