Artificial Neural Network: Some Applications in Physical Metallurgy of Steels

Abstract

Artificial neural networks are parameterized nonlinear models used for empirical regression and classification modelling. Their flexibility enables them to discover very complex relationships among large number of variables with complex interdependencies. Hence it is a very appropriate technique for developing predictive models in the short term. This article discusses the development of neural network models based on a Bayesian framework and some applications of the same. These examples include: (i) prediction of yield and tensile strengths of hot-rolled low-carbon ferrite-pearlite steel plates as a function of composition and rolling parameters, (ii) estimation of bainite plate thickness, (iii) estimation of retained austenite as a function of process parameters in Transformation Induced Plasticity aided (TRIP-aided) steels, and (iv) analysis of strain-induced transformation behavior of retained austenite during uniaxial tensile testing of TRIP-aided steels. While examples (i), (ii), and (iv) provide an overview of the work reported earlier, example (iii) is reported here in open literature for the first time. In all these four cases, it has been shown that the results are consistent with the established physical metallurgy principles.