See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274333504

Co2 corrosion of carbon steel - from mechanistic to empirical modelling

Article in CORROSION REVIEWS · June 1997

DOI: 10.1515/CORRREV.1997.15.1-2.211

CITATIONS READS

33 2,108

3 authors, including:

Miran Vrhovac
ISQ
10 PUBLICATIONS 156 CITATIONS

SEE PROFILE

All content following this page was uploaded by Miran Vrhovac on 13 January 2019.

The user has requested enhancement of the downloaded file.

 COZCORROSIONOF CARsON STEEL - FROM
MECHAMSTIC TO EMPIRICAL MODELLING

SrdjanNe5ic
Departmentof MechanicalEngineering
University of QueensIand
Brisbane,Qld 4072,Australia

JohnPostlethwaite
Departmentof Chemical Engineering
Univer si ty of Saskatchewan
Saskatoon,CanadaS7N0W0

Miran Vrhovac
Research,Developmentand TrainingDepartment
ISQ - Instituto de Soldadurae Qualidade
PortugueseWelding Institute
8N249 - Km 3, Cabanas- Laido, TagusPark
Apartado II9, 2781Oeiras- Codex,Portugal

ABSTRACT

Many different modelsfor COz corrosion are usedby engineersin the oil
and gas industry. Someare describedin the open literature and only those
are coveredby this review.In orderto classiSthe COzcorrosionmodels,we
have arbitrarily decidedto group them into three categoriesbasedon how
firrnly they are gfounded in theory: mechanistic, semi-empirical and
empirical models.A few most representative examplesare discussedfrom
each gtoup. The performanceof four different models covering all three
groupsis testedby comparingthe predictionswith a large experimentalCO2
corrosiondatabase.

KEYWORDS

review,COzcolTosion,model,experiments,carbonsteel,neuralnetwork

2rl
Vol. 15,No.s.l-2, 1997 CO 2 Corrosion of Carbon Steel -
From Afechani.smto Empirical Modelling

INTRODUCTION

Mechanistic modelling of phenomena such as CO2 corrosion is done with

the immediate objective of gaining a greater understanding of the role of the
various parameters that affect the corrosion rate. The models should be
based on fundamental equations and involve as ftw empirical relationships
as possible. The ultimate objective is to develop predictive models that can
be used for design purposes, and the more fundamentally based they are, the
greater the generality and utility of the models. When unacceptable
corrosion rates are predicted, fundamentally based models should point to
the factors controlling the corrosion rate and suggestmethods to reduce it to
acceptable levels. However, whenever the complexity of the problems
exceeds our understanding, semi-empirical and even purely empirical
models have to be used.

UNDERSTAIIDTNGOF CO2CORROSTON

Aqueous COz corrosion of carbon steel is an electrochemicalprocess

involving the anodic dissolution of iron and the cathodic evolution of
hydrogen.The overallreactionis:

Fe + CO, + HrO -> FeCO, + H, (1)

The electrochemical reactions are often accompaniedby the formation of

films of FeCOg (and/or FerOa) which can be protective or non-protective
depending on the conditions under which they are formed.

Anodic Reaction

The anodic dissolution of iron in acid solutions:

Fe -+ Fe** + 2e- (2)

has beenstudiedextensivelywith severalmulti-stepmechanismssuggested

to explainthe variousexperimentalresults.The reactionis pH dependentin
acidic solutionswith a reactionorderwith respectto OH- betweenI artd2,
decreasingtoward I and 0 at pH > 4-5.MeasuredTafel slopesare typically

2t2
S. Neiib, J. Po.stlethwaite snd M. Vrhovac Corro.sion Reviev,s

30-40 mV. This subject,which is controversiatwith respectto the mech-

anism, is reviewedin detailby Drazic /I/ and Lorenz and Heusler/2/. The
anodicdissolutionin aqueousCO2solutionshas,until recently/3/, notbeen
the subjectof detailedmechanisticstudies.
The mechanismfound by Bockris, Drazic and Despic /4/ @DD) in
strongacids:

Fe+HrOeFeOH +Hn +e

FeOH rds (3)

, Fe OH* + e-

FeOH*+H* eFe** +HrO

which predictsa Tafel slopeof 2RT|3F: 40 mv at zs"c and a reaction

orderwith respectto OH-equalto one,hasfrequenflybeenassumedto apply
in COz solutions.It was overlookedthat the experimentalresultspresented
by Bockris et al. /4/ show that that the pH dependenrydecreasesrapidly
abovepH 4 and that is exacflythe areaof COzcorrosionwhereit was sub-
sequentlyused/5-8i.
For example, Schmitt and Rothman 16l concrudedthat the BDD
mechanismappliedto the dissolutionof iron in deaeratedCOz containing
0.5 M sodiumsulfatesolutionson the basisof a measuredTafel slopeof 40
mv which was observedover the temperaturerange 25-7s"C. In fact, it
might have been expectedfrom theory that the Tafel slope given for the
BDD mechanismas2RTRFwouldhaverisento 4i mV at 75"C.
Hurlen et al. /9/ concludedthat the activedissolutionof iron appearsto
be little affectedby CO2 in aqueoussalt solutions,but that COz stimulates
iron dissolutionin the intermediateprepassivestate.They reporteda Tafel
slopeof ca. 30 mV at25oCand a first orderOH-dependency.
Gray et al. /7,8/ studied the effect of pH and temperatureon the
mechanismof carbon steel corrosionin aqueousCOz solutionsand their
"modelled" Tafel constantsfor anodic iron dissolutionincreasedfrom 50-
120 mv in the pH range 2-10 whereasthe exchangecurrent density (i,;
valueswererelativelyindependentof pH. Theyalso reportedthat the anodic
polarisation curve in the region near the corrosion potential was
significantlyincreased by the presenceof dissolvedCOzat pH 6-10. It was
alsofound that the Tafel constantsfor the anodicdissolutionof iron derived
from the studywere proportional to the absolutetemperatureT, andthat the

2t3
VoL 15,Nos.I-2, 1997 CO2 Corrosion of Carbon Steel -
From Mechanism to Empirical Modelling

l, valuesin the presenceand absenceof CO2wereproportionalto llT up to

60oC,abovewhich temperaturethe dependenrywassubstantiallydecreased.
The de Waard and Milliarns l5l model for aqueousCOz corrosion
assumedthat the BDD mechanismapplied. They obtained similar Tafel
slopes30-40 mV in the absenceand presenceof COzand no changein the
positionof the Tafel slope.
Davies and Burstein /10/ found that the formation of the complex
, ,,_
fe(CO)r- during the corrosionof iron in bicarbonatesolutiongovernsthe
anodicbehaviourof iron in the activeregion.
Videm /11/ found that the anodicdissolutionin NaCl solutionat Pcoz=
I in the active region had a reactiondependencywith respectto OIf equalto
I at pH < 4.2. At higher pH valuesthe dissolutionwas independentof pH.
At pH > 4.3 the presenceof COzincreasedthe rate of anodicdissolutionand
this effect was proposedto relate to the concentrationof the bicarbonate
ions.
In a recent study, Nesic et al. l3l have reported that the anodic
dissolutionof iron is affectedby the presenceof COzand zummarised their
experimentalfindings asfollows:
-
L

t"- kloH-T'Q*,)'ron (4)

where for:

pH<4 :+ Qt:Z ; bo:0.03Il

4< pH<5 =) 2<a1<0 ; 0 . 0 3< b o < 0 . 1 2V
pH>5 = 4r=0 ; bo:0.I2V

Pco2<I0-2bar + a z : 0
!0''< pcor< Ibar :+ a z = l
Pcor> | bar oz= 0

In order to explain the different kinetics of iron dissolutionin COz solu-

tions comparedto strong acids, a carbonic speciesacting as a chemical
ligand and catalysingthe dissolutionof iron wasproposed.SinceHzCO:and

*
The Tafel slopebois shownfor roomtemperatue(22"C).

