Professional Documents
Culture Documents
I. I NTRODUCTION
This report deals with the problem of designing robust
controller for a magnetic suspension system. Feedback control
is essential for magnetic suspension systems because they
are unstable systems naturally. The mathematical model of
these systems have various uncertainties such as parame-
ters identification errors, unmodeled dynamics, and neglected
nonlinearities. Because of these uncertainties in modeling of
the system, the controller is required to have robustness for
stability and performance.
The rest of this report is organized as follows. In section II,
Fig. 1. Schematic diagram of the system [1].
the mathematical modeling of the system with uncertainties
are discussed. In section III, H ∞ and μ-synthesis controller Multiplicative Uncertainty Model
0
designing approaches are presented and simulation results 10
10
TABLE I
PARAMETERS OF ELECTROMAGNETIC SUSPENSION SYSTEM
II. M ODEL S TRUCTURE Fig. 2. (P (jω) − Pnom (jω))/Pnom (jω) and Wm (jω) versus ω for
A. Electromagnetic Suspension System one hundred random parameters.
4
{P (jω) − P̃ (jω)}
−2
−4
−6
−8
−0.5 0 0.5 1 1.5 2 2.5
{P (jω) − P̃ (jω)} −4
x 10
Fig. 3. The Nyquist diagram for the random transfer functions and the weight Fig. 5. Interconnection Structure
function for additive uncertainty.
Nominal H∞ Controller
−6 5
10
−7
10 4
P (jω) − P̃ (jω)
−8
10
μ
3
−9
10
−10 2
10
−11
10 1
−12
10
0
−2 0 2 4
−13 10 10 10 10
10 ω
−2 0 2 4
10 10 10 10
ω
• Magnetic permeability of the electromagnet is infinity. can be neglected. In view of these equations, the linear time-
• Magnetic flux density and magnetic field have not hys- invariant mathematical model is given by
teresis, and they are not saturated. ⎡ ⎤ ⎡ ⎤
0 1 0 0
• Eddy current in the magnetic pole can be neglected.
A = ⎣ Kx 0 Ki ⎦ , B = ⎣ 0 ⎦ , (6)
1
Using these assumption and considering the physical nature 0 0 − LR0 L0
of the system, we have
2 C= 1 0 0 , (7)
d x
m = mg − f, (3)
dt2 where Kx = 2kI 2 /(m(X + x0 )3 ) and Ki = −2kI/(m(X +
2 x0 )2 ). Now, because of the uncertainties of the parameters in
i these equation, a multiplicative uncertainty model is added to
f =k , (4)
x + X + x0 the system. Using the nominal parameters for the system we
have
d −77.63
e = Ri + [L(x)i], (5) Pnom (s) = (8)
dt (s + 51.65)(s − 51.65)(s + 31.67)
where k, X, and x0 are constant parameters related to the We must select Wm (s) such that
operating point. At this point we will assume that the coil
inductance constant near an operating point (L(x) = L 0 ) P (s) − Pnom (s)
= Wm (s)Δ(s), Δ(s) ≤ 1. (9)
and the electromotive forces due to the differential of the gap Pnom (s)
3
Nominal H Controller
∞
1
2) Ideal Mathematical Model - L(x): In this subsection, I
assume that the coil inductance L(x) is a function of a gap x,
0.9
and written as follows [1]
0.8
2k
0.7 L(x) = , (12)
x + X + x0
0.6
Using the above equation we get
μ
0.5
2ki dx 2k di
e = Ri − + + L 0 ,
0.4
(x + X + x0 )2 dt x + X + x0 dt
0.3 (13)
0.2
using this equation results in the following linear, time-
invariant model of the system
0.1 ⎡ ⎤
0
0 1 0
−2
10 10
0
ω
10
2
10
4
A = ⎣ Kx 0 Ki ⎦ , (14)
0 Kv − R L
Fig. 7. μ analysis of the closed loop system using D-K iteration (second where
iteration). The robust performance (green line), robust stability (blue line), 2k
and nominal performance (red line) are shown in this figure. L= + L0 , (15)
x + X + x0
and
2ki
Step Response
Kv = . (16)
1 L(x + X + x0 )2
0.8 We call this an unmodeled nonlinearity in the system, and
0.6
we can model it with an additive uncertainty block. Figure
4 exhibits P (jω) − Pnom (jω) and Wa (jω) versus ω for
0.4
one hundred random parameters K v ∈ [35.0137, 42.7945],
and L − L0 ∈ [0.10, 0.13]. Here it should be noted that the
Amplitude
0.2
System: untitled1
Final Value: −0.00493
0 magnitude of the uncertainty weighting W a covers all the
System: untitled1
Settling Time (sec): 0.146
model perturbations shown in Figure 4. The red line is based
−0.2
on
s(s/9 + 1)(s/600 + 1)3
−0.4
A. Control Objective
at each frequency. Considering Δ(s) ≤ 1, results in
Electromagnetic suspension system is essentially unstable.
