Help for

Uncertainty calculations using the Evaluator program

1. Introduction
This help file describes how to use the uncertainty spreadsheet and equation editor for the
evaluation of uncertainties in measurement results.
This help is complementary to the help in the uncertainty menu.
The methods used comply with ISO 1702 and with ISO !"# and also provide for
uncertainty calculations usin$ #onte%&arlo simulations both from the spreadsheet or
directly from the measurin$ equation.
The uncertainty tools are located in ' windows. To $et to the main widow from the
(valuator)s *oard +main window, clic- Uncertainty. .n uncertainty spreadsheet opens.
The second tool is the uncertainty equation editor. To open it clic- Description from the
uncertainty window. The editor is at the bottom of this window. These two tools are
sufficient when doin$ routine uncertainty calculations.
. third tool opens when clic-in$ Estimator in the main uncertainty window and can be
used for some speciali/ed wor-. 0efer to the help in this window for how to use it.
In the followin$ the uncertainty spreadsheet and equation editor are described.
2. How the evaluator calculates uncertainties
The correct theoretical way to start uncertainty calculation is to write the measurement
equation. This equation has the form
12 3+415 42564n, 718
9here 41 etc are the components contributin$ to measurement outcome 15 and 3 is the
function that defines their contribution. 3rom the equation 3 one could in principle
derived how the components uncertainties affect the uncertainty in 1. This is not always
easy and appro:imations are used. The simplest appro:imation assumes that the
uncertainty in 1 can be obtained from the root some of squares of the uncertainties in the
4is. The ISO !"# supports this approach and it results in reasonable outcome in many
cases.
If there are correlations5 the ISO !"# tells how to calculate the standard uncertainty in
1.
The equation is;
u +1, 2 sqrt< = +ci>ui,
2
? =+ci>ui, +c->u-,>0i-
2
@ 728
where ui are the standard uncertainties in 4i5
ci5 c- are the derivatives of 1 with respect to 4i and 4-
and 0i- are the correlation coefficients between the components 4i and 4-.
In most cases the correlation coefficients un-nown and are i$nored and the simple 0SS
formula results.
To calculate the e:panded uncertainty5 accordin$ to the !"# one must first calculate the
de$rees of freedom of the combined standard uncertainty and then under the assumption
that the result has a student distribution5 to multiply the standard uncertainty by the -
factor appropriate for the chosen confidence level.
In this approach5 the !"# i$nores the actual distributions of the components and their
possible correlations. This is rectified by some options in the (valuator.
The uncertainty main window will do all this for you even in the presence of correlations.
The new !"# is considerin$ these points and proposes to use #onte%&arlo simulations
to overcome some of the problems. This pro$ram is already featured for doin$ #onte%
&arlo simulations.
3. Uncertainty window
The uncertainty window is comprised of A tables. These are
&omponent Table
&orrelations Table
#easurement Table +repeatability,
0esult Table
These tables are interconnected and chan$es in one of them may affect the others. They
include all required data about the components to comprise the uncertainty bud$et.
The four tables in a window can be manipulated +duplicated5 moved5 saved etc., as a
$roup. This $roup is referred to as a page. &hec- the pa$e menu for possible functions.
To review and chan$e the displayed pa$es you can clic- on the upBdown arrows or chose
a pa$e from the two dropdown list5 one by pa$e number and the other by the pa$e
description. To edit a pa$e description5 clic- Description. This will open the description
window +see more details later in this help document,.
9hen you start a new uncertainty file pa$e 1 is assi$ned automatically. Cater you can add
additional pa$es of related calculations to the same file. To add a pa$e5 in the page menu
clic- Append a new page to the file. If you want to reorder the pa$es5 use the Target
page numer in the little editor to indicate in the pa$e menu to where to move or
duplicate a pa$e. If you do not provide a number or last is indicated5 the tar$et for these
operations is the last pa$e.
1ou can flip font si/e by clic-in$ the AB. button.
1ou can copy the contents of a cell by dra$$in$ it to another cell.
3or more editin$ functions ri$ht clic- on a table.
Dra$ upBdown the Drag to si!e button to show lar$er portions of some tables.
1ou may add rows or columns to some of the tables5 Eust ri$ht clic- to find the ri$ht
command.
.ll tables can be copy copy%pasted to a spreadsheet li-e #S (:cel in both directions.
3.1 "omponents Tale
This is the startin$ point for buildin$ an uncertainty bud$et +even if you decide to use the
measurement equation,. 9hen the pro$ram runs per ISO !"# +clic- on the lue
triangle or the menu calc#I$%&, it calculates the table row by row and from left to ri$ht.
