Mechanistic Insights into Electrolyte Retention and Flow Optimization in Alkaline Electrolyzers

Abstract

This study addresses the challenges of electrolyte retention and remixing in alkaline electrolyzers, which affect electrolysis efficiency, gas yield, and equipment stability. It focuses on optimizing flow inhomogeneities to improve hydrogen production efficiency. The study investigates how varying flow rates and inlet velocity distributions impact flow states and inhomogeneities within the electrode plate. By modifying the inlet structure, the goal is to reduce low-speed stagnation areas and improve electrolyte flow uniformity, velocity distribution, and residence time. Additionally, the optimization of internal infusion structures, including vertical and horizontal configurations, will be explored to assess their effects on internal flow and electrolysis efficiency. The results will provide key insights and technical guidance for the design optimization of large-scale electrolyzers, advancing the commercialization of water electrolysis hydrogen production while enhancing efficiency and stability.

1. Introduction

Hydrogen energy (123 MJ/kg) enables carbon neutrality via renewable integration [1,2].Green hydrogen from alkaline electrolysis dominates industries due to cost and durability [3,4]. However, scaling AWE (Alkaline Water Electrolysis) systems to multi-megawatt capacities introduces critical hydrodynamic bottlenecks. Maldistributed electrolyte flow, exacerbated by geometric imperfections in bipolar plates, induces stagnation zones where gas bubbles accumulate, elevating ionic resistance and localizing current densities beyond optimal ranges [5,6]. These phenomena collectively degrade Faradaic efficiency and accelerate electrode corrosion, posing formidable barriers to gigawatt-scale commercialization [7,8]. To address these challenges, resolving flow inhomogeneity is imperative to align AWE’s operational performance with its theoretical potential.

To mitigate these hydrodynamic limitations, experimental research has progressively unraveled multiphase flow complexities in AWE systems. Early studies by Eigeldinger and Vogt [9] established quantitative relationships between electrolyte velocity (0.1–1.0 m/s) and bubble detachment frequencies, while Riegel et al. [10] correlated gas holdup (5–20%) with current density fluctuations (±15%) in vertical electrodes. Building on these foundations, modern diagnostic tools, such as laser-induced fluorescence (LIF) and synchronized high-speed imaging, now enable micron-scale resolution of transient gas–liquid interfaces [11,12]. For example, Hine and Murakami [13] demonstrated that forced convection (Re > 2000) reduces bubble-induced voltage overpotentials by 18% compared to natural convection regimes. Despite these advancements, empirical design practices—often reliant on trial-and-error iterations—remain prevalent, resulting in suboptimal flow uniformity at industrial scales. Notably, current studies predominantly focus on laboratory-scale systems (electrode diameters < 0.5 m), leaving a critical gap in understanding flow optimization for electrode plates exceeding 2 m in diameter, where inertial forces and multiphase interactions dominate flow inhomogeneity. A critical gap lies in the inability of conventional experimental setups to replicate dynamic operating conditions (e.g., fluctuating renewable energy inputs), which amplify remixing and electrolyte stratification [14,15]. This limitation underscores the necessity of integrating computational approaches to bridge laboratory-to-industry knowledge gaps.

Building on the experimental challenges in scaling to industrial dimensions, computational fluid dynamics (CFD) has emerged as an indispensable tool for deconvoluting multiscale flow phenomena in large-scale AWE systems. The Euler–Euler framework, treating gas–liquid phases as interpenetrating continua, accurately predicts macroscopic velocity profiles (error < 8% against PIV data) and residence time distributions [16,17]. Conversely, the Euler–Lagrange approach resolves discrete bubble dynamics, enabling precise modeling of coalescence and mass transfer at gas-evolving electrodes [18]. Hybrid methodologies such as Delayed Detached Eddy Simulation (DDES) integrate Reynolds-Averaged Navier–Stokes (RANS) and Large Eddy Simulation (LES), optimizing the trade-off between computational accuracy and cost. For instance, Ziegler and Evans [19] demonstrated that DDES achieves 92% agreement with experimental stagnation zone geometries, validating its efficacy in industrial-scale simulations. Notably, Mat [20] utilized DDES to optimize baffle designs, reducing dead zones by 22% while maintaining pressure drops below 5 kPa. However, persistent challenges include the prohibitive computational cost of resolving turbulence spectra in multiphase flows and the lack of universal correlations for diverse electrolyzer geometries [12]. These unresolved issues highlight the need for experimentally validated, reduced-order models tailored to industrial AWE scaling.

