About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
Presentation Title |
A Practical Deep Learning Fiber Segmentation Approach in a Manufacturing Setting |
Author(s) |
Akira Matsui, Yu Okano, Yoshihige Okuno |
On-Site Speaker (Planned) |
Akira Matsui |
Abstract Scope |
In recent years, there has been an increasing number of Deep Learning applications to the visual inspection process in manufacturing, which offers advantages such as reducing human error, providing objective diagnosis, and saving labor costs.
In this presentation, we demonstrate Deep Learning application to our manufacturing facility, where the automation of visual inspection of carbon nanofiber (CNT) is urgently needed. The challenge we faced was detecting fibers from the images of complexly entangled CNTs and measuring their width and length. In order to address this challenge, we developed a practical approach combining instance and semantic segmentation. We demonstrate the role of semantic segmentation in selecting measurable fibers from the instance segmentation output, which contributes to a feasible prediction of fiber width/length distributions. As a consequence, we accomplished establishing the image analysis system, which achieved the optimization of the visual inspection process as a whole. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, |