Observation of hydrogen trapping at dislocations, grain boundaries, and precipitates

See allHide authors and affiliations

Science  10 Jan 2020:
Vol. 367, Issue 6474, pp. 171-175
DOI: 10.1126/science.aaz0122

An anti-embrittlement roadmap

Hydrogen is important for energy applications such as fuel cells but tends to diffuse into materials and make them more susceptible to fracture. Chen et al. tackled the challenge of identifying the exact location of hydrogen atoms in two common steels. The light weight and high mobility of hydrogen creates serious problems with conventional techniques. The authors used cryo-transfer atom probe tomography to show that hydrogen is pinned to various interfaces in the steels. This direct look into hydrogen trapping should help with the development of materials that are more resistant to hydrogen embrittlement.

Science, this issue p. 171


Hydrogen embrittlement of high-strength steel is an obstacle for using these steels in sustainable energy production. Hydrogen embrittlement involves hydrogen-defect interactions at multiple-length scales. However, the challenge of measuring the precise location of hydrogen atoms limits our understanding. Thermal desorption spectroscopy can identify hydrogen retention or trapping, but data cannot be easily linked to the relative contributions of different microstructural features. We used cryo-transfer atom probe tomography to observe hydrogen at specific microstructural features in steels. Direct observation of hydrogen at carbon-rich dislocations and grain boundaries provides validation for embrittlement models. Hydrogen observed at an incoherent interface between niobium carbides and the surrounding steel provides direct evidence that these incoherent boundaries can act as trapping sites. This information is vital for designing embrittlement-resistant steels.

Although the phenomenon was first identified a century ago (1), hydrogen embrittlement (HE) remains an extremely important unsolved industrial challenge. High-strength steels are particularly prone to early fracture and reduced toughness in the presence of hydrogen (2), which limits their possible manufacturing routes and industrial applications. An example is in the automotive industry, which is looking to use high-strength steels to increase fuel efficiency by reducing weight while still meeting strength requirements (3). Another example is the use of steel for the hydrogen-pressurized pipes and storage containers that would be required to underpin hydrogen fuel as a form of energy (4). These barriers are examples of the challenges for using steel in the decarbonizing technologies that are likely required to meet the global commitments of the 2016 Paris Agreement (5). The development of steels that are resistant to HE requires a solid understanding of the interaction between hydrogen and steels.

The specific origin of HE has long been a topic of debate, and no precise proposed mechanism exists that is consistent with all the observed HE consequences. HE phenomena are specific to material and microstructure (6). For high-strength steels that display intergranular failure, HE has been attributed to decohesion caused by hydrogen segregation at grain boundaries (GBs) (7), whereas for others that display transgranular failure with a dimple-like fracture surface, embrittlement has been associated with a weakening effect in the presence of hydrogen (6). To describe this effect, Beachem proposed a model in which hydrogen enhances dislocation activity, facilitating the formation of voids, and their subsequent coalescence allows crack propagation, leading to the observed dimples (8). Beachem’s model is known as hydrogen-enhanced local plasticity and was later supported by in situ transmission electron microscopy (TEM) observations of enhanced dislocation activity in the presence of gaseous hydrogen (9). The missing piece of evidence to fully validate both models (that is, hydrogen-facilitated decohesion and hydrogen-enhanced dislocation mobility) is direct evidence of the presence of hydrogen at these critical microstructural features.

In addition to hydrogen at GBs and dislocations, the interaction between hydrogen and second phases, such as carbides, is important. These features have the potential to provide HE resistance by serving as hydrogen traps preventing the availability of hydrogen for embrittlement (10).

To understand both embrittlement and trapping, we require information about the position of hydrogen atoms in relation to microstructural features at the nanometer-length scale. However, the combination of the low level of interaction between an electron and hydrogen and the extremely fast diffusion of hydrogen in steels renders it extremely difficult to experimentally determine the location of hydrogen by electron microscopy (11). Thermal desorption spectroscopy (TDS) provides macroscale information about hydrogen desorption, but this hydrogen originates from all of the possible trapping sites, making it difficult to distinguish the contributions from the different types of hydrogen-containing features (12). This problem obscures our understanding of the specific roles of, for example, GBs, dislocations, and precipitates. Solving this problem removes the impediment to better microstructural design of HE-resistant materials.

