2021 W.R. Whitney Award: The Importance of Local Chemistry and Potential in Localized Corrosion and Stress Corrosion Cracking

Corroding surfaces are notoriously near-sighted; they react to the electrochemical potential and the chemical composition of the solution in which they are in direct contact. The electrochemical potential defines, in part, the possible reactions that can occur as well as their rate. Although the surface solution composition can be affected by reactions and events that occur at greater distances, it is the solution closest to the surface that matters most. Scanning methods have made great strides in allowing the probing of both potential^1-2  and chemistry^3-4  near surfaces that are boldly exposed and thus directly accessible. However, some of the most impactful corrosion processes occur in occluded regions such as crevices, intergranular fissures, and cracks. Access to these sites is difficult as the dimensions are small, but the potentials and solution compositions are the key factors in the stability and propagation rate of corrosion in these areas, which can determine the service life of structures. Better characterization of these conditions inside occluded sites would not only allow an improved description of the electrochemical kinetics in concentrated solutions, but also provide insights into how such damage can be mitigated.

The following sections describe the challenges of measurements in occluded sites, as well as those of modeling the corrosion processes that occur within them and on the boldly exposed surfaces outside them. Examples will be drawn from the work of my students, although this focus is not meant to disregard the superb work done by many others. The paper ends with some thoughts on the future of measurements and modeling.

The chemical composition of occluded sites has been proposed to be altered due to local chemical reactions since Hoar and Agar.⁵  Oldfield and Sutton⁶  formulated a framework for the initiation of crevice corrosion in stainless steels that has been the basis of the field’s thinking about occluded solution chemistry since. In this framework, occluded regions first have their oxygen depleted due to the restriction of diffusion into the site. Due to this oxygen depletion, a galvanic cell is created, with the primary anode inside the occluded site and the primary cathode outside. The dissolution of metal inside the crevice leads to acidification due to hydrolysis of the metal cations. Due to the need for electroneutrality and their high mobility, chloride ions migrate into the crevice to balance the positive charge. The net result is an acidic, concentrated metal chloride solution that dissolves the passive film, leading to breakdown and high-rate metal dissolution, starting an autocatalytic cascade, stabilizing the occluded cell environment, and leading to damage evolution.

Attempts to measure the chemical composition of these occluded sites have been the subject of many studies,^7-11  using both in situ^8-9  and ex situ methods.^7,10-11  Brown, et al.,⁷  are generally recognized as the first to demonstrate the acidification of occluded sites in their study of crevice corrosion on Fe-Cr alloys. They used indicator paper applied to surfaces that had been part of an occluded solution before being frozen, separated, and allowed to melt. The ex situ characterization of occluded solutions has generally followed this generic approach. A number of in situ measurements of occluded cell chemistry have also been reported,^8-9  with the pH and chloride concentration being the focus. Microelectrodes inserted into a crevice before assembly have been the method of choice, although, in order to protect the delicate tips, crevice gaps have been generally far larger than one would expect in service, thus begging the question as to the actual composition of a highly occluded site. Measurements of potential must be, by their very nature, in situ. Again, microelectrodes have been successfully used to follow the spatiotemporal evolution of the potential during crevice corrosion, albeit with the same caveats regarding crevice size.

All of these studies of occluded chemistry and potential added to our understanding, and in large part confirmed the essentials of the Oldfield and Sutton framework, including its extension to alloy systems other than stainless steel. Nonetheless, the methods used were not applicable to the measurement of minor species which can have dramatic effects on corrosion initiation and growth, despite their relatively low solution concentrations. At the other end of the concentration spectrum, none of the methods investigated the presence and movement of water into and out of occluded sites as the site reacts to changes in external temperature and relative humidity (RH). The actual thickness of electrolyte layers exposed to atmospheric conditions, including salt spray, has also resisted quantitative analysis due to a lack of methods. Finally, although previous studies had accessed occluded sites to measure chemistry, none were able to directly change the chemistry of the site and measure the reaction of the system. In the examples below, each of these unaddressed issues is addressed.

