You are on page 1of 16

The Aeronautical Journal 1

c Royal Aeronautical Society 2016


Page 1 of 16.
doi: 10.1017/aer.2016.1

Latency on a Stewart Platform


Using Washout Filter
A. Rodrigo Cristian Lemes, B. Mateus Moreira de Souza
Federal Institute of São Paulo
São Carlos
Brazil

C. Eduardo Morgado Belo, D. Jorge Henrique Bidinotto∗


jhbidi@sc.usp.br
São Carlos School of Engineering, University of São Paulo
Department of Aeronautics Engineering
São Carlos
Brazil

ABSTRACT
The aim of this work is to investigate and quantify the latency on a Stewart Platform caused
exclusively by a Classic Washout filter. This washout filter is intended to recreate the sensa-
tions of motion caused by changes of translational and rotational acceleration that an aircraft
can provide, due to changes in attitudes caused by external factors, and those caused by the
pilot’s command. The input signal was generated by a FlightGear Simulator in order to obtain
the specific forces and angular velocities of a Boeing 747 during a takeoff procedure. These
signals are then filtered by washout filter and sent to the inverse kinematics of the movable
platform, which will transform the aircraft motion sensations in platforms actuator position,
thereby causing a certain signal delay. Experiments were performed in a Stewart Platform to
obtain the latency caused by the mathematical modeling of the entire washout filter system,
and compares to the latency caused by the control and dynamics of the platform’s actuators.
Results indicate that the washout filter is the most responsible for the latency of the specific
force signals to be reproduced by the platform in this experiment, and that the natural fre-
quency and damping coefficient values must be properly estimated in order to optimize the
total latency.
∗ Corresponding Author

Received DD MM YYYY; revised DD MM YYYY; accepted DD MM YYYY.


2 The Aeronautical Journal

NOMENCLATURE
amax Maximum acceleration (mm/s2 )
DOF Degrees of Freedom
E pi Encoder position signal
F Center of the base coordinate system
fAx , fAy , fAz translational specific forces (m/s2 ) at the cockpit reference point for x, y, z axes
respectively
fAcc aircraft translational specific forces (m/s2 ) at the cockpit reference point
GPS Global positioning system
i Gear ratio
K p , Ki , Kd denote the coefficients for the proportional, integral and derivative terms re-
spectively
L1 , ..., L6 Latency of simulations
Li Inverse kinematic actuator length (mm)
LI S Rotation matrix that transforms vector components from the simulator refer-
ence frame to the inertial frame
M Movable platform coordinate system
PID Proportional-integral-derivative controller
PLC Programmable logic controller
S max Maximum actuator stroke (mm)
T MF Transformation matrix coordinates, from movable coordinate system to base
coordinate system
TS Transformation matrix from angular velocity to Euler angles rates
U Linear length per rotation (mm/U)
V Voltage
Vi Length vectors of the actuators (mm)
Vmax Maximum speed (mm/s)
W body-axis components of the angular velocities (rad/s)
W x , Wy , Wz angular velocities (rad/s) at the cockpit reference point for x, y, z axes respec-
tively
X, Y, Z linear translatory motions (surge, sway, heave)

Greek Symbol
d∆i Displacement of actuator (mm)
ζ Damping coefficient
ϕ, θ, ψ Motions of rotation (roll, pitch, yaw)
ωn Natural frequency

1.0 INTRODUCTION
The interest for Stewart platforms emerges from the demand to create flight simulators in-
creasingly realistic to be used for pilot and crew training and in the design and evaluation
of aircraft control and embedded systems (1) . The application of flight simulation in research
has increased considerably in the last decades, as hardware improvements have become avail-
able (2) . As this growth proceeds, there are many issues that are being discussed regarding
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 3

