2010 Conference on Control and Fault Tolerant Systems

Nice, France, October 6-8, 2010

WeC1.1

Fault Tolerant Control of a Quadrotor UAV using Sliding Mode

Control
Farid Sharifi, Mostafa Mirzaei, Brandon W. Gordon, and Youmin Zhang


AbstractIn this paper, the sliding mode approach is used to

control of a quadrotor unmanned aerial vehicle (UAV) in the
presence of external disturbance and actuator fault. Fault
detection unit can detect the actuator fault using a state
estimator. Then it reconfigures the structure of controller such
that some control performance is achieved. The proposed
control structure has the advantage of disturbance rejection in
the fault-free condition. Moreover it can recover some of
control performances when a fault occurs. Different
simulations have been carried out to show the performance and
effectiveness of the proposed method.

I. INTRODUCTION

nmanned aerial vehicles (UAVs) have gained

increasing interest in recent years because of a wide
area of possible applications such as security, traffic
surveillance, management of natural risks, environment
exploration, agriculture, and military. For these applications,
the ability of helicopters to take off and land vertically, to
perform hover flight, as well as their agility, make them
ideal vehicles.
One type of aircraft with a strong potential for both indoor
and outdoor flight is the rotorcraft and the so-called
quadrotor helicopter has been chosen by many researchers
as the most promising vehicle [1]-[3]. A quadrotor is an
aircraft that is lifted and propelled by four rotors. Control of
quadrotor can be achieved by varying the relative speed of
each rotor to change the thrust and torque produced by each.
Quadrotors are classified as rotorcraft, as opposed to fixedwing aircraft, because their lift is derived from four rotors.
The use of four rotors allows each individual rotor to have a
smaller diameter than the equivalent single-rotor helicopter,
allowing them to store less kinetic energy during flight and
thus reduces the damage caused by the rotors hitting any
objects. By enclosing the rotors within a frame, the rotors
can be protected during collisions [4].
Different control methods mostly designed for quadrotor
UAV are feedback linearization [5]-[7], backstepping [8],
and sliding mode control [8], [9]. Some other methods also
used for linearized model of quadrotor in literature such as
PID [10] and LQR control [11]. In this paper, sliding mode

This work was supported in part by Natural Sciences and Engineering

Research Council of Canada (NSERC) through a Strategic Project Grant
and a Discovery Project Grant.
F. Sharifi, M. Mirzaei, B. W. Gordon, and Y. M. Zhang are with the
Department of Mechanical and Industrial Engineering, Concordia
University, 1515 St. Catherine, Montreal, Quebec, H3G 1M8, Canada (emails:
[email protected],
[email protected],
[email protected], [email protected]).

978-1-4244-8154-5/10/$26.00 2010 IEEE

has chosen as controller because of its insensitivity to the

model errors, parametric uncertainties and external
disturbances. This sliding mode controller is able to make
the quadrotor reach a desired height with desired angles.
Fault tolerant control system (FTCS) is a control system
with the ability to tolerate faults automatically and continue
its operation in the event of a failure in some of its
components [12]. FTCS can be classified into two different
types which are known as Passive FTCS and Active FTCS.
Passive FTCS can tolerate a predefined set of faults by using
a specially-designed fixed controller while active FTCS
relies on fault detection and diagnosis (FDD) process to
monitor system performance and to detect and identify faults
in the system and the controller is reconfigured on-line and
in real-time [12-14]. Modern UAVs need to be designed to
achieve the desired performance under both normal and fault
conditions. A relatively small amount of existing research
has addressed the FTCS for quadrotor UAVs. Some of few
works addressing this issue are multi-observer switching
strategy [15], Thau observer for systems with sensor partial
failures [16], and gain scheduling based PID controller [17].
In this paper, with the aid of an observer, partial faults in
actuator can be detected and then controller reconfigures in
such a way to maintain stability and performance of the
faulty system.
This paper is organized as follows. In Section II, the
nonlinear model of a quadrotor UAV and dynamic of
actuators are presented. The sliding mode control for the
quadrotor is designed in Section III. The proposed fault
detection and fault tolerant control strategy is described in
Section IV. Section V is devoted to the presentation of the
simulation results obtained for various fault-free situations
and fault scenarios when the proposed scheme is applied to
the quadrotor UAV. Finally, conclusion is provided in
Section VI.
II. MODEL OF THE QUADROTOR UAV
A. Nonlinear model
In this section, the general dynamic model of a quadrotor
UAV has been studied. In lightweight flying systems, the
dynamic model ideally includes the gyroscopic effects
resulting from both the rigid body rotation in space, and the
four propellers rotation.
The dynamic model is derived using Euler-Lagrange
formalism. A body-fixed frame B and the earth-fixed frame
E are assumed to be at the center of gravity of the quadrotor
UAV, where the z-axis is pointing upwards, as seen in
Figure 1. The position of the quadrotor UAV in earth frame

