*Article* **Performance Study on Methanol Steam Reforming Rib Micro-Reactor with Waste Heat Recovery**

## **Guoqiang Wang 1,2, Feng Wang 1,2,\* and Bohong Chen <sup>2</sup>**


Received: 11 March 2020; Accepted: 26 March 2020; Published: 27 March 2020

**Abstract:** Automobile exhaust heat recovery is considered to be an effective means to enhance fuel utilization. The catalytic production of hydrogen by methanol steam reforming is an attractive option for onboard mobile applications, due to its many advantages. However, the reformers of conventional packed bed type suffer from axial temperature gradients and cold spots resulting from severe limitations of mass and heat transfer. These disadvantages limit reformers to a low efficiency of catalyst utilization. A novel rib microreactor was designed for the hydrogen production from methanol steam reforming heated by automobile exhaust, and the effect of inlet exhaust and methanol steam on reactor performance was numerically analyzed in detail, with computational fluid dynamics. The results showed that the best operating parameters were the counter flow, water-to-alcohol (W/A) of 1.3, exhaust inlet velocity of 1.1 m/s, and exhaust inlet temperature of 773 K, when the inlet velocity and inlet temperature of the reactant were 0.1 m/s and 493 K, respectively. At this condition, a methanol conversion of 99.4% and thermal efficiency of 28% were achieved, together with a hydrogen content of 69.6%.

**Keywords:** methanol steam reforming; hydrogen production; exhaust waste heat; rib microreactor

## **1. Introduction**

Motor vehicles are increasing dramatically with the rapid economic development [1,2]. However, the power used by the internal combustion engine for power output generally accounts for only 30%–45% (diesel) or 20%–30% (gasoline) of the total fuel combustion heat. A car effectively uses only a small part of the fuel's chemical energy, and most is lost through the engine's cooling water and high-temperature exhaust heat [3,4]. Therefore, the exhaust heat recovery, which is very important to improve the fuel efficiency, attracts more and more attention [5,6]. Pashchenko [7] studied thermochemical recovery of heat contained in flue gases with steam methane reforming. It was found that the enthalpy increased with increasing mole fraction of combustion products in the reaction mixture. At the same time, the greenhouse effect resulting from the burning of fossil energy has seriously affected the earth. In this regard, many countries are actively investing in the development of pollution-free clean energy and alternative energy [8–10]. Hydrogen is one of the prominent alternative energy because of its many excellent properties, especially its combustion product of water [11,12]. However, difficulties in storage and ecological environment transportation of hydrogen persist [13,14]. Liquid fuel reformation is becoming an increasingly important process of hydrogen production for on-board mobile applications [15,16]. The use of bioethanol in the schemes of thermochemical recovery of heat contained in exit flue gases is also an option that was considered [17]. It was found that the degree of ethanol conversion is near unity above the temperature of 600 K. Pashchenko [18] compared

thermochemical waste-heat recuperation through steam reforming of liquid biofuels. The maximum transformation coefficient 1.187 was observed for ethanol steam reforming, and a minimum effective temperature of about 600 K was observed for methanol. Methanol, which can be converted to hydrogen at lower temperature as it contains no carbon–carbon bonds, is an excellent hydrogen carrier and is free of storage and transportation issues [19,20]. In addition, methanol can be reformed to produce hydrogen at low temperatures, with very small amount of CO in the products [21]. Hydrogen production processes are numerous, and decisions on the choice of fuel are made based on which parameter is deemed most important for the system. Among various hydrogen production technologies, hydrogen production from methanol steam reforming (MSR) has attracted attention in the industry, due to its mild reaction and high hydrogen content of products.

Hydrogen production from endothermic MSR heated by exhaust can recover waste heat of the exhaust, increasing the fuel utilization. At the same time, the hydrogen from MSR can be sent to the internal combustion engine, which improves fuel combustion efficiency [22]. Thus, MSR heated by exhaust is considered to be an effective form of waste heat recovery [23,24]. Mishra [25] designed an experimental system for hydrogen production from MSR heated by automobile exhaust, and mainly studied the effects of hydrogen flow rate and exhaust heat exchange rate, on exhaust composition and reaction performance, under different conditions. The results showed that when the throttle opening is within 20%, the exhaust temperature and heat flow can meet the needs of hydrogen production from MSR. Methanol conversion increases with the heat exchange efficiency of reformers, and heat recovery increases with increasing engine speed. However, too high an engine speed will cause the heat exchange efficiency to decrease. Kumar [26] used flow-through tubular heat exchanger and porous ceramic reactors to enhance the heat transfer, and studied hydrogen production from MSR heated by exhausts. The results showed that the methanol conversion increased with the increasing temperature of the exhaust. At exhaust temperatures of 350 ◦C, the hydrogen volume fraction was approximately 42%. This method can provide hydrogen for on-board applications in an internal combustion engine, greatly improving the thermal efficiency of the system. Wang [27] studied the characteristics of the MSR-coupled with thermoelectric generator system heated by automobile exhausts. The results showed that when the temperature difference between the cold and hot ends of the thermoelectric module was 22 K, the output voltage of the power chip was 55 mV, the methanol conversion was 72.6%, and the molar fraction of hydrogen was 62.6%.

The packed bed is widely used for the conventional MSR method. However, the packed bed was reported to suffer from axial temperature gradients and cold spots [28,29]. These problems, which lead to thermal stresses in the channels, result from the severe limitations of mass and heat transfer. The stability and durability of the catalyst are significantly affected by the thermal stresses. Furthermore, the severe transfer resistance led to an effectiveness factor of the catalyst that is typically less than 5% in conventional steam reformers [30]. Micro-reactors can offer a higher heat transfer rate than the traditional chemical reactors, benefit from the high surface-to-volume ratio and short conduction paths [31]. Since the small diameters of the reactor channels can shorten the radial diffusion time, a high heat transfer coefficient is acquired. Moreover, the heat transfer coefficient is known to beneficial for the homogeneously catalyzed reaction [32]. Thus, microreactors have been increasingly seen as new tools for chemistry and chemical processes in recent years. Zhou [33] improved hydrogen production efficiency through sintered copper microreactors. Liang [34] studied the effect of the novel high-pressure propulsion on hydrogen production from MSR. The result showed that the methanol conversion increased by 11% in the microreactor. This behavior was attributed to the superior heat transfer in the microreactors. Pressure drop has been demonstrated to play a significant role in packed bed reformers in terms of the efficiency of the thermochemical heat recuperation systems [35]. However, the difficulty of introducing catalyst particles into the micro-channel persists when using micro-reactors in heterogeneously catalyzed gasphase reactions. Therefore, each channel must be packed identically to avoid misdistribution, because random packing would result in a high-pressure drop. The catalyst coating of regular geometry is convenient to be integrated into microreactors, compared to the packed

bed of catalyst particles. And the catalyst coating is found to be combined closely with the microreactor. This can intensify thermal conductivity from microreactor to the coating due to the decreased thermal contact resistance. The pressure drop is lower in a coated catalyst bed, because the coating catalyst provides the advantage of superior geometry. The activity of the coated catalyst was also found to be superior to that of the same catalyst in a packed bed for MSR [36]. Therefore, for this study, a microreactor coupled with catalyst coating is proposed to intensify the process due to its advantages of heat transfer. th

Previous research work has focused on the study of systems with conventional reactors, and studies on the influence of specific operating parameters on MSR is insufficient. While the vehicles are in motion, the temperature and flow of the exhaust would change at different motor conditions. In this paper, a novel rib microreactor coupled with a catalyst coating is designed for the hydrogen production from MSR heated by automobile exhausts. The exhaust provides heat to the MSR in the same rib microreactor without outside heat source, and the effect of inlet exhaust and methanol steam on reactor performance is numerically analyzed in detail. This research can create a reference significance for the comprehensive utilization of exhaust heat and hydrogen production heated by engine exhaust reforming. .

#### **2. Materials and Methods**

#### *2.1. Physical Model*

The physical model is shown in Figure 1. The exhaust heats the reactant while flowing through the rib microreactor. The mixture of methanol and water enters the reaction side from the reactant inlet, and the products flow out of the outlet. The microreactor chamber is 100 mm long, with a radius of 35 mm, and the heating side radius is 26 mm; the single reaction side angle is 10 degrees and the intermediate baffle thickness is 1 mm. As the structure, the volume, and the reaction performance of the single reaction unit of the reactor are completely same and it is a symmetric model. In order to facilitate the calculation, a half of the single reaction unit is calculated in this paper.

**Figure 1.** Schematic diagram of rib reactor.

#### *2.2. Mathematical Model*

In order to simplify the analysis, the following simplified assumptions are made for this reaction, combining the following characteristics:


(3) The system is in a stable state, and the laminar flow model is adopted;

(4) Ignoring the influence of gravity;


(7) The catalyst area is considered as a homogeneous medium.

