You are on page 1of 19

Chinese Journal of Aeronautics, (2019), 32(11): 2538–2556

Chinese Society of Aeronautics and Astronautics


& Beihang University
Chinese Journal of Aeronautics
cja@buaa.edu.cn
www.sciencedirect.com

High-efficiency aircraft antiskid brake control


algorithm via runway condition identification based
on an on-off valve array
Dong SUN a, Zongxia JIAO a,b, Yaoxing SHANG a,b,*, Shuai WU a,b,
Xiaochao LIU a,b

a
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China
b
Ningbo Institute of Technology, Beihang University, Ningbo 315800, China

Received 9 June 2019; revised 7 July 2019; accepted 29 July 2019


Available online 9 September 2019

KEYWORDS Abstract The aircraft antiskid braking system is an important hydraulic system for preventing tire
Aircraft; bursts and ensuring safe take-off and landing. The brake system adjusts the force applied on the
Brake; brake discs by controlling the brake pressure. Traditional aircraft antiskid braking systems achieve
Hydraulic system; antiskid performance by controlling the braking pressure with an electrohydraulic servo valve.
On-off valve array; Because the pilot stage of an electrohydraulic servo valve is easily blocked by carbonized hydraulic
Runway condition oil, the servo valve would become a dangerous weak point for aircraft safety. This paper proposes a
identification; new approach that uses an on-off valve array to replace the servo valve for pressure control. Based
Servo control on this new pressure control component, an efficient antiskid control algorithm that can utilize this
discontinuous feature is proposed. Furthermore, the algorithm has the ability to identify the runway
circumstances. To overcome the discontinuity in the process of using an on-off valve array, the
Filippov framework is introduced. The conditions of convergence of the system are also discussed.
The results of the digital simulations and the hardware-in-the-loop (HIL) braking experiments are
used to verify the efficiency and stability of the proposed control algorithm. The method also proves
that the on-off valve array can replace the servo valve perfectly as a new type of antiskid braking
pressure control component.
Ó 2019 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd. This is
an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction
* Corresponding author at: School of Automation Science and
Electrical Engineering, Beihang University, Beijing 100191, China. The aircraft antiskid brake system ensures that the tires will
E-mail address: syx@buaa.edu.cn (Y. SHANG). not wear excessively or burst while braking by reducing the
Peer review under responsibility of Editorial Committee of CJA. brake pressure when the wheel is locked, thereby ensuring
the safety of aircraft take-off and landing. An antiskid brake
system requires high adaptability to the runway environment
Production and hosting by Elsevier and high dynamic response speed. Although with the

https://doi.org/10.1016/j.cja.2019.08.020
1000-9361 Ó 2019 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
High-efficiency aircraft antiskid brake control algorithm 2539

development of more electric aircraft, the electrical drive brake have strong nonlinear and time-varying characteristics. In the
system appears, but due to technical maturity and reliability, field of traditional aircraft antiskid brake control, many effec-
the hydraulic brake system is still the mainstream in the field tive control methods have been proposed to solve this problem,
of aircraft brake.1 Hydraulic brake system obtains hydraulic and some methods have been applied in aircraft for actual use.
energy from the engine driven pump (EDP) through a long The pressure-bias-modulated (PBM) control method devel-
pipeline.2 The traditional aircraft brake system uses servo oped by Stubbs et al. has been applied in the Boeing 707 and
valves as brake pressure regulating components, as shown in some other kinds of aircraft; this method was presented as
Fig. 1, Ps is the supply pressure, Pb is the brake pressure, Pbr early as 1979.4 By setting different thresholds of wheel speed
is the reference brake pressure, PT is the reservoir pressure tracking error, the braking pressure is adjusted according to
or tank pressure. Because of the vibration and high the wheel skidding depth. In recent decades, many new control
temperatures in the braking process, the hydraulic oil can algorithms have been proposed. In 1995, neural networks and
easily carbonize, and carbonized hydraulic oil often blocks fuzzy logic were used in aircraft antiskid brake control by
the pilot stage of a servo valve.3 This blockage will lead to Tseng et al.5 Tseng’s method uses these two tools to learn run-
the loss of the pressure regulating function of the aircraft brake way friction characteristics. Robust control was introduced by
system, which substantially affects aircraft safety. Tunay and Fengyu Li et al. These previous studies6–8 com-
Compared with a traditional brake servo valve, an on-off pleted the control for the nonlinear characteristics of the
valve array has great advantages in response speed and oil pol- hydraulic part of the brake system. Li and Jiao8 used the
lution resistance. The port of an on-off valve can be fully ‘‘Lugre” friction model and parameter adaptive law to esti-
opened and closed in a few microseconds. Moreover, a com- mate runway friction characteristic parameters and calculate
parison of these two kinds of valves with the same maximum the optimal slip ratio. In 2017, Jiao et al. proposed a method
flow shows that the valve opening area of an on-off valve is to detect the runway friction coefficient by the slip factor, in
hundreds of times larger than that of the pilot stage of a servo which the calculation of the slip factor requires the installation
valve, and the reliability of an on-off valve is higher than that of a brake torque sensor to obtain a brake torque signal.9 In
of a servo valve in the case of hydraulic oil contamination. 2017, D’Avico et al. presented a method based on mixed-
Considering the difficulties of high-precision control caused slip-deceleration control, and they calculated ground friction
by large-flow on-off valves and the low-pressure response with a detailed model of landing gear for antiskid braking con-
speed caused by small flow valves, the control accuracy and trol.10 In 2018, Jia et al. proposed an active interference sup-
response speed of the system can be guaranteed by using differ- pression control based on the optimal slip ratio, which
ent flow-size on-off valves to form an array. The structure of controlled the working point of the system near the optimal
the on-off valve array is shown in Fig. 2. slip ratio and restrained the interference in the braking process
During the process of increasing pressure and decreasing to ensure the high efficiency and stability of the brake.11 These
pressure, many upstream or downstream on-off valves are methods can achieve good control effects under different run-
opened at the same time to realize pressure control. In the way conditions, but almost all of these methods need to intro-
stage of pressure maintenance, the on-off valves are closed, duce other sensor signals, such as aircraft speed or braking
and there is no flow loss. This structure can also increase the torque. However, in a traditional braking system, the wheel
redundancy and reliability of the system. speed signal and the brake pressure are the only inputs, and
The purpose of using an on-off valve array is to improve it is very difficult to add other sensors to an aircraft. In general,
the antipollution ability of the system while realizing efficient these traditional algorithms provide many ideas for the key
antiskid braking control. Therefore, the primary goal is to problems of runway identification in antiskid control.
achieve high-efficiency antiskid brake control. Furthermore, considering the discontinuity of on-off valve
The performance of the antiskid brake system is seriously controls, on-off valve antiskid braking control is seldom used
affected by the runway conditions (e.g., dry, wet or icy), which in aircraft brakes. In contrast, on-off valve antiskid braking

Fig. 1 Structure of a traditional aircraft hydraulic brake system.


