Hydrogen Permeability of Tectosilicate Glasses for Tank Barrier Liners

. The permeation of hydrogen gas was studied in meta-aluminous (tectosilicate) glass powders of Li 2 O ⋅ Al 2 O 3 ⋅ SiO 2 (LAS), Na 2 O ⋅ Al 2 O 3 ⋅ SiO 2 (NAS) and MgO ⋅ Al 2 O 3 ⋅ SiO 2 (MAS) systems by pressure loading and vacuum extraction in the temperatures range 210–310 °C. With this method, both the solubility S and the diffusivity D were determined, while the permeability was given by the product SD . For all glasses, S was found to decrease with temperature, while D increased. Since the activation energy of diffusion of H 2 molecules exceeded that of dissolution, permeation increased slightly with temperature. When extrapolated to standard condi-tions (25 °C), the permeability of tectosilicate glasses was found to be only 10 -22 –10 -24 mol H 2 (m s Pa) -1 , which is 8–10 magnitudes lower than most polymers. Thin glass liners of these compositions are expected to be the most effective barrier for tanks of pressurised hydrogen.


Introduction
Glasses with meta-aluminous (tectosilicate) composition such as eucryptite (Li 2 O⋅Al 2 O 3 ⋅2SiO 2 ), spodumene (Li 2 O⋅Al 2 O 3 ⋅4SiO 2 ) and cordierite (2MgO⋅2Al 2 O 3 ⋅5SiO 2 ) are the basis for commercially available glass-ceramics. Their structure consists of modifiers that fill completely polymerised silicate networks by introducing charge-balanced aluminium tetrahedra. Meta-aluminous melts are very good glass formers as they generally retain their complex polymerised structure even at high temperatures [1] and can therefore be produced by conventional melt quenching and shaped to thin (submillimetre) glasses of various geometries such as sheets and tubes by drawing.
When storing compressed hydrogen, materials with low H 2 permeability are required to minimise leakage over long standby and operating times. Hydrogen storing technologies on a volumetric basis for on-board applications, such as vehicle tanks, are currently fibre-reinforced composites with metal (type III) and polymer liners (type IV) to meet safety aspects at pressures up to 70 MPa [2], [3].
The loss of hydrogen gas through the wall of these tanks can be significantly reduced if glasses or glass liners are developed that have several orders of magnitude lower H 2 permeability than polymer matrix materials today [4]. Barth et al. [5] showed that the permeability of various polymer materials at room temperature is in the range of 10 -14 -10 -16 mol H 2 (m s Pa) -1 . It is known that inorganic oxide glasses are far below that limit, e.g. the archetypal silica glass has a lower permeability of 10 -19 -10 -20 mol H 2 (m s Pa) -1 [6]. However, from a glass-manufacturing point of view, fused silica requires the highest temperatures (about 2000 °C), which is costly and therefore less attractive for a potential H 2 barrier liner in tanks.
Previous work has shown [7] that minimising the accessible volume is a successful strategy to further hinder the migration and solubility of H 2 molecules in the network structure of silicate glasses. Two competing approaches have been identified that achieve a reduction in the total free volume firstly by modifying the network through the creation of non-bridging oxygen (NBO) and secondly by filling or stuffing polymerised (NBO-free) silicate networks through the introduction of charge-balanced network formers such as aluminium and boron tetrahedra. While in the former (modified glass structures) the reduction in free volume and thus the decrease in permeability seems to depend on the type of modifier cation (cations with high field strength such as Li are more effective than those with low field strength), in the latter (NBOfree glasses) a tendency in favour of aluminosilicates over borosilicates can be seen [7]. Note that NBO defines an oxygen that is part of a tetrahedron but is not shared with a second tetrahedron. NBO forms the ionic bonds with the modifiers.
By changing the network structure due to the introduction of modifier cations that compensate for the charges of either NBO or tetrahedrally coordinated Al 3+ and B 3+ , the melting and forming temperatures drop compared to silica glass, while thermal expansion coefficients increase. Both will favour the potential application as hydrogen barrier in multi-material composites. However, highly modified network glasses, such as alkali di-and metasilicates, have limited technological suitability due to their high kinetic fragility and strong tendency to crystallise or corrode. Therefore, fully polymerised aluminosilicate glasses were investigated here. In particular, eucryptite, spodumene and cordierite glasses are studied, since these compositions have a relatively high degree of substitution of silicon by aluminium (in eucryptite glass, nominally every second silicon tetrahedron is replaced by an aluminium tetrahedron) and therefore the lowest hydrogen gas permeation can be expected. To extend the spectrum of aluminosilicate glasses with tectosilicate composition, data from earlier measurements [8] for albite (Na 2 O⋅Al 2 O 3 ⋅6SiO 2 ), jadeite (Na 2 O⋅Al 2 O 3 ⋅4SiO 2 ) and nepheline (Na 2 O⋅Al 2 O 3 ⋅2SiO 2 ) glasses are added. To illustrate the outstanding barrier effect of a thin layer of these tectosilicate glasses (glass liner), the effective permeation through a pressurised hydrogen tank with glass lining is calculated.

