Engineering disaster by computer

I was interested in the article about the numerical errors produced by the inexactness of some computer-processing routines ("Cutting edge", THES, October 23). But I take issue with the view that all will be well when these problems are solved, particularly regarding engineering products and systems.

Since about the mid-1980s I have been concerned about such powerful analysis and design tools being used without due regard to their limits and modelling assumptions invoked in their operation. I was not the only one.

A group of European partners in industry and education obtained European Union funding to produce a distance-learning package on finite element analysis. The aim was to introduce such techniques and their place in engineering design and highlight the limitations of any software that relies on numerical analytical techniques. The need for thorough engineering understanding and appreciation of the assumptions, approximations and other limitations involved at all stages was emphasised. In a typical engineering design this could involve the following key modelling steps: 1. Assessing the loading conditions likely to be encountered, especially worst cases

2. Obtaining materials data and properties, and behaviour in the situations envisaged

3. Developing and applying mathematical modelling techniques for analysis 4. Use of approximation methods to solve mathematical models

5. Use of modern powerful computing by which such solutions are obtained.

The problems highlighted in the article relate only to step 5, in which case we are already four or five steps away from the reality of the situation being modelled. Even with the most perfect computer processing it would still be possible to obtain beautifully produced solutions that are completely wrong. Indeed it is a guiding principle of the National Association of Finite Element Methods and Standards to assume that all such results are wrong by default until sufficient interrogation and experienced judgement have been applied to the whole analysis. Unfortunately I often encounter people who say there is no need for this any more "as it is all done by computer" - rather worrying coming from tomorrow's aeroplane designers.

We wish to encourage widespread but safe use of computers in engineering. In the right hands they make a good engineer better, but in the wrong hands they could make a bad engineer dangerous.

J. K.Martin Lecturer in mechanical engineering Open University