Hydrogen embrittlement mechanisms in advanced high strength steel

Hydrogen embrittlement is increasingly important in advanced high strength steels (AHHS) as strength levels increase well above 1000MPa. This work developed a detailed understanding of the embrittling mechanism in model AHHS steels based on Fe-Ti-Mo and Fe-V-Mo, both strengthened through interphase precipitation. Hydrogen charging led to an increase in the dislocation density and an enlarged strain field around precipitates, resulting in an increase in residual stress. This was much greater for the Ti-Mo steel compared to the VMo. Important differences in the hydrogen trapping behaviour was seen between the two steels, with hydrogen believed to be trapped at the matrix/precipitate interface for the Ti-Mo steel, but within the precipitate for the V-Mo steel. The effects of hydrogen were investigated in detail for slow strain rate tensile tests and double notched tensile samples. Hydrogen charging resulted in a loss in strength and ductility, with the Ti-Mo steel failing at yield, while the V-Mo steel exhibited a ~13% loss in strength and a ~ 35% loss of ductility. Crack initiation in tensile samples occurred at high strain gradient dislocation boundaries. However, crack propagation rapidly became quasi-cleavage, along the {100} plane in ferrite, and also along the martensite/ferrite grain boundaries on the {110} plane in the martensite. Minimal plasticity was observed associated with the crack tip, which was believed to be a result of the suppression of dislocation emission at the crack tip by the hydrogen. Graphical Abstract Graphical Abstract .

There has been much drive to increase the strength of steels as part of a strategy to reduce CO 2 emissions through higher performance in, for example, ultra-light car bodies. This has led to the rapid development of Advanced High Strength Steels (AHSS) which provide ever increasing strength levels, but with excellent ductility. AHSS have several different microstructures including ferrite/martensite dual phase (DP), transformation induced plasticity (TRIP), complex phase (CP) and martensitic steels. DP steels, which offer a tensile strength (around 1.0GPa) and ductility (no less than 10%) are the most attractive AHSS steels for safe and fuel-efficient cars due to the simple production process. As the strength of the steel increases, the susceptibility to hydrogen embrittlement (HE) increases, to the point now that HE is an important consideration in the design of AHSS.
There are numerous observations of HE and many theories have been proposed . The contrasting theories of hydrogen embrittlement range from enhanced dislocation mobility leading to failure to suppression of dislocation emission at a crack tip. Hydrogen enhanced localized plasticity (HELP) [19][20][21][22][23] has received extensive attention in the literature and proposes enhanced crack growth is a highly localised plastic failure process as a result of enhanced dislocation emission and mobility. At the opposite end of the spectrum, hydrogen enhanced decohesion (HEDE) considers that atomic bonding is weakened at the crack tip by the presence of hydrogen [24,25]. However, as Song and Curtin [26] have pointed out, the stress intensity at the crack tip must be lower than the value at which dislocation emission occurs, which is physically unrealistic. Rather, their modelling indicates that the primary role of hydrogen is to suppress dislocation emission from the crack tip, leaving cleavage as the only energy dissipation mechanism available. Some theories are quite specific to the phases present.
Hydrogen embrittlement mechanisms in advanced high strength steel There are several studies of HE in DP steels (although not in precipitation hardened DP steels), which show, for example, the ability of a crack to nucleate and grow is a function of strength, size and distribution of the martensite islands after hydrogen charging, as well as the carbon content in the martensite phase [9,24,[27][28][29][30]. Sun et al. [31] reported that with the carbon exceeding 0.15%, HE was associated with crack initiation in the hard martensite regions and propagates through the ferrite grain or along the ferrite/ martensite interface due to the increase of mobile dislocations by transportation of hydrogen atoms. Koyama et al. [24] reported that in DP steels HE promotes damage nucleation and growth mechanisms, while the ability of ferrite to arrest cracks is greatly reduced. The critical strain for decohesion in the martensite was decreased through the HEDE mechanism, while the cracking in the martensite/ferrite interface and in the ferrite was enhanced through the HELP mechanism.
However, there is no research on HE in DP steels strengthened by precipitation, which of course can act as efficient hydrogen traps as well as modifying dislocation dynamics.
Moreover, the volume fraction of the martensitic in the current steels is much lower than conventional DP steels.
The present work focuses on a detailed analysis of the hydrogen induced cracking mechanism, looking at both crack initiation and propagation with the effect of volume fraction of martensite, different types and volume fraction of precipitaitons as well as the different density of dislocations. In order to do this, model steel microstructures were developed, comprising fine grained ferrite, precipitation hardened with fine coherent precipitates that are known to act as efficient hydrogen trapping sites. In addition, both steels contained fine martensite islands and high volume fraction of precipitations, with ~5% martensite and TiC in the Ti-Mo steel and ~20% martensite and VC in the V-Mo steel. Both steels exhibit a good combination of strength and ductility. Slow strain rate testing (SSRT) and double notched tensile tests were used to study the HE susceptibility, and the fracture Hydrogen embrittlement mechanisms in advanced high strength steel morphology was examined to analyse the nucleation and propagation of hydrogen assisted cracks.

