You are on page 1of 25

Exponential Mapping Controller Applied to Aircraft

Hildebrando F. de Castro1, Pedro Paglione2 and Carlos Henrique Ribeiro3


Instituto Tecnolgico de Aeronutica, So Jos dos Campos, So Paulo, 12228-900, Brazil
A novel modified exponential function to achieve tracking and regulatory direct and
indirect model-free feedback control of a class of nonlinear dynamical systems is presented.
Its simplicity was a requisite to have it run on equipment and devices with memory and
processor constraints. The control algorithm needs only two parameters and can be applied
to systems based on knowledge about its free response and expected disturbances. Its main
advantages are relative ease of implementation and intuitive tuning. The controller is
applied to the simulation of two different aircraft under stochastic winds and wind shear
and its results are compared to a classical implementation. EMC presented promising
results. Its intuitive form of parameterization allowed in some cases for immediate good
results or at least for good initial estimatives for later tuning.

Nomenclature
u
v
w
p
q
r
x
y
H

=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=

forward speed
lateral speed
vertical speed
roll rate
pitch speed
yaw speed
position along the x axis
position along the y axis
altitude
roll angle
pitch angle
heading
speed
indicated air speed
angle of attack
sideslip angle
flight path angle (FPA)
heading rate
mass
mass change rate

a
e
r

=
=
=
=

aileron position
elevator position
rudder position
throttle position

=
=
=
=

initial mass
wing area
wing span
mean aerodynamic chord

Doctoral Student, Departmento de Mecnica do Voo, Praa Mal. Eduardo Gomes, 50 Vila das Accias.
Professor, Departmento de Mecnica do Voo, Praa Mal. Eduardo Gomes, 50 Vila das Accias.
3
Professor, Departmento de Teoria da Computao, Praa Mal. Eduardo Gomes, 50 Vila das Accias.
1
American Institute of Aeronautics and Astronautics
2

=
=
=
=

moment of inertia about x axis


moment of inertia about y axis
moment of inertia about z axis
moment of inertia about xz plane

=
=
=
=
=
=
=
=
=

maximum propulsive force


velocity influence on propulsive force
density influence on propulsive force
reference speed
thrust reference speed
reference air density
specific fuel consumption
induced drag
induced drag constant

=
=
=
=
=

wind x position, world coordinates


wind y position, world coordinates
wind radius
wind strength, radial direction
wind strength, vertical direction

=
=
=
=
=
=

lift coefficient
drag coefficient
side force coefficient
rolling moment coefficient
pitching moment coefficient
yawing moment coefficient

=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=

reference drag coefficient


reference lift coefficient
aircraft lift curve slope
lift due to pitch rate
lift due to elevator
reference pitching moment
pitching moment due to pitch rate
pitching moment due to angle of attack rate
pitching moment due to elevator
side force due to sideslip
side force due to aileron
side force due to rudder
rolling moment due to sideslip
rolling moment due to roll rate
rolling moment due to yaw rate
rolling moment due to aileron
rolling moment due to rudder
yawing moment due to sideslip
yawing moment due to roll rate
yawing moment due to yaw rate
yawing moment due to aileron
yawing moment due to rudder
2
American Institute of Aeronautics and Astronautics

I. Introduction

ontrol of a dynamical system can be based on its model or on different levels of prior knowledge about its free
response. Model-based linear techniques for control abound, but systems and disturbances are usually of a
nonlinear nature and can be extremely hard to model. Imprecisions may come from uncertainties about the plant and
its external disturbances, or from a purposeful choice of model simplification1. This may lead to poor or inadequate
response in a model-based control project.
Classical (e.g. robust and adaptive control) and Artificial Intelligence-based (e.g. fuzzy logic control, neural and
neuro-fuzzy) techniques exist to deal with uncertainties in nonlinear systems. While classical techniques will rely on
a model or on a simplified model of the plant to be controlled, Artificial Intelligence (AI) techniques use heuristics
or other methods to learn or to identify the plant behavior, and then to execute control actions.
Implementation of a control system can be a time-consuming task if the system is unknown or poorly known.
The derivation of a control-oriented model can take up to 90% of a project global time, and requires much
knowledge of the system to be modelled and decisions about the physical model structure, parameter identification
and experimental validation2. Maintenance of controller parameters is also a big concern in the industry. Since more
than 90% of the installed and running controllers are of the Proportional-Integral-Derivative (PID) type3, changes in
the process materials, equipment ageing and campaigns shifting demand maintenance personnel and engineering to
continuously monitor and reprogram Distributed Control Systems (DCSs), Programmable Logic Controllers (PLCs)
and other devices to guarantee proper response of the process. Moreover, most current DCSs in use were installed in
the 1970s and 1980s, and their processing power and network capabilities are limited4. While there is a trend to
update this hardware, this is a slow task because they are usually replaced only during maintenance stoppages.
With these aspects in mind, a model-free, low memory and processor consuming, intuitive to implement and
tune controller is then desirable. We thus propose here a novel sliding mode- and fuzzy logic-inspired exponential
mapping controller (EMC) which seeks to take advantage of both the conceptual simplicity of the bang-bang-based
Sliding Mode Control (SMC) and the lack of need of a model, shape-modifiable response and use of heuristics for
tuning of the Fuzzy Logic Controller (FLC).
The rest of the paper is organized as follows. Sections 2 and 3 review concepts and bibliography related to SMC
and FLC, respectively. Section 4 presents the EMC controller. Section 5 shows its implementation and simulation
results for two different aircraft. Finally, Section 6 presents the conclusions of the paper.

