PHD Position High Performance Parallel Computing for complex and industrial multiphase flows
Computational modelling has attracted increasing interest in recent years to assess the impact of multiphase flow phenomena arising from industrial heat treatment processes. Indeed these processes in particular quenching, involve the flow of two fluids (water and vapor) separated by sharp interfaces around rigid solids. Numerically modelling the transport of these interfaces is a challenging task and a range of methods have been developed over the last decades. The Level Set method is a widely used approach that represents geometrically the interface between two fluids using a mixture law on fluid properties. It is this geometric interface that is advected with the flowing fluid. The main difficulty of the LevelSet method is, to conserve the sharp interface during fluid motion. An additional reinitialization step is then required to correct the loss of mass.
Although the LevelSet method is not necessarily CPU intensive, it imposes a further restriction on the solution of the fluid flow equations. The flow is almost exclusively unsteady and the time step generally needs to be significantly reduced by comparison with conventional fluid flow problems. This is particularly the case when surface tension must be taken into account. The sharp changes in the fluid properties at the interface also lead to consider anisotropic mesh adaptation methods to capture accurately the properties and solution variations. Unfortunately the resulted algebraic systems are generally asymmetric and ill-conditioned and need specific preconditioning techniques to overcome the convergence slowing.
Recently, there has been growing interest in hybrid architectures. Two solutions have received special attention from the high-performance computing community: GPUs, originally developed to render graphics, which are now used for very demanding computational tasks, and the Intel Xeon Phi, which employs very wide (512 bit) single instruction, multiple data (SIMD) vectors on the same X86 architecture as other Intel CPUs. Relative to CPUs, the faster growth curves of these hybrid machines in the speed and power efficiency have spawned a new area of development in computational technology. Now, many of the top supercomputers are equipped with such chips. Developing efficient multiphase codes for these hybrid architectures is a very important step toward fully utilizing the power of such supercomputers.
The objective of the PhD project is to develop a high performance computing model for solving multi- phase problems into such hybrid machines. The LevelSet advection equation is coupled to the Navier- Stokes equations and mesh adaptation techniques to simulate accurately the motion of fluids. The research focuses both on modelling moving interfaces and on improving the computational efficiency of the iterative methods used to solve the system of algebraic equations. In this project, we investigate the state-of-the-art parallel algorithms for solving efficiently the Levelset transport equation on unstructured meshes on both CPU-based and GPU-based parallel processing systems. We shall pay a particular attention to the coupling with parallel mesh adaptation as well.
Keywords: Computational Fluid Dynamics; High Performance Computing; Finite elements; Unstructured meshes; Mesh adaptation; Dynamic load balancing; C++.
R. Chiodi, O. Desjardins A reformulation of the conservative level set reinitialization equation for accurate and robust simulation of complex multiphase flows, JCP, 343, pp. 186-200, 2017
M Khalloufi, Y Mesri, R Valette, E Massoni, E Hachem High fidelity anisotropic adaptive variational multiscale method for multiphase flows with surface tension- CMAME, 307, pp. 44-67, 2016
Y Mesri, W Zerguine, H Digonnet, L Silva, T Coupez, Dynamic parallel adaption for three dimensional unstructured meshes: Application to interface tracking, IMR 17, 195-212, 2008
All numerical developments will be undertaken in the collaborative finite elements, C++ CIMlib_CFD Library, developed at CEMEF by the Computing and Fluids (CFL) research team. The candidate will work on a new HPC cluster with more than 2000 cores based on the newest technology and will have access to the national GENCI supercomputers.
Required skills and qualifications
Degree: MSc in Scientific Computing or Applied Mathematics, with excellent academic records.
Skills: Finite Element Method; Meshing; C++; HPC; ability to work within a multi-disciplinary team.
The 3-year PhD will take place at CEMEF, an internationally-recognized research laboratory of MINES ParisTech located in Sophia-Antipolis, on the French Riviera. It offers a dynamic research environment, exhaustive training opportunities and a strong connection with the industry. Annual gross salary: around 26k€. The PhD candidate will join the CFL team under the supervision of Y. MESRI.
The thesis is a part of an industrial chair program called INFINITY that gathers 10 industrial partners: ARCELORMITTAL, AREVA NP, AUBERT & DUVAL, CEFIVAL, CMI, FAURECIA, INDUSTEEL, LISI AEROSPACE, MONTUPET, SAFRAN. This chair contributes to a long-term vision of high fidelity numerical tool as a basis for reliable simulations of quenching processes, to facilitate faster decision making for delivering high quality parts while minimizing residual stresses, preventing cracking and thus optimizing heat processes.