ESR9 - Parallel scalability of the computation of quasi-cyclic turbine models using FETI-type methods

Recruiting Institution/Company: Technische Universitaet Muenchen (Germany)

PhD awarded by: Technische Universitaet Muenchen (Germany)

Objectives: The project focuses on the combination of methods from structural dynamics and computer science to establish high-efficient parallelized algorithms for the computation of the nonlinear vibration response of multistage turbines with friction contacts. Large-scale finite element models are decomposed in space using FETI domain decomposition methods, which introduce parallel scalability and enable the computation of quasi-cyclic multistage turbines. Modified versions, like the TFETI method, will be used to solve contact problems and combined with approximation methods frequently used in structural dynamics, like the Harmonic Balance Methods, to compute the nonlinear vibration response in the frequency domain efficiently.

Expected Results: Best practise strategies for the parallelization of the computation of multistage turbines will be identified, implemented and proven concerning stability, accuracy and efficiency.

Planned secondment(s):
GE-GRC: Understanding the current practice in turbines design and modeling.
IT4I: Learning specific Domain Decomposition solver techniques for problems including contact with friction
USTUTT: Learning techniques to efficiently handle high volume I/O needed in computation of large turbine models