International Conference

on
POWER ELECTRONICS AND INTELLIGENT
CONTROL FOR ENERGY CONSERVATION
Warsaw, Poland,
16-19 October, 2005

—————————————————————
Rotor Flux and Speed Observers for Induction
Motors
—————————————————————

Chunki Kwon, Stefen Hui, Scott Sudhoff, and Stan Żak

1
Outline

• Brief overview of the observer idea

• The induction motor model

• Examples of flux and rotor speed observers

– Observer architecture of Derdiyok, Güven, Rehman, Inanc and Xu
(2002)
– Sliding mode observer of Utkin, Guldner and Shi (1999)

• Our observer
– Analysis
– Simulation and experimental results

• Concluding remarks

2
Observer and its use

• Observer—an auxiliary dynamical system that estimates the

system state based on the system inputs and outputs

• Many controllers need the whole state vector of the con-

trolled system

• The observer provides a solution to the problem of incom-

plete state vector information

• An observer can also be used to augment or replace sensors,

and in fault detection and isolation (FDI) systems

3
A trivial observer

A trivial observer (open-loop observer)—a system copy as an

observer.
d (x̂ − x) = A(x̂ − x)
Observation error, e = x̂ − x; hence, dt
4
Problems with the open-loop observer

• Observation error dynamics,

d
(x̂ − x) = A(x̂ − x)
dt

• The observation error tends to zero only if the observed sys-

tem is stable

• There is no control over the observation error dynamics

• There is a fix—add an observer compensator to get the

closed-loop observer

5
General closed-loop observer

6
Luenberger’s closed-loop observer

x̂˙ = Ax̂ + Bu + L(y − ŷ )

Observation error dynamics, dtd (x̂ − x) = (A − LC )(x̂ − x)

7
Combined observer-controller compensator

• Works well for linear systems without uncertainties

• What about nonlinear or uncertain systems?

8
Induction motor model—parameters

Rr resistance of the rotor windings

Rs resistance of the stator windings
Lr self-inductance of the rotor windings
Ls self-inductance of the stator windings
Lm mutual inductance of the rotor and stator windings
np number of the pole pairs
D viscous friction coefficient
η =R Lr
r

L2m
σ =1− Lr Ls
leakage parameter
β = σLLrmLs
µ = 32 nJL
p Lm

r 
1 L2m
γ = σLs Rs + L2 Rr
r

9
State and input variables

ω rotor angular speed

λar rotor flux linkage of phase a
λbr rotor flux linkage of phase b
ias stator current in phase a
ibs stator current in phase b
uas stator voltage of phase a
ubs stator voltage of phase b
τL load torque

10
The 2-phase cage induction motor model

dω D τL
= µ(λar ibs − λbr ias) − ω −
dt J J
dλar
= −ηλar − npωλbr + ηLmias
dt
dλbr
= −ηλbr + npωλar + ηLmibs
dt
dias 1
= ηβλar + npβωλbr − γias + uas
dt σLs
dibs 1
= ηβλbr − npβωλar − γibs + ubs
dt σLs

11
Induction motor model representation for the ob-
server construction

dλr
= −Aω λr + ηLmis
dt
dis 1
= β Aω λr − γ is + v s,
dt σLs
where
" #
η np ω
Aω =
−npω η

12
The current observer of Derdiyok et al. (2002)

Recall,
dis 1
= β Aω λr − γ is + vs
dt σLs
Proposed observer
dîs 1
= β u − γ îs + vs
dt σLs

13
Sliding surface and switching functions

Sliding surface
" # " #
sa îas − ias
s= = =0
sb îbs − ibs
Switching functions
" # " #
sign(sa) sign(îas − ias)
u = −u0 = −u0
sign(sb) sign(îbs − ibs)

14
Dynamical behavior of the observer of Derdiyok
et al. (2002)

The observer
dîs 1
= β u − γ îs + vs
dt σLs

The surface s = 0 is attractive for sufficiently large gain u0.

