Prediction of hot ductility of steels from elemental composition and thermal history by deep neural networks

ABSTRACT

An artificial neural network model is used to predict high-temperature ductility of steels from their composition and thermal history. The model used literature data on reduction of area (RA). In most types of steel, RA has a U-shaped or V-shaped function of temperature; this shape was represented using a Gaussian fit. The predictive model considers conditions, including alloy composition and thermal conditions. The predicted values agreed well with most experimental values. This model can predict ductility trough for a wider composition range and thermal history than previous studies have achieved. The model also presents how fine components such as titanium (Ti) and nitrogen (N) affect changes in the hot-ductility trough. This model can be used to set steel-casting operating conditions to ensure that steel is not at the temperature in which ductility is low when the slab passes through the bending/unbending area of a continuous caster.

KEYWORDS:

• Continuous casting
• deep learning
• reduction of area
• alloying effects
• Gaussian fitting

Disclosure statement

No potential conflict of interest was reported by the authors.