Stimulation Initiative for European Neural
Applications
Esprit Project 9811
Prediction of Yarn Properties in Chemical Process
Technology
Company background
Akzo Nobel, headquartered in the
Netherlands, is one of the world's leading companies in selected areas
of chemicals, coatings, healthcare products, and fibers. More than 70,000
people in over 50 countries make up the Akzo Nobel work force.
The problem
Among other things, Akzo produces high quality yarns. To produce yarns
with the desired properties in an economic way, one need to know the relation
between the available production technologies, molecular structures and
the final yarn properties. These relations can be obtained by chemical-technological
experiments. However these experiments are costly and time consuming. Insights
in these relations could considerably save on the technological experiments.
However, these relations are too complex for a quantitative judgement by
human experts.
Neural network application
Akzo, in collaboration with the university of Nijmegen, studied the possibility
of the application of neural networks to this problem. To provide examples
to train the neural network 295 yarns were produced with different structures
and properties. Of each of the yarns, 5 structure parameters and 15 properties
were determined. With these data, the network has been trained. The trained
network is able to successfully predict the properties from the structures
of new yarns. The neural network is now used within Akzo as a tool for
the researchers to find out how the best yarns are synthesized.
Benefits
The neural network saves on chemical-technological experiments. For this
problem, neural networks perform significantly better than standard techniques
from statistics and artificial intelligence.
Generalization
This is a typical problem for which no good numerical model exists and
which is so complex that human experts can give only a qualitative judgement.
Such problems are very widespread in engineering. Quantitative knowledge
of the relations between the various parameters adjustments and product
properties is helpful to increase the quality of the product and the efficiency
of the production process. If a large number of examples can be generated,
neural networks are often capable to learn these relations successfully.
Contact person
A. de Weyer, Akzo Nobel Central Research, P.O.Box 9300, 6800 SB Arnhem,
the Netherlands
References for further reading
A. de Weyer et al.: Akzo spint garen met genetische algoritmen en
neurale netwerken (in Dutch).
PolyTechnisch tijdschrift (procestechniek),
49
8 (1994) ia1-ia5
A. de Weyer et al.: Neural Networks used as a soft modelling
technique for quantitative description of the relation between physical
structure and mechanical properties of poly(ethylene terephthalate) yarns.
Chemometrics
and Intelligent Laboratory Systems, 16 (1992) 77-86.