II. Sliding Mode Control


Sliding Mode Control (SMC) is known as a method for dealing with nonlinear systems with uncertain dynamics
and disturbances because of its order reduction property and low sensitivity to disturbances and process parameter
variations, which relaxes the burden of the need for exact modeling5. It does so by application of a discontinuous
control signal that forces the system to slide along a specified surface of the systems normal behavior.
For second order systems, the sliding surface (a line) can be defined as:
(1)

where is the system output and


is the reference to be followed,
derivative. The control law consists of two parts:

(2)
is a constant,

is the error and its first


(3)

( )

(4)

where is the control input,


is the equivalent control term,
is the corrective term, is a strictly positive
( ) is the sign function. The equivalent control6 uses the available model of the system dynamics
constant and
and the sliding surface definition to determine a control action that would keep the state on the sliding surface, if the
model were perfect. The corrective term guarantees that the state will be drawn to and remain on the sliding surface
3
American Institute of Aeronautics and Astronautics

despite model and disturbances uncertainties. However, this term can also be a cause of chattering, which occurs in
systems with delays in their actuators or limited sampling rate and can lead to high oscillatory control efforts and
ultimately instability. Slotine and Li1 proposed a way to smooth it out by means of a boundary layer neighboring the
so called switching surface. This can be accomplished by replacing the sign function by a saturation function such
that:
( )

(5)

where:
| |
( )

(6)

and is the term that defines when interpolation will occur. Song and Smith 7 observed that both and can be
derived using open loop experimental data. They proposed to apply maximal input and record the saturation result to
define . Additionally, the disturbance magnitude can be used to define . Standard linear controller design can only
linearly approximate this nonlinear curve.
A hyperbolic tangent function can also be used:
( )

(7)

Finally, a bang-bang-type SMC is a direct switching control strategy that drops the equivalent term:
( )

(8)

where
is large enough to suppress all bounded uncertainties and unstructured systems dynamics 8. The
9
Fillipov method is used for its design, by which a sufficient large local attractor is defined for the system trajectory.

III.

Fuzzy and Neuro-Fuzzy Control

10

Fuzzy logic, proposed by Zadeh in his seminal work for logic calculus and later extended11 is a way to
approach problems that are difficult or unnatural to define in a binary, crisp sense.
Fuzzy systems use fuzzy sets. A fuzzy set is a set without a clearly defined boundary. Its elements contain only a
partial degree of membership:
( )

(9)

where ( ) is the membership function for a fuzzy set ,


is an element of the set and is the universe of
discourse. It means that can have degrees of membership from 0 to 1, inclusive. Thats much different from
ordinary logic that accepts only true or false.
Fuzzy sets are subject to set operations such as union, intersection, and complement, which express logic
statements or propositions. Fuzzy relations can be represented linguistically by natural language statements in the
form of fuzzy IF or antecedents, and THEN or consequent rules. Table 1 below shows an example of a set of fuzzy
rules for a speed control system.
Table 1 Fuzzy rules for a speed control system
IF

THEN

Speed is too slow

Accelerate hard

Speed is slow

Accelerate
4
American Institute of Aeronautics and Astronautics

Speed is ok

Do nothing

Speed is fast

Break

Speed is too fast

Break hard

The first implementation of fuzzy logic to dynamic control was performed by Mamdani 12. Fuzzy Control is
currently used in many industrial applications13 and in many other fields14,15.
The process of using fuzzy logic to implement a control system consists of a fuzzification of the input variables
( and , for example, for a SISO control), application of the fuzzy operators, matching of the antecedents and
inference of consequent parts, aggregation of the consequents and finally, defuzzification16. The fuzzy aspect of the
rules tackels the imprecise definition of the system, allowing for imprecisions in the design of the control system to
be tolerated to a certain degree and eliminating the need for a well-defined mathematical model of the plant17, but
the translation of these linguistic rules into actions depend on the choice of certain parameters, for which there are
no formal methods to get. These parameters, however, can be defined by adaptive schemes, such as those defined by
neural network architectures, designed to learn from training data, but which are in general not able to profit from
structural knowledge18.
A neural network (NN) is a mathematical structure comprised of neurons and layers. A three-layer feed-forward
NN is capable of approximating any nonlinear function19. The fundamental building block for a NN is the neuron,
consisting of a weight function, a net input function and the transfer function:
(

