Professional Documents
Culture Documents
H_based_model_following_method_in_autola
H_based_model_following_method_in_autola
a r t i c l e i n f o a b s t r a c t
Article history: Probably the most important part during a flight is the landing phase because most of the accidents occur
Received 23 February 2019 in this phase. Automatic landing systems (ALS) take over the control during this phase to avoid potential
Received in revised form 24 June 2019 pilot-induced risks. However, some external disturbances such as windshear can jeopardize the safe
Accepted 2 September 2019
landing. In this paper, the flare part of ALS is handled in a different way. A combination of some useful
Available online 4 September 2019
design methods is brought together to improve the performance of the conventional ALS even under
Keywords: severe weather conditions. Model following method is combined with the H ∞ synthesis method to find
Aircraft landing out the optimal solution for a given cost function. Resultant H ∞ optimal control problem is solved using
H ∞ synthesis Linear Matrix Inequalities (LMIs) and then a dynamic controller is constructed. On the other hand, the
Model following overall system is formed into P-K configuration, thus the system can be reconfigured easily when there
Flare exists a change in the system such as addition or removal of disturbance, noise and so on. We achieved
Windshear significant performance on the system without any disturbance. In addition to that, the robustness takes
Linear Matrix Inequality (LMI)
an important role for the flight systems and needs to be handled correctly. Therefore, two kinds of
windshear are taken care of and their effects minimized in a way that the tracking performance remains
unaffected. Thus, highly considerable results are obtained using the proposed method even under severe
weather conditions.
2019 Elsevier Masson SAS. All rights reserved.
1. Introduction transmitters. One of them is for glide-slope and the other one is for
localizer. Airplane aligns and localizes itself to a suitable altitude
Most of the aircraft accidents occur during the landing maneu- level and approach angle owing to these transmitters and starts
ver due to the pilotage errors, technical issues or severe weather decreasing its altitude [3].
conditions. To increase safety and reliability of a flight, researchers Classical control methods are generally preferred in Automatic
have developed Automatic Landing System (ALS) which was firstly Landing Systems because of their simplicity and reliability. How-
developed for commercial aircraft by BAE in 1965 [1]. It has be- ever, these methods mostly do not satisfy the desired robustness
come popular since then and is used in most of the aircraft today. and performance requirements. In recent years, difficulties in the
ALS not only provides comfortable, smooth and safe landing but it implementation of modern control systems have been reduced in-
also avoids tire and landing gear wear. It is suitable for use in rainy versely to the significant developments in the hardware of digi-
and foggy weather conditions but deactivated in the strong winds, tal flight computers [4]. Therefore, the studies related to modern
thus the pilot handles the landing procedure manually. control systems in flight systems made this area mature. Many
The aircraft must get through five different phases during the methodologies from fuzzy-logic [5] to vision-based navigation con-
landing maneuver [2]: alignment, glide-slope, flare, touchdown and trol algorithms [6] have been applied by researchers to achieve
taxiing. In this paper, we focus mainly on the flare phase which is desired, smoother and safe landing. In [7], Stevens and Lewis
the most challenging one and slightly on the glide-slope phase. used the model following method to track the reference model
Automatic Landing System works in harmony with Instrumental as described in equation (2) for the flare phase of the landing
Landing Systems (ILS) which is located in the airport. The aircraft system. LQR is used as a controller type. To increase the adap-
is guided by ILS to prepare for landing. ILS consists of two beam tation of the flight controller to different environments, a fuzzy
neural network structure was used as a controller in [8]. On the
other hand, Edward and Chang [5] worked on unsteady aerody-
namic effects in landing operation to show the capability of the
*Corresponding author.
fuzzy-logic dynamic inversion technique and obtained satisfactory
E-mail addresses: f6715010@std.yildiz.edu.tr (K. Tamkaya), lucun@yildiz.edu.tr
(L. Ucun), ustoglui@itu.edu.tr (I. Ustoglu). results. Another approach to automatic landing systems is sliding
https://doi.org/10.1016/j.ast.2019.105379
1270-9638/ 2019 Elsevier Masson SAS. All rights reserved.
2 K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379
Table 1
Medium-Size Transport Aircraft Parameters.
