LABORATORY INVESTIGATION AND NEURAL NETWORKS MODELING OF DEICER INGRESS INTO PORTLAND CEMENT...

LABORATORY INVESTIGATION AND NEURAL NETWORKSMODELING OF DEICER INGRESS INTO PORTLAND CEMENT

CONCRETE AND ITS CORROSION IMPLICATIONS

Xianming Shi1'2' *, Yajun Liu1, Matthew Mooney1, Michael Berry2, BarrettHubbard1, Tuan Anh Nguyen1

' Corrosion and Sustainable Infrastructure Laboratory, WesternTransportation Institute, PO Box 174250, College of Engineering, Montana

State University, Bozeman, MT 59717-4250, USA

2 Civil Engineering Department, 205 Cobleigh Hall, Montana StateUniversity, Bozeman, MT 59717-2220, USA

* Corresponding author: Xianming Shi, Ph.D., P.E., Phone: 406-994-6486,email: xianming_s@coe. montana. edu

ABSTRACT

By exposing reinforced concrete samples to four common chloride-baseddeicers, the corrosive effect of chloride-based deicers on rebars and dowelbars was systematically investigated. The experiments were designed in sucha way that the effect of deicers on reinforced concrete can be characterized inan accelerated manner, by either ponding the concrete samples with deicersolutions at room temperature, or incorporating pressurized ingress, wet-drycycling and temperature cycling into the test regime. The chloride ingressover time was monitored using a custom-made chloride sensor embedded ineach concrete sample. Also periodically measured were the open circuitpotentials (OCPs) of the top bar in concrete. Once the chloride sensordetected the arrival of sufficient chlorides near the top bar and the OCP dataindicated the possible initiation of top bar corrosion, the corrosion rates ofrebars and dowel bars are characterized by macrocell current and corrosioncurrent density derived from electrochemical impedance measurements.From a modeling perspective, artificial neural networks (ANNs) were used toachieve better understanding of the complex cause-and-effect relationshipsinherent in the deicer/concrete/bar systems and were successful in findingmeaningful, logical results from the noisy data associated with the deicerponding experiments. According to the ANN modeling, corrosion inhibitor(and possibly other additives in the inhibited CaCl2 and MgCl2 deicers) didslow down the ingress of chloride into concrete (indicated by more positive

105

Brought to you by | Missouri University of Science and TechnologyAuthenticated | 131.151.26.29

Download Date | 5/6/14 4:40 PM

Volume 28, Nos. 3 & 4, 2010 Laboratory Investigation andNeural Networks Modeling

chloride sensor potentials),, reduce the pitting risk of rebars and dowel bars inconcrete (indicated by more positive OCP values) and reduce their corrosionrate in concrete (indicated by reduced macrocell current and corrosion currentdensity of top bar). For both concrete mixes and both types of deicerponding, the overall risk of deicers to reinforced concrete was in the order of:non-inhibited NaCl > inhibited NaCl > inhibited CaCl2 deicer > inhibitedMgCl2 deicer. While the corrosion inhibitors in deicer products provide somebenefits in delaying the corrosion initiation of rebars and dowel bars inconcrete, such benefits seem to diminish once the active corrosion of the barsis initiated. In other words, the inhibitors showed little benefits in re-passivating the actively corroding bars in concrete or in stifling the corrosionpropagation.

Keywords: corrosion-inhibited deicer, Portland cement concrete, rebar,dowel bar, NaCl, CaCl2, MgCl2

1. INTRODUCTION

In northern regions across North America and other cold-climate areas,snow and ice control operations are crucial to maintaining a safe, mobile andproductive roadway system. To keep highways clear of ice and snow inwinter, large quantities of deicers, either solid or liquid, are massivelyapplied. For the purpose of cost-effectiveness, deices utilized for wintermaintenance usually contain chloride compounds, such as sodium chloride(NaCl), magnesium chloride (MgCl2), and calcium chloride (CaCI2) /!/. Suchchemicals can cause corrosion damage to transportation infrastructure, aschloride ingress from deicer applications is one of the primary forms ofcorrosion attach for reinforced concrete structures. The mechanism by whichthe deleterious influence of chloride-based deicers on rebars in concretestructures is quite well-known. Through reactions with cement paste andaggregates, chloride-based deicers impose detrimental effects on concreteinfrastructures; thereby reducing concrete integrity and fostering rebarcorrosion. In a recent study sponsored by the U.S. National CooperativeHighway Research Program (NCHRP) 111, all chloride-based deicers wereranked equally corrosive for reinforcing steel in concrete. MgCl2 and CaCl2deicers are known to deteriorate concretes by accelerating the alkali-carbonate reaction /3-4/. MgCl2 is more aggressive to exposed metals thanCaCl2 and NaCl, which is attributable to the hygroscopic nature and thereaction between Mg2+ and hydrated products in cement paste /3-6/. CaCl2features a similar corrosion characteristic to those exposed to MgCl2, but witha slower and less severe characteristic 111. Long-time exposure of concrete toNaCl can lead to accelerated alkali-silica reaction by supplying additionalalkalis to concrete /8-14/. In addition to chloride-induced rebar corrosion,

106Brought to you by | Missouri University of Science and Technology

Authenticated | 131.151.26.29Download Date | 5/6/14 4:40 PM

X.. Shi, et al. Corrosion Reviews

chloride-based deicers can also trigger scaling problems and freeze-thawdamage of concrete.' A recent NCHRP study identified the deicer corrosion to steel rebar as theprimary concern, followed by detrimental effects to vehicles, concrete ingeneral, structural steel, and roadside structures 121. In 2007, the U.S. soldapproximately 20.2 million tons of deicing salts for use in wintermaintenance I\5I. One study has estimated that the use of road salts imposesinfrastructure corrosion costs of at least $615 per ton, vehicular corrosioncosts of at least $113 per ton, aesthetic costs of $75 per ton if applied nearenvironmentally sensitive areas, in addition to uncertain human health costs/16/. The estimated cost of installing corrosion protection measures in newbridges and repairing old bridges in the Snowbelt states is between $250million and $650 million annually /17/. As such, the growing use of chloride-based deicers raised economic concerns for transportation. It should be notedthat any repairs to the infrastructure translate to costs to the user in terms ofconstruction costs, traffic delays and lost productivity. Indirect costs areestimated to be greater than ten times the cost of corrosion maintenance,repair and rehabilitation /18/.

It is a popular practice to add corrosion inhibitors and other additives todeicer products, in an effort to reduce their corrosive effects on bare metals/16, 17, 19/. The relative corrosivity of deicers is dependent on variousfactors related to the metal/deicer system, as the cation (Na+, Ca2+, or Mg2+)and inhibitors associated with Cl" affect the pH value of the electrolyte andthe chloride diffusion coefficient in concrete, thus posing different levels ofcorrosion risk to the rebar in concrete. However, little is known about thepossible contribution or risk deicer inhibitors (and other additives) pose toinfrastructure preservation. Chloride ingress into concrete is a complexprocess, which in the highway environment is further complicated by thefreeze-thaw cycles and wet-dry cycles experienced by roadways and bridges.Therefore, research is needed to determine whether or not corrosion-inhibiteddeicers provide benefits in mitigating the corrosion of rebars or dowel bars inconcrete, relative to the "straight salt" (non-inhibited sodium chloride).

The objectives of this research are to assess the effect of chloride-baseddeicers on reinforced concrete structures such as roadways and bridges and todetermine whether or not reducing deicer corrosiveness helps preserve thetransportation infrastructure. To this end, a laboratory investigation wasconducted using reinforced concrete samples exposed to various deicers. Thelaboratory investigation aims to simulate the effect of deicers on reinforcedconcrete in an accelerated manner, by either ponding the concrete sampleswith deicer solutions at room temperature, or incorporating pressurizedingress, wet-dry cycling and temperature cycling into the test regime. Thecorrosion behavior of rebars and dowel bars in concrete subject to inhibitor-meditated deicers was studied in terms of corrosion potential. Following suchexperimental efforts, four ANN-based predictive models were establishedusing the numerous data obtained throughout the deicer ponding experiments.




For each model, the raw data first underwent a series of quality control stepsto eliminate erroneous readings or apparent outliers. Thereafter, five percentof the data that passed the quality control was randomly selected as thetesting data and the remaining data were used as the training data for theestablishment of the ANN model. The testing data were used to monitor theperformance of the model during training. The training process involvedselecting the appropriate number of hidden layer nodes and determining theappropriate limit of allowable training error.

2. MATERIALS AND METHODS

2.1. Deicers of Interest

This study involved four liquid deicers. The non-inhibited NaCl brineserved as a control and was a 23% aqueous solution prepared by the researchteam, using solid "rock salt" provided by the Montana Department ofTransportation (MDT), Bozeman District, from the salt stockpile. Thecorrosion-inhibited NaCl brine was prepared by adding a given amount ofShield GZT™ inhibitor into the non-inhibited NaCl brine and stirring to blend(as specified by the vendor). The corrosion-inhibited MgCl2 brine, FreezgardCI Plus™ was provided by the MDT-Bozeman district from their liquidstorage tanks and contained approximately 30% MgCl2 by weight of thesolution. The corrosion-inhibited CaCl2 brine, Geomelt C7™ and wasprovided by WSDOT directly through America West EnvironmentalSupplies, Inc., and contained approximately 30% CaCl2 by weight of thesolution along with some other chlorides (MgCl2, NaCl and KC1) as integralpart of the formulation. On average, these liquid deicers were further dilutedby a factor of 100:31 before being used for the deicing ponding experiments.The average daily concentrations of the deicer solutions used for the deicingponding experiments are provided in Table 1. Note that the average daily saltand Cl" concentrations for the Cyclic Exposure Test was lower that those forthe Natural Diffusion Test, since the former incorporated dry cycles featuringzero concentrations in the pond.

The Pacific Northwest Snowfighters (PNS), an Association oftransportation professionals for British Columbia, Washington, Idaho,Montana, Oregon, and Colorado, has implemented testing protocols andguidelines for new product qualification for deicers. A central feature of theserequirements is the presence of corrosion inhibitor in deicers, and thequalification and evaluation of all deicers by a modified National Association




αΒV

ιωΒ

1ΟΟ*L.

