Your readership survey forms have flooded back, resulting in a much greater response than anticipated. However, late arrivals are still welcome and will be included in the results, so if yours has not yet been posted, please spend that five minutes to complete and return it. Thank you.
I welcome any contributions you may have for the next issue, to be published in the New Year. Just the time for a New Year Resolution - YOU write the articles which YOU want to appear!
... an Object Oriented Environment
CAD and CAE systems are becoming more and more a standard tool of the engineering profession. However, underlying the use of such systems are several complex issues as yet insufficiently resolved.
These issues - such as validity, applicability, integration, the HCI and the role of IKBS - are currently being pursued in research systems being developed at a number of centres. What remains to be tackled is the means by which such systems can efficiently be built and adapted in response to changing user needs and deficiencies exposed through use. One field in which the size and complexity of state-of-the-art programs/systems is hindering the development, maintenance and distribution of software and thereby stifling the dissemination of new ideas/ techniques, is that of computerised performance prediction of existing/proposed buildings. This is the domain of the Energy Kernel System. Though focused on building performance modelling, the EKS clearly has applicability in other areas of building performance assessment - such as lighting simulation and visualisation and across a range of engineering fields such as process and control engineering.
The EKS is intended as a software development environment offering a high degree of flexibility and facilitating program development, validation and maintenance. It will be capable of constructing a range of simulation program from a range of alternative computational techniques each addressing, at a different level, the underlying building physics. From the application viewpoint, the objectives are:
The EKS is an object-oriented software environment containing a set of class definitions corresponding to the building and thermodynamic domains. These classes can be considered as basic building blocks from which a wide range of modelling programs may be built. Each class, either alone or in conjunction with a few support classes, handles one particular aspect of the building performance prediction process. The classes break into three distinct groups:
This separation of functionality provides maximum flexibility and code reuse when creating new programs and facilitates validation and maintenance of the resulting software. Where different program architectures handle the same task in different ways, derived classes provide a means to incorporate that functionality while ensuring minimal impact on the rest of the system.
The classes are organised into an EKS taxonomy, which specifies how the classes interrelate and defines the information flow between them. The ever present dilemma between extensibility, ie the ability for any existing class to use or be used even by a newly created class, and security, ie the guarantee that the given classes are compatible is solved by using Metaclasses. A Metaclass is a class which knows about and can manipulate another class. Thus, by insisting that programs can only be built by using the Metaclasses, it is possible to have a extensible but secure system without paying the performance penalty of runtime type checking.
The program building is actually carried out by a separate Template class. Given the chosen program architecture, the Template builds up a collection of Metaclass instances which define the program structure. The Metaclasses can then check that the specified program is internally consistent. Furthermore, since this Template is in effect a specification for the program, there is no need to generate a executable program at this stage. Instead, once the problem specific data is available, the Template can use these Metaclasses to create a custom built program tailored to the problem being addressed. This is all achieved internally in the Template class; the program builder or user need know nothing about the Metaclass scheme.
Three categories of use of the EKS can be identified. Firstly, there is the the EKS software developer who creates, modifies or extends the classes. This is done by specifying the class behaviour to the EKS environment and installing the class function code in the EKS library. Secondly, there is the Program Developer who will use the EKS Template class to define the structure of a given program. This is done by specifying the classes to be used, and assuming they are acceptable, saving an instance of the Template (in lieu of an executable). Exploring alternative methods of handling, eg representing plant, modelling conduction or solving state equations, is simply a matter of replacing one Metaclass with another, the Template ensuring program consistency. Lastly, there is the Model User who models a building by producing (via some interface) a description 0 the building, running the EKS defined program, ie invoking the required Template with the necessary data and analysing the resultant output (again via some interface). This process should be indistinguishable from running a conventional program. While these categories basically reflect the anticipated EKS user "roles", actual users will not be restricted in any way, and may indeed assume complete control over their EKS at any time.
In its present form the core of the EKS is an Object Oriented Database (OODB). This plays three separate roles. Firstly, it holds entities such as climatic data and material properties encapsulated as persistent objects. Secondly, it can hold the problem description and results as objects, enabling the interfaces to be separated from the body of the performance prediction engine. Thirdly, it holds the Template and Metaclass objects used internally by the EKS environment. The chosen language for the EKS was C++, primarily because of its widespread acceptance and partly for efficiency reasons. However, one serious drawback is the lack of an inbuilt mechanism for examining or manipulating classes at runtime. This feature was provided by the OODB as part of its schema definition facilities.
