Documente Academic
Documente Profesional
Documente Cultură
.,
;1,
NA
CALIFORNIA
Technical Memorandum
No, 63-1o6
24
October
1955
FACTORS AFFECTING
MEASUREMENT RELIABILITY
by
Jerry L. Hayes
Measurements Reliability Branch
1
'\
FOREWORD
(!omma.nding Officer
MBASUREMUT RBLIABILI'l'r
by
Jerry L. Bqes
Corona, California
I.
----.....,
--
.....
~--,--------
of IJandling Errors in
Testing and .easurins", Industrial Quetity ~
trol, ~, 1954
20rubbs, P.E. and Coon, H~J., "on Se'!;ting Te9t
Limits Relative to. Specitication Lilllits"" Industrial Quality Control, March, 1954
-
MCIUllUTY Of'E~LY
.. .
.b . . tM (\esign :1;o1eraaclt .. J~ 1004t.ed at
the 2 aig11a c;ootidence .l"vl... on the product
cli1tribution: curve, .
MICEPTllM l'llOOUCTIOUE
10 TEITIH lllllOlll
~l
1---~~'.t
---------1
t---1-lt-F~::+-~..q_.::_=+~=--lu
Ia
I
MI
- I T Y Of E-EOUILY
ACCEPTllll l'llllllUCll DUE TO
111t111t
EllllDM
__..._.-'---I.#
...
...
1
...
,.
.
'l'he a1111bola .used on tlul' cui:;ves (Fisures 2
throl.lih 8) a..-e described as J'bil.Ow.a:
' .
A.I,E, c Allowable lrurl;rumente.tion. Error
(the value obt<JJ.ned lrl.U be -~ ot the, total plus
to lllinua er~ e.J.lo~ 1.e. , A.t.E. ;~ mlita)
Design Toierance iviliue or the aiiowacle
variation a.bout a nominal des13n value.
ti : Fe.cto1:" to b.e multiJ?.~icd bT ft r-,g. to
2
determine locat fon at tect tolere.rice in rei.a.t19n to deei:;n tolert.nce. -+N locat~e the test
tol'!lrance outsid!I the' desi::;n, toi.ars.ilce end -:-?
loco.tes
it
inside,
j;
..
-I
~nerunuu
M,IU1-....CU ...
TutlM . . . .
To_ ..........i-4olo~L."1
t-""--"i.......,--f.;,.;._""+-J,_L~~..Ui.,'
.r
.I
~~l
It
-i---+-~i----.+-~___..
~~J
...,.__F"
:-.."; r
~ 1....
?.r:-
':
"
'~.
~,,~,
.:~
~ .... ~
~
;
: .. i_.;:
.4
'I
,! -~
~.. 1.:_
';;j,U
.,,,
'
1:
t--~+-~~:;..._.J.'---+.--.Ji!:__.Ll&.a
'
t-~-+~~~+..~~....,,J.:J,..;:,,L~~
,,_
...
,....
....,.. ....,.._,..
''*'
lUT-
;..
r-~~-:d--~...,.__"!"""--1-~~
-'~
.;{
.
~r
.;J
~
'~
'.
'I
'd'
.:~1
,.
_,_.f'-.. . . ~~~
~~~
...1
-:
~~~1
~"1
. .I
......
-'-"~-
.,! -
'!be curvea can be empl.oyed in the following manner to ensure that within 95.5fo of the
teata, a apecific teating riak or reliability
can be attained:
a. Determine ~.~9.aD!i.technical
data the proper comprca:lvblftveell Consumer and
Producer. R:l.k to yiel4 the aeiSli'ela. " testing
proceH r:Lek.
.
~. Atti~ that deiSn ~lerances have
been ',.a+,iblia'bed r8al.ia'Ucall. .and represent
rhie instrument, which is to b~ used to measure a 4oo cps. 100 volt parameter, has a range
~t 150 V. The selector assumes that at lOOY
he has a accuracy of 5'ti; however a f'u.ll scale
accuracy qi'~ results 1il an' error of 5~ x 150
7. 5 volts at 1501 ' 'l'hiB' error "at l.ooY results in all &CCUfacy'.9f:C"7~5yh~) 1;00 7.5 '/.
which is tar too great for an A;I.E. of 5.3.
t1'118
Error
:tequ:l~nt.
4.
,~e:
The selector obtained a
60 cpj 1~ Which wa& not frequency
compen1ated or calibrated tor 4oo cps. The
error Of 'the .i~tl"Ullltnt at this frequency was
2.5~ of tul1 eCl.l:e due to, the ettect ot frequency~ aione..