_α4>•οu

«2Όυ

Ή 9κ inΞ ΞΗ .σ"

_s'Sηο

JSQ

^

Ό

Λ

"ΕC.SC!ΒIΒ

Iaυ

ί

w>£|aQ

ja

2

>» c™ BQ ra

Φ c5 υ

O £.= ·.= o(0 t3 »i 3 Sc m ^1 E §Q 5 o

t ° ° „Q « £ c»uri0) C >, )

>* c

S 1Sbl lΓ9 Q

5 o<

|||C 3 f

Q i:

t I B -Q ra £ c

l f i l01 C >, (0

< o —

o> 01 r*-· oq o CN tp

0

00 CO t ΙΛCM CM CM CVJ

ob oo h- ί-ο CO O (D(D <D h- h-

τ- CO 00 <0CO CO ^ OO

g? ^ g? g?CO ^ Ϊ*· O)CO CO CM CM

vO v? vO vOtf · tf1* 0s- 0s"r- oo CM oor- h- oo oo

C 1*-ο οΟ *~" ** "^TS -t Οϊ .9 S? ci cP β^Ι § 'S 1 ^ s s s^ o 51

o —J2 -~ r- coω 4> ξ ob oo ° ΙΛo < S *° in ^ σ>Ξ

ά §

I|!| l 1 1 lu. »

fc ? ? ? "S'S c S --9 -·9 ΰ -SQ o . c a £ c o - £ a - £ o >

Z = Z = 2 = £5

s

NP

^CN

S?σ>ώ

σι

S?CO

JCO

S

109Brought to you by | Missouri University of Science and TechnologyAuthenticated | 131.151.26.29



of Corrosion Engineers (NACE) corrosion test. As such, for this study wetested the deicer corrosivity before using them to conduct the pondingexperiments of reinforced concrete samples. The corrosion tests followed thegravimetric method as specified by the PNS Association using the modifiedNACE Standard TMO169-95, aimed to assess the 72-hour average corrosivityof a deicer solution to carbon steel. The PNS Association has modified thisprocedure so that the test procedure uses 30 ml of a 3% chemical productsolution as received per square inch of coupon surface area for the corrosiontest.

The PNS/NACE corrosion test entailed cyclic immersion (10 minutes inthe solution followed by 50 minutes exposed to air) of multiple parallel steelcoupons for 72 hours on a custom-design machine (by AD-Tek™), followedby a measurement of weight loss. The weight loss result in MPY (milli-inchper year) was translated into a percentage, or percent corrosion rate (PCR), interms of the solution corrosivity relative to a eutectic reagent-grade NaClbrine. The coupons used were TSI washer steel and met ASTM F 436, Type1, with a Rockwell Hardness of C 38-45. The test results provide the baselineof the four deicers used in this study and indicate that the corrosion-inhibiteddeicers (PCR of 18.5, 19.9 and 17.0 for the inhibited NaCl, inhibited CaCl2deicer and inhibited MgCl2 deicer respectively) had much lower corrosivitythan the non-inhibited salt brine as control (PCR of 94.0). Note that thecorrosivity data correspond to the deicer solutions diluted from theirindividual eutectic concentration by a factor of 100:3 (as specified by thePNS Association).

2.2. Design of Deicer Ponding Experiments

To investigate the potential impact of deicers to rebar or dowel bar inconcrete, the following concrete/bar combinations are included.

a. One sound concrete mix representing bridge decks with uncoatedsteel rebar inside;

b. One "cracked" concrete mix representing bridge decks withuncoated steel rebar inside;

c. Same as "a" except with 0.1% NaCl (by weight of concrete)admixed in fresh concrete;

d. One concrete mix representing pavement concrete with a sawedjoint and a MMFX™ dowel bar;

e. One concrete mix representing pavement concrete with a sawedjoint and epoxy-coated dowel bar;

f. Same mix as "d" except for a stainless steel tube with epoxy-coateddowel bar insert.

Note that for the combination "b", mechanical force was not used togenerate the cracks since such methods (e.g., compression) would generate

110 Brought to you by | Missouri University of Science and TechnologyAuthenticated | 131.151.26.29



cracks of non-reproducible nature. Instead, a carbon steel strip (withthickness of 5 mm, length of 50 mm and width of 25 mm) was embedded inthe concrete sample, with its upper side directly exposed to deicers and itlower side 25 mm into the concrete. As such, during the deicer pondingexperiments, the generation of micron-scale cracks is anticipated at the steelstrip/concrete interface through the corrosion of the steel strip and thesubsequent formation of corrosion products.

Table Error! Reference source not found.2 presents the overallexperimental design used in this study. The reinforced concrete samples wereexposed to deicers by either ponding at room temperature (referred to asNatural Diffusion Test), or by incorporating pressurized ingress, wet-drycycling and temperature cycling into the test regime (referred to as CyclicExposure Test). As shown in Table 2, a total of 72 reinforced concretesamples were used in the Cyclic Exposure Test (three duplicates for eachdeicer/concrete/bar combination) and a total of 16 reinforced concretesamples were used in the Natural Diffusion Test (two duplicates for eachdeicer/concrete combination, using only uncoated rebar).

2.3. Reinforced Concrete Samples: Materials and Preparation

For preparing the reinforced concrete samples used in this study, themixing water, type I-II cement, coarse and fine aggregates, chemicaladmixtures, uncoated number 4 rebars (Pacific Steel), and epoxy-coateddowel bars were obtained from Stoneway Concrete Inc. (Renton, WA). TheMMFX dowel bars were provided by the ACME Concrete Paving, Inc.(Spokane, WA). The stainless steel tubes with epoxy-coated dowel bar insertwere provided by the American Highway Technologies, Inc. (Seattle, WA).

The reinforced concrete samples used for the Cyclic Exposure Test hadtwo bars in them and the distance between the top bar and the bottom bar wasdesigned to be 1" (25.4 mm). The reinforced concrete samples used for theNatural Diffusion Test only had one bar in them. This study used twoconcrete mixes, both obtained from Stoneway Concrete Inc., as shown inTable 3. The concrete mixes had no fly ash included, in order to simulate theWSDOT roadway pavements and bridge decks built before 1980s. Note thatthe designed water-to-cement (w/c) ratio was 0.39 and 0.38 for the pavementand bridge mixes respectively.

The optimum dimensions of the concrete samples were selectedconsidering all the constraints imposed by the optimum design, such as themaximum aggregate size, weight of the cured specimen, bar dimensions andlocations, transport distance of deicer solution from the pond to the top bar,and location of the chloride sensor. The reinforced pavement concretesamples used for the Cyclic Exposure Test had a rectangular shape and finaldimensions of 16"xl2"*9" (LengthxWidthxHeight: 406x305x229 mm),




4)E

°CuQ.XtiM

H i-^ <uo'u•Ό<l-lOB

|_

Ο)

gj- cd

83« 13 ι« * c1*2

m **• · Ste ωM 'S

Cycl

ic E

xpos

ure

'to

tally

12

spec

imen

s in

•s_^

Cl

&Ϊ.y

Q

(N

CJ00S

1

<N

CJcdCJ

l

Ϊl"<

7jcd

^

m

m

ΓΛ

ΓΛ

Unc

oate

d Re

bar

m

ΓΛ

ro

ΓΛ

Unc

oate

d Re

bar w

/ste

el str

ip in

con

crete

ΓΛ

ff^

m

ΓΛ

Unc

oate

d Re

bar w

/i. l

wt.%

adm

ixed

NaC

l

o

tt$<ω

*&.oq

ΓΛ

Γ*"ΐ

ΠΊ

Γ^Ί

Epox

y Co

ated

Dow

el

m

m

cn

ΓΛ

~^ t;<u ωg*"ο ΰ

§(Λ

ω χ

•'s R5 ωΕΛ

ΓΛ

m

m

m

00

10QXu.

S

it·*κgζa.

J= ω

Μ ΙU Qi

• - 1Q. "CJ Cu C^ ΟU *-ιa> cdH 2

Nat

ural

Diff

usio

n(to

tally

16

spec

imei

l*}

fe

. JJ

Q

r-iG

bX)^>>— 1

GcdCJ

I

1GcdZ

CN

(N

(N

(N

Unc

oate

d Re

bar

.M

^^>

|oq

<N

(N

(N

es

Unc

oate

d Re

bar

iiκs3ζ^





Volume 28, Nos. 3&4. 2010 Laboratory Investigation andNeural Networks Modeling

eoU

Ai

I

2ώ

iI

g

55

a: £ # s?

8 8ΙΟ IDΟ Ο

^_ -8.8

S u .ycB ^ <^ 'S s




with a pond 2.5" (63.5 mm) deep in its upper portion. A custom-madewooden mold used for fabricating such samples is shown in Figure la. Thedowel bars had the dimension of length 18"x diameter 1.5" (LxD:457.2X38.1 mm) and had a concrete cover of 2" (50.8 mm) over the top bar.In light of the fact that most mix designs for WSDOT roadway pavementshave a maximum aggregate size of 1.5" (38.1 mm) to guarantee sufficientstrength, a pavement mix design with a maximum aggregate size of 1.5"(38.1 mm) was used. The rebars had the dimension of length 18"x diameter0.5" (LxD: 457.2x12.7 mm) and had a concrete cover of 1.5" (38.1 mm)over the top bar. A bridge mix design with a maximum aggregate size of0.75" (19 mm) was used to minimize the sample size and to minimize therisk of hydration cracking.

Fig. 1: (a) A custom-made wooden mold used to fabricate the reinforcedpavement samples; (b) a portion of the mold to make the pond; (c) afabricated concrete sample with the ponding mold not yet removed




This designed mold gave the fabricated concrete samples walls of 3" (76mm) thickness around the pond, with reinforcing bolts in the walls. Thisdesign survived a pressure of 10 psi during the mock tests without showingany noticeable cracking and maintained the pressure in the enclosed pondrelatively well. The pond was made by embedding a wooden mold (seeFigure Ib) in the upper portion of the concrete specimen during casting, andthen removing it two days after the concrete casting, leaving a "pond" withdimension of 10"χ6"χ2.5" (254x152.4x63.5 mm) in length, width andheight respectively. Figure Ic shows a fabricated concrete sample with theponding mold not yet removed. Note that a plastic ponding mold would havegreatly facilitated its de-molding. The bridge concrete samples used for theCyclic Exposure Test had a rectangular shape and final dimensions of16"χ12"χ7.5" (LxWxH: 406x305x190 mm). The reinforced concretesamples used for the Natural Diffusion Test had a rectangular shape and finaldimensions of 16"χ12"χ?.5" (LxWxH: 406x305x190 mm) and 16"χ12"χ5"(406x305x127 mm) respectively, for the pavement and bridge mixesrespectively, with a pond 2.5" (63.5 mm) deep in its upper portion.

