Machinery Fault Diagnosis and Signal Processing

Prof. A. R. Mohanty
Department of Mechanical Engineering
Indian Institute of Technology, Kharagpur

Lecture – 16
Computer Aided Data Acquisition

In this class, we are going to talk about computer aided data acquisition. As you know in
the previously when we talked about signal analysis, particularly finding out the features
of the signal be it the RMS value, be it the mean value, we had certain integrations to be
done where we required a mathematical function for a signal as simple as that you know.

(Refer Slide Time: 00:39)

Suppose I want to find out mean of a signal, I need to integrate the signal over a time
period. Now the problem there lies that if I have a signal of this nature which is coming
out of my machinery, I do not have an expression or a mathematically equation to this
curve. If it has a nice sin wave where x(t) = Asinωt, I am sure all of us have done such
integrations even; first year electrical circuit courses and we have very happy and
comfortable to find out features of the signal, ok.
(Refer Slide Time: 01:30)

But then this could be done in a better way instead of x(t) representing this signal, I have
closely spaced point x i .

So, if I have such sequence of x i I can analyze them like nothing, but i = 1 to N / N. So,
this is a simple example how if we have this as an analog signal, and here I have a digital
equivalent of that signal. So, the job of us right now is to how to convert this analogue
signal into a digital signal. So, that all this numerical computations bit just finding out
the features, even finding out the frequency domain coefficients of such signal we can do
that and that is actually what is done in Digital Signal Processing, and just to recap as
you know in CBM our machines condition actually depends on what we infer from the
signal obtained by this transducer x(t).
(Refer Slide Time: 03:08)

What do we understand from this as signal x(t) and. So, that is why we have on a nice
module of signal analysis, but to do any mathematical signal analysis or DSP, I need to
convert this signal from analog to digital.

(Refer Slide Time: 03:54)

So, this is what we are going to focus on in today’s class. So, data acquisition is values
which are discretely sampled in time, discretely means there is a definite time interval
between the signals. So, this Δt is known as the sampling interval, a device hardware
device does this. So, it will give you an corresponding analog value to this discretized
signal x(t) and that is this hardware which we are going to talk about.

(Refer Slide Time: 04:39)

And then another quantity is resolution, what is the minimum value the signal can be
sensed I will give you an example here, because see all the signals which are digitized I
have an analog domain and then I have a digital domain.

(Refer Slide Time: 05:05)

So, in analogue domain I have a analogue voltage may be from 0 to 10 volts this is my
transducer output. So, in transducer gives me maximum value it can give to 0 to 10 volt
analog signal.

Ok, but then once I do an A to D conversion, I should be able to store this n digits as you
know in digital domain I can store a value either as a 0 or a 1. So, there are 2 options. So,
if this is a 23 bit for the sake of discussion system. So, the possible values of zeros this
guy can have is all the way from 0 0 0 to 111 So, I will have 8 values. So, whatever be
the analogue signal which is coming to the system my A to D is only be able to convert
into digital values and only have 8 positions or locations.

(Refer Slide Time: 06:40)

to store this value. So, this amplitude resolution is nothing, but your total voltage. So, in
our case 10 volts divided by 23, 10 by 28 is 1.25 volt. You see this bit size plays an
important role in the resolution of this A to D converter. So, one thing that we have to
keep in mind, one is the sampling interval I will give an example of the sampling interval
and other is the bit size apart from few other things.

So, today if I go to by hardware for the analog to digital conversion.

(Refer Slide Time: 07:49)

These are the 2 important features one has to look for all the specifications, and by the
way sometimes the sampling interval is related to the sampling frequency. F s is nothing
but 1 by Δt. So, this also creates another problem; imagine, I have an analogue signal of
this nature and I sample at few points shown by this red dots; that means, what I am
picking values equally spaced sampling intervals. So, if I join this red dots I will get my
original white signal, but instead of Δt suppose I pick another sampling interval at a
different sampling rate one is this x and so on. So, you see if I join this I am getting a
straight triangular wave.

So, and this is the new sampling interval Δt*, where Δt* > Δt. So, I have got this
triangular wave as opposed to the original sine wave. So, I have created by inadequate
sampling frequency I have created are or aliased or picked made a signal wrongly
represented in the frequency domain.
(Refer Slide Time: 10:07)

I will continue this discussion, but just to tell you this analog to digital conversion
methods there are 2 methods, one is the flash conversion, another is the sigma Δ
conversion will not go in the details of this just to know that when you specify you have
to specify either of these 2 processes.

(Refer Slide Time: 10:22)

So, the errors because of data acquisition one is the sampling errors, which we discuss is
known as signal aliasing and other is the quantization error because of the inadequate
resolution I will explain it through a better diagram here.
(Refer Slide Time: 10:38)

So, if you see here the same signal, this is my black line or continuous line in the original
signal ok.

Now, because of a defect in this machinery, I have got some glitches in the signal which
could be because of some machine faults and this has to be captured by me, but if you
see if I have a 3 bit computer. So, these are the actual values 0 0 0 to 111. So, if I have
the amplitude resolution this is nothing, but 10 divided by 23 or 10 by 8 this is 1.25. So,
the data acquisition system only understand 0 to 1.25, 2.5, 3.75 and so on volts. So, in
this case of 5 volt analog signal is represented by this line, and the next point it comes
here which is only 6.25. So, it is not able to decide the minor variations between 5 to
6.25, because the least it can sense is only 1.25. So, in that token what I get is a jagged
wave like this, which is my digital representation of my analog signal which is quite
different well how could this be improved this could be improved if I had final
resolutions.
(Refer Slide Time: 12:23)

So, this amplitude resolution which I am talking about is nothing, but the maximum
range by 2n where n is a bit size, if maximum range was 10 volts and 3 bits we saw this
as 1.25 volt.

If I increase this, bit size you know 24, 16, 25, 32 and so on this is with increase in bit
size the denominator increased. So, this will reduce. So, one way to improve this
amplitude resolution is to reduce the quantization error. So, quantization error in a data
acquisition system can be reduced by increasing the bit size.

(Refer Slide Time: 13:33)

So, if you see the numerical accuracy of a 12 bit laptop is less compared to a 24 bit or a
32 bit or a 64 bit, and you have seen nowadays in the market if you buy a computer
system with higher bit size they are highly priced because they have give you a better
resolution errors will be less as we increase, are less. Because my objective is to
faithfully represent my machine signal whatever I do, I need to get back my original
signal after the data acquisition system and this is known as x i this is one case we have to
do.

(Refer Slide Time: 14:31)

Now, another related to the sampling error is this case here. So, you see I have the
original black signal, if I now sample it at lower rate of the time period is high, I will get
back get my dash line this is actually a low frequency signal.
(Refer Slide Time: 14:57)

So, signal is aliased as a low frequency signal so in fact, there is a theorem which says
your sampling frequency should be greater than twice of f max to prevent signal aliasing.
So, this is the Shannon’s sampling theorem. So, to prevent signal aliasing we have to do
this.

(Refer Slide Time: 15:52)

But the question is many a times in this A to D devices, I have my machine here and I
have a transducer, I am giving an analogue signal and I have one A to D converter,
which comes in the market with the sampling frequency and then I have to put some
device here so that I get no aliasing in my digital signal. So, this is of course, a digital
signal x i , and this is this is the analog, this is also analog and this is my transducer. So,
what we can do is if I put a filter. So, that it does not allow this is the certain role of
cutoff frequency of this filter.

Such that it does not allow any frequency beyond F s / 2 to pass on to the analog filter
because this is known to me. So, I can set the this is a low pass analog anti aliasing filter.
So, if I go to the market and if I buy an A to D converter with the sampling frequency, I
can always have the low pass anti aliasing filter put as a front and before that A to D
process some of them have some of the A to D converter is have such filters before them
or I can put an additional low pass analog anti aliasing effected with the cutoff frequency
f c = F s /2. So, that any signal with comes in which I get x i here is never aliased, this is
very important thing which we have to keep in mind. So, as you know in CBM there are
many signals for example with time with this is a temperature signal as opposed to time
for a vibration signal.

(Refer Slide Time: 19:02)

Temperature signals do not change so rapidly unlike a vibration signal. Vibration signal
could be changing very fast, but a temperature signal is very slow slow. So, this is a very
very low frequency signal, and this is a usually high frequency signal. So, the data
acquisition system A to D converters, for a temperature recording conversion is actually
I can have a low frequency signal sampling frequency. In fact, they are much much
cheaper compared to the data acquisition devices used for dynamic signals which are
vibrations.

(Refer Slide Time: 20:23)

So, you know; obviously, people sometimes ask me what is the typical frequency range
of vibration signals for CBM, I say you know if you look into IC engines and motors you
know though the if I talk about the audio range from 20 hertz to 20 kilohertz. Vibration
signals are good up till 10 kilohertz; there is enough information to do any CBM;
temperature even less than around 100 hertz, 10 kilo hertz very high. Now there are few
other devices for example, you will see lot of ultrasonics which are used for in radars
etcetera sonars etcetera this is greater than 20 kilohertz depending on some underwater
acoustic applications they can go up to you when even, you know, 200 to 500 kilo hertz,
particularly some they underwater sonars and. If we look how acoustic emission is
something which will discuss later on, the maximum frequency of the signal could be
around 5 megahertz see you can see different signals, and there frequency content
because each, to get a meaningful representation, they are all have a sampled at F s >
2F max .

So, this has to be kept in mind. So, if today if you go to the market to buy data
acquisition systems, we will see data acquisition systems to different sampling frequency
is because of this obvious reasons one is to prevent signal aliasing and other is 2 have
higher bit size so that the amplitude resolution is fine enough.
(Refer Slide Time: 22:35)

But then there are few other features of A to D converter, one is this maximum input
range its usually you know 0 to 10 volts polarity, sometimes you know whether they will
be bipolar or unipolar gain sometimes some of this data acquisition units the least
sensitivity or amplitude resolution is so low, that even a thermocouple signal giving few
millivolts cannot be sensed.

(Refer Slide Time: 23:07)

So, we need to have a pre amplifier to give a gain, some gain, 10 x or whatever 100 x.
So, that they are greater than the amplitude resolution of the unit of course, nowadays
you know this one open loop process in which we are doing in CBM, but may of the
closed loop operations wherein after you take the signal through an A to D converter,
you get do an analysis and you want to control through a mechanical feedback and
actuator. So, we need to have what is known as the digital to analog.

So, some of the units have D 2 one that is for a closed loop control system and so on.

(Refer Slide Time: 24:14)

Some of this because you know finally, this system has showed the values 0 0 0 to 111
where do I store them, I store them in the random access memory of my computer
systems some RAM memory is required. So, this was an 8 bit system. So, each digital
sample will take one byte of storage space. So, memory also comes with the cost. So, if
my have higher resolutions the storage space would increase and then that would lead to
having more number of RAM required because you know finally, this digital data needs
to be stored somewhere . So, I have a bus digital bus on to which I put my A to D
converter digital and from this bus I can have a memory location or I can put on board
memory. So, and this bus is an architecture which is prevalent in today’s computers you
know started, I will talk about this is different bus architectures which are available today
and then there are provisions we just talked about one channel.
(Refer Slide Time: 25:46)

Imagine I have a machine where there are hundreds of analog output. So, some of them
have to be acquired simultaneously some after them one by one. So, in an A to D
converter simultaneously how many channels I can digitize is also function and so on.
Because the data storage would capability would reduce by having more channels and
many a times you will see later on only we talk about wireless data acquisition how this
signal can be transported through Wi-Fi to another device and; what are the limitations
of signal transmission over Wi-Fi frequency range and so on.

(Refer Slide Time: 26:44)

Sometimes you know when you have 1 2 converter one digital converter, but there are
many input and we can have a analogue or digital switching circuit which is known as
the multiplexing and then sometimes you know we will trigger and synchronize, and end
of the day we have who is who is controlling all this. So, there has to be a user interface
software to control all this.

So, in in the universities or in IIT is we have a very popular software Lab view which is
of the company, and then this is used to control the data acquisition units and
communicate with the computers and then today it was analog signals for CBM, it could
be vision signals, it could be in you know motion signals, and so on. And another thing is
of course, there any other features, which are available in data acquisition unit; one is
triggering when at what time do I need to sample a signal.

(Refer Slide Time: 28:00)

So, only when a max has occurred I will start acquisition. So, this is called as a trigger
level, only when a max has occurred or event has occurred I will acquire the acquire the
signal. So, this has to be taken into account when you are doing data acquisition. So, just
to recap you know there are lot of features in data acquisition units, but most important is
quantization error which can be improved by increasing the bit size, and other is the
aliasing error which can be improved by increasing the sampling frequency or having a
low pass anti aliasing filter.
(Refer Slide Time: 28:47)

Now, coming to this bus that is once I have the digital data I need to transfer them of
course, we use to have the GPIB transfer protocol, which had a speed of 8 megabytes per
second.

We had the RS 232 it is very slow, but still people use it 232 kilobits per second firewire
is a 3.2 gigabits per second you know which is very popular in the max systems today we
have the USB systems, and then there is to be PCMCIA at 20 megabytes per second. So,
faster computers have you know faster speeds you know we have 3.2 today when in the
data transfer when in wireless communications we are talking about G, 4G you know
people are talking about 5G. So, this is only how fast we can transfer the data if you can
transfer the data first you can acquire data large number of datas store large number of
datas and process dynamic signal through will talk about data recording and data transfer
later on.
(Refer Slide Time: 29:57)

And then PCI interconnect it is a clock speed of 33 megahertz, which is replaced the ISA
EISA bus.

(Refer Slide Time: 30:06)

We also use the Ethernet once the data is in digits and of course, we have the CAT 5
cable connecting devices we have the wireless Ethernet and so on. So, today is world be
it machines, computers, mobile devices, we need faster ways of transferring digital data
through this protocols which are available and then we can do our processing and make
an analysis.
(Refer Slide Time: 30:38)

Of course you know there are network based data acquisition system. So, unlike a client
server in a model. So, you could be sitting somewhere and controlling your acquisition
and so on.

(Refer Slide Time: 30:49)

We will continue more on this in the subsequent classes.

Thank you.