Questions/comments following the presentation + reactions:

1. What are your comments to the simulation tool comparison made
by Annex 21, where ESP-r was one of the programs out of the
confidence band. (Schuler)

3 points to be made.  Firstly, the IEA work - both empirical and
BESTTEST - was based on ESP-r version 6.18a (released May 1989) which
had been superseded by several new releases by the time of the IEA
study.  ESRU has no idea why the latest version was not used in the IEA
work.  Secondly, we believe that some technical aspects of the IEA
study are open to question - e.g. why the ESP-r simulations were based
on time independent thermophysical parameters when it is known that
this can be an important parameter in test cells or why the system was
operated with a convective regime which was known to be significantly
lower that those specified to other models.  It is possible to conclude
that some of the programs that performed well did so because they were
fortuitously configured, while some of the more sophisticated programs
behaved less well because they were badly configured.  This raises the
question on whether the IEA work intended to establish the effect of
the types of assumptions typical program uses might make or,
alternatively the intention was to test the predictive robustness of
the programs when expertly configured.  Thirdly, what the modelling
community and the profession at large need is a constructive approach
to program testing and refinement, not an approach which can undermine
many years of R&D in a short period of time (as may yet happen in the
UK).

2. I believe the IEA21C runs using ESP-r were made by a third
party so perhaps that is a reason for some problems with the
predictions. This raises the question of the suitability of the
present interface for third party use. Surely validation must
encompass both user and software issues. (Holmes)

We agree that "user and software issues" are interlinked but cannot see
the point in keeping them together in a scientific approach to program
proving.  What is needed is a methodology to test programs in terms of
their predictive accuracy (as developed within the PASSYS project) and
a means to introduce quality assurance and modelling know-how at the
interface.  Our view is that building simulation is a) a non-trivial
task and b) a technology which is still in its infancy.  The production
of interfaces which are truly suitable for non-specialist use is still
some way off and will only come about when technologies such as our own
intelligent front end mature and the profession becomes more involved
in the interface definition process (e.g. through organisations such as
BEPAC and IBPSA).  With respect to ESP-r's interface, we believe that
with the release of our new project manager we now have a powerful
interface of a type which will grow in popularity as more PCs become
multi-tasking.

3. Are the CFD results fed back to the program (and used)? Are
the convection coefficients modified, presuming they are still
used? Analogously, are the pressures at the interface between the
CFD domain and the "Macro" domain adjusted? Iteration between the
CFD model and the building model will likely make the
computational time requirement prohibitive. (Spittler)

The implementation of CFD within ESP-r has been enabled at two levels
of granularity in order to allow our researchers to explore the
conflation issue.  At the first level, the CFD domain is essentially
decoupled with the building/ plant solver (with fully integrated
network air flow) passing the internal surface temperatures to the CFD
solver, which passes back the surface convection coefficients.  At the
second level, one or more air flow network nodes are replaced by the
CFD domain and the conservations equations describing the entire
problem solved simultaneously with full pressure adjustments taking
place at each iteration.  Papers describing the approaches are
available from ESRU.

We agree that the approach is computationally intensive but then ESP-r
has always been a research environment and we have observed over the
years how technology often acts to make the complex routine.  When
problematic convergence does occur within ESP-r, the current rule is to
discarded the CFD domain and reinstated the previously removed network
nodes so that the user is always given a best estimate at reasonable
CPU expense.

4. I believe it is much too early to integrate CFD inside the
whole system for a "global" simulation. Even for research, I
would encourage people to deal with the items separately in order
to understand better what they are doing. (Lebrun)

Don't agree.  As we see it most developments, explicitly or implicitly,
are striving for integration - e.g. COMBINE, STEP, NMF, WINDOWS NT,
etc.  The real issue is to ensure that the different elements of the
overall problem - heat transfer theories, plant components, numerical
solvers, air movement, air quality, algorithm validation, user
interface, etc. - do not become inappropriately mixed.  This is a
software development issue and one we are very aware of within the
ESP-r project.

5. Do you have a view on why (or on what basis you could make a
comparison) your "unified" approach is better than coupling or
combining separate simulation methodologies. (Hanby)

We believe that it is better to couple separate modelling systems and
to this end are developing an Intelligent, Integrated Building Design
System (IIBDS) for COMBINE.  This allows any number of programs to
interact and be sequenced in terms of a process model which defines the
purpose of program use.  The IIBDS is currently sequence programs such
as AUTOCAD, RADIANCE, TSBI3, ESP-r and regulation compliance software
from BRE.  However, when it comes to the simulation of thermodynamic
systems, integration at the theoretical level is essential and
decoupling dangerous.  Of course, there are many ways to achieve this
integration of which the modular sequential approach of TRNSYS and the
modular simultaneous approach of ESP-r are but two.

6. I think that many potential users would be happy to play with
one component alone and see how it behaves (..... in steady-state
conditions). Does ESP offer such possibility? (Lebrun)

Yes, although there is a minimum problem resolution.  The current
version of ESP-r will process from one to many building zones, from one
to many plant components, from one to many flow networks, from one to
many control loops, from one to many CFD domains, from one to many
climates, etc.  It will also process any combination of these under
dynamic or steady-state conditions.