ISA Transactions 55 (2015) 100–115

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Adaptive sensor-fault tolerant control for a class of multivariable

uncertain nonlinear systems
Hicham Khebbache a,n, Mohamed Tadjine a, Salim Labiod b, Abdesselem Boulkroune b
a
LCP, Department of Automatic Control, National Polytechnic School (ENP), 10, Av. Hassen Badi, BP. 182, Algiers, Algeria
b
LAJ, Faculty of Science and Technology, University of Jijel, BP. 98 Ouled Aissa, 18000 Jijel, Algeria

art ic l e i nf o a b s t r a c t

Article history: This paper deals with the active fault tolerant control (AFTC) problem for a class of multiple-input
Received 17 July 2014 multiple-output (MIMO) uncertain nonlinear systems subject to sensor faults and external disturbances.
Received in revised form The proposed AFTC method can tolerate three additive (bias, drift and loss of accuracy) and one
20 September 2014
multiplicative (loss of effectiveness) sensor faults. By employing backstepping technique, a novel
Accepted 5 October 2014
adaptive backstepping-based AFTC scheme is developed using the fact that sensor faults and system
Available online 18 February 2015
This paper was recommended for uncertainties (including external disturbances and unexpected nonlinear functions caused by sensor
publication by Jeff Pieper faults) can be on-line estimated and compensated via robust adaptive schemes. The stability analysis of
the closed-loop system is rigorously proven using a Lyapunov approach. The effectiveness of the
Keywords: proposed controller is illustrated by two simulation examples.
Adaptive control
& 2014 ISA. Published by Elsevier Ltd. All rights reserved.
Backstepping approach
Fault tolerant control
MIMO uncertain nonlinear systems
Sensor faults

1. Introduction In the passive approaches, the same controller is used. In fact, the
passive fault tolerant controller can maintain the closed-loop system
Most practical engineering systems are multivariable, and uncer- stability and performance, not only when all components are opera-
tain in nature, varying in both structures and parameters, due to tional, but also under sensor faults occurrence. In the literature, many
external disturbances, harsh operational environment or plant aging, techniques have been employed and combined with each others for
etc [1]. These systems are subjected also to various faults caused by providing reliable controllers. For linear systems, a reliable H1
actuators, sensors, and system components. So, it is important for controller has been proposed in [3]. A reliable LQG controller was
these systems to be kept stable with an acceptable closed-loop designed in [4]. A solution to a reliable filtering problem was
control performance when faults occur. Ideally, in applications presented in [5] using LMI approach. A reliable H1 guaranteed cost
where continuity of operation is a key feature, the closed-loop control has been discussed in [6] also by employing the LMI technique.
system should be capable of maintaining its pre-specified perfor- Moreover, a reliable H1 controller was developed in [7,8] based on
mance in terms of quality of service, safety, and stability despite the LMI approach and static output feedback control. While in [9], a
presence of faults. This procedure is rendered possible thanks to the reliable non-fragile H1 compensation filter has been designed. On the
fault tolerant control (FTC) design [2]. The system faults such as other side, for nonlinear systems, a mixed L2/H1 fuzzy static output
sensor failures may cause severe problems such as system perfor- feedback control was proposed in [10] using iterative LMI (ILMI)
mance deterioration and instability. For this reason, the compensa- approach. A reliable H1 filter was designed in [11] for a class of
tion for these faults has been an important and challenging research nonlinear networked control systems using T–S fuzzy model and LMI
problem for FTC systems. Compared to actuator FTC, relatively fewer technique. A static output feedback fuzzy controller has been devel-
results are available for sensor FTC of linear and nonlinear systems. oped in [12] for T–S fuzzy systems with sensor faults, from which the
Indeed, the existing design techniques for sensor faults mainly stability of these T–S fuzzy systems is guaranteed thanks to a sufficient
include passive approaches and active approaches. condition based on LMIs.
In the active approaches, the control law is redesigned using
desirable properties of performance and robustness. So, an AFTC
n
Corresponding author. Tel.: þ 213 776738031. system can compensate for sensor faults either by selecting a pre-
E-mail addresses: [email protected] (H. Khebbache), computed control law or by synthesizing a new control strategy
[email protected] (M. Tadjine), [email protected] (S. Labiod), online. Speaking in literature sense, several techniques have been
[email protected] (A. Boulkroune).

http://dx.doi.org/10.1016/j.isatra.2014.10.001
0019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 101

employed and combined with fault detection and isolation (FDI) an unknown parameter function. While in [36], an anti-disturbance
blocks or adaptive sensor faults estimation and compensation FTC scheme is presented for a class of nonlinear systems under both
schemes. For linear systems, a fault-tolerant H1 controller via dynamic time-varying faults and multiple disturbances. A composite hierarch-
output feedback has been proposed in [13,14] for sensor failures ical anti-disturbance control (CHADC) scheme consisting of DOBC and
compensation using adaptive method and LMI approach. An adaptive an H1 based PID control was developed in [37] for robotic systems in
fault-tolerant static output feedback controller was presented in [15] the presence of multiple disturbances with system uncertainties. By
with on-line adaptive sensor faults estimation. On the other hand, for combining adaptive disturbance observers with backstepping techni-
nonlinear systems, an adaptive reconfigurable controller based on FDI que, a composite anti-disturbance controller is synthetized in [38] for a
system represented by nonlinear observers has been proposed in class of SISO nonlinear strick-feedback systems with multiple mis-
[16,17] for spacecraft under sensor bias occurrence. In [18], a robust matched disturbances, where the considered disturbances are
fault tolerant control based on several nonlinear state observers was described by not only a single harmonic or constant disturbances
synthesized using Lyapunov's direct method. An adaptive reconfigur- but also another unexpected nonlinear signal presented as a nonlinear
able controller has been developed in [19,20] based on multiple model function.
switching and tuning methodology under occurrence of several types Motivated by the above observations, in this paper, an active fault
of sensor faults. In [21], a robust fuzzy output feedback controller tolerant control scheme is investigated for a class of second-order
based on fuzzy state observer with time delay and sensor faults was MIMO uncertain nonlinear systems subject to sensor faults, external
designed. An AFTC system based on sliding mode has been proposed disturbances as well as system uncertainties, where the input control
in [22] for a class of nonlinear systems with sensor faults by designing signals are updated according to sensor faults occurrence. Using
one detection observer for all outputs and presenting an isol- backstepping technique, a novel adaptive fault tolerant control strat-
ation observer and an estimation observer for each output. In [23], egy is developed, for which adaptive estimators are used for on-line
a proportional-derivative (PD) sliding mode observer-based fault fault estimation and compensation. Compared with existing works,
tolerant control scheme for a class of Lipschitz nonlinear systems the main contributions of this paper can be summarized as follows:
against sensor faults was synthesized. A T–S dynamic output feedback 1) Unlike in [26–30], where any information about velocity sensor
control using two T–S fuzzy observers dedicated to provide separate faults is given, precise models of sensor faults described in [19,20] are
estimates of actuator and sensor faults for the purpose of fault considered in this paper. 2) Unlike in [3–18,21–25], where the
compensation was developed in [24] to deal with simultaneous proposed FTC systems are specifically developed for just one or two
actuator and sensor faults. Moreover, a sensor AFTC system based on types of sensor faults, which considerably limits the range of applic-
output feedback control using high-gain observers has been presented ability of these FTC approaches, in particular, when different types of
in [25] for nonlinear systems subject to input constraints. However, sensor faults occur simultaneously in the controlled system. One of the
the above FTC schemes did not use the backstepping technique. main purposes of this paper is to deal with four types of sensor faults
Since, autonomous systems such as aerial and space vehicles have including bias, drift, loss of accuracy, and loss of effectiveness. 3) Using
highly nonlinear, interacting and unstable dynamics, fault tolerant the fact that the external disturbances, the system uncertainties as
control of these MIMO systems is one of the key issues that needs to well as the sensor faults are considered to be time-varying with
be addressed. By employing velocity-free control aspects, several derivative bounded, the proposed AFTC scheme remove the restrictive
attitude FTC schemes are available in the literature: quaternion feed- assumptions in [31–38] that the disturbances are described by a
back control [26] nonlinear proportional-integral (PI) control allocation neutral stable exogenous system, or a norm bounded variable satisfy-
[27], reliable low-cost control [28] and terminal sliding mode observer ing H2 norm. 4) The on-line fault estimation and compensation can be
based control [29,30]. Although, the aforementioned works address achieved without the need of fault detection and isolation (FDI)
the actuator faults and give a practical solution to the unavailability of module. Hence, it is replaced by robust adaptive estimation schemes.
angular velocity measurement by generating the velocity estimation 5) By using the Lyapunov approach, the stability of the closed-loop
using either a first-order differentiation filter [26–28], or a terminal system is rigorously proven taking into consideration the sensor faults,
sliding mode observer [29,30] with only attitude measurement. the external disturbances and the system uncertainties.
However, the occurrence of velocity sensor faults may affect the This study is organized as follows: In Section 2, the fault
dynamics of the controlled system, cause the corruption of attitude tolerant control problem is formulated, four types of sensor faults
measurement and even to drive it to the instability. So, by providing a considered in this paper are introduced, and after performing
wrong information about attitude measurement after occurrence of some manipulations, the uncertain MIMO nonlinear system is
velocity sensor faults, it is not reasonable to obtain a good stability expressed in a suitable form. In Section 3, a non-adaptive baseline
performance using the above FTC systems. control scheme is designed. The stability analysis of the proposed
Besides, disturbances exist in all practical systems, an interesting AFTC system is detailed in Section 4. To demonstrate the various
problem should be also addressed to achieve a good FTC scheme, is features of the proposed AFTC scheme, simulation results are given
how to attenuate or reject disturbances inherent with systems ?. To in Section 5, followed by a conclusion in Section 6.
overcome this problem, disturbance estimation and compensation
technique can be introduced into controller. Recently, disturbance
observer-based control (DOBC) methods have been widely used to 2. Problem formulation
deal with such problem. Based on nonlinear disturbance observer, a
robust control schemes have been proposed for a class of single-input Consider the following class of uncertain second-order MIMO
single-output (SISO) nonlinear systems [31], and for a class of MIMO nonlinear dynamic systems:
nonlinear systems [32]. A composite control law combining the DOBC
8
_
with H1 control [33] and variable structure control [34] have been < xi;1 ¼ xi;2
>
presented for a class of MIMO nonlinear systems with multiple x_ i;2 ¼ f i ðxÞ þg i ðxÞui þ di ðtÞ; i ¼ 1; 2; :::; q ð1Þ
disturbances, where these disturbances are divided into a norm >
: y ¼ hðx; f Þ
s
bounded variable assumed to satisfy H2 norm and uncertain modeled
 T n
disturbance generated by an exogenous system. In [35], a hierarchical where x ¼ x1;1 ; x1;2  ; x2;1 ; x2;2 ; :::;
Txq;1 ; xq q;2 A ℜ with n ¼ 2q is the
anti-disturbance control scheme consisting of DOBC and a robust states vector, u ¼ u1 ; u2 ;:::; uq A ℜ is the control inputs vector,
T
adaptive controller has been designed for a class of nonlinear systems dðtÞ ¼ d1 ðtÞ; d2 ðtÞ; :::; dq ðtÞ A ℜq is the external disturbances vec-
with composites disturbances described by an exogenous model and tor, y A ℜ with p r n is the measured outputs vector, f s A ℜm with
p
102 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

