Measurement systems error

1
 Errors

1. Random errors 2. Systematic errors

Are errors which can vary in random Are errors which do not vary
manner between successive readings from one reading to another.
of the same quantity.
(a) Construction errors
(occurs while manufacturing the
(a) Operating errors (Ex. errors in instrument)
reading the output device)
(b) Approximation errors (result from
assumptions made regarding the
(b) Environmental errors (due to
relationship between quantities. As
change in temperature, pressure,
linearization)
humidity, …etc)
(c) Aging errors
(c) Stochastic errors (result from (d) Insertion errors (loading errors)
stochastic process such as noise)

2
Uncertainty (Error) analysis
Propagation of uncertainties
A quantity may be determined as a result of calculations carried out on
the results of a number of measurements, each of which has some error
associated with it. If the result O is a given function of the independent
(variables) measurements x1, x2, …, xn as

O  f (x 1 , x 2 ,K , x n )
Then, the uncertainty (overall error) in the quantity O is given as
2 1/2
 O   O  2 2
 O  
 O   e1    e2  L   en  
 x 1   x 2   x n  
Where: δO is the uncertainty in O (overall error), and
ei s the error in the individual measurement xi

3
Uncertainty (Error) analysis
Propagation of uncertainties
O  f (x 1 , x 2 ,K , x n )
Then, the uncertainty (overall error) in the quantity O is given as
2 1/2
 O   O  2
 O 
2

 O   e1    e2  L   en  
 x 1   x 2   x n  

All derivatives are evaluated at the nominal values of xi

O
The overall percentage error is:% eo  100
O
4
Propagation of uncertainties

 In the same way, the total standard deviation is

2
n
 O 
 O    x i 
i 1  x i 

5
Propagation of uncertainties
Example: Assume that we have the following measurements for the voltage E
and the current I for a certain heater:

E  12V  1.5%
I  10  0.2A
Calculate the power of the heater and its uncertainty (error)

Solution:
Nominal values are E0  12V, I 0  10A
Individual errors:
1.5
% eE  1.5  eE  12  0.18V
100
and eI  0.2A
6
Propagation of uncertainties
Power, P  EI
2 12
 P   P  
2
Overall error,  P   eE    eI  
 E   I  
P P
Where  I  10A,  E  12V
E I
 P  [(10  0.18)2  (12  0.2)2 ]1 2  3 W
P
Percentage overall error is %e P  100
P0
Where P0 is the nominal power,

P0  E0 I 0  12 10  120 W P  120  3 W

3
% eP  100  2.5%
120 7
Propagation of uncertainties
Example: The resistance of a certain copper wire is given as

R  r 1   (T  20)

where r  6  0.3% is the resistance at 20C

  0.004C-1  1% is the temperature coefficient of resistance

T  30  1C is the temperature of the wire

Calculate the resistance of the wire and its uncertainty

Solution:
Nominal values are
r0  6
T0  30C
 0  0.004C-1
8
Propagation of uncertainties
The resistance of the wire (nominal resistance) is

R0  r0 1   0 (T0  20)
R0  6 1  0.004(30  20)
 6.24
Individual errors:

0.3
er   6  0.018
100
1
e   0.004  4 105 C1
100
eT  1C
9
Propagation of uncertainties
The uncertainty in resistance of the wire is

2 12
 R   R   R  
2 2

 R   e r    e     eT  
 r      T  
where

R
 1   0 (T0  20)  1  (0.004)(30  20)  1.04
r
R
 r0 (T0  20)  (6)(30  20)  60

R
 r0 0  (6)(0.004)  0.024
T
10
Propagation of uncertainties
 R  [ 1.04  0.018   60  4 10    0.024 1
5 2
2 2 12
]

 0.0305 (Answer)

R  6.24  0.03 05

Percentage uncertainty in the resistance

R0.0305
%e R    0.49%
R0 6.24

11
Propagation of uncertainties

Power measurement across a resistor

12
Propagation of uncertainties
Solution:

R0  10, eR  0.0110  0.1

E0  100V, eE  0.01100  1V
I 0  10A, eI  0.0110  0.1A
Case 1: Calculating the power by measuring the voltage and knowing the
resistance

E2
P (is a function of E and R)
R
Uncertainty in the power is 2 12
 P   P  
2

 P   eE    eR  
 E   R  

13
Propagation of uncertainties
P 2 E 2 100 P E2 (100) 2
   20  2   100
E R 0 10 R R 0
(10) 2

Uncertainty in the power is

12
 P   20 1  100  0.1   22.36W
2 2
 
The nominal power is
E02 1002
P0    1000W
R0 10
The percentage uncertainty in the power measurement is