.fter each row is calculated there must be a value for c>u in each row5 either calculated
or edited. If you do not provide the required data to calculate c>u5 you should provide the
value for c>u. If you provide partial information and c>u cannot be calculated the
pro$ram will tell you that there is a problem in that row.
1ou may chose for each component a distribution from the supported distribution drop%
down; first chose a distribution and then ri$ht%clic- on the required row in the
&omponents Table.
1ou may chose a component description from the FSu$$ested &omponentsG drop down
that is a reminder of possible components. 0i$ht%clic- on the required row as above.
&lic- Description in the menu to fill%in additional info about the measurement.
.t this time i$nore the E'uation Editor.
To calculate a field in this table write
2e:pression
in that field.
3or e:ample 22>'
.lso5 a calculation can include the mean value of the component. 3or e:ample;
21Bmean+',
where ' is the row number and mean+', is the value of the mean column in row '.
To add a comment to a field5 ri$ht%clic- on it.
The left%most column of this table shows an inde: of the format "i5 where i is the
component number. This is used to reference that row in other tables and calculations as
will be e:plained later.
1ou can choose a supported distribution from the drop down distribution list. Do to so5
chose a distribution and then ri$ht clic- on the component row "i and chose Insert
Distriution from the popup menu.
1ou can choose a su$$ested uncertainty component from the drop down of su$$ested
components list. Do to so5 chose an item from the list and then ri$ht clic- on the
component row "i and chose Insert $uggested "omponent from the popup menu.
3.1.1 $upported distriutions
.s of today5 the supported distributions are
Data
Student
Hormal
0ectan$le
Trian$le
"%Sine
Two%pea-
The distribution names must be type e:actly +upperBlower case, as stated or chosen from
the drop down list as described above.
The Data distribution is the empirical distribution of the actual measurements. 3or more
on this see the para$raph on the (easured Data Tale below.
The "%Sine is the " distribution.
The Two%pea- distribution is a two%delta function distribution +two narrow spi-es,.
9hen choosin$ a distribution5 an icon ne:t to the distribution drop down list is displayed
showin$ the form of the probability function is shown in a little window.
3.2 (easured Data #repeataility& Tale
The #easured Data Table is used when repeated measurements are to be incorporated
into the uncertainty bud$et. To add the contribution of the repeated measurements5 first
decide which component in the "omponents Tale it will be. To do so5 in the first row
of the (easured Data Tale write the number of "i. That is5 if you chose to ma-e the
measurement component '5 write ' in the first row. 9hen you do so5 in the header the
title FCine "'G will appear. Startin$ from the 2
nd
row5 enter all the measured data as a
column. 9hen you clic- calc +any ISO5 #onte%&arlo or one of the trian$le buttons, I row
' in the "omponents Tale is )illed +note that if not all required fields are filled in the
pro$ram will complain,. The default calculation was done assumin$ that the repeated
measurements are normally distributed +it will add student distribution and de$rees of
freedom accordin$ly,. This can be overridden by fillin$ a different distribution5 e.$. Data
distribution I in which case the calculation will be different.
.lso5 the calculated component in +row "' of the above e:ample, is the standard
deviation of the mean of the used data. 1ou can choose a different statistic and even use a
calculation on the data before it is used. This is useful when the data is $iven in nonlinear
units5 li-e d*5 which should be converted before calculatin$ the standard deviation.
The available other than standard deviation of the mean +sdm, statistics are
sdm I standard deviation of the mean
ssd I sample standard deviation
rms I root mean square
.dd the above to-ens after the row number5 e.$. 2 ssd
9ill use the data in row 2 and calculate the uncertainty as the standard deviation.
To apply a formula to the data before the statistic calculation follow this e:ample;
2 ssd2+,B2
where the formula is to the ri$ht of F2G and the empty +, stand for the data.
This e:ample will first divide the data by 25 calculate ssd and put the result into row 2.
The e:pression
210J++,B10,
will convert from d* to ratio5 calculate the standard deviation of the mean and use the
result in row .
To do more complicated calculations on the data5 use instead the Data window on the
main board menu. 9hen done5 import the data in that table to the (easured Data Tale
and Eust edit the row numbers.
9hen the chosen distribution for the measured data is Data* the standard deviation +and
other statistics, are derived from the actual data. 9hen computin$ uncertainties usin$
(onte+"arlo simulation +see below, with the Data distribution5 the simulated values are
the actual measured data points.
If there is only one measured data point5 do not use the #easured Data Table. Instead edit
a row for this component and provide the uncertainty from prior or other -nowled$e. The
measured value can be written in the (ean ,alue field in the "omponent Tale.
3.3 "orrelations Tale
The "orrelation Tale is used to enter or calculate correlation coefficient for the
components. Hote; To reveal a lar$er part of this table dra$ up the button Drag up to
si!e.
