Hydrogen diffusion and the percolation of austenite in nanostructured bainitic steel

The diffusion of hydrogen in austenite is slower than in ferrite. Experiments have been conducted to study the behaviour of hydrogen in a nanostructured steel sample consisting of a mixture of thin plates of bainitic ferrite and intervening films of retained austenite, with the latter phase present in a quantity larger than the percolation threshold, i.e. it has three-dimensional connectivity. The structure was then heat treated to control the fraction of austenite, and hence to study the role of hydrogen when the austenite decomposes below the value required to sustain percolation. The experiments have involved both thermal desorption analysis and permeation, and when combined with theoretical analysis, indicate a significant influence of percolating austenite in hindering the passage of hydrogen into the steel during hydrogen charging, and its permeation through the composite nanostructure. The effect is not as large as might be expected from a simple comparison of independent data on the diffusivities of hydrogen in the two lattices, because the effective diffusivity in ferrite is found to be much smaller than in the defect-free ferrite, owing to trapping effects. The morphology of the austenite is demonstrated to play a role by comparing with a sample containing a larger volume fraction of austenite but present as isolated grains which are ineffective to the permeation of hydrogen.

The diffusion of hydrogen in austenite is slower than in ferrite. Experiments have been conducted to study the behaviour of hydrogen in a nanostructured steel sample consisting of a mixture of thin plates of bainitic ferrite and intervening films of retained austenite, with the latter phase present in a quantity larger than the percolation threshold, i.e. it has threedimensional connectivity. The structure was then heat treated to control the fraction of austenite, and hence to study the role of hydrogen when the austenite decomposes below the value required to sustain percolation. The experiments have involved both thermal desorption analysis and permeation, and when combined with theoretical analysis, indicate a significant influence of percolating austenite in hindering the passage of hydrogen into the steel during hydrogen charging, and its permeation through the composite nanostructure. The effect is not as large as might be expected from a simple comparison of independent data on the diffusivities of hydrogen in the two lattices, because the effective diffusivity in ferrite is found to be much smaller than in the defect-free ferrite, owing to trapping effects. The morphology of the austenite is demonstrated to play a role by comparing with a sample containing a larger volume fraction of austenite but present as isolated grains which are ineffective to the permeation of hydrogen.
2014 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/ by/3.0/, which permits unrestricted use, provided the original author and source are credited.

Introduction
Nanostructured bainite consists of exceptionally thin platelets of bainitic ferrite separated by films of retained austenite, a structure produced by heat treatment alone [1][2][3], and yet resulting in an interfacial area per unit volume that is greater than that associated with the majority of severe plastic deformation experiments [4]. The tensile behaviour is characterized by uniform elongation accompanied by little, if any necking [5][6][7]. The uniform elongation therefore is usually equal to the total elongation. In general, tensile tests reveal a dependence of ductility upon transformation temperature, and an inverse correlation with strength [6,8]. In carbide-free bainitic steels, such as superbainite, the volume fraction and morphology of austenite are controlling factors in the ductility [9,10]. In particular, the TRIP (transformation-induced plasticity) effect comes into play-retained austenite transforms into martensite under tensile loading. This process is influenced by various factors relating to austenite stability, such as grain orientation, morphology, carbon content and scale [7,[11][12][13][14][15]].

