Deteksi Kualitas Granulasi Klinker Berdasarkan Ukuran Menggunakan Metode YOLOv8-CNN Berbasis Website Streamlit Untuk PT. Semen Gresik

Isa, Muhamad (2026) Deteksi Kualitas Granulasi Klinker Berdasarkan Ukuran Menggunakan Metode YOLOv8-CNN Berbasis Website Streamlit Untuk PT. Semen Gresik. Diploma thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221040-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Industri manufaktur semen di Indonesia dihadapkan pada persaingan yang ketat sehingga menuntut perusahaan untuk menjaga kualitas produk, khususnya klinker sebagai bahan baku utama. Kualitas granulasi klinker diklasifikasikan berdasarkan ukuran, dengan kategori ideal berada pada rentang diameter 5 mm hingga 25 mm. Keberadaan klinker out of specification, baik yang berukuran terlalu kecil (fines <5 mm) maupun terlalu besar (boulders >25 mm), dapat mengganggu proses produksi, meningkatkan konsumsi energi, serta menambah biaya operasional. Permasalahan yang dihadapi PT. Semen Gresik adalah proses pengecekan kualitas granulasi klinker yang masih menggunakan metode saringan semi-manual, yang memiliki jeda waktu (time-lag) sehingga respons terhadap penyimpangan kualitas menjadi kurang efisien. Penelitian ini bertujuan untuk mengembangkan sistem deteksi kualitas granulasi klinker secara otomatis menggunakan metode YOLOv8 yang diimplementasikan dalam website Streamlit. Dataset penelitian terdiri dari 100 citra asli klinker yang melalui tahap preprocessing, dan augmentation data sehingga menjadi 240 citra, kemudian dibagi menjadi data training, validation, dan testing. Model YOLOv8 Small (YOLOv8s) dilatih menggunakan data training dan dievaluasi menggunakan data validation. Hasil pengujian menggunakan data testing menunjukkan bahwa model menghasilkan nilai Mean Average Precision (mAP@0.5) sebesar 77,8% dengan nilai F1-Score maksimum sebesar 0,76, yang menunjukkan kemampuan generalisasi model yang cukup baik dalam mendeteksi klinker kategori ideal. Model yang dikembangkan berhasil diimplementasikan ke dalam aplikasi webite Streamlit untuk deteksi berbasis citra dan real time. Sistem ini berpotensi mendukung proses pengendalian kualitas klinker secara lebih cepat, objektif, dan konsisten.
===================================================================================================================================
The cement manufacturing industry in Indonesia faces intense competition, requiring companies to maintain product quality, particularly clinker as the main raw material. Clinker granulation quality is classified based on particle size, with the ideal category ranging from 5 mm to 25 mm in diameter. The presence of out-of-specification clinker, either too small (fines <5 mm) or too large (boulders >25 mm), can disrupt the production process, increase energy consumption, and raise operational costs. One of the main challenges faced by PT Semen Gresik is that the inspection of clinker granulation quality is still conducted using a semi-manual sieving method, which involves a significant time lag and results in less efficient responses to quality deviations. Therefore, this study aims to develop an automatic clinker granulation quality detection system using the YOLOv8 method, implemented in a Streamlit-based web application. The dataset used in this study consists of 100 original clinker images, which were processed through preprocessing and data augmentation stages, resulting in 240 images. The dataset was then divided into training, validation, and testing sets. The YOLOv8 Small (YOLOv8s) model was trained using the training data and evaluated using the validation data. Testing results show that the model achieved a Mean Average Precision (mAP@0.5) of 77.8% and a maximum F1-score of 0.76, indicating a good generalization capability in detecting ideal clinker granulation. The developed model was successfully implemented in a Streamlit web application for image-based and real-time detection. This system has the potential to support clinker quality control processes in a faster, more objective, and more consistent manner.

Item Type:	Thesis (Diploma)
Subjects:	Q Science Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Muhamad Isa
Date Deposited:	24 Jul 2026 02:44
Last Modified:	24 Jul 2026 02:44
URI:	http://repository.its.ac.id/id/eprint/136703

Actions (login required)

View Item