Real-Time Corrosion Monitoring of Aircraft Structures with ... · corrosion monitoring (Harris,...

Real-Time Corrosion Monitoring of Aircraft Structures withPrognostic Applications

Douglas Brown1, Duane Darr2, Jefferey Morse3, and Bernard Laskowski4

1,2,3,4 Analatom, Inc., 562 E. Weddell Dr. Suite 4, Sunnyvale, CA 94089-2108, [email protected]@[email protected]

[email protected]

ABSTRACT

This paper presents the theory and experimental validation ofa Structural Health Management (SHM) system for monitor-ing corrosion. Corrosion measurements are acquired usinga micro-sized Linear Polarization Resistance (µLPR) sensor.The µLPR sensor is based on conventional macro-sized Lin-ear Polarization Resistance (LPR) sensors with the additionalbenefit of a reduced form factor making it a viable and eco-nomical candidate for remote corrosion monitoring of highvalue structures, such as buildings, bridges, or aircraft.

A series of experiments were conducted to evaluate the µLPRsensor for AA 7075-T6, a common alloy used in aircraftstructures. Twelve corrosion coupons were placed alongsidetwenty-four µLPR sensors in a series of accelerated tests.LPR measurements were sampled once per minute and con-verted to a corrosion rate using the algorithms presented inthis paper. At the end of the experiment, pit-depth due to cor-rosion was computed from each µLPR sensor and comparedwith the control coupons.

The paper concludes with a feasibility study for the µLPRsensor in prognostic applications. Simultaneous evaluationof twenty-four µLPR sensors provided a stochastic data setappropriate for prognostics. RUL estimates were computeda-posteriori for three separate failure thresholds. The resultsdemonstrate the effectiveness of the sensor as an efficient andpractical approach to measuring pit-depth for aircraft struc-tures, such as AA 7075-T6, and provide feasibility for its usein prognostic applications.

1. INTRODUCTION

Recent studies have exposed the generally poor state of ournation’s critical infrastructure systems that has resulted from

Douglas Brown et al. This is an open-access article distributed under theterms of the Creative Commons Attribution 3.0 United States License, whichpermits unrestricted use, distribution, and reproduction in any medium, pro-vided the original author and source are credited.

Figure 1. Embedded SHM system installed in the rear fuel-bay bulkhead of a commercial aircraft.

wear and tear under excessive operational loads and environ-mental conditions. SHM (Structural Health Monitoring) Sys-tems aim at reducing the cost of maintaining high value struc-tures by moving from SBM (Scheduled Based Maintenance)to CBM (Condition Based Maintenance) schemes (Huston,2010). These systems must be low-cost, simple to install witha user interface designed to be easy to operate. To reduce thecost and complexity of such a system a generic interface nodethat uses low-powered wireless communications has been de-veloped. This node can communicate with a myriad of com-mon sensors used in SHM. In this manner a structure such asa bridge, aircraft or ship can be fitted with sensors in any de-sired or designated location and format without the need forcommunications and power lines that are inherently expen-sive and complex to route. Data from these nodes is trans-mitted to a central communications Personal Computer (PC)for data analysis. An example of this is provided in Figure 1showing an embedded SHM system installed in the rear fuel-bay bulkhead of a commercial aircraft.

Corrosion sensors can be distinguished by the following cat-egories, direct or indirect and intrusive or non-intrusive. Di-rect corrosion monitoring measures a response signal, such

1

Annual Conference of the Prognostics and Health Management Society 2012

as a current or voltage, as direct result of corrosion. Ex-amples of common direct corrosion monitoring techniquesare: corrosion coupons, Electrical Resistance (ER), Electro-Impedance Spectroscopy (EIS) and Linear Polarization Re-sistance (LPR) techniques. Whereas, indirect corrosion mon-itoring techniques measure an outcome of the corrosion pro-cess. Two of the most common indirect techniques are ul-trasonic testing and radiography. An intrusive measurementrequires access to the structure. Corrosion coupons, ER, EISand LPR probes are intrusive since they have to access thestructure. Non-intrusive techniques include ultrasonic testingand radiography.

Each of these methods have advantages and disadvantages.Corrosion coupons provide the most reliable physical evi-dence possible. Unfortunately, coupons usually require sig-nificant time in terms of labor and they provide time aver-aged data that can not be utilized for real time or on-linecorrosion monitoring (Harris, Mishon, & Hebbron, 2006).ER probes provide a basic measurement of metal loss, butunlike coupons, the value of metal loss can be measured atany time, as frequently as required, while the probe is in-situ and permanently exposed to the structure. The disad-vantage is ER probes require calibration with material prop-erties of the structure to be monitored. The advantage of theLPR technique is that the measurement of corrosion rate ismade instantaneously. This is a more powerful tool than ei-ther coupons or ER where the fundamental measurement ismetal loss and where some period of exposure is required todetermine corrosion rate. The disadvantage to the LPR tech-nique is that it can only be successfully performed in rela-tively clean aqueous electrolytic environments (Introductionto Corrosion Monitoring, 2012). EIS is a very powerful tech-nique that provides both kinetic (corrosion rate) and mecha-nistic information. The main disadvantages associated withthe use of EIS are that the instrumentation is sophisticatedand sometimes difficult to use in the field due to the lengthof time required for each frequency sweep. Additionally, in-terpretation of the data can be difficult (Buchheit, Hinkebein,Maestas, & Montes, 1998). Finally, ultrasonic testing andradiography can be used to detect and measure (depth) ofcorrosion through non-destructive and non-intrusive means(Twomey, 1997). The disadvantage with the ultrasonic test-ing and radiography equipment is the same with corrosioncoupons, both require significant time in terms of labor andcan not be utilized for real time or on-line corrosion monitor-ing.

