Mikko Lyly

A method for partitioned fluid-structure interaction computation of flow in arteries.

In this paper we apply the artificial compressibility method (ACM) in strongly coupled fluid-structure interaction (FSI) computation of blood flow in an elastic artery. Previously published and here referred to as the ACM/FSI method uses the idea of artificial compressibility by Chorin 1967, except the term of pressure time derivative in the continuity equation is used to mimic the response of the walls, thereby stabilizing the iterative coupling. To reach the aim, we present a new way, the test load method, to improve ACM/FSI computations. In the test load method, the compressibility parameter is computed locally and is based on the mesh deformation of the fluid domain. The functionality of the ACM/FSI coupling with the test load method is demonstrated in an arterial flow simulation, and the combination is shown to provide a robust convergence. In order to get the test cases to correspond better to human physiology, one-dimensional FSI models are combined with the higher dimensional test models.

Efficient Parallel 3-D Computation of Electrical Machines With Elmer

After its recent improvements described here, open source finite element software Elmer is shown to be a highly efficient option for electrical machine modeling. The parallelization of computational burden is shown to be a necessity. The methods implemented enable applying fully parallel computation to electrical machine models, including rotation and electrical circuits. Computational experiments performed demonstrate that Elmer can effectively utilize several hundreds of computational cores in parallel, making it an attractive alternative when computational speed is of high importance.

Multi-slice 2.5D modelling and validation of skewed electrical machines using open-source tools

This paper describes multi-slice modelling and validation of an axially skewed synchronous machine in 2D domain. The technique involves coupling of circuit and electromagnetic domains together with carefully constructed geometry. The model size in comparison with full 3D simulation reduces considerably. Developed 2.5D multi-slice model allows fast simulation in 2D domain, still taking into account 3D effect of skew on torque and back-EMF of the machine. Simulation results in 2.5D and 3D domains are compared to measurements. All simulations are performed using free and open-source tools.

Parallel Performance of Multi-Slice Finite-Element Modeling of Skewed Electrical Machines

The multi-slice method allows approximation of the 3-D phenomena without carrying out a full 3-D analysis, e.g., in skewed radial flux electrical machines. The idea is to divide a 3-D machine into several 2-D finite-element models along the axis, connected by electrical circuits. Here we show how the multi-slice method is perfect for parallel computation; the computation efficiency is close to that of 2-D models in modern parallel hardware. The results are shown to agree with 3-D computation and experimental results.

Parallel performance of multi-slice method for skewed electrical machines

Multi-slice methods allow us to approximate the 3D phenomena without carrying out a full 3D analysis, e.g. in skewed radial flux electrical machines. The idea is to divide a 3D machine into several 2D FEM models, only connected by electrical circuits. Here we show how the multi-slice method is perfect for parallel computation; the computation efficiency is close to that of 2D models in modern parallel hardware. The results are shown to match with 3D computation and experimental results.

Applied Parallel Computing

We study the electrical field in the human body, generated by the ventricular muscle, by means of numerical simulations. The involved mathematical model consists of two partial differential equations (PDEs) that are also coupled with a system of ordinary differential equations (ODEs). Following the strategy of operator-splitting, we have devised an efficient numerical algorithm that carries out a simulation stepwise in time. At every time level, the ODE system is solved before a parabolic PDE, and then an elliptic PDE. The main focus of this paper is on the transformation of an existing sequential simulator into a parallel simulator that runs on multiprocessor platforms. Two important numerical ingredients used in the resulting parallel simulator are overlapping domain decomposition and multigrid, which together ensure good numerical efficiency. We also explain how object-oriented programming techniques enable the software parallelization in a simple and structured manner. In addition, we study the performance of the parallel simulator on different multiprocessor platforms.