Professional Documents
Culture Documents
Measurement
journal homepage: www.elsevier.com/locate/measurement
A R T I C LE I N FO A B S T R A C T
Keywords: This work deals with the design of a portable device for measuring and monitoring corrosion variables. The
Electrochemical noise measured variables are the electrochemical potential (EP) and electrochemical current (EC). Using these signals
Localization index the localization index (LI) and the corrosion rate (CR) can be estimated. Since electrochemical potential (EP) and
Corrosion rate electrochemical current (EC) are weak signals, an electronic circuit was designed and developed for measuring
Neural network
these two variables. The electronic circuit includes a signal amplifier, a physical filter, and an Analog to Digital
Extreme learning machine
Converter (ADC). To estimate the IL and CR the statistical method and Neural Network method were used and
compared. The device advantages are its low-cost of construction, its portability and the possibility to analyze
the corrosion variables in different materials. Experimental tests carried out using 6061-T6 aluminum shown the
effectiveness of the proposed device and methods and the versatility of the portable device.
1. Introduction some of them are more destructive than others. To identify the corro-
sion type is necessary to estimate the localization index (LI). Knowing
Metals are materials with different characteristics and properties, the LI will allow to take preventive actions and prevent any damage
such as hardness, malleability, ductility, among others. Because of their caused by an aggressive corrosion. The corrosion rate must be calcu-
properties metals are widely used in the industry. The metals most lated to determine the lifetime of metals. The CR can be calculated
commonly used are steel, zinc, aluminum, and copper, any of this using physical or electrochemical methods. In literature, important
metals can suffer corrosion. The corrosion is characterized by the metal works have been reported on the corrosion analysis using the EN
destruction, due to the metal exposure to different environmental ag- technique [3–5]. Authors in [3] mention that the EN is one of the most
gressive agents. These agents include humidity, temperature, air, vapor, promising in situ electrochemical methods in corrosion and electro-
pollutants, etc. chemical science.
Corrosion is present in most of the metal parts of machinery and Nowadays, exists different devices to estimate the corrosion rate
industrial equipment, as well as in pipes and containers of industrial however they are usually expensive for example the PT-6000 corros-
processes. The corrosion consequences are diverse, some of them are a ometer with a price of $1 572.45 USD, the Gill AC potentiostat with a
reduction of the pipe surfaces thickness, leakage, the presence of pol- price of $7 618.72 USD, the Checkmate-1 corrosometer with a price of
lution substances, or an equipment damage as a result of a leak. $4 649.65 USD, Aquamate corrater with a price of $2 925.24 USD. To
Furthermore, the expenditure of additional costs to repair or replace the deal with the devices cost different research works have been developed
corroded surface must be considered. Corrosion has been the cause of to propose lower cost devices as an alternative. The authors in [6–8]
accidents such as the flight 243 of Aloha Airlines in 1988 ([1]) or the presented the development of devices for measuring the EN to estimate
accident caused by a gas pipe explosion in Guadalajara, Mexico in 1992 corrosion variables.
([2]). The authors in [6] developed a method and apparatus for the
There are different corrosion types (uniform, general or localized) identification of corrosion in metal objects. The method consisted in
⁎
Corresponding author.
E-mail address: esjiri@cenidet.edu.mx (R.F. Escobar-Jiménez).
https://doi.org/10.1016/j.measurement.2018.03.008
Received 8 November 2017; Received in revised form 17 February 2018; Accepted 5 March 2018
Available online 06 March 2018
0263-2241/ © 2018 Elsevier Ltd. All rights reserved.
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
74
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
3. Corrosion estimation
Next, the electrochemical (Section 3.1 and the ANN (Section 3.2
methods are presented. The CRE-1 uses two ANNs to estimate the
corrosion type and corrosion rate. The electrochemical method was
used to validate the ANNs.
Fig. 3. Block diagram of the amplification circuit. When a metal is in contact with an electrolyte, the corrosion phe-
nomena generates changes in the potential and the current caused by
the reduction reactions. So, electrochemical noise technique uses these
EP and EC changes to describe the corrosion phenomena.
