Naslov
Deep Learning for Casting Microstructure Inspection: A Convolutional Neural Network Approach

Autori
Nikolić, Filip ; Štajduhar, Ivan ; Čanađija, Marko

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
My First Conference 2021 – Book of Abstracts / Grbčić, Ana ; Lopac, Nikola ; Strabić, Marko ; Dugonjić-Jovančević, Sanja ; Franulović, Marina ; Vukelić, Goran - Rijeka, 2021, 29-30

Skup
5th edition of annual conference for doctoral students of engineering and technology „MY FIRST CONFERENCE“

Mjesto i datum
Rijeka, Hrvatska, 23.09.2021

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Domaća recenzija

Ključne riječi
secondary dendrite arm spacing ; convolutional neural network ; casting microstructure inspection ; deep learning ; aluminum alloys

Sažetak
In the present research, we used a convolutional neural network (CNN) to determine secondary dendrite arm spacing (SDAS). The goal was to develop a Deep Learning (DL) model capable of predicting SDAS from optical microscopy images with industrially acceptable prediction accuracy. For the training of the CNN model, polished samples from several widely known casting aluminum alloys were used. In order to build a training set having distinct types of microstructures, alloys cast using different casting processes were used: high-pressure die casting, gravity die casting, and ingot casting. Additional alloys were used in various tasks in order to check model performance. We showed that our CNN structure was able to predict SDAS very well, on all different alloy groups. Additionally, it is also shown that using a higher magnification on the microscope, better accuracy could be obtained because more pixels describe the dendrite structure in that case. Consequently, it is shown that our CNN structure can be used for the task of SDAS prediction within the industry because it achieves an acceptable level of predictive accuracy.

Izvorni jezik
Engleski

Znanstvena područja
Strojarstvo, Temeljne tehničke znanosti

POVEZANOST RADA