239

is given by a vector ( x, y , z ) . The orientation of quadrotor

UAV that referred to as roll, pitch, and yaw is given by a
vector (I , T ,\ ) which measured with respect to the earth
coordinate frame E.

the rigid body rotation, the gyroscopic effect resulting from

the propeller rotation coupled with the body rotation, the
actuators action and finally the drag effects. Using the
Lagrangian method, quadrotor rotational dynamic model is
as follows:
  ( I  I )  J TZ
  lU ) I  K I
I (T\
y
z
1
x
4
  ( I  I )  J IZ
  lU ) I  K T
T (I\
z
x
2
y
5

(3)

 ( I  I )  U ) I  K \
(IT
x
y
3
z
6

\

where K i is the drag coefficient, and systems inputs

U1 , U 2 , U 3 , U 4 are defined as follows:

Fig. 1. The structure of quadrotor and its frames

The transformation of vectors from the body-fixed frame

to the earth-fixed frame can be obtained based on Euler
angles and the rotation matrix REB .
REB

C\ CT
S C
\ T
 ST

C\ SI  S\ ST CI

CI CT

 S\ CI  C\ ST SI

S\ SI  C\ ST CI

C\ CI  S\ ST SI
CT SI

(1)

1, 2, 3, 4 ,

is Fi bZi2 where b is the trust factor and Zi is the speed of

rotor i . In the body-fixed frame, the forces are defined as
follows:
FB

0
4
i 1 Fi

FxB
F
yB
FzB

REB .FB

4
i 1

F (C S C

F (S S C

FC C
4

Fx  K 1 x
Fy  K 2 y
 mg  K 3

z

i 1

i 1

U4

b(Z1  Z2  Z3  Z4 )

(4)

4
i 1

Fi

Z4  Z3  Z2  Z1 is considered as a disturbance.

b b

d d

b b

0 Z22

2
Z
: 2
2
Z32
 d Z3
2

b Z42
Z4

0
d
b

can

find

Zj ,

appropriate

while

1, ..., 4

rewritten in the state-space form x

11

11

11

11

x

12

 S\ SI )  K1 x

 mg  K 3 z

10

10

10

(2)

12

12

where K i is the drag coefficient. Note that these coefficients

are negligible at low speed.
The quadrotor dynamic model describing the roll, pitch
and yaw angles contains the gyroscopic effect resulting from

other

f ( x, u ) as follows:

(C S C  S S ). U m  K x
x
x
x
x
x

(C x S x S x  S x C x ). U m  K x

 g  (C x C x ). U m  K x

( x x ( I  I )  Jx Z  lU ) I  K x

( x x ( I  I )  Jx Z  lU ) I  K x

(x x (I  I )  U ) I  K x
7

the

inputs, U k , k z i , do not change. Finally, this model can be

 C\ SI )  K 2 y

(5)

Since : is nonsingular, it is obvious that for each U i we

i 1

d (Z1  Z2  Z3  Z4 )

U
2
U 3

U 4

Therefore, the equation of motion ( x, y, z ) in the earthfixed frame are represented as:

x
m 
y

z

U3

F2  F1

b(Z2  Z1 )

S S C S C

F ) C S  S S C

CC

The relation between systems inputs and speed of rotors

can be shown in the matrix form as follows:
2
Z12
U 1 0 0 b b Z1

In the earth-fixed frame, the forces can be described as:

F
F

F

F4  F3

U2

and Z

sin() and cos() , respectively. It is important to note that

REB R .
The trust force generated by rotor i , i

where b and d are the trust and drag factor respectively,

where the abbreviations S() and C() have been used for
T
BE

b ( Z 4  Z3 )

U1

10

where u
and x

240

12

10

(6)

(U1 , U 2 , U 3 , U 4 ) is the vector of input variables

( x, x , y, y , z , z, I , I, T , T,\ ,\ ) , x 12 is the vector

of state variables.

