Development of a SCADA system with Rescaled

Range (R/S) Analysis and Image Acquisition

Armando Campos Domı́nguez, Eliseo Hernández Martı́nez

Oscar Velázquez Camilo Xalapa Faculty of Chemistry
Veracruz Faculty of Engineering Universidad Veracruzana
Universidad Veracruzana Xalapa, Veracruz, México
Boca del Rı́o, Veracruz, México [email protected]
[email protected], [email protected]

Abstract—Time series and imaging using fractal analysis have (SCADA), in seudoline technique that was implemented to
been implemented in many areas of science and engineering, acquire micrographs of the crystallization process of sugar
economics, biology and statistics among others, for their ability cane in the first instance. This technique is implemented for
to detect auto-similarities or irregular behavior patterns. Such sugar cane analysis, but it also can be used for any particulate
advantage has represented a significant advance in the monitoring process, to analyze the image based on fractal theory and
and control of variables processes and research. But all this has
been done in most cases with the help of software trademarks and
correlating it with physical properties like crystal growth of
sophisticated equipment that often do not realize the possibility the process in question. This is because recent studies suggest
of acquiring or present the results in a user-friendly graphical that fractal analysis could be used to track the evolution of
interface to analyze online or process variables seudoline their many processes or variables that at first glance cannot be seen
fractals descriptors. Therefore, it has been developed in graphic using conventional techniques.
language (language-G) a program that performs the analysis
of process variables crystallization of cane sugar for online II. DESCRIPCTION OF THE METHOD OF
monitoring of their acquired images and evolution crystallization ANALYSYS
with fractal descriptors.
A. Fractal Analysis
Keywords—SCADA, fractal descriptors, G-language, image ac-
quisition. The geometry was first raised in the years 300 BC by
Euclid, and it has had few changes since then. Euclidean
I. I NTRODUCTION geometry is based on characteristic of that the dimensions are
integers. For example, the lines are one-dimensional, drawings
The use of modern techniques for identification, character- two-dimensional and solid three-dimensional. In general, the
ization and modeling of variables at different scales such as Euclidean forms are smooth, continuous and homogeneous.
fractal descriptors are very important for monitoring and pro- However, around us, we can see that it is not possible to
cess control; this analytically techniques still are unknown to describe the shape of the mountains, clouds, plants, animals,
much of the scientific community in many areas of science, not in terms of Euclidean geometry. Although the world seems
being arranged at present analysis software online that allow to reflect a simple and neat perfect harmony (studied by
for a better understanding of the variables of a process and Euclidean geometry), it seems to be more of the domain of
its feedback control. Therefore the development of software the irregularity and chaos.
for analyzing the uncertainty of experimental data at different
scales like fractal analysis technique applied to images that Benoit Mandelbrot in 1975 laid the foundations for a new
can develop in a graphical user interface in language-G. geometry known as fractal geometry, which is tending more
and more applications in fields of knowledge as Economy [1],
The current demand for quality commercial products, has Geophysics [2], Electrochemistry [3], Biology [4], Engineering
led to multiple schemes to monitor and control the implemen- [5], [6] and [7], Bioreactors [8], among others. The main
tation of strategies to regulate the quality of their end prod- concept of this new geometry is the fractal dimension Df ,
ucts. For example the design of feedback control techniques that is an object property indicating that occupies much of the
based on measurements online, by acquiring data and even space that contains.
high-resolution images are used to track the most important
variables. The acquisition and analysis of data with images When explaining the terms of fractal geometry [9] Man-
represents a significant advance in monitoring the trend of delbrot coined the term fractal, which comes from the Latin
these variables. However, the analysis and interpretation of this word ”fractus” that means irregular. In addition, a fractal object
information is not always easy, mainly because of the complex has a similar appearance when viewed at different scales of
distribution behavior of the data or irregular images. magnitude; this property is known as auto-similarity and is
considered the main feature of fractals.
Therefore, in the present work it is developed language-G
program and a system of Supervisory Control Data Acquisition
978-1-5090-2458-2/16/$31.00 c 2016 IEEE
 Fractal objects can be divided into three groups depending measures the intensity of the long-term dependence of a time
on their origin, mathematical, natural and human; they all have series.
the essential characteristic of having auto-similarity and fractal
dimension. The Math’s fractals are those that are defined Interpreting the results of an analysis by Hurst exponent
by formula or mathematical expressions that can generate its shows that:
image by computer and used to simulate natural objects Fig. • If 0.5<H<1 is showing that time-series processes are
1a. Natural fractals are those which are present in nature as persistent or correlated. A persistent series is characterized by
trees, leaves, corals, fruits, clouds, etc. Fig. 1b, while Human effects of long-term memory. It has been found that persistent
fractals are those human creations that cannot be studied by series are the most common in nature.
classical statistical tools. For example, some sections of the
paintings of Jackson Pollock exhibit fractal behavior Fig. 1c. • If 0<H<0.5 corresponds to an anti-past behavior leads
to an increase in the average fluctuations in the future and
vice versa. In theory what happens today will impact the
future forever; all daily changes are correlated with all anti
correlational or persistent changes. Unlike the previous case, an
increase in fluctuations past a decrease in future fluctuations.
A persistent anti system tends constantly return to the place of
origin and have the distinction of being very irregular signals.

