UNIVERSITY OF TECHNOLOGY

LULEA
Dept. of Computer Science and Electrical Engineering

Course E7021E
Date
2009-10-20
Time
15:00-19:00

Exam in:
Measurement Technology & Uncertainty Analysis
Attending teacher: Johan Carlson (070-580 82 52)
Problems:
5 (5 points per problem)
Tools allowed:
BETA (Mathematics Handbook), Physics handbook,
Language dictionary, calculator
Text books: Principles of Measurement Systems, by John Bentley
Introduction to Empirical Model Building and Parameter Estimation,
by Johan E. Carlson

1. A displacement sensor has an input range of 0.0 to 3.0 cm and a supply voltage Vs = 0.5
V. Results from a calibration experiment are given in the table below
Displacement x (cm) 0.0
Output voltage (mV) 0.0

0.5
16.5

1.0
32.0

1.5
44.0

2.0
51.5

2.5
55.5

3.0
58.0

(a) Calculate the maximum non-linearity as a percentage of the full-scale deflection

(f.s.d.), assuming the steady state sensitivity is calculated as in the book, i.e.
(2p)
K=

OMAX OMIN
IMAX IMIN

Solution: The slope of the straight line is

K=

58.0 0.0
mV/cm 19.3 mV/cm
3.0 0.0

The intersection is (page 10, below Eq. 2.2)

a = OMIN K IMIN = 0.0

58.0
0.0 = 0.0
3.0

So, the ideal straight line has the equation

OIDEAL =

58
I
3

The non-linearities, given by the outputs in the table are thus

N (I) = O (I) KI,
which becomes
1

Displacement x (cm) 0.0 0.5 1.0

1.5 2.0
2.5
3.0
Output voltage (mV) 0.0 16.5 32.0 44.0 51.5 55.5 58.0
Ideal straight line
0
9.67 19.33 29.0 38.67 48.33 58.0
Non-linearity
0.0 6.83 12.67 15.0 12.83 7.17 0.0
b = 15.0, which in terms of the full-scale
We see that the maximum non-linearity is N
deflection (O = 58.0) is
15
100 25.9%
58 0.0
(b) The performance of the system can easily be improved by instead fitting the optimal
straight line, using the principle of least-squares. Give the necessary equations for
finding the slope K and the intersection a of the straight line using the data in the
table above.
Note: You do not need to solve for the actual numerical values of K and a.
(2p)
Solution: Fitting the straight line by the method of least squares means solving the
following over-determined system of equations
O = Ia,
where

O =

0
16.5
32
44
51.5
55.5
58

,I =

1
1
1
1
1
1
1

0
0.5
1
1.5
2
2.5
3




,a = a .

This gives the least-squares estimate of a and K as






1 T
ab
7.696
T
b
a= b = I I
I O
19.393
K
(c) Explain why the method in (b) gives a better average result.
Solution: Since the principle of least-squares minimizes the sum of squares of the
errors over the entire range, the result is better than for the straight line fit in (a),
where only two of the calibration points are used.
(1p)
2. A temperature measurement system consists of the following elements

where is the true temperature and M is the measured temperature (in Kelvin). The
model equations and the corresponding uncertainties are given in the table below.
Thermistor
 

4
= 5 10 k

R = K1 exp

Model equations
Mean values

K1
= 3 103 K
Standard deviations K1 = 0.5 104
= 0

Bridge 
 Recorder
M = K2VO + a2
VO = VS 1+13.3 a1
R

V S = 3.00 V
a1 = 0.77
VS = 0.03
a1 = 0.01

K 2 = 50.0 K/V
a2 = 300 K
K2 = 0
a2 = 3.0

(a) Calculate the mean output M and the mean error E = M for an input temperature of 320 K.
(2p)
Solution: Mean output (at 320 K):
!



3 103
4
R = K 1 exp
= 5 10 exp
5.895 k

320
!
!
1
1
V O = VS
0.77 V 0.3867 V
a1 = 3
3.3
1 + 3.3
1
+
5.895
R

M = K 2V O + a2 = 50.0 0.3867 + 300 K 319.34 K

The mean error is then
E = M = 320 319.34 K 0.66 K
(3p)
(b) Calculate the standard deviation of the output error E for an input of 320 K.
Solution: The standard deviation of the error is the same as the standard deviation
of the output, since the standard deviation only depends on the uncertainty of the
elements, not the mean values of them. Doing the error propagation calculations we
get that
2R


=

V2O

R
K1


=
=

2

VO

2K1

2

 2

2

2

3 103
2
= exp
K1 = exp
0.5 104 0.3475

320

2R +

VO
Vs

2

V2s +

!2

330Vs
2

(10R + 33)

VO
a1

2

1
a1
1 + 3.3
R

2R +

2a1 =

!2
V2s + (Vs )2 2a1 0.00257




M 2 2
M 2 2
=
a2 +
VO 15.4265
a2
VO
This gives the standard deviation of the error as
q
M = 2M 3.93 K
2M

3. A force measurement system consisting of a piezoelectric crystal, charge amplifier and

recorder is shown in the figure below.

Calculate the system output and the corresponding dynamic error when the force input
signal is
(5p)


1
1
F (t) = 50 sin 10t + sin 30t + sin 50t .
3
5
Solution: The overall system transfer function is (the product of the three transfer functions in the figure)

 3 


10
50
= Inserting the numbers =
G (s) = 20
1 2
1 + s
s + 2n s + 1
2n
!