(10)

where is the input,


is the weight, is a constant bias, ( )is the transfer function, usually linear for the
output layer and nonlinear for the internal ones and
is the output. The parameters and are both adjustable
scalars. This flexibility allows for optimization algorithms to be employed in order for convergence or learning to
occur.
A. The Neo-Fuzzy Neuron
The Neo-Fuzzy Neuron (NFN) is a type of neuro-fuzzy structure that allows fast training time and relatively low
processing demands20. The input signals through the nonlinear transfer function are calculated as:

( )

(11)

where
is the number of inputs. For each nonlinear transfer function, the following fuzzy inference with a
singleton consequent is computed:
(12)
where
is a fuzzy set whose membership function is
and
is a singleton.
Since each membership function in the antecedent is triangular and complementary, an input signal activates
only two neighboring membership functions simultaneously so the sum of the grades of these membership functions
is always 1. Thus, the defuzzification taking a center of gravity does not need a division:
( )

( )

( )

(13)

The NFN then uses gradient descent to reduce the errors through the adjustment of the weights
. In the work
of Gouva21 this already relatively simple structure is further minimized in the so called ONFC (Online Neurofuzzy
Controller). The ONFC uses only one input, two linear membership functions and one output for SISO feedback
control, so only two gains have to be adjusted. This simplicity allows it to be implemented in a DCS function block.
Carvalho22 implemented a modified ONFC in a refinery coke processing unit with very good results.

5
American Institute of Aeronautics and Astronautics

IV. The EMC


The proposed EMC (Exponential Mapping Controller) seeks to make use of some of the advantages of SMC and
FLC in a simplified and intuitive way. The relative ease of implementation of a bang-bang SMC coupled with a
boundary layer to deal with chattering and the lack of need of a prior model of the system and intuitiveness of
implementation of a FLC.
Yagiz and Hacioglu23 proposed improving the response within the boundary layer in a SMC through
dynamically deciding the slope of a linear sliding surface. Their results are compared to a traditional constant slope
linear sliding surface SMC with good results. Ha et al.24 presented a way to tune a fuzzy SMC (FSMC) by rotating
and shifting a nonlinear sliding surface and achieved faster reaching times and improved tracking error. Tokat et
al.25 proposed a parabolic function coupled to a shifted sigmoid function to change the linear discontinuous control,
the equivalent term was computed with a simplifed yet model-based piecewise function. Choi and Park26 proposed a
Moving Sliding Surface (MSS) that was dependent on the initial conditions of the system and through rotations and
shifts would subsequentely move it to a predetermined switching surface. Park and Choi27 generalized the MSS to
higher-order SMCs by checking the conditions to rotate and shift the sliding surface. These approaches, however,
make use of the equivalent control and therefore need a model of the system. Tabatabaei et al.28 proposed a
sigmoidal mapping function for control of molten metal pouring, but it needs eleven parameters to be defined and
ten rules for the adaptation of the algorithm.
EMC implements a bang-bang SMC-inspired approach with a heuristically-defined nonlinear mapping function
comparable to a bounded linear sliding surface. The shape and boundary of the mapping function are defined by the
plant operator based on partial knowledge about the system to be controlled and about expected disturbances. The
boundary is derived from basic information about the open loop response of the plant and its shape is defined based
on knowledge about the system behavior.
The number and format of the membership functions in an FLC allow for altering the shape of its response
function. This can be simplified by using only one shape-modifiable function. This in turn accounts for easier
implementation since there is no need to define the number and shape of the membership functions and allows for
faster computation times and more modest memory requirements, at the cost of a limited set of shapes.
Since in practice EMC needs only two parameters its implementation and tuning are much simplified. This
simplicity allows for restoring the responsability of tuning back to the systems operator or engineer. PIDs have three
or more non-intuitive parameters to be tuned. As a matter of fact, the number of parameters to tune an industrial
scale PID controller is usually five. Typical SISO FLCs, even though intuitive to implement, typically need at least
nine parameters to be tuned. We understand, however, that tuning only a small number of intuitive parameters is
crucial for the implementation of automatic control systems in the industry. An alternative would be designing
adaptable, untunable and hands-off controllers, but those depend on a high level of confidence - usually derived
from design efforts with long duration and expensive costs - before its implementation can be carried out. Also,
controllers maintenance is another key issue for the industry which further motivate the design and use of relatively
easy-to-tune controllers.
EMC is implemented as follows. First it calculates the error, as in equation 1, redefined here:
(14)
Then a switching line is computed:

(15)

where is heuristically defined as being the acceptable error before applying full control input. For lower order
or non-oscillatory systems the term can usually be dropped. Then is restricted to vary between -1 and 1 to be
within the exponential function allowable range:

And then an exponential function is computed:


6
American Institute of Aeronautics and Astronautics

(16)

( ) (

|| |

(17)

where
is the parameter that alters the shape of the exponential function, allowing for a class of
nonlinearities and disturbances to be dealt with.
Finally, the controller output is calculated:
(

(18)

where
and
are the actuator minimum and maximum positions, respectively.
EMC can be used for direct and indirect control. The above equations are for the direct case. Direct control is
used when
for
, otherwise indirect control must be used. For the indirect case equation (18) becomes
equation (19):
(

(19)

where
and
are the controller inputs in time and
, respectively and
and
actuator minimal and maximal changes during the sampling period .
Figure 1 presents the error and controller output for the EMC direct exponential function, for
and
, for
, not considering .

