11/17/2019

Chapter Three
Analysis of experimental data, and regression

Home

Previous
Next
End

Analysis of experimental data

Introduction

In engineering practice, the experimental work has a high importance because it
is the connection between the real case and the simulate one.

when performing the experiment, engineers try hard to achieve certain level of
validity. Validity means how much we can trust the data obtained from an
experiment. The level of trust in data is called the level of confidence

 when performing the experiment – even with highly calibrated accurate

measurement devices, error will enter to the experiment without knocking.

Error mean in its simplest definitions: the deviation from the true. In this
manner, error is the opposite of accuracy. Knowing the sources of errors in the
experiment allow the experimenter to eliminate it and enhance the results.

The analysis of error is a fundamental process done on the experimental data to

reduce the gap between the true and measured values.

1
 11/17/2019

Analysis of experimental data

Errors

Experimental
data

Single Multi
sample sample

The single sample experiment means performing the experiment and obtain
one or more data results by using the same apparatus

The single sample experimental The multisample experimental

measurements is – in many time – measurements allow to the
avoided because it does not examiner to study the results
indicates the uncertainty in the statistically which insure more
experiment . confidence in the decision made
depending on the obtained results.

Analysis of experimental data

Uncertainty

Uncertainty means how much we are not confident (uncertain) about the
experimental data that we have. The main source of uncertainty is the error. In
some cases, we can say that we are uncertain in the magnitude of error.

Error

Instrument Systematic Random

error or bias error
error

Instrument error: errors in the measurement instrumentation

Systematic error: fixed shift in multi-readings taken by the same device
Random errors come from the operator (human errors), noise, …, etc.

2
 11/17/2019

Analysis of experimental data

Uncertainty analysis

Kline and McClintock establish a method to calculate the uncertainty.

They define the uncertainty as a range where the true value lies in. for
example, if a temperature measurement was
T = 103Co ± 1oC
The ± sign means that the experimenter is not sure about the results
and he/she define the range where the true value lies . In our example,
the experimenter implies that the true value lies between 102Co and
104Co.

Analysis of experimental data

Uncertainty analysis

If a set of measurements is made and we wish to calculate the total

uncertainty for the total set, we can follow these procedures:
 represent the results in terms of R. R is a function in independent
variables: x1, x2, …, xn (i.e. R = R(x1, x2, …, xn. ))
Define the uncertainties for each independent variable. For
example, w1, w2, …, wn.
Assume wR is the total uncertainty :

2 2 2
 R   R   R 
wR   w1    w2   ...   wn 
 x1   x2   xn 

3
 11/17/2019

Analysis of experimental data

Uncertainty analysis

For product functions

2
 ai wxi 
    
n
w
R  x1a1 x2a2 ... xnan  R   
i 1  xi

R 

For additive functions

 a w 
n
R  a1 x1  a2 x2  ...  an xn  wR 
2
i xi
i 1

Analysis of experimental data

Uncertainty Example [1]

The resistance of a copper wire is given as:

R  Ro 1   T  20
Where: Ro = 6Ω ± 0.3% is the resistance at a reference temperature (20 oC),
α = 0.004 oC-1 ± 1% is the temperature coefficient resistance and the
temperature of the wire T = 30oC ± 1oC.
calculate the wire resistance and its uncertainty.
Solution

The nominal resistance: R =(6)[1+(0.004)(30-20)] = 6.24Ω

4
 11/17/2019

Analysis of experimental data

Uncertainty Example [1]

Solution

The total uncertainty can be calculated using the general form

2 2 2
 R   R   R 
wR   w1    w2   ...   wn 
 x1   x2   xn 

R
 1   T  20  1  0.00430  20  1.04
Ro
R wRo  60.003  0.018
 Ro T  20  630  20  60
 w  0.0040.01  4 x105 C 1
R
 Ro  6 0.004  0.024 wT  1 C
T

Analysis of experimental data

Uncertainty Example [1]

Solution

The total uncertainty can be calculated using the general form

wR  1.040.0182  604 x105 2 0.2412

 0.0305 or 0.49%

5
 11/17/2019

Analysis of experimental data

Uncertainty Example [2]

Two resistance are connected in series as shown in the figure

If the measurements of voltage drop across these R1 and R2 (E1 and E2

respectively) were found as
E1 = 10V± 0.1V (1%)
E2 = 1.2V± 0.005V (0.467%)
And the value of R2 was found as R2 = 0.0066Ω± 0.25%
Now , find the power dissipated in resistance in R1 and its uncertainty.

Analysis of experimental data

Uncertainty Example [2]

Solution

The power dissipated (P) through a resistance R1 is given as: P = E1I and
the current passes through the resistance (I) is calculated as: I=E/R so the
power dissipated through a resistance R2 is calculated as:
E1 E2 101.2
P   1818.2W    1
R2 0.0066
The relation between P and the drop in voltage E is a product relation, so:

2
 ai wxi 
    
n
w
R  x1a1 x2a2 ... xnan  R   
i 1  xi

R 

6
 11/17/2019

Analysis of experimental data

Uncertainty Example [2]

