|
For the purpose of achieving an economical production scale,
the newly-designed chemical processes tend to be much larger
and more complex. Also, to meet the need for optimizing
performance, the demand for tighter control has become a trend
in modern plants. As a result, probability of faults and/or
operational problems in chemical industries have increased
significantly in recent years. Therefore, there is a real
incentive for the development of automatic fault detection and
diagnosis techniques to be used as an aid in plant operation.
of our research is to assess the feasibility of adopting, The
objective artificial neural networks (ANNs) in fault detection
and diagnosis for dynamic systems}. Although there is a large
volume of related publications avail- able, most of them used
steady-state data to train ANNs and, as such, the task of fault
diagnosis can only be carried out after reaching a new steady
state. To avoid this drawback,the two-stage ramework proposed
by Isermann (1982) and Panossian (1988) was utilized to
incorporate two ANNs in series in our study. On the of the
main advantages of using an ANN is its ability to perform
complex functional mapping without formulating accurate
mathematical models. Once trained, ANNs can be implemented
accord -ing to on-line data. Therefore, in order to verify the
feasibil- ity of the proposed approach, a pilot plant, which
simlulates the operation of pipeline networks, has been
assembled in our laboratory.
|