MnS inclusions are well known for their role as initiation sites in austenitic stainless steels.^11-19  Their dissolution releases chemical species that in some way initiate and stabilize a localized attack. The rich chemistry of sulfur has led to virtually every valence state being accused as the culprit.¹⁵  In the early 1980s, thiosulfate had been shown to be an aggressive species when present in concert with chloride over a range of thiosulfate-to-chloride ratios.¹⁶  Lott and Alkire¹⁷  made measurements of dissolution products from synthetic MnS inclusions using UV-Vis spectrophotometry absorbance at 214 nm, and concluded that thiosulfate was the product of MnS dissolution and thus the most likely suspect. Since that time, many authors have invoked that conclusion.^18-20  Several sulfur species absorb at 214 nm, so a means of simultaneous measurement of all sulfur species was needed. Brossia and Kelly used capillary electrophoresis (CE) to assess crevice solutions for sulfur species.¹⁵  CE uses nanoliter volumes and has a nanogram mass sensitivity. Just as important, CE uses separation before detection to allow one sample to be used for the measurement of multiple species. Most importantly, CE allows for the separation of species by valence, thus allowing differentiation among sulfide, sulfite, thiosulfate, and sulfate.

Figure 1(a) shows electropherograms for solutions extracted from crevices of 304SS with gaps of 4 μm early in the initiation stage and during steady-state propagation, as well as that from a standard containing a mixture of sulfur species.¹⁵  The standard is used to identify the species in other electropherograms via their migration time, and calibration curves are used to quantify the amount of each species from the peak height or area of the peaks. It is clear that no thiosulfate is detected in the actual crevice, nor was it in any of the crevice samples tested. Instead, HS⁻ is the only species detected at initiation, with a small amount of sulfite being observed during steady-state propagation, most likely due to homogeneous oxidation of the bisulfide. Figure 1(b) illustrates why bisulfide is so important to localized corrosion stabilization in stainless steels. Independent of the sulfur content of the stainless steel (304C had 0.027 wt% sulfur, 304SCQ had 0.0017 wt% sulfur), the presence of the concentration of sulfide estimated to be in the crevice during propagation²¹  strongly activates the surface, accelerating dissolution and preventing passivation upon anodic polarization.

A similar approach was used to understand the dissimilar metal crevice (DMC) corrosion phenomenon in which the crevice corrosion of marginal stainless steel (e.g., 316L [UNS S31603⁽¹⁾]) can lead to the crevice corrosion of a highly resistant stainless steel (e.g., 29-4C) when they constitute the two sides of the crevice.²²  By measuring the chemical composition of the solution resulting from the process and then determining the polarization behavior of the two alloys in that solution composition, the ability of the 316L to polarize the 29-4C into the active nose of its polarization curve was established, and the mechanism controlling this unusual phenomenon was established.

A different form of occluded site corrosion occurs in structures such as lap-splice joints in aircraft. An electrolyte is formed in the occluded region when water and contaminants are drawn into the structure during rain or condensation events. When the outer surface is wet, the standard model of crevice corrosion can apply, with the deoxygenated occluded region being the anode in the galvanic couple with cathodic reactions on the external surface, thereby driving corrosion in the crevice. In service, however, the majority of the time the external surface is dry, although as will be seen, an aqueous solution likely remains in the occluded site. During these times, a galvanic couple in which the majority of the cathodic reaction is outside the occluded region cannot form. Instead, the entirety of the electrochemical reactions must occur within the lap splice joint which has an important impact on the pH of the occluded site and the morphology of the corrosion damage.

After the Aloha Airlines accident,²³  a greater emphasis on hidden corrosion damage led to a large aging aircraft program across the U.S. federal government. One goal was to be able to simulate the corrosion damage observed in actual lap joints in the laboratory. In collaboration with NASA/Langley and Tinker Air Force Base, over 100 lap joints removed from KC-135 and other aircraft were disassembled, and their occluded solution reconstituted and analyzed with CE.²⁴  An impressive variety of ionic species was found, 25 in all. One example electropherogram is shown in Figure 2(a). In addition, five lap joints were exposed to high humidity before opening in order to rehydrate any soluble species present. The joints were then disassembled and pH paper was pressed onto the mating surfaces. The pH values measured were between 7.5 and 8.5. Cross sections of the service joints showed that the damage was nominally uniform corrosion with an absence of large pits, as shown in Figure 2(b). Exposure of AA2024-T351 to 0.6 M NaCl for the same amount of time creates, as expected, distinct, hemispherical pits (Figure 2[c]). Thus, it is obvious that a simple chloride solution does not simulate the attack observed.