To bridge these gaps, this study addresses the aforementioned challenges through a dual experimental–computational strategy targeting flow uniformity in industrial AWE systems. First, fluorescent tracer experiments quantify electrolyte retention dynamics across a broad operational range (240–600 L/h; 0.5–1.25 m/s inlet velocities), employing sodium fluorescein and UV-PIV to map stagnation zones and remixing intensities. Concurrently, high-fidelity CFD simulations deploy a hybrid DDES model—validated against experimental velocity fields to evaluate novel inlet restrictors and hierarchical baffle geometries (e.g., radial vs. helical deflectors). Key metrics include velocity inhomogeneity indices, defined as the standard deviation of Z-direction velocities normalized by the mean, and residence time distribution (RTD) variances, calculated via discretized Lagrangian particle tracking. The resultant framework not only advances fundamental understanding of AWE hydrodynamics but also delivers actionable guidelines for minimizing energy losses in next-generation hydrogen infrastructures.

2. Equipment and Methods

2.1. Experimental Equipment

To enhance the analysis and visualization of fluid flow patterns within a large bipolar plate, we have established a testing system to investigate flow behavior in a Large Alkaline Water Electrolyzer. This system utilizes plate data provided through collaboration with an external research institute (see Figure 1). Figure 1 shows the 3D modeling structure and Figure 1b shows the real experimental setup. The key component of this system is a full-scale, 1:1 replica of an existing Alkaline Water Electrolyzer plate. The experimental setup consists of several assembled components, including a transparent acrylic plate, external frame, rubber gasket, structured liner, and back frame plate. As shown in Figure 2, the transparent acrylic plate serves as a visual observation window, with an overall diameter of 2000 mm, enabling internal flow visualization. The external frame shares the same outer diameter (2000 mm) as the acrylic plate, with an inner diameter of 1880 mm. Circular connection holes are uniformly distributed along the circumference at 15° intervals to secure the acrylic plate to the back frame.

The rubber gasket is used to seal the assembly, matching the outer and inner diameters of the external frame. Its uncompressed thickness is 5 mm, and it compresses to approximately 4 mm after sealing, effectively ensuring both gas and liquid tightness.

The geometric parameters of the structured liner are shown in Figure 2. The liner has an overall diameter of 1880 mm and a thickness of 4 mm. The surface of the liner is uniformly covered with hemispherical protrusions, each with a radius of 5 mm, forming a disturbance structure to enhance flow behavior. There are eight inlet ports located at the lower part, with a central angle of 15° between adjacent inlets, and four outlet ports at the upper part, also spaced at 15° intervals. Fluid enters through the inlet ports, fills the entire flow structure from bottom to top, and exits through the outlet ports.

The components of the Large Alkaline Water Electrolyzer Internal Flow Visualization and Performance Test System are bolted together to ensure efficient sealing.

The replica preserves the active area (400 cm²), channel width, depth, and flow field pattern, which are critical for replicating realistic electrolyte flow behavior. The serpentine parallel-channel structure and manifold configurations align with standard industrial practices commonly observed in commercial AWE systems.

The flow configuration (vertical upward flow) and channel network are representative of commercial designs aimed at promoting effective gas–liquid separation. While the replica focuses primarily on hydrodynamic characteristics rather than electrochemical performance, ensuring geometric and hydraulic similarity enables meaningful insights into flow uniformity, potential dead zones, and preferential pathways, which are closely linked to gas purity and energy efficiency in industrial operations.

It is important to note that active electrochemical reactions, gas evolution, and long-term durability tests were not conducted in this initial study. Also, it is acknowledged that some industrial systems adopt horizontal flow setups to meet specific design and operational requirements.

Also, one limitation of the current study is that all experiments and simulations were conducted under ambient pressure conditions. While this approach ensures experimental safety and reproducibility, it does not fully replicate the pressurized environments (typically 10–30 bar) commonly used in industrial AWE operations to enhance gas purity and system efficiency. Future work will integrate electrochemical operation and performance evaluation to further validate the representativeness of the design under realistic operating conditions and will implement high-pressure experimental setups and corresponding numerical simulations, allowing for a more comprehensive evaluation of flow behavior and performance under realistic operating pressures.

2.2. Introduction to the Fluorescent Tracer Method and Experimental Procedure

2.2.1. Fluorescent Tracer Method

Given the scale of the internal flow field and the complexities in visualizing extracted data, a fluorescent tracer method was applied to facilitate detailed flow visualization, crucial for subsequent data analysis.