Motivation
It is possible to use standard calculation methods [16][17][18] to follow the variation in retained austenite during tensile tests on nanostructured bainite as a function of plastic strain, accounting for austenite composition and the temperature at which deformation occurs. It then becomes evident that the strain at the point of fracture consistently occurs when the austenite volume fraction is reduced to about 0.1, a conclusion supported by in situ tests in which phase fractions are monitored during tensile testing [15]. It has therefore been argued that in these steels, the failure in tension occurs when the retained austenite loses three-dimensional [19]. This means that regions of austenite become geometrically isolated from one another, to the extent that it is no longer possible for applied stresses and strains to 'trace a path' that passes only through the austenite phase. At this point, the stresses must additionally be borne by the more brittle martensite that results from the deformation-induced transformation of austenite, leading to fracture. Measurements of austenite volume fraction at various locations in tested tensile specimens support this idea [7]. The need to control the mechanical stability of retained austenite in automotive TRIP-assisted steels in order to ensure an optimized uniform ductility has long been recognized [20], but the important difference here is that with bulk nanostructured bainitic steel, the ductility is essentially all uniform and a different mechanism seems to govern ultimate fracture.
Our goal in this work was to use thermal desorption analysis (TDA) techniques on hydrogencharged samples to test the hypothesis that the austenite in a nanostructured steel loses percolation at a volume fraction of about 0.1. A great deal of peripheral information on the distribution of hydrogen in the two-phase ferrite plus austenite mixture is obtained at the same time. The logic of the experiments is based on the fact that the diffusivity and solubility of hydrogen is dramatically different in the two phases, with diffusion being much slower in austenite. It follows that in a percolating structure, less hydrogen should be able to penetrate the sample for a given charging time, because the hydrogen must necessarily diffuse through austenite at some stage.

Experimental methods
TDA is a technique normally used to study hydrogen trapping states [21,22]. Specimens are charged electrolytically with hydrogen and then heated at a constant rate, with any desorbed hydrogen detected using gas chromatography. Strongly trapped hydrogen is released at a higher temperature, and the total released can be detected by integrating the data over the time period of the experiment.
The chemical composition of the steels used in this experiment are given in table 1. The vast majority of experiments reported here have been conducted on Alloy 1, but because of the limited material available, some of the experiments are on Alloy 2. Both can be heat-treated to produce the required nanostructure, as shown previously [1,2,23,24]. Alloy 1 was austenitized in an argon tube furnace at 900 • C for 30 min, followed by isothermal transformation at 200 • C for 96 h. Specimens measuring 55 × 12 × 4 mm were produced from the isothermally transformed sample of Alloy 1. Alloy 2 was similarly austenitized at 850 • C for 1 h and then isothermally transformed at 200 • C for 168 h, a longer period than Alloy 1, because it does not contain the Co or Al that accelerates transformation [24]. The samples were in this case produced as 55 × 20 × 1 mm size because they were destined primarily for permeation experiments, and to establish the saturation concentrations of hydrogen.
In order to vary the volume fraction of austenite, the isothermally transformed samples were tempered to various extents as detailed in table 2. The austenite volume fraction (V γ ) was measured using quantitative X-ray analysis. After tempering, the samples were ground with progressively finer grades of silicon carbide paper, up to 4000 grit. They were then etched to remove deformed layers generated while grinding. X-ray work was carried out by step-scanning on a Philips PW150 X-ray diffractometer using unfiltered Cu K α radiation, operated at 40 kV and 40 mA. The 2θ scanning speed was 0.1 • min −1 over a range of 40-140 • . Rietveld refinement was done using Panalytic HighScore Plus software, and retained austenite carbon content was calculated from the measured lattice parameters as described by Dyson & Holmes [25].
The samples were polished with 220-grit and 800-grit silicon carbide papers, cleaned with high-purity ethanol and electrochemically charged with hydrogen in an aqueous mixture of 3% NaCl + 0.3% NH 4 SCN, using a current density of 1 A m −2 . The standard charging time used was 24 h, although this was varied from 1-6 h in the later stages of the experiment, in order to study the effect of charging time. Figure 1a shows completion of charging and placing the sample in the furnace was 15 min. Samples were heated to 700 • C at a rate of 100 • C h −1 . The sample gas was analysed at 3-min intervals using helium as a carrier gas. This transfers the desorbed hydrogen into the gas chromatograph where it is monitored by a detector. The hydrogen content was calibrated using a standard gas mixture. In the TDA results presented here, the quantity of hydrogen desorbed is expressed in units of ppmw, representing the parts per million by weight of hydrogen relative to the sample weight.
The electrochemical hydrogen permeation cell used to measure the effective diffusivity of hydrogen through the variety of structures is illustrated in figure 1b. The test was performed according to the ISO17081 standard. The detailed procedure is described elsewhere [26] but the test permits, in particular, the effective diffusivity and permeability to be determined by monitoring the current density through the sample as a function of time. The steady-state flux of hydrogen through the sample is J ss = i ss /nF, where i ss is the steady-state current density, F is the Faraday constant and n the number of electrons that participate in the electrochemical reaction. The permeability is then simply J ss L, where L is the sample thickness. A time lag t l is defined to be that required to achieve 0.63 of the steady-state current density, and the effective diffusivity as  calculated from this time lag is simply D l = L 2 /6t l . Alternatively, both the effective diffusivity and permeability may be calculated using the break-through method, in which the time t b is defined as that at which the current density begins to increase after beginning the experiment. The effective diffusivity calculated in this way is Transmission electron microscopy was carried out on a JEOL 200CX microscope. Samples were prepared by slicing thin sections from heat-treated specimens and grinding them with 340-grade silicon carbide papers until the thickness reached 70 µm. Circular samples 3 mm in diameter were punched out from these sections and ground down with 2500-grade silicon carbide papers until the thickness reached 50 µm. These samples were electropolished at 20 V in a twin jet unit with a 15 vol% perchloric acid-85 vol% ethanol electrolyte.
Vickers hardness testing was carried out at a load of 30 kg. Samples were cut from the tempered specimens and mounted in resin. They were polished to a mirror finish with 600-grade, 800-grade, 2500-grade and 4000-grade silicon carbide papers. Eight hardness measurements were taken from each sample and averaged.