B. Actuator dynamics
The thrust generated by each propeller is modeled using the
following first-order system
Z0
(7)
Fi K
ui
s  Z0

where I1

I y  Iz
Ix

( sI sI  K sI )

condition

will be used to represent the actuator dynamics, which is

defined as follows:

Z0

vi

s  Z0

ui

surface sI

0
U1

III. SLIDING MODE CONTROL

In this section, a sliding mode controller is used to control
the height and rotational angles of quadrotor.
Consider dynamic equation of roll angle
I  Iz
J 
l
 ( y
(9)
 U 1  K 4I  g1
I T\
)  TZ
Ix
Ix
Ix
where U1 is the control input, I is the output of interest,
f1

J 
TZ
Ix

is the dynamic associated with disturbance

which is unknown but assumed to be bounded by f1 , and

g1 is the dynamic which is not exactly known but assumed

to be bounded by g1 . In other words, the nominal dynamic
equation of roll angle is given as follows:
I  Iz
l
 ( y
I T\
)  U 1  K 4I
Ix
Ix

(10)

where zI
sI

Id  I

So by letting

 (
T\

Ix

)  f1 

Ix

Main drawback of sliding mode control is the chattering

effect produced by the discontinuous function sign. To
eliminate this, the sign function is replaced with a saturation
function sat defined as follows:
sign( s ) if s ! U
sat ( s )
if s  U
s U
where U is a boundary layer around the sliding surface s .
The same steps are followed to extract U 2 and U 3 as
follows:
Iy
U2
x9 d
(  x8 x12 I 2  K 5 x10  
l
 OT ( x9 d  x10 ))  k 2 sat ( sT )
(15)
I z (  x8 x10 I 3  K 6 x12  
x11d

U3

 O\ ( x11d  x12 ))  k3 sat ( s\ )

zI  OI zI

(11)

where

U 1  K 4I  g1  Id  OI zI

(12)

The best approximation U 1 of a continuous control law

that would achieve sI
U 1

Ix
l
Ix
l

 (
( T\

0 is thus
Iy  Iz
Ix

(14)

Therefore, under the control (13) the system (9) will reach
and thereafter stay on the manifold sI 0 in finite time.

x7 d  x7 . Then we have

f1  g 1  K

the sliding condition will be satisfied. It means that by

1 T
choosing Lyapunov function candidate as V
sI sI , its
2
time derivative is negative definite (V sI sI  K sI ) .

I  Id  OI zI

Iy  Iz

(13)

Then we have
sI sI sI ( f1  g1  k1 sign ( sI )) d ( f1  g 1  k1 ) sI

In order to have the system track I (t ) { Id (t ) , we define

a sliding surface s as follows:
d
sI (  OI ) zI
dt

U 1  k1 sign( sI )

k1

(8)

despite uncertainty on the

dynamics, we add to U 1 a discontinuous term across the

where ui is the input to the i-th actuator, Z0 is the actuator

bandwidth and K is a positive gain. A state variable, vi ,

) . In order to satisfy sliding reachability

)  K 4I  Id  OI zI )

x9 d  x9

z\

x11d  x11

sT

zT  OT zT

s\

z3  O\ z\

k2

f2  g 2  K ,

k3

g 3  K

I2

I3

Iz  Ix
Iy

Ix  I y

Iz

The altitude control U 4 is obtained using the same

approach described for rotational angles.
U4

m
Cx Cx
7

(  x10 x12 I1  K 4 x8  
x7 d  OI ( x7 d  x8 ))

zT

where:

241

( g  K 3 x6  
x5 d  Oz ( x5 d  x6 ))  k 4 sat ( s4 ) (16)

zz

x5 d  x5

sz

z z  Oz z z

k4

g 4  K

After the control signals U i are calculated, desired output

of each actuator can be obtained using (4) and (5) as
follows:

F
F

F
F

1

0
0

 1 2
1 2

1 4 U 1

1 2

b 4d

1 2

b 4d

1 4 U 2

b 4 d

1 4 U 3

b 4 d

1 4 U 4

By knowing the dynamic of actuators (7), the control

input of rotors ui can be easily computed such that the
outputs of actuators, Fi , follow the desired trajectory.
IV. FAULT DETECTION AND FAULT TOLERANT CONTROL
The sliding mode control that described in the previous
section is able to control a quadrotor in the presence of
external disturbances. Although that method is completely
effective in the normal mode, it is not able to control the
system when a fault occurs. Two main sources of faults in
the system are sensors faults and actuators faults. In this
section we consider the effect of actuators fault on the
system performance.
During a long time mission it is very probable that one of
rotors of a quadrotor becomes faulty and partially loses its
effectiveness. It means that the rotor cannot produce its
desired effectiveness any more but it is still working and is
not completely shut down. If one uses a tool to detect this
fault then it might be possible to choose a strategy to
overcome this faulty condition and at least land the
quadrotor safely.
Normally there are no sensors in quadrotor to measure the
outputs of rotors. With the help of these sensors, it might be
easy to determine whether a fault has been occurred in one
of rotors. In the absence of such sensors, a state estimator
can be used to estimate the outputs of rotors. We use a
Luenberger linear state estimator for this purpose [18].
Although the dynamic model of quadrotor is nonlinear, we
can use the linearizred model around the equilibrium point
to design the linear state estimator. Since our objective is to
estimate the outputs of actuators as the states of system, the
linearized model of quadrotor has to include the actuator
dynamics as well.
When the observer estimates the outputs of each rotors,
the fault detection unit has to compare these amounts with
the desired rotors outputs that calculated by system
controller. If the difference between these two amounts is
more than a threshold for a considerable period of time, it
means that the rotor is in a faulty condition.
After fault detection unit realizes that the quadrotor has
been faced with a partial loss in one of its rotors, it has to
change the control strategy to mitigate the effect of fault.
This control strategy has to be able to land the quadrotor

safely. In faulty situation, we do not change the control

signals U i , but we modify transform matrix (17) to change
the rotors trust forces Fi . For this purpose, we propose the
algorithm of controller reconfiguration in the faulty
condition which is shown in Figure 2.
The objective of this fault tolerant control strategy is to
land the quadrotor horizontally (roll 0 and pitch 0)
when an actuator fault occurs. The zeros in the 3rd column of
all matrices in Figure 2 mean that there is no control on the
yaw angle in the faulty mode. The other difference between
(17) strategies in the faulty mode and the normal mode is
control
that the faulty rotor does not participate in the control of roll
or pitch angles. However it still participates in the control of
height. For instance, in the normal mode, signal U 2 that
controls the pitch angle is equal to the difference of trust of
the motor 1 and 2. But when there is a fault in the motor 1, it
is possible that it cannot provide sufficient trust for control
signal U 2 . In order to resolve this problem, we omit the
effect of the motor 1 in control signal U 2 and provide this
control signal only with the motor 2. In fact, the relation
between the control signals U1 and U 2 and rotor outputs in
the faulty condition is as follows:
If Motor #1 is faulty

U2

 F2

U1

F4  F3

If Motor # 2 is faulty

U2

 F1

U1

F4  F3

If Motor # 3 is faulty

U1

 F4

U2

F2  F1

If Motor # 4 is faulty

U1

 F3

U2

F2  F1

But the relation between control signal U 4 and rotor outputs

is still U 4

4
i 1

Fi

If Motor #1 is faulty

Set > F1

F2

F3

0 0 .25 U1
1 0 .25 U 2
.
0 0 .25 U 3

0 0 .25 U 4

F4 @

0
0

.5

.5

F4 @

0 1
0
0

.5 0

.5 0

F4 @

0 .5
0 .5

0 0

1 0

F4 @

0 .5
0 .5

1 0

0
0

If Motor # 2 is faulty

Set > F1

F2

F3

0 .25 U1
0 .25 U 2
.
0 .25 U 3

0 .25 U 4

If Motor #3 is faulty

Set > F1

F2

F3

0 .25 U1
0 .25 U 2
.
0 .25 U 3

0 .25 U 4

If Motor # 4 is faulty

Set > F1

F2

F3

0 .25 U1
0 .25 U 2
.
0 .25 U 3

0 .25 U 4

Fig. 2. The algorithm of controller reconfiguration for fault tolerant control

The schematic of fault detection and fault tolerant control

is illustrated in Figure 3.

