Artificial neural network-genetic algorithm prediction of aluminum anode corrosion kinetics in high-concentration NaOH and KOH

1 Introduction

The global transition toward cleaner energy systems has positioned electric vehicles (EVs) at the forefront of sustainable transportation (Timilsina et al., 2024). As nations implement stricter emission regulations and invest in renewable energy infrastructure, the demand for efficient, cost-effective, and environmentally friendly energy storage and conversion technologies has accelerated (Muratori et al., 2025; Schulthoff et al., 2025; Ludin et al., 2025; Song et al., 2025). Lithium-ion batteries, which currently dominate the EV market, have achieved remarkable improvements in energy density and cost reduction (Koech et al., 2024). However, limitations related to the availability of critical raw materials, recycling complexity, and safety concerns under high-energy operation continue to restrict their scalability and sustainability (Pol, 2025; Zhang et al., 2025). This situation has stimulated research into alternative chemistries capable of complementing or extending the range of conventional EV batteries (Shabeer et al., 2025a). Among these emerging technologies, metal-air systems have gained increasing attention for their potential to deliver high energy density while maintaining simple and lightweight configurations (Qahtan et al., 2025).

Aluminum-air (Al-air) batteries represent a promising direction for the next-generation of energy systems due to aluminum’s natural abundance, low cost, and inherent recyclability (Dilshad et al., 2025). Aluminum possesses one of the highest theoretical gravimetric energy densities among metals (8.1 kWh kg⁻¹), making it an attractive anode material for vehicular energy applications (Shabeer et al., 2025b). The reaction of aluminum with oxygen in an alkaline electrolyte produces electrical energy with water and aluminum hydroxide as benign by-products, reinforcing its environmental appeal (Tan et al., 2025). These characteristics make Al-air systems highly suitable as potential range extenders for electric vehicles, where secondary Li-ion packs may be complemented by lightweight and high-energy aluminum-based cells to achieve longer driving distances with minimal additional mass (Wan et al., 2025). Furthermore, aluminum’s established industrial supply chain and recycling infrastructure provide additional advantages over more resource-limited chemistries, supporting the circular economic goals of modern transportation systems (Pahlevan et al., 2021; Tian et al., 2022).

Despite these benefits, widespread commercialization of Al-air batteries has been constrained by one critical issue: the corrosion of aluminum in alkaline electrolytes (Obot et al., 2025). Spontaneous parasitic corrosion and hydrogen evolution during open-circuit and discharge conditions lead to the loss of active material, reduced anode utilization, and gas accumulation that degrades system reliability (Li et al., 2025). These challenges are particularly severe in highly concentrated alkaline electrolytes, such as sodium hydroxide (NaOH) and potassium hydroxide (KOH), which are commonly used due to their ability to enhance ionic conductivity and dissolution kinetics (Ma et al., 2025; Wen et al., 2020a; Hu et al., 2019; Wen et al., 2020b). Therefore, a fundamental understanding of aluminum anode corrosion behavior in alkaline environments is essential for improving material utilization and guiding electrolyte selection. It should be noted that the present study focuses on corrosion behavior under open-circuit conditions, and direct correlations to discharge performance in Al-air systems are beyond the scope of this work. The corrosion behavior of aluminum is complex, influenced by alloy composition, surface passivation, electrolyte concentration, and temperature. Besides, aluminum’s corrosion behavior in alkaline electrolytes is strongly dependent on the OH⁻ concentration, resulting in distinct kinetic regimes characterized by film stability and dissolution processes. In weaker alkaline solutions, the corrosion process is dominated by the hydrolysis and subsequent dissolution of the passivating oxide layer (Wu et al., 2020; Zhu et al., 2024).

• Film Transformation and Penetration: The initial Al₂O₃ layer undergoes hydrolysis to form more porous and hydrated species, such as AlOOH or Al(OH)₃. Water molecules and OH⁻ ions then penetrate this hydrated film to initiate corrosion at the Al/film interface.
• Dissolution and Rate Control: Simultaneously, the hydroxide precipitate Al(OH)₃ at the film/electrolyte interface dissolves through complexation with OH⁻ ions (Equation 1):

A l O H 3 + O H − → A l ( O H 4 − (1)

The dissolution of the foundational Al₂O₃film also yields the soluble tetrahydroxoaluminate ion (Al ( O H 4 −) (Equation 2):

A l 2 O 3 + O H − + H 2 O → A l ( O H 4 − (2)

In this regime, the diffusion of OH⁻ ions toward the surface and the Al ( O H 4 − product away from the interface is generally considered the rate-controlling step. It is important to note that the direct ejection of Al³⁺ into the alkaline solution is thermodynamically prohibited. In highly concentrated alkaline electrolytes, the aggressive environment leads to the complete breakdown of the protective mechanism, resulting in active corrosion.

• Film Breakdown and Direct Attack: The excessive concentration of OH⁻ causes the complete dissolution of any surface oxide or hydroxide film. Consequently, OH⁻ ions directly attack the bare aluminum surface, which is the primary corrosion reaction in these highly caustic environments:

A l + 4 O H − → A l ( O H 4 − (3)

• Kinetic Dominance: Due to the high concentration of OH⁻ and the high solubility of the product, the influence of ion diffusion on the corrosion rate becomes negligible. The corrosion kinetics are instead governed by the rate of the surface chemical reaction (Equation 3). This rapid, film-free mechanism is directly responsible for the significant self-discharge and material consumption observed in Al anodes used in Al-air batteries under conditions of excessive electrolyte alkalinity (RSC Publishing, 2025a).

Among numerous aluminum systems, a commercial Al-Mg-Si alloy is frequently selected due to its favorable balance of mechanical robustness, corrosion resistance, and thermal transport properties (RSC Publishing, 2025b). Its controlled additions of magnesium and silicon enable a stable microstructure with good strength-to-weight characteristics, while maintaining relatively high thermal conductivity compared to higher-alloyed aluminum systems (Buckingham et al., 2021). This combination supports efficient heat dissipation and mitigates localized thermal gradients during operation, which is advantageous for electrochemical environments where coupled thermal and corrosion effects can influence performance and degradation (Sitanggang et al., 2025). In contrast to higher-strength aluminum alloys with elevated alloying content, Al-Mg-Si compositions exhibit improved processability and reduced susceptibility to stress-assisted corrosion and cracking. These attributes, together with their widespread industrial use, reinforce the suitability of commercial Al-Mg-Si alloys for electrochemical applications that prioritize durability, manufacturability, and materials circularity (Elboughdiri et al., 2025).