2r4
S. Neitt, J. Postlethwaite and M. Vrhovac Corrosion Reviews

dissolvedCOz are the only carbonicspecieswhoseconcentrationdoesnot

dependon pH, and since the concentrationof COz is far higher, it was
assumedthat the ligand Fet: Fe-COzis formed as an adsorbedspeciesat
the electrodesurface.Severaldetailedmulti-stepmodelswere presentedto
explain the various findings. The following mechanismwas proposedto
explain the experimentalresultsat pH > 5:

Fe+COre Fe,

Fer+HrOeFerOHou+H*+e (s)
rds
FeroH"u , FeroHj, + e
FerOH)u + HrO e Fer(OH)r"u + H*
re r(Oa)ro,t € Fer(OH),*,

r e , , ( o n ) r , o t + 2 H *e F e 2 *+ C o r + 2 H r o

Cathodic reactions

The presenceof COz increasesthe rate of corrosion of iron in aqueous

solutionsby increasingthe rate of the hydrogenevolution reaction.In strong
acids, which are completely dissociated,the rate of hydrogen evolution
cannot exceedthe rate at which tf ions can be transportedto the surface
from the bulk solution. In solutions with a pH > 4 this mass transfer
controlled limiting current is small and the presenceof HzCOsenables
hydrogenevolution at a much higher rate. Thus at a given pH the presence
of COz leads to a much higher corrosionrate than would be found in a
solution of a strongacid. It is not known whetherH2CO3is reduceddirectly
(as assumedby many workersin the field /5,7,121),or the final stepin the
reactionfollows the dissociationof the HzCOgll3l.lvlarty haveassumedthat
the two reactionsare independentand the net cathodiccurrent is the sum of
the currentsfor the two reactions17/:

2II* + 2e- -+ H, (6)

2H2CO3+2e-
+ Hr+2HCO, (7)

The limiting currentsfor hydrogenevolutionin aqueousCOz solutions

do not respondto changesin the solutionflow rate in a mannerthat would

215
Vol. 15,Nos.l-2, 1997 CO 2 Corrosion of Carbon Steel -
From Alechanism to Empirical Modelling

be expectedfor a wholly masstransfer controlled reactio& and the higher

the pH the less the responseto flow rate changes.This has been explained
on the basisthat the replenishmentof the H2CO3is controlledby the slow
hydration of dissolvedCO2. Schmitt and Rothman 16l were the first to
observethe chemicalreactioncomponentof the limiting current, suggesting
a heterogeneous reactioncontrol relating to surfaceadsorption.Since the
effect is the same on platinum and iron, homogeneouscontrol is now
accepted /12/.
Hydrogen evolution from the direct reduction of water can become
importantll4,I5lat pH > 5.

zHzO+ 2e- + H, + 2off (8)

Gray et al. /8/ have zuggestedthat in COz solutionsat higher pH the

direct reductionof the bicarbonateion dominates:

L H C O ;+ 2 e - + H r + 2 C O ; (e)

THE MODELS

Many different modelsfor COz corrosionare usedby engineersin the oil

and gas industry. Someare describedin the openliteratureand only those
are coveredby this review.Thereare quite a few proprietaryor so called in-
housemodelswhich are also in use- most are variationsof thosepublicly
available,while othersare empirical modelsbasedon the companies'own
field experience.In orderto classiS the COzcorrosionmodelsfound in the
open literature we have arbitrarily decided to group them into three
categoriesbasedon how firrnly theyare groundedin theory:
o Mechanistic models- thesemodelsdescribingthe mechanismof the
underlyingreactionshavea strongtheoreticalbackgtound.All or mostof
the constantsappearingin this typeof modelhavea clearphysicalmean-
ing. Many of the numericalvaluescan be found in the literature, such as
diffirsion coefficients,and Tafel slopes;however,it is not excludedthat
somedata have still to be obtainedby comparing the behaviour of the
model with availableexperiments.When calibratedon a reliable (and
not necessarilylarge)experimentaldatabase this type of modelshould,in
principle, enableaccurateand physicallyrealistic interpolation,as well
asextrapolationpredictions.

2t6
S. Ne.3t6,J. Postlelhwaite and M. Vrhovac Corrosiott Reviews

Semi-empiricalmodels- thesemodelsare only partly basedon firm

theoreticalhypotheses. They are for practicalpurposesoften extendedto
areaswhere insuffrcienttheoreticalknowledgeis availableso that the
additionalphenomenaare describedwith empiricalfunctions.Obviously,
some of the constantsappearingin these models will have a clear
physicalrneaningwhile otherswill bejust best-fitparameters.Calibrated
with a sufficientlylarge and reliableexperimentaldatabasezuchmodels
will enablegood interpolationpredictions.However,extrapolationcan
leadto unreliableand sometimesphysicallyunrealisticresults.
Empirical models- thesemodelshavevery little or no theoreticalback-
ground and most or all of the constantshave little physicalmeaning-
they are just best-fit parametersto the availableexperimentalresults.
However,when basedon a large and reliable experimentaldatabase,
thesemodelscan have excellentinterpolationcharacteristicsas shown
below. However, any extrapolationcan lead to incorrect predictions.
Someof the recentadvancesin this field zuchas the modellingapproach
of "evolved problems descriptors" /16/ seem to have significant
advantages over the classicallinear multivariablecurvefitting.

Mechanisticmodels

As described in the previous section, COz corrosion is a complex

phenomenonwhereelectrochemical, transportand chemicalprocesses occur
simultaneously.Thus the processes to be modelledare the electrochemical
reactionsat the metal surfaceand the transportprocessof all the speciesin
the qystemsuch as If, CO2 , HzCOgand Fe*, as well as the chemical
reactionsoccurring betweenthem. There is no single model in the open
literature which graspsthe entire complexity and interplay of the different
processes, however,thereare a few studieswhich in a physicallycorrectway
treat parts of the overall picture.
Since COz corrosionis an electrochemicalphenomenonit is not sur-
prising that a numberof researchers attemptingto constructa mechanistic
model haveelectedto start with the electrochemical processeshappeningat
the metal surface.Although there is no disputing the fact that the surface
electrochemistry is an importantcomponentof the overallprocess,it can be
arguedthat the rate limiting stepis oftenelsewherein the process(e.g.mass
transport through a surfacefilm) and that detailed and often very complex
modellingof the electrochemical processesis of little benefit.

2t7
Vol. 15.Nos.l-2, 1997 CO2 Corrosion of Carbon Steel -
Front Mechanism to Enrpirical Modelling

One of the first significantand mostwidely usedmechanisticmodelsis

the oneproposedby de WaardandMilliams l5,I7l. Basedon the assumption
of direct reductionof HzCOr(as presentedpreviouslyin the text) and their
own glass cell experiments,the authors presenteda correlation for the
corrosionratewhich is a functionof the Pco" andtemperature.

