Hydrogen purification by palladium-based membrane is one of the most well-known techniques for supplying high purity hydrogen to low temperature operated (1) PEFC in various electronic devices such as tablets, laptop computers and small vehicles. The compact methanol steam reformer that consists of a catalytic burner, reformer and hydrogen purification device in a single package was developed almost two decades ago (2). Over the years, various integrated systems, operating conditions and geometries have been investigated to obtain the optimum operating conditions for reliable performance (3–9). As a result, the compact reformer has several advantages (i.e. simple and compact) compared to complex systems with separated units (10).

Due to the difficulty and high risk of exposure to accidents, the transportation and storage of gaseous hydrogen are undesirable. Alternatively, the utilisation of alcohols is more practical due to their existence in a liquid form at ambient conditions. Methanol for instance has a relatively high hydrogen:carbon ratio and moderate reaction temperature compared to other alcohols (10, 11). Based on the two step equations of methanol steam reforming, the reaction between vaporised methanol and steam produces hydrogen and carbon dioxide along with carbon monoxide. However, the concentration of carbon monoxide in the electrode of PEFC must be equal to or less than 10 ppm (12). Otherwise, the carbon monoxide could deteriorate the electrode performance by reducing the active surface area for reaction and lowering the partial pressure of hydrogen (13–15).

Hydrogen purification by palladium based membrane is preferable rather than selective carbon monoxide methanation and selective carbon monoxide oxidation (16), in which very high purity of hydrogen and very high permeation flux can be obtained (17–19). PSA is an alternative technique to produce very high purity hydrogen (20). However, the high costs of installing PSA is not economical for small size (or small capacity) applications (21). In addition to the well-known inhibitive carbon monoxide produced from the aforementioned methanol steam reforming, carbon dioxide and excessive methanol from the same reaction (22), hydrogen sulfide (23, 24) and trace amounts of ammonia (20, 25) from coal gasification process, dehydrogenated methanol and ethanol (26) also affect adversely the performance of palladium membranes through its inhibitive mechanism.

Alloying palladium with other metals such as silver and copper is necessary during membrane fabrication to prevent embrittlement when the membrane is used under hydrogen atmosphere and below 573 K and 2 MPa (10). It was found that the maximum amount of permeated hydrogen is obtained when the silver content is 23% (27). In addition, the palladium/silver membrane shows better performance compared to the palladium/copper membrane from the viewpoint of hydrogen permeability. Consequently, there is growing interest in the palladium/silver membrane. In addition to these two types of alloys, the palladium-yttrium membrane has also been previously examined due to its superior hydrogen permeability. However, this type of membrane is not commercialised due to the expensive processes required to convert the palladium-yttrium alloy into the functional separation membrane (28).

Hydrogen permeation through palladium based membranes is based on the solution-diffusion mechanism (29). When a membrane with sufficient thickness is operated at sufficiently high temperature, the diffusion of hydrogen atoms through the metallic lattice becomes dominant, thus permeation flux can be estimated accurately by using Sieverts’ Law (30, 31) that is quantified in Fick’s First Law as follows, Equation (i):

(i)

where f is the hydrogen permeation mole flux, q is the hydrogen permeance coefficient, d is the membrane thickness, P_H2,1 is the hydrogen partial pressure at membrane surface of upstream (or retentate) side and P_H2,2 is the hydrogen partial pressure at membrane surface of downstream (or permeate) side. The Sieverts’ equation as shown by Equation (i) states that hydrogen permeation is governed by the difference in the square root of the hydrogen partial pressure between the upstream and downstream side.

Membrane temperature (22, 32) and membrane thickness (33–35) are among the important parameters that determine compliance with Sieverts’ Law. When the temperature of the membrane is sufficiently high, the adsorption and dissociation of hydrogen atoms at the membrane surface are very fast. Therefore, the diffusion of hydrogen atoms through the metallic lattice becomes a controlling step for permeation. In this case, the hydrogen permeation flux is found to be linear with respect to the difference in the square root of hydrogen partial pressures between the upstream and downstream side. For the case of the palladium/copper membrane, Ma found that Sieverts’ Law is valid only when the membrane temperature is set above 573 K (32), whereas for the palladium/silver membrane, the law can be applied even if the temperature is lower than 500 K (22).

The different findings by several groups of researchers had proven that there is no exact value of thickness that can be set as a limit for compliance with Sieverts’ Law. Ward and Dao (33) and Federico et al. (34) found that the membrane thickness should be higher than 10 μm in order to apply the Sieverts’ equation (Equation (i)). Other research groups discovered that the Sieverts’ Law is still valid even though the membrane thickness is below 10 μm (22, 35). These contrary findings are supposed to be caused by difficulty in quantifying various uncontrolled factors such as surface processes (36), surface poisoning (37) and grain boundaries (38).

For the case of pure hydrogen, hydrogen permeation is found to follow the correlation of Sieverts’ equation regardless of feed flow rate. However, for mixtures, when the feed flow rate of hydrogen becomes sufficiently high, the hydrogen permeation ratio (fraction of fed hydrogen that permeates membrane) (39) becomes low. Therefore, the hydrogen permeation flux can be predicted accurately by Sieverts’ equation (Equation (i)). Further, the term P_H2,1 in Equation (i) can be predicted from the bulk value of hydrogen mole fraction at the upstream side, which indicates that hydrogen partial pressures at the membrane surface and the bulk flow are uniform, as illustrated by Line 1 in Figure 1.

Fig. 1.

When a mixture of hydrogen with relatively low feed flow rate (or high permeation ratio) is used, the direct substitution of inlet hydrogen partial pressure (P_H2,in) value into the Sieverts’ equation causes an overestimation of P_H2,1 from the actual permeation flux. This indicates there is a nonuniformity of hydrogen mole fraction in the boundary layer near to membrane surface, as illustrated by Line 2 in Figure 1. In this case, several research groups (39–41) mentioned that the hydrogen permeation flux could start to affect the hydrogen concentration at the membrane surface, thus could trigger the phenomenon of concentration polarisation.

The concentration polarisation phenomenon causes the accumulation of the less permeable species and the depletion of the more permeable species in the boundary layer adjacent to the membrane, thus generating a concentration gradient in the boundary layer (42). Therefore, in such situation, an additional elementary step is essential for the solution-diffusion mechanism of the membrane. This involves the transportation of molecular hydrogen from (to) the bulk gas phase to (from) the gas layer adjacent to the surface at the upstream (downstream) side (33). As a consequence, if the inlet hydrogen partial pressure is directly substituted into the Sieverts’ equation (Equation (i)), the hydrogen partial pressure at the membrane surface of upstream side is overestimated, which causes a significant deviation from the actual permeation flux. Chen et al. observed that such deviation implies the level of concentration polarisation for a palladium based membrane (41). Based on the analytical study for multicomponent hydrogen mixtures, Caravella et al. confirmed that such deviation is caused by the effect of multicomponent external mass transfer (such as concentration polarisation) in addition to non-ideal diffusion through the membrane (43).

Technological advances in membrane materials, modification and fabrication (42) since the end of the 20th century have stimulated the fabrication of very thin membranes in order to substantially improve the permeation performance. Therefore, the phenomenon of concentration polarisation could not be avoided due to the use of membranes with minimal thickness. However, some researchers (44) have recommended installation of baffles in the membrane reactor to decrease the polarisation effect. For palladium based membranes, concentration polarisation has been extensively investigated in the past decade (40) although concentration polarisation had been investigated in the 1990s for separation of a gas-vapour mixture (45). Therefore, significant research interest in the theoretical understanding of the phenomenon has resulted in numerous methods for estimating hydrogen flux for such conditions.

Therefore, a comprehensive review of the transport phenomena for palladium based membranes is performed, particularly on concentration polarisation. In addition to the background of the scenario related to palladium based membranes already highlighted, current information on various parametric studies and theoretical approaches for predicting hydrogen permeation flux under the influence of concentration polarisation are covered. Therefore, this review presents critical scientific knowledge and current research on concentration polarisation. The coverage of the present review is significantly different from the published works of Adhikari and Fernando (46), Rei (47), Gallucci et al. (21), Al-Mufachi et al. (28), Li et al. (48), Conde et al. (49) and Peters and Caravella (50). Adhikari and Fernando comprehensively reviewed the classification of hydrogen purification membranes, along with the advantages and disadvantages of each type of membrane (46). The study emphasised the superior quality of palladium based membranes in producing ultra-high purity hydrogen due to very high selectivity (46). Similarly, Rei (47) reviewed the advances in permeation through palladium based membranes for the case of a hydrogen mixture based on case studies in Taiwan. Several discoveries on new phenomena of hydrogen permeation such as perturbation of hydrogen permeation due to palladium lattice expansion and hydrogen spillover in the membrane reactor have been described (47). Gallucci et al. also highlighted the problem of concentration polarisation in the membrane reactor, although this was not the main topic of the study (21). The authors mainly focused on the route of commercialisation and application of various types of membranes (21). The review of several palladium alloy membranes was presented by Al-Mufachi et al. (28). The paper highlighted the advantages and disadvantages of each type of membrane in terms of hydrogen permeability, tensile strength and fabrication costs (28). Subsequently, Li et al. reviewed the thermal and chemical stability of palladium based membranes which are considered the two most critical issues for the commercialisation of the membranes (48). In 2017, a review on palladium based membranes was performed by Conde et al. (49). The paper presented a review of the alloying elements for palladium based membranes and their effect on the membrane properties. Finally, a most recent overview by Peters and Caravella (50) has covered the scopes of manufacturing process of palladium membranes, membrane materials, membrane modules and reactor design, as well as applications of palladium based membranes.