Fig. 1. Fractal types, a) Mathematics, b) Natural and c) Humans

• If H=1 indicates a process with a deterministic behavior
and therefore must find other methods that are not based on
fractals to analyze the phenomenon.
B. R/S Analysis
• If H=0.5 (white noise) implies a completely random and
The analysis rescaled R/S is a statistical test used to independent process, with no correlation between increases in
quantify the dynamics of a time series and determine the ex- the signal [12].
istence of fractal characteristics of a system, it was developed
• If H>1 corresponds to deterministic processes reflecting
by Hurst [10] who studied the records (data set) overflows
auto-correlations.
Nile River. Analysis R/S is a widely used method for fractal
signal characterization of physical, chemical and biological Interestingly, the fractal dimension of the sequence is
processes, which has been shown to allow identification of related to its XN Hurst exponent by the relationship Df =
autocorrelations between process parameters and the fractal 2–H2 . Analysis of R/S can be extended to estimate the
indexes, which can be used to purposes of assessment, diag- standard multi-fractal q (3).
nosis and characterization.
#1/q
Autocorrelation in time series may indicate the presence
" M
1 X
of hidden deterministic mechanisms in a number fluctuations σs,q = (yk − ys )q (3)
derived from an underlying stochastic process. M
k=1

Statistical R/S measures the range of deviations of the

It is used instead of the standard deviation of normal σs .
partial sums of a time series data on the average, rescaled
As before, it is expected the statistics of (R/s) corresponds to
by the standard deviation of the series. So, for a given N-
follow a power law of the form (6)
dimensional vector XN = (xi ) consider an M-dimensional
sample sub-vector YM = (yi ), where M = sN , and s ∈ (0, 1)
as described in [11]. Using widely tool is (1 and (2)). (R/S)s,q = aq sHq (4)

1 Where Hq is the exponent of q − th scale. The large

(R/S)s = A (1)
σNs values of q magnified large fluctuations average yi − ys of
ys . Moreover, values of q −→ 0 provide a balanced weighting
of large and small deviations of the mean. If Hq is constant
for all q ∈ R, the sequence is mono-fractal, which means
" i i
#
X X
A= max (yk − y Ns ) − min (yk − y Ns ) (2) that the corresponding sequence is produced by a single
1≤i≤Ns 1≤i≤Ns
k=1 k=1 fractal generator. Otherwise, it is said that the sequence is
multifractal and has over a fractal generator involved and
C. Hurst Exponent perhaps nonlinearity.
An important result of the analysis R/S is to get the Hurst The statistic R/S can be extended to higher dimensions
coefficient, a fractal index is directly related to the fractal along similar steps. However, as we are interested in the
dimension of the time series. Statistical R/S follows a power application of analysis R/S for images, the two-dimensional
law, (R/S)s = a ∗ NsH where a is a constant, Ns sample sizes case is detailed as follows. For NR × NC - dimensional
and H is the Hurst exponent. A log-log graph of (R/S)s as matrix XNr ,Nc = xi,j , where Nr and Nc are, respectively the
a function of the scale, S ∈ (Smin , Smax ) it gives a straight number of rows and columns, consider a sample Mr × Mc
line with slope H. Hurst exponent is an indicator to determine – dimensional matrix YMr ,Mc = yi,j where Mr = sNr
whether an event or a time series, exhibit fractal behavior and and Mc = sNc (i.e., the subsample matrix preserves the
relationship of row column), and s is the scale considered. The
statistics of R/S are given by (5-7), where (8) is an average
value and (9) is the q-norm.