2540 D. SUN et al.

Fig. 2 Structure of the on-off valve array.

control is usually used in an automobile antilock brake system modeling and analysis processes, the secondary factors that
(ABS). influence the system, such as the residual thrust of engines,
In the field of aircraft braking, Huang et al. proposed a are neglected. First, the kinematic model of the aircraft and
switching control method that achieved aircraft antiskid the main wheels of the aircraft are built, the force balance dia-
brakes using two on-off valves in 2013.12,13 They defined a gram of the wheel is shown in Fig. 3. This model considers
switching surface to determine the control strategy of the on- only the longitudinal motion of the aircraft, and the rolling
off valves. However, this method needed to determine the opti- resistance acting on the wheel is neglected. The equations of
mal slip ratio of the runway in advance. In the field of automo- the model are shown as follows:
bile brakes, a four-phase control scheme was presented by Kuo 8
>
> 2Ff  Fa ¼ M0 V_ p
and Yeh in 1992.14 By adding a high-pressure holding mode >
< F ¼ lF
and a low-pressure holding mode, the system achieved prede- f N
ð1Þ
termined deceleration in a few cycles, and the road condition >
> Fa ¼ 12 qCa SV2p
>
:
was also detected by wheel angular deceleration. In 1999, Choi Jx_ ¼ rFf  Tb
and Cho proposed an antiskid braking method using pulse-
width modulation (PWM) to control on-off valves to achieve The forces acting on the aircraft mainly include the friction
slip ratio control. This control method achieved a good control between the two main wheels and the runway Ff. In addition,
effect. However, the PWM control diminished the response unlike automotive dynamics, aerodynamic drag Fa exists in the
speed and service life of the on-off valves.15 In 2003, the effects aircraft model because of the fast speed. M0 is the mass of the
of ABS antiskid control using different sliding surfaces were aircraft, Vp is the velocity of the plane, q is the density of air,
compared by Johansen et al. A new type of sliding surface Ca is the aerodynamic drag coefficient, S is the windward area
was proposed that greatly reduced the system vibration caused of the aircraft, r is the wheel radius, x is the angular velocity of
by using on-off valves.16 A sliding mode control (SMC) con- the wheel, FN represents the support force acting on the wheel,
troller was designed for the antilock control system, and the l is the friction coefficient, Tb is the brake Torque acting on
effects of the sliding surface design were analyzed by Shim the wheel, J is the wheel inertia, and G in Fig. 3 represents
et al. in 2008.17 In 2011, Jing et al. proposed a switched control the gravity acting on the fuselage.
strategy, and the algorithm was sufficiently robust to guaran-
tee good performance in consideration of the uncertainties of
the actuator characteristics and road friction coefficient.18 In
this method, the controller needs to use an acceleration sensor
to estimate the vehicle speed as an input. These control meth-
ods in the field of automotive braking provide many good
ideas for the design of aircraft antiskid braking systems based
on on-off valve arrays.
This article is organized as follows: A dynamic model of
aircraft brake is built in part 2; The algorithm structure and
details are designed in part 3; Simulation and analysis of the
algorithm are shown in part 4; In part 5, the hardware-in-
the-loop (HIL) brake experiments are carried out, and the
results are analyzed; In part 6, the characteristics of the
method are summarized.

2. Modeling for an aircraft brake system

The process of aircraft antiskid braking is complex, and many


nonlinear factors exist within this process. To simplify the Fig. 3 Force balance diagram of the wheel.
High-efficiency aircraft antiskid brake control algorithm 2541
8
> _
< Fa ¼ DMVp A large number of runway friction tests indicate that
M ¼ M0  DM ð2Þ because there are too many factors affecting the runway fric-
>
: tion, most of these factors have nonlinear characteristics. It
2Ff ¼ MV_ p
is difficult to obtain an empirical database of runway friction
DM represents an equivalent change in aircraft mass. In the from which we can match the current runway condition.
high-speed stage of aircraft landing, engine thrust reverser or Moreover, the shape of these friction-slip ratio curves under
deceleration parachute are usually used to decelerate, and different runway conditions indicates that we can judge
brake is usually intervened in the middle-speed stage. With whether the friction is close to the maximum friction. Here,
the existence of aerodynamic drag, the aircraft would deceler- we use the slope of the curves dFf/dk as an indicator. Obvi-
ate more easily, which means that the equivalent mass M of the ously, dFf/dk > 0 suggests that the friction increases with the
aircraft will be reduced, as shown in Eq. (2). It should be noted increase of slip ratio. When dFf/dk = 0, the system achieves
that due to the aircraft aerodynamic shape design and the rel- the optimal working point, which means that the maximum
atively low brake start speed, the aerodynamic drag has a low friction is achieved. When wheel begins to slip, the friction
impact on the equivalent inertia of the aircraft, and changes decreases with increasing slip ratio, and dFf/dk < 0. Accord-
slowly compared with wheel speed. It can be considered that ing to these conclusions, we only need to monitor the system
the equivalent inertia does not change in a short period of time status until dFf/dk is close to 0, which can make the working
(e.g. within 1 s). This will facilitate our further analysis. point of the system close to the optimal point but located on
l in Eq. (1) represents the friction coefficient between the the left side of the optimal point, which ensures that the wheels
runway and the tires. The change of k affects the value of l. will not easily slip. To achieve this goal, a friction point
k is the slip ratio, which can be obtained by the following determination threshold g (g > 0) should be set. When
equation: 0 < dFf/dk < g is satisfied, the system status should be
Vp  xr maintained.
k¼ ð3Þ To obtain dFf/dk, the friction force Ff and the aircraft
Vp
velocity Vp should be calculated in advance. In the process
The relationship between the friction coefficient l and the of aircraft braking, unlike the antiskid control of automobiles,
slip ratio k can be described by the ‘‘Magic Equation” which the input to the controller is only the wheel speed and braking
is expressed as follows19–22: pressure, so the aircraft speed and friction should be observed
Ff ¼ D sinðC arctanfBk  E½Bk  arctanðBkÞgÞ ð4Þ first. Many studies have investigated vehicle and aircraft brake
control.26 It can be concluded that the friction Ff can be
where B, C, D and E are the friction characteristic description observed by some methods, such as the Kalman filter. How-
coefficients. ever, to ensure the observability of the system and the accuracy
For a hydraulic brake actuator, without considering heat of the observation results, a relatively accurate aircraft speed
transfer, temperature changes, time-varying oil properties Vp is needed as the input of the observer. A simple method
and the dynamic piston process (including oil filling process), of aircraft speed estimation is presented here.
the brake actuator model can be expressed by the following According to Eq. (2), the following expression can be
equation12: obtained:

Tb ¼ kb Pb 2Ff r ¼ MV_ p r ¼ Jx_ þ kb Pb ð7Þ
ð5Þ
P_ b ¼ Ke Q
Normally, the difference between the two terms on the
where kb is the brake disc friction coefficient and Ke represents right-hand side of the equation can be hundreds of times of
the equivalent pressure-volume coefficient of the brake actua- Jx_ or even greater. Thus, the equation above can be simplified
tor, which can be obtained from Eq. (6): as follows:
 
Vba0 A2ba MV_ p r  Pb kb ) V_ p  Pb kb =Mr ð8Þ
Ke ¼ 1= þ ð6Þ
Eh Kbd
During the initial stage after landing, there is no braking
where Vba0 is the initial volume of the brake actuator; Kbd is pressure, so the linear speed of the wheel is equal to the aircraft
the comprehensive stiffness of the brake device; Q represents speed. When the pressure begins to increase without slippage,
the flow of the system, which is affected by the brake pressure Eq. (8) can be used to calculate the aircraft speed, and a rela-
and the state of the on-off valve array; Aba is the area of the tively accurate aircraft speed can be obtained.
brake actuator; and Eh is the elastic modulus of hydraulic oil.

3. Controller design

3.1. Runway circumstance identification

From the relationship between the ground friction coefficient


and the slip ratio, we can conclude that under different runway
conditions, the maximum friction coefficient and the optimal
slip ratio of runway are different,23–25 as shown in Fig. 4, Ffmax
represents the maximum friction and kopt represents the opti-
mal slip ratio that maximizes the friction. Fig. 4 Relation between slip ratio and friction.
2542 D. SUN et al.

