You are on page 1of 14

APPLI CAT1ON OF MONTE-CARL0 TECHNIQUES

TO M E 7571767 AUTOLAND DISPERSION ANALYSIS BY SIMULATION

A. Shakarian
Boeing C m e r c i a1 Airplane Canpany
. .
Seattle, Washington

Abstract
4) Provide l a r g e number (>2000) o f runs f o r
Reported i n t h i s paper i s the summary o f each l e v e l o f disturbance.
r e s u l t s and experience i n v o l v i n g Monte Carlo 5) Combine touchdown distributions for
simulations o f the 757 and 767 autoland d i f f e r e n t l e v e l s o f one disturbance
performance. Included are samples o f f 1i g h t according t o disturbance probabi 1it i e s
t e s t t o simulator c o r r e l a t i o n and model(s) 6) Combine dispersion d i s t r i b u t i o n s from
/statistical parameter descriptions. The d i f f e r e n t disturbances by " r o o t sum
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

s t a t i s t i c a l simulator r e s u l t s were used as an squaring"


i n t e g r a l p a r t o f the FAA and CAA autoland 7) Extrapolate p r o b a b i l i t y d i s t r i b u t i o n s o f
c e r t i f i c a t i o n process. each dispersion parameter

In summary, the piecewi se , incremental


Introduction approach previously used requires excessive
1i n e a r i z a t i o n of the models, ignores
The c e r t i f i c a t i o n o f autoland systems f o r s i g n i f i c a n t coupl i n g i n a i r p l a n e dynamics as
commercial j e t t r a n s p o r t h i s t o r i c a l l y requires w e l l as i n p u t disturbances and does not
some form o f simulator based studies t h a t can provide high confidence i n the predi c t a b i 1 it y
p r e d i c t the autoland performance i n r e a l is t i c o f the s t a t i s t i c a l r e s u l t s . A more h o l i s t i c ,
environments. The greater confidence p u t i n non-fragmented approach f o r t h e estimation/
simulated autolands increases the u t i l i t y o f p r e d i c t i o n o f autoland performance l i e s i n the
the simulator as a t o o l which provides Monte Carlo method.
reasonably accurate r e s u l t s t h a t influences
the number o f actual f l i g h t t e s t s required. The j u s t i f i c a t i o n o f the Monte Carlo approach
The objective for autoland performance f o r autoland performance evaluation can be
analysis i s t o show compliance t o established summarized as follows:
requirements s e t by the Federal A v i a t i o n 1) use o f s i x degree o f freedom d i g i t a l
Administration (FAA) and C i v i 1 A v i a t i o n simulation t o account f o r l o n g i t u d i n a l
A u t h o r i t y (CAA). The general requirements t o and l a t e r a l - d i r e c t i o n a l non-1 i n e a r
be met are s t a t i s t i c a l l y defined l e v e l s o f coupl ing
probabi 1it i e s o f occurrence o f hazardous and 2) simultaneous i n p u t o f a l l disturbances t o
non - hazardous events during approach f 1are, provide proper correlation between
touchdown and r o l l o u t . Simulated autoland disturbances and disturbance
statistics therefore must meet these distributions
operational and safety requirements f o r CAT 3) e l i m i n a t e post simulation s t a t i s t i c a l
I I I b autoland f o r f u l l FAA/CAA c e r t i f i c a t i o n . c a l c u l a t i o n f o r p r e d i c t i o n and estimation
The Monte Carlo technique was u t i l i z e d f o r purposes i . e . no r o o t sum squares, etc.
simulation purposes f o r p r e d i c t i o n o f the 4) reduce the t o t a l number o f simulations
757/767 autoland performance. hence c o s t and time allotments

Why Monte Carl o? I n conclusion, the Monte Carlo technique


provides a more reali s t i c simulated
Before discussing the p r i n c i p a l b e n e f i t s o f environment f o r the p r e d i c t i o n and estimation
the Monte Carlo technique a summary o f the o f dynamic autoland performance.
previous approach used f o r autoland simulator
c e r t i f i c a t i o n i s presented. The previous I. The Monte Carlo Technique
approach can be summarized as follows:
The f o l l o w i n g i s a b r i e f summary o f the Monte
1) Time scaled analog simulation Carlo statistical simulation technique
2) Separate linearized longitudinal and including computing and procedural
l a t e r a l - d i r e c t i o n a l simulation considerations. I t i s not intended t o be
3) Possible separate simulations f o r exhaustive but reflective of simulator
- headwind/tailwind experience related to 757/767 autoland
- headwi nd/tai l w i nd/shear performance analysis for FAA/CAA
- crosswind shear certification.
- l o n g i t u d i n a l turbulence
- v e r t i c a l turbulence
- glideslope noise
- l o c a l i z e r noise
Copyright @ American Institute of Aeronautics and
Astronautics. Ine.. 1983. All rights reserved.
Due t o a great number o f stochastic v a r i a t i o n s The Central L i m i t theorem applies d i r e c t l y t o
(inputs) and combinations, a large number o f l i n e a r systems through the convolution o f a
simulated autolands i s required t o establ i s h cascaded network. The output d i s t r i b u t i o n s
high confidence i n the r e s u l t a n t s t a t i s t i c a l (convolution) o f non-linear systems would n o t
d i s t r i b u t i o n s (outputs) given the 1i m i t a t i o n s necessarily be Gaussian b u t would r e f l e c t the
and approximations inherent i n the modeling response c h a r a c t e r i s t i c s o f the p a r t i c u l a r
process. The number o f autoland simulations system * e.g. autoland c o n t r o l law
required depends primarily on 1) the characteristics and resultant dynamic
s t a t i s t i c a l l y defined autoland performance performance. The shape o f the output
requirements by FAA/CAA; 2) the expected l e v e l d i s t r i b u t i o n i s also dependent on what
o f confidence and therefore predi c t a b i 1it y o f characteri s t i c o f t h e random v a r i a b l e i s
the simulated r e s u l t s , and 3) computing measured. The s t a t i s t i c s o f the peak values
time and cost considerations. Within the o f a random variable, f o r example, normally 7
framework o f the above considerations a y i e l d the Rayleigh d i s t r i b u t i o n .
qua1 it a t i v e / q u a n t i t a t i v e decision must be made
t o minimize t h e number o f simulated auto1 ands One o f the main concerns regarding Monte Carlo
without s a c r i f i c i n g the r e l i a b i l i t y o f the type simulations i s the number o f runs
estimation/prediction process. The f o l l o w i n g required t o acquire a s e t o f output
i s a b r i e f discussion o f the s t a t i s t i c a l distributions from which statistical
i m p l i c a t i o n s o f the Monte Carlo simulation inferences can be made w i t h high confidence.
technique. C l e a r l y i f c e r t a i n t y i n the s t a t i s t i c a l
r e s u l t s i s desired having p r o b a b i l i t y o f
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

