Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
AD-769 753
A FAMILY OF DISTRIBUTIONS TO DESCRIBEA MULTIPIOJECTILE ROUND
David Kennedy Hugus
Naval Postgraduate SchoolMonterey, California
September 1973
DISTRIBUTED BY:
National Technical Information ServiceU. S. DEPARTMENT OF COMMERCE5285 Port Royal Road, Springfield Va. 22151
StCsm REPORtCTn 4 P1101AENTA Mi7s POn G _0PJvfl /*1
FMOAT UU21-R GOV- ACCE$E10n !4L' 3. Ps-cila-rst ALG nt
34.~~~ TYME OFAEPORT & PEMO C0 -EAFnmilv of Distribution tttfescril zare. Texs
a tiuti~rojiectile RoE
wA~rd K. R-ucus PC~TATi EY MEN
2.Ppc-iGOItNIZA7TOn %"AE "DMO &ZZES W.UMBSASl_
Naval Po-stgaut Sco..-t
NvlPostg-raduate School [ o11A 17IMonterey California 93940[1.IIBRLASj
It. gCTrM* fl" AHGOFFICE NAME &N AOESUSmtf~Cmrlh OI t2. IL ECURITY CASS N e
Naval Postgraduate SchoolMionterey, California 93940 1e~qlaTI-tNR~GI~ _
W . WSTfIU~lO*C STATEMENT of V-14 SCSIEOJL
Approved for Punlic Rel:ease; Distribution Unliamited.
Mt WMtlrJ~T!OM StAThSEIT fta fte efrfrt Onfa ft Bt W. U d-Wurs'f LOn Ra.Pj&
S& SUPnEUETAVO--F
INFO2RMATION SERVICE
M2 61 .45 caliber- zistold ens ity s ho0t c-hi-squarLe goodness-of-f-itpolar comordanatesgamma31 distrih utio rbivarinate orm± dastri-bUt] _________
T hl i s tPaer is an analci- of th belhavlor -dp t shot-IGQ-6I1 winen fired fron the .45 - aliber itl(u9± 1 . estshot ammiinJion is a lt~';cieroulnd whic', v--nta nm 16 -small p-lsletS. Bo0th the tanard n. fleA barm' eJ
ro rA A 'a-n-flyV of- aast'ribrtf. i-ns selzztea r ac r arzetyretocC e tnB h atte o. fr:--1 essirom theXN3421G n atva~r -.
-ffs~~ -.3
------~m -- I- - -
Ri - - -Z
A Famly- Of Dcstrirr ___ '-orato Describe a utptjctl Ru nd
by
David Kennedy HugusCaptain, Uited States Am
3.5., Pr-,due Uiversity, 1-965
S7ubmittedq in orar nJal t _
reqa zeni for" F Z4i~e ofth
MASTER OF SCIErICE' TM OPERATIONSJ RESEARCU
Naval Postgradnate Schc-'o3Septemt~r 1973
Author _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Approval 'by: t c; P;?Jrnezis Advisor
~'flSecond Reader
7-7
/ Acatiejc Dear.
J-rF-w r __
4~ ~ -~p - - wp-
This paperv is an analvsis of the behavior of density
shot 9iH261 when I fire d from th-e .45 caliber pistci (Ml9l.1
Densitv shot. aiinunznton is a railiproiected roCund which
contains 16 sna-1l pellets. Both th&e standard rfe hrp
F a ~and a smocoth barrel were used to co-.lect data on the behavio~r
of the M1261 round. A tally of disribtins- is selected-
for each barrel type toC- model th*e scatter of pellets fromW
the 3M4261 round at wzrin ranges.
p
Sa
-~ n- -1
QN
TABE OF CONTSU--T
I N ITR D aION ---- -- -
"i. DATA COL*aCTIN .------- ------------- 8
III. SELCTI OF DISTPaBUT PBi O------------------
IV. ANALYSIS OF RIFLED BARREL DATA ---..---.- -- 14
A. INITITAL UT- ............... 1
A. FJBL RES ULTS----------- --------- 23
V. ANALY.1SIS Op SMOOrh BARLDT---- - -- I
Vt. PSC MVn.ME---------- --------- 3 2
APPSIDIX A List of Data .
:.PfI-DfX B Pazameter Ltimat ion by the Method ofM ents---- --. . . .. ..--------------- 43
APPENDIX C Center of zlass as a Function of Range-- -45
APPMD7X D ChAnge ii Distributien Due to Centering--48
APFNDIX S Simulation to Determine the CriticalValues for the Goodness-of-Fit Test- 51
ATPESDIX 7 Variation of Lwdda as a Fun0tionof Range --------........ .... . ---- -5-
LIST OF RFERENCES----------------..57
INITIAL DISIBU"IM ST ..--. --------------------------- 58
S- -
100M DI-1473 ------------------------------------------
3So4
tLIST OF II~~~~~~~~~ -c~i~o 4s jj~~ RL----------- ---- 9a I1. GENWIEJ OF 4ASS-IIX 3~--.~ -- 19
III SZ&LZPERCEINTIL-S FROM 7h SflliON~. .54
-tE
I.Partition. of the Plane f or -the Polar fisitribution-- -2
-=2. Partition of the Plane for the Bivariate 4.
51
ANNI=
- ~ ~ ~ ~ ~ ~ ~ ~ _V - te=- nC--=- , =~---~-~ 7-~=_
investiaatinq three different types of av=-nition for es
with the .45 calIiber pistol (M1A The three tves of1
a~nitIon beinzm consider-ed are te- standardhl W mI -rm ition,
a sgwetedprojiectile which bre-as into sixpee fe
leaving the barrel, and t-%e- aensitv zo X1261 whidch containcs
'16 small pa1liets. USMsSASA is also invuestigatina the ef fects
t-Nat a moth fra bae and dt tairr ued--'h
on each type of amni tin. tASASA is trvy na to deterulmne
if greater effectven-ess car b achieved with the 45S nSlst-
by chanzing the combination ort asm tion and narrel. This
paper will deal only withay theu. density shot a-4-10nt-Ir-. H. P.