1
(R/S)s,q = [B − C] (5)
σs,q

Mr X
X Mc
B = max (yk,l − ys ) (6)
0≤i≤Mr
1≥j≥Mc k=1 l=1

Mr X
X Mc
C= min (yk,l − ys ) (7)
0≤i≤Mr
1≥j≥Mc k=1 l=1

Mc
Mr X
1 X
ys = yk,l (8) Fig. 2. Stainless steel crystallizer batch pilot plant scale
Mr Mc
k=1 l=1
TABLE I. DEVICES OF BATCH CRYSTALLIZER
" Mc
Mr X
#1/q Quantity Device
1 X 1 Crystallizer 9-liter stainless steel, heating-cooling jacket 4.8 L.
σs,q = (yk,l − ys )q (9) Ced. 40 304 SS-360 CFM.
Mr Mc 1 Motor variable speed model RW 20D IKA s1. Digital, 110 VAC
k=1 l=1
(1-phase, 60 cycles) .
1 9 Agitator impeller blades of 4 ”diameter, arrow rigid plastic 10 inches.
As in the one-dimensional case, it is estimated that the 2 J-type Thermocouple; temperature of 0 760◦ C; cable length 2 m.
rescaling in a sufficiently large number of samples submatrices 1 Thermo circulator, recirculation bath model LCB-22D capacity 22
YMr ,Mc with different scales s ∈ (Smin , Smax ) and then liters of stainless steel (SUS304).
1 Shut-off valve 1/2 ”. (Download).
averaged over the entire domain Nr × Nc . Since the random 1 Shut-off valve 3/8 ”. (Recirculation bath).
route represented by the double sum (10). 1 Shut-off valve 1/4 ”. (Sampling).

Mr X
X Mc
(yk,l − ys ) (10) The equipment and devices of the crystallizer are showed
k=1 l=1
in Table I. The pilot plant was integrated to a data acquisition
system, which is used to supervise pH, temperature controlling
It is done in a two-dimensional Euclidean space; statistics and a system of image acquisition to record the distribution
R/S follow the power law behavior (11). of crystal size in seudoline through sampling to a trinocular
microscope coupled to a digital camera.

(R/S)s,q = aq s2Hq (11) The relationship of refined sugar cane and distilled water
was used for the preparation of 5 kg of saturated at 70◦ C,
using an equation of equilibrium in terms of ◦ Brix depending
A log-log plot of (R/S)s , q in function s2 gives a straight
saturation temperature obtained for a range of 70◦ C to 40◦ C
line with Hq slope. As in the one-dimensional case, Hq = 0.5
is presented in (12).
indicates lack of correlations, Hq > 0.5 indicates the presence
of correlations, and Hq < 0.5 reflects the anti-correlation
◦
processes. In fact, the great values of Hq indicate that XNr ,Nc Brix = 0.0005T 2 + 0.01566T + 63.021 (12)
two-dimensional array, is less complex than uncorrelated ran-
dom matrices. In this case, the fractal dimension and Hurst Subsequently, the sugar grams per gram of water are
exponent are related by Df = 3 − H2 . obtained (13), which they are based on the ◦ Brix value of
(12).
III. EXPERIMENTAL METHODOLOGY
◦
Brix
A. Experimental equipment C70◦C = (13)
100 −◦ Brix
The equipment used for the experimental development was
a stainless steel batch crystallizer, with heating-cooling jacket Equation (12) resulted 3.24 (gr sugar) / (gr water), there-
to pilot plant scale as shown in Fig. 2. fore, to prepare 5 kg of saturated solution at 70 ◦ C, 3821.6
gr of cane refined sugar and 1178.4 gr of distilled water are
The equipment has a capacity of 9 liters, recirculating
needed.
bath 22 liters, engine with variable speed agitator, trinocular
microscope, digital camera, pH potentiometer, pH adapter, The dissolution is heated in a batch crystallizer until at 70◦
computer, CompactRIO, Ethernet switch, flow valves, thermo- C in order to obtain a saturated solution. After that, 1.85 gr
couples type J and optical sensor with tachometer. of seed cores obtained with average size of a sieve No. 100
 TABLE II. ELECTRONIC DEVICES FOR DATA AND IMAGES
ACQUISITION SYSTEM
Quantity Device
1 Acer Aspire 5920 Laptop Intel Core 2 Duo processor with 4MB
L2 Cache and 800 MHz.
1 CompactRIO 9074. 8-slot chassis integrated 400 MHz
real-time and 2M gate FPGA.
1 Power supply. NI PS-15. 24 VCD y 5 A. 120 VCA.
National Instruments.
1 NI-9211 Module for type J thermocouples measuring four channels.
1 NI-9201 Module for measuring analog input signals ± 10V.
1 Potentiometer electrode pH measurement. Model ASP200-2-1m
/ BNC Phidgets Inc. pH range 0-14.
1 pH Potentiometer adapter. Model 1130-pH / ORP.
1 Steren Switch 5-port 10/100 Ethernet.
1 NI 1776C digital camera.
1 Cole Parmer Trinocular microscope.

