MKM1133 INSTRUMENTATION AND

CONTROL SYSTEM

1 Introduction
For many, digital signal processor, such as an FFT analyser, is treated as a black box for
as long as they know what to place at the inputs and what to extract from the outputs.
An average user may get away with this just fine. However, a little peep into this black
box would reveal quite a substantial amount of knowledge that allow a user, on one hand
able to extract the maximum benefit of the analyser, and on the other hand would be
fully aware of its limitations. One common situation is to know why different windowing
functions are being used - when one should use Hanning and when one should use
Flattop. Another common situation is to know why one should amplify the input signal
level to the maximum offered by the analyser. In this chapter we wish to demonstrate the
path of signal from the point it is being measured to the point where the signal is being
processed, displayed and stored. We would like to surf this area of study touching only on
areas that are considered of practical importance from the user point of view. Fig. -1
shows the signal path. Generally, all digital signal processors are based on the similar
flow pattern.

Fig1. shows the flowchart of the process.

2 Transducers

This is the point where a signature from a measured

specimen begins its journey.
A transducer fundamentally converts a physical
signal to electrical signal. The signal is in an analogue
form receptive to all types of signal processors, may
it be analogue or digital. Below are some the
examples and their respective measuring engineering
quantities: a) Accelerometer acceleration
b) Force Transducer - dynamic force
c) Eddy current probe displacement
d) Tachometer - rotational
e) Speed
f)
g)
h)
i)

Thermocouple temperature
Strain Gauge - strain
Microphone - sound pressure
Sound Intensity Probe sound intensity

For a given of operation, the electrical quantity has to

be proportional to the physical quantity, otherwise
distortions would occur. This constant of
proportionality is commonly called sensitivity.
It is normally described in terms of Volts per
Engineering Unit.
For example, if the sensitivity of an accelerometer is
100 mV/g, it means that the sensor will output
100mV for 1g of motion it senses.

Fig2a shows some of these sensors.

The most common sensors used in machine
structural dynamics are briefly explained as follows:

Accelerometers

Force transducers

Proximity Probe

Pressure Transducer

a) Accelerometer
Accelerometer is probably one of the most
commonly used sensors in vibration analysis. The
most popular one is based on piezoelectric where
the charge emitted from stressed piezoelectric
quartz is proportional to acceleration.

Microphones

Fig.-2a Various sensors

Basically, there are two types of piezoelectric accelerometers, namely the integrated
circuit piezoelectric (ICP) and the non-ICP accelerometer. The difference is that the
ICP has a built-in charge amplifier. The internal circuitry converts charges (high
impedance) emitted from the quartz into voltage (low impedance) and output through
the sensors casing. This type of accelerometer requires a power supply of between
2mA and 10mA to power up the charge amplifier. Certain accelerometers include an
integrator in the circuit so that the output is velocity instead of acceleration. Some
further include a double integrator so that the output will be displacement. However,
when using an integrator a high-pass filter is required to eliminate the constant of
integration or very low frequency signals. In normal working environment, this is
probably the neater of the two to work with, as it needs lesser accessories. The
limitation with the ICP lies in the environment temperature in which it is placed. As it
has built in electronic, it can sustain temperature up to about 120 0C. The non-ICP is
the traditional accelerometer where it requires an external charge amplifier.
It is worthwhile to mention here that for a variety of low frequency monitoring
applications, manufacturers who traditionally used quartz are now using
piezoceramics (PZT) in many of their new accelerometer designs. When monitoring a
roll at 60 rpm (1 Hz), 10 mils pp of shaft movement produces only .0005 g of
acceleration. These low amplitude levels can approach the electronic noise floor of
standard accelerometers. Low frequency accelerometers require very high sensitivity
piezoceramics to overcome the electronic noise of the sensor amplifier and
monitoring instrument.
Cabling with ICP sensors. Operation over long cables may affect frequency response and
introduce noise and distortion when an insufficient current is available to drive cable
capacitance. Unlike charge mode systems, where the system noise is a function of cable
length, ICP sensors provide a high voltage, low impedance output well suited for driving
long cables through harsh environments. Generally, this signal distortion is not a problem
with lower frequency testing within a range up to 10 000 Hz. However, for higher
frequency vibration, shock or transient testing over cables longer than 100 ft. (30 m.),
the possibility of signal distortion exists.

b) Force transducer
The common transducers for dynamic force are also based on piezoelectric where the
charge emitted is proportional to force. Similar to the accelerometer, it exists in two
types namely the ICP and the non-ICP. The impact hammer used in modal analysis is
basically a hammer with a force transducer, as shown in Fig.-2b, fixed at the
hammerhead.