White Laboratory, Beel A-ir, ma- land, was coztracted to
provide veloci tv and accura-- data forv the eval-iation of the
six ccmbinationts of' amuni tion and barrel types. Inp Fetr-rvz
1573. the laboratory w-Us a ~LnaJ I reor ictrtaiIdiflc the
data required [ Ief a.
rhe ~r po se of ti s Paper is t^ oc -men- an a t-al--s. of
that sgntof the data nrov d byie . P. Whi e !A-
tha o-_m h K16- m -- = ' .J -= -t as -i=-='
taken at three r-.j's 110 -, ao Ui.mters. -,CNNever. K
t ap-, vot anre Zu ter -- at --t
data did not .eitsel to anallwsis at rt ano - 'rc
T .y~v the analysis- wa_% condncta w ta h r. four o=Ao
6A
______ fi--=LL~-- - -=--- __ --
pea lt -ad-u h o C ---- J dismito--:Uto --o db
ud ffa -- cas siv4 3Qato to rdeem.= the r =
ives s~1 o tE2SAS ASowieaste us; teji aWrca
sh11ee abo t te 40' - Ti thi stxibutrincoii
use di fwr exa I e r== r-iseft usearion toof data
w _i - t Of a COMter Sim =3 M
Canfe ctive t o mcde t:-- t-ca=l-t± tim;=j~ e3- I e4
Ofso-.- ofar Beasn' sh pooeteduseiof thef ditribtions
in ~ther s gu1latian ofAS -i--= tha theotr
the dP--qr-tni darl-u'te to ano aas -O trat
raif ec analic i~tt ied- a-o ati e q ddsts
can n d toe adngle r sttr- ar The
utiliz er -- te sthe :2da t=vt-_ - ,escria d ::; ff the
~e aq~e -5 aszrsu!.eAS
fal ecf =- fw -z t~ .- ne tmat
4IX9 Wve 101ile ex41- 4-cN t30ad1 t- e -- M - -c= t__ I
S tXS' mse r tait- Am th ieoSta1: - a t iets S 1--rC
mzl t -cc -j r- 15 f-- r= -the wwl
czaV off mr1timnzn-1tec le r-- =b WEechpc
jecti- on eac of n tzx - I trt ISG -
a~erci wane rccn~zt in r~Z. -4a = - ~-~
t7=u thre -- r - usd- St=-' -tr mtWta -0=CM
Lav-S fket 5-0 w a elizrs thi stme--a-
co" taIh zx~mfcier-zejr A-amml onflJ
v -1 ramsme
l -ler rm w 4 a smn-
- di At~;tr! -7- kit Z, o= -t 7 A
R~a = -
III. SELECTION OF DISTRIBUTIONS
For the present analysis, the basis for selecting
distributions to model the scatter of pellet strikes from
the XM261 was a letter from the United States Army Material
Systems Analysis Agency (USAMSAA) (Ref. 2]. USAMSAA was
replying to a request from USASASA to analyze data obtained
from firing the XM261 round. The data analyzed by USAMSAA
was prepared by the Frankford Arsenal. As a result of their
analysis, USAMSAA suggested that the distribution of pellet
strikes from the XM261 round when fired from a rifled barrel
could be modeled by using a polar coordinate system with the
angle havir.g a uniform distribution, the radius vector having
a normal distribution, and the angle and radius vector being
independent. (The model based on these assumptions will
hereafter be called the Polar Model.) When the XM26i round
was fired from a smooth bore barrel, USAMSAA suggested that a
bivariate normal distribution with independent X and Y would
serve as an adequate model. This student felt that these two
models could serve as an adequate basis for an initial investi-
gation of the XM261 data provided by H. P. White Laboratory.
The plane had to be partitioned into regions so that a
chi-square goodness-of-fit test could be performed. The
objective for the partition was to allow the( maximum nurabar of
regions consistent with the requirement to have an expected
frequency of observation of pellets of approximately five
[Ref. 3 and 4]. In connection with testing the fit of the
10
-~~~~~ ~~ 'A-- - lW---=-~~
Polar Model, the plane was originally divided into 30 nearly
equally probable regions. Since the angle was asvumed to
have a uniform distribution, the plane was first divided into
=six regions by constructing six rays from the origin at 60
degree intervals. Then four circles were superimposed over
the rays to further divide the plane. Tha radii of the A
circles corresponded to the 20, 40, 60, and 80 percentiles
of the fitted distribution of the radius v*ector. This
resulted in a parti tion of the plane into 30 nearly equally
probable regions (see Figure 1), Since there were 160 data
points, the expected frequency for each of tb: regions was
5.33.
Wher tiesting the fit of the b --riate normal distribution,
the plane was divided into 36 rectangles. The mean and 1
standard deviation of the X variable were estimated usini
the maximum likelihood estimates. Then values for the 16.67,
33.33, 50r 66.67, and 83.33 percentiles of the X variable
were estimated using the estimates for the mean and standard
deviation. A vertical line was constructed on t/he X value
which corresponded to each ot these estimated percentiles.
Then using the same method horizontal lines were constructed
on the same sample percentiie. of the Y variable. The
resulting partition prrvided 36 nearly equally probable
rectangles (see Figure 2). The expected frequency based on
the 160 data points was 4.45. Although this was less than
the desired frequency of five, it was sufficient lRef.3).
1a 1A
I II I - -T* p-A rn w
'15
I 10ra 2
U7
FU I
I1
PA TiTi _- __F T-I PL , ' F-', --r:,,.
T I I t P!iI s i
54- 35 36
!9 20 21 22 23 24
1314 !516 17 18 ;
7 8 9I0 !i 2 -
F I
F- 3
rQW
IV. AN.LYSIS OF DATA FROM THE RIFLED BARREL
I- A. INITIAL " tSU-S
When fitting the bitaria t norm dist±ibution, four
paramzeters, the mean and standard deviation of both the
X and Y variables, had to be estimated. Since the X and Y
variables were assimed to be independent, there was no need
to estimate a correlation coefficient. Because there were
36 regions and four parameters estimated, this chi-square
test had 31 degrees of freedom. This yielded a critical
value of 45.0 for a fite percent level of significance and
a critical value of 41.5 for a ten perc2n Thvel of signifi-
cance. The test statistic had a value of 95.6 for the 10
meter data and a value of 93.8 for the 30 meter data.
ience, the hypothesis t-hat either set of data was distributed
according to a bivarate normal was rejected.
When fitting the Polar Model, only two par.eters, the
mean and standard deviation of the distribution of the
radius vector, were estimated. Because the Polar Model had
30 regions and two parameters were estimated, this chi-square
statistic had 27 degrees of freedom. The critical values for
five L-id ten percent levels of significance are 40.1 and
36.7, respectively. The test statistic for the 10 meter data
was 55.25, that for the 30 meter data was also 55.25. This
led to the rejection of the hypothesis that either set of AV
data was distributed according to the Polar Model.
al14A
Since both dLstributiuns had been reiectad as methods
for adequately describing the data, an additional distr ibu'-n
-war soucht. A histogram of the values for the radius vector
suggestod that the radius vector might be distri!huted -
according to a gam-ma distribution. Therefore, an additional
distribution was introduced. This distribution utilized the
polax coordinate description of the data withthe asmptions
that the. angle was uniformly distribut d between zero and
two pi, the radius vector was distributed according to a _
gamma distribution, and the angle and the radius vector
were independent. This distribution will be referred to as
the Polar Gama distribution. The plane was partitioned for
the chi-square goodness-of-fit test in the same way as for
the Polar Model. The values of the radii of the cireles
corresponded to the appiopriate percentiles of the gamma-
distribution of the radius vector.
If a gamma distribution was to be used to represent the .
dis-tribution of the radius vector-, the parameters would
have to be estim-ated from the data. The maximum liklihood
method of estimating the parameters has some appeal, but
it is cin barsome because it involves the gamma function.
