Trends and developments

Industrial, practical requirements of data modelling

Industry looks at data modelling procedures from a commercial point of view. Industry looks for tasks, which have certain, directly identifiable – and practically useful – purposes, e.g. we may need to calibrate an on-line measurement instrument, predict a quality parameter, evaluate a new sample or monitor/control a complex production process etc. If the data modelling procedure used fulfils the practical task at hand, the concern is not focussed on the statistical assumptions of the procedure, say. Industrial concern is focussed on methods that guarantee a practical functioning of the procedure(s) used. It is not overly concerned if the procedure is “scientifically valid” or not. There is sometimes a tendency towards a conflict between 'proper researched methods' – and the immediate needs in practical, industrial data modelling.

Better/faster computers

The development in computers has been very fast for the last 10-15 years. The processor speed has doubled every 18 months, and there are no signs of slowing down next five years at least. The computer, which in 1998 won over the world chess champion, is today available on an ordinary high-end PC-level. During a few seconds such a computer can compute billions of strategic possibilities of the chess game.

Similarly in the industrial setting we can expect to have giga-bytes from the company database in the memory of the computer and in seconds carry out a huge amount of specific tasks or simulations etc. The message to the software developers of tomorrow is “make algorithms - do not to be concerned about how to carry them out”.

We are aware though that this characterisation only applies to the “high-end” user levels, and that this segment will always be relatively narrow. So there will always also be a demand for less computer-intensive approaches, existing in parallel with the above front-line demands.

On-line measurements

The Toyota automotive company has the motto “all components are to be measured”. Car makers have had great success in securing that all components have absolutely correct dimensional tolerances – for very obvious reasons. Many companies, also outside the automotive sector, are now investing in similar on-line measurement equipment etc. Another important aspect is that experience has shown that it is important to detect as soon as possible, when the processes have become off-spec, or downright defect. Industry is increasingly interested in methods that can be used on-line for optimal process monitoring and control. Clearly such approaches are both data-intensive as well as heavily computer demanding under all circumstances.

Requirements for monitoring and control

From the point of view of today’s computer facilities it is fully possible to set up the necessary advanced procedures for monitoring and control of industrial productions. Many companies, especially within the process industry, have made large investments with the purpose of automatic processes control. But industry is expecting more from the present technology.

Software engineering – increased emphasis on reliability
When a construction engineer computes the strength of a specific construction, his company relies upon the results of the method employed. Similarly, when decisions are made on the basis of data modelling, one wants, indeed one needs to be able to rely on the recommendations of the software engineer or data analyst, who has carried out the analysis. It is expected from the software engineer to make arrangements such that the tasks are correctly carried out.

Future developments
Industry worldwide is investing heavily in monitoring data production chains a.o. Today the ultimate focus is on the possibilities for monitoring entire production plants etc. There have been established standards for the collection of such data. The equipment needed is relatively inexpensive. An example would be the recent developments of optical spectroscopic on-line instruments. Chemical process companies have invested in e.g. NIR instruments for process control in more than 10-15 years. These instruments cost less than 10% of traditional chemical process control equipment. The present, and especially the future data flow from this development are literally enormous, and continually growing. This development is requiring a paradigme shift that the H-method is addressing.