Fig.-2b illustrates the cross-section

of a typical quartz force sensor
ICP Low Impedance Quartz Force Sensors ICP force sensors incorporate a built-in
MOSFET microelectronic amplifier to convert the high impedance charge output into a
low impedance voltage signal for recording. ICP sensors, powered from a separate
constant current source, operate over long ordinary coaxial or ribbon cable without signal
degradation. The low impedance voltage signal is not affected by triboelectric cable
noise or contaminants. Power to operate ICP sensors is generally in the form of a low
cost, 24-27 VDC, 2-20 mA constant current supply. Many data acquisition systems now
incorporate constant current power for directly powering ICP sensors.
For Dynamic and not Static force Measurements The quartz crystals of a piezoelectric
force sensor generate an electrostatic charge only when force is applied to or removed
from them. However, even though the electrical insulation resistance is quite large, the
electrostatic charge will eventually leak to zero through the lowest resistance path. In
effect, if you apply a static force to a piezoelectric force sensor, the electrostatic charge
output initially generated will eventually leak back to zero. The rate at which the charge
leaks back to zero is dependent on the lowest insulation resistance path in the sensor,
cable and the electrical resistance/capacitance of the amplifier used.

c)

Eddy current probe (proximity

probe)
Eddy current or commonly called proximity
probe is mainly used in rotor dynamics to
study the motion (displacements) of shaft
axis. This is normally called orbit analysis.
This is performed with two probes located at
90o to each other to give the orbital motion
of the shaft axis while the machine is in
operations. This allows for the analysis of
shaft related problems such as shaft critical
speeds and their related modes, shaft
Fig.-2c Eddy current probe are placed at
90o to each other to measure the orbit of
the centre of the shaft

cracks, whirl and whip problem of journal

bearings.

d) Tachometer
Tachometer is used to measure the speed of a rotating component such as a shaft
and also used as trigger for synchronising data acquisition. There are two type
commonly used, namely, the fibre-optical and eddy current as in proximity probe. In
both cases it measures pulses generated from a marker mounted on the rotating
shaft. For fibre optic type it requires a reflector while the eddy current type requires a
notch on the shaft.

-2.1 Linearity Range

For sensitivity to remain constant,
Related are
websites:
transducers
required to operate within
http://sensorsits linearity
range. The transducers
transducers.globalspec.com/ProductFinder/Sensors_Transducers_Detectors
Linearity
Range, as shown in Fig.-3,
is the operating range where the response
is linear, i.e. the sensitivity remains
constant. Linearity can be affected by
frequency, temperature, distance, etc.

Fig.-3b The transducers Linearity Range

Fig.-3a Deviation from linearity affected by frequency,

temperature, etc.

3 Signal Conditioner & Amplifier

After leaving the transducer, the signal is conditioned in order to minimise noise during
transmission and error during digitisation process. All transducers have their respective
electronic signal conditioning units associated with each of them. In most cases,
transducers generate analogue signals (voltage or current) not conducive for
transmission over the cable length. It conditions the signal from the transducer into
voltage proportional to the measured physical quantity and transmits it. Also, these units
are sometimes used to provide

power-supply to the transducer. As mentioned earlier, in ICP transducers, it has signal

conditioner and amplifier built into the units.

4 Aliasing
Aliasing is a phenomenon where high
frequency components are shifted to a lower
frequency, which produces an inaccurate
representation of data. It occurs when
converting analogue (continuous) data to
digital (discrete) data. A classic example of
such happening is when we are in front of
our television screen watching a cowboy
wagon moving off from its station. We would
observe that as the wagon speeds up, the
wagon wheels appear to be slowing down
and eventually turning in the opposite
direction.
This is because the supposedly continuous
moving images are in reality made up of
discrete frames being scanned at a certain
rates per second.
Fig. -4 Effect of aliasing

--1 Nyquists Sampling Theorem

According to the Sampling Theorem, any signal can be accurately reconstructed from
values sampled at uniform intervals as long as it is sampled at a rate at least twice the
highest frequency present in the signal. Failure to satisfy this requirement will result in
aliasing of higher-frequency components, meaning that these components will appear to
have frequencies lower than their true values.
As indicated in Fig.-4, the sampling frequency must be at least two times the highest
frequency present in the data - Nyquists Sampling Theorem
For example, if the data contains data up to a maximum of 500Hz, the data need to be
sampled at least 1000 times per second, or 1000Hz. The question is, what is the highest
frequency contained in the signal to be sampled?

--2 Anti-Aliasing Filter

One way of avoiding the problem of aliasing is
to apply a low-pass filter to the signal, prior to
the sampling stage, to remove any frequency
components above the "folding" or Nyquist
frequency (half the sampling frequency). Such
anti-aliasing filters are commonly built into the
analog interface chips and codecs, which
convert analog input signals into digital form
for processing, by a digital signal processor
(DSP). In many cases, these anti-aliasing
filters are implemented using conventional
analog circuitry.
Hence, a
nti-aliasing filters are low-pass filters whose
cut-off frequencies are set at or below onehalf of the sampling frequency. For example,
as shown in Fig.-5, if the sampling frequency is
1000Hz, then it will require a low-pass filter of
cut-off frequency of 500Hz or less. However,
in a real filter, the cut-off characteristic is
rather tapered as shown in Fig. -5. Hence, in
selecting the frequency range
, fr, if you need to work on a signal within a
frequency range of 400Hz, say, you need a
low pass anti-aliasing filter of around 500Hz,
and your sampling frequency, fs has to be at
least 1000Hz or 1000 sample per second, as
described in Fig.-6.

Fig.-5 Anti-aliasing filter characteristics

5 Sample and Hold & Analogue to Digital Converter

After the signal is anti-aliased, a Sample and Hold circuit acquires and stores the
analogue voltage on a capacitor for a short period of time
Fig.-6ends
Selecting
frequency
range
. This is the point where the analogue form of the signal
and the
digital form
begins.
This storage time will allow the continuous analogue signal to be digitised into discrete
time series digital signal. The time between two sampling points is called sampling
period Ts and the sampling frequency fs is the inverse of Ts.

65536 grid

6 Dynamic Range
The dynamic range, as shown in Fig.-8, is
defined as the ratio between the largest and
smallest signals that can be analysed at the
same time.
It is affected by a number of components
namely:
1) ADC resolution (linearity and bits)
2) Input amplifier noise floor
3) Anti-aliasing filter performance
4) DSP chip performance
A 16-bit ADC will discretise the amplitude
range of the signal to a resolution of 65536
grid.
In most FFT analyser there is a function
called Auto-ranging where the analyser will
automatically amplify the input signal to
utilise these grids to the maximum.

Fig.-7 Digitisation of continuous analogue signal

Amplitudes are normally displayed on either

linear or logarithmic dB scale
Linear scale can be converted to logarithmic
dB scale by the formula:

dB 20log10

V
Vref

The idea of presenting a spectrum in dB

scale is to compress the low and high
amplitudes so as that they could be easily
visualised in a single display.
Fig.-8 Dynamic range

7 Time Domain

After ADC the continuous time signal are discretised into sequence of values commonly
known as Discrete Time Series. From hereon, data processing is performed digitally.
The signal can then be:
1) Displayed on digital screen
2) Stored in disk or
3) Further digitally processed
There
a)
b)
c)
d)

are various treatments that can be performed on the time series including:
Integration, for example, from acceleration to velocity
Differentiation
Band-pass Filtering
Statistical operation, for example, RMS, Average, Max, Min, Std. Deviation,
Variance
e) Arithmetic
f) Logical

8 Transforming from Time Domain

To Frequency Domain
A signal normally consists of many frequency
components superimposed. Hence, in many
cases cannot be visualise or interpreted
effectively. There is a requirement to
decompose the signal into its various
frequency components. One common way is
to use Fourier Transform to perform this task.
-8.1 Fast Fourier Transform
The algorithm for Fast Fourier Transformation
is derived from Discrete Fourier
Transformation, as described in Section 2.5.1.

9 Leakage
The requirement of FFT is that the signal has
to be periodic or infinitely long. For a time
frame, this implies that the signal has to start
at zero and to end its number of cycles at
zero. This does not happen in practice. As a
result, energy leaks out of one line to all the
other lines. This phenomenon is known as
leakage. To overcome this, we
utilise windowing functions. Windowing is a
process of forcing the end points of a time
frame to zero, as shown in Fig.-12

Fig.-12 Windowing process

10 Windowing
Windowing is a process of forcing the end
points of a time frame to zero. This will get rid
of the leakage, but at the expense of
broadening the bandwidth. This not too much
of concerned for high amplitude distinct peaks
but may affect the visibility of two close
frequency components. Zooming the
spectrum can be performed to overcome this.
10.1 Types Of Windowing Functions

Fig.-13a Frequency resolution for

Hanning and Flat top windows

There are a number of window functions developed, the

common ones used, especially in our scope of work, are:
a) Hanning
b) Flat Top
c) Exponential
d) Rectangular
Hanning and flattop windows are used mainly
for steady-stead signals. The choice depends on
the results required. Hanning gives narrow band
shape but maximum amplitude errors of
16%(1.5dB).
Flat Top gives broader band shape but
maximum amplitude error of 1%(0.1dB). These
are shown in Figs.5.13a, b and c.
Rectangular and exponential windows are
normally used in modal analysis, namely for
impulse and response signals, as shown in Fig.13d. Rectangular windows are used for signals
that start and end at zero. Typical example of
these signals is obtained from the impulse of
force transducer and the response from the
accelerometer for a highly damped structure.
Here, the signal decays to zero before reaching
the end of the time record. However, in low
damped structure, it takes a long time for the
response to decay to zero. For this, an
exponential window is being used to suppress
the tail end of the response to zero.

Fig.-13b Amplitude error for

Hanning

Fig.-13c Amplitude error for Flat

Top

Fig.-13d Rectangular windows on impulse and fast decaying

response signals and exponential window on slow decaying
response

11 Real-Time Bandwidth
Real Time operation can be described as a data processing with no loss of information
between input and output. In FFT analysis, time series have to be collected into time
record of n samples before it can be processed. For as long as the FFT processing time is
shorter than the time record, than there will be no loss of information. The length of time
record depends on the sampling frequency. It can therefore be seen that at lower
frequency range settings, implying lower sampling frequencies, the operation is real-time
as the FFT processing time is shorter than the time record. However, at higher frequency
range setting, implying high sampling frequencies, the time record becomes shorter. FFT
processing time is longer than the time record. This results in loss of a portion of input
time series. Hence, the operation is no longer real time. This is shown diagrammatically
in Fig.-1With the speed of computers rapidly increasing, the FFT processing time is reduced
allowing for the real-time operation at higher frequency range.

Fig.-14 Real and Non-Real Time operation

12 Overlapping Processing
At a lower frequency range, we need not wait for the time record to complete collecting
the n required samples before performing the FFT operation. We can fill up the current
time record with part of the preceding record so that the FFT operation can be re-initiated
immediately after completing the preceding cycle. This process is termed as overlapping.
This is aimed at reducing the processing time, especially when performing averaging of
low frequency range spectrum where the time record is longer due to inevitable lower
sampling frequency.

Fig.-15 Overlapping
Frequency Domain Analysis consists of single channel analysis and dual or multi-channel
analysis. Single channel analysis is sometimes termed as signal analysis that includes:
a) Forward FFT
b) Inverse FFT
c) Waterfall Spectrum
d) Colour Intensity Waterfall Spectrum
Dual or multi-channel analysis is sometimes termed as system analysis that involved
the study of input-output relationships. These include:
a) Auto-correlation
b) Cross-correlation
d) Transfer Function
e) Coherence
f)

Auto Power Spectrum and Spectral Density

g) Cross Power Spectrum and Spectral Density

13 FFT Analysers
Fundamentally, all FFT analysers in the market
consist of hardware and software to perform the
flowchart processes as shown in Fig.-1. They
come in different configurations to suit different
applications, such as:
a) Handheld analyser
b) PC-based analyser
c) Desktop analyser
Most of them have at least two channels
minimal required for system analysis. In most
analysers, the input terminal has the capability
to generate the bias current to power up the ICP
accelerometer and impact hammer.
The PC-based analyser normally comes with an
external DSP unit and is normally linked to the
PC using PCMCIA card. The speed of current PC
is sufficiently fast that we do not need the
dedicated FFT microprocessor chip. The
advantages of having a PC based analyser
running in Microsoft Windows environment
includes:
1)
Portability of data. The data can
be transferred from one program or
hardware to another quite easily for
further processing or for report
generation.
2)
Easy upgrading. Where the
system require software upgrading it can
be quite easily done without having to
take the unit back to the factory.
3)
Provide easy learning curve, as
most are us are quite familiar with the
intuitive flow of Windows environment.

Fig.-16a Handheld FFT

analyser

Fig.-16b PC-based FFT analyser

13.1 FFT Analyser versus Digital Frequency Analyser

The difference between these two types of analysers has been a confusing affair and
many engineers has mistakenly regarded both as one of the same kind. The fact of the
matter is that both differ in software as well as in hardware usage. The table -1 below
summarises the differences between the two.
FFT Analyser
Generally used for steady-state
signature such as constant speed
machine vibration

Digital Frequency Analyser

Generally used in situation where
there is a rapid change in the
signature such as vocal sound
pressure
Each sample of time series is
processed once it is available.

Requires a complete collection of

record of n samples of time series
before it is transformed into
frequency domain
Generate a spectrum of linear
Generate a spectrum of constant
frequency axis where the bandwidth
percentage bandwidth. For a 1/3is constant and equal to 1/T, where T
octave spectrum each bandwidth is
is the period of record
33% of its centre frequency.
Uses Discrete Fourier Transform
Uses a series of digital filtering
algorithm
operation
Table -1 Comparison between FFT and Digital FrequencyFig.-16c
analysers
Desktop FFT analyser with
accelerometer and impact hammer
used in Modal analysis

14 Field Tips - Instrument Specification

-1-1 Accelerometers
Fig.-17 shows typical specification sheets of an ICP accelerometer.
Among other things, one would like to pay attention to the following:
a) Sensitivity
Probably the most commonly used sensitivity is 100mV/g. It is most suited for general
machine and structural vibration. The voltage generated is well within the range to
take full advantage of the Analogue to Digital converter in the FFT analyser, hence
generating minimum error. In application where the low vibration is sought, one
would go for a higher sensitivity.
b) Linearity Range
Machine structure upper frequency response would not exceed 20kHz, and in most
application much less than that. However, in special application where one would
desire seek response at low frequency i.e. near to DC (0Hz) one has to pay more
attention to the low limit in the specification. General accelerometers would tend to
attenuate at frequencies below 2Hz.

c) Power requirement
Most analysers inputs have a constant current supply for ICP accelerometer as well
as for microphones. It is worthwhile to check ones analyser before purchasing the ICP
unit.
d) Output connection and cabling
In 2-pin connection, for the purpose of easy availability, one would be recommended
to purchase an accelerometer with a simple BNC connection. In situation when the
cable gets faulty, BNC terminated cable are normally easy to obtain. Microdot
connection requires microdot cables, which is slightly hard to seek, depending on
ones location. In cases where one has to use a 3-pin connection, then one has no
choice but to use cable terminals as specified by manufacturer.
-1-2 FFT Analysers
Fig.-18 show typical specification sheets of an FFT analyser. Among other things, one
would like to pay attention to the following:
a) Dynamic range
Most analysers have a dynamic range about 90dB when using a 16-bit AD converter.
This is good enough for a most mechanical vibration signature.
b) Frequency range
Maximum frequency of 20kHz is all we need for mechanical vibration.
c) ICP output
Although most analyser today comes with the constant current supply, it is
worthwhile to ensure this before purchasing one. It makes life in the field much easier
when one does not have to handle with too many cable connections and accessories.
d) Number of channels
A four-channel unit would be quite an asset if you were concern with the length of
data collection. In modal analysis, one can use the first channel as input and the
other 3 channels for the x, y, and z direction response. In this way, one can perform
the measurement in one-third the time if one were to use just two channels. For ODS
one can use the first channel as the stationary point input.
e) Windowing functions
The minimum windowing functions should be the Hanning, Flattop, Force, and
Exponential (Response) windows.
f) Portability versus Display functions
One needs to define the use of the analyser before purchasing one. If one needs to
carry the unit up and down an 8-storey structure then one would obviously go for a
light, battery powered handheld unit. However, such a unit normally sacrificed the
display capabilities. Hand held unit normally comes with an LCD display and limits the
number of traces displayed at one time. On the other hand, a Window-based analyser
is capable of displaying at least 8 traces at a time.
g) Compatibility with the post-processing software such as modal analysis
and ODS analysis software
This is crucial. One must ensure that the post-processing software can read data from
the analyser with minimal hassle. One is advised not to entirely rely on the
specification chart given by the analysers maker, but to have a complete article
describing a case study that made use of the software with an analyser.

15 Review

Digital signal processing has been a breakthrough in providing an instrumentation

solution to vibration analysis. The size and speed of processors have allowed advanced
analysis to be performed on site. When analogue analysers were used, it was normally
possible to carry out analysis in the laboratory, and has been time consuming. Today,
modal analysis can be performed on a complex structure in a plant with equipment that
can be carried in a briefcase.

Fig.-17 Typical FFT specification sheet

Fig.-17 Typical accelerometer specification sheet