Understanding and controlling the corrosion behavior of a commercial Al-Mg-Si alloy in alkaline environments is essential for advancing Al-air battery and other aluminum-related electrochemical systems. Thus, extensive research on Al-Mg-Si alloy corrosion needs to be systematically evaluated (Rani et al., 2024). Previous corrosion modeling studies have demonstrated that electrochemical and environmental descriptors, including electrolyte concentration, pH, applied potential, and current density, exert strong influence on corrosion kinetics and interfacial stability due to their direct control over oxide-film dissolution, charge-transfer processes, and hydrogen evolution reactions. Recent interpretable machine-learning approaches, including feature-importance analysis, SHAP-based explainability, and correlation mapping, have further confirmed that electrolyte-related variables are among the dominant predictors governing corrosion potential and corrosion current density across metallic systems. These findings highlight the importance of incorporating physically meaningful electrochemical descriptors into data-driven corrosion models to accurately capture the nonlinear interactions governing degradation behavior in alkaline environments (Xiong et al., 2024; Song et al., 2023; Wang et al., 2025a; Wang et al., 2025b).

Existing studies often employ limited or inconsistent electrolyte concentrations, making it difficult to establish a clear relationship between electrolyte strength and corrosion activity. Comparative evaluation in NaOH and KOH solutions is particularly valuable, as these two electrolytes differ in ionic characteristics, solubility, and aggressiveness toward passive film dissolution (Che et al., 2025; RSC Publishing, 2026; Liu et al., 2017). A structured investigation across a wide concentration range is therefore needed to generate a coherent understanding of electrolyte-dependent corrosion mechanisms and to guide the selection of optimal electrolyte formulations that minimize parasitic corrosion while maintaining electrochemical efficiency. Moreover, accurately predicting corrosion parameters such as corrosion potential (Icorr) and current density (Icorr) remain challenging because of the nonlinear and coupled effects of electrolyte composition, temperature, and surface state. These complexities highlight the need for data-driven approaches capable of modeling multivariate nonlinear systems beyond the capability of conventional empirical or regression-based models.

Artificial neural networks (ANNs) have shown increasing promise in electrochemical modeling, materials characterization, and corrosion prediction due to their ability to learn complex relationships directly from data. When combined with optimization techniques such as scalar genetic algorithms (GA) and non-dominated sorting genetic algorithm (NSGA-II), ANN models can be further tuned to achieve superior prediction accuracy and generalization performance (Madhavan et al., 2024). Previous studies have demonstrated the potential of ANN models for corrosion prediction in several alloy systems (Amodu et al., 2022; Coelho et al., 2022); however, no systematic effort has yet focused on modeling the corrosion kinetics of aluminum alloys in alkaline environments. Considering the growing global interest in aluminum based electrochemical technologies and the extensive datasets generated from electrochemical testing, integrating ANN-GA frameworks into corrosion studies provides a powerful avenue to predict kinetic parameters, optimize electrolyte composition, and support future digital-twin models for metal-air battery design and control.

This study systematically investigated the corrosion of a commercial Al-Mg-Si alloy in 1 M–10 M NaOH and KOH electrolytes using an integrated experimental and computational approach. The key aspects of the work are summarized below:

• Experimental Techniques and Structural Analysis: Electrochemical measurements, including open-circuit voltage, potentiodynamic polarization, and electrochemical impedance spectroscopy, were employed to determine corrosion potential and corrosion current density across varying electrolyte concentrations. Complementary phase identification of corrosion products was conducted using X-ray diffraction (XRD) to assess structural evolution during alkaline exposure.
• Comparative Analysis: The work provides a detailed comparative understanding of Al corrosion mechanisms between NaOH and KOH electrolytes.
• Predictive Modeling: An ANN model, optimized via scalar GA and NSGA-II, was developed and trained on experimental data to predict corrosion potential and current density as a function of electrolyte parameters.
• Novelty and Impact: This work is the first to couple systematic alkaline corrosion experiments of a commercial Al-Mg-Si anode across a wide concentration range with ANN-assisted genetic and multi-objective optimization, enabling quantitative identification of corrosion-optimal electrolyte conditions rather than empirical trial-and-error selection.

2 Methods and modeling

2.1 Experimental methods

2.1.1 Materials

The working electrode material consists of samples of a commercial Al-Mg-Si alloy, machined to provide an exposed surface area of 0.79 cm² for electrochemical testing. The bulk alloy composition is summarized in Table 1. Prior to each experiment, the samples are mechanically sanded to remove surface impurities and native oxide layers, followed by cleaning with ethanol and air-drying at room temperature to ensure a reproducible and contaminant-free surface, as shown in Figure 1. Analytical-grade sodium hydroxide (NaOH, Sigma-Aldrich) and potassium hydroxide (KOH, Sigma-Aldrich) are used to prepare alkaline electrolytes in concentrations ranging from 1 M to 10 M using deionized water. All electrochemical tests are conducted at 25 °C.

TABLE 1. Chemical composition of the commercial Al-Mg-Si alloy used in this study.

2.1.2 Cell setup

Electrochemical characterization is conducted using a conventional three-electrode corrosion test cell (Figure 1). The alloy sample functions as the working electrode, a carbon rod is used as the counter electrode, and an Ag/AgCl electrode serves as the reference. All potentials are measured relative to VAg/AgCl. Tests are performed under quiescent conditions to ensure uniformity across measurements. All electrochemical measurements were repeated at least three times to ensure reproducibility, and consistent trends were observed across independent experiments.

2.1.3 Procedure

Prior to each experiment, the working electrode is immersed in the electrolyte and allowed to stabilize under open circuit voltage (OCV) for 15 min. The OCV stabilization period was limited to 15 min to establish consistent interfacial condition while minimizing prolonged exposure to highly aggressive alkaline environments, which may induce continuous surface dissolution and alter subsequent polarization and impedance responses. This stabilization period removes transient oxide films formed during sample preparation and ensures potential equilibrium. Measurements are carried out in NaOH and KOH electrolytes across concentrations ranging from 1 M to 10 M. Although the native oxide layer was removed during surface preparation, reformation of an oxide/hydroxide layer is expected during the OCV stabilization period prior to electrochemical measurements.