- 7.96-ry-
logV,o, 5.55 pco,
x l0-3t + 0.67lo1 (10)
t +273

The 0.67 factor for the Pcorwas obtainedby assumingthat the pH is a

function only of pcor; that is: all the Ff ions in solution come from
dissociationof carbonicacid. While this assumptionholds for pure water-
COz solutions,the presenceof other speciessuchas the corrosionproducts
and NaHCO:, which are nonnally presentwhen steel corrodesin brines,
meansthat this is no longer true. The temperaturefunction in (10) was
obtainedby assumingan Arrhenius-tlpe dependence for a chargetransfer
controlledprocess.In their derivation,the authorsassumedthat the anodic
processof iron dissolutionproceedsvia a pH-dependentBDD mechanism
l4l.Itis knowntodaythat this assumption doesnot hold l3,Il,l5l sothat the
whole validity of the proposed model can be questionedon this basis. In
addition, the authors later realisedthat the original experimentaldatabase
for their model was somewhatunreliable so that new model constantswere
determined/18,19/from experimentsof Dugstadet al. /201.T\e original
model of de Waard and Milliarns lITl was revised on severaloccasions
/I8,l9l to extend its validity into areaswhere protectivefilms form, to
accountfor velocity, water wetting, glycol, etc. However,this was doneby
using correction factors (multipliers to the original equation 10) which
ignoredthe assumptions underwhich this equationwasderived.This neces-
sarily took their modelling strategy into the area of semi-empirical
modelling which is discussedin the following section. Despite all the
theoreticalshortcomings,the modelof de Waardand Mlliams stoodas one
of the important referencepoints for any CO2corrosionresearchover the
pasttwo decadesand in its mostrecentlyrevisedform lI9l is still usedas an
informal industrial standard.
Bonis and Crolet /I3/ briefly mentionedan electrochemicalmodel but
have not presented it in detail, downplaying themselvesits overall
importancefor predictingCOzcorrosionrates.
Gray et al. 17,8/ alsopresentedan electrochemicalmodel as a part of

218
S. Neit6, J. Po.stlethwaiteand M. Vrhot,nc Corrosiott Revieu,^s

their experimentalstudyof COzmechanisms.A numberof mechanismsfor

the electrochemicalreactionsoccurring at the metal surfacewere adopted
from literature and composedinto an overall model. On the cathodicside
they accountedfor If reduction,ffid the direct reductionof HzCOg,HCO:-
and HzO. For the anodic dissolution of iron they adopted the BDD
mechanism.The modelconstantsweredeterminedfrom their own glasscell
experimentsand flow cell experiments(for higher temperatures).In its
scopeand approachthis was a "breakthrough"study in the field of CO2
corrosion modelling, despitethe very ambitious attempt to cover a very
broad range of parameters(pH 2-ll, t-25-125"C)which sometimesled to
oversimph&ing assumptions.Despite its qualities, zurprisingly few re-
searchersin the field havecontinuedalong this well establishedtheoretical
path, selectingrathera semi-empiricalmodellingstrategy.
One of the rarefollow-up studiesis the oneby Nesicet al. /15/ wherethe
authors also presentedan electrochemicalmodel of COz corrosion.They
consideredthe following cathodicreactions:Ff reduction,and the direct
reduction of HzCO: and HzO. The BDD mechanism,which predicts a
reactionorderof I with respectto OH-for the anodicdissolutionof iron, was
not usedas it was clear from the authors'experimentsthat it doesnot hold
under CO2 corrosionconditionsat pH>4. This was noticed previouslyby
Gray et al. l8l and Videm /11/; however,nothing was done to clari$ the
deficiency.Physicalconstantsappearingin the modelwhich are functionsof
the pH, temperatureand other environmentalvariableswere determined
from the literature or, when missing,found from mrmerousrotating cylinder
glasscell e4periments.Most of the experimentalobservationsand physical
constantsfound in this studyconfirmedthe findings of Grayet al. 17,8/.The
predictionsmadewith the model were successfirllycomparedwith indepen-
dent pipe-flow glass-loopexperimentsdemonstratingthe ability of mechan-
istic modelsfor extrapolation.The authorsalso comparedtheir predictions
with two other models:the semi-empiricalmodelsof de Waard et al. /I8/
and Dugsladet al. l20l(presentedbelow)obtaininggoodagreementwith the
former and more significantdiscrepancies with the latter. Thesediscrepan-
cieswereascribedto the dtfferentdatabases which wereusedfor determina-
tion of individual modelconstantsfor the two models.
Both in the modelsof Gray et al. /8/ andNesic et al. /15/ the cathodic
currenti" wasfound as:
1
r l l
11
- Ir - (l l)
i, i ,, i,'t

2t9
Vol. 15,No.v.l-2, 1997 CO2 Corro.sion of Carbon Steel -
From A.lechanismto Empirical Modelling

where the first term reflects the charge transfer controlled current:
_n
h" (r2)
i,, = io'10

while the secondterm is the limiting current.In the caseof Ff reduction,the

limiting currentcomesfrom the masstransferlimitation for Ff reduction:

= k,,F'
t,1*(n.) (13)
[o. ]
of
and in the caseof HzCO:reductionthe limiting currentis a consequence
a slow COzhydrationstepand canbe found asl2ll:

-
u,co., F . lcor\, . (o rroK oruk{,ru)'
ili^1 (14)

Nesic et al. l22l later proposeda theoretical multiplier f for this

expressionwhich is significantly different from unity only when the
thicknessof the masstransferboundaryis of the sameorder of magnitudeas
the reactionlayer for HzCO3hydration(at lower temperaturesand higher
velocities).

/ - coth(
(1s)

which takesinto accountthe effectof flow on the chemicalreactionlimiting

current. Parameter( is the ratio of the thicknessesof the difrrsion and
reactionboundarylayers.
For the caseof direct H2Oreductionthereis no limiting current.For the
anodic dissolutionof iron pure Tafel behaviourwas assumedclose to the
corrosionpotentialand the anodiccurrentis:
n
io = .
i,,1rr1 7 0 h " (16)

i':irt :;
The corrosion potential nnd ir.r..rls:r1 ,* equaiity:

\-; L.
-;
- 0.,
(r7)
,/
4.

Both the models of Gray et al. 18/ and Nesic et al. ll5l can be criticised
on the grounds that while they described in detail the electrochemical
processes occurring on the metal surface, the treatment of the transport

22,0
 S. Ne.'tit, J. Po.rtlethwaite antl M. Vt.hotac Conosion Reviews

processesin the boundary layer, which lead to occurrenceof limiting

currents,was oversimplified.This was also the casein an otherwiseuseful
paper by Liu et al. l23l wherethe scaling correlationsfor different flow
geometriesusedin investigatingCo2 corrosionwereidentified.
Turgooseet al. /24/ werethe first to pavethe way for a more realistic
way of describingthe transportprocesses in the boundarylayer for the case
of CO2corrosion.This approach,long establishedin other areasof electro-
chemistry125/ andcrevicecorrosion126/,wasin the areaof CO2corrosion
carried further in the subsequentstudiesof Nesic /27t and especiallyPots
/28/. The essentiallogic underlyingall threetransportmodelsis similar and
briefly describedbelow.
In a corrosionprocesssome of the componentsare consumedat the
metal surface(e.g.rD while othersare produced(e.g.Fea. Thus gradients
of componentsare establishedin a boundarylayer and transport processes
are initiated. However, the components do not move independenfly
throughoutthe boundarylayer. They are involved in chemicalreactionssuch
as hydration, dissociation"dissolutionand precipitation.As nuny of the
componentsare chargedthey can be transportedonly in such a way that
electroneutrality is maintained locally and globally in the solution at all
times. The transport equation to be solved for the concentratior c; of
componenti in the solutionis:

. ^t 1 '] (18)
6', = - a I:,2:,-t- D,+l + 9[r, A^ + R
g ariI:;D,c,?-'-'a,I dr\ or| *,-
'-/-
L , _j sourceorsrnk
accumulationJi]JttsuntJueroachemicalreactron

along with the electroneutralityequation:

F
) z,c =o
L-,/ t I (1e)

which substitutes a much more complicated Poisson equation. Since

equation (18) is a partial differential equation, one must specr$ the initial
and boundary conditions. For initial conditions, one can, for example, select
a solution which is in equilibrium. However, any other physically realistic
initial conditions are possible. For a one-dimensional model, the boundary
conditions have to be defined at the metal surface and in the bulk solution.
For the bulk it can be assumed that all species are thoroughly mixed by

221
V oL 15,Nos l.-2 , 1 9 9 7 CO2 Corrosion of Cat'bon Steel'
Fronr Mechanisnt to Empirical Modellhry

convectionand thus the concentrationis known. At the metal zurfacethe

flux of a componenti which is involved in an electrochemicalreaction is
relatedto the current density:

N, = -\i, (20)

wherei; is the portion of the corrosioncurrentcreatedby the electrochemical

reactionof componenti at the metal surface.If a componentis not involved
in an electrochemicalreaction at the surfacethen the no-flux condition
appliesat the metal surface.Obviouslythe flux of the Fe* ions at the metal
surfacerepresents the corrosionrate.In orderto definethe currentdensityl;
one mustdefinethe kineticsof the electrodeprocesses wherean exponential
Tafel relationship (12) or (16) can be used. The unknown surface
overpotential tl" can be found from the condition that the total anodic
current must be equal to the total cathodiccurrent (17). This is the link
betweenthe electrochemical and transportmodels.The samecanbe saidthe
other way around: in order to be able to model properly the surface
electrochemistryof COzcorrosionone must know the surfaceconcentrations
of speciesand thosecanbe found only by using the firll transportmodel.
According to our physical picture, th€ species in the solution are
involved in chemicalreactions.In terms of the transportequation(18) this
can be modelled via the source term R; which describesthe rate of
productionor consumptionof a componenti by a homogeneous chemical
reaction.This does not apply for the electrochemicatprocessesat the metal
surface (heterogeneous reaction) which are taken into account via the
boundaryconditionsas describedpreviously.In general,the rate of change
of the concentratiorc; of componenti due to a chemical reactioncan be
written as:

(2r)

where a; is the stoichiometriccoefficient,ki is the reactionrate constant,n is

the order of the reaction and i representsother speciesinvolved in the
reactionwith componenti. This equationappliesto the hydrationand disso-
ciation reactionsin the presentsystem.
When the solubility productfor a salt is reachedor exceed* (", t "i )
,

the processof precipitationis initiated.The rateof precipitationis relatedto

the so-calledrelativesupersaturationrrr=ci lcf , where cf is the equi-

222
S. Neit6, J. Postlethwaite and M. Vrhovac Corrosion Reviews

librium solubility of componenti. The precipitationrate can be calculated

asl29l:

R,: k', ftr) (22)

wherek" is the rate constantwhile m is a kinetic parameterwhich also indi-

cates the morphology of the precipitating film (m : y' nucleation, m = I
crystal growth). It can be aszumedthat the rate of redissolutionof a soluble
film is governedentirely by masstransfer.
This very fundamentalapproachbriefly describedhere is really the only
onewith the potentialto capturethe essentialfeaturesof the COzcorrosion
process.It integratesall the important processesoccurring: electrochemical,
chemicaland transport.Most of the constantsappearingin zucha model are
physically meaningflrl parameters found in engineering and chemical
handbooks. Others can be determined in customdesigned experiments.
According to the reviewers' best knowledge,this kind of approach is
currently being further developedat a few researchcentres:IFE Kjeller,
Norway,and SIIELL in Amsterdam,Holland.
Achour et al. /30/ haveuseda similar modelto simulatepit propagation
of carbonsteelin COzenvironmentsunderhighly turbulentconditions.The
initial direct comparisonof the modelpredictionswith the field data was not
satisfactory,which is not surprising grven the complexity of processes
occurring.
Dalayanet al. /3Il havepresenteda similar approachto COzcorrosion
modelsto the onejust described,simplified by using an establishedmass
transfer pipe flow correlationinsteadof the transport equation (18) and
neglecting the kinetics of the chemical reactions.The latter could lead to
erroneouspredictions,especiallyas it is known that someof the reactions
(e.g. CO2hydration)arevery slow andcanbecomerate-determining.
In the numerouspapersof Adamsand collaboratorsl32l thereis a men-
tion of a fundamentalCOzcorrosionpredictionmodel (Model F). However,
due to the proprietary nature of the model, very little is known about the
assumptionsother than that it is a mass transfer based procedure for
calculatingthe corrosionrate.

Semi-empiricalmodels

Most of the modelspresentedaboveare the so-called"worst-case"models.

This meansthat they addressthe problemof bare steelzurfacescorroding in

223
VoL 15,Nos.1-2,1997 CO2 Corrosion of Carbon Steel -
From Mechanism to Empirical Modelling

water-CO2Eystemwith no protectivesurfacefilms present.A number

a p111e
of other effects have to be accountedfor, in order to bring the predictions
closer to realistic conditionsfound in oil and gas production and trans-
portation.They are listedbelow:

. formationof protectivesurfacefilms,
. presenceofthe hydrocarbonphase,
o presenceof corrosioninhibitors,
. presenceof glycol or methanol,
. condensationin wet gastransport,
o high total pressure,
o multi-phaseflow,
. steelt5'pe,
. corrosionhistorv

Most of the listed efflectsinteract in a complex fashion with the basic

corrosionmechanismsdescribedin the prwious section.The physicalunder-
standing of the ongoing processesis far below the lwel required for any
mechanisticmodelling. Thus most of the modelsaccountingfor theseeffects
use empirical functions which capture the trends noticed in the sparse
experimentaldatabase.
The studiesof de Waard and collaboratorsll8,l9l following the initial
landmark study/5/ representthe most completeattemptto addressthe afore-
mentionedeffects.Improvementsto the original model were mademostly by
calibrating the model constantsagainstthe emerging,more reliable experi-
ments of Dugstad et al. t20t andby adding numerouscorrection factors to
the original correlation(10), which accountfor mostof the prwiously listed
additional eflects.
Dugstad et al. t20l have, in a similar approactr' presenteda 'semi-
empirical model of COz corrosion based on a large body of their own
experiments(the sameonesusedin the last studyof de Waardet al. ll9l)- In
that sense,the two modelsdifrer only in the stratery of identifing the "basic
equation" and the set of correctionfactors. The approachof Dugstad et al.
l20l was to determine a temperature-dependent basic equation (a best-fit
polynomial function) and then multiply it with the correction factors for
pcor, pH, velocity(shearstress)and steelCr content'
Other semi-empiricalmodels can be found in the literature 133,341.
However,they can be seenmost often just as anothervariation of the two

224
 S, NeSi6,J. Postlethwaite and M. Vrhovac Corrosiott Reviews

previously describedmodels,where the experimentaldatabaseis diferenf

and the curyefitting strategyis moreor lessadvanced.

Protectivefilms: In the work of de Waard and Lotz /18/ the formation of

protective films (scales)was modelledby a correction factor obtained
with a best-fit multi-dimensibnalregressionof the high temperatureCOz
corrosionexperimentsof Ikedaet al. /35/. The authorschoseto neglect
the fact that the initial correlation (10) which they were trying to
"correct" is valid only underpure chargetransfercontrolconditions.The
presenceof any protectivefilms implies mass transfer control of the
corrosionprocess- thus renderingthe previouscorrelationinvalid (no
amount of correction factors can change this). Dugstad et al. /20/
accountedfor film formation at higher temperaturesvia the empirical
temperature-dependent functionbasedon their own experiments.
Hydrocarbons: When the walls of the pipeline are wetted with oil
(hydrophobic)no corrosionis possible.In addition,it hasbeensuggested
that somecomponentsin the crudeoil could haveinhibitive properties,
as well as helping to obtain more protectivefllms. However,experimen-
tal support for these effects is still not conclusive.In the work of de
waard and Lotz /l8l the presenceof the hydrocarbon phase was
accountedfor by a so-called water-wetting factor. From the original
experimentsof Wicks and Fraser /36/, a binary prediction factor was
extracted zuggesting that oil-wetting will occur only for water cuts
(contents)lessthan 30Yoand velocitieslarger than I nr/s when all water
can be entrained in the oil phase.Adams et al. l32l judgedthat below
30Yowater cut the tubing will be oil-wef from 30-50yo,intermittent
water wetting occurs,and over 50% the tubing is water wet. Theseare
very crude criteria which neglectthe varying propertiesof the oil and
waterphases,the flow regimeand flow geometry.Howeverthere are no
other relevantmodelsin the openliteraturewhich satisfactorilyaddress
the questionof water wetting and thus previous"rules of thumb" are the
only criteria availablefor the "prediction" of water wetting. Work is now
under way at IFE, Nonvay on a specially designedloop which wiu
contributeto our better understanding,and henceprediction of the effect
of waterwetting on COzcorrosion.

