Professional Documents
Culture Documents
Control of Centrally-Powered Variable Pitch Propeller Quadcopters Subject To Propeller Faults
Control of Centrally-Powered Variable Pitch Propeller Quadcopters Subject To Propeller Faults
a r t i c l e i n f o a b s t r a c t
Article history: Research into variable pitch propeller (VPP) quadcopters has seen a marked increase in recent years
Received 25 May 2021 which is due to their enhanced dynamic capabilities compared to conventional fixed pitch propeller
Received in revised form 1 November 2021 quadcopters. Adding actuators to control the pitch angles of the propellers increases the mechanical
Accepted 24 November 2021
complexity and hence may increase the risk of faults. In this paper, the flight control of a centrally-
Available online 29 November 2021
Communicated by Xiande Fang
powered VPP quadcopter in the presence of a propeller fault is studied. This problem has not been
studied in the literature. In this paper firstly the balance trajectory is analysed. The uncontrollable mode
Keywords: is identified next. Finally, a linear controller is proposed. It is shown that the yaw angle and angular
Variable pitch propeller quadcopter velocity become uncontrollable in the presence of a VPP fault, yet the quadcopter can still accurately track
Flight control a desired trajectory. It is also discovered that the quadcopter exhibits different and favourable behaviour,
Fault control such as slow self-spinning speed. The relationship under certain parameter conditions is analysed and
the parameter conditions that lead to zero self-spinning are identified. Our analysis could contribute to
the development of high-performance quadcopters that are both agile and robust with respect to faults.
Simulation results are presented to verify the theoretical findings.
© 2021 Elsevier Masson SAS. All rights reserved.
1. Introduction The difference between a VPP and a fixed pitch propeller is that
an actuator is used to control the pitch angle. Adding such actua-
Recent years have witnessed the rapid development of com- tors remarkably enhances the flight capability of a quadcopter in a
mercial micro unmanned aerial vehicles (UAVs) across the world. number of ways. Firstly, a VPP can generate forces in either upward
Due to the simplicity and low cost of their mechanical and dynam- or downward directions. This novel property brings benefits to the
ical systems, quadcopter UAVs have been successfully applied in flight control performance. For example, a VPP quadcopter will be
many domains [1,2]. In the future, quadcopter UAVs are expected able to fly upside down steadily, which is not possible to achieve
to be routinely utilised in common, day-to-day tasks, such as par- with a conventional quadcopter. It also enables VPP quadcopters
cel delivery or passenger transportation [3]. Many of these tasks to recover swiftly from largely disturbed attitudes back to stable
are safety-critical and require UAVs to achieve high flight perfor- hovering conditions, which is important considering the safety of
mance in terms of both agility and safety [4–7]. the system. Secondly, as the thrust magnitude of a VPP is con-
Variable pitch propeller (VPP) quadcopters are generalised vari- trolled by adjusting the pitch angle, the control bandwidth is much
ants of fixed pitch propeller quadcopters. They have attracted in- higher than with spinning speed control, the method used by con-
creasing attention in recent years [8–11]. One advantage of them is ventional fixed pitch propellers to adjust their thrust magnitude.
that they have the same overall control structure as conventional The drawback of adding actuators to adjust pitch angles is that it
fixed pitch propeller quadcopters. Hence, they inherit the simplic- increases the mechanical complexity and hence may increase the
ity of the overall dynamical structure from fixed pitch propeller fault rate. It is therefore important to study flight control of VPP
quadcopters, and therefore have the potential to be widely applied quadcopters in the presence of actuator faults.
in practice. Although fault-tolerant control of quadcopters has been studied
extensively [12–15], the case of VPP quadcopters remains largely
unexplored. Two recent works [11,16] have studied control of
* Corresponding author. separately-powered VPP (SVPP) quadcopters with one faulty pro-
E-mail address: zhaoshiyu@westlake.edu.cn (S. Zhao). peller. If one VPP is faulty, an SVPP quadcopter still has six in-
https://doi.org/10.1016/j.ast.2021.107245
1270-9638/© 2021 Elsevier Masson SAS. All rights reserved.
