An Experimental Study for the Construction of an AI-based Image Diagnosis System for Recycled Automobile Parts
-
- HAO Hu
- Waseda University
-
- YANG Wenbo
- JARA Corporation
-
- CHENG Tianhao
- Waseda University
-
- ONODA Hiroshi
- Waseda University
Bibliographic Information
- Other Title
-
- 自動車リサイクル部品を対象としたAIによる画像診断システムの構築に向けた実験的検討
Abstract
<p>In the production process of end of life vehicles (ELVs), skilled workers inspect the ELVs visually. The dependency on skilled workers is high, causing problems related to ageing, knowledge transfer, and quality variation. In this study, we investigated the possibility of improving the production process of recycled automobile parts by automating a part of the parts inspection process. Specifically, we investigated the possibility of introducing an AI-based image diagnosis system and developed and evaluated a prototype of the system. We conducted a preliminary study at a production plant of recycled automobile parts and found that the time required for the production and inspection process of doors was relatively large among the exterior parts. In addition, we found that “scratches” accounted for a large proportion of the damage to exterior parts. Therefore, we conducted an experimental study to determine the presence or absence of scratches on doors using AI. The results showed that the loss function decreased, and the accuracy of the AI system was about 97% after 100 training sessions. As a result of considering the extension to the damage other than “scratches”, it was suggested that the method could be applied to other damage such as “dents”, “rust”, and “stepping stones” on the premise that the accurate teacher data of 500 to 1,000 pieces are secured. This means the system can cover about 65% of the total damage. On the other hand, one of the challenges for this system’s practical use is establishing a method of collecting accurate supervisory data, and it is expected that the system will be upgraded on the assumption that AI will perform the image diagnosis.</p>
Journal
-
- ENVIRONMENTAL SCIENCE
-
ENVIRONMENTAL SCIENCE 35 (5), 276-281, 2022-09-30
SOCIETY OF ENVIRONMENTAL SCIENCE, JAPAN
- Tweet
Details 詳細情報について
-
- CRID
- 1390293589392536960
-
- ISSN
- 18845029
- 09150048
-
- Text Lang
- ja
-
- Data Source
-
- JaLC
-
- Abstract License Flag
- Disallowed