Mass spectroscopy techniques are suited to identifying the location of hydrogen because it can be identified by its ratio of mass to charge state (11). The high spatial and mass resolution of atom probe tomography (APT) makes it useful for mapping hydrogen in nanoscale features in three dimensions (1316). APT is regularly used for both precipitate analyses and identification of microstructural defects such as GBs and dislocations through associated elemental segregation (1719). Deuterium has almost the same chemical properties as hydrogen but a slightly lower diffusion coefficient in steel due to its larger mass (20). The use of deuterium circumvents several potential hydrogen artefacts and, combined with cryogenic sample transfer, allows us to obtain precise observations of the distribution of hydrogen around nanoscale features. Here, we provide direct evidence of hydrogen localization to achieve a mechanistic understanding of hydrogen-materials interactions and facilitate the development of hydrogen-resistant steels.

We subjected a niobium-bearing low-carbon steel to two different heat treatments to produce ferritic and martensitic states (21). The ferritic state was well annealed and contained few dislocations, allowing uncomplicated analysis of carbide precipitates. By contrast, the fine-grained martensitic state had a high density of dislocations and GBs (both high and low angle). We examined the microstructures of both materials by using transmission Kikuchi diffraction (TKD) (22) and TEM (Fig. 1). We confirmed that the martensitic sample consisted of much finer grains and that these grains also contain low-angle lath boundaries (Fig. 1, A and D). We made a high-resolution image of the niobium carbide (NbC) precipitate (Fig. 1C) indicated with white arrows in the bright-field TEM image (Fig. 1B). We took both images under the [1–10]ferrite incident beam condition (Fig. 1C, inset). We did not observe an orientation relationship between the precipitates and the matrix, indicating that they were incoherent, as suggested by their round shape. We observed an extremely high density of dislocations in the martensitic sample (Fig. 1, E and F). We took the dark-field image with the use of the (220)ferrite diffracted vector (23) (Fig. 1F, arrow). We examined atom probe needles prepared from the martensitic sample (fig. S3) and found that they also contained GBs and a high density of dislocations.

Fig. 1 Microstructure of the ferritic and martensitic steels.

(A and D) TKD inverse pole figure maps of the ferritic and martensitic samples, respectively, showing the grain size and orientation. (Scales differ for these two panels.) (B and C) Low- and high-magnification bright-field images of the NbC precipitates in the ferritic sample. The precipitate shown in (C) is marked in (B). (E and F) Bright- and dark-field images of the same area showing the high density of dislocations in the martensitic sample.

We confirmed with thermal desorption experiments that hydrogen was present in samples that had been electrolytically precharged with hydrogen (Fig. 2). For the martensitic sample, we found a substantial hydrogen desorption peak when the sample reached between 100° and 300°C (Fig. 2A). We believe this peak corresponds to hydrogen that was absorbed at the abundant dislocations in the martensite, consistent with other studies that used the same technique (12, 24, 25). On magnifying the scale of Fig. 2A, we found that the data from the ferritic sample contain two peaks in Fig. 2B, one at ~100° to 200°C and the other at ~300° to 350°C. We attribute these peaks to hydrogen at the ferrite GBs (26) and at NbC precipitates (27), respectively. Of course, the martensitic sample might also be expected to contain hydrogen at GBs. However, any signal from the GB hydrogen in the martensitic sample was likely buried in the large peak associated with dislocations. Resolving the exact hydrogen-trapping behavior would require peak deconvolution, which is challenging because we do not have enough information to get a statistically robust deconvolution (28).

Fig. 2 TDS data showing the temperature at which trapped hydrogen is released for the two types of steel.

(A) Full dataset. The large peak in the data for the martensitic steel is thought to be the result of trapping at dislocations. (B) Details of the region shaded in gray in (A). The peak at 100° to 200°C is thought to correspond to trapping at the GBs, and the peaks at higher temperatures are thought to be due to trapping at carbides.

We performed APT on both samples. We charged needle-shaped samples with deuterium and transferred them via a complete cold chain into the atom probe chamber (21). Because the needle-shaped sample is small, and hydrogen-deuterium is prone to recombination and desorption if it reaches the specimen surface (29), it is crucial that diffusion is minimized by maintaining the cold chain. We collected data from a deuterium-charged ferritic sample (Fig. 3) and a deuterium-free reference sample, confirming that the deuterium signal, at a mass-to-charge ratio of 2, is only present for the deuterium-charged sample (fig. S5, A and B). The charged sample contains three NbC precipitates in a ferritic matrix (Fig. 3A). Slices that were 5 nm thick and taken through the center of the precipitates (Fig. 3B, left) allow us to visualize the deuterium (Fig. 3, C and D, red spheres) in the vicinity of the carbide precipitates. We found that the deuterium atoms are mostly located at the interfaces between NbC and ferrite.