While the EKS demonstrator currently approaching completion is targeted at building performance prediction, it is clear that most of the techniques employed, and hopefully a lot of the code, will be directly relevant to other engineering disciplines. The project, funded by the SERC Built Environment SubCommittee and the EASE Programme, is a collaborative venture between ESRU, University of Strathclyde, Department of Architecture, Newcastle University, Department of Architecture, Bath University and RAL. Further information about the system can be had from the above, or from me.
The Collaborative Computational Project looking at the application of novel architecture computing to engineering problems (CCP12) and the Computational Fluid Dynamics (CFD) Community Club held a workshop on Parallel Computing in CFD at SERC Daresbury Laboratory, Warrington on 13 May 1991. It was chaired by Mr S P Fiddes (Bristol) and was attended by over 80 people. The purpose of the meeting was to review some of the hardware and software aspects of parallel computers applied to problems in CFD, outline current research interests and discuss areas for collaboration in this rapidly developing field. Dr R J Blake (Daresbury) opened the meeting by reviewing the background to the establishment of CCP12 and the related Parallel Computing Hardware Initiative. Prof G P Hammond (Bath) then went on to describe the activities of the Community Club which concerns itself with the full range of computational issues of relevance to CFD.
Dr M J Guest (Daresbury) opened the scientific show with a tour-de-force overview of parallel processors in scientific and engineering computation. He described developments in the conventional range of shared memory: multiprocessor flagship supercomputers, mini-supercomputers and superworkstations, and reviewed the more scalable distributed memory systems which employ message passing or virtual shared memory environments. The peak performance of the new generation of single instruction multiple data systems was highlighted. The cost-performance, integration, usability and production capabilities of the various systems for scientific and engineering calculations were discussed.
The focus of the meeting then switched to software issues with a presentation from Prof P M Dew (Leeds) who outlined some general principles of parallel program design for CFD codes and illustrated the concepts with reference to the SPRINT + project. The goals of the project are to provide reliable numerical solutions to incompressible flow CFD problems using adaptive unstructured meshes, differential algebraic equation time integrators, exploiting parallelism and integrating visualisation. Portability is achieved by assuming a logically full interconnected distributed memory virtual network which can be realized on both shared and distributed memory machines. Work on developing scalable algorithms for the common vector and matrix operations found in CFD is under way and preliminary results indicated that grid scattering techniques are simple but effective for adaptive meshes.
Dr J Reeve (Southampton) went on to review work at Southampton aimed at developing a CASE tool for domain decomposition. The tool allows a graphical definition of the geometry followed by dissection into an unlimited number of subdomains and a configuration file to load the resulting processes onto a network of distributed memory processors. The implementation of a complex geometry spectral element Poisson equation solver on parallel systems was discussed.
Dr D R Emerson (Daresbury) opened the afternoon session by describing the parallel implementation of a finite volume code TVD code which solves for the laminar Navier-Stokes flow over a cavity. The explicit time marching code had been parallelised using standard domain decomposition techniques with the communication of halo data between the subdomains expressed in terms of the portable message passing harness Fortnet. Timings for both shared memory and distributed memory machines illustrated the effectiveness of machine independent harnesses.
Prof M Cross (Thames Polytechnic) discussed the merits and demerits of vector versus parallel processing for large scale commercial CFD codes. Parallel processing offered more scaleable increases in performance, and linear speedups with numbers of processors were illustrated for the commercial finite volume code FL03D. Professor Cross advocated the use of inert communications layers to aid portability and the development of software engineering tools to assist in the porting of dusty deck codes to parallel systems.
Dr M Wilson (Bath) described the porting of multigrid codes to their distributed memory Meiko i860 system. The aim was to port a 3D TEACH like Navier-Stokes code which solved for heat transfer in rotor-stator systems with pre-swirl cooling but current work focused on implementing a 2D version of the code. The domain decomposition, and halo data transfer procedures for the multigrid implementation were described. A 4 strip decomposition was shown to reproduce the convergence acceleration features for the serial solution for 3 level multigrid.
