Pengembangan Implementasi Convolutional Neural Network Dalam Inspeksi Visual Permukaan Drill Cutting Tools

Nazaruddin, Risaf Nazaruddin (2025) Pengembangan Implementasi Convolutional Neural Network Dalam Inspeksi Visual Permukaan Drill Cutting Tools. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022231022-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Perusahaan yang memproduksi drill cutting tool harus menjaga kualitas agar tidak mengirimkan barang yang defect ke pelanggan. Di dalam proses pengecekan tim QC masih memiliki kekurangan seperti ketajaman penglihatan dikarenakan perbedaan intensitas cahaya antar waktu, konsentrasi pun berkurang ketika bekerja mendekati batas waktu kerja satu hari, dan ketepatan dari setiap member QC pun berbeda-beda. Jadi terkadang terdapat barang yang lolos uji QC yang terkirim ke pelanggan dan dapat mempengaruhi kepuasan pelanggan. Maka dari itu dibuatlah sebuah sistem inspeksi visual yang berisi sebuah algoritma yang mendeteksi cacat menggunakan Convolutional Neural Network dengan arsitektur YOLOv11. Pendeteksian menggunakan metode YOLOv11 diharapkan mempunyai akurasi yang tinggi dan deteksi yang lebih cepat. Dalam proses perancangannya diawali dengan melakukan desain hardware maupun software. Dikarenakan drill cutting tool berbentuk silinder, maka digunakan motor stepper untuk memutar benda tersebut sehingga dapat diinspeksi dari dua posisi sisi tajam mata carbide yang dikontrol oleh PLC (Programmable Logic Control), untuk pengambilan gambar yang normal dan defect diambil 770 gambar dari cutting tool yang kemudian dimasukkan kedalam aplikasi Roboflow untuk memandai bagian yang defect dan kemudian dilakukan training menggunakan program dengan bahasa python. Setelah itu program di deploy pada komputer dengan mengaktifkan kamera dan mengambil gambar langsung dari Drill Cutting Tools, setelah selesai pada satu sisi maka rotary table memutar untuk melakukan inspeksi pada sisi yang lainnya. Pada pengujian dilakukan perbandingan antara tipe dengan ukuran yang berbeda pada YOLOv11 yaitu YOLOv11n (nano), YOLOv11s (small) dan YOLOv11m (medium). Untuk implementasi model dengan device yang sudah terinstalasi maka YOLOv11n merupakan model yang dipilih dikarenakan waktu proses yang paling cepat dan akurasinya yang masih diatas 90%. Dengan waktu proses antara 110.34 ms - 123.24 ms maka identifikasi dapat dilakukan secara realtime. Solusi ini diharapkan mampu mendeteksi defect secara akurat sehingga dapat menjaga kualitas produk.
=================================================================================================================================================================================================
The company that manufactures drill cutting tools must maintain quality in order to avoid providing defective products to clients. In the checking process, the QC team still has shortcomings such as visual acuity because of differences in light strength at different times, concentration lowers when working near the ending of the workday, and each QC member's precision varies. As a result, some items pass quality control inspection and are supplied to clients, affecting customer satisfaction. As a result, a visual inspection system was developed, complete with a defect detection algorithm based on Deep Learning with the YOLOv11 architecture. It is expected that the YOLOv11 approach will detect objects more faster along with high accuracy. Hardware and software design are the first steps in the design process. Because the drill cutting tools are cylindrical, an actuator that is controlled by a PLC (Programmable Logic Control) was created to rotate the object so that it may be inspected from different perspectives. 770 photos of the cutting tool were taken in order to capture both normal and defective photographs. The images were subsequently entered into the Roboflow application, which labeled the parts that were defective. The program written in Python was then used for training. The software is then installed on a computer by turning on the camera and capturing images straight from the Drill Cutting Tools. After that, the program is deployed on the computer by activating the camera and taking pictures directly from the Drill Cutting Tools. Once one side is finished, the rotary table rotates to inspect the other side until a full 360 degrees is completed. During the testing process, a comparison was made between types with different sizes on YOLOv11, such as YOLOv11n (nano), YOLOv11s (small), and YOLOv11m (medium). For type implementation with installed devices, YOLOv11n was chosen because of its fastest processing time and accuracy which is still above 90%. With a processing time between 110.34 ms - 123.24 ms, the inspection can be done in real time. This solution is expected to accurately detect the defects, thereby maintaining product quality.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Quality Control, Drill Cutting Tool, Convolutional Neural Network, YOLOv11
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Risaf Nazaruddin
Date Deposited:	24 Jul 2025 07:47
Last Modified:	24 Jul 2025 07:47
URI:	http://repository.its.ac.id/id/eprint/121287

Actions (login required)

View Item