Results and discussion (a) Transmission electron microscopic characterization
Both of the nanostructured alloys were characterized using transmission electron micrography and hardness testing. Much more detailed structural characterization has been reported elsewhere [3,15,27] and so is not repeated here. The as-transformed structure of Alloy 1 is shown in figure   structure also contains some regions of fine blocky-austenite; the reported ratio of film-austenite to blocky-austenite in this alloy, when transformed at 200 • C, is approximately 3 : 1 by volume [3]. The as-transformed structure of Alloy 2, shown in figure 3a, is coarser, with an average plate thickness of 83 nm. Bainitic steels, such as the alloys studied here, are resistant to tempering, with little reduction in hardness or austenite volume fraction below 450 • C and significant softening only taking place at temperatures above 500 • C [27][28][29]. This is reflected in the fact that the hardness remains high in both alloys, even after tempering at 500 • C for 1 h. However, the reduction in hardness following the tempering treatments signals the onset of austenite decomposition, as confirmed by X-ray analysis (table 3). Attempts to temper Alloy 1 at temperatures below 450 • C did not produce a particularly significant change in austenite volume fraction. Therefore, no such treatments were applied to Alloy 2.
Tempering that leads to carbide precipitation from the decomposition of austenite, also hinders the coarsening of the ferrite plates since the carbides in fact are located precisely at the plate boundaries, thus preventing significant coarsening [27]. Therefore, the plate structure is preserved after all the tempering heat treatments, as illustrated in figures 2 and 3. All the tempering treatments listed in table 3 led to carbide precipitation in Alloy 1 (figure 2b-d). The carbides were confirmed to be cementite using electron diffraction (figure 4); note that the as-transformed material is carbide-free because of the large silicon concentration in the steel. Cementite was also detected in Alloy 2 after all tempering treatments at 500 • C (figures 3d and 4c,d). Although no carbides were found after tempering Alloy 2 at 450 • C, it is possible that the volume fraction of any carbides is smaller than can be easily detected.    A topologically different kind of a mixture of ferrite and austenite, where the latter is distributed as non-percolating grains was generated in a dual-phase steel of composition Fe-0.11C-6.17Mn-0.98Al wt%, by isothermal heat treatment at 620 • C for 24 h, as illustrated in figure 5. This alloy is interesting because the morphology is such that austenite grains are isolated, with continuity for any hydrogen to permeate through the ferrite where the hydrogen is much more mobile.