Dr A G Chalmers (Bristol) described a range of CFD applications code which were being parallelised. Efficient parallelisation of problems with large global data dependencies. For example, panel methods, free lagrange codes, DSNS and time accurate Euler codes were being investigated. Efficient parallelisation of conventional CFD codes using pressure correction methods and demand driven paradigms were also being studied. Impressive speedups for a number of linear equation solvers were demonstrated.
Dr G Robinson (AEA, Dounreay) kindly stepped in at the last minute to give a talk on the parallel implementation of a simple algebraic multigrid scheme. This method uses the matrix of equations on a fine grid of points to generate the equation matrices for coarser grids of points by agglomerating equations at clusters of neighbouring points into super-nodes. A domain decomposition procedure is used to parallelise the scheme. Two coarsening algorithms, one of which incorporates halo nodes and the other of which does not allow grid coarsening across subdomain boundaries were described. The latter algorithm was shown to be the more efficient. Prof P Bettess (Newcastle Upon Tyne) discussed parallel computing applied to problems in marine technology. Codes to solve the Morison equation to solve diffraction problems using the panel method and a discrete vortex code had been implemented on a transputer system. Typical efficiencies ranged from 50% to 80% for systems using up to eight processors.
The subsequent discussion focused on areas of common interest between the various groups. Obvious areas of overlap included the development of portability tools, the development of generic domain decomposition harnesses and the implementation of different algorithms such as multigrid or iterative conjugate gradient solvers on parallel systems.
CCPl2 is in the process of arranging a series of more focused technical meetings, the first of which will be held on 25 November 1991 at the University of Bath and will discuss Multiblock and Multigrid Methods for High Performance Computers. If you would like a copy of the proceedings of the May workshop, or a registration form for the November workshop, please contact either of us.
This meeting, one of a series held by the Modelling and Management of Engineering Data Community Club (ClubMmed) on 26 September at the IEE headquarters in London, had an attendance of 62, including 23 from industry or independent consultancies.
Chairman Colin McGreavy (Leeds) explained that the activities of the Club have had to be reduced since the SERC funding cuts last year, which hit this part of the EASE programme particularly hard. However the Education and Awareness part of the programme still provides for the continuation of this Club.
In setting the scene for the day's programme Professor McGreavy explained that the intention was to examine the role that data management and data modelling could be expected to play in SERC supported work. Not only is it now an important issue in industry, but with the increasing number of interdisciplinary research groups and active collaboration both nationally and internationally there is a need for the provision of a range of infrastructure software support tools. With the continuing move to integrated design and operations, the availability of such resources is reaching the stage where further progress would be compromised without these facilities.
The presentations were designed to expose the problems which need to be addressed and to try to summarize the present situation.
Peter Winter (Prosys Technology) gave an overview of the data exchange and management problems faced by the process industries. Typically process plants are complex, represent a considerable investment, and are operational for a long time. Managing several views of a process plant from initial concept through design and operation to decommissioning, while coping with maintenance and revisions over the years, presents a challenge to both corporate management and IT specialists.
Paul Sawyer (Independent Consultant) listed the initiatives presently addressing these concerns. They include an IChemE/IMechE joint working party, special interest groups in the Engineering Equipment and Material Users Association and the British Chemical Engineering Contractors Association, the work of BSI, the DTI, CADDETC (University of Leeds), the Engineering Design Centre at Newcastle, and, of course, SERC. There are also projects within the major operators and CAE system vendors.
Geoff Parker (BP Engineering) gave us an entertaining look at the business management view, pointing out the benefits (and disincentives) to developing data management and exchange systems, and the importance of the long term viability of methods and technologies developed today.
Paul Preece (University College Swansea) summarised recent activities in the UK academic world. While there is certainly a lot of research effort in the UK, the Japanese are considering launching a substantial effort in the area, with wide industrial support and a time frame of 10 years, which would dwarf the present UK commitment.
Susan Bloor (Leeds), who was to talk about the ISO standard for engineering data exchange (STEP), could not attend due to travel disruption. This was unfortunate because there was considerable interest in this area. There was a general feeling that the community as a whole would benefit from some further briefing on this topic.