10

uEMC

-5

-10
-15

-10

-5

Figure 1. Plot of

e
and

Figure 2 emphasizes the difference in response due to

10

15

7
American Institute of Aeronautics and Astronautics

are the
,

10

uEMC

-5

-10
-15

Figure 2. Plot of

-10

-5

and

10

(solid),

15

(dashed) and

(dotted).

For systems that need a high control effort even with small errors,
should be defined in the range
, or
otherwise. Systems with increasing levels of disturbance should alleviate the actuators
response by decreasing .
One issue that arises with this heuristic approach is for systems where instability happens before saturation. In
this case, knowledge about the system behavior before instability must be used. One example is the control of
aircraft altitude with the elevator as the sole actuator. In this case when full elevator is applied continuously, the
aircraft will eventually stall or even spin so the ratio between full elevator and altitude gain cannot be determined
directly. Song and Smith29 used the same approach for heading control of an AUV using only the vertical rudder.
With application of full rudder, the AUV will eventually reach a maximal turn speed. The resulting function is the
sliding, or switching, line.

V. Aircraft Control
Two aircraft, very different dynamically, were chosen for implementation of the EMC. Autopilots were built for
a widebody jet airliner and for a small single-engine aircraft. The reader is referred to Stevens and Lewis30 for a
thorough discussion of aircraft control.
The vector for the states of the aircraft is shown in (20). The control variables are shown in (21).
[

]
[

(20)
(21)

Two different flight control systems were implemented. One controls ascent and descent
and coupled
with a Stability Augmentation System (SAS) and the other is a cruise autopilot with speed hold, altitude hold and
roll angle hold also coupled with an SAS. The SAS for both control systems stabilize , , and . They will be
referenced as CAP, for Cruise Autopilot and IACAP, for Constant Indicated Airspeed Climb and Descent Autopilot,
respectively. Both flight control systems were implemented using a Linear Quadratic Tracker with Output
Feedback30 (LQTOF) and EMC.
The controllers were implemented using MATLAB and Simulink, an environment to perform generic matrixbased mathematical calculations and simulations. The aircraft were initially trimmed to a specific flight condition.
For the CAP trimming was performed to maintain constant ,
, , and relative position of the center of
gravity (C.G.) and solving for , , , , , and .
For the IACAP trimming was performed to maintain
, , and and solving for , , , , , and .
8
American Institute of Aeronautics and Astronautics

After trimming, disturbances were applied to check the performance of the control systems. Stochastic winds
were generated using the continuous Dryden wind turbulance model, available in the Aerospace Blockset in the
MATLAB environment and a wind shear was generated using a sinusoidal turbulence function for the horizontal
aspect and a semi-sinusoidal turbulence function for the vertical aspect. The wind shear model uses five parameters,
namely
,
, ,
and
. All aircraft were simulated with both disturbances and all responses were
satisfactory but for the sake of brevity not all results are shown. Fuel consumption is considered in the models.
The main characteristics and coefficients for the simulated aircraft, including mass, geometry, inertia and
propulsion properties are given in the Appendix. The adimensional force and moments coefficients are calculated
according to equations (22) to (27).
(22)
(23)
(24)
(
(

(25)

(26)

(27)

The propulsive force is calculated according to equation (28).


( )

( )

(28)

Finally, the loss of mass due to fuel consumption is calculated according to equation (29).

(29)

A. The Widebody Jet Airliner


A widebody jet airliner (comparable to an Airbus A310) was used to simulate a cruise autopilot. The actuators
for the jet are modeled using first order transfer functions, saturation and rate limiters. Their values and
configuration are available in the Appendix.
B. Widebody Jet Airliner LQTOF CAP
To hold and a longitudinal compensator was coupled with LQR gains and control inputs to
state-space compensator is shown in (30). The LQR gains are shown in (31).

[
A PID controller was used to control

through

and

. The

(30)

(31)

. Its parameters were:


(32)

9
American Institute of Aeronautics and Astronautics

and

The states , ,
.

and

were stabilized in the SAS inner loop using the LQR gains shown in (33) through

(33)

Figure 3 presents a diagram of the control architecture.


is a vector containing the reference states, which are
constant for the CAP but become the reference trajectory for the IACAP.

