You are on page 1of 6

2021 IEEE 4th International Conference on Electronics Technology

Model and Adaptive Control of Rotor/Wing


Compound UAV Based on Derivative-Free Adaptive
NDI
2021 IEEE 4th International Conference on Electronics Technology (ICET) | 978-1-7281-7673-4/20/$31.00 ©2021 IEEE | DOI: 10.1109/ICET51757.2021.9451096

Lin Qing Liu Juanxia


School of Automation China Aviation General Aircraft Research Institute Co.,Ltd.
Northwestern Polytechnical University Zhuhai, China
Xi'an, China ljx_xgd@163.com
linqingbh@163.com

Abstract—With the upsurge of urban air mobility, compound analysis software CAMRAD II. Later, they examined the
V/STOL vehicle enter a new boom round. There is a long stability and control of slowed-rotor compound aircraft,
research and development history for compound vehicle, but particularly at high advance ratios [7, 8]. Rigsby [9] built an
remain rare studies on the full-mode flight control. In this work, articulated rotor model based on the FLIGHTLAB software
a UAV compound with double-mode rotor and wing is and used the model to predict stability and control
introduced and typical flight profile and flight stage division of characteristics of lightly loaded autorotating rotor of a
the UAV are described. Then flight dynamics model of the UAV conceptual compound gyroplane in high advance ratio flight.
is built based on explicit blade element method. And a robust
adaptive full-mode control scheme is designed based on adaptive In the current literature, there remain rare studies on the
neural network augmented nonlinear dynamic inversion. The full-mode flight control of rotor/wing compound vehicle. This
derivative-free adaptive strategy is adopted to achieve fast work focuses on multimode flight dynamics and control of an
adaption. A control allocator is designed based on the dynamic example rotor-wing compound UAV.
allocation method to coordinate the rotor and the aerodynamic
surfaces. Simulation results illustrate superior full-mode flight The contributions of the paper are described as follows: (1)
control performance of the proposed scheme. a high-fidelity six/seven-degree-of-freedom model of the
compound UAV is built based on blade element method
Keywords—rotor/wing compound UAV, derivative-free adaptive, (BEM); (2) flight characteristics of all three modes and mode
nonlinear inversion control, dynamic control allocation transitions are analyzed in detail; (3) a robust adaptive full-
mode controller based on derivative-free adaptive neural
I. INTRODUCTION
network and control allocator is designed and verified with
Recently compound aircrafts have become a hot spot in numerical simulations. The control scheme could be adopted to
both academy and industry due to the performance limits of design flight control system of rotor/wing compound UAV or
conventional helicopter and aircraft. With novel powerplants e-VTOL.
and aerodynamic configurations, compound vehicles could
combine the vertical/short takeoff and landing (V/STOL) II. THE COMPOUND UAV
capacity of rotorcraft and the speed/altitude performance of
aircraft.
There is a long research and development history for
compound V/STOL vehicles, which produced some successful
military types such as V22 “Osprey”, AV-8B “Harrier” and the
(c)
F-35B “Lighting” [1]. Recently, with the upsurge of Urban Air (a) (b)

Mobility(UAM), compound V/STOL vehicle enter a new


boom round [2]. The World eVTOL Aircraft Directory lists
almost all known eVTOL designs. Among all the prototypes, Fig. 1. The DMRW UAV.
tiltrotor and slow rotor compound(SR/C) are two of the most
promising technologies, adopted by Joby Aviation and Jaunt The rotor/wing compound UAV studied in this work is
Air Mobility Journey respectively [3,4]. Long after the depicted in Fig. 1, which is called double mode rotor and wing
McDonnell XV-1 and the Fairey Rotodyne, Carter Aviation UAV, DMRW UAV for short. Tip-jet technology is adopted to
Technologies [5] proposed and demonstrated the slow drive the rotor in helicopter flight mode, which is mechanically
rotor/compound (SR/C) technology, which is the key source of simpler than a shaft-driven rotor by eliminating gearbox and
nowadays SRC compound vehicle. transmission of both main rotor and tail rotor with weight,
complexity and life-cycle costs significantly reduced.
In the aspects of flight dynamics of rotor/wing compound
vehicle, Floros and Johnson [6] analyzed the performance a The typical flight profile and flight stage division of the
slowed-rotor compound aircraft using the comprehensive rotor compound UAV are described in Fig.2. And flight mode are