To obtain a bound on u0, use the generalized Lyapunov function

candidate
1 T
V = s s.
2
Find the Lyapunov derivative of V , that is, the time derivative
of V evaluated on the trajectories of the observer.
15
Analysis of the current observer of Derdiyok et
al. (2002)

V̇ = sT ṡ
where
d d
ṡ = îs − is
dt dt
1 1
= β u − γ îs + v s − β Aω λr + γ is − vs
σLs   σLs
= β (u − β Aω λr ) − γ îs − is

16
Lyapunov’s derivative

Therefore, V̇ < 0 if
 T  2
  γ
β  îs − is Aω λr  − β îs − is
 
u0 >    
îsa − isa + îsb − isb
   

We showed that the right-hand side is bounded in sliding mode.

17
Flux and rotor speed observer of Utkin, Guldner
and Shi (1999)

Assume that ω = constant. Consider a copy of the induction

model subsystem,
dλ̂r
= −Âω λ̂r + ηLmîs
dt
dîs 1
= β Âω λ̂r − γ îs + v s,
dt σLs
where
" #
η npω̂
Âω =
−npω̂ η

18
Rotor speed observer of Utkin et al. (1999)

Let
ωe = npω
Then, Utkin, Guldner and Si (1999) propose to use as the rotor
speed estimate
ω̂e = ω0sign(sn),
where
   
sn = îbs − ibs λ̂ar − îas − ias λ̂br
and
" # " #" # " #
d λ̂r −Âω ηLmI 2 λ̂r O 2×2
= + 1 I vs
dt îs β Âω −γ I 2 îs σLs 2

19
Our observer—another representation of the in-
duction motor model
 
dω  
0 0
  
dt −D
J 0 0 0 0 ω
dλar
 
 0 0
 
 0 −η 0 ηLm 0 λar
    
dt
    " #
  u
dλbr as
 0 0
    
  =  0 0 −η 0 ηLm  λbr + 
dt  ubs

      1
 dias 
 0

ηβ 0 −γ 0  ias   σL
s
0 
dt
   
1
 
dibs 0 0 ηβ 0 −γ ibs 0
 
dt σLs
 
µ 0
0  
 0 −np − − 1 τ
0  λ i λ i
 
  ar bs br as µJ L 
+
 0 0 np  
 ωλbr 

 0 n β 0 
p ωλar


0 0 −npβ
= Ax + B 1u1 + B 2u2(x)

20
Analysis of the fifth order model

• The pair (A, C ) is detectable but not observable

• The matrix A has all its eigenvalues in the open left-half

complex plane

• We assume that the function u2 (representing the nonlin-

earities of the model) satisfies a Lipschitz condition in some
neighborhood of the origin, that is, there exists a positive
constant κ such that for any x and x̃ in a neighborhood of
the origin,
ku2(x̃) − u2(x)k ≤ κkx̃ − xk

21
Proposed observer architecture

• Induction motor model: ẋ = Ax + B 1u1 + B 2u2(x)

˙ = (A − LC )x̂ + B 1u1 + Ly + B 2u2(x̂)

• Observer: x̂

• Estimation error: e(t) = x̂(t) − x(t)

• Estimation error dynamics:

ė(t) = x̂˙ (t) − ẋ(t)

= (A − LC )e(t) + B 2 (u2(x̂) − u2(x))

22
Analysis of the observer

• (A − LC ) is asymptotically stable, hence for any Q = QT > 0,

there exists P = P T > 0 such that

(A − LC )T P + P (A − LC ) = −Q

• Consider the Lyapunov function candidate for the error sys-

tem,
V (e) = eT P e.

• Compute the Lyapunov derivative of V

23
Computing the Lyapunov derivative of V

• Evaluate the time derivative of V on the dynamics of the

observation error,

V̇ (e) = 2eT P ė
= 2eT P (A − LC )e + 2eT P B 2 (u2(x̂) − u2(x))
≤ −eT Qe + 2kekkB 2kλmax(P )ku2(x̂) − u2(x)k

• Take into account the Lipschitz condition,

V̇ (e) ≤ −λmin(Q)kek2 + 2κkB 2kλmax(P )kek2

= − (λmin(Q) − 2κkB 2kλmax(P )) kek2

24
Making the Lyapunov derivative negative definite

• Recall,

V̇ (e) ≤ −λmin(Q)kek2 + 2κkB 2kλmax(P )kek2

= − (λmin(Q) − 2κkB 2kλmax(P )) kek2

• If
λmin(Q)
κ< ,
2kB 2kλmax(P )
then V̇ (e) < 0 and e = 0 is an asymptotically stable equilib-
rium state of the error