The µLPR presented in this paper improves on existing LPRtechnology by reducing the form-factor of the sensor to thesize of a United States postage stamp, shown in Figure 2,to enable real-time sensing of remote and hard-to-access ar-eas where conventional sensors cannot due to limitations insize and form-factor. Further improvements are realized bynarrowing the separation distance between electrodes, which

(a)

(b)

Figure 2. One-to-one scaled comparison of (a) the µLPR sen-sor attached to a flexcable and (b) a United States postagestamp.

minimizing the effects due to solution resistance. This en-ables the µLPR to operate outside a controlled aqueous envi-ronment, such as an electro-chemical cell, in a broad range offielded applications (eg. civil engineering, aerospace, petro-chemical).

The remainder of the paper is organized by the following.Section 2 describes the general theory governing LPR. Sec-tion 3 presents the µLPR discussing the benefits of miniatur-izing the sensor from a macro-scaled LPR. Section 4 outlinesthe experimental setup and procedure used to validate theµLPR sensor. Section 5 presents the experimental measure-ments with the accompanying analysis which demonstratesthe effectiveness of the µLPR sensor. Section 6 evaluatesthe feasibility for using the µLPR sensor in prognostic ap-plications. Finally, the paper is concluded in Section 7 with asummary of the findings and future work.

2. LPR THEORY

The corrosion of metals takes place when the metal dissolvesdue to oxidation and reduction (electrochemical) reactionsat the interface of metal and the (aqueous) electrolyte so-lution. Atmospheric water vapor is an example of an elec-trolyte that corrodes exposed metal surface and wet concreteis another example of an electrolyte that can cause corrosionof reinforced rods in bridges. Corrosion usually proceedsthrough a combination of electrochemical reactions; (1) an-odic (oxidation) reactions involving dissolution of metals inthe electrolyte and release of electrons, and (2) cathodic (re-duction) reactions involving gain of electrons by the elec-trolyte species like atmospheric oxygen O2, moisture H2O,or H+ ions in an acid (Bockris, Reddy, & Gambola-Aldeco,2000). The flow of electrons from the anodic reaction sites tothe cathodic reaction sites constitutes corrosion current andis used to estimate the corrosion rate. When the two reac-tions are in equilibrium at the equilibrium corrosion poten-

2


tial, Ecorr, the net current on the metal surface is zero with-out an external source of current. The anodic reactions pro-ceed more rapidly at more positive potentials and the cathodicreactions proceed more rapidly at more negative potentials.Since the corrosion current from the unstable anodic and ca-thodic sites is too small to measure, an external activationpotential is applied across the metal surface and the currentis measured for electrochemical calculations. The resultingEa vs. Ia curve is called the polarization curve. Under exter-nal activation potential, the anodic and cathodic currents in-crease exponentially and so when log10 Ia is plotted againstEa (a Tafel Plot), the linear regions on the anodic and ca-thodic curves correspond to regions where either the anodicor cathodic reactions dominate and represent the rate of theelectrochemical process. The extrapolation of the Tafel linearregions to the corrosion potential gives the corrosion current,Icorr, which is then used to calculate the rate of corrosion(Burstein, 2005).

2.1. Anodic and Cathodic Reactions

Electrochemical technique of Linear Polarization Resistance(LPR) is used to study corrosion processes since the corrosionreactions are electrochemical reactions occurring on the metalsurface. Modern corrosion studies are based on the conceptof mixed potential theory postulated by Wagner and Traud,which states that the net corrosion reaction is the result oftwo or more partial electrochemical reactions that proceed in-dependently of each other (Wagner & Traud, 1938). For thecase of metallic corrosion in presence of an aqueous medium,the corrosion process can be written as,

M + zH2Of←→b

Mz+ +z

2H2 + zOH−, (1)

where z is the number of electrons lost per atom of the metal.This reaction is the result of an anodic (oxidation) reaction,

Mf←→b

Mz+ + ze−, (2)

and a cathodic (reduction) reaction,

zH2O + ze−f←→b

z

2H2 + zOH−. (3)

It is assumed that the anodic and cathodic reactions occurat a number of sites on a metal surface and that these siteschange in a dynamic statistical distribution with respect tolocation and time. Thus, during corrosion of a metal surface,metal ions are formed at anodic sites with the loss of electronsand these electrons are then consumed by water molecules toform hydrogen molecules. The interaction between the an-odic and cathodic sites as described on the basis of mixedpotential theory is represented by well-known relationshipsusing current (reaction rate) and potential (driving force). Forthe above pair of electrochemical reactions (anodic (2) andcathodic (3)), the relationship between the applied current Ia

and potential Ea follows the Butler-Volmer equation,

Ia = Icorr

exp

[2.303 (Ea − Ecorr)

βa

]− . . .

exp

[−2.303 (Ea − Ecorr)

βc

], (4)

where βa and βc are the anodic and cathodic Tafel parametersgiven by the slopes of the polarization curves ∂Ea/∂ log10 Iain the anodic and cathodic Tafel regimes, respectively andEcorr is the corrosion potential (Bockris et al., 2000).