In this study, fluorescent dye visualization was employed primarily to qualitatively identify the dominant flow paths, recirculation zones, and large-scale distribution patterns within the electrolyzer channels. Sodium fluorescein was dissolved in deionized water at a mass ratio of 1:200 and stirred at 600 rpm for 30 min using a magnetic stirrer to ensure complete dissolution, producing a homogeneous fluorescent tracer solution. This tracer emits green fluorescence when excited by a 365 nm ultraviolet light source. Once the fluid flow within the bipolar plate reached a stable state, the fluorescent tracer was introduced at a steady rate, matching the bulk flow velocity, enabling clear visualization of internal flow dynamics, and supporting precise tracer analysis. Observations focused on the early stages of dye propagation, where convective transport dominates over molecular diffusion. It is acknowledged that while dye visualization provides valuable qualitative insights, it does not offer fully quantitative flow measurements. To achieve more detailed and quantitative flow field characterization, future work will incorporate particle image velocimetry (PIV) techniques. It should be noted that in this study, the experimental setup was operated under steady-state conditions to establish a baseline understanding of the flow distribution and to enable direct validation with steady-state numerical simulations. While dynamic operating conditions—such as fluctuating inlet flow rates representing renewable energy variability—are recognized as important factors affecting electrolyte remixing and stratification, they were not investigated in the current work. Future studies are planned to incorporate dynamic boundary conditions to evaluate system robustness under transient operating scenarios. In addition to steady-state tests, future work will explore dynamic operation scenarios to further validate the system’s performance.

2.2.2. Experimental Procedure

The experimental setup of this study is shown in Figure 2, which includes a pure water storage tank, a fluorescent dye reservoir, a pump, a controller, a camera, a flow control mechanism, and an ultraviolet lamp. The controller is designed to independently regulate the pump, enabling precise control over the injection of pure water from the storage tank and fluorescent dye from the dye reservoir at specific flow rates and velocities into the infusion-style flow field assembly. In this configuration, the liquid enters the flow field through eight inlet ports located at the bottom of the bipolar plate, traverses the flow field in a vertical, bottom-up direction, and exits through four outlet ports positioned at the top. This bottom-to-top flow arrangement allows for a comprehensive assessment of flow dynamics within the flow field.

During experimental validation, pure water was initially pumped into the flow field to fully saturate the internal channels. Once a stable flow was established, the fluorescent dye was introduced using the flow control device, and the UV lamp was activated under dark conditions. Photographs were then captured through the camera to observe and document the distribution of the fluorescent dye, providing a simulated visualization of electrolyte flow behavior within the conductive-type flow field component.

2.3. Evaluation Methods for Analyzing the Results of Experiments

Accurately evaluating non-ideal flow phenomena, including stagnation zones and remixing, is essential for enhancing flow uniformity in electrolytic cells. In this study, the analysis of experimental imaging results plays a pivotal role. We evaluated the flow characteristics inside the electrode plate by analyzing the occupied area and distribution characteristics of green fluorescent dye in the experimental photos frame by frame and then presented the residence time distribution. This method is called the area analysis method, and the detailed steps are as follows:

• Determination of scale;

As shown in Figure 3, the first image in the set to be processed is used for scale calibration to determine the true size of the area being analyzed, and this scale is applied to the subsequent processing methods.

2.

Determination of the maximum area and calibration of the fluorescence generation area;

In order to qualitatively analyze the internal flow state, the theoretical maximum flow area and fluorescence appearance area were ticked off and recorded using the ticking tool, as shown in Figure 4, taken from the first image where the green fluorescent dye appeared.

Subsequently, the time interval, $\Delta t$, was determined, and the fluorescence emergence area was determined at the same time interval until a set of experimental photo data were analyzed.

3.

Data processing methods;

To quantitatively analyze the area data, the fluorescence area recorded in each measurement image was compared to the theoretical maximum flow area to obtain the ratio of fluorescence area to total flow area. This ratio was then plotted against time, with the ratio as the vertical axis and time as the horizontal axis, to characterize the temporal evolution of the dyed region within the system. The resulting analysis was used to assess the impact of different structural designs on the optimization of internal flow behavior.

3. Simulation Details

The model described in this study is based on the following assumptions: (1) This study focuses solely on the flow conditions of the main liquid within the polar plate, considering only the liquid phase in the system. (2) According to the experimental methodology employed, the physical properties of the dye used are essentially identical to those of pure water; therefore, the two liquids considered in the calculations share the same physical properties. (3) The flow is assumed to be adiabatic, with negligible heat loss.