m r p is the sensor failures vector, f i and g i ; i ¼ 1; 2; :::; q are smooth Therefore, system (6) can be rearranged in the following form
nonlinear functions. 8
< y_ i;1 ¼ yi;2 f si
The model (1) includes a large class of second-order systems ; i ¼ 1; 2; :::; q ð9Þ
such as: inverted pendulum system [39–41], induction motor drive : y_ i;2 ¼ f i ðyÞ þ g i ðyÞui þ Δi ðy; f s ; f_ si ; ui ; di ðtÞÞ
[42–44], single-link robot arm [45–47], mass-springer-damper system
[48,49], flexible spacecraft [50,51], 3-DOF helicopter [52–54], 6-DOF
Remark 1. It is worth noting that, before sensor faults occurrence
quadrotor helicopter [55–62], and many other systems. In the afore-
(i.e. for t ot Fi ), the lumped uncertainties only represent the
mentioned mechanical systems, the states xi;1 ; i ¼ 1; 2; :::; q denote
external disturbances (i.e. Δi ¼ di ðtÞ for i ¼ 1; 2; :::; q).
positions or angles, and xi;2 ; i ¼ 1; 2; :::; q denote linear or angular
velocities. Control objective: Design control laws ui ðt Þ for i ¼ 1; 2; :::; q
In this paper, we will focus on the case when the velocity sensors based only on measurement vector yðt Þ in order to follow a desired
(e.g. gyroscopes in aerial vehicles) are corrupted by several failures trajectory vector yd ðt Þ, while ensuring the stability of the closed-
(i.e. p ¼ n; yi;1 ¼ xi;1 and yi;2 ¼ hi ðxi;2 ; f si Þ for i ¼ 1; 2; :::; q). Thus, the loop system in the presence of different types of sensor faults with
proposed AFTC system is designed to deal with three additive (bias, external disturbances. The block diagram of the proposed AFTC
drift, and loss of accuracy) and one multiplicative (loss of effective- system structure can be depicted by Fig. 1.
ness) sensor faults, mathematically expressed as [19,20]: Throughout this paper we make the following assumptions:
8
>
>
> xi;2 ðt Þ þ bi ; b_ i ðt Þ  0; bi ðt Fi Þ a 0; Bias
>
>
< xi;2 ðt Þ þ bi ðt Þ; bi ðt Þ ¼ λi t; 0 o λi {1; for all t Z t Fi ; Drift
yi;2 ðt Þ ¼ ð2Þ
>
> xi;2 ðt Þ þ bi ðt Þ; bi ðt Þ o b0i ; b_ i ðt Þ-0; for all t Z t Fi ; Loss of accuracy
>
>
>
: κ i ðt Þxi;2 ðt Þ; 0 o κ i r κ i ðt Þ r 1; for all t Z t Fi ; Loss of effectiveness

where t Fi denotes the time instant of failure of the ith sensor, and bi
Assumption 1. [40,41,63]. There exist
unknown positive con-
denotes its accuracy coefficient such that bi A ½ b0i ; b0i , where
stants g i;0 such that: 0 o g i;0 r g i ðxÞ ; for i ¼ 1; 2; :::; q.
b0i 4 0. Also, it is seen that κ i A ½κ i ; 1, where κ i 4 0 denotes the
minimum sensor effectiveness, in which bi and κ i are slowly varying Assumption 2. The desired trajectory vector yd A ℜq , and its time-
respectively within ½  b0i ; b0i  and ½κ i ; 1. derivatives y_ d ; y€ d are assumed to be known, smooth and bounded.
From (2), we have
Remark 2. Assumption 1 is employed to guarantee the controll-
yi;2 ðt Þ ¼ κ i ðt Þxi;2 ðt Þ þ bi ðt Þ ð3Þ
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |ffl{zffl} ability of system (1). Assumption 2 is used to ensure the bound-
ðloss of ef f ectivenessÞ ðbias; drif t; loss of accuracyÞ edness of the differentiation of virtual controls during the stability
By some manipulations, (3) can be rearranged as analysis in backstepping design.
 
yi;2 ðt Þ ¼ xi;2 ðt Þ þ f si t; xi;2 ðt Þ ð4Þ 3. Baseline control scheme
 
where f si t; xi;2 ðt Þ ¼ ðκ i ðt Þ 1Þxi;2 ðt Þ þ bi ðt Þ is an equivalent sensor
fault including all types mentioned above. According to (4), the In the ideal case (i.e. the case without sensor faults), a baseline
output vector becomes y ¼ ½y1;1 ; y1;2 ; y2;1 ; y2;2 ; :::; yq;1 ; yq;2 T A ℜn control scheme is to be designed for system (1) by combining
such as backstepping approach with sliding-mode control (SMC) method,
( as discussed below:
yi;1 ¼ xi;1
; i ¼ 1; 2; :::; q ð5Þ
yi;2 ¼ xi;2 þ f si Step 1. Let us consider the first tracking-error ei;1 ¼ yi;1  yi;1d ¼
xi;1  yi;1d , in which the time-derivative of ei;1 can be written as
Substituting (5) into (1) yields
8 e_ i;1 ¼ xi;2  y_ i;1d ð10Þ
< y_ i;1 ¼ yi;2  f si
    ; i ¼ 1; 2; :::; q
: ẏi;2 ¼ f i ðyÞ þ δf i ðy; f s Þ þ g i ðyÞ þ δg i ðy; f s Þ ui þdi ðtÞ þ f_ si
The stabilization of ei;1 can be obtained by introducing a new
ð6Þ virtual control xi;2d as

where xi;2d ¼ y_ 1;d  ci;1 ei;1 ; ci;1 4 0 ð11Þ

8
>
> f i ðy1;1 ; y1;2  f s1 ; y2;1 ; y2;2 f s2 ; :::; yq;1 ; yq;2  f sq Þ ¼ f i ðy1;1 ; y1;2 ; y2;1 ; y2;2 ; :::; yq;1 ; yq;2 Þ
>
>
>
< þ δf i ðy1;1 ; y1;2 ; y2;1 ; y2;2 ; :::; yq;1 ; yq;2 ; f s1 ; :::; f sq Þ
ð7Þ
>
> g i ðy1;1 ; y1;2  f s1 ; y2;1 ; y2;2  f s2 ; :::; yq;1 ; yq;2  f sq Þ ¼ g i ðy1;1 ; y1;2 ; y2;1 ; y2;2 ; :::; yq;1 ; yq;2 Þ
>
>
>
: þ δg i ðy1;1 ; y1;2 ; y2;1 ; y2;2 ; :::; yq;1 ; yq;2 ; f s1 ; :::; f sq Þ

f i and g i denote the nominal parts, δf i and δg i denote the Substituting (11) into (10) yields
uncertain parts (related to sensor faults) of nonlinear functions f i
e_ i;1 ¼ ci;1 ei;1 þ ei;2 ð12Þ
and g i respectively.
If we choose the lumped uncertainties as where ei;2 ¼ xi;2  xi;2d is the second tracking-error.
Define the first Lyapunov function candidate as
Δi ðy; f s ; f_ si ; ui ; di ðtÞÞ ¼ δf i ðy; f s Þ þ δg i ðy; f s Þui þ f_ si þ di ðtÞ; i ¼ 1; 2; :::; q   1 q
ð8Þ V 1 ei;1 ¼ ∑ e2i;1 ð13Þ
2i¼1
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 103

Estimated sensor faults fˆs

Sensor faults
estimator

Reference Control input Output

yd u y
Controller Nonlinear system Sensors

d fs
External disturbances Sensor faults

Uncertainties
estimator
Estimated uncertainties ˆ

Fig. 1. Block diagram of the proposed AFTC system.

The time-derivative of (13) is given by Assumptions 1–2 are satisfied. Then, the proposed control scheme
  q q   (18) can guarantee that the closed-loop system is globally asympto-
V_ 1 ei;1 ¼ ∑ ei;1 e_ i;1 ¼ ∑  ci;1 e2i;1 þ ei;1 ei;2 ð14Þ tically stable.
i¼1 i¼1

Proof. Since V_ 2 ðt Þ is negative semi-definite, that is V 2 ðt Þ r V 2 ð0Þ.