2.2. Electrode Configuration

An electrode is a (semi-)conductive solid that interfaces withan electrolytic solution. The most common electrode con-figuration is the three-electrode configuration. The commondesignations are: working, reference and counter electrodes.The working electrode is the designation for the electrode be-ing studied. In corrosion experiments, this is the material thatis corroding. The counter electrode is the electrode that com-pletes the current path. All electrochemistry experiments con-tain a working–counter pair. In most experiments the counterelectrode is simply the current source/sink comprised of in-ert materials like graphite or platinum. Finally, the referenceelectrode serves as an experimental reference point, specifi-cally for potential (sense) measurements. The reference elec-trode is positioned so that it measures a point very close tothe working electrode.

The three-electrode setup has a distinct experimental advan-tage over a two electrode setup: only one half of the cell ismeasured. That is, potential changes of the working elec-trode are measured independently of changes that may occurat the counter electrode. This configuration also reduces theeffect of measuring potential drops across the solution resis-tance when measuring between the working and counter elec-trodes.

2.3. Polarization Resistance

The corrosion current, Icorr, cannot be measured directly.However, a-priori knowledge of βa and βc along with a smallsignal analysis technique, known as polarization resistance,can be used to indirectly compute Icorr. The polarization re-sistance technique, also referred to as “linear polarization”,is an experimental electrochemical technique that estimatesthe small signal changes in Ia when Ea is perturbed byEcorr±10 mV (G102, 1994). The slope of the resulting curveover this range is the polarization resistance,

Rp ,∂Ea

∂Ia

∣∣∣∣|Ea−Ecorr|≤10 mV

. (5)

Note, the applied current, Ia, is the total applied current andis not multiplied by the electrode area so Rp as defined in (5)

3


has units of Ω. Provided that |Ea − Ecorr| /βa ≤ 0.1 and|Ea − Ecorr| /βc ≤ 0.1, the first order Taylor series expan-sion exp (x) u 1 + x can be applied to (4) and (5) to arriveat,

Rp =1

2.303Icorr

(βaβcβa + βc

). (6)

Finally, this expression can be re-written for Icorr to arrive atthe Stern-Geary equation,

Icorr =B

Rp, (7)

where B = 12.303 [βaβc/ (βa + βc)] is a constant of propor-

tionality.

2.4. Pit-Depth

The pit-depth due to corrosion is calculated by computing thepitting current density, ipit,

ipit (t) =icorr − ipvNpit

, (8)

where icorr = Icorr/Asen is the corrosion current density,ipv is the passive current density, Npit is the pit density forthe alloy (derived empirically) and Asen is the effective sur-face area of the LPR sensor. One critical assumption is thepH is in the range of 6-8. If this cannot be assumed, then ameasurement of pH is required and ipassive is needed overthe range of pH values. Next, Faraday’s law is used to re-late the total pitting charge with respect to molar mass loss.Let the equivalent weight (EW ) represent the weight of themetal that reacts with 1 C of charge, thus contributing to thecorrosion and overall loss of material in the anodic (oxida-tion) reaction given in (2). The total pitting charge, Qcorr,and molar mass loss, M , can be related to the following,

Qpit (t) = zF ·M (t) , (9)

where F = 9.650 × 104 C/mol is Faraday’s constant, and zis the number of electrons lost per atom in the metal in thereduction-oxidation reaction. The EW is calculated from theknown Atomic Weight (AW ) of the metal,

EW =AW

z. (10)

Next, the number of moles of the metal reacting can be con-verted to an equivalent mass loss, mloss,

mloss (t) = M (t) ·AW. (11)

Combining (9) through (11), the mass loss mloss is related toQpit by,

mloss (t) =EW ·Qpit (t)

F. (12)

With the mass loss calculated and knowing the density ρ,the pit-depth modeled using a semi-spherical volume with adepth (or radius) d is expressed as,

d (t) =

(3mloss (t)

2πρ

)1/3

. (13)

Now, note that Qpit can be found by integrating ipit over thetotal time,

Qpit (t) =

ˆ t

0

ipit (τ) dτ, (14)

Substituting (12) and (14) into (13) gives,

d (t) =3

√3EW

2πρF

ˆ t

0

ipit (τ) dτ. (15)

Next, by substituting (7) and (8) into (15), the expression ford can be rewritten as,

d (t) = 3

√3EW

2πρNpitF

ˆ t

0

(B

AsenRp (τ)− ipv

)dτ. (16)

In practice, Rp is not measured continuously, rather, periodicmeasurements are taken every Ts seconds. If its assumed overthis interval the Rp values changes linearly then the meanvalue theorem for integrals can be applied to arrive at an al-ternative expression for d,

d (t) =3

√3TsEW

2πρNpitF

N−1

Σk=0

(B

AsenRp (kTs)− ipv

). (17)

2.5. Standard Measurements

2.5.1. Polarization Resistance

ASTM standard G59 outlines procedures for measuring po-larization resistance. Potentiodynamic, potential step, andcurrent-step methods can be used to computeRp (G59, 1994).The potentiodynamic sweep method is the most commonmethod for measuring Rp. A potentiodynamic sweep is con-ducted by applyingEa betweenEcorr±10 mV at a slow scanrate, typically 0.125 mV/s. A linear fit of the resulting Ea vs.Ia curve is used to compute Rp.