A universal finite rate model and the homogeneous model for the reactor in fluent software is used. The model's control equations are as follows:

Continuity equation:

$$\frac{\partial(\rho V\_j)}{\partial \mathbf{x}\_j} = \mathbf{0} \tag{1}$$

Component equation:

$$
\rho V\_{\dot{j}} \frac{\partial Y\_s}{\partial \mathbf{x}\_{\dot{j}}} = \frac{\partial}{\partial \mathbf{x}\_{\dot{j}}} (\rho D \frac{\partial Y\_s}{\partial \mathbf{x}\_{\dot{j}}}) + R\_s \tag{2}
$$

Momentum equation:

$$\frac{\partial(\rho V\_{\dot{j}}V\_{\dot{i}})}{\partial \mathbf{x}\_{\dot{j}}} = -\frac{\partial p}{\partial \mathbf{x}\_{\dot{i}}} + \frac{\partial}{\partial \mathbf{x}\_{\dot{j}}} \Big(\mu \frac{\partial V\_{\dot{i}}}{\partial \mathbf{x}\_{\dot{j}}} - \mu \frac{\partial V\_{\dot{j}}}{\partial \mathbf{x}\_{\dot{i}}}\Big) \tag{3}$$

Energy equation:

$$
\langle V\_{\rangle} \frac{\partial (\rho h)}{\partial \mathbf{x}\_{\rangle}} = \frac{\partial}{\partial \mathbf{x}\_{\rangle}} (\mathbf{k} \frac{\partial T}{\partial \mathbf{x}\_{\rangle}}) + \frac{\partial}{\partial \mathbf{x}\_{\rangle}} (\sum\_{s} \rho D\_{i} \frac{\partial Y\_{s}}{\partial \mathbf{x}\_{\rangle}} h\_{i}) + q \tag{4}
$$

$$h\_{\rm s} = h\_0 + \int \mathcal{C}\_{\rm ps} dT\tag{5}$$

The ideal gas state equation:

$$p = \rho RT \sum \frac{Y\_s}{M\_s} \tag{6}$$

where p, T, ρ are the pressure, the temperature, and the density of the mixed gas, respectively, *X<sup>j</sup>* is the direction, and V<sup>j</sup> is the mixed gas velocity. D, λ, µ are the diffusion coefficient, thermal conductivity, and viscosity coefficient of the mixed gas, respectively, and the ideal gas mixing law is used for the calculations. Y<sup>s</sup> is the mass fraction of component s, s = 1–5, respectively, for CH3OH, H2O, H2, CO2, CO. Cps is the constant pressure specific heat, M<sup>s</sup> is the molar mass of component s, and h<sup>s</sup> is the enthalpy of component s.

#### *2.3. Solving Method*

The fluent software is used for simulation calculation. Three-dimensional symmetry and laminar flow model are adopted, The speed inlet is used for the inlet of reactants and exhaust use, and the pressure outlet is used for the outlets. Fluid-structure coupled heat transfer is adopted for the interface between the reactor and the heater, a symmetrical model is adpoted for the symmetrical surface adopts.All the outer wall surfaces are set as adiabatic, and the copper-based catalyst is uniformly loaded inside the reactor.

#### *2.4. Model Validation*

In this paper, an experimental platform is built, and the MSR heated by the waste heat of exhaust is studied in a plate-type. After a comprehensive comparison, the reaction mechanism and kinetic model in the literature [24] are selected for calculation. The same boundary conditions and parameters as the experiment are adopted. The results are shown in Figure 2.

**Figure 2.** Comparison of the experimental and simulation results.

From the figure, it can be seen that methanol conversion changes of the simulation and the experiment are the same. The methanol conversion increases gradually with the increase of the exhaust temperature, and the maximum difference is only 0.8%. After the verification, it can be confirmed that the reaction mechanism and kinetics adopted in this paper are feasible.

#### **3. Results and Discussions**

#### *3.1. E*ff*ects of Inlet Exhaust Velocity on MSR*

At inlet reactant temperature of 453 K, inlet reactant velocity of 0.1 m/s and inlet exhaust temperature of 673 K, the characteristics of the MSR are shown as figures when the inlet exhaust velocity increases from 0.3 m/s to 1.9 m/s. Figure 3a–d shows the temperature distribution, methanol conversion, and hydrogen and carbon monoxide in the direction of the central axis on the reaction side, when the inlet velocity of the exhaust increases from 0.3 m/s to 1.7 m/s. As can be seen from the figure, the temperature, methanol conversion, the hydrogen mole fraction along the axis all increase gradually with inlet exhaust velocity. The results agree with the literatures [27,37], and suitable for heterogeneous catalytic hydrogen production from MSR in microreactor. This is because the total amount of heat supplied to MSR increases as the inlet exhaust velocity increases, so the temperature of MSR increases. Since the MSR is endothermic, the methanol conversion increases. The temperature is lower at the front section on the reaction side due to the lower inlet temperature of the reactants. With the increase of the temperature in the axial direction, the MSR is favored, and the reaction intensity increases initially and decreases afterwards, along the axis. The axial temperature does not change much before 30 mm from the inlet, and then increases gradually. The hydrogen molar fraction increases slightly before 30 mm, is comparatively larger from 30 mm to 85 mm, and tends to be gentle after 85 mm. Before 80 mm, the molar fraction of carbon monoxide increases slowly with the inlet exhaust velocity and increases at a higher rate, from 80 mm to 100 mm. The same is the trend of the hydrogen mole fraction, as the methanol conversion increases at 30 mm from the inlet and then flattens.

Figure 4a shows the outlet temperature change with the inlet exhaust velocity. With an increase in the inlet exhaust velocity, the outlet temperature of the MSR and the exhaust, and the temperature difference between the MSR and the exhaust increases. The outlet exhaust temperature is always higher than that of MSR. This is because the increase of the inlet exhaust velocity leads to a direct export of some heat, without participating in the MSR. Therefore, the outlet exhaust temperature becomes higher. As shown in Figure 4b, there is an increase in methanol conversion and thermal efficiency as the inlet exhaust velocity increases. This is because the heat absorption from the exhaust increases with an increase in the inlet exhaust velocity, and the methanol conversion. The increase of the outlet exhaust velocity leads to an increase in the output heat and a decrease in the thermal efficiency. As shown in Figure 4c, an increase in the methanol conversion causes an increase of the product, so the mole fraction of hydrogen, carbon monoxide, and carbon dioxide also increases with the increase of the

inlet exhaust velocity. When the inlet exhaust velocity is lower than 1.1 m/s, the product increases with the increasing of inlet exhaust velocity, and then tends to be stable. This is consistent with the change trend of the methanol conversion and thermal efficiency. The highest efficiency is achieved when the inlet exhaust is 1.1 m/s, and a methanol conversion and waste heat utilization ratio of 88.07% and 21.93% is obtained, respectively.

**Figure 3.** Temperature in the axial direction (**a**), methanol conversion (**b**), hydrogen mole fraction (**c**), and carbon monoxide mole fraction (**d**), as a function of inlet exhaust velocity.

**Figure 4.** Outlet temperature on the reaction and exhaust side (**a**), methanol conversion and thermal efficiency (**b**), and reaction product mole fraction (**c**) as a function of inlet exhaust velocity.

#### *3.2. E*ff*ects of Inlet Exhaust Temperature on MSR*

At inlet reactant temperature of 453 K, inlet reactant velocity of 0.1 m/s and inlet exhaust velocity of 0.1 m/s, the characteristics of the MSR are studied when the inlet exhaust temperature increases from 573 K to 873 K. Figure 5a shows the axial temperature distribution at different inlet exhaust temperatures. As the inlet exhaust temperature increases, the heat from the exhaust to MSR increases, so the axial temperature increases. The results also agree with the literatures [37]. The temperature of the reactants increases slowly before 20 mm, and then increases rapidly from 20 mm to 80 mm, and tends to be stable after 80 mm. Due to the low inlet temperature of the reactants, it needs to absorb the heat before reaching the reaction temperature, so the temperature at the front section of the entrance side is lower. As the reactant temperature increases, the MSR deepens, causing the amount of unreacted reactants to decrease. Consequently, the amount of heat absorption along the axial direction decreases, and the axial temperature increases gradually and tends to be stable. Figure 5b–d shows the axial distribution of methanol conversion, and hydrogen and carbon monoxide mole fraction, with different inlet exhaust temperature. All of these increase with an increase of the inlet exhaust temperature and increases gradually along the axis. It can be seen that the hydrogen mole fraction changes little before 18 mm and then increases gradually. When the inlet exhaust temperature is greater than 773 K, it stabilizes at about 80 mm from the entrance. This is because the reaction is almost completed at the position of about 80 mm when the exhaust temperature is 773 K, so the amount of the product changes little. Since hydrogen is the main product of the MSR, the change trends of the hydrogen and methanol conversion are similar. The molar fraction of carbon monoxide is almost 0 before 30 mm, and increases gradually after 30 mm. This is because the temperature is lower than that of methanol decomposition, 30 mm before the entrance.

**Figure 5.** Axial temperature (**a**), methanol conversion (**b**), hydrogen mole fraction (**c**), carbon monoxide mole fraction (**d**), as a function of the exhaust temperature.

As shown in Figure 6a, the outlet temperature of the reactant and the exhaust increases with the inlet exhaust temperature. The higher the inlet exhaust temperature, the smaller the outlet temperature difference between the reactant and the exhaust, which decreases from 14.2 K to 6.88 K. The main reason is that the inlet exhaust temperature increase causes the heat absorbed to increase, and the main reaction section of the MSR moves parallel. In the latter part of the reaction side, the heat absorbed by the reactant is mainly used to raise the temperature of the reactant rather than supplying to the reaction. This leads to the outlet temperature rise. Figure 6b shows the change of thermal efficiency and methanol conversion with the inlet exhaust temperature. The methanol conversion increases from 61% to 99.9% and the thermal efficiency increases from 16% to 26%, with the inlet exhaust temperature. This is because the heat absorbed by the reactant increases with the increase of the inlet exhaust temperature, causing the methanol conversion to increase, as a result, more heat is utilized and the resulting thermal efficiency is higher. When the exhaust temperature is higher than 773 K, the increase of methanol conversion and thermal efficiency increases slowly. Since the heat increase caused by the inlet exhaust temperature is not supplied to MSR, the impact of the increase of the exhaust temperature on the MSR reaction becomes weak. Figure 6c shows the change of the mole fraction of the reaction product with the exhaust temperature. The products have the same change tendency, as the methanol conversion increases with the exhaust temperature. Therefore, when the inlet exhaust is 773 K, the best performance is achieved. At this time, the methanol conversion is 98%, the thermal utilization is 24.6%, and the mole fraction of hydrogen is 69%.