Potentiodynamic polarization tests were performed over a potential range of −0.5 V to +0.5 V vs. Ag/AgCl at a scan rate of 10 mV s^-1 referenced to the Ag/AgCl electrode. The corrosion potential (Ecorr) and corrosion current density (Icorr) were estimated from the polarization curves. It is noted that the polarization scans were conducted as continuous sweeps across both cathodic and anodic regions for consistency; therefore, the extracted Ecorr and Icorr values are used for comparative analysis rather than strict Tafel-based kinetic determination. Under sufficiently high overpotential conditions, the polarization response may be approximated using Tafel behavior for trend interpretation. Polarization curves are presented in semi-logarithmic form (log current density vs. potential) to enable extraction of corrosion parameters.

Electrochemical impedance spectroscopy (EIS) measurements were carried out using a 10 mV AC perturbation over a frequency range of 10⁵ to 10⁻¹ Hz with 10 points per decade. The impedance response was analyzed using an equivalent circuit consisting of solution resistance (Rs) in series with a parallel combination of charge-transfer resistance (Rct) and a constant phase element (CPE), representing non-ideal double-layer behavior. The extracted parameters provide insight into interfacial processes, corrosion resistance, and surface heterogeneity of the alloy in alkaline electrolytes. Impedance values are reported in Ω·cm², normalized to the electrode surface area.

2.1.4 X-ray diffraction

X-ray diffraction (XRD) is performed on post-corrosion samples to identify crystalline corrosion products formed during immersion in NaOH and KOH. Diffraction patterns allow detection of aluminum hydroxides, oxide phases, and other species that form because of alloy dissolution and surface film transformation in alkaline environments. Comparing XRD profiles across electrolyte concentrations provides insight into differences in film stability and phase composition between NaOH and KOH systems.

2.2 Modeling methodology

The machine-learning model was developed using a curated dataset comprising electrochemical measurements collected across a range of alkaline electrolyte conditions (Section 2.1.2). Each entry in the dataset represents a unique electrolyte configuration and its corresponding polarization response, encoded through five descriptive input features: electrolyte type (encoded as KOH = 1, NaOH = 2)), electrolyte concentration (M), solution temperature (°C), applied potential (V), and measured current density (A cm^-2). These variables capture the physicochemical environment governing anodic dissolution and serve as the predictors in the ANN framework. The associated outputs consist of the experimentally derived Ecorr and Icorr, which are key indicators of alloy degradation and performance in alkaline media. The dataset was refined through consistency checks and removal of incomplete entries, resulting in a final set suitable for supervised learning. Predicting Ecorr and Icorr directly from easily measurable electrochemical inputs provides an efficient pathway for rapid corrosion assessment and supports the development of optimization strategies for alkaline aluminum-based energy systems. The overview of the modeling methodology is represented in Figure 2. All mathematical operations associated with data preprocessing, ANN training, optimization, and model evaluation are summarized in Table 2 to consolidate the computational framework used throughout this study (Mehdizadeh Chellehbari et al., 2025; Chepino et al., 2023; Kunstner et al., 2024; Liu et al., 2025; Chicco et al., 2021; Zhang et al., 2024; Pereira et al., 2021; Giovannelli et al., 2025; He et al., 2024; Wu et al., 2025a; Wu et al., 2025b; Wu et al., 2025c).

TABLE 2. Summary of all equations used for data normalization, ANN modeling, hyperparameter optimization, GA search, and performance evaluation.

2.2.1 Artificial neural network structure and data

ANNs are data-driven computational frameworks designed to learn nonlinear relationships from experimental observations. They consist of interconnected processing units (neurons) arranged in input, hidden, and output layers, where each connection is associated with a trainable weight. During model training, these weights are iteratively updated to minimize prediction error, allowing the network to identify complex patterns within the data (Xia et al., 2012). Nonlinearity within the network is introduced through activation functions, with the rectified linear unit (ReLU, Equation 10), being one of the most commonly used due to its computational efficiency and its ability to mitigate vanishing-gradient issues in deeper architectures (Salem et al., 2022). During forward propagation, the ANN maps its inputs to predicted outputs through the functional relationship defined in Equation 11. Model optimization is performed using the Adam algorithm, an adaptive gradient-based optimizer that adjusts the learning rate for each weight by estimating the first (Equation 12) and second moments (Equation 13) of the gradient (Wang et al., 2022). This results in stable and rapid convergence, even in high-dimensional and nonlinear datasets such as those associated with corrosion behavior. Several hyperparameters, including learning rate, number of hidden layers, number of neurons, number of epochs, β₁ and β₂ decay coefficients, and loss function, are tuned through systematic search.

2.2.2 Grid-based hyperparameter optimization

To identify the most effective ANN configuration, a structured grid-based hyperparameter search was performed. Three model attributes were systematically varied: the number of hidden layers (1, 2, or 3), the number of neurons per layer (20, 25, or 30), and the Adam optimizer parameters, including three learning rates (0.00100, 0.00085, and 0.00070) and a range of β₁ values. The β₁ parameter was found to strongly influence convergence behavior, with β₁ = 0.92 consistently yielding the highest predictive accuracy.

2.2.3 Data normalization and performance assessment

During preprocessing, all input features are scaled using min-max normalization to prevent variables with larger magnitudes from dominating the learning process (Equation 9). This method rescales each feature to a dimensionless range between 0 and 1, improving numerical stability and convergence during model training (Shantal et al., 2025). Model performance is subsequently assessed using four standard regression metrics that quantify the accuracy of predicted corrosion parameters relative to experimentally measured values. These include mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE) and the coefficient of determination (R ²) shown in Equations 16–19 respectively. Lower MAE, MSE, and RMSE values indicate closer agreement between predicted and experimental corrosion parameters, while R ² values approaching unity reflect high predictive capability and strong model generalization (Hodson, 2022; Wang and Lu, 2018).

2.2.4 Dataset visualization and input-output interactions

To assess the structure of the experimental dataset and evaluate potential nonlinear relationships among variables, a pairwise scatter and kernel-density matrix was generated for all key electrochemical parameters (Figure 3). The matrix includes electrolyte concentration, corrosion potential (Ecorr), corrosion current density (Icorr), applied voltage, and log-transformed current density. Scatter plots along the lower triangle illustrate how the distributions of Ecorr and Icorr shift between NaOH and KOH, while the diagonal kernel-density curves highlight differences in the statistical behavior of each variable. The scatter plots also reveal concentration-dependent clustering and asymmetric relationships, especially the strong electrolyte-dependent separation in both Ecorr and Icorr. This visualization provides an integrated overview of how experimental variables covary and demonstrates the presence of distinct nonlinear trends, supporting the motivation for developing an ANN-based predictive model.