Total pressure: The effect of high total pressureem be modelledby using a

fugacity coefficient which accountsfor non-ideality of the natural gas
/18/. The fugacity coefficientcan be obtained,for example,by soMng
the equationof statefor a COz/naturalgasmixture
225
Vol. 15,Nos.l-2, 1997 CO2 Con'osion of Carbon Steel -
From Mechanism to Empirical Modelling

Condensationand glycol: for the effectsof condensationand glycol present

in wet gas pipelines, de Waard and Lotz llSl suggestedempirical
correction factors determinedfrom a limited experimentaldatabase.In
the very few studiesavailableit hasbeenassumedthat the main "inhibi-
tive" effect of glycol (which is addedprimarily as a hydrate preventer)
comesfrom dilution of the water phase.Howwer, there is no firm proof
for this and the mechanismof CO2 corrosion in water/glycol mixtures
has yet to be discovered.Such studiesare hamperedby the practically
unavoidableanduncontrolledpresenceof impuritiesin the commercially
available glycol which often act as effective corrosion inhibitors
themselves.

Velocity: Velocity can affect COz corrosion only if the colrosion rate is
under partial or total mass transfer control' (the effect of velocity on
water wetting has alreadybeen discussed).If there are no zurfacefilms
present on the steel surface velocity affects the transport of species
through the masstransferboundarylayer. In the caseof the protective
surfacefilms a high velocity can be responsiblefor their removal or
inhibition of their ggowthoftenleadingto localisedattack.In the rvork of
de Waard and Lotz /18/ the effectof velocity in the absenceof surface
films wasmodelledvia a so-calledresistance model:

- =1 1 1 (23)
V
' c oV
r V
r n,

wherethe first term on the right-hand side representsthe rate of electro-

chemical processes,while the second denotes mass transfer. It is
interestingthat this equationproposedobviouslyfrom a simple electrical
circuit analogy actually resemblesthe cathodic equation (11) which is
derived directly from theory. This explainsthe increasedsuccesswhich
the resistancemodels had in approximatingthe experimentalresults,
comparedwith the original equation(10) which did not include any
provisionsfor the effectof velocity.Dugstadet al. l2olhavemodelledthe

*
There is onerather exotic casewhere velocity actuallyhas a diflerent effect. When
iron carbide films (which achrally increase the corrosion) are present on the
surfaceof the steel,high velocitiescan lead to their removaland thus decreasethe
corrosionrate.

226
S. Neiit, J. Po.rtlethwaite and M. Vrhot,ac Corrosiott Reviews

eflectof flow by defininga correctionfactorwhich is a combinedpower-

law-exponentialfunction of the wall-shear stress.Efird et al. 133/
suggested an empiricalpower-lawrelationshipof the shearstress.

Multi-phaseflow: An importantproblemis the modellingof COzcorrosion

in multi-phaseflow sincemany of the pipelinesand flowlines carrying
oil and gas are operatingunder two- or three-phaseflow conditions.
Modelling of multi-phaseflow aloneis a difficult task, evenmore so is
its effecton COzcorrosion.In the absence of protectivefilms multi-phase
flow can lead to very high fluctuating masstransfer rates(particularly in
slug flow) which, in turn, can affect corrosion.Associatedwith this are
even larger fluctuations of the surface shear stresswhich can lead to
removal of protective film and/or inhibitors. There is very little
experimentaldata publishedin this area. One of the rare attemptsof
modelling the multi-phaseflow effect on COz corrosionis the study of
Pots /28/ where the mass transfer was modelledby using a standard
multi-phase flow simulator coupled with a single-phaseflow mass
transfer correlation and the hydraulic diameter concept. In their
empirical studiesJepsonet al. 134/suggested that the Froudenumberis
important for characterising the effect of multi-phase flow on COz
corrosion.

Steel composition: Most of the :iir:dels presentedhere were based on

experimentsrvith only one steel Thebasicmodelof de Waardet al. /l9l
involvedthe n*rmatisedDIli Si5?.steel.Nesicet al. /15/ havestudied
predominantlythe X65 icw carborrsteel,while the modelof Gray et al.
;'8/wasbasedon experirrients witli i.i;,,,APIMN-80 productionrun casing
ste*i. Fr*ru thc resultsof Dugstads; ui. 120/it was clear that different
s:ter:ist]*iLii,u"'i,.
vtrv differentlyunder the sameenvironmentalconditions.
=:i *:iis de Waardet al. /l9lhave groupedall the testedsteels
Tc ;,rcur,r-r:::
in ir.l'i;L:ai";${}rirrs:
(a) normalisedsteelsand O) quenchedand tempered
steeis,linding a different set of model constantsfor eachgroup which
irnprovedthe correlationwith the experimentaldata.Dugstadet al. /201
proposeda correction factor only for the steel Cr content which was
applicableto their baselineformula. Nesic et al. /151,in their electro-
chemical model, have identified exchange current densities for iron
dissolutionwhich differed an order of magnitudefor the X65 and St52
steels.

227
Vol. 15.Nos.I-2, 1997 CO2 Corrosion of Carbon Steel -
From Mechani,sm to Empirical Modelling

All the aboveconclusionsapply to the COz corrosionof "bare" steel

surfaces.The effectof steelcompositionon protectivefilm formation,mesa
attackand adsorptionof inhibitors is evenmore complicated.For example,
it is known that a high Cr content can be beneficial for formation of
protective films but detrimental for performanceof certain corrosion
inhibitors. Obviously,this is an areawheremuch of the new researcheffort
shouldbe directed.

Inhibitors; The effectof corrosioninhibitorsis a storyof its own which goes

far beyondthe scopeof the presentreview. It shouldbe mentionedthat
the typical fashionof incorporatingan inhibitor effect on the corrosion
rateis throughyet anotherfactotfi,;, oftenexpressed in termsof inhibitor
efficienryqi,r,which is:

v,",)*o
=l- (24)
l,rn=I-
v,.,) -fu,n

Corrosionhistory: A steadystatecorrosionrate is very much a function of

the conditionswhich the metal surfacehas experiencedduring its lifetime.
In complex interactions,eventssuch as: changein the production rates,
water chemistry,chemical treatments,presenceof a mill scale or oxide
films, and acid cleaning,can lead to very different steadystatecorrosion
rates.Obviouslythis "history" effect is not easily understoodor modelled.
Nevertheless, evenon a very zuperficiallevel, it is easyto understandthat
there is only one type of model which has a chanceof capturing the history
effect: the transient mechanisticmodel. It is the only one which can capture
the time developmentof the masstransferrate and the kinetics of growth of
a surfacefilm. The mechanisticmodel presentedin the previousheading
(equationslS-22) has the potential to capturethe history effects.Models
which do not includetime as a variableare not suitedfor this purpose.