All concrete specimens according the design of experiments in Table 2were fabricated with the received rebars, dowel bars, as well as the custom-made specimen molds and custom-made chloride sensors. To allow someredundancy, a total of 78 samples were cast for the pavement mix and 63were cast for the bridge mix. Slump and air content measurements wereperformed to check the workability and quality of the freshly mixed concrete.For the pavement and bridge mixes, the slump was tested as 3.5" and 4.75"respectively, and both values fell in the specified ranges as provided byStoneway (2.5" - 4.5" for pavement mix and 4.0" - 5.0" for bridge mix). Theair content was tested as 5.7% and 5.1% for the pavement and bridge mixesrespectively, and both values fell in the specified ranges as provided byStoneway (4.5% - 7.5% for pavement mix and 3.5% - 6.5% for bridge mix).Six cylinder specimens for compressive strength test were also prepared forquality control of each mix, which showed average 7-day strength of 3943psi and 4318 psi for the pavement and bridge mixes respectively, indicatingreasonable quality of these fabricated concrete samples. The deicer pondingexperiment did not start until the concrete specimens were at least 60 daysold to allow sufficient hydration.

2.4. Chloride Sensors

The chloride ingress over time was monitored using a custom-madechloride sensor embedded in each concrete sample. Chloride sensors made ofsilver/silver chloride (Ag/AgCl) electrodes were fabricated in the WTICorrosion & Sustainable Infrastructure Laboratory (CSIL) using an electro-deposition process. First, the silver wire was sanded and cleaned, followed by




the measurement of its diameter. Then, it is mounted in a 0.2M potassiumchloride (KC1) solution along with a clean graphite rod, with them connectedto the positive and negative end of a computer-controlled Princeton AppliedResearch™ Potentiostat (serving as the power source). A constant current isapplied by the Potentiostat to provide a 1 mA/cm2 current density on thesilver wire for one hour (to deposit a silver chloride layer on it), followed bythe rinsing of the electroplated silver wire.

As shown in Figure 2, each chloride sensor (in dark color) was welded toa copper wire at each end (with the welding points sealed with epoxy resin)and then carefully rinsed before being embedded above the top bar in theconcrete samples. The chloride sensor was fixed at 5 mm above the top barinstead of at the same vertical height as the top bar, as a result of the specificconfiguration we chose and the effort to avoid any possible physical contactof the sensing layer with the top bar. The quality of these custom-madechloride sensors were calibrated by being embedded into concrete samplesfeaturing a mix design similar to Table 3 but with known amounts of NaCladmixed. The electrochemical potential of the custom-made Ag/AgCl sensorwas monitored to indicate the arrival of chlorides, with lower OCP valuescorresponding to higher concentrations of free chloride ions. The OCP of thetop bar in concrete were also monitored to provide additional indication ofchloride arrival and possible corrosion initiation at the top bar.

Fig. 2:. Custom-made chloride sensors (dark color) welded to copper wires(red color), before being embedded in the concrete samples.

2.5. Deicing Ponding Experiments

The Cyclic Exposure Test established and utilized a pressurized transportmethod, which aimed to accelerate the ingress of deicers into concrete andalso to simulate the field scenario involving the traffic flow on pavementsand bridge decks. This method forces the flow of chlorides and corrosioninhibitors (and other additives) into concrete by exposing one face of theconcrete to the deicer solution that is under pressure. All other faces of theconcrete are sealed.




To maintain a constant pressure in the deicer ponds, all specimens wereconnected to an air compressor via a manifold, vinyl tubing, and hose barbs.Figure 3 shows the schematics and experimental setup for the pressurizedtransport of deicers into concrete, with concrete samples attached to thepressure manifold within an environmental chamber, in which airtemperature was precisely regulated and closely monitored. In this study, theapplied pressure was chosen to be maintained at 3 psi during the wet cyclesof the Cyclic Exposure Test. No pressure was applied during the dry cycles,during which we assumed that the externally applied pressure would makelittle difference in the transport of species in concrete. To allow good electriccontact for the monitoring of chloride sensor and top bar, the epoxy coatingfor a small area at the side of each concrete sample was removed in order toplace the wet sponge.

Additionally, the temperature cycles were defined and maintained usingthe environmental chamber. The temperature cycles were designed asfollows: warm (52T) - cold (20°F) - warm (52°F) - hot (84°F) cycling on aweekly basis, intended to simulate the varied seasons in Washington in anaccelerated manner. Along with varying temperatures, the specimensunderwent wet-dry cycles. The deicers were added to each specimen for 3weeks, and then removed for 1 week. Such cycles were repeated untiltermination of the specimen with little interruption. Note that the relativehumidity inside the environmental chamber fluctuated as a function of airtemperature, that is, lower temperature tended to cause lower relativehumidity of the air. For samples subject to temperature and wet/dry cycles, anenvironmental chamber was used for this study, the photo of which ispresented in Figure 4. The designed temperature and wet/dry cyclesexperienced by the reinforced concrete samples during the cyclic exposuretest are provided in Figure 5.

For each concrete sample subjected to either Cyclic Exposure or NaturalDiffusion Test, the custom-made chloride sensor (Ag/AgCl) was embedded 5mm above the top bar. During the deicer ponding experiments, the OCP ofthe top rebar or dowel bar and that of the chloride sensor were bothperiodically measured, using an Ag/AgCl/saturated KC1 reference electrodein a wet sponge placed on the concrete exterior surface adjacent to the topbar. The chloride sensor was used to detect the temporal evolution of freechloride concentration near the top bar, since its potential (£cr) indicates thechloride ion activity at that specific concrete depth. In the early stage ofdeicer ponding experiments, chloride sensor potential and top bar OCP weremeasured for each concrete sample once per week. Then the frequency ofreadings were reduced to one reading every other week for the samples underCyclic Exposure Test and every 4th week for the samples under NaturalDiffusion Test. According to the ASTM C876 guidelines, the probability ofcorrosion initiation is greater than 90% when the OCP of steel in concrete ismore negative than -270 mV relative to the saturated calomel electrode




(SCE), which equals to -235 mV relative to the Ag/AgCl/saturated KC1electrode. This rule, however, may not be applicable for water-saturatedconcrete or dry concrete, where the availability of dissolved oxygen ormoisture can significantly affect the measured OCP of steel in concrete. Thethreshold value will also change as a function of the temperature of serviceenvironment.

EnvironmentalChamberw/ControlledTemperature

Chloride sensorChloride solutionunder pressure

V V » V >;>

Concrete specimen

(a)

(b)

Fig. 3:. (a) Schematics and (b) experimental setup for the pressurizedtransport of deicers into concrete



Volume 28, Nos 3 & 4, 2010 Laboratory Investigation andNeural Networks Modeling

Fig. 4: (a) The environmental chamber used for this study; (b) The reinforcedconcrete samples for the Cyclic Exposure Test on the shelvesinside the chamber.

90

n60

50

II 40

JO

20

10

Ο

Cycle vs Time

17 35 56 77 98 119 140 161 182 203 224 24S 266 287 308 329 350 371

Days Since Average Start Date (7/02/07)

Fig. 5: Temperature and wet/dry cycles experienced by the concretesamples during the Cyclic Exposure Test




For concrete samples subjected to Cyclic Exposure Test, macro-cellcurrent was also monitored to confirm whether any passive-to-activepotential shift was in fact a consequence of top bar corrosion. Two rebars ordowel bars of the same type were embedded in the concrete samples, using1" (25.4 mm) vertical spacing. Periodically, the current between the pair ofbars was monitored with a zero resistance ampermeter. When the chlorideconcentration reached the threshold level at the top bar depth and initiatedactive corrosion, the top bar would feature a lower OCP than the bottom bar.As such, the dynamics of macro-cell current was used as a detection of theonset of active corrosion of the steel, and provided an indirect measure ofcorrosion rate, similar to a previous research /20/ as shown in Figure 6. Forthis study, a macrocell current of 10 μΑ and 40 μΑ was used as the thresholdfor closer monitoring for the rebars and dowel bars respectively, whichtranslated to a current density of 0.19 μΑ/cm2 and 0.09 μΑ/cm2 (by dividingtheir exposed surface area in concrete: 51.5 cm2 and 463.3 cm2) respectively.These thresholds worked relatively well according to the subsequent LP andEIS measurements.

MacroceU-current in μ-AmpereChloride application after 42 dConcrete cover; 5 mm-·- w/c = 0.70— w/c = 0,60

70 $0 90 100Time in days after concreting

Fig. 6: Time-dependent behavior of macrocell current /20/.

Once the chloride sensor detected the arrival of sufficient chlorides nearthe top bar and the OCP data indicated the possible initiation of top barcorrosion, the corrosion rate of the top bar was also periodically measured




using non-destructive, electrochemical techniques. A previous studyindicated that in the mortar when clean reinforcements were passivated, theyexhibited corrosion current density (/corT) values below 0.1 μΑ/cm2, typical ofthe passive state. On the other hand, strongly corroded reinforcementsmaintained /corr values in the region of 10 μΑ/cm2, typical of the active stateI2\l. For this study, a value of 1 μΑ/cm2 was used as the threshold value fordetecting active corrosion of top bar in concrete.

In addition to the Cyclic Exposure Test, some reinforced concretesamples were subjected to the Natural Diffusion Test, featuring airtemperature of 73±3T and relative humidity of 50±5 percent. A portablepotentiostat was used to measure the instantaneous corrosion rate of the topbars in concrete, which was expected to provide more reliable snapshot of thebar conditions than the macro-cell current readings. The approach utilized isdescribed as follows: first, the corrosion rate of each top bar in the concretesamples was measured using the linear polarization (LP) method. For theCyclic Exposure concrete samples, the top bar, the bottom bar andAg/AgCl/saturated KC1 in wet sponge placed on the concrete exterior surfaceadjacent to the top bar served as the working electrode, the counter electrode,and the reference electrode, respectively. For the Natural Diffusion concretesamples, a platinum mesh was placed in the ponding solution to serve as thecounter electrode (in lieu of the bottom bar used in the Cyclic Exposureconcrete samples). The /corr value was then calculated using the measuredlinear polarization resistance T?LP and the bars with /cori of higher than 1μΑ/cm2 were further tested using the electrochemical impedancespectroscopy (EIS) method and the aforementioned three-electrode system.The EIS data were then analyzed using an equivalent circuit (shown in Figure7) to obtain the polarization resistance (R2), which was then used to calculatethe EIS-based corrosion current density, /corr'. The bars with both ;'corTand /con.'of higher than 1 μΑ/cm2 were considered to be truly actively corroding, andthe corresponding concrete samples were terminated for further analyses.

