Optimization of Glucose Dehydration to 5-Hydroxymethylfurfural over Fe₃O₄–SiO₂/Al₂O₃ Catalyst Using Response Surface Methodology in a Biphasic System

1. Introduction

5-Hydroxymethylfurfural (HMF) has emerged as an important biomass-derived platform molecule in the transition from a fossil fuel-based chemical economy to one that relies on renewable carbon resources, driven by the depletion of fossil fuels and environmental concerns [1,2]. HMF can be produced from carbohydrates derived from lignocellulosic biomass and subsequently upgraded into a wide range of high-value chemicals, making it a key intermediate in biorefinery systems designed for both materials and energy applications [3]. HMF contains a furan ring with both aldehyde and hydroxymethyl groups, making it highly reactive and enabling it to undergo various transformations, such as esterification, oxidation, reduction, hydrogenation, and halogenation [4,5,6]. These reactions produce derivatives including 2,5-diformylfuran, 2,5-furandicarboxylic acid (FDCA), 2,5-dimethylfuran, 5-ethoxymethylfurfural, levulinic acid, and other furan compounds intermediates [7]. These compounds find applications in biofuels, plasticizers, pharmaceuticals, polyesters, and chemicals, making HMF a crucial link between renewable biomass chemistry and the petrochemical industry [6,8,9]. As a result, the U.S. Department of Energy (DOE) has listed HMF among the top 10 value-added chemicals derived from biomass, emphasizing its importance for future bio-based value chains [10,11].

HMF can be produced from C6 carbohydrates, including glucose and fructose, which are derived from lignocellulosic biomass. The typical process for converting biomass into HMF involves three main steps: hydrolyzing cellulose into glucose, isomerizing glucose into fructose, and subsequently dehydrating fructose to form HMF [6]. Among these sugars, dehydration of fructose is relatively simple and generally yields higher HMF selectivity and yield. However, fructose is relatively expensive and occurs in limited amounts in nature, which limits its industrial importance. Conversely, glucose accounts for 40-60% of lignocellulosic biomass, making it a primary component of cellulose and economically appealing. Therefore, creating effective methods for converting glucose into HMF has become a crucial goal in biomass valorization research [12]. Despite its availability, glucose conversion remains significantly more challenging than fructose dehydration due to low selectivity, a multistep reaction process, and the formation of several undesired byproducts [8].

To address these challenges, various catalytic systems and solvents have been investigated. Solvent selection plays a critical role in both selectivity and activity during HMF production, influencing reaction rates, sugar solubility, product stability, and the formation of undesired by-products [1,2,13]. Water is widely regarded as a green and sustainable solvent due to its safety, abundance, and ability to dissolve carbohydrates [9]. However, purely aqueous systems often yield low HMF selectivity due to rapid rehydration, and formation of formic acid, levulinic acid, and humin [12]. Ionic liquids (ILs) have become attractive alternative solvents because they can dissolve cellulose, offer high thermal stability, and have low volatility. Additionally, several IL systems have shown enhanced HMF production [14]. Organic solvents, including tetrahydrofuran (THF) and dimethyl sulfoxide (DMSO), can effectively suppress by-products and enhance product selectivity [15]. However, concerns about cost, environmental impact, toxicity, and separation complexity restrict their large-scale implementation [16]. As a result, biphasic solvent systems composed of water and an immiscible organic phase have been extensively employed. In these systems, HMF produced in the aqueous phase is rapidly extracted into the organic phase, minimizing its degradation in water [17]. The presence of inorganic salts further improves HMF extraction by a salting-out effect, resulting in better product selectivity [4,6]. Several organic phases such as ethyl acetate [1], n-butanol [9], THF [6], DMSO [16], acetone [18] and methyl isobutyl ketone (MIBK) have been used in previous studies. MIBK is considered one of the best solvents due to its lower miscibility with water, chemical stability, and effectiveness in phase separation [4,19].

