Você está na página 1de 11

5.

8 COMBINING UNCERTAINTIES 203

Example 6
The mass, m, of a wire is found to be 2.255 g with a standard uncertainty of 0.032 g. The
length, l, of the wire is 0.2365 m with a standard uncertainty of 0.0035 m. The mass per unit
length, , is given by
m
: (5:36)
l
Determine the:
(a) best estimate of ;
(b) standard uncertainty in .

ANSWER
(a) Using equation 5.36, ml 0:2365
2:255
9:535 g=m.
(b) Writing equation 5.33, in terms of , m and l, gives
 2  2
@ @
u2c um ul : (5:37)
@m @l

From equation 5.36,

@ 1 1
4:2283 m1 ;
@m l 0:2365
@ m 2:255
and  2  40:317 g=m2 :
@l l 0:23652
Substituting values into equation 5.37 gives,
u2c 4:2283  0:0322 40:317  0:00352
0:01831 0:01991 0:03822g=m2 ;
therefore,

uc 0:1955 g=m:

Exercise J
(1) Light incident at an angle, i, on a plane glass surface is refracted as it enters the glass.
The refractive index of the glass, n, can be calculated by using
sin i
n ;
sin r
where r is the angle of refraction of the light.
The best estimate of i is 52.0 with a standard uncertainty of 1.0 . The best estimate
of r is 32.5 with a standard uncertainty of 1.5 .
Use this information to calculate the best estimate of n. Assuming errors in i and r
are not correlated, determine the combined standard uncertainty in n.
204 5 MEASUREMENT, ERROR AND UNCERTAINTY

y dy
Note, for y = sin x, the approximation x  dx is valid for x expressed in radians.
(2) The focal length, f, of a thin lens is related to the object distance, u, and the image
distance, v, by the equation
1 1 1
:
f u v
In an experiment the best estimate of u is 125.5 mm with a standard uncertainty of
1.5 mm. The best estimate of v is 628.0 mm with a standard uncertainty of 1.5 mm.
Assuming that the errors in u and v are not correlated,

(i) calculate the best estimate of the focal length and its standard uncertainty;
(ii) calculate the best estimate of the linear magnication, m, of the lens and the standard
uncertainty in m, given that
v
m :
u
(3) The electrical resistivity, , of a wire may be determined by using
RA
; (5:38)
l
where R is the electrical resistance of the wire measured between its ends, l is the length
of the wire and A is the cross-sectional area of the wire.
For a given metal wire it is found that the:

best estimate of its length is 1.21 m with a standard uncertainty of 0.012 m;


best estimate of its diameter is 0.24 mm with a standard uncertainty of 0.015 mm;
best estimate of its resistance is 0.52 with a standard uncertainty of 0.022 .
Using this information determine the

(i) best estimate of A;


(ii) standard uncertainty in A;
(iii) best estimate of ;
(iv) standard uncertainty in .

(4) Figure 5.11 shows a at belt wrapped around a drum. If the tension in the belt, T1, is
greater than T2 then the drum will rotate clockwise so long as the friction between the belt
and the drum is large enough. The relationship between the coecient of static friction, s
between belt and drum, and the two tensions, T1 and T2, can be written33
 
1 T1
s ln ; (5:39)
T2
where is the angle dened in gure 5.11.

T1 = 412 N with a standard uncertainty of 52 N;


T2 = 81 N with a standard uncertainty of 13 N;

33
See Blau (2009).
5.8 COMBINING UNCERTAINTIES 205

= 2.15 rad with a standard uncertainty of 0.15 rad.

Assuming errors in T1, T2 and are uncorrelated, use this information to nd the
best estimate of s and the combined standard uncertainty in the best estimate of s.

