signals that are generated by the high-level control module among

the available redundant actuators. In the case of actuator faults,

the proposed adaptive scheme can seamlessly adjust the control pa-

Adaptive Fault-Tolerant Control rameters to compensate the virtual control error and reconfigure
the distribution of control signals among the available redundant
actuators. A significant feature of this study is that the stability of
of a Hybrid Canard Rotor/Wing the closed-loop system is guaranteed theoretically in the presence of
both actuator faults and model uncertainties and overestimation of the
UAV Under Transition Flight adaptive control parameters can be avoided. The effectiveness of the
proposed control strategy is validated through comparative simulation
tests under different faulty and uncertain scenarios.
Subject to Actuator Faults and
I. INTRODUCTION
Model Uncertainties
Over the past few decades, unmanned aerial vehicles
(UAVs) have been a persistent focus due to the increasing
requirements raised by various practical applications, such
as surveillance [1], monitoring [2], and transportation [3].
To fulfill these needs, a variety of UAVs have been devel-
BAN WANG , Member, IEEE
DEHAI ZHU
oped ranging from a few grams to several tons. In terms of
LINYING HAN the different ways of providing lift, UAVs can be generally
HONGGANG GAO classified into four types, namely fixed-wing, rotary-wing,
ZHENGHONG GAO flapping-wing, and hybrid UAVs. Rotary-wing UAVs have
Northwestern Polytechnical University, Xi’an, China the advantages of vertical take-off and landing (VTOL),
YOUMIN ZHANG , Fellow, IEEE hovering at fixed positions. However, they cannot fly at
Concordia University, Montreal, QC, Canada high speeds and long endurance since they are not energy
efficient [4]. On the contrary, fixed-wing UAVs are more
power-efficient with long flight endurance. Nevertheless,
fixed-wing UAVs need a runway for take-off and landing. In
This article proposes an adaptive fault-tolerant control scheme order to integrate the advantages of fixed- and rotary-wing
for an overactuated hybrid vertical take-off and landing canard ro- UAVs, hybrid VTOL UAVs that can operate in a wider
tor/wing unmanned aerial vehicle (UAV) to simultaneously compen-
sate actuator faults and model uncertainties without the requirement
envelope are drawing more attention in the last decade [5],
of fault information and uncertain bounds. The proposed control especially for small-scale UAVs. As a result, a number of pi-
scheme is constructed with two separate control modules. The high- oneer research works have been done and documented in the
level control module is developed with a novel adaptive sliding-mode literature [6], [7]. For now, the development of hybrid VTOL
controller, which is employed to maintain the overall system tracking UAVs is still the research focus for both academic and
performance in both faulty and fault-free conditions. The low-level
control allocation module is used to distribute the virtual control
industrial communities. In terms of hybrid VTOL UAVs,
various designs have been proposed, including tilt-rotor
Manuscript received 17 December 2021; revised 12 April 2022, 17 Au- UAV [8], tilt-wing UAV [9], tail-sitter UAV [10], quad-plane
gust 2022, and 27 December 2022; accepted 2 February 2023. Date of UAV [11], and canard rotor/wing (CRW) UAV [12]. Among
publication 9 February 2023; date of current version 9 August 2023. these different types of hybrid VTOL UAVs, the CRW
DOI. No. 10.1109/TAES.2023.3243580 UAV is favored in this article because it does not need a
Refereeing of this contribution was handled by J. Marzat.
complicated transition tilting mechanism like tilt-rotor and
tilt-wing UAVs and it possesses low disc loading under heli-
This work was supported in part by the National Natural Science Foun- copter mode over tail-sitter and quad-plane UAVs, which
dation of China under Grant 62003266 and Grant 61833013, in part by
the Industry-University-Research Innovation Foundation for the Chinese
can improve flight efficiency.
Ministry of Education under Grant 2021ZYA07002, in part by the Fun- The concept of CRW aircraft originated in the United
damental Research Funds for the Central Universities, and in part by the States in the early 1990s [13], [14]. In 2003, Boeing demon-
Natural Sciences and Engineering Research Council of Canada. strated flight tests of a CRW UAV under the Dragonfly pro-
Authors’ addresses: Ban Wang, Dehai Zhu, Linying Han, and gram [15]. Following these works, the flight dynamics mod-
Zhenghong Gao are with the School of Aeronautics, North- eling and control of CRW UAV with an H-tail aerodynamic
western Polytechnical University, Xi’an 710072, China, E-mail: layout as proposed by Boeing company are investigated
([email protected]; [email protected]; hchanliny- in [16] and [17]. In addition, a CRW UAV with a T-tail layout
ing@ mail.nwpu.edu.cn; [email protected]); Honggang Gao is with
the School of Civil Aviation, Northwestern Polytechnical University,
is proposed in [12], which aims to reduce the adverse influ-
Xi’an 710072, China, E-mail: ([email protected]); Youmin Zhang is ence of downwash flow on the horizontal tail. In [18], the
with the Department of Mechanical, Industrial and Aerospace Engineer- aerodynamic characteristics of elliptic airfoil used for the
ing, Concordia University, Montreal, QC H3G 1M8, Canada, E-mail: main rotor of the CRW UAV is investigated. In [19] and [20],
([email protected]). (Corresponding author: Youmin Zhang.) the characteristics of the T-tail CRW UAV’s flight dynamics
in transition and fixed-wing flight modes are studied, which
0018-9251 © 2023 IEEE lays the foundation for the design of the flight control system

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023 4559

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
in this work. In terms of flight control, the difficulties of and CA module, respectively [40]. In [41], a constrained
transition control for hybrid VTOL UAVs have long been the CA method is first proposed for the allocation problem
focus of researchers due to their high dynamic uncertainty of a variety of high-performance aircraft flight controls.
and the serious aerodynamic interference between rotors Following this work, the CA problem has been intensively
and wings. In [21], a nonlinear robust controller is presented studied and different methods on CA are investigated in the
for flight mode transition control of a tail-sitter VTOL UAV literature. In [42], an overview of different CA methods,
based on H∞ control and nonlinear disturbance observer to such as redistributed pseudoinverse and direct allocation,
deal with model uncertainties. In [22], an altitude controller is presented. Most of these methods assume that all the
is designed based on active disturbance rejection control control effectors are present. Due to the overactuated feature
to achieve fast flight mode transition for a tilt-rotor UAV of hybrid VTOL UAVs, they have an inherent capability to
during landing phase. In [23], an adaptive neural network tolerate certain level of actuator fault or even failure. In [43],
dynamic inversion control strategy is proposed to compen- an FTC scheme is proposed for a tilt-rotor UAV developed
sate model uncertainties for transition flight control of a by Korea Aerospace Research Institute to compensate the
CRW UAV. In general, model uncertainties inevitably exist, adverse effect of actuator faults. In [44], a closed-loop CA
thus, the designed flight controller is required to be robust scheme based on nonlinear dynamic inversion is presented
against such uncertainties. Sliding-mode control (SMC) is a to accommodate actuator faults for longitudinal control
robust control approach that handles uncertainties with dis- of a CRW aircraft in fixed-wing mode. In [45], a hierar-
continuous control strategy, which demonstrates invariance chical nonlinear controller based on backstepping control
to matched uncertainties when the sliding variable stays is proposed for a tilt-rotor UAV to maintain the cruise
on a predefined sliding surface. In [24], a sliding-mode capability in the presence of actuator stuck fault. However,
controller is presented for a miniature unmanned helicopter real-time fault information is required for the presented FTC
to compensate unmodeled flapping dynamics and external strategies in order to achieve control reallocation. To solve
disturbances. In [25], a sliding-mode controller is proposed this problem, adaptive CA method [46] is investigated by
for a tilt-wing UAV to stabilize its attitude and altitude. Most many researchers. In [47], an adaptive CA approach is de-
of the works use the inherent robustness of the SMC, which veloped for overactuated systems with actuator uncertainty.
may induce serious control chattering. To overcome this In [48], a fault-tolerant adaptive CA scheme is presented
issue, many adaptive SMC schemes have been investigated for overactuated systems subject to loss of effectiveness
in the literature [26], [27]. However, the sliding variable actuator faults. Additionally, some researchers solve this
is usually employed to construct the adaptive schemes in problem from the closed-loop control point of view. Instead
these works, which may not guarantee the convergence of of designing new CA schemes, they focus on the design of
adaptive control parameters due to the inevitable tracking the robust high-level motion controller to complement the
errors in practical applications. The overestimation of the CA module in the presence of actuator faults/saturation.
adaptive control parameters may induce severe control chat- In [49], a sliding-mode controller with a time-varying slid-
tering even if a boundary layer is used to construct the ing surface is proposed to guarantee stability and tracking
discontinuous control portion of SMC. This presents one performance of the closed-loop system in the presence of
of the motivations of this article which is to develop an CA error. In [50], an FTC scheme is designed based on
adaptive SMC to compensate for model uncertainties of the robust SMC for overactuated uncertain linear systems to
CRW UAV while suppressing control chattering. compensate the CA error induced by the imperfection in
In addition, the transition flight is extremely critical for the estimated effectiveness gain.
hybrid VTOL UAVs, any fault or failure that occurs during Although extensive research works have been carried
this mode may result in a catastrophic crash, which calls out to address the problem of transition control of hybrid
for the development of fault diagnosis and fault-tolerant VTOL UAVs, there are still some problems that need to be
control (FTC) [28], [29], [30]. As stated in [31], [32], [33], further solved.
[34], [35], and [36], the increasing demands for safety
and reliability of safety-critical systems have stimulated 1) Most of the research works dealing with transition
research in the area of FTC. Successful applications of control of hybrid VTOL UAVs in the literature
FTC on aerospace systems can be found in [37], [38], and mainly consider the uncertain aerodynamic infer-
[39]. Moreover, the control variables of hybrid VTOL UAVs ences. If actuator faults occur during the transition
in helicopter mode and airplane mode are both involved flight phase, it may lead to system instability.
under transition flight condition. This makes the hybrid 2) In order to deal with actuator faults for overactu-
VTOL UAV physically redundant, which can be naturally ated systems, accurate fault information is usually
used to develop advanced FTC schemes. Control allocation required for most of the FTC allocation schemes.
(CA) is one of the effective approaches for dealing with To solve this problem, some research works propose
overactuated systems. The modular design of the control adaptive CA schemes. A few research works solve
algorithm by employing CA approach allows the design of the FTC problem from the closed-loop control point
high-level motion control algorithm to be independent of ac- of view, but most of them use the inherent robustness
tuator configuration by introducing virtual control module of SMC to compensate the CA error, which may