Based on these published review articles, it can be said that the subject of transport phenomena, particularly on the concentration polarisation, is the novelty for the present review. This paper presents coverage of recent research features and technological advances on the transport phenomena of palladium based membranes. It also presents the various prediction methods applicable to hydrogen permeation under the influence of concentration polarisation that could serve as a future reference for researchers and industrial practitioners.

In this section, a review of case studies on concentration polarisation is presented. The various studies and types of membrane used for each study are presented in Table I, whereas the parameters considered for studying such phenomena are listed in Table II. Based on Table I, it is evident that most common membranes are tubular, as illustrated by the various configurations in Figures 2(a)–2(c). The studies by Faizal et al. (55, 60) examined the phenomenon of concentration polarisation for flat sheet type membranes, which are widely used in compact reformers for hydrogen production (4–5, 65–66). The common configuration for a flat sheet type membrane is illustrated by Figure 2(d). It is interesting to note that various configurations and fluid flow conditions have been considered for studying concentration polarisation. Based on Table II, it is evident that the most common varied parameters for concentration polarisation are operating pressure, hydrogen mixture composition, feed flow rate and Reynolds number as well as membrane temperature. However, several studies have focused on geometry improvement to suppress concentration polarisation, as elaborated in this section.

Fig. 2.

In the early 21st century, Hou and Hughes were among the earliest groups who observed the concentration polarisation during an experiment involving hydrogen permeation with a palladium based membrane (67). The authors confirmed the existence of the concentration polarisation phenomenon during their experiment for a membrane with a thickness of 5 μm to 6 μm. However, the effect was not severe due to the relatively high feed gas velocity, which was a 5% decrease in hydrogen concentration for the various mixing ratios of the binary mixture of hydrogen and nitrogen (67).

Zhang et al. (51) were one of the pioneer groups who comprehensively investigated the concentration polarisation phenomenon specifically for palladium based membranes. The effect of various parameters such as pressure, temperature, feed gas flow rate and permeability was investigated experimentally. Also, the parameters were interpreted through mathematical modelling particularly for tubular membranes with porous ceramic supports. The authors found that when the feed gas flow rate is increased, the concentration polarisation is weakened, resulting in higher hydrogen permeation. For instance, for the case of membrane temperature of 723 K, when the feed flow rate was 5 ml s⁻¹ (equivalent to 5 × 10⁻⁶ m³ s⁻¹), the concentration polarisation degree for hydrogen (ratio of hydrogen permeation flux with concentration polarisation to hydrogen permeation flux without concentration) was around 0.54. However, when the feed flow rate was increased to around 14 ml s⁻¹ (equivalent to 14 × 10⁻⁶ m³ s⁻¹), the concentration polarisation degree for hydrogen became 1, that is no effect of concentration polarisation on hydrogen permeation flux was found. Furthermore, the authors reported that the observed phenomena are mainly due to the higher removal rate of accumulated nitrogen in the boundary layer at higher feed flow rates (51). However, an increase in pressure at the retentate or upstream side (at constant permeated pressure) increases concentration polarisation, as clearly described by the mathematical modelling developed in the study. Based on their study, at constant temperature of 723 K, for the case of pressure of 2 atm (equivalent to 202.6 kPa), the concentration polarisation degree for hydrogen already reached a value of 1 (no effect of concentration polarisation) when the feed flow rate was around 14 ml s⁻¹ (equivalent to 14 × 10⁻⁶ m³ s⁻¹). However, for the case of higher pressure of 4 atm (equivalent to 405.2 kPa), the concentration polarisation degree for hydrogen still not reached value of 1 even though feed flow rate was increased to 32 ml s⁻¹ (equivalent to 32 × 10⁻⁶ m³ s⁻¹). Based on the model, the observed trend is related to the proportional relation between the mass transfer coefficient of the retentate side and the diffusion coefficient, which is reciprocal to operating pressure (51). Therefore, the findings of Zhang et al. corroborated the previous findings by Morguez and Sanchez (68), which reported that the effect of selectivity is less significant compared to feed gas flow rate, pressure and permeability (68). The authors also observed that the temperature range used in their experiment did not trigger the concentration polarisation phenomenon. However, the authors did not rule out the possibility of concentration polarisation when the hydrogen permeation rate is enhanced due to the increase in temperature (51).

Pizzi et al. performed an experimental study for ultra-thin (~2.5 μm thickness) palladium/silver membranes deposited on ceramic supports. The authors observed that pronounced concentration polarisation occurred regardless of the composition mixture (52). As a result, the phenomenon was evident despite the concentration of nitrogen in a binary mixture of H₂:N₂ being relatively very low (12 vol%) (52). For this case, it was found that when Sieverts’ driving force is set to 0.8 bar^0.5 (equivalent to 253 Pa^0.5), the hydrogen permeate flux is only 0.26 mol s⁻¹ m⁻², that is significantly lower if compared to the permeate flux obtained when the concentration polarisation effect is negligible (permeate flux of 0.68 mol s⁻¹ m⁻²).

The lack of extensive studies on the detailed mechanism of concentration polarisation specifically for palladium based membranes prompted Catalano et al. (40) to explore this research area. The findings of Catalano et al. demonstrated a similar trend with that of Zhang et al. (51) in terms of the effect of feed flow rate, and Pizzi et al. (52) in terms of the effect of mixture composition on the concentration polarisation phenomena. Compared to previous research, a fundamental investigation on the ternary mixture of H₂:N₂:CH₄ (volume ratio of 50:25:25) was also conducted for the first time. The findings showed that the permeation flux was marginally less than the flux for the binary mixture of H₂:N₂ (volume ratio of 50:50) but almost similar to the H₂:CH₄ (volume ratio of 50:50) mixture. Therefore, the findings reveal that there was a very slight decrease in permeation flux, which occurred when nitrogen was replaced with methane (40). In this case, previous researchers have confirmed that the inhibitive effect caused by methane is very minimal, thus can be neglected (69). Based on the findings of Catalano et al. (40) and Jung et al. (69), it is evident that the severity level of concentration polarisation is independent on number of species (binary or ternary) in the non-inhibitive hydrogen mixture. Catalano et al. also have interpreted the level of concentration polarisation for various operating conditions using a dimensionless polarisation number, which is defined as the ratio of gas phase to membrane sensitivity factor (40). The polarisation number of much higher than 1 (S>>1) means concentration polarisation is dominant, whereas when the number is much less than 1 (S<<1), resistance by the metal membrane controls the entire permeation process. The authors discovered that for both cases of ternary and binary hydrogen mixtures with feed flow rates of 1 Nl min⁻¹ to 3 Nl min⁻¹ and relatively low inlet hydrogen concentration (50 vol% H₂), the concentration polarisation becomes dominant, that is S>>1. However the value of S is reduced when hydrogen concentration or feed flow rate is increased. As example, for the case of binary mixture of H₂:N₂ (50 vol% H₂ and 50 vol% N₂) with operating temperature and total pressure of 673 K and 600 kPa, respectively, when the feed flow rate was increased from 1 Nl min⁻¹ to 3 Nl min⁻¹, S reduced significantly from 6 to 2.5.

Most of the studies on concentration polarisation elaborated previously used palladium based membrane with support that influences the hydrogen permeation process (70). Conversely, Caravella et al. (53) examined the concentration polarisation phenomenon on self-supported palladium-based membrane in which case the effect of support was eliminated. In addition to the previous studies, other researchers (40, 51–52), have analysed the broader range of upstream hydrogen molar fraction, total upstream pressure, downstream total pressure, operating membrane temperature, membrane thickness and permeability to develop polarisation maps as a very useful guide for membrane reactor designer. The term concentration polarisation coefficient (CPC) was introduced in the maps as a demonstration for the level of concentration polarisation. Here, when the value of CPC is 0, it means no polarisation occurs while the value of 1 indicates the occurrence of total or maximum polarisation. Additional parameters were considered in this study that were not covered by the other previous researchers, namely: operating temperature, permeance, total downstream pressure and membrane thickness. Furthermore, the analysis demonstrated that the severity of concentration polarisation increases when the temperature and permeance are increased whereas the total downstream pressure and membrane thickness are reduced (53). As example, when Reynolds number, hydrogen retentate molar fraction, temperature, pressure at retentate side and pressure at permeate side were set to 2100, 0.40, 500°C, 1000 kPa and 200 kPa, respectively, the CPC increased significantly from around 0.13 to around 0.65 (thus concentration polarisation effect become stronger) when membrane thickness was reduced from 50 μm to 5 μm. Similar to the assertion by Morgues et al. (68), Caravella et al. (53) also found that the hydrogen flux itself plays a significant role during concentration polarisation.

