EUROTEX Workshop on Internet- and Web-based Computing

Tuesday April 13th, 1999, Crowne Plaza Hotel, Dallas Market Center, Texas

Computational Fluid Dynamics on the Internet

Brian Spalding, of CHAM Ltd

Summary of the argument

The nature of CFD

CFD, ie Computational Fluid Dynamics, is a technique for predicting quantitatively:

• the performance of equipment, (e.g. automobiles, gasoline engines or electric motors) and
• the behaviour of the:
  • man-made (e.g. buildings, towns, bridges, aqueducts, harbours), and
  • natural (e.g. winds, rivers, forest fires) environments,
• in circumstances involving:
  • the flow of fluids (e.g. steam , combustion gases, blood, oil) ,
  • the transfer of heat (by conduction, convection or radiation),
  • associated phase changes (e.g. melting, vaporisation, dissolution), and
  • other concomitant effects (e.g. thermal stresses, chemical reactions, the generation of noise).

CFD requires, if its predictions are to be reliable:

1. user-friendly software which embodies the best that is known about the laws of physics and chemistry;
2. very-high-performance computer hardware;
3. high-level human expertise.

History

General-purpose CFD software first became commercially available in 1981, when CHAM's PHOENICS code was first released. Creare's FLUENT code followed two years later; and by now there must be a dozen or more packages offering similar capabilities.

From its beginnings in "high-tech" industry (e.g. nuclear power and aerospace), the use of CFD has spread widely into more-traditional sectors (e.g. metallurgical and food processing) and to some extent into lower-technology professions such as architecture and building services.

Obstacles to the wider use of CFD

Nevertheless, three obstacles stand in the way of its use by all who could benefit from it (e.g. town planners, surgeons), namely:-

1. the cost of purchasing or leasing the software;
2. the cost of the high-powered computer hardware; and
3. the scarcity of adequately-trained and experienced personel.

A way around the obstacles

Use of the Internet provides, potentially, a way around the obstacles; for it allows the software and hardware to be used remotely, on a pay-by-use basis; and, in principle, appropriate human expertise can be obtained from anywhere in the world.

Once such a service has been made available, it is to be expected that the number of persons making use of CFD will increase rapidly, with great economic and social advantage, for it saves them the expenses of:

• employing specialised staff;
• purchasing expensive software;
• purchasing and maintaining expensive hardware;

CHAM has been engaged for several years in developing such a service, by way of:

• the EC-supported MICA project, now completed;
• the EC-supported ADELFI project, now starting; and
• the commercial Simuserve operation, which will start to offer its services generally before the end of 1999.

The presentation will describe these developments.

Contents

1. The Internet opportunity for CFD
2. The MICA project
3. The ADELFI project
4. Simuserve
5. Some helpful CFD developments
6. Questions still to be answered

1. The Internet opportunity for CFD

• In earlier centuries, domestic water was drawn from wells in buckets; nowadays "piped water" is taken for granted.

• Similarly, light was generated locally by candles or oil lamps; now we simply click a switch.

• Before the creation of modern health services, whatever persons and implements were near at hand were use in dealing with
  • broken legs,
  • child-birth and
  • bullet wounds.
  Most of us nowadays receive the appropriate attentions in hospitals.

• Do you remember (we may be saying before long) how, in the early years of CFD,users had to create or buy expensive software, use it on their own never-powerful-enough computers, and undergo training so as not to do too much damage. Fortunately nowadays Internet enables us to treat CFD like any other utility.

• CFD-code vendors can be likened to sellers of buckets; and hardware vendors to the diggers of wells; the user turns the handle. Together they satisfy currently-admitted needs; but:
  • the buckets seldom have the required capacity (the software lacks desired features);
  • the well rarely has enough water (the computer cannot handle the necessary fine grid); and
  • turning the handle needs too much strength (the user lacks training).

• CFD-code vendors can also be likened to sellers of do-it-yourself surgical instruments. With a knife like this, they say, Christian Barnard performed the first heart transplant. The purchaser is tacitly encouraged to believe that he can do a similar operation himself, indeed on himself.

1.3 The reliability of today's CFD

The truth is that almost all currently-made CFD predictions are unreliable, because:

1. the grid used is too coarse for numerical accuracy (being limited by the memory and speed of the computer):

2. the physical models built into the software (e.g. for turbulence or chemical reaction) are not the best known to science; and

3. the user has had too little time to become an expert CFD-code user as well as being one in his/her own profession.

It is common to allude to the tendency of CFD users to believe in the realism of the beautiful graphical displays which their software produces by saying that CFD is really an acronym for "Colourful Fluid Dynamics".

Harsher critics say that "Cheats, Frauds and Deceivers" would be nearer the mark.

Proof that this suggestion is not wholly unfair is provided by the results of the recent EC-supported EMU Project, in which four different users of the same widely-used computer code produced predictions for comparison with some carefully-conducted experiments.

The differences between their predictions were very much larger than the differences of between all of them and the experiments.

Yet the geometry (a building made up of rectangular blocks) and the physical phenomena (steady turbulent flow of a uniform-density gas mixture) were among the simplest which CFD ever has to deal with!

