Sunteți pe pagina 1din 4

Image Processing Approach for Estimating the Degree

of Surface Degradation by Corrosion


Elena Crina Cringasu, Andrei Dragomirescu, Carmen-Anca Safta
University Politehnica of Bucharest, Power Engineering Faculty
Bucharest, Romania
safta.carmenanca@gmail.com

Abstract—Hydraulic machines are permanently exposed to vanes, and any other metallic structure which comes into
degradation by erosion. Erosion corrosion can be a very direct contact with the water flow. Grecu [5] proposed the
aggressive one, being caused by the environment in which the usage of digital image processing techniques for monitoring
machine operates. This paper proposes a method for assessing the surface degradation of hydraulic turbine runners.
the degree of degradation by erosion corrosion based on image
processing using open source software. Such a method, which is In this paper, a quantitative method to analyze the
cheap and easy to implement, offers the possibility of monitoring evolution of the rust spots over time by “measuring” the area
the surface degradation in time and it can be useful in the of the eroded/corroded surface is presented. The method is
maintenance of hydraulic turbines or wind turbines. To based on an image processing technique for analyzing the rust
illustrate the proposed method, a case study for a wicket gate
blade is presented.
spots and can be used in case of big structures that cannot be
easily dismantled and transported. It is the case of hydraulic
Index Terms—erosion, corrosion, image processing, preventive turbine runners which are subjected to cavitational and/or
maintenance abrasive erosion.

I. INTRODUCTION In Section II, the proposed method is thoroughly


described. To illustrate the usage of the method, a case study
Corrosion is the result of the of the interaction between is presented in Section III. Conclusions and final remarks are
materials and their environment. Corrosion of metals appears formulated in Section IV.
mostly in electrochemical processes. There are many forms of
corrosion [1]. Erosion corrosion is “the result of a combination II. IMAGE PROCESSING METHOD
of an aggressive chemical environment and high fluid-surface Digital image processing techniques are based on a
velocities” [2]. From this point of view, abrasive erosion and specific signal processing of the input image to get an output
cavitational erosion are types of erosion corrosion often found image with certain characteristics and/or a set of parameters
in hydraulic machines-hydraulic turbines and pumps [3]. related to the image [6]. Usually, the images produced by
common cameras show a slight lack of contrast, clarity, and
Abrasive erosion, in case of hydraulic turbines, is caused sharpness. Therefore, these image characteristics could require
by the impact of water and silt particles against solid surfaces. a certain boost prior to applying any advanced processing
General corrosion and erosion corrosion often act techniques, like those described below.
simultaneously on metallic surfaces. When they act separately,
the damages are similar and the true cause is established after The method presented hereinafter for evaluating the
a thorough study. Corrosion and abrasive erosion produce evolution of corrosion is based on using the Python Imaging
changes at the structure level of the materials used in the Library (PIL) [7] and the Open Source Computer Vision
Library (OpenCV) [8]. Both libraries are free, yet very
manufacturing of important components of hydraulic
powerful and efficient. It should be pointed out from the very
machines [3]. The first effect of erosion corrosion against a
beginning that the method can be efficient only if the corroded
metallic surface is the rust stain. (or rusted) areas stand clearly out against the surrounding,
Since 1950, in the scientific literature, the corrosion and uncorroded, surface. Rust appears normally in different shades
abrasive erosion have been identified as important issues of of orange, red, and brown. A protective paint coating should
hydraulic turbines [4]. have such a color that allows rust spots to be clearly visible
when they appear. Under these circumstances, the image
For the maintenance of hydraulic machines in general and processing can be carried out in four steps, starting from an
of hydraulic turbines in particular, it is important to have a original image obtained by properly photographing the area
method of monitoring the surface of the runner blades, guide affected by corrosion.