The p r i n c i p a l feature o f a Monte Carlo exceedance requirements in the range


simulation i s t h a t N i n p u t random variables -6 -9
represented by p r o b a b i l i t y density functions 10 5PS10 , the number o f runs would be on
are u t i l i z e d t o generate the stochastic 6 9
environment. The probabi 1it y density the order 10 t o 10 . This can be shown by
functions are assumed t o be independent and p a r t i t i o n i n g a l l t e s t s o r output p r o b a b i l i t y
t h e i r parameter values randomly selected d i s t r i b u t i o n r e s u l t s as f a i l u r e s o r successes
using a uniformly d i s t r i b u t e d random number depending on whether o r not the output
generator. parameter value exceeds s p e c i f i e d exceedance
l e v e l s by the use o f the binomial d i s t r i b u t i o n
The Central L i m i t Theorem states t h a t i f there (Ref. 2) :
are N random variables XI.. . . . ,XN w i t h
p r o b a b i l i t y density functions

fl(X,) . . . . .fN(XN), the r e s u l t a n t


sum (Ref. 1)

where

r - number o f f a i l u r e s
(exceedances) i n n t e s t s
o r the convolution o f t h e i r respective density
functions b - average r a t e o f
occurrences ( f a i J ures) in n
F(X) = fl(X1) * f2(X2) * . . . - .*fN(XN) tests
p - probability of r
failures i n n tests
tend toward normal density (Gaussian) n - number o f t e s t s
(simulations)
functions under general conditions as N
increases. The normal density f u n c t i o n i s
defined as:
By set.ting r=O (no f a i l u r e s i n n t e s t s ) and
s o l v i n g f o r n:

log P
n =
log(1-b)
where p = mean , a = standard deviation.

* An example from autoland performance i n


touchdown response is the statistical
d i s t r i b u t i o n o f the s i n k r a t e which r e s u l t s i n
a log-normal type d i s t r i b u t i o n .
I f the p r o b a b i l i t y o f exceedance requirement d i s t r i b u t i o n as p r e d i c t e d by the Central L i m i t
-9 Theorem. The .Chi-squared P r o b a b i l i t y density
i s lo-', i.e. b=10 , and the p r o b a b i l i t y o f f u n c t i o n therefore can be used i n evaluating
having a f a i l u r e r a t e b i s t o be greater than the behavior o f the sample variance s2 w i t h
5%, i.e. P = 0.05 then: respect t o the actual variance u2 a t some
confidence level (1-a) e.g. for 95%
confidence, a = 0.05:

S i m i l a r l y f o r b = 10- 6 ,

Figures 1 and 2 show the 95% and 99%


l o g (0.05) 6 confidence boundaries f o r s/o derived from
n = =3x10
1 q (1-10'~) the above expression. The RMS convergence
r a t e dramatically decreases a t about ~ 1 0 0 0
i n d i c a t i n g a diminished e f f e c t on the sample
RMS c e r t a i n t y o r accuracy w i t h respect t o an
Obviously, the required number o f runs i s increased l e v e l i n the degrees o f freedom
p r o h i b i t i v e i n computing time * and cost.
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

As mentioned previously the decision regarding


a reasonable number o f simulations has t o have
a q u a l i t a t i v e as w e l l as a q u a n t i t a t i v e
component. A reasonable approach t h a t was
u t i l i z e d f o r the 757/767 autoland performance
analysis was t o observe the rate of
convergence o f the RMS value(s) o f key output
parameters such as l o n g i t u d i n a l and l a t e r a l
touchdown dispersions.

The t h e o r e t i c a l basis i n t h i s approach l i e s i n


the estimation o f the variance o f the sample
distribution+ . The d i s t r i b u t i o n o f the
variance y i e l d s the Chi-squared d i s t r i b u t i o n
(Ref. 3) given by:

I 7 nl 3 I
a =s
1. I 1.2 , .
RATIO OF TRUE W ( 0 ) TO S M P L E iPIS(S1
95% CONFIOENCE INTERVAL (CHI-SQUARE1 FOR 5
FIGURE 1

where I' (v/2) = dt


0
i s the Gamma f u n c t i o n i n which theargument i s
(v/2) and v i s the degrees o f freedom
(-samples). As n - O D .

fx2 ( v ) approaches a normal


V

* Average autoland simulation completion time


o f 5 minutes ( r e a l time) = 28.5
,/% 6
years/CPU f o r n=3x10 !