978-1-7281-7673-4/21/$31.00 ©2021 IEEE 960

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.
described as follows:(1)The helicopter mode: The speed of this Where Vb >u, v, w@ and Ωb = > p, q, r @ are the velocity vector
T T

mode ranges from 0m/s to 40m/s. The rotor is driven by the


tip-jet engines which are power by chemical reactions of and angular velocity vector in the body frame, m is the mass of
hydrogen peroxide. The main rotor speed is designed as 600 the DMWR UAV, QM is the aerodynamic torque about the
r/min.(2)The gyroplane mode: When the airspeed reaches the rotor shaft, J M is the polar moment of inertia of the main rotor.
threshold (40m/s), the tip-jet engines are closed, the propeller
begin to work and the rotor enters autorotating regime. DMRW Fb and M are the external forces and moments vectors in the
UAV can transmit into the gyroplane mode by adjusting the body frame. More specifically, Fb =[Fx ,Fy ,Fz ]T and
angle of attack of tip-path plane D s and the collective pitch M =[L , M , N ]T are calculated as follows:
T 0 .(3)The aircraft mode: the autorotating rotor can be
unloaded by the wing with the airspeed increasing, resulting in ­ Fx Fxrotor  Fxplane  FxP  Gx ­ L M xrotor  M xplane  M xP
lowered rotor speed and dropping of the lift produced by the °
° °
rotor. ® Fy Fyrotor  Fyplane  FyP  Gy and ® M M yrotor  M yplane  M yP
° °
Fzrotor  Fzplane  FyP  Gz ¯N M z  M z  M z
rotor plane P
Cruise °̄ Fz

Here rotor , P stand for the main rotor and propeller, plane
stands for the fixed-wing aerodynamic parts including
Cl
im

fuselage, vertical fin, horizontal fin and rudder. Gi , i x, y, z


b
nt
sce

Driven-Autorotate represent the gravity force components in the body frame.


De

Transition

lerati
on Based on the output of the autorotating rotor model, rotor
Glide window Acce
Approach are calculated according to the geometrical relationship
off between the rotor hub and the center of gravity of the DMWR
ke-
Ta
Land
ing UAV. The derived results are as follows: Where E is the
Autorotate-Driven sideslip angle; x, y, z is the rotor hub coordinate in the body
Transition
Airport frame.
Fig. 2. Typical flight profile and stages of the DMRW UAV.
­ Fxrotor  H cos iM  T sin a1s  iM
°
III. MODEL OF THE DMWR UAV ° Fyrotor  H sin( E )  T sin b1s
° Fzrotor T cos a1s  iM
A. The Kinematics °
®   
Two sets of kinematics equations are used to describe the ° Mx
rotor
Fyrotor ( z )  Fzrotor y  Q sin a1s  iM
relative motion between the body frame and the north-east- ° M yrotor Fxrotor z  Fzrotor ( x)  Q sin b1s
earth (NED) frame. ° rotor
°̄ M z Fyrotor x  Fxrotor ( y )  Q cos a1s  iM cos b1s
­° X TbVb
 ®   
°̄ Ξ Sb Ωb
C. The Rotor Model Based on EBEM
where X =[ x, y, z ] and Ξ
T
[I ,T ,\ ] are the position vector
T
The double-model rotor is modeled based on the explicit
blade element method (EBEM).It is noted that the expressions
and Euler angle vector in the NED frame; Vb [u, v, w]T and are mainly based on the blade element model integration
Ωb [ p, q, r ]T are the velocity vector and angular velocity results presented in [11] and partial modified to model the
driven/autorotating rotor.
vector in the body frame; Tb and Sb are the transformation
matrices, which can be found in [10]. The double-model rotor model is an iterative calculation
process, which can be described as follows: First, the induce
B. Rigid-Body Dynamics velocity is converged(with iteration) using the following
Driven/autorotating rotor is key feature of the DMWR equation
UAV. The rotor angular velocity : should be introduced as a
new state variable. The seven-degree-of-freedom (7-DOF) ­ CT :R
rigid-body dynamics of the DMWR UAV can be expressed by ° 2P Vk ! 25m / s
the following Newton-Euler equations: °
vi = ® 2 2

° Vk
2
§ Vk2 · § T ·
­ Vb  Ωb u Vb  Fb / m   ¨ ¸ ¨ Vk d 25m / s
° 2 ¸
° 1 ¯ 2 © 2 ¹ © 2 US R ¹
 ® Ωb I M  Ωb u I ˜ Ωb  

° : Q / J M
¯ With the steady induced velocity, the thrust T , the horizon
force H and the torque Q are calculated as follows:

961

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.
­T CT US R 2 :R 2 The desired attitude dynamics is designed as follows:
°
° ª Id º ª kI
ªk 0 0 º ª Ic  I º
® H CH US R :R
2
 2
 