Caravella et al. extended their analytical study to cover the concentration polarisation phenomena for a hydrogen mixture that contains well-known inhibitive species of carbon monoxide (22, 71–76). In this study (54), the authors simultaneously considered the effects of concentration polarisation and inhibition by carbon monoxide by merging their previously introduced approach (53) and the approach by Barbieri et al. (77). Similar to their previous study (53), the authors introduced a parameter so-called permeation reduction coefficient (PRC) that includes both polarisation and carbon monoxide inhibitive effects simultaneously. Interestingly, it was found that when the polarisation and carbon monoxide inhibition occur at the same time, the hydrogen permeation flux obtained is lower compared to the flux obtained when both phenomena occurred separately. This is mainly due to the polarisation of the inhibitor carbon monoxide toward the membrane surface, which increases the carbon monoxide partial pressure at the surface (54). The researchers found that for binary mixture of hydrogen and carbon monoxide, when operating temperature, membrane thickness, Reynolds number, hydrogen upstream molar fraction, total upstream pressure and downstream pressure are set to approximately 647 K, 60 μm, 1200, 0.60, 1000 kPa and 200 kPa, respectively, the permeating flux for the cases of polarisation only, inhibition by carbon monoxide only, and simultaneous polarisation and inhibition are around 85 mmol m⁻² s⁻¹, 71.5 mmol m⁻² s⁻¹ and 67 mmol m⁻² s⁻¹ (54). The inhibition of carbon monoxide under the influence of concentration polarisation is expected since the low feed flow rate is generally applied, for instance, during steam reforming for application in small portable electrical devices (8, 66).

The phenomenon of concentration polarisation was featured by the two-dimensional numerical method for tubular type (39) and flat sheet type (78) membranes. In general, for permeation with tubular type membrane under concentration polarisation influence, the direction of hydrogen concentration decrease for binary mixture of H₂:N₂ is from the region around the leading edge (inlet part) to the tailing edge (outlet part) (39). This phenomenon is featured by Figure 3, in which dimensionless hydrogen concentration gradient decreases from the leading edge to the tailing edge of the membrane surface regardless of hydrogen volume percentage. However, for the case with vertical flow towards flat sheet type membrane surface, the hydrogen concentration is highest at the centre of the membrane, and decreases in radial direction, as shown by Figure 4 (78). These studies investigated the hydrogen concentration distribution around the membrane surface for various important parameters such as operating pressure, hydrogen molar fraction, feed flow rate (or Reynolds number) and membrane permeance. The qualitative simulated results reveal that the hydrogen concentration decreases at the membrane surface due to the effect of hydrogen flux itself during the phenomena. However, the severity level of concentration polarisation for the various parameters above is generally similar to the values described quantitatively by previous experimental and analytical studies (40, 51–54). Chen et al. introduced an important parameter called hydrogen permeation ratio (HPR) to indicate the level of concentration polarisation (39). The HPR is defined as the ratio of hydrogen permeation rate across the membrane to the hydrogen feed rate at the inlet. The authors concluded that a decrease in value of HPR indicates that the severity of concentration polarisation is diminished. As an example, for the case of binary mixture of H₂:N₂ (H₂ mole fraction of 0.50) with pressure difference and membrane permeance of 30 atm and 10⁻⁴ mol m⁻² s⁻¹ Pa^−0.5, respectively, HPR decreased from 96 to 3 when Reynolds number was increased from 20 to 2000, thus severity of concentration polarisation is significantly reduced (39). It is interesting to note that when the aforementioned four important parameters (operating pressure, hydrogen molar fraction, feed flow rate and membrane permeance) were set in such a way to cause the effect of concentration polarisation to become very significant, the hydrogen concentration gradient is very high at the leading edge of the membrane (in the region near to the inlet) and then decays faster. Due to this phenomenon, there was almost no driving force for permeation in most of the remaining membrane length, as demonstrated by Figure 5 (refer to the case of permeance (K) of 10⁻³) (39). Figure 5 also demonstrates that when the permeance is increased, the tendency for concentration polarisation to occur increases. Based on Nagy et al. (79), a convex shape can be observed for the hydrogen concentration curve in a boundary layer during concentration polarisation phenomena, once the convective flow starts to play a role in the diffusive flow of the layer (80, 81). Based on these numerical simulation studies, Chen et al. also asserted that the concentration polarisation phenomenon is not significant when the hydrogen permeation ratio (H₂ permeation rate:H₂ feed rate) is less than 30%. It is important to note that even if the concentration polarisation can be improved (and more hydrogen flux can be obtained), the HPR that indicates hydrogen recovery becomes smaller, and this becomes a shortcoming for the membrane performance (39).

Fig. 3.

Fig. 4.

Fig. 5.

Subsequently, Chen et al. extended their numerical simulation to the tubular type membrane with simultaneous use of feed flow and sweep flow (41) under concentration polarisation influence. The advantage of using sweep flow to improve hydrogen permeation flux (and to abate concentration polarisation) has been confirmed by previous researchers (6, 7, 82–84). When the sweep flow rate at the permeated side is increased, the hydrogen partial pressure at the membrane surface of the permeated side can be reduced, thus hydrogen permeation flux increases (and concentration polarisation is weakened) due to increase in hydrogen partial pressure difference (7, 83). For instance, in the case of countercurrent mode with the use of sweep gas in the shell side (outside tubular membrane and inside shell), the improvement in hydrogen flux was improved by 12.3% when the sweep flow rate was increased from 2.713 mol s⁻¹ to 4.3408 mol s⁻¹, thus indicating the importance of sweep flow rate in improving hydrogen flux. In this case, the optimum flow rate of sweep gas can be estimated from the arctangent function (41, 85) of feed gas flow rate, once the flow rate or Reynolds number of the feed gas is specified. Here, the flow rate of sweep gas is considered optimum when the flow rate can give the maximum hydrogen permeation flux and sufficiently high hydrogen recovery (up to 95% H₂ recovery) could be maintained (41). It is interesting to note that the coupling between feed gas and sweep gas will give better separation performance when countercurrent mode is applied (5, 41). It is also interesting to note that whether the feed gas is issued in the lumen side (inside tubular membrane) or the shell side (outside tubular membrane and inside the shell) during countercurrent mode, the difference in hydrogen flux for both configurations is negligible. Therefore, this implies the independence of hydrogen flux on the position of the feed flow. Also, the difference in hydrogen permeation flux between cocurrent mode and countercurrent mode can be reduced by increasing the feed flow rate. Similar to the feed flow rate, a decrease in the sweep flow rate causes the effect of concentration polarisation to become stronger.

Part II (86) continues the discussion and provides the conclusions.

Miguel et al. examined the decrease in the hydrogen concentration along the membrane length of a finger-like configuration (1–3) for the case of binary hydrogen mixtures with inhibitive carbon monoxide or carbon dioxide, by replacing the terms of feed partial pressure of inhibitive species and the difference in the square root of hydrogen partial pressure in the Sieverts’ Langmuir equation (4) with the average partial pressure of inhibitive species and logarithm mean driving force (5), respectively. In this case, the authors obtained an excellent concordance between the predicted results obtained from their rearranged Sieverts’ Langmuir equation with the actual hydrogen permeation flux (2), which proved the existence of concentration polarisation during the permeation.

With the apparent advantage of using sweep gas during hydrogen permeation (6), Chen et al. have further investigated the concentration polarisation phenomena under sweep gas and baffles implementation (7). The flows of feed gas and sweep gas were in the form of countercurrent mode. It is interesting to note that higher hydrogen flux can be obtained from the membrane when a smaller diameter of the shell (smaller distance between the shell and tubular membrane) is used. As an example, when the pressure difference, temperature, mass flow rate of feed gas and Reynolds number of flow at the permeate side were set to 9 atm, 623 K, 267.48 mg s⁻¹ and 1000, respectively, the hydrogen flux for the cases of large, medium and small shell were 0.88 mol m⁻² s⁻¹, 0.96 mol m⁻² s⁻¹ and 1.03 mol m⁻² s⁻¹, respectively. This is due to the reduction of boundary layer thickness, as demonstrated by the numerical results (7). Besides, the introduction of baffles to the shell side causes disturbance to the boundary layer and more hydrogen is directed towards the membrane surface. Therefore, concentration polarisation is weakened and more permeated hydrogen can be obtained. Interestingly, due to the trade-off between the installation cost and slight improvement in permeation performance when more baffles are installed, one baffle installation has been recommended (7). In this case, Coroneo et al. also have asserted there should be an optimum number of baffles installed after observing just a slight improvement in permeated flow when the number of baffles is increased, from around 38% (two baffles configuration) to just around 46% (three baffles configuration) (8).

Further investigation by Chen et al. (9) then discovered the optimum baffles configuration for minimising concentration polarisation while obtaining maximum hydrogen recovery. In this case, the authors emphasised the importance of concentrating hydrogen at the membrane surface through the flow contraction mechanism. The optimum conditions for baffle installations are as follows: (a) installation of single baffle at shell wall; (b) installation at the leading edge of the membrane and (c) use of a sufficiently high ratio of baffle length to shell radius (ratio of 0.75) (9).

