Klasifikasi Kemasan Pakan Kucing Berdasarkan Karakteristik Visual Antarvarian untuk Robot Palletizer Menggunakan Edge Learning dan Rule-Based Vision

Maulana, Mahesa Ilham (2026) Klasifikasi Kemasan Pakan Kucing Berdasarkan Karakteristik Visual Antarvarian untuk Robot Palletizer Menggunakan Edge Learning dan Rule-Based Vision. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221108-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Proses palletizing pada lini produksi industri memerlukan sistem klasifikasi produk yang mampu mengenali varian kemasan secara akurat agar robot palletizer dapat menentukan jalur outfeed secara otomatis tanpa bergantung pada proses pemilahan manual. Namun, perbedaan karakteristik visual antarvarian kemasan, meliputi warna dominan, logo, ilustrasi kucing, tipografi, dan tata letak elemen grafis, serta kondisi kemasan yang mengalami kerutan plastik, perubahan posisi, dan pantulan cahaya dapat menurunkan keandalan klasifikasi apabila hanya menggunakan satu metode inspeksi. Untuk mengatasi permasalahan tersebut, penelitian ini mengimplementasikan sistem klasifikasi kemasan pakan kucing menggunakan metode Hybrid Vision yang mengkombinasikan Edge Learning dan Rule-Based Vision pada kamera Cognex In-Sight dan diintegrasikan Programmable Logic Controller. Sistem melakukan klasifikasi awal menggunakan alat ViDi EL Classify berdasarkan visual global kemasan, kemudian memvalidasi area karakteristik berupa ilustrasi kucing menggunakan alat Detect Pattern sebelum menghasilkan keputusan akhir valid berupa, kemasan warna pink dengan Chicken & Tuna Flavor serta ilustrasi anak kucing oranye untuk varian 1, kemasan warna ungu dengan varian Tuna Flavor serta ilustrasi kucing dewasa oranye untuk Varian 2, kemasan warna oranye dengan varian Chicken & Tuna Flavor serta ilustrasi kucing dewasa abu-abu Varian 3, atau Reject untuk produk yang tidak memenuhi kriteria validasi. Pengujian dilakukan terhadap 448 citra yang terdiri atas 142 citra Varian 1, 235 citra Varian 2, dan 71 citra Varian 3. Hasil pengujian menunjukkan akurasi klasifikasi oleh sistem Hybrid Vision sebesar 99,33% pada Varian 1 dan Varian 2, serta 100% pada Varian 3 tanpa terjadi kesalahan klasifikasi menuju varian lain. Selain itu, hasil integrasi antara Cognex dan PLC menunjukkan tingkat match sebesar 98,66%, di mana 442 dari 448 hasil klasifikasi berhasil sesuai dengan recipe aktif sehingga produk diarahkan menuju jalur outfeed, sedangkan 6 data yang tidak memenuhi proses validasi secara otomatis dikategorikan sebagai Reject. Hasil tersebut menunjukkan bahwa metode Hybrid Vision mampu menghasilkan sistem klasifikasi yang andal untuk mendukung proses robot palletizer di lingkungan industri.
===================================================================================================================================
The palletizing process on industrial production lines requires a product classification system capable of accurately identifying packaging variants so that the palletizing robot can automatically determine the outfeed path without relying on manual sorting. However, differences in visual characteristics among packaging variants—including dominant colors, logos, cat illustrations, typography, and the layout of graphic elements—as well as packaging conditions such as plastic wrinkles, positional shifts, and light reflections can reduce classification reliability if only a single inspection method is used. To address these issues, this study implements a cat food packaging classification system using the Hybrid Vision method, which combines Edge Learning and Rule-Based Vision on a Cognex In-Sight camera and is integrated with a Programmable Logic Controller. The system performs an initial classification using the ViDi EL Classify tool based on the global visual appearance of the packaging, then validates characteristic areas—specifically the cat illustrations—using the Detect Pattern tool before generating a final valid decision: pink packaging with the Chicken & Tuna flavor and an illustration of an orange kitten for Variant 1; purple packaging with the Tuna flavor and an illustration of an orange adult cat for Variant 2; orange packaging with the Chicken & Tuna Flavor variant and a gray adult cat illustration for Variant 3, or “Reject” for products that do not meet the validation criteria. Testing was conducted on 448 images, consisting of 142 images of Variant 1, 235 images of Variant 2, and 71 images of Variant 3. The test results showed that the Hybrid Vision system achieved a classification accuracy of 99.33% for Variants 1 and 2, and 100% for Variant 3, with no misclassifications into other variants. Additionally, the integration results between Cognex and the PLC showed a match rate of 98.66%, where 442 out of 448 classification results successfully matched the active recipe, directing the products to the outfeed lane, while the 6 data points that did not meet the validation criteria were automatically categorized as Rejects. These results demonstrate that the Hybrid Vision method is capable of produce a reliable classification system to support the robotic palletizer process in an industrial environment.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Palletizer, Hybrid Vision, PLC, Cognex In-Sight, Integrasi. Palletizer, Hybrid Vision, PLC, Cognex In-Sight, Integration.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T58.62 Decision support systems T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Mahesa Ilham Maulana
Date Deposited:	12 Aug 2026 01:02
Last Modified:	12 Aug 2026 01:02
URI:	http://repository.its.ac.id/id/eprint/144268

Actions (login required)

View Item