Code and Data: Efficient multiscale simulations of incremental sheet forming using machine learning surrogate models for crystal plasticity

Code and data for "Efficient multiscale simulations of incremental sheet forming using machine learning surrogate models for crystal plasticity" by John S. Weeks and Aaron P. Stebner at Georgia Institute of Technology. Journal article DOI: https://doi.org/10.1007/s40192-025-00427-0 Multiscale crystal plasticity modeling of metal forming is expensive due to complex loading conditions and high-cost microscale models. However, these simulations can be significantly accelerated through use of low-cost machine learning surrogate models. This work develops machine learning surrogate models for the constitutive response and texture evolution of a microscale crystal plasticity model using recurrent neural networks, and embeds them into efficient multiscale simulations of single-point incremental forming that demonstrate up to a 63.6x increase in speed compared to conventional techniques. Two concurrent multiscale crystal plasticity simulation workflows are demonstrated for random textured Al5052-H32 with truncated pyramid and truncated cone forming paths which show good agreement between results of forming force, thickness variation, and texture evolution. Surrogate models for the microscale constitutive response are embedded and take strain path history inputs from macroscale loading conditions and output deviatoric stresses at each material point. Surrogate models for texture evolution take resultant plastic strain history inputs and output a reduced order representation of texture using generalized spherical harmonic coefficients and principal component analysis. These workflows permit extraction of the full-field time-history response for texture evolution over the entire formed part which enables multiscale trade studies using crystal plasticity simulations as well as optimization of local microstructures in various forming applications.