Empirical models

Not many purely empirical modelsexist in the openliterature.A statistical

l37l anda Markov descriptionstochasticapproach138/were presentedfor
modelling of pitting corrosion;however,this topic is outsidethe scopeof
this review.Anotherexampleis the linear multiple regressionmodel shown
by Adams et al. l32l; however,it is doubtfrrlthat one zuch simple linear
model (evenif it is multivariable)can representa satisfactorydescriptionof

228
S. Ne|ib, J. Postlethwaite and M, Vrhovctc Corrosiotr Review,y

the highly nonlinearand coupledprocesses occurringin COzcorrosion.On

the other hand, it is clear that there is a large knowledgegap betweenour
mechanisticmodelsand the realisticprocesses occurringin the field, which
has to be bridged, for now, with some kind of empirical or statistical
approach.One such attempt is the artificial intelligencecorrosion model
basedon neural networks(NI\i) which is under developmentand is briefly
describedbelow.
In this approachit is assumedthat the actual problem is depictedby
indirect, crudeor noiqyinput parameters,calledprimitive descriptors,such
as: t, pH, pcor, Fe*, HCOr-, v. These are often weakly linked to the
relevantconclusio,n, basedon predictedoutputvaluesor outputcategory,in
this case the corrosion rate V*n Primitive descriptors are usually
measurable.They containvaluableinformation(often masked)aboutcausal
relationswithin the problem descripfors,which are actually the subjectof
the research.It is possibleto discovertheserelations,or at least to bring
them near to the surface,by introductionof additionalproblemdescriptors
called evolveddescriptors.Someevolveddescriptors,very well known in
engineering,are the Reynolds,Schmidt and Sherwoodnumbers.A parti-
cular problem solution can depend on some very specific evolved de-
scriptors,which have to be discoveredin order to improve the empirical
correlation.
The presentmodel was developedusing the e4perimentaldatabaseof
Dugstadet al. /20l describedbelow. It initially consistedof primitive de-
scriptorsr, pH, pcor, Fe*, HCOr-,v and output V*,. A completeindepen-
dence of velocity, v, from the other primitive descriptorsand its strong
influence on V*, was discovered.During the analysis, parameter v w:ls
substitutedwith the group v0'ox ln(v) (in the descriptorslist) which is, on
average,linearly correlatedto V*r.It was also found that there is another
important evolveddescriptorwhich is / x pH x HCO3-.Using this approactq
other causalrelations were defined betweenprimitive descriptorsand V*,,
resultingin derivationof the following evolveddescriptors,whosemeaning
and influenceis still beinganalysed:
n++
n
['t'()- re vot . ln(r')
el: -
-,1
--
et:
t .pH'HCO; t .pH.HCO, , ug. 11gg;

o tn(rico;)
vn . ln(v) P <'o.
€ ra =r J+ + e6:-
0,-r,
'l€
J

pnf f "pH

229
VoL 15,Nos.I-2, 1997 CO2 Corro.sionof Carbon Sl,:el-
Front Mecheni,vm to Enpirical Modellhrg

The correlation, developedby an Artificial Intelligence (AI) model,

consistedof the "Back Propagation"neuralnetwork(NI'I) combinedwith a
geneticalgorithm (GA) 139,401. Generally,neural networksare aimed for
predictionor patternrecognitionproblems,for which it is difficult to devel-
op a suitableanaly'ticalmodel. The "Back Propagation"neural network is
used for correlationdevelopmentbetweentwo known sets of output and
input data. It consists of three layers of interconnectedcomputational
elementscalledneurones,shownin Figure l. Eachneuroneis characterised
by a sigmoidal transfer function, reqponsiblefor non-linear interpretation
("learning") of input signals,scaledbetween0 and 1. The correlation is
achieved by an adaptive adjustment procedure ("training") of network
connectionconductivitiescalledweights(wi and w6, Figure 1). Training is
performedby repetitivecomparisonbetweennetworkpredictionand correct
answer,followedby backwardprocessingof networkpredictionerror.
Non-linear function of the parameterV*, is stnrcturedas a "Back Propa-
gation" neuralnetwork.The networkinputs aret, P\ Pcor, F€*, HCOr-,v

N e u raNl e tw o rk

l l u t p u t5 l a b

T r a n s f eFr. l

GeneticAlgoritham

F'ig. 1: Artificial intelligencemodelfor determinationof V-, correlation.

230
S. Ne.i;6, J. Po^stlethwaiteand M. Vrhovac Corrosion Reviews

and the harmonicsof Fourier cosineseriesdefinedfor evolveddescriptorse1,

a2, 83,04,€s and ec. The Fourier cosineseriesapproximatethe unknown
analytical form of evolved descriptorscorrelations to the network output
V"or-
The networkinputsare presentedin increasingharmonicorder:Pr, Pz...,
p, wherer : l0 is the total numberof harmonicsarbitrarily chosenfor each
evolved descriptors. The compressiveneural network architecture for
function V*, is definedby 66 input neurones,51 hidden neuronesand one
output neruone.
During the training, the network operatesinteractively with a genetic
algorithm (Figure 1). Genetic algorithms are stochasticsearch tools for
determinationof the fittest problem solution accordingto somepre-defined
criteria. The purposeof the genetic algorithm, shown in Figure 1, is to
acceleratethe network training by determinationof the most representative
sequenceof training pattemsfrom the training dataset. The fittest training
sequence generatesthe minimal training error.
NN is a convenienttool for testingthe suitability of different assumed
evolveddescriptors,becauseof its flexibility in their accommodation into the
correlationmodel. The best set of chosendescriptorsis the one for which
NN performs the best extrapolation.On the basis of the knowledge of an
extracted set of evolved descriptors,the correlation developmentcan be
performedin an even simpler way. Under the aszumptionthat a physically
correctanalyticalexpressionis proposedfor the correlationbasedon evolved
descriptors,its multivariablenon-linearregressioncoefficientscan be easily
determinedby simple GA. The model quality is certainly not basedon the
modelling technique used, but on the proper selection of the problem
descriptors.

Comparison

It is challengingat this point to comparethe performanceof someof the

modelsdiscussedin the previoussections.Four modelswereselectedfor this
task:
r the mechanistic(electrochemical) model of Nesic et al. ll5l (in further
text referredto asthe EC model),
o the semi-empiricalmodelof de Waardet al. /19/ (SIIELL model),
. the semi-empiricalmodelof Dugstadet al. l20l (WE model),and
. the empirical neural network model presentedin the previous section
(NN model).

23r
VoL 15,Nos.l-2, 1997 CO2 Conosion of Cat"bon Steel -
Front Mechanism to Empirical Modelling

It is not easy to find a sufffrcientlywide and reliable experimental

databasewhere all of the relevantparameterswere measuredand controlled
so that it could be usedas a test case.The study that satisfiesthis criterion
bestis the KSC2 database presentedby Dugstadet al. 1201. k coversa broad
rangeof e4perimentalconditions:temperaturet 20-90oC,ptpe velocity v =
=
0.1-13m/s, pH 3.5-7,COzpartial pressurePco, = 0.3-26bar, ferrousion
concentrationFe* = l-230 ppm. The corrosionrate was measuredon flat
diagonallymountedSt52 steelcouponsusing a radioactivemeasuringtech-
nique in experimentswhich lastedfrom a few daysto a few weeks.Long
duration experimentswere neededto obtain stablecorrosionrates due to
growth of iron carbidefilms. Most of the experimentswere conductedunder
conditionswhereprotectivefilms did not form; however,in someof the high
temperatureexperimentsprecipitation of iron carbonatescalesmay have
occurred.The total numberof experimentsusedis 330. For more detailsof
the experimentaltechniquesand the resultsseethe original paper 1201.
It shouldbe pointed out that the KSC2 databasewas usedby two of the
four modelscomparedbelow (SHELL and IFE) for calibrationof the model
constants.In that sensethe comparisonsshownbelowcan be seenprimarily
as a test of interpolation capabilitiesof the mentionedmodels.The EC
modelwasnot calibratedagainstthis database and the presentedcomparison
representsan extrapolationpredictionfor this model. The NN model was
actually trained with 300 of the total 330 cases.The other 30 caseswere
randomlysampledand extractedin orderto testthe model'sability to extra-
polateby comparingthe predictionswith an "unknown" data-set.
As shown in Figures 24, the SIIELL, fFE and NN models all show
reasonableinterpolationcapabilities.By far the bestagreementbetweenthe
predictionsand the KSC2 experimentalvalueswas obtainedfor the purely
empirical NN model with a correlationcoefficientof f = 0.99. The IFE
modelwassecondbest1l :0.82) followedby the SIIELL model(f : A.72).
The main discrepancies for the lasttwo modelscomefrom the high tempera-
ture experimentswhereprotectivefilms probablyformed.With thesepoints
removedfrom the analysis(alongwith someother outliers)de Waard et al.
/l9l reportedan improvedagreement (r2:0.91) for the SIfiLL model.
The predictionsmade with the EC model (Figure 5) showeda worse
agreementwith the experimentalda:a(f = 0.62).This is not really sur-
prising since the model constantswere calibratedwith a different database
(relatively short glass cell experimentswhere lower corrosionrates were
measuredthan in the KSC2 experimentsdue to the absenceof iron carbide

232
S. Neiit, J. Postlethwaite and M. Vrhovac Corrosion Review.s

100
0
o

80 o

0
o9 o
E I
-oo
o
-? 60 o
0 o
tr o o
0
oo
O 0 o $
rc o ./'
oo oo 'oo
940 8
.o or,o A
o
- $9
o
E to
q) o o o
":s:'
o- o o
Oo
20 oO^.4"
o9 S H E L Lm o d e l

0 20 40 60 80 100
measuredCR / (mm/y)
Fig. 2: Comparison
of the SFIELLmodelwith the KSC2experimental
data.