2.6. Artificial Neural Networks

To study the complex cause-and-effect relationships inherent in thedeicer/concrete/bar systems, artificial neural networks (ANNs) was elected asa modeling alternative to establish predictive models correlating potentialinfluential factors (e.g., deicer type, mix design, exposure duration, averagedaily temperature, wet percent time during deicer ponding, average dailychloride concentration of the ponding solution, bar type, and test type) andtarget output factor (Ec\-, OCP of top bar). ANNs are powerful tools to modelthe non-linear cause-and-effect relationships inherent in complex processes1221, as they provide non-parametric, data-driven, self-adaptive approaches toinformation processing. ANNs offer several advantages over traditionalmodel-based methods. First, ANNs are robust and can produce



X.. Shi, el al. Corrosion Reviews

generalizations from experience even if the data are incomplete or noisy,given that over-fitting is avoided with expert intervention. Second, ANNs canlearn from examples and capture subtle functional relationships among casedata. Prior assumptions about the underlying relationships in a particularproblem, which in the real world are usually implicit or complicated, neednot be made. Third, ANNs provide universal approximation functionsflexible in modeling linear and nonlinear relationships. ANNs have beensuccessfully utilized to predict the compressive strength of concrete 123,241,to predict the electrochemical behavior of steel in various chloride solutions7257 and the chloride binding 1261, chloride profiles 7277, and chloridepermeability 7287 in concrete, to recognize the OCP behavioral pattern ofsteel in concrete 7297, and to predict the time to onset of rebar corrosion 7307and the life of concrete structures 7317.

Pore solution/Bulk Concrete

Interface

Concrete/ SteelInterface

Steel/ ElectrolyteInterface

Fig. 7: Equivalent electric circuit used in the EIS data analysis

The ANN paradigm adopted in this study was the multiplayer feed-forward neural network, of which a typical architecture is shown in Figure 8.The nodes in the input and output layers consist of independent variables andresponse variable(s), respectively. One or two hidden layers are included tomodel the dependency based on the complexity of relationship(s). For a feed-forward network, signals are propagated from the input layer through thehidden layer(s) to the output layer, and each node in a layer is connected inthe forward direction to every node in the next layer. Every node simulatesthe function of an artificial neuron. The inputs are linearly summatedutilizing connection weights and bias terms and then transformed via a non-linear transfer function.

In this study, a modified BP algorithm was employed for the ANNtraining, in which a sigmoid function in Equation (1) was used as the




nonlinear transfer function and the sum of the mean squared error (SMSE) inthe output layer as the convergence criteria.

(D

All the data for input and output were normalized based on Equation (2),where X, and NX, are the /4h value of factor X before and after thenormalization, and Xmm and Xmw: are the minimum and maximum value offactor X, respectively.

NX (*,-*·„. +0.1)

Table 4 shows the parameters and performance of the two ANN models inthis study. Note that for modeling purpose, some of the qualitative influentialfactors were given a numerical value for each level as presented in Table 5.When the ANN models were established, all the predictions were made withthese qualitative factors strictly fixed at the levels given in

Exposure Timnoon·wmmmmmAvg. Daily Τ

Fig. 8:. Typical multi-layer feed-forward neural network architectureTable 5, without any attempt for interpolation or extrapolation(which would have been unreasonable). Each ANN model wastrained to allow for a reasonable training error and a reasonabletesting error.



P

ο

<n ,™O) Li-cr:

o

s

8

S

ϊ

Ο)δ

Q.

8

00p

°

συ

&οc

ο.

.gο

s•Sa Ό05 Ob ε

u "·

MSi o

1^Ο Τ3U %£> 3'-· (L>

3 gσ· 5*J s?(Λ β^

«^|1* -33 E• 3 5u j;

a -al§^ sc u

i§"Uo o

cd O.

- 2u i;l §f~ O



Laboratory Investigation andNeural Networks Modeling

οu

«2!s%e<u

(A

*uce«Λ υ^ 5

o·*·»•a0)

X4lΌ

.a>1§·

|Γ*Q·"<3'S*j!§

*S·,§'•5<§saδ

Uj

Q.

H*

COm

c'<oCDQ.X

φCL

^8φQ

<DCL

r-1o<D

^_

Iι- £ro cu ωΐ 8ω υro >,ο χυ οc α.

Ζ) UJι ιI I

τ — ·

•σ^^•σωΕωο.£l1/5£

χ χΈ Έω ωTO TO

^& ^3'C 'Cm m

I IIo coτ-1 0

ro

"g

ro '·£•z. E•r- CN

.y 2>. ·"o g.

l Xτ- LU

1OX

^—211

IO

T30)X'j^

T3roOπ

S?0

•.X

(DO).•gm

lIOh~

CM

0roO"g

|

co

ro3"J51

CNI

XoCL0)

*i

0)JD

W φ0) ^Έ ~S2(n co

ι OCD O

XΈ

ωω>roΟ-

ΙΟΟΟΟοο0

00)

"g

i§4

Q

(Λ!tQ

1•"flr

3K)

1ocC7J

<5<^»— J'5D

i

Ί

cuUcουυ2I'SDc

. ·13LH

^^ dl

ω 3ϋ Λ(Λ υ1 1in Mο c'S "3ra Λ

> Τ3i §'.S vT

I«ω 31 1-Sw_3 _;"υ S.Ξ °Q '""'

^ "I·* Q




All the connection weights and bias terms for nodes in different layerswere initially randomized and then iteratively adjusted based on certainlearning rules. For each given sample, the inputs were forwarded through thenetwork until they reached the output layer producing output values, whichwere then compared with the target values. Errors were computed for theoutput nodes and propagated back to the connections stemming from theinput layer. The weights were systematically modified to reduce the error atthe nodes, first in the output layer and then in the hidden layer(s). Thechanges in weights involved a learning rate and a momentum factor and wereusually in proportion to the negative derivative of the error term. The learningprocess was continued with multiple samples until the prediction errorconverged to an acceptable level.

The number of hidden layers and nodes in the ANN models are generallyrelated to the complexity of the relationship. The more complex therelationship, the more layers and nodes are necessary. Usually one or twohidden layers are enough to approximate the reality. In our work, theselection of layers and nodes took into consideration driving the SMSE assmall as possible and the training process as efficient as possible. From amodeling perspective, artificial neural networks were used to achieve betterunderstanding of the complex cause-and-effect relationships inherent in thedeicer/concrete/bar systems and were successful in finding some meaningful,logical results from the noisy data associated with the deicer pondingexperiments.

3. RESULTS AND DISCUSSION

3.1. ANN Model for the Chloride Sensor Potential

An ANN model was trained and tested to establish the chloride sensorpotential, Ec\-, as a function of eight parameters defining the deicer pondingexperiments', including: Test Type, Deicer Type, Mix Design, ExposureDuration, Daily Average C/~ Concentration, Daily Average Temperature, WetDays/ Total Exposure Days, and Reading Temperature. Then, the trainedmodel was used to predict the dependencies of Ec\- on the Exposure Duration

' In this research, the extremely low EC\- values (e.g., lower than -100 mV vs. Ag/AgCI)were filtered through a data quality check step prior to the modeling process, since they areunlikely attributable to actual increase in chloride concentration in the concrete poresolution. Instead, they may also be the result of a malfunctioning sensor or of the evolutionof surface chemistry in the chloride-sensing layer that defines the electrochemical potentialof the sensor in concrete (e.g., oxidation of AgCl by the hydroxyl ions). This could beaddressed by future research to enhance the reliability and longevity of chloride sensor inconcrete.




and Reading Temperature, with the other six factors assumed at a reasonablelevel. The pattern of such dependencies was used to determine whether theANN model was properly trained.

Figure 9 shows the relationship between experimental and modeled Ec\-values. From the training and testing results, it appears that the establishedANN model has relatively good "memory" and the trained matrices ofinterconnected weights and bias reflect the hidden functional relationshipwell. As such, the ANN model was reasonably suitable for predicting the Ec\-value of unknown samples within the ranges of the modeling data. Once theempirical ANN model was trained and tested, it was used to predict thechloride sensor potential associated with different deicer/concrete scenarios.Figure 10 presents the predicted Ec\- value a function of deicer type, test typeand mix design, with the other input factors fixed as follows: exposureduration at 298 days, daily average Cl" concentration at 1.31 M, both dailyaverage temperature and reading temperature at 72°F, and wet percent time at87% for Cyclic Exposure and at 100% for Natural Diffusion.

Electrochemical Potential of the Chlroide Sensor

!350

300

250

200

150

100

50

0

-50

• Training Dataα Testing Data

Linear (Training Data)

-50 0 50 100 150 200 250 300 350Actual reading (mV, vs. Ag/AgCI)

Fig. 9:. Relationship between experimental and modeled chloridesensor potential

298 days of the Natural Diffusion Test (data shown with dashed lines inFigure 10), the chloride sensor potential in the bridge mix samples withpremixed NaCl or embedded steel strip is predicted to be generally lowerthan that in the bridge mix samples without them whereas the chloride sensorpotential in the pavement mix samples generally fall in between. This holdstrue for concrete samples exposed to all deicers except the inhibited MgCl2,




suggesting the corresponding order of chloride penetration resistance in thesehardened concrete mixes. At 298 days of the Cyclic Exposure Test (datashown with solid lines in Figure 10), the chloride sensor potential in thepavement mix samples is predicted to be significantly lower than that in thebridge mix samples, again confirming the higher chloride penetrationresistance of the latter. For both mixes, the chloride sensor potential data forthe Cyclic Exposure Test are predicted to be significantly lower than thosefor the Natural Diffusion Test, suggesting the acceleration of chloride ingressby temperature and wet/dry cycling. For the bridge mix samples withpremixed NaCl or embedded steel strip, however, such acceleration effect isreversed for unknown reasons.

