Yanbo Wang
2007
Computational fluid dynamics (CFD) based Fire Field Modelling (FFM) codes
offer powerful tools for fire safety engineers but their operation requires a
high level of skill and an understanding of the mode of operation and
limitations, in order to obtain meaningful results in complex scenarios. This
problem is compounded by the fact that many FFM cases are barely stable and poor
quality set-up can lead to solution failure. There are considerable dangers of
misuse of FFM techniques if they are used without adequate knowledge of both the
underlying fire science and the associated numerical modelling. CFD modelling
can be difficult to set up effectively since there are a number of potential
problems: it is not always clear what controls are needed for optimal solution
performance, typically there will be no optimal static set of controls for the
whole solution period to cover all stages of a complex simulation, there is the
generic problem of requiring a high quality mesh – which cannot usually be
ascertained until the mesh is actually used for the particular simulation for
which it is intended and there are potential handling issues, e.g. for
transitional events (and extremes of physical behaviour) which are likely to
break the solution process.
In order to tackle these key problems, the research described in this thesis has
identified and investigated a methodology for analysing, applying and automating
a CFD Expert user’s knowledge to support various stages of the simulation
process – including the key stages of creating a mesh and performing the
simulation. This research has also indicated an approach for the control of a
FFM CFD simulation which is analogous to the way that a FFM CFD Expert would
approach the modelling of a previously unseen scenario. These investigations
have led to the identification of a set of requirements and appropriate
knowledge which have been instantiated as the, so called, Experiment Engine
(EE). This prototype component (which has been built and tested within the
SMARTFIRE FFM environment) is capable, both of emulating an Expert users’
ability to produce a high quality and appropriate mesh for arbitrary scenarios,
and is also able to automatically adjust a key control factor of the solution
process.
This research has demonstrated that it is possible to emulate an Experts’
ability to analyse a series of simulation trials (starting from a simplified,
coarse mesh test run) in order to improve subsequent modelling attempts and to
improve the scenario specification and/or meshing solution in order to allow the
software to recover from a complete solution failure. The research has also
shown that it is possible to emulate an Expert user’s ability to provide
continual run-time control of a simulation and to provide significant benefits
in terms of performance, overall reliability and accuracy of the results.
The instantiation and testing of the Experiment Engine concept, on a chosen FFM
environment – SMARTFIRE, has demonstrated significant performance and stability
gains when compared to non Experiment Engine controlled simulations, for a range
of complex “real world” fire scenarios. Preliminary tests have shown that the
Experiment Engine controlled simulation was generally able to finish the
simulations successfully without experiencing any difficulty, even for very
complex scenarios, and that the run-time solution control adjustments, made to
the time step size by both the Experiment Engine and by the Expert, showed
similar trends and responses in reacting to the physical and/or numerical
changes in the solution. This was also noticed for transitional events seen
during the simulation. It has also been shown that the Experiment Engine (EE)
controlled simulation demonstrates a saving of up to 40% of simulation sweeps
for complex fire scenarios when compared with non-EE controlled simulations.
Analysis of the results has demonstrated that the control technique, deployed by
the EE, have no significant impact on the final solution results – hence, the
Experiment Engine controlled simulations are able to produce physically sound
results, which are almost identical to Expert controlled simulations.
The research has investigated a number of new methods and algorithms (e.g. case
categorisation, case recognition, block-wise mesh justification, local adaptive
mesh refinements, etc.) that are combined into a novel approach to enhance the
robustness, efficiency and the ease-of-use of the existing FFM software package.
The instantiation of these methods as a prototype control system (within the
target FFM environment – SMARTFIRE) has enhanced the software with a valuable
tool-set and arguably will make the FFM techniques more accessible and reliable
for novice users.
The component based design and implementation of the Experiment Engine has
proved to be highly robust and flexible. The Experiment Engine (EE) provides a
bi-directional communication channel between the existing SMARTFIRE Case
Specification Environment and the solution module (the CFD Engine). These key
components can now communicate directly via status- and control- messages. In
this way, it is possible to maintain the original Case Specification Environment
and the CFD Engine processes completely independently. The two components
interact with each other when the EE is operating. This componentization has
enabled rapid prototyping and implementation of new development requirements (as
well as the integration of other support techniques) as they have been
identified.