Materials
The chemical composition of the laboratory cast Fe-Ti-Mo (Ti-Mo), Fe-V-Mo (V-Mo) steels are listed in Table 1. The alloys were made by vacuum induction melting and cast into ingots having dimensions 620 mm × 105 mm × 35 mm at Tata Steel, IJmuiden. The ingots were homogenized at 1250 °C for 2 h and hot worked through several passes to 8 mm thick plates.
Dilatometer specimens, 120 mm long × 12 mm wide × 6 mm thick, were machined from the plate along the rolling direction. The heat treatment for producing interphase precipitated carbides was performed using the Dilatronic dilatometer at The University of Sheffield.
Samples were first reheated to 1250 °C in a tube furnace, held for 30 min, and then water quenched to room temperature. The specimens were then austenitised at 1200 °C for 3 min and rapidly cooled to 630 °C at a rate of 10 °Cs -1 , held isothermally for 90 min, and finally water quenched to room temperature. The rolling and heat treatment schedule are illustrated in Fig. 1.

Hydrogen charging and mechanical tests
Tensile test specimens were prepared according to ASTM standard (E 8M-04) with the long axes parallel to the rolling direction (shown in Fig. 2). Smooth, tensile specimens were 2.66 mm thick with a gauge length of 12.5 mm ( Fig. 2 (a)). The double notched tensile specimens were 2.66 mm thick, with a notch of 60°, 1 mm deep and a root radius of 0.15 mm (Fig. 2 (b)). For the double notch tensile samples, failure occurred from one notch, allowing the study of damage accumulation at the second unbroken notch.
Hydrogen charging was undertaken using standard electrochemical techniques. The tensile test specimens were charged using 1 g/L in an aqueous solution of 3 wt% NaCl and 0.3 wt% NH 4 SCN with a current of 10 mAcm -2 at room temperature. Charging was undertaken for 1, 2, 4, 16 and 48h and TDS (thermal desorption spectroscopy) spectra were taken from each to determine the time at which the specimen was saturated with hydrogen (see section 2.3 for TDS procedure). Based on this, all samples for further investigation were charged for 48 h, which was shown to be the fully saturated condition. Following charging, samples were immediately tensile tested using a constant slow strain-rate tensile of 10 -5 s -1 at room temperature using a Zwick (BTC T1-FR020 TN A50) universal testing machine.

Hydrogen measurement and observation of hydrogen trapping sites
To measure the concentrations of diffusible H in the H-charged specimens, thermal desorption spectroscopy (TDS) was conducted using a gas chromatography system (Agilent Technologies 7890A) and a tubular furnace in a He gaseous atmosphere. The samples was transfer to TDS rig within 10 min after H charging, but purged for 20 min before the start of the TDS experiment. Continuous heating from room temperature to 700 °C was used. Heating rates of 50, 100 and 150 °Ch -1 were used to determine the activation energy. The detailed understanding of hydrogen charging was undertaken at a heating rate of 100 ºCh -1 .

Microstructural and crystallographic investigation
Residual stress measurements were made using X-ray diffraction using the sin 2 ψ method [32] on a Siemens D5000. A number of XRD spectra are obtained with the sample systematically tilted through angles, ψ, (in this case, 10 spectra were obtained a different tilt Hydrogen embrittlement mechanisms in advanced high strength steel angles). The inter-planar spacing of the chosen crystallographic plane, i.e. the 2 peak position, is measured and is plotted against sin 2 ψ. In this case, the (211) plane was selected, since the errors are smaller with the higher angle planes. The strain at an angle  to the principal stress axis  1 , for a sample tilt ψ, is given by: Where  is Poisson's ratio,  1 and  2 are the principal stresses along the axes in the sample plane and E is Young's modulus. If the strains are considered in terms of the inter-planar spacing, the stress at an angle  is given by: Where d n is the stress free lattice spacing and d ψ is the lattice spacing at specimen tilt angle ψ.
The residual stress for a given crystallographic plane is then obtained by obtaining the gradient in a plot of d ψ against sin 2 ψ.
The general microstructures of the two steels were examined using a field-emission scanning electron microscope (FE-SEM; FEI NOVA, The Netherlands) equipped with electron backscattered diffraction (EBSD; Oxford Instruments, UK). Specimens for SEM and EBSD observation were prepared by standard metallographic methods which included grinding from P400 to P1200 and polishing up to 0.04 um colloidal silica particles.
EBSD was undertaken on an FEI-Nova 600 at 20 kV with 14 mm working distance. A step size of 150 nm was used for all samples. Post-processing of the EBSD data was undertaken using the Oxford Instruments-channel 5TM software package. The EBSD maps were first cleaned by removing wild orientation spikes, extrapolating zero solution using a minimum of Hydrogen embrittlement mechanisms in advanced high strength steel 6 neighbouring pixels. Several orientation imaging microscopy (OIM) maps, namely inverse pole figure (IPF), band contract (BC), kernel average misorientation (KAM), were obtained.
Crystal orientation maps were employed to select certain grain orientations from an overall microstructure. The kernel average misorientation, which represents the numerical misorientation average of a given pixel with its neighbours, was used to characterize the local strain distribution in the microstructure and at cracks. The maps were calculated imposing a maximum misorientation of 5 (small local deformations) and with third neighbour misorientation averaging. Such maps allow a qualitative evaluation of small local strain gradients, since these are related to orientation gradients in the material.

