Diki Fahrizal, - (2024) ANALISIS KOMPARATIF ALGORITMA FASTER R-CNN DAN YOLOV8 UNTUK DETEKSI OBJEK CACAT PCB LAYOUT: Implementasi Graphical User Interface. S1 thesis, Universitas Pendidikan Indonesia.
Abstract
Seiring dengan perkembangan teknologi elektronika, banyaknya komponen yang diintegrasikan pada papan PCB dengan tata letak yang kompleks dan rumit. Permasalahan cacat kecil pada jejak jalur menyebabkan kegagalan pada fungsi elektronika, maka pemeriksaan permukaan PCB layout adalah salah satu proses kontrol kualitas paling penting. Pada kondisi sekarang, keterbatasan manual inspection untuk mengidentifikasi cacat pada PCB semakin sulit dan kompleks yang tidak bisa dijangkau oleh mata manusia, dari tantangan ini muncul adanya kebutuhan sistem pemeriksaan permukaan PCB dengan memanfaatkan deep learning untuk melakukan Automated Optical Inspection (AOI) berbasis deteksi objek. Tujuan dari penelitian ini mengembangkan dua algoritma deep learning yaitu Faster R-CNN dan YOLOv8, serta melakukan analisis komparasi performa algoritma, menentukan algoritma yang terbaik untuk diintegrasikan berbasis Graphical User Interface sebagai solusi yang ditawarkan dalam kebutuhan AOI untuk deteksi cacat pada permukaan PCB layout. Penelitian ini menggunakan model pengembangan ADDIE (Analysis, Design, Development, Implementation, Evaluation). Hasil dari penelitian ini, algoritma YOLOv8 memiliki performa terbaik dibandingkan Faster R-CNN dengan varian model YOLOv8x menjadi pilihan yang terbaik untuk melakukan tugas deteksi objek cacat pada permukaan PCB, dengan skor performa 0,962 (mAP@50), 0,503 (mAP@50:95), 0,953 (Precision), 0,945 (Recall), dan 0,949 (F1-Score). GUI yang dikembangkan menunjukkan performa rata-rata waktu inference model sebesar 7198,96 millisecond dan rata-rata waktu untuk post-process sebesar 2104,88 millisecond dari percobaan 30 gambar yang memiliki cacat PCB. Hasil temuan ini diharapkan menjadi inovasi baru bagi industri manufaktur elektronika untuk kebutuhan sistem pemeriksaan papan PCB dengan metode AOI berbasis GUI. Selain itu, menjadi referensi bagi peneliti lainnya untuk memperluas aplikasi teknologi AOI. Along with the development of electronic technology, many components are integrated on PCB boards with complex and complicated layouts. The problem of small defects in the trace path causes the failure of electronic functions, so the surface inspection of PCB layout is one of the most important quality control processes. In the current condition, the limitations of manual inspection to identify defects on PCBs are increasingly difficult and complex that cannot be reached by the human eye, from this challenge comes the need for a PCB surface inspection system by utilizing deep learning to perform Automated Optical Inspection (AOI) based on object detection. The purpose of this research is to develop two deep learning algorithms, namely Faster R-CNN and YOLOv8, and conduct a comparative analysis of algorithm performance, determining the best algorithm to be integrated based on the Graphical User Interface as a solution offered in the AOI needs for defect detection on the PCB layout surface. This research uses the ADDIE development model (Analysis, Design, Development, Implementation, Evaluation). As a result of this research, the YOLOv8 algorithm has the best performance compared to Faster R-CNN with the YOLOv8x model variant being the best choice for performing defect detection tasks on PCB surfaces, with performance scores of 0,962 (mAP@50), 0,503 (mAP@50:95), 0,953 (Precision), 0,945 (Recall), and 0,949 (F1-Score). The developed GUI shows the performance of the average time for inference model of 7198,96 milliseconds and the average time for post-process of 2104,88 milliseconds from the experiment of 30 images that have PCB defects. These findings are expected to be a new innovation for the electronics manufacturing industry for the needs of PCB board inspection systems with the GUI-based AOI method. In addition, it is a reference for other researchers to expand the application of AOI technology.
![[img]](http://repository.upi.edu/style/images/fileicons/text.png) |
Text
S_TE_2001691_Title.pdf Download (6MB) |
|
Text
S_TE_2001691_Chapter1.pdf Download (2MB) |
|
Text
S_TE_2001691_Chapter2.pdf Restricted to Staf Perpustakaan Download (12MB) |
|
Text
S_TE_2001691_Chapter3.pdf Download (4MB) |
|
Text
S_TE_2001691_Chapter4.pdf Restricted to Staf Perpustakaan Download (15MB) |
|
Text
S_TE_2001691_Chapter5.pdf Download (1MB) |
|
Text
S_TE_2001691_Appendix.pdf Restricted to Staf Perpustakaan Download (28MB) |
| Item Type: | Thesis (S1) |
|---|---|
| Additional Information: | https://scholar.google.com/citations?user=IINnkjIAAAAJ&hl=en ID SINTA Dosen Pembimbing : Jaja Kustija : 5993968 |
| Uncontrolled Keywords: | Faster R-CNN, YOLOv8, Graphical User Interface, Defect Object Detection, Automated Optical Inspection (AOI). Faster R-CNN, YOLOv8, Graphical User Interface, Defect Object Detection, Automated Optical Inspection (AOI). |
| Subjects: | Q Science > QA Mathematics T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Fakultas Pendidikan Teknik dan Industri > Jurusan Pendidikan Teknik Elektro > Program Studi Pendidikan Teknik Elektro |
| Depositing User: | Diki Fahrizal |
| Date Deposited: | 04 Sep 2024 03:54 |
| Last Modified: | 04 Sep 2024 03:54 |
| URI: | http://repository.upi.edu/id/eprint/122395 |
Actions (login required)
 |
View Item |