Parameter Value Unit Description
W 162,000 lb Mass
S 2,170 ft2 Wing Reference Area
b 140 ft Wing Span
l 90 ft Length
c̄ 17.5 ft Mean Aerodynamic Chord
ρ 0.00232 slug/ft3 Air Density
I yy 41 × 106 slug ft2 Pitch Axis Inertia
g 32.17 ft/s2 Gravitational Acceleration
CDO 0.08
CLO 1
C Lα 4.87 1/rad Lift-Curve Slope
CMO −0.2 Fig. 2. Vortex-Ring Model.
C Mα −1.26 1/rad Pitch Stiffness Derivative
C M δe −0.916 1/rad Elevator Control Power
C Mq −16 1/rad Pitch Damping Derivative
to one. δe represents the elevator deflection angle changing be-
C M α̇ −6 1/rad Acceleration Derivative for Moment tween the limits of −25◦ to 25◦ . We will also need an acceleration
C L α̇ 0 1/rad Acceleration Derivative for Lift equation at z-axis to interpret the situation when the tire of the
A0 60 × 106 lb Static Thrust from Two Engines
aircraft touches the ground. Body acceleration of the aircraft cen-
A1 −38 lb/(ft/s) Thrust for each ft/s
K 0.042 Constant ter of gravity (cg) along the z-axis,
a z = V γ̇a = V (q − α̇ ) (7)
In (4), total aerodynamic coefficients can be described as fol-
lows:
2.3. Windshear model
C L = C L O + C Lα α
Windshear is a rapid variation in wind speed and/or direction
CD = CDO + K CL2 (5)
over a short period of time or distance. If the diameter of the
C M = C M O + C M α α + C M δe δe + C Mq q windshear is less than 4 km, it is called microburst, otherwise
macroburst. Microbursts may last a few minutes but their effects
In this paper, a medium-size transport aircraft and correspond-
such as extreme speed and variation can be very dangerous for the
ing parameters and coefficients [7] as given in Table 1 are used for aircraft.
the analysis and simulations. A piece of code is written to trim and Researchers have developed many different models to simulate
linearize the nonlinear equations in (3) according the described the microburst. One approach is the vortex-ring model which is
values in Table 1. These equations are trimmed for the conditions firstly presented by Woodfield and Wood [15]. The vortex-ring in-
of V T = 250 ft/s, h = 750 ft and γa = −2.5◦ , where V T is the trim duces the velocity wind and two symmetrical of them are used
velocity and h is the aircraft altitude. Then, they are linearized by to represent the microburst to satisfy boundary conditions. One is
using the trim values obtained. Linearized state equations can be above the ground and called the primary vortex-ring. The other
written as follows [7]: one is below the ground with the same strength and called imagi-
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤ nary vortex-ring. The center of the vortex-ring is called the vortex
V̇ V filament. Vortex-ring model is illustrated in Fig. 2. Multiple pairs
⎢ α̇ ⎥ ⎢ ⎥ ⎢ α ⎥ ⎢ ⎥ of vortex-rings are preferred for accuracy in practice. Simplified
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ θ̇ ⎥ ⎢ ⎥ ⎢ θ ⎥ ⎢ ⎥ δt vortex-ring downburst model [16] is used for the simulations in
⎢ ⎥ = ⎢A⎥ ⎢ ⎥ + ⎢B⎥
⎢ q̇ ⎥ ⎢ ⎥ ⎢ q ⎥ ⎢ ⎥ δe (6)
⎢ ⎥ ⎣ ⎦⎣ ⎦ ⎣ ⎦ this paper. R , r C and Ŵ represent the radius of the vortex-ring,
ḣ
⎣ ⎦ h the radius of the finite core and vortex-ring model circulation,
ẋ x respectively. X , Y and H are the coordinates of the primary ring
⎡ ⎤ center.