3.1.Microstructure of starting materials
Hydrogen embrittlement mechanisms in advanced high strength steel The full details of the microstructure are given in [33]. The optical microstructure of the steels is shown in Fig. 3 (a, b), revealing a predominantly ferritic microstructure with martensite islands. XRD demonstrated the absence of any retained austenite, Fig. A1. The V-Mo samples had a higher volume fraction of martensite (~20%) than the Ti-Mo sample (~5%). The average ferrite grain size of the Ti-Mo steel was 9.7 μm and the V-Mo steel was 19.1 μm.
Representative TEM micrographs complete with diffraction patterns from electropolished samples are shown in Fig. 3 (c-f). Precipitates were present in parallel rows showing classical interphase precipitation. For the Ti bearing steel, the two types of precipitate were observed throughout the sample, namely TiC (finer) and Ti 2 C (coarser), while for the V bearing steels, VC (finer) and V 4 C 3 (coarser) were observed. The TiC and VC formed by classical planer interphase precipitation while the Ti 2 C and V 4 C 3 , that had a much wider row spacing, formed through curved interphase precipitation. Each adopted one variant of the Baker-Nutting orientation relationship [33]. The smallest precipitates of both TiC and VC were shown to be fully coherent with the matrix. The larger precipitates were plate-shaped, with the larger interfaces being semicoherent with the matrix. The size distribution of the precipitates is given in Fig. 4. The number density of interphase precipitates in the Ti-Mo steel was 550±38 μm -2 , which was slightly larger than in the V-Mo steel with 420±31 μm -2 . The average size of the precipitates in the Ti-Mo steel was 5.8±0.1 nm. For the V-Mo steel the average size of the precipitates was 7.9±0.1 nm.

3.2.TDS
TDS was undertaken to investigate the hydrogen trapping behaviour of the two steels. In the first instance hydrogen charging was undertaken for various different times and the TDS spectra are presented in Fig. A2. This demonstrated that after 48 h the sample was fully Where  is the heating rate (K/min), T max is the temperature of the peak in the TDS spectrum, . The activation energies measured from these peaks, without any deconvolution to the peaks, is presented in Table 2.
In order to determine the specific microstructural consistent contribution to hydrogen trapping TDS were acquired at a heating rate of 100 ºCh -1 , with the spectra given in Fig. 5.
The low temperature peak is associated with several different trapping sites including dislocations, grain boundaries, martensite lath boundaries, coherent precipitates and vacancies, convoluted into a single experimental peak. The activation energies for these trapping sites have been widely reported in the literature and will be discussed later.
Following what has now become standard procedure (the procedures are given by Lee & Lee [35] and Depover and Verbeken [36]), the low temperature peak was deconvoluted into three peaks. The results for each steel are presented in Fig. 6 and corresponding activation energies for the different peaks are summarised in Table 2. The hydrogen concentration for each of the Hydrogen embrittlement mechanisms in advanced high strength steel simulated peaks was calculated and is shown in Table 3 along with total hydrogen concentration in the low temperature peak. The second, high temperature, peak for the Ti-Mo steel covered a wider temperature range and with a lower peak temperature than for the V-Mo steel. However, the total hydrogen under the high temperature peak was similar for both steels (0.10±0.01wppm for the Ti-Mo steel and 0.14±0.01wppm for the V-Mo steel), Table 3. The differences in temperature range and peak temperature were probably a result of the differences in precipitate type, discussed later.

3.3.Microstructural changes on hydrogen charging
The microstructures of charged samples were analysed by TEM. All samples were imaged under the same conditions, namely close to the <100> zone axis. Fig. 7 Table 4. precipitates. Similar images were obtained for the finer precipitates in the V-Mo steel. Fig. 8 (b) shows a bright field image of the V-Mo steel after charging, taken under the same imaging conditions as Fig. 8 (a). There has been significant changes to the image contrast on charging. It was not possible to obtain the classic lobe diffraction contrast from the precipitates after charging. In addition the size of the strain field, as shown by diffraction contrast, was much larger after charging, Fig. 8(b). To further explore the effect of charging on the strain field around the finer precipitates, high resolution imaging was undertaken. Fig.   9 show high resolution TEM (HRTEM) and associated FFT on the precipitates without and with hydrogen charging. In the uncharged state, Fig. 9 (a-c), it is relatively straight forward to image both the precipitate and the matrix and to get an understanding of the interface between the two. This was true for both steels and was identical from grain to grain. However, on charging, it proved impossible to get a clear HRTEM image from the matrix around the precipitate in the strain field region, Fig. 9 (d-f). The images show that the distortion occurred in the strain field in the matrix, but there was no evidence of distortion in the precipitate itself ( Fig. 9 (f)).
The residual stresses were measured in the uncharged and charged states using the method described in the experimental procedure. Compressive stresses were found in both alloys in the uncharged state, with similar magnitude, Table 4. The value of this stress increased on charging for both alloys, with a much more significant increase for the Ti-Mo alloy compared to the V-Mo alloy.   Table 5. The susceptibility of the Ti-Mo steel to HE was much greater than in the V-Mo steel. The work hardening rate as a function of true strain for the two steels are shown in Fig. 10  immediately after yield when compared to the uncharged samples. As straining proceeded, the work hardening rate of the charged samples became similar to that of the uncharged samples, but failure occurred soon after.