Faizal et al. (10) investigated the effect of hydrogen partial pressure and feed flow rate on the level of concentration polarisation for flat sheet palladium/silver membrane, despite widespread research interest in tubular type membranes. A third degree polynomial equation has been introduced as a tool to predict hydrogen permeation flux for such geometry. Based on the predicted profile of hydrogen mole fraction at the membrane surface, the difference between the predicted average hydrogen mole fraction at the membrane surface and hydrogen mole fraction at the inlet becomes larger at higher inlet hydrogen partial pressure. For the case of a binary mixture of H₂:N₂ (inlet hydrogen mole fraction of 0.75), when operating temperature, feed mole flux and hydrogen partial pressure at downstream (permeate) side were set to 623 K, 0.40 mol m⁻² s⁻¹ and 0.10 MPa, respectively, the aforementioned difference increased from 9% to 20% when inlet hydrogen partial pressure was increased from 0.150 MPa to 0.225 MPa. Therefore, the concentration polarisation was strengthened. Compared to the previous studies, the changes in the concentration polarisation level concerning the inlet hydrogen partial pressure and feed flow rate are similar qualitatively (10).

Chen et al. performed experimental studies on a H₂:N₂ mixture permeation test using high permeance tubular palladium based membranes (palladium and palladium/copper membrane with porous stainless steel) (11). Here, the thickness applied was from 6–7 μm. Similar to the previous study on ultrathin high permeance palladium/silver membrane with ceramic support (thickness of 2.5 μm) (12), the authors revealed that concentration polarisation was most affected by the concentration of the hydrogen feed, notably when the hydrogen concentration was decreased from 75 vol% to 50 vol%. The severity of concentration polarisation becomes higher even though the hydrogen partial pressure difference has been set to the same value. In this case, it can be noticed that in order to obtain the same hydrogen partial pressure difference, when the hydrogen partial pressure at the permeated side is set constant, higher total upstream pressure is necessary for smaller feed hydrogen fraction, and this increases the levels of concentration polarisation. Within their selected operating condition, feed flow rate and hydrogen partial pressure difference cause concentration polarisation as well, but with minor influence compared to the feed hydrogen concentration factor (11). Zhao et al. performed a permeation test for a mixture that was almost similar to coal gasification product (<40% H₂ and <40 ppm H₂S) to simultaneously determine the effect of sulfur contamination and concentration polarisation. The authors found that the influence of concentration polarisation was dominant for the mixture with lower hydrogen composition (50% mole fraction) especially at a low feed flow rate due to the minor effect of sulfur poisoning in the specified condition (13).

As a continuation of their previous study (10), Faizal et al. (14) investigated the concentration polarisation phenomena for various binary hydrogen mixtures with different inlet hydrogen mole fraction (0.70–0.80) and species (nitrogen, argon, helium and carbon dioxide). It is interesting to note that a mixture of hydrogen and argon was used due to different chemical characteristics, whereas the mixture of hydrogen and helium was used due to the different binary diffusivity compared to the hydrogen mixture that contained nitrogen (14). The authors compared the analytical results calculated by using their previously introduced theoretical equation that takes into account the effect of hydrogen permeation itself (10) with the actual hydrogen permeation flux. The study demonstrated an excellent concordance between the estimated hydrogen permeation flux and the actual flux regardless of inlet hydrogen mole fraction and species, thus elucidates the significant effect of hydrogen permeation itself on the decrease in hydrogen concentration at the membrane surface during concentration polarisation. Therefore, it is interesting to note that the severity of concentration polarisation is determined by feed flow rate and inlet hydrogen mole fraction, but not the different chemical characteristics and binary diffusivity of the mixtures (14).

Nakajima et al. reduced the boundary layer thickness to abate concentration polarisation by improving physical geometry of a reactor vessel containing a tubular palladium/silver membrane (15). In this case, the reduction of boundary layer thickness was performed by narrowing the path between the membrane surface and the inner surface of the vessel shell (15). For instance, by narrowing the path from 23.9 mm to 16.6 mm, the amount of produced hydrogen can be improved by around 25% even though there is only a 2% increment in methane conversion during hydrogen production from natural gas (for feed flow rate of 9 Nml min⁻¹ cm⁻²) (15). Such improvement is due to the reduction in distance between the inner surface of the vessel and the membrane surface, which has also been observed through the previous numerical simulation performed by Chen et al. (7).

As an alternative to the numerical simulation technique performed by Chen et al. (6, 16), the profiles of average axial concentration and concentration at the membrane surface can also be obtained by determining an analytical solution of the governed ordinary differential equation (ODE) (17). Further discussion on the analytical solution is presented in the following section.

In 2018, Kian et al. investigated the concentration polarisation phenomenon for the case of various ternary mixtures (18). They found the hydrogen permeation behaviour for ternary mixtures is similar to the case of binary mixtures, in which competitive adsorption becomes dominant when strong inhibitive carbon monoxide is used in the ternary hydrogen mixture (H₂:CO₂:CO) while concentration polarisation starts to play a significant role when methane is used instead of carbon monoxide. In the same year, Helmi et al. also proved the concentration polarisation phenomenon for the case of a fluidised membrane reactor (19). The good agreement between the developed model that considers concentration polarisation (so called ‘1D/k_d’) with experimental results elucidates that this phenomenon becomes more significant when lower inlet hydrogen mole fraction and higher inlet velocity are used (19). Based on the comprehensive review of the research scenarios in the previous paragraphs, it is evident that concentration polarisation is an undesirable phenomenon. This is because the ability of the membrane to permeate a very high amount of hydrogen cannot be fully utilised. However, it is unavoidable due to the advances in membrane technology that cause the fabrication of very thin membranes. Consequently, the external mass transfer becomes the permeation controlling step instead of diffusion in the metallic lattice. Although the development of compact devices with sufficiently high hydrogen recovery is possible, the concentration polarisation level becomes high due to the significant effect of permeation itself during the permeation.

In brief, several techniques have been introduced to address such disadvantages and effectively reduce the severity of concentration polarisation. For example, baffles may be implemented (7, 8) in the membrane reactor and the path between membrane surface and inner vessel wall may be narrowed (7, 15) to reduce the boundary layer at the membrane surface (15). The implementation of a spherical particles bed between tubular membranes in a membrane reactor has been confirmed to reduce the effect of concentration polarisation, due to the increase in the interparticle velocity and the increase in the Reynolds number between the particles and membrane surfaces (20). Helmi et al. have used fluidising particles inside a fluidised membrane reactor to significantly reduce the concentration polarisation effect due to better mixing of gases (21). The permeation system with microchannel configuration (22, 23) also has been suggested as a technique for concentration polarisation abatement due to the ability to decrease boundary layer thickness near to the membrane surface (22, 24). The combined usage of baffles and perforated pipe to reduce concentration polarisation effect has been demonstrated by Peters et al. (25) due to the creation of turbulence. Recently, an integrated compact system that consists of combustor, prereformer, reformer and hydrogen separator (palladium membrane) in a single module was developed by Wunsch et al. for small hydrogen demand applications (26). Since the height of the integrated system was relatively very small (12.4 mm) with eleven plates arranged in a stack, the concentration polarisation effect could be minimised, thus improving hydrogen yield as well as hydrogen productivity (26).

In 2008, Chen et al. introduced a constant concentration method with the aim to characterise the membrane by eliminating the effect of concentration polarisation (27). Based on this method, the tubular membrane reactor was filled with non-H₂ species first and followed by H₂ to obtain the desired hydrogen partial pressure without allowing any permeation. Once permeation began, the hydrogen that permeated out of the membrane was made up or replaced by the fresh hydrogen from the pressurised tank to maintain the hydrogen partial pressure at the retentate side. Therefore, the variation of hydrogen concentration at the membrane surface could be eliminated (27). However, this technique is not practical for industrial application since the flow stream at the upstream side of industrial membrane reactor is usually in plug flow, and not perfectly mixed flow, in which concentration polarisation usually could be triggered.

Extensive studies on concentration polarisation phenomena have been performed by Caravella et al. (28) and Catalano et al. (12). Caravella et al. (28) have created concentration polarisation maps which are very useful for hydrogen purification system design. Here, the maps are a two-dimensional graph of concentration polarisation coefficient versus hydrogen retentate molar fraction that was derived based on Equation (i):

(i)

where H₂flux (elementary steps) is the hydrogen permeation flux that is calculated by considering all the permeation elementary steps (external mass transfer, superficial adsorption, diffusion through the palladium-based bulk and superficial desorption) (29), in which in this case, involved complex procedures. Meanwhile, the concentration polarisation coefficient (CPC) is a concentration polarisation coefficient that becomes as an indicator for concentration polarisation level. The values of CPC were obtained by solving Equation (i) for various operating conditions and the values were plotted with respect to hydrogen retentate molar fraction. Here, when the value of CPC is 0, it means no concentration polarisation occurs while the CPC value of 1 indicates maximum level of concentration polarisation that has been defined as ‘total polarisation’ in their study. Π^Membrane is the membrane permeance that can be obtained from permeation test for pure hydrogen. Meanwhile, DF^Bulk is bulk driving force that can be obtained from the difference in square root of hydrogen partial pressure between the feed side and permeate side when the effect of concentration polarisation is negligible. For those who want to evaluate the concentration polarisation level by using these maps, only knowledge of the operating condition of the membrane reactor is required. The CPC can be determined manually from the maps. Finally, hydrogen permeation flux can be predicted by substitution of the CPC value into Equation (i) and followed by solving the equation. Despite the simplicity of this prediction method, the determined CPC actually does not account for the remaining length (and remaining area) of the tubular membrane where no permeation occurs anymore due to very fast decay of driving force at the region around the inlet. This situation occurs when concentration polarisation becomes significant (16), thus hydrogen concentration is overestimated when the aforementioned CPC value is used. This weakness was then solved through the introduction of the powerful parameter so called effective average CPC (EAC) (30). The determination of EAC is stated as follows, Equation (ii):

