ME444: Analysis and Design of Mechanisms

Lab 4

(Vibration Data Analysis)

Group 8
Date: March 27, 2025

Sr. No Name Roll Number

1 Ayush Lonakadi 23B2253
2 Varun Inamdar 23B2203
3 Aagam Kasaliwal 23B2215
4 Lakshaditya Singh 23B2230
5 Anirudh Maindargikar 23B2219

Code implementation: MATLAB code

1.​Introduction:

●​ Objectives

The primary objective of this lab is to analyze vibration data using signal
processing techniques, focusing on understanding the effects of sampling
frequency and windowing on the acquired data. Specifically, the experiment
aims to:

1.​ Investigate how altering the sampling rate impacts the accuracy and
quality of the reconstructed signal.

2.​ Examine the role of windowing functions in mitigating

discontinuities and aberrations in Fast Fourier Transform (FFT)
analysis.

3.​ Apply FFT to analyze the frequency content of vibration data and
interpret the results.

●​ Background of Experiment

Vibration data analysis is critical in engineering applications, as it helps

identify frequencies and amplitudes associated with mechanical oscillations.
However, acquiring and processing such data involves challenges, including
noise contamination and signal discretization. Digital measurement systems
sample continuous signals at discrete intervals, and the choice of sampling
frequency is governed by the Nyquist-Shannon theorem to avoid aliasing.
Additionally, signal processing techniques like filtering and windowing
enhance data quality before analysis.
●​ Methods and Terminologies

Sample and Sampling Frequency

Sample: A sample is a discrete value of a continuous signal captured at a

specific instant. This experiment sampled the vibration data at different rates
(e.g., 1385 Hz, 27700 Hz) to observe its impact on signal reconstruction.

Sampling Frequency: It defines the number of samples taken per second

(Hz). Higher sampling frequencies approximate the continuous signal more
accurately but require more computational resources.

Figure 1: Low sampling rate v/s high sampling rate

Aliasing

This is a phenomenon where a signal is misrepresented due to insufficient

sampling. If the sampling frequency is too low, high-frequency components
in the signal appear as lower frequencies, distorting the output. The
experiment demonstrates aliasing by sampling below the Nyquist rate.
 Figure 2: Aliasing

Nyquist-Shannon Sampling Theorem

This theorem states that a signal can be perfectly reconstructed if the

sampling frequency (fs) is at least twice the highest frequency (B) present in
the signal (fs > 2B). The experiment validates this by comparing sampled
signals at frequencies below, at, and above the Nyquist rate (e.g.,2 × 1385Hz
= 2770Hz).

Fast Fourier Transform

FFT: An algorithm to convert time-domain signals into frequency-domain

representations. It decomposes a signal into its constituent sine and cosine
waves, revealing amplitude vs. frequency relationships. The FFT experiment
identifies the four frequency components (f1 = 1385Hz, f2 = 100Hz, f3 =
275Hz, f4 = 495Hz) in the given function -

f(t) = A1sin(2πf1t) + A2sin(2πf2t) + A3sin(2πf3t) + A4sin(2πf4t)

Windowing

Fourier analysis inherently assumes that signals are infinitely periodic,

extending from −∞ to ∞. However, real-world measurements capture signals
over finite time intervals, meaning the FFT algorithm must artificially
extend the recorded data by repeating it periodically. Since signal acquisition
starts and stops at arbitrary points, the captured segment may not contain
complete cycles of all frequency components. This mismatch between the
initial and final values of the finite signal introduces discontinuities when the
FFT replicates the data. These abrupt transitions distort the frequency
spectrum, producing artifacts and spectral leakage.

To counteract this issue, windowing is applied. A window function smoothly

tapers the signal's edges to zero, eliminating abrupt jumps when the FFT
replicates the data. By enforcing continuity at the boundaries, windowing
 minimizes spectral leakage and ensures a more accurate representation of the
signal’s true frequency content.

Mathematical Basis of Windowing

A finite-length signal x[n] of length N is multiplied by a window function

w[n] before applying the FFT:

xwindowed[n] = x[n] . w[n], 0 ≤ n ≤ N - 1

The window function w[n] is designed to taper the signal smoothly to zero at
both ends, reducing abrupt transitions when the FFT assumes periodicity.

●​ Different Windowing Functions used:

We have used the following Window functions in this experiment - including

the Hann, Welch, Hamming, and Tukey windows.