The effect of the hydrogen charging on crack propagation and the deformation substructure in tensile samples tested to failure
The morphology of the tensile fracture surfaces on the two steels are shown in Fig. 11. In the absence of hydrogen, the fracture surface of both steels showed the expected microvoid coalescence typical of tensile ductile failure, Fig. 11 (a) and (b). When charged for 48h a mixed morphology fracture surface was observed with some microvoid coalescence, but otherwise the surface dominated by quasi-cleavage facets, Fig. 11 (c) and (d). The extent of microvoid coalescence was greater for the V-Mo steel than the Ti-Mo steel, which was dominated by quasi-cleavage fracture. Cross-sections of the tensile fracture surface showed that any secondary cracking seen in the plan view specimens did not extend far below the fracture surface and that there was no evidence of subcritical crack growth, Fig. A4.
The quasi-cleavage fracture was largely in the ferrite. Moreover, the morphology of the quasi-cleavage fracture allowed crack initiation sites to be identified, Fig. 12 (a) and (d).
Crack initiation was found to be entirely in the ferrite with no evidence of the martensite playing a part. The initiation site often had a spherical appearance, suggesting that it was associated with a second phase particle such as MnS. EDS analysis of the crack initiation sites found no evidence of second phase particles. Multiple FIB sections were removed from the initiation sites to investigate them in more detail. In none of these were second phase particles found. Selected images are given in Fig. 12 (b) and (c) for Ti-Mo steel and (e) and (f) for V-Mo steel. The initiation sites in both steels were similar in characteristics. No evidence of second phase particles was found at any initiation site. Rather the shape of the feature arises from a dislocation wall that separates a region of low dislocation density from the surrounding microstructure Fig. 12 (b, e). A detailed analysis of the misorientation across these boundaries was presented in [58]. It was shown that the misorientation across these boundaries was typically 15-20°, but the one shown in Fig. 12 (e) was measured as 57°.
Secondary cracking was also evident, Fig. 12(f).  Interestingly, crack propagation was clearly between the rows of precipitates. The crack adopted a tortuous crack path at the atomic scale by switching from {100} to {110} planes to ensure that the overall propagation route fell in between the precipitate rows. This effect is further shown in Fig. 16 (d) and (e), which is taken from the crack wake. Thus, crack propagation was clearly affected by the strain field around the precipitates.
Although failure in hydrogen charged samples appeared to be associated with initiation in the ferrite grains, cracking was also found associated with the martensite and martensite/ferrite boundaries. Fig. 17 shows what is perhaps best described as voiding in the martensitic regions in the V-Mo steel. Fig. 17 (a) shows voiding at a ferrite/ martensite boundary (top), with voiding along the martensite lath boundary in the lower part of the image. The latter martensite region is magnified in Fig. 17 (b). There is also extensive voiding at triple junctions, as shown in Fig. 17 (c) and (d), with small cracks growing into the ferrite.
Although these features were significant, we found no evidence that they played a role in the failure mechanism. It would appear that these were more ductile aspects of the damage accumulation in the samples, while failure was associated with the cleavage fracture.

The crack growth behaviour from notched tensile samples
To understand the effect of hydrogen on crack growth from a notch, two notched tensile samples were used, whereby failure occurred from one notch, while the other notch exhibited extensive damage, but no failure. EBSD was used to understand the crack growth behaviour from this second notch.   19 shows EBSD maps taken from the edge of the notch from the hydrogen charged Ti-Mo steel. Cracks were observed that initiated from the notch and propagated in a transgranular manner in a direction perpendicular to the tensile axis. Fig. 19 (a-c) shows one of many similar such cracks. The initiation region was difficult to study but it was clear that the crack propagated across a grain along the trace of the {100} planes. The crack was arrested at the grain boundary, with evidence of crack tip plasticity in the adjacent grain as shown by the KAM image in Fig. 19 (a). The KAM map gives a measure of the geometrically necessary dislocation (GND) distribution in the sample. Interestingly, the evidence of residual lattice curvature around the crack and crack tip was minimal. Fig. 19 (d-f) shows a crack in the middle of the specimen, equidistant between the two notches. Again, the crack was transgranular and followed the {100} trace in going from grain to grain, Fig. 19 (f). Fig. 19 (d) is a KAM map. Interestingly, the cracking does not occur in regions of the highest GND density, rather it appears to propagate in grains with minimal GND.