**Figure 6.** Outlet temperature of reactant and exhaust (**a**), methanol conversion and thermal efficiency (**b**), and mole fraction of the reaction product (**c**), as a function of inlet exhaust temperature.

#### *3.3. E*ff*ects of Reactant Inlet Velocity on MSR*

The characteristics of the MSR are studied when the inlet reactant velocity increases from 0.01 m/s to 0.3 m/s, at inlet exhaust velocity of 1.1 m/s, inlet exhaust temperature of 673 K, inlet reactant temperature of 453 K. As the inlet reactant velocity increases, the heat absorption by the reactant increases, resulting in a decrease of the temperature. The change laws agree with the literatures [37,38], but the increasing range is larger, because the temperature is higher than literature one. As shown in

Figure 7a, the axial temperature increases gradually along the axis and decreases with the reactant inlet velocity. When the inlet reactant velocity is 0.01 m/s, the axial temperature increases rapidly and tends to stabilize at about 55 mm from the entrance. When the inlet reactants velocity is 0.05 m/s, the axial temperature tends to be stable at about 80 mm. When the inlet reactant velocity is more than 0.05 m/s, the axial temperature increases slowly, without being stable before the outlet. The heat supplied by the exhaust can meet the needs of the MSR with the inlet reactant velocity being less than 0.05 m/s, and the reaction starts at the entrance, with the temperature rising rapidly. When the heat supplied by the exhaust cannot satisfy the reaction with the inlet reactant velocity by more than 0.05 m/s, the reaction moves in the opposite direction and the reactant temperature side increases slowly. Figure 7b–d indicate the axial distribution of methanol conversion, hydrogen, and carbon monoxide, with the inlet reactant velocity. The methanol conversion, and the mole fractions of hydrogen and carbon monoxide decrease with the increase of inlet exhaust velocity and gradually increases along the axis. As can be seen from the figure, when the inlet velocity of the reactants are 0.01 m/s and 0.05 m/s, respectively, the methanol conversion and the mole fraction of hydrogen increases rapidly and become stable near the outlet. When the inlet velocity is greater than 0.05 m/s, the methanol conversion and the hydrogen mole fraction keep increasing along the axis. When the reactant inlet velocity is small, the heat supplied is sufficient for the MSR on the reaction side. Therefore, the methanol conversion and the products are already stable before the outlet. When the inlet reactant velocity increases, the heat absorption increases, resulting in the MSR moving in the opposite direction.

**Figure 7.** Temperature in the axial direction (**a**), methanol conversion (**b**), hydrogen mole fraction (**c**), and carbon monoxide mole fraction (**d**), as a function of reactant inlet velocity.

As shown in Figure 8a, when the reactant inlet velocity increases, the outlet temperature on the reactant and the exhaust decreases and the temperature difference between each other increases from 1 K to 41 K. At the constant amount of heat supplied from the exhaust, the heat required by the reactants increases when the reactant inlet velocity increases, so the outlet temperature decreases greatly. At this time, the heat is mainly used to supply the endothermic reaction. Moreover, the reactant temperature decreases with the increasing inlet reactant velocity. Figure 8b shows the change of the

thermal efficiency and methanol conversion with the inlet reactant velocity. With the increase of the inlet reactant velocity, the thermal efficiency increases from 6% to 31.7%, and the methanol conversion decreases from 99.6% to 45.7%. This is because with the increase of the inlet velocity of the reactants, the contact time becomes shorter and the total heat cannot satisfy the heat absorbed, so the methanol conversion decreases. The thermal efficiency increase is caused by the increase of the temperature difference between the reactant and the exhaust. When the exhaust inlet velocity is 1.1 m/s, and the thermal efficiency is also considered, the reactant inlet velocity of 0.1 m/s is found to be optimal. Although methanol conversion is enhanced, the actual mass of hydrogen produced is indeed small at this condition, and the throughput can be increased by integrating a certain amount of rib microreactors. As shown in Figure 8c, the mole fractions of hydrogen, carbon monoxide, and carbon dioxide vary with the inlet velocity of the reactant. It indicates that the hydrogen mole fractions decreases with the increase of the inlet reactant velocity.

**Figure 8.** Outlet temperature of product and exhaust (**a**), methanol conversion and thermal efficiency (**b**), and mole fraction (**c**), as a function of reactant inlet velocity.

#### *3.4. E*ff*ects of Reactant Inlet Temperature on MSR*

At inlet exhaust velocity of 1.1 m/s, temperature of 673 K and inlet reactant velocity of 0.1 m/s, the characteristics of the MSR are are illustrated in figures when the inlet reactant temperature increases from 359 K to 573 K. Figure 9a shows the axial temperature change with the inlet reactant temperature. The wall temperature on the reactant side increases as the inlet reactant temperature increases. When the inlet reactant temperature is higher than 493 K, the axial temperature begins to decrease, and then increases with the observed minimum temperature, at about 25mm. When the inlet reactant temperature is below 493 K, the axial temperature increases along the axis" and increases slower after about 70 mm. At lower inlet reactant temperatures, the MSR reaction is relatively moderate without the temperature dropping significantly, and the "cold spot" appears at about 25 mm. At the lower inlet reactant temperature, the MSR reacts relatively gently without the temperature dropping significantly, so the "cold spot" is not observed.The "cold spot" temperature difference is smaller than that of the literature [28,29] because a microreator coupled with catalyst coating which has advantages

of efficient heat transfer is adoptd in this study. Most of the reaction is completed at about 70 mm, after which the temperature increases rapidly. Figure 9b–d shows the axial distribution of methanol conversion and the mole fractions of hydrogen and carbon monoxide at different inlet temperatures.. It can be seen from the figure that the methanol conversion rate and the molar fraction of hydrogen and carbon monoxide gradually increase in the axial direction, and increase with the increase of the inlet reactant temperature. When the inlet reactant temperature is 533 K, the MSR reaction starts at the entrance. The main reason is that the heat carried by the reactants can reach MSR at a relatively high inlet reactant temperature, and absorbs a large amount of heat, which results in a "cold spot" at the entrance. In contrast, the MSR reaction is relatively gentle at lower inlet temperature. At this time, the methanol conversion and the mole fractions of hydrogen and carbon monoxide steadily increase along the axis.. The temperature of 359 K is the vaporization temperature of the reactants, and the reactants need to absorb heat.

**Figure 9.** Temperature along the axial direction(**a**), methanol conversion (**b**), hydrogen mole fraction (**c**), and carbon monoxide mole fraction (**d**), as a function of reactant inlet temperature.

As the inlet temperature of the reactants increases, most of the reactions are completed before the outlet. At this time, the heat absorption from the exhaust reduces. Since the total amount of the exhaust is constant, the outlet temperature of the exhaust and the reaction side increases, and the temperature difference between the two sides decreases gradually. As shown in Figure 10a, the outlet temperature increases from 549 K to 607.5 K at the reaction side and the outlet temperature difference between the exhaust and the reaction side decreases from 21.9 K to 6.8 K.

Figure 10b is the change of the methanol conversion and the thermal efficiency with the reactant inlet temperature. As the reactant inlet temperature increases, the methanol conversion increases and the thermal efficiency decreases. When the inlet temperature is 359 K and 453 K, the methanol conversion is 74.5% and 88%, respectively. There is a big difference between the two conversions. The main reason is that the reaction temperature of the MSR based on the copper catalysts is higher

than 359 K. The reactant is in a state of vaporization at a temperature of 359 K, and the temperature needs to be increased before the reaction. When the reactant inlet temperature is 453 K, the reactants react as soon as it contacts the catalyst, the methanol conversion increases and the thermal efficiency decreases. As shown in Figure 10c, the mole fraction of the product increases with the increasing inlet reactant temperature, and the hydrogen mole fraction increases from 59% to 68%.

**Figure 10.** Outlet temperatures of the reaction side and the exhaust side (**a**), methanol conversion and thermal efficiency (**b**), and product mole fraction (**c**), as a function of reactant inlet temperature.

#### *3.5. E*ff*ects of W*/*A on MSR*

The characteristics of the MSR are shown as figures when the W/A increases from 1.1 to 1.6, at inlet reactant velocity of 0.1 m/s, temperature of 453 K and inlet exhaust velocity of 1.1 m/s. W/A (water-to-alcohol) indicates the molar ratio of water/methanol. Figure 11a shows the change of the axial temperature with the W/A. As can be seen from the figure, the axial temperature increases gradually with the increasing of W/A along the axis. When the W/A is 1.1 and 1.6, the outlet temperatures are 575 K and 582 K, respectively, with a little temperature difference observed. This indicates that W/A is not the most important factor for the MSR under the flow reaction conditions. The result agrees with the literature [33]. This is also confirmed by the change of the methanol conversion and the mole fractions of carbon monoxide and hydrogen with the W/A. As shown in Figure 11b–d, MSR does not start before about 25 mm, in all cases. This indicates that the heat is the main factor of influencing MSR. With the increase of W/A, the methanol conversion increases, as the methanol content in the unit mass of the reactant decreases at a constant heat. At the same time, as the total amount of reactant decreases, the products decrease and the mole fraction of hydrogen and carbon monoxide decreases as the W/A increases.

**Figure 11.** Axial temperature (**a**), methanol conversion (**b**), hydrogen mole fraction (**c**), and carbon monoxide mole fraction (**d**) as a function of water-to-alcohol (W/A).