100

Et

80 Et
tr E ../'

E o
tr
e60
n f
EGr EI
EF
t trO
trrrtr
O
]C tr
-Ftrro t r f t
940
.c) B a-o8s#
io tr
o Po
o- s
20 b fiffi-o I F Em o d e l

0 20 40 60 80 100
measuredCR / (mm/y)
F'ig.3: Comparison
of theIFE modelwith theKSC2experimental
data.

233
V oL 15,Nos l.-2 , 1 9 9 7 CO2 Corcosion of Carbon Steel -
From A4echanisntto Empirical Modelling

Interpolation
Extrapolation
E
tr
e60
t
O
ro
940
.o
E
q)
t-
o-
20 Neuralnetworkmodel

02040 60 80 100
measured CR / (mm/y)
Fig.4: Comparison of the neural network model with the KSC2
experimental data.

A
A

AA
80 6A A
6A

E AA
tr
vou A A
Aa
AA

6A iA
t A

O of,o 4oo
a A

E a
daa aA A
940 A A AAA A
.o
E
a a o^o.oAo1-o A
0)
o- ^ 4oA Aoligiaoef"
20 "ffi"44; ^ EC model

20 40 60 80 100
measuredCR / (mm/Y)
Fig. 5: Comparisonof the EC modelwith the KSC2 experimentaldata.

234
S, NeYib,J. Postlethwaite and M. Vrhovac Corro.siott Reviews

films normally developedduring more prolonged corrosion tests). It is

interestingthat the extrapolationattempt of the NN on 30 caseswas nearly
I = 0.59@igure4).
as successfirl:

CONCLUSIONS

There is no doubt that modelsof COzcorrosionlag significantly behind the

needsof the industry. Someof the mechanisticmodelspresentedhere, and
which are under developmenf have a realistic chance of improving our
capabilityfor predicting the complexprocesses underlyin1 COzcorrosion.
Indeed, such models can only be as good as our understandingof the
processes modelled.The latter can be improvedby more and better experi-
mentation,which is plannedand executedtogetherwith the developmentof
mechanisticmodels.The modellingactivity would feedon the experimental
resultsfor obtainingnew ideasand for determiningvaluesof constantsbut
would alsoposenew questionswhich would determinethe courseof future
experiments.Sucha combinedapproachwould lead us ultimately to better
COzcorrosionpredictionmodels.
In essence,the transition from mechanisticto semi-empiricalmodels
marksthe point wherethe searchfor understandingbecomes lessimportant,
while the searchfor immediateanswersbecomesdominant. It is a point
wherewe admit that the complexityof the processes surpasses our ability to
understandthem. At that point we resignourselvesto obtaining the answers
in the form of arbitrary functionsconvenientlycalled "correctionfactors".
Even if the individual correctionfactorswere correct,their simultaneoususe
could lead to wrong resultsdue to the complexinterplay of the processes
involved. In addition, we know that these factors are backed with very
limited mechanisticunderstandingand often sparseexperimentaldatabases,
leading to an inevitable conclusionthat their wide and uncritical usecan be
seriouslymisleading.Any extrapolationof the model which includes the
"conection factor approach" is unwise. The shortsightedness of such a
modelling strategyis obvious.On the other hand, it can be acceptedthat
each individual correction factor mentioned here actually points to a
deficiency in our knowledge of COz corrosion and indicates the direction
where increasedresearchefforts shouldbe directed.
Someof the purely empirical modelsconsistingof simple curve fitting
using polynomial, powerJaw, and exponentialfunctions, or even multiple

235
VoL 15,Nos.I-2, 1997 CO 2 Corrosion of Carbon Steel -
From Alechani,wt to Enrpirical Modelling

linear regression,havevery little chanceof successflrlprediction due to the

completelack of mechanisticunderstandingin them. However,other more
sophisticatedempirical techniquesinvolving neural networkscoupledwith
genetic algorithms,when trained with suffficienflylarge and good experi-
mental databases, appearto be more promising, althoughthey too are not
basedexplicitly on theoreticalunderstandingof the processes. Theseadap-
table algorithmscan produce modelswhich have excellentinterpolation but
also good extrapolationcapabilities.Even more important they are able to
recognisepatterns and regularities presentin complex empirical databases
which might otherwisego unnoticed.In that sensethey are a very valuable
tool which provides input and ideasfor further mechanisticmodelling and
the full circle is closed.

NOMENCLATIIRE

b* b" anodicand cathodicTafel slope,V/decade;

ci concentrationof speciesi, mol/m3;
[COrlt bulk concentrationof dissolvedcarbondioxide,mol/m3;
D diftrsion coefficient,m'ls;
E,,u reversiblepotential,Z;
F Faradayconstant(96490C/equiv);
f correctionfactor;
[F']", Vt1, zurfaceandbulk hydrogenion concentration" mol/m3;
i srrrent density,A/m2;
i" anodiccurrent density,A/*'i
i" cathodiccurrent density,A/m2;
i"t chargetransfercomponentof the total current density, A/m2;
io exchangecurrentdensity,A/m2;
i(r^ difr.rsionlimiting currentdensity,A/m2;
ti!- chemicalreactionlimiting currentdensity,A/m2;
k- masstransfercoefficient n/s;
.f
ktnra fonrard reactionrate for the COzhydrationreaction,//s;
Kn a equilibrium constantfor the COzhydrationreaction;
Nr flux of speciesi;
Pco, partial pres$ueof carbondioxide gas,bar;
R universalgasconstant(8.3143J/(molK));

236
S. Ne\it, J. Postlethwaite and M. Vrhovac Corrosiort Reviews

Re = pvl/p Reynoldsnumber
t temperature,oC;
T absolutetemperature,K;
v velocity,m/s;
V*, corrosionrate,mm/y;
z charge;
n: E - Erru overpotentiil,V',
p viscosity,kg/(ms);
p density,kg/m3;
'c, time, s,
(, ratio betweenthe thicknessof the masstransfer and chemical
reactionboundarvlaver.

REX'ERENCES

1. Dr?rjc, D.M., "Iron and its electrochemistryin an active strate",in:

Aspectsof Electrochemistry,Vol. 19,PlenumPress,1989;p.79.
2. Loretu, W. andHeusler,K., *Anodic dissolutionof iron groupmetals",
in CorrosionMechanisms,Mansfeld,F. (ed.), Marcel Dekker, New
York, 1987.
3. Nesic,S., Thevenot,N. and Crolet,J.L., "Electrochemicalpropertiesof
iron dissolutionin COz solutions- basicsrevisited",PaperNo. 3 in:
Corrosion/96,Houston,TX. NACE International,1996.
4. Bockris,J.O.M.,Drazic,D. andDespic,A.R., Electrochemica Acta, 4,
32s (Le6r).
5. de Waard,C. andMilliams,D.E., Corrosion,31, 13l (1975).
6. Schmitt, G. and Rothman, 8., Werl<sto/fe und Korrosion, 28, 816
(1e77).
7. Gray, L.G.S., Andersorq8.G., Danysh,M.J. and Tremaine,P.G.,
"Mechanism of carbon steel corrosion in brines containing dissolved
carbondioxide at pH 4", PaperNo. 464 in: Corrosion/89,Houston,
TX: NACE International,1989.
8. Gray, L.G.S., Anderson"8.G., Danysh,M.J. and Tremaine,P.G.,
*Effect of pH and temperature
on the mechanism of carbon steel
corrosionby aqueouscarbondioxide", PaperNo. 40 in: Conosion/91,
Houston,TX: NACE International,1990.