−38.5 18984 −32139 0 .1325 0
We can take Y = 0 because the lateral movement is ignored
⎢ −1.02 −632.53 5.61 1000 .00376 0 ⎥
⎢ 0 0 0 1000 0 0 ⎥ in our study. If x and h horizontal and vertical point of interest
−3 ⎢
A = 10 ⎥ respectively, then the induced velocities are computed as follows
⎢ .0785 −759.05 .793 −518.3 −.00031 0 ⎥
⎣ ⎦ [17],
−43.6 −249760 249760 0 0 0
999 −10905 10905 0 0 0
⎡ ⎤ x1 = x − X − R x2 = x − X + R
10.1 0
⎢ 15.445 × 10−5 0 ⎥ hp = h − H hm = h + H
⎢ ⎥
⎢
B =⎢
0 0 ⎥
⎥ r1p = x21 + h2p r2p = x22 + h2p (8)
⎢ 0.024656 −0.01077 ⎥
⎢ ⎥ r1m = x21 2
+ hm r2m = x22 2
+ hm
⎣ 0 0⎦
−(r0 /rc2 )
0 0 r0 = min{r1p , r2p } ζ =1−e
In equation (3), there were four states, however, the states can r xp = (x − X )2 + h2p + R 2
be expanded to include α – AoA, h – altitude, x – horizontal dis-
tance as shown in (6). r xm = 2
(x − X )2 + hm + R2
3/4
α̇ = θ̇ − γ̇a ḣ = V sin γa ẋ = V cos γa
rhp = (x − X )2 /4 + h2p + R 2
We will handle two control inputs which are δt and δe . δt rep- 3/4
resents the throttle position changing between the limits of zero rhm = (x − X )2 /4 + hm
2
+ R2
4 K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379
Table 2
Downburst Parameters.
Parameter Moderate Severe Unit
Ŵ1 200,000 400,000 ft2 /s
R1 5,500 5,000 ft
H1 2,000 2000 ft
R c1 500 500 ft
Ŵ2 120,000 280,000 ft2 /s
R2 4,000 3,500 ft
H2 2,500 2000 ft
R c2 500 300 ft
ẋ(t ) = Ax(t ) + B 1 w (t ) + B 2 u (t )
z(t ) = C 1 x(t ) + D 11 w (t ) + D 12 u (t ) (14)
Fig. 4. Generalized P-K Form.
y (t ) = C 2 x(t ) + D 21 w (t ) + D 22 u (t )
If we take D 22 = 0, then the system dynamics becomes, Now a controller can be constructed using,
D K = ( I + D K 2 D 22 )−1 D K 2
⎡ ⎤
A B1 B2
P (s) = ⎣ C 1 D 11 D 12 ⎦ (15) B K = B K 2 ( I − D 22 D K )
C 2 D 21 0 (25)
C K = ( I − D K D 22 )C K 2
The dynamic controller is represented as follows:
A K = A K 2 − B K ( I − D 22 D K )−1 D 22 C K
ẋk (t ) = A k xk (t ) + B k y (t ) where
(16)
u (t ) = C k xk (t ) + D k y (t ) −1 −1
AK2 BK2 X2 X1 B 2 An Bn X 1 AY 1 0 Y 2T 0
CK2 DK2
= 0 I Cn Dn
− 0 0 C2 Y1 I
Ak Bk
K = K (s) = (17) (26)
Ck Dk
If u (t ) in (16) is substituted into (14) and y (t ) in (14) is sub- and where X 2 and Y 2 are any matrices which satisfy
stituted into (16), then the closed loop system can be written as
follows: X2 Y 2 = I − X1 Y 1
⎡
ẋ(t )
⎤ ⎡
A + B 2 Dk C2 B 2 Ck B 1 + B 2 D k D 21
⎤⎡
x(t )
⎤ To construct a dynamic controller K as shown in (25), LMIs in
⎣ ẋk (t ) ⎦ = ⎣ Bk C2 Ak B k D 21 ⎦ ⎣ xk (t ) ⎦ (23) and (24) have to be solved. In this paper, LMIs are solved by
z(t ) C 1 + D 12 D k C 2 D12C k D 11 + D 12 D k D 21 w (t ) (18) using YALMIP and SEDUMI which is a semidefinite programming
solver. γ is initialized to a value and LMIs are solved. γ is iterated
S(P,K)
downward until an optimal solution is found. Once this iteration
A cl B cl process is done, we have everything to construct the controller as
S(P , K ) = (19)
C cl D cl described in (25).
Performance criterion which is desired to be minimized can be 4. Obtaining the generalized plant P
taken as H ∞ norm of the closed loop transfer function. Thus the
cost function takes the following form [26,27], All the models and methods used in this paper have been de-
fined so far. Simulations are conducted for two different cases us-
J ∞ ( K ) = S ( P , K )∞ (20) ing these models and methods. In the first case, a simple design
without any exogenous inputs other than a reference is handled.
