AKBAR, ZAIDANE KHAIRUL (2026) IMPLEMENTASI YOLOv8x UNTUK DETEKSI CACAT LAS SECARA REAL-TIME PADA APLIKASI WEB BERBASIS STREAMLIT. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
COVER.pdf Download (1MB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
BAB1.pdf Restricted to Registered users only Download (699kB) |
|
|
Text (BAB II)
BAB2.pdf Restricted to Registered users only Download (1MB) |
|
|
Text (BAB III)
BAB3.pdf Restricted to Registered users only Download (1MB) |
|
|
Text (BAB IV)
BAB4.pdf Restricted to Registered users only Download (2MB) |
|
|
Text (BAB V)
BAB5.pdf Restricted to Registered users only Download (541kB) |
|
|
Text (DAFTAR PUSTAKA)
DAFTARPUSTAKA.pdf Restricted to Registered users only Download (633kB) |
|
|
Text (LAMPIRAN)
LAMPIRAN.pdf Restricted to Registered users only Download (1MB) |
Abstract
Visual inspection of welding quality in industrial environments still faces challenges due to inspector subjectivity and limitations in detecting small-scale defects. This study aims to implement YOLOv8x to detect six welding condition classes, namely Good Weld, Bad Weld, Porosity, Crack, Spatter, and Undercut, and to integrate the model into a Streamlit-based web application to support real-time detection. The dataset consists of 2,828 images that underwent re-labeling, preprocessing, and augmentation. The preprocessing stages included Auto-Orient, resizing to 640 × 640 pixels, and grayscale conversion, while the augmentation techniques included flip, rotation, brightness, blur, and noise. The dataset was divided into 80% training, 10% validation, and 10% testing data. The YOLOv8x model was trained for 300 epochs with a batch size of 4 and evaluated using precision, recall, F1-score, mAP@50, and mAP@50–95. The evaluation results achieved a precision of 85.6%, recall of 84.2%, F1-score of 84.9%, mAP@50 of 86.2%, and mAP@50–95 of 62.1%, with an inference time of 17.1 ms/image. Good Weld achieved the highest F1-score of 91.1%, while Porosity obtained the lowest F1-score of 76.2%. The model was subsequently integrated into a Streamlit web application and achieved an average of 3.03 FPS with a latency of 135.75 ms/frame during real-time testing. The results demonstrate that YOLOv8x can be utilized as a visual inspection assistance tool, although further optimization is required to improve the performance of certain classes and the application's processing speed before deployment in an industrial environment. Kata kunci: Welding Defects, YOLOv8x, Single-Stage Pipeline, Computer Vision, Image Augmentation. Proses inspeksi kualitas hasil pengelasan secara visual pada industri masih menghadapi kendala berupa subjektivitas pemeriksa dan keterbatasan dalam mengenali cacat berukuran kecil. Penelitian ini bertujuan mengimplementasikan YOLOv8x untuk mendeteksi enam kelas kondisi pengelasan, yaitu Good Weld, Bad Weld, Porosity, Crack, Spatter, dan Undercut, serta mengintegrasikannya ke dalam aplikasi web berbasis Streamlit untuk mendukung deteksi secara real-time. Dataset yang digunakan terdiri atas 2.828 citra yang melalui proses re-labeling, preprocessing, dan augmentasi. Preprocessing meliputi Auto-Orient, resize menjadi 640 × 640 piksel, dan grayscale, sedangkan augmentasi meliputi flip, rotation, brightness, blur, dan noise. Dataset dibagi menjadi 80% data latih, 10% data validasi, dan 10% data uji. Model YOLOv8x dilatih selama 300 epoch dengan batch size 4 dan selanjutnya dievaluasi menggunakan precision, recall, F1-score, mAP@50, dan mAP@50–95. Hasil evaluasi menunjukkan nilai precision 85,6%, recall 84,2%, F1-score 84,9%, mAP@50 86,2%, dan mAP@50–95 62,1%, dengan waktu inferensi sebesar 17,1 ms/image. Kelas Good Weld memperoleh F1-score tertinggi sebesar 91,1%, sedangkan Porosity memperoleh F1-score terendah sebesar 76,2%. Model kemudian diintegrasikan ke dalam aplikasi Streamlit dan menghasilkan rata-rata 3,03 FPS dengan latency 135,75 ms/frame pada pengujian real-time. Hasil penelitian menunjukkan bahwa YOLOv8x dapat digunakan sebagai alat bantu inspeksi visual, meskipun masih diperlukan optimasi untuk meningkatkan performa kelas tertentu dan kecepatan aplikasi sebelum diterapkan pada lingkungan industri. Kata kunci: Cacat Las, YOLOv8x, Single-Stage Pipeline, Computer Vision, Augmentasi Citra.
| Item Type: | Thesis (S1) |
|---|---|
| NIM/NIDN Creators: | 41521010064 |
| Uncontrolled Keywords: | Cacat Las, YOLOv8x, Single-Stage Pipeline, Computer Vision, Augmentasi Citra. |
| Subjects: | 000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 004 Data Processing, Computer Science/Pemrosesan Data, Ilmu Komputer, Teknik Informatika 700 Arts/Seni, Seni Rupa, Kesenian > 770 Photography and Photographs/Seni Fotografi dan Foto > 771 Techniques and Procedures/Teknik Seni Fotografi dan Foto, Prosedur Seni Fotografi dan Foto > 771.3 Cameras and Accessories/Kamera dan Asesoris Kamera > 771.31 Specific Makes Brands of Cameras/Pembuatan Merek Kamera |
| Divisions: | Fakultas Ilmu Komputer > Informatika |
| Depositing User: | khalimah |
| Date Deposited: | 04 Sep 2026 03:15 |
| Last Modified: | 04 Sep 2026 03:15 |
| URI: | http://repository.mercubuana.ac.id/id/eprint/103616 |
Actions (login required)
 |
View Item |