4560 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 induce control chattering if the discontinuous control
gain is chosen to be large to compensate large virtual
control errors.
3) To compensate for the adverse effect of both actuator
faults and model uncertainties by using adaptive
SMC technique, the sliding variable is usually em-
ployed to formulate the adaptive schemes for esti-
mation of uncertain control parameters. This may
result in overestimation of the control parameters
if the tracking errors cannot be zero, which will
induce control chattering and cause instability of the
closed-loop system.
Fig. 1. Schematic diagram of the studied hybrid CRW UAV.
In an attempt to solve the aforementioned difficulties to
improve system safety and reliability, this article proposes
a novel adaptive FTC scheme for a hybrid CRW UAV TABLE I
to simultaneously accommodate actuator faults and model Physical Parameters of the Studied Hybrid CRW UAV
uncertainties based on adaptive SMC and dynamic CA. The
main contributions of this article are briefly summarized as
follows.
1) In order to accommodate multiple actuator faults,
the proposed control strategy is developed from the
closed-loop control point of view and no fault infor-
mation is required for the proposed control strategy
compared to the existing works [43], [44], [45] in
the literature. By employing the proposed adaptive
schemes, the control parameters can be adjusted in
real time. In this sense, the virtual control error can be II. PRELIMINARIES
effectively compensated to ensure the desired system A. Modeling of a Hybrid CRW UAV
tracking performance, and the control signals will be
redistributed to the available redundant actuators. An overview of the studied hybrid CRW UAV is illus-
2) The proposed control strategy can greatly reduce trated in Fig. 1 and its specifications are listed in Table I.
the use of discontinuous control portion of SMC by It has a main rotor (which is the same as a conventional
incorporating the adaptive control parameters in both helicopter) driven by an electrical motor, a canard and a
continuous and discontinuous control portions for horizontal tail. The change of flight mode of this UAV
compensating actuator faults. In addition, the over- mainly depends on the status of the main rotor. Under
estimation of the control parameters is avoided with helicopter mode, the CRW UAV operates in the same way as
the proposed adaptation law in contrast to [26], [27], a conventional single-rotor helicopter. The collective pitch
which further contributes to avoiding unexpected is used for height control, the cyclic pitch is employed
control chattering. Only when the sliding variable for longitudinal and lateral control, and the tail rotor is
is outside the defined boundary layer, the adapta- utilized to balance the antitorque generated by the main
tion will be triggered with the proposed adaptive rotor, whereas the motor mounted on the front of the CRW
schemes. UAV does not work under this flight mode. If the CRW UAV
3) Unlike the studies in [19], [21], and [22], the pro- operates under airplane mode, the main rotor will be locked
posed control strategy can simultaneously compen- as a fixed wing, which is similar to a three-surface aircraft.
sate both actuator faults and model uncertainties for The forward flight power comes from the motor mounted
a hybrid CRW UAV without knowledge of the uncer- on the front of the CRW UAV. The roll and pitch motion
tainties. Furthermore, the stability of the closed-loop of the CRW UAV are controlled by canard control surface
system is guaranteed theoretically. and elevator together, and the yaw motion is controlled by
rudder. If the CRW UAV works under transition mode, the
The rest of this article is organized as follows. The main rotor, the tail rotor, the front motor, and all the control
dynamic modeling of the studied hybrid CRW UAV and surfaces for both helicopter and airplane mode will work
problem formulation are presented in Section II. Section III together to achieve the desired movement.
addresses the detailed design procedures of the proposed In order to model the CRW UAV, a body-fixed reference
control strategy. The comparative simulation results are frame is employed and denoted as (ob , xb , yb , zb ). The origin
followed in Section IV to demonstrate the effectiveness of of the body-fixed frame coincides with the center of gravity
the proposed control strategy. Finally, Section V concludes of the CRW UAV. By employing the Newton–Euler formu-
this article. lation, the dynamic equations of the CRW UAV with respect