'1qe -ewer-all accuraoy ot a 5~
selectt<l .. inatl'Ullleat woulA be reduced to as
much ao. 71 ~ full stale a,'t an operating frequency ~i' ~cps.
lllt.ablish
~ Deeign ~
Arfn!'!Qleran&e
'
2 x Dis
t ...\ '
\, ~
Since the fantily.of curves was established for a desiGD tolerance" .location at the 2
sigma confidence 1*1/el of' i;l::!e product d:!.strtbution clurve, a design to~$(:e which deviates i'roii this 2 sigma locatii:xl.~Quld require
a new i'alllilY of cur\tes.
" "
or
---wen
J'lbe
can b! rea4U1.obt~.)
.
.Frequency
:~ample:
The "fnstt'iliiit!ttt will be
operated in a field condition where extremes
of heat, hwhl.di ty, and. rough hendling are
e:>:pec~d.
Unless ruggidize.tion and temp2rature end humidity compensatiqn measures are
taken 9r alloved for, groaa errors can result.
I'he h:l.gb amount of heat dissipated by the
shunts and reai.stors of some instruments when
operating . ne&.r' ~h~ir rat~ current value may
produce significant
errors of indicatfon.
Unlesi adequate ~t dissipation measures have
been provided)' 1hstruments should be selected
to opero.te well below thc~r rated cut-rent .
. )
v,
are
1~ffects
Disc~ssion:
The selection of an
:lnetrtinent ~~hose resistance we.a lolr cnou~h to
draw exceasiVe current from the meo.sured pare.meter ~ ~;tGuld redui:e 'the .meaau:i.-ed vbl tage and
thus introduce further '11\rrol',
The 1.r.ipedancc
ot the m$aouring syiitems inueit be so established as to load the 'Circuit t.o a r.>inimwn a."ld
concurrenU.y produce a desired sens~.tivity.
1Woodaon, W.E.,
1 Indicates
uuta
...
DIVlllON WIDTll
,...
'~ ....._
..
I
I
"'"'',..
_,,,_.,
that probably 95,5~ of instruments meet accuracy figure, Limited investigation has indicated that the confidence
leve~ of 2 sigma is norinally exceeded,
instrument s accuracy wa.s :!:5~~ and the calibrat ion instruments accuracy was ~~' !t is logical. to assume that a portion, of the ';If, inetrument s error would be transmitted to the 5~
instrument,
In the most extreme case, the
two could combine' to yield an inaccuracy at
the seleeted instrument of 5+2 7'/.. However,
the combined accuracy computed to a 2 sigma
confidence levei (that which waa originally
eatabliehed for the instruments) is mathematically the square root of the awn of the
squares of the contributing errors or (52+~)t
5.~. Further contribution to the error
would be made by the instrument used in calibrating the
~ instrument, although its
e~ect would be
lesa .pronounced. Thus, it
would be necesaaey that the. calibration system
into which the selected instrument
18.
placed be considered from the standpoint of
ita effect on the aa1umed accuracy of the
inatrument.
Since the aim is to remain within the Allowable Instrumentation Error at all
times, such consideration should be
given
cautiously, It is understood that many alternate plans such as calibration carc1.s, checking
to t rated accuracy throughout, etc, could be
employed; however, it is believed that for
ease of operation, consideration of personnel
factors, dependability, etc,, the method set
forth is the most practicable for the majority
of instruments involved.
If any correlations
between like tests perf'o:nned at various locations arc expected,it is also necesear;r that
all selected instruments 'have a common reference .and be so integrated int:> the program as
to provide compa.ra.1lJ.~ t\a.ta. at var:l.ous levels
of tent and inspection.
este.blisl~
Condition . c. would be
tlil.filled
following
an analysis of the results of
conditions a and b. This analysis would yield
the data necessary to the selection of calibration instruments.
The instrument acc\lI'acy at the test
set" eschelon can be denoted as C. From the
results of the compatability program, the
value of the corresponding Allowable Instrumentation Error, A.I.E., would be provided.
'!'he seJ.ectiori ot , an e.pi)roprlate calibration
instrument
at
the "eeeonde.ry
transfer"
eschelon (whose aecura~t can be denoted ao D)
and the "secondary reference" eschalon (whose
acc\.\rany caa be denoted as A) would be !11a4e
by inltially considering the factors affectins
reliability th&t have been previously discussed in section III.
lie at the Haval Ordnance Lal)ore.tor;, believe that the problem can be contained within
reasonable limits tlu-oueh the implemt\ntation
of two carefully planned rrogrs.~s, clocely
coordinated for ma;:imum cooperation and il1terc;lwlge of int'orniation, and delicated to the
t~k of aasurillG reliable ;nco.surements.
These
are a calibration program and a compatability
progrolll,
l. Calibration Pro~ram
file miaaion . o the calibr~tion program is to ensure the !11&in~enance of accnracy
within the desiJn limits, of the r.ieasuring
systems emplo;o!ed in the test of all Dureau of
Ordnance guided missnea.