 3
50
10 0.1
= 20
=
1 2
1 + 0.1s
s + 20.2
50 s + 1
502
0.1
=
(1.0 + 0.1s) (.0004s2 + .008s + 1.0)
Now, replacing s with j where is the angular frequency of the input, we can compute
the output by the principle, sinusoid in gives sinusoid out, i.e. the output signal is given
by
|G ( j)| sin (t + ]G ( j)) ,
where |G ( j)| is the absolute value of the transfer function and ]G ( j) is the phase of
the transfer function. Doing this for each of the sinusoids in the input signal we obtain






0.1

  =
|G ( j)| = 
 (1.0 + 0.1 j) .0004 ( j)2 + .008 j + 1.0 
0.1


= 

2
(1.0
+
0.1
j)
.0004
(
j)
+
.008
j
+
1.0


4

n

o
Re (1.0 + 0.1 j) .0004 ( j)2 + .008 j + 1.0

o
]G ( j) = tan1 0.1 tan1 n
Im (1.0 + 0.1 j) .0004 ( j)2 + .008 j + 1.0

o
n
2
Re (1.0 + 0.1 j) .0004 ( j) + .008 j + 1.0

o
= tan1 n
2
Im (1.0 + 0.1 j) .0004 ( j) + .008 j + 1.0

4. An ultrasonic transit-time flowmeter is used to measure the average flow velocity v (m/s)
through a pipe.
Transducer B

average velocity, v (m/s)

Transducer A

A short ultrasonic pulse is first transmitted from Transducer B to Transducer A and then
from Transducer A to Transducer B. The cross-correlation function R() between the two
pulses is shown in the figure below.
8
6

R()

4
2
0
2
4
6
30

20

10

0
10
lag, [samples]

The flow velocity is given by

v=
5

T c2
,
2D cot

20

30

40

where is the angle between the transducers and the center axis of the flow, c is the speed
of sound through the fluid, and D is the pipe diameter.
Assuming that = /6 rad, c = 1480 m/s, D = 10 cm, and the sampling frequency of the
A/D converter used to measure the pulses is fs = 10 MHz:
(a) Determine the average flow velocity, v.
(2p)
Solution: We know that the cross-correlation has its maximum at the lag corresponding to the time delay between the pulses. From the figure we see that
Tsamp = 6 samples. This corresponds to a time delay of
T =

Tsamp
6
= 7 seconds = 6 107 s = 0.6 s.
fs
10

Using this together with the numbers given in the problem, we obtain the average
flow velocity as
v=

6 107 14802
T c2
=
m/s 3.79 m/s
2D cot 2 0.1 cot (/6)

(b) In reality, the speed of sound, c, is temperature dependent. Assuming it can be

modeled as a second-order polynomial function of temperature, make the necessary
modifications to the equation for the flow velocity.
(1p)
Solution: Assuming that the speed of sound can be modeled as
c (T ) = a0 + a1 T + a2 T 2 ,
the expression for the average flow velocity becomes
T a0 + a1 T + a2 T 2
v=
2D cot

2

(c) Assuming there is some uncertainty regarding the mounting of the transducer, modeled as = 0.01 rad, what is the effect on the overall uncertainty of the measured
flow velocity?
Note: You do not need to estimate the total uncertainty, only the contribution of the
transducer angle uncertainty.
(2p)
Solution: The sensitivity of the velocity measurement to the angle of the transducer
is given by


2

2

T c2
2D cot

!2
T c2 1 + cot2
2
2
2D cot

2 =

For a deviation of 2 = 0.02 rad, this would mean (for the example in (a)), that the
uncertainty is increased by

!2
7 14802 1 + cot2 (/6 + 0.02)
6

10
2v =
(0.01)2 8.05 103 ,
2
2 0.1 cot (/6 + 0.02)
meaning that the total standard deviation is increased by
v 8.97 102 m/s
5. We have a differential pressure flowmeter setup designed for incompressible fluids. The
volume flow rate is given by
s
2 (P1 P2 )
A2
Q= r
,
 2

A2
1 A1
where A1 and A2 are the cross-sectional areas of the pipe where the pressures P1 and P2
are measured, respectively. The problem now is that we do not know the fluid density.
Assuming we can control the volume flow rate, Q to within some uncertainty, i.e. the
flow rate can be said to be normally distributed as N (Q, Q ), derive an expression for the
least-squares estimator of the flow rate, using the Gauss-Newton linearization method,
that based on the calibration measurements also estimates the unknown fluid density .
(5p)
Solution: Given that the cross-sectional areas are constant, and only the pressures change,
lets assume that the volume flow rate we are actually able to set is
ym = Qm (P1 , P2 ; ) + em ,
where
em N(0, Q ).
b so that the sum of
We then have the problem of finding an estimate of the flow rate, Q
squares of the errors is minimized. For repeated calibration measurements, this can be
written on vector form as

y1
Q1 (X1 ;)
e1
y2
Q2 (X2 ;) e2

y = .. = f (X;) + e = ..
+ .. ,
.
.
.
yM
QM (XM ;)
eM
where


T
X = P1,m P2,m
.

The subscript n denotes pressures measured for experiment m, where m = 1, 2, . . . , M.

Now, assuming we have an initial guess of the density, lets say b
= 0 . We can then use
7

the Gauss-Newton linearization method to find the estimate of that miminizes the sum
of squares of the errors

T 

b
b .
J = yQ
yQ
The iteration is given by (see text book)



1 T
b (b
b
H (b
i ) y Q
i ) ,
i+1 = b
i + HT (b
i ) H (b
i )
where H () is the gradient of the model with respect to the unknown parameter , i.e.

h1
Q (X;)
h2
,
H () =
=

hM
where

2 (P1,m P2,m )
d
r A2
=

 2
d

1 AA21
s
d
Lm
Lm K
=
K
= 2q ,
Lm
d

hm =

where
A2
 2
1 AA21

K=r

Lm = 2 (P1,m P2,m )
So, for each element in the gradient vector, the only things that vary are the measured
pressures.