As shown in Figure 12a, the outlet temperature of the reaction and exhaust side increases with the increase of W/A. The temperature difference between the reaction and exhaust side does not change significantly while the maximum and the minimum temperature difference are 14.1 K and 11.4 K, respectively. This indicates that the change of the W/A has a slight effect on the MSR reaction. Simultaneously, it is verified that the heat is the main influencing factor at this time. Figure 12b shows the methanol conversion and thermal efficiency as a function of the W/A. The methanol conversion increases from 81.7% to 93.6% with the increase of W/A, and the thermal efficiency decreases from 22.3% to 21%. With the increase of the W/A, the MSR is conducive to hydrogen production, and the methanol conversion and hydrogen production rate increase. Sine water has a greater latent heat of vaporization and heat capacity, the increase of water content in the reactant leads to more heat consumption, which causes a drop in thermal efficiency. As shown in Figure 12c, the molar concentrations of carbon monoxide, hydrogen, and carbon dioxide decrease with increasinge W/A, and the optimal W/A in this work is 1.3.

#### *3.6. E*ff*ects of Parallel and Counter Flow on MSR*

Figure 13 shows the temperature distribution of the parallel flow and counter flows when the inlet exhaust temperature is 773 K. Compared with the parallel flow, the temperature difference in the adjacent area of the reactor is smaller than that of the counter flow. The internal temperature increases on the reaction side, and the heat transferred from the exhaust to the MSR, increases.

592

596

21.8 22.0 22.2 22.4

/%

反反反反反反反 热热反反反反 热热反反反反 Outlet temperature of MSR Outlet temperature of exhaust

**(a) (b)**

**Figure 12.** Outlet temperature on the reaction and exhaust side (**a**), methanol conversion and thermal efficiency (**b**), and mole fraction of reaction product (**c**), as a function of the W/A.

reactant inlet reactant outlet exhaust inlet exhaust inlet As shown in Figure 14a, the outlet temperatures on the reaction and the exhaust side increase along with increase of the inlet exhaust temperature, under the parallel and counter flow. At the parallel flow, the outlet temperature on the exhaust side is higher than that of the reaction side, and the outlet temperature difference decreases as the inlet exhaust temperature increases. At the parallel flow, the outlet temperature on the reaction side is higher than that of the exhaust side, and the temperature difference increases as the inlet exhaust temperature increases. At tehe parallel flow, the outlet of the exhaust is adjacent to the inlet of the lower temperature reactant, and the outlet of the reactants is adjacent to the inlet of high temperature exhaust, so the outlet temperature of reactants is higher than that of the exhaust. The heat supply of the exhaust is not enough for vigorous MSR in the front section, but the heat is sufficient in the rear section on the reaction side. However, for the parallel flow, ,

the temperature of the reaction and the exhaust side both decrease as the reaction proceeds, so the outlet temperature on the reaction side is higher.

When the inlet temperature of the exhaust increases, the heat supplied to the reactant in the rear section increases. Meanwhile, the outlet reactant temperature and the temperature differencebothincrease. As shown in Figure 14b, the methanol conversion increases from 61% to 98% at parallel flow, and increases from 64.8% to 99% at counter flow. The methanol conversion of the counter flow was slightly higher than that of the parallel flow. This is possibly because the MSR is relatively gentle during the counter flow. There is some difference between this result and the literature [39]. The methanol conversion of the counter flow was higher than that of the parallel flow, often higher by 5%. This is probably because the model size of literature is larger than that of this study. The temperature difference between the exhaust and the reaction side is slightly larger in this study, causing a little more heat transfer amount. Therefore, methanol conversion and the thermal efficiency both increase slightly. As shown in Figure 14c, the thermal efficiency increases with an increase of the inlet exhaust temperature in the parallel and the counter flow. The thermal efficiency of the parallel flow increases from 16% to 24% and that of the counter flow increases from 18% to 28%. It can be known that the reactor performance is a little better at the counter flow.

**Figure 14.** Outlet temperature of the methanol steam reforming (MSR) and the exhaust (**a**), methanol conversion (**b**), and the thermal efficiency (**c**), with inlet exhaust temperature under different inlet exhaust temperatures.

#### **4. Conclusions**

A rib microreactor for MSR heated by automobile exhaust was designed to study the effects of inlet exhaust and methanol steam on the reactor performance. The results showed that the inlet temperature of the reactants is the most influential factor for MSR. The total amount of heat supplied to MSR increased as the inlet exhaust velocity increased. The methanol conversion and hydrogen mole along the axis all increased with the inlet exhaust velocity. Since the heat absorbed by the reactant increased with increasing inlet exhaust temperature, methanol conversion increased with increasing inlet exhaust temperature. The axial temperature increased gradually along the axis and decreased with the reactant inlet velocity. The methanol conversion, the mole fractions of hydrogen and carbon monoxide decreased with the increase of inlet exhaust velocity. The W/A slightly influenced the reactor performance of MSR. The best parameter performance of of MSR was observed with inlet exhaust velocity at 1.1 m/s, inlet exhaust temperature at 773 K, inlet reactant velocity at 0.1 m/s, inlet reactant temperature at 493 K, and W/A at 1.3, under counter flow. In addition, the methanol conversion of 99.4% was achieved with a thermal efficiency of 28%. Research results are beneficial for the developments of microreactor in comprehensive utilization of waste heat from heterogeneous catalytic reaction, and provides theoretical support for designing microreactor for waste heat utilization.

**Author Contributions:** Funding acquisition, F.W.; Investigation, G.W.; Methodology, F.W.; Project administration, F.W.; Visualization, G.W.; Writing—original draft, G.W.; Writing—review & editing, B.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Natural Science Foundation of China (50906104).

**Acknowledgments:** The authors acknowledge data sources supported by Yanyun Li.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Effects of Impurities on Pre-Doped and Post-Doped Membranes for High Temperature PEM Fuel Cell Stacks**

**Samuel Simon Araya 1,\* , Sobi Thomas <sup>2</sup> , Andrej Lotriˇc 3,4 , Simon Lennart Sahlin <sup>1</sup> , Vincenzo Liso <sup>1</sup> and Søren Juhl Andreasen <sup>4</sup>**


**Abstract:** In this paper, we experimentally investigated two high temperature polymer electrolyte membrane fuel cell (HT-PEMFC) stacks for their response to the presence of reformate impurities in an anode gas stream. The investigation was aimed at characterizing the effects of reformate impurities at the stack level, including in humidified conditions and identifying fault features for diagnosis purposes. Two HT-PEMFC stacks of 37 cells each with active areas of 165 cm<sup>2</sup> were used with one stack containing a pre-doped membrane with a woven gas diffusion layer (GDL) and the other containing a post-doped membrane with non-woven GDL. Polarization curves and galvanostatic electrochemical impedance spectroscopy (EIS) were used for characterization. We found that both N2 dilution and impurities in the anode feed affected mainly the charge transfer losses, especially on the anode side. We also found that humidification alleviated the poisoning effects of the impurities in the stack with pre-doped membrane electrode assemblies (MEA) and woven GDL but had detrimental effects on the stack with post-doped MEAs and non-woven GDL. We demonstrated that pure and dry hydrogen operation at the end of the tests resulted in significant recovery of the performance losses due to impurities for both stacks even after the humidified reformate operation. This implies that there was only limited acid loss during the test period of around 150 h for each stack.

**Keywords:** PEM; fuel cell; fault; diagnosis; electrochemical impedance spectroscopy; distribution of relaxation times; reformate

#### **1. Introduction**

In recent years, proton exchange membrane fuel cells (PEMFC) have become one of the most researched and most mature fuel cell technologies [1]. However, despite the tremendous research efforts and the technological advancements thus far achieved, further optimization and improvements are still needed to reduce their cost, enhance their durability, and accelerate their commercialization [2,3].

PEMFC technology has evolved into two sub-types; one operating at low-temperature (LT-PEMFC) and the other at high-temperature (HT-PEMFC). In general, the two types consist of the same core components; bipolar plates with flow-field channels, a gas diffusion layer made of carbon fiber, a catalyst layer based on Pt particles and a carbon support, and a proton exchange membrane. The main difference between the two is the material used for the proton exchange membrane. In an LT-PEMFC the Perfluorosulfonic acid (PFSA) membranes, such as Nafion®, are used [4], which require liquid water to achieve good proton conductivity. Hence, their operation is limited to temperatures of up to 100 ◦C if no over-pressure is used. Therefore, liquid water is crucial for the proper operation of a

**Citation:** Simon Araya , S.; Thomas, S.; Lotriˇc, A.; Sahlin, S.L.; Liso, V.; Andreasen, S.J. Characterization of Humidified Reformate Impurities on Pre-Doped and Post-Doped Membranes for High Temperature PEM Fuel Cell Stacks. *Energies* **2021**, *14*, 2994. https://doi.org/10.3390/en14112994

Academic Editor: Francesco Lufrano

Received: 21 April 2021 Accepted: 18 May 2021 Published: 21 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Nafion®-based LT-PEMFC stack, and different techniques are being explored to enhance the humidification process [5,6].

On the other hand, an HT-PEMFC uses a polybenzimidazole (PBI) membrane impregnated with phosphoric acid (H3PO4) to facilitate the transfer of protons, which eliminates the need for liquid water, thereby, allowing for higher temperature operation—typically around 160 ◦C. This higher operating temperature comes with advantages, including easier cooling, more efficient utilization of excess heat, reduced or no water management issues, and a higher tolerance to impurities [7].