Glass preparation
Three glasses of target composition Li 2 O⋅Al 2 O 3 ⋅2SiO 2 (eucryptite), Li 2 O⋅Al 2 O 3 ⋅4SiO 2 (spodumene) and 2MgO⋅2Al 2 O 3 ⋅5SiO 2 (cordierite) were melted from Al(OH) 3 , MgCO 3 , Li 2 CO 3 and SiO 2 industrial grade raw materials (~500 g batch) in an electric furnace and stirred using PtRh20 crucibles at Schott AG, Mainz, Germany. To avoid unwanted crystallisation during pouring, the eucryptite and spodumene melts were roller quenched to receive ≤ 1 mm thick glass ribbons. The cordierite melt was less prone to crystallisation and was therefore poured onto a steel plate to form a ~15 mm thick glass block and cooled to the estimated glass transition temperature with 30 K h −1 . The chemical composition of the glasses was not analysed. However, glasses of a similar composition (+ nucleating agents), previously melted from the same raw materials and following the same time-temperature protocol, showed only minor deviations from the target composition (see Tab. 1 of Ref. [9]). For the permeation experiments, parts of the glass blocks and ribbons were crushed (steel mortar) with one or two strokes and then sieved. This process was repeated with new starting pieces until about 1.9 g of the fraction 100-125 µm was obtained, while unground parts of the ribbons and glass block were used for buoyancy measurements. The surface area of the sieved powder fraction was not studied, as preliminary tests showed that the influence of the grain shape and size distribution within the sieved 100-125 µm fraction on diffusivity and solubility experiments is negligible. In addition, eight sodium aluminosilicate glasses of a previous study were used [8]. The composition of these glasses (reports on the analyses in [8]) ranged from ~50-90 mol% SiO 2 (all with Na 2 O/Al 2 O 3 = 1), including compositions close to the stoichiometric tectosilicates Na 2 O⋅Al 2 O 3 ⋅2SiO 2 (nepheline), Na 2 O⋅Al 2 O 3 ⋅4SiO 2 (jadeite) and Na 2 O⋅Al 2 O 3 ⋅6SiO 2 (albite). Furthermore, a silica glass from the same study was also examined as a reference material. Glasses were named after the element and molar percentage of the modifier cation that balances the charge of the tetrahedrally coordinated Al 3+ . In the lithium aluminosilicate system (LAS), glasses of the eucryptite and spodumene compositions were named Li25 and Li16.7, respectively, and the cordierite glass of the magnesium aluminosilicate system (MAS) was named Mg22.2. The glasses of the sodium aluminosilicate system (NAS) were: Na5, Na10.5, Na12.1, Na14, Na16.1, Na20, Na22 and Na23.8. The density of the prepared glasses was determined according to the buoyancy method at room temperature. The weight of individual pieces (0.2-0.5 g mass) was measured in air and in ethanol. The error of this method (based on repetitive measurements) was ≤ 0.2% and resulted in the values 2.363 g cm -3 (Li25), 2.374 g cm -3 (Li16.7) and 2.620 g cm -3 (Mg22.2). Density of the eight sodium aluminosilicate glasses and the silica reference glass was determined previously and reported in [8].