+ Not an unreasonable assumption due t o


Central L i m i t .Theorem and the approximately 99% CONFIOLNCE INTERVAL ( C H I - S W A R L ) FOR 5

Gaussian sample d i s t r i b u t i o n ( f o r +
20) o f the
FIGURE 2

key touchdown dispersion parameters.


(samples). The convergence o f the RMS t o i t s Uniform Density To General Density
expected f i n a l value* establishes the f a c t Transformation
t h a t w i t h an increasing sample s i z e the
probabi 1it y of encountering an extreme I n order t o generate any p r o b a b i l i t y density
parameter value we1 1 beyond the c u r r e n t f u n c t i o n from a sequence o f uniformly
minimum o r maximum i s remote assuming t h a t the d i s t r i b u t e d s e t o f pseudo-random numbers the
distribution is approximately normal f o l l o w i n g transformation i s made.
representing s t a b i l i t y o f system response t o
probable environmental and configuration I f the p r o b a b i l i t y d i s t r i b u t i o n o f the uniform
states. Figures 3 and 4 show actual RMS density f u n c t i o n i s given by
s t a t i s t i c s from the 757 autoland performance
simulation as a f u n c t i o n o f sample s i z e
(runs). I t can be seen t h a t a t approximately
1000-1200 runs the d e v i a t i o n o f the RMS values fx (X) = 0 elsewhere,
f o r both l o n g i t u d i n a l and l a t e r a l touchdown
dispersions are w i t h i n *2.5% o f and converging then the mapping o f the p r o b a b i l i t y element dP
t o the respective estimated * f i n a l values o f o f fx(x) t o any other density f u n c t i o n f ( y )
S, = 218 and S = 8.95. comparison o f these is: Y
Y
statistics to the Chi-square predicted
confidence/RMS convergence p l o t s shows t h a t dP => f x ( x ) d x = f (y)dy
Y
there e x i s t s a very high confidence (99% o r
greater) i n the accuracy o f the estimated RMS i n t e g r a t i n g the above we have
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

statistics. F i n a l l y , i t should be mentioned


t h a t the convergence o f the mean value
r e l a t i v e t o the RMS i s much more r a p i d and
t h e r e f o r e does n o t impact the simulation Ifx(x) b = fy(y) dy + C where
sample s i z e assessment procedure.

gives the transformatioh o f any p r o b a b i l i t y


density f u n c t i o n fv(y) w i t h random v a r i a b l e y
t o the uniformly i i s t r i b u t e d random v a r i a b l e
X.

A useful transformation i s from the uniform t o


the normal density function. This involves
the s o l u t i o n o f the i n t e g r a l :
7 5 7 AUTO1 AND MONTE CARLO S I M I I L A T I O N B S U l 1 5
rlGURF 3

which has no a n a l y t i c s o l u t i o n and i s commonly


r e f e r r e d t o as the error-function. I n order
t o d e r i v e an a n a l y t i c s o l u t i o n a mapping o f
the random v a r i a b l e y i s made from Cartesian
t o p o l a r coordinates i.e. y-->(r,8) where

XI , Xz are two independent normally -7


A- LONG LMOIWG
0 - SHORT LANDING d i s t r i b u t e d random variables. The d e r i v a t i o n
157 lUTOLfflD WOllTE U R L O S l W L A T l M RESULTS
FIWW 4
i s q u i t e involved and s h a l l not be presented
here, however, the f i n a l s o l u t i o n turns o u t t o
be :

* Second order least squares regression x1 = vziG cos e


x2 = K%i sin 0 where processing. Typical input
d i s t r i b u t i o n s used were 1) uniform, 2) Normal
probability

o r 3) s p e c i f i c d i s t r i b u t i o n s defined by t a b l e
u i s a uniformly d i s t r i b u t e d random variable. look-up functions and weighted uniform
distributions. The a i r p l a n e was i n i t i a l i z e d
and trimmed a t some l o c a l i z e r i n t e r c e p t angle
The above then generates two normally up t o 90 deg. w i t h an o f f s e t from the runway
d i s t r i b u t e d random variables w i t h mean y = 0 c e n t e r l i n e and a capture distance between 5
and standard deviation s = 1. and 10 miles from the runway threshold. The
For any uniformly d i s t r i b u t e d random v a r i a b l e i n i t i a l a i r p l a n e position/heading was randomly
we therefore have: chosen though bounded t o prevent missed
r- glideslope o r l o c a l i z e r capture conditions
Vi = ui(b-a) where caused by some unique combination of
i - 1,2 . . .n(No o f samples)
g l ideslope, a l t i t u d e and capture distance.
Randomly selected weight, airspeed, c.g. and
body moments o f inertia provided the
ui - uniform random number sequence, c o n f i g u r a t i o n a l variations.* Each autoland
simulation was executed w i t h multichannel
autopilot, autothrottle engaged and
a,b - d i s t r i b u t i o n lower and upper l i m i t s g e a r / f l a p s down. Autoland completion and
re- in i ti a1 i z a t i o n was made a f t e r the maximum
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

Vi - transformed uniform random number l a t e r a l d e v i a t i o n i n r o l l o u t was achieved.