Discussion
In this study, Ti-Mo and V-Mo high strength steels have been analysed in detail before and after hydrogen charging to understand the mechanisms of hydrogen embrittlement and how changes in precipitate type effect the mechanisms. For both steels, hydrogen charging had a significant effect on the ductility of tensile samples, Fig. 10. The TDS curves in Fig. 5, quantified in Tables 2 & 3, show two peaks for both steels. TDS measurements are accurate to better than ±5%, so the differences between the two steels is significant. The first peak is known to be associated with trapping at a number of different sites, namely dislocations, grain boundaries, martensite lath boundaries and at coherent precipitates [38][39][40][41][42][43][44][45][46][47][48][49]. In order to determine the origin of the low temperature peak, it was deconvoluted into 3 peaks, Fig. 6, with activation energies and trapped hydrogen content given in Table 3. The best fit yielded three peaks with activation energies of 18, 35 and 55kJ/mol for the Ti-Mo steel and 18, 35 and 58kJ/mol for the V-Mo steel.
TDS gives indirect information on trapping and is a combination of several different factors, such that interpretation is complicated. Moreover, for carbide traps, small changes in the composition of the carbide or the nature of the interface with the matrix have significant effects on the trapping behaviour. In the current steels, charging was found to increase the dislocation density, which is a new observation as the vast majority of investigations did not compare the uncharged and ascharged microstructures, which will be discussed in detail in a subsequent section. The Ti-Mo exhibited a much higher dislocation density after charging (23.9x10 13 m -2 ) compared to the V-Mo steel (11.6x10 13 m -2 ), Table 4. The hydrogen trapped in this 1st peak was much larger for the Ti-Mo steel (0.21±0.02wppm) compared to the V-Mo steel (0.14±0.02wppm), Table   3. These values are consistent with the much larger dislocation density in the Ti-Mo steel compared to the V-Mo steel. There was a difference in the grain size and in the volume fraction of martensite between the two steels. The Ti-Mo steel had the finest grain size (9.7μm) compared to the V-Mo steel (19.1μm) and so had approximately 4x larger grain boundary area, which would lead to greater trapping at these sites and therefore a larger TDS peak. However, the V-Mo contained a greater fraction of martensite (~20%) compared to the Ti-Mo (~5%), which would have contributed to the hydrogen trapped in the second peak. The differences in grain size and martensite lath boundary content between the two steels will have offset each other in the size of peak 2 leading to the similar amount of hydrogen trapped in peak 2, Table 3.
The third peak obtained by deconvolution of the low temperature peak is believed to be associated with reversibly trapped hydrogen at fine coherent precipitates. The measured activation energy of this peak of ~55kJ/mol for the Ti-Mo steel and 58kJ/mol for the V-Mo steel is consistent with values in the literature. Depover and Verbeken [43] report activation energies of 52 kJ/mol, associated with the carbide/matrix interface in a quench and tempered In the current work the carbides were not formed through a quench and temper, as investigated in the literature, rather through interphase precipitation. Interphase precipitation resulted in a bimodal size distribution of the precipitates (Fig.3), with finer coherent carbides and slightly coarser semi-coherent particles (i.e. coherent TiC and semicoherent Ti 2 C for the Ti-Mo steel and coherent VC and semicoherent V 4 C 3 for the V-Mo steel). Peak 3 was therefore believed to be associated with the finer, coherent TiC in the Ti-Mo steel and VC in the V-Mo steel. This is consistent with Takahashi et al. [54] who used atom probe to show no trapping of deuterium at fine coherent VC (believed to be a reversible trap), but trapping occurred at semi-coherent V 4 C 3 , which was believed to be an irreversible trap. Similarly, Echeverri Restrepo et al. [55] used DFT calculations to show only weak trapping at coherent VC/ferrite interfaces.
The second high temperature peak, Fig. 5, has been seen by numerous authors [40,47,[56][57][58][59][60]. The second peak has been universally assigned to irreversible trapping of hydrogen at carbides, including titanium, molybdenum and niobium based carbides. In the current work, the activation energy, Ea, associated with the high temperature peak was measured as 73.8kJ/mol for the Ti-Mo steel and 66.9 kJ/mol for the V-Mo steel. Drexler et al. [39] reported binding energy of 76kJ/mol for trapping at carbon vacancies at the precipitate/matrix (011) interface for TiC and 73.4kJ/mol for the same interface with V 4 C 3 . Similarly, Yoo et al. [47] report an activation energy of 83.4kJ/mol for TiC precipitates in a 'low carbon' tempered martensitic steel. Wei et al. [56], report that the the E a of TiC in martensite can be in the range of 46-86.9 kJ/mol, with the variation arising from differences in interfacial coherency. Thus, the current values are in-line with those reported in the literature, with any differences associated with the specific structure of the interface between precipitate and matrix. In any event, trap binding energies of these values are generally considered to be irreversible traps and not likely to contribute to hydrogen embrittlement [43].
The hydrogen content in the high temperature peak was ~1.4x higher in the V-Mo steel compared to the Ti-Mo steel. Interpretation of the high temperature peak is complicated by having a bimodal carbide size distribution in both the Ti-Mo and V-Mo steels. In each case, the finer, coherent, precipitates led to the simulated peak 3 of the low temperature peak, while the coarser semi-coherent precipitates led to the high temperature peak. The proportion of coarser particles was greater for the V-Mo steel compared to the Ti-Mo steel, Fig. 4, consistent with the greater proportion of semi-coherent precipitates in the V-Mo steel.  [55] show that the perfect coherent interface traps H only weakly, but much stronger trapping can be expected at C vacancies, both at the interface and in the bulk of the VC.
The work of Kawakami and Matsumiya [61] is consistent the recent 3D atom probe studies of Chen et al. [65] on deuterium charging of a steel that was nearly identical to that used in the current work. They carefully analysed the location of the deuterium and clearly demonstrated that it was trapped within the V 4 C 3 precipitates [65]. In contrast, Takahashi et al. [54] found no trapping of deuterium at VC precipitates in an underaged steel, but believed that deuterium was trapped at misfit dislocations at the ferrite/V 4 C 3 interface in a peak aged sample. In a similar study on relatively large TiC plate-like precipitates in a ferritic steel, Takahashi et al.