« » « »
°  « Td » «0 kT 0 » «« T c  T »»  
°̄ M K CM K US R :R
3 2

«\ d » «0 0 k\ »¼ «¬\ c \ »¼
¬ ¼ ¬
where thrust coefficient CT , the torque coefficient CM K and the
where Ic , T c and \ c are the commanded attitude angles; kI ,
horizontal force coefficient CH are calculated as follows:
kT and k\ are the desired bandwidth parameters; Id , T d and
Va ­
°§ 2 2 2 3 4 · § P2 · ½
° \ d are the desired attitude rates. Then we can derive the
CT 2 ®¨
 P  P ¸T 0  ¨1  ¸O¾ 
4 ª¬1  1.5P º¼ ¯
°© 3 3 2 ¹ © 2 °
¹ ¿ T
desired angular rates [ pd , qd , rd ] as follows:
V cd 1  P 2 ­
° aO ½°
° ­T0 (2  P 2 ) § P 2 ½
· ° 1
CM K V ® ¾®  ¨1  ¸O¾ ª pd º ª1 tan T sin I tan T cos I º ª Id º
8  1.5P 2 ° 3 © 2 °
¹ ¿ «q » «0 « »
° °¯
 sin I »»
¯ 4 1 ¿  « d» « cos I «Td »  
V ª cd CT CM º «¬ rd »¼ «¬0 sin I / cos T cos I / cos T »¼ «\ d »
CH « 1  3P  O
2
 »
K
¬ ¼
P¬8 V V ¼
The fast dynamics is described in detail:
As the paper length limit, the full description of rotor model
can be referred to [11]. ª pº ª f p º ª c3 0 c4 º ª M xrotor  'M xwing º
« » « » « « »
 «q» « fq »  « 0 c7 0 »» « M yrotor  'M ywing »  
IV. CONTROL LAW BASED ON ADAPTIVE NDI
¬« r ¼» ¬« f r ¼» ¬«c4 0 c9 ¼» «¬ M zrotor  'M zwing »¼
For the rotor/wing compound UAV, the aerodynamics of
the double mode rotor is fast time-varying due to the changes
where C is the aerodynamic derivatives. 'M
wing
is the surface
of the airspeed, attitude and rotor cyclic controls. Especially,
rotor
the periodic rotor aerodynamics has remarkable influence on control moments M is the rotor moments . f p , f q and f r are
the attitude tracking, resulting in cyclical fluctuation during defined by
steady tracking and nonlinear strong coupling between the
pitch and the roll channels. To tackle these problems, a ­ fp c1r  c2 p q  c3 L0  c4 N 0
derivative-free adaptive dynamic inversion controller is °°
designed in this section. And a rotor-surfaces allocator is ® fq c5 pr  c6 p 2  r 2  c7 M 0 
designed based on dynamic control allocation method to °
coordinate the rotor cyclic pitch controls and the aerodynamic °̄ f r c8 p  c2 r q  c4 L0  c9 N 0
surfaces. Define x > p, q, r @ T
, u > A , B ,G
1 1 e
, Ga , Gr @
T
,
Different from the conventional derivative-based adaptive T
f x ª¬ f p , f q , f r ¼º and
control (DBAC), derivative-free adaptive control (DFAC) [12]
relaxes the assumption of constant unknown ideal weights to
the existence of time-varying weights, such that fast and ª c3 0 c4 º ª M xrotor  'M xwing º
«0 « »
discontinuous variation in weights are allowed. Besides, g x,u « c7 0 »» « M yrotor  'M ywing » 
derivative-free weight-update law uses delayed weight law «¬c4 0 c9 »¼ «¬ M zrotor  'M zwing »¼
instead of the derivative-based integration strategy, which can
avoid the conflict between the nonadaptive integration. Equation (8) can be expressed as follows:
Therefore, derivative-free adaptive control is particularly well
suited for systems with fast time-varying or sudden-changing  x f x  g x,u  
dynamics. Considering the model uncertainty, the fast dynamics
A. Adaptive Dynamic Inversion Control should be expressed
Adaptive nonlinear dynamic inversion(NDI) has great  x fˆ x  gˆ x,u  Δ1   
potential to design full-envelope control law without gain-
scheduling; the adaptive neural network is designed to where f̂ x , ĝ x,u is the nominal model; Δ1 is the
compensate errors occurring in the inversion process,
improving the performance and robustness of the controller. In modelling error due to unmodeled dynamics and parameter
this section, an adaptive control scheme is developed for full- uncertainty and defined by
mode control law of the DMRW UAV based on neural net-
work augmented nonlinear dynamic inversion and dynamic  Δ1 f x  fˆ x  g x,u  gˆ x,u    
control allocation. A derivative-free update law is designed to
At equilibrium points, the model can be linearized and
compensate fast time-varying uncertainty and a dynamic
simplified as
control allocator is design to coordinate the rotor cyclic
controls and the aerodynamic surfaces.