P22.36
%e P   100  2.236
P0 1000

14
Propagation of uncertainties
Case 2: Calculating the power by measuring the voltage and the current

P  EI (is a function of E and I)

Uncertainty in the power is

2 12
 P   P  
2

 P   eE    eI  
 E   I  

P
Where  I 0  10
E
P
 E0  100
I
15
Propagation of uncertainties
Uncertainty in the power is

12
 P  10 1  100  0.1   14.14W
2 2
 
The nominal power is

P0  E0 I 0  100 10  1000W

The percentage uncertainty in the power measurement is

P
14.14
%e P   100  1.414
P0 1000

16
Propagation of uncertainties
Conclusion:

First method: P  E R2

P  1000  22.36 W
Second method: P  E I  P  1000  14.14 W
Note that the second method of power determination provides considerably less
uncertainty than the first method, even though the primary uncertainties in each
quantity are the same. In this example the utility of the uncertainty analysis is
that it affords the individual a basis for selection of a measurement method to
produce a result with less uncertainty.

Therefore, it is better to select the second measurement method to

determine the power

17
Loading Effects as a source
of systematic error
 loading
When we insert an instrument into the position to measure a quantity, an error
in the measurement may occur due to loading effect.

Examples:

inserting an ammeter into a circuit to measure the current changes the value of
the current due to the ammeters own resistance which changes the total
resistance of the circuit.

Also inserting a thermometer at room temperature into a hot water to measure

its temperature changes the temperature of the water which leads to error in
the temperature measurement.
Loading Effects and Two-port Networks

 One element of the system may modify the characteristics

of the previous one (for example by drawing current).
 Electrical loading
 Thévenin equivalent circuit
 Norton equivalent circuit
 Electrical Loading

in order to get maximum voltage transfer from the network to the load, the load
impedance should be far greater than the Thévenin impedance for the network. In
order to get maximum power transfer from network to load, the load impedance
should be equal to the network impedance
Therefore, the effect of connecting a load across the network yields a loading
error of magnitude
Loading Error  E  V Th L
 ZL 
 ETh 1  
 Z Th  Z L 
Example
Loading a voltmeter

Assume that a voltmeter of resistance Rm is connected across the shown active

circuit to measure the voltage between terminals A and B.

Therefore, the reading indicated by the

instrument (voltmeter) is

Rm
 Rm 
Vm    ETh
 Rm  RTh 

Voltage before the meter

Active Voltmeter was connected
network
23
This means that the error in the reading is

Error  ETh  Vm
 Rm 
 ETh 1  
 Z Th  Rm 
and the accuracy of the voltmeter is

Vm
Accuracy  100%
ETh
 Rm 
   100%
 Z Th  Rm 
Note that if Rm is very large, the error goes to zero and the accuracy goes to
100%
24
Example 3.1
 Reduction of Systematic Errors

Careful Instrument Design

Calibration valid for everything

Method of Opposing Inputs invalid for everything

The method of opposing inputs
compensates for the effect of an
environmental input in a
measurement system by introducing Rcoil
an equal and opposite environmental
input that cancels it out.

compensating resistance Rcomp into the circuit,

where Rcomp has a temperature coefficient equal in
magnitude but opposite in sign to that of the coil
 Reduction of Systematic Errors
High-Gain Feedback
• The effect of environmental inputs on the motor and spring constants
is represented by variables Dm and Ds. In the absence of
environmental inputs,

Xo = KmKs Ei

• in the presence of environmental inputs, both Km and Ks change, and the

relationship between Xo and Ei can be affected greatly it becomes difficult
or impossible to calculate Ei from the measured value of Xo
 Reduction of Systematic Errors

 High-Gain Feedback

Because Ka is very large (it is a high-gain amplifier), Kf Ka Km Ks >> 1,

The sensitivity of the gain constants Ka, Km, and Ks to the environmental inputs
Da, Dm, and Ds has thereby been rendered irrelevant, and we only have to be
concerned with one environmental input, Df.
 Reduction of Systematic Errors

 Signal Filtering (chapter 6)

 Intelligent Instruments (chapter 11)

Intelligent instruments contain extra sensors that measure the value of

environmental inputs and automatically compensate the value of the output
reading.
Calculation of Overall Systematic Error
Statistical Analysis of Measurements Subject to Random Errors

