Experimental Analysis of Liquid Sloshing Using an Image Processing Technique

Ehsan Badri Koohi1, Pouria Khalaj2, Elmira Zohrevandi3, Roohollah Dehghani Firouzabadi4
Sharif University of Technology, Azadi Ave., Tehran, Iran

Abstract
In this study the graph of vibration of individual components of
a fluid sloshing versus time, is provided using image
processing and coding techniques on the captured video of the
fluid motion. The main stage of image processing has been
developed in Simulink workspace. In this level the components
of the surface are detected and distinguished from the other
layers and their related pixels’ coordinates are achieved.
During the next stage using concise professional coding
Figure1 - Simulink Block set arrangement
techniques noise reduction has been performed and then the
absolute data of each frame is coordinated precisely and the
As the main feature it uses a decimation number (n) to
graph of longitudinal vibration of a specific component is
capture (n) frames per second. It transfers the data in a four-
achieved. Although most of the research has been progressed
dimensional array in which the first two dimensions specify the
with matlab various image processing features, but the main
pixel coordinates. According to what “edge detection” block
body of noise reduction method has been done by self-
has done the third dimension identifies whether the pixel is a
innovative solutions. To check the result accuracy another
surface component or not (a surface pixel gets a “1” value
graph is drawn representing the position of all particles in a
while a non-surface one gets a 0 value).the last dimension
specific moment of the fluid motion; both “particle position in
provides the number of captured frames in the video that’s been
the x direction” of the first graph and the “specific moment” of
specified by the user due to the fluid’s natural frequency. A
the fluid motion in the second graph will be provided by the
video viewer block has been used to examine edge detection
user meaning that the code is capable of providing both “the
accuracy and the result of noise reduction techniques.
longitudinal vibration of all particles for the whole captured
movie duration” and “the position of all fluid particles in every
Noise Reduction Methods
individual moment of the motion”.
In this stage noise reduction method operates on the transferred
Keywords: Liquid Sloshing-Image processing-Matlab
matrix data to omit the non-surface components and also
R2009a-Simulink-Noise Reduction
noises. The first method of noise reduction is described as
bellow;
Introduction:
Among those pixels with “1” value the number of the
With the recent trend to use vibration analysis of fluids during
pixels with “1” value in their neighborhood will be checked. If
every individual moment of their fluctuation, it is neither
there are less than specified threshold scale factor value number
feasible to capture each moment of the motion manually nor
in ones neighborhood, that pixel is undoubtedly a noise. The
economically efficient to use high-tech multi-frame capturing
threshold scale factor value has been determined in a way that
cameras with noise reduction technology even though if that
among all the pixels in a column, the pixel with the highest
technology is affordable it may have not been designed for this
number of “1” pixels in its neighborhood will be considered as
specific purpose. Simulink library block set of video and image
the surface pixel and its coordinates will be saved. The radius
processing provides limited number of frames out of a captured
of this neighborhood has been selected to be equal to the width
video. Then pixels data for each frame will be transferred to a
of the captured video, in this way a set of noise pixels which
matrix in Matlab workspace. The number of frames should be
may have coincidently formed the shape of a line will be found
specified by the user depending on the natural frequency of the
and omitted. All other noise pixels with different shapes will be
fluid, meaning that for the results to be more accurate; the
omitted using other methods which are stronger. In the first
higher the natural frequency of the fluid is, the more the
method the coordinates and the frame number of the only
number of captured frames per second should be. This feature
remained pixel with the highest number of “1” value pixels in
has been used in order to comprise microscopic movements of
its neighborhood will be saved. Then the second method will
the fluid particles for applied coding techniques to do the noise
be applied on the video. In the second method both the
reduction precisely, for this to be done, a couple of image
weighted variance ( and weighted mean ( will be
processing methods such as noise reduction and component
calculated for all edge pixels (white pixels). The pixels with the
verifying should be done first.
number of “1” value pixels in their neighborhood, less than the
value of ( , will be omitted. The program has been run for a
Image Processing Method
test and as a result of first method application, the pixel with
In order to analyze the surface vibration, in Simulink library,
the highest number of “1” value pixels in its neighborhood is
"edge detection" as a function block detects the surface of the
selected as the surface pixel of the 160th column. The pixels
fluid. For this to be done, a specific threshold scale factor is
location respect to one another is shown as bellow (the gray
determined to measure the threshold value of the edge thinning.
pixels are the ones in the 160th column);
To transfer the pixel data of surface components, “video to
In this column the second gray pixel from the top with is
workspace” block has been used.
selected by the first method as the surface component of the
column “160” with the highest number of “1” value pixels in
1- Research fellow its neighborhood. In this method the longitudinal distance of
2- Research fellow the surface pixel from the top (the first dimension of the
3- Research fellow transferred matrix) is considered as a criterion for vibration of
4- Assistant professor the fluid particle in y direction. To draw the graph of
longitudinal vibration of a specific particle, plotting variations
of y coordinates (the first dimension of the transferred matrix)
versus the frame number of a specific pixel defined by its x
coordinate (the second dimension of the transferred matrix) is
only needed.

Figure 2 – First method of noise reduction, counting the pixels in the

neighborhood of edge candidate pixels Figure 5 - Position of the surface in the captured movie

Analysis Result
To run an experimental test in order to do a comparison
between the results, a cubic reservoir is built and a video of the
water sloshing vibration has been captured.

Figure 3 - Self-built reservoir for experimental analysis

Figure 6 - Detected Surface edge of water sloshing

The program is run for 10 seconds and the longitudinal

vibration of a component with the distance equal to “160”
pixels from the left wall, (the width of the frame is 320 pixels)
has been analyzed versus time.
200
Distance Form The Bottom In Pixels

150

100

50

0
0 50 100 150 200 250 300
Distance Form The Left Wall In Pixels
Figure 4 - Longitudinal vibration of a surface component of water
sloshing versus time
Figure 7 - Noise reduced surface of water sloshing
To assure about the results, the graph of the position of
all surface particles in the 10th second of the motion needs to be
References
drawn; the 10th second of the motion is inserted from the movie
1. R.A. Ibrahim, Liquid Sloshing Dynamics: Theory and
To check the accuracy of edge detection method, the
Applications. Cambridge University Press, Cambridge, 2005.
motion has been paused at the 10th the second on Simulink
2. Advanced Digital Signal Processing and Noise Reduction,
video viewer and the noisy detected edge is shown as bellow
3rd ed., Wiley, Sussex, 2006.
Finally to check the accuracy of noise reduction method,
3. Matlab R2009a Help Documentation
the graph of the status of all surface components in the 10 th
second of the motion, has been drawn by the program and the
result is shown as below; as a result of the low quality of the
captured video in addition to general trend one can see small
fluctuations.