PASIROM – A Success Story for Mathematics in Industry
Electro-mechanical energy converters, such as electrical motors, are an essential part of our everyday lives. We use them in household appliances, power tools, e.g., electric drills, and nowadays for e-mobility, like in e-bikes or e-cars. Many of these electrical devices are operated by induction motors. The wide exploitation range of such asynchronous machines is caused by their relatively cheap costs, due to the absence of expensive permanent magnets, and their high efficiency.
Fig. 1. Exemplary results of the parallel simulation of an induction machine using getDP
Prior to construction of physical prototypes, modern engineering takes advantage of computed-aided design. Modeling, simulation and optimization allow for novel designs of machines, which save energy, are more environment-friendly or have a longer lifetime. However, high-fidelity simulations are time consuming, especially when one is interested in the steady-state behavior of induction machines. This often leads to the application of less accurate models which may result in suboptimal design choices. For this reason, novel efficient numerical algorithms are of paramount importance for further progress in engineering.
Parallel-in-time methods, such as the Parareal algorithm, have great acceleration potential due to the easy work distribution among available multiple processing units of modern computer architectures. Parareal allows to parallelize the classical sequential time stepping and obtain the solution in a shorter time. In the framework of the BMBF project PASIROM (TU Darmstadt, U Koblenz-Landau and BU Wuppertal) several new time parallelization (e.g. doi:10.1137/18M1175653) approaches were developed and applied to speed up the simulation of induction machines. Recently, the methods were implemented into the in-house finite element solver of the Robert Bosch GmbH by Denys Bast, TU Darmstadt in cooperation with Oliver Rain from Bosch and Stefan Kurz (Bosch/TU Darmstadt). Thereby, the steady state behavior of an induction motor can now be obtained up to 28 times faster by exploiting 80 parallel processors. For more details see the Preprint arXiv:1902.08277.
Contact:
Sebastian Schöps and Iryna Kulchytska-Ruchka
Computational Electromagnetics Group, Technische Universität Darmstadt