The Kalman filter is used to observe and estimate the run- V_ p Pbmax kb
x_ d ¼ x_ max   ð15Þ
way friction force.27–30 Usually, after doing several differential r Mr2
operations on friction Ff, it can be assumed that the result is a
Controlling the aircraft deceleration at the maximum decel-
random value. The state vector can be defined as follows:
eration can stabilize the working point near the maximum fric-
F_ f F€f
T
X ¼ ½ x Ff Vp  ¼ ½ x1 x2 x3 x4 x 5 T ð9Þ tion working point without skidding. The next step is to use
this maximum deceleration x_ d to generate the wheel reference
The state equations can be expressed as follows: speed xd and control the wheel speed.
( It should be noted that the whole optimization process
X_ ¼ fkf ðXÞ þ BU þ wðtÞ
ð10Þ depends on a stable pressurization process. To make the pres-
Y ¼ CX þ nðtÞ sure increase process as smooth as possible, the minimum flow
Eq. (10) can be divided into the following expressions: on-off valve group is used for controlling pressure increases,
8 and the appropriate error band is set to avoid the data fluctu-
>
> fkf ðXÞ ¼ ½f1 ðXÞ f2 ðXÞ f3 ðXÞ f4 ðXÞ f5 ðXÞT ation caused by frequent switching of valves, as shown in
>
>
>
> f1 ðXÞ ¼ rx2 =J Fig. 5. Because of the gap between the brake actuators and
>
>
>
> brake discs, all the valves are opened at the same time at first.
>
> f 2 ðXÞ ¼ x3
>
> When the pressure increases to a lower pressure Pbm, the large-
< f ðXÞ ¼ x
3 4
ð11Þ flow on-off valve is closed and the valve with minimum flow is
> f4 ðXÞ ¼ 0
>
>
> opened. The pressure Pbm should be lower than Pbmax under
>
>
> f5 ðXÞ ¼ ð2x2  1=2qCx Sx5 Þ=M
2
> all runway conditions.
>
>
>
> B ¼ ½ kb =J 0 0 0 0 T
>
>
: 3.2. State feedback based wheel speed control
U ¼ Pb
The measurement equation can be expressed as follows: 3.2.1. System phase plane analysis
2 3
x To use the on-off valve array to control the wheel deceleration
6 7 according to the maximum deceleration, the phase plane anal-
 6 Ff 7
1 0 0 0 0 6 _ 7 ysis of the system is carried out. The dynamic of the wheel
Y ¼ CX ¼ 6 Ff 7 þ nðtÞ ð12Þ
0 0 0 0 1 6 6 € 7
7 speed is determined by the braking pressure, and the braking
4 Ff 5 pressure is determined by the flow Q through the on-off valve
Vp array. The influence of Q on the tracking error of the wheel
speed and wheel deceleration is analyzed in Fig. 6.
where w(t) is the Gaussian white noise produced by the state The flow characteristic of an on-off valve array control is
estimating process, n(t) is the measurement of the Gaussian discontinuous. To simplify the problem, three kinds of flow
white noise. input (Q > 0, Q = 0 and Q < 0) are analyzed. It is easy to
The equations for estimating the state and the error of the determine from Fig. 6 above that to reduce the tracking error
covariance matrix are as follows: of wheel speed and deceleration, it is necessary to switch the
(
X^_ ¼ FX
^ þ BU þ K½Y  CX ^ flow Q at the appropriate time.
ð13Þ Define x = [x1,x2]T as the system state variables,
P_ ¼ FP þ PF þ Qkf  KRkf KT
T
x1 = x-xd and x2 = x_  x_ d . Fig. 7 shows that the tracking
where ^ indicates the estimated value of the variable, P is the error can be gradually reduced by switching the direction of
estimated error covariance matrix, F is the Jacobian matrix the flow with the switching surfaces f(x) = e1 and f(x) = e2.
fkf(X), Qkf and Rkf are covariance matrixes, the subscripts of Note that f(x) is the state feedback function we needed.
letters represent Kalman filter; K is the Kalman filter gain,
which is determined by Eq. (14):
K ¼ PCT R1
kf ð14Þ
The change rate of friction dFf/dt and the friction Ff can be
obtained with the Kalman filter. Using the aircraft velocity Vp,
the slip ratio can be obtained, and the differential of the slip
ratio can also be calculated. The key parameter dFf/dk can
be obtained. During the initial stage of braking, the braking
pressure is controlled by the on-off valve array, which causes
the pressure to increase. The runway condition is identified
by the condition 0 < dFf/dk < g. Once this condition is met,
the current braking pressure is recorded as Pbmax, which repre-
sents the maximum brake pressure that does not slip on the
current runway condition. According to Pbmax and the condi-
tion above, we can obtain the maximum deceleration that the
runway can provide to the aircraft. The calculation method is
as follows: Fig. 5 Control process of pressure growth with a constant slope.
High-efficiency aircraft antiskid brake control algorithm 2543

Fig. 6 Phase plane analysis of wheel speed control.

By using this state feedback for switching control of the on-


off valve array, the phase trajectory of the system can be grad- Fig. 8 Phase trajectory after introducing f(x).
ually approached to the set of equilibrium points of the system,
as shown in Fig. 8. In the set of equilibrium points, the track- the system swings back and forth in a very small range near
ing error of the wheel deceleration can be guaranteed to be 0, the optimal working point. During this process, the following
which is also the purpose of our wheel speed control. linearization process is performed.
Note that when the system reaches the set of equilibrium F_ f ¼ Kk k_ ð18Þ
points, the tracking error of the wheel speed is not zero, which
does not affect the tracking of wheel deceleration. Experiments Note that x2 can be written as follows:
show that in the actual system, as long as the tracking error of 1
the wheel speed is limited and sufficiently small, stable tracking x_ 2 ¼ ðrKk k_  kb Kc QÞ ð19Þ
J
control of wheel deceleration can be guaranteed. Hence, the
Since the slip ratio k = 1-xr/Vp, the differential of the slip
system stability can be guaranteed by choosing appropriate
ratio is as follows:
parameters e1 and e2.
V_ p xr xr
_
k_ ¼  ð20Þ
3.2.2. State feedback function design V2p Vp
The key point of the control is to design an appropriate state
If the system does not slip, the slip ratio is very small.
feedback function f(x). First, the state equation of the system is
Moreover, it can be stated that the aircraft speed is approxi-
constructed.
  mately equal to the wheel speed.
x 1 ¼ x  xd x_ 1 ¼ x2
) ð16Þ V_ p xr
_
x2 ¼ x_ 1 ¼ x_  x_ d €
x_ 2 ¼ x k_   ð21Þ
Vp Vp
For the state variable x2, according to the previous model-
The validity of this approximation has been verified by sim-
ing process, we can obtain the following result:
ulation. Fig. 10 shows that the differential values of the slip
1
€ ¼ ðrF_ f  kb P_ b Þ
x_ 2 ¼ x ð17Þ ratio before and after approximation are very close during
J the process of braking pressure variation.
In the last section, we used dFf/dk as a criterion to deter- Combining Eq. (1), Eq. (15), Eq. (16), and Eq. (21) pro-
mine whether the peak value of friction was found. With duces the following expression:
Kk = dFf/dk, Kk represents the slope of the friction curve near V_ p xr
_ V_ p x_ d r x2 r x2 r
the optimal point, as shown in Fig. 9. k_   ¼   ¼ x_ d ð22Þ
Vp Vp Vp Vp Vp Vp
After finding the optimal working point of the system, the
system controls the wheel speed. The actual working point of Bringing the above results into the system state equation
produces the following expressions:
(
x_ 1 ¼ x2
2 ð23Þ
x_ 2 ¼  1J KVkpr x2  1J kb Ke Q

Fig. 7 Phase diagram after introducing f(x). Fig. 9 Practical significance of Kk.
2544 D. SUN et al.
8 @f
>
> 0
> @x1
>
>
< @f ¼ a2
@x1
ð28Þ
>
>
@f
0  @x  q12 jx2 j @f
x2 ¼ 0; @x ¼0
>
>
>
2 2
: @f @f
0  @x 2
 q11 jx2 j x2 ¼ 0; @x 2
¼0