the use of simulation for experimental research and development. One of the main issues
surrounding human factors flight simulation research involves setting what level of simulator
fidelity is needed to provide acceptable experimental results (3) . The entire structure motion
of a flight simulator system can be understood in a way that, when the pilot responds to mo-
tion sensations and to tasks given by training, it represents the inputs for the dynamic model
of the aircraft which will have as output, the specific forces and angular velocity from the
aircraft center of gravity (CG). Then, due to the limitations of the motion mechanism, the
specific forces and angular velocity of the aircraft model is filtered by the washout filter. The
trajectories generated by the filter are the reference inputs of the motion controller. The con-
troller drives the actuators servo motors with appropriate control signals, and receives signals
from encoders as feedback signals (4) . So, understanding the latency behavior generated by
the washout filter becomes very important to ensure greater simulation fidelity.
Also, it is important to understand the visual system on this fidelity process. The visual sys-
tem provides a number of scenes for real-time images seen from the pilot eye position (5) . As
the aircraft maneuvers, the position and orientation of the pilot’s eye are computed in the mo-
tion equations and the scene is rendered every frame, typically 60 Hz (6) . Depending on the
imaging system, there is latency between obtaining a new eye position and the pilot seeing
the projected image. This latency, commonly related to as visual latency, should be kept to
a minimum but can extend to three or four frames (four frames at 50Hz is 80ms). A value
of 100 ms is commonly accepted as the worst admissible latency (depending on the aircraft
dynamics) and figures of 20 ms to 50 ms are more commonly quoted for current flight simu-
lators (6) .
It is obvious that a Stewart Platform, cannot generate the same forces on the human sense as
an aircraft flying, and understanding the human motion sensing system is essential to explain
the way which the human responds to motion. Also, it is important to identify the human
limitations sensing systems that enable forces to be simulated in a way that is indiscernible to
the pilot in a flight simulator (7) .
The vestibular human system consists of two sets of sensors, which measure the angular and
specific forces. Angular accelerations are sensed by the semicircular canals, organized in
three perpendicular plans (8) . At low angular frequencies, below 0.1 Hz, they measure accel-
eration, at high frequencies (above 5 Hz), they measure displacement and in the mid-range,
they measure velocity. This behavior can be considerably similar to a second order filter (6) .
While the semicircular canals feel angular acceleration, the otoliths sense specific forces.
The transfer function is almost the same to the semicircular canals. For combined motion of
angular and specific forces, the semicircular canal signals dominate with the otoliths provid-
ing measurements of predominantly linear motion (6) .
Flight simulators must be constructed using real-time systems. It means that the equations
are solved at a fixed frame rate, which must be sufficiently fast in a way that they do no
perceptible discontinuities in the simulation. We know that inevitable latency occurs in real-
time simulation, these latency are related to lags or latency and if they are enough large, they
can introduce many problems in real-time simulation. A simplified diagram of the primary
sources of latency in flight simulation is shown in Fig. 1.
Latency L1 emerge from the pilots response, that is, the time to acquire information, process
the information (cognition) and then activate the muscles to implement the desired control
function, typical pilot latency is the order of 300ms to 500ms (6) . Although the latency of a
pilot is likely to be the same in a simulator as it is in an aircraft, it may be affected by the
visual feedback cues, motion feedback cues and tactile cues provided in the simulator. L2
4 The Aeronautical Journal

Figure 1. Latency in Flight Simulation Platform.

occurs due to a finite time to acquire the inputs needed by the flight model and engine model,
which will include analogue to digital conversion sampling times.
The latency L3 is the time needed to compute the aircraft dynamics before the updated air-
craft position and attitude can be passed to the visual system. The latency of the real-time
connection between the serial port data dynamics and motion control ensures that the arrival
of this data is not significantly delayed. Testing of this connection during the commissioning
of a simulator indicated the average time delay between the sending of dynamic data and its
use by the washout filter to be less than 0.5ms (2) .
Finally, the visual system needs to complete its current frame, render the image for the next
frame and then project the image for the pilot. L4 can vary from one to four frames, depending
on the rendering methods used by the image generator. The total latency L5 is mainly caused
by phase error due to the motion drive algorithm which includes the third and second order
equations present in the filters models and the integration techniques. Rounding out the total
delay L5, delays caused by the motion drive algorithm, are added to the delays caused by the
data acquisition system and the computational cost of the inverse kinematics of the platform.
The control system and the dynamics of the actuators used are responsible for the L6 delay
caused in flight simulation platform.
In simulators with both visual and motion systems, it is particularly important that the latency
of the motion system not be greater than that the visual system, or symptoms of simulator
sickness may result (9) . This is because in the real world, motion cues are those of acceleration
and are quickly transmitted to the brain, typically in less than 50 milliseconds (8) ; These are
followed some milliseconds later by a perception of change in the visual scene. The visual
scene change is essentially the change of perspective and/or displacement of objects such as
the horizon, which takes some time to build up to discernible amounts after the initial ac-
celeration that causes the displacement. A simulator should therefore reflect the real-world
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 5

situation by ensuring that the motion latency is equal to or less than that of the visual system
and not the other way round.