Figure 3. LQTOF Control Architecture


Results for the LQTOF controller are shown in Fig. 4, with Dryden wind and wind shear. The equilibrium points
for the simulation were
,
,
and
. The parameters for the wind shear were
,
,
,
and
, meaning that the aircraft reaches the
wind at approximately 10 seconds of flight.
Speed

230.5

H [m]

V [m/s]

230
229.5
229
228.5

20

40

60

8000
7995
7990

80

t [s]
Roll Angle

10

20

40

60

80

t [s]
Sideslip

[]

[]

5
0
-5
-10

Altitude

8005

-2
-4

20

40

60

-6

80

t [s]

20

40

60

t [s]

a)

10
American Institute of Aeronautics and Astronautics

80

Throttle

50

20

40
t [s]

60

20

40

60

80

t [s]
Rudder

4
2

r []

a []

-2
-4

80

0
-5
-10

Aileron

Elevator

p []

[%]

100

0
-2

20

40

60

-4

80

20

t [s]

40

60

80

t [s]

b)
Figure 4. LQTOF responses for the widebody jet, Dryden wind and wind shear. a) States. b) Actuators.
Results for the LQTOF controller with wind shear only are shown in Fig. 5.
Speed

230
229
228

20

40

60

8000
7995
7990

80

t [s]
Roll Angle

10

20

40

60

80

t [s]
Sideslip

[]

[]

5
0
-5
-10

Altitude

8005

H [m]

V [m/s]

231

-2
-4

20

40

60

-6

80

t [s]

20

40

60

t [s]

a)

11
American Institute of Aeronautics and Astronautics

80

Throttle

50

20

60

-2

20

40

60

80

t [s]
Rudder

4
2

r []

0
-5
-10

-4

80

Aileron

a []

40
t [s]

Elevator

p []

[%]

100

0
-2

20

40

60

-4

80

t [s]

20

40

60

80

t [s]

b)
Figure 5. LQTOF responses and actuators for the widebody jet, wind shear only. a) States. b) Actuators.
C. Widebody Jet Airliner EMC CAP
As in the classical controller,
was controlled through . , and were controlled through ,
and
were controlled through
and was controlled through . The corresponding signals were added for each
individual actuator.
Figure 6 presents the control architecture for the EMC CAP and the EMC array of controllers, where is the
error vector for each state and is the controller output vector, containing , ,
and , respectively. The Error
Calculation block computes the errors from the equilibrium states.

a)

12
American Institute of Aeronautics and Astronautics

EMC_V
e

V
EMC_H
e

EMC_phi

u
e

phi

EMC_theta
1
e

theta

EMC_q
e

EMC_p
e

EMC_r
e

b)
Figure 6. Simulink model for the EMC CAP. a) Control architecture. b) EMC array of controllers.
The parameters for the EMC CAP implementation are shown in Table 2.
Table 2 EMC parameters for the CAP
State
-2.5m/s

-1

-25

+25

35m

-25

+25

25

-25

+25

-25

+25

45/s

-0.5

-25

+25

-30/s

0.25

-25

+25

-150/s

-30

+30

Results for the EMC controller are shown in Fig. 7, with Dryden wind and wind shear. The equilibrium points
and wind shear parameters are the same as for the LQTOF controller.

13
American Institute of Aeronautics and Astronautics

Speed

230.2

H [m]

V [m/s]

230.1
230
229.9
229.8

Altitude

8005

20

40

60

8000
7995
7990

80

t [s]
Roll Angle

20

40

60

80

t [s]
Sideslip

[]

[]

1
0

0
-1

-5

20

40

60

-2

80

20

t [s]

40

60

80

t [s]

a)
Throttle

70

65

20

40
60
t [s]

-0.5
-1
-1.5

80

Aileron

10

r []

a []

0
-5
0

20

40

60

20
-3

10

-10

Elevator

p []

[%]

75

t [s]

60

80

t [s]
Rudder

5
0
-5

80

x 10

40

20

40

60

80

t [s]

b)
Figure 7. EMC responses for the widebody jet, Dryden wind and wind shear. a) States. b) Actuators.
Results for the EMC controller with wind shear only are shown in Fig. 8.

14
American Institute of Aeronautics and Astronautics

Speed

229.95

229.9

20

40

60

8000
7995
7990

80

t [s]
Roll Angle

20

40

60

80

t [s]
Sideslip

[]

[]

2
0

-1

-2
-4

Altitude

8005

H [m]

V [m/s]

230

20

40

60

-2

80

20

t [s]

40

60

80

t [s]

a)
Throttle

71

p []

[%]

70
69
68
67

20

-0.9
-1

80

r []
0

20

40

60

20
-3

-5

-0.8

Aileron

a []

40
60
t [s]

Elevator

-0.7

80

t [s]

x 10

40

60

80

t [s]
Rudder

2
0
-2

20

40

60

80

t [s]

b)
Figure 8. EMC responses for the widebody jet, wind shear only. a) States. b) Actuators.
D. Widebody Jet Airliner LQTOF IACAP
To hold during a climb or descent maneuver a longitudinal compensator was coupled with LQR gains and
control inputs to . The state-space compensator is shown in (34). The LQR gains are shown in (35).
[ ]

[ ]
[

[ ]

[ ]

(34)

A PID controller was used to control through . Its parameters were the same as for the CAP.
States , , and were stabilized in the SAS inner loop using the LQR gains shown in (36) through
.
15
American Institute of Aeronautics and Astronautics

