1. Introduction
Low-temperature carburization is a potent method of surface engineering. It can be applied to alloys that contain significant fractions of alloying elements with a high affinity for carbon. The prototype of such an alloy is the austenitic stainless steel AISI-316L (UNS designation S31603) with an atom fraction
of Cr as the high-carbon-affinity, carbide-forming element. The properties of finished AISI-316L alloy parts can be improved significantly by exposing them to an ambient that infuses carbon into a subsurface zone to generate a solid solution of interstitial carbon. However, owing to the high stability of Cr carbides, the equilibrium-solubility limit of carbon in this alloy is very small, corresponding to a carbon atom fraction
at 300 K. While
increases with increasing temperature, solutionizing at high temperature and subsequent quenching does not provide an easily realizable processing route: The high stability of Cr carbides substantially reduces
compared to e.g., the binary system Fe–C. Even at a high temperature,
is significantly smaller than the technically desired levels of e.g.,
. Moreover, with an increasing carbon fraction, the “C” curve of the time–temperature–transformation diagram that represents carbide precipitation moves to higher temperatures and shorter times. Therefore, (hypothetical) carburization at high temperature and subsequent quenching, unless carried out at cooling rates difficult to realize, causes the precipitation of Cr-rich carbides [
1,
2]. Generally, carbide formation is undesirable because it compromises important properties, especially fatigue and corrosion resistances (“sensitization”).
The desired outcome of a concentrated yet carbide-free subsurface solid solution of interstitial carbon can be obtained by carburizing at a low temperature. “Low,” in this context, means low enough to suppress carbide precipitation by immobilizing Cr, but still high enough to enable carbon diffusion into technically useful depths within industrially feasible processing times. Idealizing these nonequilibrium conditions as a state with no metal atom mobility at all, the system is then confined to a trajectory leading to a “carbon paraequilibrium”, rather than conventional thermodynamic (or “ortho”) equilibrium [
3]. With this approach, treatment times of order
Ms (1 day) can infuse carbon to an average depth
m. Within an
m deep zone below the alloy surface, a carbide-free interstitial solid solution forms in which the (smoothly graded) carbon profile significantly exceeds the equilibrium solubility of carbon. At the surface, the carbon fraction can reach
, corresponding to
at
K. Similar considerations apply to the low-temperature infusion of nitrogen or a combination of nitrogen and carbon.
The interstitial solute-rich subsurface zone, denoted as a “case”, can dramatically improve the performance of alloy parts in technical applications. Typical benefits include a fourfold increase of surface hardness [
4], the reduction of wear volume by a factor of order
[
5], an increase in high-cycle fatigue life by a factor of order
[
6], and a significant enhancement of corrosion resistance, particularly in seawater [
7,
8,
9].
For industrial applications of low-temperature carburization, gas-phase processes are important, i.e., exposing alloy parts contained in a furnace to a gas that provides elemental carbon or nitrogen to diffuse into the alloy. Swagelok Company (Solon, OH, USA) invented a low-temperature atmospheric pressure gas-phase carburization process for AISI-316L stainless steel [
4,
10,
11]. However, for alloys with a Cr fraction suitably high for low-temperature carburization (or nitridation or nitro-carburization), successful carburization from a gas phase first requires surface “activation”: The surface of AISI-316L and related “stainless” alloys is typically covered by a
nm thick Cr-rich oxide layer [
12] that (i) passivates the alloy against low-temperature corrosion and (ii) constitutes a diffusion barrier to carbon and nitrogen (
Figure 1).
Another potential obstacle to carbon or nitrogen infusion into alloy parts can originate from machining-induced severe plastic deformation, poor crystallinity, and contamination in a zone directly below the surface, known as the “Beilby” layer (
Figure 1) [
13]. This has been confirmed by the results of Ge et al. [
14,
15] on acetylene-based low-temperature carburization of AISI-316L.