5.8.3 Combining Type A and Type B evaluations of uncertainty

Equation 5.33 may be used when we need to combine uncertainties deter-


mined by Type A and Type B evaluations.
Consider the situation in which a digital thermometer with a resolution of
0.1 C is used to measure the temperature of air in a sealed enclosure.
Table 5.6 contains eight repeat measurements of the air temperature.
The best estimate of the true temperature can be written

X Z ; (5:40)
where X is the mean of the values of temperature in table 5.6. This is equal to
the best estimate of the true temperature in the absence of any systematic
errors. Using data in table 5.6,
X 32:148 C:
The standard deviation, s, of the value in table 5.6 is found using equation 1.16,
s 0:1408 C:
p p
Standard uncertainty in X, uX s= n 0:1408= 8, i.e.
uX 0:0498 C:
Z in equation 5.40 is a correction term which takes into account errors due
to, for example, calibration error or resolution error. Here we restrict the

Table 5.6. Values of air temperature inside a sealed container.

Temperature ( C) 32.3 31.9 32.1 32.3 32.1 32.2 32.2 32.0

T1

T2

Figure 5.11. Flat belt wrapped around a drum.


206 5 MEASUREMENT, ERROR AND UNCERTAINTY

Table 5.7. Values of the mass of a steel ball.

Mass (g) 7.054 7.053 7.053 7.054 7.053 7.053 7.052 7.054

determination of Z and the standard uncertainty in Z to consideration of the


resolution error. The resolution error could be positive or negative. As there is
no reason to favour either sign, we take Z = 0, so that the best estimate of the
temperature is

32:148 0 C 32:148 C:
The resolution, , of the thermometer is 0.1 C, so the half-width a = /2 = 0.05 C.
Using equation 5.20,
a 0:05
uZ p p 0:0289 C:
3 3
@ @
We now apply equation 5.33, noting that @X
@Z
1;

u2c u2 X u2 Z 0:04982 0:02892 0:00331 C2 :


It follows that

uc 0:058 C:
To summarise, the best estimate of the true value of the temperature is 32.148 C
with a combined standard uncertainty of 0.058 C.

Exercise K
A steel ball of nominal diameter 6.35 mm is weighed on a top-loading balance which has
a resolution of 0.001 g. The balance is zeroed between each measurement, so that you
can assume that any zero error is negligible.
The values obtained are shown in table 5.7.
Using the data in table 5.7, determine the best estimate of the true value of the mass
of the steel ball, and the combined standard uncertainty in the best estimate.

5.8.4 Do we need to combine all uncertainties?

Strictly, uncertainties in all input quantities need to be established and brought


together when determining a combined standard uncertainty. Careful consid-
eration of all sources of uncertainty might lead to a redesign of an experiment to
reduce the uncertainty. However there are situations, for example when the
variability observed when making repeat measurements of a quantity is far
larger than the resolution of the instrument used to make the measurements,
5.8 COMBINING UNCERTAINTIES 207

45

Percentage increase in uc (y)


40
35
30
25
20
15
10
5
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
u(z)/u(x)

Figure 5.12. Increase in uc(y) resulting from including u(z) in the calculation represented by
equation 5.42.

that the contribution to the combined uncertainty due to instrument resolution


can safely be ignored. Consider equation 5.41 which relates the best estimate of
a quantity, y, to the best estimate of quantities x and z:

y x z: (5:41)

Assuming uncertainties in x and z are uncorrelated, then the combined stand-


ard uncertainty can be calculated by using
u2c y u2 x u2 z: (5:42)
A consequence of combining uncertainties by summing the squares of uncer-
tainties as given by equation 5.42 is that if one uncertainty is markedly smaller
than the other, then the smaller uncertainty may be neglected.
Assuming that u(z) < u(x), at what stage can we neglect u(z) altogether in
the calculation of uc(y)? Figure 5.11 shows how much larger is uc(y), expressed
as a percentage when uncertainties in both x and z are included, compared
when the uncertainty in z is neglected. The horizontal axis of gure 5.11
represents the ratio u(z)/u(x), which ranges from zero (when u(z) = 0) to 1
(when u(z) = u(x)).
Inspection of gure 5.12 shows that including u(z) into the calculation of
uc(y), will only increase uc(y) by about 12% when u(z) is half of u(x). Though
there are no international guidelines about when to neglect an uncertainty
when calculating the combined standard uncertainty (and we could argue for
completeness, all uncertainties should be included), the example given here
shows that for u(z)/u(x) <0.25, the eect on the combined standard uncertainty
of neglecting u(z) is negligible.
208 5 MEASUREMENT, ERROR AND UNCERTAINTY