2.5.2. Tafel Coefficients

ASTM standard G59 outlines the procedure for measuringthe Tafel slopes, βa and βc (G59, 1994). First, Ecorr is mea-sured from the open circuit potential. Next, Ea is initial-ized to Ecorr − 250 mV. Then, a potentiodynamic sweepis conducted by increasing Ea from Ecorr − 250 mV toEcorr + 250 mV at a slow scan rate, typically 0.125 mV/s.Next, a Tafel curve is plotted for Ea vs log10 Ia as illustratedby the example in Figure 3. Values for βa and βc are esti-mated from the slopes of the linear extrapolated anodic andcathodic currents, which are identified in Figure 3.

4


10−3 10−2 10−1 100 101 102−0.1

−0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Ia [µA]

Ea[V

]

Tafel CurveExtrapolated Anodic CurrentExtrapolated Cathodic Current

Figure 3. Illustration of a typical Tafel curve identifying theextrapolated anodic and cathodic currents.

3. µLPR CORROSION SENSOR

In this section, a micro-LPR (µLPR) is presented which usesthe potential step-sweep method to compute polarization re-sistance. The µLPR works on the same principle as themacro-sized LPR sensors and is designed to corrode at thesame rate as the structure on which it is placed. AlthoughLPR theory is well established and accepted as a viable cor-rosion monitoring technique, conventional macro-sized LPRsensor systems are expensive and highly intrusive. The µLPRis a micro-scaled LPR sensor inspired from the macro-sizedversion discussed in the previous section. Scaling the LPRsensor into a micro-sized package provides several advan-tages which include,

• Miniature form factor

• Two-pair electrode configuration

• Faster LPR measurements

3.1. Form Factor

Expertise in semiconductor manufacturing is used to micro-machine the µLPR. Using photolithography it is possible tomanufacture the µLPR sensor from a variety of standard engi-neering construction materials varying from steels for build-ings and bridges through to novel alloys for airframes. Themicro sensor is made up of two micro machined electrodesthat are interdigitated at 150µm spacing. The µLPR sensoris made from shim stock of the source/sample material thatis pressure and thermally bonded to Kapton tape. The shimis prepared using photolithographic techniques and ElectroChemical Etching (ECM). It is further machined on the Kap-ton to produce a highly ductile and mechanically robust microsensor that is very sensitive to corrosion. Images of the µLPRshown bare and a fitted sensor underneath a coating are shownin Figure 4.

(a)

(b)

Figure 4. Thin film µLPR sensor (a) exposed and (b) quasi-exposed with the lower-half underneath a coating.

3.2. Electrode Configuration

The µLPR differs from conventional macro-sized LPR sen-sors in two major ways. First, the µLPR is a two electrodedevice. The reference electrode is eliminated as the sepa-ration distance between the working and counter electrodes,typically 150µm, minimizes any voltage drop due to the so-lution resistance, Rs. Second, both electrodes are composedof the same working metal. This is uncommon in most elec-trochemical cells where the counter electrode is made of aninert material. The benefit is the electrodes provide a moredirect measurement of corrosion than techniques which useelectrodes made of different metals (eg. gold).

3.3. LPR Measurements

Potential step-sweeps are performed by applying a series of30 steps over a range of ±10 mV spanning a period of 2.6 s.This allows eight µLPR sensors to be measured in less than30 s. However, the effective scan-rate of 7.7 mV/s generatesan additional current, Idl, due to rapid charging and discharg-ing of the capacitance, referred to as the double-layer capaci-tance Cdl, at the electrode-electrolyte interface,

Idl = CdldEa

dt. (18)

Let the resulting polarization resistance that is computedwhen Idl is non-zero be represented by Rp. It can be shownthat Rp is related to Rp by the following,

R−1p = R−1

p + Ydl, (19)

5


0 2 4 6 8 103.6

3.7

3.8

3.9

4

4.1

4.2

Scan Rate [mV/s]

R−1

p[Ω

−1]·10−6

bC

bC

bCbC

bCbC

bC

MeasurementLinear Fit

bC

Figure 5. Plot of inverse polarization resistance vs. scan-ratefor a µLPR sensor made from AA 7075-T6 submersed in tapwater.

such that Ydl is defined by the admittance,

Ydl =

(Cdl

20 mV

)dEa

dt(20)

where dEa/dt is the scan rate. An example of this relation-ship is provided in Figure 5. In this example Cdl/20 mVandR−1

p correspond to the slope and y-intercept; these valueswere computed as 5.466×10−8 Ω−1·s/mV and 3.624×10−6 Ω,respectively. For a scan rate of dEa/dt = 7.7 mV/s, Ydl iscomputed as 4.209×10−7 Ω−1. Finally, for a given solution,Rp can be compensated by,

Rp =Rp

1− YdlRp

for YdlRp < 1. (21)

4. EXPERIMENT

4.1. Setup

The experiment consisted of twenty-four (24) µLPR sensorsand twelve (12) control coupons. The coupons and µLPRsensors were made from AA 7075-T6. Each coupon wasplaced next to a pair of µLPR sensors. Each sensor washeld in place using a non-reactive polycarbonate clamp witha nylon fitting. All the sensors and coupons were mountedon an acrylic plexiglass base with the embedded hardwareplaced on the opposite side of the frame, shown in Figure 6.An electronic precision balance (Tree HRB-203) with a cali-brated range of 0 − 200 g (±0.001 g) was used to weigh thecoupons before and after the experiment. Finally, a weather-ing chamber (Q-Lab QUV/spray) promoted corrosion on thecoupons and µLPR sensors by applying a controlled streamof tap water for 10 seconds every five minutes.