(b) Thermal desorption analysis
There are two essential categories of hydrogen in steel: that which is diffusible and can in principle escape from the steel, and the remainder, which is said to be trapped and is much more difficult to remove. The concept of hydrogen-trapping comes from Darken & Smith [30], who established that the passage of hydrogen through a steel is hindered by lattice imperfections (dislocations, interfaces) that tend to attract and bind it, thus rendering it immobile at temperatures where it should diffuse readily. The binding energy is defect-dependent [31]. In two-phase mixtures of austenite and ferrite, of the type relevant to this studies, the diffusivity of hydrogen in perfect austenite is far smaller than in ferrite, although its solubility in austenite is much greater than in ferrite. The austenite can, in this sense, be regarded as a macroscopic trap for hydrogen, relative to the ease with which hydrogen can diffuse in ferrite [32]. Therefore, in a thermal desorption experiment, the hydrogen in austenite is expected to be detected at a higher temperature during heating than that from ferrite, which may be released at close to ambient temperature.
Given the complexity of the problem, several experiments were carried out to understand the response of the hydrogen-charged and -uncharged samples when subjected to TDA; these are reported first, before a discussion of the main results.

(c) 'Calibration' experiments
A sample of Alloy 1 was subjected to TDA before deliberate charging with hydrogen. Figure 6a shows the TDA output from an uncharged untempered sample with V γ = 0.21, heated at 100 • C min −1 . The little hydrogen evolved does so at temperatures below 50 • C, which means that it is not dissolved in the steel since it is easily desorbed. The origin of this low-temperature evolution is not clear, but the quantity desorbed between room temperature and 300 • C amounts to just 0.17 ppmw and can be neglected. Figure 6b details the experiments done to assess the reproducibility of data from physically distinct samples that have been identically heat-treated and charged. The variability is likely to be due to specimen rather than instrument scatter, since the structures involved are complex.
A significant goal of this work was to see whether the films of austenite present in the nanostructure impede the motion of hydrogen by percolating through the structure. Clearly, in order to assess this, the structure must not be saturated with hydrogen prior to the TDA. At the same time, sufficient hydrogen must be charged to ensure a good signal-to-noise ratio. The samples of Alloy 1 were therefore made 4 mm in thickness, so as to allow sufficient hydrogen to enter the specimen surface, but leaving the core with a reduced concentration. Both samples were charged using the same current density. In this way, the amount of hydrogen that effuses during TDA is a reflection of its permeation into the sample. Figure 6 compares the thickness effect for two samples, both transformed into nanostructured bainite at 200 • C, albeit from slightly different alloys. The change in charging time from 24 to 72 h did not alter the hydrogen content of the 1 mm thick sample, proving saturation. The total hydrogen desorbed 1 mm thick sample was measured to be 11.1 ppm, whereas that for the 4 mm thick sample was 19.6 ppm. The latter value should be much larger (at least 67 ppm given that the 4 mm sample is six times heavier) to match the saturation level of the thinner sample. The higher hydrogen content of the 4 mm sample indicates a somewhat greater penetration depth, as the penetration in the thinner sample cannot exceed 0.5 mm.