Step 2. By employing (10) and (11), the time-derivative of Therefore, ei;1 and ei;2 are uniformly bounded. This implies the
ei;2 ¼ xi;2  xi;2d is then boundedness of the closed-loop states. Since V 2 ð0Þ is bounded,
 
e_ i;2 ¼ f i ðxÞ þ g i ðxÞui  y€ i;1d þ ci;1  ci;1 ei;1 þ ei;2 þ di ðtÞ ð15Þ and V 2 ðt Þ is non-increasing and bounded from below, the
lim V 2 ðt Þ ¼ V 2 ð1Þ exists. By integrating (20) from 0 to 1, we obtain
t-1
The augmented Lyapunov function is selected as Z 1 q  
∑ ci;1 e2i;1 ðt Þ þci;2 e2i;2 ðt Þ dt rV 2 ð0Þ  V 2 ð1Þ o1 ð21Þ
  1 q   0 i¼1
V 2 ei;1 ; ei;2 ¼ ∑ e2i;1 þ e2i;2 ð16Þ
2i¼1

Its time-derivative can be expressed as By employing Barbalat's Lemma [63], one can conclude that
    ei;1 ðt Þ-0 and ei;2 ðt Þ-0 as t-1, which implies the asymptotic
q
 
V_ 2 ei;1 ; ei;2 ¼ ∑ ei;1 e_ i;1 þ ei;2 e_ i;2 converge of tracking-errors ei;1 ; ei;2 to zero for i ¼ 1; 2; :::; q. This
i¼1
q 
completes the proof.
  
¼ ∑ ei;1  ci;1 ei;1 þ ei;2 þ ei;2 f i ðxÞ þ g i ðxÞui  y€ i;1d
i¼1
   Remark 3. To deal with the chattering effect caused by the
þ ci;1  ci;1 ei;1 þ ei;2 þ di ðtÞ Þ ð17Þ
discontinuous term
 in the baseline control law (18), the sign
function sign ei;2 is replaced by the continuous saturation func-
By assuming that the external disturbances are bounded tion defined as
by unknown positive constants di 4 0; i.e. di ðt Þ r di ; for
þ þ
( ei;2
i ¼ 1; 2; :::; q, and choosing the control laws ui for i ¼ 1; 2; :::; q as Φ if ei;2 r Φ
satðei;2 Þ ¼   ; i ¼ 1; 2; :::; q ð22Þ
1      sign ei;2 if ei;2 4 Φ
ui ¼ y€ ei;1 ci;2 ei;2 ci;1 ci;1 ei;1 þ ei;2 f i ðxÞ  ki sign ei;2 ð18Þ
g i ðxÞ i;1d
where Φ 40 is the boundary layer thickness.
where, ci;2 and ki are positive constants, signð:Þ denotes the sign
function.
4. Adaptive backstepping-based sensor-AFTC scheme
It results that
  q   To achieve the control objective in the faulty case, a novel
V_ 2 ei;1 ; ei;2 r  ∑ ci;1 e2i;1 þ ci;2 e2i;2 þ ei;2 ðki di Þ
þ
ð19Þ
i¼1
adaptive backstepping-based AFTC scheme will be developed for
system (9) under sensor faults and system uncertainties with the
if we choose the design parameter ki large enough such that following procedures.
þ
ki 4 di for i ¼ 1; 2; :::; q, then (19) can be reduced to
  q  
V_ 2 ei;1 ; ei;2 r  ∑ ci;1 e2i;1 þ ci;2 e2i;2 ð20Þ Step 1. Initially, we consider the first tracking-error ei;1 ¼ yi;1  yi;1d
i¼1 (like in fault-free case). The time-derivative of this tracking-error
is selected as
Theorem 1. Consider the uncertain system (1) in fault-free
case (without faults) with bounded disturbances. Suppose that e_ i;1 ¼ yi;2  f si  y_ i;1d ð23Þ
104 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

Table 1
Comparison between our controller and that proposed by Sun et al. (i.e. [38]).

Comparison Control scheme in [38] Our control scheme

The class of the systems considered Nth order SISO nonlinear system Second order MIMO nonlinear system
The methods employed in control Disturbance observer technique, adaptive control and Adaptive control and backstepping technique
design backstepping method
The study problem Stabilization Tracking
The external disturbances Described by exogenous neutral stable systems Described by free-models, but time-varying with derivative
bounded
The uncertainties Composed by the multiplication of unknown differentiable Composed by the sum of external disturbances with three
time-varying parametric uncertainties with known bounded unexpected nonlinear functions caused by sensor faults, and
nonlinear functions considered to be time-varying with derivative bounded
The sensor faults Not considered, but with appropriate changes in control design, Having precise models and including four different types,
and if sensor faults are viewed as output disturbances, they can three additive (bias, drift and loss of accuracy) with one
be modeled as the sum of exogenous neutral stable systems multiplicative (loss of effectiveness) sensor faults
with nonlinear functions
The functions estimated External disturbances and parametric uncertainties (not system System uncertainties (external disturbances in fault-free
uncertainties) case) and sensor faults
The on-line estimation is performed Disturbance observers Adaptive schemes
using
The location of compensation schemes Each external disturbance with parametric uncertainty in each – Sensor faults in the first step
in backstepping design step – System uncertainties in the second step

Stability of the closed-loop system It is locally asymptotically stable Two cases can be distinguished:
– It is UUB stable under time-varying with derivative
bounded sensor faults and system uncertainties
– It is globally asymptotically stable under slowly time-
varying sensor faults and system uncertainties

Practical systems such those handled in Not belong to the class of nonlinear systems considered Belong to the considered nonlinear systems class
simulation results (i.e. inverted
pendulum and quadrotor helicopter)
Sensor noises in simulation results Not considered Considered
Conclusions The advantages of our control scheme compared with that in [38]:
- Taking into account MIMO systems, the class of systems considered in our work is relatively large with respect to that of
Sun et al.
- Our study problem is more general than that of Sun et al.
- The assumption considered for sensor faults and system uncertainties in our control design is less restrictive than the others
required in [38] for external disturbances and parametric uncertainties.
- Unlike in [38] with appropriate changes, if sensor faults are considered as output disturbances, they can be modeled as
constants, harmonics, or unexpected nonlinear functions. Our work deal with four common sensor faults with precise models
including bias, drift, loss of accuracy and loss of effectiveness.

The advantages of controller in [38] compared with that in our work:

– Unlike in our control design where either sensor faults or system uncertainties are handled at each step, the control scheme
in [38] can deal with an external disturbance and parametric uncertainty at each step.
– Unlike in our work where the tracking-errors can converge to a small neighborhood of the origin (They are only UUB) under
time-varying sensor faults and system uncertainties. The tracking-errors correspond to the work of Sun et al. can
theoretically converge to the origin in finite time.

The stabilization of ei;1 at zero can be obtained by choosing yi;2d Table 2

as virtual control, its desired value is given by Inverted pendulum parameters.

yi;2d ¼ y_ i;1d þ f^ si  ci;1 ei;1 ; ci;1 4 0 ð24Þ Parameter Value

mc 1 kg
m 0.1 kg
Such as f^ si is the estimates of f si , which leads to l 0.5 m
g 9.8 m/s2

e_ i;1 ¼  ci;1 ei;1 þ ei;2  f~ si ð25Þ

Step 2. Using (24), the time-derivative of ei;2 ¼ yi;2  yi;2d , can be
expressed as
where ei;2 ¼ yi;2 yi;2d is the second tracking-error, and f~ si ¼ f si  f^ si
is the sensor fault estimation-error.   _
e_ i;2 ¼  y€ i;1d þ ci;1  ci;1 ei;1 þ ei;2 þf i ðyÞ þ g i ðyÞui  ci;1 f~ si  f^ si þ Δi ð28Þ
The correspond Lyapunov function is chosen as
  1 q
1 ~2

The augmented Lyapunov function is described by
V 1 ei;1 ; f~ si ¼ ∑ e2i;1 þ f si ; αi;1 40 ð26Þ
2i¼1 αi;1 q


~ Þ ¼ V ðe ; f~ Þ þ 1 ∑ e2 þ 1 Δ
V 2 ðei;1 ; ei;2 ; f~ si ; Δ ~ 2 ; α 40 ð29Þ
2 i ¼ 1 i;2 αi;2 i
i 1 i;1 si i;2
Then, the time-derivative of (26) is given by
 


q 1 _ _  where Δ ^ is the estimates of the lumped uncertainty Δ , and
V_ 1 ei;1 ; f~ si ¼ ∑  ci;1 e2i;1 þ ei;1 ei;2 þ f~ si f si  f^ si  e1 ð27Þ i i
i¼1 αi;1 ~
Δ i ¼ Δi  Δ^ i is its estimation-error.
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 105

The time-derivative of (29) is then If we choose the control input ui as

1   
ui ¼ y€ ei;1  ci;2 ei;2  ci;1  ci;1 ei;1 þ ei;2
g i ðyÞ i;1d

  
^ ; c 40
 αi;1 ei;1 þci;1 ei;2  f i ðyÞ  Δ ð31Þ
i i;2
ð30Þ

40
1
Angle y1 [rad]

0.5
20
0

-0.5 0

Applied Force u [N]

-1
-20
0 2 4 6 8 10 12 14 Reference
16 18 20
α 1 =0.1 and β1 =80
3 -40
α1 =0.5 and β 1 =16
2 α 1 =1 and β 1 =8
Velocity y2 [rad/s]

-60
1
α 1 =0.1 and β 1 =80
0
-80
α 1 =0.5 and β 1 =16
-1
α 1 =1 and β 1 =8
-2 -100
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