The afternoon began with an open session, where 18 people had the opportunity to give a short introduction to their work in the field. The topics covered include: modelling of the design process itself, computer tools for information modelling, integrating NC machines and automatic inspection through product models, data modelling in building services, data modelling in control engineering and dynamic systems, the work of the Petrochemical Open Systems Corporation and data modelling for hypermedia. This session showed the variety of disciplines facing problems managing complcx data, and the underlying similarity of these problems. I believe a number of useful contacts were made.
The final discussion centred on the future activities of the Club; the Steering Group was given a clear mandate to continue to run a programme of Education and Awareness events, particularly in the area of the nascent ISO STandard for Exchange of Product data (STEP).
What conclusions can we draw from the meeting? The enthusiasm of the attendees, a significant proportion of them non-academics, shows that UK industry is, and should be, taking this subject SERIOUSLY. The number of academic and industry-led initiatives over a broad range of engineering disciplines; from CAD drawings to enterprise-wide information systems, indicates a growing willingness to address Management of Engineering Data. It is important that a way ahead be found which will draw together these resources and encourage academic-industry collaboration in this developing field.
SERC's Application of Computers to Manufacturing Engineering (ACME) Directorate was established in 1984 to promote, at Higher Education Institutes (HEIs), high quality research and training in advanced manufacturing closely linked to the needs of industry.
Since then the Directorate has built up a comprehensive portfolio of projects in each of its major theme areas:
ACME insists that all of the projects it supports are carried out in partnership with industrial collaborators, ideally a consortium of a vendor and end user/s. Through its network of regional coordinators, ACME assists the HEIs and their industrial partners to develop grant applications, and once grants are awarded they are monitored at regular intervals to ensure the work is on target.
ACME holds an annual conference which gives its community, both academics and industrialists, the opportunity to hear about projects currently underway and to discuss issues of mutual interest. The latest in this very successful series was held at Leicester University at the end of August. At previous conferences presentations have tended to focus on ACME-sponsored projects. This year however the Directorate broadened the programme to include talks from outside its immediate community. For example, speakers from DTI presented on a range of activities the Department now supports in advanced manufacturing. And presentations from industrialists ranged across topics such as the management of change in a manufacturing environment to how to get the best out of industry/academic collaboration. The format was very well received and future conferences will build on this experience. A copy of the latest edition of ACME's quarterly newsletter is enclosed, which gives fuller details on the conference.
If you would like to hear more above ACME and its work please contact Gay Ford at SERC's Swindon Office.
The first Summer School in Computational Fluid Dynamics (CFD) to be held by the Rutherford Appleton Laboratory was jointly sponsored by SERC, under the EASE Programme, and Ercoftac (European Research Community on Flow Turbulence and Combustion).
The main speakers were Professor Derek Causon (Manchester Polytechnic), Mr Steve Fiddes (University of Bristol), Professor Jim McGuirk (Loughborough University of Technology) and Dr Elie Znaty (Bertin et Cie). Together they presented a well-coordinated programme of lectures, skilfully leading us from an introduction to CFD to a fascinating insight into their current research interests. The early lectures dealt with the mathematical model and elementary finite difference schemes. The ideas of stability, consistency and convergence were introduced against this background and then we moved on to more sophisticated approaches including Total Variation Diminishing (TV D) and Pressure Correction schemes. Shock capturing performance was related to a number of interesting applications including the design of aircraft and oil rigs.
Throughout the week there was a range of hardware available on site, including a Stardent GS2000 superworkstation, nine SUN 3s and two SUN SPARCstations.
In addition to a number of codes provided by the lecturers we were introduced to two commercial packages, FLUENT and PHOENICS. Each of the participants was asked to tackle a particular problem and the week ended with the presentation of results (or an explanation of what went wrong!).
The course was residential with excellent accommodation provided at The Cosener's House in Abingdon. In such pleasant surroundings, with good company and, above all else, an extremely good course, it is hardly surprising that the week was voted a great suceess by all concerned. We all know how much work is involved in making this kind of event run smoothly and our special thanks go to a number of staff at the Rutherford Appleton Laboratory, namely Pam Peisley, Manjit Boparai, Chris Greenough and, in particular, Conor Fitzsimons.
