This issue describes the reshaped EASE Programme. including detailed examples of some of the remaining activities. namely the Education & Awareness events, AI activities, examples of community club work, etc. There has been a great demand for EASE technical reports. New reader? If you wish to receive your own copy. please let me know.
On 19 March 1991 the Computing Facilities Committee (CFC) decided on the activities to be carried out under the much reduced EASE Programme for 1992/93 and beyond. The funding will be 20% of that previously planned.
CFC decided to retain a viable Education and Awareness Programme, together with one major focus activity, rather than attempt to spread the effort and funding across a number of activities. Visualisation, including the essential graphics and user interaction components, was selected from a wide range of possible options as the major focus. The details, including deliverables, of the residual programme from 1992193 onwards are as follows:
CFC also looked at priorities for the 1991~2 Programme, depending on the funds allocated. Unfortunately the final allocation for 1991/92 is still not decided so the final details of the programme cannot yet be settled. In priority order the activities for 1991/ 92 were decided as follows:
CFC decided that all other activities planned for 1991/92 should stop. These include Assessment, Data Exchange, Parallel Processing, and Finite Element Library. More details on the 1991/92 activities and the 1991 events diary will be in the next issue.
Despite the cut in the EASE programme announced by Bob Hopgood, work is continuing to support you, the Engineering Community, albeit at a reduced level of effort. One project, the Engineering Information Directory (EID), has recently scored a notable success - with a prototype database and user interface now available to Informatics users (for alpha-testing).
The EID, also known as the EASE Information Directory, is a database system facility being developed at RAL with the following three features:
As indicated above, we are now well down the road of implementation, currently offering prototype access internally to the EASE Schedules and RAL expertise database. This has involved several tasks, as follows.
A data analysis of the EASE Schedules has been performed, and a method devised for the automatic transfer of the data from Displaywriter (word-processor) form into database tables using embedded SQL in an INGRES UNIX environment. Similar data analysis and table building has been undertaken for the RAL expertise facility, with some of the data being regularly updated from the RAL Personnel/Accommodation database on the IBM. Using INGRES facilities, a standard menu-based screen interface has been designed to incorporate both of these (and later other) applications, and has been implemented using forms in INGRES 4GL.
The production configuration will involve a Sparcstation 2 accessible over JANET to EID users, who will be able to define their own terminal type to exploit normal INGRES functionality. This machine will also enable EID to communicate with externally held databases and will act as the client node running an INGRES front-end process. The INGRES server will be another Sparcstation 2, closely coupled to this front-end.
This development work is being spearheaded by a team of four staff in the Data Engineering Group, under the leadership of Keith Jeffery, Read of the Systems Engineering Division, together with input from members of other divisions of Informatics Department at RAL. We expect the production platform, based on INGRES DBMS and the Sparcstation 2 server, to be accessible from the network in July. Later, it is intended that the INGRES relational DBMS will be integrated with STATUS freetext search DBMS, to provide a powerful search engine suitable for Engineering users.
This three-day course, specifically produced at Computing Facilities Committee's request, will be held at Salford University on 16-18 July and RAL on 23-25 September 1991.
The course is intended to provide you with an awareness of what it takes to develop a knowledge based system (KBS). To achieve this we focus on the life-cycle of a KBS, with the intention of making you aware of the steps that the knowledge engineer should be taking, the decisions that have to be made and some of the pitfalls that should be avoided.
We will discuss the different ways of developing KBSs, taking particular note of the emerging basis for a model-driven approach.
To get the most out of this course you should already have a good grounding in the fundamentals of the popular knowledge representations and inferencing methods. Ideally you will have some knowledge of the kinds of support environments available to AI programmers and have done at least a small amount of programming using object oriented systems and rule based systems with forward and backward chaining. AlAI runs associated courses for those without this experience.
The content of the course is designed to make you aware of how KBSs have been designed and developed, what attempts are being made to formalise the process and the difficulties inherent in formalising it.
There are seven course components:
We examine differences between conventional life-cycle models and prototyping methodologies as applied to the development of KBS, highlighting some of the problems encountered with each approach. We then present a recently developed life-cycle model that claims to overcome many of the problems encountered in other approaches.
In our discussion of selecting an application we review fairly well established and accepted criteria for deciding whether a KBS solution is appropriate for a particular problem. The criteria fall into the broad categories of the problem characteristics, the available human expertise, the economic justification and the feasibility of establishing a project team.
Scoping is the estimation of the resources needed to execute a KBS solution and of course is essential in determining the economic justification. In a discipline as young as KBS development the guidelines are perhaps not as well established as the selection criteria.
