Chapter 4
SISO IMC DESIGN FOR STABLE
SYSTEMS
‘The IMC design procedure consists of two steps
STEP 1: Nominal Performance
Z is selected to yield a “good” system respouse for the input(s) of interest,
without regard for constraints and model uncertainty.
STEP 2: Robust Stability and Performance
is augmented by a low-pass filter f (q = Gf) to achieve robust stability and
robust performance.
‘We will proceed with a detailed discussion of these two steps.
4.1 Nominal Performance
For SISO systems we will generally choose G such that it is Hy optimal for a
particular input v where v = dor v = —r. Thus has to solve
aja = mu 0 — Bolle (1-1)
subject to the constraint that @ is stable and causal. Problem (4.11) reaches its
absolute minimum (zero) for
a= (41-2)
However, the model inverse is an acceptable solution only for MP systems. For
NMP systems the exact inverse (4.1-2) is unstable and/or noncausal. The ob-
jective function cannot be made zero and an “approximate inverse” of j has to
be found such that the weighted 2-norm of the sensitivity function is minimized.
Note that for NMP systems the optimal solution depends on the weight (the
input v), while for MP systems (4.1-2) is optimal independent of the weight.
a8 CHAPTER 4. SISO IMC DESIGN FOR STABLE SYSTEMS
4.1.1 Hy-Optimal Control
In the next chapter we will derive a fairly simple general analytic procedure to
solve the problem (4.1-1). In this chapter we will just state the solution and
demonstrate the results for some important special cases.
‘Theorem 4.1-1. Assume that p is stable. Factor p into an allpass portion Ba
and a MP portion pig
Badu (41-3)
so that fy includes all the RHP zeros and delays of and
Waliw)|=1 Ww (41-4)
In general a has the form
op aatG
tet
‘where the superscript H denotes complez conjugate.
Factor the input v similarly
Pals) = Re(G), @>0 (41-5)
v= vary (41-6)
The controller which solves (4.1-1) is given by
= Gun) (Baew}, (41-7)
where the operator {:}, denotes that after a partial fraction expansion of the
operand all terms involving the poles of 3! are omitted.
In general the optimal controller g is not proper. This is of no concern at this
point because it is to be augmented by a low-pass filter in Step 2 of the IMC
design procedure.
Note that the objective function (4.1-1) and the optimal controller (4.1-7)
are unaffected by NMP elements in v. For example, for the design of G and
the optimal closed-loop performance it, makes no difference if a step disturbance
centers directly at the plant output or after passing through a time delay. We will
earn in Chap.6 that this is exactly the situation when feedforward control can
be used effectively.
‘The analytic expression (4.1-7) allows us to make some general observations re-
‘garding the pole-zero configuration of the closed-loop system with the Hz-optimal
controller q.
‘Theorem 4.1-2. The Hy-optimal complementary sensitivity function
a
has the following properties:
aval {y'vm}, (41-8)4.1, NOMINAL PERFORMANCE 59
1. The poles of are at the mirror images of the plant RHP zeros and at the
zeros of tw.
2. has RHP zeros at the plant RHP zeros. If ij has any additional RHP zeros
they are at mirror images of zeros of vss.
9. If'vy has no finite zeros, then f} has no RHP zeros except those of the plant.
This implies that g is MP.
Proof.
1. The potential poles of 7 are the poles of pa, the zeros of vyy and the poles of
{#;'vw}s, which are the poles of vy. However, the poles of vyy are cancelled
by zeros of uj} premultiplying {-}..
2, Assume that qi is stable and minisizes
IQ - Baul (41-9)
‘The RHP zeros of j appear in #] = ii in the form of an allpass. Define
new plant = fiji. The stable Gy which minimizes
MW = Bee)ulle (4.1-10)
is unity (if it were not, di could not be optimal). The RHP zeros of p have to
appear in P= 7 im the form of an allpass. Because of 1. all RHP
zeros in addition to the plant RHP zeros have to be located at the mirror
images of the zeros of uur.
3. Follows directly from 2. o
RHP zeros are generally considered as “bad” for performance. However, 2. im-
plies that depending on the input viz the optimal controller g can sometimes add
REP zeros,
Example 4.1-1. p = 9,
how =
s [et2st+1) 5 1 =s41
MASE aed
5 (ce4 2-041) taht
(e+ 26+)
‘The controller j added a RHP zero at (+1,0). As predicted by Thm. 4.1-2 it is
at the mirror image of the disturbance zero at (-1,0). o