The latter also means that the HT-PEMFC systems can use (in addition to pure hydrogen) a variety of fuels, which can be converted into hydrogen-rich gas, without the need for purification. HT-PEMFC systems are commonly integrated with a reformer and use liquid methanol as a fuel, which is easier to transport and has a higher volumetric energy density compared to compressed hydrogen at 700 bar [8,9].

The advantage of using fuels that are more manageable than pure hydrogen when compared with LT-PEMFCs, especially liquid fuels, such as methanol, makes HT-PEMFCs ideal candidates to replace diesel generators for various applications, including as backups for telecom applications and auxiliary power units (APU). In addition, they are considered for combined heat and power (CHP) applications due to the efficient utilization of excess heat, while the infrastructure advantage of liquid fuels provides an edge over LT-PEMFCs for applications in heavy duty transportation both as a main power source and as range extenders [9].

However, there are still challenges that need to be addressed in order to optimize HT-PEMFCs. Durability and stability issues are some of the factors that hinder their wide spread commercialization and that are being studied to optimise fuel cell components and operating conditions [10,11]. For instance, in the steam reforming of hydrocarbons to hydrogen-rich syngas, the steam is supplied with over the stoichiometric ratios. Thus, in addition to hydrogen, CO2, and CO, there is always some water vapor present in the reformed gas. The presence of water can enhance the fuel cell performance by alleviating the CO poisoning effect [12].

On the other hand, Zhou et al. [13] proposed that the water content in the anode gas should be minimized to avoid the performance loss when the HT-PEMFC is operated at lower operating temperatures (i.e., 140–160 ◦C), and researchers reported that this may also cause faster degradation due to increased acid loss [11]. According to Park et al. [14], the performance loss due to a humidified atmosphere can be reversed by precise acid-dosing of the degraded membrane electrode assemblies (MEA).

When the doping level of a PBI membrane exceeds two H3PO<sup>4</sup> molecules per PBI repeat unit, free and mobile acid molecules are present in the membrane, as only two phosphoric acid molecules can bond with each PBI repeat unit [7]. This acid can leach out of the membrane by various mechanisms, such as diffusion, capillary transport, membrane compression, evaporation, and especially, can be washed out by condensed water during shutdowns and/or cold starts [15].

The phosphoric acid anions can adsorb on the Pt particles on the cathode side and occupy the electrochemical surface area (ECSA) of the catalyst for the already slow oxygen reduction reaction (ORR) [16,17]. The low permeability of oxygen in the phosphoric acid and phosphate anion adsorption on the Pt catalyst are considered as the main causes for the lower performance of HT-PEMFC compared to LT-PEMFC under pure hydrogen operation [7,18]. As the acid tends to leach out of the membrane, it can also block some of the pores for reactant gases in the catalyst and gas-diffusion layer (GDL) and, thus, may cause significant mass transport resistance [18].

To alleviate these issues several approaches are taken to optimise the MEA core components, such as improving the design of the PBI membrane by tweaking its structure, acid doping levels, and thickness [19]. There are typically two acid doping methods for PBI membranes. The first method is called pre-doped, in which PBI is dissolved in polyphosphoric acid and the solution is then cast into a membrane. Moisture from the surrounding

environment is sufficient to induce a sol-gel transition by the hydrolysis of polyphosphoric acid to phosphoric acid resulting in phosphoric acid-doped PBI membranes.

The second method is called a post-doped membrane, where the PBI is first dissolved in an organic solvent and cast into a membrane, where the acid doping is then achieved by immersing the PBI membrane into highly concentrated phosphoric acid. The doping levels are usually higher for pre-doped membranes at up to 70 molecules of H3PO<sup>4</sup> per PBI repeat unit [20], while the post-doped membranes have a doping level of around 9–12 molecules of H3PO<sup>4</sup> per PBI repeat unit [21]. Due to swelling, higher doping levels in pre-doped membranes lead to these membranes being thicker compared to post-doped ones. To increase the membrane durability or acid uptake, various fillers (e.g., SiO2, TiO2, aluminium silicate, and graphene oxide) or different synthesis techniques (e.g., sulfonation, cross-linking, or the electrospinning of nanofibers) are used [19,22].

The preparation of the catalyst layer along with the pore sizes and tortuosity of the GDL are important factors when it comes to supplying reactants to the triple phase boundary (TPB) [23], which is the crucial contact point among the Pt catalyst, the PA/PBI electrolyte, and the reactants for the electrochemical reactions to take place. The amount of electrolyte (phosphoric acid) around the Pt particles is of paramount importance, where too little electrolyte will not create sufficient paths for proton transfer and will cause charge transfer issues, while too much electrolyte might cover the active sites for the reactants.

To optimize the catalyst layer, various binders (e.g., polytetrafluoroethylene (PTFE), and polyethylene oxide) can be added to make the layer more hydrophobic and/or better adhere to the membrane and the GDL [24]. Often a sub-layer, called a mesoporous layer (MPL), is used to create better contact between the catalyst layer and the GDL, which also improves the redistribution of acid in the catalyst layer, thus, also enhancing the electrochemical surface area and preventing the substantial intrusion of acid into the GDL [25,26].

Traditionally, the catalyst loading in HT-PEMFCs is higher compared to LT-PEMFCs, and current state-of-the art electrodes have loadings of around 1 mgPt/cm<sup>2</sup> [27]. Some recent studies have investigated MEAs with low platinum loading [28,29], where it was observed that the traditional MPL or catalyst layer compositions may have to be changed to allow increased Pt utilization. Martin et al. [28] showed that electrodes without any binder or ionomer in the catalyst layer and catalyst loading of only 0.1 mgPt/cm<sup>2</sup> yielded a maximum performance of 0.42 W/cm<sup>2</sup> . Yao et al. [29] also investigated an HT-PEMFC without the MPL and with Pt loading of 0.2 mgPt/cm<sup>2</sup> and achieved a power density of 0.32 W/cm<sup>2</sup> .

The GDL is a porous material that is traditionally made of carbon fibers and serves multiple purposes. It provides an electrically conductive pathway for current collection, passage for transport of reactants and removal of the produced heat and water, mechanical support to the MEA, and protection of the catalyst layer from corrosion or erosion caused by flows or other factors. Carbon fibers can either be woven in a so-called carbon cloth or non-woven in a form of carbon paper. The main difference between the two types is that non-woven carbon tend to be thinner, more brittle, and less compressible, compared to the woven forms.

Kannan et al. [30] compared four types of commercially available non-woven GDLs from Freudenberg that were assembled with the same type of a post-doped membrane into the MEAs. The main difference between the investigated GDLs is the composition of the MPL and surface treatment of the GDL. The lowest degradation rate was demonstrated with the GDL that most efficiently retained the phosphoric acid. Therefore, each type of GDL can be tailored to have different porosity, hydrophobicity, and conductivity characteristics.

As with all fuel cell types, the initial activation phase or break-in is crucial to allow the HT-PEMFC to reach its optimal performance and to avoid fast degradation [11]. The break-in period depends on the structure of the components comprising the MEA and on the acid doping method. In a study [21], researchers demonstrated that pre-doped MEAs required at least 30 h to achieve peak performance while post-doped MEAs underwent minor changes in the break-in period, which may indicate that this type of MEA can be used directly without the need for a break-in period. The changes in performance during the break-in can be ascribed to redistribution of the phosphoric acid within the membrane, catalyst layer, and the GDL [25].

In this work, we study two HT-PEMFC stacks assembled with the same components but with different types of MEAs. The first MEA type consists of a post-doped membrane and a non-woven GDL, while the second consists of a pre-doped membrane and a woven GDL. These two types of stacks were exposed to various operating conditions, namely, N<sup>2</sup> dilution and CO and CO<sup>2</sup> poisoning both with and without humidification. For the experimental characterization of the effect of the different operating conditions on the stacks, polarization curves and electrochemical impedance spectroscopy (EIS) were used. The data obtained by EIS were then analyzed by using two approaches: an equivalent circuit model (ECM) and distribution of relaxation times (DRT).

#### **2. Methodology**

#### *2.1. Experimental Setup*

A schematic of the test setup used for the experiments in the current work as well as a photo of one of the fuel cell stacks are shown in Figure 1. The experimental characterization was carried out using a GreenLight Innovation fuel cell test station. Two oil-cooled HT-PEMFC stacks of 37 cells, each with active area of 165 cm<sup>2</sup> , were used. The first stack employed a post-doped Dapozol membrane by Danish Power systems (DPS), which utilized meta-PBI as polymer materials and a typical acid doping level of 8–10 phosphoric acid molecules per repeat unit of PBI. The second stack used pre-doped MEAs by Serenergy A/S that utilized a direct cast membrane of the para-polybenzimidazole type with an acid doping level of 30–40 phosphoric acid molecules per polymer repeat unit.

While both membranes utilized phosphoric acid for proton conduction, the high acid content leads to less mechanical stability of the membranes resulting in differences in the chosen thicknesses of the membranes used, with the post-doped membranes being thinner than the pre-doped membranes. This thinner format of the post-doped membranes partially compensates for the lower acid doping level when it comes to the proton conductivity. Finally, the DPS MEAs employ a non-woven gas diffusion layer (GDL), whereas the Serenergy MEAs use a woven type GDL, and both MEAs have a Pt loading of the electrodes of ∼1 mgPt/cm<sup>2</sup> .

**Figure 1.** Test setup (**a**) A schematic of the test setup. (**b**) One of the short stacks used in the current work.