Direct minimization of J ∞ ( K ) cost function is a very hard prob-
The simulation results are compared with the design having the
lem. Instead, it is easier to investigate if there is a stabilizing con-
same parameters but utilizing a different method, described in [7].
troller that provides the H ∞ norm constraints,
In the second case, the windshear effect is taken into consideration
as a disturbance in order to demonstrate the disturbance rejection
J ∞(K ) < γ , γ >0 (21)
property and the ease of adaptation of this design methodology for
These constraints can be used to check if the inequalities in new additional inputs and outputs.
(21) are assured for different γ values. Case 1:
The purpose here is to find the controller K that solves the H ∞ The basic configuration of the system is shown in Fig. 5a. In
optimal control problem for a minimum γ using Linear Matrix In- the system, “Ideal Model” and “Aircraft Model” are defined in sec-
tion 2.1 and 2.2, respectively. There is only reference input r as an
equalities (LMIs) without losing convexity. In other words, to solve
exogenous input. The main purpose here is to ensure that the alti-
H ∞ optimal control problem, we solve [28]
tude of the aircraft follows the ideal model closely. The exogenous
outputs, z1 and z2 , are used to achieve this constraint. In this case,
min γ (22)
γ , X 1 , Y 1 , A n , B n ,C n , D n z1 is the altitude error signal to be minimized between the air-
craft altitude and the desired altitude and z2 is used to decrease
such that the amplitude of the control signal to an achievable level in terms
X1 I of the elevator deflection angle. For instance, the elevator deflec-
≥0 (23) tion angle has to be kept within its limits which are ±25◦ in this
I Y1
⎡ ⎤ case and W u is the weighting function used for this purpose. The
AY 1 + Y 1 A T +
T T ∗T ∗T ∗T ⎥ best W u can be found after a couple of simulation experiments
B 2 CT n + C n B 2
⎢ and selected as a constant number ( W u = 6) for this simulation
⎢ ⎥
⎢ A + An + X1 A + A T X1 + ⎥
∗T ∗T ⎥ scenario. M (s) and L (s) are dynamic controllers. Elevator-Actuator
B n C 2 + C 2T B nT
⎢
⎢ [ B 2 D n C 2 ]T ⎥ is modeled as a first-order transfer function with a time constant
[ B 1 + B 2 D n D 21 ]T [ X B 1 + B n D 21 ]T ∗T
⎢ −γ I
⎥
⎢ ⎥ of 0.1 s.
⎣ D 11 + ⎦
The generalized plant P is obtained from Fig. 5a. If M (s) and
C 1 Y 1 + D 12 C n C 1 + D 12 D n C 2
D 12 D n D 21
−γ I
L (s) are kept outside as in Fig. 4, the generalized plant P structure
<0 (24) becomes as in Fig. 5b.
6 K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379
⎡ ⎤
0.00 0.00 0.56 −0.02 −0.02 the same altitude. P-K configuration was shown in Fig. 6. K is the
⎢ 0.00 0.00 −0.13 0.01 0.01 ⎥ sixth-order controller obtained in Section 4 – Case 1. P is the block
⎢ ⎥
⎢ 0.30 −0.07 238.37 −10.33 −9.66 ⎥ diagram already shown in Fig. 5b. There is only one exogenous
⎢ ⎥
⎢ −0.45 0.10 −363.57 15.76 14.74 ⎥ input r which is a steady state altitude level. In this case, it has to
B K = 105 × ⎢ ⎥
⎢ −0.08 0.02 −67.40 2.92 2.73 ⎥ be zero since the landing phase is considered.
⎢ ⎥
⎢ 0.00 0.00 −0.90 0.04 0.04 ⎥ In Fig. 9a, “Actual Path” and “Reference Path” are compared.
