Enabling intelligent Mg-sheet processing utilizing efficient machine-learning algorithm

Abstract

Process – property relationship control during magnesium sheet manufacturing is demanding due to the complexity of involved physical parameters and the sensitivity of the system to small changes. Here, data science might help to extract crucial information on interdependencies between processing parameters and sheet quality. In this paper we suggest a dedicated machine learning framework, which enables the possibility of correlating material property determining concepts such as pole figure to processing parameters, namely temperature and deformation degree without knowledge on prior dependencies of physical variables. Despite the impacts that using a relatively small data set can have, for Mg-AZ31 alloy we show that some projections of crystallographic texture can be reliably predicted from mechanical measurement data set. In general, the framework is useful for those processing parameters, which conventionally can be represented by a mathematical basis in the context of interpolation. In the future with access to more data it is proposed that applying our approach might allow predicting and controlling in-situ the rolling process route.