S t a t i s t i c a l data was c o l l e c t e d on d i s k during
sequence s i x major events: 1) a t the beginning o f each
autoland ( i n i t i a l conditions) 2) d u r i n g
For any normally d i s t r i b u t e d random v a r i a b l e approach on the ILS between 700 t o 100 f e e t
we have: (Max g l ideslope, l o c a l i z e r and airspeed), 3)
a t f l a r e engage (@ an a1ti tude o f -50 ft.) , 4)
Yi = sX.I +y where a t t h e runway threshold crossing, 5) a t
touchdown, defined by e i t h e r main gear
i - 1,2 ...n(No o f samples) compression and 6) during r o l l o u t . In
Xi - normal standard d i s t r i b u t i o n p r i n c i p l e o f course, s t a t i s t i c s could be
compiled a t any time during the approach/flare
y=o, s=1. phase o f autoland.
p - mean value f o r y, s - standard
deviation f o r y
Yi - transformed Gaussian (normal )
I n p u t Random Variables
distribution
Table 1 l i s t s a l l the i n p u t random variables
Using the above transformations, the and type o f s t a t i s t i c a l d i s t r i b u t i o n s f o r the
probability densities of airplane 767 airplane. The only missing i n p u t random
c o n f i g u r a t i o n and environmental disturbances variables are the steady wind, shear, heading,
were randomly sampled t o generate the autoland and turbulence. The wind model as w e l l as the
stochastic environment. The next section ILS beam noise models s h a l l be described i n
describes i n some d e t a i l the Monte Carlo some d e t a i l l a t e r .
simulation o f the 757/767 autoland system.
Weight D i s t r i b u t i o n

The airplane gross weight d i s t r i b u t i o n i s


uniformly d i s t r i b u t e d w i t h a lower l i m i t s e t
11. A p p l i c a t i o n o f Monte Carlo - 757/767 a t empty weight and an upper l i m i t s e t a t the
Auto] and Performance maximum landing weight. Since a l l weights are
given equal value due t o the uniform
The f o l l o w i n g gives a b r i e f summary o f the d i s t r i b u t i o n , the s t a t i s t i c a l v a r i a t i o n s tend
757/767 Monte Carlo autoland performance t o give a more severe s e t o f weights f o r the
analysis simulated on the HARRIS computer Monte Carlo simulation. The value o f weight
system a t BOEING's Renton F l i g h t Simulator i s randomly picked a t the beginning o f each
Center. The analysis r e s u l t s were used i n the autoland run.
FAA/CAA autoland c e r t i f i c a t i o n e f f o r t .
Moments o f I n e r t i a and Loading Configurations
General Description
The moments o f i n e r t i a are c a l c u l a t e d as a
The 757/767 autoland performance analysis f u n c t i o n o f weight and various loading
simulation model was designed t o generate a configurations. For each weight condition,
r e a l i s t i c autoland environment composed o f the i n e r t i a s vary between a minimum and a
steady winds, shears and turbulence, ILS beam maximum value which depend on a p a r t i c u l a r
variations, bends and biases, runway loading scheme. For the purposes o f the Monte
c h a r a c t e r i s t i c s and a i r p l a n e c o n f i g u r a t i o n a l Carlo simulation, the p i t c h , r o l l , yaw and
variations. The stochastic environment was product of inertias are randomly
created by the random s e l e c t i o n o f i n p u t
--
parameter values w i t h predefined p r o b a b i l i t y
distributions. Statistical results were * Each random i n p u t v a r i a b l e d i s t r i b u t i o n
stored on d i s k f o r post run data p l o t t i n g and w i l l be described i n the next section.
selected w i t h a uniform random number Airspeed D i s t r i b u t i o n
generator t o have e i t h e r a minimum, nominal o r
a maximum value a t the randomly selected The i n i t i a l airspeed value i s uniformly
weight condition. The s e t o f i n e r t i a s remain d i s t r i b u t e d w i t h a lower l i m i t o f 1.3Vs and an
constant f o r the duration o f an autoland run. upper l i m i t o f 1.3Vs + 10 kts. The nominal
value 1.3Vs + 5 i s calculated as a f u n c t i o n o f
weight and f l a p s e t t i n g as follows:
1.3Vs = [ (2z:s4 w)14 where
INPUT TYPE OF
RANDOM VARIABLE DISTRIBUTION Vs = s t a l l airspeed, kts.
WEIGHT UNI FOH CL = l i f t C o e f f i c i e n t
MOMENTS OF INERTIA UNIFOH
S = wing surface area
CENTER OF GRAVITY UNIFOR4 W = operating weight l b s .
A1 RSPEED UNIFOR4
The selected i n i t i a l airspeed value i s s e t
FLAPS SPECIAL i n the TMS (Thrust Management System).
GLI DESLOPE SPECIAL
Glideslope D i s t r i b u t i o n
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