0.2
0
Tracking-error e 1
Tracking-error e 1

0 -0.1
-0.2
-0.2
-0.3
-0.4 -0.4

-0.6 -0.5
0 2 4 6 8 10 12 14 16 18 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

1 0.5
0
Tracking-error e 2
Tracking-error e 2

0
-0.5
-1 α 1 =0.1 and β 1 =80 -1 α 1 =0.1 and β 1 =80

-2 α 1 =0.5 and β 1 =16 -1.5 α 1 =0.5 and β 1 =16

α 1 =1 and β 1 =8 -2 α 1 =1 and β 1 =8
-3
0 2 4 6 8 10 12 14 16 18 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [s] Time [s]

1 0.5

0.5
sensor fault f s

sensor fault f s

0 0

-0.5
Actual
-1 α 1 =0.1 and β 1 =80 -0.5
0 2 4 6 8 10 12 14 16 18 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α 1 =0.5 and β 1 =16

0.4 α 1 =1 and β 1 =8
Lumped uncertainty

Lumped uncertainty

0.2
0.2
Actual
0 α 1 =0.1 and β 1 =80
0
-0.2 α 1 =0.5 and β 1 =16
α 1 =1 and β 1 =8
-0.4 -0.2
0 2 4 6 8 10 12 14 16 18 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [s] Time [s]

Fig. 2. Evolution of inverted pendulum system without sensor faults. (a) Trajectory-tracking (b) Applied force (c) Tracking-errors (d) Transient process of tracking-errors over
[0,1] s (e) Sensor fault and lumped uncertainty estimation (f) Transient process of sensor fault and lumped uncertainty estimation over [0,1] s.
106 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

and the adaptation laws of f^ si and Δ ^ as

i 1 ~ _ 1 ~2 2

f si f si r f þ f 1i ð35Þ
8_     αi;1 2αi;1 si
< f^ si ¼ αi;1  ei;1 þ ci;1 ei;2 þ βi;1 f~ si
>
  ð32Þ

> _^ ~ 1 ~ _ 1
Δ~ i þ Δ21i
2
:Δ ¼α e þβ Δ Δ i Δi r ð36Þ
i i;2 i;2 i;2 i
αi;2 2αi;2

where β i;1 and βi;2 are positive constants. We obtain Substituting (34)–(36) into (33) yields
q 
V_ 2 ¼ ∑  ci;1 e2i;1  ci;2 e2i;2 αi;1 βi;1 ei;2 f~ i 1 q    
i¼1 V_ 2 r  ∑ 2ci;1 e2i;1 þ 2ci;2  αi;1 βi;1 e2i;2 þ 2βi;1
 2i¼1
~ þ 1 ~ _ 1 ~ _
 βi;1 f~ si  βi;2 Δ ΔΔ
2 2
f f þ ð33Þ


i
αi;1 si si αi;2 i i 1 1 ~2
f~ si þ 2βi;2 
2
 αi;1 βi;1 þ Δi Þ þ λ ð37Þ
αi;1 αi;2
The following assumption is required in the analysis:
where
Assumption 3. The time-derivatives of sensor faults f si and

lumped uncertainties Δi for i ¼ 1; 2; :::; q are assumed to be 1 q 1 2 1 2
λ¼ ∑ f 1i þ Δ1i :
bounded by unknown positive constants f 1i 40 and Δ1i 40 2 i ¼ 1 αi;1 αi;2
_
respectively, i.e. jf_ si j r f 1i and Δ i r Δ1i ; for i ¼ 1; 2; :::; q.
Let
From Assumption 3 and Young’s inequality, we have
αi;1 βi;1 1 1
ci;1 4 0; ci;2 4 ; βi;1 4  ; β i;2 4
αi;1 βi;1  2 ~ 2  2 αi;1 2  αi;1 2αi;2
 αi;1 β i;1 ei;2 f~ si r ei;2 þ f si ð34Þ
2

0.2
1
Tracking-error e 1

0.5 0
Angle y1 [rad]

0
-0.2
-0.5
-0.4
-1

-1.5 -0.6
0 2 4 6 8 10 12 14 16
Reference 18 20 0 2 4 6 8 10 12 14 16 18 20
BSMC system
3
AB-based AFTC system
1
Velocity y [rad/s]

2
Tracking-error e 2

0
1
2

-1
0
-2
-1 BSMC system
AB-based AFTC system
-2 -3
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

40 2
sensor fault f s

20 1

0 0
Actual
Applied Force u [N]

Estimated
-20 -1
0 2 4 6 8 10 12 14 16 18 20

-40 0.4
Lumped uncertainty Δ

0.2
-60
0

-0.2
-80
uBSMC system
-0.4
uAB-based AFTC system
-100 -0.6
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

Fig. 3. Evolution of inverted pendulum system in the presence of velocity sensor bias. (a) Trajectory-tracking (b) Tracking-errors (c) Applied force (d) Sensor fault and
lumped uncertainty estimation.
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 107

and choosing Integrating (39) over ½0; t  results in




 
γ ¼ min 2ci;1 ; 2ci;2  αi;1 βi;1 ; 2βi;1  αi;1 βi;1 þ
1
; 2βi;2 
1 λ λ
0 r V 2 ðt Þ r þ V 2 ð0Þ  e  γ t ð40Þ
αi;1 αi;2 γ γ
4 0; i ¼ 1; 2; :::; q;
This indicates that all signals in the closed-loop system are
then (37) can be rewritten as  
bounded. Thus, the tracking-errors ei;1 ; ei;2 , the sensor faults
V_ 2 r  γ V 2 þ λ ð38Þ estimation-errors f~ si , and the lumped uncertainties estimation-
~ for i ¼ 1; 2; :::; q are uniformly ultimately bounded (UUB).
errors Δ i
According to the above analysis, the following theorem states
the stability and control performance of the closed-loop system. According to (26) and (29), V 2 ð0Þ can be defined as


1 q 1 ~2 1 ~2
Theorem 2. Consider the uncertain system (9) with sensor faults V 2 ð0Þ ¼ ∑ e2i;1 ð0Þ þ e2i;2 ð0Þ þ f si ð0Þ þ Δ i ð0Þ ð41Þ
described in (2). Suppose that Assumptions 1–3 are satisfied. Then,
2i¼1 αi;1 αi;2
the proposed AFTC scheme (31) with the adaptive estimators (32) can From (40), we can obtain
guarantee that the closed-loop system is UUB stable and that the sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q  

 ffi
tracking-errors converge to a small neighborhood of the origin by λ λ
‖E‖ ¼ ∑ e2i;1 þ e2i;2 r 2 þ V 2 ð0Þ  e  γ t ð42Þ
appropriately choosing design parameters. i¼1 γ γ
 
Then, it is easy to show that the tracking errors ei;1 ; ei;2 for
Proof. Multiplying (38) by eγ t yields
i ¼ 1; 2; :::; q exponentially converge to the following compact set:
d  ( sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi )
V 2 eγ t r λeγ t ð39Þ q   qffiffiffiffiffiffiffiffiffiffiffi
dt ΩE ¼ e1;1 ; e1;2 ; e2;1 ; e2;2 ; :::; eq;1 ; eq;2 ‖E‖ ¼ ∑ e2i;1 þe2i;2 r 2λ=γ ð43Þ
i¼1

0.2
1
Tracking-error e 1

0.5 0
Angle y1 [rad]

0
-0.2
-0.5
-0.4
-1

-1.5 -0.6
0 2 4 6 8 10 12 14 16
Reference 18 20 0 2 4 6 8 10 12 14 16 18 20
BSMC system
AB-based AFTC system
3 1

2
Velocity y2 [rad/s]

Tracking-error e 2

0
1
-1
0
-2
-1 BSMC system
AB-based AFTC system
-2 -3
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

40 1.5

1
sensor fault f s

20
0.5

0
0
-0.5 Actual
Applied Force u [N]

Estimated
-20 -1
0 2 4 6 8 10 12 14 16 18 20

-40 0.4
Lumped uncertainty Δ

0.2
-60

0
-80 uBSMC system -0.2
uAB-based AFTC system
-100 -0.4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

Fig. 4. Evolution of inverted pendulum system in the presence of velocity sensor drift. (a) Trajectory-tracking (b) Tracking-errors (c) Applied force (d) Sensor fault and
lumped uncertainty estimation.
108 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

Notice that, ΩE can be made arbitrarily small, which means that applications, there is a certain tradeoff between the choice of the
the tracking-errors can stay arbitrarily close to zero. This ends the design parameters and the numerical precision of the tools involved in
proof. the MIMO control design [64–66].