We distinguish between knowledge elicitation (getting knowledge from an expert) and the more encompassing process of knowledge acquisition (getting knowledge from many sources, including experts, and structuring the knowledge). We stress that good knowledge engineering requires not only good knowledge elicitation but a lot more besides.
We briefly review knowledge elicitation techniques but concentrate more on the overall process of knowledge acquisition.
A feature of more formalised knowledge acquisition techniques is that not only do they assist the knowledge engineer in acquiring, analysing and structuring knowledge, but they also result in working design documents called intermediate representations. These provide documentation for the designer, implementer and maintainer of the KBS and can be maintained as and when the KBS software itself is changed. This is in contrast with current ad hoc methods, where the KBS software itself provides the only working document for the designer, implementer and maintainer who interact with it.
We concentrate on one of these more formal knowledge engineering techniques (the KADS methodology) and through paper exercises, give you the opportunity to try some parts of the methodology for yourselves.
Automatic knowledge acquisition methods are intended to reduce the role of the knowledge engineer by having some of the analysis and structuring of knowledge done by a machine.
An intermediate representation by no means fully determines the exact contents of the final knowledge base and we present some of the many options and difficulties which face the implementer in the creation of a knowledge base. We look at the task of transforming analysed knowledge into an implementation and identify some of the things to look out for when deciding which shell or toolkit to use. Again we use paper exercises to encourage you to think about some of these options and difficulties yourselves.
Most knowledge based systems perform some function which is only a part of a complete software solution. We present an overview of the kinds of integration that might be required of a knowledge based system, such as the ability to access and update data in a spreadsheet or relational database. Integration with existing work practices and procedures is also considered.
In conventional software engineering the process of verifying and validating a design and its implementation is a fundamental part of the development process. We look at how these activities may be incorporated into the development of knowledge based systems and how the more formalised approaches to knowledge engineering help in validation.
Welook at the task of on-going maintenance of a knowledge based system. We also consider the environment in which the knowledge based system is installed and used.
After each component of the course, students should be able to perform:
The SERC Community Club in Computational Fluid Dynamics (CFD) was founded in March 1990 as one of four set up under the EASE Programme.
Since its Inaugural Meeting in March 1990 it has grown and now has over 400 registered members. It has held three specialist meetings and held its annual Plenary Meeting at the end of April 1991. Other activities of the Club this year include: sponsoring development work in visualisation for CFD, a common software library, software assessment and planning for the Club Summer School to be held in September 1991.
In 1986 the Electro-Mechanical Engineering Committee (EMEC) set up the Advisory Group in CFD, chaired by Prof Phil Hutchinson (Cranfield), to review the state of SERC-supported research in CFD. The Group's remit covered a review of the methods of coordinating research and disseminating information and to advise on future policy on research, education and training. Arising from their deliberations, EMEC requested the Computing Facilities Committee (CFC) to support a computational modelling activity whose main objectives are
The registered membership has grown dramatically during the first year of operation. Its composition reflects the wide range of application interests of the members, eg the computation of flows around aircraft and in combustion engines, the mixing of chemicals in the process industry, the slow motion of viscous materials in the production of glass, and the flow of rivers and tides. The operation of the Club is overseen by a Steering Group, whose membership reflects the interdisciplinary nature of CFD:
By holding joint meetings and by having suitable representation on the Steering Group the Club cooperates with complementary organisations in CFD in the UK, eg:
This helps to co-ordinate activities in the UK.
The Club held three technical meetings during the year, on visualisation, accuracy in numerical modelling, and mesh generation. A feature of these meetings has been the active participation of industry, both in attendance and in presenting papers. The Visualisation Workshop, held in July 1990atRAL, which was attended by more than 70 people, served two purposes: to review the current state and future trends in visualisation, and to give examples of the effective use of visualisation in CFD. Applications included flow around buildings, aerospace applications, combustion in engines and the motion of plasma in space.
A two-day workshop on the important subject of Accuracy in Numerical Modelling was held in Abingdon in November 1990 in collaboration with the ICFD. Eighty people attended a meeting which discussed accuracy in the various stages of numerical modelling in CFD: the choice of physical models, numerical formulation, problem solution and validation. The workshop recommended that the Club should
The Club held a seminar on Mesh Generation in February 1991 at RAL. The Club proposed two activities in visualisation during the year: the assessment of the use of a superworkstation in engineering applications and the initial development and assessment of advanced visualisation techniques. The superworkstation placement was part of the CFC programme to assess these types of machine in a research environment. A Stardent Titan was placed in the Mechanical Engineering Department at Imperial College for six months. The work uses an object-oriented approach and is based on the SPEED package for engine combustion in Professor Gosman's group. The contract for a pilot development of visualisation in CFD was awarded to University College Swansea. The objective of the study was to exploit recent advances in visualisation hardware and software to improve the processing of CFD analysis, in a form which would be available to the community. The SERC financial cutbacks resulted in this contract being cancelled before work could start.