When a set of measurements of the same quantity is

taken, the individual readings will vary somewhat from
each other. If each reading is denoted by xi and there
are n readings, then we use the arithmetic mean (xm)
as an estimate for the measured quantity such that
1 n
xm   xi
n i 1
The deviation (error) di of each measurement xi from
the mean value xm is defined as

di  xi  xm
32
 Statistics
The variance (V) of the measurements is given by
d 12  d 22  L  d n2
V 
n 1
1 n
V  
n  1 i 1
(x i  x m ) 2

The standard deviation (σ) of the measurements

is define as the square root of the variance (V)
12
 1 n

  
 n  1 i 1
(x i  x m ) 2 


33
Statistics
Example: Calculate xm, Vx and σ for measurement sets A, B, and C, and discuss the
results.
Measurement set A, n = 11
398 420 394 416 404 408 400 420 396
413 430
Measurement set B, n = 11

409 406 402 407 405 404 407 404 407

407 408
Measurement set C, n = 23

409 406 402 407 405 404 407 404 407

407 408 406 410 406 405 408 406 409
406 405 409 406 407
34
Statistics
Solution: measurement set A, n = 11. A
xi 398 420 394 416 404 408 400 420 396
413 430
1 n
Mean, xm   xi  409.0
n i 1
di -11 11 -15 7 -5 -1 -9 11 -13
4 21

(di)2 121 121 225 49 25 1 81 121 169

16 441
1 n 1370
V   i
n  1 i 1
(d ) 2

10
 137,
  V  11.7

35
Statistics
Solution: measurement set B, n = 11. B
xi 409 406 402 407 405 404 407 404 407
407 408
1 n
Mean, xm   xi  406.0
n i 1
di 3 0 -4 1 -1 -2 1 -2 1
1 2

(di)2 9 0 16 1 1 4 1 4 1
1 4
1 n 42
V  
n  1 i 1
(d i ) 2

10
 4.2,
  V  2.05
36
Statistics
Solution: measurement set C, n = 23. C
409 406 402 407 405 404 407 404 407
xi
407 408 406 410 406 405 408 406 409
406 405 409 406 407
1 n
Mean, xm   xi  406.5
n i 1
2.5 -0.5 -4.5 0.5 -1.5 -2.5 0.5 -2.5 0.5
di
0.5 1.5 -0.5 3.5 -0.5 -1.5 1.5 -0.5 2.5
-0.5 -1.5 2.5 -0.5 0.5

1 n 77.6
V  
n  1 i 1
(d i ) 2

22
 3.53,   V  1.88

37
Statistics
Results:

Set n xm V σ
A 11 409.0 137 11.7
B 11 406.0 4.2 2.05
C 23 406.5 3.53 1.88

A plot of the
readings
(reading)

38
(i)
Statistics
Conclusions:

Note that the smaller values of V and σ for measurement set B compared with A
agree with the respective size of the scatter (smaller scatter) in the readings as
shown in the plot

Therefore, as V and σ decrease for a measurement set, we are able to express

greater confidence that the calculated mean value is close to the true value, i.e.
that the averaging process has reduced the random error value close to zero.

Comparing V and σ for measurement sets B and C, V and σ get smaller as the
number of measurements increases, confirming that confidence in the mean value
increases as the number of measurements increases.

39
Graphical data analysis technique-frequency
distributions
Graphical techniques are a very useful method of analyzing the way in which random
measurement errors are distributed. The simplest way to do this is to draw a
histogram, in which intervals of equal width across the range of measurement values
are defined and the number of measurement (or frequency) within each interval is
counted.

An estimate of the number of intervals (K) required for statistical analysis

can be calculated from

K  1.87(n  1)0.40  1

number of intervals Number of readings

40
 Graphical data analysis technique-frequency
distributions
Example: Plot the histogram (frequency distribution) of measurement set
given in the table bellow

41
Graphical data analysis technique-frequency
distributions
Solution:
K  1.87(n  1)0.40  1
 1.87(20  1)0.40  1
 7 intervals
Range  xmax  xmin
 1.34  0.68
 0.66
Range 0.66
Interval width (x)    0.1
K 7
42
 Graphical data analysis technique-frequency
distributions
Data distribution on the intervals

number of occurrence frequency of occurrence

43
Graphical data analysis technique-frequency
distributions

Histogram and frequency distribution for the given data 44

 Graphical data analysis technique-frequency
distributions
Comment:

The sum of the number of occurrences must equal to the total number of
measurement n,

K
 n j  (1  1  3  7  4  2  2)  20
j 1

The sum of the frequency of occurrences will always equal 1.0

7
 f j  (2  5  15  35  20  2 10) / 100
1
 1.0
45
Graphical data analysis technique-frequency
distributions
Since the measurement error is usually of most concern, it is often more
useful to draw a histogram of the deviations of the measurements (Di) from
the mean value (xm) rather than to draw a histogram of the measurements
themselves. The resulted graph is known as histogram of deviations.
This histogram has exactly the
histogram of deviations
same shape as the histogram of
the row measurements except
that the scaling of the horizontal
axis has to be redefined in
terms of the deviations (Di)
instead of xi as shown