Therefore, it was decided to use the method of moments to
estimate the gamma parameters. This decision was made in
spite of the fact that cAe method of moments produced a
biased estimate of the parameters. Appendix B contains the
r~ method of moments estimates and a justification that they
are biased. Even after the para:eters were estimatee some
J.15
SA
- -+
a
difficultr is experie.nced in estt:atinq the 20. 40. 60, Z
and 80 percntiles of a aeneral cam- dia stribution to use a
values for radii of the circles se-arating the different __
regions it the plane. However, if Che shape narameter
alpha is a positive integer, then a convenient relationship
exists between the gara and the Poisson distributions. I
Specifically, if X is distributed gai-na !G1,) and Y is
distributed Poisson { X ) then P(x<-_)=P (Y>X) [Ref. 5].
Therefore, the percentile values for the ganm-a distribution -
with an integer alpha value can be interpolated fran' a tableI
of cumulative values for the Poisson distribution.
One of the considerations for this nroblem was that
the distribution used to describe the data should lend
itself to easy generation by computer si-ulatiun. Hence,
it was decided that after the gaw-na para-reters were estimated iA
using the method of moments, the integers on either side of
the estimated alpha value would be useti inztead of the
estimated alpha value. This approach would permit the use
of the relationship between the gaam and the Poisson f Idistributions, Each data set would then yield tvo chi-snTaare
test statistics for each set of qamma parameters estimated:
One test would use the integer value anmediatelv below the j
estimated alpha value, the second would use the integer A
value immediately above the estimated alnha value. Both 3
tests would use the estimated lambd-a value. Z
Since fitting the polar gamma also requires the estimation
of two parameters and the plane had been partitioned into 30
-a6f
_ z- -- - - -- - -_- - ---_--_ - x w - - . - -
regions, the chi-quare goodhness-of-fit test fo- the Polar
C-amma had 27 degrees of freedom. Consecuently. the critical A
values for the test statistic were the a am as for the Polar
Mcdel. The values for the two test statisrcs generated
using the 10 meter data with kat rifled barrel were 62.0 and
4.l. TIhose for the 30 meter data with the rilEd barrel
were 58.6 and 52.6. Therefore, both of the null hvnotheses
that the two sets oct data were distributed according to the
Polar Gam-ma distribution were rejected. Although the
bivariate normal, Polar, and Polar Gamicwa distributions bad
been rejected as models for te data, the Polar Gamma
distribution had produced the lowest value for the test
statistic for both the 10 and 30 meter data. Tis indicated
that the Polar Gamma distribution might have sone potential i
t , flescribe the data.
Be-ause all the distributions had failed to produce
an accentable description of the data, a check was made
to see if the data contained anv identifiable variation
that had not been accountedfor. The physical character- =
istics of the method of data collection led to the belief
that such a variation might exist. The recoil and rifling 3
of the barrel were suspected of causina he center of mass 2
(sample mean) of a pellet pattern to drift up and to the
left as the range increased. The recoil was suspected of 0
causing the drift up. The mceMent -& the barrel of one
half of one degree is sufficient to cause th center of mass
to move more than 10 inches at 30 meters. Te riflina was
17
___~~~. iT -- ==- =
susnected 0 causing the drift to the left. Since the
left handed rifllna produces a cou'nterclockwise rotation
4Lof the projectilker t4LtJLJC=-i rwJzla -0 -5tto~
left whce. meeting air resistance. if the center of mass did
in fact drift fro. the adm noint and if the drift was ran-dom,
then tis variation in the location of the centers of mass
would introduce an uncontrolled variation ang the rounds.
This would be considered undesirable since the chi-sqruare
goodness-of-fit test assumes that all observations ishots)
come from the same distribution.
In order to determine whether the =ata would tend to
support or refute the hypothesis that i-he centers of mass
did drift as range increased, the center of mass of each
shot was calculated from the data at ranges of 10 and 30
meters. Each center of mass was based -n the 16 pellets
in one XC2ol r.and. An examination of the centers of mass
calculated in this fashion tended to support the hypothesis
that the center of mass drifted up and t'o the left as
range increased, For both the 10 and 30 meter data eight
out of ten I cordinates of the centers of mass were negativeand nine out of ten Y coordinates were positive.Fuand ineoutof en cordiate wee iie Fuarthermore,
a comparison of the centers of mass at 10 and 30 meters
for the sar -ound revealed that eight out of ten centers
of mass had drifted up and- to the left. it was also noted
that the actua! distance that the center of mnass moved
varied siuaificantly a-ng the rours. The calculated Cen-
ters of mass are pro_ided in table 1.
1 iI34
_ - - _ . -ST . . . .____ - - - -W-
TABLE I
Rou. 10 Meters 30 Meters1 (0.25,1.!4) (-0.46,2.25)
2 (-0 .10,Q -1S1 (-1.27,2.80)-t':e
3 (-0.41,2.88) (0.50,8.35)
4(0C,1.3&) (-3.03,2.83)55 (-l. 9 2 ,-1.3l) 1-3.76,-4.30) --
S (-0.43,2;33) (-8.43,3.92)
7 (-0.85,0.64) 1.45,2.15)
8 (-0.53,1L.48) (t9,.l
(i-2fl0,2.27) (-5.37,2.58)
10(a1.29,0.05) (-4.8 20 C33'
OverNall Mean (-0.52,1.161 (-2.82,2.36)
$ I
-- - --
Inaenl orcer te1 o'- ifmir
m the centers of mass, t~h c~ao .a .- s fo eah rl-- of
th A xa-dY coordinates &-1- that shot'2 cen- of s fram
the X and YF co-rd'-tates of each pellet0- inm- thaG-t grouPP. Ti
Ct~i ~ - - -EL ~~ ~ ~.-ae -- v a
center- of: iass at the oriuin.
A.inte-cesting pheomno miv occur wnien each- gcny- of
J4 '.et has a center ofl -mass at te ori-in. TV-e data may
rappear to hav.e an are-a orf relatively low pronhjai.Li;%t-y at the
orig in (i.e., a 'hole' may appear in -het data) . This may' be
Ftrue evenx though the original data may not have exhibiteua
this trait. A possible explanation for the 'no0-le' appearaing
Is t-hat when the centers of mass are not constrained to be
at the %erigin, the -area or ;L io pr-obability %r one sht ay
lie in an area of high probabiliJZty% for a different sh4ot.
The overlap o-f the differentVI patterns fir-r tra different
rokir my Prevent am 'ho1le' from being wbsern4-r. Hmoever,
when the centers org mass are all a'- the origtn,. the areas
of ilow prcbab'Ulty may all coincide to nnodue aoe'in
thedata.
