Model Order Reduction
Student projects covering the research topics can be found here.
Numerical simulations are a key technology in the design and development of new products and the realization of innovative technical ideas. As the computational power has increased dramatically while the price for powerful hardware has decreased simultaneously in the last decades, numerical simulation has become both, an affordable and at the same time extremely powerful tool for developers and designers. Albeit this developments, the need for large-scale, detailed and powerful simulation models is still unbroken and exceeds in many cases the capability of current hardware. Thus one wants to have both, more detailed and better methods while reducing the computational effort. One answer to this conflicting goals of detailed models and concurrently little computational effort is Model Order Reduction. The goal of Model Order Reduction is – similar to picture- or music compression methods like jpeg or mp3 – the reduction of computational effort and memory consumption while keeping the loss of information minimal and controlled. In the Finite Element Technology – the core technology for the computation of elastic deformations of mechanical parts – the high dimensional and fine meshed models with many degrees of freedom can be reduced to simulation models with few degrees of freedom, which enable detailed solutions while keeping the computational effort and the memory consumption minimal. At the Chair of Applied Mechanics, we develop sophisticated Model Order Reduction Techniques for linear and nonlinear Finite Element Models. The focus is thereby on simulation-free methods which do not need expensive training simulations of an unreduced model. Apart from projection methods, which make huge reduction ratios between full and reduced model possible, we also do active research on so called Hyper Reduction Techniques, which only enable huge speedups of nonlinear models. For the development of Model Order Reduction Methods, we developed the nonlinear Finite Element Code AMFE at the Chair of Applied Mechanics. The code, written in Python/FORTRAN enables both, the rapid prototyping of models and methods as well as the fast execution of high dimensional models. We plan to open source the software in the beginning of 2017
- Dipl.-Ing. Johannes Rutzmoser