The model equations were solved in a transient mode in the CFD
solver FLUENT. As the flow is incompressible, a pressure-based segregated
solver employing a modified SIMPLE algorithm known as Phase-
Coupled SIMPLE (PC-SIMPLE) was used. Since the flow calculation of
multiphase flows is susceptible to divergence, and since spurious
oscillations/under- and over-shoots of flow variables may be produced
by high order differencing schemes, all advection terms were treated
with first-order upwind differencing while the diffusion-like terms
were treatedwith central differencing. A fully first-order implicit scheme
was applied for time marching; although this scheme is expected to be
unconditionally stable in its linear limit, a small time step size of 10−4 s
was used to avoid any divergence. The model equations along with the
boundary and initial conditions were solved on a server on a 64-bit
LINUX platform. The server had two quad AMD processors with a clock
speed of 2.17 GHz each. All the 8 cores were utilized simultaneously in
a parallel computing mode. Considering the computational expense,
each run was monitored for steady state. It is found that on an average
it took 20–25 s of real time for the solution to reach steady state. A typical
simulation run took 10–20 days. Due to the large computational time, it
has not been possible to conduct a systematic grid independence study.
In order to examine the sensitivity of the present results to grid size, simulations
were carried out for one case using 25% more cells, i.e., about
0.815 million cells instead of 0.651 million cells, with selective refinement
in the riser section which is where there is a lot of recirculation.
The predicted values of important flow variables such as the solids holdup
proved to be nearly identicalwith those of the coarser grid.While this
cannot be used to formally claim grid independence—this would require
several-fold (not just 25%) increase in the grid density—, the results show
a measure of grid insensitivity of the calculated flow variables. Similarly,
although second-order accurate central differencing was used for the
discretization of the diffusion terms, only first-order accurate upwind
schemes were used for the advective terms; use of a second-order accurate
scheme led to the divergence of the iterative calculation procedure.
Due to the current non-availability of sufficient experimental local
and in situ data in the literature, no effort has been made to fine tune
the constants in the variousmodels. The very fact that a converged solution
has been obtained for amultiphase,multiphysics problemin a fairly
complicated three-dimensional geometry can itself be considered as a
matter of achievement. The focus of the present work is therefore on
the robustness of the calculation procedure (so that a numerical solution
is obtained) and on its capability to take account of the physics of the
phenomena. To this end, a number of runs with different primary and
secondary liquid velocities and solids inventories were done as summarized
in Table 1. All of these cases were simulated with the same set of
model constants; only the liquid velocities at the inlets and the initial
solids height in the downcomer were changed as appropriate for each
case. The results from these simulations are discussed below with
specific focus on the reasonable correctness of the computed solution.