(35)

and

(36)

Results for the LQTOF controller are shown in Fig. 9, with wind shear only. The equilibrium points for the
simulation were
,
,
,
and
. The parameters for the wind shear
were
,
,
,
and
, meaning that the aircraft reaches
the wind at approximately 23 seconds of flight.
Altitude

500

20

40

160
140
120

60

t [s]
Flight Path Angle

40

20

40

0
-20

5
0

20

40

-5

60

20

40

t [s]
Elevator

e []

0
-50

10

20

30

40

50

60

70

50

60

70

50

60

70

t [s]
Aileron

a []

60

t [s]

50

0
-5

10

20

30

40

t [s]
Rudder

r []

60

t [s]
Roll Angle

10

[]

[m]

15

20

-40

Indicated Airspeed

180

VIAS []

H [m/s]

1000

0
-5

10

20

30

40

t [s]

Figure 9. LQTOF responses for the IACAP for the Widebody Jet Airliner, wind shear only. a) States. b)
Actuators.

16
American Institute of Aeronautics and Astronautics

E. Widebody Jet Airliner EMC IACAP


As in the classical controller,
was controlled through . , and were controlled through ,
and
were controlled through
and was controlled through . The corresponding signals were added for each
individual actuator. In this case, the reference condition changes with time.
The IACAP EMC parameters are shown in Table 3.
Table 3 EMC parameters for the IACAP
State
8m/s

3.5

-25

+25

2.5m

-0.75

-25

+25

-25

-25

+25

15

-25

+25

20/s

-25

+25

30/s

-25

+25

10/s

-30

+30

Results for the EMC controller are shown in Fig. 10, with wind shear only. The equilibrium points and wind
shear parameters are the same as for the LQR controller.
Altitude

1000

Indicated Airspeed

180

VIAS []

H [m/s]

170
500

160
150

20

40

140

60

t [s]
Flight Path Angle

15

40

60

[]

[m]

20

t [s]
Roll Angle

10
5
0
-5

0
-1

20

40

-2

60

t [s]

20

40

t [s]

a)

17
American Institute of Aeronautics and Astronautics

60

Elevator

e []

20
0
-20

10

20

a []

10

40

50

60

40

50

60

40

50

60

0
-10

10

20

30

t [s]
Rudder

10

r []

30

t [s]
Aileron

0
-10

10

20

30

t [s]

Figure 10.

b)
EMC responses for the widebody jet, wind shear only. a) States. b) Actuators.

F. The Small Single-Engine Aircraft


A small single-engine aircraft (comparable to a North American Aviation Navion) was used to simulate a cruise
autopilot. The actuators for the aircraft are modeled using first order transfer functions, saturation and rate limiters.
Their values and configuration are available in the Appendix.
G. Small Single-Engine Aircraft EMC CAP
EMC was used to implement the CAP for a small single-engine aircraft. Its parameters (previously shown in
Table 2) and control architecture are the same as for the widebody jet airliner. The equilibrium points for the
simulation were
,
,
and
. The parameters for the wind shear were
,
,
,
and
, meaning that the aircraft reaches the wind at
approximately 43 seconds of flight.
It is worth noting that even though the dynamics of a widebody jet airliner and a small single-engine aircraft are
much different, the EMC parameters did not have to be changed. This presents some robustness that is still under
analysis.
Results for the EMC CAP controller are shown in Fig. 11, with Dryden wind and wind shear.

18
American Institute of Aeronautics and Astronautics

Speed

61

H [m]

V [m/s]

60
59

1000

58
57

50

980
960

100

t [s]
Roll Angle

50

100

t [s]
Sideslip

[]

[]

0.5
0

-2

-0.5
-1

Altitude

1020

50

-4

100

t [s]

50

100

t [s]

a)
Throttle

100

p []

[%]

90
80
70
60

50
t [s]

2
1
0

100

Aileron

Elevator

50

100

t [s]
Rudder

r []

a []

2
0

-2
-4

50

t [s]

Figure 11.

100

-5

50

100

t [s]

b)
EMC responses for the single-engine aircraft, Dryden, wind and wind shear. a) States. b)
Actuators.

Results for the EMC controller with wind shear only are shown in Fig. 12.

19
American Institute of Aeronautics and Astronautics

Speed

59
58
57

50

1000
980
960

100

t [s]
Roll Angle

0.6

50

100

t [s]
Sideslip

[]

[]

0.4
0.2

-2

0
-0.2

Altitude

1020

H [m]

V [m/s]

60

50

-4

100

t [s]

50

100

t [s]

a)
Throttle

90
80
70

50
t [s]

r []

a []

50

100

t [s]
Rudder

0
-0.5

0
-2

50

100

t [s]

Figure 12.

1.5

0.5

-1

100

Aileron

Elevator

2.5

p []

[%]

100

-4

50

100

t [s]

b)
EMC responses for the single-engine aircraft, wind shear only. a) States. b) Actuators.