Remark 5. If we consider the case correspond to bias, drift, and

Remark 4. From (42), we have
loss of accuracy sensor faults (i.e. f si ðt Þ ¼ bi ðt Þ). Assuming that, the
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q   pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sensor faults f si and lumped uncertainties Δi for i ¼ 1; 2; :::; q are
‖Eð0Þ‖ ¼ ∑ e2i;1 ð0Þ þ e2i;2 ð0Þ r 2V 2 ð0Þ ð44Þ _ ffi 0 for i ¼ 1; 2; :::; q, and using
slowly time-varying, i.e. f_ si ffi 0 and Δ i
i¼1
(34), Eq. (33) can be upper bounded as


    2 ~ 2 ; χ ¼ αi;1 βi;1
q
So, the transient performance of the tracking errors depends also
~ ð0Þ and design parameters V_ 2 r  ∑ ci;1 e2i;1 þ ci;2  χ i e2i;2 þ βi;1  χ i f~ si þ βi;2 Δ
on the initial estimation-errors f~ si ð0Þ; Δ
i i
i i¼1 2
αi;1 ; αi;2 . The closer the initial estimates f^ si ð0Þ; Δ^ i ð0Þ, to the true values ð45Þ
f si ð0Þ; Δi ð0Þ, the better the transient performance. Although, increasing
Let ci;1 40; ci;2 4 χ i ; αi;1 o 2 and β i;2 4 0; then expression (45)
the estimation gains αi;1 ; αi;2 will help to reduce the effects of the
becomes
initial estimation-errors on the transient performance as well as the

constant λ. However, choosing αi;1 Z 2 can result in unbounded
q 2
~2
V_ 2 r  ∑ ηi;1 e2i;1 þ ηi;2 e2i;2 þ ηi;3 f~ si þ ηi;4 Δ ð46Þ
 not making γ 4 0. On the other hand, increas-
i
closed-loop
 system by i¼1
ing ci;1 ; ci;2 ; β i;1 ; β i;2 might lead to larger γ and thus, leads to smaller
where ηi;1 ¼ ci;1 ; ηi;2 ¼ ci;2  χ i ; ηi;3 ¼ βi;1  χ i ; and ηi;4 ¼ βi;2 are
λ=γ , this choice can result in a high gain control scheme. Therefore, by positive constants.
employing (38), (43) and (44) and in order to achieve both suitable
transient performance with control action, it is suggested to fix the By following the same reasoning as in Theorem 1, and by
terms αi;1 βi;1 for i ¼ 1; 2; :::; q to some acceptable values and carefully using Barbalat’s Lemma [63] it can be shown that all signals
adjust the other design parameters. This well explains that, in practical in the closed-loop system are bounded and that the signals

0.4
1
0.2
Tracking-error e 1
Angle y1 [rad]

0.5
0
0
-0.2
-0.5
-0.4
-1
-0.6
0 2 4 6 8 10 12 14 16
Reference 18 20 0 2 4 6 8 10 12 14 16 18 20
BSMC system
AB-based AFTC system 1
3

2
Velocity y2 [rad/s]

Tracking-error e 2

0
1
-1
0
-2
-1 BSMC system
AB-based AFTC system
-2 -3
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

40 1.5
Actual
1 Estimated
sensor fault f s

20
0.5

0
0
-0.5
Applied Force u [N]

-20 -1
0 2 4 6 8 10 12 14 16 18 20

-40 0.4
Lumped uncertainty Δ

0.2
-60
0
-80
uBSMC system -0.2
uAB-based AFTC system
-100 -0.4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

Fig. 5. Evolution of inverted pendulum system in the presence of velocity sensor loss of accuracy. (a) Trajectory-tracking (b) Tracking-errors (c) Applied force (d) Sensor fault
and lumped uncertainty estimation.
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 109

~ Þ asymptotically convergence to zero, which means

ðei;1 ; ei;2 ; f~ s i ; Δ However, the terms αi;1 β i;1 y_ i;1 and αi;2 βi;2 y_ i;2 raises the question
i
that the closed-loop system is globally asymptotically stable. of the availability of signals y_ i;1 and y_ i;2 . To overcome this, the
Eq. (48) must be integrated, therefore
~ in
Remark 6. The motivation of using the terms β i;1 f~ si and βi;2 Δ 8   R
i
< f^ si ðtÞ ¼ f^ si ð0Þ  αi;1 βi;1 yi;1 ðtÞ  yi;1 ð0Þ þ t Li;1 ðtÞdt
the adaptive estimators (32) is to guarantee the robustness against  
0
ð49Þ
sensor faults and lumped uncertainties as well as the asymptotic :Δ ^ ðtÞ ¼ Δ ^ ð0Þ þ α β y ðtÞ  y ð0Þ þ R t L ðtÞdt
i i i;2 i;2 i;2 i;2 0 i;2
~ to zero in the case of additive bias, drift,
convergence of f~ si and Δ i  
and loss of accuracy sensor faults (as it is mentioned in Remark 5). where Li;1 ¼ αi;1 βi;1 ðyi;2  f^ si Þ  αi;1 ei;1 þ ci;1 ei;2 and Li;2 ¼ αi;2 βi;2
~ ^
ðf i ðyÞ þg i ðyÞui þ Δi Þ þαi;2 ei;2 .
However, we recall here that the estimation-errors f~ si and Δ i
are strictly unknown. To handle this problem, we will propose a Consequently, the estimates of sensor faults f^ si and lumped
^ by using only the available uncertainties Δ ^ for i ¼ 1; 2; :::; q can be computed without the need
simple algorithm to obtain f^ si and Δ i
i

measurements, as shown below. of any unavailable signal.

Consider the estimation laws (32), which can be written as Remark 7. In order to show the advantages of the proposed AFTC
8 system according to the recent and closely related works, a
 
< f_^ si ¼ αi;1 βi;1 f si  αi;1 βi;1 f^ si  αi;1 ei;1 þ ci;1 ei;2
> comparison between our controller and that of Sun et al. [38] is
ð47Þ summarized in Table 1 (see Appendix A).
>
:Δ ^_ ¼ α β Δ  α β Δ ^
i i;2 i;2 i i;2 i;2 i þαi;2 ei;2

Now, using (9) to compute the values of f si and Δi , we get

8 5. Application examples
_    
< f^ si ¼  αi;1 βi;1 y_ i;1  yi;2  αi;1 βi;1 f^ si  αi;1 ei;1 þci;1 ei;2
>
_^   ð48Þ To show the effectiveness of the proposed AFTC scheme with
> Δ _ i;2  f ðyÞ  g ðyÞui  αi;2 βi;2 Δ ^ þα e
: 1 ¼ αi;2 βi;2 y i i i i;2 i;2 respects to bias, drift, loss of accuracy and loss of effectiveness sensor
faults which occur separately or simultaneously in the controlled

0.2
1
Tracking-error e 1

0
Angle y1 [rad]

0.5

0 -0.2

-0.5 -0.4
-1
-0.6
0 2 4 6 8 10 12 14 16
Reference 18 20 0 2 4 6 8 10 12 14 16 18 20
BSMC system
AB-based AFTC system
3 1
Velocity y2 [rad/s]

2
Tracking-error e 2

0
1
-1
0
-2
-1 BSMC system
AB-based AFTC system
-2 -3
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

40
1
sensor fault f s

20 0.5

0
0
-0.5
Applied Force u [N]

-20 -1
0 2 4 6 8 10 12 14 16 18 20

-40
1 Actual
Lumped uncertainty Δ

Estimated
-60 0.5

0
-80
uBSMC system -0.5
uAB-based AFTC system -1
-100
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [s] Time [s]

Fig. 6. Evolution of inverted pendulum system in the presence of velocity sensor loss of effectiveness. (a) Trajectory-tracking (b) Tracking-errors (c) Applied force (d) Sensor
fault and lumped uncertainty estimation.
110 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

system, two application examples are considered. The first one is an convergence of tracking-errors after occurrence of sensor faults
inverted pendulum, while the second one is a quadrotor helicopter. (see Figs. 3–6 (a and b)). Showing that, the control input corre-
In addition, to obtain realistic simulations, additive
 sensor noises sponding to the baseline controller becoming much sensitive to
modeled as Gaussian random variables N μi ; σ 2i with means μi , and velocity sensor faults, which can even to drive the inverted
variances σ i for i ¼ 1; 2; :::; q are introduced in angular velocities pendulum system to the instability under this situation (see
measurements during all simulations. Figs. 3–6 (c)). It is worth noting that the obtained control input
signal given by the proposed AFTC system is bounded, acceptable
5.1. Fault tolerant control of inverted Pendulum and physically realizable. On the other hand, Figs. 3–6 (d) show
respectively the estimation of velocity sensor bias, drift, loss of
Let x1 ¼ θ be the angle of the pendulum with respect to the accuracy and loss of effectiveness with corresponding uncertain-
vertical line and x2 ¼ θ_ its velocity. The uncertain nonlinear model ties, where the good estimation performance can be achieved for
of inverted pendulum is given by [39–41]: this AFTC scheme despite the undesired effect of sensor noise. For
8 which, the importance of the proposed adaptive estimation
< x_ 1 ¼ x2 schemes is highlighted and the consideration of different types
mlx2 cos x1 sin x1  ðmc þ mÞg sin x1 ð50Þ
: x_ 2 ¼ 2 ml cos 2 x  4ðm þ mÞl
1 c
þ ml cos 2 xcos 4xð1m þ mÞlu þ dðtÞ
1 c
of velocity sensor faults with the presence of external disturbances
3 3
in the analysis is justified.
where mc is the mass of the cart, m is the mass of the pendulum,
l is the effective length of the pendulum, g is the acceleration due 5.2. Attitude fault tolerant control of quadrotor
to gravity, and u is the applied force.
It is assumed that the external disturbance dðt Þ is a square wave The attitude dynamical  model of quadrotor helicopter is repre-
having an amplitude 7 0:1 with a frequency of 0.2 Hz. Thus, the sented by Euler angles ϕ; θ; ψ under the conditions ( π =2 o
lumped uncertainty can be written as ϕ o π =2) for roll, (  π =2 o θ o π =2)  and (π r ψ o π ) for
 for pitch,
  yaw. Define x ¼ ϕ; ϕ _ ; θ; θ_ ; ψ ; ψ_ T ; u ¼ u ; u ; u T : Thus, the
2 ϕ θ ψ
f s  2y2 f s cos y1 sin y1 uncertain nonlinear quadrotor attitude model is described by [58,59]:
Δ¼   þ f_ s þ dðtÞ ð51Þ
cos 2 y1  43 1 þ mc =m 8
> x_ 1;1 ¼ x1;2
>
>
>
> x_ 1;2 ¼ a1 x2;2 x3;2  a2 Ωr x2;2  a3 x1;2 þ b1 uϕ þ dϕ_ ðtÞ
The control objective is to force the output y1 ¼ θ to track the >
>
>
>
reference trajectory yd ¼ sin ðt Þ under occurrence of velocity < x_ 2;1 ¼ x2;2
ð52Þ
> x_ 2;2 ¼ a4 x1;2 x3;2  a5 Ωr x1;2  a6 x2;2 þ b2 uθ þ dθ_ ðtÞ
sensor bias, drift, loss of accuracy and loss of effectiveness with >
the presence of external disturbance. >
>
> x_ ¼ x
>
> 3;1
> 3;2
The system parameters are given in Table 2 (see Appendix B). >
: x_ ¼ a x x  a x þ b u þ d _ ðtÞ
The initial conditions are chosen as: xð0Þ ¼ ð 0:5; 0Þ rad, f^ s ð0Þ ¼ 3;2 7 1;2 2;2 8 3;2 3 ψ ψ

