Research on recognition method of wear debris based on YOLO V5S network
Industrial Lubrication and Tribology
ISSN: 0036-8792
Article publication date: 11 April 2022
Issue publication date: 30 May 2022
Abstract
Purpose
The purpose of this paper is to identify smaller wear particles and improve the calculation speed, identify more abrasive particles and promote industrial applications.
Design/methodology/approach
This paper studies a new intelligent recognition method for equipment wear debris based on the YOLO V5S model released in June 2020. Nearly 800 ferrography pictures, 23 types of wear debris, about 5,000 wear debris were used to train and test the model. The new lightweight approach of wear debris recognition can be implemented in rapidly and automatically and also provide for the recognition of wear debris in the field of online wear monitoring.
Findings
An intelligent recognition method of wear debris in ferrography image based on the YOLO V5S model was designed. After the training, the GIoU values of the model converged steadily at about 0.02. The overall precision rate and recall rate reached 0.4 and 0.5, respectively. The overall MAP value of each type of wear debris was 40.5, which was close to the official recognition level of YOLO V5S in the MS COCO competition. The practicality of the model was approved. The intelligent recognition method of wear debris based on the YOLO V5S model can effectively reduce the sensitivity of wear debris size. It also has a good recognition effect on wear debris in different sizes and different scales. Compared with YOLOV. YOLOV, Mask R-CNN and other algorithms%2C, the intelligent recognition method based on the YOLO V5S model, have shown their own advantages in terms of the recognition effect of wear debris%2C the operation speed and the size of weight files. It also provides a new function for implementing accurate recognition of wear debris images collected by online and independent ferrography analysis devices.
Originality/value
To the best of the authors’ knowledge, the intelligent identification of wear debris based on the YOLO V5S network is proposed for the first time, and a large number of wear debris images are verified and applied.
Keywords
Acknowledgements
Financial support from Guangdong Provincial Science and technology project(2020B1212070022)and Post-doctoral Program of Guangzhou Mechanical Engineering Research Institute, Co, Ltd (17300065).
Citation
Shi, X., Cui, C., He, S., Xie, X., Sun, Y. and Qin, C. (2022), "Research on recognition method of wear debris based on YOLO V5S network", Industrial Lubrication and Tribology, Vol. 74 No. 5, pp. 488-497. https://doi.org/10.1108/ILT-08-2021-0334
Publisher
:Emerald Publishing Limited
Copyright © 2020, Emerald Publishing Limited