242

Angle (Deg)

20
10
0

Angle (Deg)

-10

Angle (deg)

V. SIMULATION RESULTS

10

12

14

10

12

14

10

12

14

10

12

14

Yaw Angle

20
0

Height

1.5

Height (m)

1
0.5
0
-0.5

Time (Sec)

Fig. 5. The outputs of the quadrotor: solid line is sliding mode control, and
dashed line is backstepping control
Roll Angle

Angle (Deg)

30
20
10
0
-10

10

12

14

10

12

14

10

12

14

Pitch Angle

Angle (Deg)

30

20

10

6
Height

Height (m)

1.5
1
0.5
0
0

6
Time (Sec)

Fig. 6. The outputs of the quadrotor UAV in faulty situation

Roll Angle

10

Rotor 1

10

12

Fault Level

60

14

Pitch Angle

30

40
20
0

20

10

12

Fault Level

10
14

Yaw Angle

30

10

12

14

Fault Level
2

10

12

14

Fig. 4. The desired trajectories of angles and height of the quadrotor UAV

10

10

10

Rotor 4

0.5
0
-0.5
2

Time (Sec)

Fig. 7. The outputs of fault detection unit

243

0.5

-1

Time (Sec)

Rotor 3

10

-0.5

Height

0.5

-1

0.5

10

-0.5
-1

20

Rotr 2

Fault Level

Angle (Deg)

Pitch Angle

20

Height (m)

20

-20

In order to show the performance of sliding mode control

and effectiveness of fault detection and fault tolerant control
we simulate a quadrotor UAV in MATLAB/SIMULINK
environment. The desired trajectories for angles and height
are shown in Figure 4. In all simulations, a pulse of
disturbance is inserted to the system in the interval of [3-6]
seconds.
In the first scenario, all rotors works properly and there is
no fault in the actuators. The outputs of system using sliding
mode controller are shown in the Figure 5. In order to show
the performance of sliding mode control in disturbance
rejection, the output of system using a backstepping method
[8] is also shown in Figure 5.
In the next scenario, after 7 seconds the rotor 1 faces with
faulty condition. In this situation, it partially losses its trust
and it can only produce half of its normal maximum trust.
The outputs of system are shown in Figure 6. The outputs of
fault detection unit are depicted in Figure 7. As we can see
in Figure 7, the fault detection unit detects the faulty
condition in rotor 1 almost immediately. After detection of
the fault, the fault tolerant control unit changes the control
strategy which enables the quadrotor to follow the desired
trajectory for a safe landing. It is worth to mention that the
fault detection unit is able to distinguish between external

Angle (Deg)

40

Fig. 3. The schematic of fault detection and fault tolerant control

Angle (Deg)

40

-20

30

Roll Angle

30

Force (N)

5
4
3
2
1
0

Force (N)

10

12

14

10

12

14

5
4
3
2
1
0

REFERENCES

Rotor 3

Force (N)

Rotor 2

5
4
3
2
1
0

10

12

14

10

12

14

Rotor 4

Force (N)

modifies the structure of controller such that some of control

performances are achieved. It has been shown that the fault
detection unit can distinguish between actuator fault and
external disturbance. So it is possible to use the advantage of
disturbance rejection and fault detection and fault tolerant
control at the same time. The simulation results verify the
effectiveness of the proposed method.

Rotor 1

5
4
3
2
1
0

Time (Sec)

Fig. 8. The output of the actuators in faulty situation

Roll Angle

50

Angle (Deg)

40
30
20
10
0

Pitch Angle

300

Angle (Deg)

250
200
150
100
50
0

4
Height

1.5

Height (m)

1
0.5
0
-0.5
-1
-1.5

4
Time (Sec)

Fig. 9. The outputs of the quadrotor in faulty situation without using fault
detection and fault tolerant control strategy

disturbance and actuator fault. It means that we can use the

disturbance rejection ability of sliding mode control while
using the advantage of fault detection and fault tolerant
control strategy. The output of actuators in the faulty
situation is depicted in Figure 8. As shown in this figure, the
output of first motor has been faced a partial loss at time 7
second. Outputs of system in the faulty situation without
using the proposed fault detection and fault tolerant control
is also shown in Figure 9 for comparison.
VI. CONCLUSION
The sliding mode approach is able to control the
quadrotor in the presence of external disturbance. When an
actuator fault occurs, the fault detection unit can detect it
using the state estimator. Then the control configuration unit
that is a residual-based method using a linear observer