The choice of catalyst support is equally critical for achieving high 5-HMF yields and facilitating catalyst recovery. Initial studies on sugar dehydration primarily relied on homogeneous mineral acids, including H₂SO₄, HCl, and H₃PO₄ [20]. While these acids provide strong catalytic activity, they suffer from several drawbacks such as reactor corrosion, difficulties in catalyst recovery, energy-intensive separation and environmental hazards [6]. Subsequently, metal halide catalysts such as SnCl₄, CrCl₃, and AlCl₃ were introduced due to their enhanced catalytic efficiency and strong Lewis acidity [21]. However, these catalysts may generate corrosive acidic species, suffer from hydrolysis in aqueous media, and pose sustainability challenges, which have driven the development of more environmentally benign heterogeneous acid catalysts [22]. To address these challenges, many diverse solid acid catalysts have been developed, including heteropolyacids, supported metal catalysts, zeolites, metal phosphates, metal oxides, sulfonated resins, functionalized silica-based materials, and carbonaceous acids [23]. Many of these catalysts are designed to incorporate both Lewis and Brønsted acid sites within a bifunctional framework, which facilitates the isomerization of glucose and the dehydration of fructose. These heterogeneous catalysts offer notable benefits in ease of separation, recyclability, and process stability. Their catalytic performance mainly depends on the strength, density, and accessibility of acid sites, which influence interactions with sugar molecules and mass transfer [9,24].

Among heterogeneous catalysts, silica- and aluminosilicate-based materials (such as silicon dioxide, montmorillonite, and zeolites) have received significant attention. These materials possess high surface area, excellent thermal stability, tunable pore structures, and versatile surface chemistry, enabling easy functionalization [25]. Recent research has focused on modifying silica and aluminosilicate materials to incorporate Lewis-acidic metal centers such as Nb, Zr, Sn, Cr, and Fe, along with Brønsted-acid groups, such as sulfonic groups, on the surface [3,26]. This approach enables the creation of bifunctional catalysts capable of performing isomerization and dehydration simultaneously in a single reactor. Moreover, silica-based catalysts perform well in biphasic systems, where solvent synergy and optimized acidity enhance HMF selectivity and reduce side reactions [27]. In a recent study, Zhang et al. successfully synthesized the Sn-OH/SBA-15 catalyst under acidic conditions and achieved a 70.6% HMF yield with a 96.9% glucose conversion 180℃ [28]. Similarly, Mercedes Moreno-Recio et al. reported a 55.8% HMF yield using H-Beta zeolite in a biphasic system (water: NaCl/MIBK) at 195℃ [29]. In another approach, Bharath Velaga et al. employed an H-MOR zeolite catalyst with a high silica-to-alumina ratio, achieving 87% HMF selectivity and 76% glucose conversion [30]. Furthermore, Elsayed et al. obtained 70% HMF yield in a water/MIBK biphasic system by using a magnetic nanoparticle catalyst (Fe₃O₄@SiO₂-SO₃H) prepared by using sulfonic acid [17].

Traditional methods for evaluating parameters usually focus on one factor at a time by keeping all others constant. However, this approach has limitations, as it fails to account for the effects and interactions among variables. Response Surface Methodology (RSM) and Design of Experiments (DoE) offer a framework that optimizes complex systems [31]. RSM measures the relationship between variables and responses to improve process efficiency while reducing the number of experiments. At the same time, DoE ensures that the smaller experimental set is statistically sound and provides reliable, reproducible results [32].

This study, for the first time, reports the synthesis of a novel and cost-effective iron-modified silica–alumina catalyst, in which iron introduces Lewis acid sites to complement the inherent Brønsted acidity of silica–alumina, enabling efficient glucose dehydration to HMF in a biphasic water/MIBK system. The catalyst was characterized using FTIR, TGA, XRD, SEM–EDX, XPS, and N2 adsorption-desorption isotherm (BET). Reaction parameters such as temperature, time, and catalyst loading were optimized through an RSM-BBD study. Reusability and regeneration of the catalyst were also evaluated.

2. Materials and Methods

2.1. Materials

Silica alumina catalyst (Grade-135), D-(+)-glucose (≥99.5%), D-(-)-fructose (≥99%), 5-hydroxymethylfurfural (5-HMF) (≥99%), and methanol (99.8%) were purchased from Sigma-Aldrich. Iron (III) nitrate nonahydrate (>98%) and methyl isobutyl ketone (MIBK) (HPLC grade, 99+%) were obtained from Alfa Aesar. Nickel (II) nitrate hexahydrate (99%) and glacial acetic acid were purchased from ACROS Organics. All chemicals were used without further purification.

