A Case Study on Evaluation of Defect Characteristics for Practical Application of Appearance Inspection Work Support System Utilizing Deep Learning-Reference-Cited by-同舟云学术

A Case Study on Evaluation of Defect Characteristics for Practical Application of Appearance Inspection Work Support System Utilizing Deep Learning

Published:2023 Issue: Volume: Page:536-541
ISSN:2195-4356
Container-title:Lecture Notes in Mechanical Engineering
language:
Short-container-title:

Author:

Nakakura Yuta,Temizu Kosuke,Nishino Mana,Nakajima Ryosuke

Abstract

AbstractTo prevent outflow of defective products to customers, many manufacturing industries have focused not only on processing and assembling, but also on product inspection. In appearance inspection, a work support system using deep learning has been proposed, and its usefulness was experimentally shown in model images of industrial product in recent years. Therefore, in this study, aiming for practical application of work support system, the relationship between the lighting angles and the visibility of defects is experimentally evaluated using 80 actual automobile parts as a case study. As results, it is found that the visibility of the defect greatly differs depending on the angle of lighting, and the conditions for high visibility differ depending on the defect. Furthermore, it is found that it is possible to improve the visibility of defects in about 24% of all 80 automobile parts, but, it is difficult to improve the visibility of defects for the remaining 76%. From the above, for the practical application of the work support system, it is clarified that the importance of constructing of lighting condition optimization method to improve the visibility of defects, and examining the input image considering the visibility of defects for deep learning.

Publisher

Springer International Publishing

Link

https://link.springer.com/content/pdf/10.1007/978-3-031-28839-5_60

Reference5 articles.

1. Chugoku Regional Innovation Research Center.: [Survey on the feasibility of promoting automation of inspections at production sites of manufacturing companies.]Monodukuri kigyou no seisan genba ni okeru kensa no jidouka sokushin kanousei chousa (2016)

2. Akio, S.: [Peripheral visual inspection method] Shuhen shi mokushi kensa hou. [I] In: IE review, The Japan Institute of Industrial Engineering, vol. 46, No. 4, pp. 65–75 (2005)

3. Chengkan, L., Zengtao, Z., Fei, S., Feng Z., Hu. S.: A fast surface defect detection method based on background reconstruction. Int. J. Prec. Eng. Manufac. 21(3), 363–375 (2019)

4. Hidenori, K.M., Ryosuke, N.: [Basic research on the development of a work support system for visual inspection using Deep Learning] Deep Learning wo mochiita gaikan kensa no sagyou shien system no kaihatu ni kansuru kiso kenkyu. In: Japan Industrial Management Association Spring Conference 2020, pp. 6–7 (2020)

5. Ryosuke, N., Keisuke, S., Toshiyuki M.: A study on the effect of irradiation angle of light on defect detection in visual inspection. Int. J. Indust. Eng. Theory Appl. Pract. 22(1), 525–537 (2015)