The Club conducted a survey of its membership in March 1991, the results of which were presented to the Plenary Meeting. Arising from this, several offers of packages for the Club software library were received. These are in the process of being evaluated and the initial release of the library will be available later in 1991. It is intended to have a phased release of software.
The Club will hold a Summer School on 16-20 September 1991 in conjunction with the UK Pilot Centres of ERCOFTAC. The school will focus on problem solution strategies, equation discretisation and mesh generation. The school will include tutorial lectures and research presentations with extensive practical sessions.
The first year of operation of the SERC Community Club in CFD has been very successful. The Club has organised series of very well attended meetings with presentations of a high standard and lively and well-informed discussion. Good foundations have been laid for the responsible dissemination of common academic software through the Club library. The Steering Group of the Club is developing an exciting programme of activities to build on the success of this year.
A final total of 173 abstracts have been submitted to the Third International Conference and Exhibition being run by the SERC/DTI Engineering Applications of Transputers Initiative.
Seventy-seven of these abstracts are from 26 different countries outside the UK (see table for a detailed breakdown). Transputer Applications '91 will be held at the Moat House International Hotel in Glasgow on 28- 30 August 1991. A registration form for the Conference and the Tutorial programme on 27 August are included with this edition of the Newsletter.
| Country submitting abstract |
Number Submitted |
|---|---|
| United Kingdom | 96 |
| Germany | 10 |
| France | 9 |
| India | 7 |
| Italy | 6 |
| Japan, South Africa (each) | 5 |
| Australia, Spain (each) | 3 |
| Belgium, Israel, Singapore, USA, USSR, Ireland, Brazil, Yugoslavia, Canada, Bulgaria, Greece (each) | 2 |
| New Zealand, Algeria, Pakistan, Poland, Switzerland, Hong Kong, The Netherlands (each) | 1 |
| Total | 173 |
The Conference and Exhibition are being jointly organised by the Strathclyde University based Scottish Transputer Centre and the Transputer Initiative.
Refereeing of abstracts started on 18 April and was completed by the end of the month. A detailed Conference Programme will be distributed in late May/early June.
Inmos will be giving details (both general and technical) on the newly announced T9000 (previously referred to as the H1) at the Conference. The Keynote Speakers will be David May (Inmos) and Douglas Stevenson (Parsys).
The Community Club in Computational Fluid Dynamics (CFD) held a seminar on Mesh Generation Applied to CFD at RAL on Thursday, 7 February 1991. In spite of the treacherous weather conditions more than 75 from Higher Education Institutes and industry attended a very lively meeting chaired by Dr Nigel Weatherill (Swansea). The morning session was devoted to presentations on the more established methods used in mesh generation, while in the afternoon the meeting was told about new approaches under development.
Prof Ken Morgan (Swansea) delivered a talk on the advancing front method. He began by explaining the algorithm and showed how it could be used to mesh geometrically complex domains in two or three dimensions. He outlined the data structures which give the method its flexibility and power. Professor Morgan finished by showing how the method could be used in mesh adaption, and by indicating areas requiring further research, eg the development of reliable error indicators.
Mr Bryan Colyer (Vector Fields Ltd) presented a paper on Delaunay triangulation in three dimensions. He explained Delaunay's original criterion and described the algorithm in detail in two dimensions, and how to handle numerical round-off errors in the implementation. Mr Colyer went on to describe the method in three dimensions and showed how to deal with degenerate tetrahedra and the problems associated with meshing non-convex domains.
Mr Jonathon Shaw (Aircraft Research Association) made a presentation about the multi-block approach in which the domain is subdivided, and each subdomain is meshed using a structured grid with a pre-determined topology. He compared the elliptic and algebraic approaches to generating the mesh in each of these blocks. Mr Shaw then outlined a current investigation into the development of a hybrid multi- block method in which some subdomains are meshed using a Delaunay algorithm.
Mr Malcolm Sabin (PEGS Ltd) introduced the transfinite interpolation technique, showed its relevance to a variety of analysis methods, and indicated its operational capabilities. He explained the basic idea underlying the method, in two dimensions, and described its application to three-dimensional mesh generation problems.