Our current understanding is that the passivating oxide layer must be removed before carbon or nitrogen can diffuse into the alloy. Removal of the Beilby layer may not be critical, but can boost the efficacy of low-temperature carburization/nitridation/nitrocarburization. Accordingly, both layers should be removed prior to the infusion of interstitial solute. An established method for such “surface activation”, i.e., removing the passivating oxide layer and potentially the Beilby layer, is the application of hot gaseous HCl (hydrochloric acid). For example, the low-temperature carburization process established by Swagelok uses gaseous HCl at 520 K directly before the infusion of interstitial solute [
4]. However, it was empirically found that a single in situ exposure to gaseous HCl does not sufficiently activate the surface of AISI-316L. Activation by gaseous HCl is substantially more effective after some carbon has already been infused into the specimen. For this reason, the Swagelok process operates with “double activation”, consisting of two subsequent gaseous HCl activation steps interlaced with an initial step of low-temperature carburization [
4,
11]. The total time for activation is 32 ks (9 h), one-third to one-half the low-temperature carburization time of
Ms (
h). Moreover, gaseous HCl is aggressive and can damage non-corrosion-resistant processing furnace hardware.
In this article, we report the discovery that the industrially established in situ surface activation process using gaseous HCl can effectively be replaced by the following ex situ procedure: (i) immersing the alloy part in a liquid etchant (concentrated aqueous HCl) at room temperature, (ii) stopping the etch process by directly transferring the specimen into a reservoir of CHOH (without prior drying), (iii) loading the specimens into a furnace for gas-phase low-temperature carburization, purged with dry N gas, (iv) and heating them and exposing them to streaming carburizing gas. In this way, alloy parts can be directly carburized by exposure to carburizing gas (CH, H, N) at 723 K without prior application of gaseous HCl. The total additional processing time for activation, Step (i) and (ii), is less than 1 ks (17 min)—one-hundredth of the low-temperature carburization time and times shorter than in the Swagelok process.
That low-temperature carburization can be successfully carried out directly after ex situ etching and cleaning/storing leads to the hypothesis that the specific cleaning/storing procedure we are applying inhibits the spontaneous reformation of the regular passivating oxide layer. Instead, this procedure seems to provide “provisional” passivation that inhibits the formation of the usual Cr-rich passivating oxide film and is readily removed when the alloy is subjected to low-temperature carburization, by the applied temperature or by reaction with the carburizing gas.
The alloy for which we demonstrate the efficacy of “provisional passivation” is AISI-AL-6XN (UNS designation N08367,
Table 1), a superaustenitic stainless steel containing higher fractions of Ni and Mo than AISI-316L. Enhanced fractions of Ni and Mo provide superaustenitic stainless steels with better corrosion resistance than martensitic-, ferritic-, and 300-series austenitic stainless steels. At the same time, they have higher strength than austenitic steels AISI-304 (UNS designation S30403) and AISI-316, making them attractive for demanding applications. Additionally, superaustenitic stainless steels have excellent strength and oxidation resistance at elevated temperatures [
16]. The hardness of AISI-AL-6XN, on the other hand, is not significantly higher than that of e.g., AISI-304 and AISI-316 [
8]. The lack of higher hardness is particularly limiting in applications demanding high wear resistance. Previous work has demonstrated that the surface hardness of AISI-AL-6XN can be significantly improved by low-temperature carburization [
17] while not compromising their excellent corrosion resistance [
18]. Therefore, low-temperature carburization provides potential for expanding the range of technical applications in which AISI-AL-6XN can serve.
3. Results
Figure 2 presents cross-sectional light-optical microscopy images of all samples at the same magnification. The left column contains the micrographs from specimens that were activated by etching in aqueous HCl, whereas the right column shows micrographs from specimens prepared without activation. Micrographs in the same row are from specimens with the same mechanical surface preparation. The two micrographs in the top row are from electropolished samples, followed by samples ground by SiC paper with increasing coarseness (P4000, P800, P180).
All micrographs reveal similar grain structures in the alloy core, i.e., in depths to which no significant amounts of carbon could diffuse within the low-temperature carburization processing time. (Differences in appearance are merely the result of under- or overetching.) While a Beilby layer introduced by surface machining typically involves a subsurface zone of reduced grain size, this is not observed here: Regarding grain morphology, the zone directly below the surface appears to be no different from the alloy core. Some of the micrographs exhibit a bright band below the surface, or fragments thereof (
Figure 2d,f). This indicates the formation of a “case”, i.e., a zone with a carbon fraction
high enough to provide corrosion resistance against the metallographic etchant, preventing it from developing the grain structure. However, what appears to be a uniform layer with a sharp boundary to the alloy core does not properly reflect the properties of the underlying carbon profile, which is actually smoothly graded, as expected for a diffusion profile (see scanning Auger microprobe data below). Rather, the apparently sharp border between the case and the core indicates that a critical level of carbon is required to make the alloy resistant against the metallographic etchant and a surface step develops in this region as a result of local galvanic element formation. Nevertheless, comparing the apparent thickness of the nonetched case layer in micrographs obtained with the same procedure of metallographic etching provides a useful (monotonic relative) measure for the level of carbon uptake and penetration during low-temperature carburization.