mesh ASTM NBS and ASA were introduced to the saturated

solution. Natural cooling profile was applied in a range of 70◦
C to 40◦ C, the stirring speed was adjusted constant to 250
rpm for the whole processing time (120 min).
The acquisition of micrographic images for monitoring
crystal growth was carried out every 10 minutes, starting at
time zero until process is completed. For each sample from
the bulk solution were taken 5 ml. In Fig. 3, the sequence of
micrographs taken are presented, where crystals formed during
growth, show complex irregular grouping at the final step.

B. System SCADA
A SCADA system for the crystallization process of cane re-
fined sugar in a language-G was implemented using LabVIEW
by National Instruments Inc. The software acquires online,
variables temperature inside the crystallizer, jacket temperature
and pH; and offline, rpm, concentration in ◦ Brix and density
in gr/cm3 .
For the above the following data acquisition cards were
needed: Arduino card one conditioning ORP/pH-1300, Series
C module: NI-9201, NI-9211, CompactRIO Chassis 9074
National Instruments Inc. and NI-1776C digital camera and
microscope trinocular Cole-Parmer. Electronic devices used
are presented in Table II.
In Fig. 4, the tabs for the front control panel to the R/S
analysis of the images of the online process, Fig. 4a represents
the main panel, which features the presentation of credits. In
Fig. 4b the tab is presented to the selection of the image by
specifying the path or direction where it is stored for process
monitoring for fractal analysis R/S and its corresponding Hurst
exponent. Fig. 4c shows the original and binarized captured
image, and Fig. 4d the log-log graph R/S.

IV. RESULTS AND DISCUSSION

After the process of crystallization it concluded at 120
minutes, it proceeded to centrifuge the product in a basket
centrifuge at 4000 rpm for 10 minutes, and dried in a drying
chamber with forced convection at 60◦ C for 36 hours to give
1,267 gr of Formed Crystal Mass (FCM). The corresponding
D(4,3) with standard deviation S(4,3) was obtained with the
software of Vision assistant 2012 and the DTC Adq-im,
obtained from fitting a log-normal distribution to the Crystal
Size Distribution (CSD), as described [13], the results are Fig. 3. Sequence micrographs taken the bulk solution.
presented in Fig. 5.
 TABLE III. HURST EXPONENT AND FRACTAL DIMENSION

Image (minute) Hurst Exponent Df

10 1.7783 1.2217
50 1.2761 1.7239
80 1.0816 1.9184
110 1.0221 1.9779

Therefore the program created to relate the mass fractal

dimension Df and increased average diameter D(4,3), ob-
taining the following results in Fig. 6, showing the result
of running the program to acquire the Hurst exponent and
fractal dimension mass Df . From 80 minutes image proposed
program was run, measurements are summarized in Table III.
Fig. 4. Front panel. a) Credits and initial tab panel, b) Tab to select the
image and value of the Hurst exponent and Df, c) Images binarized original
process and d) log-log graph power law R/S.

Fig. 6. Hurst Exponent an Df for image at minute 80.