Hydrogen gas permeation through glass
The permeability of hydrogen gas P was calculated from hydrogen diffusivity D and hydrogen solubility S according to [10]: D and S were measured with the vacuum extraction device equipped with mass spectrometry (VHE) as described in detail in [8], [11]. For this purpose, glass powders (80-105 mg) were exposed to pressurised hydrogen at different temperatures for different exposure times to ensure that the hydrogen loading was fully saturated. D was then determined by fitting the VHE hydrogen degassing rate dQ/dt with the classical diffusion model for isothermal degassing from a sphere [12]: with t = isothermal degassing time, r = radius of the glass sphere, Q(t) = amount of hydrogen released at t and Q 0 = total amount of dissolved hydrogen. Herein, a uniform radius of glass spheres was assumed, which was taken from d 50 of the measured particle size distribution [11]. As the ratio Q(t)/Q ∞ is dimensionless, D was fitted in terms of I H2 , the VHE hydrogen ion current (in A) (Fig. 1). S was obtained from the same VHE degassing run as the total hydrogen amount dissolved per unit volume per unit external (loading) pressure according to [13]: with p 0 = standard pressure (≈ 1 bar), p L = loading pressure, C = the VHE calibration constant (mol H 2 A -1 s -1 ) and t = total degassing time (s) including the ramp from room temperature to the isothermal section, the waiting time at this temperature and the final ramp to 825 °C (to release the remaining, previously non-degassed hydrogen) (Fig. 1).
For the glasses Li25, Li16.7, and Mg22.2, powders of the sieve fraction 100-125 µm were exposed to hydrogen at 6 MPa and 210 °C, 260 °C and 310 °C. 24 h exposure was thoroughly applied to ensure complete saturation of the loading. The experiments were repeated up to 2 times. Since the diffusivity of the eight sodium aluminosilicate glasses and the silica reference glass had been determined previously [8], only solubility experiments were performed with powders of the sieve fraction 180-200 µm at 268 °C and the same H 2 loading pressure of 6 MPa for 16-52 h. The latter experiments were repeated up to seven times. The uncertainties of D and S determined according to the experimental powder method were factors of up to 3 and 0.7, respectively. Preliminary measurements showed that at exposure temperatures below 200 °C the amount of detectable hydrogen was too low, i.e. the lower detection limit of our instrument was reached, so that measurements at temperatures < 200 °C were not carried out.
The temperature dependence of the solubility S, diffusivity D and permeability P of hydrogen gas is best described by Arrhenius-type equations [14]: and where D 0 , S 0 and P 0 (P 0 = S 0 D 0 ) are the coefficients of hydrogen gas diffusion, dissolution and permeation at infinite temperature, respectively and E D , E S , E P (E P = E S + E D ) are the corresponding activation energies. R is the gas constant (8.314 J K -1 mol -1 ). It should be noted that for gas diffusion in glasses the temperature dependence of the pre-exponential terms in Eqs. (4) and (5) must be taken into account [14].

Effective hydrogen gas permeation through a glass lined tank
The effective hydrogen gas permeation coefficient P through a double layer tank wall (polymer of thickness d 1 with glass lining of thickness d 2 ) with the intrinsic hydrogen gas permeation coefficients P 1 (polymer) and P 2 (glass lining) was calculated by assuming additivity of the layer thickness d = d 1 + d 2 and the associated decrease in the hydrogen partial pressure p = p 1 + p 2 . For the flux J of the diffusing gas with the surface area A one has: The conservation of J leads to: The transformation of Eq. (8) leads to the following expression for the total hydrogen partial pressure p: Finally, for the effective permeation coefficient of the double-layered wall, we obtain: For d 2 = 0 (no glass lining) and thus d 1 = d, the result is e.g. P = P 1 (polymer).    2 shows Arrhenius plots of the solubility and diffusivity data of Tab. 1. For all three glasses, the hydrogen gas solubility decreases slightly with temperature, while the diffusivity increases more strongly. In comparison, the glass with the spodumene composition (Li16.7) shows the highest solubility and diffusivity values, whereas hydrogen molecules are dissolved least in the eucryptite glass (Li25) and can diffuse least in the cordierite glass (Mg22.2). Fitting Eqs. (4) and (5) to the data resulted in the pre-exponential constants and activation energies listed in Tab. 2. (4) and (5) to the data. Fig. 3 shows the hydrogen gas solubility together with the previously measured diffusivity [8] at 268 °C as a function of Na 2 O content of the NAS glasses and the reference silica glass. The dependence on the composition of S and D with increasing SiO 2 content is strongly non-linear with a sharp increase (of solubility) with increasing silica content and a local maximum of S (shoulder for D) at albite composition. The permeability P was calculated from Eq. (6). P was smallest (6.9 × 10 -20 mol H 2 (m s Pa) -1 ) for the Na23.8 glass close to the nepheline composition, about three orders of magnitude smaller than the reference silica glass. Since hydrogen solubility was only measured at this temperature, the unknown dependence on temperature, i.e., the parameters S 0 and E S of NAS glasses, was calculated from the relation between the activation energies for dissolution and diffusion of NBO-free glasses (E D was taken from [8]). For E S and E D in the units kJ mol -1 one has [7]:  The D data were added from an earlier diffusivity study [8] to highlight the similar composition trend of both parameters. P is given by the product SD. Compositions for quartz (Qz) and the tectosilicates albite (Ab), jadeite (Jd) and nepheline (Np) are shown. Lines are for visual guidance only. Table 2. Parameters of Arrhenius Eqs. (4)-(6) of hydrogen gas solubility, diffusivity and permeability, i.e. S 0 (mol H 2 m -3 Pa -1 ), E S (kJ mol -1 ), D 0 (m 2 s -1 K -1 ), E D (kJ mol -1 ), P 0 = S 0 D 0 (mol H 2 m -1 s -1 Pa -1 K -1 ) and E P = E S + E D (kJ mol -1 ).Specific tectosilicate compositions: spodumene (Spd), eucryptite (Ecp), albite (Ab), jadeite (Jd), nepheline (Np) and reference quartz (Qz). Key: *Data from [8].