To use the electrochemical noise technique the EP and EC mea-
surements have to be stored and recorded in data sets. Then, the
75
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
resistance (Rn) is calculated based on each data set or record. The Rn is k: constant in different units (Table 1).
the relation between the standard deviation of potential noise (σV ) and ρ : metal or alloy density.
the standard deviation of current noise (σI ), Eq. (1). icorr : current corrosion density.
σv Weq: metal or alloy equivalent weight (dimensionless).
Rn = .
σI (1)
The corrosion type can be identified using the corrosion
Another important parameter is the polarization resistance (Rp), Eq. Localization Index (LI) which is calculated with σI and the root mean
(2) square of the current noise (Irms ) [24], Eq. (6). According to Table 2 the
B corrosion type can be classified in localized, mixed and passive or
Rp = ,
icorr (2) uniform. [25].
⎝ ⎝ ρ ⎠⎠ (5) ANN was trained to predict the CR. Both ANNs are feedforward type
with one hidden layer, inputs and outputs of the hidden layers are
where
weighted. The ANNs inputs are the standard deviation of the potential
Table 1
Table 2
Values of k for the CR calculus.
LI value ranges.
Units Values of k
Corrosion type LI values
CR i corr ρ k
Max. Min.
mpy μ A/cm2 g/cm3 mpy g/μ A cm 0.1288
Localized corrosion 1.0 0.1
mm/year A/m2 kg/m3 mm kg/A m year 327.20 Mixed corrosion 0.1 0.01
mm/year μ A/cm2 g/cm3 mm g/μ A cm year 3.27 × 10−3 Uniform corrosion 0.01 0.001
76
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
and the standard deviation of the current. The hidden layer had 15 an experimental current noise dataset. Each dataset contained 3000
neurons. The ANNs architecture is described in Fig. 5. data. The ANNs training was carried out for the CRE-1 as well as the
The Extreme Learning Machine (ELM) algorithm was used for ANNs commercial devices. The datasets were begins sampled when the cor-
training [34]. Defining the training data set as rosion measurements reached the steady state (after two hours of the
D = (x i,ti )|x i ∈ Rn,ti ∈ Rm,i = 1,…,N , one activation function g (x ) and a electrodes immersion in a 3.5% NaCl solution). It is important to
∼
hidden layer of N neurons. Where N is the number of data for neural mention that the experimental tests between the CRE-1 and the com-
training, x i are the inputs and ti the desired outputs, the training is mercial devices didn’t carry out simultaneously. So, small variations in
described in the next steps: the variables predictions may occur.
Step 1: Random assignment of the input weighs wi and the bias 4.2. Electrochemical measurements for training and test the ANNs
∼
bi,i = 1,…,N .
Step 2: Calculation of the matrix H of the output of the hidden layer The tests were carried out by measuring the potential and current
where from the 6061-T6 aluminum electrodes immersed in a 3.5% NaCl so-
lution. The measures were recorded using the sampling time of one
H (w1,…,w N∼,b1,…,b ∼
N ,x1,…,xN )
second per sample having a total record of 27,270 data (7 h 35 min).
⎡ g (w1·x1 + b1) ⋯ g (w N ·x1 + b N ) ⎤
∼ ∼
The experimental measures of the commercial devices and the CRE-1,
=⎢ ⋮ ⋯ ⋮ ⎥ . are shown in the following figures.
⎢ g (w ·x + b ) ⋯ g (w ∼·x + b ∼) ⎥ ∼
⎣ 1 N 1 N N N ⎦N × N Fig. 6 shows the potential noise measurement comparison between
the commercial device and the CRE-1 and Fig. 7 shows the current noise
measurement comparison between the commercial device and the CRE-
Step 3: Calculation of the output weights β
1.