4.2. Procedure

First, the surface of each coupon was cleaned using sandblast-ing. Then, each coupon was weighed using the analyticalbalance. The entire panel of coupons and µLPR sensors were

(a)

(b)

Figure 6. Experimental setup showing (a) all 24 µLPR sen-sors, 12 coupons and three AN101 instrumentation boardsand (b) a close-up view of one of the panels used in the ex-periment.

placed in the weathering chamber for accelerated testing. Theexperiment ran for approximately 60 days. During the exper-iment, a set of coupons were periodically removed from theweathering chamber. Throughout the experiment, the SHMembedded hardware was logging Rp from each µLPR sen-sor. The sample rate was set at one sample per minute. Onceaccelerated testing was finished, the coupons were removedand the LPR data was downloaded and archived for analysis.The corrosion byproducts were removed from each couponby applying micro-bead blasting to the coupon surface. Fi-nally, the cleaned coupons were weighted using the analyticalscale to compute the relative corrosion depth during the ex-periment. Figure 7 shows images of coupons before and aftermicro-bead blasting. Also shown are images comparing twosets of three coupons after 15 days and 57 days of corrosionproceeding micro-bead blasting.

5. RESULTS

5.1. Coupon Corrosion

The corrosion byproducts were carefully removed usingmicro-bead blasting. The pitting depth, d, of each couponwas calculated using the formula,

d = 3

√3mloss

2πρNpitAexp, (22)

6


(a)

(b)

(c)

Figure 7. Image of the three AA 7075-T6 coupons (ID 2.01,2.03 and 2.04) after approximately 15 days of corrosion test-ing showing (a) the condition of the coupons before cleaningand (b) after cleaning using micro-bead blasting. Also shownfor comparison are (c) three AA 7075-T6 coupons (ID 2.09,2.10 and 2.11) after 57 days and cleaning using micro-beadblasting.

where values for the mass loss mloss, exposed surface areaAexp, resulting pit-depth, d, and total time of exposure ofeach coupon is provided in Table 1 in the Appendix. Valuesfor the pitting density and ρ were set at Npit = 11 cm−2 andρ = 2.810 g/cm3, respectively. The pitting density was com-puted by counting the average number of pits over the surfacefor coupons 2.06 and 2.08. The measurement uncertainty inthe pit-depth due to uncertainty in the mass loss, ∆mloss andpit density, ∆Npit, is approximately,

∆d ≈ d

3

(∆mloss

mloss

+∆Npit

Npit

), (23)

where ∆mloss = ±0.001 g is the minimum resolution of thescale and ∆Npit = ±3 cm−2 was the standard deviation ofthe measured pit density over 1 cm2 sample areas for coupons2.06 and 2.08.

5.2. µLPR Corrosion

The linear polarization resistance measurements were usedto compute corrosion pit-depth for each µLPR sensor. Thecomputed pit-depth for each of the 24 µLPR sensors over aperiod of approximately 60 days is provided in Figure 2(a).

0 10 20 30 40 50 600

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time [days]

Pit

dep

th[m

m]

b

bbb

b

b

b

b

bbbbSensor Avg.

Measurementb

(a)

0 10 20 30 40 50 600

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time [days]

Averagepit

dep

th[m

m]

b

bbb

b

b

b

b

bbbbSensor Avg.

Measurementb

(b)

Figure 8. Comparison of the measured and computed pit-depth over a period of approximately 60 days for (a) eachµLPR sensor and (b) the average of all µLPR sensors.

Superimposed in the plot are the pit-depth measurements ofthe corrosion coupons. The average computed pit-depth foreach of the 24 µLPR sensors over the same period is shownin Figure 2(b).

Next, the corrosion coupons were used to evaluate the perfor-mance of the µLPR sensor. Figure 9(a) compares the mea-sured pit-depth with the pit-depth computed from each ofthe 24 µLPR sensors during the removal of each corrosioncoupon. A solid line is used to identify an ideal one-to-onerelationship between the measured and computed pit-depth.Figure 9(b) shows the distribution of the computed residu-als from Figure 9(a). The residuals follow a Gaussian dis-tribution with a computed mean and standard deviation of0.0153 mm and 0.0272 mm, accordingly. Figure 9(c) com-pares the measured pit-depth with the pit-depth computedfrom the average of the 24 µLPR sensors during the removalof each corrosion coupon. Also provided in the plot are theerror bars corresponding to one standard deviation in the cor-rosion measurements.