ILS REF DATUM GAUSSIAN


GLI DESLOPE XMTR GAUSS IAN The glideslope distribution is a
GAUSSIAN three-valued distribution with the
GLI DESLOPE
glideslope values o f 2.5 deg., 2.75 deg.
ALIGNMENT ERROR
and 3.00 deg. d i s t r i b u t e d t o represent
GLI OESLOPE RCVR GAUSSIAN t y p i c a l world wide f a c i l i t i e s (Ref. 4).
CENTERING ERROR The nominal value o f the glideslope i s held
f o r each run b u t i s modified by noise,
GLI DESLOPE BEAM GAUSSIAN
alignment and receiver errors.
NOISE
LOCALI ZER ALIGNMENT GAUSSIAN ILS Reference Datum
AT THE THRESHOLD
The ILS reference datum i s a Gaussian
LOCALI ZER RCVR GAUSSIAN d i s t r i b u t e d i n p u t derived from data found
CENTERING ERROR i n Ref. 4. The ILS reference datum i s a
LOCALIZER BEAM GAUSSIAN f u n c t i o n o f the glideslope t r a n s m i t t e r
NOISE l o c a t i o n GSTRW, and t h e glideslope GS.
I n a c t u a l i t y since GS has a unique
INTERCEPT ANGLE GAUSSIAN d i s t r i b u t i o n , then i s d i s t r i b u t e d as
CAPTURE DISTANCE
(LONGITUDINAL) UNI FORM GSTRW = D I L S / ~ ~ ~ - '(GS)
( LATERAL) UNIFORM
A t the beginning o f each autoland run a
RADIO ALTITUDE GAUSSIAN value o f GSTRW i s c a l c u l a t e d based on
ERROR the d i s t r i b u t i o n o f the ILS reference datum
ALTITUDE UNI FORM and the g l idesl ope.
RUNWAY SLOPE TABLE LOOK-UP Glideslope Alignment E r r o r
RUNWAY LENGTH TABLE LOOK-UP
The glideslope alignment e r r o r i s a
RUNWAY FRI CTI ON UNI FOW Gaussian d i s t r i b u t e d i n p u t based on I C A O
RUNWAY ALTITUDE TABLE LOOK-UP CAT I 1 s p e c i f i c a t i o n s . A t the beginning o f
each autoland run the value o f the
glideslope GS i s biased based on the
TABLE 1 INPUT RANDOM VARIABLES randomly picked alignment e r r o r ; i.e. GS =
GS + e r r o r . The value o f the glideslope
misalignment i s held constant u n t i l the
completion o f the autoland.

Glideslope Receiver Centering E r r o r


Center o f G r a v i t y D i s t r i b u t i o n bistribution

The a i r p l a n e center o f g r a v i t y v a r i a t i o n i s The g l ideslope receiver centering e r r o r i s


modeled as a uniform d i s t r i b u t i o n w i t h the a Gaussian d i s t r i b u t e d i n p u t on ICAO CAT I 1
lower and upper l i m i t s s e t a t the forward s p e c i f i c a t i o n s . The randomly picked value
aerodynamic l i m i t and the a f t aerodynamic i s added as a bias e r r o r on the nominal
l i m i t respectively. They are referenced t o glideslope value; i . e . , GS = GS + e r r o r .
the quarter chord (.25c). A value o f c.g. i s The receiver centering e r r o r i s held
randomly selected and remains constant f o r constant u n t i l t h e completion o f the
each autoland simulation run. autoland.
~ -
Glideslope and Localizer Beam Noise Models

The ILS beam noise models (see Reference 5)


for both g l ideslope and local izer
t r a n s m i t t e r s are composed o f two basic
capture conditions, the i n i t i a l l a t e r a l
o f f s e t i s l i m i t e d w i t h i n the boundaries
defined by the maximum l o c a l i z e r angle,
LOCAL,, the i n i t i a l value o f SXO and the
value o f SXLOC.
parts: 1) the amplitude d i s t r i b u t i o n and
2 ) the frequency spectrum. The amplitude
i s a normally d i s t r i b u t e d f u n c t i o n w i t h
zero mean and standard deviation Radar A l t i m e t e r E r r o r
proportional t o the RMS noise l e v e l . The
noise spectrum i s a f i r s t order l a g w i t h a The radar a l t i m e t e r e r r o r i s modeled as a
time constant . The glideslope and Gaussian d i s t r i b u t i o n . The r a d i o a l t i m e t e r
r l o c a l i z e r beam noise are independently and e r r o r i s updated every simulation i t e r a t i o n
randomly updated each frame u n t i 1 the and the random values selected based on the
completion o f the Monte Carlo runs. varying values o f the standard deviations
The beam noise RMS requirements are found d e f i n i n g the normal d i s t r i b u t i o n . This
i n I C A O Annex 10 and CAA paper 575 suggests e r r o r i s added t o the nominal value o f the
the model c h a r a c t e r i s t i c s . a l t i t u d e , h.

Localizer Alignment Runway S l ope

The l o c a l i z e r misalignment, i s modeled as a The a i r p o r t runway slope c h a r a c t e r i s t i c s


Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

Gaussian distribution at the runway data i s shown as a histogram i n Figure


threshold (Ref 4). The misalignment 5 f o r world-wide Category I 1 and I 1 1
l i n e a r l y approaches zero a t the l o c a l i z e r facilities. With the exception o f one
transmitter. At the beginning o f each a i r p o r t having a 1.2% slope runway, the
autoland run the value o f misalignment i s d i s t r i b u t i o n i s l i m i t e d t o ?0.8% slopes.
randomly picked and i s constant throughout The data i s normalized and stored as a
the run. t a b l e o f slope values. Runway slope values
are randomly selected a t the beginning o f
Localizer Receiver Centering E r r o r each autoland.

The l o c a l i z e r receiver centering e r r o r i s


modeled as a Gaussian d i s t r i b u t i o n (ICAO
Annex 10). The l o c a l i z e r i s o f f s e t by the U.S. OPERATIONAL
randomly picked e r r o r a t the beginning o f
each autoland run and held constant
throughout the run.

Localizer I n t e r c e p t Angle

The l o c a l i z e r i n t e r c e p t angle I A i s defined


by
I A = PSIGT - PSIR where

PSIGT = Ground t r a c k heading, deg.


( I n i t i a l condition)

~ PSIR = Runway heading, deg.