0:3 rad/s and Δ ^ ð0Þ ¼  0:2. The controllers parameters are: c ¼

1 where
5; c2 ¼ 6; k ¼ 0:11 and Φ ¼ 0:01. The estimation parameters are :
α2 ¼ 1 and β2 ¼ 6 (α1 and β1 will be fixed later after performing Iy  Iz J kϕ Iz  Ix J
a1 ¼ ; a2 ¼ r ; a3 ¼ ; a4 ¼ ; a5 ¼ r ; a6
some simulations). The velocity sensor noiseis characterized
 by a Ix Ix Ix Iy Iy
mean μ ¼ 0:01, and variance σ ¼ 0:05 (i.e. N 0:01; 0:052 ). k Ix  Iy kψ d d 1
¼ θ ; a7 ¼ ; a8 ¼ ; b1 ¼ ; b2 ¼ ; b3 ¼ ;
In order to verify the effectiveness of the proposed method, five Iy Iz Iz Ix Iy Iz
tests are performed as follows:
and Ωr is defined as a disturbance, with Ωr ¼ ω1  ω2 þ
ω3  ω4 :
Case. 1 The first test is performed without sensor faults by
The system inputs control are selected as
fixing α1 β1 ¼ 8, three
 simulations are made in the same time 8
by choosing α1 ; β1 ¼ ð0:1; 80Þ; ð0:5; 16Þ and ð1; 8Þ. > u ¼ bðω24  ω22 Þ
>
< ϕ
Case. 2 In the second test, a velocity sensor bias with an
uθ ¼ bðω23  ω21 Þ ð53Þ
amplitude of 1 rad/s is considered at t¼ 8 s. >
>  
: u ψ ¼ κ ω2  ω2 þ ω2  ω2
Case. 3 For this test, we consider a velocity sensor drift with 1 2 3 4

a coefficient λ ¼ 0:05 at t ¼8 s.
   
Case. 4 Here, it is assumed that the loss of accuracy occurring in where I x ; I y ; I z ; kϕ ; kθ ; kψ ; J r ; b; κ; d; and ωi ; i ¼ 1; ::; 4 are
the velocity sensor is a square wave having an amplitude respectively: body inertia, aerodynamic friction coefficients, rotor
7 0:5 rad/s with a frequency of 0.3 Hz occurring at t ¼8 s. inertia, thrust factor, drag factor, distance from center of mass to
Case. 5 In the last test, an effectiveness loss of 75% is simulated rotor shaft, and angular velocity of each rotor i. The exter-
in the velocity sensor at t¼8 s.
Noted that, the tests 2, 3, 4 and 5 are performed after choosing
α1 ¼ 1 and β1 ¼ 8. Table 3
Quadrotor parameters.

The obtained results are shown respectively in Figs. 2–6. Parameter Value
It is concluded from the simulations that, besides the rejection
of external disturbance and the attenuation of measurement noise d 20.5 cm
in all tests (see Figs. 2-6 (a)), an appropriate choice of design Jr 2.8  10  5 kg m2
3.82  10  3 kg m2
parameters α1 and β1 , can lead to the best transient perform- Ix
Iy 3.82  10  3 kg m2
ance for tracking-errors and estimation-errors (see respectively Iz 7.13  10  3 kg m2
Fig. 2 (c)–(f), specifically for α1 ¼ 1 and β 1 ¼ 8). Thus, it clear to see kϕ 5.56  10  3 N m s/rad
that, the proposed AFTC system gives satisfactory results in term of kθ 5.56  10  3 N m s/rad
stability and tracking performance for the inverted pendulum kψ 6.35  10  3 N m s/rad
b 2.98  10  5 N s2/rad2
system. Unlike for the baseline control system (i.e. BSMC system), κ 3.23  10  7 N m s2/rad2
which provides a poor trajectory tracking, with very bad
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 111

10 20

Pitch angle y2,1 [deg]

Roll angle y1,1 [deg]

5 10

0 0

-5 -10

-10 -20

-15 -30
0 5 10 15 20 25 30 35Reference
40 45 50 0 5 10 15 20 25 30 35Reference
40 45 50
α i,1 =0.1 and βi,1 =40 α i,1 =0.1 and βi,1 =40
15 α i,1 =0.5 and βi,1 =8 30 α i,1 =0.5 and βi,1 =8

Pitch velocity y2,2 [deg/s]

Roll velocity y1,2 [deg/s]

10 α i,1 =1 and βi,1 =4 20

α i,1 =1 and βi,1 =4

5 10

0 0

-5 -10
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time [s] Time [s]

0.04
Yaw angle y3,1 [deg]

uφ [N.m]
20 0.02

0 0

0 5 10 15 20 25 30 35 40 45 50
-20
0.1

35Reference
40 45
uθ [N.m]

0 5 10 15 20 25 30 50
0.05
α i,1 =0.1 and βi,1 =40
40 α i,1 =0.5 and βi,1 =8 0
Yaw velocity y3,2 [deg/s]

α i,1 =1 and βi,1 =4 0 5 10 15 20 25 30 35α 40 and 45

βi,1 =4050
i,1 =0.1
20 0.2 α i,1 =0.5 and βi,1 =8
uψ [N.m]

0.1
α i,1 =1 and βi,1 =4
0

0
-20
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time [s] Time [s]

5
0
0
-2
e1,1

e1,1

-5
-4
-10
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5 2 2.5 3
10 0
0
e2,1

e2,1

-5 α i,1 =0.1 and βi,1 =40

-10 α i,1 =0.1 and βi,1 =40
α i,1 =0.5 and βi,1 =8
α i,1 =0.5 and βi,1 =8 -10
-20
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5 2 2.5 βi,1 =4
α i,1 =1 and 3
α i,1 =1 and βi,1 =4
10 0
0 -5
e3,1

e3,1

-10 -10
-15
-20
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5 2 2.5 3
Time [s] Time [s]

Fig. 7. Evolution of quadrotor helicopter system without sensor faults. (a) Trajectory-tracking of roll (b) Trajectory-tracking of pitch c) Trajectory-tracking of yaw (d) Control
inputs (e) Tracking-errors (ei,1, i¼1, 2, 3) (f) Transient process of tracking-errors (ei,1, i¼1, 2, 3) over [0,3] s (g) Tracking-errors (ei,2, i¼ 1, 2, 3) (h) Transient process of tracking-
errors (ei,2, i¼ 1, 2, 3) over [0,1.5] s (i) Rate gyros faults estimation (j) Transient process of rate gyros faults estimation over [0,1.5] s (k) Lumped uncertainties estimation
(l) Transient process of lumped uncertainties estimation over [0,1.5] s.

nal disturbances are given as [55–57]: ðdϕ _ ðtÞ;

T
dθ_ ðtÞ; dψ_ ðtÞÞT ¼ respect to the earth-fixed inertial frame. In addition, these wind
_ ; θ_ ; ψ_
diagða3 ; a6 ; a8 Þη_ air , where η_ air ¼ ϕ represents the disturbances are modeled by square waves having velocities
air air air
wind disturbances defined by the rotation velocity of the air with 7 ð30; 45; 60ÞT deg/s, with a frequency of 0.1 Hz. Hence, the
112 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

5 2
0
0 -2

e1,2
e1,2

-4
-5
-6
-10 -8
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
10
0
0
-5

e2,2
e2,2

-10 α i,1 =0.1 and βi,1 =40 -10 α i,1 =0.1 and βi,1 =40

-20
α i,1 =0.5 and βi,1 =8 -15 α i,1 =0.5 and βi,1 =8
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
α i,1 =1 and βi,1 =4 α i,1 =1 and βi,1 =4
20 10
0
0
-10

e3,2
e3,2

-20 -20
-40 -30
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
Time [s] Time [s]

5 2
0
fs1 [deg/s]
fs1 [deg/s]

0
-2 Actual
-5 -4 α i,1 =0.1 and βi,1 =40
-6 α i,1 =0.5 and βi,1 =8
-10
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
α i,1 =1 and βi,1 =4
5
fs2 [deg/s]
fs2 [deg/s]

0 0
Actual -5
-5
α i,1 =0.1 and βi,1 =40
-10 -10
0 5 10 15 20 25 30 35α i,1 =0.5 βi,1 =8 50
40 and 45 0 0.5 1 1.5
10 α i,1 =1 and βi,1 =4 10
fs3 [deg/s]
fs3 [deg/s]

0 0
-10
-10
-20
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
Time [s] Time [s]

5 6
4
Δ1
Δ1

0 2
0
-5 -2
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5

10 8
6
5 4
Actual
2
Δ2
Δ2

0 Actual 0 α i,1 =0.1 and βi,1 =40

-5 α i,1 =0.1 and βi,1 =40 -2
-4
α i,1 =0.5 and βi,1 =8
0 5 10 15 20 25 30 35α i,1 =0.5 βi,1 =8 50
40 and 45 0 0.5 1 1.5
α i,1 =1 and βi,1 =4
α i,1 =1 and βi,1 =4
6
5 4
2
Δ3
Δ3