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4561

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
to the body-fixed frame can be expressed as follows [51]:
⎧ Fx
⎪
⎪ u̇ = − qw + r v − g sin θ
⎪
⎪ m
⎪
⎨
Fy
v̇ = − ru + pw + g cos θ sin φ (1)
⎪
⎪ m
⎪
⎪
⎪
⎩ẇ = Fz − pv + qu + g cos θ cos φ
m
⎧
⎪ M x Izz + Mz Ixz + (Ixx Ixz − Iyy Ixz + Izz Ixz ) pq
⎪
⎪ ṗ =
⎪
⎪ Ixx Izz − Ixz
2
⎪
⎪ Fig. 2. Wind tunnel test setup.
⎪
⎪ (Iyy Izz − Ixz − Izz )qr
2 2
⎪
⎪ +
⎪
⎪ Ixx Izz − Ixz 2
⎪
⎨ My + (Izz − Ixx ) pr − Ixz ( p2 − r 2 ) of the resultant forces and torques, such as the aerodynamic
q̇ = (2)
⎪
⎪ Iyy inference forces and torques induced by the inferences
⎪
⎪
⎪
⎪ Mz Ixx + Mx Ixz + (Ixz 2
+ Ixx
2
− Ixx Iyy ) pq among the main rotor, the fuselage and the horizontal tail.
⎪
⎪ ṙ = Specially, during the transition mode, there exist large
⎪
⎪ Ixx Izz − Ixz2
⎪
⎪ inferences among the main rotor, the fuselage, and the
⎪
⎪ (Ixz Iyy − Ixz Izz − Ixz Ixx )qr
⎩ + horizontal tail, which is hard to be calculated accurately, and
Ixx Izz − Ixz2
⎧ therefore, included in Fx1 , Fz1 , and My1 . For the purpose
⎨φ̇ = p + q sin θ sin φ/ cos θ + r sin θ cos φ/ cos θ of validating the effectiveness of the proposed control
θ̇ = q cos φ − r sin φ (3) scheme under such uncertainties, the uncertain forces and
⎩
ψ̇ = q sin φ/ cos θ + r cos φ/ cos θ torques are modeled according to both wind tunnel test
and computational fluid dynamics (CFD) calculation data.
where u, v , w are the translational velocities of the CRW
Fig. 2 shows the setup of wind tunnel test, and the
UAV defined in the body-fixed frame, p, q, r are the angular
demonstrated aircraft is mainly built for wind tunnel test.
velocities of the roll, pitch, and yaw motion, respectively.
The main rotor and the fuselage are placed in different
Ixx , Iyy , Izz are the moments of inertia of the UAV, Ixz is
measurement systems. First, the testing is carried out with
the product of inertia, and m is the total mass of the
the fixed main rotor, and then the testing is carried out
UAV. Fx , Fy , Fz and Mx , My , Mz are the resultant forces
with the rotating main rotor. The interfering forces and
and torques acting on the UAV along xb -, yb -, and zb -axis,
torques are calculated based on the difference between
respectively.
these two tests. In addition, since the interfering forces
Since during the transition flight mode, the CRW UAV is
and torques are related to the thrust generated by the
desired to move with small Euler angles, it is reasonable to
main rotor, it is fixed to 15 N in this test. By changing
assume that the angular velocity of the aircraft is equal to its
the advance ratio of the main rotor, different testing data
angular rate. In this sense, in order to facilitate the controller
can be obtained. However, due to the limitation of the
design, (2) and (3) can be combined and rewritten as
experimental instrument, the testing can only be carried
⎧
⎪ Mx Izz + Mz Ixz + (Ixx Ixz − Iyy Ixz + Izz Ixz )φ̇ θ̇ out up to the advance ratio of 0.16. In order to obtain more
⎪
⎪ φ̈ =
⎪
⎪ Ixx Izz − Ixz
2 data, CFD calculation is conducted. Finally, based on both
⎪
⎪
⎪
⎪ (Iyy Izz − Ixz − Izz )θ̇ ψ̇
2 2 the experimental and CFD calculation data, fitted curves
⎪
⎪ +
⎪
⎪
are obtained and shown in Fig. 3.
⎪
⎪ Ixx Izz − Ixz 2
⎨ My + (Izz − Ixx )φ̇ ψ̇ − Ixz (φ̇ 2 − ψ̇ 2 )
The expressions for the fitted curves are given as fol-
θ̈ = (4) lows:
⎪
⎪ Iyy ⎧
⎪
⎪ ⎨Fx1 = (−0.003 − 0.234μ + 0.17μ2 )Tmr
⎪
⎪ M I + Mx Ixz + (Ixz 2
+ Ixx
2
− Ixx Iyy )φ̇ θ̇
⎪ψ̈ = z xx
⎪ Fz1 = (0.22 − 0.255μ − 0.1616μ2 )Tmr
⎪
⎪ Ixx Izz − Ixz
2
⎩
(5)
⎪
⎪ My1 = M
⎪
⎪ ( I I − I Izz − Ixz Ixx )θ̇ ψ̇ ⎧
⎪
⎩ +
xz yy xz
.
Ixx Izz − Ixz 2 ⎪
⎪ 0 if μ ≤ 0.02235
⎨
(4.206μ − 0.094)Tmr R if μ ≤ 0.124
M =
Focusing on the longitudinal control of the CRW ⎪
⎪ ( −1.588 μ + 0 .624) Tmr R if μ ≤ 0.3931
UAV, it is reasonable to assume that p = r = 0 ⎩0 if μ > 0.3931
and v = 0. In addition, considering the fact that in (6)
practice the forces and torques acting on the UAV may
not be obtained accurately, the resultant forces and where μ is the advance ratio of the main rotor, R is the radius
torques are formulated as Fx = Fxheli + Fxplane + Fx1 , of the main rotor, and Tmr is the thrust generated by the main
Fz = Fzheli + Fzplane + Fz1 , My = Myheli + Myplane + My1 . rotor.
Fxheli , Fxplane , Fzheli , Fzplane , Myheli , Myplane are denoted as During transition mode, the forward speed is mainly
the known portions of the resultant forces and torques, controlled by the motor mounted on the front of the UAV,
whereas Fx1 , Fz1 , My1 are denoted as the unknown portions the vertical speed of the UAV is mainly controlled by the

4562 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 
fi (x1 , x2 ) = f0i (x1 , x2 ) + f1i (x1 , x2 )
(9)
νi = Bi Lc (t )uc (t )

T
where i = 1, 2, 3, x2 = [u, w, θ̇ ] , uc = [T, θ0 , θB , δc , δe ]T ,
f0i (x1 , x2 ), and hi (x1 , x2 ) are the certain portions of the
model, f1i (x1 , x2 ) is the uncertain portion of the model,
ν = [νu , νw , νθ ]T is the virtual control input vector, B is the
control effectiveness matrix that is composed of the control
input coefficients, and Lc (t ) = diag([lc1 , lc2 , . . . , lcm ]) is a
diagonal matrix that represents the faulty level of the actu-
ators. If lc j = 1, the jth actuator works perfectly, whereas
0 ≤ lc j < 1 denotes that the jth actuator suffers a certain
level of fault.

REMARK 1 The diagonal matrix Lc (t ) is unknown and time-

dependent, which is used to model the investigated actuator
faults. In the context of active FTC framework, after fault
occurrence, this unknown matrix Lc (t ) is usually estimated
Fig. 3. Experimental and CFD calculation data and fitted curves. and used to formulate the corresponding control law. How-
ever, in the context of adaptive FTC framework, the main
priority is to compensate the fault and may not need to
know the actual faulty level of the actuators. Therefore, the
unknown matrix Lc (t ) is not used in the developed control
law in this article and the following equation νi = Bi uc (t )
is employed for distributing the virtual control signals.

With this definition and the state vector x2 , the

Fig. 4. Illustrative example for positive pitch motion control.
dynamic equations of the studied CRW UAV in
(7) can be resolved into the following subsystems.
collective pitch of the main rotor, and the pitch angle of Forward velocity subsystem: ẋ11 = x21 , ẋ21 = f01 +
the UAV is controlled by deflections of the canard control h1 ν1 + f11 with f01 = (Fxheli + Fxplane )/m − qw − g sin θ,
surface, elevator, and the cyclic pitch of the main rotor. h1 = 1/m, ν1 = T , f11 = Fx1 /m; Vertical velocity
An illustrative example for positive pitch motion control is subsystem: ẋ12 = x22 , ẋ22 = f02 + h2 ν2 + f12 with f02 =
demonstrated in Fig. 4. (Fzheli + Fzplane )/m + qu + g cos θ, h2 = 1/m, ν2 = Lθ0 θ0 ,
In this sense, the control-oriented model can be obtained f12 = Fz1 /m; Pitch motion subsystem: ẋ13 = x23 ,
as ẋ23 = f03 + h3 ν3 + f13 with f03 = (Myheli + Myplane )/Iyy ,
⎧ h3 = 1/Iyy , f13 = My1 /Iyy , ν3 = LθB θB + Lδc δc + Lδe δe .
⎪ F + Fxplane
⎪u̇ = xheli
⎪ − qw − g sin θ + mT + Fmx1
⎪
⎪ m In this article, for the pitch motion control of the
⎨ Fzheli + Fzplane CRW UAV, there are three kinds of different control inputs
L θ
ẇ = + qu + g cos θ + θm0 0 + Fmz1 (7)
⎪ m θB , δc , and δe . The CA problem refers to the distribution
⎪
⎪ M + M
⎪
⎪ yheli yplane L θ +L δ +L δ M of the virtual control input νθ over the available redundant
⎩θ̈ = + θB B Iδyyc c δe e + Iyyy1
Iyy actuators. In a faulty condition where lc j < 1, given the
desired virtual control signal that is generated from the
where T is the thrust generated by the front motor, θ0
high-level controller, the solution uc (t ) is searched such that
is the collective pitch control input of the main rotor, θB
νd (t ) = Buc (t ) is satisfied. In addition, the position and rate
is the longitudinal cyclic pitch control input of the main
constraints of the actuators need to be considered, which
rotor, δc is the deflection of the canard control surface, δe
can be formulated as
is the deflection of elevator, and Lθ0 , LθB , Lδc , Lδe are the
corresponding coefficients related to the generated forces  min
uc ≤ uc (t ) ≤ ucmax
and torques. (10)
|u̇c (t )| ≤ ucrate
B. Problem Formulation
where ucmin and ucmax are the lower and upper position con-
Consider a nonlinear affine CRW UAV with actuator straints of the actuators, and ucrate is the maximum rate for
faults and model uncertainties as the individual actuators.