(ii)

where L, z and CPC are the membrane length, membrane axial abscissa and concentration polarisation coefficient, respectively. Here, the local value of CPC for each position on the membrane in z-direction is determined analogously as was introduced previously (28). Meanwhile, the hydrogen concentration profile and the respective profile of hydrogen permeation flux can be obtained by simultaneously solving the external mass transfer and hydrogen permeation equations (30). Here, the calculation of mass transfer coefficient is as reported by Caravella et al. (29). Finally, Equation (ii) can be solved to obtain the value of EAC. Based on the previous individual elaboration on the prediction techniques using CPC (28) and EAC (30), it can be said that the use of EAC is more desirable, since it has the ability to represent the real behaviour of a hydrogen permeation device. It is interesting to note that once the EAC maps have been prepared, similar to the previous technique of using the CPC maps (28), the hydrogen permeation flux can be estimated by simply substituting the value of the EAC obtained from the maps into Equation (ii).

Similarly, Catalano et al. (12) concluded the existence of non-negligible resistance to hydrogen transport in the gaseous phase itself, in addition to resistance caused by the membrane. For the case of a hydrogen mixture, the authors demonstrated a significant deviation from Sieverts’ Equation (Equation (iii)) when the hydrogen partial pressure of the bulk gas is substituted into the equation. To compensate for this situation, semi-empirical equations were developed for a tubular type membrane (membrane thickness of 2.5 μm) as follows (12), Equations (iv) and (v):

(iii)

(iv)

(v)

where N_H2,int is the hydrogen flux crossing the membrane interface and is defined as the hydrogen flux within the gas-metal interface, k_G is the mass transport coefficient, p_ret is the pressure at the retentate side, p_H2,int is the hydrogen partial pressure at gas-metal interface, p_H2,ret is the hydrogen partial pressure at retentate side, p_H2,per is the hydrogen partial pressure at permeate side and is the hydrogen permeance obtained from pure hydrogen experiment. It is interesting to note that once the value of k_G is obtained by solving Equations (iv) and (v) simultaneously and by using experimental data of N_H2,int, the same value of k_G can then be used to estimate hydrogen permeation flux for the cases with different hydrogen partial pressure difference.

As a continuity of the previous study (28), Caravella et al. (31) considered simultaneously both concentration polarisation and inhibition by carbon monoxide species in their model, by introducing the permeation reduction coefficient. Similar to the prediction technique introduced previously (28), the permeation reduction coefficients were plotted for different operating conditions, or so called ‘permeation reduction maps’. The simple relation for the permeation reduction coefficient as shown by Equation (vi) was derived from definitions of CPC and inhibitive coefficient (IC), that were obtained from previous studies by Caravella et al. (28) and Barbieri et al. (4), respectively through complex calculation steps, Equation (vi):

(vi)

where PRC is the permeation reduction coefficient, CPC is the concentration polarisation coefficient and IC is the inhibition coefficient. To predict the hydrogen permeation flux for certain operating conditions, the value of PRC is determined manually from the ‘permeation reduction maps’ and then the value is substituted into Equation (vii) as follows:

(vii)

where J_H2 is the hydrogen permeation flux, PRC is the permeation reduction coefficient, Π^Sieverts is the permeance which is similar to the hydrogen permeance coefficient, obtained from pure-hydrogen test. Meanwhile, is the bulk driving force for hydrogen permeation, that is bulk difference in square root of hydrogen partial pressures between the retentate and permeate side.

As one of the solutions for the difficulty in obtaining a general relation that consists of several interdependent parameters as has been mentioned by Morgues et al. (32), Faizal et al. (10, 14, 33) have introduced a theoretical approach for hydrogen permeation through a flat sheet palladium based membrane after observing a significant deviation between the actual permeation flux and the estimated flux by Sieverts’ equation (Equation (iii)) when inlet hydrogen concentration was used (33). As asserted by previous researchers on the significant effect of permeation flux during concentration polarisation phenomena (6, 12, 16), the term of hydrogen partial pressure at membrane surface of upstream side in the Sieverts’ equation (Equation (iii)) has been modified to consider the effect of hydrogen permeation flux during permeation. The modification of Equation (iii) leads to the formation of Equation (viii) as follows (14):

(viii)

where f_p is the estimated hydrogen permeation mole flux, q is the hydrogen permeance coefficient, d is the membrane thickness, f_H2,in is the mole flux of hydrogen from inlet (feed flow rate of hydrogen divided by effective membrane surface area), f_in is the mole flux of the mixture from inlet (feed flow rate of mixture divided by effective membrane surface area), P_in is the total pressure at inlet (total upstream pressure) and p_H2,2 is the hydrogen partial pressure at membrane surface of the downstream side. In order to predict f_p, the values of operating parameters are substituted into Equation (viii) and followed by solving the equation for f_p. Surprisingly, the modified equation as shown by Equation (viii) can estimate accurately hydrogen permeation flux for any noninhibitive binary hydrogen mixture with different chemical characteristics and binary diffusivity, along with any hydrogen concentration and with any mole flux of mixture from the inlet. For instance, when the mole flux of mixture from the inlet is increased, the effect of concentration polarisation is weakened. Therefore, the estimated flux obtained from Equation (viii) approaches the flux obtained from Equation (iii) due to the weaker effect of f_p. The introduced method is supposed to be very useful for reactors with similar type of membrane used (34–36).

To prevent membrane damage due to mechanical stress, the palladium/silver tubular membrane was created in the form of a ‘finger-like’ configuration, thus allowing the free elongation and contraction of the membrane (2). For this kind of configuration, the way to predict hydrogen permeation flux is supposed to be similar to that for the tubular type membrane with common configurations (Figures 2(a) and 2(b) in Part I (37)) because the hydrogen mixture similarly flows horizontally along the membrane length for both cases. In the research performed by Miguel et al. (2), a model to simultaneously consider both concentration polarisation and inhibition by carbon monoxide or carbon dioxide has been introduced based on the logarithm-mean driving force (for considering concentration polarisation effect) and correction factor due to inhibitive effect (4). Compared to the approach by Barbieri et al. (4), the model introduced by Miguel et al. could provide more accurate results for the simultaneous occurrence of both phenomena. This is because the previous approach by Barbieri et al. (4) only considers the feed hydrogen partial pressure for prediction. However, the model introduced by Miguel et al. is semi-empirical and therefore, experimental data is necessary in order to determine certain parameters in the correction factor. The combination of correction factor and rearranged Sieverts’ equation forms the rearranged Sieverts’-Langmuir equation as shown below (2), Equation (ix):

(ix)

where the term is the correction factor due to adsorption of inhibitive species on the membrane surface and the term is the rearranged Sieverts’ equation.

Here, is the hydrogen permeation flux, α is the Sieverts’-Langmuir reduction factor, K_i is the Langmuir’s adsorption constant for species (CO or CO₂), is the average partial pressure of species (CO or CO₂) between the feed and retentate sides, L_H2 is the hydrogen permeance or hydrogen permeation coefficient, δ is the membrane thickness and ΔP_ln is the logarithm mean driving force that is determined based on theory of heat exchanger for parallel flow case (2, 5). It is important to note that the values of α and K_i for carbon monoxide and carbon dioxide are dependent on operating temperature, thus these values need to be fitted with experimental data first before using Equation (ix) to estimate hydrogen permeation flux.

A prediction method for hydrogen permeation capacity (length of membrane for hydrogen permeation) has been introduced by Xie et al. (38) through computer programming. The investigation was performed analytically for seven important scenarios with a different flow pattern on both sides (upstream and downstream side). In this case, the concept is similar to that performed by Faizal et al. (10, 14, 33), in which the effect of hydrogen permeation rate is taken into account when determining hydrogen partial pressure at the membrane surface of the upstream side, as shown by the following Equation (x) (example for the scenario with plug flow at upstream side, and no sweep gas):

(x)

where P_H (x) is the hydrogen partial pressure of an infinitesimal permeation capacity (infinitesimal membrane length for permeation), dx in the high pressure (upstream) side, M₁ is the feed flow rate of hydrogen at upstream side, M₂ is the feed flow rate of nonpermeable gas at upstream side, P₁ is the total pressure of upstream side and M_x is the hydrogen permeation rate through a membrane for length from 0 to x. Then, the local hydrogen permeation rate through the infinitesimal permeation capacity dx can be predicted by substituting Equation (x) into the Sieverts’ equation as follows, Equation (xi):

(xi)

where C is the constant for membrane and P₂ is the total pressure of downstream side. In their study, Equation (xi) is rearranged and then followed by integration of x with respect to M_x in order to predict the permeation capacity (membrane length for separation) x as shown by Equation (xii):

(xii)

Differently to other techniques, the technique introduced by Xie et al. (38) is used to estimate the value of x instead of M_x.

