Towards a Data-Driven Evolutionary Model of the Cyclic Behaviour of Austenitic Steels

Understanding the cyclic behaviour of austenitic stainless steels is important in assessing the integrity of pipework in a wide range of applications including in the nuclear industry. Design and assessment codes such as R5 address the cyclic evolution through phenomenological approaches which can also be informed by models such as Armstrong-Frederick, Ohno-Wang and Chaboche-Lemaitre. They aim to predict the evolution of combined isotropic and kinematic hardening under cyclic loading. More recently, advances have been made in the development of both phenomenological and physically based crystal plasticity models in modelling the cyclic behaviour of steels, however the application of these is limited by their computational expense.

In this study, the application of data-driven approaches for modelling the cyclic evolutionary behaviour is investigated. A synthetic cyclic dataset is produced through crystal plasticity finite element simulations to represent experimental cyclic data. A functional principal component analysis technique is proposed as a reduced order representation of the cyclic data to act as a data-driven parameterisation of the hysteresis curve. This method is shown to be able to accurately recreate the hysteresis loops with only a small number of scalar parameters.