Neural Network-Based Modeling of the Interplay between Composition, Service Temperature, and Thermal Conductivity in Steels for Engineering Applications

Abstract

The present study presents an artificial neural network (ANN) model developed to predict the thermal conductivity of steels at different service temperatures based on their composition. The model was developed using a comprehensive database of 413 datasets, spanning diverse steel compositions and pure iron across a temperature spectrum from 0 degrees C to 1200 degrees C, extracted from literature. The ANN model, with steel composition and temperature as inputs and thermal conductivity as output, underwent meticulous experimentation, resulting in an optimal architecture among 291 variations. The model was trained using 253 datasets and validated against an unseen dataset of 160 data points. The model exhibited superior predictive accuracy, boasting an R2 of 98.42%, Pearson's r of 99.21%, and a mean average error of 1.165 for unseen data. The user-friendly software derived from this model facilitates the accurate prediction of thermal conductivity for a wide range of steels, providing a valuable source for industry professionals and researchers.