Fig. 3 APT analysis of a deuterium-charged ferritic steel sample containing NbC.

(A) Full reconstructed atom map showing atoms with mass-to-charge ratios that match deuterium (large red spheres), carbon (small blue spheres), and niobium (small brown spheres). (B) Schematic describing the data in (C) to (F). (C and D) y Axis slices 5 nm thick through the center of the top (NbC#1) and bottom (NbC#2) precipitates, respectively, showing carbon, niobium, and deuterium atoms. (E and F) One-dimensional (1D) z axis composition profiles through the centers of NbC#1 and NbC#2, respectively, extracted from a 7-nm-diameter cylindrical region of interest with a bin width of 0.5 nm and no overlapping of bins.

We then quantitatively analyzed the NbC precipitates from top to bottom in the z direction (Fig. 3B, right), which is the direction that provides the highest spatial resolution (30). We obtained data from cylindrical regions of interest through the precipitate centers (Fig. 3, E and F). Peaks in deuterium concentration were aligned with the edges of the NbC profile, confirming that the deuterium was mostly located at the precipitate-matrix interface. Proximity histograms based on a 1% niobium isoconcentration surface (isosurface) (31) also showed the localization of the deuterium at the interface (fig. S6).

Carbides are well known to trap hydrogen and reduce embrittlement, but the mechanism is not clear. Conflicting arguments have emerged about whether the trapping is in the carbide bulk, at the interface, or both. Previous atom probe studies showed that deuterium was present in the vicinity of carbides for a peak-aged vanadium carbide (14), including along the direction of the coherent (001) surface, but were unable to resolve details of whether it is located at the interface or the carbide bulk. For much smaller, coherent vanadium carbides, deuterium was found in the bulk (13), presumably because of strain in the precipitates and a high resulting fraction of carbon vacancies (32, 33).

In this study, we show deuterium trapping at carbide interfaces in larger, incoherent precipitates. Our results are consistent with the earlier findings that carbide interfaces can effectively trap hydrogen (28, 34) and that, for equivalent volume fractions of vanadium carbide, alloys with smaller precipitates, and hence a higher effective carbide-matrix interface area, trap more hydrogen (35).

Our martensitic sample contains a much higher volume-fraction of dislocations and GBs than our ferritic sample. Two atom probe datasets from this sample (Fig. 4) both contained a high density of dislocations that were enriched with carbon, identifiable by their linear morphology. The first dataset contains only dislocations (Fig. 4, A and B, and movies S2 and S3), and the second dataset also contains three carbon-enriched GBs connected by a triple line, which are most likely low-angle martensitic lath boundaries (Fig. 4, E and F, and movie S4). Solute carbon is known to segregate at the lattice defects of iron (19), so we were able to highlight the GB and dislocations by 2% carbon isosurfaces. The dislocations are highlighted by transparent blue isosurfaces, whereas the GBs are highlighted by transparent red isosurfaces. We confirmed the presence of deuterium by comparing the mass spectrum from this sample (fig. S5C) with that of a martensitic sample that had also been deuterium-charged and cryogenically transferred but then left to desorb in the buffer chamber at room temperature overnight (fig. S5D).

Fig. 4 APT analyses of deuterium-charged martensitic steel samples containing GBs and dislocations.

(A and B) Views at two different angles of a reconstructed atom map showing deuterium (red), iron (gray), and 2% carbon isosurfaces in blue. This sample contains a high density of linear, carbon-decorated dislocations (movie S2). (C) Slice, 5 nm thick, from the center of the dataset shown in (B) (indicated with the black dashed rectangle) with carbon atoms (blue) and deuterium atoms (red) showing the spatial correlation between the two elements. Movie S3 shows this 5-nm slice moving across the full dataset. (D) Collective proxigram analysis of the 70 dislocations and carbon isosurfaces contained in (A), showing that the dislocations have carbon and deuterium at their cores but little niobium. (E and F) Two views of a different dataset from the same sample. Again, features are highlighted by 2% carbon isosurfaces, with one of these surfaces that encompasses a GB region highlighted by a transparent red isosurface. (G) z Axis, 5-nm-thick slices from the GB region, indicated in (F) (movie S4). (H) 1D composition profile with a 1-nm step size from a volume of 20 nm by 20 nm by 18 nm across the GB marked in (E).