Figure 2 provides the following important results: Regardless of Beilby layer depth, all samples etched in HCl developed a continuous case (carbon-rich layer) with excellent thickness uniformity.
Figure 2a,c,e,g consistently exhibits an apparent average case thickness of
mm. This means that etching in aqueous HCl has properly removed the passivating oxide layer, regardless of Beilby layer thickness.
In the micrographs in the right column of
Figure 2, a continuous case is only observed for the electropolished specimen. While significantly thinner than the case thickness obtained with HCl etching (left column), this observation shows some ability of the carburizing gas itself to activate the surface by removing the passivating oxide layer. The average apparent case thickness is only about 1/3 of the case thickness observed after HCl etching. Additionally, the case thickness seen in
Figure 2b exhibits strong local variations, indicating that the activation is slow and locally incomplete.
The nonetched specimens prepared with a Beilby layer (
Figure 2d,f,h) exhibit no continuous case at all.
Figure 2d (P4000) features semiglobular fragments of a case, apparently growing in from the surface. In
Figure 2f (P800), similar features appear, but are even less developed.
Figure 2h (P180), obtained from the specimen with the thickest Beilby layer, does not exhibit any trace of case formation. These observations demonstrate that both the passivating oxide and the Beilby layer can be similarly effective diffusion barriers to carbon infusion at low temperature: The inhibiting effect of the passivating oxide layer follows from the fact that the carbon penetration of the electropolished nonetched sample is less than for the HCl-etched samples (assuming that the oxide layer that forms during or after electropolishing is equivalent to the passivating oxide layer that “normally” forms on AISI-AL-6XN). The inhibiting effect of the Beilby layer follows from the decrease of carbon penetration with increasing Beilby layer thickness.
Given that the Beilby layer constitutes an effective carbon-diffusion barrier, the fact that the apparent case thickness in the left column of
Figure 2 does not significantly differ between specimens with different mechanical pre-preparation indicates that the severely plastically deformed subsurface zone generated by the respective procedures (electropolishing, mechanical grinding) has been completely etched away by the application of aqueous HCl.
The efficacy of activation by aqueous HCl is further demonstrated by X-ray diffractometry and scanning Auger microprobe carbon-fraction–depth profiling.
Figure 3 shows a comparison between the the X-ray diffractogram of low-temperature carburization AISI-AL-6XN, activated with HCl, and non-carburized AISI-AL-6XN. The 111 and 200 peak positions of nontreated AISI-AL-6XN are consistent with the anticipated lattice parameter of
nm. The peaks are sharp and there are no (e.g., carbide) peaks in addition to those expected from austenite.
After low-temperature carburization with HCl activation, the corresponding peaks are shifted to lower diffraction angles, indicating increased interatomic distances between the metal atoms. The peak shift is almost twice the peak shift observed for AISI-316L reported in Reference [
4], consistent with the maximum carbon fraction also being almost twice as large. The peaks are also broadened. This can be attributed to two effects: (i) As the carbon profile is graded, the peak shape corresponds to a weighed average over lattice plane spacings that decrease with increasing depth below the alloy surface. (ii) A high dislocation density caused by stress that builds up between the region of expanded interatomic distances at the surface and the noncarburized alloy core [
20].
Figure 4 shows a carbon profile measured on the cross-sectional AISI-AL-6XN specimen by scanning Auger microprobe. Defining the case depth as the depth at which the signal vanishes in the background noise, the case depth is
m, which is comparable to the case depth suggested by the metallographic images in
Figure 2. As expected, the maximum carbon fraction is observed at the surface:
. The carbon profile gradually decreases with increasing
z, exhibiting a concave shape characteristic of a diffusion coefficient that increases with increasing
[
21].
Comparison of
Figure 4 and the apparent case thickness observed in the left column of
Figure 2 shows that a carbon fraction
is needed to establish corrosion resistance against the metallographic etchant.
