A FORTRAN code was developed to numerically simulate the mixed convective flow over a three-dimensional horizontal backward-facing step. The momentum and energy equations under the assumption of the Boussinesq approximation were
discretized by means of a finite volume technique. The SIMPLE algorithm scheme was applied to link the pressure and velocity fields inside the domain while an OpenMP
parallel implementation was proposed to improve the numerical performance and to accelerate the numerical solution. The heating process corresponds to a channel heated from below at constant temperature keeping insulated all the other channel walls. In addition, the back-step was considered as a thermally conducting block and its influence in the heating process was explored by holding different solid to fluid thermal conductivity ratios. The effects over the velocity and temperature distribution of buoyancy forces, acting perpendicular to the mainstream flow, are studied for three different Richardson numbers Ri=3, 2, and 1 and the results are compared against those of pure forced convection Ri=0. In these simulations the Reynolds number is fixed at 200 while the bottom wall temperature is adjusted to fulfill the conditions for the different Ri. Under this assumption, as Ri increases the buoyancy effects are the dominant effects in the mixed convective process.
The numerical results indicate that the velocity field and the temperature distribution for pure forced convection are highly distorted if compared with the mixed
convective flow. If the Ri parameter is increased, then the primary re-circulation zone is reduced. Similarly, as the buoyancy forces become predominant in the flow, the
convective rolls, in the form of spiral-flow structures, become curlier and then higher velocity components are found inside the domain. The temperature field distribution showed that as the Ri is increased a thicker layer of high temperature flow is located at the channel??s top wall as a result of the higher rates of low-density flow moving to the top wall. The flow is ascending by the channel sidewalls, while descending by the channel span-wise central plane. The parallel numerical strategy is presented and some results for the performance of the OpenMP implementation are included. In this sense, linear speedup was obtained when using 16 possessors in parallel.
- Anand, N. K. Executive Associate Dean