Project Details
Projekt
Machine learning and high-performance computing for efficient simulation of thermo-elastoplastic solidification processes
Applicants
Professor Dr. Axel Klawonn; Dr. Martin Lanser
Subject Area
Mathematics
Joining and Separation Technology
Joining and Separation Technology
Term
since 2021
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 434946896
As a flexible and contact-free joining technology, laser beam welding has increasingly gained importance. Processing of alloys with large melting range poses a challenge due to their solidification cracking tendency. Solidification cracks form due to critical stress and strain states of the dendritic microstructure with inter-dendritic melt. Despite the high industrial relevance, there are only approaches addressing single aspects of the problem, metallurgically or structurally oriented. The research unit “Solidification Cracking during Laser Beam Welding – High Performance Computing for High Performance Processes” aims at developing quantitative process understanding of the mechanisms of solidification cracking and their relation to process parameters. This subproject is dedicated to multi-scale simulations of the processes in the mushy zone right before the solidification front - the critical zone for the initiation of solidification cracks. This is done in a strong co-design with TP4. One focus is on the extension of the homogenization method FE$^2$ within the software package FE2TI, as well as a multiscale method based on machine learning (ML). While the FE$^2$ method solves many localized elastoplastic boundary value problems to compute the homogenized stress, we alternatively rely on ML-based surrogate models that predict the stresses directly. Another focus is the further development of robust and highly scalable implicit solvers for thermo-elastoplastic problems - in particular nonlinear two-level Schwarz methods. As in the first phase of the project, Schwarz domain decomposition methods with monolithic GDSW coarse-grid spaces are used to solve the thermo-elastoplastic problems. In the second phase, the focus is now on the integration of nonlinear variants of the Schwarz method, which are primarily intended to improve the nonlinear convergence. This class of methods has not yet been used for thermo-elastoplastic problems, but they are promising for enabling larger time steps for large, plastic deformations. The implementation is based on our own PETSc library, which has already proven its parallel scalability for realistic single-scale laser beam welding processes in the first phase. An interface to the finite element and material model library FEAP was developed in the first phase to solve thermoplastic problems using the FE2TI software package. Considering the standards for sustainable software development defined in TP7, including the continuous benchmarking principle, the interface to FEAP is to be further developed together with TP4. In addition, crack indicators developed in various TPs are to be integrated into FE2TI, for example the local stability analysis proposed by TP4, in order to test the various indicators in high-resolution simulations.
DFG Programme
Research Units