ẋ1i = x2i For simplicity of expressions, the notation t is omitted
(8)
ẋ2i = fi (x1 , x2 ) + hi (x1 , x2 )νi in the following sections, e.g., uc (t ) is expressed as uc .

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4563

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 Fig. 5. Schematic of the proposed adaptive FTC strategy.

III. ADAPTIVE FTC STRATEGY Considering the uncertain portion f1i , the adaptive
In this section, an adaptive FTC strategy is proposed model uncertainty compensation term is designed by mak-
to simultaneously accommodate actuator faults and model ing ṡi = 0. In this case, the following equation can be
uncertainties for a CRW UAV with redundant actuators. The obtained as
schematic of the proposed control strategy is illustrated in ṡi = ẍ1i − ẍ1di + kc2i ẋei + kc1i xei
Fig. 5. First, an adaptive high-level SMC scheme is pro-
= fi + hi νi − ẍ1di + kc2i ẋei + kc1i xei = 0.
posed to compensate model uncertainties, and the generated
virtual control signals are distributed to the actuators with (14)
the low-level CA module. Then, in the case of actuator From (14), the adaptive model uncertainty compensa-
faults, the generated virtual control signal from the low-level tion term is designed as
CA module will not meet the desired virtual control signal
demanded from the high-level SMC scheme. Thus, an adap- νmi = hi−1 (ẍ1di − kc2i ẋei − kc1i xei − f̂i ). (15)
tive scheme is constructed to redistribute the control signals Then, the robust feedback term that ensures the sliding
to the available redundant actuators for compensating the motion is designed as
virtual control error induced by actuator faults. In such a
way, the control performance of the overall system can be νri = −hi−1 kci sat(si / i) (16)
maintained.
where kci is an adjustable control gain and the sat(· ) function
A. Design of Adaptive SMC With Model Uncertainties is defined as
⎧
  T ⎨1 if si > i
Consider the states x1 = [ u, w, θ ] and
sat(si / i ) = −1 if si < − i (17)
T
x2 = [u, w, θ̇ ] of the CRW UAV, and denote x1i as
d ⎩
si / i if |si | ≤ i
the reference command of the system, then the tracking
error vector can be defined as with i denoting the thickness of the defined boundary
⎡  ⎤ layer.
 u −  ud Finally, the developed high-level control law can be
xei = x1i − x1di = ⎣ w − wd ⎦ . (11) formulated as
θ − θd
νi = hi−1 (ẍ1di − kc2i ẋei − kc1 xei − f̂i ) − hi−1 kci sat(si /
i ).
With consideration of the defined tracking error vector (18)
in (11), the desired sliding surface for the studied CRW The corresponding adaptation law for estimating the
UAV is constructed as uncertain parameter f̂i is designed as
t
si = ẋei + kc2i xei + kc1i xei (τ )dτ − kc2i xei (t0 ) − ẋei (t0 ) ˙f̂ = a2 s (19)
i i i
t0
(12) where ai is a positive parameter that governs the adaptation
where t0 is the initial time instant, kc1i and kc2i represent the rate, si denotes the algebraic distance between the current
control gains. sliding variable si and the defined boundary layer i , which
In order to analyze the sliding motion associated with is defined as si = si − i sat(si / i ).
the defined integral sliding surface as shown in (12), the
derivative of the sliding surface is calculated as follows: REMARK 2 The variable si used for constructing the
adaptation law has the following features: when the sliding
ṡi = ẍei + kc2i ẋei + kc1i xei . (13) variable is inside the defined boundary layer, the variable

4564 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 will be equal to zero, i.e., si = 0; when the sliding variable THEOREM 1 Consider a nonlinear system with model un-
is outside the boundary layer, where the system tracking certainties as shown in (8), and the integral sliding surface
performance is not satisfactory, the adaptation will be trig- is defined in (12). By employing the feedback high-level
gered. In such a way, as long as the sliding variable is inside control law in (18) that is updated by the adaptation law
the boundary layer, the adaptation will be stopped. This in (19), the expected sliding motion can be achieved and
can contribute to avoiding overestimation of the uncertain maintained inside the defined boundary layer with satisfac-
parameter. tory tracking performance if the discontinuous control gain
is chosen as kci ≥ ηi with ηi denoting a positive small value.
Taking pitch control of the CRW UAV as an example, to
track a time-varying desired pitch angle θ d (t ), the tracking PROOF OF THEOREM 1 Consider the Lyapunov candidate
error states are defined as θe = [θe1 , θe2 ]T with θe1 = θ − θ d function as
and θe2 = θ̇ − θ̇ d . Then, the integral sliding surface based  
1 2
3
on the defined tracking error states can be designed as 1
V1 = si2 + 2 f̂i − fi . (30)
t i=1
2 ai
sθ = θe2 + kθ 2 θe1 + kθ 1 θe1 (τ )dτ − kθ 2 θe1 (t0 ) − θe2 (t0 )
t0 First, the time derivative of si , where si = 0, can be
(20) calculated as
where kθ 1 and kθ 2 are adjustable control gains.
Considering the uncertain portion My1 , the adaptive ṡi = ṡi = ẍei + kc2i ẋei + kc1i xei
 
model uncertainty compensation term is designed by mak- = hi hi−1 ẍ1di − kc2i ẋei − kc1i xei − f̂i
ing ṡθ = 0 as 
− kci sat(si / i )) + fi − ẍ1di + kc2i ẋei + kc1i xei
νθm = Iyy (θ̈d − kθ 2 θ̇e1 − kθ 1 θe1 ) − M̂y . (21) = fi − f̂i − kci sat(si / i ). (31)
Then, the robust feedback term that ensures the sliding Then, the time derivative of V1 is calculated as
motion is designed as
3  
νθr = −Iyy kcθ sat(sθ / θ) (22) V̇1 = si ṡi + ( f̂i − fi ) ˙f̂ i /ai2
i=1
where kcθ is an adjustable control gain. 3    
The uncertain portion is updated as
= si fi − f̂i − kci sat(si / i) + ( f̂i − fi ) ˙f̂ i /ai2
˙ = a 2 s
M̂ (23) i=1
y θ θ
3    
where sθ = sθ − θ sat(sθ / θ ), aθ is a positive parameter. = si − ˙f̂ i /ai2 fi − f̂i − kci sat(si / i )si .
In a similar fashion, the derivative of the sliding surface i=1
for u and w can be defined as (32)
ṡu = u̇e2 + ku2 ue2 + ku1 ue1 (24) Finally, substituting (19) into (32) leads to
ṡw = ẇe2 + kw2 we2 + kw1 we1 (25) 3
 
  d   V̇1 = −kci sat(si / i )si
where ue1 = u − u , ue2 = u − ud , we1 = w − wd ,
i=1
we2 = w − w d . ku1 , ku2 , kw1 , and kw2 are adjustable control
gains.
3
 
≤ −ηi |si | . (33)
The corresponding control law is designed, respectively,
i=1
as

νu = m (u̇d − ku2 ue2 − ku1 ue1 + qw + g sin θ )
Since there are three kinds of different actuators for pitch
− F̂x − mkcu sat(su / u) (26)
motion control under transition flight mode, CA approach
νw = m (ẇd − kw2 we2 − kw1 we1 − qu − g cos θ ) needs to be employed to appropriately distribute the de-
− F̂z − mkcw sat(sw / w) (27) sired virtual control signal νi over the available redundant
actuators. Given the system in (8), the control input uc is
where kcu and kcw are positive control gains, u and w computed employing a quadratic optimization approach,
represent boundary layer thickness. such that the conditions as shown in (9) and (10) can be
The uncertain parameters are then estimated as satisfied.
F̂˙ x = au2 su (28) LEMMA 1 The dynamic CA approach based on sequential
F̂˙ z = aw2 sw (29) quadratic programming can be described as [52]
 
uc = arg min W1 [uc − us ]2 + W2 [uc − uc (t − T )]2 (34)
where su = su − u sat(su / u ), sw = sw − w sat(sw / uc ∈
w ). au and aw are positive parameters that govern the
= arg min Wν [Buc − ν ] (35)
adaptation rate. umin ≤uc ≤umax