β = H+T, (7) The level of the signals generated by the corrosion processes is
higher than the level of the thermal (Johnson) noise processes signals
[35]. In this sense, it is possible to ensure that the monitored signals
where H+ is the generalized inverse of Moore–Penrose commonly correspond to a corrosion process.
known like pseudo-inverse and T contains the desired outputs
T = [t1,…,tN ]T . 4.3. Training of the ANNs
The input weighs (w), the bias (b) and the output weighs (β ) are
obtained with the neural network training, then the neural network For each artificial neural network training, 27,270 data were di-
output y is calculated with the following equation vided into 2727 records of 10 samples each one. These records were
used to calculate the standard deviation of current (σI ), the standard
y = Hβ. (8)
deviation of voltage (σV ), and be able to estimate by means the statis-
tical method the Corrosion Rate (CR), and the Localization Index (LI).
4. Results Once obtained the CR and the LI the training inputs were CR,σI and σV
for one ANN. And LI ,σI and σV for the other. The ANN design has 15
4.1. Testing the device and the ANNs neurons in the hidden layer, and the activation function was the hy-
perbolic tangent function.
The measures obtained from the CRE-1 were used to compare its The training errors (Root Mean Square Error) with 100,000 epochs,
performance versus two commercial devices. One commercial multi- were 0.1288 for the CR approximation (Fig. 8) and 0.0042 for the LI
meter(Keysight™ model 34461A) was used to measure the current noise approximation (Fig. 9). The obtained bias, weighs and thresholds were
and the second one (Agilent™ 34410A) was used to measure the po- saved to test the networks.
tential noise. A LabVIEW™ interface was used to record the data of the As a result of the device development and the implementation of the
two multimeters and save the data in a text file. Two artificial neural neural networks, it is possible to estimate every ten seconds the alu-
networks were implemented to develop the performance comparison minum CR and the corrosion LI. Fig. 10 shows the corrosion rate cal-
between the CRE-1 device and the commercial devices. As mentioned culation using the statistical method (blue1 line) and the corrosion rate
previously, the LI and the CR are estimated by each one of them. Both estimated by the ANN (red line) of the 3000 data measured, the error
ANNs were trained using an experimental potential noise dataset and between both methods was 0.5627.
77
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
Fig. 11 shows the comparison between the CR obtained with the 3000 data measured, the error between both methods was 0.0017.
multimeter data using the statistical method (blue line) and the CR Fig. 13 shows the comparison between the LI obtained with the
obtained with the developed device data using the ANN (red line). multimeter data using the statistical method (blue line) and the LI ob-
Fig. 12 shows the LI estimation using the statistical method (blue tained with the device data using the ANN (red line).
line) and the localization index estimated by the ANN (red line) of the The LI reported in almost all cases was mixed-type corrosion with a
slight tendency toward passivation.
1
For interpretation of color in Figs. 10–13, the reader is referred to the web version of
this article.
78
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
Fig. 10. Testing the ANN for CR estimation with data of aluminum inmersed in 3.5% NaCl.
Fig. 11. Comparison between the CR obtained by the multimeters and the device.
79
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
Fig. 12. Testing the ANN for LI estimation with data of aluminum inmersed in 3.5% NaCl.
Fig. 13. Comparison between the LI obtained by the multimeter and the device.
training, in order to estimate the metal’s CR and LI. thesis. The authors would like to thank PRODEP, Tecnológico Nacional
Using the multimeter measures, the CR calculated average value de México and CENIDET for the support to develop this work. José
was 0.4207 mm/year. While using the CRE-1 measures, the CR esti- Francisco Gómez Aguilar acknowledges the support provided by
mated average value was 0.3703 mm/year. Considering a thickness of CONACyT: cátedras CONACyT para jovenes investigadores 2014.
five millimeters the metal’s useful lifetime is 11.8 years according to the
estimation using the multimeters data and 13.5 years according to the References
neural network estimation. Also, the methods comparison (ANNs and
statistical) showed slight differences between the estimation of the CR. [1] B.V. Salas, M.S. Wiener, Corrosión y preservación de la infraestructura industrial,
OmniaScience, 2013. http://dx.doi.org/10.3926/oms.36.
[2] P. Roberge, Handbook of Corrosion Engineering, McGraw-Hill, 1999.
6. Conclusions [3] D.H. Xia, S.Z. Song, Y. Behnamian, Detection of corrosion degradation using elec-
trochemical noise (en): review of signal processing methods for identifying corro-
sion forms, Sci. Technol. 51 (7) (2016) 527–544.