While
Figure 2 indicates the “case depth” (i.e., depth of the zone where
), the carbon profile of
Figure 4 and the magnitude of the peak shift observed in
Figure 3 prove that the case actually has the expected high level of carbon. Further,
Figure 3 proves that the volume fraction of potentially existing additional phases (carbides) is below the volume fraction that X-ray diffractometry can detect, estimated as
.
4. Discussion
The empirically developed surface activation by HCl gas that is part of the Swagelok (industrial) low-temperature carburization process is carried out at 520 K (250 ) for ks (3 h), and it is carried out twice with a 743 K (470 ) ks (3 h) low-temperature carburization step in between. Room-temperature etching in aqueous HCl for ks (10 min) and immersing in CHOH for ks (5 min) is a much more effective way of surface activation. The efficacy of aqueous HCl in removing the passivating oxide layer and Beilby layer, compared to gaseous HCl, can be rationalized by the times higher density of aqueous HCl.
The experimental observations can be understood in terms of the model shown in
Figure 5: Generally, as-received AISI-AL-6XN exhibits a passivating oxide layer at the surface and a Beilby layer in the zone below. Typically, this includes a zone of nanosized grains or subgrains directly below the surface [
14], as well as an increased density of line and point defects. Contrary to what the terminology suggests (“Beilby layer”), this layer has no well-defined thickness. Rather, the number-density–depth profiles of the defects that characterize the Beilby layer are smoothly graded, i.e., a steady function of depth
z, and the profiles of different defect species fall off differently with increasing
z.
Figure 5a illustrates the process without external activation: If no HCl (or other etching) is applied, and the material is exposed directly to low-temperature carburization, an underdeveloped carbon profile results, or none at all (
). This is because of two reasons: (i) While C
H
can activate the surface by removing the passivating oxide layer, it does not remove the passivating oxide layer as efficiently as e.g., gaseous HCl. (ii) C
H
does not etch the alloy. Therefore, a potentially existing Beilby layer is not removed and remains as a diffusion barrier to carbon.
Figure 5b shows the corresponding stages with external activation: External etching in concentrated aqueous HCl removes the passivating oxide layer and the Beilby layer—or at least the (presumably uppermost) subzone of the Beilby layer that inhibits carbon diffusion. Storing in C
H
OH results in a provisional passivating layer. This provisional layer protects the alloy surface against the reformation of the Cr-rich passivating oxide film. At the beginning of the low-temperature carburization process, the increasing temperature and interaction with C
H
readily remove the provisional passivating layer, exposing a fully activated surface to the carburizing gas. This results in a fully-developed carbon profile.
Our results and the model of
Figure 5 are consistent with the findings of Somers et al. [
22], who reported that C
H
has the ability to activate and carburize Fe–Cr–Ni alloys. The work of Ge et al. [
14] confirmed that C
H
can carburize AISI-316L at low temperatures, but only if a potentially existing Beilby layer introduced by machining was first polished off. AISI-316L specimens with a coarse surface finish did not carburize well. These observations indicated that C
H
can remove the passivating oxide layer, but does not etch the alloy and therefore cannot remove the Beilby layer.
The detailed mechanisms by which the Beilby layer, i.e., a severely plastically deformed layer, inhibits the diffusion of interstitially dissolved carbon, are complex and currently not understood. It is known, however, the Beilby layer contains high dislocation density. Owing to locally increased free volume, dislocations can accelerate the diffusion of interstitial atoms (although the concept of “pipe” diffusion may not apply [
23]). On the other hand, the interaction of the stress fields caused by dislocations and interstitial atoms causes dislocations to bind carbon atoms in “Cottrell clouds” [
24]. Quantitatively, simulations (for the body-centered cubic structure) suggest that, for 70% of the total time, carbon atoms diffuse within a distance of ≤1 nm from the dislocation core [
25]. This effectively reduces the normally three-dimensional diffusion to one-dimensional diffusion along the dislocation line. Moreover, the carbon fraction within the Cottrell clouds is higher than in dislocation-free regions. For the high carbon fractions considered here in the context of low-temperature carburization, the enhanced
in Cottrell clouds may lead to a configuration in which the availability of free interstitial sites to jump into becomes limiting for diffusivity. This effect has been observed earlier by Christiansen et al. in experiments revealing the concentration dependence of the nitrogen-diffusion coefficient in AISI-304 and AISI-316 [
26]. Reduced availability of vacant interstitial sites combined with the reduced dimensionality of diffusion explains why austenite regions with high dislocation density act as a carbon diffusion barrier, rather than enhancing diffusion.