For further information or details about literature please contact: Conor Fitzsimons.
I have no hesitation in recommending similar events in the future. You won't be disappointed.
Software Validation, Evaluation and Quality Assurance
The Computational Fluid Dynamics (CFD) Community Club held a Seminar on Software Validation, Evaluation and Quality Assurance in CFD at RAL on Tuesday, 9 July 1991.
It was chaired by Dr B R Williams (RAE) and was attended by more than 75 people from industry and higher education. The meeting was organised in response to a recommendation arising from the Community Club's Workshop on Accuracy in Numerical Modelling held in November 1990. It was recognised that the validation, evaluation and quality assurance of software were important in the provision of good quality computational data for the designer. The purpose of the seminar was to present a range of approaches to these topics and to provide a forum where other methods and appropriate software tools could be identified.
Dr Williams welcomed the attendees and set the scene by defining what he meant by the terms validation, ie, the accurate implementation of the mathematical model, and evaluation, ie, the degree to which the CFD method models the physical problem.
Dr C Albone (RAE) introduced a methodology for validating Euler codes, covering a number of widely applicable topics. For example: does the code justify properties of the solution not explicitly satisfied in the numerical formulation, is the solution independent of the choice of time step, does the code provide the predicted order of accuracy for the algorithm. He gave a number of examples, and showed how useful information could be extracted from results which at first sight indicated that there was a problem in the code. Dr Albone finished by noting that although validation was time-consuming it was essentiaL
Dr M F Webster (Swansea) made a presentation on the validation of software for Newtonian and non-Newtonian flows. He discussed a sequence of test problems to test aspects of the transient and steady solutions obtained by the code. These ranged from couette and cavity-driven flows up to industrial examples, eg a glass flow (body force driven) problem with temperature dependent viscosity.
Dr G J Page (UMIST) spoke about the evaluation of shock wave and boundary layer interaction flows in the EUROVAL Project. He outlined the work of the EUROVAL project and then moved on to consider two test cases used within the project to evaluate turbulence modelling and numerical practices for transonic bump flows and transonic aerofoil flows, and compared the computational results with the experimental data.
Dr B A Younis (City University) presented work on the evaluation of turbulence models for non-circular ducts and channels. He identified shortcomings in a number of turbulence models which had been identified through careful their evaluation against a range of experiments, and showed the sequence of test cases used in code validation for duct and channel problems.
Dr I P Jones (CFD Services) discussed a number of issues pertaining to validation, evaluation and quality assurance. He highlighted methods for validating new releases of software and the value of comparison against external software and experimental results. He stressed the need for agreed solution databases for validator. purposes. Dr Jones finished by noting that no matter how good the software, there was no replacement for good, skilled people.
Dr P R Voke (Surrey) made a presentation on the evaluation of and using direct numerical simulation (DNS). He showed the role of DNS in the evaluation of turbulence models. He gave examples where DNS had been used to identify shortcomings in experimental data. He concluded that the evaluation of turbulence codes could only be carried out effectively by those with understanding of real turbulence.
Dr T R Hopkins (Kent) surveyed a number of techniques of tools for the quality assurance of Fortran software. He provided an introduction to the subject, and explained a number of the commonly-used QA metrics applied to software. He went on to survey a number of public domain or inexpensive software tools which could be used to help improve software quality.
Mr A G Hutton (Nuclear Electric) discussed the development of CFD software to comply with quality management within the nuclear industry. The talk focused on the theory and implementation of the quality management procedures used and how they were applied flexibly to reflect the relative importance of the software in question.
Mr R Doe (British Aerospace) presented an overview of the quality assurance of CFD software in the aerospace industry. He showed where this fitted into the software development and maintenance life cycle, and the role for validation and evaluation in the process. He finished by giving an example of some software developed using the methodology.
In the ensuing discussion session a few points emerged clearly. The quality of much experimental data is insufficient for the purpose of evaluating CFD software, and that it would be helpful if the Community Club gathered information on sources of good quality data. It was agreed that there is a need for a QA system which doesn't stifle initiative or innovation in CFD software development and research.
The chairman then brought an informative and enjoyable meeting to a close. A copy of the presentations at the Seminar can be obtained by contacting me.
