Stimulation Initiative for European Neural Applications
Esprit Project 9811
Production Planning for Client Specific
Transformers
Company background
Smit Transformatoren B.V. is
specialist in the development and the manufacture of large
transformers - up to 1000 MWatt / 525 KV - for industry and
generation of energy.
Transformers of Smit are supplied over the whole world. Smit
has about 350 employees.
The problem
Transformers manufactured by Smit vary in prices from
a hundred thousand guilders to 15 million guilders. They take 6 months
to 2 years manufacturing, including engineering time.
Since each client of Smit has his own specifications with regard to power,
voltage, efficiency etc. the transformers delivered by Smit
are custom designed and largely made by hand.
As each transformer is different, each transformer has its
own production time. In the planning of production time, not only
the delivery date and budget play a role, but also
questions with respect to the planning of the total
needed man hours, the progress of the project,
the planning between the different departments,
the optimal utilization of personnel capacity and the planning of
delivery of the raw materials.
When the production time of a transformer is planned,
only global specifications by the client and the delivery date are available.
For these reasons, even with Smit's experience
and know-how, production planning remains complicated.
About 15 years ago, Smit started to systematically sort out
the production data of previous projects. On the basis of these
data and on human experience, a
mathematical model - with about 150 parameters - has been designed
to estimate the production time.
However, since production methods and experience of personnel
changes continually, and moreover, since each time a different
transformer is produced, the model should kept up-to-date.
So the laborious and time-consuming work for the continual
adaptation (by hand) of the model remained.
Neural network application
In collaboration with Biologica - a neural network technology supplier,
some pilot neural network experiments with the data have been performed.
After a relatively short training period, the neural network
showed a better performance than the mathematical model.
Using a commercial neural network
package together with a spread sheet, the application is automized
and a user-friendly neural network is obtained.
The development, implementation and maintenance of
the neural network is kept within Smit.
Benefits
With the neural network, the quality of the prediction of the production time
is increased. But more importantly, the neural network is easier
in use. It saves much time in maintenance compared to the mathematical model
and it reduces the dependency on human experts.
Generalization
The key feature of this application is the use of a neural network
to make a prediction of the production time of a customized product based on a number of specifications
by the client. This has been possible thanks to the large number of representative example-data.
Similar methods can be used for cost calculation and planning of personnel.
Contact persons
H. Brockmeyer, Production Planning,
Smit Transformatoren B.V., P.O.Box 9107, 6500 HJ Nijmegen,
The Netherlands. Tel +31 24 3568649, Fax +31 24 3567636.
D.J.N. Egberts, Biologica, Julianalaan 35, 6721 ED Bennekom, The Netherlands.
Tel +31 318 413954, Fax +31 318 413944.
References for further reading
H. Brockmeyer. Calculation for Client Specific Transformers.
In:
Neural Networks: Artificial Intelligence and Industrial Applications.
Proceedings of the 3rd Annual SNN Symposium on Neural Networks, Nijmegen,
The Netherlands, 14-15 September 1995, Nijmegen. B. Kappen and S. Gielen (eds).
Springer-Verlag, London.