Mandag 6. nov. 1995
kl. 12-14: Besok ved teknisk kybernetikk (Stein Wasboe)
kl. 14.15: Seminar "On-line estimation", Lunsjrom kjemiteknikk
(kjemiblokk 4). ALLE VELKOMMEN!
kl. 15-17: Besok ved institutt for kjemiteknikk.
15.00.15.15: S. Skogestad
15.15-15.30: M. Hovd
15.30-16.00: Y. Zhao / Y. Cao
16.00-16.15: B. Wittgens
16.15-16.30: A. Christiansen
16.30-17.00: Open
Kort bibliografi:
(En del av dere husker ham sikkert fra Aaland i august).
David Wilson received his PhD from the University of Queensland,
Australia in 1990 where he investigated control of raw sugar
crystallisation. For the past three years, he has been a post-doc at
the Swiss Institute of Technology (ETH) in Zurich working jointly with
Ciba-Geigy. Currently he is now lecturing at the Hogskolan i Karlstad,
Sweden.
Current interests lie in broadly the
simulation and modelling field of chemical processes, and I am currently
working with a new proposed simulation center attached to one of our local
pulp & paper mills who is expanding by constructing a new 3.5 Billion
SEK paper machine.
I am trying to get money for a feasibility study for paper machine dynamics,
and also a CFD-style environmental model of the mill effulent treatment
& lake surrounds.
I am also teaching electrical engineering subjects, DSP, assembler
programming, etc, and hence the course in RISC programming. However I am
still very much a beginner in this field still.
ABSTRACT FOREDRAG:
On-line estimation in an Industrial Semibatch Reactor
(A critical afternote)
David I. Wilson
Swiss Federal Institute of Technology
Zurich
Modern control techniques are not yet routinely implemented in
the chemical processing industries. Problems hindering application
are discussed, and will focus on a specific example from the
Swiss pharmaceutical industry. It will be argued that model
based estimation techniques applied at the industrial scale
present many difficulties that are typically inadequately
addressed in simulation studies. This work assesses candidly
the feasibility of on-line state estimation applied to an
industrial production 1m3 semibatch reactor.
In a joint venture with CIBA-Geigy, we applied estimation
techniques, (in the form of an extended Kalman filter),
devised to predict the batch end-point using only standard
instruments coupled with a simple fitted model. This venture
was successful, and was the first application of model based
estimation at CIBA. However the reasons for success were not
so much due to the quality or suitability of the theory, but
more the quality of our open loop model and attention to detail.
We believe that these conclusions hold also for other reported
applications and therefore caution that it may be difficult to
find suitable industrial estimation applications where errors
in process dynamics and measurements are of similar magnitude
allowing significant estimator improvement potential.
Other application considerations are that all models should
be kept simple, for ease of maintenance and portability, and
to ensure robust numerical properties. One must install
sufficient instrumentation to log both inputs and outputs
and that extensive preprocessing of these measurements is
required. Minor modifications to the estimator algorithm
are necessary to prevent possible divergence and these
modifications were found to be more effective than the
traditional divergence control approaches such as artificially
elevating filter tuning parameters.
.......................................................................
David Wilson, PhD tel: +46 54 838267
Lecturer in Control Engineering fax: +46 54 838165
Department of Technology email: david.wilson@hks.se
University of Karlstad
S-651 88 Karlstad, SWEDEN
.......................................................................