Changes in microstructure on charging
The charging clearly modified the strain field around the precipitates in both steels, Figs. 8, 9. Diffraction contrast TEM images, Fig. 8 (b) showed that the strain field around the precipitates was greatly increased by hydrogen charging. HRTEM images indicated that changes occurred in the matrix adjacent to the precipitate, with changes in the precipitate itself too small to detect, Fig. 9 (d-f)). In addition, the residual stress in the samples measured by XRD increased substantially on charging (Table 4), which is consistent with this enlargement of the strain field (this specific point is discussed later). This is the first observation of hydrogen significantly changing the strain field around precipitates. As discussed later, the strain field around the precipitates played a significant role in determining the crack propagation at the atomic scale.
The hydrogen charging increased the dislocation density in both steels quite substantially, steel, around 3x. Not only was the dislocation density increased but also the dislocation structures were different. After hydrogen charging dislocations were generally pinned by precipitates, and in several places arranged in tangles forming rudimentary cells structures ( Fig. 7 (c) and (d)). In contrast, in the uncharged state random dislocation distributions were observed ( Fig. 7 (a) and (b)). Similar increases in dislocation density have been observed on charging in the literature [67,68]. Barnoush et al. [68] used classical dislocation theory to model homogeneous dislocation nucleation and showed that hydrogen reduces the activation energy for homogeneous dislocation nucleation (HDN). Thus, they demonstrated that hydrogen reduces the required stress for HDN by a reduction in the shear modulus, dislocation line energy and stacking fault energy. Moreover, they showed that the hydrogen induced activation energy for HDN could be used to rank the sensitivity to hydrogen embrittlement of various metal alloys. The residual stresses could reach levels sufficient for triggering dislocation activity [69]. In the current work, it is probable that the increase in dislocation density on charging was associated with the increase in strain in the strain field around precipitates. As quantified in Section 4.2.3, this factor was much greater for the Ti-Mo steel than the V-Mo steel. On the other hand, the formation of dislocation cells is regarded as a sign of a reduction in the interaction energy of dislocations, also due to hydrogen. Of course, dislocations pinned by precipitates in this way will provide a local source of hydrogen, with hydrogen accumulations being key to failure.

Tensile work hardening
For the V-Mo steel, the work hardening rate immediately after yield was lower for the hydrogen charged specimen compared to the hydrogen free, Fig. 10, although work hardening rates became similar at higher strains. While in many steels, the work hardening rate is the same for charged and uncharged specimens, there have been several observations of different work hardening rate after charging [70][71][72][73], although there was little attempt to explain the observations. A reduction in flow stress could be explained by cracking and/or voiding in the sample during tensile deformation. However, we undertook detailed investigations and confirmed that there was no cracking in the samples, except at the final failure site. While there was some voiding at subgrain walls, Fig. 17, this was too small to account for the differences in flow stress and was only prevalent at higher strains when work hardening rates were similar in charged and uncharged samples. Therefore, the lower flow stress is more likely to be associated with the higher density of more mobile dislocations in the charged specimen, before normal work hardening processes set in. Indeed, calculations for the same steels investigated here, presented in detail elsewhere [37], indicate that hydrogen leads to an increase in dislocation mobility of 14 to 17 times.