The 1991 edition of the CHEST Software Directory has been printed and distributed widely throughout the academic community. Electronic versions (accessible via JANET on NISSPAC and Lancaster University's NPDSA service) are also available and kept as up to date as possible. The Directory lists details and prices of over 800 software products, with contact details of more than 200 software suppliers. Your departmental secretary should have received a complimentary copy of the 1991 edition of the Directory - enquire at the CHEST office in Bath on 0225-826282 if this is not the case.
To encourage further communication amongst users of particular products, CHEST sets up an electronic discussion forum for each of its major deals.
The lists are based upon Newcastle University's MailBase service, and each has a name beginning with chest- with a suffix which identifies the particular product, eg brs for BRS/Search.
In addition to the software specific discussion lists shown below, there is also a chest-directory list for general comments about the CHEST Software Directory, and a read-only chest-news list, via which the latest news from CHEST is circulated.
chest-arc-info chest-oxmol chest -bartholomews chest-pc-fortran chest-brs chest-pc-graphics chest-cgm chest-pcnfs chest-dbase chest-pro-pascal chest-dtp chest-sas chest-emutek chest-saspac91 chest-geoplan chest-spss chest-ingres chest-uniras chest-isi chest-wordperfect chest-novell chest-wordstar chest-oracle chest-xwindows
Subscribe to a CHEST discussion list by sending an electronic mail message to mailbase@newcastle which has a one-line command as its text in the form of
subscribe <name of list> <your first name> <your surname>
where <name of list> is replaced by the name of the list you wish to join. Thus if someone named John Smith wished to join the Ingres list, he would send the message :-
subscribe chest-ingres John Smith
The following products are currently available as CHEST centrally-funded deals:
Confirmation of a deal for Guide, an authoring package for PCs and Macs, has also recently been received.
A number of other products, such as DOS 5 and WordPerfect, are subject to special offers and available at discount prices for limited periods of time. Improved offers on WordStar, Ventura and Micrografx products are also available. Anti-virus software is now being issued (no charge to sites) via the NPDSA service. Your institution's CHEST Site Contact has further details about all CHEST agreements, and copies of the mailings are held in Section D3 on NISSBB.
More than 70 representatives of computer hardware and software suppliers attended a CHEST Suppliers Seminar in London earlier this Summer. Speakers from various computing fields in education gave the delegates a valuable insight into the world of education and the marketing opportunities created by the academic community's computing needs. CHEST is planning similar events in the future.
The growing range of services available via the NISS Gateway currently includes the NISS Bulletin Board, NISS Public Access Collections, the National Public Domain Software Archive (NPDSA), and British Telecom's electronic yellow pages (EYP). Access to three European services - ASK (a German information service with English and German interface), STN International (a scientific and technical information network), and DIMDI (a German biosciences online service) has been available for some time, and three other European services - EuroKom (the EC's conferencing/mail service), ECHO (EC information service), and DataStar (an online information service) - have recently been incorporated. We are currently testing the service with BUBL (a bulletin board for libraries), and expect to add it to the Gateway in the very near future.
A second gateway has been implemented which will load-share with the original one. Users should normally be unaware of which machine they are being routed through, but will benefit from this improvement to the service since it effectively doubles the number of simultaneous users which can be supported.
The two word command call niss typed at a PAD prompt should access the NISS Gateway, and it is then a simple matter of making choices from the menus and selecting the services you require. All subsequent connections are made automatically, and when you logout from a service the Gateway menu will be restored allowing access to other services if wished.
New sections of information continue to be incorporated into the NISS Bulletin Board (NISSBB) - recent additions include details of ex-demonstration hardware on offer from Hewlett-Packard (Section S10), brief details (Section D3F6) of any changes to CHEST Directory data (complete versions of which are available on NISSPAC and NPDSA) and other Sections (eg Y2, Y3) have been reorganised.
Plans are underway to enhance the appeal of the NISSPAC Service by broadening still further the range of subjects and disciplines it encompasses. Other datasets will be added (eg the SUN Catalyst collection), and descriptions of datasets (eg the ESRC data archive) will also be incorporated. NISSPAC is also collaborating with the German academic information service (ASK) on information exchange and the development of a joint English German thesaurus.