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
⎡ ⎤
⎡ ⎤ u
ẋ
⎢ v ⎥
⎢ ẏ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ w ⎥
⎢ ż ⎥ ⎢ (cos ψ sin θ cos φ+sin ψ sin φ)u −k u |u | ⎥
⎢ ⎥=⎢ 1 β ⎥, (1)
Fig. 1. The schematic diagram of a quadcopter. ⎢ u̇ ⎥ ⎢ m ⎥
⎣ v̇ ⎦ ⎢ (sin ψ sin θ cos φ−cos ψ sin φ)u 1 −kβ v | v | ⎥
⎣ m
⎦
ẇ (cos θ cos φ)u 1 −kβ w | w |
dependent control inputs (i.e., three motor spinning speeds and m
−g
three pitch angles), and the entire system remains controllable. and the rotational motion of the quadcopter is described by
By contrast, our present work addresses the control of centrally-
⎡ ⎤
powered variable pitch propeller (CVPP) quadcopters. For a CVPP ⎡ ⎤ p + q sin φ tan θ + r cos φ tan θ
φ̇ ⎢ ⎥
quadcopter, all propellers spin at the same constant speed; they
⎢ θ̇ ⎥ ⎢ q cos φ − r sin φ ⎥
are driven by the same motor located at the centre of the body ⎢ ⎥ ⎢ q sin φ sec θ + r cos φ sec θ ⎥
⎢ ψ̇ ⎥ ⎢ ⎥
[17,18]. If one propeller is faulty, there would only be three inde- ⎢ ⎥ = ⎢ qr I z − I y + u l − p | p | kγ ⎥, (2)
⎢ ṗ ⎥ ⎢ 3 Ix ⎥
pendent control inputs, and the entire system becomes uncontrol- ⎢ ⎥ ⎢ Ix Ix
⎥
lable and hence more challenging to handle than the SVPP case. ⎣ q̇ ⎦ ⎢ rp I x − I z + u 2 l − q|q| kγ ⎥
⎣ Iy Iy Iy ⎦
The main theoretical contribution of this paper is to discover ṙ kγ
u 4 Il − r |r | I
that a CVPP quadcopter can exhibit different but favourable be- z z
haviour such as a slow self-spinning speed in the presence of a where I x , I y , and I z are the moments of inertia, l is the half length
propeller fault. The relationship between the self-spinning speed of the diagonal wheelbase of the CVPP quadcopter, m is the mass
and the parameter conditions is analysed and the parameter con- of the CVPP quadcopter, and g is the gravitational acceleration con-
ditions that lead to zero self-spinning in the presence of one pro- stant. The inputs u 1 , u 2 , u 3 , u 4 in (1) and (2) are given by
peller fault are identified. Such a discovery is favourably surprising.
⎡ ⎤ ⎡ ⎤
The theoretical findings are supported by analysing the equilib- u1 f1 + f2 + f3 + f4
rium trajectory, identifying the uncontrollable modes, proposing ⎢ u2 ⎥ ⎢ f3 − f1 ⎥
an H ∞ controller and simulating it in two validation experiments. u=⎢ ⎥ ⎢
⎣ u3 ⎦ = ⎣
⎥,
⎦ (3)
f4 − f2
The aforementioned properties make CVPP quadcopters an attrac-
u4 −τ1 + τ2 − τ3 + τ4
tive platform and may stimulate further research and application
in the future. where f 1 , f 2 , f 3 , f 4 and τ1 , τ2 , τ3 , τ4 are the thrust forces and
torques generated by the four rotors, respectively. The element-
2. Problem setup wise form of the model in (1) and (2) could be written in a
matrix-vector form as ẋ = f(x, u), where f represents the right-
Consider a CVPP quadcopter with a “plus” configuration where hand side of (1) and (2).
propellers 1 and 3 rotate clockwise and propellers 2 and 4 ro- The key difference between VPP and conventional fixed pitch
tate counter-clockwise as shown in Fig. 1. There are two coordi- propeller quadcopters is the model of their thrust forces and
nate frames: a global inertial frame and a body frame. The ori- torques. In particular, the force and torque generated by a VPP are,
gin of the body frame expressed in the global frame is [x, y , z] T . respectively,
The yaw-pitch-roll angles [φ, θ, ψ ] T represent the rotational trans-
formation from the body frame to the global frame. The an- f i = k1 ωi2 αi ,
(4)
gular rate of the quadcopter expressed in the body frame is τi = k2 ωi2 + k3 ωi2 αi2 + k4 ωi αi ,
[p , q, r] T . The linear speed of the quadcopter in the global frame
is [u , v , w] T . The whole state vector of the quadcopter is x = where ωi is the spinning rate of propeller i, αi is the pitch angle of
T
x, y , z, φ, θ, ψ, p , q, r , u , v , w . the propeller, and k1 , k2 , k3 and k4 are drag coefficients [23]. Fig. 2
The translational and rotational air resistance F D and τ D , that illustrates the structure of a VPP.