2.0 THE EESC-USP STEWART PLATFORM


A Stewart platform, Fig. 2(a) was designed in the Flight Dynamics and Control Laboratory
of the Department of Aeronautics of the Engineering School of São Carlos (EESC - USP)
and it is composed of two platforms connected by six parallel linear actuators. The relative
position between the two platforms can be changed by extending or retracting the actuators.

(a)

(b) (c)
Figure 2. The Stewart platform of the Department of Aeronautics Flight Dynamics and Control Laboratory
EESC-USP (a); Power units and PLC incorporated into the drives (b); Acquisition, transmission, and
processing systems dSPACE (c).

Gough, the first person to develop and utilize this type of parallel structure, had previously
suggested a tire test machine (10) and Stewart adapted it for flight simulation platforms. This
type of structure allows movement of the movable platform in six degrees of freedom, the
three linear translatory motions X, Y, Z (surge, sway, heave), and the three rotational motions
6 The Aeronautical Journal

φ (roll), θ (pitch) and ψ (yaw) (11) .The axes adopted for the movable platform, correspond to
the axes usually adopted for the airplanes. On an aircraft, the frame X-axis typically points
out the nose, the Y-axis points out the right side of the fuselage, and the Z-axis points out the
bottom of the fuselage. It has six electromechanical actuators that are used to control the po-
sition and attitude of the movable platform. To measure the variation of the actuators lengths
encoders in the shaft of the actuators engines were used to measure the number of rotations of
the engine, and then a calibration curve was applied to obtain each actuator length variation.
The acquisition, transmission, and processing system dSPACE c , Fig. 2(c) was used to send
signal of voltage in the range of -10V to +10V to the power units and PLC incorporated
into the drives to control the motors, Fig. 2(b). For this platform, synchronous servomotors
(Bosch-Rexroth) were used that are composed of the drive, the electric motor and the feed-
back sensor installed in the electric motor housing. All this set of elements forms a closed
loop control which input is the current (or its equivalent voltage) that according to the drive
configuration may represent a desired torque or a desired speed.

3.0 THE WASHOUT FILTER


Motion cueing algorithm also called washout filter, transforms the motion from real vehicles
to simulators. The objective of the washout filter is to reproduce the angular rates and forces
that a pilot would feel if the simulation was real, using a 6 DOF Stewart Platform, which has
limited workspace (12) . To do this, the filters inputs are the specific forces and angular rates
that would be sensed by the pilot’s head, where the human vestibular system is located. Such
input is then reproduced, in the most realistic way, at the pilot’s head position by moving the
platform in which the pilot is seated. Its major function is to wash out unnecessary signals
and pull the position of, or move, the simulator back to its neutral position. Other functions
of the washout filter are designed to increase the efficiency of the platform workspace. The
senseless maneuver is developed to move the platform toward its original position beyond the
threshold of human perception (13) .

Classical washout filters Fig. 3, is composed of linear low-pass filters and high-pass filters
and is featured by its simplicity and ease of adjustment according to eqs.(1) to (5) (12) . The
inputs to the algorithm are the aircraft translational specific forces fAcc = [ fAx , fAy , fAz ]T and
angular velocities W = [W x , Wy , Wz ]T while the outputs are the simulator displacements lin-
ear translatory motions (surge, sway, heave) and the motions of rotations (roll, pitch, yaw).
Before filtering, the input channels signals are scaled and limited, then transformed to the
inertial reference frame by transformation matrices LIS and T S . Both matrices depend on the
actual state of the simulator and are continuously updated. After that, gravity is added to the
specific forces to create the inertial frame accelerations. The inertial accelerations are high-
pass filtered at the first section called High-Frequency Translation Motions (eq. 1) in order to
remove the large displacement low-frequency motions. Finally, twice integrating the filtered
accelerations, one obtains the desired translational displacements of the actuators.
The aircraft translational specific forces signals fAcc , are also sent to a second section called
Tilt-Coordination where are low-pass filtered in order to remove the large displacement high-
frequency motions and then subjected to tilt coordination to transform the translational mo-
tions into rotational motion (eq. 1). These signals are then limited to 3 deg/s to keep the tilt
rate and acceleration motions below the human perception thresholds, so that only the gravity
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 7

Figure 3. Classical washout algorithm.

component of the tilt angle can be detected (14) .