A slice taken from the center of the dataset shown in Fig. 4B shows the spatial correlation between the carbon and deuterium (Fig. 4C) in the vicinity of the dislocations (movie S3). We conducted a proxigram analysis based on the 2% carbon isosurface (Fig. 4D), which provides the average composition as a function of the distance from the isosurface (31). We found that deuterium was concentrated at the core of the dislocations. Very little niobium was present in these regions, distinguishing them from the carbides we observed in the ferritic sample (Fig. 3, E and F).

For the GB region, we found that the amount of carbon segregation to the GBs varied across the boundary plane. For this reason, the transparent red isosurface in Fig. 4, E and F, does not capture the full area of the boundary. To show the continuous nature of the boundary, z axis slice views (Fig. 4G) were taken from the positions indicated in Fig. 4F. We found that three GBs were connected by a triple line. Deuterium atoms in the martensitic sample are also concentrated in the GB region, demonstrating that the GBs also act as trapping sites. We created a composition line profile across the GB area (Fig. 4H, taken from the area shown in Fig. 4E, bottom right corner). We found carbon and deuterium at the GB but very little niobium segregation.

All the features we analyzed contain carbon. This observation is consistent with experimental evidence by TDS showing the hydrogen trapping in a specimen that is known to contain carbon-decorated lattice defects in a martensitic steel (36). A reasonable speculation is that an affinity between carbon and hydrogen might lead to the hydrogen trapping. However, a theoretical study showed that the carbon atoms in lattice defects have only a weak influence on the hydrogen trapping (37), and other modeling has successfully predicted the presence of trapped hydrogen at dislocations in a body-centered cubic lattice (38). Therefore, we believe that the observed hydrogen is directly related to the presence of lattice defects, rather than the result of attraction by segregated carbon atoms.

In this study, cryogenic APT has enabled the measurement of the distribution of hydrogen at dislocations, GBs, and precipitates. We anticipate that this method will continue to be used in the future as a complement to detailed TEM analysis (including in situ tests), mechanical testing in the presence of hydrogen, and first-principles calculations to provide further insights into mechanisms of embrittlement and its prevention.

Our observation of hydrogen atoms at dislocations and GBs confirms the existence of excess hydrogen at defects, an assumption that underpins critical theories of HE. Knowledge of hydrogen trapping at incoherent interfaces provides a pathway for the design of alloys in which embrittlement is prevented or delayed.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

Table S1

References (39, 40)

Movies S1 to S4

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: The authors thank T. Paxton (King’s College London), P.-C. Hsu (Duke University), H.-W. Yen (National Taiwan University), Y.-C. Lo (National Chiao Tung University), D. Xie (Xi’an Jiaotong University), H. Jiang (University of Western Australia), B. Gault (Max Planck Institute for Iron and Steel Research), T. Sato, G. Derrick, X. Cui, B. Pace, and M. He (all University of Sydney) for useful discussions. Y.-S.C. thanks his family members for support. Funding: The authors acknowledge funding from Australian Research Council Linkage Project (LP180100431), Future Fellowship (FT180100232), CITIC-CBMM Nb Steel Award Fund Program (2018FWNB30064), the 2019 University of Sydney Postdoctoral Fellowship, and USYD-SJTU Partnership Award, as well as the research facilities supported by Microscopy Australia and the University of Sydney. Author contributions: Y.-S.C. drafted the manuscript, supported the development of the cryo-transfer protocol, conducted the APT experiment, and analyzed all the data. H.Lu conceptualized the study, designed the samples, and acquired research funds. J.L. produced the samples and conducted the TDS experiment. A.R. designed and produced the cryo-transfer system and developed the cryo-transfer protocol. H.Liu and G.S. conducted the TEM imaging and the TKD mapping, respectively. I.M. developed the cryo-transfer protocol. Z.Z. provided the facilities for sample production and processing. W.L. provided the TDS facility. A.G. provided the research funds. J.M.C. provided the microscope facilities and cryo-transfer analyses, supervised the project, and finalized the manuscript. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials and can be accessed online via

Stay Connected to Science

Navigate This Article