oppose the motion of an aircraft, are given by For a CVPP quadcopter the spinning rates of the four propellers
are identical. That is, it has ωi = ω0 for all i. To simplify (4),
⎡ ⎤ ⎡ ⎤
u |u | p| p| the drag coefficients are combined with the fixed spinning speed
F D = −kβ ⎣ v | v | ⎦ , τD = −kγ ⎣ q|q| ⎦ , which gives b1 = k1 ω02 , b2 = k2 ω02 , b3 = k3 ω02 , and b4 = k4 ω0 ,
w|w| r |r | where (4) becomes
2
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
Consider the scenario where the pitch angle of propeller 1 be- Table 1
comes stuck at a constant value c. Substituting α1 = c and (5) to CVPP quadcopter parameters.
0 = b3 (−c 2 + α22 − α32 + α42 )+ Along the equilibrium trajectory, the quadcopter would rotate
b4 (−c + α2 − α3 + α4 ). at a constant rate of r ∗ as given in (12). It is of great interest to
see the specific values of r ∗ , especially whether r ∗ could be zero.
This is a system with 4 equations and 3 unknowns, which cannot Substituting b1 = k1 ω02 , b3 = k3 ω02 , and b4 = k4 ω0 into (11)
be solved. Only 3 constraint equations can be satisfied in the sys- gives
tem, and one cannot. Compared with the loss of attitude, pitch and
roll control which is likely to cause the quadcopter to crash, the k 3 m2 g 2 k4mg 2mgk3
loss of yaw control seems acceptable. Therefore, we have u ∗1 = mg,
u ∗4 = + − + 4k4 ω0 c , (13)
2k21 ω02 k1 ω0 k1
u ∗2 = 0, and u ∗3 = 0, substituting which into (6)-(8) gives
substituting which into (12) yields
mg
α2∗ = α4∗ = − c, α3∗ = c . (10)
2b1
k3m2 g 2 k4mg 2mgk3
∗ l
Furthermore, substituting (10) into (9) yields r = 2 2
+ − + 4k4 ω0 c . (14)
2k1 ω0 k1 ω0 k1 kγ
u ∗4 = 2b3 α42 − c 2 + 2b4 (α4 − c ) . (11)
The parameters could be selected according to [23], as shown in
Moreover, substituting ṙ ∗ = 0 to (2) gives Table 1.
Although it is assumed that ω0 is fixed in each control process,
|u ∗4 |l we would like to analyse the impact of different values of ω0 on
r ∗ = sgn(u ∗4 ) , (12) r ∗ . First, suppose that the physically achievable interval of ω0 is
kγ
[0, ωmax ]. Second, since ω0 must satisfy (10), it follows that
where sgn(·) denotes the sign function. When α2∗ = α4∗ = c, (11)
gives u ∗4 = 0 and r ∗ = 0 which indicates that the CVPP quadcopter mg
would not be able to remain at any static equilibrium point but ω0 = . (15)
2k1 (α2∗ + c )
can follow a state trajectory with constant self-spinning speed.