The calibration
'program must have flexibHity to provide tor
clw.ngine; iequireMents; must keep pace lTi th
!'.:lld evaluate the 1:t0at rccet cleveJ.opmenta in
~tandards, especially in arcan \There start<:.ards
are nebuloun; and r.1\1.at :irovide e. cott.1on reference for each cha:r>.cteriati-.: mee.sureQ.,
It would now appear that some syste111ati c method of relating A.I.E. to A,B, and C
would simplify the selection process lthen the
effects of calibration error transferral were
tmder consideration. Ae v~.s prevfously stated
i.f it. is e.sSt'J!IM that the hstI",U'.!!1tS I errors
have a nor;1l.l distribution fl."ld t;hc1r acet~O.C;f
re.t1!lge arc o.H l!etablis'1ed at least at ~he 2
sii;111 co.1:~:!.dcnce level, then th:;i \'oot sum of
The
inotrument
selection
graph
resulting fro:n the equation .is presented in
FiGUre 10. Onoo the value of R is established,
any point on the corresponding radius can be
selected to arrive at values for A and B which
are r.easonable and in lreepins with availabJ.e
calibration standards.
Condition
e
could be fulfilled
initially t}lrough the recommendations l!lBde b~
instrument
manufactur~rs,
statistical
interpretat:i.on of 3.Il.Y~ previous ce.libra:i;ion
data available, o.nd the experience of other
activities using similar instru.111ents in lilte
applications.
The expected environment into
lThich the instrwnents "1ll be placed weighs
heavily
in
the detel'l!lination
of initial
calibration periods.
The selection of proper
calibration periods is also highly ir~:portant
to the success of the prosrmn.
The safe
approach would be to establish peria<:.s between
calibrations which were shorter than estimates
actually ipdica.tc.
Analysis of calibration
data would provide information
tor cny
necesse.r,y nvill:l,ons to the periods,
..
--...
. . . , _ T llUCllOll-1
... +.1
~..AtJl - ~t
Pll.10
,,
Condition t is the ''we\teh do<:" tunction of the proaram and provides to~ rovisions
to calibration periods, methods and :;roce<lureo,
csl.ibration policies, stsnda.rV.s, and a.i.i other
elements associa.ted with the pro;p.aut, E::t91\
sive ll8e of calibratj,on data report fol'llU! and
other simpl!fying report fornl8 would provide a
be.sis for. these "revisions when screene4 tan4
analyzed by a central egency. 60111& 1'orm ot
periodic publication or CO!lllllllnica'vion from the
CC11;1.tral aaency to using calibration or~an1Z$
tions would advise as to ch?Jtges and revisions
in effect or forth-coming fl.a well aa other
information intended for co11tl:1itou::: improvement of th:;; proe;rar.1.
C,
2
'l'hef!., A.I.E22 - KA2c:+KB~c2+c
and, A.I.E. -1 lCA +KB
c2
.,
'
...
....
...
...
0'4 ;
'
I
I
,IG.12
...
I
0
.. .. ..
.I
A.C.6.
a:rr.2.
Compatability Program
The success of the calibration program depends on the proper selection of test
instruments. A perfect calibration program
will not provide asurement reliability
unless controls are provided to guarantee the
adequacy of test instruments employed in measuring processes,
The calibration program
ensures the common referral and accuracy of
instruments used in test sets, but the efforts
involved will be wasted unless the instruments
are correctly chosen,
14
I ..
II/ .....
J
II
..... /
.....
.. !!
.. .. .. ..
IV,
Conclusion
The factors which have bean presented
here as affecting the reliability of' measure. ments are not considered to be the only ones
which influence measurements, but they are
considered to represent the more important of
those generally encountered,
The systems of' selection and measurement
controls as established by the calibration
and compatability programs nave been set forth
in this paper as a means of pinpointing and
containing measurement problems 1 rather tha.'l
as a means of' eliminating the:u.
Neither the
compatability program nor the calibration procan operate separately, since oaa is dependent upon the other. The need f'or close
coordination of' ef'f'orts a.a well as for active
cooperation is essential to the development of
reliable measurements,
gram
MATHEMATICAL APPENDIX
The family ot curves . shown 1n tigures 2
through 8 were determined by use ot a computer.
A more tedious method involves the use ot an
autoistlc calculator and a table ot areaa under
the normal curve.
1.
lt
-t2
" eT
-1
Where:
r(k..t)-b
dtT.
-k
-(t2+s 2)/2
dsdt
-r{k+t )+b
ot A.I.E.
to design tol-
b-N, the number ot standard deviations ot A.I.E. the test tGJ.erance is displaced trom the design tolerance.
t and s selected tor specific
increments to attain desired accuracy ot computation.
ID