250 η

οοιI

n2

200 -

150 -

100

50 -

—*-- Cyclic, bridge mixw/steel strip

Λ Natural, bridge mixw/steel strip

—·— Cyclic, bridge mixw/premixed NaCl

- -a- - Natural, bridge mixw/premixed NaCl

—φ— Cyclic, bridge mix

- ·χ· - Natural, bridge mix

—·— Cyclic, pavement mix

- -o- · Natural, pavement mix

NaCl 1-NaCI l-CaCI2

De/eer Typel-MgClj

Fig. 10: Predicted chloride sensor reading as a function of deicer type, testtype and mix design

At In both type of tests, the predicted ECr values suggest that the threeinhibited deicers would lead to higher chloride sensor readings in concrete,i.e., lower free Cl" concentrations at the sensor depth. In other words,corrosion inhibitor (and possibly other additives in the case of the inhibitedCaCl2 and MgCl2 deicers) did slow down the ingress of chloride intoconcrete. This benefit of corrosion inhibitor in deicer held true for allconcrete mixes and was especially significant for concrete samples subjectedto the Cyclic Exposure Test or with premixed NaCl or embedded steel strip.The rate of chloride ingress into concrete was in the order of: non-inhibitedNaCl > inhibited NaCl > inhibited CaCl2 deicer > inhibited MgCl2 deicer.




The established £Cr ANN model was also used to build three-dimensional response surfaces, in order to graphically illustrate thedependencies of £Cr on two selected influential factors, with the other sixfactors assumed at a reasonable level. Note that each presented Ec\- data pointwas an average of eight predicted Ec\- values corresponding to 2 test typesand 4 deicer types.

Figure 11 a presents a predicted response surface illustrating thedependency of Ec\- on exposure duration (100 - 298 days) and wet percenttime (80% - 100%). The predictions were made with the other input factorsfixed as follows: pavement mix, daily average Cl" concentration at 1.31 M,both daily average exposure temperature and reading temperature at 72°F. Itcan be seen that Ec\- tends to decrease drastically as the wet percent timedecreases from 100% to 80%, suggesting that the incorporation of dry cyclesin the deicer ponding experiments accelerates the ingress of chlorides intoconcrete, when the daily average Cl" concentration was maintained constant(i.e., more chlorides are provided in the wet cycles to compensate for the zerosurface Cl" concentration during dry cycles). This is consistent with thegeneral consensus that wet-dry cycles generally accelerate the ingress ofchloride into concrete 732-347. Figure l l a also shows that EC\- tends todecrease as the exposure duration increases from 100 days to 298 days,suggesting continued ingress of more chlorides into concrete.

Figure 1 Ib presents a predicted response surface illustrating thedependency of Ec\- on reading temperature (20 - 72°F) and daily averageexposure temperature (44 - 72°F). The predictions were made with the otherinput factors fixed as follows: pavement mix, exposure duration at 298 days,daily average Cl" concentration at 1.31 M, and wet percent time at 87% forthe Cyclic Exposure Test and at 100% for the Natural Diffusion Test. It canbe seen that Ea- generally increases with the daily average temperature,suggesting the important role of cold temperature (20°F) exposure inaccelerating the chloride ingress into concrete, through freeze-thaw damageof the concrete. The daily average temperature might also have affected theEc\- through the processes altering the physicochemical properties of thechloride-sensing layer. Figure l i b also shows that Ec\- tends to decrease asthe reading temperature increases from 20°F to 72°F, especially when thedaily average exposure temperature is high. This is consistent with theexpected working mechanism of the Ag/AgCl sensor, for which therelationship between electrochemical potential and temperature (along withfree Cl" concentration, [Cl"]) is governed by the Nernst Equation. When thedaily average exposure temperature is low, however, the effect of readingtemperature on the Ec\- value is less significant for unknown reasons. Three-dimensional response surfaces were also established using the ANN modelpredictions for chloride sensor embedded in the bridge mix, which featuredvery similar patterns to those for the pavement mix (as shown in Figure 12).




3.2. ANN Model for the Top Bar OCPAn ANN model was trained and tested to establish the OCP of top bar in

concrete as a function of nine parameters defining the deicer pondingexperiments, including: Test Type, Bar Type, Deicer Type, Mix Design,Exposure Duration, Daily Average Cf Concentration, Daily AverageTemperature, Wet Days/ Total Exposure Days, and Reading Temperature.

Predictedchloride sensorreading, mV (vs.

Ag/AgCI)

1.00,0.95 298

Wet days / totalexposure days

Exposureduration, days

(a)

Predictedchloride sensorreading, mV (vs.

Ag/AgCI)

£- 1 U

-180

-150

-120

m 180-210• 150-180D 120-1 50OJ 90-1 20B 60-90

-90

-60

-30

-o2

Daily averageexposure

temperature, °F

20

Readingtemperature, °F

(b)Fig. 11:. Predicted response surface of chloride sensor reading in pavement

mix as a function of: (a) exposure duration and wet percent time;(b) reading temperature and daily average exposure temperature



Volume 28, Nos. 3&4, 2010 Laboratory Investigation andNeural Networks Modeling

Predictedchloride sensorreading, mV (vs. 21°

Ag/AgCI)


0.89oo

298

Exposureduration, days

(a)

Predicted 210-,

180chloride sensorreading, mV (vs.

Ag/AgCI) 150

Β 180-210

• 150-180

D 120-150

Ξ 90-120

Β 60-90

58

Reading temperature, °F

._ . ·65 Daily average

exposuretemperature,

°F

(b)

Fig. 12:. Predicted response surface of chloride sensor reading in bridge mixas a function of: (a) exposure duration and wet percent time; (b)reading temperature and daily average exposure temperature



X. Shi. et al. Corrosion Reviews

Figure 13 shows the relationship between experimental and modeled topbar OCP values. From the training and testing results, it appears that theestablished ANN model has relatively selective "memory" and the trainedmatrices of interconnected weights and bias aim to capture the hiddenfunctional relationship while filtering the noise in measured data. The low R-square in the ANN model can be attributed to the inherently high variabilityof the measured potential of the top bar, which is not only affected by thebulk properties of the metallic bar and hardened concrete and the poresolution chemistry at the bar surface, but also affected by the surfacecondition of the top bar (e.g., presence of macro- and/or micro-scale defectson the bar surface or its coating layer) and many other factors (e.g., presenceof air voids and aggregates at or near the bar/concrete interface, moistureavailability at the bar surface, Ohmic drop between the top bar and thereference electrode, and probabilistic nature of corrosion process). Tomitigate the noisy data issue, the trained model was used to predict thedependencies of top bar OCP on the Exposure Duration and Daily Average€Γ Concentration, with the other seven factors assumed at a reasonable level.The pattern of such dependencies was used to determine whether the ANNmodel was properly trained. Thereafter, the ANN model was reasonablysuitable for predicting the top bar OCP value of unknown samples within theranges of the modeling data.

οD)

510)

200 -

150 -

100

50

0 -

-50 -

-100 -

-150 -

-200

Open Circuit Potential of the Top Bar

« Training dataα Testing Data

Linear (Training data)

y = 0.7858XR2 = 0.5107

-200 -100 0 100 200Actual reading (mV, vs. Ag/AgCI)

300

Fig. 13:. Relationship between experimental and modeled top bar OCP




Once the empirical ANN model was trained and tested, it was used topredict the top bar OCP associated with different deicer/concrete/barscenarios. Figure 14a and Figure 14b present the predicted top bar OCP as afunction of deicer type, test type and bar type in the bridge and pavementmixes respectively, with the other input factors fixed as follows: exposureduration at 298 days, daily average Cl" concentration at 1.31 M, both dailyaverage temperature and reading temperature at 72°F, and wet percent time at87% for Cyclic Exposure and at 100% for Natural Diffusion.

υσιS 50 -Dl

m 0 -

I 'SO'

§-100-

S -150 -

ί -200 -

-250 -

*"'.·'K:^ ·-···*

NaCI l -NaCI l-CaCI2 l-MgCI2

Deicer Type

» Cyclic, bridge mix w/steel strip, uncoatedrebar

·,·,··· Natural, bridge mix w/steel strip, uncoatedrebar

* Cyclic, bridge mix w/premixed NaCI,uncoated rebar

···»·· Natural, bridge mix w/premixed NaCI,uncoated rebar

• Cyclic, bridge mix,uncoated rebar

·--·-- Natural, bridge mix.uncoated rebar

(a)

100-o>

!«i 50 ->>£a.8 °-sa.ί -.-50 -

100

,--"'-·>'' --•""''ν*" ,-' ' f //</ />// ''""^

,^'"'

NaCI l-NaCI l-CaCI2 l-MgClz

Deicer Type

* Cyclic, pavementmix/epoxy coated dowel

— ώ — Natural, pavementm ix/epoxy coated dowel

* Cyclic, pavementm b(/MMFX dowel

--O-- Natural, pavementm ix/MMFX dowel

• Cyclic, pavementmix/S.S.tube

··-·**· Natural, pavementmix/S.S. tube

(b)

Fig. 14:. Predicted top bar OCP as a function of deicer type, test type andbar type in the (a) bridge mix; (b) pavement mix




Figure 14 illustrates several important points. First of all, the ANN modelpredictions suggest that in both concrete mixes and for all bar types, the threeinhibited deicers would lead to more positive top bar OCP values at 298 days,implying the beneficial role of inhibitors in reducing the pitting risk of bars inconcrete. Second, the predicted OCP of top bar in both mixes generallyfollows the order of: non-inhibited NaCl < or » inhibited NaCl > inhibitedCaCl2 deicer > inhibited MgCl2 deicer. Finally, in most cases, the top barOCP values for the Cyclic Exposure concrete samples were more positivethan those for the Natural Diffusion ones, indicating a more disrupted state ofthe bar surface. This is consistent with the finding that temperature andwet/dry cycling generally accelerates the ingress of chlorides into concrete, asdiscussed earlier (also shown in Figure 10).

The established OCP ANN model was also used to build three-dimensional response surfaces, in order to graphically illustrate thedependencies of top bar OCP on two selected influential factors, with theother seven factors assumed at a reasonable level. Note that each presentedOCP data point was an average of eight predicted OCP values correspondingto 2 test types and 4 deicer types. Figure 15a presents a predicted responsesurface illustrating the dependency of top bar OCP on daily average Cl"concentration (0.35 - 1.83 M) and wet percent time (80% - 100%). Thepredictions were made with the other input factors fixed as follows: epoxycoated dowel bar/pavement mix, exposure duration at 298 days, both dailyaverage exposure temperature and reading temperature at 72°F. It can be seenthat the top bar OCP changes little as the wet percent time decreases from100% to 80%, especially when the daily average Cl" concentration is high.This is likely the result of several mechanisms offsetting each other. Figure15a also shows that the top bar OCP decreases with the increase in the dailyaverage Cl" concentration, confirming the role of chloride in disrupting thepassive state of the bar surface.