The value o f IA i s randomly picked a t the


beginning o f each autoland run based on a
-0.8 -0.4 0
WNWY SLOPE
0.4
- I
0.8 1.2

Gaussian d i s t r i b u t i o n w i t h a mean o f zero


and standard d e v i a t i o n o f 30 deg. The FIGURE 5 RUIMAY SLOPE AT YORLO-YIDE CATEGORY I 1 AND 111 FACILITIES
i n t e r c e p t angle i s l i m i t e d t o +
90°(+3a).

Capture Distance

Capture distance i s i n i t i a l i z e d by randomly Runway F r i c t i o n


p i c k i n g the i n i t i a l l o n g i t u d i n a l component
SXO and computing SYO, the i n i t i a l l a t e r a l Runway f r i c t i o n i s simulated by a uniformly
position. The value o f SXO i s randomly d i s t r i b u t e d f u n c t i o n t h a t selects the value
o f the f l a g LANCON. A dry runway occurs
chosen v i a a uniform d i s t r i b u t i o n described
by SXOmni = 5 miles t o SXOmax = 10 miles. when LANCON = 1 and an i c y runway when
LANCON = 4. Intermediate values o f LANCON
The uniformly d i s t r i b u t e d i n i t i a l l a t e r a l represent wet o r snowy conditions. The
value o f runway f r i c t i o n i s determined a t
o f f s e t SYO ,is a f u n c t i o n o f SXO, the
l o n g i t u d i n a l p o s i t i o n , SXLOC the l o c a l i z e r the beginning o f each run by the values o f
t r a n s m i t t e r l o c a t i o n , and the l o c a l i z e r SMUD (friction coefficient) and RWSN
(surface factor).
angle LOCAL. I n order t o prevent missed
Runway Length
The runway length i s shown i n Figure 6.
The p r o b a b i l i t i e s and values o f runway
length were compiled from a world-wide
a i r p o r t survey o f CAT I1 and I11 runways.
The runway length distribution is
approximately a Gaussian d i s t r i b u t i o n w i t h
a mean o f 9438 f t and a standard d e v i a t i o n
1373 ft. At the beginning o f each autoland
run, the runway length i s randomly
selected.
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

2000 4000 6000 8000 10000


AIRPORT ELEVATION - FEET

FIGUFf 7 UNYAV ELEVATIffl A T YORLD-YIDE CATEGORY I 1 bNO 1 1 1 FACILITIES

The wind turbulence used i n the Monte Carlo


analysis generates the vertical and
h o r i z o n t a l RMS magnitudes as follows:

UV = 0.1061 VZ0 , v e r t i c a l RMS


Runway A l t i t u d e turbulence

The runway a l t i t u d e d i s t r i b u t i o n i s shown


i n Figure 7. The d i s t r i b u t i o n represents
data compiled from worldwide CAT 11 and I11 uH = uVf(h) , h o r i z o n t a l RMS
facilities turbulence

Wind and Turbulence Model


The wind model used f o r s i m u l a t i o n consists
p r i n c i p a l l y o f 1) steady o r constant wind
component, 2) a logarithmic wind shear
component 3) turbulence component and 4)
heading. Wind magnitude and d i r e c t i o n were
considered t o be independent random
variables. The source o f the t o t a l wind
magnitude i s the reference wind magnitude
p r o b a b i l i t y d i s t r i b u t i o n shown i n Figure 8.
This wind magnitude, V i s referenced a t
20 ft. above the r u a a y a l t i t u d e . The
t o t a l wind magnitude Vw i s decomposed
i n t o i t s body a x i s components UW, VW and WW
namely the h o r i z o n t a l , l a t e r a l and v e r t i c a l
components. I n order t o simulate the wind
turbulence the mean wind must be calculated
as a f u n c t i o n of the mean reference wind
PROBABILITY O F EXCEEOANCE (
and a l t i t u d e . The wind model used i s CUMULATIVE PROBABII.ITY D I S T R I B U T I O N OF REFERENCE MEAN *!NO M f f i N I N L Y
defined i n Ref. 2. FIGURE B
Figure 9 shows the a l t i t u d e e f f e c t , f(h) on
the h o r i z o n t a l turbulence RMS magnitude.
The turbulence scales are generated as
follows:
Lv = ( 1000, h
h,
==- 1000 it
1000 f t ,v e r t i c a l turbulence
scale

3
L,, = ~ ~'( h)][ f , horizontal turbulence scale
/--
<
Turbulence above 1000 f t a l t i t u d e i s
isotropic. The turbulence magnitudes and
scales are continuously generated w i t h a
"white noise" source through f i l t e r s which
produce the proper spectral shapes.
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

hI = 1000 Feet

F I W E 10 VIND DIRECTION HISTOGRM. 24 U.S. AIRPORTS (PERCENT FREQUENCY


OF OCCURRENCE OF DIRECTIONS FRm YHlCH MIND 15 CMING)