To create an appropriate simulant solution, a downselection of the 25 species detected to only those species that impacted corrosion behavior was needed. A design-of-experiments approach was used, and a lap joint simulant was proposed²⁴  containing four anionic species in addition to hydroxyl. The likely concentration range of the species was determined, and a solution composition for accelerated testing of occluded joints of aluminum alloys of 20 mM NaCl, 4 mM NaNO₂, 4 mM NaHCO₃, and 2 mM NaF at pH 9.0 was recommended as a test solution. When AA2024-T351 is exposed to this test solution, a scalloped surface appearance develops as shown in Figure 2(d), very similar to the attack morphology of the actual lap joint retrieved from service (Figure 2[b]).

This alkaline pH observed in the lap splice joints would seem to be in direct contradiction to the standard Oldfield and Sutton framework for occluded cell corrosion in which an acidic environment is expected, including for aluminum alloys.^25-26  The key to understanding the mildly alkaline pH lies in consideration of the effects of local cathodic reactions inside the occluded region. Both common cathodic reactions in aqueous solution, the oxygen reduction reaction (ORR) and the hydrogen evolution reaction (HER), raise the pH by the creation of OH⁻ or consumption of H⁺, respectively. These reaction products then interact with the H⁺ created by the hydrolysis of the metal cations. If the hydrolysis reactions for metal cations were complete to their terminal hydroxides, the pH would fall, independent of the local cathodic reactions. However, the metal cations of interest (e.g., Fe²⁺, Cr³⁺, Ni²⁺, and Al³⁺) only partially hydrolyze. Therefore, an equilibrium pH is created involving all of the homogeneous reactions in the occluded site. This equilibrium can be visualized with the diagram shown in Figure 3.

Figure 3 shows that when all of the cathodic reaction is local (i.e., a fraction of electrons consumed locally = 1), then the equilibrium pH of corroding aluminum is close to 7.7.²⁵  Thus, for the majority of the service life when the outside surface of the joints is dry, the pH inside the joints would be expected to be mildly alkaline and more or less uniform corrosion would result, as observed. The use of metal chloride salts to simulate localized corrosion site conditions is appropriate if virtually all of the cathodic reaction occurs outside the occluded site. In essence, the metal salts hydrolyze in the absence of any local cathodic reaction, leading to low pH values. Consistent with this idea, Figure 3 shows that in order to have pH values at 3 as often cited for active Al pits,^26-27  no more than 0.1% of the cathodic reaction can be local.

The reason for the movement of water into an occluded region is straightforward; capillary action draws a solution at the mouth of the crevice well into any lap joint. The speed at which that occurs, the impact of that solution on corrosion activity, and the difficulty in removing that solution had not been explored using in situ methods before the work of Cooper, et al.²⁸  The linking of two technology developments allowed these issues to be addressed. Fiber-optic-based moisture sensors²⁹  were able to determine when an area had condensed water via a change in the refractive index of coatings on the optical fiber. The Superconducting Quantum Interference Device (SQUID) has an extraordinarily high sensitivity to magnetic fields and has been used to measure ionic currents in living organisms³⁰  as well as corrosion currents³¹  among many other uses.

Combining these two methods allowed new insights into the corrosion that occurs in isolated occluded regions, that is, those that do not have wetted outside surfaces. Figure 4(a) shows the experimental arrangement of a simulated lap splice joint instrumented with four low-profile optical fibers with their sensing elements at different positions within the joint. Water was introduced at the joint mouth and both the ingress of water and the corrosion currents as measured by the SQUID were recorded. Figure 4(b) shows one SQUID image of the currents as a function of position. Integration of the response over the surface has been shown to correlate with corrosion rate.²⁸  Figure 4(c) shows both the fiber optic measurement of wetting and the simultaneous SQUID measurement, indicating that water rapidly enters the lap joint, moving as much as 44 mm in less than 20 min. The first SQUID measurement after the introduction of the water at the joint mouth shows that the water led to corrosion activity which reached a quasi-steady state for at least 15 h. Thus, tight occluded regions wet very quickly and start corroding immediately.

Although the wetting behavior is not overly surprising, the response of the system to drying showed the strength of the capillary action. After removing the source of the water and introducing dry nitrogen gas into the chamber, two of the fibers sense some drying within 4 h to 12 h. These are the two fibers farthest from the mouth which was the location of the water source. All of the fibers sensed water for the entire testing time (>30 h), with the fiber closest to the mouth showing no indication of drying. The corrosion activity, integrated over the entire occluded surface, showed complex dynamics. After an initial decrease, the corrosion activity spiked for approximately 5 h before decreasing to a lower steady state.