Figure 15b presents a predicted response surface illustrating thedependency of top bar OCP on reading temperature (20 - 72°F) and dailyaverage exposure temperature (44 - 72°F). The predictions were made withthe other input factors fixed as follows: epoxy coated dowel bar/pavementmix, exposure duration at 298 days, daily average Cl" concentration at 1.31M, and wet percent time at 87% for the Cyclic Exposure Test and at 100%for the Natural Diffusion Test. It can be seen that top bar OCP generallydecreases with the increase in daily average temperature, likely due to theaccelerating effect of warmer temperature on the disruption of coating orpassive film on the top bar surface. This effect on COP seems to overshadowthe accelerated chloride ingress by concrete freeze-thaw damage at coldertemperatures that we discussed earlier (shown in Figure 1 Ib and Figure 12b).Figure 15b also shows that the top bar OCP has a highly nonlineardependency on the reading temperature, suggesting two or more mechanismsat work.



Volume 28, Nos. 3&4, 2010 Laboratory Investigation andNeural Networks Modeling

Predicted topbar OCR

reading, mV (vs.Ag/AgCI)

Dailyaverage[Gil, M

Wet days/total exposure days

(a)

Predicted topbar OCP

reading, mV(vs.Ag/AgCI)

125

85-

45-

51

-3544

*>T

58Daily average

temperature, °F

Readingtemperature, °F

(b)

Fig. 15:. Predicted response surface of OCP of epoxy coated dowel bar inpavement mix as a function of: (a) daily average Cl- concentrationand wet percent time; (b) reading temperature and daily averageexposure temperature




Three-dimensional response surfaces were also established using theANN model predictions for the top bar (uncoated rebar) embedded in thebridge mix, which featured similar patterns to those for the epoxy coateddowel bar in the pavement mix (as shown in Figure 16).

Predicted topbarOCP

reading, mV(vs.Ag/AgCI)

-60


Daily average[CI1 Μ

(a)

Predicted topbar OCP

reading, mV (vs.Ag/AgCI)

45

-3544

Daily averagetemperature, °F

33— Readingtemperatur

e,°F

(b)

Fig. 16:. Predicted response surface of OCP of uncoated rebar in bridge mixas a function of: (a) daily average Cl- concentration and wetpercent time; (b) reading temperature and daily average exposuretemperature




3.3. ANN Model for the Macrocell Current

Ίη this research, only the Cyclic Exposure concrete samples providedmacrocell current data. An ANN model was trained and tested to establish themacrocell current flowing between top and bottom bars as a function of eightparameters defining the deicer ponding experiments, including: Bar Type,Deicer Type, Mix Design, Exposure Duration, Daily Average CfConcentration, Daily Average Temperature, Wet Days/ Total Exposure Days,and Reading Temperature. Then, the trained model was used to predict thedependencies of macrocell current on the Exposure Duration and DailyAverage Temperature, with the other six factors assumed at a reasonablelevel. The pattern of such dependencies was used to determine whether theANN model was properly trained. Note that an increase in macrocell currentcould result from the active corrosion of top bar, or from reduced concreteresistivity due to salt contamination or concrete microcracking, or both.

Macrocell current flowing between upper and lower bars

70

60

50

40

30

20

10

0

-10

-20

Training Data

Testing DataLinear (Training Data)

y = 0.6549XR2 = 0.6364

-20 -10 0 10 20 30 40 50 60

Actual reading (uA)70

Fig. 17: Relationship between experimental and modeled macrocell current



X.Shi, et αϊ. Corrosion Reviews

Figure 17 shows the relationship between experimental and modeledmacrocell current values. From the training and testing results, it appears thatthe established ANN model has relatively good "memory" and the trainedmatrices of interconnected weights and bias reflect the hidden functionalrelationship well2. As such, the ANN model was reasonably suitable forpredicting the macrocell current of unknown samples within the ranges of themodeling data.

Once the empirical ANN model was trained and tested, it was used topredict the macrocell current associated with different deicer/concrete/barscenarios. Figure 18 presents the predicted macrocell current as a function ofdeicer type, mix design and bar type, with the other input factors fixed asfollows: exposure duration at 254 days, daily average Cl" concentration at1.31 M, both daily average temperature and reading temperature at 52°F, andwet percent time at 87%.

30 ·

— 20

acro

cell

curr

ent (

u

ο

Ε

\

ΛΜ\\\\•:::--,·! ^

"""· -ο

NaCI 1-NaCI l-CaCI2 MMgCI:

Deicer Type

— * — Bndge mix w/crack,uncoated rebar

- · - - Bndge mix w/premixedNaCI, uncoated rebar

« Bridge mix, uncoatedrebar

Δ Road mix, epoxy-coateddowel

- --α - - - Road mix, MMFX dowelbar

β - Road mix, S.S tube

Fig. 18: Predicted macrocell current as a function of deicer type, mix design,and bar type

2 The less-than-ideal R-square in the ANN model can be attributed to the inherently highvariability of the measured macrocell current, which is not only affected by the bulkproperties of the metallic bar and hardened concrete and the pore solution chemistry at thebar surface, but also affected by the surface condition of both top and bottom bars (e.g.,presence of macro- and/or micro-scale defects on the bar surface or its coating layer) andmany other factors (e.g., presence of air voids and aggregates at or near the bar/concreteinterface, moisture availability in concrete, and probabilistic nature of corrosion process).




Figure 18 illustrates several important points. First of all, there wasstatistically significant difference in the macrocell current between the rebarsin bridge mix and the dowel bars in pavement mix. The macrocell currentvalues for the dowel bars were consistently lower than those for the rebars,despite their larger exposed surface area in concrete (463.3 cm2 for dowel barvs. 51.5 cm2 for rebar). As such, under the specific conditions investigated,Figure 18 suggests more active corrosion of top rebars than top dowel bars inconcrete at 254 days. This is likely due to the higher corrosion resistance ofthe dowel bars (epoxy-coated dowel bar, MMFX dowel bar, and stainlesssteel tubes with epoxy-coated insert) than that of the uncoated rebar, and alsoto the fact that the concrete cover over the top dowel bar in the pavement mixwas 2" (50.8 mm) vs. 1.5" (38.1 mm) for the top rebar in the bridge mix.Second, the predicted macrocell current values suggest that the threeinhibited deicers would lead to lower corrosion rates of the top bar. In otherwords, corrosion inhibitor (and possibly other additives in the case of theinhibited CaCl2 and MgCl2 deicers) helped to reduce the corrosion of the topbars in concrete. This benefit is not significant for the dowel bars in thepavement mix since none of them are actively corroding, based on the ANNmodel predictions for the given conditions. Finally, the predicted macrocellcurrent of top rebar in concrete generally follows the order of: non-inhibitedNaCl > inhibited NaCI > inhibited CaCl2 deicer or inhibited MgCl2 deicer.

3.4. ANN Model for the Top Bar Corrosion Rate

An ANN model was trained and tested to establish the corrosion rate oftop bar in concrete (corrosion current density, /'„„, derived from the EISmeasurement3) as a function of nine parameters defining the deicer pondingexperiments, including: Test Type, Bar Type, Deicer Type, Mix Design,Exposure Duration, Daily Average Cf Concentration, Daily AverageTemperature, Wet Days/ Total Exposure Days, and Reading Temperature.Then, the trained model was used to predict the dependencies of /cori on theDaily Average Temperature and Reading Temperature, with the other sevenfactors assumed at a reasonable level. The pattern of such dependencies wasused to determine whether the ANN model was properly trained.

Figure 19 shows the relationship between experimental and modeled topbar /'con. values. From the training and testing results, it appears that the

3 The /'corr values were obtained by dividing the corrosion current values by the estimatedexposed top bar surface area in concrete, which was 51.5 cm2 and 463.3 cm2 for rebar anddowel bar respectively. Note that corrosion potential (£&„) and /'^data were also obtainedfrom the LP measurements of top bar in concrete samples during the deicer pondingexperiments. We elected to choose EIS data over LP data for estimating the corrosion rateof top bar in concrete since they were less affected by the presence of coating layer or thehigh resistivity of the concrete matrix and thus more accurate.




established ANN model has very good "memory" and the trained matrices ofinterconnected weights and bias reflect the hidden functional relationshipwell. As such, the ANN model was reasonably suitable for predicting the topbar ;'COIT value of unknown samples within the ranges of the modeling data.

Corrosion Current Density of Top Bar (Determined by EIS)3.0

• Training dataD Testing data

^— Linear (Training data)

00 05 10 1 5 2 .0Actual reading (μΑ/cm2)

25 30

Fig. 19:. Relationship between experimental and modeled corrosion rate oftop bar in concrete

Once the empirical ANN model was trained and tested, it was used topredict the top bar imn associated with different deicer/concrete/bar scenarios.Figure 20a and Figure 20b present the predicted top bar /COII as a function ofdeicer type, test type and bar type in the bridge and pavement mixesrespectively, with the other input factors fixed as follows: exposure durationat 298 days, daily average Cl" concentration at 1.31 M, both daily averagetemperature and reading temperature at 72°F, and wet percent time at 87%for Cyclic Exposure and at 100% for Natural Diffusion. Figure 20illustrates several important points. First, there was statisticallysignificant difference in the top bar /corr (indicative of its corrosion rate)between the rebars in bridge mix and the dowel bars in pavement mix.The /corr values for the dowel bars were consistently lower than those forthe rebars. For this study, an ;'corr of 1 μΑ/cm2 was used as the thresholdfor the threshold value for detecting active corrosion of top bar inconcrete. As such, under the specific conditions investigated, Figure 20suggests the active corrosion of top rebar at 254 days in the bridge mix



Volume 28, Nos, 3 & 4, 2010 Laboratory Investigation andNeural Networks Modeling

Corro

sion

curre

nt de

nsity

(μΑ

/cm

2) by E

ISO

->

· N

J W

3U

i-

*o

ir

ou

ic

ou

i

d & 6 Λ

" °* \^

* \ * χ\ *\\

\\^\* «

""•-•β

NaCI 1-NaCI l-CaCI2 l-MgCI2

Deicer Type

-*-- Cyclic, bridge mix w/ steelstnp

Λ Natural, bridge mix w/ steelstrip

— Cyclic, bridge mix w/premixed NaCI

π Natural, bridge mix w/premixed NaCI

-»-Cyclic, bndgemix

< Natural, bndgemix

(a)