In a third section called High-Frequency Rotational Motions, the rotational motions signals
are first scaled, limited and transformed into Euler rates using the matrix TS and then are high-
pass filtered (eq. 3) to yield the high-frequency component of the simulator angular motion.
The values of natural frequency ωn and damping coefficient ζ can be determined by eqs.(4)
and (5), as well as from a set of values tuned by trial and error through experiments reported
by the flight simulator pilot relative to a set of maneuvers (15) .
 
s2 s
= 
 
BX,Y,Z × . . . (1)
s + 2ζ x,y,z ωnx,y,z s + ωnx,y,z s + ωnx,y,z
2 2


ω2nx,y ωnx,y 


 
= 

Btiltφ,θ × . . . (2)
s2 + 2ζ x,y ωnx,y s + ω2nx,y s + ωnx,y 


 
s2
=   φ, θ, ψ
 
Bφ,θ,ψ . . . (3)
s2 + 2ζφ,θ,ψ ωnφ,θ,ψ s + ω2nφ,θ,ψ

amax
ω2n = . . . (4)
Vmax

amax
2ζωn = . . . (5)
Vmax
8 The Aeronautical Journal

4.0 THE INVERSE KINEMATICS


Once defined the desired position and attitude of the Stewart platform from the dynamic and
washout filter, the length of the six actuators can be obtained utilizing the inverse kinematics of
the platform (16) . Joints of actuators are known for a given platform, and their relative positions
can be written in relation to the center of each platform in two coordinate systems. The base
platform coordinate system utilizes the center of the base platform F as origin, the xF-axis
pointing between joints with actuators 1 and 6, zF-axis is perpendicular to the platform plane,
and yF-axis completes the right hand rule. The movable platform coordinate system center
M and its axis xM, yM, and zM are defined in a similar way. The positions of joints of the
base fixed and movable platforms in its coordinate systems are represented in Eq. (6) and (7),
respectively (17) .
 T
[Fi ]F = Fi1 Fi2 0 , i = 1, 2, ..., 6 . . . (6)

 T
[Mi ] M = Mi1 Mi2 0 , i = 1, 2, ..., 6 . . . (7)
The transformation matrix [T MF ] to obtain coordinates from the movable coordinate system
to the base coordinate one can be obtained utilizing three rotations in sequence (18) . The first
rotation is applied in xM-axis until yM-axis is parallel to the base platform plane; this angle
of rotation φ is known as roll angle. Then it is applied a rotation in the yM-axis until the
movable platform is parallel to the base platform; the pitch angle θ is obtained. And the last
rotation is applied in the zM-axis generating the yaw angle ψ (17) . The transformation matrix
is show in Eq. (8). Where, C = Cosine and S = S ine.

 CψCθ CψS θS φ − CφS ψ CψCφS θ + S ψS φ 


 
[T MF ] =  CθS ψ CψCφ + S ψS θS φ CφS ψS θ − CψS φ  . . . (8)
 
−S θ CθS φ CθS φ
 

The position of the movable platform can be written in the base platform system as shown
in Eq. (9) and the length vectors Vi of the six actuators can be obtained utilizing Eq.(10).
n oT
{M}F = x y z . . . (9)

n oT
{Vi }F = Vi1 Vi2 Vi3 = {M} + [T MF ] ∗ {Mi } − {Fi }, i = 1, 2, ..., 6 . . . (10)
The actuators length Li is the module of the vector [Vi ] as shown in Eq. (11)
 0.5
Li = Vi12 Vi22 Vi32 , i = 1, 2, ..., 6 . . . (11)

5.0 EXPERIMENTAL SETUP


The experimental setup was implemented using Matlab/Simulink, Fig. 4. In order to in-
vestigate the latency on a Stewart Platform caused by the Washout filter, it is necessary to
reproduce the specific forces and angular velocities as input to the filter. These signals were
created using the FlightGear software by reproducing a Boeing 747-400 take-off procedure,
saved into Matlab/Simulink function block and converted from the center of gravity (CG) of
the aircraft to the cockpit pilot seat.These data of the specific forces and angular velocities
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 9

are then compared before the entry of the washout filter, with the specific forces and angular
velocity developed by the movable platform. These results can be seen from Figures 5 and
8. Then the signals are washout filtered and inserted in the inverse kinematics of the movable
platform, which will transform the correct motions of the actuators according to the motion
sensations created by the input data, see Fig. 4.