2.2.5 Genetic algorithm optimization

A genetic algorithm (GA) was implemented to identify an optimal electrolyte condition that simultaneously promotes a high Ecorr while minimizing the Icorr. Because standard GAs are designed for single-objective searches (Lambora et al., 2019), the two corrosion objectives were incorporated into a scalarized fitness function such that a single cost value could represent the combined corrosion response predicted by the trained ANN model. After ANN training, the model function (Equation 11) was used to estimate Ecorr and Icorr from candidate input vectors representing electrolyte type and concentration, with the remaining operational variables (temperature, applied potential, and instantaneous current density) set to representative operating values.

To embed the two corrosion objectives into a single scalar score, Ecorr and Icorr were first normalized to the range [0,1] using min-max scaling (Equation 9). To maintain a consistent minimization direction, Ecorr was inverted such that higher Ecorr values corresponded to lower normalized cost, while Icorr was used directly since lower Icorr values naturally indicate reduced corrosion. The scalar objective function for a candidate electrolyte condition was expressed as a weighted sum in Equation 24, where w I = w E = 0.5 reflect an equal preference for reducing corrosion rate and increasing corrosion potential. This formulation transforms the bi-objective corrosion problem into a single-objective optimization that can be solved using standard GA operations (selection, crossover, and mutation; Equations 21–23). A continuous search space was defined for concentration (1–10 M) and electrolyte type, which was discretized during evaluation to maintain its categorical nature (KOH = 1, NaOH = 2). The GA iteratively refined the population until convergence, and the candidate with the lowest objective value was designated as the scalar GA optimum. This optimum represents the best solution according to the chosen weights, rather than the full Pareto trade-off structure of the corrosion system.

2.2.6 Multi-objective optimization using NSGA-II

To obtain a weight-free and physically interpretable characterization of the trade-off between Ecorr and Icorr, a non-dominated sorting genetic algorithm (NSGA-II) was implemented. Unlike scalar GA, NSGA-II preserves the independence of both corrosion objectives and identifies the complete Pareto front, defined as the set of electrolyte conditions for which no further improvement in one objective is possible without degrading the other (Al-Majali and Zobaa, 2025). The two objectives were defined as (Equation 25).

• Objective 1: minimize Icorr
• Objective 2: shift Ecorr toward less negative values to reduce corrosion tendency under open-circuit conditions
• To maintain a uniform minimization framework, the second objective was evaluated through its inverse, where larger Ecorr corresponds to lower objective value. For each candidate solution, the ANN model predicted Ecorr and Icorr, from which the objective pair was obtained. Pareto dominance between solutions followed the standard NSGA-II criterion (Equation 26), and the formation of successive fronts was determined using fast non-dominated sorting (Equation 27).

The NSGA-II algorithm proceeded through (Al-Majali and Zobaa, 2025; Ver et al., 2021).

• Fast non-dominated sorting, which partitions the population into hierarchical fronts based on Pareto dominance (Equation 27).
• Crowding-distance assignment, which preserves diversity along the front (Equation 28).
• Tournament selection, crossover, and polynomial mutation to generate the next population, with selection comparing solutions by Pareto rank and crowding distance (Equation 29).
• Elitist replacement, ensuring that high-quality Pareto solutions are retained across generations.

The final NSGA-II front represents the complete corrosion trade-off landscape across electrolyte type and concentration. From this front, a knee point was identified using the minimum-distance-to-ideal method (Equation 30), where the ideal point corresponds to the maximum Ecorr and minimum Icorr present on the Pareto front. The knee point therefore represents the least-compromise solution, where simultaneous improvements in Ecorr and Icorr become mutually limiting due to the curvature of the trade-off surface.

3 Results and discussion

The electrochemical behaviour of the commercial Al alloy across the full 1–10 M concentration range in KOH and NaOH electrolytes was evaluated using OCV, potentiodynamic polarization, and EIS. To present these data without excessive crowding and to preserve clarity in trend interpretation, the results are organized into four concentration windows: 1–5 M KOH, 6–10 M KOH, 1–5 M NaOH, and 6–10 M NaOH. Each window consolidates the complete electrochemical response, including OCV evolution, Tafel behaviour, Ecorr and Icorr values, estimated anode lifetime, and impedance characteristics. This structured presentation enables direct comparison within each electrolyte across its full molarity range, followed by cross-electrolyte interpretation that highlights mechanistic differences between KOH and NaOH.

3.1 Experimental corrosion trends

Across both electrolytes and the full concentration range, the open-circuit voltage profiles exhibit a consistent two-stage behavior, characterized by an initially rapid shift toward more positive potential followed by stabilization to a quasi-steady plateau within the first 200–400 s of immersion. This early-time evolution reflects rapid interfacial adjustment upon exposure to alkaline media, during which the native surface film is chemically modified, and the mixed potential evolve as anodic aluminum dissolution and cathodic hydrogen evolution approach a steady balance.

For KOH in the low-to-moderate concentration range (1–5 M, Figure 4a), the stabilized OCV displays a clear but non-monotonic dependence on concentration. The most positive steady potential is observed at 2 M, while both lower (1 M) and higher concentrations (4–5 M) stabilize at more negative values. This behavior suggests that intermediate hydroxide concentrations may temporarily favor partial surface stabilization, whereas higher concentrations promote more aggressive dissolution of surface films and a shift toward a more active corrosion state (Moghadam et al., 2017). In the higher concentration KOH regime (6–10 M, Figure 5a), the OCV curves converge closely and stabilize at similarly negative plateau values, indicating that once a sufficiently high hydroxide activity is reached, the surface rapidly transitions to a comparable actively corroding steady state across concentrations, with further increases in molarity exerting a limited influence on the final mixed potential. All curves exhibit an upward drift in potential as the native Al₂O₃/AlOOH surface layer gradually dissolves and the interface transitions toward an actively corroding state. This monotonic shift indicates that higher OH⁻ activity accelerates removal of the initial, thereby promoting earlier exposure of metallic Al and facilitating the onset of anodic dissolution (Boukerche et al., 2014).

