Development of a dual serial-parallel multiphase CFD code for application to industrial combustor/reactor systems. Page: 6 of 9
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
The following text was automatically extracted from the image on this page using optical character recognition software:
be very difficult even for optical techniques. Combustors and FCC risers are two examples of such environments.
CFD analysis, then becomes an important tool to augment and make sense of what data is available. These systems
generally require a lot of computational resources to resolve flow structures and local processes in different regions
of the system with size scales that may range over several orders of magnitude. The use of parallel processing can
facilitate multiphase process and local flow structure resolution in large systems within a reasonable amount of
computing time. The CFD codes can then be used for a variety of tasks including: optimization of system operation,
system problem isolation, system management, analysis of system design improvements for retrofit, and investigation
of design alternatives for new systems.
A large scale up of problem size that involves moving to grids that may use an order or two more computational
points is done with spatial decomposition of the computational domain for all three material phases in a
gas/droplet/particulate or liquid/bubble/particulate flow system. The grid partitions may then be assigned to tens or
hundreds of processors in a large parallel computer or distributed computing environment. Under these
circumstances the computation for each phase can be assigned to a large set of processors to gain the advantage of the
natural partitioning of the problem between the material phases.
The problem size of multiphase reacting flow problems is not solely a function of the need for spatial resolution
of hydrodynamic flow features. The problem size for these types of problems is also a function of the number of
chemical reactions included in the reaction model, and the kind of mathematical modeling applied to the discrete
phases (particles, droplets, or bubbles). The chemical kinetics and multiphase formulation dramatically increases the
number of coupled partial differential equations (PDEs) to solve beyond the few that govern the hydrodynamics. In
an Eulerian formulation for the discrete phases, the size spectrum of droplets, bubbles, or particles is discretized into
size groups, and the conservation laws of mass, momentum, and energy yield a set of PDEs for each size group. In
some cases studies may be done using mono-sized particles, however, for vaporization or condensation, droplets or
bubbles grow or shrink in size creating an evolving size spectrum . The resolution of this size spectrum is
determined by the number of size groups. The momentum equation is a vector equation with three component PDEs
in three dimensional space. Therefore, there are 5 PDEs to solve for each size group. Using as few as five to ten size
groups for the discrete phases in a three phase problem leads to 50 to 100 PDEs for the basic conservation laws of the
discrete phases. Chemical reaction also greatly adds to the number of PDEs to solve. For each chemical species in
the reaction set, one PDE transport equation governing chemical specie concentration with reaction source and sink
terms is added. Often a highly reduced chemical kinetics model is used to limit the number of these chemical species
PDEs, but the number can range from 10 to 1000 or more depending on the level of detail needed in the chemistry
model. On a serial computer with 128 to 768 megabytes of memory, memory size limits the problem size to a range
of approximately 20,000 to 100,000 spatial grid nodes and 50 to 100 coupled nonlinear PDEs. Problems that
consume the maximum amount of memory available on a serial machine may take days to reach a converged solution
if convergence is achieved. Note that a 30x30x100 grid is 90,000 nodes.
APPROACHES TO PARALLELIZATION
Two primarily different approaches to parallelization were investigated. An important goal in both approaches
was to provide a means to do continuing development of the CFD code on single processor personal computers or
workstations. Development work on these single processor platforms can be done very rapidly at relatively low cost,
and therefore preserving this capability was considered to be very important. In addition some engineering level
applications can be run on these inexpensive and nearly universally available platforms, so preserving the capability
to run the CFD code in a production mode for parametric studies on a fast personal computer was also considered to
be necessary. The first approach to parallelization was to use an expert system that allows user interaction to setup a
process for automatically converting a single processor CFD code to a parallel version. CAPTools (Computer Aided
Parallelization Tools) developed at the University of Greenwich [91 was chosen as the software package to use for
trying to automate the conversion process. After extensive testing, CAPTools was judged to be of benefit primarily
for the one time conversion of legacy serial codes in porting them from serial to parallel computer platforms. Using
CAPTools, as contemplated, to frequently convert a complex, serial, multiphase CFD code undergoing continuous
development to a parallel code for parametric studies on a highly parallel computer was judged to require too much
work each time the code needed to be reconverted to parallel (possibly monthly or more).
The second approach to parallelization was to use a sequestered set of inter-process communication subroutines
to implement parallel processing capability directly in the CFD code. The communication subroutines are set up to
handle any number of processors including the degenerate case of one processor. The MPI set of communication
functions was chosen to implement the communications . MPI is an industry standard portable library of inter-
processor data communication and control functions for programming parallel algorithms. Nearly all MPI function
calls are contained within the sequestered communication subroutines. A set of simple tie-off functions was
developed to provide the correct trivial return values when the CFD code is in use on a personal computer or
Here’s what’s next.
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Article.
Lottes, S. A.; Fischer, P. F. & Chang, S. L. Development of a dual serial-parallel multiphase CFD code for application to industrial combustor/reactor systems., article, May 16, 2000; Illinois. (https://digital.library.unt.edu/ark:/67531/metadc712529/m1/6/: accessed May 23, 2019), University of North Texas Libraries, Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.