Figure 4. experimental Setup.

A PID controller was used with transfer function presented in eq. (18), and was applied
to all actuators. The controller takes as input the signal from the error between the desired
actuator length and real length (Encoder signals block, Fig. 4) and then defines the control
action that is the sum of proportional, derivative and integrative actions.The displacement
of each actuator (d∆i ) is measured by absolute encoder type sensors (E pi ). The ratio of the
encoder signals to the linear motion of the actuators is given by the eqs.(12) and (17) and
were obtained experimentally.

−834.986 + E p1
d∆1 = + 0.399 . . . (12)
25

−835.447 + E p2
d∆2 = + 0.417 . . . (13)
25.001

−833.397 + E p3
d∆3 = + 0.336 . . . (14)
25

−827.575 + E p4
d∆4 = + 0.106 . . . (15)
24.998

−833.615 + E p5
d∆5 = + 0.347 . . . (16)
24.998

−834.222 + E p6
d∆6 = + 0.368 . . . (17)
25.001
10 The Aeronautical Journal

A PID controller continuously calculates an error value e(t) as the difference between a
desired set point and a measured process variable and applies a correction based on propor-
tional, integral, and derivative terms. The controller attempts to minimize the error over time
by adjustment of a control variable u(t), such as the position of the actuators.
Z t
de(t)
u(t) = K p e(t) + Ki + e(τ)dτ + Kd . . . (18)
0 dt
To compare the input signals with the motions produced by the movable platform, it is
used an inertial measurement unit with an attitude and heading reference system processor
at the center of the movable platform. This measurement unit low-power signal processor
provides real time inertial enhanced 3-D position and velocity estimates. The inertial sensor
also provides drift-free, GPS enhanced, 3-D orientation estimates, as well as calibrated 3-D
accelerations, 3-D rates of turn, 3-D earth-magnetic field data and static pressure (barometer).
The dynamics of the actuators were defined according to the maximum payload and maximum
acceleration of 700 kg and 15 m/s2 respectively. Table 1 shows the dynamics of actuators used
for this platform.

Table 1
Dynamic actuator characteristics - Rexroth EMC 80-32
S max (mm) U (mm/U) Vmax (m/s) amax (m/s2 ) i
600 10 0.5 15 1

6.0 RESULTS
The experimental flight data of the simulation takeoff process was recorded and reproduced
using the system presented in Fig. 4. The recorded data has 50Hz acquisition and trans-
mission frequency. For the experimental results a Boeing 747-400 was used during takeoff
procedure using FlightGear open source software. The linear translatory motions X, Y, Z
(surge, sway, heave), and the three rotations φ, θ, ψ (roll, pitch, yaw) are transmitted to
the data acquisition board through the computer serial port. These data are converted from
aircraft CG to the pilot’s seat base, to ensure that the same inertial forces that a pilot would
feel inside the cockpit would be reproduced by the movable platform.
Fig. 5 shows the specific forces felt by the pilot on the X axis aircraft direction, according
to the take-off procedure already scaled and limited by the washout filter input (Fig. 3) and
compares its with the acceleration created by the movable platform. One also can observe the
total latency (L5+ L6) present in this experimental procedure. One may also notice the takeoff
procedure used. In the beginning, the engines are turned on and the aircraft has the brakes
activated, around 15s the brakes are deactivated and the aircraft starts running on the runway
until the takeoff, around the time of 56s. The aircraft remains for approximately 8 seconds at
takeoff angle, and then some unconventional disturbances are created during takeoff so that
it is possible to analyze and quantify the movable platform latency in reproducing aircraft
accelerations. The signals A1, B1, C1 and D1 are the accelerations developed by the aircraft
cockpit, while A2, B2, C2 and D2 are the accelerations reproduced by the movable platform.
The movable platform latency in reproducing the aircraft fAx acceleration are A2 = 440ms,
B2 = 240ms, C2 = 180ms and D2 = 260ms. The data show an average total latency (L5 + L6)
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 11

of about 280ms for the movable platform to reproduce the fAx , even at the points of greatest
variation of acceleration as shown in Fig. 5(a) and the errors presented by the Fig. 5(b). It is
also possible to say that the smaller the acceleration error, the smaller the signal delay.