In contrast, NaOH exhibits a more systematic concentration dependence across both regimes. In the 1–5 M range (Figure 6a), the stabilized OCV shifts progressively toward more negative values with increasing concentration, indicating a steady enhancement in corrosive activity as hydroxide concentration increases. This monotonic trend continues in the high-concentration NaOH window (6–10 M, Figure 7a), where the most negative steady potentials are consistently observed at the highest molarities. Compared with KOH at equivalent concentrations, NaOH generally stabilizes at more negative OCV values, particularly in the high-concentration regime, implying differences in the steady-state balance of anodic and cathodic reactions and in the nature or persistence of surface hydroxide films formed in the two electrolytes. Overall, the OCV behavior reveals that while both electrolytes promote rapid activation of the aluminum surface, KOH shows convergence toward a concentration-independent active state at high molarity, whereas NaOH maintains a stronger concentration sensitivity across the full range studied.

The potentiodynamic polarization curves obtained for the commercial Al alloy in 1–5 M KOH (Figure 4b) exhibit systematic concentration-dependent changes in both anodic and cathodic branches, reflecting progressive activation of aluminum dissolution with increasing hydroxide concentration. As the KOH concentration increases from one to 5 M, the corrosion potential shifts toward more negative values (Figure 4c), indicating enhanced thermodynamic driving force for anodic dissolution. Similar shifts in Ecorr with increasing KOH concentration have been reported for aluminum and Al-Mg-Si alloys and are commonly attributed to accelerated chemical dissolution of the native Al₂O₃/AlOOH surface film under high OH⁻ activity (Reena Kumari et al., 2016).

The corrosion current density increases markedly with concentration up to 4 M KOH (Figure 4d), confirming a transition from partially inhibited dissolution at low concentration to a fully active corrosion regime at intermediate concentrations. This increase in Icorr is consistent with previous studies showing that elevated KOH concentrations enhance ionic conductivity, reduce charge-transfer resistance, and promote faster anodic aluminum oxidation coupled with hydrogen evolution (Musicco et al., 2025; Porciúncula et al., 2012). When interpreting the KOH concentration dependence, it is also important to separate “OH⁻ availability” from interfacial water activity and double-layer effects. Cation-controlled solvation and electric double-layer composition can modulate the activity of “free water” participating in anodic corrosion pathways and can promote formation of more corrosion-resistant surface compounds, which directly affects the apparent activation observed in polarization responses (Chung et al., 2025). The slight reduction in Icorr observed at 5 M relative to 4 M suggests the onset of secondary effects, such as accumulation of soluble aluminate species or partial surface blockage by corrosion products, phenomena that have also been observed at high alkalinity in earlier investigations (Harchegani, 2023; Buryakovskaya and Vlaskin, 2022).

The coupling between Ecorr and Icorr shown in Figure 4f indicates that anodic activation dominates the corrosion kinetics in this concentration window (Xiao et al., 2022). Comparable Ecorr-Icorr correlations have been reported in alkaline corrosion studies, where increasingly negative corrosion potentials coincide with accelerated uniform aluminum dissolution rather than localized attack (Correa et al., 2014; Wei et al., 2016; Nyby et al., 2021).

In the 6–10 M KOH concentration window, the electrochemical behavior shifts into a regime where aggressive alkaline dissolution is governed by mass-transport and saturation-limited kinetics rather than pure charge transfer. While the rapid stabilization of Ecorr at strongly negative values in Figure 5c indicates the immediate removal of the native oxide layer, the uniformly high Icorr with modest variation observed in Figure 5d suggests that the reaction has reached a kinetic limitation (Wang Z. et al., 2025). This behavior is driven by the accumulation of soluble aluminate species A l O H 4 −, coupled with increased electrolyte viscosity at higher concentrations which in-turn leading to the increased thickness of the Nernst boundary layer (δ ≈ 10–100 μm), thereby hindering the diffusion of fresh hydroxide ions to the metal surface. Simultaneously, intense hydrogen evolution leads to significant bubble adsorption, with an estimated surface coverage of 20%–50%, inducing gas evolution impedance and physically blocking the active surface area (Shao et al., 2004; Chen et al., 2015; Niverty et al., 2023). This transition from anodic activation to transport control is further evidenced by the weak Ecorr-Icorr coupling in Figure 5f, confirming that the reaction rate is throttled by the physical movement of species and interfacial blockage once a critical hydroxide threshold is exceeded.

The polarization behavior of aluminum in 1–5 M NaOH (Figure 6b) exhibits qualitatively similar trends to KOH but with clear quantitative differences. Increasing NaOH concentration drives Ecorr to more negative values (Figure 6c), reflecting progressive destabilization of the surface oxide film. However, at equivalent molarities, Ecorr values in NaOH remain consistently less negative than those measured in KOH, indicating weaker anodic activation. The corrosion current density shows a non-monotonic dependence on concentration (Figure 6d), with a pronounced increase at intermediate concentration followed by partial suppression at higher molarity. This behavior can be attributed to competing processes, including oxide dissolution, surface reorganization, and partial re-passivation at elevated hydroxide concentrations. The longer estimated lifetimes observed at low NaOH concentrations (Figure 6e) further support the interpretation that NaOH provides a comparatively milder corrosion environment than KOH in the low-to-moderate concentration regime. The broader scatter in the Ecorr-Icorr relationship (Figure 6f) suggests greater sensitivity of corrosion kinetics to local surface conditions, indicating less stable interfacial behavior in NaOH (Porciúncula et al., 2012).

At higher NaOH concentrations (6–10 M), the polarization curves (Figure 7b) indicate a transition toward more active corrosion behavior. Ecorr becomes increasingly negative with concentration (Figure 7c), although the magnitude of this shift remains smaller than that observed in KOH at comparable molarities. This observation indicates that NaOH requires higher concentrations to achieve corrosion rates comparable to KOH due to differences in cation size, hydration, and interfacial transport properties (Das et al., 2020). The corrosion current density increases substantially at 9–10 M NaOH (Figure 7d), indicating that sufficiently high hydroxide activity can overcome surface stabilization effects and promote rapid aluminum dissolution. However, the broader Ecorr-Icorr distribution (Figure 7f) suggests less stable electrochemical behavior relative to KOH, a trend previously attributed to weaker disruption of surface passivation and less efficient charge transport in Na⁺-based electrolytes (Fadhilah et al., 2025; Li et al., 2020).