978-1-5386-3943-6/17/$31.00 ©2017 IEEE


In the first step, the original image is desaturated, i.e. it is
converted to a grayscale image, that contains only gray tones.
The condition to go to the next step is that on the grayscale
image the corroded areas are clearly visible against the
uncorroded surface. This means that the corroded areas should
be either significantly darker or lighter than the uncorroded
surface. When this condition is not met, the color of the
protective painting must be appropriately changed at post-
processing, prior to desaturating the image.
In the second step, the histogram of the grayscale image is
calculated. The histogram is a graphical representation of the
tonal distribution in the image, which plots the number of
pixels for each tonal value. The horizontal axis represents the
tonal variation, while the vertical axis represents the number
of pixels for each tone. For a grayscale image, the left side of
the histogram corresponds to black and dark tones, the middle
side corresponds to medium gray tones, and the right side
corresponds to light and white tones. Normally, the tonal
Figure 1. Guide vane of a Bulb turbine that operated 30 years on Olt River,
range is divided into 256 bins having values that span from 0 Romania.
(pure black) to 255 (pure white) with an increment of 1.
In the third step of the method, thresholding is applied to The guide vane was monitored during an interval of six
the grayscale image to obtain a binary (or bi-level) image, that months. Photos of a small area affected by corrosion, that were
contains only pure black and pure white pixels. The threshold taken once every three months, are presented in Fig. 2a. It can
value required in this step should be carefully chosen based on be seen that the corroded area evolved quite rapidly during the
the histogram calculated in the previous step. A method for monitoring period. At time t0, a large rust spot is clearly visible
choosing the threshold value is indicated in the subsequent on the left side of the monitored area. It can also be seen that
case study. Depending on whether the corroded areas are the protective paint started to flake off on the right side of the
darker or lighter than the uncorroded surface on the grayscale area, exposing to weather conditions the metallic surface
image, rust becomes black or white, respectively, on the beneath. After three months, at time t0 + 3 months, the flaked
binary image. At this step, it becomes easy to assess the paint disappeared and a second patch of rust appeared on the
degree of corrosion by counting the number of pixels that guide vane surface. Additionally, the area of the rust spot to the
correspond to rust. left has visibly increased. After another three months, at
time t0 + 6 months, the rusted areas have increased in such a
Depending on luminosity, when applying thresholding to manner that the two rust spots have merged, denoting a quite
the grayscale image the possibility exists for some pixels on aggressive corrosion process. To better understand how
the corroded area to get the same value as the pixels on the aggressive the corrosion was, it should be emphasized that the
uncorroded surface. Therefore, the degree of corrosion materials used for manufacturing the parts of a hydraulic
assessed at the third step might not be accurate enough. To turbine are corrosion and cavitation resistant, with high
mitigate this problem, a fourth step is recommended, in which hardness and high mechanical strength [10].
the contours of the corroded areas are found together with
their moment invariants [9]. Based on the moment invariants, The entire image processing of the monitored surface was
other important information regarding the corroded surfaces carried out with a custom-made program written in Python,
can be derived: geometrical areas, centroids, and orientations. that uses specialized functions from PIL and OpenCV. The
The degree of corrosion assessed based on the areas obtained processing followed the steps of the image processing
in this step are expected to be more accurate, provided that the methodology presented in Section II.
algorithm of finding the contours keeps only the exterior First, each original image from Fig. 2a was desaturated
contours of the rust spots. with the convert function from the Image module in PIL. The
III. CASE STUDY: EVOLUTION OF CORROSION ON A GUIDE results are presented in Fig. 2b. As it can be seen, each
VANE OF A BULB TURBINE corroded area is represented by dark tones of gray, while the
protective coating is represented by medium and light tones of
The usage of the method proposed for estimating the gray.
evolution of corrosion is illustrated for a guide vane of a bulb
turbine. This guide vane is currently showcased in the Second, the histogram of each desaturated image was
courtyard of University Politehnica of Bucharest. Its photo is calculated. The histograms obtained are presented in Fig. 2c.
presented in Fig. 1. The guide vane is in a good technical Each histogram has two main regions. One is located in the
condition. To keep the guide vane from corroding, its surface range of dark tones and corresponds to the corroded surfaces.
was coated with a gray protective paint. But even under the The other is located mainly in the range of light tones and
circumstances of being no more in operation, the guide vane corresponds to the still intact portions of protective coating.
has suffered from the action of the environment (or weather The two regions are separated by thresholds located at grey
conditions) and rust spots have appeared. levels of around 100.
Figure 2. Illustration of the image processing workflow for obtaining the evolution of corroded surafces.

