6 Measurement System Errors

Electronic Instrumentation
Lecturer Touseef Yaqoob

1
Measurement System errors
All measurements contain error. This may be difficult for
you perfectionist types to come to grips with, but you will
have error, and it is not a sin. The sin is not knowing how
big your error can be. Or as Clint Eastwood says A mans
got to know his limitations, or something like that.

In this chapter, we will learn how to estimate the size of the
error in a given measurement. The theory is obtuse, but
important, and will be clarified with hands-on examples in
the lab.

This is an extremely important topic (perhaps the most
important topic in the course). It is also one that many
people, including many experimentalists, do not fully
understand.

2
What we will study?

Measurement Uncertainty
Error Analysis
Propagation of Error
3
Background
Error exists in all experiments and can only be minimized.
Errors can be random or fixed in value, frequency of
occurrence, linear or non-linear in relation to magnitude of
measurements.
Some errors just skew the results of the data in one
discernable direction and can be accounted for, other times,
they just cause random scattering of the data.
Error can hide real trends in the data, or give the perception
of a real trend. It is important in interpreting data to account
for error and its effect on the relationship presented by the
data.

4
Types of Error
Gross Experimental Error
Related to mistakes made by the experimenter.
Equipment failure or faulty set-up or miscalibration.
Fixed Errors (systematic and bias errors)
Caused repeated readings to vary by the roughly the
same amount, usually for an unknown reasons
Usually due to poor set-up; voltage bias; not setting
indicators to zero. Can usually be detected by
experienced experimenter.
Random Errors
Usually follow a statistical distribution, caused by
random electronic fluctuations, mechanical play and
friction.

5
Three Common Errors in Instrumentation
Zero Offset Error
The ability of an instrument to display zero in the
absence of an input, or when the input is zero.
Non-linear response
It is desirable for an instrument to have a linear
response, so error is uniform over the measurement
range.
Resolution
The smallest unit of measurement that can be
indicated by an instrument.
Instrument error is typically associated with the span
chosen on the instrument.
Span - The difference between the highest and
lowest value for which a linear response exists for
the instrument.

6
Hysteresis
Many sensors have the
undesirable
characteristic of giving
a different value when
the input is increasing
than when it is
decreasing. This is
called hysteresis.

7
Linearity
Many types of sensors have linear input/output
behavior, at least within a narrow range of
inputs. The sensor thus follows an input/output
relation like
yL(x) = a0 + a1x. These will often be marketed
as linear, and the only calibration data you get is
the slope of the input/output relation (a1) and the
zero input value (a0). For these types of
sensors, the deviation from linear behavior
should be reported in the specifications. This
deviation can be calculated: eL(x) = y(x) - yL(x).
The spec is usually the percentage error relative
to full scale.

8
Linearity error
9
Zero and Sensitivity errors
Variations in the linearity parameters a0
and a1 are called zero
errors and sensitivity errors, respectively .
Zero errors are handled rather easily by
measuring the zero input response before
measurements are started. These two
errors are often sensitive to temperature
fluctuations in electronic equipment.

10
Sensitivity error/Zero shift
error
11
Resolution Example
Given an instrument with an accuracy of 0.1%

Voltmeter set to mV range shows 30.1 mV
0.1% gives instrument accuracy of 0.01 mV

so the measured value could be 30.09 to 30.11 mV

Voltmeter set to V range shows 0.030 V
0.1% gives instrument accuracy of 0.01 V = 10 mV
so the measured value could be 0.020 to 0.040 V