Another phencaenon may oroduc a 'ho]0' in the data even
when th-e dist-ribution of the data does nolt ehibit an area
at low prc-babilitv. For- examrpe, fr one =tew paikrs of
numbers ft= a urd-form d-Listrilion an n-1o+tsA the i-33--tn'e
of each point fro the center of ma s of its par of po-i nts,
then he would obsFerve an ara of low pret- ailiy at the
-- - -
- - - -, ~ = - ~ - __ _ ___ __7
-OR Cal, a
r-e art F-aue a~ll m- ~ te '- --i iimnts ea- sin-m
st salv for iltsinnc isthics Prr-c-m- t -ntr O fleS adt
flt-IL IS occurringd tht zzp to- atth
'Ut-rke a - diefre orn tz cente ora f =- tiemsno
Alh ro- nc -ey odificatio of tne dat subsd convenent
toabif. .sana~sswvs4 e.lettsnec4 thi ro elre antouce thee
cener f m-, b- wold-o-es h -F cth -e~e
of mass kfcam &'-e aem. ponrsr Mit caiC r--j the cet __o as
Jv'-4 utlin towb cretr odti ed to mare th efmaiect vss ohe
aciofthes roun. rr-' *b- rbzi dosr1 iht P-=--,C
arrnlysis estiate o~flec the sffctte of thes abounth
of mas fljro the accnt Uon. h drift of teCenter of ss
worl navuoe to be rentoue an aCs.altooh
actins ofl thi-s tfLsh'nd Fave t doe so migtit. Tproduce
ero EcnS irtes of-c the Ce -~e ofms- a:--ta h -
Pro cdre to, a- otfrthe driftD--- on!cha-f thes '-ene of msast
arlxteo-noet in Aredj- PC - eaidlfenth
whenomm al is of shots abova. ther sa-m orin measurng
may t 4oft-gtre di allina cohaactertc of f mae scate
ij~21
-t f~i rc ntt zr za 1::ie Value Zra
L; sa ste --n - a setea
a nellet wi-ll strike vditin the sam distance of -t'he -e~nter
of mas- See axni =l M a rate ics o ftis
taIC t Since messur ing amte -S5e jaS does n~f
the distribution of t1 data~e a bivari ae norma± -_
disri uti n4s used, one- w-%- a susnset tr=- a e n a
S___ ri-=4, .. LOLOS .1.c-L
phenonono micht occu-.r wttn 0hrr "ons t-i t' tcp
en t e rin g th t e Pellets at th1-er -oespecti.ve etrof mass
may appear to increa~a Ue urobabiit-wr oEf sitrikin 0 Cse
to the origapn. mI
The~n fia prbempoduceci nv modifin the data '--nerfls
the degree to chich the data conf orms to the asl szti ens
recuirec for the -hsur oocs-ness-of-f it tl;est aferi
has been odlified. Even if the data b-ad been ix--enendent -
'A
iits orliinal form, it is inoortant to note- that after the
viata has been nmdified, the coordi-na tesoth es i-- One
shot are no lan1cer :ndevendenut. This is so because al I '!7
coor'dinates of the pellet Iocationc denapend on the coorda-tC5
of the Cen-r of mass. Since thue cez-ter Of _mss is a
function of althe data poilts- in- oe c-o'n al-I of the
pellet -crLd.ates in a singde sihot a-re statist canv
2~.
e v. -- ft
t_ te .ia 12e= crU t4th sa13as 7bleassa- - as the best c to
7. aa s c- -el -AU - r = Z' - _-~ a-
Set O l-d- 1y- -l t byac-eFac st o dta -a sujetm tot=e- ssOf a:- -
p__- o e th - -4. 1 -1-mitt tierattis-i m Sbr - h
or-rat test s-ta are odred fr= os-nzIlzc toIgLOJ.
tesoivesa ale was scetlete as the test: rtanistcero
-rnlt I Ae alue of Vsne 900 =~n 95 -tst stmsc wer
Sach se ofe data.e w y he cte t ci-i-cdbu-ines v-of fIev
pxa-Bcae offv r-=z- tn tetsttsicwste. Te
'dsvera woamyne nea~iles ofbm s statistic war
prted h e vrorates ofnte 9of mass nd 950- a- sietzss of th
used s+_rb t ioe r oa vales tolli te-- th e ctests-imte Us -
da.t e 'LOd and ~-0eter -vt tetdsriuinwth2lo
-- gmate d~-'iaere used e ve-n at-wsto warna ostaat f-or
Z3
- g
IM met t-k .-L-s- -M-4 C-i Dcx t e pi:
E, - 4C-
sirtv-- the~ v -l les - h
ati= _ wra-- 3 a=Q44- f-c- I&A= .O i d ~th - -ai andth
A smia etr emr wc- n. t- 31 - e aa a
~~zua1t aa equal- Si5c andtiS -~
the~~~ Pola r3=:= iitn e1 R+=-
tenua to 0.-7 rened a:t- sto-nr-i of -3
troa £o -1.r 7 G M est conEsti o' 2be 12ted -tas-5
AuMctv scflrnatepr trg- whereun ZOM- 3CRe dt
distribution wa!s;ch has4 ?hzt Ror ethe Po-Iar the wasi 3P.
the:n- Polar_ Gcedsriuteion wih-~h- 10b eMi 15oth
PCI-p e e to M
I
distribution. For the 30 meter data the Polar Gamma distri-
bution with alpha parameter equal to 16 would be preferred
above all other distributions. However, when choosing
between the t-wo Polar Gamma distribution for the 10 meter
data, an additional consideration should be included. It
seemed reasonable that the distribution selected to model
the scatter of pellets at 10 mtzters should resemble the 30
meter distribution as closely as possible4 The addition of
the consideration favors the selection of the Polar Gamma
distribution with alpha equal to 16, even though the test
statistic for the other Polar Gamma distribution is sliqhtly
smaller. Therefore, this student would select the Polar
Gamma distribution with alpha equal to 16 as a model for the
scatter of pellets fired from a rifled barrel. Additional
justification for preferring the Polar Gamma distribution
over the Polar distribution is found in the characteristics
of the different distribution. Since the Gamma distribution
is defined only for positive values, using a Gamma distri- J-. 41bution to mode] the distribution of values of the radius
vrector is appealing. The normal distribution, on the other
hand, -is urbounded in both the positive and negative direc-
tions. This always allows the theoretical possibility of
having a negative value for the radius vector. if tle mean U
of the distribution of the radius vectcr is not large with
rcspect tc the standard deviation, using a normal di--:ribu-
tion could produce a significant nuiber of negative values.
25
-~ WE-
Moreover, the histogram of data values for the radius
vector appeared to.resemble a gam.. diaribution more than
a normal distribution. Finally, the gamma distribution is
easily modified to allow for variation in ranges. Only
the scale parameter lambda need be modified to reflect a
A change in range. Possible procedures to acount for the
change in lambda as the result of a change in range are
discussed in appendix F.
NI0
NI
-NN I
inn
26
'-_ s'o = .t
__--_g - ==
l£
V. ANALYSIS OF SMOOTH,1 BAR-EL DATA
The analysis of the smooth barrel data was sim.ilar to
thr analysis of the rifled barrel data. ThMe paramfters for
t'N. Garaa distribution were estimated using the wethcd of
moments. The three distributions, bivariate normal, Polar,
and clar Gamma, were then used to ieteraine if they couldaequately explain the data. All three failed to provide an
acceptably siaall value tor the test statistic. For example,
the 10 meter data- produced test statistics of 145.5, 291-0,
359.2, and 24S.5 for the bivariate norma±, polar, and the
-two versions of the Polar Ga-.maa distributions, respectively.