A device for measuring metal corrosion rates (6061-T6 aluminum) [4] C. Ma, S. Song, Z. Gao, J. Wang, W. Hu, Y. Behnamian, D.H. Xia, Electrochemical
was developed by recording and analyzing the EN data generated when noise monitoring of the atmospheric corrosion of steels: identifying corrosion form
using wavelet analysis, Sci. Technol. 52 (6) (2017) 432–440.
a metal comes into contact with an electrolyte. [5] D.H. Xia, C. Ma, S. Song, L. Ma, J. Wang, Z. Gao, C. Zhong, W. Hu, Assessing at-
The developed device is economical compared to commercial po- mospheric corrosion of metals by a novel electrochemical sensor combining with a
tentiostats or corrosometers, which in addition to having a high cost thin insulating net using electrochemical noise technique, Sens. Actuators B: Chem.
252 (2017) 353–358.
they require a computer or there are difficult to handle or move to the
[6] S. Reid, D. Eden, Assessment of corrosion, uS Patent 6,264,824 B1, Jul. 24 2001.
testing area. In addition to providing online information, the CRE-1 also URL <https://www.google.ch/patents/US6264824>.
saves the records made since they could be required for further analysis [7] J.A.R. Enciso, A.R. Salinas, Sistema que supervisa la corrosión en tiempo real por
or collect them as a history. análisis del ruido electroquímico, Sci. Tech. 1 (36) (2007) 937–940.
[8] J.M. Laria, Electronic system for monitoring at distance the electrochemical cor-
rosion processes., MX 2009013718 A, 2011. URL <https://worldwide.espacenet.
com/publicationDetails/biblio?CC=MX&NR=2009013718A&KC=A&FT=D&
Acknowledgments ND=4&date=20110621&DB=lp.espacenet.com&locale=es_LP>.
[9] ASTM, Standard Guide for Electrochemical Noise Measurement: ASTM G199 – 09,
José Hugo Arellano Pérez and Oscar Jaime Ramos Negrón would American Society for Testing and Materials, 2014.
[10] U. Bertocci, C. Gabrielli, F. Huet, M. Keddam, Noise resistance applied to corrosion
like to thank to CONACYT (Consejo Nacional de Ciencia y Tecnología measurements i. theoretical analysis, J. Electrochem. Soc. 144 (31-37) (1994)
de México) for the support given during the development of their Ph.D’s
80
J.H. Arellano-Pérez et al. Measurement 122 (2018) 73–81
155–157. [23] ASTM, G102-89(2015), Standard practice for calculation of corrosion rates and
[11] J.R. Kearns, R. Jeffery, J.R. Scully, P.R. Roberge, D.L. Reichert, L. David, J.L. related information from electrochemical measurements, Tech. Rep., ASTM
Dawson, Electrochemical Noise Measurement for Corrosion Applications, vol. 1277, International, West Conshohocken, PA, 2015. http://dx.doi.org/10.1520/G0102-
ASTM International, 1996. 89R15E01.
[12] G.P. Bierwagen, Calculation of noise resistance from simultaneous electrochemical [24] J.A. Hernández, R.J. Romero, D.J. Romero, R.F. Escobar, J. Siqueiros, A neural
voltage and current noise data, J. Electrochem. Soc. 141 (11) (1994) 155–157. network approach and thermodynamic model of waste energy recovery in a heat
[13] E.D. Kenny, R.S. Paredes, L.A. de Lacerda, Y.C. Sica, G.P. de Souza, J. Lázaris, transformer in a water purification process, Desalination 243 (1) (2009) 273–285.