Here
1
xm   xi  1.1
n -.45 -.35 -.25 -.15 -.05 .05 .15 .25

Di  xi  xm Deviation, D
46
 Graphical data analysis technique-frequency
distributions
As the number of measurements increases, smaller intervals can be defined for
the histogram, which retains its basic shape but then consists of a larger number
of smaller steps on each side of the peak. In the limit, as the number of
measurements approaches infinity, the histogram becomes a smooth curve
known as a frequency distribution curve as shown in the figure. The ordinate of
this curve is the frequency of occurrence of each deviation value, F(D), and the
abscissa is the magnitude of deviation, D.
F(D)

The height F(D) at any particular

deviation magnitude D is known
as the probability density function
(p.d.f.). 0 D

47
 Graphical data analysis technique-frequency
distributions
The area under the curve is unity and can be expressed mathematically as


 F ( D)dD  1 F(D)

The probability that the error in anyone
particular measurement lies between two
levels D1 and D2 can be calculated by
measuring the area under the curve
contained between two vertical lines drawn
through D1 and D2
D2
P( D1  D  D2 )   F ( D)dD D0 Dp D1 D2 D
D1

48
 Graphical data analysis technique-frequency
distributions
The probability of observing a value less than or equal to D0 is expressed
mathematically as
D0
P( D  D0 )   F ( D)dD F(D)


The deviation magnitude Dp
corresponding with the peak of the
frequency distribution curve is the
value of deviation that has the greatest
probability. If the errors are entirely
random in nature, then the value of Dp
will equal zero. Any non-zero value of
Dp indicates systematic errors in the D0 D p D1 D2 D
data, in the form of a bias that is often
removable by recalibration
49
Graphical data analysis technique-frequency
distributions
Gaussian distribution

- Measurement sets that only contain random errors usually conform to a

distribution with a particular shape that is called Gaussian.

-The shape of a Gaussian curve is such that the frequency of small

deviations from the mean value is much greater than the frequency of
large deviations.

- Alternative names for the Gaussian distribution are the Normal

distribution or Bell-shaped distribution.
- A Gaussian curve is formally defined as a normalized frequency
distribution that is symmetrical about the line of zero error and in which
the frequency and magnitude of quantities are related by

F (D ) 
1
e D 2 2 2
, where D  x  xm
 2
50
 Graphical data analysis technique-frequency
distributions
D2
1
P (D1  D  D 2 )   e D 2 2 2
dD ,
D1  2

F(D)
Solution of this expression is simplified by
the substitution
D
z 

Therefore

P( D1  D  D2 )  P( z1  z  z2 ) D0 Dp D1 D2 D
z2
1
 e z 2 2
dz , Difficult to be solved

z 1  2
analytically !

51
Standard Gaussian tables z
1
F ( z)   Sx e  z 2 dz
2

 2

0 z

F(z) gives the proportion

of data values that are
less than or equal to z.


1
  2
z 2 2
e dz  1

52
Therefore

P( D1  D  D2 )  P( z1  z  z2 )
 F ( z2 )  F ( z1 )
For negative values of z, we can make use of the following relation

F ( z )  1  F ( z )
This is true because the total area the frequency distribution curve is equal to 1.

F(-z) is the area under the curve to the left of –z, i.e. it represents the
proportion of data values less than or equal to –z.

53
Example:

How many measurements in a data set subject to random errors lie

outside deviation boundaries of + σ and –σ . How many measurements
have a deviation greater than | σ |?

Solution F(D)

We know that D
z 

For D   , z  1
For D   , z  1

P (D   )  P (z  1)

-σ 0 +σ D
 F (1)  1  F (1)  1  0.8413  0.1587

54
Similarly

P (D   )  P (z  1)

 1  F (1)  1  0.8413  0.1587

Now:
P (D   or D   )  P (D   )  P (D   )

 0.1587  0.1587
 0.3174
 32%
i.e. 32% of the measurements lie outside  σ boundaries, then 68% of the
measurements lie inside

55
The previous analysis shows that, for Gaussian-
distributed data values, 68% of the measurements
have deviations within the bounds σ. Similar
analysis shows that

Probability of any
% of the points lie within
Deviation boundaries particular data point
the boundary
being outside boundary