H. Small Single-Engine Aircraft EMC IACAP


The EMC IACAP implementation uses the same parameters shown in Table 3. The architecture is the same as
previously shown.
Results for the EMC controllers are shown in Fig. 13, with Dryden wind and wind shear. The equilibrium points
for the simulation were
,
,
,
and
. The parameters for the wind
shear were
,
,
,
and
, meaning that the aircraft
reaches the wind at approximately 194 seconds of flight.

20
American Institute of Aeronautics and Astronautics

Altitude

1000

Indicated Airspeed

54

VIAS []

H [m/s]

52
500

50
48

100

200

46

300

t [s]
Flight Path Angle

0.6

200

300

[]

[m]

100

t [s]
Roll Angle

0.4
0.2

0
-0.2

100

200

-1

300

100

t [s]

200

300

t [s]

a)
Elevator

p []

50
0
-50

50

100

a []

200

250

300

250

300

250

300

t [s]
Aileron

10
0
-10

50

100

150

200

t [s]
Rudder

r []

150

0
-5

50

100

150

200

t [s]

Figure 13.

b)
EMC responses for the single-engine aircraft, Dryden wind and wind shear. a) States. b)
Actuators.

VI. Conclusion
An exponential-function based controller inspired by the bang-bang SMC and NFN was proposed. It seeks to
ease the task of implementing and maintaining a nonlinear control system. EMC exhibits the property of allowing its
tuning using only two parameters. The first ( ) is the most important one, since during the simulations it was seen
that its value can be critical in terms of stability, but a heuristic helps estimating its initial values. The second
parameter ( ) allows for specifying a smoother or a more aggressive response. Some of the implemented EMCs
showed good results with the first estimate of their parameters, obtained from the simulation. The classical
controllers required the model of each aircraft and each controller system used 43 parameters. EMC managed to
achieved roughly the same results without any analytical information about the aircraft and used 14 heuristicallydefined parameters. The authors relied only on the aircraft open loop response and behavior. Using Lyapunov to
check its stability properties, robustness characteristics and parameter range validation are intended for further
21
American Institute of Aeronautics and Astronautics

development of the proposed controller. The EMC has already been implemented in a real laboratory system,
namely the Quanser 3D Helicopter, with good results.

Appendix
Aircraft Main Characteristics
Parameter

Widebody Jet
Airliner

Single-engine
Aircraft

Length

46.66m

8.3m

Height

15.80m

2.6m

Maximum Takeoff
Weight

157,000kg

1,247kg

Cruise Speed

861km/h

250km/h

Aircraft Geometry, Mass, Inertia and Propulsion Properties


Variable

Widebody Jet

Single-engine

120,000kg

1,123.7kg

260m2

17.09m2

43.89m

10.18m

6.61m

1.74m

5.55e6 kg/m2

1415.4 kg/m2

9.72e6 kg/m2

3999.7 kg/m2

14.51e6 kg/m2

4756.7 kg/m2

-3.3e4 kg/m2

-142.4 kg/m2

240,000N

2,200 N

-1

0.75

0.75

240m/s

45m/s

1.225kg/m3

1.225 kg/m3

1.8e-5

6e-6

0.06

0.075

-0.045

22
American Institute of Aeronautics and Astronautics

Widebody Jet Airliner and Small Single-Engine Aircraft Actuators


Actuator

First Order Dynamics

Saturation Limits
[

Rising and Falling Rate Limits

Aircraft Coefficients
Variable

Widebody Jet

Single-engine

0.0175

0.0468

4.982

4.44

-1.4

3.8

0.435

0.353

-0.025

-30

-4.98

-10

-2.18

-1.46

-0.923

-1.5

-0.564

0.05

0.3

0.157

1.3

-0.074

-3.9157

-1.2012

0.8735

0.3135

-0.33

-0.134

0.25

0.107

1.75

0.071

-0.4518

-0.1685

23
American Institute of Aeronautics and Astronautics

-2.2591

-0.3662

-0.125

-0.0035

-1

-0.072

Actuator Dynamics Simulation Block

The reference values for the actuators are constant for the CAP. For the IACAP they are constantly updated,
except for the throttle, which is constant.
Aircraft and Wind Simulation Block

References
1

Slotine, Jean J. and Li, W. Applied Nonlinear Control. Prentice Hall, New Jersey, 1991.
Gdouin, P., Delaleau, E., Bourgeot J., Join, C., Chirani, S. A. and Calloch, S. Experimental comparison of classical PID and
model-free control: position control of a shape memory alloy active spring. Control Engineering Practice, 2011, no. 19, pp. 433441.
3
Yamamoto, S., Hashimoto, I. Present Status And Future Needs: The View From Japanese Industry. Chemical Process
Controls CPC IV: Proceedings of the Fourth International Conference on Chemical Process Control; Arkun, Y., Ray, W. H.,
Eds.; AIChE: New York, 1991.
4
Ward, S. Electrical Engineering. Global Media, UK, 2007.
5
Utkin, V. I. Guldner, J. and Shi, J. Sliding Mode Control in Electromechanical Systems. Taylor and Francis, London, 1999.
6
Utkin, V. I. Sliding Regimes and their Applications in Variable Structure Systems. MIR Publications, Moscow, 1978.
7
Song, F. and Smith, S. M. Design of Sliding Mode Fuzzy Controllers for an Autonomous Underwater Vehicle without System
Model. OCEAN 2000 MTS/IEEE Conference and Exhibition, 2000.
2