Following the technique of the previous section, the
centers of mass- w*ere cal.ulated usinc the data po-ints in
each round. These calculated centers of mass were then
examined. As with the rifled barrel, the smooth bore data
centers exhibited a random drift. However, the characteris-
tics of the drift v4re slighLzy different. Since the barrel
was no longer rifled and since the rifling was suspected of
producing the drift toward the left, on- would expect the
smooth barrel to fail to consistently produce a movement to S
the left. ..e smooth barrel data was, in fact, free frcm.q
this tendency. Essentially half cf the centers of mass
drifted to the left and half to the right. Th'e.--re ccntiznued
to be a tendency for the centers of mass Co rise as V,-&
range increased. As a matter of fact, all ten centers of 4
-Mass rose as the range increased. In order to remove this
_ 27
V= random rift, the data was again modified by subtracting the
coordinates of tte center of mass from the coordinates of
each impact poin: in that round. The centers of mass for
the smooth barrel data are provided in Table 2.
However, evm wieh the data -.o'Jifi ed in this way, the
values of the tast statistic were unacceptati-y !arge. Onlythe Polar Gaz distrbutior produced a test statistic less
than 100. WAile this stuAent was preparing a histogram of
vaues for thc:dius "vector, he noticei that rounds 6, 7, Iand B proance! -an inrax-dinate nrnter of large values fer the
radiua vector. This caused him to suspect hat these ror.As
might be classified as outliers. ::onsequfi(ntly, he calcula-
ted the sam,:le variance for the tree rounds in questxor and
the sam.le variance for the remainings seven rounds. Whe2.
the ratio of these sample variances was calculated, an F
statistic was produced. The F stetistic for tue 10 meter
data was 21.9, for the 30 meter data 3A0.3. The critical
value fcr a]ee ofsis .fiaen.oe of five percent and
degrees of freedomh 46,112 is approximately 1.55. Since the
test stttistics were large with respect to the critical
value, this student classified rounds 6, 7, and P ogtliers
and remirved them frcr. tle data set. This variation among
the rotunds may have been produced by variation in the nanu-
facture or performance of the ammunition. In any e%.ent,
there pat-IS to be sufficient reason to desire a repeat of
this exp':riment to see if this variation is c'bserved again. 4
-4 V-
TABLE Ii
CENTERS OF MhSS-SIM Or , BAR&REL
Round 10 Meters 30 Meters "
1 (-0.16,2.2-1 (1.11,3.85)
2 (0.303,3.431 01,.5
3 (-0.98,2.40) {-2.65,38-66). I_ < °]
(-1.73,2.581 (-349,8.35)5(-0.30,2.53) .&-1.23,6.21I) ]
9 (0 .11i, 356) (i. 4-0, 0.73)-1-
I. 2
B9 61II
IiNIIN
r29ma-_---- - - -- -f.-~
If 30 regions had been used for the goodness-of-fit test
after the data se had been reduced by the rcluion of the
three ou-ling rounds, this would have resulted in an expected
- frequency of only 3.73, much less than the desired f"ve.
Consequently, Cm nwnner of regions used when ttng the fi t
of the Polar ad- Polar Gaj-a distributions was reauged to 24
biy superimosing only t-ree cxrcles over rihe = r5.--j frmf the
origin. The radii corresponded to the estimated 25, 50, and
75 perce-Mtiles of the distribution of the radiuc vector. The
use nif 24 regions raised the exoected frequency to 4.67 A
which is acceptable [Ref. 31 .
After e11--ina the atypical rounds, reducina the number
of regions in the partition of the plane, and reestimating
the ga parameers based on the re--ining seen rounds,
the three 'is -ibutions -were tested for -fit. Te bivariate
no_-al aLd Foiar distributions resulted in relatively large
test statistic all were over 30. The Polar Gamna nroduced
a test statistL PF 23. for parx eter values of aipha itA
equal to 4 and Lamxa equal to '4, using the 10 ter data.
r The :LOWestE.-t statistic for the 30 me ata was 2.
; obtaird t he Polar Gamma distribution with parameter
estimates of alpha eraa to 4 and il ,t ezua i to 0.5-352-
Since u. ooth barrel data had been modified r. the same
manner as the rifled barrel data, the same problems were enconz-
Stered. Since th e P ar Ga: distributon with an aipha value
of 4 produced thr 1-est value of tn - test statitict for
both sets of data, it was recomended to model the behavior orf
tc smooth barrel dat-a.30
V aam
=.- wa Che caser~ = wih h ife. aa h ai
distibutionwas ---eer-ed Orerthe ~olr dzst-lbuton l-e1aws
was th e a e wt h ilddaa i od
bii
Of ass Or e-nc ru*.5Tw eNDam-t S 7U~ -m.eaaSold- est iJaaIsdf = h nat SU
i lTe drift of the center- mass as afucinoraw
h e c -- ntzo~duceo in-o simulationls ofi t -- Se ro=ns. hc
en-nd n the Ont - - a- ____l
fmcn -nro te d-e
aJA4-omldata- shoul bn taken- to- allow a recX.rae
investiga-tion of t-he Ai str-ibution c-f the center mass Orf
a single rouand'a a functaon of -rnge. Aditional data
a should &l c be taken to aliro a car zete investicat-ion
cf the variation of the scale Parameter iat *da a function
of range.
ACT
2S
APPENDIX A
A-CCURACY DATA
S hBoot Bare rre.1
Shot iact at 10 Yues 'rct at 30 mers
a -0.1 +0.4 +32-2be -C--4 -0.5 - '1 -c l... 4-0. 6 4i -1.7 .9 -. 41.7
e -0.8 1.- +5.8 +2.2_ f -0.4 i3+292.
g 4~ 1~ +1.4+-0.3 -2.3 +0.4 +3.7-1.0 42.1 -2. +4.7
j-0. As+2.8 - +-4,8k+2.4 A- .74
2.0 +- -2.± '7.5-+1.7*~ 1. +7.7n-3.9 +3. • 94.9
OF 04 .9 -1-2 -0.5c -. 1.. . . -0.
2a.-1 --- 3,-. +.7 --
d -0.6 +1 0. +J0.e -0.9 +2.6 +2-3 +2.9
C' A -4 5,
--U . 9 2.9 +0.: +5.4A- 5 - A + -. 4I
A1 ~~--..4. 74
-5 7 - -+1
Se2unce cf sub-oectiles list- at- 0 m-ees is m t
n-e cePs sariy i e nticzalI w i th t e A s -me n Ce -Iis - - d a t I u .- ters.
- -A-A
*-.L + .i -- a£f
Sno -Imact at, 10- leesia t at et ersXlin2 Yn XWnjYin) I
3-3-
-1.5 +1.4-3.6
5 -7*7
h -0Z-2.:-.
:2-3.02 9
4a 2
32_ o -3.
:.242 (D* ±:-2
esaa ze - ii. ea a3 ----Ial -e -1.3 s =c d t-0m"e9pa sa Z r %
imac a* gee- -npaCl at 3m C M. e t e -r s-
-Z
c ~~ -!.L.