2.2. Synthesis of Fe₃O₄-SiO₂/Al₂O₃ Catalyst

The Fe₃O₄-SiO₂/Al₂O₃ catalyst was prepared by co-precipitation, as outlined in previous studies, with some modifications [33]. First, 0.6 g of silica alumina was added to a 100 mL beaker, and the beaker was filled to 50 mL with deionized water. The solution was sonicated for 30 min at 50℃ to form an emulsion. A 10 wt.% iron solution, based on the weight of silica-alumina (1:10, m/m), was prepared in a separate beaker by dissolving 0.4347 g Fe(NO₃)₃·9H₂O in 10 mL of water. The salt mixture was added dropwise to the silica-alumina solution under continuous stirring. The solution was sonicated for 1 h at 80 ℃, then placed in an oven at 80℃ overnight. The dried mixture was transferred to a crucible and placed in a muffle furnace for calcination at 550 °C for 5 h. A monometallic catalyst was formed and stored in a closed vial for further use. The monometallic NiO-SiO₂/Al₂O₃ and the bimetallic NiO-Fe₃O₄-SiO₂/Al₂O₃ were also prepared using the same method described above, and their catalytic abilities were compared with the Fe₃O₄-SiO₂/Al₂O₃ catalyst. Results shown in Table S1 indicate that Fe₃O₄-SiO₂/Al₂O₃ performed better, so we used this catalyst in our study. Additionally, we compared different iron loading ratios (5, 10, and 20wt%) on a silica-alumina catalyst, and the results are presented in Table S2.

2.3. Catalyst Characterization

The functional groups of the catalyst were characterized by Fourier Transform Infrared (FTIR) spectroscopy using a PerkinElmer Spectrum Two instrument. Spectra were recorded in transmission mode using KBr pellets over the wavenumber range of 450–4000 cm⁻¹, with a spectral resolution of 4 cm⁻¹ and accumulation of 10 scans. The thermogravimetric analysis (TGA) was performed using an SDT Q600. The sample was heated from 25 °C to 800 °C at 5 °C/min under a nitrogen flow rate of 50 mL/min. X-ray diffraction (XRD) patterns were recorded on a RINT Ultima III diffractometer (Rigaku Corp., Japan) with Cu Kα₁ radiation (λ = 1.54 Å). The instrument was operated at 40kV and 44 mA to examine the crystal structure of the synthesized catalyst. The samples were ground finely and placed in recessed glass holders to ensure a flat surface. X-ray photoelectron spectroscopy (XPS) measurements were carried out using a Kratos Axis Ultra to analyze the surface chemical composition and oxidation states of the elements (Kratos Analytical, Inc., Manchester, UK). The specific surface area, pore size distribution, and pore volume of the catalyst were obtained from N₂ adsorption-desorption isotherms measured at −196 °C using a Quantachrome Autosorb iQ instrument (Quantachrome, USA). The surface area was calculated using the Brunauer–Emmett–Teller (BET) method, while the pore size distribution and pore volume were obtained using the Barrett–Joyner–Halenda (BJH) model. Prior to analysis, the samples were degassed under vacuum at 105 °C for 10 h to remove physically adsorbed species.

2.4. Dehydration of Glucose into 5-HMF

A total of 50 mg of glucose was dissolved in 10 mL of deionized water in a 100 mL stainless steel hydrothermal autoclave reactor (Huanyu). Subsequently, 40 mL of MIBK and the catalyst were added, with catalyst loadings of 25, 50, and 75wt% relative to the initial mass of glucose. The sealed autoclave was then placed in a furnace and heated to 170, 180, and 190 °C for reaction times of 10, 12, and 14 h. Upon completion, the reactor was cooled to room temperature, and the aqueous and organic phases were separated using a separating funnel. The catalyst was recovered by centrifugation at 4150 rpm. Both phases were subsequently filtered through 0.22 µm syringe filters prior to HPLC analysis, as illustrated in Scheme 1. The experimental conditions for the 17 runs conducted prior to applying the response surface methodology based on the Box–Behnken design (RSM-BBD) are summarized in Table 1. The selected temperature range was established from preliminary experiments using a silica–alumina catalyst (Table S3), while the water-to-MIBK volume ratio (1:4, v/v) was adopted based on previous studies [17,34].

2.5. HPLC Analysis

The residual glucose concentration in the reaction mixture was determined using high-performance liquid chromatography (HPLC) on an Agilent 1200 system equipped with a refractive index (RI) detector and a UV detector (λ = 278 nm). Separation was achieved using a Bio-Rad HPX-87P column maintained at 80℃, with deionized water as the mobile phase at a flow rate of 0.6 mL/min for 65 min. The concentration of HMF in both aqueous and organic phases was determined using an Agilent 1100 system equipped with an Agilent Eclipse XDB-C18 column maintained at 25℃ and a UV detector set at λ = 282 nm. Methanol-to-water (1:5, v/v) with 5wt% acetic acid were used as the mobile phase at a flow rate of 1 mL/min for 30 min. Quantification of glucose and HMF was performed by external calibration using standard solutions, with concentrations determined from the corresponding chromatographic peak areas. Representative chromatograms obtained from the Agilent 1100 and Agilent 1200 systems are shown in Figure S1 (a-c).