1.4 Internet to the rescue!

It is therefore no exaggeration to state that the CFD-over-the-Internet facility is arriving in the nick of time, because:

• on the one hand, engineers and environmentalists have started to believe seriously in CFD as a predictive tool; and

• on the other hand, experience is showing how dangerously misleading can be the predictions when:
  • the grid is too coarse,
  • the physical models are inadequate,
  • the user of the software lacks, inevitably, sufficient expertise.

• moreover it is increasingly recognised that, constrained by lack of resources, CFD practitioners have been using no-longer-acceptable short-cuts, for example:
  • treating as steady the essentially unsteady turbomachinery flows;
  • presuming (often wrongly) the shapes of the probability-density functions of a turbulent flow rather than, as is now possible, calculating them;
  • neglecting totally the turbulent "sifting" process whereby faster-moving fluid fragments are "centrifuged outwards in a rotating flow field".

There are thus two benefits which Internet has to offer; and it is important that it does indeed provide both, namely:

1. the lowering of costs, which will greatly enlarge the number of CFD users; and
2. the raising of quality, by the provision of sufficient computer power and of the best possible human advice.

2. The MICA project

The project, conceived and coordinated by CHAM, involved fourteen partners (mostly industrial but some academic) from nine European countries. It started at the beginning of 1996 and finished at the end of April 1998.

Its aim, which was successfully achieved, was to show that serious CFD calculations could be set up, and their results inspected, on PCs equipped only with "front-end" software of non-immersive "virtual-reality" character, while the computations were performed remotely.

A full description can be found on CHAM's website, www.cham.co.uk, the relevant content of which can be seen by clicking HERE.

3. The ADELFI project

The aim of the ADELFI project, which is again EC-supported, and conceived and coordinated by CHAM, is similar to that of MICA, but with two significant differences, namely:

1. data-mining is included as an application type of equal weight with CFD; and

2. the only software which the user will need is a JAVA-enabled browser.

The project started at the beginning of December 1998; a report of the "kick-off" meeting can be found on CHAM's web-site by clicking HERE.

The first step has been, while still adhering to the MICA concept in which the user has "front-end" software on his own machine, to enable users to interact with, and to some extent control, the execution of the CFD package on the remote machine.

It is intended to make a "live" demonstration of this capability at the end of the workshop lecture.

4. Simuserve

It is still in the formative stage. At the present time, it employs only CHAM's proprietary software for communication and CFD simulation, but the plan is to enlarge the provision so as to enable customers of Simuserve to access "best-of-class" software, hardware and human expertise, wherever in the world they are to be found.

Some information about Simuserve can be found on CHAM's web-site, by clicking HERE.

However, the situation changes rapidly; the period between now and June 16, 1999, which is when submissions to the European Commission's "Fifth Framework Programme" for Information Society Technology are due, may see some significant re-groupings.

It may perhaps interest non-European participants in the EUROTEX Workshop to know that, under the IMS (i.e. Intelligent Manufacturing Systems) section of this programme, North-American partners are sought (and may receive funding, but from their own governments) for projects submitted to EC.

5. Some helpful CFD developments

To them, therefore, it seems proper to point out that (in the author's opinion) several recent CFD-technique developments make it both more necessary and more easy to move to Internet-based computing.

There will no time during the lecture to discuss them in detail; but they can be listed; and more about them can be learned from CHAM's web-site. They are:

• The use of Virtual-Reality user-interfaces makes problem-setup and results-interpretation easier than ever before.

• It has become effortless to turn a CAD-generated solid model into the geometrical starting point of a CFD simulation.

• Domain-decomposition has provided a flexible means of "parallelising" CFD codes, and so enabling them to use massively-parallel machines and work-station or PC clusters.

• The use of the "cut-cell" and "fine-grid-embedding" techniques, which can be easily automated, have greatly reduced the necessity for using body-fitted-coordinate grids.

• It has been shown that solid-stress and fluid-flow analyses can be performed simultaneously by a single computer code, indeed a single algorithm.

• Advances in turbulence modelling, and specifically the multi-fluid turbulence model, are providing new opportunities to simulate industrially important processses realistically.

The conventional method of new-ideas dissemination is this:

• someone innovates;
• others copy, sometimes making improvements but sometimes the reverse;
• all contend that their variants are best;
• some lowest-common denominator version becomes popular, because it is easiet to describe.

All this takes many years; but an Internet-based service will enable the best innovations to be made available to all immediately.

6. Questions still to be answered

• What will be the constituents and configuration of that which ultimately proves to be the winning (i.e. most world-pleasing) business model?

• Since this will almost certainly require significant financial investment, if it is is to be brought quickly into being, how will investors be persuaded that they will earn adequately from what may seem to them a highly risky (because unprecedented) enterprise?

• Where will one find the people to manage an enterprise combining such diverse elements, and necessarily involving specialists of many different kinds?

• How long will it take, once a reliable and cost-effective service has been brought into existence, for customers to avail themselves of it sufficiently to make it an economic success?

However CFD predictions are not like books; for each one is the unique product of:-

• the poser of the flow-simulation question (who may over-look important elements);
• the software which interprets it, as well as it can;
• the not-always-perfect science which is built into the software;
• the numerics which always introduce some discretization error;
• the hardware and its limitations;
• the human adviser, of whom the very best is fallible;
• the constraints of time and money which pertain to the particular persons and organizations involved.

That the service will come into being, the present author feels convinced. But how?

Most probably by some combination of the respective strengths of North American business and European inventiveness.