Recently, algebraic functions that can be used to predict the profiles of hydrogen permeation flux under the influence of the concentration polarisation phenomenon have been introduced (17). Concentration polarisation is accounted through the use of an effectiveness factor which was derived in the previous study (39). The effectiveness factor is the ratio of actual permeation flux over the calculated flux based on the average concentration, and it is a function of separation parameter that represents the ratio of diffusive to permeation flux. The effectiveness factor and separation parameter that obeys Sieverts’ Law can be expressed as Equation (xiii) and Equation (xiv), respectively (2):

(xiii)

(xiv)

where (Equation (xv)):

(xv)

Here, η is the effectiveness factor, is the hydrogen partial pressure at membrane surface of retentate side,  is the hydrogen partial pressure at membrane surface of permeate side, is the average hydrogen partial pressure at retentate side, Γ is the separation parameter, D is the diffusion coefficient of hydrogen in the gaseous phase, Sh is the Sherwood number, d is the characteristic length, p^ret is the pressure at retentate side, K_H2 is the hydrogen permeability and c_tot is the total molar density. To predict the profile of hydrogen permeation flux, the algebraic functions presented by Nekhamkina et al. (17) need to be solved.

The background of the scenarios related to palladium based membranes has been elaborated. It was found that concentration polarisation becomes unavoidable in parallel with advances in membrane technology. The scenario of parametric studies on the concentration polarisation phenomenon specifically for palladium based membranes was reviewed comprehensively. Based on the present review, it is evident that an increase in total upstream pressure, membrane temperature and permeance promotes concentration polarisation. The same trend is also achieved when the feed flow rate, inlet hydrogen concentration, total downstream pressure and membrane thickness are reduced. When the ratio of hydrogen permeation rate to hydrogen feed rate at the inlet becomes sufficiently high, the effect of hydrogen permeation flux on the hydrogen concentration decrease at the membrane surface becomes significant, thus concentration polarisation becomes strong. Therefore, it can be said that an increase in hydrogen recovery percentage leads to a stronger tendency for concentration polarisation to occur, and as a consequence, larger deviation from the hydrogen permeation flux estimated by Sieverts’ equation could be observed. For both tubular type and flat sheet type membranes, when concentration polarisation is triggered, the hydrogen concentration decreases in the horizontal direction (from the leading edge to the tailing edge) and radial direction, respectively. Furthermore, the existence of inhibitive species such as carbon monoxide in the hydrogen mixture somehow causes the membrane performance in terms of hydrogen permeation flux to become worse due to the simultaneous occurrence of concentration polarisation and inhibition by carbon monoxide. Meanwhile, the concentration polarisation level does not depend on the number of noninhibitive species, chemical characteristics of noninhibitive species or binary diffusivity in the hydrogen mixture (binary or ternary mixture).

Several techniques have been identified to effectively abate concentration polarisation such as coupling of upstream flow with sweep flow in countercurrent mode, installation of baffles in the appropriate number, size and position, and by narrowing the space for upstream flow, that is reduction of the distance between the shell and membrane. Basically, the aforementioned techniques were implemented to increase the hydrogen concentration at the membrane surface of upstream side, thus concentration polarisation could be reduced.

Finally, several estimation methods for hydrogen permeation flux have been reported for different membrane configurations. Several methods are empirical, in which experimental data is necessary to obtain certain coefficients while some of the methods can be used by just substituting the operating parameters into the introduced equation.

For future work, it is suggested that the methods to estimate hydrogen permeation flux for the application of steam reforming should be intensively developed for various operating conditions. In this case, the detailed chemical kinetics must be considered to obtain the accurate mixture composition near the membrane surface. The competitive adsorption by excessive steam and excessive vaporised alcohols (methanol for instance) in addition to carbon monoxide during the occurrence of steam reforming reaction should also be taken into account in the future development of estimation methods.

The production of nitric acid, a key industrial process, requires the synthesis of nitric oxide as a precursor to the desired end product. Nitric acid, with a current annual production of 65.9 million tonnes, is primarily used in the production of ammonium nitrate for fertiliser applications, with the remainder (around 20%) used in industrial applications including the production of explosives (1).

Johnson Matthey has manufactured pgm ammonia oxidation catalysts for over 100 years, selling the first platinum gauze pack to the UK Munitions Invention department in October 1916. This pack contained woven wires manufactured from pure platinum and required 1 kg of platinum to manufacture 1 tonne of nitric acid (2). In the century that followed, a number of technological advances resulted in lighter and more efficient catalyst packs and reduced the installed metal content per tonne of nitric acid produced from kilograms to milligrams.

The first advancement in technology was the substitution of pure platinum with a 10% rhodium 90% platinum alloy, first patented by DuPont, USA, in 1929 (3). Rhodium increased the mechanical strength of the alloy and reduced in situ metal loss, allowing the catalyst to be operated at higher temperatures and benefit from the associated increase in selectivity to nitric oxide (4).

Despite the reduction in metal loss from the optimisation in alloy, the metal lost from the catalyst in situ remained high and formed a significant part of the plant operating costs. To reduce the cost of the metal loss, gauzes comprising woven palladium-alloy wires were developed in 1968, using a gold-palladium alloy, with up to 20% gold (5). This catchment pack sacrificially collects platinum whilst losing palladium, reducing the cost of the total metal loss. Further research to reduce the cost of the catchment system resulted in the development of nickel-palladium and tungsten-palladium alloys (6, 7) with research showing that a low concentration of the base metal was required for high collection efficiency (8). As a result, alloys used in catchment systems today typically contain 95% palladium.

The 1990s saw a significant change in the design and manufacture of pgm catalysts, with Johnson Matthey introducing new technology where knitted gauzes superseded woven. As well as introducing more flexibility to catalyst design, the new technology was observed to have a positive impact on the catalyst selectivity to nitric oxide (6). This technology is now the standard for ammonia oxidation catalysts and is offered by all major catalyst suppliers.

Gauze manufacturers make use of different technologies, i.e. warp and weft knitting, to manufacture the catalyst gauzes. Recent advances driven by knitting include the manufacture of very high-density gauzes to concentrate metal towards the top of the pack. There are different methods to achieve this: Johnson Matthey has achieved this by minimising the open area of a weft-knitted gauze, using high density structures that increase the metal content per gauze by over 90% compared to the basic knit structure. Other manufacturers such as Umicore, Belgium, limited to greater open areas, have developed three-dimensional knit structures to achieve a high-density gauze (9).

The mid-2000s saw a fundamental change in ammonia oxidation catalyst design, with ternary rhodium-platinum-palladium alloys, containing high concentrations of palladium, replacing traditional rhodium-platinum alloys in the lower portion of the pack. Johnson Matthey developed ECO-CAT^TM gauzes using these principles, and this new technology saw a reduction in installed weight required to produce a set amount of nitric acid. The length of time the plant could run with a catalyst pack (known industrially as the ‘campaign’) was increased. Lower density palladium-based alloys in the lower layers collect volatilised platinum from the top layers, with the captured platinum being used later in the campaign as the reaction progresses further into the pack. Variations on this technology are now the standard offering by most catalyst suppliers for medium-pressure nitric acid plants.

Historically, catalyst packs were designed using rules of thumb and best estimates. An increased understanding of ammonia oxidation and primary metal losses within pgm catalysts has resulted in the creation of bespoke design tools, which allow Johnson Matthey to optimise the metal content and distribution required within a catalyst pack. Using these tools, catalyst designs can be tailored to plant operating conditions and KPIs. A tailored design can result in higher selectivity to nitric oxide, along with increased campaign lengths and reduced metal content in catalyst packs.

Improvements have also been observed in nitric acid plant designs over the last 100 years. The first industrial nitric acid plants operated at atmospheric pressure and were restricted to low production rates. In the late 1920s, nitric acid plants were constructed that could operate under pressure, resulting in more compact plants that could achieve higher rates (10).

Today, new nitric acid plants are typically dual-pressure plants or ultra-high mono-pressure plants (10). Dual-pressure plants utilise a low operating pressure for the ammonia oxidation reactor to ensure high selectivity to nitric oxide (typically 3–6 barg) and increase the pressure in the downstream absorption column to improve the column efficiency. Ultra-high mono-pressure plants operate at a single pressure, typically 10–13 barg, benefiting from lower capital costs but achieving lower plant efficiencies.

2.1 Reaction Fundamentals

Ammonia reacts with oxygen to produce nitric oxide, nitrogen and nitrous oxide. The selectivity of ammonia to each reactant is dependent on process conditions and catalyst specification. Nitrogen is the favoured product at lower temperatures, whilst nitric oxide formation is favoured at higher temperatures and at lower pressures. At low temperatures, NO is formed but is bound strongly to the platinum surface. This allows the NO to react with ammonia to form N₂ which will then desorb. At high temperatures, the NO formed will desorb from the catalyst surface before this secondary reaction occurs, increasing the catalyst selectivity to NO (11).