Across the full concentration range, KOH consistently produces more negative corrosion potentials and higher corrosion current densities than NaOH at equivalent molarities (Table 3). This behavior cannot be explained solely by hydroxide concentration, as both electrolytes supply identical OH⁻ activity at the same molarity. Instead, the observed differences are attributed to cation-specific effects, where the larger ionic radius and higher mobility of K⁺ facilitate faster charge transport, enhanced interfacial reactions, and accelerated hydrogen evolution kinetics (Xhanari et al., 2019). K⁺ ions exhibit lower hydration energy and higher mobility compared to Na⁺, which can influence interfacial transport and double-layer properties, thereby affecting corrosion kinetics and hydrogen evolution behavior. The present results extend these observations across a broader concentration range and provide a unified electrochemical interpretation linking electrolyte identity to corrosion kinetics through combined Ecorr-Icorr analysis. However, these effects are inferred based on electrochemical trends, and direct experimental isolation of cation-specific contributions was not performed in the present study.

TABLE 3. Comparison of corrosion potential (Ecorr) and corrosion current density (Icorr) of the commercial Al-Mg-Si alloy in KOH and NaOH electrolytes over the concentration range of 1–10 M.

3.2 Mechanistic interpretation from electrochemical impedance

Electrochemical impedance spectroscopy was employed to elucidate the interfacial mechanisms governing alkaline corrosion of the commercial Al alloy and to complement the trends observed from OCV and potentiodynamic polarization measurements. The impedance response across all electrolyte concentrations and both alkaline media is well represented by an equivalent circuit consisting of the solution resistance in series with a parallel combination of charge-transfer resistance and a constant phase element representing the double-layer capacitance (Figure 4g inset). The same circuit was used for all cases (Rs​ + (Rct​∥CPEdl​)). This circuit captures the non-ideal capacitive behavior associated with surface roughness, evolving oxide films, and gas evolution at the aluminum-electrolyte interface. Given the complexity of aluminum corrosion in alkaline media, alternative interpretations involving mixed kinetic and film-controlled processes are also possible.

Table 4 shows the values of the fitted electrochemical impedance parameters for the commercial Al alloy in KOH and NaOH electrolytes over the concentration range of 1–10 M. Across the full concentration range, the impedance spectra exhibit depressed semicircles in the Nyquist plots and broad phase-angle maxima in the Bode phase plots, indicating significant deviation from ideal capacitive behavior. This non-ideality is quantified by the CPEdl-P exponent, which remains consistently below unity for both electrolytes, confirming a heterogeneous and dynamically evolving interface. Such behavior is characteristic of aluminum corrosion in strongly alkaline environments, where continuous oxide dissolution, aluminate formation, and hydrogen evolution lead to spatially and temporally varying electrochemical conditions at the surface (Fabris et al., 2024; Khan et al., 2022). The charge-transfer resistance decreases markedly with increasing hydroxide concentration in both KOH and NaOH, consistent with the progressive activation of anodic aluminum dissolution inferred from polarization measurements. In KOH, Rct decreases rapidly from high values at low concentration to very low values at moderate-to-high molarity, reflecting the strong ability of KOH to destabilize the native Al₂O₃/AlOOH surface layer and promote facile charge transfer. This trend aligns with the increasingly negative corrosion potentials and elevated corrosion current densities observed in the polarization data, confirming that anodic activation dominates the corrosion kinetics over a broad concentration range (Gelman et al., 2015). At higher alkaline concentrations, particularly in KOH, the interpretation of Rct requires careful consideration. Extremely low fitted Rct values do not necessarily imply an absence of interfacial resistance but rather indicate that classical charge-transfer resistance becomes masked by other impedance contributions. The impedance response reflects combined contributions from charge-transfer processes and the dynamically evolving surface oxide/hydroxide layer. Under these conditions, vigorous hydrogen evolution, accumulation of soluble aluminate species, and increased electrolyte viscosity contribute to gas bubble coverage, interfacial blockage, and transport limitations. These effects distort the impedance response and reduce the apparent semicircle diameter, leading to artificially low Rct values when interpreted using conventional equivalent circuit models. The persistence of non-ideal CPE behavior and the flattening of phase-angle responses at low frequencies further support a transition from pure charge-transfer control to a regime dominated by mass transport and interfacial obstruction (El-sayed et al., 2024). It should be noted that the impedance response may arise from coupled charge-transfer and surface film processes, and alternative interpretations of the observed time constants are possible.

TABLE 4. Fitted electrochemical impedance parameters for the commercial Al alloy in KOH and NaOH electrolytes over the concentration range of 1–10 M.

Rs denotes solution resistance, Rct represents charge-transfer resistance, and the double-layer capacitance is represented using a constant phase element (CPEdl), where T is the magnitude parameter and P is the phase exponent.

The magnitude parameter of the constant phase element, CPEdl-T, provides additional insight into interfacial evolution. Variations in CPEdl-T across concentration reflect changes in effective double-layer capacitance, which is influenced by surface roughness, oxide film thickness, and hydrogen bubble coverage. In general, higher CPEdl-T values correspond to increased electrochemically active surface area and thinning of surface films, consistent with enhanced anodic dissolution at elevated hydroxide concentrations. However, pronounced fluctuations in CPEdl-T at high molarity suggest that competing processes, including surface roughening and partial blockage by corrosion products or gas bubbles, act simultaneously to modulate the interfacial capacitance (Zhao et al., 2024). Comparison between KOH and NaOH reveals systematic electrolyte-dependent differences in impedance behavior that cannot be attributed solely to hydroxide concentration. At equivalent molarities, NaOH consistently exhibits higher charge-transfer resistance than KOH, indicating reduced anodic activation and slower corrosion kinetics. This finding corroborates the less negative corrosion potentials and lower corrosion current densities measured in NaOH, particularly at low-to-moderate concentrations. The CPEdl-P values in NaOH also display greater variability and, in some cases, approach or exceed unity, suggesting a less stable and more heterogeneous interfacial structure compared with KOH. Such behavior is consistent with weaker disruption of surface passivation and less efficient charge transport in Na⁺-based electrolytes (Chung et al., 2025; El-sayed et al., 2024). These electrolyte-specific impedance trends highlight the critical role of cation identity in governing alkaline aluminum corrosion. Although both KOH and NaOH provide identical hydroxide activity at a given molarity, the larger ionic radius, lower hydration energy, and higher mobility of K⁺ facilitate enhanced interfacial transport, stronger double-layer perturbation, and more aggressive hydrogen evolution. As a result, KOH promotes lower Rct, more negative corrosion potentials, and higher corrosion currents than NaOH across the full concentration range investigated. The impedance results therefore reinforce the conclusion drawn from polarization and OCV analyses that corrosion kinetics in alkaline aluminum systems are governed by coupled hydroxide activity and cation-mediated interfacial effects. Overall, the EIS analysis provides mechanistic confirmation of the corrosion regimes inferred from DC electrochemical measurements. At low-to-moderate concentrations, corrosion is dominated by anodic activation and charge-transfer control, whereas at high alkalinity, transport limitations, surface saturation, and gas evolution increasingly govern the electrochemical response. The consistency between impedance-derived parameters and polarization behavior demonstrates the robustness of the experimental interpretation and underscores the value of EIS in resolving the complex, non-ideal interfacial processes underlying alkaline aluminum corrosion.