111. Autoland Touchdown S t a t i s t i c s

The 767 Monte Carlo autoland simulation


touchdown performance i s presented i n
Figures 11-18. The histograms and
cumulative probabi 1i t y distributions
represent 1500 simulated autolands i n the
stochastic environment previously
discussed. T r i p l e channel a u t o p i l o t , a c t i v e
a u t o t h r o t t l e w i t h gears and f l a p s down were
employed. Figures 11 and 12 show the
s t a t i s t i c s f o r the l o n g i t u d i n a l touchdown
dispersion. The 1ongi t u d i n a l touchdown
s t a t i s t i c s i s referenced from the runway
threshold and includes v a r i a t i o n s i n the
glideslope t r a n s m i t t e r l o c a t i o n w i t h mean
o f 1075 ft. and RMS o f 9 1 f t . The
FIGURE 9 TURBULENCE ALTITUDE FUNTION d i s t r i b u t i o n i s approximately Gaussian to *
k20 b u t deviates s l i g h t l y a t the t a i l s .
Figures 13 and 14 show the l a t e r a l
touchdown dispersion. This d i s t r i b u t i o n i s
v6ry c l o s e l y Gaussian as seen i n Figure 14.
Wind Heading D i s t r i b u t i o n Based on FAA autoland certification
* criteria shown in Figure 19 the
Figure 10 shows the wind d i r e c t i o n 1ongi t u d i nal and 1a t e r a l touchdown
histogram o f the wind heading varying from dispersion s t a t i s t i c s are w e l l w i t h i n the
0' (head wind) back t o 360'. The values o f performance requirements as defined i n
wind heading versus the p r o b a b i l i t y o f Reference 6.
exceedance are stored i n a table. The
t a b l e look-up i s performed w i t h a uniform
distribution 0 , The independent
combination o f wind magnitude and heading
generate a l l possible t o t a l wind vectors.
* The t h e o r e t i c a l Gaussian curve i s shown
as the s t r a i g h t l i n e .
VlWC I 1 L O U G I l W I N n L TWCWOUN DISPERSION I F T )
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

Not a l l the output random variables however


have approximate normal d i s t r i b u t i o u s .
Examples o f such d i s t r i b u t i o n s are the peak
o r maximum/minimum value s t a t i s t i c s such as
maximum glideslope deviations on approach
and the touchdown s i n k r a t e d i s t r i b u t i o n .
The resulting distributions fit the
Rayleigh and log-normal distributions
respectively. Figures 15 and 16 show the
touchdown sink rate distributions.
Estimation o f performance i n extreme
conditions could be made by p l o t t i n g the
r e s u l t s on log-normal scale r a t h e r than
normal p r o b a b i l i t y scale. Figures 1 7 and
18 show the maximum l a t e r a l d e v i a t i o n i n
rollout. P r e d i c t i o n and estimation o f
autoland performance should be evaluated
w i t h f l i g h t t e s t data r e s u l t s t o confirm
and increase confidence i n the Monte Carlo
analysis and the airplane simulation model.
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

F l i g h t Test C o r r e l a t i o n

The t r u e t e s t o f any model i s i t s a b i l i t y


t o p r e d i c t accurately the behavior o f
the r e a l process. Good c o r r e l a t i o n between
system t e s t r e s u l t s and simulation i s
essential f o r acceptance o f the model as a
v i a b l e representation o f the system.

Table 2 shows 757 f l i g h t t e s t vs the


simulator test results for specific
autolands. Actual time h i s t o r y data were
entered i n t o the simulator. The a i r p l a n e
equations o f motion were perturbated by the
wind and ILS beam s t r u c t u r e w i t h t e r r a i n
p r o f i l e s used t o a l t e r the beam geometry.
Initial conditions including airplane
weight, c. g. , l o c a l i z e r i n t e r c e p t angle and
l o c a t i o n s were entered. The selected
airspeed was i n p u t t o the simulated
autothrottle.

Table 3 shows the s t a t i s t i c a l c o r r e l a t i o n


o f the simulator Monte Carlo r e s u l t s t o
approximately 100 f l i g h t t e s t autolands.
These numbers should be compared i n l i g h t
o f mismatches t h a t e x i s t between f 1i g h t
t e s t and simulator conditions tested
coupled w i t h configurational d i s t r i b u t i o n
differences. P r i n c i p a l focus therefore
should be on t h e mean and RMS values r a t h e r
the extremes (min. , max. ). Good
c o r r e l a t i o n e x i s t s f o r the mean values o f
l o n g i t u d i n a l touchdown (about 8%). The RMS
values o f 217 ft. f o r the simulator t o 186
ft. f o r f l i g h t t e s t shows a greater
Lr
d i f f e r e n c e (17%) b u t i s reasonable given
the above considerations. The l a r g e s t
discrepancy e x i s t s between the s i n k r a t e
values i n both the mean and RMS. Upon
i n v e s t i g a t i o n , other than possible modeling
1i m i t a t i o n s , two primary causes were found
which a f f e c t the simulater s i n k r a t e
response; 1) the simulator frame time,
which was nominally s e t a t 80 msec and 2)
ground e f f e c t s aero mismatches between the
- simulator and f l i g h t t e s t . Lower frame
PROBLRILITY OF E XCI time and readjusted ground e f f e c t s provided
167 l r O L A N D PERFOR
FIGUPS 18
an even b e t t e r c o r r e l a t i o n w i t h f l i g h t t e s t
results. This post Monte Carlo analysis
X-DS
I T:
FROM THRESHOLD
I y-DIsTy:
FROM CENTERLINE NOTES

1
FLIGHT TEST SIMULATOR (FLIGHT TEST FLIGHT TESl

3 CHANNELS

3 CHANNELS

3 CHANNELS
3 CHANNELS
3 CHANNELS

ADC FA1 L U R . MANUAL THROTTLES I N


R I G H T , A/T I N SIMULATION.

