Bayesian Calibration of a Physical Model for Plastic Flow Behavior of Trip Steels

Abstract

Rivera’s physical model [Modeling Simul. Mater. Sci. Eng. 22 (2014) 015009 (22pp)] based on irreversible thermodynamics of dislocation evolution has been used in this work to predict the stress-strain behavior of TRIP steels during plastic deformation. This model has been applied to account for plastic deformation of individual phases, and the iso-work approximation has been used as a homogenization framework to simulate the behavior of the composite system. Contribution to plastic behavior due to strain-induced martensitic transformation in austenite has also been taken into account. Calibration and analysis of parameter uncertainty is performed using a Bayesian method based on Metropolis-Hastings Markov Chain Monte Carlo algorithms. Using this approach, the model has been trained with different experimental data to estimate parameters and their uncertainties. The parameter posterior probability distribution obtained from is considered as the prior probability distribution for subsequent training. The stress-strain curves obtained from the model with new estimated parameters show good agreement with the experimental data in literature.