3.3 X-ray diffraction analysis

X-ray diffraction was employed to assess whether the electrochemical corrosion processes identified from OCV, polarization, and EIS measurements are accompanied by crystallographic degradation or formation of stable corrosion products on the commercial Al alloy.

The XRD patterns obtained after exposure to KOH and NaOH electrolytes across the full 1–10 M concentration range are dominated by the characteristic reflections of face-centered cubic aluminum at approximately 38.5°, 44.7°, 65.1°, and 78.2°, corresponding to the (111) (200), (220), and (311) planes as shown in Figure 8. The persistence of these reflections with invariant peak positions across all concentrations and both electrolytes indicates that the bulk aluminum lattice remains structurally intact despite substantial electrochemical activity. No peak shifts or broadening indicative of lattice distortion or phase transformation are observed, confirming that alkaline exposure does not induce measurable crystallographic degradation of the alloy. Notably, no additional diffraction peaks associated with crystalline aluminum hydroxides, oxyhydroxides, or aluminate phases are detected under any condition. This absence suggests that corrosion products formed during alkaline exposure are predominantly amorphous, nanocrystalline, or highly hydrated, falling below the detection limit of XRD. This finding is consistent with electrochemical evidence for continuous oxide dissolution, high corrosion currents, and the lack of sustained passivation, particularly at moderate-to-high hydroxide concentrations. Minor variations in relative peak intensity observed at higher concentrations are attributed to surface roughening, preferential dissolution, and changes in near-surface texture rather than to bulk compositional changes. Such effects are expected under aggressive alkaline corrosion, where rapid anodic dissolution and hydrogen evolution continuously renew the exposed surface and inhibit the accumulation or crystallization of corrosion products. Comparison between KOH and NaOH reveals no electrolyte-specific crystalline phases, despite pronounced differences in corrosion kinetics and impedance response. The retention of the aluminum crystalline framework across all conditions provides structural validation for the mechanistic interpretations derived from polarization and EIS analyses and confirms that corrosion proceeds through continuous surface dissolution with transient, non-protective corrosion products.

3.4 ANN model performance and predictive accuracy

A comprehensive grid search comprising 351 ANN configurations was conducted to identify an accurate and robust predictive model for Ecorr and Icorr across NaOH and KOH concentrations. The search space systematically varied network depth, neuron allocation per layer, learning rate, and Adam β₁, while training length and batch size were held constant. Validation performance across the grid revealed substantial variability in model accuracy, with mean R ² values ranging from 0.785 to 1.000 and an average of 0.972. Notably, 267 configurations, corresponding to 76 percent of all tested models, achieved mean validation R ² values exceeding 0.97, indicating that the underlying electrochemical relationships are highly learnable yet sensitive to architectural choices.

Performance trends were consistent with expected nonlinear structure in the dataset. Increasing hidden-layer depth generally improved Ecorr prediction accuracy, whereas Icorr remained robust across a wider range of architectures. Networks with non-uniform layer widths outperformed uniform configurations, demonstrating that asymmetric feature compression and expansion better captured the coupled effects of electrolyte identity, concentration, and polarization behavior. Learning rates above 0.001 and β₁ values below 0.90 tended to produce unstable convergence, confirming the importance of carefully tuned optimization dynamics when modeling strongly nonlinear corrosion responses.

The best-performing configuration used a three-layer architecture with 20, 25, and 25 neurons, a learning rate of 0.001, and β₁ = 0.92. This model achieved R ² values of 0.999,888 for Ecorr and 0.999,945 for Icorr, yielding a mean validation R ² of 0.999,917. Such accuracy represents an almost perfect mapping between the experimental variables and both corrosion indicators, validating the suitability of the ANN approach for predicting corrosion behavior in alkaline media. The optimized hyperparameters are summarized in Table 5, confirming that relatively compact architectures can achieve high fidelity when hyperparameters are tuned systematically.

TABLE 5. ANN hyperparameter search results, including tuning grid and selected optimal configuration.

The optimized three-layer ANN model demonstrated excellent predictive capability for both target variables, achieving near-perfect agreement between measured and predicted values across all data partitions. The training history (Figure 9a) shows rapid and stable convergence of the loss function, with monotonic reduction of MSE for the training, validation, and test sets throughout 500 epochs. No divergence or oscillatory behavior was observed, confirming that the selected hyperparameters ensured stable optimization. Table 6 shows the Summary of the optimized ANN model architecture and training configuration used for Ecorr and Icorr prediction.

TABLE 6. Summary of the optimized ANN model architecture and training configuration used for Ecorr and Icorr prediction.

Scatter plots of predicted versus target Icorr values for the training, test, and validation subsets (Figures 9b–d) reveal extremely strong predictive performance, with R ² values of 0.99975, 0.99975, and 0.99970, respectively. The points lie tightly along the line 1:1, indicating negligible systematic bias and excellent generalization across previously unseen samples. When predictions were transformed back to physical units, the experimental-predicted comparison (Figure 9e) also maintained a very high coefficient of determination (R ² = 0.99974), demonstrating that model accuracy remains robust in real units, not only in normalized space.

The consistently high R ² values across all subsets and the absence of overfitting behavior confirm that the optimized ANN architecture successfully captures the nonlinear dependence of Ecorr and Icorr on electrolyte type, concentration, temperature, and polarization conditions. Together with the grid-search analysis, these results validate the ANN as a powerful predictive tool for corrosion modeling in alkaline environments.

3.5 GA and NSGA-II optimization outcomes

Following ANN training, genetic algorithm-based optimization was applied to identify electrolyte conditions that balance corrosion thermodynamics and kinetics. A scalar GA was first employed to obtain a single optimal solution under equal weighting of Ecorr maximization and Icorr minimization, followed by NSGA-II to resolve the full trade-off structure between these two objectives without imposed weights. The resulting NSGA-II Pareto front and associated knee point are shown in Figure 10. The NSGA-II results reveal a smooth and continuous Pareto front for alkaline corrosion behavior, indicating a well-resolved trade-off between Ecorr and Icorr across electrolyte concentration. As shown in Figure 10, solutions with lower Icorr are associated with more negative Ecorr values, whereas improvements in Ecorr are accompanied by a rapid increase in corrosion current density. This monotonic but nonlinear trend reflects the fundamental electrochemical coupling between anodic dissolution kinetics and corrosion thermodynamics in alkaline aluminum systems. It should be noted that the Pareto front is generated from ANN-predicted values across a continuous input space, whereas the experimental data are discrete and are presented separately in the Results section.

