ACADSTAFF UGM

CREATION
Title : Deep Learning-based Spot Welding Segmentation using Modified UNet with Various Convolutional Blocks
Author :

Oskar Natan, S.ST., M.Tr.T., Ph.D. (1) Diyah Utami Kusumaning Putri, S.Kom., M.Sc., M.Cs. (2) Dr. Andi Dharmawan, S.Si., M.Cs. (3)
Date : 1 2021
Keyword : Image Segmentation,Deep Learning,UNet,Convolutional Blocks,Weld Spot Image Segmentation,Deep Learning,UNet,Convolutional Blocks,Weld Spot
Abstract : Welding inspection is an absolute need for an industrial factory to ensure the quality of weld joins. However, most of the industry still uses manual inspection which can be subjective, full of bias, and it will lead to inconsistency of the quality standard. Therefore, an intelligent system that can check the quality of welding automatically is needed. As the first step in developing this system, this research aims to create a knowledge model based on deep learning and computer vision that used to segment weld spots of iron. A convolutional neural networks (CNN) model that adopts the architecture of UNet is used as the main model with plenty of modification on its architecture and hyperparameter tuning. The study of using several convolutional blocks is also conducted to achieve the best model configuration. The model works by segmenting the captured image or video frame to determine the region of weld spots. As a result, a modified UNet model with dense convolutional blocks on its bottleneck has the best performance according to the percentage of Intersection over Union (IoU) which reaches 75.45% on the validation set.
Group of Knowledge : Sistem Informasi Geografi (SIG)
Original Language : English
Level : Internasional
Status :
Published
Document
No Title Document Type Action
1 2020 Surat Tugas Hibah Dosen Muda.pdf
Document Type :
View
2 similarity check ICICELB2021.pdf
Document Type : [PAK] Cek Similarity
[PAK] Cek Similarity View
3 Halaman Editorial (1).pdf
Document Type : Dokumen Pendukung Karya Ilmiah (Hibah, Publikasi, Penelitian, Pengabdian)
Dokumen Pendukung Karya Ilmiah (Hibah, Publikasi, Penelitian, Pengabdian) View
4 Bukti korespondensi new ICICELB.pdf
Document Type : [PAK] Bukti Korespondensi Penulis
[PAK] Bukti Korespondensi Penulis View
5 Artikel.pdf
Document Type : [PAK] Full Dokumen
[PAK] Full Dokumen View