(a)

(b)

Figure 5. Comparison of fAx created by the aircraft and reproduced by the movable platform (a); Experimental
acceleration error in fAx (b).

To find out the washout filter latency influence (L5) to reproduce the motion sensations
by the movable platform, it is necessary to know the dynamic and control delays of the
platform (L6) and subtract from the total latency (L5 + L6). Figs. 6(a) and 6(b) compares the
second actuator position indicated by the kinematics to the real actuator position and how
to obtain the latency (L6) caused by the dynamic system of the motion platform and control
system used. The average latency obtained by the author in several points along the time axis
indicated by Fig. 6(a) is about 40ms, with the maximum latency of 80 ms indicated by Fig.
6(b) during the moment of highest peak acceleration required. As a result of this, the average
delay L5 influenced by the Washout filter is about 240ms for the fAx .
Still in Fig.6(a), it may be noted close to the time of 40s, a small correction is applied by
the pilot to the rudder to correct an aircraft position on the runway, causing a rotation of the
aircraft’s Z axis and creating small disturbances that are reproduced by the movable platform.
12 The Aeronautical Journal

Fig. 7 shows the position error of the platform actuators for the entire take-off procedure,
where the efficiency of the control applied to each actuator can be noted.

(a)

(b)

Figure 6. Kinematics and actual second actuator position(a); Maximum latency indicated (b).

Figure 7. Platform actuator position error.


Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 13

Fig. 8 shows the Y-axis angular speed of the aircraft cockpit generated during the takeoff
procedure and the angular speed reproduced by the movable platform on the Y-axis. The
details of latency can be well observed in Fig. 8(b) which is the zoom at time C. Time A
had an average latency of 40 ms, 100ms for instant B, 60ms for instant C and 80ms for D.
According to the data, the average total latency (L5+L6) for Wy was 70ms and with an average
error below 0.8 degrees/s as shown on the Fig. 9. As a result of this, the average delay L5,
influenced by the Washout filter, is about 30ms for the Wy .

(a)

(b)

Figure 8. Angular velocity on the Y-axis of the aircraft cockpit during the entire take-off process and the
movable platform angular velocity of the Y-axis(a); Latency of the angular velocity in the Y-axis at time C (b).

Fig. 10 shows the pitch angle formed by the washout filter output which is composed
mainly of the input of specific force fAx and the angular velocity of Wy and compares with the
pitch angle formed by the movable platform. Signals A1, B1, C1 and D1 are the angular pitch
positions developed by the aircraft, while A2, B2, C2 and D2 are the angular pitch positions
reproduced by the movable platform. The latency observed by the platform in reproducing
the Pitch angle of the aircraft are A2 = 40ms, B2 = 60ms, C2 = 60ms and D2 = 40ms. These
data presents an average total latency (L5 + L6) of approximately 50 ms for the movable
platform to reproduce the pitch angles considerably.
14 The Aeronautical Journal

Figure 9. Experimental error of the angular velocity (Wy ).

Figure 10. The pitch angle formed by the washout filter output and the pitch angle formed by the movable
platform.

7.0 CONCLUSIONS
Many works in recent years have been produced to optimize these filter models within the
platform work area, without worrying about the latency that these models may cause to sim-
ulation quality. This results, seeks to draw attention to the importance of latency caused by
these filters models, in the simulation procedure of this kind of platform.
As the takeoff process flight data were recorded and are reproduced directly in the data ac-
quisition system, there is no signal delay between the operating hardware FlightGear and
hardware data acquisition system. Obviously the natural delays (L1, L2, and L4) shown in
Fig. 1, are already contained in the recorded flight data to be reproduced by the movable plat-
form. To know the dynamic and control latency platform (L6) and the Washout filter latency
(L5), the frequency acquisition and data transmission (L3) of 50 Hz is subtracted from the to-
tal latency (L5 + L6) experimental results data presented in Figs. 5 to 10. The results showed
that the parameterization used for the washout filter (eqs. 1 to 5), and the control system used
(eq. 12), was able to allow the movable platform to accurately reproduce the accelerations
and angular speeds from the Boeing 747 cockpit, during the take-off procedure performed in
the FlightGear software. With a total latency (L5 + L6) of 280 ms presented by the specific
force fAx , the washout filter was responsible for 86% of the total latency in this case and 43%
Lemes et al. Latency on a Stewart Platform Using Washout Filter . . . 15