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4565

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
where uc ∈ Rm is the actual control input, us ∈ Rm is the Accordingly, the online adaptive scheme is formulated
desired steady-state control input, v ∈ Rn is the virtual as
control command generated from the high-level controller,
B ∈ Rn×m is the control effectiveness matrix. W1 , W2 , Wv ˆ˙ i = b2i si (40)
represent the weighting matrices with proper dimensions, ˆ˙ i = ci2 (−ẍ1di + kc2i ẋei + kc1i xei + kci sat(si / i ))si (41)
and  ·  denotes the Euclidean 2-norm.
where bi and ci are positive parameters that govern the
B. Design of Adaptive FTC Scheme With Both Actuator adaptation rate and can be tuned based on the actual tracking
Faults and Model Uncertainties control performance.
After actuator faults occurrence, the diagonal matrix Lc
in (9) becomes a nonidentity matrix. In this condition, the THEOREM 2 Given an overactuated nonlinear system (8)
following equation can be obtained: with actuator faults and model uncertainties, by employ-
ing the developed high-level control law (39), the online
νi = Bi Iuc − Bi (I − Lc )uc (36) adaptation laws (40) and (41), and the dynamic CA scheme
where I is an identity matrix. (34) and (35), the expected sliding motion can be achieved
Let νei = −Bi (I − Lc )uc , the system dynamics in (8) can and kept inside the boundary layer even in the presence
be rewritten as of both actuator faults and model uncertainties with the
discontinuous control gain chosen as kci ≥ ηi .
ẋ2i = fi (x1 , x2 ) + hi (x1 , x2 )(vdi + νei ) (37)
PROOF OF THEOREM 2 Consider the Lyapunov candidate
where νdi = Bi Iuc denotes the desired virtual control signal
function as
from the high-level controller.
When actuator faults occur, since the CA module does 3  
1 1 2 1 2
not have the postfault information, there exist virtual control V2 = si + 2 ( i − i ) + 2 ( i − i ) .
2 ˆ ˆ
2 bi i ci i
errors. This will degrade the overall system tracking perfor- i=1
mance or even make the system lose its stability. In order (42)
to maintain the closed-loop system tracking performance, Then, the derivative of the selected Lyapunov candidate
the high-level controller needs to be reconfigured in this function can be calculated as
circumstance. Observed from (37), when there is an error νei 3
 
between the desired virtual control signal νdi and the actual V̇2 = si −1
i i + −1
i
ˆ i (ẍ1di − kc2i ẋei − kc1i xei )
virtual control signal νi , in order to maintain the high-level i=1
control tracking performance, the parameter hi needs to be −1 −1 ˆ
− i
ˆ i kci sat(si / i) − i i − ẍ1di
adaptively adjusted for eliminating the virtual control error. −1
For this reason, let hi νei = h̃i νdi , thus, (37) can be rewritten + kc2i ẋei + kc1i xei ] + i
(ˆi − i) i
ˆ˙
as b2i
−1 
ẋ2i = fi (x1 , x2 ) + (hi + h̃i )νdi = fi (x1 , x2 ) + ĥi νdi . (38) + i
( ˆ i − i ) ˆ˙ i
ci2
REMARK 3 After actuator faults occurrence, the actual vir-
tual control signal will not be equal to the desired virtual
3
 −1
control signal generated from the high-level controller. In = si (ẍ1di − kc2i ẋei − kc1i xei )( i
ˆ i − 1)
i=1
this circumstance, by adaptively changing the parameter −1
i ( i −
ĥi , more virtual control signals will be generated from + si ˆ i ) − si kci sat(si / i)
−1 ˆ
the high-level controller. This can help to compensate the − si ( i i − 1)kci sat(si / i)
adverse effect caused by actuator faults and maintain the −1 
˙ 1 −1 ˆ ˙
original tracking performance of the system. + i
− i) i + 2 ( i
(ˆi ˆ i − 1) i
ˆ
b2i ci
In addition, to fully use both control portions of SMC,  
the estimated parameter ĥi is added to make adaptive change
3
ˆ˙ i
−1 ˆ
= ( i i − 1) 2
− (−ẍ1di + kc2i ẋei + kc1i xei
of both continuous and discontinuous control portions of i=1
c i
SMC in the presence of actuator faults. Moreover, by recall- 
+ kci sat(si / i ))si − kci sat(si / i )si
ing the previously designed control law for compensating  
model uncertainties, both estimations f̂i and ĥi should be ˆ˙ i
used for deriving the corresponding control law. + i−1 ( ˆ i − i ) 2 − si . (43)
bi
In this case, for the ease of controller design, by denoting
ˆ i = ĥ−1 f̂i , ˆ i = ĥ−1 , i = h−1 fi , i = h−1 , and consider- Substituting (40) and (41) into (43) leads to
i i i i
ing the sliding surface in (12), the high-level control law
can be redesigned as ˆ˙ i  
− −ẍ1di + kc2i ẋei + kc1i xei + kci sat(si / i) si = 0
νi = ˆ i (ẍ1di − kc2i ẋei − kc1i xei ) − ˆ i − ˆ i kci sat(si / i ).
ci2
(39) (44)

4566 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 Fig. 6. Pitch motion tracking performance of the hybrid CRW UAV in Fig. 8. Vertical velocity tracking performance of the hybrid CRW UAV
Scenario 1. in Scenario 1.

Fig. 7. Forward velocity tracking performance of the hybrid CRW UAV

in Scenario 1.
Fig. 9. Deflection of canard control surface of the hybrid CRW UAV in
 
  ˆ˙ i Scenario 1.
−1 ˆi −
i i − si = 0. (45)
b2i
In this sense, the following equation can be obtained:
3
 
V̇2 = −kci sat(si / i )si . (46)
i=1

If si > 0, then sat(si / i ) = sign(si ) = 1. If si < 0,

then sat(si / i ) = sign(si ) = −1. Therefore, the following
equation can be obtained as sat(si / i )si = |si |. In ad-
dition, considering that the discontinuous control gain is
chosen as kci ≥ ηi , (46) can be rewritten as
3
 
V̇2 ≤ −ηi |si | . (47)
i=1

Therefore, with the proposed adaptive control strategy,

the system stability and tracking performance can be main-
tained under both actuator faults and model uncertainties.
REMARK 4 Although actuator faults are additionally
considered in this section, with the help of the developed Fig. 10. Deflection of elevator of the hybrid CRW UAV in Scenario 1.

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4567

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
Fig. 11. Deflection of longitudinal cyclic pitch of the hybrid CRW UAV Fig. 13. Thrust input of the hybrid CRW UAV in Scenario 1.
in Scenario 1.

Fig. 12. Collective pitch input of the hybrid CRW UAV in Scenario 1.

Fig. 14. Change of adaptive control parameters for forward velocity in

adaptive schemes, the value of the discontinuous control Scenario 1.
gain is not increased for compensating both actuator faults
and model uncertainties compared to the one that is only
for accommodating model uncertainties. This feature can
preserve the original tracking performance and prevent the
unexpected chattering effect under fault-free conditions.

IV. SIMULATION RESULTS AND DISCUSSIONS

In this section, in order to validate the performance
of the proposed adaptive fault-tolerant control scheme
(PAFTCS), a series of simulations based on the studied
hybrid CRW UAV under model uncertainties and different
faulty scenarios are carried out. The related physical
parameters of the studied hybrid CRW UAV are set
as m = 16.13 kg, Ixx = 2.4 kg · m2 , Iyy = 7.1 kg · m2 ,
Izz = 6.9 kg · m2 . The position saturation value for cyclic
pitch control is [−10, 10] deg, and the position saturation
values for canard control surface and elevator are both set
as [−20, 20] deg. The rate saturation values for the control
surfaces are all set as [−60, 60] deg/s. The solver used
for simulation is chosen as ode4, and the sampling time is Fig. 15. Change of adaptive control parameters for vertical velocity in
chosen as 0.01 s. For the purpose of comparison, the tracking Scenario 1.

