Machine learning for extending capability of mechanical characterization to improve springback prediction of a quenching and partitioning steel

Abstract

Quenching and partitioning (QP) steels exhibit apparent tension-compression asymmetry and evolving Bauschinger effect as plastic deformation proceeds, which brings challenges to accurately predict springback of QP steel parts. Loading-reverse loading tests are conducted to characterize the Bauschinger effect of a QP steel with a strength grade of 1500 MPa (QP1500), where the maximum achievable strains are limited. A machine-learning method is developed to extend the capability limit of physical tension-compression (TC) or compression-tension (CT) tests and virtually generate stress vs. strain curves at larger strains (up to 0.18), which cover the strain history of an actual forming part (e.g. M-shaped part). Then a data-driven method is proposed to calibrate parameters of a kinematic hardening model (Yoshida-Uemori model), which adopts all TC and CT stress vs. strain curves from physical tests and machine learning. This new set of Yoshida-Uemori model parameters is used in finite element simulations to predict springback of an M-shaped QP1500 steel part, and an obviously improved agreement is reached between simulation and experimental measurements of the M-shaped part. The mean error of predicted local springback distance was reduced to 0.23 mm. It is demonstrated that machine learning method is capable to capture the asymmetric and evolving Bauschinger effect within a much broader strain range and improve springback prediction of QP1500 steel. • QP1500 steel exhibits evolving Bauschinger effect as a function of strain. • QP1500 steel shows obvious tension-compression asymmetry between TC & CT loading. • Machine learning (ML) is employed to acquire TC/CT curves at larger strains. • A data-driven calibration method for YU parameters uses data from experiment & ML. • Data-driven calibration enhances springback prediction of QP1500 steel part.