According to the above restrictions, it is easy to find a state


feedback function f(x) that can ensure the stability of the sys-
tem. As shown below, the feedback control law of the system
Q = U(f(x)) is designed.
1
fðxÞ ¼ a1 x1 þ jx2 jx2 ; q ¼ minðq1 ; q2 Þ ð29Þ
q
The obtained state feedback function f(x) can be directly
applied to wheel speed control using a single group of on-off
Fig. 10 Real and approximate dk/dt. valves. Note that the wheel deceleration is used in the state
feedback function. To obtain this variable, the wheel speed
must be differentiated. The tracking differentiator (TD) is used
for the differential filter32 and achieves good results.
The inertia of the wheel and the mass of the aircraft differ
enormously in magnitude, usually up to thousands of times. 3.2.3. Runway change detection based on state feedback
Hence, compared with the dynamic change in the wheel speed, In Section 3.1, an optimization method for runway recognition
the dynamic change in the aircraft speed is very small. By let- is proposed. This method is effective in the initial stage of
ting a1 = Kkr2/JVP and a2 = kbKe/J, the following expres- braking. However, the environment of the runway is not con-
sion can be obtained: stant throughout the braking process. Water may accumulate

x_ 1 ¼ x2 in some sections of the runway. In other words, the road con-
ð24Þ dition will switch among dry, wet or other runway conditions.
x_ 2 ¼ a1 x2  a2 Q
If the optimization result of the initial braking stage is used
Let f(x1, x2) = f(x) be the state feedback equation. The during the whole braking process, frequent wheel skidding or
input of the system is selected according to the state feedback: low efficiency braking would occur. This realization also
8
> fðxÞ > e1 means that the controller needs to detect the changes in the
< q1
runway condition.
Q ¼ /ðfðxÞÞ ¼ 0 e2 < fðxÞ < e1 ) x_ ¼ gðx; /ðfðxÞÞÞ
>
: First, the impact of changing runway conditions is
q2 fðxÞ < e2 analyzed.
ð25Þ
A. Low friction coefficient runway to high friction coeffi-
Note that the system is a right-hand discontinuous system.
cient runway.
To analyze the characteristics of this discontinuous system, the
Filippov framework is introduced here. This framework is
Eq. (1) in the modeling section shows that the dynamics of
introduced in detail in the literature.31 The existence and
the wheel are related only to the braking torque and the fric-
uniqueness of the right-hand discontinuous system solution
tion torque. When the aircraft enters the high friction coeffi-
are analyzed in Appendix A.
cient runway, the change in the working point of the system
In addition, according to the previous analysis, we define
is shown by the green arrow in Fig. 11, wherein only the final
the equilibrium point set of the system:
stable state of the system is considered. Because the braking
XEq ¼ fxjx2 ¼ 0; e2  fðxÞ  e1 g ð26Þ torque does not change in this process and there exists the rela-
The method of system stability is used to analyze and calcu- tionship that rFfTb, we can conclude that the friction remains
late the system state feedback f(x). Because the system is dis- approximately constant in this process, which means that the
continuous on the right-hand side, we choose the Lur’e slip ratio of the system will decrease from k1 to k2. In this pro-
Lyapunov function to analyze the stability of the system. This cess, because the deceleration of the aircraft is approximately
method is widely used for similar problems.32 constant, the wheel deceleration and wheel speed will rapidly
8 increase.
< 1=2x2 þ q1 ðfðxÞ  e1 Þ
> fðxÞ > e1
2
Z fðxÞ The changes in the above process are reflected in Fig. 12.
1
VðxÞ ¼ x22 þ /ðgÞdg ¼ 1=2x22 e2  fðxÞ  e1 When this process occurs, the system working point originally
2 0 >
:
1=2x22  q2 ðfðxÞ þ e2 Þ fðxÞ < e2 located in the shaded area will enter the first quadrant due to
ð27Þ the rapid increase in the wheel speed and wheel deceleration.
Using this feature, runway change detection can be
To ensure that the generalized derivative of V(x) is less than achieved by using the state feedback function and setting the
or equal to 0, the state feedback function f(x) needs to satisfy error limit ec (ec > e1) when Q  0 (pressure maintenance or
the following conditions. The detailed proof and derivation of increase). If the working point suddenly enters the area of f
the system stability are given in Appendix B. (x) > ec, then we can determine that the maximum friction
High-efficiency aircraft antiskid brake control algorithm 2545

coefficient, this deceleration is too large for the runway with a


low friction coefficient, which will eventually lead to wheel
skidding. The working point enters the third quadrant from
the shaded area.
Using this feature, runway change detection can be
achieved by using the state feedback function and setting the
error limit ec (ec < e2) when Q  0 (pressure maintenance
or decrease) and Pb < cpPbmax. If the working point suddenly
enters the area of f(x) < ec, then we can determine that the
maximum friction coefficient of the runway decreases. Note
Fig. 11 Changes during runway switching (low to high). that cp is a threshold coefficient used to determine the differ-
ence between the actual braking pressure Pb and Pbmax. This
kind of detection needs to be maintained for a period of time
to eliminate disturbances, such as local water accumulations
on the runway. After this judgment, the following work
involves decreasing the pressure to a pressure that does not slip
under any runway conditions and using the optimization
method in Section 3.1 to find the maximum friction point of
the low coefficient runway.

3.2.4. Multivalve array control designs


To further explore the characteristics of this control algorithm
and derive control strategies of multiple groups of on-off
valves, simulations of on-off valve control with different flow
Fig. 12 Runway state switching detection (low to high). Q are carried out. During the simulation process, the threshold
e1 and e2 of the state feedback function remain constant.
The inlet flow of the switch valve in the simulations is q1
coefficient of the runway increases. After this judgment, the and 4q1, and the outlet flow is q2 and 4q2, i.e., the flow
following work involves using the optimization method in Sec- of the high flow on-off valve group is four times that of the
tion 3.1 to find the maximum friction point of the high coeffi- other group.
cient runway and controlling the wheel speed using the state Figs. 15 and 16 show the speed control capability of the sys-
feedback function. tem at a constant deceleration under two kinds of on-off valve
flow. The preset deceleration and the wheel reference speed
B. High friction coefficient runway to low friction coeffi- curve are the same in the simulations. The only difference is
cient runway. the flow of the on-off valve. The control signal of the on-off
valve in the simulation is displayed. When the value of the sig-
Unlike the first case, when the lmax of the runway the run- nal is +1, the upstream on-off valve is controlled to open and
way condition becomes worse. Because the braking torque the downstream on-off valve is controlled to close, which will
does not change in a short period of time and the friction tor- cause the brake pressure to increase. When the value of the sig-
que decreases rapidly, the wheel will be locked quickly and the nal is 1, the downstream on-off valve is controlled to open
slip ratio will change very rapidly from k2 to k1 (k1 = 1), as and the upstream on-off valve is controlled to close, which will
shown in Fig. 13. cause the brake pressure to decrease. When the value of the
In the system diagram analysis shown in Fig. 14, in the signal is 0, both valves are closed, which will cause brake pres-
presence of wheel speed control, the controller will actively sure to maintain.
reduce the pressure and stabilize the working point in the Note that to reflect the actual situation as accurately as pos-
shaded part. However, because the preset deceleration of wheel sible, the gap between the initial position of the brake actua-
speed control is determined by the runway with a high friction tors and the brake discs is added in the simulation, which

Fig. 13 Changes during runway switching (high to low). Fig. 14 Runway state switching detection (high to low).
2546 D. SUN et al.