4568 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 Fig. 16. Change of adaptive control parameters for pitch motion in Fig. 19. Vertical velocity tracking performance of the hybrid CRW
Scenario 1. UAV in Scenario 2.

Fig. 17. Pitch motion tracking performance of the hybrid CRW UAV in
Fig. 20. Deflection of canard control surface of the hybrid CRW UAV
Scenario 2.
in Scenario 2.

Fig. 18. Forward velocity tracking performance of the hybrid CRW

UAV in Scenario 2. Fig. 21. Deflection of elevator of the hybrid CRW UAV in Scenario 2.

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4569

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
Fig. 22. Deflection of longitudinal cyclic pitch of the hybrid CRW UAV Fig. 24. Thrust input of the hybrid CRW UAV in Scenario 2.
in Scenario 2.

Fig. 23. Collective pitch input of the hybrid CRW UAV in Scenario 2. Fig. 25. Change of adaptive control parameters for forward velocity in
Scenario 2.
performance of a conventional adaptive sliding-mode
control (CASMC) and a nominal sliding-mode control
without adaptation (NSMCWA) are also demonstrated
as the high-level controllers that are combined with
the same low-level CA module as the proposed
control scheme. The control parameters are chosen as
kc11 = 2, kc21 = 1, kc1 = 1, kc12 = 10, kc22 = 25, kc2 = 1,
kc13 = 20, kc23 = 100, kc3 = 1, 1 = 2 = 3 = 0.01, b1
= 10, b2 = 5, b3 = 5, c1 = 10, c2 = 5, c3 = 5, η1 = η2 =
η3 = 0.1, W1 = diag([1, 1, 1]), W2 = diag([1, 1, 1]),
Wν = 1. The control effectiveness matrix B = [B1 , B2 , B3 ]
is given as B1 = 25, B2 = 9 u, B3 = −11 u with u denoting
the forward speed. Two different scenarios are tested and
demonstrated, which are detailed in Table II. In Scenario
1, the CRW UAV first hovers in helicopter mode. Then, the
UAV is commanded to increase the forward speed to transit
from helicopter mode to airplane mode. During this stage,
the UAV is controlled to keep the vertical speed as zero and
track the desired pitch command. A 50% loss of control
effectiveness fault is injected into elevator at 20 s, then, Fig. 26. Change of adaptive control parameters for vertical velocity in
the severity of injected fault becomes worse as 80% loss Scenario 2.

4570 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 canard, and elevator are almost the same for the developed
PAFTCS and the compared CASMC and NSMCWA
when the first fault occurs at 20 s. However, as the
considered fault becomes larger at 30 s, the use of control
inputs for the developed PAFTCS is the smallest. This
indicates that the developed PAFTCS does not use more
control efforts for accommodating the faults. However,
as can be observed from Figs. 12 and 13, the compared
NSMCWA uses more thrust and collective pitch inputs
than the developed PAFTCS for maintaining the desired
velocities.
In Scenario 2, in order to further investigate the per-
formance of the proposed control strategy, in addition to
the partial loss of control effectiveness fault in elevator as
examined in Scenario 1, a lock-in-place failure is injected
into the canard control surface at 30 s with the stuck
Fig. 27. Change of adaptive control parameters for pitch motion in position at −3◦ . Moreover, the white noise with a mean
Scenario 2. of 0 and covariance of 0.01 is injected into each measure-
TABLE II
ment channel. The corresponding tracking performances of
Faulty Scenarios Considered in Simulation Tests pitch motion and longitudinal velocities are demonstrated in
Figs. 17–19. Similar to Scenario 1, the tracking performance
of longitudinal velocities are not heavily affected in the pres-
ence of both elevator and canard actuator faults. For the pitch
motion control, a 50% loss of control effectiveness fault is
injected into elevator at 20 s. With the adaptive change of
control parameters as shown in Figs. 25–27, the developed
PAFTCS can effectively compensate the faults, whereas the
of control effectiveness at 30 s. The tracking performances compared NSMCWA fails to maintain the original system
of pitch motion and longitudinal velocities are shown in tracking performance. After that, the elevator fault becomes
Figs. 6–8. As can be observed from Figs. 7 and 8, the larger as 80% loss of control effectiveness whereas the ca-
tracking performances of forward and vertical speed for nard control surface gets stuck with the amplitude of −3◦ at
the developed PAFTCS and the compared CASMC and 30 s. Under this faulty condition, the compared NSMCWA
NSMCWA schemes are almost the same. At the initial demonstrates a large tracking error for the pitch motion and
stage of the transition process when there is no fault, with the compared CASMC can track the desired pitch angle but
the increase of forward speed, the inference between the with small tracking errors. On the contrary, with the help of
main rotor and the horizontal tail will have an adverse the developed adaptive control schemes and the redundant
effect on pitch motion control. The developed PAFTCS elevator and longitudinal cyclic pitch control, the developed
can adaptively change the control parameter to strive for PAFTCS can instantaneously increase the control inputs to
accurate tracking performance, and the desired tracking the redundant available control effectors to accommodate
performance of pitch motion can be quickly maintained. the faults, as shown in Figs. 20–22. Compared to Scenario
On the contrary, the compared CASMC and NSMCWA 1, due to the additional fault in canard control surface, the
schemes track the desired pitch angle with tracking errors. use of elevator and longitudinal cyclic pitch control are
When actuator faults occur at 20 s and 30 s, respectively, increased. In addition, the collective pitch and thrust inputs
the existence of the virtual control error caused by actuator are also affected that can be observed from Figs. 23 and 24.
faults will trigger the synthesized adaptive schemes. By The values of the corresponding adaptive control parameters
adaptively changing the control parameters as shown are also increased, which can be observed from Figs. 16 and
in Figs. 14–16, the developed PAFTCS can make a 27.
quick compensation to maintain the desired tracking Furthermore, in order to quantitatively evaluate the
performance of pitch motion. Nevertheless, the compared demonstrated control performance, the average use of the
CASMC scheme stimulates unexpected control chattering control inputs (T̄ , θ̄0 , θ̄B , δ̄c , δ̄e ) is selected as one of the cri-
and the compared NSMCWA fails to maintain the desired teria, that is mainly for evaluating the energy consumption.
tracking performance and the tracking error increases as In addition, considering that the longitudinal velocity track-
the fault getting worse. The corresponding control inputs ing performances are almost the same for the demonstrated
are illustrated in Figs. 9–13. When elevator experiences control schemes, define the following criteria for evaluating
faults, the deflections of longitudinal cyclic pitch, canard, the pitch motion tracking performance as
and elevator are all increased accordingly to eliminate 
t1
the adverse impact of faults. In addition, in order to 1 2
eRMSE = (θ d − θ ) dτ (48)
compensate the faults, the use of longitudinal cyclic pitch, t1 − t0 t0

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4571

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 TABLE III
Performance Indices