P (jω) − Pnom (jω)
We must design a robust controller to stabilize the closed-loop
≤ Wm (jω), (10)
Pnom (jω)
system; furthermore, we would like to design a controller to
maintain the performance against uncertainties. Figure 5 ex-
for all ω ∈ R. Figure 2 exhibits (P (jω) − hibit the interconnection structure of the system. To follow the
Pnom (jω))/Pnom (jω) and Wm (jω) versus ω for one command at low frequency band, the performance weighting
hundred random parameter K x ∈ [3.09 × 103 , 3.78 × 103 ], function Wp (s) is chosen as
Ki ∈ [−16.42, −13.43], R ∈ [25.6, 27.6], and
20
L0 ∈ [1.08, 1.23]. Here it should be noted that the Wp = . (18)
magnitude of the uncertainty weighting W m covers all the s+1
model perturbations shown in Figure 2. Figure 3 shows the
B. H∞ Design
Nyquist diagram for the random transfer functions and the
weight function. The blue one is based on random transfer An H∞ nominal performance is designed for the plant in
functions and in each frequency these blue circle exhibit a this subsection. For designing the controller we assume that the
measure of the uncertainty in that frequency. The red circles uncertainties are zero. With assumption the closed loop system
are based on satisfies the nominal performance, but it fails to satisfy robust
stability and robust performance. As you can see in figure
(s/60 + 1) 6, this controller does not have robustness for stability and
Wm (s) = .175 . (11)
(s/30 + 1) performance.
4
(s + 1.52 × 102 )(s + 1.092 )(s + 8.30 × 10−1 )(s + 4.52 × 10−2 )(s + 4.26)(s + 2.49 × 101 )
×(s + 2.12 × 101 )(s + 3.00 × 101 )(s + 5.86 × 101 )(s + 5.00 × 101 )
×(s + 4.99 × 101 + i2.31 × 10−2 )(s + 4.99 × 101 − i2.31 × 10−2 )(s + 4.77 × 101 )
K2 (s) = −1.56 × 106 . (24)
(s + 3.36 × 102 + i3.99 × 102 )(s + 3.36 × 102 − i3.99 × 102 )(s + 1.24 × 102 + i2.53 × 101 )
×(s + 1.24 × 102 − i2.53 × 101 )(s + 5.14 × 101 + i2.09)(s + 5.14 × 102 − i2.09)
×(s + 4.83 × 101 )(s + 2.84 × 101 )(s + 5.97)(s + 1.01)(s + 1.76 × 10−1 )(s + 4.52 × 10−2 )
C. μ-Synthesis 2
ΔP = ⎣ 0 Δm 0 ⎦ : Δa ∈ C, Δm ∈ C, Δp ∈ C ,
⎩ ⎭ 0.5
0 0 Δp
(19)
e(t)
0
where C denotes the set of complex numbers. Next we
consider the generalized plant partitioned as −0.5
−1
P11 P12
P = . (20)
P21 P22 −1.5
Fl (P, K) := P11 + P12 (I − P22 K)−1 P21 . (21) Fig. 9. The closed loop system step response (from input to error) for
the initial H∞ controller (red color) and the step response for the closed
loop system with μ-synthesized controller (blue one) with the worst-case
Considering the equation (21) the robust performance and perturbation.
stability is equivalent to the following μ test