Dr David Mayers (Oxford) introduced some current research on adaptive refinement in multigrid. The structured quadrilateral mesh is refmed locally and the multigrid is applied so that a comparison of the residuals of the same solution on the fine and coarse mesh is used as a local error indicator for the adaption process.
Dr J Z Zhu (Swansea) presented an algorithm for adaptive remeshing by quadrilateral elements. After a brief survey of quadrilateral mesh generation, Dr Zhu showed how his method used many of the ideas in the triangular, advancing front algorithm. A quadrilateral is generated by first generating two triangles in the front which share a common side, and then combining them. After showing how the method is used in an adaption strategy, he went on to discuss how element quality is assessed on these meshes.
Mr Clive Albone (Royal Aircraft Establishment) described his feature-associated mesh embedding algorithm. The method meshes the domain using a strict hierarchy of 6 overlapping meshes. A feature is a simple geometric entity, eg the point of intersection of three surfaces. Configurations of any complexity can be broken down into a combination of features and meshed accordingly. There is one local mesh for each feature, thus many meshes may cover the same region of space. He finished by saying that the method aims to combine all the natural advantages of structured meshes with the flexibility of unstructured meshes.
Dr Cecil Armstrong (Belfast) introduced his work on mesh generation by medial axis subdivision. In this method the domain is subdivided into simple subregions using the medial axis of the associated Voronoi diagram. Additional subdivisions are used to remove concavities and to simplify the subregions further. Each of these is meshed using standard patterns of quadrilaterals. Dr Armstrong finished by indicating how he was extending this method to three dimensions.
Dr Conor Fitzsimons (RAL) made a short presentation on measuring mesh quality. He proposed a set of measures which might provide a source of objectivity in assessing the quality of a mesh used in solving as flow problem. He divided these into two classes: a priori, ie those which depend on the geometric properties of the mesh, eg element aspect ratio and angle size, and a posteriori, ie those which use information from the solution. He finished by citing examples of meshes to which these measures had been applied.
The chairman then brought a most lively and enjoyable meeting to a close, and announced that the next seminar would be held on 22 May 1991 at Daresbury Laboratory, in collaboration with Collaborative Computational Project 12, on Parallel Computing in CFD. A copy of the presentations made at the Mesh Seminar can be obtained by contacting me.
SERC/ERCOFTAC UK SUMMER SCHOOL 16-20 September 1991 The Cosener's House, Abingdon. Full details in next issue.
The Steering Committee of the AI in Engineering Community Club has produced a paper reviewing the current state of AI tools available to the engineering community. The aim of this review was to identify AI software development tools which should be adopted as standard throughout the academic engineering community.
This report concludes that the current knowledge relating engineering applications to AI-based solutions is still embryonic. As such,no clear statement can be made on the suitability of currently available AI environments or shells. However the role of such systems in initial training is supported and the evaluation reports on such systems produced by the Artificial Intelligence Applications Institute (AIAI) are considered to be very valuable. However consideration should be given to adopting versions of three languages as standard: Common Lisp, Prolog and C++, running on Unix based workstations supporting X11 and TCP/IP. It is further recommended that strong support be given to establishing a mechanism for software exchange between academic researchers. This would encourage the pooling and comparison of a number of tools in a range of application domains. This is seen as a means of accelerating the progress towards the principled use of AI techniques in engineering applications, and the formation of valuable library of methods and tools.
The report is entitled AI Tools Selection for Applications in the Academic Engineering Community (AIAI-TR-91) and is available from AIAI (address given below).
This evaluation report of MUSE is now completed and is available from AIAI. Muse is an AI toolkit designed for real-time applications. It includes rule-based and object-oriented programming, a general AI programming language (PopTalk) and agenda-based control. The system runs on Sun workstations.
Prototype applications in muse exist for data fusion, command and control, on-board fault diagnosis on helicopters, monitoring and control of a paint plant and flight monitoring. Prototypes currently under development include computer network management, control of manufacturing cells and for electronics manufacture.
We are currently preparing a paper based around the tutorial which was to be given as part of the EASE91 symposium. We have extended the range of the paper and hope that by making this available as part of our report series we will be able to reach a a larger proportion of the academic engineering community.
This paper will highlight the importance of using methods when building a computer system and the special requirements of methods for knowledge based systems development. In addition to a review of the KBS-specific methodologies, the approaches used in the development of conventional computer systems will be outlined, along with some of the more formal methods used, and the applicability of these approaches to KBS development will be discussed.
It is hoped to make this paper available by June 1991.
The courses schedule for 1991 is as follows:
