Signal Processing
Topics
o Introduction
o Filtration: low pass, band pass, high pass
o Integration
o Sampling & aliasing
o The FFT process
o Sampling & Resolution
o Dynamic Range
o Windowing & leakage
• Hanning, Rectangular/uniform
• Force & exponential
o Averaging
• Linear averaging, overlap averaging & peak hold averaging
Quick Overview
Signal processing
o What is signal processing?
o Science of analyzing time varying physical processes.
o Types of signal processing
• Analog & Digital signal processing
OR
• Continuous and Discrete Signal Processing
What happens inside the data collector?
o The data collector takes the “analog” signal from the sensor and
provides us with data we can use:
• Waveform, spectra, rms readings, and more…
o There are lots of setting, so we need to have an idea what is
happening “inside the box”
What happens inside the data collector?
o Step-1:
• Supply power to the transducer (ICP)
• The signal from the accelerometer (sensor) is an electrical signal- its is an
analog “signal”
o Step-2:
• Convert the analog signal into a digital signal
• The data collector needs “number” to work with
What happens inside the data collector?
o There are a number of challenges:
• Does the voltage from the sensor match the input range of the data collector?
• How quick should we sample the signal?
• How many samples do we keep?
o Step-3:
• We need to create other data we can work with: spectra, rms overall
readings, envelope spectra.
o Challenges: Filtering, integration, sampling, windowing, and
averaging.
Signal Processing Techniques
Transformation
Windowing
Filtering Windowing
✓ Signal Addition
✓ FFT & its use
✓ Spectral Leakage
✓ Need to filter?
✓ Use of Windows
✓ Types of filters
Sampling Averaging
✓ Application
✓ Importance of Sampling SP techniques aid in ✓ Types of Averaging
✓ Terms to know the collection,
extraction & analysis
of data
Filters
Understanding filters
o Filters are used extensively in vibration analysis
o There are four type of filters:
Band Pass Low Pass
Types of
Band Stop Filters High Pass
Low pass filter
Data from a machine, no filter applied.
Same data but with a low pass filter applied.
High pass filter
Data from a machine, no filter applied.
High pass filter below the specified frequency
Band-pass filter
Data from a machine, no filter applied.
Band pass filter applied
Example (time-domain)
Filters application
o Anti-aliasing filters
• Handled by the data collector
• Either a serios of analog filters or one or two analog filters with digital
resampling.
o Integration
• Convert acceleration to velocity
o Tracking Filters
• Focus on 1X ( or 2X, etc.)
• Trach machine speed
o Demodulation/Peakvue/Acceleration enveloping
Fast Fourier Transform (FFT)
Fourier transform
o The Fast Fourier Transform or FFT is a digital algorithm that
converts a time waveform into a spectrum.
o It converts the time domain into the frequency domain.
o The formula for the FFT is shown in the figure below.
o When we compute the FFT, each line in the spectrum contains a
phase component
• Unless we perform cross channel measurement, or specifically take a phase
measurement, the phase is discarded.
Time record and the FFT records
o A time record is defined to be N consecutive, equally spaced
samples of the input.
• N is a power of two; typically 1024, 2048, or 4096
o The FFT is a calculation that creates a spectrum from the time
record.
• From the time waveform we get amplitude & phase
• The resultant spectrum actually contains N/2 lines (i.e. 512, 1024, 2048).
FFT “challenges”
o For the FFT to work correctly we have to deal with three
problems:
• Aliasing
• We need at least wo samples per cycle if we wish to resolve that frequency
• Leakage
• A loss of resolution if the cycles are not perfectly periodic in the time reord.
• Resolution
• We only have at specific frequencies - we don’t know what happened
between those frequency.
Sampling & Aliasing
Sampling the signal
o The electrical signal from the sensor is called the “analog” signal.
It is continuous signal.
o It represents vibration at that location in terms of velocity,
displacement or acceleration.
o The signal must be digitized to record and store the vibration
information
Sampling the analog signal
o We digitize the signal.
o We have individual readings at discrete tin interval.
o The rate at which the waveform is sampled is called “sample
rate” or sample frequency.
• “Number of samples per second”
Time domain
o The result is a series of numbers.
o This is the time waveform that the data collector can store and
use for the FFT.
o This is called the “ time domain”.
o Anything that happened between the sample is lost.
Sampling
o Th total amount of time of the recorded data is defined by
variable “T”.
o “N” samples or data points will have been collected within that
period of time.
o The sample period Ts and the sample frequency Fs are
interrelated as below.
Aliasing
o Aliasing is loss of information which occurs due to inadequate
sampling rate.
o It’s an effect that causes different signals to become
indistinguishable when sampled
Figure : A 50Hz signal sampled at a rate of 50Hz
Aliasing
Figure : A 50Hz signal sampled at a rate of 100Hz
Nyquist Theorem / Criterion
o Nyquist Criterion states that the waveform must be sampled at a
rate greater than twice the input signal.
o This frequency is often referred to as Nyquist Frequency.
Sampling time = ts
Sampling Frequency/Rate = fs
Nyquist Theorem: Fs > 2 X Fmax
o If we don’t sample quick enough, we get aliasing.
Anti-aliasing
o To eliminate the effect of aliasing, all signals too high for the
sampling rate must be filtered out
• A low pass filter is put in place that filters out any signal above the
sampling rate
o The low pass filter is set at Fmax to correct this phenomenon
o This filtering is called an Anti-Aliasing Filter.
o “Digital” filters or “Sigma-delta” method will be used for anti-
aliasing.
o Most modern data collector no employ the sigma-delta method,
as they have better performance.
Sampling & Resolution
Sample rate
o We can control two parameter:
• Sample rate (# samples per second)
• Total number of samples collected
o If we sample more quickly we can resolve higher frequency
signals.
Number of samples
o However, if we want to resolve two frequencies that are close
together, we need to collect more samples.
Resolution
o Each line in the spectrum is referred as cell or bin.
o We have a certain number of data points or “lines” and we
cannot see between the lines.
o The distance between two lines in the spectrum is called the
Resolution.
R = Fmax / LOR
Examples
o The following waveform contains 3 signals, a 6 Hz, an 8 Hz, and
380 Hz signal.
Examples
Sample time
Example
Windowing
Windowing & leakage
o When the FFT is used to measure the frequency component of a
signal, the analysis is based on a finite set of data.
• It assumes the time record before and after are identical.
o When the measured signal is periodic and an integer number of
periods fill the acquisition time interval, the FFT turns out fine as it
matches this assumption.
o n this example, Four complete cycles live within the time record.
Spectral leakage
o However it is seldom true that the time record starts and ends at
zero. More commonly they are similar to Figure.
o These artificial discontinuities show up in the FFT as high-
frequency components not present in the original signal
o The spectrum you get by using a FFT, therefore, is not the actual
spectrum of the original signal, but a smeared version.
Spectral leakage
o The real data in below Figure shows that the ends of each
sample do not have the starting and ending amplitude at zero.
o This phenomenon is called Leakage. The result is a broadening
of peaks within the spectrum.
Windowing
o The effects of performing an FFT over a non-integer number of
cycles can be minimized by using a technique called windowing.
o Windowing reduces the amplitude of the discontinuities at the
boundaries of each finite sequence acquired by the digitizer.
Windowing
o The window has a minimal effect on the frequency content, but it
does affect the shape of the spectral peaks and the amplitude
levels.
Comparison
o Comparison of non periodic sine wave and FFT with leakage
(left) to windowed sine wave and FFT showing no leakage.
Resolution & accuracy
o As we know each line in the spectrum is referred as cell or bin.
o We have signal of 30Hz of amplitude “1”
• We have a bin at 30 Hz
• Our spectrum will have a peak of “1”
Resolution & accuracy
o What if the vibration was at 30.5 Hz or directly between two
lines?
• If we are using a Hanning window we will get peaks at 30 and at 31 Hz,
each with an amplitude of 0.85 meaning our maximum amplitude error with
the Hanning window is 15%. ”
Window type & bandwidth
o Although Hanning window has best frequency accuracy, it still
affects it. To correct this phenomenon a window factor is used.
o The “window factor” for the Hanning window is 1.5.
o Our calculation of resolution must be altered to include the
window factor. We use the term bandwidth.
• Example: 1600Hz, 800Lines: line spacing 2Hz
• Hannning window: window factor = 1.5
• Bandwidth = 2 * 1.5 =3Hz
Types of windows
Flat Top Window- has better amplitude accuracy but the
frequency accuracy is poor.
Rectangular, Uniform, or No Window - the Rectangular
and Uniform are the same as having no window at all.
These can be used for impact tests for modal analysis
Hamming Window– similar to Hanning window except
the ends do not go to zero amplitude.
Exponential Window - A special windowing function for
minimizing leakage in lightly damped structures that is
used in impact testing.
Importance of resolution
o If the Fmax is not high enough you can miss fault.
o If the resolution is insufficient you cannot distinguish between
peaks.
o If the resolution is too high, you will spend too much time
collecting data.
o We will discuss how o sent Fmax in the analysis session.
Averaging
Averaging
o Observing only one section of time waveform may include some
peak caused by a random vibration influence.
o To minimize this, it is common to look at several sections of the
time waveform, calculate several FFTs, and display an average
result.
o Four averages are commonly taken.
o Averaging is available in most FFT analyzers to assist in
interpreting data.
o Averaging also provides more repeatable results in data
collection for early warnings of machine deterioration.
Types of averaging
Linear
Overlap
➢ Works by collecting a time
record, windowing the data, In case of overlap averaging,
converting it to an FFT and instead of using an entirely
storing it in temporary memory new time record for each
➢ Then collecting another time average, part of the previous
record, producing the FFT and record is used.
adding it to the average Time Synchronous
➢This is repeated for each
average. The average spectrum ➢Linear averaging does not
is then displayed. actually remove noise , it just
reduces noise
Peak Hold ➢ Averaging the time data can
actually reduce the level of
➢ Holds the highest peak for each line of the spectrum noise by removing noise from
➢ Used for special tests such as Run-up, Coast down, and the data
bump tests ➢ It can therefore be used to
➢ This method is very useful in viewing transients or for uncover low level signals that
stress analysis calculations. may have been obscured by
noise
Linear averaging
Figure: Linear Averaging with four samples
Overlap averaging
Figure: Overlap Averaging with four samples
Peak-hold averaging
Exercise
o Frequency Range= 10KHz
o No. of Samples = 1024
o Data Collection Time = ? LOR= No. samples/2.56
Time= LOR/Fmax
✓ 40msec
o Average = 04
o Total data collection time = ?
✓160 msec
Exercise
o Overlap = 75%
o Data collection time after the first block = ?
✓ 10msec
o Total data collection time?
✓ 70msec
o Time saved (in case of 4 averages) = ?
✓ 90msece
Exercise
o Frequency = 100Hz
o No. of Samples = 1024
o Data collection time = ?
o Overlap =75%
o Data Collection time after 1st block?
o Time saved (incase of 16 averages) = ?
Conclusion
Analog Signal
Anti- Windows
A/D
Input Aliasing & Input
Converter
Filter Buffer
Display & Averaging FFT
Storage
Thanks