Another effect of the Beilby layer may arise from the fact that it is likely to contain passivating oxide layer fragments that initially formed on the alloy surface in contact with air and were kneaded into the Beilby layer that formed during subsequent machining. In earlier work, we concluded that the oxide particles can retard or prevent the ingress of carbon [
14].
The formation of a provisional passivating layer on or after immersion in C
H
OH is a compelling hypothesis: The bare metal surface is not stable in air. On the other hand, the formation of the regular passivating layer is known to inhibit low-temperature carburization. This means that a passivating film must form, but it is not the usual Cr-rich oxide film. The observed phenomenon resembles the provisional passivation of Si
surface with aqueous HF solution, which is common in semiconductor technology [
27,
28]. In that case, dangling bonds of Si are saturated by a monolayer of hydrogen, preventing the formation of the usual passivating SiO
layer. Different from a SiO
layer, the provisional passivating layer formed by hydrogen readily desorbs upon heating to expose a pristine Si surface, e.g., for epitaxy.
Previous work [
17,
29] demonstrated that aqueous HCl can also effectively activate AISI-316L for low-temperature carburization. Comparative experiments in which the etching by aqueous HCl was followed by rinsing in C
H
OH in one case and rinsing in H
O in the other case demonstrated that immersion in H
O after aqueous HCl etching does not enable subsequent low-temperature carburization, and, thus, does not accomplish provisional passivation.
Based on these observations, we now discuss three alternative hypotheses for the mechanism of provisional passivation of the alloy in the period between removing the passivating oxide layer by aqueous HCl and the beginning of carbon infusion.
1. Provisional passivation by liquid ethanol. In principle, it is possible to protect a bare metal surface against oxidation by an oxygen-free liquid. As described under “Experimental Methods”, the specimens were covered by liquid ethanol for the entire period between HCl etching and drying in streaming dry N in the carburization furnace. This could explain the success of our procedure if covering the alloy surface with ethanol were sufficient to reduce contact with oxygen to a level at which reformation of the passivating oxide layer is impossible within the given amount of time. However, the literature data on the solubility and diffusivity of oxygen in ethanol do not support this hypothesis.
Based on the literature data [
30], nominally pure C
H
OH at room temperature (293 K) can contain dissolved oxygen up to a mole fraction of
. This means that an C
H
OH layer with a thickness of 2
m contains enough oxygen to build up the regular passivating oxide layer with its thickness of 1 nm. In C
H
OH–C
H
CH
solution at room temperature, the diffusion coefficient of oxygen is
m
·s
[
31]. Assuming the same diffusivity for oxygen in C
H
OH implies that within the
ks (5 min) for which the specimens are stored in C
H
OH, the typical diffusion distance
mm. Thus, the diffusion distance is about 500 times larger than the thickness of the C
H
OH layer that could provide enough oxygen to build up a regular passivating oxide layer. Ergo, the observation that a regular passivating oxide layer does not reform while the specimens are stored under C
H
OH cannot be explained by a lack of oxygen.
Within the ks of transferring the specimens into the furnace, the diffusion distance mm. As the CHOH film on the specimens becomes thinner than , oxygen from the ambient air can reach the metal surface to form a regular passivating oxide layer. Summarizing, it seems unlikely that the effect of CHOH is simply to “seal” the metal surface against contact with oxygen.
2. Provisional passivation by ethoxides. Here, the hypothesis is that contact of the HCl-etched bare metal surface with C
H
OH causes the formation of a provisional passivating layer consisting of ethoxides. As the pyrolysis temperature of metal ethoxide can be as low as 520 K [
32], they may readily pyrolyze during heating for low-temperature carburization, leaving oxides of all involved metals on the surface [
32,
33,
34]. It is known that Fe and Ni oxides are transparent to the infusion of carbon at low temperature [
35]. However, a detailed analysis, sketched in the following, does not support this model:
The conjugate base of an alcohol (R–OH) is the alkoxide ion (RO–). Typically, alkoxides readily form ligands, act as strong bases, and are nucleophilic when they have nonbulky R-groups [
36]. Metal alkoxides can be monomeric or polymeric and can form complex macromolecules [
33,
36,
37,
38]. Highly electropositive metals may react directly with an alcohol solution, leading to the formation of an alkoxide [
36]. Such reactions may dissolve metal in the alcohol, forming hydrogen gas and an insoluble alkoxide, following the general reaction:
where the alcohol may be methanol, ethanol, etc. However, this route is unlikely for transition-metal alkoxide formation [
36].