Solution

aE1 =1, aE2 = 1 and aR2 = -1

2 2 2
wP a w  a w   aR 2 wR 2 
  E1 E1    E 2 E 2     0.0111
P  E1   E2   R 2 

Then
wP  0.0111 1818 .2   20.18

Analysis of experimental data

Uncertainty: further analysis

It is noted from the uncertainty calculations that: if there is a single

uncertainty in one of the measured parameters or variables much larger than
the uncertainties in the other parameters, the total uncertainty will be much
affected by this uncertainty rather than the other uncertainties due to the
presence of the square power. For example, assume you perform an
experiment and the value of the term (∂R/∂x)wr were: 10.00, 0.50, 1.00,
0.75, then the total uncertainty (wR) will equal :

wR  102  0.52  12  0.752  10.09  10.00

As you can see, the total uncertainty tends to the site of the largest
uncertainty in the experiment. This type of analysis is useful when
performing the experiment because it permits to the experimenter to focus
on reducing the largest uncertainty in the experiment which reduce the
effort and time

7
 11/17/2019

Analysis of experimental data

Selection of measurement method and measuring instrument

When an experiment includes many variables to measure and we have

many options to perform the experiment, the method of measurement
must be selected carefully. The selection of measurement method
depends, mainly, on:

 the level total uncertainty desired

Time and effort
Cost
Environmental conditions and technical difficulties

Analysis of experimental data

Example [3]: selection of measurement method

A resistor (R) has a nominal stated

value of 10Ω ± 1%. The circuit shown
in the figure was used to measure the
current (I) passes through R and the
voltage drop (E) across R. To find the
power (P) dissipated in R, we have tow formulas:
1. P=E2/R
2. P=EI
The first method implies a single measurement for (E) while the second
needs another measurement for (I). If the measurements for I and E
were:
E = 100 V± 1%
I = 10A ± 1%
Find the uncertainty in the power calculations using both methods

8
 11/17/2019

Analysis of experimental data

Example [3]: selection of measurement method

Solution
P 2 E
Case 1:  2 2
R  w   2 E  w2    E  w2
2
E
P E2 P E  R2  R
 2  R   
R R
2 2
w  w 
Divide by P=E2/R
wP
 4 E    R   40.012  0.012  2.236%
P  E  R 
Case 2:
P
I
I w
 P  0.012  0.012  1.414%
P P
E
I

Analysis of experimental data

Example [3]: selection of measurement method

Comments

You may see from the previous example that the uncertainty
decreased when in the 2nd case although there were two uncertainties
even there was one uncertainty in 1st case.

This is not necessarily correct for all cases. However, we can conclude
that the selection of method depends on the 4 factors we mentioned
before. In our example, reducing the uncertainty means more effort,
more time and more cost.

9
 11/17/2019

Analysis of experimental data

Statistical Analysis of Experimental Data

Fundamental concepts

Mean value is average of a set of collected measurements and it is

1 n
given by: xm   xi where xm is the arithmetic mean, n is the number
n i 1
of readings and xi is the reading.

Deviation is difference between the threading and the mean and given
by : d i  xi  xm . Note that the average of deviations equal zero

1 n
Absolute deviations average is given by: d i   xi  xm
n i 1

Analysis of experimental data

Statistical Analysis of Experimental Data

Fundamental concepts

Standard deviation (σ) or root-mean-square deviation is defined by:

1 n
  xi  xm 2
n i 1
The variance is defined as σ2. This is sometimes called the population or
biased standard deviation because it strictly applies only when a large
number of samples is taken to describe the population.

To obtain reliable estimates of standard deviation, it is desired to have at

least 20 measurements. In many cases, the engineer is not able to have
this number of readings. In such cases (small set of data), unbiased or
sample standard deviation is used instead

10
 11/17/2019

Analysis of experimental data

Statistical Analysis of Experimental Data

Fundamental concepts

Unbiased or sample standard deviation is given by

1 n
  xi  xm 2
n  1 i 1

The median is defined as the value that divide the readings in half

Analysis of experimental data

Example [4]:

Find the mean reading, standard deviation, variance, and average of the absolute
value of the deviation for the given data in the table below
Reading No. Pressure, kPa
1 1.25
2 2.45
3 1.10
4 2.03
5 3.11
6 2.95
7 2.36
8 3.42
9 3.01
10 2.10

11
 11/17/2019

Analysis of experimental data

Example [4]:

Solution

1 n
xi  23.78  2.378kPa
1
1. xm xm  
n i 1 10

1 n
2. σ   xi  xm 2  0.7388
n i 1

3. σ2 = 0.5458 kPa2

1 n
4. d i di   xi  xm  0.61
n i 1

Analysis of experimental data

Example [4]: solve using table

No. xi di = xi – xm d i2 |di|
1 1.25 -1.128 1.272384 1.128
2 2.45 0.072 0.005184 0.072
3 1.10 -1.278 1.633284 1.278
4 2.03 -0.348 0.121104 0.348
5 3.11 0.732 0.535824 0.732
6 2.95 0.572 0.327184 0.572
7 2.36 -0.018 0.000324 0.018
8 3.42 1.042 1.085764 1.042
9 3.01 0.632 0.399424 0.632
10 2.10 -0.278 0.077284 0.278
Sum 23.78 ------ 5.45776 6.10

12
 11/17/2019

Regression- Method of Least

Squares

25

Least square method (cont….)

26

13
 11/17/2019

Least square method (cont…..)

27

Quadratic regression

28

14
 11/17/2019

Quadratic reg.

29

Example

30

15
 11/17/2019

Example (cont……)

31

32

16
 11/17/2019

Table 3-8 (cont…..)

33

Practice and HW
TRY TO SOLVE THE FOLLOWING PROBLEMS
3.4
3.5
3.6
3.12
3.27
3.28
3.29
3.30 34

17
 11/17/2019

35

36

18
 11/17/2019

37

19