6. APPLICATIONS IN PROGNOSTICS

The data generated from this experiment can be used todemonstrate the feasibility of the µLPR sensor for use inprognostics. According to (Vachtsevanos, Lewis, Roemer,Hess, & Wu, 2006), prognostics is the ability to predict ac-

7


0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Measured Pit Depth [mm]

LPR

Pit

Dep

th[m

m]

bCbCbCbC

bCbCbC

bC

bCbCbCbCbC

bC

bC

bC

bC

bCbC

bCbCbCbCbCbCbCbC

bCbCbCbC

bCbCbCbCbC

bC

bCbCbCbCbCbCbCbCbC

bCbCbCbC

bCbCbC

bC

bCbCbCbCbC

bC

bC

bC

bC

bCbC

bCbCbCbCbCbCbCbC

bCbCbC

bC

bCbCbCbCbC

bC

bC

bC

bC

bCbC

bCbCbCbC

bCbCbCbC

bCbCbCbC

bC

bCbC

bC

bCbC

bCbC

bC

bCbC

bCbC

bCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bC

bC

bC

bCbC

bC

bCbC

bCbCbCbC

bCbCbCbC

bCbCbCbC

bC

bCbC

bC

bCbC

bCbC

bC

bCbC

bCbC

bCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bC

bC

bC

bCbC

bC

bCbC

bCbCbCbC

bCbCbCbC

bCbCbCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bCbC

bCbCbCbCbCbC

bCbCbCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bCbC

bCbC bCbCbCbC

bCbCbCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bCbC

bCbC bCbCbCbC

bCbCbCbC

bCbCbCbCbCbCbCbC

bC

bCbC

bCbC

bCbCIdealMeasurementbC

(a)

−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.10

2

4

6

8

10

12

14

16

Pit Depth Residuals [mm]

Den

sity

bC bC

bC

bC

bC bC bC

bCbC

bC

bC

bC

bC

bCbC

bC

bCbC

bC

bC

bC

bC

bC

bCbC

bCbC

bC bCbC

ResidualsGaussian Fit

bC

(b)

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Measured Pit Depth [mm]

AverageLPR

Pit

Dep

th[m

m]

bCbC

bCbC

bCbCbCbC

bCbC

bCbC

bCbC

bCbC

bCbCbCbC bCbC bCbCIdealMeasurementbC

(c)

Figure 9. Direct comparison of the measured versus com-puted pit-depth for (a) each µLPR sensor measured at eachcoupon removal, (b) the corresponding distribution of thecomputed residuals and (c) average µLPR sensor measure-ment at each coupon removal.

curately and precisely the Remaining Useful Life (RUL) ofa failing component or subsystem. In this application thefailing component is the metallic structure being monitored,more specifically AA 7075-T6, by the µLPR sensor. Thephysical quantity to be predicted, otherwise referred to as thefault dimension, is the pit-depth of the metallic structure.

Typically the fault dimension is not a quantity that is directlymeasured. Rather, the fault dimension is commonly com-puted from a mapping of one or more indirect measures, orfeatures. In this application, the features are temperature andpolarization resistance. In this particular experiment, tem-perature remained at (or near) room temperature. As a re-

Figure 10. Histograms of the fault dimension (pit depth) after1, 10, 30 and 57 days into the experiment.

sult, the primary feature is polarization resistance. The faultdimension was computed using the mathematical model de-rived earlier in Section 2.4 as a mapping function.

Of course, the fault dimension does not stay constant withtime. The fault grows with many factors including the currentfault dimension and the operating environment. In this exam-ple, the operating environment is the environmental chamberwith the 5 minute spray cycle operating at room temperature.Figure 10 shows the distribution of the fault dimension for 24µLPR sensors at four different time intervals correspond tothe end of day 1, 10, 30 and 57. As time progresses, the dis-tribution of the fault dimension changes. In practice, a fault-growth model is used to project the fault dimension over afuture time interval. The fault-growth model usually includesmultiple factors. In this example, the fault-growth, or corro-sion rate, is dependent on factors that include the solution,temperature, pH and time of wetness. Please note: in thescope of this paper no fault-growth model was studied. Allof the data analysis for prognosis feasibility was performeda-posteriori using the µLPR data only.

Knowledge of the fault dimension distribution at any momentin time is important in prognostics. The RUL is defined as aprojection of the fault dimension onto the time domain fora fixed value, referred to as the failure threshold. The fail-

8


Figure 11. Prognosis feasibility demonstrated by (top) plots of (top) fault-dimension (pit-depth) vs. time identifying the timewhen fault-dimension computed from each of the 24 µLPR sensors first exceeds the failure threshold of (from left-to-right)0.150 mm, 0.200 mm and 0.250 mm, respectively. Also shown (bottom) are histograms of the RUL corresponding to eachrespective failure threshold.

ure threshold value is usually determined a-priori by an en-gineer with experience in the particular area of application.In this example, three failure thresholds are selected for thefault dimension: 0.150 mm, 0.200 mm and 0.250 mm. Theupper plots in Figure 11 show the fault-dimension measuredfrom each µLPR sensor over a period of approximately 60days. Starting from left to right, the failure threshold valuesof 0.150 mm, 0.200 mm and 0.250 mm are identified with ared horizontal line. The intersection of the failure thresholdand each of the 24 fault-dimension curves are identified witha superimposed stem plot. The lower plots in Figure 11 showthe probability density function (pdf) of each respective stemplot on the time-axis. Each pdf represents the RUL startingfrom the beginning of the experiment at time zero. Note, asthe failure threshold is raised, both the mean value and vari-ance of the RUL grows. This demonstrates the uncertainty inthe RUL prediction increases over an increasing time interval,which is to be expected.

