To read this content please select one of the options below:

Research on recognition method of wear debris based on YOLO V5S network

Xinfa Shi (Department of Equipment Lubrication and Testing Research Institute, Guangzhou Mechanical Engineering Research Institute Co., Ltd., Guangzhou, China)
Ce Cui (Department of Equipment Lubrication and Testing Research Institute, Guangzhou Mechanical Engineering Research Institute Co., Ltd., Guangzhou, China)
Shizhong He (Department of Equipment Lubrication and Testing Research Institute, Guangzhou Mechanical Engineering Research Institute Co., Ltd., Guangzhou, China)
Xiaopeng Xie (Department of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou, China)
Yuhang Sun (Department of Equipment Lubrication and Testing Research Institute, Guangzhou Mechanical Engineering Research Institute Co., Ltd., Guangzhou, China)
Chudong Qin (Department of Equipment Lubrication and Testing Research Institute, Guangzhou Mechanical Engineering Research Institute Co., Ltd., Guangzhou, China)

Industrial Lubrication and Tribology

ISSN: 0036-8792

Article publication date: 11 April 2022

Issue publication date: 30 May 2022

233

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

Related articles