«2111

ο•ο

I

sυ

0.3 -

0.25 -

0.2 -

0.15 -

0.1 -

0.05 -

Ο

^Α- Cyclic, pavement mix w/epoxy coated dowel

3 Natural, pavement mixw/epoxy coated dowel

-•—Cyclic, pavement mix w/MMFX dowel

α Natural, pavement mixw/MMFX dowel

-·— Cyclic, pavement mix w/S S tube

x Natural, pavement mix w/S S tube

NaCI l-NaCI l-CaCI2 l-MgCI2

Deicer Type

(b)Fig. 20: Predicted top bar icorr as a function of deicer type, test type

and bar type in the (a) bridge mix; (b) pavement mix




Predicted topbar/cornμΑ/cm2

0.0

Wet days /totalexposure days Daily avg. [CI],M

(a)

Predicted topbar/comμΑ/cm2

2.0 r"

0.0282 305 327

Exposure duration, days '·'" 56

(b)Fig. 21: Predicted response surface of /corT of epoxy-coated dowel bar in

pavement mix as a function of: (a) daily average Cl- concentrationand wet percent time; (b) exposure duration and daily averageexposure temperature




exposed to most deicers. For both Cyclic Exposure and NaturalDiffusion tests, on the other hand, all top dowel bars seem to remainpassivated in the pavement mix. This is consistent with the finding fromthe predicted macrocell current data. Second, the predicted /corr valuessuggest that the three inhibited deicers would generally lead to lowercorrosion rates of the top bar in concrete, when the bar is not veryactively corroding (with ;corr lower than 1.5 μΑ/cm2). In most cases,corrosion inhibitor (and possibly other additives in the case of theinhibited CaCl2 and MgCl2 deicers) did reduce the corrosion rate of barsin concrete. The predicted corrosion rate of top bar in concretegenerally follows the order of: non-inhibited NaCl > inhibited NaCl >inhibited CaCl2 deicer > inhibited MgCl2 deicer. Figure 20a also showsa few cases where the inhibitors did not seem to reduce the corrosionrate of uncoated rebar at 254 days, especially in the bridge mixes withpremixed NaCl or embedded steel strip where the bar is activelycorroding (with /corr close to 3 μΑ/cm2).

In light of these findings, we propose that the corrosion inhibitor(and possibly other additives in the case of the inhibited CaCl2 andMgCl2 deicers) can affect the chloride ingress into concrete and thechloride-induced corrosion of rebar or dowel bar in concrete through ahost of mechanisms, either individually, synergistically, or offsettingeach other. Some possible mechanisms that merit further investigationare as follows, inhibitors may: 1) physically or chemically interact withthe concrete matrix and slow down the ingress of chlorides intoconcrete; 2) change the water permeability of concrete; 3) alter thechemistry of concrete pore solution; 4) affect the chloride binding; 5)reduce the electrical conductivity of concrete; 6) mitigate the freeze-thaw damage of concrete by changing the freezing/thawing dynamics;and 7) mitigate the corrosion of rebar or dowel bar in concrete byforming a protective film on the bar surface. While the corrosioninhibitors in deicer products provide some benefits in delaying thecorrosion initiation of rebars and dowel bars in concrete, such benefitsseem to diminish once the active corrosion of the bars is initiated. Inother words, the inhibitors showed little benefits in re-passivating theactively corroding bars in concrete or in stifling the corrosionpropagation.

The established /corr ANN model was also used to build three-dimensional response surfaces, in order to graphically illustrate thedependencies of /corr on two selected influential factors, with the otherseven factors assumed at a reasonable level. Note that each presented/corr data point was an average of eight predicted 7corr valuescorresponding to 2 test types and 4 deicer types divided by the exposedbar surface area in concrete.




Figure 21 a presents a predicted response surface illustrating thedependency of top bar /corr on daily average Cl" concentration (1.07 -1.86 M) and wet percent time (80% - 100%). The predictions were madewith the other input factors fixed as follows: epoxy coated dowelbar/pavement mix, exposure duration at 298 days, both daily averageexposure temperature and reading temperature at 72°F. In general, thecorrosion rate of epoxy coated rebar in concrete increases with the dailyaverage Cl" concentration and slightly increases with the decrease in wetpercent time (especially at high Cl" concentrations). At 298 days, most'corr values are predicted to be below 1 μΑ/cm2 for the epoxy coateddowel bar.

Figure 21b presents a predicted response surface illustrating thedependency of top bar /corr on exposure duration (282 - 372 days) anddaily average exposure temperature (51 - 72°F). The predictions weremade with the other input factors fixed as follows: epoxy coated dowelbar/pavement mix, reading temperature at 72°F, daily average Cl"concentration at 1.31 M, and wet percent time at 87% for the CyclicExposure Test and at 100% for the Natural Diffusion Test. There was noactive corrosion of the epoxy coated dowel bar until day 350 or so,where the top bar corrosion rate started to show significant increaseswith the daily average temperature (from 51°F to 72°F) and theexposure duration. This is reasonable since in this temperature rangethere is no freeze-thaw damage and higher temperature contributes tohigh reaction rates of concrete deterioration and steel corrosion.

Three-dimensional response surfaces were also established using theANN model predictions for Stainless Steel tube (with epoxy coatedinsert) embedded in the pavement mix (as shown in Figure 22), whichfeatured very similar patterns to those for the epoxy coated dowel bar(as shown in Figure 21) except significantly lower ;corr values. Responsesurfaces were also established using the ANN model predictions for theMMFX dowel bar in the pavement mix, which also featured very similarpatterns to those in Figures 21 with /'corr values between the other twodowel bars.

Figure 23a presents a predicted response surface illustrating thedependency of top bar ;corr on daily average Cl" concentration (1.07 -1.86 M) and wet percent time (80% - 100%). The predictions were madewith the other input factors fixed as follows: uncoated rebar in thebridge mix (in the absence or presence of embedded steel strip oradmixed 0.1% NaCl), exposure duration at 298 days, both daily averageexposure temperature and reading temperature at 72°F. The uncoatedrebar in the bridge mix featured /corr values generally higher than 1μΑ/cm2 (indicative active corrosion of the top bar), which have a highlynon-linear dependency on the daily average chloride concentration andare not significantly affected by the change in wet percent time.



Volume 28, Nos. 3 & 4. 2010 Laboratory Investigation andNeural Networks Modeling

2.0 ιPredicted top

bar/comμΑ/cm2

0.0

Wet days /totalexposure days

D 1.5-2.0α 1.0-1.5• 0.5-1.0m 0.0-0.5

Daily avg. [CI-],M

(a)

Predicted topbar/cort,μΑ/cm2

0.0282 305

Exposure duration, days327 Daily avg. temp., °F

(b)

Fig. 22: Predicted response surface of /C0rr of stainless steel tube in pavementmix as a function of: (a) daily average Cl- concentration and wetpercent time; (b) exposure duration and daily average exposuretemperature




3.0Predicted top

baricorn 2.5μΑ/cm2

02.0-2.5α 1.5-2.0D 1.0-1.5• 0.5-1.0m 0.0-0.5

0.0

Wet days / totalexposure days Daily avg. [Ch],M

(a)

Predicted to pbar/comμΑ/cm2

0.0282 305 327

Exposure duration, days 35° 372 56 Daily avg. temp., °FΟΊ

(b)

Fig. 23: Predicted response surface of /con· of uncoated rebar in bridge mix asa function of: (a) daily average Cl- concentration and wet percenttime; (b) exposure duration and daily average exposure temperature




3.0Predicted top

baricom 2.5μΑ/cm2

2.0

B 2.0-2.501.5-2.001.0-1.5• 0.5-1.0• 0.0-0.5

0.0

Wet days / totalexposure days Daily avg. [CI-],M

(a)

Predicted topbaricomμΑ/cm2

0.0

282 305Exposure duration, days

6772

Daily avg. temp., "F

(b)

Fig. 24: Predicted response surface of /con· of uncoated rebar in bridge mixwith embedded steel strip, as a function of: (a) daily average Cl-concentration and wet percent time; (b) exposure duration anddaily average exposure temperature




Figure 23b presents a predicted response surface illustrating thedependency of top bar /'corr on exposure duration (282 - 372 days) anddaily average exposure temperature (51 - 72°F). The predictions weremade with the other input factors fixed as follows: uncoated rebar in thebridge mix (in the absence or presence of embedded steel strip oradmixed 0.1% NaCl), reading temperature at 72°F, daily average Cl"concentration at 1.31 M, and wet percent time at 87% for the CyclicExposure Test and at 100% for the Natural Diffusion Test. Thecorrosion rate of top dowel bar increases significantly as the dailyaverage temperature increases from 51°F to 72°F, confirming theimportant role of temperature in accelerating metallic corrosion inconcrete. The corrosion rate of dowel bar shows a non-linear behavioras the exposure duration increases from 282 days to 372 days, possiblyreflecting the formation and subsequent disruption of a compact rustlayer as corrosion progresses.

Three-dimensional response surfaces were also established using theANN model predictions for the uncoated rebar in bridge mix withembedded steel strip (as shown in Figure 24), which featured verysimilar patterns to those without the steel strip (as shown in Figure 23)but slightly higher /corr values. Response surfaces were also establishedusing the ANN model predictions for the uncoated rebar in bridge mixwith admixed 0.1% NaCl, which also featured very similar patterns and'corr values to those in Figure 23 and Figure 24.

Note that the presence of air voids and aggregates at or near thebar/concrete interface, the presence of macro- and/or micro-scaledefects on the bar surface or its coating layer, and the probabilisticnature of corrosion process could have all contributed to the variabilityin the top bar ;'corr readings obtained in this study.

4. CONCLUSIONS

Prior to this research, little was known about the possible contribution orrisk deicer inhibitors (and other additives) pose to infrastructure preservation.Therefore, research was needed to determine whether or not corrosion-inhibited deicers provide benefits in mitigating the corrosion of rebars ordowel bars in concrete, relative to the "straight salt". To this end, extensivelaboratory tests were conducted to assess the effect of chloride-based deicerson reinforced concrete structures such as roadways and bridges operated byWSDOT and to determine whether or not reducing deicer corrosiveness helpspreserve the transportation infrastructure. The key findings and conclusionsare presented as follows:




1. Great variability was observed in the numerous data obtained fromthe periodical monitoring of chloride sensor potential, top barpotential, macrocell current, and top bar corrosion rate during thedeicer ponding experiments. As such, two ANN-based predictivemodels were established to filter the noises from the signals in thedata and to identify meaningful patterns. From the training andtesting results, it appears that the established ANN models haverelatively good or selective "memory" and the trained matrices ofinterconnected weights and bias reflect the hidden functionalrelationship well.