⎣ 0.00 0.00 −0.20 0.01 0.01 ⎦ For this study, we also compare our results with the study in [7],
0.00 0.00 0.63 −0.03 −0.03 where LQR based controller design method is used. It is clearly
seen that when compared with the LQR method in [7], the actual
3.16 −3.62 −0.02 −0.02 0.18 2.92 −0.45 16.18
CK = path follows the reference path more closely and smoothly using
0.02 −0.01 −0.01 −0.01 0.06 0.01 −0.01 2.43
H ∞ method. Based on this result, it is also expected that the flight
−1.0002 0.00 0.0003 0.00 0.00 path angle and elevator deflection angle have to be smoother. The
DK =
−0.0001 −1.00 −0.0004 0.00 0.00 flight path angle γa and elevator deflection angle δe are shown in
Fig. 9b. γa starts with −2.5◦ initial condition which is the flight
5. Simulation results path angle of the glide slope phase and goes to zero as the aircraft
approaches the touchdown point. During landing, elevator deflec-
Case 1: tion angle stays within the limits. As seen from Fig. 9b, when com-
It is assumed that the glide-slope phase is just ended at the pared with [7], the transient behaviors of γa and δe are smoother
altitude level of 36.37 ft and the flare phase has just started at as expected.
K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379 9
Case 2: by manipulating the elevator and throttle controls. Even if the al-
To examine the effect of the windshear, it is assumed that the titude and angle of attack decrease as a first reaction due to the
aircraft is still on the glide-slope phase and its altitude level is intervention of the controller, the altitude gets back on the desired
800 ft. The flare phase starts at the altitude level of 36.37 ft. The level with a little increase in the airspeed and the angle of attack.
desired touch down point beyond the glide slope transmitter point After ten seconds, the headwind loses its effect and the down-
(D) is 2, 500 ft. In ideal conditions, the distance the aircraft should draft starts being effective. The angle of attack decreases to keep
take until the glide slope transmitter point is calculated as x = the altitude of the aircraft on the track as expected. The aircraft
800 ÷ tan 2.5◦ = 18, 323 ft. Total distance the aircraft should take starts encountering an increasing tailwind after about thirty-three
is x = 18, 323 + 2, 500 = 20, 823 ft. seconds. As the tailwind gets close to the maximum level, the an-
A comparison of the altitude behaviors between the ideal gle of attack start decreasing while the airspeed is increasing, but
model and the actual aircraft in a moderate downburst environ- the controller takes the control again. Then, the airspeed decreases
ment is shown in Fig. 11a. The aircraft encounters an increasing and the angle of attack gets increased by manipulating the elevator
headwind through the first ten seconds as shown in Fig. 10a. This and throttle controls. Thus, a small increase in the altitude occurs.
causes an increase in the airspeed of the aircraft in consequence The flare phase starts when the tailwind starts losing its effect. As
of the horizontal component of the windshear, an increase in the seen in Fig. 11a, the aircraft continues to follow the ideal model
angle of attack in consequence of the vertical component of the very closely during both the glide-slope and flare phases, and lands
windshear and therefore an increase in the gamma. As a result, very close to the calculated touchdown point smoothly as shown
the altitude of the aircraft starts increasing. On the other hand, in Fig. 11c. Fig. 11b visualizes the altitude error between actual
the controller starts compensating the altitude and velocity errors and reference model. The rest of the aircraft behaviors mentioned
10 K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379
above in the moderate downburst environment are illustrated in the aircraft in the desired path by controlling variables V , θ and
Fig. 11d, 11e, 11f, 11g. Note that the velocity deviation from the hence γa by taking the windshear components into account. This
trim velocity of 250 ft/s is plotted in Fig. 11d. It is worth not- dependence can be seen in Fig. 11e. Fig. 11f and 11g depict, corre-
ing that, during the glide-slope phase of the landing, contrary to spondingly, the elevator and throttle actuator displacements, which
the conventional landing systems, it is not necessary to keep the remain within the limits, as the aircraft follows the reference land-
flight path angle (γa ) constant. A reference model was defined ing trajectory. Body acceleration at the Aircraft’s C G (along z-axis)
for both the glide-slope and the flare phases, and it is expected is shown in 11h. As seen from the plot, the acceleration on z-axis
from the controller that it enables the aircraft to follow (in lon- is very close to zero near the touchdown point. This means that
gitudinal direction) the reference model by utilizing throttle and the aircraft’s tires smoothly touch the ground.
elevator control surfaces. Dynamic equations including the winds- P-K configuration is also tested for a severe downburst envi-
hear components for the aircraft were introduced in (10). As seen ronment. Similar attitudes are obtained but the magnitudes are
from the equation, the altitude depends on V , γa and W h . W h greater as shown in Fig. 12a, 12b, 12c, 12d, 12e, 12f, 12g, 12h. The
is a disturbance input as the windshear component, and cannot altitude returns to the reference path after the maximum head-
be controlled. However, γa and V can be used by the controller wind but it goes little down and stays there through downdraft.