where [t0 , t1 ] covers the overall time frame of the test. [5] Z. Liu, Y. He, L. Yang, and J. Han, “Control techniques of tilt rotor
The performance indices are listed in Table III. The unmanned aerial vehicle systems: A review,” Chin. J. Aeronaut.,
quantitative analysis confirms that the developed PAFTCS vol. 30, no. 1, pp. 135–148, 2017.
[6] A. S. Saeed, A. B. Younes, C. Cai, and G. Cai, “A survey of hybrid
not only ensures a satisfactory tracking performance of the unmanned aerial vehicles,” Prog. Aerosp. Sci., vol. 98, pp. 91–105,
CRW UAV but also preserves a low energy consumption 2018.
with relatively small control inputs. [7] G. J. Ducard and M. Allenspach, “Review of designs and flight
control techniques of hybrid and convertible VTOL UAVs,” Aerosp.
Sci. Technol., vol. 118, 2021, Art. no. 107035.
V. CONCLUSION [8] D. N. Cardoso, S. Esteban, and G. V. Raffo, “A new robust
In this article, an adaptive FTC scheme is proposed adaptive mixing control for trajectory tracking with improved for-
ward flight of a tilt-rotor UAV,” ISA Trans., vol. 110, pp. 86–104,
for an overactuated hybrid CRW UAV to simultaneously 2021.
compensate actuator faults and model uncertainties without [9] M. Sato and K. Muraoka, “Flight controller design and demonstration
the requirement of information on fault and uncertainty. of quad-tilt-wing unmanned aerial vehicle,” J. Guid. Control Dynam.,
The proposed control scheme is divided into two separate vol. 38, no. 6, pp. 1071–1082, 2015.
control modules: the high-level adaptive SMC module and [10] Y. Ke, K. Wang, and B. M. Chen, “Design and implemen-
tation of a hybrid UAV with model-based flight capabilities,”
the low-level CA module. The low-level CA module is used IEEE/ASME Trans. Mechatron., vol. 23, no. 3, pp. 1114–1125,
to distribute the virtual control signals that are generated by Jun. 2018.
the high-level control module among the redundant avail- [11] F. Li, W.-P. Song, B.-F. Song, and H. Zhang, “Dynamic modeling,
able actuators. The high-level control module is constructed simulation, and parameter study of electric quadrotor system of quad-
by an adaptive sliding-mode controller, which is employed plane UAV in wind disturbance environment,” Int. J. Micro Air Veh.,
vol. 18, pp. 1–23, 2021.
to maintain the closed-loop system tracking performance [12] H. Gao, A. He, Z. Gao, Y. Na, and Y. Deng, “Flight dy-
in both faulty and fault-free conditions. In the case of namics characteristics of canard rotor/wing aircraft in helicopter
actuator faults, the adaptive scheme will be triggered to flight mode,” Chin. J. Aeronaut., vol. 32, no. 7, pp. 1577–1587,
generate more virtual control signals. With the help of the 2019.
synthesized adaptive scheme, the control parameter can be [13] J. W. Rutherford, S. M. Bass, and S. D. Larsen, “Ca-
nard rotor/wing: A revolutionary high-speed rotorcraft con-
changed adaptively to compensate the virtual control error. cept,” in Proc. AIAA/AHS/ASEE Aerosp. Des. Conf., 1993,
The comparative simulation results show the effectiveness Art. no. 1175.
and advantages of the proposed strategy in the presence of [14] S. Bass, T. Thompson, J. Rutherford, and S. Swanson,
both partial loss actuator fault and lock-in-place actuator “Low-speed wind tunnel test results of the canard rotor/wing
failure and model uncertainties. concept,” in Proc. 11th Appl. Aerodynam. Conf., 1993,
Art. no. 3412.
[15] C. Mitchell and B. Vogel, “The canard rotor wing (CRW) aircraft-A
REFERENCES new way to fly,” in Proc. AIAA Int. Air Space Sym. Expo.: Next 100
Years, 2003, Art. no. 2517.
[1] J. Scherer and B. Rinner, “Multi-UAV surveillance with minimum [16] W. Gai, H. Wang, T. Guo, and D. Li, “Modeling and LPV flight
information idleness and latency constraints,” IEEE Robot. Autom. control of the canard rotor/wing unmanned aerial vehicle,” in Proc.
Lett., vol. 5, no. 3, pp. 4812–4819, Jul. 2020. 2nd Int. Conf. Artif. Intell. Manage. Sci. Electron. Commer., 2011,
[2] C. Yuan, Y. M. Zhang, and Z. X. Liu, “A survey on technologies pp. 2187–2191.
for automatic forest fire monitoring, detection, and fighting using [17] W. Gai, H. Wang, and D. Li, “Flight dynamic modeling and analysis
unmanned aerial vehicles and remote sensing techniques,” Can. J. for the canard rotor/wing UAV,” Acta Aerodynam. Sinica, vol. 30,
For. Res., vol. 45, no. 7, pp. 783–792, 2015. no. 2, pp. 244–249, 2012.
[3] F. Munoz, S. Zuniga, L. R. G. Carrillo, E. S. Espinoza, S. Salazar, [18] W. Sun, Z. Gao, Y. Du, and F. Xu, “Mechanism of unconventional
and M. A. Marquez, “Adaptive fuzzy consensus control strategy for aerodynamic characteristics of an elliptic airfoil,” Chin. J. Aeronaut.,
UAS-based load transportation tasks,” IEEE Trans. Aerosp. Electron. vol. 28, no. 3, pp. 687–694, 2015.
Syst., vol. 57, no. 6, pp. 3844–3860, Dec. 2021. [19] H. Gao, Z. Gao, Y. Na, and C. Pang, “Trim strategy, control
[4] J. Zhong and W. Chen, “Transition characteristics for a small tail- model, and flight dynamics characteristics of canard rotor/wing
sitter unmanned aerial vehicle,” Chin. J. Aeronaut., vol. 34, no. 10, aircraft in transition mode,” IEEE Access, vol. 7, pp. 66042–66054,
pp. 220–236, 2021. 2019.