+0 al.. %F.. -4
C 20 +-8-A
4, 2 .s --- 7a
ol r. -
F0 t. #±
+42 +12 __
l7nU - j -
I~ *2
e 410.4
+j 4 j~
44 - S
ase- Z -
JC
4.4
Sar I v mt
- A---
- -- gL
7 - -. C
is a -77.
~t
- =4
-- -- - ~ ~ =
_ - -. -- - -~
S
-r.~ . a .. ?J rt tt~Z. ~ at -
- - - -
I. a---- ~ ~. - - I
-; rr- ~ ~ ~
-- C -
-~ ~ t = --
. t -~
-- 3I z-i;? - - -~
I~37 3=
-~ - -~
----
- - - - ---- -I.4. ~. - -~ -
.r .4. - -. -
- .4- -
-s -1~ e -r - -- -~
-4. -S 4
w - -
a I;I- - - -- - a
ft -~ ~,
- -~
a .j~
- 4..
- - = - - 4.-- -
I i:t' I[ _
IrI I
- r~~-- -
Shot Impact at 10 Meters Impact at 30 MetersN umb ey X (in. ) Y ( in. ) X (in. ) Y (in.)
3~a +3,4 -7.4 +35. 3 +6.3b -0.9 -8.6 +33.4 + ,3.5
c -1.8 -8.1 +33.0 +33.4d -3.7 -S.5 +18.3 +36.5
e -7.4 +1.8 48.1- +38.4f -6.5 +3.2 +10.3 +4.8g -11. 3 +1.8 +2.7 +34.3h -12.6 +3.7 -1.0 +40.0i -5,2 +5.4 -21.4 +5.j -3.6 +13.1 -23.1 +2.5k +019 +13.8 -37.2 +1.31 -;4.-8 +15.7 -40.2 +7.2m +2.2 +15.7 -2.7 -28.5.A +4.0 +1.9 -4.9 -27.0o +12.3 +3.0 -11.6 -31-6p +14.6 +4.3 +9.1 -23.1
4a 412.4 -7.G +29.5 -t32.4b +12.2 -6.6 +24.0 +35.7c %'7.0 -10.0 +21.5 +29.5d1 +7.5 -4.6 +16.2 +15.9e -4.0 -9.4 +38.4 -19.0 if -6.4 -5.9 +35.1 -16.0g -8.0 -6.6 +21.1 -12.3h -10.9 -7.5 +6.1 -10.1 Ii -90 +7.4 -7.5 -32.3
j -919 +9.8 -25.2 -- 23.8k -4.7 -12.2 -38.8 -26.51 -13.3 +14.4 -30.3 -39.7m +8.4 +11.i -31.2 +21.6n +8.0' +9.5 -33.5 +28.6o +10.3 +10.3 -33.5 +26.2p +7.1 +4,7 -38.4 +35.1 Z
Sequence of sub-projectiles 1isteP at E t r w rs is not
necessarily identical with the seouence %Ae at 10 meters.
39
Shot Impact- at 10 Meters Infact at 30 Meters :Nun~ber V~in ) (in.) X (in.) -In.
-5a +12 8 -1 +12.0 417.2
"b 7.-4.1 +2.5 +24.0c+1.4 -2.5 +18. 6 +0.1°
d +7, . +0. 2 +14.0 +21.5e -5. -1!1.5 +2.5 -6.7
-i" -15.6 +22.5 -9.5
g-4.8 -8.3 +402 -39.5h1 -25.3 -5.9 +13.8 -35.5i -14.8 2.0 -4.4 -33.5
Sj-16.7 -22 -14.;2 -37.0
k -23+1.5 -14.1 -294 .01 -0.1 +13.0 -35.8 -47.5 :m +4.2 +15.3 -42.0 +7,7
S+3.6 +7.3 -30.5 +32.5U +6a0 +5.4 -26.7 +30 2
Nu+6e0 +20 -23.6 +31
b +9.5 -0. 2.4 +37.2 SC +7.9 -0 3 +23.8 +2.9
d +7.9 -0.8 +14.0 +21.5e - 2.9 -1.2 +24.7 -6.0
f -3.9 -. 6 +28.5 -5.6q -4.8 -6.5 +4.2 -39
h -I.3 -96 -. 5 -21.2i -10 . +4.4-30.0
j -. 7 2+4.2 -142 -3.79.
k -12.4 +4.9 -31.6 -2.7--1 +4.0 -35.8 +9.2
1m +3.65 +_ .3 -30.5 12.5 '
n +6.0 +13.0 -44.6 +12.2a +13,2 +14.8 -44.3 +42.6
p +5.5 +15.3 -351 +320
Sequence of su-.-projeiies listed at 130 1ters is -ot.
nlecessarily identical with tte sequ~ence listed at 10 meters.
. 4G
-C 7
Shot Irapact at 10 Meters Impact- atC 30 MetCersNumber X(in.) Y in.) XMin.) Y (in.)'
fa +9.5 -30+116 +32.6=b +6.0 -3.0 +6.1; +38.8
c +4.2 -3.8 +11.0 +41 5d +0.9 -10.3 +.29.5 1±t-9e -3.1 -94.2 +28.1 -8.9
t-4.0 -9.3 m1. -.g -4.5 -0.4 +411.0 -9.4h .7 +. +24.6 -34.7
i -1.0 +?.8 +2.6-3.+e 31. +.," -15,8 -i4.0
kc -7.7 +3.5 -12.0 -25.91 -1.3 +9.8 -29.3 43.1at +1.4 -rlO. 1 - 33.8 +8.4
+.-+12.3 -36.)1 +10.4o +3 3.4 -23.8!9p ±03 +0.3 -4.0 +30.2
Ba +7.0 -4.7 +23.3 +20.3b +6.2 -7.3 +14.6 +210.0o+3.'7 _-7.0 +16.0 6.
d +1.0 -7.4 +i5.6 +13.0e -6.2 -5.1 +21. 0 -14.8
f-7.8 - 3.2 +19 .9 -- 23.6
I -7.9cc a-t+3.
Sequence -2f su-rjcieslse +t34 -25r s 5ot
k1 -A* -2 .-
Shot Impact at 10 Meters Impact at 30 1ieters
Nuer X(in.) Vin.) X (in.) Y (in.)
9a +0.9 +1.7 +3.7 +4.INb -1A-9 +i.3 --7.1i +5. !Sc -0. 8 +2.8 +8.4 +7.9
d -4.6 +2.8 -2.8 +8 .6e -1.4 +2.9 -13.8 0e.er +2.2 +3.0 +7.9 8,7g +0.4 +3.1 +.2 r9.0h +2.3 +3.2 +7.2 +9.5i +2.0 +3.3 -.3.7i +2.1 +3.4 +4.9 10.3k +1.2 +3.5 +3.2 - 1.61 -!.3 +4.2 +. 2m -0.2 +4.3 -4.5 ±12.5n -2.! +5.2 -4.3 +15.7o +0.8 +5.7 +. +i.7p 4-2.3 +6.1 +8 A6ir4
10a -0.5 -1.] +1.5 -9-4-04 +0.3 +2.8 -3.7c 0+I1 9 +249 -l,23 27 -1.3 +3.7
.0+.8 3 4.4f 0.8 +2.9 -37 +
- =. +3-0 +6.9 +4.zh -2,1 +3.1 -3.4 -4.9
- . A-4 .5 +5 ,
~~~ 4±. 46j-I.S +3.@ .t-
-i L!. '7 -4.2 ~*. - .k -A-8 +2.7 +-
n-3.5 +56 -4.7 14. 2- -2 .2+ 5 . 8 -3 .7- + 4 3
S- 4+ 5 1- . 9 1 6 . 3 !