For some transition-metal alkoxides, a metal–halide feed stock reacts with sodium alkoxides in an alcohol solution. A metathesis reaction (mutual ion exchange) occurs with the precipitation of the less-soluble product [
33,
36,
38,
39,
40,
41]. An example is:
where THF is tetrahydrofuran [
41].
Another route for transition-metal alkoxide formation is oxidative synthesis. Here, H–OR additions are made to the metal centers of ligands.
-ligand metathesis reactions are such a pathway [
37]. They only involve one metal, more credible as an avenue for the formation of a provisional passivating layer. An example reaction is:
where only hydrogen atoms are exchanged between
-ligands [
37]. However, these reactions are not likely to readily proceed with the transition metals of interest in this study, especially considering the lack of an alkoxide feed stock (e.g., Na-ethoxide) for metathesis reactions.
3. Provisional passivation by chloride. On exposure to aqueous HCl, alloy samples react with it corresponding to typical corrosion reactions, leading to the dissolution of metal ions and the formation of hydrogen gas:
However, it is also energetically favorable to form a metal chloride, especially as the sample is removed from the solution with residual HCl adsorbed on the surface. This would lead to a metal- (Fe-, Cr-, Ni-) chloride layer on the alloy surface, which could have multiple impacts on the efficacy of the process. If the surface layer is adherent and covers the majority of the sample surface, the chloride, as such, may act as a provisional passivating layer. Binding the Cr at the alloy surface to Cl, so it is not available to form Cr oxide when exposed to ambient oxygen later, effectively inhibits formation of the regular, Cr-oxide-rich passivating oxide layer, which (presumably) poses the strongest barrier to carbon infusion.
The resulting condition of the alloy surface in C
H
OH depends on the solubility of the hypothetical Cr chlorides.
Table 2 classifies the solubility of various Fe, Cr, and Ni chlorides [
42,
43,
44,
45,
46]. The most common modification of chromic chloride, CrCl
, is hexahydrate. It is soluble in both water and C
H
OH. Another important modification is anhydrous CrCl
, which, in pure form, is mostly insoluble in both water and C
H
OH. However, if traces of CrCl
(or another reduction agent) are present, this can catalyze dissolution in H
O, which is strongly exothermal [
47].
If the hypothetical provisional passivating layer mainly consisted of the hexahydrate modification of CrCl
, it would be less susceptible to which rinsing agent is used [
42]. Since in our process samples rinsed in ethanol performed considerably better than samples rinsed in water, the provisional passivating layer much more likely consisted of chlorides that are soluble in water, but insoluble in C
H
OH. The only chlorides matching this requirement are chromium dichloride (chromous chloride), CrCl
, and the anhydrous modification of CrCl
. So, if an alloy part covered with (Fe, Cr, Ni) chlorides were rinsed in cold water, and at least traces of CrCl
were present, all chlorides would be removed. Rinsing in C
H
OH, in contrast, would remove all chlorides except CrCl
and CrCl
. Thus, the observation that the process works with rinsing in C
H
OH, but not with rinsing in H
O, indicates that the critical compounds that needs to form to generate a chloride-based provisional passivating layer are CrCl
or anhydrous CrCl
. Thermodynamic data for similar conditions [
48] indicate that the Gibbs energy of formation is lower for CrCl
than for CrCl
. Cr and HCl readily form CrCl
, while forming CrCl
from CrCl
usually requires significant activity of oxygen (2 CrCl
+
O
+ 2 HCl → 2 CrCl
+ H
O). Accordingly, the provisional passivating layer most likely consists of CrCl
. While the low-temperature carburization temperature is significantly below the melting temperatures of chlorides, the thin layers considered here may be removed by sublimation [
49] or via reduction by the H
in the low-temperature carburization gas mixture.