To avoid the extremely high grid resolutions typically required in Large Eddy Simulations, this study employs a hybrid Delayed Detached Eddy Simulation (DDES) model. This model integrates elements from both Reynolds-Averaged Navier–Stokes (RANS) and Large Eddy Simulation (LES) approaches. Within this framework, the wall boundary layer is modeled using RANS, while the larger separation regions are addressed using LES, enabling the resolution of parts of the turbulence spectrum in both time and space. The DDES model facilitates the transition between RANS and LES by comparing turbulence length scales with the grid spacing. For the k-ε, BSL, and SST models, the DDES function has been recalibrated to enhance the protection of the boundary layer. In this study, the hybrid Delayed Detached Eddy Simulation (DDES) framework was selected to balance computational cost and flow structure resolution. While DDES does not achieve the full fidelity of DNS or LES methods, it captures key turbulent features with significantly lower computational expense, making it suitable for simulating large-scale industrial flow fields such as those in AWE cells.

A detailed benchmarking of computational cost against other modeling strategies (e.g., RANS, LES, DNS) was not conducted, as this would require an extensive separate study. Future work will explore a systematic cost–performance comparison across different turbulence modeling approaches to further validate the computational advantages of DDES.

The specific equations are provided below.

Continuity equation:

$${\displaystyle \frac{\partial \rho }{\partial t}}+\nabla \cdot \left(\rho u\right)=0$$

(1)

Momentum conservation equation:

$${\displaystyle \frac{\partial }{\partial t}}\left(\rho u\right)+\nabla \cdot \left(\rho uu\right)=\nabla P+\nabla \left({\mu }_{eff}\left(\nabla u+\nabla u^T\right)\right)+\rho g$$

(2)

Species transport equation:

$${\displaystyle \frac{\partial }{\partial t}}\left(\rho Y_i\right)+\nabla \cdot \left(\rho u{Y}_i\right)=-\nabla \cdot J_i+R_i+S_i$$

(3)

where $R_i$ is the net rate of production of species i by the chemical reaction (described later in this section) and $S_i$ is the rate of creation by addition from the dispersed phase plus any user-defined sources. An equation of this form will be solved for the N-1 species where N is the total number of fluid phase. In this simulation, both $R_i$ and Y are $S_i$.

In turbulent flows, ANSYS 2023 R1 Fluent computes the mass diffusion in the following form:

$$J_i=-\left(\rho D_{i,m}+{\displaystyle \frac{{\mu }_t}{S{c}_t}}\right)\nabla Y_i-D_{T,i}{\displaystyle \frac{\nabla T}{T}}$$

(4)

where $S{c}_t$, is the turbulent Schmidt number (${\displaystyle \frac{{\mu }_t}{\rho D_t}}$, where ${\mu }_t$ is the turbulent viscosity and $D_t$ is the turbulent diffusivity). The default $S{c}_t$ is 0.7 and $D_{i,m}$ is 2 × 10⁻⁹ m²/s. Note that turbulent diffusion generally overwhelms laminar diffusion, and the specification of detailed laminar diffusion properties in turbulent flows is generally not necessary.

We have simulated the flow within the polar plate using a low Reynolds number corrected SST k-ω turbulence model. The turbulence kinetic energy, k, and the specific dissipation rate, ω, are obtained from the following transport equations:

$${\displaystyle \frac{\partial }{\partial t}}\left(\rho k\right)+{\displaystyle \frac{\partial }{\partial x_i}}\left(\rho k{u}_i\right)={\displaystyle \frac{\partial }{\partial x_j}}\left({\Gamma }_k{\displaystyle \frac{\partial k}{x_j}}\right)+G_k-Y_k+S_k$$

(5)

and

$${\displaystyle \frac{\partial }{\partial t}}\left(\rho \omega \right)+{\displaystyle \frac{\partial }{\partial x_i}}\left(\rho \omega u_i\right)={\displaystyle \frac{\partial }{\partial x_j}}\left({\Gamma }_{\omega }{\displaystyle \frac{\partial \omega }{x_j}}\right)+G_{\omega }-Y_{\omega }+S_{\omega }$$

(6)

In these equations, $G_k$ represents the generation of turbulence kinetic energy due to mean velocity gradients. $G_{\omega }$ represents the generation of ω. ${\Gamma }_k$ and ${\Gamma }_{\omega }$ represent the effective diffusivity of k and ω, respectively. $Y_k$ and $Y_{\omega }$ represent the dissipation of k and ω due to turbulence.