Discussion
To explore the potential of glasses with tectosilicate composition for hydrogen gas barrier linings, the permeability in Fig. 4 was extrapolated to the ambient temperature range and compared with that of conventional polymers. Despite the considerable uncertainty (approximately ±1 log unit) of such a wide extrapolation, it can be seen that linings of these glass compositions can contribute to a significant reduction in hydrogen permeation and thus minimize hydrogen volume losses of after pressurised refuelling over long periods of time. Most striking, the cordierite composition was found to be particularly suitable because its permeation coefficient at 298 K is about 9 orders of magnitude lower than the lowest permeable polymer material (polyamide).   5 illustrates the possible reduction in hydrogen permeation that can be achieved with a glass lining. As an example, a spherically capped cylinder (radius r = 0.1 m, cylinder height h = 0.5 m, polymer wall thickness d 1 = 2 cm), a tank pressure of p = 70 MPa, and a glass liner thickness of d 2 =1 µm is considered. The hydrogen permeability of an example polymer (HDPE) and example glass (Li 16.7) at 273 K were taken from Fig. 4 as P 1 ≈ 10 -15 mol H 2 (m s Pa) -1 and P 2 ≈ 10 -23 mol H 2 (m s Pa) -1 , respectively. Simple calculations give a tank volume of V = 0.02 m 3 , a surface area of A = 0.44 m 2 . Without glass lining (dashed curves), the initial hydrogen loss during the first day is only 0.268 g (constant pressure during the first day). After 50 days, however, ∆Q reaches 13.3 g H 2 , which is already ~149 litres. In this way, dangerous amounts of hydrogen could be accumulated close to the tank environment. On the other hand, hydrogen dissolves into the polymer material at high pressure, which could cause materials degradation phenomena, e.g. promoting formation of microcracks during loading and reloading. With a glass lining, permeation and the hydrogen partial pressure to which the polymer wall is exposed are greatly reduced. Already, a 1 µm thick glass lining of a 2 cm polymer wall would reduce hydrogen leakage by about 4 orders of magnitude to ∆Q = 5.4 × 10 -5 g H 2 after the first day or 2.68 × 10 -3 g H 2 after 50 days. Furthermore, according to Eq. (8), the maximum hydrogen partial pressure to which the polymer wall is exposed is about 14 × 10 3 Pa, i.e. 4 orders of magnitude less than the inner tank pressure.

Conclusions
Eucryptite, spodumene and cordierite glasses with fully polymerised aluminosilicate network structures showed ultra-low hydrogen gas permeation at 210-310 °C. When extrapolated to room temperature, the permeability was found to be as low as 10 -22 -10 -24 mol H 2 (m s Pa) -1 , 8-10 orders of magnitude lower than that of most polymers. Thin glass linings of this composition are likely to be the most effective barrier for tanks containing compressed hydrogen.

Data availability statement
All original data supporting the results are given in Table 1 and Table 2.