Experimental Techniques

in Thermofluids
MEL7070
Error Analysis

By: Dr. Shrutidhara Sarma

Error

The difference between the measured value and true value is called “Error”.

It is important to know how close the measured value is to the true value.

True Value=Measured value ±Errors

Types:

1. Systematic Error/Bias: Have some identifiable source and may be corrected by

calibration. It is consistent, repeatable and is associated with faulty instrument or
a flawed experimental design. E.g. a worn out instrument, am incorrectly
calibrated instrument or a person consistently taking an incorrect measurement.

2. Random Error: These errors can not be sourced out and hence can not be
corrected. Due to temperature change, humidity, wind, vibrations,
electromagnetic field, etc. Since, this error arises from a multiple sources, it is
impossible to quantify. They can be both positive or negative.
Systematic & random error
Estimation of random error
• Since random errors are not deterministic, a probabilistic approach is used.
• Probabilistic approach involve a probability distribution.
• Gaussian or Normal distribution is commonly used. If μ and σ are the mean and
variance, then the probability distribution is given by:
1  x 
2

1  
2   
f ( x)  e
 2
Cumulative density function

Deviation in terms of ±σ

Confidence Intervals
Cumulative 0 0.95 0.99 0.999
probability
Interval 0 ±1.96σ ±2.58σ ±3.29σ
Cumulative probability density
The total cumulative probability is given by:
1  v 
2
x  
1 2   
F( x) 
 2 
x
e dv

Because random errors have equal likelihood to be positive and negative:

1  v 
2
x  
2 2   
F( x) 
 2 0
e dv
Properties of Gaussian Function
• The area under the Gaussian Curve is 1, hence the maximum
probability is 1. I.e., the probability of any real value to lie between -∞
to + ∞ is 1.

• The area lies under [-σ, +σ] is 68%: the change that a measurement
will be out of the given limits is 1 in 3 times

• The area lies under [-2σ, +2σ] is 95.4% ; in every 20 measurements,

one may fall out of the given limits

• The area lies under [-3σ, +3σ] is 99.7% ; one measurement in every
300 measurements may fall outside of the limits:
Gaussian function narrows down with decreasing σ
Best estimate of true value
Let’s x be the value measured for n times i.e. xi = 1:N
Then cumulative probability:
   x1 2    x1 2    x1 2
1 2 2 2 2 2 2
e e .......e dx1dx2 ....dxN
 
N
2
N
   x1 2
1  2 2
 e i1 dx1dx2 ....dxN
 
N
2
Considering the values x1, x2 etc. are highly probable we maximize
N
   x1 2
1  2 2
P(  ,  )  e i1
 
N
2

with respect to the two unknowns μ and σ

Mean and Variance
N
1

N
x1
i

N
1
    x 
2 2
i
N i 1

Variance is the least possible value. Variance w.r.t. mean is the smallest one
can imagine.

Q. Value of a thermocouple output under repeated trials are given below.

Find the mean, variance and the error with 95% confidence.
Expt 1 2 3 4 5 6 7 8 9
V (mV) 1.22 1.23 1.26 1.21 1.22 1.22 1.22 1.24 1.19

Ans. Mean=1.22 kΩ, Variance = 3.33x10-4 kΩ, Error with 95% confidence= 0.04kΩ
Uncertainty Analysis:
Kline and McClintock’s Method 1953
• Since the total error includes random error, which are uncertain, errors
are usually referred to as uncertainties.
• For any experimental study, uncertainties analysis must be performed
and reported along with the measurements

Consider a variable N (dependent) which is calculated from various

measurements such as u1, u2, u3,u4…….un and governed by the function
N  f (u1 , u2 , u3 ,......, un )
The total uncertainty (in N) Δn would include uncertainties of ui i.e.u1 , u2 , u3 ,....
Therefore
N  N  f (u1   u1 , u2   u 2 , u3   u 3 ,......, un   u n )
Expand f (u1 , u2 , u3 ,......, un ) by Taylor series:
 Kline and McClintock’s Method
f
f (u1   u1 , u2   u 2 , u3   u 3 ,......, un   u n )  f(u1 , u2 , u3 ,......, un )   u1
u1
f f
  u 2 ............   u n  O( u1 ) 2  O( u1 )3 ......
u2 un

Ignoring the higher order terms:

f f f
N   u1   u 2 ............   un
u1 u2 un
Sometimes the estimated error may be wrong . Some time systematic errors may
have equal magnitude in OPPOSITE sign; they may tend to cancel out each other.
So, a reasonable estimate would be the Root Sum Square (RSS) error given by:

2 2 2
 f   f   f 
N    u1     u 2  ............    un 
 u1   u2   un 
Uncertainties for product functions

N  (u .u2 .u3 ........un )