0
0
-5 -2
-4
0 5 10 15 20 25 30 35 40 45 50 0 0.5 1 1.5
Time [s] Time [s]

Fig. 7. (continued)
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 113

Fig. 8. Evolution of quadrotor helicopter system in the presence of rate gyros bias, drift, loss of accuracy and loss of effectiveness. (a) Trajectory-tracking of roll (b) Trajectory-
tracking of pitch (c) Trajectory-tracking of yaw (d) Tracking-errors (ei,1, i ¼1, 2, 3) (e) Tracking-errors (ei,2, i¼ 1, 2, 3) (f) Control inputs (g) Rate gyros faults estimation
(h) Lumped uncertainties estimation.
114 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115

lumped uncertainties can be written as uncertainties thanks to the updated control input signals adjusted
8     via an adaptive algorithm providing a correct on-line estimation of
>
>
> Δ1 ¼ a1 f s2 f s3  f s2 y3;2  f s3 y2;2 þ a2 Ωr f s2 þ a3 f s1 þ ϕ_ air þ f_ s1 occurring faults. The effectiveness of the proposed AFTC system is
>
<  
  _ _ verified by controlling two mechanical systems, an inverted
>
Δ 2 ¼ a4 f s1 f s3  f s1 y3;2  f s3 y1;2 þ a5 Ωr f s1 þ a6 f s2 þ θ air þ f s2
>
>     pendulum and a quadrotor helicopter. The simulation results
>
: Δ3 ¼ a 7 f f  f y  f y _ air þ f_ s3
s1 s2 s1 2;2 s2 1;2 þ a8 f s3 þ ψ prove that the proposed AFTC scheme can guarantee the good
tracking performance as well as the stability of the closed-loop
ð54Þ
system against the simulated faults.
For this application example, it is assumed that the three rate gyros
of the Inertial Measurement Unit (IMU) embedded in quadrotor
system to provide measurement of attitude velocity are subject to
Appendix A
bias, drift, loss of accuracy and loss of effectiveness.  Therefore,Tour
control
 objective is to force the outputs y1 ¼ y1;1; y2;1 ; y3;1 ¼  See appendix Table 1
ϕ;θ; ψ T to track the desired trajectories y1d ¼ y1d ; y2d ; y3d T
¼ ϕd ; θ d ; ψ d
T
under simultaneously occurrence of four types of
sensor faults in rate gyros with the presence of wind disturbances.
The physical parameters of the used quadrotor are given in Table. 3 Appendix B
[67,68] (see Appendix B). The desired trajectories are of sinusoidal
form with magnitude equal to π =18 rad (101) for roll, π =9 rad (201) for See appendix Tables 2 and 3
pitch and π =6 rad (301) for yaw. The initial conditions are: xð0Þ ¼
ð  51; 0;  101; 0;  151; 0Þ, f^ s ð0Þ ¼ ð  6;  9;  12Þ1=s and Δ ^ ð0Þ ¼
 References
ð  2:15;  3:25;  2:5Þ. The controllers parameters are: ci;1 ;
ci;2 Þi ¼ 1;2;3 ¼ ð2; 3Þ; ðk1 ; k2 ; k3 Þ ¼ ð4:5; 6:6; 5Þ and Φ ¼ 0:01. The esti-
    [1] Liu XY, Song YD, Song Q. Fault-tolerant control of dynamic systems with
mation parameters are: αi;2 ; β i;2 i ¼ 1;2;3 ¼ ð1; 3Þ ( αi;1 ; β i;1 i ¼ 1;2;3 unknown control direction-input nonlinearities-actuator failures. Proceedings
will be also fixed later after performing some simulations). The of the 50th IEEE Conference on Decision and Control and European Control
Conference (CDC-ECC) 2011:4973–8.
considered noises are Gaussian distributed with means [2] Patton RJ. Fault-tolerant control systems: the 1997 situation. Proceedings of
μ1 ; μ2 ; μ3 ¼ ð0:005; 0:015; 0:025Þ and variances ðσ 1 ; σ 2 ; σ 3 Þ ¼ the 3rd IFAC symposium on fault detection, supervision and safety for techinal
ð0:01; 0:03; 0:05Þ. processes 1997:1033–54.
Two tests are performed, one without and the other with [3] Yang G-H, Wang JL, Soh YC. Reliable H1 controller design for linear systems
with sensor failures. Proceedings of the 37th IEEE Conference on Decision and
sensor faults.   Control (CDC) 1998:2822–7.
In the first test, by fixing αi;1 βi;1 i ¼ 1;2;3 ¼ 4, three simulations [4] Yang GH, Wang JL, Soh YC. Reliable LQG control with sensor failures. IEE Proc –
 
are performed in the same time by choosing αi;1 ; βi;1 i ¼ 1;2;3 ¼ Control Theory Appl 2000;147(4):433–9.
[5] Liu J, Wang JL, Yang G-H. Reliable guaranteed variance filtering against sensor
ð0:1; 40Þ; ð0:5; 8Þ and ð1; 4Þ. The corresponding results are shown in failures. IEEE Trans Signal Process 2003;51(5):1403–11.
Fig.
 7, where  it very clear to see that the choice of design perameters [6] Zhang D, Su H, Chu J, Wang Z. Satisfactory reliable H1 guaranteed cost control
αi;1 ; βi;1 i ¼ 1;2;3 ¼ ð1; 4Þ can lead to thebest transient
performance. with D-stability and control input constraints. IET Control Theory Appl 2008;2
In the second test, after choosing αi;1 ; βi;1 i ¼ 1;2;3 ¼ ð1; 4Þ, it is (8):643–53.
[7] Fujun C, Yu C, Wu W, Fuwen Y. Robust reliable H1 control in the presence of
considered that the three rate gyros are corrupted respectively by: sensor failure. Proceedings of the 7th World Congress on Intelligent Control
1) a bias with 31/s at t¼ 10 s, 2) a drift with coefficient λ ¼ 0:005 at and Automation (WCICA) 2008:8143–8.
[8] Bo Y, Fuzhong W, Jianhua W. Reliable output feedback for linear systems with
t¼12 s, and 3) a loss of accuracy represented by a square wave
sensor mixed faults. Proceedings of the 8th World Congress on Intelligent
having an amplitude 7 51/s with a frequency of 0.15 Hz at t ¼15 s, Control and Automation (WCICA) 2010:509–13.
followed by a loss of effectiveness with 70% at t¼30 s. The [9] Jin X-Z, Yang G-H, Ye D. Insensitive reliable H1 filtering against sensor failures.
simulation results are depicted in Fig. 8. Inform Sci 2013;224:188–99.
[10] Wu H-N, Zhang H-Y. Reliable mixed L2/H1 fuzzy static output feedback control
Fig. 8 (a-c) shows the Euler angles with their angular velocities, for nonlinear systems with sensor faults. Automatica 2005;41(11):1925–32.
Fig. 8 (d and e) represents the corresponding tracking-errors, and [11] Zhao X, Tian E, Wei J, Yuan Y. Reliable H1 filter design for nonlinear
Fig. 8 (f) depicts the control inputs of quadrotor system. The actual networked control systems with probabilistic sensor failure. Proceedings of
the 30th Chinese Control Conference (CCC) 2011:4170–5.
and the estimates of rate gyros faults with corresponding uncer- [12] Wang C, Dong J, Yang G, Kang H. Fuzzy fault tolerant control for nonlinear
tainties are shown respectively in Fig. 8 (g and h). When faults are systems with sensor faults. Proceedings of the 32nd Chinese Control Con-
introduced in rate gyros, it is clearly shown that the baseline ference (CCC) 2013:6214–9.
[13] Yang G-H, Ye D. Fault-tolerant H1 control against sensor failures with adaptive
control scheme becoming ineffective, and may cause the instabil- mechanism. Proceedings of the IEEE International Conference on Control
ity of quadrotor system. Unlike for the proposed AFTC system, Applications 2007:646–51.
where the desired performance can be achieved despite the [14] Yang GH, Ye D. Adaptive fault-tolerant H1 control against sensor failures. IET
Control Theory Appl 2008;2(2):95–107.
corruption of angular velocities thanks to the adaptive estimators
[15] Wang Z, Jin X, Wang Z. Adaptive controller design for linear time-invariant
used for on-line estimation and compensation. Thus, in spite of the systems with sensor failures. Proceedings of the Chinese Control and Decision
sensitivity of the estimates of rate gyros faults and system Conference (CCDC) 2011:2459–62.
[16] Boskovic JD, Sai-Ming L, Mehra RK. Fault tolerant control of spacecraft in the
uncertainties to measurement noises, the correct estimation can
presence of sensor bias. Proceedings of the American Control Conference
be always achieved for the proposed AFTC scheme. Where, it (ACC) 2000:1205–9.
obvious to show the importance of the proposed estimators. [17] Boskovic JD, Sai-Ming L, Mehra RK. A globally stable scheme for Spacecraft
control in the presence of sensor bias. Proceedings of the IEEE Aerospace
Conference 2000:505–11.
[18] Zhihua Q, Ihlefeld CM, Jin Y, Saengdeejing A. Robust control of a class of
6. Conclusion nonlinear uncertain systems. Fault tolerance against sensor failures and
subsequent self recovery. Proceedings of the 40th IEEE Conference on Decision
and Control (CDC) 2001:1472–8 (pp.).
In this paper, an adaptive AFTC scheme for a class of uncertain [19] Bošković JD, Mehra RK. Stable adaptive multiple model-based control design
nonlinear MIMO systems subject to sensor bias, drift, loss of for accommodation of sensor failures. Proceedings of the American Control
accuracy and loss of effectiveness with external disturbances has Conference (ACC) 2002:2046–51 (Anchorage, AK).
[20] Bošković JD and Mehra RK., Failure detection, identification and reconfigura-
been developed. The proposed AFTC controller can compensate for tion in flight control, in Fault Diagnosis and Fault Tolerance for Mechatronic
effects of sensor faults, external disturbances and system Systems: Recent Advances, ed Berlin: Springer, 2003, pp. 127–167.
 H. Khebbache et al. / ISA Transactions 55 (2015) 100–115 115

