Escolar Documentos
Profissional Documentos
Cultura Documentos
Smith predictor
You may have a process tf G ( s )
G F
1 G F H
Q
1
G 1 Q H
1
G0 e
ds
Q0 e
1 Q0 e
G0 F0
1
1 G0 F0 H
G0 F0
G0 1
e
1 G0 F0 H
ds
ds
ds
F0
1 G0 F0 H G0 F0 e
d s
F0
1 F0 ( G0 H (1 e
d s
))
This is now presented below within the overall system between e and u in a possible form, the
content of the inner clauses for comparison of 1 and e-ds implemented within a box bounded by
dashed (- -) lines:
u
F0
(d)
w
G0
G0
H
198
y
(d)
F0
G0
(d)
G0
Above the delay is removed from the open loop system. This could allow use of Routh test to test
the stability of the left sub-system which includes the feedback. Root Locus could also be easily
used? However, in practice things are not so beautiful but modelling errors may imply a controller
not able to remove the delay from inside the loop. In addition, for some systems there may be other
disappointments. Note also that this was tf play without too much promised for natural responses ...
In an ad hoc derivation of a non-minimal Smith predictor one first cancels the true negative
feedback by a corresponding positive feedback from the controller output and then implements a
new negative feedback from the controller output without a delay sub-system:
u
F0
y
(d)
G0
H
H
w
G0
G0
(d)
H
Here H and G0 appear twice in the controller. Hence this implementation of the controller is not a
minimum realization. However, interchanging the order of (d) and H in the middle feedback path
opens a possibility to reach the implementation initially presented:
(d)
w
G0
F0
G0
G0
y
(d)
H
Now first remove the latter sharing point of the controller but pick the input of (d) from the output
of the H block in the innerst feedback path & remove the blocks H and G0 of the middle path &
destroy the positive feedback but subtract the output of (d) from output of the remaining H block.
199