Table 5.8. Coverage factors in Type A evaluations for degrees of freedom when the level of
condence is 95%.

1 2 3 4 5 6 7 8 9 10

k 12.706 4.303 3.182 2.776 2.571 2.447 2.365 2.306 2.262 2.228

5.9 Expanded uncertainty


The combined standard uncertainty, uc, is the basic measure of uncertainty in
the GUM. However, there are situations in which it is preferable to quote a
coverage interval within which there is a specied probability (often 0.95) that
the true value will lie. When this probability is expressed as a percentage, it is
referred to the level of condence.
Specically, we can write

y  U Y y U; (5:43)

where Y is the true value of the quantity, y is the best estimate of the true value,
and U is the expanded uncertainty. If there is a probability of, say, 95% that the
true value lies in the interval given by equation 5.43, then it is referred to as the
95% coverage interval.
Equation 5.43 may be written

Y y  U: (5:44)
The expanded uncertainty, U, is related to the combined standard uncertainty,
uc, by the equation:

U kuc ; (5:45)

where k is the coverage factor.

5.9.1 Coverage factor and Type A evaluations of uncertainty

If uc is determined through a Type A evaluation of uncertainty, then it is usual to


assume that the t distribution may be applied when determining the coverage
factor, k.
When the level of condence is 95% (i.e. the probability that the true value
lies within a specied interval is 0.95), then table 5.8 gives the coverage factor
for various degrees of freedom, . For values of > 10, k tends towards a value of
close to 2. When > 10, experimenters often use 2 as the coverage factor when
the level of condence is 95%.
5.9 EXPANDED UNCERTAINTY 209

Example 7
In an experiment to calibrate a 1 mL pipette, the mass of water dispensed by the
pipette was measured using an electronic balance. Ten measurements were made
of the mass of water dispensed by the pipette. The mean mass of water was found
to be 0.9567 g, with a standard error in the mean of 0.0035 g. Using this information,
determine the:

(i) best estimate of the true value of the mass of water dispensed by the pipette;
(ii) standard uncertainty in the mass, u;
(iii) coverage factor, k, for a 95% level of condence;
(iv) expanded uncertainty, U, for a 95% level of condence;
(v) 95% coverage interval for the true value of the mass of water.

ANSWER
(i) The best estimate is taken to be the mean of values obtained through repeat measure-
ments, which in this example is 0.9567 g.
(ii) When Type A evaluations are carried out, the standard uncertainty is equal to the
standard error in the mean, i.e. u = 0.0035 g.
(iii) In this example, the number of degrees of freedom, = 10 1 = 9. Using table 5.8, the
corresponding value for k is 2.262.
(iv) U = ku = 2.262 0.0035 = 0.0079 g.
(v) Using equation 5.44, the 95% coverage interval is written (0.9567 0.0079) g.

Exercise L
The calcium content of ve samples of powdered mineral is determined. The mean of the
ve values is (expressed as percent composition) 0.02725 with a standard uncertainty
(obtained through a Type A evaluation) of 0.00012. Determine the expanded uncertainty
at the 95% level of condence.