24
American Institute of Aeronautics and Astronautics

8
Yu, X., Kaynak, O. Sliding-Mode Control With Soft Computing: A Survey. IEEE Transactions on Industrial Electronics,
2009, Vol. 56, No. 9.
9
Filippov, A. F. Differential Equations with Discontinuous Right-hand Sides. Kluwer Academic, Dordrecht, 1988.
10
Zadeh, L. Fuzzy sets. Information and Control. Vol. 8, pp. 338-353, 1965.
11
Zadeh, L. Fuzzy logic and approximate reasoning. Synthese, 1975, vol. 30: 407-428.
12
Mamdani, E. H. Application of fuzzy algorithms for control of simple dynamic plant. In Proceedings IEEE vol. 121, no. 12,
pp. 1585-1588, 1974.
13
Yasunobu and Miyamoto, S. Automatic train operation system by predictive fuzzy control. In M. Sugeno, editor, Industrial
Applications of Fuzzy Control, pp. 1-18. North-Holland, 1985.
14
Hirota, K. Industrial Applications of Fuzzy Technology. Springer, Tokyo, 1993.
15
Terano, T., Asai, K. and M. Sugeno. Applied Fuzzy Systems. Boston: Academic Press, Inc., 1994.
16
Passino, K. M. and Yurkovich, S. Fuzzy Control. Addison-Wesley, US, pp. 444-446, 1998.
17
Sakthivel, G. , Anandhi, T. S. and Natarajan, S. P. Real Time Implementation Of A Fuzzy Logic Controller On FPGA
Using VGDL For DC Motor Speed Control. International Journal of Engineering Science and Technology, vol. 2(9), pp. 45114519, 2010.
18
Nauck, D., Klawonn, F. and Kruse, R. Combining Neural Networks and Fuzzy Controllers. In Proceedings, Fuzzy Logic
and Artificial Intelligence, Springer-Verlag, Berlin, 1993, pages 35-46.
19
Cybenko, G. Approximation by Superposition of a Sigmoidal Function. Mathematics Control Signals Systems, Vol. 2, pp.
303-314, 1989.
20
Yamakawa, T., Uchino, E., Miki T. and Kusanagi, H. A Neo Fuzzy Neuron and Its Applications to System Identification
and Prediction of the System Behavior. Proc. of the 2nd Int. Conf on Fuzzy Logic and Neural Networks, 1992, pp. 477-483.
21
Gouva, M. R. Induction Motor Neurofuzzy Control with Parameter Estimation and Estator Flow (Controle Neurofuzzy de
Motor de Induo com Estimao de Parmetros e Fluxo de Estator). PhD Dissertation, Universidade Federal de Minas Gerais,
Brazil, 2005.
22
Carvalho, Mateus A. Neurofuzzy Controller with On-Line Learning: Theory and Application in the Petroleum Industry.
(Controlador Neurofuzzy com Aprendizagem On-Line: Teoria e Aplicao na Indstria de Petrleo). MSc. Dissertation,
Universidade Federal de Minas Gerais, Brazil, 2010.
24
Yagis, N. and Hacioglu, Y. Fuzzy Sliding Modes with Moving Surface for the Robust Control of a Planar Robot. Journal of
Vibration and Control, 2005, Vol. 11 No. 7 pp. 903-922.
25
Tokat, S., Eksin I., Guzelkaya, M. Sliding Mode Control Using A Nonlinear Time-Varying Sliding Surface. Proceedings of
the 10th Mediterranean Conference on Control and Automation - MED2002, 2002, Lisbon, Portugal.
26
Choi, S. B. and Park, D.W. Moving sliding surfaces for fast tracking control of second-order dynamical systems. ASME
Journal of Dynamic Systems, Measurement, and Control, 1994, 116, pp. 154-158.
27
Park, D. and Choi, S. Moving sliding surfaces for high-order variable structure systems. International Journal of Control,
1999, 72:11, pp. 960-970.
28
Tabatabaei, E., Guez, A. and Choi, H. Adaptive Sigmoidal Molten Metal Pouring Control. IEEE Transactions on Control
Systems Technology, 1998, Vol. 6, No. 2.
29
Song, F. and Smith, S. M. Combine Sliding Mode Control and Fuzzy Logic Control for Autonomous Underwater Vehicles.
Advanced Fuzzy Logic Technologies in Industrial Applications, Advances in Industrial Control, 2006, pp. 191-205.
30
Stevens, B. L. and Lewis, F. L. Aircraft Control and Simulation, John Wiley & Sons Hoboken, NJ, 2003.

25
American Institute of Aeronautics and Astronautics

You might also like