Font Size:
GPU-based numerical solution of thermal multiphase flow in porous media
Last modified: 2016-05-24
Abstract
In this work, a numerical thermal multiphase flow in porous media is studied. We considera porous media filled with three phases (water, oil and gas) and steam injection to investigatethe effects of temperature rise on viscosity reduction. The mathematical model is based on themass conservation equations for water, oil and gas, and an equation for the energy is consideredto account for heat transport. These conditions give us a system of four non-linear andcoupled partial differential equations that need to be solved using some linearization strategy.The finite volume technique is used to develop a numerical model and to linearize the equationswe apply the Newton-Raphson method. The linear systems generated by the combination ofboth numerical techniques are large and difficult to solve. Some problems arise in the Jacobianconstruction, which takes considerable time to be generated. Also, the solution of the systemsis expensive and requires a large amount of time for high number of unknowns. For this reason,it is necessary to apply new technologies to implement solution algorithms in parallel, sinceaccelerating the computation time allows us to save time. Within these new technologies is thecomputation using Graphics Processing Units (GPU’s) which allows to construct massively parallelnumerical algorithms. The numerical solution of the systems was implemented in paralleland the strategy was to reduce the amount of information that needs to be exchanged betweenCPU and GPU memories. We used libraries that already include GPU Krylov methods. Also,we study the possibility of to construct the Jacobian directly on the GPU, which could reducethe communication information. Numerical results indicate very good acceleration using theGPU parallelization for our steam injection numerical code and we present some examples forregular domains and structured meshes.
An account with this site is required in order to view papers. Click here to create an account.