The effective diffusivities for the k-ω model are given by the following:

$$\begin{array}{l}{\Gamma }_k=\mu +{\displaystyle \frac{{\mu }_t}{{\sigma }_k}}\\ {\Gamma }_{\omega }=\mu +{\displaystyle \frac{{\mu }_t}{{\sigma }_{\omega }}}\end{array}$$

(7)

where ${\sigma }_k$ and ${\sigma }_{\omega }$ are the turbulent Prandtl numbers for k and ω, respectively, and ${\mu }_t$ is the turbulent velocity in the SST model. The computational equations are as follows:

$${\mu }_t={\displaystyle \frac{\rho k}{\omega }}{\displaystyle \frac{1}{\max \left[\frac{1}{{\alpha }^{*}},\frac{S{F}_2}{a_1\omega }\right]}}$$

(8)

The coefficient ${\alpha }^{*}$ damps the turbulent viscosity causing a low-Reynolds number correction. It is given by the following:

$${\mu }_t={\displaystyle \frac{\rho k}{\omega }}{\displaystyle \frac{1}{\max \left[\frac{1}{{\alpha }^{*}},\frac{S{F}_2}{a_1\omega }\right]}}$$

(9)

where

$${\mathrm{Re}}_t={\displaystyle \frac{\rho k}{\mu \omega }}$$

(10)

$$R_k=6$$

(11)

$${\alpha }_0^{*}={\displaystyle \frac{{\beta }_i}{3}}$$

(12)

$${\beta }_i=0.072$$

(13)

where $S$ is the strain rate magnitude and ${\alpha }^{*}$ is defined in Equation a. $F_2$ is given by

$$F_2=\tanh \left({\Phi }_2^2\right)$$

(14)

$${\Phi }_2=\max \left[2{\displaystyle \frac{\sqrt{k}}{0.09\omega y}},{\displaystyle \frac{500\mu }{\rho y^2\omega }}\right]$$

(15)

where y is the distance to the next surface.

In this study, a single-phase liquid flow model was employed to characterize the baseline hydrodynamic behavior within the electrolyzer channels. The modeling approach intentionally excluded gas generation and electrochemical reactions in order to isolate the effects of geometry and inlet conditions on flow distribution under controlled conditions.

While this simplification enables a clearer analysis of the fundamental flow structures, it is recognized that gas bubbles generated during electrolysis can influence flow uniformity, pressure drop, and mass transport by altering the effective flow area and introducing buoyancy-driven effects.

Future work will extend the current model to incorporate multiphase flow dynamics, including gas evolution, coalescence, and detachment mechanisms, to more accurately replicate real-world operational scenarios in industrial-scale AWE cells.

4. Results and Discussion

4.1. Mesh Independence Verification

To evaluate numerical discretization errors induced by truncation of higher-order Taylor series terms, a systematic mesh independence study was conducted. Three distinct grid resolutions were implemented with spatial refinement concentrated in the inlet region, as illustrated in Figure 5. The corresponding grid parameters are quantitatively summarized in

Table 1. Detailed information of mesh size.

	Coarse Mesh	Medium Mesh	Fine Mesh
Minimum grid size (mm)	1	0.5	0.2
Maximum grid size (mm)	10	5	1
Growth rate	1.2	1.2	1.2
Normal angle (°)	5	5	5
Number of grids	3,149,825	13,573,397	34,318,148
Minimum quality standard	0.300	0.353	0.489
Number of tetrahedral grids	2,866,964	6,802,644	16,333,886

. Computational comparisons revealed non-negligible discrepancies between coarse-mesh results and those obtained from medium/fine meshes. However, progressive refinement from medium to fine mesh demonstrated diminishing returns. This convergence behavior suggests that medium-resolution discretization achieves optimal balance between computational accuracy and resource efficiency. The medium mesh configuration was adopted for subsequent simulations of all foam structure models, ensuring numerical reliability while maintaining computational economy.

4.2. Total Flow

In order to investigate the effect of the total liquid flow rate on the flow state inside the electrolyzer, we analyzed the law of its effect on the internal flow behavior by adjusting the total liquid flow rate into the electrolyzer. The experiment uses an internal electrode plate structure as shown in Figure 6a, and the initial flow direction is toward the center. The specific experimental implementation is detailed in

Table 2. Experimental parameters under different total flow conditions.

Total Flow (L/h)/ Inlet Flow (L/min)	1	2	3	4	5	6	7	8
240	0.500	0.500	0.500	0.500	0.500	0.500	0.500	0.500
300	0.583	0.583	0.583	0.583	0.583	0.583	0.583	0.583
360	0.750	0.750	0.750	0.750	0.750	0.750	0.750	0.750
420	0.875	0.875	0.875	0.875	0.875	0.875	0.875	0.875
480	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
540	1.125	1.125	1.125	1.125	1.125	1.125	1.125	1.125
600	1.250	1.250	1.250	1.250	1.250	1.250	1.250	1.250

, which lists the experimental parameters under different total flow conditions.

