TY - JOUR

T1 - Neuromelt model for estimating mold flux melting behaviour

AU - Vargas Hernandez, M

AU - Mapelli, Carlo

AU - Cho, Jungwook

AU - Kölbl, Nathalie

AU - Marschall, Irmtraud

AU - Alloni, Marco

AU - Carli, Ricardo

PY - 2022/1

Y1 - 2022/1

N2 - The existing models and methods used to determine the melting temperature of the mold powders used in the continuouscasting process remain inaccurate in the case of equations reported in current literature or consider for commercial softwa-re only an equilibrium state. In this work, a new approach has been implemented using neural networks, which will act asa “black box” to predict the melting temperature of a mold powder with a given composition within an acceptable range.The proposed neural network will be working as a regression neural model whose inputs will be the composition of eachof the chemical species contained within the powder. A database provided by a research net comprising multiple countries'research institutes will be fed as a training set for network learning. Such data comes from experimental measurementsperformed mainly through the high-temperature microscope test. The correct implementation and training of the networkshould provide a new alternative to develop new products and verify existing products' melting properties. In future mo-dels, further considerations should be made towards a better understanding of these phenomena, which should considerfactors such as the formation of mineral phases, interaction among some specific components of the powder, or even theparameters used at the time of experimental measurement.

AB - The existing models and methods used to determine the melting temperature of the mold powders used in the continuouscasting process remain inaccurate in the case of equations reported in current literature or consider for commercial softwa-re only an equilibrium state. In this work, a new approach has been implemented using neural networks, which will act asa “black box” to predict the melting temperature of a mold powder with a given composition within an acceptable range.The proposed neural network will be working as a regression neural model whose inputs will be the composition of eachof the chemical species contained within the powder. A database provided by a research net comprising multiple countries'research institutes will be fed as a training set for network learning. Such data comes from experimental measurementsperformed mainly through the high-temperature microscope test. The correct implementation and training of the networkshould provide a new alternative to develop new products and verify existing products' melting properties. In future mo-dels, further considerations should be made towards a better understanding of these phenomena, which should considerfactors such as the formation of mineral phases, interaction among some specific components of the powder, or even theparameters used at the time of experimental measurement.

KW - neural network

KW - mold powder

KW - Liquidus temperature

M3 - Article

VL - 2022

SP - 23

EP - 31

JO - Metallurgia Italiana

JF - Metallurgia Italiana

SN - 0026-0843

IS - 1

ER -