V. C ONCLUSIONS
The application of fractal analysis method R/S to images
taken from direct sampling of the crystallization process was
explored in this paper. The results showed that binarized
images, extracted from the images in gray scale can provide
information on the evolution of the crystallization process,
including the effects of the increase in size during the initial
stages and the possible occurrence of secondary nucleation’s
and wear mechanisms due to changes in high-profile cooling
and stirring. While the reported results of image processing,
Fig. 5. Average Diameter D(4,3) and standard deviation S(4,3). focused on the trajectory of size and shape of isolated crystals,
this approach can be used to complement the evolution of
suspended crystals with the crystallization.
The diameter increases continuously from the start of the
crystallization process in which seed crystals were added until In this paper we have focused only on the scaling H
110 minutes, wherein the diameter peaked. After this time, the exponent and the mass fractal dimension Df for monitoring
average diameter showed a slight decrease probably caused by the crystallization using the R/S analysis. However, if one
the large number of crystals formed and their interaction with recognizes that the clear images added are complex geometries,
a high and constant agitation may also be due to the formation additional fractality rates can also be explored, such as multi
of new nuclei. fractality and lacunarity. The first is indicative of nonlinear
mechanisms, and the second form inhomogeneous structures.
As described in section III, micrographs they were obtained
The results in this line should lead to a more complete charac-
every 10 minutes to monitor and control the advance of crys-
terization on-line crystallization processes. Finally, given the
tallization. Figure 3 shows 12 images taken every 10 minutes.
complexity of the crystal size and the ways in real processes
As the amount and size of crystals increase with time expected
of the above, it is understood that the online image analysis is
to cover the surface of the image. For small time crystallization
possible, but should be used only as a quick indicator of the
(Fig. 3a-d) the crystals are scattered and individually, but in
dynamics of crystallization and the resulting determinations
the final part of the process (Fig. 3i-l) the size and the amount
(fractals) should complement offline such as the CSD.
of crystals formed is largest. In this final part of the process,
overlapping crystals are presented leaving only a portion of
ACKNOWLEDGMENT
empty space in the image. These crystals form aggregates
of complex and irregularly even appear superimposed small The authors we thank to Laboratory of Processes and
crystals and large, this difference in irregular shapes and sizes Sustainable Development, Faculty of Engineering, Veracruz of
due to its nature is evident in the patterns formed, whereby the Universidad Veracruzana, for letting us use their facilities,
the use of fractal analysis tool R/S for quantifying fractality as well as to Secretary of Higher Education SES, of Academic
image is suitable. Improvement Program, the granting of scholarship UV-545.
 R EFERENCES
[1] C. Leon, and F. Vivas. Long-term dependency and the rule of root of
time to scale the volatility in the Colombian market. Drafts of economy,
No. 603. Bank of the Republic of Colombia, 2010.
[2] V. Arguello, and H. Gustavo. Design and implementation of a fractal-
based algorithm for analysis of re-scaling range well logs. Sartenejas:
Simon Bolivar University Engineering Geophysics, 2005.
[3] R. Pidaparti, B. Aghazadeh., A. Whitfield, A. rivers, and G. Mercier.
Classification of corrosion defects in NiAl bronze-through image analysis.
Corrosion Science, 52 (11), pp. 3661-3666, 2010.
[4] A. G. Salvador, L. G. Salvador, and A. S. Alcaide. Is Hurst coefficient a
good indicator of species extinction?. Mathematical Thinking, 11, 2011.
[5] G. Olivieri, A. Marzocchella, J. R. Van Ommen, and P. Salatino. Local
and global hydrodynamics in a two-phase internal loop airlift. Chemical
Engineering Science, 62(24), pp. 7068-7077, 2007.
[6] W. Zhang, and X. Li, Origin of presure fluctuations in an internal loop
airlift reactor and its application in flow regime detection. Chemical
Engineering Science 64, pp. 1009-1018, 2009.
[7] O. Velázquez, E. Bolaños, E. Rodrı́guez, and J. Álvarez, Characterization
of cane sugar crystallization using image fractal analysis. Journal of
Food Engineering 100, pp. 77-84. 2010.
[8] E Hernández, H. Puebla, H. O. Méndez and J. Álvarez. Fractality in pH
Time Series of Continuous Anaerobic Bioreactors for Tequila Vinasses
Treatment. Chemical Engineering Science 109, pp. 17–25, 2014.
[9] Mandelbrot, B. The Fractal Geometry of Nature. Freeman New York,
pp. 477-487. 1982.
[10] Hurst, H.E. Long-Term Storage Capacity of Reservoirs.Transactions of
the American Society of Civil Engineers 116, 770-880. 1951.
[11] Alvarez-Ramirez, J., Echeverria, J. C., and Rodriguez, E. Performance
of a high-dimensional R/S method for Hurst exponent estimation. Physica
A: Statistical Mechanics and its Applications, 387(26), pp. 6452-6462,
2008.
[12] O. Quintero-Delgado, and J. Ruiz-Delgado. Estimated Hurst exponent
and fractal dimension of a topographic surface through extraction
profiles. Geomatics 5, pp. 84-90, 2011.
[13] N. M. Cordova. Determining planting conditions to maximize the growth
of particles obtained by crystallization by cooling batch. Master Thesis
of Science in Chemical Engineering, Instituto Tecnológico de Orizaba,
Mexico. 2004.