Sequence of sub-projecta:.s i at 30 meters is srot
necessarilt, identical with the sc ence listed at '0 meters.
- 4
II
F4
?AA Mi. EzTIMATION,, BYl~ YZM-hOD- OF M-i4EflWS
Xi~a xX: for s-ach i I,.., x~ii
The expected value of a oau-ta. Aist-ibated randot, variable
is a/A, the varianc.c at/-.2 To use the methof. of mn-'ents
set the sa:tple :Eiear. eqxaUj to the theoretical mean -and -the
Sample variance equal t, the t-hcor.et-ica± variance. Thu- as
-I a / end S A.
Slubstituting xfor it- equivalen-t exnrecSSion in the reltnU shp btven s~andthedistritution nararmetev yi'eld:Beve s IIi P
Substituting 23 value fcor 2aLda in the exnr-esSion foDr z
X 0
Thlerefore, the -jetloc or noens ro-2uces &h..flThin
esttzatces -for the parame-e~rg
Since &aa.,.& both denen on x and 2s*, -e are petvi
cxrrela ted.
The bias b (9-i of an estimate iLs defined a lb (8)
E (T) where T is the stati~stic used t%-o esti-ate -A, the
true parameter value. Approximate th3,e bias by using a firs t
order Taylor- expansion:
43
= X~ X-2:x (s-04)
2 6
C; 0
2 'ad
22
A ~ Zcv (six)
ED)SI <Q c (,ic since aa ar~e positivek-V
-= correated.
~~44
APPENDIX C
CP =MR OF MASS AS A RXNCIEWCN 0F RANGE
The information iout t.e center -f mass of any om round
is limited. r location of The center of mass is knVwn at
th-ree ranges: 0, fl, and - meters. At zero range the cewnter
of mass is at the origin since the pellets would not have had
an opportunity to dispere. -he location at 10 and 30 meters
can be calculated fr the data. Wth., this limited informa-
tIon it is difficalt to acc-urately derict the ction of the
center of mass. Additional experimentation IZo rovLe more
data on the motion of the -enter o r Mass would allow a more
thorouh examation of thle mtion of the center - -ass as a
fucaen of range. A desirable situation would ba to have
sufficient data to allow an analyst to determine a distribu-
tion of t-he center ,f mass. in t.'-e absence of furtler
info0mation, the data at hand sh-Uld he used. Therefore, two
possible anor-oachos to model the motion of the center of ra-s
are presented here. Users of the results of this stu---
should deteIrine what procedure best satisfies their view of
the drift of the center of mass.
One meathod to mdel the drift of the center of mass is
for each one of the coordinates to fit the lowest degree
polynomial that appears reasonable. An exam ination of the Z
data w uld sewa to eliminate a ccnstant or a linear relation-
ship between either coordinate of the center of mass and
range. Since a parabola contains three parameters and coordinate '
_ 45
NA
$
values are available at three ranges for each shot fired,
it is possible to fit a parabola through the coordL.nate
values for each coordinate of each center or mass Usin the
parabolas determined for the two coordinates of a center of
mass, the position of the center of mass ca- be determined
for any range. However, the procedure does not provide ary
insight into the physical pilhenoenon procucing the ntion.
It does provide a method of m ideling the motion of the
center of mass as a function of range which will reroduce
the experimental data.
A model of the drift of the c enter of ass can -h based
a the physics of motion if one can accept three assemptions
the forces acti.g on the projectile (the reco-i an airresIstance) act for only a short period of time, once the
covering on lhe pellets is aerodynamically stripped away
air resistance is negligible, and gravity does not signi-
ficantlv affect the pellets their maxim= effective
range. Under these assum-ptions, .e forces causLnq the centtur
of mass to drift can be visualized as acting instantaneously
as the projectile leaves the barrel. After leaving the
barrel, the projectile would travel in a straight line. One
can then use the two data points at 10 and 30 meters to
estimate the parameter of a straight line tkrough the origin
- usiz.: the least si-ua.-.es method. The line rast go through
the orjgin because at zero range the center of mass is at
the origin.
46
UK. CR T ---
T-hi m--odel has thie asxeal o-f being b-ase-d on rohvsica
principLleso motion,.Uteei has the Aisadvalnte
-not rCDr-dC2cfl the- C actu-al- e3-merimn tal resultsecas it
-is unlikely that a le squares fitted liewil1o2 roz
either of tlhe data points.
Once thne usr has se lec ted a m-tbo-d of moe'g th I e
motion of t-he cen-ter of n=-ass as a functi-;on or- ranoes, he can-
use th-at model to cnrtarutinfor each- of the extanr-
mental rounds- When a rom4n is efireciaL in a simlulaton of
one -of these functions hasad _an Den-..otz1A
randw-.lV Selected to%. XrvdG u nsUno the center of
mas at* the range desire-d fr t'. sirnulation r =L
the drift ofa the cente-*r- of nass can b-ein'cue to the
Ss:*'fat1 On.
__ __ 471
it'.LIL 1&±Z coLiz,1-p-- -t fs t rie ri-qn' daab
rtlnr' frcxm "e ~JLates -. r e-k --- #_ e nt_ rr=t
Wc--fLa have- Cx the 2stribr-on of pe ee stnxkes.
Atbnucn g the bosa--mri4ate no ea-. L ~~ir~v aescribe
the~_J~ tabt± it was afcinl I- cw- tt ' -
~~~~~~~~ nn__ c-rW-e :ix ue ~cetei na Gflra the na5,_ - thr e --- % -.
ttae cr-ana r-oielzto~s-=--~x use e--..
no~ua ~'n eCar. us fy hiTheni~ 00 nnLIL cr- -is a±=.
distrbton.