In industrial production, the reaction is mass transfer limited due to the relatively slow diffusion of reactants to the wire surface compared to the almost instantaneous surface kinetics (12). In industrial operation, the catalyst operates between 850–930°C and plant operating pressures range from atmospheric pressure to 13 barg. Typical selectivity to nitric oxide ranges from 90–97% (Equations (i)–(iii)):

(i)

(ii)

(iii)

In addition to these primary reactions, ammonia can react with nitric oxide to produce nitrogen and nitrous oxide, which reduces the efficiency of the plant (Equations (iv) and (v)):

(iv)

(v)

As a result, an effective catalyst will be designed to convert 100% of ammonia in the top few layers of the catalyst pack at the beginning of a campaign; although the reaction will progress further through the pack as platinum is lost and the top layers of the catalyst lose activity.

The surface chemistry and metallurgy of the catalyst used impacts the selectivity achieved. Several precious metals have been shown to catalyse the ammonia oxidation reaction, including platinum, rhodium, palladium (13), silver and iridium (14). Platinum has the greatest selectivity to nitric oxide, making it the most suitable catalyst for nitric acid plants. Rhodium-platinum alloys, containing 3–10% rhodium, are the most common alloys offered in ammonia oxidation catalysts today, although often in conjunction with a rhodium-platinum-palladium alloy.

While rhodium-containing alloys are primarily used to increase the alloy strength during manufacture and operation (15), rhodium-platinum alloys have been observed to have a higher selectivity to nitric oxide than pure platinum (4), and palladium-containing alloys have a reduced selectivity to nitrous oxide at all temperatures as demonstrated through ab initio modelling of the selectivity of ammonia oxidation to NO, NO₂, N₂ and N₂O on pure platinum and palladium-platinum alloys (Figure 1). The benefit of lower selectivity to N₂O is partially offset by an increase in selectivity to N₂: palladium-based alloys can help reduce N₂O emissions but at the cost of reduced selectivity to NO (16). The benefit of using palladium-based alloys has been demonstrated within the nitric acid industry, with catalyst packs using ECO-CAT^TM technology or other catalyst manufacturers ternary alloy technologies, shown to reduce N₂O emissions by up to 30% compared to standard rhodium-platinum catalyst packs (17).

Fig. 1

In-house modelling by Johnson Matthey using ab initio kinetics and density functional theory (DFT) demonstrates the significant reduction in selectivity to N₂O for: (a) a pure platinum alloy; (b) a palladium-doped platinum alloy across a range of temperatures. An increase in selectivity to N₂ at higher temperatures is also demonstrated (16)

2.2 In situ Restructuring and Metal Loss

2.3 Industrial Operation

The reaction fundamentals, restructuring and loss of metal and constraints in industrial operation must all be considered when designing a catalyst pack for ammonia oxidation in a nitric acid plant. The variables available to a catalyst designer and manufacturer include optimising the metal content and metal placement within the pack and selecting the most appropriate alloys and optimising their placement in the pack.

3.1 Metal Content

3.2 Alloy Selection

3.3 Catchment System

Optimisation of metal content is applicable to all categories of nitric acid plants and Johnson Matthey continually reviews existing catalyst designs to ensure they are optimal for plant operation and prevailing pgm market conditions. In recent years, significant steps have been taken to optimise metal content for ultra-high-pressure plants, with design options tailored to the key drivers of each plant, for example maximising daily production rates or maximising plant efficiency from a set level of ammonia feed.

For a significant proportion of nitric acid producers operating ultra-high-pressure plants, the key driver is to produce the maximum amount of nitric acid from a gauze pack. In these cases, the introduction of palladium alloys is not recommended due to the increased pressure drop. The catalyst pack is instead optimised by increasing the platinum weight at the top of the pack by using larger wire diameters in the top layers. To maintain a constant pack weight, the total number of layers within the pack is reduced. As a result, ammonia oxidation is completed higher in the pack and the N₂ and N₂O-producing side reactions are reduced. In addition to selectivity benefits, increasing the wire diameter in the top layers provides sufficient metal for further cauliflower formation in the event of mechanical metal loss following a plant trip.

For other ultra-high-pressure producers, typically operating in locations with higher raw material costs, the key driver is to maximise the efficiency of the gauze pack over a set campaign length. As a slight increase in pressure drop is tolerable for these plants, pack optimisation includes the use of ECO-CAT^TM design principles. Initial design optimisation work for a plant in this category combined the use of high-density knit structures in the top portion of the pack with high palladium alloys in the lower portion of the pack. The changes resulted in improving the conversion efficiency towards the end of the campaign, and as a result increasing the average campaign efficiency. Further design changes, including the reduction of palladium content whilst increasing the concentration of palladium higher in the pack, resulted in a reduction in N₂O emissions from the gauze system.

The catalysis of ammonia oxidation using pgm gauzes has been used industrially for over 100 years. Step changes in performance have been achieved throughout the decades, reducing the required pgm metal content required in a catalyst pack and introducing a range of alloys and knitted structures from which to construct a catalyst pack. Research into ammonia oxidation has continued to support the development of catalysts with higher selectivity to NO and reduced N₂O emissions, giving rise to the current binary and ternary alloys used in catalysts today.

The underlying reaction kinetics and metal movement provide a basis for designing the optimal ammonia oxidation catalyst and have been used within Johnson Matthey to develop plant category specific design rules. Achieving a high selectivity to nitric oxide can be achieved by promoting near 100% ammonia conversion in the top layers, and the use of ternary alloys in the lower catalyst layers promotes recycling of platinum to maintain a high efficiency towards the end of the campaign. A wide range of knit structures, wire diameters and alloys are available to tailor the catalyst to the plant operating conditions and to distribute the metal within the pack in a way that promotes the highest selectivity to NO throughout the campaign. The concentration of rhodium and palladium in the top catalyst layers can be optimised to address plant-specific issues, including the formation of rhodium oxide and helping the catalyst to light-off at a lower temperature.

Other factors influence the optimal design, with the acceptable pressure drop over the catalyst pack being key. Despite having significant benefits, including reduced N₂O emissions, ECO-CAT^TM and catchment technologies are not always suitable for ultra-high-pressure plants due to the increased pressure drop.

Optimisation of metal content within the catalyst pack is now the expected standard for all medium-pressure plants, with the development of high-density knit structures and ECO-CAT^TM technology resulting in improvements to conversion efficiency, reduction in installed metal content and the extension of campaign lengths. In recent years, further progress has been made in optimising catalyst packs for ultra-high-pressure nitric acid plants operating at >10 bar and with nitrogen loading values in excess of 60 te_N m⁻² day⁻¹, designing the pack to minimise unwanted side reactions. While the high intrinsic metal loss continues to constrain the maximum campaign length for these plants, the optimisation of pgm content within the pack has resulted in an improvement in plant efficiency and, where targeted, a reduction in N₂O emissions.

Despite being a mature and conservative industry, the operational targets for nitric acid plants continue to increase to support the global demand for fertiliser. Many plants will seek to operate in excess of the nameplate capacity and carry out de-bottlenecking studies to support this. Changing operational targets should always be communicated to the catalyst supplier, as a change in plant loading, temperature or pressure is likely to require an updated catalyst design. The dynamic pgm market also influences the optimal pack design; catchment packs see a reduction in profitability when the price of palladium exceeds that of platinum. Additionally, at the time of writing rhodium was seeing extremely high prices which form a significant portion of the pack cost, leading to increasing interest in low-rhodium alloys.

The optimisation of a catalyst pack is never a complete process, and continual monitoring of catalyst performance and an open dialogue between plant operator and catalyst supplier are critical in creating and maintaining an optimal pack design.

ECO-CAT^TM is a trademark of Johnson Matthey PLC.

The thermodynamic properties of osmium were reviewed by the author in 1995 (1) with a further review in 2005 (2) to estimate a most likely value for the melting point at 3400 ± 50 K to replace the poor quality experimental values which were being quoted in the literature. More recently Burakovsky et al. (3) have estimated a value of 3370 ± 75 K in good agreement with the above selected value. In the 1995 review the enthalpy of fusion was unknown but was estimated from a relationship between the entropy of fusion and the melting point which showed a high degree of correlation for the platinum group metals (pgms). However the derived entropy of fusion value for osmium was based on values for the other pgms available at that time but since then the values for both palladium and platinum have been revised so that the entropy of fusion value for osmium would also be revised leading to a new estimate of 68.0 ± 1.7 kJ mol⁻¹ for the enthalpy of fusion. This would then require the thermodynamic properties of the liquid phase to also be updated. A comment is included on an independent much lower estimate of the enthalpy of fusion. Wherever possible measurements have been corrected to the International Temperature Scale (ITS-90) and to the currently accepted atomic weight of 190.23 ± 0.03 (4).

Selected values in the normal and superconducting states are based on the specific heat measurements of Okaz and Keesom (0.18 K to 4.2 K) (5) including a superconducting transition temperature of 0.638 ± 0.002 K, an electronic specific heat coefficient (γ) of 2.050 ± 0.003 mJ mol⁻¹ K⁻² and a limiting Debye temperature (Θ_D) of 467 ± 6 K. Specific heat values up to 5 K in both the normal and superconducting states are given in Table I.