In this study, total flow rates of 240 L/h, 300 L/h, 360 L/h, and 420 L/h were selected for the experiments, and the simulation results, along with the corresponding experimental validation results, are presented in Figure 6b. Once the flow state reached a steady state, the phase distribution cloud diagram inside the polar plate closely matched the experimental results, thereby verifying the accuracy of the simulation. Figure 6a illustrates the comparison between the simulation outcomes and experimental data for the liquid content rate at the polar plate cross-section. The trend in the liquid content rate is largely consistent with the experimental results, further substantiating the reliability of the simulation method. Additionally, we analyzed the residence time distributions at four inlet locations (one for inlets 1 and 8 in Table 2, two for inlets 2 and 7 in Table 2, three for inlets 3 and 6 in Table 2, and four for inlets 4 and 5 in Table 2), with the results displayed in Figure 6c. As the total flow increases, the residence time distribution gradually concentrates toward the center, indicating a decrease in the low-speed stagnation region of the internal flow. Moreover, there are significant differences in the residence times among the different types of inlets. The average residence time distributions for inlet types 1 to 4 were further analyzed, as shown in the table in Figure 6c. The results indicate that inlet types 1 and 4 exhibit a longer residence time trend because they are furthest from the outlet.

To quantitatively assess the flow uniformity at different horizontal cross-sections, we utilized the velocity inhomogeneity parameter,$M_{u,f}$, defined as follows:

$$M_{u,f}={\left[{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{(\frac{u_i-\overline{u}}{\overline{u}})}^2}\right]}^{0.5}$$

(16)

where $u_i$ denotes the velocity magnitude of a single fluid particle in the Z-direction at the position, and $\overline{u}$ denotes the velocity distribution of a fluid particle in the XY-plane at the position.

Figure 7a shows the velocity distribution in the center cross-section of the polar plate, and Figure 7b shows the velocity cloud distribution in the XZ-plane of the polar plate. As shown in Figure 7a, the velocity distribution inside the polar plate tends to homogenize as the velocity increases; Figure 7b shows that the internal low-speed stagnation region gradually decreases as the velocity increases, and the overall velocity distribution is closer to uniform. Furthermore, we counted the XY-plane average velocity distribution of the pole plate at different positions (from inlet to outlet), and the results are shown in Figure 7c. Figure 7c shows that the average velocity inside the plate shows a trend of “decreasing and then increasing” from the bottom inlet position to the top outlet position, with the lowest average velocity at the center of the plate. Based on Figure 7c, we further calculated the distribution law of the velocity inhomogeneity of the plate at different positions, and the results are shown in Figure 7d. From the figure, it can be seen that the velocity inhomogeneity has an overall decreasing trend along the flow direction. The inhomogeneity is larger at the initial position, while there is a weak increase in inhomogeneity at some positions, which may be due to the fact that all intakes at this position have started feeding, which causes the fluid from the outermost intakes 1 and 8 to mix with the main flow for the first time, thus increasing the unevenness. In addition, when the flow rate was increased in the range of 240 L/h to 360 L/h, it contributed to a significant reduction in the velocity inhomogeneity, which led to the homogenization of the internal flow velocity distribution; when the flow rate exceeded 420 L/h, further increase in the flow rate had a lesser effect on the improvement of the velocity inhomogeneity. This is because in Figure 7a,b, it can be seen that when the total flow rate exceeds 420 L/h, the cross-section velocity and velocity cloud image obtained by increasing the total flow rate do not change significantly, indicating that further increasing the total flow rate cannot further optimize the internal flow.

4.3. Import Flow Allocation

By analyzing the residence time distributions and cross-sectional velocity profiles of various inlets, it is evident that differences in residence time exist for each inlet, and the cross-sectional velocity distributions are not uniform under the same total flow conditions. Consequently, this section uses a total flow rate of 360 L/h as a baseline and adjusts the flow distribution among different inlets to optimize the internal velocity uniformity. The specific implementation details are presented in

Table 3. Flow data for different inlets.

Total Flow (L/h)/ Inlet Flow (L/min)	1	2	3	4	5	6	7	8
1	0.3	0.6	0.9	1.2	1.2	0.9	0.6	0.3
2	0.3	0.6	1.2	0.9	0.9	1.2	0.6	0.3
3	0.6	1.2	0.9	0.3	0.3	0.9	1.2	0.6
4	1.2	0.9	0.6	0.3	0.3	0.6	0.9	1.2
5	0.6	1.2	0.3	0.9	0.9	0.3	1.2	0.6
6	1.2	0.6	0.3	0.9	0.9	0.3	0.6	1.2

.