The results and phase diagrams in Fig. 17 show that a small


flow will cause a slow convergence process; however, the con-
trol process is smoother under a small flow than a large flow.
The system would converge faster if the flow of the brake valve
is increased, but it will lead to the fluctuations in the system
that are difficult to suppress. Similar conclusions can be drawn
from the system phase trajectories of the two simulation pro-
cesses. Under a large flow, the state feedback function can
restrict the error within a small range, but it will lead to a lar-
ger control step, which means that the system trajectory is
more likely to cross the system equilibrium point set and oscil-
late near the equilibrium point set. These oscillations will result
in many unnecessary actions of the valve during the control
process. When the flow is small, the system converges slowly
due to the smaller control step, and the error range may be lar-
ger, but it is easier to converge to the set of equilibrium points.
Through the above simulations and analyses, we hope to
combine the control advantages of the two kinds of on-off
valves so that the whole wheel speed control process can be
fast and smooth. In addition, the number of valve actions in
the whole control process should be minimized to prolong
the service life of the on-off valve.
The basic idea of wheel speed control using an on-off valve
array is to select the valve with appropriate flow according to
the value of the state feedback function. The greater the differ-
ence is between the value of the state feedback function and the
Fig. 15 Wheel speed control effect (Q = {q1,-q2}). state feedback threshold (e1, e2), the greater the flow of the on-
off valve array will be applied.
Take the on-off valve array composed of three groups of
on-off valve arrays as an example. The flow of upstream on-
off valves is Quv = {q1u,q2u, q3u}(q1u < q2u < q3u), and the
flow of downstream on-off valves is Qdv = {q1d, q2d,
q3d}(q1d > q2d > q3d), the subscript u represents
upstream, d represents downstream. Flow combination can
be achieved by controlling multiple on-off valves to open
and close simultaneously. Therefore, in the process of wheel
speed control, the upstream flow of the system is Qu = {q1u,
q2u,q3u, q1u + q2u,q1u + q3u,q2u + q3u,q1u + q2u + q3u} and
the downstream flow is Qd = {q1d,q2d,q3d,q1dq2d,
q1dq3d,q2dq3d, q1dq2dq3d}. The positive threshold of
the state feedback function is defined as ep = {ep1,ep2,ep3,. . . ,
ep8}, and the negative threshold is defined as en = {en1,en2,
en3, . . . ,en8}, the subscript p represents positive, n represents
negative. The control law can be designed as follows:
8
< Qu ðiÞ
> ep ðiÞ < fðxÞ < ep ði þ 1Þ
Q¼ 0 en ð1Þ < fðxÞ < ep ð1Þ i ¼ 1; 2; 3; :::; 7
>
:
Qd ðiÞ en ði þ 1Þ < fðxÞ < en ðiÞ
ð30Þ

Fig. 16 Wheel speed control effect (Q = {4q1,-4q2}).

means that the on-off valve controlling the increase pressure


needs to be kept open for a period of time to fill the chamber
so that the actuators contact the brake discs. After this pro-
cess, the brake pressure will increase rapidly. This is the reason
why the pressure is approximately constant while the upstream
valves remain open at the beginning of the simulation curves. Fig. 17 Wheel speed control phase diagram under different flow.
High-efficiency aircraft antiskid brake control algorithm 2547

3.3. Overall structure design of the controller system and the maximum friction point is obtained. On this
basis, the threshold g is set to determine whether the working
The design in the above two parts mainly solves the problem of point is sufficiently close to the runway maximum friction
runway condition identification and wheel speed control by point. When the judgment condition is satisfied, the brake
using the on-off valve array. The antiskid control law of the pressure Pb is sampled as Pbmax. According to Pbmax, the pre-
aircraft braking system is composed of two parts. The overall set wheel deceleration x_ is calculated under current runway
structure of the controller is shown in Fig. 18. conditions. Then, the optimization control is switched, and
The whole control algorithm consists of two parts: runway the wheel speed control begins to dominate the braking
identification and wheel speed control based on state feedback. control.
The two parts are controlled at different times, and the control In the process of wheel speed control, the system samples
law needs to be switched. the wheel speed x and uses the TD to calculate the state feed-
The two control laws should be switched depending on the back f(x), and the error is judged by the state feedback thresh-
system status, and the function block ‘‘Switching of control old e. The output flow Q is determined according to the error.
laws” in Fig. 18 is a function designed to achieve this goal. Then, the flow Q is converted to the array on-off signal and
The switching of the control laws can be triggered by two con- output to the on-off valve array. Moreover, Q and Pb are com-
ditions. One condition is that the maximum friction is found bined with the state feedback function f(x) to monitor the
by the optimization control in the initial stage of braking; change in the maximum friction coefficient of the runway.
the control law is converted from optimization control to Through the above processes, aircraft antiskid brake con-
wheel speed control. The other condition is that the change trol based on an on-off valve array is achieved.
in the lmax is detected; the wheel speed control will be switched
to optimization control. Note that when the latter condition is 4. Simulation and analysis of the control algorithm
satisfied, it is necessary to determine the baseline value of
brake pressure increase for optimization control according to To verify the effectiveness of the antiskid control of the aircraft
the change direction of lmax. by using the switch valve array, the control algorithm is simu-
In the optimal control, the controller collects the actual lated and analyzed first. In this paper, simulations are carried
pressure signal Pb and wheel speed x from the brake system out on the MATLAB/Simulink platform, and the step size is
and controls the brake pressure to increase according to the fixed at 1  104 s. The sampling period and control calcula-
preset fixed slope. Then, the aircraft speed Vp is estimated. tion period are both 5 ms. The main parameters of the simula-
The Kalman filter is used to estimate the ground friction Ff tion model are listed in Table 1.
and the friction change rate dFf/dt. The estimated aircraft This algorithm is designed to achieve antiskid control. To
speed and wheel speed are used to calculate the slip ratio k, evaluate the effectiveness of the control algorithm, the concept
and the TD is used to calculate the slip ratio differential of the friction coefficient efficiency of the brake is introduced
dk/dt. By combining dFf/dt and dk/dt, the variable dFf/dk to evaluate the effectiveness of the control algorithm in the
reflecting the distance between the working point of the antiskid process.

Fig. 18 Overall control structure diagram.


2548 D. SUN et al.

maximum friction coefficient of runway, and L is the braking


Table 1 Main parameters list of the simulation. distance. The efficiency of the friction coefficient is obtained by
Parameters Symbols Values integrating the utilization ratio of the friction coefficient in the
whole braking process with respect to the braking distance.
Aircraft mass M0 18000 kg
Wheel radius r 0.35 m
The ‘‘PBM” algorithm, which has no aircraft speed signal as
Gravity acceleration G 9.8 m/s2 an input, is used for comparison. Many simulations and exper-
Wheel load FN 50 kN iments show that the friction coefficient efficiency is 50%–80%
Wheel Inertia J 9.5 kg/m2 under wet runway conditions and 80%–90% under dry run-
Initial brake velocity Vp0 50 m/s way conditions when using the PBM control method.10
Brake disc torque-pressure coefficient kb 1000 kNm/MPa Using the above parameters and evaluation methods, we
Oil source pressure Ps 14 MPa simulated four different working conditions. To simulate the
Initial volume of brake actuator VL 100 mL braking process, the braking instructions are given after the
Flow of upstream on-off valves Quv {1, 2, 4} L/min aircraft landed for 1 s. In simulations, the pressure increase
(DP = 10 MPa)
slope of the optimization process is set to 8 MPa/s. Since the
Flow of downstream on-off valves Qdv {1, 2, 4}
(DP = 10 MPa) L/min
flow of the on-off valve has been defined in the above param-
eter table, we quantify the on-off signal of the on-off valve
array as {7,6,. . .,6,7} in the simulation process to open or
close the on-off valve or a combination of on-off valves to pro-
duce a particular flow.
RL I Dry runway condition. The maximum friction coefficient
lb ðlÞdl
gl ¼ R L0 ð31Þ of the runway is 0.5, and the runway state does not change
0
lmax ðlÞdl during the whole process. The simulation results are shown
in Fig. 19.
where gl represents the friction coefficient efficiency, lb is the The simulation results above show that the algorithm is very
actual ground friction coefficient used in braking, lmax is the effective under dry runway conditions. Fig. 19(a) and (b) show

Fig. 19 Brake simulation on dry runway.


