Mesh generation for the simulation of Li-ion batteries on supercomputers using GPUs
We are developing a simulation software that will help significantly improve the performance of next-generation Li-ion batteries. These batteries are widely used as portable power sources, in your smartphone or laptop as well as in Boeing’s 787 Dreamliner or Tesla automobiles. Our software will be designed to run on Piz Daint, Europe’s fastest supercomputer.
Your task is to investigate algorithms and/or develop a software to generate the geometrical model of electrodes in Li-ion batteries for the simulation on supercomputers equipped with GPUs. Depending on your interests, we can define a more theoretical or a hands-on programming project.
- Looking for 1-2 students in Electrical Engineering, Computer Science or related fields
- Project duration can be defined according to your needs
- Contact: Michael Bieri
- Be fluent in a programming language, preferably Matlab, C, C++ or Java
- Background in computer engineering or computer science
- No knowledge about numerical simulation or batteries required
- 10-40% Research about existing approaches
- 20-80% Programming
- 10-40% Theory
Detailed Task Description
Over the last couple of years, Graphics Processing Units (GPUs), originally designed for video games, have evolved into high performance parallel processors. GPUs provide a peak performance up to 10x higher than a high-end 8-core CPU and are therefore widely used in current supercomputer architectures. However, in order to fully leverage their computational power, novel approaches for numerical simulation have to be developed.
The first step in many simulation techniques such as the Finite Element Method (FEM) is to approximate the geometry of the model by a (triangular) mesh. Even though most batteries are manufactured in simple cylindrical or cubic shapes, the performance-critical inner electrodes are complicated structures with scales going from nanometers to several 100 micrometers. It is a challenging task to build a useful mesh on such a multi-scale structure. Furthermore, the topology of the mesh has to be highly regular in order to achieve good performance on GPUs.
Depending on your interests, you will work on one or more of the following sub-tasks:
- Generation of realistic electrode structures either by simulation or from real-world data in collaboration with experimental groups.
- Research and evaluation of currently existing algorithms and libraries for the construction of regular meshes.
- Implementation and investigation of your own or existing algorithms. The focus could be set on theoretical considerations as well as on a high performance implementation.
- Integration and adaption of an existing mesh generator into our simulation software..
- Investigation of mesh generation directly on GPUs.