Andy Brightmore studied Mechanical & Production Engineering at what used to be Sheffield City Polytechnic and has worked in the Arc Welding Department at TWI for seven years.
Now Head of the Microcomputer Section in the Arc Welding Department, and responsible for all arc welding software, Andy used to work on the development of expert systems, including programs on procedure generation and welding machine fault diagnosis.
Development of large scale expert systems is always done in a modular way, but not often to the extent that it is carried out in a number of different countries. Andy Brightmore describes the development of an integrated welding engineering expert system under the EUREKA initiative.
The Arc Welding Department at TWI has been developing PC software for over seven years, resulting in 11 off-the-shelf software packages, like PREHEAT, WELDCOST, and WELDSPEC PLUS, and many more one-off customised programs, developed for Industrial Members. The development team in the Microcomputer Section is always on the lookout for new ideas, and when the DTI suggested, in late 1987, that the EUREKA initiative might be a good mechanism for developing expert systems with a wider European base, the opportunity was thought too good to miss. A number of other welding research organisations in Europe were approached, they thought the idea was a good one, and in June 1988 the project was formally announced at the EUREKA Ministers' meeting in Copenhagen.
The original title of the project was EU259: Expert systems in welding, although the project collaborators adopted the name Euroweld, both for the project and for the software developed in the project. Within the project, each collaborator would lead one or more task groups, each of which would develop a PC-based expert system on a specific aspect of welding engineering. Then, Task Group 1, led by Lincoln Norweld in Norway, would integrate each program with a sophisticated database program called XWeld, giving the largest welding engineering expert system in the world.
The collaborators, and the systems they have developed, are as follows:
- Integration
- Lincoln Norweld, Norway
- Fatigue design and crack growth
- TNO, The Netherlands
- Process selection
- TWI/Nuclear Electric, UK
- Parent metal and consumable selection
- IVF, Sweden
- Edge preparation
- FORCE Institutes, Denmark
- Welding parameter generation
- TWI UK
- Mechanised MIG welding procedures
- TNO, The Netherlands
- Welder qualification and performance management
- FORCE Institutes, Denmark
- Quality control
- FORCE Institutes, Denmark
- Aluminium alloys
- TNO, The Netherlands
- Heat treatment
- TNO, The Netherlands
- Fatigue design and improvement
-
EPFL, Switzerland
The Euroweld user
The Euroweld systems are being developed primarily to give guidance to welding engineers responsible for writing and managing welding procedures. Most of the expert systems generate part of a welding procedure, although as indicated above there are also systems on fatigue design, welder qualification, and quality control.
The project collaborators recognised that personal computers are the most common form of computer hardware used by welding engineers, and indeed many welding engineering software packages are available on PCs. Thus, the Euroweld systems will all function as stand-alone PC-based programs. However, the integrated system is too large to run effectively on a PC, so the expert systems have all been modified, and integrated with Xweld, to run on Unix workstations, such as Sun computers. The integrated system is a complete welding procedure development and management program, which also takes advantage of workstations' multiuser and communications capabilities, so that a single Euroweld system can be run on many sites of a large organisation.
European Standards and 1992
Few welding engineers do not now appreciate the implications of new European standards (ENs). Indeed, EN 288 and EN 287 have already replaced BS 4870 and BS 4871 on welding procedures and welder qualifications. At the start of the project, however, these standards were in an initial draft stage, and it was not easy to predict how they would change. Welding standards have had a significant impact on development of the Euroweld systems:
- The output from the systems is often a welding procedure sheet. The content and layout of the procedure sheet is defined by EN 288.
- Different national standards have different terminology, even in English, and common European terminology would have been an advantage, if a document had been available early enough.
- Classification of parent materials and consumables is defined in a number of welding standards. Even in European standards there are three different grouping systems for parent materials. Thus, the developers of the consumable selector (called Eurosel) had to be aware of the EN groupings, as well as national grouping systems.
- Designation of welding positions and qualification ranges of welding positions were very important to a number of the expert systems, including the welding parameter generator and the welder qualification module. Even though these are defined in EN 288 (as PA, PB, PC, etc), not all welding engineers are familiar with these designations, and it was felt necessary even to include ASME designations (as 1F, 2F, 3G, 6G, etc) alongside EN (and BS) designations.
There was probably more discussion on how to incorporate information from standards in the project software than on any other subject. Having taken such care on this subject, however, it is felt that as well as being valuable technical aids to welding engineers, the systems will fulfil a valuable educational function on familiarisation with the new standards.
Project software
Another challenge has been use of state of the art programming techniques. Computer technology is advancing very rapidly, and software can become obsolete very quickly as computers become faster and graphical display techniques more powerful. This is especially true of long term developments such as Euroweld, which began in 1989.
The project collaborators, therefore, chose what was a very new technology at the start of the project, to ensure that the software would still be usable for many years after the end of the project. The tool chosen for system development was the Egeria expert system toolkit.
Use of a new technology has also had some drawbacks. Egeria was itself being developed, and was not considered robust until fairly late in the project. Potential users who are evaluating systems during development can easily be discouraged if systems occasionally crash.
The 'knowledge representation' is the way that the rules, facts and other information in the system are expressed in computer code. Conventional programming languages, such as Pascal, Fortran, and C, embed the knowledge in data statements which can be very difficult to read for the non-computer specialist. Expert system shells, however, are good at representing rules in English-like statements which a welding engineer can understand. This makes support and modification of the program easier.
Egeria also features the use of object orientated programming techniques, which can model some types of information much more efficiently than conventional programming, and also make the program more modular. This type of knowledge representation requires much memory, however, and PC systems running under MS-DOS are still restricted to 640KB RAM.
As a result of this, Egeria originally suffered from memory management problems, which caused some collaborators difficulties and which necessitated restructuring of some expert systems.
One of the prime requirements of the collaborators was that the systems should function on both PCs and workstations. Most Egeria code written for PCs can simply be recompiled on Unix workstations, but graphics images cannot be automatically transferred. Consequently, without a special conversion routine, those systems which use graphics to convey important information, such as joint preparation or welding position, would lose a very important feature.
Development of Egeria has now ceased, following acquisition of the company which produced it by the developer of another AI software toolkit. This may mean that in the long-term, the Euroweld systems may need re-programming to keep up-to-date with knowledge representation techniques, and user interface technology.
However, there is no doubt that some problems would be experienced with any programming tool on such a complex and large scale computer system. The fact that the Euroweld systems successfully function as both PC systems and Unix workstations, is a result of substantial collaboration between all the project partners.
A common mode of use
As well as using the same software development tool, the collaborators felt that a common user interface should be adopted. This means that users become familiar with all the systems, after having used one. Also, all of the systems would then be recognisable as being part of the Euroweld suite of programs.
This has led to development of common screen designs and a common use of function keys ( e.g. Fl - help, F2 - reset). Industry standards have been used wherever possible, and it is hoped that other welding software developers will follow the standards set in the project.
The future
A number of lessons learnt in the project may help other software developers in future:
- Development of a number of different, but related, software packages by geographically remote organisations is a challenging task. In EU259, the collaborators have found that full integration, where each software module communicates directly with each other, is impracticable. In fact, care has been taken not to make the systems too clever and take control away from the user.
- Great care has to be taken when selecting a tool for such a long term development project. The stability of the technology, the tool, and the supplier are key concerns, but are not always easy to predict.
- Software should be delivered using traditional programs wherever possible, as this technology tends to be more stable.
- The overheads of collaboration are high, but necessary to develop successful integrated software such as Euroweld.