237
VoL 15,Nos.I-2, 1997 CO2 Corrosion of Carbon Steel -
From Mechanism to Empirical Modelling

9. Hurlen, T., Gunvaldsen"S., Tunold, R, Blaker,F. and Lunde,P.G.,/.

Electroanal.Chem.,180,511 (1984).
10. Davies,H. andBurstein,G.T.,Corrosion,36, 385(1980).
11. Videm, I(, "Fundamental studies aimed at improving models for
prediction of COz corrosion", Progress in the Understanding and
Preventionof Conosion, Proceedingsof the l0th EuropeanCorrosion
Congress,Vol. l. Instituteof Metals,London"1993;p. 513.
*Efect of flow on CO2colrosionratesin
12. Eriksrud,E. and SonWedqT.,
real and syntheticfonnation waters",Advancesin COzConosion, Vol.
I of Proceedingsof the Corrosion/83Symposiumon CO2Corrosion in
the Oil and Gas Industry, Hauser,R.H. and Goddard,P.H. (eds.),
Houston" TX: NACE, 1984;p.20.
13. Bonis,M.R. and Crolet,J.L., "Basicsof the predictionof the risks of
COzcorrosionin oil and gaswells", PaperNo. 466 in: Corrosion/89,
Houston,TX: NACE Internationd, 1989.
14. Delahay, P.,J.Am.Chem.$oc.,74,3497 (1952).
15. Nesic, S., Postlethwaite,J. and Olsen, S., "An electrochemicalmodel
for prerliction of COz conosion", Paper No. 131 in: Corrosion /95,
Houston"TX: NACE International,1995.
16. Zal/ra,J. andNavidi, P., "Cleveroptimal designof materialsand stnrc-
tures", Proceedings, French-Korean Conference on Numerical
Analysisof Structures,Seoul,Korea, 1993.
17. de Waard, C. and Mlliams, D.E., "Prediction of carbonic acid
corrosionin natural gas pipelines",PaperFl in: First International
Conferenceon the Internal and bternal Corrosionof Pipes,Univer-
sity of Durham,UK, 1975.
18. de Waard, C. and Lotz, IJ., "Predictionof COz corrosionof carbon
steel", Paper No. 69, in: Corrosion/93,Housto& TX: NACE
International,1993.
19. de Waard, C., l-otz, U. and Dugstad,A., "Influence of liquid flow
velocity on CO2corrosion- a semi-empiricalmodel", PaperNo. 128
in: Corrosion/9i, Houston,TX: NACE International,1995.
20. Dugstad,A., Lund€, L. and Videfl, K., "Parametricstudy of COz
corrosionof carbonste€I",PaperNo. 14, irf' Corrosion/94,Houston,
TX: NACE International,1994.
21. Vetter,K.J., "Electrochemicalkinetics,theoreticalaq)@ts",Sections1,
2 and 3 of ElectrochemicalKinetics: Theoreticaland Experimental
Aspects,New York: AcademicPress,1967; p. 235 [translatedfrom
Germanl.

238
S. Neiib, J. Postlethwaite and M. Vrhovac Corroston Reviews

22. Nesic, S., Pots, B.F.M., Postlethwaite, J. and Thevenot, N.,

"Superlnsition of difrrsion and chemical reaction limiting currents -
Application to COz corrosion", Journal of Cotosion Science and
Engineering, Vol. l, Paper No. 3, World Wide Web,
http:/hvww.cp.umist.ac.uAJCSMVolI/PAPERS3/VIP3int.htm.The
corrosion Information Server, The Corrosion & Protection Crntre at
UMIST, Manchester,IIK 1995.
23. Liu, G., Tree,D.A, andHigtL M.H., Corrosion,50, 584(1994).
24. Turgoose,S., Cottis, RA. and Lawsoq K., "\dsdslling of electrode
processes and surface chemisby in carbon dioxide containing
solutions", ASTM Symposiumon ComputerModelling of Corrosion,
SanAntonio, Tlq 1990.
25. Newman, J.S., ElectrochemicalSystems,SecondEditioq Part C,
PrenticeHall, 1991.
26. Watson,M.K. andPostlethwaite, 1., Corrosion,46,522 (1990).
27. Nesic, S., "Prediction of transportprocessesin COz corrosion",Pro-
ceedingsfromthe I0th EuropeanCorrosionCongress,Progressin the
Understandingand Prevention of Corrosion, Vol. 1. Institute of
Metals,London"1993;p. 539.
28. Pots,B.F.M., "Mechanisticmodelsfor the predictionof CO2corrosion
rates under multi-phase flow conditions", Paper No. 137 in:
Corrosion/9| Houston,TX: NACE International, 1995.
29. Johnson,M.L. and Tomson, M.B., "Ferrous carbonateprecipitation
kinetics and its impact on COz corrosion", Paper No. 268 in:
Corrosion/9/,Houston,TX: NACE International,1991.
30. Achour, M.H., Kolts, J., Johannes,A.H. and Liu, G., *Mechanistic
modelling of pit propagationin COz environmentsunder high turbu-
lence effects", Paper No. 87 in: Conosion/93, Houston"TX: NACE
International,1993.
31. Dayalan,E., Vani, G., Shadl€V, J.R, Shirazi,S.A. and Ryticki, E.F.,
"Modelling COzcorrosionof carbonsteelin pipe flof', PaperNo. 118
in: Corrosion/9f,Houston,TX: NACE International,1995.
32. Adams,C.D., Garber,J.D.,Walters,F.H. and Singh,C., *Verification
sf ssmputer modelledtubing life predictionsby field data", PaperNo.
82 in: Corrosion/93,Houston,TX: NACE International,1993.
33. Efird, K.D., Wrighg E.J., Boros,J.A. and Hailey, T.G., "Wall shear
stressand flow acceleratedcorrosionof carbon steel in sweetproduc-
tion", Paper No. TS 14 194, in: I2th International Corrosion
Congress,Houston,TX: NACE International,1993.
239
 VoL 15,Nos.I-2, 1997 CO2 Cono,sion of Carbon Steel -
From Mechanisnt to Enrpirical Modelling

34. Kanwar, S. and Jepson"W.P., "A model to predict swe€tcorrosionof

multiphase flow in horizontal pipelines", Paper No. 24 in:
Corcosion/94,Houston,TX: NACE International,1994.
35. Ikeda, A., Mukai, S. and Ueda,M., "Preventionof COz corrosionof
line pipe and oil country tubular goods", Paper No. 289 in:
Corrosion/84Houston,TX: NACE International,1984.
36. Wicks,M. andFraser,J.P.,MaterialPerformance, 14,9 (1975).
37. Sheikh, A.K., Boah, J.K. and Hansen,D.A., Corrosion, 46, 190
(lee0).
38. Provan"J.W.andRodriguez,E.S.U, Corrosion,45,178(1989).
39. Vrhovac, M., "Combined empirical and analytical modelling of
hardness softening due to particle coarsening using artificial
intelligencet@ls", ASME 95 Congress,IMECE Proceedings, MD-Vol.
69-2,SanFrancisco,CA, USA,Novemberl2-I7, 1995.
40. Vrhovac, M., Jo6o, C.R. and Rodrigues, R., "Automation of
microstructureanalysisby artificial intelligenceimageprocessing",Int.
Conf. SMIRT 13,PostConferenceSeminar:Applicationsof Intelligent
Sofh,uareSystems in Power Plant, Process Plant and Structural
Engineertug, S5oPaulo,Brasil,August2I-23,1995.

240

View publication stats