The conclusions drawn in the previous section, based on the analysis of the experimental data, were verified by comparing the experimental results with simulation outcomes. The last two inlet liquid arrangement schemes (Plan 5 and Plan 6) discussed in the simulation section further optimized the homogeneity of the internal flow (Figure 8a). As shown in Figure 8b, the liquid content ratios for these two scheduling schemes are close to 1, indicating that the internal low-velocity stagnation region has been largely eliminated. This demonstrates that the internal flow performance can be effectively optimized by appropriately adjusting the inlet liquid arrangement while maintaining a constant total fluid residence time. Additionally, Figure 8c illustrates that, despite keeping the total residence time unchanged, the residence time distributions for Plans 5 and 6 are more concentrated, resulting in a more uniform internal flow compared to the previous four inlet arrangement schemes. Further analysis reveals that, due to adjustments in the inlet velocity distribution, variations in the average residence time occur for each inlet, with the smallest difference in average residence time observed for scheme 6 (Plan 6), which demonstrates superior flow uniformity.

The analysis highlights the effect of different liquid inlet arrangements on the velocity distribution within a polar plate configuration. As shown in Figure 9a, both inlet schemes 1 and 2 exhibit high-velocity regions concentrated at the center of the polar plate. This phenomenon is attributed to an excessive intake of internal fluid, which enhances the primary flow but fails to diffuse effectively outward, leading to low-velocity stagnation zones around the periphery.

The velocity cloud plot in Figure 9b further supports this observation, revealing stagnation regions across all tested conditions from Plans 1 to 4. Notably, the stagnation regions created by Plans 1 and 2 are positioned opposite to those of Plans 3 and 4. Despite the differing locations, the underlying mechanisms leading to the formation of these stagnation zones remain consistent with the earlier analysis. In Figure 9c, the phenomenon of average velocity inhomogeneity is examined in relation to different liquid inlet methods. The results indicate that as the flow process progresses, the velocity inhomogeneity tends to decrease and stabilize. This suggests that the flow becomes more uniform over time. Further investigation in Figure 9d reveals that the velocity inequality values associated with Plans 5 and 6 are significantly lower and converge more effectively than those of the other four inlet arrangements. This finding highlights the effectiveness of these two schemes in optimizing flow uniformity, suggesting that they may be preferable choices in applications where even fluid distribution is critical.

4.4. Import Structure

Based on the results obtained, various structures of inlet homogenizers have been designed to improve internal stagnation regions and optimize the uniformity of mean residence time within the fluid. These structures are collectively referred to as IS (Import Structure). For the corresponding simulation conditions, the total inlet flow rate is controlled, ensuring that the flow rate is uniform across all eight inlets. The simulation results and their corresponding experimental validation are presented in Figure 10a. Once the flow state reaches a steady condition, the phase distribution cloud diagram within the polar plate aligns well with the experimental results. The total residence time distribution is depicted in Figure 10b, where the residence time distributions for IS5 and IS6 converge towards the center, indicating a reduction in the size of the low-speed stagnation region within the internal flow. Additionally, the average residence time distribution for each inlet in IS3 exhibits greater uniformity.

Figure 11a demonstrates that IS5 achieves the most uniform velocity distribution. The velocity cloud plots in Figure 11b indicate the absence of significant low-velocity stagnation regions within the polar plate when utilizing IS5 and IS6 structures, likely due to the implementation of two inverse curvature designs. Analysis of the particle average velocity map (Figure 11c) and the velocity inhomogeneity (Figure 11d) reveals that IS5 and IS6 exhibit lower inhomogeneity in the initial state, with velocity inhomogeneities for these structures being closely aligned and consistently at the lowest values across all areas of the polar plate. Based on the assessments of average residence time distribution and velocity inhomogeneity, it can be concluded that IS5 is the most effective design for optimizing internal flow in the combined case.

4.5. Internal Flow Field Components

Building on the experimental results obtained, we further optimize the flow within the internal area of the polar plate, excluding the inlet. An initial inflow-type flow field assembly is designed, and additional expanded assemblies are constructed based on this design. The experimental results are then compared, and the visual data are quantified to assess the effectiveness of the various flow field assemblies.

4.5.1. Vertical Structure Design (VBS) for Flow Field Assemblies

VBS1 serves as the initial structure, while VBS2 and VBS3 are modifications based on VBS1. As shown in Figure 12a, VBS1 exhibits the largest low-speed retention region, whereas VBS3, which is derived from VBS2, has the smallest low-speed retention region and a more concentrated residence time distribution (Figure 12b). Additionally, the average residence time distributions among the individual inlets of VBS3 show minimal variation.

The velocity distributions at the center cross-section of the three structural pole plates, along with the steady-state velocity cloud plots, are presented in Figure 8a. These indicate that VBS3 promotes a more uniform internal velocity distribution. Analysis of the mean velocity distribution (Figure 13b) and the velocity inhomogeneity (Figure 13c) confirms that VBS3 effectively restrains internal flow compared to the other two structures. However, the increased position inhomogeneity observed in VBS3 may result from flow mixing due to fluid bypassing the end of the vertical restraining zone after completing the flow in each interval. Nonetheless, the overall flow velocity inhomogeneity follows the trend VBS1 > VBS2 > VBS3, consistent with our quantitative experimental data.

