Stimulation Initiative for European Neural Applications
Esprit Project 9811
Application of Neural Networks for Evaluating the
Reaction State of Penicillin-Fermenters
Company background
No information available
The problem
Phenoxymethylpenicillin (PMP) is produced in large fermentation
reactors. Adding phenoxyacid (POA) and various mineral salts,
penicillin converts sugar into phenoxymethylpenicillin (penicillin V).
To ensure an optimal adjustment of important parameters such as
ph-value, temperature and the addition of nutrient solutions, a sample
is taken from the reactor regularly. This sample is analysed in a
laboratory to qualify the PMP-content. Depending on the result of the
analysis, experts decide on the adjustment of new parameters. A
problematic factor is that this procedure causes a deadline of about one
hour passing by between the sampling and new parameter adjustments which
very often results in a low yield in penicillin V.
Neural network application
A neural network has been trained by process data from several
reaction-processes including temperature, ph-value and time. The trained
net provides an accurate prediction of the fermenter's yield in
penicillin V.
Benefits
In comparison to laboratory analysis, the information provided by the
neural net is far more exact. The remaining minimal error is compensated
by its capability of realtime analysis. In addition to this, process
control is improved essentially.
Generalization
The application illustrates the importance of neural networks for the
control of non-linear production processes in chemical industry: Many
chemical processes are charcterized by a great amount of complex
parameters with a great influence on the product's quality. Customary
methods for an optimal process control, such as laboratory analysis and
evaluation of data by experts, very often cause stoppages or even
deadlines. Neural network techniques can provide a means for a more
exact analysis of chemical processes in realtime.
Contact person
Roland Kühne GmbH.
Hellersbergstraße 10b,
D-41460 Neuss.
Phone: +49 2132 163-3