th2- bivrate rcz=i - idectndent* Variales. 4 A Without
for ro%-th var 4 ab1 es was asslzed t bezc=adteviae
{ -- *its a .I'a~ of : s ze La - .a -
'litrbui 'nwit
ASA
At
2-C --- v Ia -
A -rI +V
diacona]_ and- t
ahe~~LV.4' ptJ6 e)n for, lt V rit'xha eC
CLe r ~--Xz C- .i e r. --_ _AassA
VXY VA X
1 ~ (.,zI=CcaI7ZNEach± t r-. te g -asteult
r Lre"-e assc theraanrc - of 4p t snriad cp ncr ; Henceztant2
2-
t~~~~~~~ t.. S2=tt j --
mv ' Arc - :r~ewr v 70,
!.I. 7 - i Nd C -0 0.rCIM teS S
at=± --d LAA Is oi t-r o--nD
n -
wc *-_ rzn ohl' a vc-real ptrrtrt
it-: foorw al .o a'meoe ' _ _ - the abov ro_
lo a X=- %rw -Q U
f reae all K>-. thcte t-a--iv.- ththat'n
wit-h' ------ amw 4 - -t heo -I A -- %r
hl -r Catrt of Z__
shou -af - C a D Q -- -;7
A~4,
Wrcai o- samew disz-3 atn-ze 7-c tz fran I M' e f arg -
A4=;taha *'-a gr Pz -,s- - --
th dta i rdued e n tala inc et-&a-:.X-. -.. .-- dat~'are U or-
T.Aent gir-v -Sa rc-1nt D-er-sac a ' ete s
e~~- x; ~ x~ PC'c t~ e dv iuet r rcen s or mn s~t~ s and -Jrasws e
assises h o- f tne th-tsls nav *-sAe madit
to have tei centers of ms cicie a Thl~may be
preOtc Q
n -ne
r- rc...A
-r~ ~~qs- - -- - ,
~ M-~-~m
The data ct.3J by RIZN-ill. Vr_'SZICt
nTh exAct dtenlt-ated be ....ne a... ::ae U-Sice
ral data, the cxeviat .on was net;- S~-n- * icet
~ZiZ n~i istitit had pnroc- ce
tZ ir 41 1tion was bae on a P ola_
rS-- t_ L-
Were proaducea. A aa. r .. A&rf nZZWS
the relaaonswt-s: X=z- CO- -- and - ~
data. n'-* the" ac p-- 7- -
thr u eic ts paibS of2 am2 Zda- z
L-: ah ar I .*. a
*the radii Of the ocice- were then rea !rzm a re-ctor.Th
valuesselecte, were base-d o the val-' of o tne aJ 'a aaee
produces- hrO t-.nin h a7- alu~ at the d-ecimal 7.V -m
PO,
-=4-. -7;l
- -. --- ~ 4--- -4 ~ 4-- S
am-.- em th a=Ead______u-_= =
- -a t7~ o~- bee -z 1. z t C* Vt
ZlzF-- te m~- t - t - -
the t r b-;n ew .-or -h ur-~e- 1= 0= as
Se i b =- - y-._ =_-n~ snaC- ;CTlmC
_ -ate SCA____
tim ___s i-_ -Z=u
- - ore
nse the~=
3--s- 4--~ r- -e - - 4
on-e ~ ~ -ean q -onM - -t -- M~t
ec~ns-o -o e ~ s~aiU 7__ a- Ih~ a 0ecXno.t.
wa s s t Calec
- - = cF
AZter 1000 runs through the program , the stored test
statistics were sorted from smallest +-c, largest. The 5, 10,
25, 50, 75, 90, and 95 sample percentiles were printed. The[
90 and 95 sample percertiles were used to represent the
critical values for this goodness-of-fit test. The sample
percentiles are given in Table 3 as a function of the alpha
value.
rI53i
TABLE IV
SA-.MPLE PERCENTILES FROM GOODNESS-OF-FIT SIMULttTION
SAMPLE PERCENTILE
ALPHA 1 10 25 50 ?5 an 95
3 21.8 24.5 33.1 84.5 157.2 135.3 199.24 26,0 31.2 42.2 65.3 105.1 135.5 162.85 20.0 22.2 28.6 36.8 45.8 56.0 65.36 20.0 23.0 28. C 36.1 46.2 57.5 65.77 21.8 24.5 30.1 41.0 54.8 60,.0 75.18 n-.5 24.5 30.1 39.5 55.6 69.5 76.29 i6 22.2 27.8 34.2 43.2 54.5 63.5
10 19.6 21.5 26.3 32.0 38.7 45.8 50.711 19.6 21.5 26.0 31.5 36.8 42.8 46.637- 19.2 21.8 25.6 31.2 37.2 42.5 46.613 18.5 21.1 25.2 30.8 36.5 42.5 47.014 19.2 21.5 26.0 30.5 36.1 41.7 45.1
15 19.2 21.1 24.8 30.1 38.7 41.3 46.21 1- 18.5 20.7 24.8 29.J 35.3 41.3 44.7 "I17 18.5 20.7 24.5 30.1 35-7 41.7 45.119 177 20.0 21-5 30.1 35.3 41.0 44.7
19 19.2 21.5 25.2 30.5 36,1 42.5 45.820 18.8 20.7 25.6 30.1 36.5 41.1 45.1
54
APPENDIX F
VARIATIO2 0-7 LAMFDA AS A FUNCTION OF RANGE
Since data for only two ranges was analyzed, only two
values of lambda are available at different ranges. An
additional value for lambda can be deduced by observing that
the variance of a gamma random variable is ct/@ and the
variance of the pellet strikes at zero range is essentially
zero. Since a is a constant in the models of interest, A
must increase without bound as the range decreases to zero.
This is the only way to produce a zero variance at zero
range within Lhis family of distributions.
Since lambda is urbounded, it is more convenient to
determine the reciprocal of lambda as a function of range
and recaire that function to go through the origin. This
sit3ation bears a strong resemnlence to the prcblem of
deter--inn the drift of the center of rass as a function
of range. Is in that case, it would have been desirable to
have additional information about the way lambda varied as
range increased. If data had been taken every five or ten
meters, a more detailed analysis of the var'ation of lambda
could have been accczplished.
A parabola can be used to fit the three values of the
reciprocal of ±ambda. This method has the advantage or
exactly duplicating the experinental results at the known
ranges. R'mu-ever, it offers no insight into the reason why
lambda should be expected to vary tn this hay.
55
An alternative methed can be nased on the dimensionality
of the reciprocal of Lambda and the moments of the cyvnma
distribution. Since lambda varies with the ran-e and the
reciprocal of lambda times a constant is the expected value
of the range, then the reciprocal of lambda must have dirn-
sions of length. Since the dimensions of the reciprocal of
lambda and range are the same, this suggests that a linear
relationship between the reciprocal of lambda and range might,
be appropriate. Therefore, it may be reasonable to use the
least squares techunique to estimate the slope of a line
through the origin and use that line to determine the recipro-
cal of lambda, and hence lambda, as a function of the range.
I56
LIST OF REFEPENCES
1. . P. Uxit-.- Laboratory Reporth a ri c-
Tests of .45 Caiber Ball, Segmented-and Shot Cartrides,p.,2 February 1971',
2. U. S. Army material Systems Analysis Agency Unclassified iletter to U. S. Army all Al s Systems Agency, Subject:comparative Effectiveness of the Cal .45 _14261 Shot -Rouand when Fired fr a: a Rifled Barrel Versus a Su-othBarrel, 6 September 1972.
3. Hoel, P. G., Port, S. C., and Stone, C. J., Introductionto Statistical Theo y, p. 95, Houghton Xifflin, 1971.
4. Meyer, P. L., Introduito, Probaility and StatisticaI J5k icapns 2a -ed., p. 333 :dison-Wesley, 1970.
5. Hoel, P. G., Port, S. C., and Stone, C. J., Introductionto Probabiily The__ry, p. 130, Houghton Mif fl- -n, -1971.
57