Artificial neural network for inferring radiative property variations across advanced high strength steel coils

Abstract

This work proposes a parametric, surrogate artificial neural network (ANN) for inferring the radiative properties of advanced high-strength steel (AHSS) alloys in their cold-rolled and pre-annealed condition using surface optical imagery and topography parameters. The ANN proposed in this study may serve as a complementary method to physics-based models for the in-situ prediction of radiative property variations across cold-rolled AHSS coils with emerging on-line strip imaging tools. The ANN is trained and tested on empirical data from 164 industrially processed AHSS samples whose spectral emissivities and surface topographies are analysed using FTIR spectroscopy and optical microscope imagery with 3D depth mapping, respectively. Further insights into the physical relevance of the model input parameters is obtained through a global sensitivity analysis (GSA) using a Fourier Amplitude Sensitivity Test (FAST).