#### *2.2. Test Procedures*

The tests in the current work consisted of different gas compositions in the anode feed stream of two fuel cell stacks. At the beginning of each test (BOT), the stacks were operated under a pure and dry hydrogen feed on the anode side as a reference for the study. This was then followed by the main tests of the current work—namely, nitrogen dilution, simulated dry reformate and simulated wet reformate. Finally, pure and dry hydrogen in the anode feed was used at the end of the tests (EOT) to check the reversibility of the effects of the impurities on the fuel cell stacks. The test procedures along with the compositions of the anode feed stream for each test step are given in Table 1.

**Table 1.** The test procedures.


Both stacks underwent a 50-h break-in procedure in the beginning of the test at 0.2 A/cm<sup>2</sup> under pure hydrogen operation. Successively, the different fuel compositions were tested on the anode side of the stack, and each fuel composition was operated for 24 h. Polarization curves and EIS spectra were recorded 1 h after the start and at the end of each 24 h test period. Galvanostatic EIS sweeps were recorded between 4 kHz and 0.1 Hz at 20 points per decade using an in-house-built frequency analyzer.

For the EIS measurements, an AC amplitude of 2.5 A was used for all set points, which corresponds to 7.5% of the operating current of 65 A. To minimize measurement errors, three impedance spectra were recorded at each test condition with 15 min of relaxation time before each measurement. The repeatability of the EIS measurements is shown in Figure 2, where it can be seen that the three EIS measurements on pure hydrogen lay on top of each other.

**Figure 2.** Repeatability of the EIS measurements.

For the polarization curve measurements, the current was increased from 0 to 75 A, with a step increase of 2.5 to 10 A to capture the activation losses better and a wider step of 5 A for the remainder of the curve. A dwell time of 3 min was used at each current during the polarization measurements. An anode stoichiometric ratio of 1.3 and cathode stoichiometric ratio of 2.5 were used for all the tests.

An air-bleed step was used in order to recover the effects of CO before proceeding to the next step whenever a test step contained CO in the anode gas composition. The air-bleed step consisted of running the fuel cell at the anode fuel composition of 2% air and 98% H<sup>2</sup> for five minutes at 65 A. The same test matrix was followed for both stacks to carry out the characterization study.

#### *2.3. Data Analysis*

In this work, a combination of polarization curves and EIS measurements were used to monitor the performance and investigate the effects of the different reformate compositions on two stacks. The EIS data was analyzed by means of equivalent circuit model (ECM) fits and DRT peaks in order to characterize both the post-doped and pre-doped stacks.

While ECM fits can provide a quick physical interpretation of the impedance measurements, there are still ambiguities surrounding their interpretation, and different ECMs can fit the same EIS data. Therefore, in the current work, the shapes of the impedance spectra in the Nyquist plots, the DRT peaks, and the understanding of the fuel cell stacks was used as a basis for the choice of ECM and analysis of the results. A typical ECM for a PBI-based HT-PEMFC [31,32], composed of a series connection of an inductor (L), a resistor (R*ohm*), and three parallel resistors and constant phase elements (RkCPE), was used to analyze the different losses of the fuel cell stacks.

A constant phase element is pseudo-capacitive element used to mimic the depressed EIS data of real systems for a better fit by accounting for interface inhomogeneities [33]. In the literature, CPE has been attributed to the DC conductivity and the capacity of an ion conductor [34] as well as to the surface roughness and electrode porosity [33]. However, there are still uncertainties regarding its physical meaning [33].

Therefore, in the current work, CPE was only used for better fits by keeping the exponential coefficient *α* constant in the expression for the impedance of a CPE (*Z* = 1 *<sup>Q</sup>*×*jω<sup>α</sup>* , where *Q* is the pre-factor of the CPE and *α* is the exponent). Only changes in the resistances due to the different operating conditions were monitored and analyzed. A more detailed discussion on the advantages and disadvantages of analyzing impedance measurements via ECM fitting along with the typical models for HT-PEMFCs and their physical interpretations is summarized in our previous work [7].

The ECM used in this work is shown in Figure 3. In the interpretation of fitted resistances, it is generally accepted that the ohmic resistance (R*ohm*) represents all the contact resistances, including the proton conduction across the electrolyte and its changes reflect the changes in proton conductivity [35,36]. High frequency resistance (R*HF*) and intermediate frequency resistance (R*IF*) are associated to charge transfer losses, with the former dominated by the anode charge transfer losses and the latter by those in the cathode [37]. Finally, the low frequency resistance (R*LF*) is ascribed to the mass transport losses [38].

**Figure 3.** The equivalent circuit model used in this work.

Distribution of relaxation times (DRT) analysis is another method of analyzing impedance data, in which the impedance spectra are represented by an infinite number of infinitesimal parallel RkC-elements in series [39]. The impedance data is then resolved on the basis of the time constants and presented as a distribution of these time constants that represent the electrochemical processes in the fuel cell [40]. Further description of DRT analysis and its use in fuel cells can be found in the literature [39,41–44].

The DRT deconvolution of the impedance spectra was performed with a freely available MATLAB application (DRTtools). The deconvolution takes place by discretizing the complex impedance response given by Equation (1) into a finite number of time constants, which gives Equation (2) [39,41–43]:

$$Z(\omega) = R\_0 + R\_{pol} \int\_0^\infty \frac{g(\tau)}{1 + j\omega\tau} d\tau \tag{1}$$

$$Z(\omega) = R\_0 + R\_{pol} \sum\_{k=1}^{N} \frac{\mathcal{G}k}{1 + j\omega \tau\_k} \tag{2}$$

where *R*<sup>0</sup> represents the ohmic resistance, *Rpol* represents the overall polarization resistance of the fuel cell, *j* is the imaginary number, *ω* is the frequency, *τ* = RC is the time constant of the single RkC—element, and *g*(*τ*) represents the distribution function. The term *<sup>g</sup>*(*τ*) <sup>1</sup>+*jωτ dτ* in Equation (1) represents the fraction of the overall polarization with relaxation times between *τ* and *τ* + *dτ* and the term *g<sup>k</sup>* in Equation (2) represents the relative share of each *τ<sup>k</sup>* on the overall polarization resistance. Therefore, in order to account for the absolute resistance distribution and compare the DRT analysis more easily with the information in the Nyquist plot, *g<sup>k</sup>* was scaled by multiplying it with the overall polarization resistance (*h<sup>k</sup>* = *Rpol* × *g<sup>k</sup>* ) [39].

Since Equation (2) cannot be solved numerically, a Tikhonov regularization can be used to stabilize the solution numerically, which, in this work, was set to 10−<sup>5</sup> , based on recommendations from the literature on fuel cells [39,43]. Even though the attribution of the different DRT peaks to the losses in the fuel cells is not straightforward, based on the literature and experience, in this work, peaks below 1 Hz were associated with mass transport losses, peaks between 10 and 50 Hz were attributed to oxygen reduction reaction (ORR) losses, and peaks above 100 Hz were considered to be due to hydrogen oxidation reaction (HOR) losses and proton conduction losses [44]. In a DRT plot, taller peaks indicate higher losses.

#### **3. Results**

#### *3.1. Nitrogen Dilution*

Nitrogen can be present in the anode feed stream of an HT-PEMFC in cases where the fuel cell is fed with reformed natural gas or decomposed ammonia. Nitrogen is known to have a dilution effect when fed into the anode of an HT-PEMFC along with hydrogen [45]. Therefore, nitrogen dilution in the current work was studied for two purposes.

First, given its presence in some reformate mixtures, it is important to investigate its effects at the stack level for the two types of MEAs. Secondly, as an easily available inert gas, nitrogen was used as a buffer gas to complete the anode gas mixtures to 100% when comparing the effects of dry and wet reformate gases on the two stacks. Therefore, nitrogen dilution analysis also has the purpose of distinguishing the effects of dilution from those of humidification. The gas recipes used for the different stages of the tests are shown in Table 1.

The results of the current work show that the performance decreased for both stacks with nitrogen dilution, Figure 4. The initial dilution effects remained unaltered for the 24 h of tests for the pre-doped stack; however, they continued to exacerbate for the post-doped stack. The initial performance of the post-doped stack was higher than that of the predoped stack under the same operating conditions, Figure 4a, while the polarization losses

were higher for the post-doped stack as can be seen from the slopes of the polarization curves in Figure 4a and the impedance spectra in Figure 4b.

The two types of MEAs used in the two stacks had different properties as described in the Section 2, and therefore their performances cannot be compared directly with each other. Other than the MEA properties, the stack assembly can also influence the overall performance. Hence, the current analysis will focus on how the different operating condition affected each stack and what fault features can be identified.

**Figure 4.** The effects of N2 dilution on both stacks. (**a**) Polarization curves. (**b**) EIS spectra.

All the high frequency real axis intercepts in the Nyquist plot lay on top of each other, implying that the ohmic resistance was not affected by dilution. However, there is a size increase of the impedance spectra in all frequency regions, which implies performance losses, perhaps due to reduced triple phase boundaries as a result of the acid redistribution and eventual loss of excess acid.

In Figure 5, the fitted equivalent circuit resistances and the DRT analyses of the two stacks are shown. The ohmic resistances remain almost unaltered in both cases, with a slight decrease for the pre-doped stack, whereas R*HF* is the resistance that is the most impacted by dilution in both cases, increasing both in the beginning and the end of the dilution tests for both stacks, Figure 5a,b.

This can be ascribed to the fact that the dilution was done by reducing the hydrogen concentration of the anode feed and substituting it with nitrogen, which significantly reduced the amount of hydrogen available for the electrochemical reaction in the active sites. The cathode charge transfer loss that dominated R*IF* also increases for both stacks in the beginning of the dilution tests; however, while it continued to increase for the postdoped stack, it recovered for the pre-doped stack. The mass transport losses appeared unaffected for the post-doped stack, and an overall slight increase was observed for the pre-doped stack.