These data demonstrate that occluded regions not connected to external cathodes wet quickly, are difficult to dry, and maintain corrosion activity in the partially dried state. Such results have important implications for many structures in service as it is impossible to determine the state of wetness of an occluded region that was not previously instrumented. Thus, selection of a time of wetness of such regions is very difficult, with the assumption of constant wetness after exposure to a liquid being of reasonable conservatism.

The most occluded site one can envision is a crack, with the opening at crack tips being less than 1 µm in some cases.³²  Nonetheless, the crack tips are not immune from the importance of chemistry and potential. Making measurements of important parameters has posed a tremendous challenge not only because of the dimensions but also because of the need to combine mechanical loading with the electrochemical control. Cooper and Kelly³³  were able to achieve measurements of potential, pH, and chloride at and near a crack tip during stress corrosion cracking of AA7050 in a chloride/chromate solution. Figure 5(a) shows the apparatus: holes were drilled from the side of a compact tension sample to the plane on which the crack would propagate. These holes were then fitted with miniature combination electrodes capable of measuring potential, pH, and chloride. These static positions were then passed by the crack causing immediate wetting of the electrodes and the ability to measure the parameters of interest.

Figure 5(b) shows a subset of the results in which the sample was held potentiostatically and at a constant stress intensity (K). The top graph shows the crack length as a function of time as well as the position of the borehole. The middle graph shows both the applied and the crack tip potential, demonstrating the large increase in ohmic drop as cracking accelerated. The bottom graph shows that the crack tip chloride concentration increase occurred simultaneously with the drop in crack tip potential, but preceded the decrease in pH by nearly 24 h which occurred once the crack velocity increased substantially.

Although measurement of crack tip conditions was important and led to important conclusions regarding the cracking process,³³  the experimental arrangement also allowed for the crack tip solution chemistry to be directly controlled by replacing the mini-electrodes with a syringe pump. Figure 6(a) shows that a crack tip condition that led to a 100× increase in crack velocity could be established by injecting a small amount of crack tip simulant solution at the tip of an AA7050-T651 crack that was only slowly propagating. Figure 6(b) is a demonstration of the inherent resistance of overaged tempers of AA7XXX alloys to SCC. Although the initial conditions were more severe than that for the -T651 material in Figure 6(a), the crack velocity for the -T7451 upon injection had only a very small and short transient increase until an even higher potential was applied. These data show that SCC resistance of these alloys is not due to an inability to create a critical chemistry, but rather due to their inherent resistance to that chemistry causing hydrogen embrittlement.³³ 

Another measurement conundrum in corrosion science has been the determination of the thickness of thin electrolyte layers relevant to atmospheric corrosion. Whether in a laboratory accelerated test such as ASTM B117 or in the field, the water layer (WL) thickness has a great impact on the processes that control corrosion including chemistry, electrochemical kinetics, and current and potential distributions. Control of the WL thickness is important in these tests, although none of the standard test methods address it. Tokuda, et al.,³⁴  developed a means of measuring WL thickness in conditions where the electrolyte composition is known, such as B117 testing. The principle is simple: measure the resistance of the WL using electrochemical impedance then, knowing the solution composition (and hence the conductivity), the WL thickness can be calculated as long as the cell constant is known. The challenge was how to implement the electrodes to measure the resistance without impacting the thickness of the WL that one is trying to measure. By using finite element method (FEM) analysis to guide the design of the device, Tokuda and Katona used small wires embedded in epoxy which was then plasma cleaned to ensure wettability before being placed into the test chamber. The FEM analysis results were validated using known geometries and constrained WL thicknesses.

Figure 7 shows the application of the device to a B117 test environment. As expected, the WL thickness increases with an increase in the angle of the wetted surface from the vertical, but what was not expected was that above 25°, the WL thickness did not reach a steady state, instead undergoing repeated, dramatic decreases followed by slow increases. The origin of these time-varying WL thicknesses is clearly the accumulation of water exceeding the ability of the surface to hold it against gravity, so that upon reaching a critical thickness nearly the entire WL runs off, and the accumulation process restarts. For B117, the standard test method allows any angle of inclination between 15° and 30°, which overlaps with the conditions of sporadic runoff. Some standard test methods allow angles from 6° to 45°.³⁴  This phenomenon not only could be part of the explanation for the well-known variability in B117 testing but also creates concerns about the lack of tight control of WL in the lab and the connection to service conditions.