In particular, consider a state trajectory Suppose the physically achievable pitch angle is saturated by
∗
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ T [−αsat , αsat ], where αsat = 0.314 rad following [9]. Since α2∗ > −c
x (t ) = x , y , z , φ , θ , ψ (t ), p , q , r , u , v , w
T by (10), weknow α2 ∈ (−c , αsat ]. It then follows from (15) that
= x∗ , y ∗ , z∗ , 0, 0, ψ ∗ (t ), 0, 0, r ∗ , 0, 0, 0 , ω0 ∈ 2 = mg
, +∞ . Thus, ω0 must satisfy
2k1 (αsat +c )
3
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
mg mg
c= ≤ = αsat , (21)
4k1 ω02 2
4k1 ωmin
mg mg
c= ≥ := c † ≈ 0.05. (22)
4k1 ω02 2
4k1 ωmax
c ≥ c† ,
4
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
Substituting (24) into (23) leads to x̃˙ = T−1 ATx̃ + T−1 Bū F . Let à =
T−1 AT and B̃ = T−1 B. Then, it follows that x̃˙ = Ãx̃ + B̃ū F , whose
Fig. 4 shows a block diagram of the control scheme for a faulty
CVPP quadcopter. Based on the decomposed controllable system
sub-block matrix form is
(25), a linear state feedback controller can be designed. Consider a
x̃˙ c Ãc Ã12 x̃c B̃c simple feedback design H ∞ controller is presented to enhance the
= + ū F , (25)
x̃˙ c̄ 02×10 Ãc̄ x̃c̄ 02×3 robustness of the system. Other control methods like LQG could
also be used. According to the faulty system (25), the H ∞ con-
where troller is designed as
⎡ ⎤
03×3 03×3 03×2 03×2 x̃˙ = Ãx̃ + B̃1 w + B̃2 u,
⎢ I3×3 03×3 03×2 03×2 ⎥
Ãc = ⎢
⎣ 02×3
⎥, z∞ = C̃1 x̃ + D12 u,
H2×3 02×2 02×2 ⎦ (27)
02×3 02×3 I2×2 02×2 ỹ = C̃2 x̃,
1 0 −1 where
H2×3 = , Ã12 = 010×2 ,
0 1 0 ⎡ ⎤
⎡ ⎤ 03×3 03×3 03×2 03×2
I3×3 ⎢ I3×3 03×3 03×2 03×2 ⎥
⎢ 03×3 ⎥ Ã = ⎢
⎣ 02×3
⎥,
Ãc̄ =
0 0
, B̃c = ⎢ ⎥ H2×3 02×2 02×2 ⎦
1 0 ⎣ 02×3 ⎦ .
02×3 02×3 I2×2 02×2
02×3 ⎡ ⎤ ⎡ ⎤
I3×3 I3×3
Equation (25) shows that the system is decomposed into an un- ⎢ 03×3 ⎥ ⎢ 03×3 ⎥
controllable subsystem x̃˙ c̄ = Ãc̄ x̃c̄ and a controllable subsystem B̃1 = ⎢ ⎥
⎣ 02×3 ⎦ , B̃2 = ⎢ ⎥
⎣ 02×3 ⎦ ,
x̃˙ c = Ãc x̃c + Ã12 x̃c̄ + B̃c ū F . The transformed states are 02×3 02×3
5
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
Fig. 5. Trajectory tracking results. The blue dotted curves represent the actual flying
trajectory and the red line is the reference square trajectory with 10 m length on
each side. The faulty propeller is highlighted in red.
⎡ ⎤
03×3 Qc2 03×2 Qc4
⎢ 02×3 S 02×2 02×2 ⎥
⎢
C̃2 = ⎣ ⎥,
S 02×3 02×2 02×2 ⎦
Qc2 03×3 Qc4 03×2
I 10×10 010×3
C̃1 = , D12 = ,
03×10 I 3×3
b1 l
0 Ix
0
S= .
− bI1yl 0 b1 l
Iy
5. Simulation validation
6
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
Fig. 7. Results from physics-based simulations for 100 random initial states. The x-
axis and y-axis show the initial pitch (θ0 ) and roll (φ0 ) angles respectively. A black
circle indicates that the faulty quadcopter achieved stability, whereas the red cross
indicates that it lost stability.
References
[1] C.A. Wargo, G.C. Church, J. Glaneueski, M. Strout, Unmanned aircraft systems
(UAS) research and future analysis, in: Proceedings of 2014 IEEE Aerospace
Conference, IEEE, 2014, pp. 1–16.
[2] G. Skorobogatov, C. Barrado, E. Salamí, Multiple UAV systems: a survey, Un-
manned Syst. 8 (02) (2020) 149–169.
[3] A. Mohiuddin, T. Tarek, Y. Zweiri, D. Gan, A survey of single and multi-UAV Fig. 8. Simulation results of random initial angle scenario where the initial state
aerial manipulation, Unmanned Syst. 8 (02) (2020) 119–147. is [x, y , z, φ, θ] = [0, 0, 0, −0.7, −1.26] and the target state is [x∗ , y ∗ , z∗ , φ ∗ , θ ∗ ] =
[4] A. Devos, E. Ebeid, P. Manoonpong, Development of autonomous drones [0, 0, 0, 0, 0].
for adaptive obstacle avoidance in real world environments, in: Proceedings
of 2018 21st Euromicro Conference on Digital System Design (DSD), 2018,
pp. 707–710. [14] Z. Liu, C. Yuan, Y. Zhang, J. Luo, A learning-based fault tolerant tracking control
[5] Z. Yu, Y. Zhang, B. Jiang, C.Y. Su, J. Fu, Y. Jin, T. Chai, Decentralized fractional- of an unmanned quadrotor helicopter, J. Intell. Robot. Syst. 84 (1–4) (2016)
order backstepping fault-tolerant control of multi-UAVs against actuator faults 145–162.