(d) Role of retained austenite
We first note that heating at the rates considered here, to temperatures up to 300 • C, is sufficient to detrap hydrogen that is dissolved in austenite, and that from defects such as dislocations and interfaces. Such hydrogen is said to be reversibly trapped and is of particular interest in embrittlement phenomena. TDA experiments were conducted on all the samples described in table 3, after charging with hydrogen for 24 h, at 100 • C min −1 . In figure 7, the three samples that contain essentially similar behaviour; the two samples tested with V γ = 0.07 show different behaviour, with one curve replicating the larger austenite data and the other consistent with a larger quantity of hydrogen penetrated into the sample during the charging process. It follows that these data alone cannot reveal differences due to the austenite content nor phenomena associated with the percolation threshold.
Many further experiments were conducted to verify the general impression regarding the influence of retained austenite; the data are summarized in table 4. It should be noted that the amount of hydrogen desorbed during heating over an appropriate temperature range is a reflection of the amount charged into the sample, as long as saturation does not set in. Figure 8a shows that the amount of hydrogen desorbed from the sample over the range 25-300 • C, as expected, increases with the charging time. The retained austenite content has minimal influence in the early stages of charging, presumably because there is sufficient exposed ferrite at the sample surface, of the 4 mm thick specimens. However, further penetration during charging is clearly more difficult for the sample with V γ = 0.21 because the percolating austenite is a barrier to the diffusion of hydrogen in the sample.
Retained austenite should be a hindrance to the diffusion of hydrogen in a two-phase structure, and more so if its fraction is above a percolation threshold since there would then be no easydiffusion paths. The data plotted in figure 8b as a function of retained austenite could be argued to further support the estimate [8] that austenite loses percolation in the bainitic sample when V γ 0.1. The desorbed hydrogen content (a reflection of the hydrogen that managed to enter the specimen) is clearly large when the percolation is lost.
However, this conclusion is not entirely established. It is not known whether there should be an abrupt change in hydrogen penetration into the sample when the percolation of austenite is lost, or whether the change should be gradual. A simple approach to this problem is illustrated in figure 9, where three scenarios are illustrated. 1 The first is the case where percolating film-like austenite dominates the movement of hydrogen, the second where both ferrite and austenite are exposed directly to the hydrogen and the third where the austenite films have been reduced to a fraction below the percolation threshold. Consider first the non-percolating structure (figure 9b) where both γ and α are exposed to the free surface. structure' is then where A is the fraction of the free surface occupied by ferrite, φ = J α /J γ . By contrast, the flux J p through the structure illustrated in figure 9a, where the austenite in effect is a significant barrier to the infusion of hydrogen, is simply J γ , which will vary inversely with the thickness t of the austenite film because that determines the magnitude of the concentration gradient given fixed boundary conditions. The thickness is here assumed to be proportional to the volume fraction of V γ of austenite as long as the latter is above the percolation threshold The situation illustrated in figure 9c is where the fraction of the film of austenite at the surface is less than unity, i.e. because the total fraction of austenite is less than the percolation threshold. For that case, the flux J np is given by The fluxes naturally depend on diffusion coefficients, and if it is assumed that undeformed ferrite and austenite have the diffusivities at 300 K of D α = 1.26 × 10 −8 m 2 s −1 [31] and D γ = 2.65 × 10 −16 m 2 s −1 [34], then J α J γ and there should be an overwhelming increase in the net flux as soon as percolation is lost. This does not in fact happen ( figure 8). The reason is that the austenite-free structure obtained by tempering the mixture of bainitic ferrite and austenite still has a fine structure with dislocation and interface traps (the cementite itself is not a prominent trap in tempered structures [35]). Therefore, the appropriate D α is that measured using permeation experiments on this structure, reported later in this paper to be D α = 10 −11 m 2 s −1 ; note also that this value is consistent with assessed data [36, fig. 12]. The gradient of concentration is also determined by the solubility of hydrogen in the phase as a result of charging, the atom fractions taken to be x α = 2.8 × 10 −8 and x γ = 7.0 × 10 −5 [37]. Using these values, the ratio φ = 15 so equations (4.1) and (4.3) were used to calculate the variation in net flux as a function of the retained austenite content assuming φ = 15 when V γ = 0.21, the largest volume fraction of austenite in the bainitic samples. Figure 10 shows the results, where for the case with V γ = 0.07, the area fraction A exposed to ferrite was taken to be 0.3, on the basis that no ferrite is exposed when V γ = 0.1 since this is at the percolation threshold. It is evident that although there will be a significant change in the passage of the hydrogen through the sample, the change is not too abrupt, consistent with the data illustrated in figure 8.