[1] G. Homan, H. Huang, C. J. Tomlin, S. L. Waslander, Quadrotor

helicopter flight dynamics and control: theory and experiment, In Proc. of
the AIAA Guidance, Navigation, and Control Conference, Hilton Head, SC,
August 2007.
[2] C. Canttea, J. Chin, S. Mehrabian, L. Montejo, and H. Thompson,
Quad-rotor unmanned aerial vehicle, Report, Columbia University, New
York, 2007.
[3] H. Voos, Nonlinear state-dependent Riccati equation control of a
quadrotor UAV, In Proc. of the 2006 IEEE International Conference on
Control Applications Munich, Germany, pp 2547-2552, 2006.
[4] M. Chen, Formation and flight control of affordable quadrotor
unmanned air vehicles, M.Sc. Thesis, Dept. of Elect. and Comp. Eng.,
University of British Columbia, Vancouver, BC, 2003.
[5] A. Benallegue, A. Mokhtari and L. Fridman, Feedback linearization
and high order sliding mode observer for a Quadrotor UAV, In Proc. of the
2006 International Workshop on Variable Structure Systems, Alghero, Italy,
June 5-7, 2006, pp. 365-370.
[6] E. Altug, J. P. Ostrowski, R. Mahony, Control of a quadrotor helicopter
using visual feedback, In Proc. of 2002 IEEE International conference on
Robotics and Automation, Washington, DC, May 2002, pp 72-77.
[7] H. Voos, Nonlinear control of a quadrotor micro-UAV using feedbacklinearization, In Proc. of the 2009 IEEE International Conference on
Mechatronics, Malaga, Spain, April 2009.
[8] S. Bouabdallah and R. Siegwart, Backstepping and sliding-mode
techniques applied to an indoor micro Quadrotor, In Proc. of the 2005
IEEE International Conference on Robotics and Automation, Barcelona,
Spain, pp 2259-2264, 2005.
[9] R. Xu and U. Ozguner, Sliding mode control of a Quadrotor
helicopter, In Proc. of the 45th IEEE Conference on Decision & Control,
San Diego, USA, December 13-15, 2006, pp. 4957-4962.
[10] S. Bouabdallah, A. Noth, and R. Siegwart, PID vs. LQ control
techniques applied to indoor micro quadrotor, In Proc. 2004 IEEE/RSJ
International Conference on Intelligent Robots and System, Sendai, Japan,
2004, pp. 2451-2456.
[11] I. D. Cowling, O. A. Yakimenko, J. F. Whidborne and A. K. Cooke, A
prototype of an autonomous controller for a quadrotor UAV, In European
Control Conference 2007, 2-5 July, Greece, 2007.
[12] Y. M. Zhang, J. Jiang, Bibliographical review on reconfigurable faulttolerant control systems, Annual Reviews in Control, Vol. 32, No. 2, 2008,
pp. 229-252.
[13] G. Ducard, Fault-tolerant flight control and guidance systems:
practical methods for small unmanned aerial vehicles, London: Springer,
2009, pp. 1-20.
[14] M. Benosman, A survey of some recent results on nonlinear fault
tolerant control, Mathematical Problems in Engineering, 2010, pp. 1-25.
[15] C. Berbra, S. Lesecq and J. J. Martinez, A multi-observer switching
strategy for fault-tolerant control of a quadrotor helicopter, In Proc. of the
16th Mediterranean Conference on Control and Automation, France, 2008,
pp. 1094-1099.
[16] A. Freddi, S. Longhi, A. Monteri, A model-based fault diagnosis
system for a mini-quadrotor, presented at the 7th workshop on Advanced
Control and Diagnosis, Poland, November 2009.
[17] A. B. Milhim, Y. M. Zhang and C. A. Rabbath, Gain scheduling based
PID controller for fault tolerant control of a quad-rotor UAV, AIAA
Infotech@Aerospace, Atlanta, Georgia, April 2010.
[18] D. G. Luenberger, Observing the state of a linear system, IEEE
Transactions on Military Electronics, April 1964, pp. 74-80.

244