Sistem Klasifikasi Baja Karbon Berdasarkan Karakteristik Morfologimikrostruktur Berbasis Deep Learning

Septiadi, Ivan Azwar (2025) Sistem Klasifikasi Baja Karbon Berdasarkan Karakteristik Morfologimikrostruktur Berbasis Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241040-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (5MB) | Request a copy

Abstract

Baja karbon adalah material dasar yang digunakan dalam berbagai industri karena sifat mekanisnya yang beragam dan biaya yang efektif. Sifat dan kinerja baja karbon sangat dipengaruhi oleh struktur mikrograpinya, yang bervariasi tergantung pada proses pembuatan dan komposisinya. Berdasarkan kandungannya baja karbon dapat dikategorikan menjadi tiga jenis, baja karbon rendah, sedang dan tinggi. Klasifikasi jenis baja karbon biasanya dilakukan secara manual dengan pengamatan visual oleh ahli material dengan menganalisa mikro strukturnya. Digitalisasi dan otomatisasi pada proses klasifikasi ini perlu di lakukan untuk meningkatkan proses klasifikasi ini sangat penting untuk kontrol kualitas dan prediksi karakteristik serta prilaku material. Penelitian ini mengadopsi teknologi deep learning untuk mengotomatiskan klasifikasi baja karbon ke dalam tiga kelas yaitu baja karbon rendah, baja karbon menengah dan baja karbon tinggi dengan menggunakan citra hasil metalografi yang diperoleh melalui mikroskop optik. Lima model deep learning yaitu Convolutional Neural Network, ResNet50, EfficientNet-B5, DenseNet-121, dan VGGNet-19 diuji menggunakan dataset yang terdiri dari 17,616 gambar metalografi yang diambil secara mandiri. Evaluasi kinerja model dilakukan dengan menghitung metrik seperti Accuracy, Precision, Confusion Matrix, Recall, dan F1-Score. Hasil evaluasi model menunjukkan performa yang memuaskan, dengan semua model mencapai nilai akurasi di atas 90%. Model ResNet50 mendapatkan nilai akurasi tertinggi dengan akurasi 99.22%, diikuti oleh EfficientNet-B5 dengan 97.63%, VGGNet-19 dengan 96.62%, DenseNet-121 dengan 95.25% dan model CNN mendapatkan hasil 91.55%. hal ini dapat menegaskan potensi otomatisasi sistem klasifikasi baja karbon untuk membantu proses inspeksi yang selama ini dilakukan secara manual oleh ahli metalurgi.
===================================================================================================================================
Carbon steel is a fundamental material used across various industries due to its diverse mechanical properties and cost-effectiveness. The properties and performance of carbon steel are greatly influenced by its microstructure, which varies based on the manufacturing process and composition. Carbon steel can be categorized into three types based on its content: low, medium, and high carbon steel. Typically, the classification of carbon steel types is conducted manually through visual inspection by materials experts who analyze its microstructure. However, digitalization and automation of this classification process are essential for enhancing quality control and predicting the characteristics and behaviour of the material. This research adopts deep learning technology to automate the classification of carbon steel into three classes: low, medium, and high carbon steel, using metallography images obtained through optical microscopy. Five deep learning models, namely Convolutional Neural Network, ResNet50, EfficientNet-B5, DenseNet-121, and VGGNet-19, were tested using a dataset consisting of 17,616 independently captured metallographic images. The performance evaluation of the models was conducted using metrics such as Accuracy, Precision, Confusion Matrix, Recall, and F1-Score. The model evaluation results showed satisfactory performance, with all models achieving accuracy rates above 90%. The ResNet50 model achieved the highest accuracy rate of 99.22%, followed by EfficientNet-B5 with 97.63%, VGGNet-19 with 96.62%, DenseNet-121 with 95.25%, and the CNN model achieving 91.55%. These results affirm the potential of automating the carbon steel classification system to assist the inspection process, which has traditionally been performed manually by metallurgy experts

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Carbon steel, Deep Learning, Convolutional Neural Network, Resnet50, EfficientNet-B5, VggNet-19, DenseNet-121, Baja karbon, Deep Learning, Convolutional Neural Network, Resnet50, EfficientNet-B5, VggNet-19, DenseNet-121
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7888.3 Digital computers
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Ivan Azwar Septiadi
Date Deposited:	22 Jan 2025 03:56
Last Modified:	22 Jan 2025 03:56
URI:	http://repository.its.ac.id/id/eprint/116580

Actions (login required)

View Item