(e) Permeation experiments
The raw permeation data from experiments conducted on Alloy 2 samples heat-treated as described in table 2 are presented in figure 11a. The derived effective-diffusivity data plotted in figure 11b confirm the general expectation that the presence of the films of retained austenite reduces the effective diffusivity of hydrogen through the sample very significantly, irrespective of the analysis method. The two methods give different answers because the time-lag method treats the problem essentially as a random walk of hydrogen through the steel, whereas the breakthrough technique equation is derived on the ability of the hydrogen to begin arriving on the other (output) side of the sample. The smaller value of the time-lag diffusivity D l < D b indicates that the hydrogen transport is governed by diffusion through the sheet rather than by surface kinetics. Permeability represents the ability of a material to act as a hydrogen barrier, determined from the steady-state current through the sample. This is in contrast to diffusivity, which describes the kinetics of atomic motion. The permeability increases dramatically as the retained austenite content is reduced and percolation is lost, as shown in figure 11.
In all of the TDA experiments on charged specimens, no significant hydrogen-desorption was observed in the temperature range 300-400 • C.

(f) Decomposition of retained austenite
It was also intended to study what happens to the hydrogen dissolved in the austenite, when the austenite associated with the bainitic structure decomposes into a mixture of cementite and ferrite. Previous work on nanostructured bainite has shown that the decomposition occurs during isothermal heat treatment at temperatures of 400 • C or above [27,29,38,39]. However, these experiments involve continuous heating, so dilatometric experiments were done to measure the temperature range over which the substantive decomposition of austenite occurs. Given that the retained austenite is enriched in carbon to at least 1.2 wt% [40], its decomposition in the manner described was calculated to lead to a net contraction using the equations given in [41]. Figure 12 shows that for both heating rates studied, the austenite decomposes in a temperature range of approximately 420-530 • C.
At the kind of heating rates used in this work, even strongly trapped hydrogen (binding energy ≈ 50 J mol −1 ) should escape from the sample by the point where it reaches 300 • C [33,42]. Any hydrogen which is even more strongly trapped is unlikely to contribute to the deterioration of mechanical properties [43]. The TDA curves presented earlier show that much of the harmful hydrogen is removed by 300 • C. However, to investigate whether the decomposition of austenite into a mixture of cementite and ferrite releases strongly trapped hydrogen, many of the experiments were continued to temperatures as high as 800 • C.
Unfortunately, the experiments revealed rather strange results, both over the temperature range of interest from the point of view of austenite decomposition, and beyond (figure 13). It was suspected that oxidation plays a role in creating these effects. The 99.9995% pure gas contains, according to the supplier, some 2 ppm of water. The thermodynamic analysis presented in appendix A shows that with an assumed value of p H 2 /p H 2 O = 0.1 given the very low concentration of water in the gas, it is possible in all the steels studied here to reduce the water into hydrogen, with the formation of Si, Al and Mn oxides (figure 13b). FeO can also form but the driving force for oxidation is rather small so it is likely that the other oxides form preferentially.
To test this experimentally, a sample of the duplex steel was polished, charged and tested to generate a TDA curve (figure 13c). The same sample was then polished and reheated without  hydrogen charging in the TDA, and on the third occasion, the same sample was reheated without charging or polishing in the TDA. The data indicate that the high-temperature effects are associated with oxidation, and that the presence of an existing oxide film tends to diminish the effect, as seen in the third experiment ( figure 13c(iii)).

Conclusion
The essential purpose of the work was to study the passage of hydrogen through a mixture of bainitic ferrite and films of austenite, in which the austenite fraction and morphology was such that it forms a continuous barrier between the ferrite plates, i.e. the austenite percolates through the mixture. The fraction of austenite could be controlled by heat treatment so that corresponding samples where the fraction is less than that required for percolation could be studied. The following conclusions can be reached on the basis of thermal desorption experiments, permeation experiments, theoretical analysis and comparison between samples containing radically different morphologies of austenite at approximately the same volume fraction: (i) The study of partially hydrogen-charged specimens of nanostructured bainite using thermal desorption shows that the amount of hydrogen absorbed by the steel, at long charging times, is smaller when the austenite fraction is beyond the percolation threshold. where 'Me' represents the metal, x and y depend on the type of oxide as listed in table 5, a represents the activity of the material identified by the subscript and R is the gas constant. The activity of the pure oxide was always taken to be unity, and those of the metallic elements in the steel concerned calculated using ThermoCalc [46].