High-efficiency aircraft antiskid brake control algorithm 2549

that the whole braking process is smooth without skidding. The In the simulation results of working condition II, Fig. 20(a)
braking pressure remains high and stable after finding the max- and (b) show that the whole braking process is also smooth
imum friction point. During the optimizing process, the pres- and steady. The braking pressure keeps high and stable after
sure fluctuates slightly while the slope increases. Fig. 19(c) finding the maximum friction point. Brake pressure is main-
shows that although there is pressure fluctuation, the differen- tained at a low level after the optimization process, but
tial tracker takes into account both signal smoothness and Fig. 20(d) shows that lb is close to lmax during the whole pro-
tracking speed. Although the estimated value of dFf/dk fluctu- cess. Fig. 20(c) shows that the estimated value of dFf/dk can
ates slightly, the overall trend is consistent with the actual value still follow the true value of dFf/dk accurately, and the maxi-
of dFf/dk, and the maximum friction point is successfully mum friction point is successfully detected at the threshold g
detected at the threshold g without slipping. Fig. 19(d) shows without slipping. Fig. 20(e) shows that, the whole braking pro-
that the maximum friction coefficient of the runway is fully uti- cess hardly requires the action of the on-off valve. The braking
lized during the braking process and that lb is close to lmax. friction coefficient efficiency reaches 85.77%. The proposed
Fig. 19(e) shows that, except for pressure following in the opti- control algorithm is more efficient than ‘‘PBM” under this
mization stage, there is little need for the on-off valve to act on condition.
the whole control process. The braking process is smooth and III. Wet-dry-wet runway condition. Conditions of dry and
efficient, and the braking friction coefficient efficiency reaches wet runways are consistent with those of the first two condi-
91.54%. The proposed control algorithm is more efficient than tions. At 7 s, the runway changes from wet runway to dry run-
the ‘‘PBM” algorithm under this condition. way, and at 15 s, from dry runway to wet runway. The runway
II. Wet runway condition. The maximum friction coeffi- state does not change in the rest of the time. The simulation
cient of the runway is 0.25, and the runway state does not results are shown in Fig. 21.
change during the whole process. The simulation results are In the simulation results of working condition III, Fig. 21(a)
shown in Fig. 20. and (b) show that when the runway changes at 7 s and 15 s, the

Fig. 20 Brake simulation on wet runway.


2550 D. SUN et al.

Fig. 21 Brake simulation on wet-dry-wet runway.

Table 2 List of experimental parameters.


Parameters Symbols Values
Aircraft mass M0 18000 kg
Wheel radius r 0.35 m
Gravity acceleration g 9.8 m/s2
Wheel load FN 50 kN
Wheel Inertia J 9.5 kg/m2
Initial brake velocity Vp0 50 m/s
Brake disc torque-pressure coefficient kb 1000 kNm/MPa
Oil source pressure Ps 14 MPa
Initial volume of brake actuator VL 100 ml
Flow of upstream on-off valves Quv {1, 2, 4} L/min
(DP = 10 MPa)
Flow of downstream on-off valves Qdv {1, 2, 4}
(DP = 10 MPa) L/min

process is smooth and steady. Fig. 21(b) shows that the braking
pressure can automatically change into a suitable value accord-
ing to the maximum friction coefficient. It suggests that the
Fig. 22 HIL braking test bench. wheel speed control can achieve accurately runway condition
identification. According to the variation trend in the runway
wheel speed fluctuates. At 15 s, due to the decrease in the run- condition, the initial pressure of the first optimization is the
way friction coefficient, a slight slip occurs. The whole braking pressure value of the wheel speed control at the last moment,
High-efficiency aircraft antiskid brake control algorithm 2551

while the initial pressure of the second optimization is 1.8 MPa. experiments. This method connects the hydraulic components
Fig. 21(c) shows that the estimated value of dFf/dk is still very of the brake system with the dynamic model of the wheel and
close to the real value after the state feedback detects the road the aircraft. In the field of braking, this kind of experimental
condition change at 7 s, and the maximum friction point with method is often used to verify the effectiveness of hydraulic
high friction coefficient is successfully detected without slip- components and control algorithms in the brake system. In
ping. Fig. 21(d) shows that lb is close to lmax during the whole this paper, a new hydraulic component and a new algorithm
process under different runway conditions. At 15 s, the runway are provided. It is difficult to accurately model the hydraulic
friction coefficient suddenly decreases. To prevent excessive part because of many uncertainty factors, such as leakage.
slipping, and because the value of the state feedback function The HIL experiment directly provides the physical object of
exceeds the preset threshold, the controller gives an on-off sig- this part, so the component and control algorithm can be val-
nal that triggers a valve combination that produces a flow of idated effectively.
4 L/min, as shown in Fig. 21(e). This signal rapidly reduces Fig. 22 shows the experimental facility for the HIL experi-
the pressure, causing the slip to become very small. Moreover, ments. Aside from a reduction in pipeline length, the experi-
due to the existence of the state feedback function, the braking mental facility has a hydraulic component configuration that
pressure does not decrease to 0 but maintains a relatively low is exactly the same as that of the aircraft. At the beginning
level without slipping until condition Q  0, Pb < cpPbmax of the experiments, the dynamic model of the aircraft and
and f(x) < -ec are satisfied. The braking friction coefficient effi- the wheel starts to run on the computer. The brake pressure
ciency reaches 86.83%. in the model is collected by the pressure sensor in the brake
chamber. The brake control unit causes the pressure to
increase and decrease by controlling the on-off valve array,
5. Experiments and the brake pressure is transmitted directly to the wheel
brake actuator. This process will continue until the aircraft
The control algorithm proposed in this paper has achieved speed in the computer model drops to 0.
good results in simulations. In addition, the effectiveness and The prototype of the designed on-off valve array is also
stability of the algorithm are further validated by HIL shown in Fig. 22. The upper three valves are used to control

Fig. 23 HIL experiment on a dry runway.


2552 D. SUN et al.

the pressure increase upstream, and the lower three valves are decreases. Hence, it can be concluded that there exists a slight
used to control the pressure decrease downstream. Six on-off leakage in the brake chamber. The decrease in pressure will
valves are divided into three groups according to the lead to a change in the state feedback function. The control
flow. The flow of each group from left to right is 4 L/min, algorithm opens the upstream on-off valves according to this
2 L/min, and 1 L/min (DP = 10 MPa). change and increases the braking pressure to maintain the
The experimental parameters are listed in Table 2. The stable tracking of the reference wheel velocity. Leakage was
experimental results are provided in Fig. 23, Figs. 24, and 25. not considered in the simulation, which explains why the num-
Fig. 23 shows the results of the dry runway test. Fig. 23(c) ber of actions of the on-off valve in the experiment is larger
shows that by setting the threshold and calculating the dFf/dk, than that in the simulation. In addition, the braking process
the controller successfully detected the maximum friction of is smooth, and there is no slippage. Fig. 23(d) shows that the
the runway. During the tests, the motor of the hydraulic source maximum friction coefficient of the runway is 0.5, and the
produces electromagnetic interference to the sampling of pres- actual utilization of the friction coefficient during the braking
sure sensors, which also exists in the aircraft landing process. process is very close to 0.5. According to the calculation
This problem is difficult to accurately repeat in simulations. method of braking efficiency mentioned above, the braking
This will lead to nonsmooth pressure sampling. These efficiency reaches 90.72%.
disturbances introduce a substantial amount of noise to the The wet runway HIL braking experiment results are dis-
calculation of wheel deceleration, which increases the noise played in Fig. 24. The controller successfully detected the max-
in the dFf/dk signal. However, these disturbances do not affect imum friction of the runway. The number of actions in this
the detection of the maximum friction point. The target decel- experiment is larger than that in simulations because of the
eration is generated according to the Pbmax. In contrast to leakage of the system. In addition, the braking process is
Fig. 23(b) and (e), when both upstream and downstream on- smooth and there is no slippage. Fig. 24(d) shows that the
off valves are closed, the brake chamber pressure still slowly maximum friction coefficient of the runway is 0.25, and the

Fig. 24 HIL experiment on a wet runway.