Measurements of local solution chemistry can provide important insights and input data for the development and exploration of models. All corrosion models, whether analytical or numerical, require descriptions of the electrochemical kinetics for the reactions of interest. The kinetics must be generated in the appropriate chemistry. The recent work at UVA on intergranular corrosion (IGC) and stress corrosion cracking (IGSCC) of AA5XXX alloys provides examples of this approach. There had been a great deal of debate as to the correct scaling law for crevice geometry. The ability to make larger, more experimentally accessible crevices relies on a clear scaling law. Using microfabrication methods along with FEM modeling, a combined measurement and modeling study clarified the correct scaling law, as described below. Selection of the governing mass transport equation for corrosion modeling has long been a challenge. The most comprehensive governing equation (Nernst-Planck) is computationally intensive to the point of being of limited utility, so there has existed a need to assess how well the Laplace Equation could calculate the parameters of interest. Comparison of calculated and measured damage in geometrically-complex galvanic structures showed that under some circumstances, the Laplace Equation approach is valid. Finally, any long-term prediction is challenging, but in corrosion such predictions have been fraught with the difficulty of accumulation of error and the evolution of the electrochemical kinetics and local environment. An alternative approach is to develop a physics-based bounding solution. An example of this approach for pitting under thin electrolytes is described.

Al-Mg (AA5XXX) alloys are susceptible to IGC and IGSCC when they are sensitized by the precipitation of the β phase at grain boundaries which can occur at temperatures as low as 40°C. This precipitation creates an active path along the grain boundaries as the β phase is highly reactive in salt solutions. One interesting aspect of the IGC is that full coverage of the grain boundaries by β is not required to create susceptibility, which might be expected when the active phase is the precipitate. In sensitized stainless steel, the precipitate Cr₂₃C₆ is not active. Rather, it is the adjacent areas that are low in Cr. For AA5XXX alloys, the rate of IGC is strongly dependent on potential, increasing as the potential becomes more positive.

To investigate how this may occur, Lim, et al.,³⁵  described the use of electrodes made with the same chemistry as the main solid solution (α) and the β phase. During IGC, the majority of the surface of the grain boundary fissure itself is α, so it would be expected that the fissure solution chemistry would reflect the α composition. Thus, the electrochemical kinetics of the two phases in that α-based solution should be able to be combined appropriately to capture the potential dependence of the IGC kinetics. Figure 8(a) shows the polarization curves generated for the α and β phases in the α-based solutions of different levels of saturation. Figure 8(b) shows the results of the incorporation of those kinetics into a simple analytical model of IGC propagation in which the only fitting parameter is the β coverage (θ_β) that is related to the NAMLT³⁶  measurement of IGC susceptibility. The models show that although it would tempting to represent the dependencies as linear, the nonlinear nature of the dependence is real.

Crevice corrosion modeling has a long history, from conceptual models^5-6  to analytical models³⁷  to FEM models.^38-41  One conflict that existed was what geometric law allowed similitude between crevices of different size. Such a scaling law would be useful, as the very small dimensions of actual crevices make many measurements of interest either extremely difficult or impossible. With a scaling law, experimental apparatus could be designed and constructed that were scaled-up crevices with the assurance that measurements made could be directly applied to the smaller scale crevices of relevance to service.

The two geometric parameters of importance in a crevice are its gap (g) and the position of maximum attack (x_crit). Early crevice model frameworks predicted that the maximum attack would occur at the most occluded position in the crevice, that is, the position farthest from the mouth as it is at this position that the altered chemistry inherent in crevice corrosion⁶  would be most severe. Such predictions flew in the face of observations that the position of maximum attack was at some distance from the mouth far less than the deepest part of the crevice. A debate developed as to whether the scaling law for crevice corrosion was linear or quadratic. That is, was the key parameter x_crit/g (linear)³⁷  or x_crit²/g (quadratic).⁴² 

In an effort to address this debate, Lee, et al.,⁴³  used microfabrication techniques common in microelectronic manufacture to create crevices with rigorously controlled dimensions at the scale of actual crevices in service. Using the Ni/H₂SO₄ system, Lee performed experiments with different gaps and assessed the position of maximum attack (Figure 9[a]), and he performed computations using the Laplace Equation as the governing law, determining where the active/passive transition for the Ni would occur, as this would be x_crit. Figure 9(b) shows that for crevices in the micrometer range, the quadratic law clearly applied. The deviation from the quadratic law observed experimentally was also captured computationally and shown to be due to the finite length of the crevice.

