Gaussian process modelling of austenite formation in steel

The present paper introduces the Gaussian process model for the empirical modelling of the formation of austenite during the continuous heating of steels. A previous paper has examined the application of neural networks to this problem, but the Gaussian process model is a more general probabilistic model which avoids some of the arbitrariness of neural networks, and is somewhat more amenable to interpretation. It is demonstrated that the model leads to an improvement in the significance of the trends of the Ac₁ and Ac₃ temperatures as a function of the chemical composition and heating rate. In some cases, these predicted trends are more plausible than those obtained with the neural network analysis. Additionally, it is shown that many of the trace alloying elements present in steels are irrelevant in determining the austenite formation temperatures.