12
Bias or Systematic error
An offset error; one that remains with repeated
measurements (i.e. a change of indicated pressure
with the difference in temperature from calibration to
use).
Systematic errors can be reduced through
calibration
Faulty equipment--such as an instrument which
always reads 3% high
Consistent or recurring human errors-- observer
bias
This type of error cannot be evaluated directly from the
data but can be determined by comparison to theory or
other experiments.
13
Sources of Systematic error
System disturbance due to
measurement
Modifying inputs in measurement
systems
Wear in instrument components
Connecting leads
Thermal e.m.f.s
14
System disturbance due to
measurement
Disturbance of the measured system by
the act of measurement is one source of
systematic error. The magnitude of the
disturbance varies from one measurement
system to the next and is affected
particularly by the type of instrument
used for measurement. Ways of
minimizing the disturbance of measured
system are an important consideration in
instrument design.
15
Examples
Measuring temperature for a beaker
of hot water
Measuring car tyre pressure
Measurement in electric circuits
16
Modifying inputs in measurement systems
Variations of environmental conditions
away from the calibration conditions are
described as modifying inputs to the
system and are a further source of
systematic error. The environmental
variation is described as a measurement
system input because the effect on the
system output is the same as if the value
of the real input (measured quantity) had
changed by a certain amount.
17
Examples
Measurement of mass inside
a box
Measurement by strain
gauges
Measurement by
galvanometer
18
Techniques used to deal with
modifying inputs
Careful instrument design
Design of strain gauges with a material
whose resistance has a very low
temperature co-efficient.
Method of opposing inputs
Introducing a compensating resistance
into a circuit used for the design of
millivoltmeter.
19
High-gain Feedback
Design of a feedback device which is
insensitive to modifying inputs is much
easier than trying to make a motor or
spring insensitive. Thus a high gain
feedback techniques are often a very
effective way of reducing a measurement
system sensitivity to modifying inputs.
The one disadvantage can be the system
instability.

20
Signal filtering
One frequent problem in
measurement system is corruption
of the output reading by periodic
noise.The amplitude of all noise
components must be substantially
attenuated by the inclusion of
filtering of an appropriate form in the
system
21
Other sources of systematic
errors
Wear in instrument components
Systematic errors can frequently developed over a period of
time. Recalibration can provide a full solution to this
problem.
Connecting leads
A common source of error is the failure to take care proper
account of the resistance of connecting leads (pipes in the
case of hydraulic or pneumatic). Careful consideration must
include, choice of connecting leads, adequate cross section,
adequately screened, and also by selecting the proper route
of cables between the transducers and control room.
22
Thermal e.m.f.s
Whenever metals of two different types are
connected together, a thermal e.m.f. is
generated according to the temperature of the
joint. This is known as thermoelectric effect and
is the physical principle on which temperature-
measuring thermocouples operate. One
instrument can use to monitor the output of
several thermocouples at different points in a
process control system. This requires a means
of automatically switching the output Nickel-iron
reed relays with copper connecting leads are
commonly used to provide this function. This
introduces a thermocouple effect of magnitude
40uV/C between the reed relay and the copper
connecting leads.

23
Reduction of systematic errors
using intelligent instruments
Intelligent instruments contain all the
usual elements of a measurement system
with the inclusion of microprocessor
within the instrument. This enables them
to apply preprogrammed signal
processing and data manipulation
algorithms to measurements. The ability
of intelligent instruments to achieve
desire performance, require that the
following preconditions be satisfied.

24
1. The physical mechanism by which a
measurement transducer is affected by
changes in ambient conditions must be fully
understood and all physical quantities which
affect the transducer output must be identified.
2. The effect of each ambient variable on the
output characteristic of the measurement
transducer must be quantified.
3. Suitable secondary transducers for monitoring
the value of all relevant ambient variable must
be available for input to the intelligent
instrument.
25
Uncertainty Analysis
The estimate of the error is called the uncertainty.
It includes both bias and precision errors.
We need to identify all the potential significant errors
for the instrument's).
All measurements should be given in three parts
Mean value
Uncertainty
Confidence interval on which that uncertainty is based
(typically 95% C.I.)
Uncertainty can be expressed in either
absolute terms (i.e., 5 Volts 0.5 Volts)
or in
percentage terms (i.e., 5 Volts 10%)
(relative uncertainty = DV / V x 100)
We will use a 95 % confidence interval throughout this
course (20:1 odds).