One of the challenges which dogged FEM modeling of the mass transport important to corrosion was the need for each research group to write their own code for the calculations from scratch as no commercial code existed that could expressly handle the highly nonlinear boundary conditions inherent in these situations. Transference of code between and among researchers was uncommon, so progress was restricted. In the early 2000s, this situation changed with several commercial packages becoming available that had modules explicitly designed for electrochemical systems, including corrosion.

The availability of commercial packages did not solve the challenges associated with solving the Nernst-Planck equation in its full form. Those issues are discussed in detail elsewhere,⁴⁴  but suffice to say that reduced-order models which considered only one transport mode to be dominant have many advantages. Under certain conditions,^44-47  the Laplace Equation can be used to accurately calculate the current and potential distributions. The key to applying the Laplace Equation in systems in which its restrictions are not strictly met is the use of electrochemical boundary conditions that account for the altered chemistry at the surface. In this way, steady-state calculations of corrosion processes can be made and interpreted.

The atmospheric galvanically-induced IGC of AA5XXX alloys has been identified as an important failure mode for these alloys when in marine service. The galvanic couple is formed when high-strength steel fasteners are used to connect AA5XXX structures to ship decks. In order to model this phenomenon, Mizuno and Kelly⁴⁵  determined the polarization behavior of AA5083-H131 with different levels of sensitization in different chloride concentrations, with one example shown in Figure 10(a). The kinetics were abstracted and used along with similarly generated kinetics for the 4130 steel as the boundary conditions for the modeling. The different chloride concentrations simulate the equilibrium composition of NaCl solutions at different RH.

The top image of Figure 10(b) shows a cross section of a sensitized AA5083-H131 after exposure to a thin electrolyte layer while coupled to 4130 steel. Quantitative analyses of the IGC damage are shown just below the micrograph. The bottom image of Figure 10(b) shows the potential distribution across the sample as measured with a scanning Kelvin probe.⁴⁸  At distances greater than 4 mm from the steel/aluminum alloy interface, there is no influence of the galvanic couple. No IGC is observed at these longer distances despite the fact that the material is highly sensitized. This phenomenon is due to the insufficient cathode current available in this small area.⁴⁹  The potential distribution was then used to predict the IGC damage⁴⁵  based on the known dependence of IGC rate on potential.⁴⁸  Figure 10(c) shows the comparison of the modeling using the Laplace equation as the governing law with chloride-concentration-dependent electrode kinetics. Considering the limited input data and the simplicity of the model, the agreement between measurement and model is satisfactory. Figure 10(d) illustrates how the model can be used to assess the impact of salt loading density (LD), RH, and degree of sensitization (DOS) on the IGC damage expected.

Noble metal fasteners are often used in aluminum alloy aerospace structures. Unfortunately, although these fasteners provide high strength, they also represent a large cathodic driving force for localized corrosion of the precipitation-hardened aluminum alloys they join. Matzdorf, et al.,⁵⁰  designed a test sample that allowed rapid screening of the impact of these fasteners on primed and coated aluminum alloys, as shown in Figure 11(a). This sample has been used to assess the impact of different coatings, primers, and surface pretreatments.^51-52  Recently, Marshall, et al.,⁴⁶  successfully modeled this sample design and its performance in B117 testing. To do so, they used electrochemical kinetics for the AA7075-T6 generated by first creating extensive localized attack using polarization to an elevated potential prior to potentiodynamic scanning. In this way, the appropriate surface chemistry was created and then the electrochemical kinetics were measured for that chemistry. For the stainless steel and Ti-6Al-4V bolts and washers, they used cathodic kinetics which took into account the acceleration of diffusion-limited oxygen reduction that occurs in thin electrolyte layers. After validating the model on B117 results from their own testing, they applied it to the altered Matzdorf geometry used by Feng, et al.,^51-52  in which the total galvanic current between the aluminum alloy plate and the stainless steel fasteners was measured. As shown in Figure 11(b), the steady-state model well captures the steady-state current that was measured for two different configurations, one with two stainless steel fasteners, the other with four stainless steel fasteners. The ability of the model to accurately capture these different configurations provides confidence in its use.