The curvature of the Pareto front is particularly pronounced at low Icorr values, where relatively small increases in Icorr lead to diminishing gains in Ecorr. This behavior gives rise to a clearly identifiable knee region, which represents the transition from corrosion-limited to dissolution-dominated regimes. Beyond this point, further enhancement of Ecorr becomes increasingly costly in terms of corrosion rate, suggesting that the system enters a regime where hydroxide-driven dissolution kinetics dominate over interfacial stabilization effects. Due to the nonlinear mapping captured by the ANN model, small variations in Ecorr may correspond to larger variations in Icorr within the predicted design space.

Using the minimum-distance-to-ideal criterion, the knee point on the Pareto front was identified at a KOH concentration of approximately 1.35 M, with a predicted Ecorr of −1.6199 V and an Icorr of 0.0876 A cm^-2. This solution represents the least-compromise operating condition, where further reductions in corrosion current would require a disproportionate loss in corrosion potential stability. From a mechanistic perspective, the knee point corresponds to a moderate alkaline environment in which oxide film disruption and anodic dissolution are sufficiently activated to stabilize Ecorr, but transport limitations, surface saturation, and solution property effects constrain further acceleration of corrosion kinetics. This interpretation is consistent with the experimentally observed plateauing behavior in corrosion metrics at higher hydroxide concentrations, where increases in alkalinity no longer yield proportional electrochemical benefits.

Notably, the scalar GA optimum coincides almost exactly with the NSGA-II knee point, both identifying KOH at 1.35 M as the identified as a representative trade-off condition within the modeled design space. The scalar GA predicts an Ecorr of −1.619,876 V and an Icorr of 0.087859 A cm^-2, yielding a combined objective score of 0.003179. The near-identical values obtained from two fundamentally different optimization formulations provide strong validation of the ANN-based optimization framework. This agreement confirms that the equal-weight scalarization adopted in the GA does not bias the solution toward artificial extrema, but instead naturally converges toward the physically meaningful compromise identified by the Pareto geometry. In practical terms, this indicates that the knee point represents a robust optimum that is insensitive to the choice of optimization strategy, further reinforcing its relevance for corrosion-aware electrolyte selection. The combined GA and NSGA-II results demonstrate that optimal corrosion performance is achieved at intermediate hydroxide concentrations rather than at the most aggressive alkaline conditions. This finding highlights the importance of accounting for nonlinear electrochemical trade-offs when designing aluminum electrochemical systems, particularly in applications where parasitic corrosion losses must be minimized without sacrificing electrochemical stability.

4 Conclusion

In this study, the corrosion behavior of a commercial Al-Mg-Si alloy in alkaline sodium hydroxide and potassium hydroxide electrolytes was systematically investigated using combined electrochemical characterization and data-driven machine learning techniques. Electrochemical measurements, including open-circuit voltage, potentiodynamic polarization, and electrochemical impedance spectroscopy, were conducted across a wide concentration range from 1 to 10 M to quantify corrosion potential and corrosion current density and to elucidate underlying corrosion mechanisms. The experimental results reveal distinct corrosion regimes governed by anodic activation at low-to-moderate alkalinity and by transport- and interfacial-limited processes at high hydroxide concentrations, with KOH consistently exhibiting more aggressive corrosion behavior than NaOH at equivalent molarities due to which may be attributed to cation-dependent interfacial effects, including differences in ionic mobility, hydration, and double-layer structure. X-ray diffraction analysis confirms that corrosion proceeds without bulk phase transformation, indicating continuous surface dissolution accompanied by predominantly amorphous or transient surface films. Future studies incorporating surface characterization techniques such as SEM or optical microscopy are required to further elucidate corrosion morphology. To model the nonlinear dependence of corrosion behavior on electrolyte parameters, an artificial neural network was developed and trained using the experimental dataset. Systematic hyperparameter optimization resulted in an optimized ANN architecture capable of predicting both corrosion potential and corrosion current density with near-perfect accuracy, achieving coefficients of determination exceeding 0.999 for both outputs. Building on this predictive capability, genetic algorithm-based optimization was applied to identify electrolyte conditions that balance corrosion thermodynamics and kinetics. Multi-objective optimization using NSGA-II revealed a well-defined Pareto front and a robust knee-point solution corresponding to approximately 1.35 M KOH, yielding a predicted corrosion potential of −1.6199 V and a corrosion current density of 0.0876 A cm^-2. These results demonstrate the effectiveness of integrating electrochemical experimentation with ANN-assisted evolutionary optimization for corrosion modeling and electrolyte selection. The proposed framework enables rapid, quantitative evaluation of corrosion behavior without exhaustive experimentation and provides a foundation for corrosion-aware electrolyte optimization in aluminum-air battery systems. Future work will extend this approach to include additional operating parameters and degradation metrics, supporting the development of digital-twin platforms for the design, optimization, and control of next-generation aluminum-based electrochemical energy systems.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mohammad Waseem, Jamia Millia Islamia, India

Reviewed by: Yulong Wu, Helmholtz Centre for Materials and Coastal Research (HZG), Germany

Jodie Yuwono, Adelaide University, Australia

Footnotes

Ag/AgCl, Silver/Silver chloride reference electrode; Al-air, Aluminum-air battery; Al-Mg-Si, Aluminum-magnesium-silicon alloy system; ANN, Artificial neural network; Cdl, Double-layer capacitance; CPE, Constant phase element; Ecorr, Corrosion potential; EIS, Electrochemical impedance spectroscopy; EV, Electric vehicle; FCC, Face-centered cubic lattice structure; GA, Genetic algorithm; Icorr, Corrosion current density; KOH, Potassium hydroxide; MAE, Mean absolute error; MSE, Mean squared error; NaOH, Sodium hydroxide; NSGA-II, Non-dominated sorting genetic algorithm II; OCV, Open-circuit voltage; R², Coefficient of determination; Rct, Charge-transfer resistance; ReLU, Rectified linear unit activation function; RMSE, Root mean squared error; Rs, Solution resistance; XRD, X-ray diffraction.