962

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.
 x Ax  BΔu  Δ     uad [uadp uadq uadr ]T
where Δ Δ1  Δ2 is the lumped model error between the T
ª 256 256 256 º
linear design model (12) and the nonlinear model (9), which is = «¦ w ˆ j1 E j C x ¦ wˆ j 2 E j C x ¦ ˆ j3E j C x »
w
composed of modelling error Δ1 and the linearization error Δ2 . ¬j 1 j 1 j 1 ¼
Wˆ T t E [ X
The linear dynamic inversion controller is designed as
  
 BΔu  Ax  ud  uad     where Wˆ  R 256u3 is the interconnection weight matrices;
E [ X  Wˆ 256u1 is the basis function vector built as
where BΔu is the pseudo control for control allocation; uad is
the adaptive output of the neural network, which is designed in E x t ,u t kron C1 , C2 , C3 , C4 . To form the basis
the next section; ud is the output of linear Proportion- function vector, the input data is preprocessed and normalized
Integration controller as follows: between -1 and 1 with [ ( x) (1  e0.1x ) / (1  e0.1x ) .

 ud xc  K I ³ xc  x dt K P xc  x     To tackle the fast time-variant uncertainties, the weight-


update law of the neural network is designed based on
where K I diag K Ip , K Iq , K Ir and K P diag K Pp , K Pq , K Pr derivative-free adaptive control as follows:
are chosen according to the desired response performance;  Wˆ t D1Wˆ t  W  D 2 E x t eT PBe   
xc [ pc , qc , rc ]T and xc [ pc , qc , rc ]T are desired angular rate
where Wˆ t is an estimate of W t , D1  1 , W ! 0 and
vector and derivative of desired angular rate vector respectively,
which are ggenerated by a second order command-shaping filter. D2 ! 0 , P P T is the positive-define solution of the
Define x ³ xc  x dt , then the error dynamics of the fast Lyapunov equation Ae P  PAe
T
Q where Q QT ! 0 .
dynamics can be expressed as
Theorem 1. Consider the feedback control law given by
 x  KP x  KI x uad  '     (13) and (14), and the adaptive component given by (17), the
T
closed system errors e and W t are uniformly ultimately
Define e ª xT , xT º , where
¬ ¼ bounded, which can be derived in a similar way as in [12] with
>03u3 I3u3˗ K I  K P @ >03u3 I3u3 @ a Lyapunov-Krasovskii function as follows:
T T
Ae ˈBe ,
 W T G W G dG   
t
³
T
uad  Δ ª¬uad uadr  'r º¼ and (15) can be eT Pe  ltr
dp  ' p uadq  'q V
adp t W

rewritten as C. Rotor-Surfaces Dynamic Control Allocation


 e Ae e  Be uad  Δ    To coordinate the rotor and the aerodynamic surfaces, a
control allocator is designed based on the dynamic allocation
B. Derivative-Free Adaptvie Neural Network method [13]. This method can make different actuators operate
ª 1 º in different frequency domain by weight matrices tuning. The
«[ u » dynamic control allocation can be expressed as a sequential
C1 « » quadratic-programming problem if no actuators are saturated
«[ v »
«[ w »
¬
ª 1 º
¼
π u t ^ 2
arg min W1 ª¬ u t  us t ¼º  W2 ª¬ u t  u t  T ¼º
u t ?:
2
`
«[ p »
 
π Ω
2
C2 « » arg min Wv Bu t  v t
«[ q » π u t du t du t
«[ r
¬
»
¼
6 ª¬uadp , uadq , uadr º¼
where v t  R k is the virtual control vector; us t  R m is the
mum mum
desired steady-state control input; W1  R , W2  R and
ª 1 º π Variable
Ŵ Wv  R k uk are square weight matrices with proper dimensions.
«[ A1 »
weights
« »
C4 «[ B1 » V. SIMULATION
Kronecker
«[ G r »¼ product
¬ Based on the adaptive robust control scheme based on
Fig. 3. Sigma-Pi neural network. derivative-free neural network nonlinear dynamic inversion
and dynamic control allocation method, an automatic
To compensate model errors Δ in (16), a Sigma-Pi neural multimode flight control system for the DMRW UAV is
network is designed, which is described in Fig. 3. The output designed and verified through simulation. Besides, the linear
of the network takes the form real-time inverse model (LRIM) algorithm is proposed to
tackle the severe time-varying characteristics of multimode
flight.

