Introduction to Instrumentation Engineering

Estimation of Errors and Reliability

Department of Electrical and Computer Engineering

Goals of this Chapter
 Differentiate the types of error
 Every measurement involves an error
 Give an overview of data analysis techniques in
instrumentation systems
 Understand basic mathematical tools required for this
purpose

2
Overview
 Measurement error analysis
 Types of errors and uncertainty
 Statistical analysis
 Gaussian and Binomial distributions
 Method of Least Squares

3
Measurement Error
 Types of errors: Systematic and random errors
 Systematic error
 Cause repeated readings to be in error by the same amount
 Consistent, or fixed error component

 May arise due to instrument short coming & environmental effects

 Related to calibration errors and can be eliminated by correct
calibration

 Or human error such as consistent misreading and arithmetic error

such as incorrect rounding off
 Or by using an inadequate measurement methods
 Example unjustified extrapolation of experimental data

 Accuracy is related to such type of errors

4
Measurement Error ….
 Random errors
 Due to unknown cause and occurs when all systematic errors have
been accounted for
 Caused by random electronic fluctuations in instruments,
unpredictable behavior of the instrument, influences of friction,
etc…

 Random fluctuations usually follow certain statistical distribution

 Treated by statistical methods
 Characterized by positive and negative errors

 Such errors are related to precision

5
Measurement Error ….
 Systematic errors analysis can be divided into
 Worst-cases analysis and RMS error analysis

 Worst-case analysis: Let Qm be the measured quantity and

Qt be true quantity
 Error:
Ea  Qt  Qm
 Relative error: Ea Ea Qt  Qm
Er   
Qt Qm Qm

 E.g., if the measured value is 10.1 when the true value is

10.0, the error is -0.1. If the measured value is 9.9 when
the true value is 10.0, the error is +0.1

6
Error & Uncertainty
 Uncertainty
 Since the true value cannot be known, the error of a measurement
is also unknown
 Thus, the closeness of the value obtained through a measurement
to the true value is unknown
 We are uncertain how well our measured value represents the true
value
 Uncertainty characterizes the dispersion of values

Qt  Qm  Ea
 ±Ea is the assigned uncertainty of Ea

7
Error and Uncertainty …
 Differentiate between error and uncertainty
 Error indicate knowledge of the correct value
 May be either positive or negative!

 Uncertainty indicate lack of knowledge of the correct value or may

be either positive or negative!
 Is always a positive quantity, like standard deviation

8
Combined Uncertainty …
 Given a function f  g ( x, y , z )

 The RMS error is given as

2
 f x   f y   f z 
2 2

E   Ea    Ea    Ea 
f

 x   y   z
a

f
 And E
E 
r
f a

Qm

9
Overview
 Measurement error analysis
 Types of errors and uncertainty
 Statistical analysis

10
Statistical Analysis
 Allows analytical determination of uncertainty of test result
1. Arithmetic mean (or most probable value) of n readings x1
to xn is given by 1 n
xm 
n
 x
i 1
i

 With a large sample, frequency distribution of the individual

xi’s can be used to save time
 If a particular value of xi occurs in the sample fj times, the mean
value can be determined as
1 1 k
xm  (f 1 x1  f 2 x2  ...  f k xk )   f j x j
n n j 1
 The sample frequency fj/n is an estimate of the probability Pj that x
has the value of xj in the population sample

11
Statistical Analysis …
2. Deviation xi-xm is the difference of all readings or
observations from the mean reading
 Is a good indicator of the uncertainty of the instrument
3. Average deviation: Sum of the absolute value of all
deviations, i.e., 1 n
n
 x x
i 1
i m

 Tends to zero and gives an indication of the precision of the

instrument (low value shows that the instrument is highly precise)
4. Standard deviation: deviation from the mean & is given as
1/2
1 n 
σ    (xi  xm )2 
 n i 1 
 σ is called the population or biased standard deviation
 Measure the extent of expected error in any observation
 Variance: σ2
12
Statistical Analysis …
 Using the probability distribution Pj and noting that Pj=fj/n
1/2
1 k  k
σ    f j(x j  xm )2    P (x j j  x m )2
 n j 1  j 1

 For most distributions (both real and theoretical) met in

statistical work, more than 94% of all observations in the
population are within the interval xm  2
 Generally, it is desirable to have about 20 observations in
order to obtain reliable estimate of 
 For smaller set of data, the expression for  modifies to
1/2
 1 n
2
s
 n  1

i 1
(x i  x m ) 


 Called unbiased or sample standard deviation

13
Overview
 Measurement error analysis
 Types of errors and uncertainty
 Statistical analysis
 Gaussian distributions
 Method of Least Squares

14
Gaussian Distribution
 Measurements will always have random errors
 For a large number of data, these errors will have a normal
distribution which follows
1  ( x  x m ) / 2 2
P( x )  e
 2
 P(x) is the probability density function
 It gives the probability that the data x will lie between x and x+dx
 Is called the Gaussian or Normal error distribution
 Xm is the mean and  is the standard deviation

15
Gaussian Distribution …
 Gaussian error distribution for
σ=0.5 and 1 and xm=3
 Probability density function has
the property



P(x)dx  1

 Xm is the most probable reading

 The value of the maximum
probability density function is
1
P(x m ) 
σ 2π
16
Gaussian Distribution …
 The standard deviation is a measure of the width of the
distribution curve about the mean
 Smaller σ produces larger value of the maximum
probability
 For a measurement, this tends to go to more precision
 The probability that a measurement will fall within a certain
range x1 of the mean reading is given by
x m  x1
1
P  e  ( x  x m ) / 2 2
dx
x m  x1  2

17
Overview
 Measurement error analysis
 Types of errors and uncertainty
 Statistical analysis
 Gaussian distributions
 Method of Least Squares

18
Method of Least Squares
 In the operation of an instrument, input parameter is varied
over some range
 Could be in increments or decrements
 Happens during calibration or measurement

 Least square can be applied to determine an equation for a

measured data
 Used to fit the data into a line (cure) to give a working relation
between input and output
 This relation will help to determine the characteristics of
the instrument

19
Method of Least Squares …
 Example: Linear Least Square Analysis (LLSA)
 Suppose xi and yi be the input and measured values, respectively
such that the data points (x1 , y1), (x2 , y2), ….. (xn , yn) are obtained
 If the expected straight line is of the form
y = mx + b
 where m is the slope and b is the intercept
 The error, which is the difference between the actual and
measured data, summed for all points is given as
n
S    yi  (mx i  b)
2

i 1

 Minimizing S using
S S
0 and 0
m b

20
Method of Least Squares …
 Will give

m
 x y x  y
i i i i

 x   x 
2 2
i i

 And
 y  x    y  x 
2 2

b i i i i

 x   x 
2 2
i i

21
Method of Least Squares …
 Example: Assume that the input and output are related by
a second order equation
y = b2x2 + b1x +b0
 The error will take the form
 
n
S   yi  (b2 x i2  b1 xi  b0 )
2

i 1

 Minimizing the error with respect to b0, b1, and b2 yields

  yi   n
    Xi  i  b0 
X 2

 
  X i yi     X i X 2
 X i  b1 
3 

  
4  b 
i

  X 2 yi    X 2 X 3
 X i  2 
 i   i i

 n = total number of data points

22