σ 68.0 32%
2σ 95.4 4.6%
3σ 99.7 0.3%

56
P

D
z   D  z


68%

95.4%
99.7%

z =-3 z =-2 z =-1 z =0 z =1 z =2 z =3 z

57
Example:

It is known that the statistics of a well-defined voltage signal are given by

x m  8.5V,and   1.5V
If a single measurement of the voltage signal is made, determine the probability
that the measured value indicated will be between 10.0 and 11.5 V

Solution:

Assume that the signal has normal distribution about the mean value xm

It is required to find
P(10.0  x  11.5)

58
Now, find Di

D1  x1  xm  10  8.5  1.5 V
D2  x2  xm  11.5  8.5  3 V
Find zi

D1 1.5
z1    1
 1.5

D2 3
z2    2
 1.5
Now:

P(10.0  x  11.5)  P(1  z  2)

59
But we know that

P(1  z  2)  F ( z 2 )  F ( z1 )
 F (2)  F (1)
From the Standard Gaussian tables, we get

F (1)  0.8413
F (2)  0.9772

P(1  z  2)  0.9772  0.8413

 0.1359
This means that there is 13.59% probability that the measurement will yield a
value between 10.0 and 11.5 V

60
 Example 3

P (x  20.2)  P (x  19.8)

61
Example 4

The results of a test that follows a normal

distribution have a mean value of 10 and a
standard deviation of 1. Find the probability
that a single reading is
a) Between 9 and 12

b) Between 8 and 9.55

c) Less than or equal to 9

d) Greater than 12

62
 Solution 4

a) Between 9 and 12
 9  10   1 z2 
12  10   2
z1  1
1

P (1  z  2)  P (z  2)  P (z  1)
P (1  z  2)  0.9772  1  P (z  1) 
P (1  z  2)  0.9772  1  .8413
P (1  z  2)  0.8185  81.85%

63
 Solution 4

b) Between 8 and 9.55

8  10   2 z2 
 9.55  10   0.45
z1  1
1

P (2  z  0.45)  P (z  0.45)  P (z  2)

P (2  z  0.45)  1  P (z  0.45)   1  P (z  2) 
P (2  z  0.45)  1  0.6736   1  .9772 
P (2  z  0.45)  0.3036  30.36%

64
 Solution 4

c) Less or equal to 9

z1 
 9  10   1
1

P (  z  1)  P (z  1)

P (  z  1)  1  P (z  1) 
P (  z  1)  1  0.8413  0.1587

65
Solution 4

d) Greater than 12

z1 
12  10   2
1

P (2  z  )  1  P (z  2)
P (2  z  )  1  0.9772  0.0228

66
Confidence Interval

area   / 2
area   / 2
In general, a 95% confidence level
is used with 1.96 confidenceinterval Confidence
interval

Confidence level=1-

 :level of significance z  /2 z  0 z  /2

67
Interval estimation of the mean µ

 Averaging a number of measurements will only yield the

true value if the number of measurements is infinite.
 For a finite number of samples measurements n>30,
different samples from the same measurements will lead
different sample mean x values.

n
 The measurement mean µ is:
  x   x   xi
n x  i 1
 : The standard error of the mean
x
n
 For this case x 
z 
x
68
Interval estimation of the mean µ
x     x 
 We are interested in the minimum and maximum of µ.

  x  z  /2 x with confidence level 1-

69
Example 6

 We would like to determine the confidence interval of the

mean of a batch of resistors made using a certain process.
Based on 36 readings, the average resistance is 25Ω, and
the sample standard deviation is 0.5 Ω.

Determine the 90% confidence interval of the mean

resistance of the batch

70
 Solution 6
x  25,   0.5, n  36 area   / 2
area   / 2
 Confidence level is 90%,
Confidence
  1  Confidence level interval
=1-0.9=0.1

 0.05
2
 We should find z  /2 z  /2 z  0 z  /2
P (  z  z  )  1  0.05  0.95  z  1.645
  x  z  /2 x
2

0.5
 25  1.645   25  0.14
36
The average resistance must be in this range with confidence level 90%

71
Least-Squares Linear fit

 Straight line equation

Y  ax  b
 If you have n measurements with x as an input
and y as an output, then
n  x i y i    x i   y i 
a
n  x   x i 
2 2
i

b
 i  y i    x i   x i y i 
x 2

n  x   x i 
2 2
i

72
Example

. 73
Solution:

74
75
 Combined Effect of Systematic and
Random Errors

Error in a sum
If the two outputs y and z (each measurement is entirely independent of
the others) of separate measurement system components are to be added together,
we can write the sum as .
If the maximum errors in y and z are ± ay and ± bz, respectively,

76