1
a1 a2 a3 an

N ai 1
 u1 .u2 .u3 ...(a i u i ).....un
a1 a2 a3 an

ui
1 N a i 1 f ai
 or 
N ui u i N ui u i
Inserting the above equation into this:
2 2 2
 f   f   f 
N    u1     u 2  ............    un 
 u1   u2   un 
2 2 2
1  a1   a2   an 
N    u1     u 2  ............    u n 
N  u1   u2   un 
2
1  ai 
N    u  ui 
N  i 
Uncertainties for additive functions

N  (a1 u1  a2u2  a3u3 ........  a n un )

N
 ai
ui
Inserting the above equation into this:
2 2 2
 f   f   f 
N    u1     u 2  ............    un 
 u1   u2   un 

N   a1 u1    a2  u 2  ............   an  u n 
2 2 2

N  a  u 
2
i i
Examples
Q1. Calculate the uncertainty in head loss h1 expressed as:
flV 2
h1 
2 gd
Given uncertainties in l, v and d are 2%, 4% and 1% respectively. Ignore the
uncertainties in f and g.

Solution

 h1   h1   h1

2 2 2

h1   l    v  ............   d 
 l   v   d 

h1 fV h12
h1 2 flV 2h1 h1 flV 2 h1
  ;   ;  
l 2 gd l v 2 gd v d 2 gd 2
d

  l  2  v   d  
2 2

h1  h1     4      
  l   v   d  
 Examples

h1  h1 0.02  4 0.04  0.01 

2 2 2

h1
 0.083
h1

Total error= 8.3%

Q2. Calculate the uncertainties in the following quantities:

pd 4
Flow rate in a fully developed laminar region: Q 
128 l
g   (Ts  T )l
2 3
Grashoff’s number: Gr 
 2

Given uncertainities in Δp, d, μ, l, ρ, β and T are 4%, 2%, 3%, 2%, 4%, 5%
and ±2˚C
 Examples

Q3. Airflow rate of 17m3/h through a pipe of 60 mm ID at 20oC is measured

using a square edged orifice (β = 0.4). A pressure drop observed is 157.85
N/m2 with ± 0.4%. If the area of orifice is maintained within 0.2 %, estimate
the design stage uncertainty in the flow rate. Assume accuracies of Cd and ρ
are ± 0.5%. Estimate the total error in the measurement for Cd = 0.63 and P
= 0.97 bar abs and R= 287 J/kg K.
0.5
 2p 
Q  Cd A  
  
 Graphical presentation of data

1. Graph usually serves to communicate knowledge from the author to the

readers.

2. Graphs plays a vital role in testing the theoretical calculations against real
experiments results.

3. There are some situation, where it is very difficult to conduct experiments

through out the range of parameters. Hence, graphs are highly useful to
interpolate / extrapolate the output in the missed regions ( where the exact
details of output is not known)
 General rules for plotting
1. The graphs should be plotted in such a way that the reader should understand
the conveyed information with out any difficulties.
2. The axes should have clear labels i.e. name of quantities, units, and their
symbol if necessary.
3. Axes should be clearly numbered and should have clear tick marks with
significant numerical divisions (Sub division also should be clearly mentioned)
4. Use scientific notation to avoid placing of too many digits on the graph.
5. When plotting on log coordinates, use real logarithmic axes. Log scales should
have tick marks at powers of 10.
6. Choice of the scales in axes should be based on the relative importance of the
variations shown in the results and the also based on the availability of data
7. Use specific symbols for the experimental data points. Analytical results are
expressed by using straight lines.
8. Indicate the maximum uncertainty of the estimated quantity.
 General rules for plotting

9. When the two different dependent variables are compared with a common
parameter, a secondary Y axis should be used.
10. When several curves are plotted on a single graph, use different pattern of line
such as solid line, dotted line, dashed line etc. In case of experimental results
use different legends for differentiating the curves.
11. Other independent parameters should be specified in the graphs. Lettering on
the graphs should be clearly visible.
12. Title of the graph should be placed at the bottom and the titles should be
properly numbered. Choice of the scales in axes should be based on the relative
importance of the variations shown in the results.
Some plotting examples
Some plotting examples
Some plotting examples
Choosing X-Y axis

Linear - Linear
Liner – Logarithmic ( Semi log )
Logarithmic – Logarithmic (Full log)
Polar coordinates ( Quantities varies with an
angle)
General considerations in data analysis

Elimination of inconsistent data: Analysis the data carefully. Try to eliminate the
inconsistent data or the error data. Suppose the measurement itself consists of many
inconsistent data, the entire experiments may be repeated.

Estimate the Uncertainty : Detailed uncertainty analysis should be performed

before doing the data analysis.

Anticipate the results from the theory : Before trying to obtain the correlations of
the experimental data, investigators should carefully review the theoretical
background of the estimated quantity. This may be useful for determining the
graphical
formats, etc. e.g. First order instrument

Correlate the data: Critical review of the results.