SFO28L / STRONG HEACWIND. ADC FAILURE. PITCH


EXCURSIONS W E TO G/S OVER-FLIGHTS. I
SF(e8L ( STRONG HEAOWINO. TEMPORARY PILOT
OVERRIDE OF A/T.
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

TABLE 2 757 FLIGHT TEST/SIMULATOR TEST CORRELATION

f l i g h t t e s t touchdowns i s q u i t e good i n
terms o f the shape o f the d i s t r i b u t i o n s .
HEAN -
O H I J ~
The e f f e c t s o f ground e f f e c t s mismatch can
SIMULATOR RESULTS ( 1 5 0 0 LANDINGS)
be seen i n the b i a s t h a t e x i s t s between
f l i g h t t e s t and simulator CPD1s. Figure 22
LONGITUOINAL TWCHWWN (FEET) shows the l a t e r a l touchdown dispersion
LATERAL TWCHOWN (FEET) CPD1s w i t h the runway c e n t e r l i n e a t zero
and l e f t and r i g h t o f the c e n t e r l i n e being
SINK RATE ( FTISEC)
negative and p o s i t i v e respectively. The
LATERAL MAX ROLLOUT (FEET) d i s t r i b u t i o n s have the same basic t r e n d and
f l i g h t t e s t deviations from predicted
r e f l e c t distortions i n the CPD due
FLIGHT TEST (61 - 1 0 0 LANDINGS)
principal l y to asymmetric crosswind
d i s t r i b u t i o n shear and turbulence l e v e l s
LONGITUDINAL T W C H W m (FEET) experienced during the autolands. Further
LATERAL TWCHWWN (FEET) f l i g h t t e s t s should show a b e t t e r match i n
the CPD1s due t o increased symmetry i n the
SINK RATE (FTISEC) f l i g h t t e s t l a t e r a l dispersion s t a t i s t i c s .
LATERAL MAX ROLLOUT (FEET) Figure 23 compares touchdown s i n k r a t e
CPDS. Figure 24 compares the CPDs f o r the
maximum l a t e r a l deviation i n r o l l o u t
TABLE 3 7 5 7 STATISTICAL CORRELATION measured r e l a t i v e t o the runway center1 ine.
The r e s u l t s compare favorably w i t h f l i g h t
t e s t results.

showed an increase i n the simulator


longitudinal touchdown mean and a
corresponding decrease i n the touchdown
s i n k rate.

F l i g h t t e s t vs Monte Carlo comparisons o f


cumulative p r o b a b i l i t y d i s t r i b u t i o n , (CPD)
( p r o b a b i l i t y o f exceedance) p l o t s are shown
i n Figures 20 t o 24. Flight test result
CPD's r e f l e c t sample sizes ranging from
61-100. Simulator CPD1s represent 1500
autolands. Figures 20 and 2 1 are t h e CPD1s
o f the l o n g i t u d i n a l touchdown dispersion
w i t h respect t o the runway threshold
( i n c l u d i n g glideslope t r a n s m i t t e r l o c a t i o n
v a r i a t i o n s ) and r e l a t i v e t o the g l ides1ope
transmitter. . Touchdowns beyond the
t r a n s m i t t e r are p o s i t i v e while those before
are ne a t i v e a s s h o w n i n Figure 21. The
c o r r e h e t w e e n simulator predicted and
F I W R 20 757 AUTOLAND FLIGHT TEST vr SIMJUTOR CORRELATION

192
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

757 AUTOUND F L l O I T TE-ST vr SIMULATOR CORRELATION


F l l E 24

CONCLUSIONS
The 757/767 autoland performance evaluation
v i a Monte Carlo simulation proved t o be a
success from the standpoint o f 1) computing
efficiency, 2) modeling viability
( s t a t i s t i c a l l y v a l i d procedure) and 3) good
f l i g h t t e s t correlation. The design/
development o f the simulation required the
monitoring, coordination and c o n f i g u r a t i o n
of technical information which was
generally i n f l u x . I t i s important t o
h i g h l i g h t possible p i t f a l l s which may a r i s e
during the model development process:
7 5 7 WTOLAllO FLIGHT TEST vr SIMULATOR CORRELATION - basic a i r p l a n e model updates
F l W C 22
and/or possible changes which
sometimes elude recognition i.e. ,
computing c o n f i g u r a t i o n c o n t r o l
1i m i t a t i o n s
- proper v e r i f ic a t i o d v a l id a t i on
procedures o f the e x i s t i n g models
i.e., preliminary s t a t i s t i c a l
evaluation techniques
- parameter sensitivities to
simulator frame time
- ground effects aerodynamic
adjustments
The above impact the e f f i c i e n c y and
c r e d i b i 1it y o f a Monte Carl o autoland
simulation. A constant e f f o r t was made t o
minimize the degree o f impact o f the above
as w e l l as other f a c t o r s which were n o t
controlled directly by the author.

757 PUTOUWD F L I O I T TEST vr SIMULATOR CORRELATION


F I W R L 23
t

REFERENCES I

1) Probabi 1it y , Random Variables


and Stochastic Processes, A. Papoulis,
McGraw H i l l , New York, N.Y. 1965.

2) Report No. FAA-RD-74-206, "Wind Models


f o r F l i g h t Simulator C e r t i f i c a t i o n
o f Landing and Approach Guidance and
Control Systems ,'I 1974, Schaeffer,
Barr and Gangsass.

3) Spectral Analysis, G.M. Jenkins and


D. G. Watts, Holden-Day San Francisco,
Ca. 1968.

4) D6-44635 (Boeing Doc.), Vol. I1


"Background Data f o r Automatic Landing
System Design Requirements and
Downloaded by Indian Institute of Technology on October 1, 2012 | http://arc.aiaa.org | DOI: 10.2514/6.1983-2193

Objectives", E. F. Weener.

5) British CAA Paper No. 575,


"Autopi l o t s " .

6) Department o f Transportation, FAA AC


No. 20-57A, "Automatic Landing
Systems".

You might also like