Dislocation substructure and crack initiation
For the uncharged case, both steels exhibited classic ductile failure with microvoid coalescence, as expected, Fig. 11  FIB samples were removed from the fracture surface, both from the initiation sites and propagation regions (the latter discussed in the next section). Hydrogen charging promoted a higher dislocation density prior to testing (Table 4) and that was also reflected in the failed specimens. The crack initiation sites are shown in detail in Fig. 12. The shape of the crack initiation, with its "fish eye" appearance suggested that second phase particles, such as MnS or carbides would be the source of initiation. However, despite extensive investigation, no second phase particles were seen. Instead, a unique effect was observed, whereby cracks were initiated at low-energy dislocation walls. It is well established that hydrogen charging leads to an increase in dislocation mobility in steels [74]. Calculations for the same steels investigated here, presented in detail elsewhere [37], indicate that hydrogen leads to an increase in dislocation mobility of 14 to 17 times. This results in the accelerated dislocation cell and subgrain formation observed in this work. The dislocation walls at the crack initiation sites exhibited a pronounced misorientation between a region of low dislocation density (light contrast in Figs. 12 (b), (e)) and the surrounding region. Detailed modelling [37] showed that the crack initiated as a result of this strain partitioning. In forming low energy dislocation structures, the hydrogen is released to produce substantial local hydrogen concentrations [75][76][77], originating cracks. Propagation after initiation was by quasi-cleavage, discussed in the next section.
The observation of crack initiation at low energy dislocation walls could be taken as consistent with the HELP mechanism. HELP is explained in terms of a hydrogen shielding model in which the presence of hydrogen atmospheres around dislocations and elastic obstacles decreases the interaction energy between them [20]. The decreased barrier to dislocation motion and the increased mobility leads to localised deformation, with reduced dislocation separation at dislocation pile-ups which results in the nucleation of microcracks.
While the current observations show similarities to the HELP mechanism, the observation of cracking associated strain partitioning at low energy dislocation structures is unique. Fig.13 gives a general view of the fracture surface in a region that was clearly associated with crack propagation, but within 50 m of where the crack was known to initiate. FIB samples were removed from these regions to examine the dislocation structures, the crack propagation and the crack tip associated with secondary cracks. The crack propagation is clearly strongly influenced by the precipitate structures, with ridges on the surface of the crack associated with the coarser precipitate rows, Fig. 16. High resolution plan views of the cleavage surface made it look like potential surface voids, but the detailed cross-sectional TEM of the cleavage surface clearly showed that the surface topography was associated with small changes in crack propagation in the vicinity of precipitates, but that the crack remained crystallographic at all times.

Crack propagation
A detailed analysis was made of these crack surfaces and where possible, crack tips associated with secondary cracking. In the uncharged case, the cracking had no specific crystallographic nature, with the features better described as voids. In contrast, in the hydrogen charged samples, crack propagation was clearly shown to be crystallographic, with growth along {100} and {110} planes, Fig. 16. Interestingly, the cracks grew specifically in between the precipitate rows. The change from {100} to {110} and back again to {100} allowed the crack to avoid the compressive strain field associated with the precipitates. Thus, the strain field around the precipitates, which was enlarged by hydrogen charging, had a key role in determining the quasi-cleavage crack propagation.
The dislocation density associated with the crack tip of a propagating crack was low, as shown in Figs. 15, 16 and 19. Indeed, no evidence was found of dislocation activity at the crack tip that would be consistent with a HELP type mechanism. In addition, the dislocation density observed in FIB specimens taken from the crack surface, away from the crack tip (i.e. the crack wake), also exhibited low dislocation density. This is in contrast to the observations of Nagao et al. [80] who observed intense plasticity below the crack wake surface in a quenched and tempered medium carbon martensitic steel. At first sight, the multiple observation of low dislocation density in the current work might be interpreted as being consistent with a HEDE mechanism [78][79][80]. However, Song and Curtin [26] have demonstrated that the HEDE mechanism is physically unlikely. They showed that a ductile to brittle transition occurs by the suppression of dislocation emission at the crack tip due to the aggregation of hydrogen driven by the crack tip stress fields. The absence of dislocation emission at the crack tip, as observed in the current work, removes the crack blunting mechanism. Cleavage then becomes the only available energy relieving mechanism. This mechanism requires hydrogen in the lattice around the crack tip. The TEM observations, Fig.   13, showed that dislocations induced on charging were trapped by the precipitate structures in the vicinity of the propagating crack. These would have provided a plentiful source of weakly trapped hydrogen. Nevertheless, the current observations could be taken as being consistent with a HEDE mechanism.