2. In both the Cyclic Exposure and the Natural Diffusion tests, theANN model predictions suggest that the three inhibited deicerswould lead to higher chloride sensor readings in concrete, i.e., lowerfree Cl" concentrations at the sensor depth. In other words, corrosioninhibitor (and possibly other additives in the case of the inhibitedCaCl2 and MgCl2 deicers) did slow down the ingress of chloride intoconcrete. This benefit of corrosion inhibitor in deicer held true forall concrete mixes and was especially significant for concretesamples subjected to the Cyclic Exposure Test or with premixedNaCl or embedded steel strip. The rate of chloride ingress intoconcrete was in the order of: non-inhibited NaCl > inhibited NaCl >inhibited CaCl2 deicer > inhibited MgCl2 deicer.

3. The ANN model predictions suggest that in both concrete mixes andfor all bar types, the three inhibited deicers would lead to morepositive top bar OCP values at 298 days, implying the beneficialrole of inhibitors in reducing the pitting risk of bars in concrete. Thepredicted OCP of top bar in both mixes generally follows the orderof: non-inhibited NaCl < or » inhibited NaCl > inhibited CaCl2deicer > inhibited MgCl2 deicer. In most cases, the top bar OCPvalues for the Cyclic Exposure concrete samples were more positivethan those for the Natural Diffusion ones, indicating a moredisrupted state of the bar surface.

4. According to the ANN modeling, the dowel bars show consistentlylower macrocell current values at 254 days than the rebars, despitetheir larger exposed surface area in concrete. This is likely due to thehigher corrosion resistance of the dowel bars (epoxy-coated dowelbar, MMFX dowel bar, and stainless steel tubes with epoxy-coatedinsert) than that of the uncoated rebar, and also to the fact that theconcrete cover over the top dowel bar in the pavement mix was 2"(50.8 mm) vs. 1.5" (38.1 mm) for the top rebar in the bridge mix.The predicted macrocell current values suggest that the threeinhibited deicers would lead to lower corrosion rates of the top bar.




The predicted macrocell current of top rebar in concrete generallyfollows the order of: non-inhibited NaCl > inhibited NaCl >inhibited CaCl2 deicer or inhibited MgCl2 deicer.

5. The ANN modeling shows active corrosion of top rebar at 254 daysin the bridge mix exposed to most deicers, whereas all top dowelbars seem to remain passivated in the pavement mix. The predicted/cori values suggest that the three inhibited deicers would generallylead to lower corrosion rates of the top bar in concrete, when the baris not very actively corroding (with /corr lower than 1.5 μΑ/cm2). Thepredicted corrosion rate of top bar in concrete generally follows theorder of: non-inhibited NaCl > inhibited NaCl > inhibited CaCl2deicer > inhibited MgCl2 deicer.

6. While the corrosion inhibitors in deicer products provide somebenefits in delaying the corrosion initiation of rebars and dowel barsin concrete, such benefits seem to diminish once the active corrosionof the bars is initiated. In other words, the inhibitors showed littlebenefits in re-passivating the actively corroding bars in concrete orin stifling the corrosion propagation.

ACKNOWLEDGEMENTS

The authors acknowledge the financial support provided by theWashington State Department of Transportation (WSDOT) as well as theResearch & Innovative Technology Administration (RITA) at the U.S.Department of Transportation for this project. The authors would like tothank the WSDOT Research Manager Kim Willoughby and the technicalpanel consisting of Rico Baroga, Tom Root, DeWayne Wilson, Linda M.Pierce, Jeff S. Uhlmeyer, and Douglas Pierce, for providing the continuedsupport throughout this project. We owe our thanks to the Stoneway ConcreteInc. for assisting with the fabrication and quality assurance of all the concretespecimens used in this study. We appreciate the following professionals whoprovided assistance to this research: Jim Weston (WSDOT), Mary Gilmore(WSDOT), and Mark Peterson (Montana Department of Transportation).Finally, we owe our thanks to the following individuals at the WesternTransportation Institute for providing help in various stages of the laboratoryinvestigation: Laura Fay, Dr. Tongyan Pan, Steven Anderson, Eric Schon,and Tristan J. Dunlap.




REFERENCES:

/!/ L. Fay, K. Volkening, C. Gallaway, and X. Shi, in Proceedings (DVD-ROM) of the 87th Annual Meeting of Transportation Research Board(held in Washington D.C., January 2008), eds. Transportation ResearchBoard, (2008), Paper No. 08-1382.

Ill Levelton Consultants Ltd., Guidelines for the selection of snow and icecontrol materials to mitigate environmental impacts. NCHRP Report577. National Cooperative Highway Research Program, NationalResearch Council, Washington, D.C. http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_rpt_577.pdf, accessed in July 2008.

/3/ H. Lee, R.D. Cody, A.M. Cody, and P.O. Spry, in Mid-ContinentTransportation Symposium Proceedings 2000 (held in Ames, Iowa,May 15-16, 2000), eds. Center for Transportation Research andEducation, Iowa State University. (2000), 151-155.

/4/ J. Deja and G. Loj, in Proceedings: Infrastructure Regeneration andRehabilitation Improving the Quality of Life through BetterConstruction - A Vision for the Next Millennium (held in Sheffield, 28June - 2 July 1999), eds. R. Narayan Swamy. Sheffield Academic Press,Sheffield, ROYAUME-UNI (1999), 603-620.

151 R.D. Cody, A.M. Cody, P.G. Spry, and G.-L. Gan, Environm. Eng.Geosci. 2(4) (1996), R.D. Cody, P.G. Spry, A.M. Cody, and G.-L. Gan,The role of magnesium in concrete deterioration. The Iowa HighwayResearch Board, Final Report - Iowa DOT HR-355, 1994.

161 L. Sutler, K. Peterson, G. Julio-Betancourt, D. Hooton, T. Vam Dam,and K. Smith, The deleterious chemical effects of concentrated deicingsolutions on Portland cement concrete. Final report for the SouthDakota Department of Transportation, 2008.

Ill S. Monosi and M.S. Collepardi, // Cemento 87(1-3) (1990), 3-8./8/ S. Chatterji, N. Thaulow, and A.D. Jensen, Cem. Cone. Res. 17(5)

(1987), 777-783.191 P.J. Nixon, C.L. Page, I. Canham, and R. Bollinghaus, Adv. Cement

Res. 1(1) (1988), 99-105./10/ M. Kawamura, K. Takemoto, and M. Ichise, in Proceedings of the 8th

International Conference on AAR, (held in Kyoto, Japan, August 1989),eds. K. Okada, S. Nishibayashi and M. Kawamura. Society of MaterialsSciences, (1989), 115-120.

/I I/ M. Kawamura, K. Takeuchi, and A. Sugiyama, Cem. Cone. Res. 24(4)(1994), 621-632.

/12/ M. Kawamura and K. Takeuchi, Cem. Cone. Res. 26(12) (1996), 1809-1819.



X. Shi, et al. Corrosion Reviews

/13/ R.G. Sibbick, "The Susceptibility of Various UK Aggregates to Alkali-Silica Reaction", Ph.D. Dissertation, Aston University, Birmingham,UK, 1993.

/14/ R.G. Sibbick and C.L. Page, in Proceedings of the 10th InternationalConference on AAR (held in Melbourne, Australia, 1996), eds. A.Shayan. (1996), 822-829.

/15/ Salt Institute, http://www.saltinstitute.org/33, last accessed on October22, 2008.

/16/ D. Vitaliano, J. Policy Anal. Mgmt. 11(3) (1992), 397-418.l\ll Transportation Research Board, Highway de-icing: comparing salt and

calcium magnesium acetate. National Research Council. Special Report235, 1991.

/18/ X. Shi, L. Fay, Z. Yang, T.A. Nguyen, and Y. Liu, Corros. Rev. 27(1-2)(2009), 23-52.

/19/ M.Yunovich, N.G. Thompson, T. Balvanyos, and L. Lave, Corrosioncosts of highway bridges, 2002. http://www.corrosioncost.com/pdf/-highway.pdf, accessed in July 2007.

/20/ M. Raupach and P. Schiessl. Constr. Bldg. Mater. 11(4) (1997), 207-214.

/21/ J.A. Gonzalez, A. Cobo, M.N. Gonzalez, and E. Otero, Mater. Corro.51(2000), 97-103.

/22/ X. Shi, P. Schillings, and D. Boyd, Inter. J. Prod. Res. 42(1) (2004),101-118.

/23/ L.-C. Yeah, Cent. Con. Res. 28(12) (1998), 1797-1808./24/ L.-C. Yeah, J. Mat. in Civ. Engrg. 18(4) (2006), 597-604./25/ A. M. Nor and R.A. Cottis, in Proceedings of CORROSION/2004

Symposium (held in New Orleans, LA, USA. March 23-April 1), eds.National Association of Corrosion Engineers, (2004), Paper No. 04063.

/26/ G. K. Glass, N. M. Hassanein, and N. R. Buenfeld, Magazine ofConcrete Research 49(181) (1997), 323-335.

/27/ J. Peng, Z. Li, and B. Ma, J. Mat. in Civ. Engrg. 14(4) (2002), 327-333./28/ E. Güneyisi, M. Gesoglu,T. Özturan, and E. Özbay, Constr. Bldg,

Maier 23 (2009), 469-481./29/ T. Parthiban, R. Ravi, G.T. Parthiban, S. Srinivasan, K.R.

Ramakrishnan, and M. Raghavan, Corro. Sei. 47 (2005), 1625-1642./30/ G. Morcous and Z. Lounis, Comp. Aid. Civ. Infra. Engrg. 20(2) (2005),

108-117./31/ N. R. Buenfeld and N. M. Hassanein, Proc. Instn. Civ. Engrs Structs &

Bldgs 128 (1998), 38-48./32/ C.L. Page, P. Lambert, and P.R.W. Vassie, Mater. Struct. 24(4) (1991),

243-252./33/ K. Hong and R.D Hooton, Cem. Cone. Res. 29(9) (1999), 1379-1386./34/ A. Boddy, E. Bentz, M.D.A. Thomas, and R.D Hooton, Cem. Cone.

Res. 29(6) (1999), 827-837.





Date post:	28-Jan-2023
Category:	Documents
Upload:	wsu
View:	0 times
Download:	0 times

LABORATORY INVESTIGATION AND NEURAL NETWORKS MODELING OF DEICER INGRESS INTO PORTLAND CEMENT...

Documents