963

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.
the dynamic control allocator used larger elevator and aileron
deflection, which is beneficial to the stability of the rotor and
the vehicle. The output of the Sigma-Pi neural network
compensator is given in Fig. 7. It can be seen that the pitch
channel neural network was adjusted more to compensate for
the large model error during the take-off, landing and the
rotation-autorotation transition stages; The roll channel neural
network was adjusted rapidly in the larger roll maneuver in
order to improve the decoupling effect of the roll and pitch
channels; The yaw channel neural network was adjusted
sharply to compensate strong external interferences during
takeoff, landing and big roll maneuver.

Fig. 4. Altitude, airspeed and deviation in the full-mode simulation.


Fig. 4 shows the height, airspeed and lateral deviation in
the full-mode flight simulation. We can see that the during the
whole flight, the airspeed and height have no obvious
fluctuation. Besides, the lateral deviation converges quickly
with excellent path tracking performance. Fig. 5 shows that the
attitude keeps stable and varies smoothly during the full-mode
flight. The pitch angle varies between -15° and 15°; the
maximum roll angle is less than 40° in cruise phase and less
than 13° in approach stage.

Fig. 6. Cyclic pitch control and surface control.

Fig. 5. Attitude tracking performance in the full-mode simulation.

Fig. 6 shows the periodic pitch controls with surface Fig. 7. Outputs of the sigma-pi neural networks.
controls (aileron and elevator) coordinated by the dynamic
control allocator. It can be seen that during the take-off, VI. CONCLUSION
landing and the rotation-autorotation transition stages the This work introduced a compound UAV with double-mode
controls had rapid adjustment to compensate for the larger rotor and wing. Then a high-fidelity six/seven-degree-of-
model error. During the stages with driven rotor or with low freedom model of the compound UAV was built. And a robust
airspeed, the periodic pitch controls played a major part adaptive full-mode controller based on derivative-free adaptive
because of low efficiency of fixed-wing surfaces. When the neural network and dynamic control allocator was designed.
rotor is unloaded and autorotated in the high airspeed stage ,

964

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.
The simulation results illustrated the superior performance of Helicopter Society 60th Annual Forum, Baltimore, MD, June 7–10,
the proposed control scheme. 2004.
[8] Floros M W, Johnson W, “Stability and Control Analysis of the
REFERENCES Slowed ̺ Rotor Compound Helicopter Configuration” Journal of the
American Helicopter Society, 2007, 52(3): 239-253.
[1] Vertipedia, “the VSTOL Wheel” https://vertipedia-
[9] Rigsby J M, “Stability and control issues associated with lightly loaded
legacy.vtol.org/vstol/wheel.htm.
rotors autorotating in high advance ratio flight” Georgia: Georgia
[2] Vertical flight society, “eVTOL Aircraft Directory” Institute of Technology, 2008.
https://evtol.news/aircraft.
[10] Stevens, Brian L., Frank L. Lewis, and Eric N. Johnson, “Aircraft
[3] Joby Aviation, “All-Electric Air Mobility” http://www.jobyaviation.com. control and simulation: dynamics, controls design, and autonomous
[4] Journey Jaunt Air Mobility LLC., “All-electric vertical take-off and systems” John Wiley & Sons, 2015.
landing aircraft” http://www.jauntairmobility.com. [11] Prouty R W, “Helicopter performance, stability, and control” Boston:
[5] Carter Aviation Technology, “the slow rotor/compound technology” PWS Engineering, 1986.
http://www.cartercopters.com. [12] Yucelen, Tansel, and Anthony J. Calise, "Derivative-free model
[6] Floros, Matthew W., and Wayne Johnson, "Performance Analysis of the reference adaptive control." Journal of Guidance, Control, and
Slowed̺Rotor Compound Helicopter Configuration." Journal of the Dynamics 34.4 (2011): 933-950.
American Helicopter Society 54.2 (2009): 22002-22002. [13] Härkegård, Ola, "Dynamic control allocation using constrained
[7] Floros, M.W., and Johnson, W, “Stability analysis of the slowed-rotor quadratic programming." Journal of Guidance, Control, and Dynamics
compound helicopter configuration” Proceedings of the American 27.6 (2004): 1028-1034

965

Authorized licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded on July 06,2021 at 02:11:02 UTC from IEEE Xplore. Restrictions apply.

You might also like