Third, a Savitzky-Golay filter was applied to each number of black pixels was counted with the function
histogram, using third order polynomials on successive histogram from the PIL library. The value obtained was
subsets of 25 bins. The smooth curves obtained, plotted in red subsequently divided to the total number of pixels in the
in Fig. 2c, have local minima between the two main regions of image to get the relative area of the corroded surfaces. As it
each histogram. These minima were extracted from the curves can be seen in Fig. 2d, this area increased in six months from
and were used for thresholding the grayscale images in Fig. 2b 10.94% to 37.90%. However, these estimated values cannot be
to obtain the binary images in Fig. 2d. In each binary image, considered as being very accurate, since it is clearly visible
the corroded surfaces are rendered by black pixels. The that the thresholding process leaves on the areas that
correspond to the corroded surfaces some white pixels that According to the ASTM G65 [11], resistance to abrasive
should be but are not accounted for. erosion of different metals and their degree of wear can be
determined as a ratio between the mass of the specimen after
Fourth, the binary images obtained in the third step were the experiment and the mass of the specimen before the
additionally processed with the findContours function from experiment. The method allows the experimental
OpenCV to find the contours of the corroded surfaces. The measurement of the amount of material lost through abrasive
results are presented in Fig. 2e, were the contours are rendered erosion.
in light green. The function findContours returns a contours
hierarchy having a tree structure, where each contour has a By comparison with the standardized method we will be
parent, into which it is contained, and children, that are able to find a connection between the standard method and the
contours contained inside it. The contours of interest for the proposed method of image processing analysis. This will
methodology described here are only those that mark the allow us to determine the error and uncertainty in using the
edges of the rusted spots. They can be easily found at the root proposed preventive maintenance method.
of the hierarchy, i.e. they are those contours that have no
parents. Beside the total corroded area, Fig. 2e shows the areas REFERENCES
of the largest rust spots as well. Based on the contours, it can [1] Roberge, P. R., Handbook of Corrosion Engineering, Ed. McGraw-
be estimated that the corroded area increased in six months Hill, ISBN 0-07-076516-2, 1999.
from 11.41% to 39.87%. For the case study presented here, the [2] ***, TM-58C (Revision C), Corrosion control and treatment Manual,
differences between the corroded areas estimated only by November 1, 1994. Available: http://www.corrosion.ksc.nasa.gov
thresholding and those estimated by finding the contours are [3] Karelin, V.Y., Denisov. A.I., Wu, Y.L., “Fundamentals of
Hydroabrasive Erosion Theory”. Book series on Hydraulic machinery,
in a range from roughly 0.5% to 1.1%. It can be noticed that Vol. 2, Duan C. G. and Karelin V. Y. Ed., 2002,
the thresholding consistently offers lower estimates of the
[4] Padhy M.K., Saini R.P., “A review on silt erosion in hydro turbines”,
corroded areas. Renewable and Sustainable Energy Reviews, 12, pp.1974-1987, 2008.
IV. CONCLUSIONS AND FINAL REMARKS [5] Grecu M. C., “Cavitation and cavitation erosions in Francis Turbines”,
Ph.D. dissertation, Power Eng. Faculty, Univ. Poltehnica of Bucharest,
The method proposed to evaluate the degree of 2015.
degradation of a surface caused by corrosion can give a [6] Gonzalez C. R., Woods R. E., Digital Image Processing, 3rd Ed, ISBN-
quantitative information about the process, such as the area 13: 978-0131687288, 2008.
and the rate of corrosion. Unfortunately, the volume of the lost [7] Python Imaging Library Handbook. [Online]. Available:
material cannot be determined. Hence, the wear of the http://effbot.org/imagingbook/pil-index.htm.
structure cannot be estimated. Nevertheless, for hydraulic [8] Image Processing in OpenCV. [Online]. Available:
turbines the method can be applied in situ as a preventive http://docs.opencv.org/3.0-
beta/doc/py_tutorials/py_imgproc/py_table_of_contents_imgproc/py_ta
maintenance method. ble_of_contents_imgproc.html#py-table-of-content-imgproc.
To preserve the accuracy of the method, when [9] Hu, M.K., “Visual Pattern Recognition by Moment Invariants”, IRE
Transactions on Information Theory, IT-8, pp. 179-187, 1962.
subsequently taking the photographs it is required to keep both
the same distance between the camera and the surface and the [10] Ghiban, B., Safta, C. A., Ion,M., Crângașu, C. E., Grecu, M. C.,
"Structural aspects of silt erosion resistant materials used in hydraulic
same focal distance of the lenses. This requirement can be machines manufacturing", Sustainable Solutions for Energy and
easily fulfilled by using a tape rule and a camera with prime Environment, EENVIRO 2016, 26-28 October 2016, Bucharest,
lenses (i.e. lenses with fixed focal distance). The brightness of Romania, in Energy Procedia, 112, pp. 75-82, 2017.
the subsequent photographs should be preserved as well. For [11] ASTM G65 - Standard Test Method for Measuring Abrasion Using the
this purpose, the usage of a flash is recommended. Dry Sand/Rubber Wheel Apparatus http://www.astm.org/cgi-
bin/resolver.cgi?G65

S-ar putea să vă placă și