4572 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 [20] H. Gao, Z. Liu, B. Wang, and C. Pang, “Flight dynamics and control [40] H. Alwi and C. Edwards, “Fault tolerant control using sliding
of a new VTOL aircraft in fixed-wing mode,” in Proc. Int. Conf. modes with on-line control allocation,” Automatica, vol. 44, no. 7,
Unmanned Aircr. Syst., 2020, pp. 1650–1657. pp. 1859–1866, 2008.
[21] Z. Li, L. Zhang, H. Liu, Z. Zuo, and C. Liu, “Nonlinear robust [41] W. C. Durham, “Constrained control allocation,” J. Guid. Control
control of tail-sitter aircrafts in flight mode transitions,” Aerosp. Sci. Dyn., vol. 16, no. 4, pp. 717–725, 1993.
Technol., vol. 81, pp. 348–361, 2018. [42] T. A. Johansen and T. I. Fossen, “Control allocation—A survey,”
[22] N. Liu, Z. Cai, Y. Wang, and J. Zhao, “Fast level-flight to hover mode Automatica, vol. 49, no. 5, pp. 1087–1103, 2013.
transition and altitude control in tiltrotor’s landing operation,” Chin. [43] S. Park, J. Bae, Y. Kim, and S. Kim, “Fault tolerant flight control
J. Aeronaut., vol. 34, no. 1, pp. 181–193, 2021. system for the tilt-rotor UAV,” J. Franklin Inst., vol. 350, no. 9,
[23] W. Gai, J. Zhang, L. Huang, and Y. Li, “Transition flight control using pp. 2535–2559, 2013.
adaptive neutral network dynamic inversion for canard rotor/wing [44] W. Gai, J. Liu, J. Zhang, and Y. Li, “A new closed-loop control
UAV,” in Proc. 26th Chin. Control Decis. Conf., 2014, pp. 4210– allocation method with application to direct force control,” Int. J.
4214. Control Autom. Syst., vol. 16, no. 3, pp. 1355–1366, 2018.
[24] B. Xian, J. Guo, Y. Zhang, and B. Zhao, “Sliding mode tracking [45] Z. Liu, D. Theilliol, L. Yang, Y. He, and J. Han, “Mode transition
control for miniature unmanned helicopters,” Chin. J. Aeronaut., and fault tolerant control under rotor-tilt axle stuck fault of quad-
vol. 28, no. 1, pp. 277–284, 2015. TRUAV,” IFAC-PapersOnLine, vol. 51, no. 24, pp. 991–997, 2018.
[25] K. T. Öner, E. Çetinsoy, E. Sırımoğlu, C. Hancer, T. Ayken, and [46] J. Tjønnås and T. A. Johansen, “Adaptive control allocation,” Auto-
M. Ünel, “LQR and SMC stabilization of a new unmanned aerial matica, vol. 44, no. 11, pp. 2754–2765, 2008.
vehicle,” World Acad. Sci. Eng. Technol., vol. 3, no. 10, pp. 367–372, [47] S. S. Tohidi, Y. Yildiz, and I. Kolmanovsky, “Adaptive control
2009. allocation for constrained systems,” Automatica, vol. 121, 2020,
[26] A. L’afflitto, R. B. Anderson, and K. Mohammadi, “An introduction Art. no. 109161.
to nonlinear robust control for unmanned quadrotor aircraft: How to [48] A. Casavola and E. Garone, “Fault-tolerant adaptive control allo-
design control algorithms for quadrotors using sliding mode control cation schemes for overactuated systems,” Int. J. Robust Nonlinear
and adaptive control techniques,” IEEE Control Syst. Mag., vol. 38, Control, vol. 20, no. 17, pp. 1958–1980, 2010.
no. 3, pp. 102–121, Jun. 2018. [49] S. S. Tohidi, Y. Yildiz, and I. Kolmanovsky, “Sliding mode control
[27] Y. Zou, “Nonlinear robust adaptive hierarchical sliding mode control for over-actuated systems with adaptive control allocation and its
approach for quadrotors,” Int. J. Robust Nonlinear Control, vol. 27, applications to flight control,” in Proc. IEEE Conf. Control Technol.
no. 6, pp. 925–941, 2017. Appl., 2021, pp. 765–770.
[28] Y. M. Zhang and J. Jiang, “Bibliographical review on reconfigurable [50] A. Argha, S. W. Su, Y. Liu, and B. G. Celler, “Control allocation
fault-tolerant control systems,” Annu. Rev. Control, vol. 32, no. 2, based sliding mode fault tolerant control,” in Proc. Amer. Control
pp. 229–252, 2008. Conf., 2019, pp. 3752–3757.
[29] J. Marzat, H. Piet-Lahanier, F. Damongeot, and E. Walter, “Model- [51] W. Durham, Aircraft Flight Dynamics and Control. Hoboken, NJ,
based fault diagnosis for aerospace systems: A survey,” Proc. Inst. USA: Wiley, 2013.
Mech. Eng. G., J. Aerosp. Eng., vol. 226, no. 10, pp. 1329–1360, [52] O. Härkegård, “Dynamic control allocation using constrained
2012. quadratic programming,” J. Guid. Control Dyn., vol. 27, no. 6,
[30] K. Rudin, G. J. Ducard, and R. Y. Siegwart, “Active fault-tolerant pp. 1028–1034, 2004.
control with imperfect fault detection information: Applications
to UAVs,” IEEE Trans. Aerosp. Electron. Syst., vol. 56, no. 4,
pp. 2792–2805, Aug. 2020.
[31] M. Mahmoud, J. Jiang, and Y. M. Zhang, Active Fault Tolerant Con-
trol Systems: Stochastic Analysis and Synthesis. Berlin, Germany:
Ban Wang (Member, IEEE) received the B.S.
Springer, 2003. degree in flight vehicle propulsion engineering
[32] G. J. Ducard, Fault-Tolerant Flight Control and Guidance Systems: and the M.S. degree in flight vehicle design
Practical Methods for Small Unmanned Aerial Vehicles. Berlin, engineering from Northwestern Polytechnical
Germany: Springer, 2009. University, Xi’an, China, in 2011 and 2014,
[33] H. Alwi, C. Edwards, and C. P. Tan, Fault Detection and Fault- respectively, and the Ph.D. degree in mechan-
Tolerant Control using Sliding Modes. Berlin, Germany: Springer, ical engineering from the Department of Me-
2011. chanical, Industrial and Aerospace Engineering,
[34] B. Wang and Y. M. Zhang, “An adaptive fault-tolerant sliding mode Concordia University, Montreal, QC, Canada, in
2018.
control allocation scheme for multirotor helicopter subject to simul-
He is currently an Associate Professor with
taneous actuator faults,” IEEE Trans. Ind. Electron., vol. 65, no. 5, the School of Aeronautics, Northwestern Polytechnical University, Xi’an,
pp. 4227–4236, May 2018. China. His research interests include fault detection, fault diagnosis, and
[35] X. Yu, L. Guo, Y. M. Zhang, and J. Jiang, Autonomous Safety Control fault-tolerant control with applications to aircraft and unmanned aerial
of Flight Vehicles. Boca Raton, FL, USA: CRC, 2021. vehicles, and real-time path planning/replanning with obstacle avoidance.
[36] Q. L. Hu, B. Li, B. Xiao, and Y. M. Zhang, Control Allocation for
Spacecraft Under Actuator Faults. Berlin, Germany: Springer, 2021.
[37] Q. Zong, X. Zhang, S. Shao, B. Tian, and W. Liu, “Disturbance
observer-based fault-tolerant attitude tracking control for rigid space-
craft with finite-time convergence,” Proc. Inst. Mech. Eng. G., J.
Aerosp. Eng., vol. 233, no. 2, pp. 616–628, 2019. Dehai Zhu received the B.S. and M.S. degrees
[38] W. Ren, B. Jiang, and H. Yang, “Fault-tolerant control of singu- in flight vehicle design engineering from North-
larly perturbed systems with applications to hypersonic vehicles,” western Polytechnical University, Xi’an, China,
IEEE Trans. Aerosp. Electron. Syst., vol. 55, no. 6, pp. 3003–3015, in 2019 and 2022, respectively.
Dec. 2019. He is currently with the China Academy of
[39] Z. Liu, Z. Han, and W. He, “Adaptive fault-tolerant boundary Launch Vehicle Technology, Beijing, China. His
research interests include flight vehicle design,
control of an autonomous aerial refueling hose system with pre-
flight control and simulation, and reinforcement
scribed constraints,” IEEE Trans. Autom. Sci. Eng., vol. 19, no. 4, learning control.
pp. 2678–2688, Oct. 2022.

WANG ET AL.: ADAPTIVE FAULT-TOLERANT CONTROL OF A HYBRID CANARD ROTOR/WING UAV UNDER 4573

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.
 Linying Han received the B.S. and M.S. degrees Youmin Zhang (Fellow, IEEE) received the
in flight vehicle design engineering from North- Ph.D. degree in automation science and engi-
western Polytechnical University, Xi’an, China, neering from Northwestern Polytechnical Uni-
in 2019 and 2022, respectively. versity, Xi’an, China, in 1995.
She is currently with the Luoyang Electro- He is currently a Professor with the Depart-
Optical Equipment Research Institute, China ment of Mechanical, Industrial and Aerospace
Aviation Industry Corporation, Luoyang, China. Engineering and the Concordia Institute of
Her research interests include flight control and Aerospace Design and Innovation, Concordia
simulation, flight dynamics, and fire control and University, Montreal, QC, Canada. His current
command control. research interests include condition monitoring,
fault detection and diagnosis and fault-tolerant
Honggang Gao received the B.S. degree in au- control, and cooperative guidance, navigation, and control of single- and
tomation from the Xi’an University of Technol- multiple-unmanned aerial/space/ground/marine vehicles.
ogy, Xi’an, China, in 2010 and the M.S. and Dr. Zhang is currently the President of the International Society of
Ph.D. degrees in control theory and control en- Intelligent Unmanned Systems and (advisory) member of the technical
gineering and flight vehicle design from North- committees for several scientific societies. He is also an (advisory) editorial
western Polytechnical University, Xi’an, China, board member, (deputy) editor-in-chief, and editor/associate editor for
in 2013 and 2020, respectively. several international journals. He was the honorary general chair, general
He is currently an Assistant Research Fellow chair, program chair, and an international program committee member of
with the School of Civil Aviation, Northwest- several international conferences. He is a Fellow of the Canadian Society
ern Polytechnical University, Xi’an, China. His for Mechanical Engineering and a senior member of the American Institute
research interests include modeling, flight iden- of Aeronautics and Astronautics.
tification, flight dynamics, flight control, and simulation.

Zhenghong Gao received the B.S., M.S., and

Ph.D. degrees in aerodynamics from Northwest-
ern Polytechnical University, Xi’an, China, in
1981, 1984, and 1989, respectively.
From 1992 to 1994, she was a Postdoctoral
Researcher with the Institute of Aerodynamics
and Fluid Mechanics, Technical University of
Munich, Munich, Germany. She is currently a
Professor with the School of Aeronautics, North-
western Polytechnical University, Xi’an, China.
Her research interests include the multidisci-
plinary design optimization of aircraft, flight mechanics and flight control,
and aircraft aerodynamics.
Dr. Gao is a Committee Member of the International Council of the
Aeronautical Sciences, and the Executive Director of Chinese Society of
Aeronautics and Astronautics.

4574 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 59, NO. 4 AUGUST 2023

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on September 11,2023 at 02:21:22 UTC from IEEE Xplore. Restrictions apply.