7. SUMMARY

A micro-sized LPR (µLPR) sensor was presented for cor-rosion monitoring in Structural Health Management (SHM)applications. An experimental test was performed to com-pare corrosion measurements from twenty-four µLPR sen-sors with twelve corrosion coupons. Both the corrosioncoupons and sensors were constructed from the same mate-rial, AA 7075-T6. According to the results, the pit-depthcomputed from the µLPRsensors agreed the pit-depth mea-sured from the corrosion coupons to within a statistical con-

fidence of 95%. The results indicate multiple µLPR sensorscan be used to provide an accurate measurement of corrosion.The paper concluded with a feasibility study for the µLPRsensor in prognostic applications.

Future work includes a combination of prognostic model de-velopment, extensive laboratory testing and field testing. Be-fore the µLPR sensor can be used in prognostic applica-tions, a prognostic algorithm, more specifically a fault-growthmodel, must be developed. Additional laboratory testing isrequired to: evaluate the µLPR sensors for different alloys(eg. AA 2024-T3); perform experiments using standardizedprotocols (eg. SAE J2334 & ASTM G85); perform a blindstudy evaluating µLPR-based prognostic algorithms; and cer-tify the AN101 SHM system for electromagnetic interfer-ence (EMI), electromagnetic conductance (EMC) and envi-ronmental conditions following MIL-STD-461F and MIL-STD-810G specifications. Field testing is necessary to evalu-ate the performance of the sensor and SHM system in a real-world environment. The system has been certified for initialflight testing on a C-130 legacy aircraft. Follow-on testing in-cludes performing a smaller-scale validation experiment forthe µLPR in-flight.

ACKNOWLEDGMENT

All funding and development of the sensors and systems inthe project has been part of the US government’s SBIR pro-grams. In particular: 1) In preparing the initial system designand development, funding was provided by the US Air Forceunder SBIR Phase II contract # F33615-01-C-5612 monitored

9


by Dr. James Mazza, 2) Funding for the development and ex-perimental set-up was provided by the US Navy under SBIRPhase II contract # N68335-06-C-0317 monitored by Dr. PaulKulowitch, and 3) for further improvements and scheduledfield installations by the US Air Force under SBIR Phase IIcontract # FA8501-11-C-0012 monitored by Mr. FeraidoonZahiri.

NOMENCLATURE

βa Anodic Tafel slope V/decβc Cathodic Tafel slope V/decρ Density g/mm3

d Corrosion depth cmk LPR sample index –icorr Corrosion current density A/cm2

ipit Pitting current density A/cm2

ipv Passive current density A/cm2

mloss Mass loss due to corrosion gz Number of electrons lost per atom –∆d Corrosion depth uncertainty cm∆mloss Mass loss uncertainty g∆Npit Pit density uncertainty cm−2

Aexp Exposed coupon area cm2

Asen Effective sensor area cm2

AW Atomic Weight g/mol

B Proportionality constant V/decCdl Double-layer capacitance FEa Applied potential VEcorr Corrosion voltage VEW Equivalent weight g/mol

F Faraday’s constant C/mol

Ia Applied current AIcorr Corrosion current AIdl Scanning current from Cdl AM Number of moles reacting molN Total number of µLPR samples –Npit Pit density cm−2

Qcorr Charge from oxidation reaction CRp Polarization resistance Ω

Rp Measured polarization resistance Ω

Rs Solution resistance ΩTs Sampling period sYdl Scanning admittance from Cdl s

REFERENCES

Bockris, J. O., Reddy, A. K. N., & Gambola-Aldeco, M.(2000). Modern Electrochemistry 2A. Fundamentalsof Electrodics (2nd ed.). New York: Kluwer Aca-demic/Plenum Publishers.

Buchheit, R. G., Hinkebein, T., Maestas, L., & Montes,L. (1998, March 22-27). Corrosion Monitoring ofConcrete-Lined Brine Service Pipelines Using AC and

DC Electrochemical Methods. In CORROSION 98.San Diego, Ca.

Burstein, G. T. (2005, December). A Century of Tafel’s Equa-tion: 1905-2005. Corrosion Science, 47(12), 2858-2870.

G102, A. S. (1994). Standard Practice for Calculation ofCorrosion Rates and Related Information from Electro-chemical Measurements. Annual Book of ASTM Stan-dards, 03.02.

G59, A. S. (1994). Standard Practice for Conducting Po-tentiodynamic Polarization Resistance Measurements.Annual Book of ASTM Standards, 03.02.

Harris, S. J., Mishon, M., & Hebbron, M. (2006, Octo-ber). Corrosion Sensors to Reduce Aircraft Mainte-nance. In RTO AVT-144 Workshop on Enhanced Air-craft Platform Availability Through Advanced Mainte-nance Concepts and Technologies. Vilnius, Lithuania.

Huston, D. (2010). Structural Sensing, Health Mon-itoring, and Performance Evaluation (B. Jones &W. B. S. J. Jnr., Eds.). Taylor and Francis.