Above 4 K selected specific heat values are initially based on the measurements by Naumov et al. (6 K to 316 K) (6). However above 280 K these measurements show an abrupt increase of 0.5 J mol⁻¹ K⁻¹ and a further abrupt increase of 0.3 J mol⁻¹ K⁻¹ above 300 K. Naumov et al. attempted to accommodate these values but the selected specific heat curve showed an unnatural sharp change in slope above 270 K. Therefore the selected values of Naumov et al. above 250 K were rejected and instead specific heat values to 298.15 K were obtained by joining smoothly with the high temperature enthalpy measurements of Ramanauskas et al. (7). In the original review of the low temperature data only the specific heat values were given consisting above 50 K of 10 K intervals to 100 K and then 20 K intervals above this temperature as well as the value at 298.15 K. This minimalist approach is now considered to be unsatisfactory and therefore comprehensive low temperature thermodynamic data are now given at 5 K intervals from 5 K to 50 K and at 10 K intervals above this temperature up to 290 K and then the value at 298.15 K as given in Table II.

In the high temperature region, after correction for temperature scale and atomic weight, the enthalpy measurements of Ramanauskas et al. (1155 K to 2961 K) (7) were fitted to the following equation with an overall accuracy of ± 200 J mol⁻¹ (0.4%) (Equation (i)):

(i)

This equation was used to represent selected enthalpy values from 298.15 K to 3400 K. Equivalent specific heat and entropy equations corresponding to the above equation are given in Table III, the free energy equations in Table IV, transitions values associated with the free energy functions in Table V and derived thermodynamic values in Table VI. The actual equation given by Ramanauskas et al. to represent the enthalpy measurements over the experimental temperature range agrees with Equation (i) to within 0.2%.

The only other enthalpy measurements were obtained by Jaeger and Rosenbohm (693 K to 1877 K) (8) and compared to the selected values vary from 1.6% low at 693 K to an estimated 1.2% low at 1600 K to 1.4% low at 1877 K.

Selected values of the enthalpies and entropies of fusion of the Groups 8 to 10 elements with a close-packed structure are given in Table VII. Only the enthalpy of fusion of osmium is unknown. References (10–14) represent the latest reviews on the thermodynamic properties of the pgms by the present author. From an evaluation of the entropies of fusion of the elements, Chekhovskoi and Kats (15) proposed that the entropy of fusion (ΔSº_M) and the melting point (T_M) could be related by the equation ΔSº_M = A T_M + B. In the previous review (1) different values were proposed for the entropies of fusion of palladium (8.80 J mol⁻¹ K⁻¹) and platinum (10.45 J mol⁻¹ K⁻¹) leading to an estimate of the entropy of fusion for osmium of 20.6 J mol⁻¹ K⁻¹. With the revised values it is clear that, although of the right order, the entropy of fusion of nickel is discrepant and has therefore been disregarded. The other six values were fitted to the equation with A = 6.6954 × 10⁻³ and B = −2.7630 and a standard deviation of the fit of ± 0.193 J mol⁻¹ K⁻¹. However in order that the derived entropy of fusion of osmium has a similar accuracy to those of the input values then the accuracy is expanded to a 95% confidence level leading to an entropy of fusion of 20.0014 ± 0.387 J mol⁻¹ K⁻¹ and based on a melting point 3400 ± 50 K to an enthalpy of fusion of 68,005 ± 1653 J mol⁻¹. Based on neighbouring elements then a liquid specific heat of 50 J mol⁻¹ K⁻¹ was proposed in the original paper (1) and therefore the enthalpy of liquid osmium can now be expressed as Equation (ii):

(ii)

Equivalent specific heat and entropy equations corresponding to the above equation are given in Table III, the free energy equation in Table IV and derived thermodynamic values in Table VI. It should now be possible to accurately determine the melting point and enthalpy of fusion of osmium since the metal is available in high purity in a coherent form whilst the enthalpies of fusion of other high melting point elements such as rhenium (3458 K) and tungsten (3687 K) have been successfully determined.

Based on a standard state pressure of 1 bar the thermodynamic properties of the monatomic gas were calculated from the 295 energy levels listed by Van Kleef and Klinkenberg (16) and Gluck et al. (17) using the method outlined by Kolsky et al. (18) together with the 2018 Fundamental Constants (19). Derived thermodynamic values are given in Table VIII.

No temperature scales were given with the measurements of the vapour pressures by Panish and Reif (20) and Carrera et al. (21). Normally the experimental temperature values would therefore be accepted but in the case of such values above 2000 K the difference from the current scale, ITS‐90, becomes significant. Since the measurements were carried out in 1962 and 1964 then they would ultimately be associated with the International Practical Temperature Scale (IPTS-1948) and were therefore corrected to the ITS-90 scale on this basis. Derived enthalpies of sublimation are given in Table IX. The selected enthalpy of sublimation of 788 ± 4 kJ mol⁻¹ is basically an unweighted average but slightly biased towards the measurements of Carrera et al. (21).

The vapour pressure equations are given in Table X. For the solid the evaluation was for free energy functions for the solid and the gas at 50 K intervals from 1700 K to 3400 K and for the liquid at 50 K intervals from 3400 K to 5600 K and were fitted to Equation (iii):

(iii)

A review of the vapour pressure data is given in Table XI.

Based on various assumptions Fokin et al. (22) proposed that the enthalpy of fusion for osmium was only in the range 30 kJ mol⁻¹ to 40 kJ mol⁻¹ or half of the above derived value. One of the main arguments was that by using the Chekhovskoi-Kats equation the entropy of fusion for rhenium was estimated to be 20.0 J mol⁻¹ K⁻¹ whereas the actual value is only 9.85 J mol⁻¹ K⁻¹ (23) and therefore if the estimate for rhenium was so completely wrong then it would also be possible that the estimate for the neighbouring element osmium at 19.0 J mol⁻¹ K⁻¹ could also be wrong. However, Fokin et al. completely misunderstood how the estimated values were arrived at. It was initially assumed that Group 7 rhenium would behave like Groups 8 to 10 (the pgms) whereas all that the experimental value proved was that Group 7 elements behaved completely independently of Groups 8 to 10 and therefore showed the same deviations as other transition metal groups. For example, the entropies of fusion of Group 5 elements vanadium, niobium and tantalum at 10.46 J mol⁻¹ K⁻¹, 11.13 J mol⁻¹ K⁻¹ and 10.25 J mol⁻¹ K⁻¹ (24) showed no trend with temperature whilst the entropies of fusion of the Group 6 elements chromium, molybdenum and tungsten at 13.89 J mol⁻¹ K⁻¹, 13.53 J mol⁻¹ K⁻¹ and 13.66 J mol⁻¹ K⁻¹ (24) were virtually identical. Therefore it would not be surprising if Group 7 elements would also behave completely independently. In fact for the transition metals only the Groups 8 to 10 elements showed a high degree of correlation with the Chekhovskoi-Kats equation. However in order to prove their point that osmium does behave differently to the other pgms, Fokin et al. used the equation: σ_M = Z ΔH_M ρ_SM d where σ_M is the surface tension at the melting point, ΔH_M is the enthalpy of fusion, ρ_SM is the density of the solid at the melting point and d is the interatomic distance. This equation was applied to a number of elements but there is virtually no correlation for the values of Z with values varying between 1.2 to 3.3. For osmium Fokin et al. selected an arbitrary rounded value of Z = 2 for osmium and values of surface tension and liquid density determined by Paradis et al. (25) to arrive at an enthalpy of fusion of only 32 kJ mol⁻¹ which is considerably less than the value of 39.0 ± 1.4 kJ mol⁻¹ (9) selected for the analogue element ruthenium whereas for the other pgms the enthalpy of fusion is always greater for the heavier analogue. This much lower value for the enthalpy of fusion would suggest that the thermal properties of osmium should then be distinct from those of the other pgms but this is not the case. For example, the specific heat values of ruthenium (10) and osmium at reduced temperature (T/T_M) as indicated in Figure 1 are very similar and show virtually the same behaviour suggesting that they are genuine analogues of each other whilst the extrapolated melting point of osmium obtained by applying the same incremental difference as between iridium and platinum agrees closely with the selected value and again suggesting a common Groups 8 to 10 behaviour.

Fig. 1

Further, the chemical properties of ruthenium and osmium are virtually identical forming the same type of compounds with similar properties. These are examples where osmium behaves exactly like the other pgms and on these grounds it is suggested that the very low value for the enthalpy of fusion as suggested by Fokin et al. is inconsistent with this behaviour and that osmium would obey the same periodic trend as suggested by the other pgms and that its entropy of fusion can be determined by the Chekhovskoi-Kats equation. This would suggest anomalies in the input values selected by Fokin et al., especially in the selection of Z = 2 for osmium since the value for the analogue ruthenium is only 1.5 whilst the value for the neighbouring element iridium is only 1.2 where the selection of such values would lead to higher enthalpies of fusion for osmium. It is suggested that in view of the lack of any real correlation for Z that the value for osmium may well be independent and could even be 1.0 leading to an enthalpy of fusion similar to that obtained from the Chekhovskoi-Kats equation. Therefore until the actual enthalpy of fusion of osmium is determined it is assumed that it behaves as a normal Groups 8 to 10 element.

Estimated entropy and enthalpy values of fusion of osmium have been revised leading to corrections of the thermodynamic properties of the liquid phase and therefore to the vapour pressure curve above the melting point. The revisions are based on the assumption that osmium behaves as a normal Group 8 to 10 element and contradicts recent suggestions that its behaviour could be abnormal.