25
Simulation and the first experimental setup

idc Inverter Induction Motor

A
+ + CH1
iai ias
u abi +
T1 T2 T3 -
-
u dc Cdc A w
+ CH2
ibi +
ibs
T4 T5 T6 + ics
ubci
-
-

s*abci

odrv
i edqi
* i edqi
**
i *abci
0 Te* Current Delta
MTPA SCR
Limit Modulator
Te,ref
o drv w s*
o drv

(
iabi

w np å
ò qe
we reset
2p
26
( Proposed Observer
æ u abi ö
çç ( ÷÷ K ss,v Y-Ä
è ubci ø u1 x&ˆ = Axˆ + B1u1 + B 2 u 2 (xˆ , ôˆ L ) x̂

tˆL
Simulation studies—Same initial conditions for the observer and
the motor model

4000 50

speed error (rpm)

Speed (rpm)

3000

2000 0
measured
estimated
1000

0 −50
0 1 2 3 4 0 1 2 3 4

2 1

1 error (vs) 0.5

(Vs)

0 0
αr
λ

αr

−1 −0.5
λ

−2 −1
0 1 2 3 4 0 1 2 3 4

50
50
error (A)
iαs (A)

0 0
αs
i

−50
−50 27
0 1 2 3 4 0 1 2 3 4
Simulation studies—Different initial conditions for the observer
and the motor model

1000 10

speed error (rpm)

Speed (rpm)

500 0
measured
estimated −5

0 −10
0 1 2 3 4 0 1 2 3 4

2 2

1 1
λαr error (vs)
(Vs)

0 0
αr
λ

−1 −1

−2 −2
0 1 2 3 4 0 1 2 3 4

100 50

50
error (A)
(A)

0 0
αs
i

αs

−50
i

−100 −50 28
0 1 2 3 4 0 1 2 3 4
Experimental verification—rotor speed varies from 500 rpm to
900 rpm

1000 50

speed error (rpm)

Speed (rpm)

500 0

0 −50
0 5 10 15 20 0 5 10 15 20

2
1
λαr (Vs)

0 measured
estimated
−1
−2
0 5 10 15 20

100
50
50
error (A)
iαs (A)

0 0
αs

−50
i

−50 29
−100
0 5 10 15 20 0 5 10 15 20
Second experimental setup

idc Inverter Induction Motor

A
+ + CH1
iai ias
u abi +
T1 T2 T3 -
-
u dc Cdc A w
+ CH2
ibi +
ibs
T4 T5 T6 + ics
ubci
-
-

s*abci

odrv
i edqi
* i edqi
**
i *abci
0 Te* Current Delta
MTPA SCR
Limit Modulator
Te,ref
o drv w s*
o drv

(
iabi
Proposed Observer

ŵ np å
ò qe x&ˆ = Axˆ + B1u1 + B 2 u 2 (xˆ , ôˆ L )
we reset
2p
30
(
æ u abi ö
çç ( ÷÷ K ss,v Y-Ä
è ubci ø u1

tˆL
Experimental results—performance study of MTPA control strat-
egy with rotor speed sensor

1.06
0.9 times ωs*
Measured torque / torque command

1.0 times ωs*

1.04
1.1 times ω *
s

1.02

0.98

0.96

0.94

50 100 150 200 31

Torque command, T * (Nm)
e
Experimental results—performance study of MTPA control strat-
egy with the proposed observer incorporated into the controller

1.06 0.9 times ω *

s
1.0 times ω *
s
1.1 times ω *
Measured torque / command torque

1.04 s

1.02

0.98

0.96

0.94

50 100 150 200 32

Torque commanded, T * (Nm)
e
Experimental results—combined performance study of MTPA
control strategy with rotor speed sensor

0.9 ωs* with ω measured

1.06 1.0 ωs* with ω measured
1.1 ω * with ω measured
s
1.0 ω * with ω estimated
Measured torque / torque command

1.04 s

1.02

0.98

0.96

0.94

20 40 60 80 100 120 140 160 180 200 33

Torque command, Te* (Nm)
Concluding remarks

• Utkin et al. (1999), Kubota et al. (1993) and Derdiyok et al.

(2002) use fourth-order model to construct their observers
for the flux and then separately construct speed estimators

• Analysis of the guaranteed performance needed

• We propose to use the fifth-order model

• The problem od the zero speed is still open.

34