[21] Tong S-C. Robust fault tolerant fuzzy control for nonlinear systems with sensor [44] Lin FJ, Chang CK, Huang PK. FPGA-based adaptive backstepping sliding-mode
failures. Proceedings of the International Conference on Machine Learning and control for linear induction motor drive. IEEE Trans Power Electron 2007;22
Cybernetics 2007:604–9. (4):1222–31.
[22] Wang YQ, Zhou DH, Qin SJ, Wang H. Active fault-tolerant control for a class of [45] Marino R, Tomei P. Adaptive observers with arbitrary exponential rate of
nonlinear systems with sensor faults. Int J Control, Autom Syst 2008;6 convergence for nonlinear systems. IEEE Trans Autom Control 1995;40
(3):339–50. (7):1300–4.
[23] Ming L, Lixian Z, Peng S, Karimi HR. State feedback control against sensor [46] Kim YH, Lewis FL, Abdallah CT. A dynamic recurrent neural-network-based
faults for Lipschitz nonlinear systems via new sliding mode observer techni- adaptive observer for a class of nonlinear systems. Automatica 1997;33
ques. Proceedings of the 50th IEEE Conference on Decision and Control and (8):1539–43.
European Control Conference (CDC-ECC) 2011:7635–40. [47] Li XJ, Yang GH. Fault diagnosis for non-linear single output systems based on
[24] Sami M, Patton RJ. Active fault tolerant control for nonlinear systems with adaptive high-gain observer. IET Control Theory Appl 2013;7(10):1969–77.
simultaneous actuator and sensor faults. Int J Control, Autom Syst 2013;11 [48] Park JH, Park GT. Adaptive fuzzy observer with minimal dynamic order for
uncertain nonlinear systems. IEE Proc – Control Theory Appl 2003;150
(6):1149–61.
(2):189–97.
[25] Du M, Mhaskar P. Isolation and handling of sensor faults in nonlinear systems.
[49] Boulkroune A, Tadjine M, M’Saad M, Farza M. Design of a unified adaptive
Automatica 2014;50(4):1066–74.
fuzzy observer for uncertain nonlinear systems. Inform Sci 2014;265:139–53.
[26] Xiao B, Hu Q, Zhang Y. Fault-tolerant attitude control for flexible spacecraft
[50] Hu Q, Xiao B, Zhang Y. Robust fault tolerant attitude stabilization control for
without angular velocity magnitude measurement. J Guidance, Control, Dyn
flexible spacecraft under partial loss of actuator effectiveness. Proceeding of
2011;34(5):1556–61.
the Conference on Control and Fault-Tolerant Systems (SysTol) 2010:263–8.
[27] Hu Q, Li B, Wang D, Poh EK. Velocity-free fault-tolerant control allocation for
[51] Xiao B, Hu Q, Zhang Y. Adaptive sliding mode fault tolerant attitude tracking
flexible spacecraft with redundant thrusters. Int J Syst Sci 2013. http://dx.doi. control for flexible spacecraft under actuator saturation. IEEE Trans Control
org/10.1080/00207721.2013.803634. Syst Technol 2012;20(6):1605–12.
[28] Xiao B, Hu Q, Peng S. Attitude stabilization of spacecrafts under actuator [52] Zheng B, Zhong Y. Robust attitude regulation of a 3-DOF helicopter bench-
saturation and partial loss of control effectiveness. IEEE Trans Control mark: theory and experiments. IEEE Trans Indus Electron 2011;58(2):660–70.
SystTechnol 2013;21(6):2251–63. [53] Ferreira de Loza A, Ríos H, Rosales A. Robust regulation for a 3-DOF helicopter
[29] Xiao B, Hu Q, Zhang Y, Huo X. Fault-tolerant tracking control of spacecraft via sliding-mode observation and identification. J Franklin Inst 2012;349
with attitude-only measurement under actuator failures. J Guidance, Control, (2):700–18.
Dyn 2014;37(3):838–49. [54] Boubakir A, Labiod S, Boudjema F, Plestan F. Design and experimentation of a
[30] Xiao B, Hu Q, Wang D. Spacecraft attitude fault tolerant control with terminal self-tuning PID control applied to the 3DOF helicopter. Arch Control Sci
sliding-mode observer. J Aerosp Eng 2014 (doi:10.1061/(ASCE)AS.1943- 2013;23(3):311–31.
5525.0000331). [55] Madani T, Benallegue A. Backstepping sliding mode control applied to a
[31] Chen W-H. Disturbance observer based control for nonlinear systems. IEEE/ miniature quadrotor flying robot. Proceeding of the 32nd IEEE Annual
ASME Trans Mechatron 2004;9(4):706–10. Conference on Industrial Electronics (IECON) 2006:700–5.
[32] Guo L, Chen W-H. Disturbance attenuation and rejection for systems with [56] Zuo Z. Trajectory tracking control design with command-filtered compensa-
nonlinearity via DOBC approach. Int J Robust Nonlinear Control 2005;15 tion for a quadrotor. IET Control Theory Appl 2010;4(11):2343–55.
(3):109–25. [57] Zuo Z. Adaptive trajectory tracking control design with command filtered
[33] Wei X, Guo L. Composite disturbance-observer-based control and H1 control compensation for a quadrotor. J Vibr Control 2013;19(1):94–108.
for complex continuous models. Int J Robust Nonlinear Control 2010;20 [58] Li T, Zhang YM, Gordon BW. Passive and active nonlinear fault-tolerant control
(1):106–18. of a quadrotor UAV based on sliding mode control technique. Proc Inst Mech
[34] Wei X, Guo L. Composite disturbance-observer-based control and terminal Eng, Part I–J Syst Control Eng 2013;227(1):12–23.
sliding mode control for non-linear systems with disturbances. Int J Control [59] Zhang YM, Chamseddine A, Rabbath CA, Gordon BW, Su CY, Rakheja S, et al.
2009;82(6):1082–98. Development of advanced FDD and FTC techniques with application to an
[35] Guo L, Wen X-Y. Hierarchical anti-disturbance adaptive control for non-linear unmanned quadrotor helicopter testbed. J Franklin Inst 2013;350
systems with composite disturbances and applications to missile systems. (9):2396–422.
Trans Inst Measur Control 2011;33(8):942–56. [60] Khebbache H, Tadjine M. Robust fuzzy backstepping sliding mode controller
[36] Cao S, Guo L, Wen X. Fault tolerant control with disturbance rejection and for a Quadrotor Unmanned Aerial Vehicle. J Control Eng Appl Inform (CEAI)
2013;15(2):3–11.
attenuation performance for systems with multiple disturbances. Asian J
[61] Xiong J-J, Zheng E-H. Position and attitude tracking control for a quadrotor
Control 2011;13(6):1056–64.
UAV. ISA Trans 2014;53(3):725–31.
[37] Wen X-Y, Guo L, Yan P. Composite hierarchical anti-disturbance control for
[62] Zheng E-H, Xiong J-J, Luo J-L. Second order sliding mode control for a
robotic systems with multiple disturbances. Int J Control, Autom Syst 2014;12
quadrotor UAV. ISA Trans 2014. http://dx.doi.org/10.1016/j.isatra.2014.03.010.
(3):541–51.
[63] Slotine JE, Li W. Appl Nonlinear Control. Englewood Cliffs, NJ: Prentice-Hall;
[38] Sun H, Guo L. Composite adaptive disturbance observer based control and
1991.
back-stepping method for nonlinear system with multiple mismatched dis- [64] Pozo F, Ikhouane F, Rodellar J. Numerical issues in backstepping control:
turbances. J Franklin Inst 2014;351(2):1027–41. sensitivity and parameter tuning. J Franklin Inst 2008;345(8):891–905.
[39] Wang LX. Adaptive Fuzzy Systems and Control: Design and Stability Analysis. [65] Chang Y. Block backstepping control of mimo systems. IEEE Trans Autom
Englewood Cliffs, NJ: Prentice-Hall; 1994. Control 2011;56(5):1191–7.
[40] Boulkroune A, Tadjine M, M’Saad M, Farza M. How to design a fuzzy adaptive [66] Chen Z, Ge SS, Zhang Y, Li Y. Adaptive neural control of mimo nonlinear
controller based on observers for uncertain affine nonlinear systems. Fuzzy systems with a block-triangular pure-feedback control structure. IEEE Trans
Sets Syst 2008;159(8):926–48. Neural Netw Learning Syst 2014. http://dx.doi.org/10.1109/
[41] Boulkroune A, Bounar N, M0 Saad M, Farza M. Indirect adaptive fuzzy control tnnls.2014.2302856.
scheme based on observer for nonlinear systems: a novel SPR-filter approach. [67] Derafa L, Madani T, Benallegue A. Dynamic modelling and experimental
Neurocomputing 2014;135(5):378–87. identification of four rotor helicopter parameters. Proc IEEE Int Conference
[42] Lin FJ, Lee CC. Adaptive backstepping control for linear induction motor drive on Industrial Technology (ICIT) 2006:1834–9.
to track period references. IEE Proc –Electr Power Appl 2000;147(6):449–58. [68] Derafa L, Benallegue A, Fridman L. Super twisting control algorithm for the
[43] Lin FJ, Shen PH, Hsu SP. Adaptive backstepping sliding mode control for linear attitude tracking of a four rotors UAV. J Franklin Inst 2012;349(2):685–99.
induction motor drive. IEE Proc-Electr Power Appl 2002;149(3):184–94.