5.9.2 Coverage factor and the WelchSatterthwaite formula

The combined uncertainty in the best estimate of a quantity will, in general, be


determined from Type A and Type B evaluations of uncertainty.
It is possible that the best estimate of each quantity is obtained from a
dierent number of repeat measurements. For example, the resistivity of a
metal is obtained from knowledge of the resistance, length and diameter of
a wire made from the metal. Suppose ve measurements are made of the
resistance of the wire, seven of its diameter and four of its length. We are able
210 5 MEASUREMENT, ERROR AND UNCERTAINTY

to determine the combined uncertainty using equation 5.33, but how are we
able to calculate the coverage factor for the expanded uncertainty? In this
situation we calculate the eective number of degrees of freedom, e, using
the WelchSatterthwaite formula,34 which can be expressed:

u4c y
eff ; (5:46)
P
N 4
ui y
i
i1

where i is number of degrees of freedom associated with each quantity con-


tributing to the calculation of the combined standard uncertainty.
We can calculate u4c y with the aid of equation 5.34. Expanding the
denominator of equation 5.46 gives

X
N
u4 y u41 y u42 y u43 y
i
:
i1
i 1 2 3

If y depends on quantities a and b, and c then


 4  4  4
@f @f @f
u41 y ua ; u2 y
4
ub and u3 y
4
uc :
@a @b @c

To avoid overestimating e and thereby underestimating the coverage factor,


e is rounded down to the nearest whole number.

Example 8
The optical density (o.d.) of a uid is given by:
o:d: Cl; (5:47)
where is the extinction coecient, C is the concentration coecient of the absorbing
species in the liquid, and l is the path length of the light. The best estimate of each
quantity, standard uncertainty and number of degrees of freedom are as follows.

14:9 L  mol1  mm1 ; u 1:2 L  mol1  mm1 ; v 5


C 0:042 mol  L1 ; uC 0:003 mol  L1 ; vc 7
l 1:42 mm; ul 0:21 mm; vl 8

Determine:

(i) the best estimate of the optical density, o.d.


(ii) the combined standard uncertainty;
(iii) the eective number of degrees of freedom, e;
(iv) the coverage factor, assuming that the level of condence is 95%.

34
See Dietrich (1991).
5.9 EXPANDED UNCERTAINTY 211

ANSWER
(i) o:d: Cl 14:9  0:042  1:42 0:889 (note that optical density has no units).
(ii)
2  2  2
@f @f @f
u2c y
u uC ul
@ @C @l
@f @f @f
Cl; l; C:
@ @C @l
Now we write
 2
@f
u21 y u Clu2 0:042  1:42  1:22 0:0051219;
@
 2
@f
u22 y uC luC2 14:9  1:42  0:0032 0:0040289;
@C
 2
@f
u23 y ul Cul2 14:2  0:042  0:212 0:017271:
@l
Using equation 5.34, this gives

u2c y u21 y u22 y u23 y 0:0051219 0:0040289 0:017271 0:026422:

It follows that uc = 0.16255.


Note that u1= 0.071568, u2= 0.063474, and u3= 0.131418.
(iii) To nd the eective number of degrees of freedom, e, we use equation 5.46, such
u4c y 4
0:16255
that35 eff 0:071568 0:063474 0:131418 15:57round to 15:
P
N
u4 y
i
4

7
4 4

5 8
i
i1
(iv) When the number of degrees of freedom is 15, the coverage factor for a level of
condence of 95% is 2.131 (see table 2 in appendix 1).

5.9.3 eff and Type B evaluation of uncertainties

What number of degrees of freedom do we assign to a standard uncertainty that


is determined by a Type B evaluation based on an assumed probability dis-
tribution, such as the triangular or rectangular distribution? A rectangular
distribution of half width a, has a standard uncertainty given by
p
u a= 3: (5:48)

The standard uncertainty given by equation 5.48 has no uncertainty, i.e. u = 0.