High-efficiency aircraft antiskid brake control algorithm 2553

Fig. 25 HIL experiment on dry and wet runway.

actual utilization of the friction coefficient during the braking controller gives the maximum pressure relief instructions
process is very close to 0.25. According to the calculation immediately to prevent deep slippage. If the features of the
method of braking efficiency mentioned above, the braking on-off valve are better, the algorithm can also achieve better
efficiency reaches 83.27%. Through the analysis of the two performance. The braking efficiency reaches 80.69%.
working conditions above, it can be concluded that the pro-
posed algorithm can effectively adapt to the features of the 6. Conclusion
actual hydraulic system.
Similar to the simulation above, the HIL braking experi- In this paper, the on-off valve array is used to replace the tra-
ment for dry-wet runway switching condition is also performed ditional servo valve used in the aircraft braking process to
in Fig. 25 When the wet runway is switched to the dry runway, overcome the oil contamination problem. Based on this new
the switch process is smooth and there is no slippage. The algo- control component, an efficient aircraft antiskid braking con-
rithm accurately finds the maximum friction points of the two trol is proposed. Under the condition that the controller only
runways in the optimization process. The maximum friction has the wheel speed and the pressure signals as inputs, by
coefficients of the two runways are 0.5 and 0.25. The runway introducing the Filippov framework, the discontinuous feature
friction coefficient used in braking process is also very close of the on-off valve is used to design the control algorithm. By
to these two values. It should be noted that in the process of observing the slope of the friction curve with Kalman filter, the
switching from dry runway to wet runway, there exists a deep maximum friction of the runway is detected, and the state
slippage. From Fig. 25(a) and (b), the reason is that although change of the runway is detected by constructing the state feed-
the controller gives the control signals of all downstream on- back function. These two methods together constitute a precise
off valves opening at the same time, the pressure drop is not and stable runway recognition algorithm. The stability and the
timely due to the relatively slow response of large flow on- disturbance resisting capacity of the system are also verified.
off valves. This is also difficult to simulate accurately in the The results of the simulation analyses and HIL braking exper-
simulations. On the other hand, this process also verifies iments show that the proposed control algorithm can accu-
the effectiveness of the algorithm. When the slip occurs, the rately identify the maximum friction of the runway without
2554 D. SUN et al.

skidding and achieve higher braking efficiency than the @f/ox2  0 is a sufficient condition for our system to have
‘‘PBM” method. It can be concluded from this paper that an unique solution.
the on-off valve array can replace the servo valve perfectly as
a new type of antiskid braking pressure control component, Appendix B. Analysis of system stability
and achieve high braking efficiency.
The method of system stability is used to analyze and calcu-
Acknowledgments late the system state feedback f(x). Because the system is dis-
continuous at the right-hand side, we choose the Lur’e
The authors would like to thank the Science and Technology Lyapunov function to analyze the stability of the system, this
on Aircraft Control Laboratory and the National Nature method is widely used in the field of similar problems33–35.
Science Foundation of China (Nos. 51775014 and 51890882). 8
< 1=2x2 þ q1 ðfðxÞ  e1 Þ
> fðxÞ > e1
2
Z fðxÞ
1
VðxÞ ¼ x22 þ /ðgÞdg ¼ 1=2x22 e2  fðxÞ  e1
Appendix A. Analysis of the existence and uniqueness 2 0 >
:
1=2x22  q2 ðfðxÞ þ e2 Þ fðxÞ < e2
ðB1Þ
According to Filippov theory, the definition of the Filippov
solution of a differential equation with right-hand discontinu- We choose the Lyapunov function as shown above. Obvi-
ous can be understood intuitively as follows: if x(t) is the solu- ously, V(x) is regular whether it is on the switching surface
tion of a right-hand discontinuous differential equation g(x, or not. In addition, we find that V(x) is continuous, but not
u), its tangent vector at the discontinuous point must belong smooth. In the process of stability analysis, we need to derive
to the convex hull of the limit vector field in the infinitesimal V(x). For non-smooth V(x), Clark’s generalized time deriva-
neighborhood of discontinuous points.32 tive method is used. In this method:
8 dVðxÞ a:e: _
>
> gðx; q1 Þ fðxÞ > e1 2 VðxÞ ¼ \ fT K½gðxÞ ðB2Þ
>
> dt f2@VðxÞ
>
< gðx; pþ q1 Þ
> fðxÞ ¼ e1
In order to ensure the stability of the system, i.e. dV/dt < 0,
x_ 2 K½gðxÞ ¼ gðx; 0Þ e2 < fðxÞ < e1 ðA1Þ
>
> Clark’s generalized derivative must be guaranteed to be less
> gðx; p q Þ
> fðxÞ ¼ e2
>
> 2 than or equal to 0. The stability analysis is needed in different
:
gðx; q2 Þ fðxÞ < e2 regions.
Case (f(x) = e1):
The existence and uniqueness of the system solution in On the switching surface, the generalized differential of V
Filippov framework are proved here. The system state vector (x) can be written as:
x and the input Q are bounded, so the value of piecewise  
_
x2
continuous function g(x,Q) is locally bounded. Through the VðxÞ ¼ \ fT ðB3Þ
mapping of functional K[], the set K[g] is a non-empty locally f2@VðxÞ a1 þ a2 pþ q1
bounded closed convex set in the solution space of the system. Similar to the previous analysis, @V can be expressed as:
Moreover, the function K[g] is semi-continuous to x, that is, 
at any point x0. @VðxÞ ¼ cof½0; x2 T ; ½q1 @f=@x1 ; x2 þ q1 @f=@x2 T g ðB4Þ

8d > 0; 9r > 0; if jx  x0 j < r ) K½gðx; /Þ Owing to f2@V, the f can be written as:
   
< K½gðx0 ; /Þ þ d ðA2Þ 0 @f=@x1
f¼ þ q1 d d 2 ½0; 1 ðB5Þ
x2 @f=@x2
The system satisfies the basic conditions of the existence
theorem of solutions in reference. The generalized derivative of V(x) can be written as:
For uniqueness, we mainly discuss the uniqueness of solu-  T   " @f #T  
tions at discontinuous switching surfaces. Using the f(x) _
0 x2 @x x2
VðxÞ ¼ þ q1 d @f1
= e1 as an example. The vector function g+ is defined as the x2 a1 þ a2 pþ q1 @x
a1 x2 þ a2 pþ q1
2
limiting value of the region (f(x) > e1). The same method _
¼ x2 ða1 x2 þ a2 pþ q1 Þ þ q1 dfðxÞ
defines g-, corresponding region (f(x) < e1).
8   ðB6Þ
>
> x2
> gþ ¼ gðx; q1 Þ ¼
< Because in this case, f(x) satisfies the condition f(x) = e1:
a1 x2 þ a2 q1
  ) h ¼ gþ  g ðA3Þ
>
> x2 _ @f @f
> g ¼ gðx; 0Þ ¼
: fðxÞ ¼ e1 ) fðxÞ ¼ x2 þ ða1 x2 þ a2 pþ q1 Þ ¼ 0
a1 x2 @x1 @x2
_

The normal vector of the switching surface is defined as N. ) VðxÞ ¼ x2 ða1 x2 þ a2 pþ q1 Þ ðB7Þ
According to the sufficient condition of uniqueness of the solu-
tion, the projection of h to N hN should satisfy the condition In order to make the generalized derivative of V(x) less than
that hN  0. or equal to 0, conditions @f/ox1  0 must be satisfied:
8   _
@f 2 @f
< N ¼ @f=@x1 VðxÞ  0 ) f x 0) 0 ðB8Þ
@x1 2 @x1
@f=@x2 ) @f=@x2  0 ðA4Þ
:
hh; Ni  0 Case (-e2 < f(x) < e1):
High-efficiency aircraft antiskid brake control algorithm 2555