Comparison of the two steels and the role of residual stress
The significant difference in embrittlement between the Ti-Mo and the V-Mo steels has to be explained. The Ti-Mo repeatedly exhibited failure at yield, while the V-Mo steels retained significant ductility. There were several differences between the two steels on charging. TDS showed that the Ti-Mo steel trapped more hydrogen in low temperature, reversible, trapping sites than the V-Mo steel. The Ti-Mo steel had a higher grain boundary area (~4x the V-Mo steel) which resulted in ~2x the amount of hydrogen being trapped reversibly at these sites compared to the V-Mo steel. Charging promoted a much higher dislocation density in the Ti-Mo steel (23.9x10 13 m -2 ) compared to the V-Mo steel (11.6x10 13 m -2 ). Hydrogen was also reversibly trapped at coherent precipitates, with much more hydrogen trapped at TiC than at VC, Table 3. Moreover, the hydrogen charging increased the strain field around the precipitates.
The combination of these factors led to a significant increase in residual stress on charging, with the increase far greater for the Ti-Mo steel compared to the V-Mo steel, With above considerations, the following contributions to residual stress are calculated: Referring to Table 6, most total residual stress values are very close to those experimentally obtained ( Table 4). The UTS -YS is the stress amount available for plastic deformation.
Values of UTS and YS are given in Table 5. For the Ti-Mo steel, the UTS -YS is 220 MPa, lower than the stress due to the expansion of precipitates by trapping hydrogen, H  , of 230 MPa, Table 6

The crack propagation and growth mechanism from notched samples after hydrogen charging
To further understand the mechanism of early stage crack propagation, double notched samples were used whereby failure occurred from one notch, but the other exhibited some limited crack propagation. boundaries (e.g. Fig. 19 (c,f)) or propagated through the next grain with a different orientation to the stress axis, but still following the trace of the {100} plane, (Fig. 19 (c,f)).

The crack propagation mechanism in the ferrite Ti-Mo steel
KAM maps give an indication of the lattice curvature. The KAM map of the crack propagating from the notch, Fig. 19 (a) exhibited very little lattice curvature in the vicinity of the crack. The exception is a small amount of plasticity in the grain adjacent to the crack tip, but in this case the crack appears to have arrested at the grain boundary. The situation was more complex at secondary cracks in the centre of the specimen, Fig. 19 (d-f). The KAM map showed regions of extensive lattice curvature, but these tended not to be associated with cracking. The flanks of the cracks showed virtually no evidence of plasticity. However, the crack tips were associated with regions of high lattice curvature, but again, the curvature appears to be in the grain adjacent to the crack tip, with the crack, at least temporarily, arrested at the grain boundary. The overall conclusion is that there was no evidence of enhanced crack tip plasticity, rather the opposite. Therefore, these results are consistent with tensile tests, which suggests that enhanced dislocation mobility and dislocation pile-ups (i.e. HELP mechanisms) were not important in crack propagation. Rather the observations of propagation are consistent with the model of Song and Curtin [26] or indeed HEDE.

The crack propagation mechanism in V-Mo steels
The cracks in the charged V-Mo steel are shown in Fig. 20. Similar behaviour was observed to the Ti-Mo steel. However, secondary cracks were also extensively observed at the phase boundary between the ferrite and martensite colonies and propagating through the martensite, Koyama et al. [24]. This is a result of hydrogen preferentially diffusing to the inter-phase grain boundaries and the lath boundaries. Koyama et al. [24] take this as evidence of a HEDE mechanism operating both within the martensite and also at the martensite/ferrite boundaries, although they also found that HELP was also important. As discussed earlier with respect to the tensile results, this result can also be taken as consistent with the model of Song and Curtin, who demonstrate that embrittlement results from the suppression of dislocation emission from the crack tip and so crack tip blunting does not occur.
Surprisingly, the crack propagated parallel to the stress axis in several cases, which is similar to observations of Koyama et al. [24]. In the vicinity of the notch, Fig. 20 (a-c) this might be a result of the complex stress system. However, this phenomenon was also observed in the middle of the specimen, Fig. 20 (d-f). It clearly shows the preferential crack path along lath boundaries that is preferential to propagation perpendicular to the stress axis through the martensite.

Conclusions
In the present work, the hydrogen embrittlement has been analysed through the mechanical properties, TDS curves and microstructures such as dislocation activity and cracks in the Ti-Mo and V-Mo steels.
1. Thermal desorption spectroscopy showed two distinct peaks for both steels, with the low temperature peak associated with trapping at grain boundaries, phase boundaries, dislocations and coherent precipitates, and a high temperature peak associated with trapping at semicoherent precipitates. The Ti-Mo steel exhibited a larger low temperature peak, which was associated with the finer grain size but mainly the large increase in dislocation density on charging (which was much greater than in the V-Mo steel). The V-Mo steel exhibited a larger high temperature peak, which was believed to be associated with hydrogen irreversibly trapped within the precipitates. In contrast, in the Ti-Mo steel, the hydrogen was believed to be irreversibly trapped at the precipitate/matrix interface. The increase in dislocation density and the increase in strain field around the precipitates led to a large increase in residual stress on charging, with the extent far larger for the Ti-Mo steel than the V-Mo steel.

Declaration of interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.        (010)) showing the strain field around the precipitates in the Ti-Mo steel. a) Uncharged state, where the classic lobe contrast from coherent precipitates is observed. b) Charged state, where the diffraction contrast from the strain field is seen to greatly increase, and the classic lobe constrast is lost.