and wind effects, Aerosp. Sci. Technol. 104 (2020) 105939. [15] Z. Hou, P. Lu, Z. Tu, Nonsingular terminal sliding mode control for a quadrotor
[6] S. Xu, H. Wen, Z. Huang, Robust fuzzy sampled-data attitude control of space- UAV with a total rotor failure, Aerosp. Sci. Technol. 98 (2020) 105716.
craft with actuator saturation and persistent disturbance, Aerosp. Sci. Technol. [16] Z. Wang, R. Groß, S. Zhao, Controllability analysis and controller design for
101 (2020) 105850. variable-pitch propeller quadcopters with one propeller failure, Advanced Con-
[7] S. Allison, H. Bai, B. Jayaraman, Wind estimation using quadcopter motion: a trol for Applications: Engineering and Industrial Systems (2020) e29 (Invited
machine learning approach, Aerosp. Sci. Technol. 98 (2020) 105699. special issue).
[8] N. Gupta, M. Kothari Abhishek, Flight dynamics and nonlinear control design [17] S. Sheng, C. Sun, Control and optimization of a variable-pitch quadrotor with
for variable-pitch quadrotors, in: Proceedings of 2016 American Control Con- minimum power consumption, Energies 9 (4) (2016) 232–249.
ference (ACC), IEEE, 2016, pp. 3150–3155. [18] T. Pang, K. Peng, F. Lin, B.M. Chen, Towards long-endurance flight: design and
[9] M. Cutler, J. How, Actuator constrained trajectory generation and control for implementation of a variable-pitch gasoline-engine quadrotor, in: Proceedings
variable-pitch quadrotors, in: Proceedings of AIAA Guidance, Navigation, and of 2016 12th IEEE International Conference on Control and Automation (ICCA),
Control Conference, 2012, pp. 4777–4792. IEEE, 2016, pp. 767–772.
[10] M.W. Mueller, R. D’Andrea, Relaxed hover solutions for multicopters: applica- [19] E. Gopalakrishnan, Quadcopter flight mechanics model and control algorithms,
tion to algorithmic redundancy and novel vehicles, Int. J. Robot. Res. 35 (8) Masters’ Thesis, Department of Control Engineering, Lulea University of Tech-
(2016) 873–889. nology, Czech Technical University, Prague, Czechia, 2017.
[11] A.F.S.l. Alessandro Baldini, Riccardo Felicetti, A. Monteriù, Actuator fault toler- [20] A. Letalenet, P. Morin, Identification and evaluation of a force model for multi-
ant control of variable pitch quadrotor vehicles, in: Proceedings of 2021 21st rotor uavs, in: 2020 IEEE International Conference on Robotics and Automation
IFAC World Congress, IFAC, 2020. (ICRA), IEEE, 2020, pp. 4280–4286.
[12] M. Tahavori, A. Hasan, Fault recoverability for nonlinear systems with applica- [21] G.P. Rible, A.A. Nicolette, R. Manuel, Modeling and implementation of quad-
tion to fault tolerant control of UAVs, Aerosp. Sci. Technol. 107 (2020) 106282. copter autonomous flight based on alternative methods to determine propeller
[13] R.C. Avram, X. Zhang, J. Muse, Nonlinear adaptive fault-tolerant quadrotor al- parameters, Adv. Sci. Technol. Eng. Syst. J. 5 (5) (2020) 727–741.
titude and attitude tracking with multiple actuator faults, IEEE Trans. Control [22] A. Freddi, A. Lanzon, S. Longhi, A feedback linearization approach to fault tol-
Syst. Technol. 26 (2) (2018) 701–707. erance in quadrotor vehicles, IFAC Proc. Vol. 44 (1) (2011) 5413–5418.
7
Z. Wang, R. Groß and S. Zhao Aerospace Science and Technology 120 (2022) 107245
[23] M. Cutler, J.P. How, Analysis and control of a variable-pitch quadrotor for agile [25] M. Petreczky, R. Tóth, G. Mercère, Realization theory for LPV state-space rep-
flight, J. Dyn. Syst. Meas. Control 137 (10) (2015) 101002–101016. resentations with affine dependence, IEEE Trans. Autom. Control 62 (9) (2016)
[24] D. Boley, Computing the Kalman decomposition: an optimal method, IEEE 4667–4674.
Trans. Autom. Control 29 (1) (1984) 51–53.