1.​ Hann window function

 Figure 3: Hann window function

2.​ Welch window

Figure 4: Welch window function

3.​ Hamming window function

 Figure 5: Hamming window function

4.​ Tukey window function

Figure 6: Tukey window function
2. Methodology:
The methodology involves several key steps:

●​ The values assigned to our group were:

Table a: Assigned values of amplitudes

A1 (m/s2) A2 (m/s2) A3 (m/s2) A4 (m/s2)

95 165 80 250

Table b: Assigned values of frequencies

f1 (Hz) f2 (Hz) f3 (Hz) f4 (Hz)

1385 100 275 495

●​ The function given is:

f(t) = A1sin(2πf1t) + A2sin(2πf2t) + A3sin(2πf3t) + A4sin(2πf4t)

●​ ​
For changing the sampling rate, the original sampling frequency was
 converted to a rational function as per requirement, and arrays of time
and signal were generated using this function.
●​ For the first part, the change in the sampled signal with the sampling
rate was observed. The initial frequency was considered to be the
maximum frequency given, f = f1 = 1385 Hz.
●​ A suitable sampling frequency to prevent aliasing was identified by
sampling the data at various sampling rates using the data given.
●​ Different windowing functions (Hann, Welch, Hamming, and Tukey)
were used for the second part, and FFT was obtained for both the
original and the sampled functions.
●​ The windowing function which gave the best FFT was found by
comparing the plots obtained.

3. Results
​ Data plots

Figure 7: Raw unfiltered data

 Figure 8: Filtered (orange) and Unfiltered(blue) data

Part 1: Variation of sampled data plots with sampling

frequencies
Plots at different sampling frequencies

The data plots and their corresponding FFTs have been plotted for various
sampling frequencies (f, 2f, 5f, 10f, 20f, and 30f), where f = 1385Hz.
Figure 9: Sampled data plot and FFT at f = 1385Hz
Figure 10: Sampled data plot and FFT at f = 2770Hz
Figure 11: Sampled data plot and FFT at f = 6925Hz
Figure 12: Sampled data plot and FFT at f = 13850Hz
Figure 13: Sampled data plot and FFT at f = 27700Hz
Figure 14: Sampled data plot and FFT at f = 41550Hz
Part 2: Variation of windowing functions at a given sampling frequency

Plotting the FFTs for different windowing functions

The sampling frequencies greater than 5f do not show appreciable differences in

peak values; hence, the plots have been generated for frequency = 5f (optimum
frequency) with different windowing functions.

Figure 15: FFT at frequency = 6925Hz for Hann window function

Figure 16: FFT at frequency = 6925Hz for Hamming window function

Figure 17: FFT at frequency = 6925Hz for Welch window function

Figure 18: FFT at frequency = 6925Hz for Tukey window function

 Figure 19: Signal after applying one of the windowing functions

4. Discussion​

●​ In the first part of the assignment, the sampling frequency was analyzed by
analyzing and comparing graphs at various sampling rates. The peak
frequencies show negligible deviation for frequencies greater than 5f
(6925Hz). Thus, the optimum frequency of sampling is taken to be 5f. This
is a slight deviation from the Nyquist - Shannon sampling theorem. The
possible causes include non-ideal filtering, noise and quantization errors, and
windowing effects.
●​ The choice of the "best" window function depends on the specific
application and requirements in signal processing, such as reducing spectral
leakage, improving frequency resolution, or maintaining amplitude accuracy.​
​
Observing the given plots and data, the Tukey window gives the most
optimum results. It gives comparable amplitudes at two frequency values,
better than the other three windows offer.​
Tukey Window is best for flexible applications, allowing tuning between
rectangular and Hann characteristics.

Tukey windowing is useful when you adjust the trade-off dynamically based
on signal properties.
 ●​ In the second part, the frequencies obtained do not change with a change in
the windowing function employed. The change is observed only in
amplitude, owing to the amplitude scaling effects of windowing and the
energy redistribution.

Based on these criteria, we determined that the Tukey window was the most
appropriate choice.

5. Conclusions:
●​ The best-estimated sampling frequency is 6925 Hz.
●​ The Tukey window function proved to be the most effective
windowing method.
●​ Vibration analysis is extensively used in predictive maintenance,
utilizing vibration data to assess the performance of bearings, gears,
and electric motors, thereby evaluating overall equipment health.
●​ This technique can identify issues such as imbalance, misalignment,
resonance, deformed shafts, and pump cavitation.

​ Sources of error

●​ Since the graphs are visually inspected, pinpointing the sampling

frequency is challenging.
●​ Extremely steep slopes near the peaks can heighten the likelihood of
errors during the FFT process.
●​ Additionally, any inaccuracies during the execution of the
methodology can contribute to errors.​

6. Contributions:

Name Roll Number Contributions

Varun Inamdar 23B2203 MATLAB coding, plot generation
Aagam Kasaliwal 23B2215 Introduction, methodology, tables
Ayush Lonakadi 23B2253 Methodology, plots
Lakshaditya Singh 23B2230 Results, Discussion
Anirudh Maindargikar 23B2219 Report verification, Conclusion

7. References:

●​ https://www.sciencedirect.com/topics/computer-science/windowing
(Windowing - an overview, 25th March, 2025)
●​ https://www.mathworks.com/help/matlab/ref/fft.html (FFT -
MATLAB, 25th March, 2025)
●​ https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_samplin
g_theorem#:~:text=The%20theorem%20states%20that%20the,a%20d
igital%20signal%20processing%20function. (Nyquist - Shannon
Sampling theorem, 26th March, 2025)
●​ https://moodle.iitb.ac.in/mod/assign/view.php?id=57512 (Lab 4 -
Moodle, 20th March, 2025)