Additionally, the area analysis data obtained during the experimental validation process generated plots that illustrate area growth trends. These plots vary the total flow rate while maintaining equal inlet feeds, as well as keeping the total flow rate constant while varying each inlet feed.

In the plots, the slope of the curve, denoted as ka, represents the growth rate of the tracer area within the total area of the polar plate flow field. The maximum value achieved by each curve indicates the highest area ratio that can be distributed at the conclusion of a complete experimental procedure. The tracer area ratio is theoretically expected to have a positive correlation with time in flow conditions absent of back mixing. Consequently, a smaller change in the slope of the curve in this experimental study suggests a reduced degree of back mixing and a more homogeneous internal flow. As illustrated in Figure 13d, VBS3 demonstrated a superior ability to accommodate high flow rates up to 400 L/h before experiencing a significant change in slope, compared to the other two structures. Additionally, Figure 13e shows that when the velocity distribution at the inlet is altered, VBS3 exhibits a higher optimization effect under the same conditions.

4.5.2. Horizontal Structure Design (HBS) of Flow Field Components

We propose a new intercepted flow field assembly, referred to as the HBS series, with HBS2 and HBS3 based on the structure of HBS1. As shown in Figure 14a, HBS2 exhibits the smallest velocity retention region. The statistical residence time distribution results presented in Figure 14b indicate that HBS3 has a more concentrated residence time distribution. Further analysis of the average residence time distributions for inlets 1–4 reveals that HBS2 optimizes the residence time distribution of internal fluid particles, as the residence time distributions for each inlet are nearly identical, as illustrated by the results of the two residence time measurements.

The velocity distribution at the center cross-section of the pole plate and the corresponding velocity cloud map are shown in Figure 15a. The velocity distributions of HBS1 and HBS2 at the center of the pole plate are more uniform, while the uneven velocity distribution of HBS3 is attributed to its blocking structure. Figure 15c demonstrates that the internal flow velocity under the HBS structure generally trends upward from bottom to top. Figure 15d, based on velocity inhomogeneity analysis, indicates that HBS2 exhibits the smallest inhomogeneity throughout the overall flow process, consistent with the findings from Figure 15c.

During the experimental validation, Figure 15d,e show that the HBS structure significantly differs from other flow conditions at low velocities, particularly when large flow strands concentrate at the internal inlet. The flow behavior closely resembles that of the inflow-type assembly. Moreover, the stagnation area is slightly smaller than that observed at the minimum flow rate, even though the slopes of the area growth curves for 280 L/h and 240 L/h are nearly identical. The morphological change in the flow trend area is more similar to that of VBS3.

5. Conclusions

This study establishes a systematic framework for optimizing hydrodynamic uniformity in industrial-scale Alkaline Water Electrolyzers (AWE), combining fluorescent tracer experiments with high-fidelity DDESs. Three critical design aspects were investigated to enhance electrolyte flow dynamics and mitigate stagnation/gas retention:

Increasing total electrolyte flow rates (240–360 L/h) substantially reduces velocity inhomogeneity (35–40%) and shrinks low-speed stagnation zones by enhancing forced convection. Beyond 420 L/h, marginal gains in flow uniformity occur due to turbulent energy saturation. Symmetric inlet scheduling (e.g., alternating high/low velocities in Plan 6) further refined flow allocation, lowering velocity disparities by 28% and balancing residence time distributions. These findings highlight the dual role of total flow magnitude and dynamic inlet adjustments in suppressing remixing and gas holdup.

Radially deflecting inlet homogenizers (IS5) and helical flow distributors (IS6) demonstrated superior performance, eliminating 95% of stagnation regions by optimizing momentum transfer. IS5 achieved a 22% reduction in residence time variance by redirecting inertial forces toward lateral zones, while IS6’s swirling flow enhanced gas detachment. Both designs reduced current density localization, offering a scalable strategy for industrial electrodes exceeding 2 m diameter.

Staggered baffle arrays in the VBS3 (Vertical Baffle Systems) configuration suppressed vertical recirculation, reducing velocity inhomogeneity by 45% through compartmentalized flow channels. The HBS2 (Horizontal Baffle Systems) design minimized lateral flow deviations, achieving near-uniform residence times (σ < 0.8 s) via segmented flow paths. Its symmetric blocking structures redirected stagnant fluid to active zones, improving gas–liquid separation efficiency.