Figure 5c,d show the DRT analyses for the two stacks, where up to five peaks for the post-doped stack and three distinct peaks for the pre-doped stack can be seen. The additional peaks of the post-doped stack are a small one at low frequency right below 10 Hz and another at high frequency, between 1 and 10 kHz. However, none of the peaks are below 1 Hz, which highlights that the mass transport losses as result of nitrogen dilution were not significant in either stack. All the peaks were affected negatively in the beginning of the dilution tests with successive recovery for most of the peaks for both stacks, with an exception at the high frequency region for the post-doped stack, where the DRT peak size continue to increase.

**Figure 5.** EIS data analysis for the effects of N2 dilution on both stacks. (**a**) EC model fitted resistances for the post-doped MEA-based stack. (**b**) EC model fitted resistances for the pre-doped MEA-based stack. (**c**) DRT analysis of the post-doped MEA-based stack. (**d**) DRT analysis of the pre-doped MEA-based stack.

#### *3.2. Poisoning Effects of Dry and Wet Reformate Impurities*

The effects of CO and CO<sup>2</sup> on the performance and durability of an HT-PEMFC are fairly well investigated in the literature [7,27,46]. However, most of these studies were performed at the single cell level and often investigated only simulated dry reformate composition. In the current work, both dry and wet reformate impurities in the anode feed were studied at the stack level for two types of MEAs in order to understand not only the effect of impurities but also the effects of water vapor in the anode feed, which is an inevitable by-product of the methanol steam reforming process. Water has positive effects in low temperature PEM fuel cells, where it is used as a proton transport medium. Nonetheless, its effects on an HT-PEMFC are not fully understood, with some reporting advantages [47] and others recommending that it is avoided under certain operating conditions [13].

#### 3.2.1. Poisoning Effects on an HT-PEMFC Stack with Post-Doped MEAs

The poisoning effects of both dry and wet reformate impurities on the fuel cell stack with a post-doped membrane and non-woven GDL-based MEAs are reported in Figure 6. From the polarization curves in Figure 6a, it can be seen that the performance decreased with the introduction of reformate impurities. The effects of both dry and wet reformate impurities are similar.

The effects for both cases are the highest in the beginning of test and there is a slight recovery, especially at higher current densities at the end of the 24 h tests. This is inline with certain reports that, even though the net effect of N<sup>2</sup> dilution on the thermodynamic,

kinetic, and mass-transport driving forces is approximately independent of the relative humidity, both the relative humidity and N<sup>2</sup> dilution affect the anode potential in the same way that water vapor also causes dilution effects [13,48,49].

**Figure 6.** The effects of dry and wet reformate impurities on an HT-PEMFC stack with an MEA composed of a post-doped PBI membrane and non-woven GDL. (**a**) Polarization curves. (**b**) EIS spectra.

Similar effects can be seen from the Nyquist polts in Figure 6b, which were recorded at the higher current density end of the polarization curves at 0.4 A/cm<sup>2</sup> . The overall increase in the polarization resistance was clear for both dry and wet reformate impurities. While the dry reformate operation appeared to stabilize after the 24-h tests, the wet reformate operation resulted in a significant increase in polarization losses at the end of the tests, despite a slight recovery at the beginning of the tests.

The impedance data was fitted to an equivalent circuit model, and DRT analysis was performed to further analyze the effects of the impurities, Figure 7. As can already be noticed from the high frequency intercept of the Nyquist plots, the ohmic resistance remained almost unaltered throughout the test, with a slight decrease with the introduction of wet reformate, which, however, then increased to the initial values, R*ohm* in Figure 7a. R*HF* and R*IF*, which are mainly associated with the charge transfer losses in the two electrodes increased significantly with the introduction of the dry reformate mixture and continued to increase slightly until the end of the tests.

Wet reformate partly recovered both of the above mentioned resistances; however, they increased back again at the end of tests. In particular, a striking increase in R*IF* is seen at the end of the wet reformate operation, which also corresponds to what is seen in the Nyquist plots. This means that humidification exacerbated the charge transfer losses, especially those at the lower frequency region dominated by ORR. However, both R*HF* and R*IF* returned almost to the initial values under pure dry hydrogen operation at the end of test, which implies that the effects are reversible.

This performance recovery also means that the reason for the losses during the wet reformate operation was not related to acid loss but rather to the interaction between water vapor, CO, CO2, and phosphoric acid. Daletou et al. [47] studied the interaction of water vapor and phosphoric acid and found that water reacted with pyrophosphoric acid in order to maintain the equilibrium concentration of phosphoric acid at a high level, thereby, improving the proton conductivity and fuel cell performance. However, their work assessed only performances with pure hydrogen operation. The conductivity and performance improvement in this work was only observed in the initial period of operation after the introduction of water vapor into the stack.

Finally, the mass transport losses increased slightly for the dry reformate operation in the beginning of test but returned to the initial values at the end of the test. The increase in mass transport losses was more visible for the wet reformate operation; however, this too recovered slightly with time, and the overall losses can be considered mild.

**Figure 7.** EIS data analysis of the poisoning effects of reformate impurities in the anode feed of the post-doped MEA-based HT-PEMFC stack. (**a**) Equivalent circuit model fitted resistance data. (**b**) DRT analysis of dry reformate. (**c**) DRT analysis of wet reformate.

The DRT analysis in Figure 7 shows that all the spectra had at least four distinct peaks at around 10 Hz, 100 Hz, 1 kHz, and 4 kHz. All the peaks increased in size with the introduction of dry CO and CO<sup>2</sup> into the anode stream, Figure 7b. While the peaks below 10 Hz decreased slightly after 24 h, the peak above 100 Hz continued to increase. The peaks remained similar in size with the introduction of humidification instead of N<sup>2</sup> in the anode stream along with CO and CO2, Figure 7c.

However, after the 24 h of testing under the wet reformate conditions, a fifth peak appeared around 2 Hz. This region is on the border between the ORR processes and mass transport; however, it is more likely due to cathode charge transfer losses as it is reflected in the increase in R*IF* in the fitted resistances in Figure 7a. Pure and dry hydrogen operation at the end of the test resulted in significant recovery as already shown from the polarization curves and Nyquist plots. While some residual losses are seen on the peak sizes between 10 Hz and 100 Hz, the high frequency peaks above 100 HZ recovered almost fully.

#### 3.2.2. Poisoning Effects on an HT-PEMFC Stack with Pre-Doped MEAs

Similarly to the previous stack based on post-doped MEAs, the effects of CO and CO<sup>2</sup> were clearly visible on the stack with the pre-doped MEAs, as can be seen from the polarization curves and Nyquist plots in Figure 8ab, respectively. The performance of the stack was slightly higher for the wet reformate operation, especially above 0.25 A/cm<sup>2</sup> , Figure 8a.

Considering that the reformate conditions were tested after the dry reformate conditions, this shows that the humidification alleviated the poisoning effects of CO and CO<sup>2</sup> on the pre-doped MEAs, and it did so until the end of the tests. This is in clear contrast with what was observed for the post-doped MEAs, where humidification did not have performance enhancing effects, and the observed recovery was mainly due to stabilization with time in both the dry and wet reformate conditions. As with the post-doped stack, the recovery with pure hydrogen at the end of tests was significant in the case of the pre-doped stack as well.

**Figure 8.** The effects of dry and wet reformate impurities on an HT-PEMFC stack with an MEA composed of a pre-doped PBI membrane and woven GDL. (**a**) Polarization curves. (**b**) EIS spectra.

The equivalent circuit model fitted resistances are shown in Figure 9a. It can be seen that, similarly to the post-doped stack, the ohmic resistances remain unaltered throughout the tests for the pre-doped stack as well. However, both R*HF* and R*IF* increase with the introduction of dry reformate and continue to increase until the end of tests. Therefore, in the case of the pre-doped stack, the recovery and stabilization seen during the dry reformate tests in the post-doped stack are not observed. The recovery happens during the wet reformate tests, where both R*HF* and R*IF* decrease and remain almost unaltered until the end of the wet reformate operation. The R*HF* for the wet reformate operation was lower than that of the pure hydrogen operation, and, for R*IF*, the values are similar to the resistance of the pure hydrogen operation. This implies that the water in this case limits the poisoning effects of CO on the stack.

Nonetheless, the effects of the dry reformate impurities on the mass transport losses were more significant in the case of pre-doped stack compared to the post-doped stack. An increase in mass transport losses was seen in the beginning of the test with dry reformate. At the end of dry reformate test, it appears, from Figure 9a, that the mass transport recovered slightly. However, this is due to the shape of the Nyquist plot of the "Dry reformate EOT" in Figure 8b, which is different from the other impedance spectra at the low frequency end and could not be fitted properly using the EC model used for the other spectra. Since changing the EC model for just one spectrum would not allow for proper comparison of the other circuit elements, it was preferred to underestimate R*LF*.

Nonetheless, when adding the information obtained from the DRT analysis in Figure 9b, one can see that the mass transport losses indeed increased at the end of the dry reformate operation. In fact, the dry reformate operation at the end of the 24 h of test was the only

spectra in the current work that resulted in a peak below 1 Hz, which, as mentioned, is associated with mass transport losses.

Both the Nyquist plots in Figure 8b and the ECM fitted resistances in Figure 9a show that all the losses were recoverable when operating with pure hydrogen again. In fact, Nyquist plots under pure hydrogen operations before and after the tests almost overlapped each other, and, overall, the fitted resistances were similar. However, as shown in the polarization curves in Figure 8a, there was slight performance degradation at the end of the tests compared to the beginning of life. This is perhaps due to the fact that, unlike EIS, polarization curves are recorded over a range of current densities at conditions that are not completely steady state.