Hote that the rows in this table are inde:ed usin$ the format ri5 where I is the inde:
number.
To enter a -nown correlation coefficient in the table5 write in the first two columns the
component row numbers and in the '
rd
the correlation coefficient value. 3or e:ample ;
Ui Uj rij(correlation) x
1 2 0.
To compute the correlation coefficient of two data sets5 e.$. two columns in the #easured
Data Table5 which correspond to two components +e.$.1 and 2, write;
Ui Uj rij(correlation) x
1 2 :
where : indicates that a computation is required.
The pro$ram will use the correlation data in the calculations.
3.- .esult Tale
The .esult Tale is where the calculated results are written by the pro$ram.
It includes two parts5 the upper is the real data +the re$ular outcome, and the lower is the
comple: part. In most simple and non%comple: calculations only the real part is
calculated. See below more about comple: uncertainties.
The result table includes also the calculated deviation or mean value of the final result. If
the ISO !"# calculation or the #onte%&arlo simulation from the table +see below, is
used5 the deviation from the mean value is calculated. 9hen the #onte%&arlo is
simulated directly from the measurement equation5 the mean value of the outcome is
calculated.
1ou may choose the confidence level for which the result uncertainty will be calculated.
&hose the supported confidence from the drop down window above the 0esults Table.
3./ 0rinting and e1porting data
9hen printin$ you may choose which parts of the uncertainty window to print and which
pa$es to print. To chose clic- 0rinting20review options in the file menu. Tic- the
options. He:t5 in the file menu chose between a fast print and a preview on the #ain
*oard. 3ast Krint is for a standard simple printin$. 9hen previewin$ on the #ain *oard5
a formattin$ can be used with the &haracters and *oard menus.
.nother possibility is to $elect all from the edit menu of the main board +after
previewin$ to the board,5 copy%paste the te:t to a word processor and print from there.
.ll tables can be copy copy%pasted to a spreadsheet li-e #S (:cel in both directions.
-. "omple1 Uncertainty "omponents
&omple: components may be used in the calculations. 3or each such a component you
must devote two rows5 one for its real part and one for the ima$inary part. In each row of
the &omponents Table only the c sensitivities and therefore also &>" can be comple:. .ll
other numbers must be real. This restriction is used to facilitate the computation of
uncertainty directly from the measurement equation.
To indicate that a c value +a derivative, is comple: use E ne:t to its ima$inary part5 e.$.
'?2E.
If the pro$ram encounters comple: or ima$inary numbers5 the results are calculated
twice5 for the real and ima$inary parts5 as described above in the 0esult Table section.
/. (onte+"arlo $imulation )rom the uncertainty tales
(onte+"arlo simulation is a calculation based on $eneratin$ possible outcomes for each
of the components in the "omponents Tale in many iterations. This process is repeated
many times and a distribution of the outcomes is obtained. .ll the uncertainties are
calculated directly from this distribution.
9hen usin$ the tables for a #onte%&arlo Simulation5 The actual distributions $iven in the
table are used. Lowever the assumption is that the measurement equation is Eust the sum
of all components. 9ith this respect5 this type of #onte%&arlo is a compromise between
. simulation directly from the measurement equation and the ISO !"# simple
approach.
3. (onte+"arlo $imulation directly )rom the measurement e'uation
The measurement equation +see equation 1 above, relates the components to the final
measurement result. . mote%&arlo simulation inserts possible values for the components
and calculates the outcome of the equation many times5 for many possibilities of each
component. . probability distribution for the outcome is $enerated from these trials and
the uncertainty is calculated from that distribution.
To write a measurement equation represent the components in the equation usin$ u+i,5
where i is the component row number in the &omponents Table.
3or e:ample5 if the measurement equation is
y2:1?:2B:'
you should write
u+1,?u+2,Bu+',
To be able to do the calculation5 you must provide in the &omponents Table the mean
value for each component. 3or e:ample if a component is the temperature and it is
specified as 2'M N'M&5 the uncertainty is ' +with the ri$ht distribution, and the
temperature mean value is 2'.
To write the measurement e'uation clic- on Description button. . window opens and
its lower part is the equation editor.
To calculate the uncertainty clic- "alculate in this window.
Once the equation is there5 you can calculate the c sensitivities. &lic- on c%Sensitivity in
the menu +of the Description window,.
The two clic-s +&alculate and c%Sensitivities, are combined in the yellow trian$le button
ne:t to the Description button in the main uncertainty window.
9hen there is a measurement equation in the (quation (ditor5 then when calculatin$ per
ISO !"# or #onte%&arlo from the table5 the c%sensitivities are calculated first from the
equation.