Corrosionists have long been accused of being like economists; we can explain everything and predict nothing. Although not completely true of either profession, there is some truth in the statement when applied generally. The prediction of the time-course of corrosion, especially under service conditions, has been and continues to be a vexing problem. The service environment is rarely constant, and the corroding surface can evolve as well. The most daunting corrosion prediction challenge is that for long-term storage of high-level nuclear waste. The canisters which hold the waste need to last hundreds of thousands of years to prevent the release of unacceptable quantities of radioactive material into the biosphere. Developing a transient model that could be accurate to that kind of time frame seems impossible. The accumulation of even small errors would compound to the point that the uncertainty would make the result of limited utility. The additional challenge of validating such a model would seem to close the door to the possibility of predicting this critically important corrosion problem. If the door is closed, go through the window, as poker legend Doyle “Texas Dolly” Brunson said. Rather than predicting the time course, Chen, et al.,^53-54  decided to take a bounding approach. That is, they sought to determine what is the maximum pit size that could form under a thin electrolyte film. The rationale for the idea of the existence of such a bounding size is illustrated in Figure 12.

A pit (or any localized corrosion site) is essentially a galvanic couple between a passive external surface serving as the cathode, and the pit serving as the anode. There are constraints on both parts of the galvanic couple. Galvele formulated^55-57  the constraint on the pit in that it needs a certain minimum current (I_LC) to maintain the aggressive chemistry required for propagation. When the cathode is covered by a thin electrolyte layer, the current passing from different locations on the surface causes ohmic drop which increases with distance from the pit. At some point, the cathode potential approaches its open circuit potential. The open-circuit potential provides a bound on the cathode which leads to a bound on the maximum current (I_cath) that the cathode can supply. Combining that maximum cathode current with the minimum anode current from the pit naturally leads to the existence of a maximum pit size (Figure 12[b]).

The literature contains numerous examples of limiting pit sizes^58-60  formed under open-circuit conditions, including in natural environments. A laboratory validation⁶¹  using exposure of stainless steels to thin electrolyte layers of ferric chloride demonstrated both the existence of a limiting pit size, and the fact that naturally occurring pits do not grow under a salt film. If the pit grew under a salt film, the maximum pit size possible would have been 96 µm as shown in Figure 12(c). The observed limiting pit size was twice that. That result shows that either propagating pits do so in a concentrated chemistry greater than 50% of saturation, but less than 80% of saturation, or there is more than one pit active at the same time within proximity of one another. The inset image of an area of the pitted surface confirms that the proximity of other pits occurs. The idea of limiting sizes for localized corrosion continues to be explored.^62-63 

Looking forward with regards to measurements, developments in measurement methods will certainly continue. Operando measurements will likely become more common, including expanded use of liquid-phase TEM^64-65  to monitor nanoscale morphology, alloy composition, and electrochemistry.⁶⁶  That said, the high precision of ex situ methods for chemical composition and the better matching to microstructural scales will ensure that optical microscopy, including profilometry, will continue to be important tools.

For modeling of local chemistry and electrochemistry, there is a critical unmet need for broader education in the use of computational modeling, from ab initio calculations^67-68  to kinetic Monte Carlo^69-70  to FEM modeling.^71-73  Just as the use of electrochemical techniques applied to corrosion expanded dramatically in the latter part of the 20th and early part of the 21st centuries, the field is poised to have a similar expansion in the use of modeling. As more and more commercial software becomes available, the accessibility of modeling tools will encourage wider adoption. For this expansion to occur, short courses need to be developed which can help introduce scientists and engineers to the various modeling methods, with special emphasis on their limitations. Analogous short courses in electrochemical techniques were important in expanding their use.

New means of validating the results of models will no doubt develop and be applied to a wider array of models described above. As modeling becomes more prevalent, better connections between models and experiments will be possible. Continuing increases in computing power will also play a critical role in models being able to be applied to more realistic scenarios in terms of geometry, size scale, and complexity of boundary conditions, including dependencies on variables (e.g., pH, [Cl⁻]). This increased computing power will also help the field realize the impact of true multiscale models which are automatically linked, allowing ab initio, atomic-scale models to be linked through the microscale to the mesoscale and beyond to the macroscale. Such models will allow the rapid development of new alloys, inhibitors, and other corrosion protection systems.

I have been blessed with an unbroken chain of wonderful mentors, colleagues, and students, not to mention the sponsors whose support made all of the work possible. Space limitations will prevent me from naming all of them, but Glenn Stoner, Pat Moran, John Scully, and Rick Gangloff deserve special mention for the decades of friendship and mentorship.