Unlike the post-doped stack, which had up to five DRT peaks, the pre-doped stack exhibited only three distinct peaks for pure hydrogen operation. An additional peak was then observed between 100 Hz and 1 kHz when CO and CO<sup>2</sup> were added to the mix. Since the nitrogen dilution tests in Figure 5d also exhibited only three peak, this additional peak in the charge transfer loss region was attributable to the presence of poisoning agents in the anode feed. It is well documented that CO covers active electrode sites, while CO<sup>2</sup> has dilution effects similar to those seen for N<sup>2</sup> with possible small poisoning effects due to CO production via the reverse-water-gas-shift reaction of CO<sup>2</sup> with H<sup>2</sup> [7,45].

With the introduction of the dry reformate mixture into the anode feed, all the DRT peaks increased in size with the exception of the peak at around 100 Hz. Continued operation under dry reformate conditions then stabilized the stack, and it regained some of the losses. However, at the end of the 24 h of dry reformate tests, a fifth peak was observed below 1 Hz. As already mentioned, even though this was not captured in the EC model fit, it can be attributed to mass transport losses, considering the shape of the EIS spectrum and the DRT peak below 1 Hz.

This shows the advantage of adding DRT analysis and not relying solely on ECM fits for the EIS analysis. In Figure 9c, it can be seen that, in the presence of water, the DRT peak sizes decreased throughout the spectrum except at the highest frequency point. However, after the end of the 24 h, the peaks around 10 Hz increased back close to the initial values.

#### **4. Discussion**

Characterization of the effects of reformate impurities in an HT-PEMFC stack is crucial not only for understanding the poisoning and degradation mechanisms to optimize the fuel cell components but also to identify the fault features for diagnostics and fault mitigation purposes. The effects of the tested operating conditions on the different EIS parameters are summarized in Table 2.

**Table 2.** The characteristic features of dry and wet reformate in HT-PEMFC stacks. The arrows indicate the overall trends of the parameters, where ↑ represents an increase and → represents no significant change.


N<sup>2</sup> dilution negatively affected only the charge transfer losses for both stacks, especially R*HF*, which is mainly associated with the anode charge transfer losses. The proton conductivity and mass transport remained mostly unaltered by N<sup>2</sup> dilution for the stack with post-doped MEAs, while an overall slight increase in the mass transport loss was seen for the pre-doped stack. It is reported in the literature that, with the addition of an inert gas, such as nitrogen, the diffusion resistance becomes more substantial because the effective diffusion coefficient of hydrogen in the gas mixture is reduced by the Maxwell–Stefan effect [48].

However, in the current work, the effects of N<sup>2</sup> dilution were manifested predominantly on the reaction kinetics-dominated high and intermediate frequency resistances rather than the diffusion-dominated mass transport resistance. This could be because dilution reduces the amount of hydrogen on the reaction sites, and this effect appears to be more severe than the mass transport issue caused by N<sup>2</sup> dilution. This is in contrast with what some researchers have observed in half-cell and single cell tests, where the nitrogen dilution effects were limited to mass transport losses [48,50].

Similarly, the effects of CO and CO<sup>2</sup> poisoning on the charge transfer losses were more dominant for both stacks, particularly on the anode-dominated high frequency resistance losses, while the ohmic resistance remained unaltered during the poisoning with or without humidification. While the poisoning effects on the high frequency resistances were to be expected due to the adsorption of CO on the anode Pt particles and the consequent loss of ECSA, the effects on the intermediate frequency resistance were peculiar.

Bevilacqua et al. [50] reported similar effects on the cathode due to CO on the anode feed, which they attributed solely to the drop in voltage that caused an exponential drop in the exchange current density according to the Butler–Volmer equation and led to higher impedance. They excluded any CO cross-over from the anode to the cathode. This could explain the increase in the intermediate frequency resistances of both stacks during the poisoning tests.

The presence of water during CO poisoning, which is reported to alleviate the degrading effects of CO [12], was only observed for the stack with pre-doped MEAs. In fact, upon the introduction of humidification, both R*HF* and R*IF* recovered for the stack with post-doped MEAs, but then continued to increase significantly. This could be because the phosphate anions from the more mobile phosphoric acid in post-doped stacks adsorbed on the cathode [16,17] and could also impede mass transport by blocking some of the pores of the GDL while leaching out [18]. The latter could also be the reason for the mass transport loss observed during dry reformate tests on the stack with pre-doped MEAs, which has significantly higher doping levels compared with the post-doped MEAs.

Performance losses in the presence of water are usually associated with acid loss. However, the fact that there was a significant performance recovery with the pure H<sup>2</sup> operation at the end of the tests indicates that the reason for the losses during the wet reformate operation were not related to acid loss. Acid loss is reported to mainly happen at lower operating temperatures and during shutdown/startup cycles due to the condensation of water that can wash out the acid [13,15]. This is not the case in the current work, as there were only a limited number of shutdown/startup procedures, and the operating temperature was kept constant at 160 ◦C.

Studies suggest both positive and negative effects of humidification in HT-PEMFCs, where some propose that the interaction of water with phosphoric acid improves the proton conductivity and fuel cell performance and others reported negative effects of water vapor, including dilution and acid loss, especially at lower temperatures [11,13,47]. Therefore, it can be said that water vapor in an HT-PEMFC has the dual opposite effects of dilution and that of maintaining the equilibrium concentration of phosphoric acid.

The positive effects of humidification compared to dry poisoning in this work were observed in the initial period of operation after the introduction of water vapor into the stack with post-doped MEAs. However, successively, the negative effects of humidification prevailed for this type of MEAs. On the other hand, humidification alleviated the effects of CO poisoning throughout almost the entire test period and across the frequency spectrum for the stack with pre-doped MEAs. Therefore, we conclude that humidification has a beneficial effect on a reformate operated stack with pre-doped MEAs but has detrimental effects on a stack with post-doped MEAs.

Finally, some of the characteristic features shown in Table 2 can be regarded as fault features, where sudden and continuous increases in the fitted resistances and DRT peaks can be a sign of the presence of impurities. For instance, the size and number of DRT peaks increase when dry CO and CO<sup>2</sup> are introduced to either stack. The stack with pre-doped MEAs exhibited only three DRT peaks under pure H<sup>2</sup> operation and under N<sup>2</sup> dilution. However, when the impurities were introduced, an additional peak appeared in the charge transfer region, which can be used to identify the presence of poisoning agents in the anode feed for a stack with pre-doped MEAs.

#### **5. Conclusions**

In this paper, experimental characterization was performed for two 37-cell HT-PEMFC stacks with active areas of 165 cm<sup>2</sup> . The effects of nitrogen dilution and reformate impurities were studied using polarization curves and EIS measurements. CO and CO<sup>2</sup> were chosen as the reformate impurities, and their effect on the stacks were investigated both with and without humidification. The two stacks used in this work employed two different types of MEAs, one with a post-doped PBI membrane and non-woven GDL and the other with a pre-doped PBI membrane and woven GDL type.

We found that N<sup>2</sup> had a reversible dilution effect, which was mainly manifested by the increase in the anode charge transfer losses for both stacks. This could be because dilution reduces the amount of hydrogen on the reaction sites, and this effect appeared to be more severe than the mass transport issue caused by N<sup>2</sup> dilution. N<sup>2</sup> dilution did not affect the ohmic resistance negatively in either stack. Generally, from the Nyquist plot and the DRT peaks analysis, the stack with pre-doped MEAs was less susceptible to dilution effects and had more stable operation over the dilution test period.

The effects of CO and CO<sup>2</sup> poisoning were also mainly seen on the charge transfer losses for both stacks, while the ohmic resistance remained unaltered even during the poisoning tests irrespective of the presence of water vapor. We also found that humidification had a beneficial effect on a reformate-operated stack with pre-doped MEAs but had detrimental effects on the stack with post-doped MEAs. Nonetheless, there was significant performance recovery for both stacks with the pure hydrogen operation at the end of the tests, implying that the poisoning effects are reversible even in the presence of humidification.

A combination of the magnitude and trend of the changes in the various parameters (equivalent circuit elements and DRT peaks) due to the different operating conditions in the current work can be used for fault matrix creation and for diagnostics purposes. The increase in the size and number of DRT peaks when dry CO and CO<sup>2</sup> were introduced to either stack can be considered as a CO poisoning fault feature. In particular, the stack with pre-doped MEAs exhibited only three DRT peaks under pure H<sup>2</sup> operation and under N<sup>2</sup> dilution. Therefore, the additional DRT peak in the charge transfer region during impurities operation can be used to identify the presence of poisoning agents in the anode feed for a stack with pre-doped MEAs.

**Author Contributions:** Conceptualization, S.S.A. and S.T.; methodology, S.S.A. and S.T.; software, S.S.A., S.T. and S.L.S.; validation, S.S.A. and S.T.; formal analysis, S.S.A.; investigation, S.S.A. and A.L.; resources, S.S.A.; data curation, S.S.A.; writing—original draft preparation, S.S.A. and A.L.; writing—review and editing, S.S.A., S.T., A.L., S.L.S., V.L. and S.J.A.; visualization, S.S.A.; supervision, S.S.A.; project administration, S.S.A. and S.J.A.; funding acquisition, S.S.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Danish Energy Technology Development and Demonstration Program (EUDP) through the project COBRA Drive (grant number—64018-0118). Part of the research conducted for this article received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 893919.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data are contained within the article.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Abbreviation**

The following abbreviations are used in this manuscript:



#### **References**