The generalized derivative calculation is performed directly. controlled, pressure-bias-modulated system. NASA Langley
Obviously, this derivative satisfies the condition that it is less Research Center, Hampton; 1979. NASA TP-1051.
than or equal to 0 and no restriction is needed to be imposed 5. Tseng HC, Chi CW. Aircraft antilock brake system with neural
on the state feedback function f (x). networks and fuzzy logic. J Guidance Control Dyn 1995;18
(18):1113–8.
 T  
_
0 x2 6. Tunay I, Amin M, Beck AA. Robust control of a hydraulic valve
VðxÞ ¼ ¼ a1 x22  0 ðB9Þ for aircraft anti-skid operational. Proceedings of the 1998 IEEE
x2 a1 x2
International Conference on Control Applications (Cat.
Case (f(x) > e1): No.98CH36104), Trieste, Italy, 1998. p. 689-93.
In this case, the derivative of V(x) can also be calculated 7. Tunay I, Rodin EY, Beck AA. Modeling and robust control
directly. design for aircraft brake hydraulics. IEEE Trans Control Syst
Technol 2001;9(2):319–29.
 T  
_
q1 @f=@x1 x2 8. Li FY, Jiao ZX. Robust control for aircraft anti-skid
VðxÞ ¼ braking system based on dynamic tire/road friction force model.
x2 þ q1 @f=@x2 a1 x2 þ a2 q1
Proceedings of the 2nd International Conference on Computer
@f @f @f
¼ q1 x2 þ a1 x22 þ a2 q1 x2 þ a1 q1 x2 þ a2 q21 Science and Electronics Engineering (ICCSEE). Atlantis Press;
@x1 @x2 @x2 2013.
@f @f @f 9. Jiao ZX, Liu XC, Li FY, Shang YX. Aircraft antiskid braking
¼ ðq1 þ a2 q1 Þx2 þ ðx2 þ q1 Þa1 x2 þ a2 q21
@x1 @x2 @x2 control method based on tire–runway friction model. J Aircraft
2016;54(1):75–84.
ðB10Þ
10. D’Avico L, Tanelli M, Savaresi SM, Airoldi M, Rapicano G. An
In order to make the generalized derivative of V(x) less than anti-skid braking system for aircraft via Mixed-Slip-Deceleration
or equal to 0, the following restrictions need to be imposed on control and Sliding Mode Observer2017 IEEE 56th Annual
the state feedback function f(x): Conference on Decision and Control (CDC). p. 4503–8.
( @f 11. Jia JH, Jiao ZX, Sun D, Shang YX. Aircraft anti-skid braking
_
@x1
¼ a2 active disturbance rejection control based on optimal slip
VðxÞ  0 ) @f @f
ðB11Þ ratioCSAA/IET International Conference on Aircraft Utility Sys-
0  @x 2
 q11 jx2 j x2 ¼ 0; @x 2
¼0 tems. p. 163–9.
12. Huang C, Jiao ZX, Shang YX, Yao JY. Antiskid braking control
Using the same analytical method to analyze the remaining
with on/off valves for aircraft applications. J Aircraft 2013;50
two cases.
(6):1869–79.
Case (f(x) = -e2): 13. Jiao ZX, Liu XC, Shang YX, Huang C. An integrated self-
_
@f energized brake system for aircrafts based on a switching valve
VðxÞ  0 ) 0 ðB12Þ control. Aerosp Sci Technol 2017;60(1):20–30.
@x1
14. Kuo CY, Yeh EC. A four-phase control scheme of an anti-skid
Case (f(x) < -e2): brake system for all road conditions. Proc Inst Mech Eng, Part D:
( @f J Automobile Eng 1992;206(4):275–83.
_
@x1
¼ a2 15. Choi S, Cho DW. Control of wheel slip ratio using sliding mode
VðxÞ  0 ) @f @f
ðB13Þ
0  @x 2
 q12 jx2 j x2 ¼ 0; @x 2
¼0 controller with pulse width modulation. Veh Syst Dyn 1999;32
(4):267–84.
In order to make the generalized derivative of V(x) less than 16. Johansen TA, Petersen I, Kalkkuhl J, Ludemann J. Gain-
or equal to 0, the state feedback function f(x) needs to satisfy scheduled wheel slip control in automotive brake systems. IEEE
the following conditions. Trans Control Syst Technol 1992;11(6):799–811.
8 @f 17. Shim T, Chang S, Lee S. Investigation of sliding-surface design on
>
> 0 the performance of sliding mode controller in antilock braking
> @x1
>
>
< ¼ a2
@f systems. IEEE Trans Veh Technol 2008;57(2):747–59.
@x1 18. Jing H, Liu Z, Chen H. A switched control strategy for antilock
ðB14Þ
>
>
@f
0  @x  q1 jx2 j x2 ¼ 0; @x@f
¼0 braking system with on/off valves. IEEE Trans Veh Technol
>
>
>
2 2 2
: 2011;50(4):1470–84.
@f @f
0  @x 2
 q11 jx2 j x2 ¼ 0; @x 2
¼0 19. de Wit CC, Tsiotras P, Velenis E, Basset M, Gissinger G. Dynamic
tire friction models for vehicle traction/braking control. Veh Syst
Dyn 2003;39(3):189–226.
References 20. Jacquet A, Chamaillard Y, Basset M, Gissinger G, Frank D,
Garcia JP. Anti-lock braking system using predictive control and
1. Jean-Charles M, Jian FU. Review on signal-by-wire and power- on-line tire/road characteristics estimation. IFAC Proceedings
by-wire actuation for more electric aircraft. Chin J Aeronaut Volumes 2008;41(2):2099–104.
2017;30(3):857–70. 21. Lin G, Pang H, Zhang W, Wang D, Feng L. A self-decoupled
2. Guo SR, Chen JH, Lu YL, Wang Y, Dong HK. Hydraulic piston three-axis force sensor for measuring the wheel force. Proc Inst
pump in civil aircraft: Current status, future directions and critical Mech Eng, Part D: J Automobile Eng 2014;228(3):319–34.
technologies. Chin J Aeronaut 2020;33(1). https://doi.org/10.1016/ 22. Bakker E, Nyborg L, Pacejka HB. Tyre modelling for use in
j.cja.2019.01.013. vehicle dynamics studies. SAE Technical Paper. 1987. Report No.:
3. Jeong HS, Kim HE. Experimental based analysis of the pressure 870421.
control characteristics of an oil hydraulic three-way on/off 23. Armstrong-Hélouvry B. Control of Machines with Friction.
solenoid valve controlled by PWM signal. J Dyn Syst, Meas, Kluwer Academic Pub 1991;12–76.
Control 2002;124(1):196–205. 24. Armstrong-Hélouvry B, Dupont P, Wit CCD. A survey of models,
4. Stubbs SM, Tanner JA, Smith EG. Behavior of aircraft antiskid analysis tools and compensation methods for the control of
braking systems on dry and wet runway surfaces. a slip-velocity- machines with friction. Automatica 1994;30(7):1083–138.
2556 D. SUN et al.

25. Coetzee E, Krauskopf B, Lowenberg M. Application of bifurca- 30. Doumiati M, Victorino A, Lechner D, Baffet G, Charara A.
tion methods to the prediction of low-speed aircraft ground Observers for vehicle tyre/road forces estimation: experimental
performance. J Aircraft 2010;47(4):1248–55. validation. Veh Syst Dyn 2010;48(11):1345–78.
26. Wilkin MA, Manning WJ, Crolla DA, Levesley MC. Use of an 31. Filippov AF. Differential Equations With Discontinuous Right-
extended kalman filter as a robust tyre force estimator. Veh Syst hand Sides. Norwell, MA: Kluwer; 1988. p. 10–117.
Dyn 2006;44(1):50–9. 32. Guo BZ, Zhao ZL. On convergence of tracking differentiator. Int
27. Ray LR. Nonlinear tire force estimation and road friction J Control 2011;84(4):693–701.
identification: Simulation and experiments. Automatica 1997;33 33. Shevitz D, Paden B. Lyapunov stability theory of nonsmooth
(10):1819–33. systems. IEEE Trans Autom Control 1994;39(9):1910–4.
28. Mendonça CBD, Hemerly EM, Góes LCS. Adaptive stochastic 34. Akasaka D, Liu KZ. Stabilization of a class of on-off control
filtering for online aircraft flight path reconstruction. J Aircraft systems by nonlinear state feedback. IEEE European Control
2015;44(5):1546–58. Conference (ECC) 2007;667–72.
29. Jiang Q, Chen Q. Dynamic model for real-time estimation of 35. Devaud FM. Feedback/feedforward matrices for optimal follow-
aerodynamic characteristics. J of Aircraft 2015;26(4):315–21. ing of a forced model. J Aircraft 2015;12(8):688–9.

You might also like