26
How to Estimate Bias Error
Manufacturers Specifications
If you cant do better, you may take it from the
manufacturers specs.
Accuracy - %FS, %reading, offset, or some
combination (e.g., 0.1% reading + 0.15 counts)
Unless you can identify otherwise,
assume that these are at a 95%
confidence interval
Independent Calibration
May be deduced from the calibration process

27
Use Statistics to Estimate Random
Uncertainties
Population: The entire collection of
measurements being analyzed.

Sample: A subset of the population which
is analyzed statistically.

Statistic: A numerical attribute of the
sample, e.g. the standard deviation, the
mean, the confidence interval.
28
Basic Statistics
Mean: the sum of measurement
values divided by the number of
measurements.

Deviation: the difference between
a single result and the mean of
many results.

Standard Deviation: the smaller
the standard deviation the more
precise the data
Large sample size

Small sample size (n<30)
Slightly larger value

x =
1
N
x
i
i =1
N
x x d
i i
=
o
s
=
1
n 1
x
i
x ( )
2

(

1
2
o =
1
n
x
i
x ( )
2

(

1
2
29
Probability Distributions
Distributions (pdf):
characterizes the
probability that an error of a
given size will occur
Normal (Gaussian, bell
curve)
30
Distributions
Two Dice Rolled
31
Gaussian Distribution
Given a set of repeated measurements which
have random error.
For the set of measurements there is a mean
value.
If the deviation from the mean for all the
measurements follows a Gaussian probability
distribution, they will form a bell-curve
centered on the mean value.
Sets of data which follow this distribution are
said to have a normal (statistical) distribution of
random data.
32
Gaussian Distribution
For a normal distribution have the follow properties:
68.3% of the measurements will fall within 1o of the mean

For an interval of -1o < x < +1o, one can be confident that
any random measurement of x will lie within one standard
deviation of the mean value 68.3% of the time.

An interval of -2o < x < +2o, one can be confident that 95.4%
of the time, x will lie within two times the standard deviation
of the mean value.

33
95% Confidence Interval
The 2o interval, or the 95% confidence interval is the
engineering standard.
It is the implied uncertainty of a value even when not
written explicitly, e.g.
Resistor value is given as 100 10 O means that the
true value of a marked 100 ohm resistor wil be
between:
91 and 109 ohms, 95% of the time
95 and 105 ohms, 68% of the time
but that 4.6% of the time, it can be between 87 and
104

34
Tolerance and Uncertainty
Dont confuse tolerance and uncertainty.

Uncertainty can be stated as a:
Percent Accuracy, e.g. 10 %, which is a 95%
confidence interval relating to the magnitude of the
measurement.
Given 95% confidence interval magnitude, e.g. 10
ohms

Tolerance implies a tested range for a specific
component or part. It is a guarantee that the value lies
within the tolerance value from the marked value.
A resistor (with 5% tolerance) marked as 100 O will
have a true value between 95 and 105 O.

35
Tabulation of Normal Data

f z
( )
=
1
o 2t
e
z
2
/ 2
Make z =(x - m)/ s
36
Connection to Error Analysis
37
Finite Samples

x =
1
n
x
i
i=1
n
| |
1 1
1
2
1
2
1
2
|
.
|
\
|
=
=
=
n
x n x
x x
n
S
n
i
i
n
i
i x
Estimate of the
Standard Deviation
Estimate of the
mean
Based on a finite sample, we would like to:

Estimate the mean and standard deviation, and their
uncertainty

Infer the distribution of the data (pdf).
38
Velocity Example
30
35
40
45
0 0.002 0.004 0.006 0.008 0.01
u
[
m
/
s
]
t
The velocity in a turbulent flow varies randomly
39
0
1000
2000
3000
4000
5000
6000
7000
13 16 19 22 25 28 31 34 37 40 43 46 49 52
C
o
u
n
t
m/s
50,000 samples
x = 36.727 m/s
S
x
= 2.898 m/s
40
Central Limit Theorem

o
x
=
o
n
41
Confidence Intervals
Given a large but finite sample, we estimate that true mean
is

x =
1
n
x
i
~
i=1
n
Wed like to be able to say how sure we are of this estimate. Lets look at the
probability that our estimate of the mean is within some bound. We can say
that there is a c% chance that our estimate of the mean lies within

z
c / 2
o
n
The larger we make the confidence interval c, the larger z becomes (look at
Table 3.2) and therefore the larger the range that our measurement may
land in. A wider dispersion in the population (large s) also makes this
interval larger. Sampling more data (increasing n) makes the interval
smaller.
42
43
Confidence Intervals
This means that we are c% confident that the true
mean m lies within the interval about our
measurement:

x z
c / 2
o
n
< < x + z
c / 2
o
n
The only trouble is that we dont know the value of s
either. If n is large enough, we can use our estimate
Sx, so

x z
c / 2
S
x
n
< < x + z
c / 2
S
x
n
44
Standard Error of the Sample
Mean

S
x
=
S
x
n
45
Example
Find the 99% Confidence Interval
46
Solution
47
Solution
48
Solution
Find the 99% confidence interval for the
mean pressure.
c = 0.99, c/2 = 0.495

49
Solution
Z0.495 = 2.575

z
c / 2
o
n
m = 4.008 2.575 (0.014)/10 = 4.008 3.605 E -3 (99%)
50
Propagation of Error
Given a parameter is calculated from two experimentally
determined values, each with its own parametric uncertainty.
Area = Length x Width
The uncertainty of the final calculated parameter is related to the
elemental uncertainties of the variables used in the calculation.
The error of the parametric components is said to propagate into
the error of the final parameter.
How the error propagates depends on the relationship of the all
the parameters.

Three common methods are used to determine the propagated
uncertainty of a calculated parameter.

51
Propagation of Error (Analytical Methods)
Maximum Uncertainty:

=
c
c
=
n
i
i
x
x
R
U
i
1
max
e
RSS (Root Sum of Squares)
Uncertainty:
=
|
|
.
|
\
|
c
c
=
n
i
i
x rms
x
R
U
i
1
2
e
52
Propagation of Error (sensitivity
analysis)
An alternative to taking the partial derivatives of
the objective function relative to each of the
independent parameters is to perform a first
order sensitivity analysis for the calculated
parameter.
The uncertainty of each measurement parameter
is individually perturbed by its maximum
uncertainty, while holding all the other
parameters constant. The error caused in the
final calculated parameter is found by using a
RMS method on the individual perturbed error
contributions.

53
Elemental Uncertainties
For a single measurement parameter, the total
(parametric) uncertainty may be a function of a
one or more elemental uncertainties.
All elemental uncertainties must:
have the same units
be estimated with the same confidence interval
be entered in the +/- format
2
3
2
2
2
1
e e e
i
+ + + = e
54
Example
( )
rpm 07 . 3 1 90 . 2
rpm 1
rpm 90 . 2 45 . 1 2 2
2 2
= + =
=
= = =
e
o
device
t measuremen
e
e
Consider a measurement of rotational speed, N, of a shaft, obtained
using a stroboscope. Several measurements were made for statistical
analysis.
Mean rotational speed: 1734 rpm
Std. Deviation of measurements: 1.45 rpm
Uncertainty of stroboscope: 1 rpm

First, get the 95% confidence interval uncertainty for measurements:
thus, the shaft speed should be reported as: N = 1734 3 rpm

6 Measurement System Errors

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

6 Measurement System Errors

Uploaded by

Copyright:

Available Formats

Electronic Instrumentation

Lecturer Touseef Yaqoob

You might also like