to compensate for the windshear effects and to guide the aircraft On the other hand, although the tailwind is being slightly effec-
on the desired flight path. Providing that the aircraft’s velocity is tive during the flare phase, the aircraft follows the ideal model
constant, the AoA (α ) depends on the wind (i.e., turbulence level) very closely. The elevator deflection angle and the throttle position
if the atmosphere is not still. As seen in equation (10), γa also don’t exceed the limits.
depends on the windshear components, V and θ (γa = θ − α ). The proposed method is also compared with the study in [7]
Thus, the controller dynamically makes fine adjustments to keep for the windshear effect. For this comparison, the system in [7] is
K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379 11
modified to have the windshear components. It is assumed that trol methods on the flight systems, this approach exploits the com-
the aircraft is under the severe downburst effect as shown in bination of the model following method, the general P-K configura-
Fig. 10b. It is also added a feedback loop for the throttle control to tion, H ∞ synthesis, LMI solution techniques and finally windshear
make this system as similar as possible to ours. In Fig. 13a, “Actual effects as the disturbance. A reference model for the flare phase
Path” and “Reference Path” are compared. At first glance, “Actual was predefined and the model following method was used to force
Path” always fluctuates around the “Reference Path”. It seems that the altitude state of the aircraft to behave like the state of the ref-
the controller is having a hard time to keep the state around the erence model. P matrix was created for the overall system to form
“Reference Path”. The divergence between “Actual Path” and “Ref- general P-K control configuration. Therefore, this problem could be
erence Path” sometimes gets higher values when compared with transformed into H ∞ optimal control problem. Using the LMI in-
Fig. 12a. In the flare phase, “Actual Path” doesn’t follow the “Ref- equalities and optimization to minimize γ , a dynamic controller
erence Path” and the aircraft has hard landing on the ground. This was constructed for the optimal solution. Finally, various scenarios
landing may be called a crush. In Fig. 13b, it is seen that the air- were simulated and studied. Thus, the simulation results demon-
craft lands little before the calculated touchdown point. As a result, strate the correctness and effectiveness of the proposed method
it is obvious that the proposed method provides more stabilized, by comparing the results with the LQR technique. Then, the wind-
more accurate and smoother landing. shear effect was taken into consideration as a disturbance to test
the limitations of the proposed technique. The simulations were
6. Conclusion conducted to test the disturbance rejection property of the system
for moderate and severe downbursts. The simulation results also
This paper brought a different approach to the landing control demonstrate us the effectiveness of the proposed method even un-
systems of a passenger aircraft. Differently from the existing con- der the severe downburst.
12 K. Tamkaya et al. / Aerospace Science and Technology 94 (2019) 105379
References
[1] D. Siegel, R.J. Hansman, Development of an Autoland System for General Avia-
tion Aircraft, Tech. Rep., 2011.
[2] P. Serra, R. Cunha, T. Hamel, C. Silvestre, F. Le Bras, Nonlinear image-based vi-
sual servo controller for the flare maneuver of fixed-wing aircraft using optical
flow, IEEE Trans. Control Syst. Technol. 23 (2) (2015) 570–583.
[3] W.T. Miller, P.J. Werbos, R.S. Sutton, Neural Networks for Control, MIT Press,
1995.
[4] Z. Zhen, S. Jiang, K. Ma, Automatic carrier landing control for unmanned aerial
vehicles based on preview control and particle filtering, Aerosp. Sci. Technol.
81 (2018) 99–107.
[5] C.E. Lan, R.C. Chang, Unsteady aerodynamic effects in landing operation
of transport aircraft and controllability with fuzzy-logic dynamic inversion,
Aerosp. Sci. Technol. 78 (2018) 354–363.
[6] Y. Meng, W. Wang, H. Han, J. Ban, A visual/inertial integrated landing guidance
method for UAV landing on the ship, Aerosp. Sci. Technol. 85 (2019) 474–480.
[7] B.L. Stevens, F.L. Lewis, E.N. Johnson, Aircraft Control and Simulation: Dynam-
ics, Controls Design, and Autonomous Systems, John Wiley & Sons, 2015.