Introduction to Corrosion Monitoring. (2012,August 20). Online. Available fromhttp://www.alspi.com/introduction.htm

Twomey, M. (1997). Inspection Techniques for Detect-ing Corrosion Under Insulation. Material Evaluation,55(2), 129-133.

Vachtsevanos, G., Lewis, F., Roemer, M., Hess, A., & Wu, B.(2006). Intelligent Fault Diagnosis and Prognosis forEngineering Systems. Hoboken, NJ, USA: John Wileyand Sons.

Wagner, C., & Traud, W. (1938).Elektrochem, 44, 391.

BIOGRAPHIES

Douglas W. Brown is the senior systems engineer atAnalatom with eight years of experience developing and ma-turing PHM and fault-tolerant control systems in avionics ap-plications. He received the B.S. degree in electrical engi-neering from the Rochester Institute of Technology in 2006and the M.S/Ph.D. degrees in electrical engineering from theGeorgia Institute of Technology in 2008 and 2011, respec-tively. Dr. Brown is a recipient of the National DefenseScience and Engineering Graduate Fellowship and has re-ceived several best-paper awards in his work in prognosticsand fault-tolerant control.

Duane Darr is the senior embedded hardware engineer atAnalatom with over 30 years of experience in the softwareand firmware engineering fields. He completed his under-graduate work in physics, and graduate work in electrical en-gineering and computer science at Santa Clara University.Mr. Darr’s previous work at Epson Imaging TechnologyCenter, San Jose, California, as Senior Software Engineer,

10


Data Technology Corporation, San Jose, California as Se-nior Firmware Engineer, and Qume Inc., San Jose Califor-nia, as Member of Engineering Staff/Senior Firmware Engi-neer, focused on generation and refinement of software andfirmware solutions for imaging core technologies as well asdigital servo controller research, development, and commer-cialization.

Jefferey Morse is the director of advanced technology atAnalatom since 2007. Prior to this, he was a senior scientistin the Center for Micro and Nano Technology at LawrenceLivermore National Laboratory. He received the B.S. andM.S. degrees in electrical engineering from the University ofMassachusetts Amherst in 1983 and 1985, respectively, anda Ph.D. in electrical engineering from Stanford University in1992. Dr. Morse has over 40 publications, including 12 jour-nal papers, and 15 patents in the areas of advanced materi-

als, nanofabrication, sensors and energy conversion technolo-gies. He has managed numerous projects in various multidis-ciplinary technical areas, including electrochemical sensorsand power sources, vacuum devices, and microfluidic sys-tems.

Bernard Laskowski is the president and senior research sci-entist at Analatom since 1981. He received the Licentiaatand Ph.D. degrees in Physics from the University of Brus-sels in 1969 and 1974, respectively. Dr. Laskowski has pub-lished over 30 papers in international refereed journals in thefields of micro physics and micro chemistry. As presidentof Analatom, Dr. Laskowski managed 93 university, govern-ment, and private industry contracts, receiving a U.S. SmallBusiness Administration Administrator’s Award for Excel-lence.

11


Table1.E

xperimentalm

easurements

ofcouponcorrosion.

Coupon

IDTim

eE

xposed[m

in]

Area [cm

2 ]InitialM

ass[g

]FinalM

ass[g

]M

assL

oss[g

]pit-depth

[mm

]

Control

05.8

01×

10

17.6

870×

10

17.6

869×

10

11×

10−

3N

/A

2.012.1

198×

104

5.8

05×

10

17.7

253×

10

17.7

215×

10

13.8×

10−

22.232

×10−

1

2.021.1

160×

104

5.7

98×

10

17.6

842×

10

17.6

818×

10

12.4×

10−

21.915

×10−

1

2.032.1

198×

104

5.7

99×

10

17.6

927×

10

17.6

896×

10

13.1×

10−

22.068

×10−

1

2.042.1

198×

104

5.8

05×

10

17.6

897×

10

17.6

869×

10

12.8×

10−

22.016

×10−

1

2.056.0

090×

104

5.8

01×

10

17.6

967×

10

17.6

886×

10

18.1×

10−

22.873

×10−

1

2.063.8

510×

104

5.7

98×

10

17.6

884×

10

17.6

828×

10

15.6×

10−

22.540

×10−

1

2.076.0

090×

104

5.8

00×

10

17.6

945×

10

17.6

856×

10

18.9×

10−

22.964

×10−

1

2.083.8

510×

104

5.8

03×

10

17.6

921×

10

17.6

810×

10

15.4×

10−

22.509

×10−

1

2.098.3

010×

104

5.8

02×

10

17.7

005×

10

17.6

885×

10

11.20

×10−

13.275

×10−

1

2.108.3

010×

104

5.8

02×

10

17.7

165×

10

17.7

057×

10

11.08

×10−

13.162

×10−

1

2.118.3

010×

104

5.8

02×

10

17.6

865×

10

17.6

751×

10

11.14

×10−

13.219

×10−

1

2.128.3

010×

104

5.8

02×

10

17.6

902×

10

17.6

769×

10

11.33

×10−

13.389

×10−

1

12

Real-Time Corrosion Monitoring of Aircraft Structures with ... · corrosion monitoring (Harris,...

Documents

Transcript of Real-Time Corrosion Monitoring of Aircraft Structures with ... · corrosion monitoring (Harris,...