for the case of the angular velocity in Wy . The greater latency presented by the specific force
fAcc can be explained by the characteristic that the washout filter presents, where the values
of the latency fAcc is composed by the sum of the latencies caused by the third-order low-pass
filter tilt coordinates channel with the second-order high-pass filter rotation motion channel,
thus forming the resulting Euler angles for the transformation matrix LIS that will produce
the inertial coordinates vector to the third-order high-pass translation motion channel.
or future work, it would be recommended to compare the latency L5 caused by different mo-
tion drive algorithms. It would also be interesting to develop, test, and apply better filters
of lower order equations to high and low pass filters, or apply mathematical models to the
motion drive algorithms to reduce the latency caused by L5, without causing greater losses in
reproducing the specific forces felt by the pilot in the simulator.

REFERENCES
1. Page, R.L. Brief History of Flight Simulation, 2000, The SimtechT Organizing and
Technical Committee.
2. Jamson, A.H.J. Motion cueing in driving simulators for research applications, 2010,
University of Leeds.
3. Rehmann, A.J. A Handbook of Flight Simulation Fidelity Requirements for Human Fac-
tors Research, 1995, Report, DOT/FAA/CT-TN95/46.
4. Lemes, R.C., Breganon R., Souza, M.M., Salvi, F.T.B., Angelico, R.A., Barbosa,
C.A.Z. and Belo, E.M. Implementation of a Washout Filter Used in Stewart Platform,
COBEM, December 2015, ABCM.
5. Buffett, A.R. Visual Cueing Requirements in Flight Simulation, Advances in Flight
Simulation Visual and Motion Systems, April 1986, The Royal Aeronautical Society.
6. Allerton, D. Principles of Flight Simulation, 2009, Aerospace Series: Wiley.
7. Lee, A.T. Flight Simulation: Virtual Environments in Aviation, 2005, Burlington: Ash-
gate Publishing Company.
8. Previc, F.H. and Ercoline, W.R. Spatial Disorientation in Aviation, Progress in Astro-
nautics and Aeronautics, 2004, AIAA.
9. Dasgupta, B. and Mruthyunjaya, T.S. The Stewart platform manipulator: A review,
Mechanism and Machine Theory, 2000, 35, (1), pp 15–40.
10. Gough, E. Contribution to discussion of papers on research in automobile stability,
control and tyre performance, Proc. Auto Div. Inst. Mech. Eng, 1956–1957, pp 392–
394.
11. Stewart, D. A Platform with six Degrees of Freedom, Proceedings of Institution of
Mechanical Engineers, 1965–1966,180, (15), pp 371–386.
12. Reid, L.D. and Nahon, M.A. Flight simulation motion-base drive algorithms: Part 1 -
Developing and testing the equations, Report no. 296, December 1985, UTIAS.
13. Chen, S.H. and Fu, L.F.C. An Optimal Washout Filter Design for a Motion Platform
with Senseless and Angular Scaling Maneuvers. American Control Conference, 2010.
14. Ko, S.F. Investigation of Simulator Motion Drive Algorithms for Airplane Upset Simu-
lation, 2012, University of Toronto.
16 The Aeronautical Journal

15. Nahon, M.A. and Reid, L.D. Simulator Motion Drive Algorithms A Designers Per-
spective. Journal of Guidance, Control, and Dynamics, 1990 13, (2), pp 356–362.
16. Nguyen, C.C., Antrazi S.S., Zhou, Z.L. and Campbell Jr, C.E. Adaptive Control of a
Stewart Platform-Based manipulator. Journal of Robotic Systems, 1993 10, (5), pp657–
687.
17. Souza, M.M., De Salvi, F.T.B., Carlos, S.R., Moreira, E.E.T., Schwening, G. S., Belo,
E.M., Breganon, R. and Lemes, R.C, A Fuzzy Technique to Control Attitude and Posi-
tion of a Stewart Platform. In: COBEM, 2013, ABCM.
18. Advani, S.K. The Kinematic Design of Flight Simulator Motion-Bases, PhD Thesis,
1998, Delft University of Technology.

You might also like