A surge arrester is a component that protects electronic equipment
from harmfully voltage/current transients (e.g. stroke of lightning,
cross-talk, etc.). It is of
importance that the ignition-voltage is stable on certain level.
A couple of already manufactured surge-arrester-prototypes
have been used
to train a neural net, in
order to optimise component properties. The final net consists of
20 different input-parameters which influences 2 outputs indicating the
first-ignition voltage
(UTLA) and the difference between the first and the
second-ignition voltage (dUTL).
It is shown in the analysis part of the
work that it is theoretically possible to
design which stands up to the world-wide ITU-T K.12 recommendation
by matching the inputs to this specific net in a certain configuration.
An implementation of
an artificial surge arrester has been made which does not fulfill
the requirements. The difference between the theoretical and the
practical results may be derived to
the noisiness of the tubes. The spark that causes the ionization
process within the surge arrester is very difficult to control. It is
out of the frame of this work to study
what s affecting the ionization process. The neural network
designed may be useful in the initial part of future development work.