Artificial Neural Network Corrosion Modeling for Metals in an Equatorial Climate, [25] R.F. Escobar, J. Uruchurtu, D. Juárez, J. Siqueiros, J.A. Hernández, On-line indirect
Corros. Sci. 51 (10) (2009) 2266–2278. measures estimation for the performance of an absorption heat transformer in-
[14] M. Kamrunnahar, M.M. Urquidi, Prediction of corrosion behavior using neural tegrated to a water purification process, Measurement 42 (3) (2009) 464–473.
network as a data mining tool, Corros. Sci. 52 (3) (2010) 669–677. [26] F. Mansfeld, Z. Sun, Technical note: Localization index obtained from electro-
[15] Z. Jančíková, O. Zimnỳ, P. Koštial, Prediction of metal corrosion by neural net- chemical noise analysis, Corrosion 55 (10) (1999) 915–918.
works, Metalurgija 52 (3) (2013) 379–381. [27] R.A. Cottis, Interpretation of electrochemical noise data, Corrosion 57 (3) (2001)
[16] R. Vera, S. Ossandón, On the prediction of atmospheric corrosion of metals and 265–285, http://dx.doi.org/10.5006/1.3290350.
alloys in chile using artificial neural networks, Int. J. Electrochem. Sci 9 (2014) [28] R.A. Cottis, M.A.A. Al-Awadhi, H. Al-Mazeedi, S. Turgoose, Measures for the de-
7131–7151. tection of localized corrosion with electrochemical noise, Electrochim. Acta 46 (24)
[17] S. Mousavifard, M. Attar, A. Ghanbari, M. Dadgar, Application of artificial neural (2001) 3665–3674.
network and adaptive neuro-fuzzy inference system to investigate corrosion rate of [29] A. Nagiub, F. Mansfeld, Evaluation of microbiologically influenced corrosion in-
zirconium-based nano-ceramic layer on galvanized steel in 3.5% NaCl solution, J. hibition (MICI) with EIS and ENA, Electrochim. Acta 47 (13) (2002) 2319–2333.
Alloy. Compd. 639 (2015) 315–324. [30] E.M. Esparza-Zúñiga, M.A. Veloz-Rodríguez, J.U. Chavarín, V.E.R. Cruz, Corrosion
[18] L.A. Garita, Evaluation of electrochemical monitors for atmospheric corrosion stu- of carbon steel in sour water from the oil industry: the effect of temperature, Int. J.
dies of carbon steel, Tecnología en marcha 27 (2) (2014) 3–13. Electrochem. Sci 6 (2011) 5016–5030.
[19] ASTM, G103(2011), Standard practice for preparing, cleaning, and evaluating [31] G. Suresh, U. Kamachi-Mudali, Electrochemical noise analysis of pitting corrosion
corrosion test specimens, Tech. rep., ASTM International, West Conshohocken, PA, of type 304l stainless steel, Corrosion 70 (3) (2014) 283–293, http://dx.doi.org/10.
2015. http://dx.doi.org/10.1520/G0001-03R11. 5006/1003.
[20] A. Torres, J. Uruchurtu, J. González-Rodríguez, S. Serna, Correlation between [32] E. Garcia-Ochoa, F. Corvo, Using recurrence plot to study the dynamics of re-
electrochemical noise resistance and noise impedance for mild and stainless steel as inforcement steel corrosion, Protect. Met. Phys. Chem. Surf. 51 (4) (2015) 716–724.
a function of ph, Corrosion 63 (9) (2007) 866–871. [33] R.A. Rodríguez-Diaz, J. Uruchurtu-Chavarin, A.M. Cotero-Villegas, S. Valdez,
[21] S. Mabbutt, D.J. Mills, C.P. Woodcock, Developments of the electrochemical noise J.A. Juárez-Islas, Corrosion behavior of AlMgSi alloy in aqueous saline solution, Int.
method (enm) for more practical assessment of anti-corrosion coatings, Prog. Org. J. Electrochem. Sci. 10 (2015) 1792–1808.
Coat. 59 (3) (2007) 192–196. [34] G.B. Huang, Q.Y. Zhu, C.K. Siew, Extreme learning machine: theory and applica-
[22] J.M. Sanchez-Amaya, R.M. Osuna, M. Bethencourt, F.J. Botana, Monitoring the tions, Neurocomputing 70 (1) (2006) 489–501.
degradation of a high solids epoxy coating by means of EIS and EN, Prog. Org. Coat. [35] R. Cottis, S. Turgoose, Electrochemical Impedance and Noise, vol. 7, National Assn
60 (3) (2007) 248–254. of Corrosion Engineers, 1999.
81