[8] J.-G. Juang, J.-Z. Chio, Fuzzy modelling control for aircraft automatic landing
system, Int. J. Syst. Sci. 36 (2) (2005) 77–87.
[9] D.V. Rao, T.H. Go, Automatic landing system design using sliding mode control,
Aerosp. Sci. Technol. 32 (1) (2014) 180–187.
[10] J.-G. Juang, S.-T. Yu, Disturbance encountered landing system design based on
sliding mode control with evolutionary computation and cerebellar model ar-
ticulation controller, Appl. Math. Model. 39 (19) (2015) 5862–5881.
[11] J.-G. Juang, H.-K. Chiou, L.-H. Chien, Analysis and comparison of aircraft landing
control using recurrent neural networks and genetic algorithms approaches,
Neurocomputing 71 (16–18) (2008) 3224–3238.
[12] S. Singh, R. Padhi, Automatic path planning and control design for autonomous
landing of UAV using dynamic inversion, in: American Control Conference,
ACC’09, IEEE, 2009, pp. 2409–2414.
[13] S.-P. Shue, R.K. Agarwal, Design of automatic landing systems using mixed h/h
control, J. Guid. Control Dyn. 22 (1) (1999) 103–114.
[14] R. Lungu, M. Lungu, Application of H 2 / H ∞ and dynamic inversion techniques
to aircraft landing control, Aerosp. Sci. Technol. 46 (2015) 146–158.
[15] A. Woodfield, J. Woods, Worldwide Experience of Wind Shear During
1981-1982, Tech. Rep., Royal Aircraft Establishment Bedford, England, 1983.
[16] Y. Zhao, Optimal Control of an Aircraft Flying Through a Downburst, PhD thesis,
Stanford University, 1989.
[17] Y. Zhao, A. Bryson, Optimal paths through downbursts, J. Guid. Control Dyn.
13 (5) (1990) 813–818.
[18] G. Alag, R. Kempel, J. Pahle, J. Bresina, F. Bartoli, Model-following control for an
oblique-wing aircraft, in: Astrodynamics Conference, 1986, p. 2244.
[19] G. Zames, Feedback and optimal sensitivity: model reference transformations,
multiplicative seminorms, and approximate inverses, IEEE Trans. Autom. Con-
trol 26 (2) (1981) 301–320.
[20] G. Zames, On the input-output stability of time-varying nonlinear feedback sys-
tems part one: conditions derived using concepts of loop gain, conicity, and
positivity, IEEE Trans. Autom. Control 11 (2) (1966) 228–238.
[21] G. Zames, On the input-output stability of time-varying nonlinear feedback
systems–part ii: conditions involving circles in the frequency plane and sec-
tor nonlinearities, IEEE Trans. Autom. Control 11 (3) (1966) 465–476.
Fig. 13. Lqr Control Method under Severe Downburst. [22] J.C. Doyle, K. Glover, P.P. Khargonekar, B.A. Francis, State-space solutions to
standard H 2 and H ∞ control problems, IEEE Trans. Autom. Control 34 (8)
(1989) 831–847.
[23] P. Gahinet, P. Apkarian, A linear matrix inequality approach to H ∞ control, Int.
In this research, it is assumed that the system under test is J. Robust Nonlinear Control 4 (4) (1994) 421–448.
known and its parameters do not change in time. But real systems [24] P. Apkarian, D. Noll, Nonsmooth H ∞ synthesis, IEEE Trans. Autom. Control
51 (1) (2006) 71–86.
usually do not work that way. Thus, the future work will be con-
[25] S. Skogestad, I. Postlethwaite, Multivariable Feedback Control: Analysis and De-
centrated on the robustness of the system by taking into account sign, vol. 2, Wiley, New York, 2007.
some forms of uncertainties and turbulence. [26] K. Zhou, J.C. Doyle, K. Glover, et al., Robust and Optimal Control, vol. 40, Pren-
tice Hall, New Jersey, 1996.
[27] K. Zhou, J.C. Doyle, Essentials of Robust Control, vol. 104, Prentice Hall, Upper
Declaration of competing interest Saddle River, NJ, 1998.
[28] C. Scherer, P. Gahinet, M. Chilali, Multiobjective output-feedback control via lmi
optimization, IEEE Trans. Autom. Control 42 (7) (1997) 896–911.
There is no competing interest.