35
Because we are raising numbers to the fourth power, it is prudent to keep numbers to extra
gures in the calculations in order to reduce the eect of rounding errors.
212 5 MEASUREMENT, ERROR AND UNCERTAINTY

It is possible to show that the number of degrees of freedom can be


determined from knowledge of the standard uncertainty, u and standard devia-
tion of the standard uncertainty, u (i.e. the uncertainty in the uncertainty!)
If the uncertainty in u is u, then the number of degrees of freedom, , can
be written36

1 u 2 1 h u i2
: (5:49)
2 u 2 u

If u = 0 as is the case for an assumed rectangular distribution, then using equation


5.49, it follows that = . The fact that = does not cause a diculty when
determining the eective number of degrees of freedom using equation 5.46, as
any term in the equation with an innite number of degrees of freedom is zero.

Example 9
A 3 digit voltmeter is used to measure the output voltage of a transistor. The mean of ten
repeat measurements of voltage is found to be 5.340 V with a standard deviation of 0.11
V. Given that the voltmeter is operating on the 20 V range and the operating manual
accompanying the voltmeter states that the accuracy of the voltmeter is 0.5% of the
reading + 1 gure, determine:

(i) the standard uncertainty in the voltage determined through a Type A evaluation of
uncertainty;
(ii) the standard uncertainty in the voltage determined through a Type B evaluation of
uncertainty;
(iii) the combined standard uncertainty;
(iv) the eective number of degrees of freedom using the WelchSatterthwaite formula;
(v) the coverage factor;
(vi) the expanded uncertainty at the 95% level of condence.

ANSWER
We begin by recognising that both the variability in the values and the stated accuracy of
the instrument will contribute to the combined standard uncertainty in the voltage. We
begin by writing the best estimate of the true value of the voltage as

V X V Z V :
XV is the mean of repeat measurements (and so the standard uncertainty in XV will be
determined by a Type A evaluation of uncertainty). ZV is a correction term to account
for errors due, to for example, osets or instrument accuracy. In this example we will use

36
See the GUM, section E.4.3.
5.10 RELATIVE STANDARD UNCERTAINTY 213

data supplied by the manufacturer to determine the standard uncertainty in ZV (and so


this is a Type B evaluation of uncertainty).

(i) The standard uncertainty in XV, u(XV) psn. In this example, s = 0.11 V and n = 10, so
p 0:03479 V:
u(XV) 0:11V
10
(ii) We assume that the accuracy of the voltmeter can be represented by a rectangular
distribution with half width of 0.5% of the reading + 1 gure, 0.5% of 5.340 V is 0.0267 V,
and the least signicant gure for a 3 digit voltmeter on the 20 V range is 0.01 V. The
half width of the distribution, a = 5.340 0.005 + 0.01 = 0.0367 V.
The standard uncertainty u(ZV) is found by using equation 5.20,
i.e. uZV pa3 0:0367
p 0:0212V
3
(iii) The combined standard uncertainty is found using equation 5.33,
i.e. u2c V u2 XV u2 ZV 0:034792 0:02122 0:00166V 2 .
So uc(V) = 0.04073 V.
u4c V 0:040734
(iv) eff 16:9 round to 16:
P
N
u4i V 0:03479
4

0:0212
4

9 1
i
i1
(v) When the number of degrees of freedom is 16, then coverage factor, k, for a level of
condence of 95% is 2.120.
(vi) The expanded uncertainty, U = kuc(V) = 2.120 0.04073 = 0.086 V (to two signicant
gures).
To summarise: the best estimate of the true value of the voltage is 5.340 V and the
expanded uncertainty at the 95% level of condence is 0.086 V.

5.10 Relative standard uncertainty


In some situations we may prefer the uncertainty to be expressed as a fraction
of the best estimate of the true value of the quantity. This is referred to as the
relative uncertainty. For example, if the best estimate of a quantity is a with a
standard uncertainty of u(a), then the relative standard uncertainty is given by

ua
relative standard uncertainty : (5:50)
j aj

The modulus of the best estimate is used to avoid the fractional uncertainty
assuming a negative value.
Suppose the best estimate of the mass of a body is 45.32 g with a standard
uncertainty of 0.15, then the
0:15 g
relative standard uncertainty 3:3  103 :
45:32 g

Você também pode gostar