2.6. Determination of Glucose Conversion and HMF Yield

Conversion of glucose is calculated using Equation (1):

$$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{\%}={\displaystyle \frac{\left[\mathrm{g}\mathrm{l}\mathrm{u}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{e}\right]\mathrm{i}\mathrm{n}\times \left[\mathrm{V}\right]\mathrm{i}\mathrm{n}-\left[\mathrm{g}\mathrm{l}\mathrm{u}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{e}\right]\mathrm{a}\mathrm{q}\times \left[\mathrm{V}\right]\mathrm{a}\mathrm{q}}{\left[\mathrm{g}\mathrm{l}\mathrm{u}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{e}\right]\mathrm{i}\mathrm{n}\times \left[\mathrm{V}\right]\mathrm{i}\mathrm{n}}}\times 100$$

(1)

Yield of HMF is calculated using Equation (2):

$$\mathrm{Y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}\mathrm{\%}={\displaystyle \frac{\left[\mathrm{H}\mathrm{M}\mathrm{F}\right]\mathrm{a}\mathrm{q}\times \left[\mathrm{V}\right]\mathrm{a}\mathrm{q}+\left[\mathrm{H}\mathrm{M}\mathrm{F}\right]\mathrm{o}\mathrm{r}\mathrm{g}\times \left[\mathrm{V}\right]\mathrm{o}\mathrm{r}\mathrm{g}}{\left[\mathrm{g}\mathrm{l}\mathrm{u}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{e}\right]\mathrm{i}\mathrm{n}\times \left[\mathrm{V}\right]\mathrm{i}\mathrm{n}}}\times {\displaystyle \frac{180}{126}}\times 100$$

(2)

Here, “in”, “aq”, and “org” are initial, aqueous and organic phase, respectively. [glucose] and [HMF] are the concentrations of glucose and HMF in solution. [V] is the volume of solution, and “180” and “126” are the molecular masses of glucose and HMF, respectively.

2.7. Experimental Design and Statistical Analysis

Response Surface Methodology (RSM) was applied to optimize the dehydration process. A Box–Behnken design (BBD) was adopted to systematically evaluate the effects of three independent variables: reaction temperature (T), reaction time (t), and catalyst loading (C) on the HMF yield. Each factor was examined at three coded levels (−1, 0, +1), corresponding to low, intermediate, and high values, respectively, as summarized in Table 2. The BBD approach is based on a second-order polynomial model incorporating linear, interaction, and quadratic terms, enabling efficient exploration and optimization within the experimental space. A total of 15 experimental runs were conducted, including three replicates at the center points to evaluate the pure error, as presented in Table 3. Statistical analysis and model fitting were performed using Design-Expert 13 software and optimization was achieved through analysis of variance (ANOVA) to assess the statistical significance and adequacy of the model using the coefficient of determination (R²), adjusted R², and lack-of-fit tests. The statistical significance of individual factors and their interaction was assessed using F-values and corresponding p-values. The experimental data were fitted to a full quadratic model to establish the relationship between the response (HMF yield) and the independent variables.

2.8. Experimental Design and Statistical Analysis

After each reaction, the catalyst was easily recovered by centrifugation, thoroughly washed with 100 mL of deionized water to remove residual reactants and products, and then reused under the optimized reaction conditions. The catalytic performance was evaluated over four consecutive cycles to assess stability and reusability. Catalyst regeneration was carried out by calcination at 550℃ for 5h in a muffle furnace, followed by reuse in subsequent reaction cycles to evaluate the recovery of catalytic activity.

3. Results and Discussion

3.1. Characterization of the Catalyst

FTIR spectra of fresh and used Fe₃O₄-SiO₂/Al₂O₃ monometallic catalyst are shown in Figure 1a. Both spectra exhibit a broad band around 3439 cm^-1, which corresponds to the stretching vibrations of surface hydroxyl (‒OH) groups. A small peak at 1634 cm^-1 represents the bending vibration of adsorbed water, consistent with the hygroscopic nature of silica-containing materials. A sharp peak at 1091 cm^-1 is assigned to the asymmetric stretching vibration of the Si-O-Al, confirming the formation of silica-alumina framework of the catalyst. Additionally, the peak at 800 cm^-1 indicates the symmetric stretching vibration of the Si-O-Si bond, characteristic of silica linkage [17,35]. Both the fresh and used catalysts exhibit the same spectra, with differences in peak intensity confirming that the catalyst’s structural backbone remains intact after the reaction. Minor variations in band intensity suggest deposition of reaction byproducts (e.g., humin) on the catalyst surface.

Figure 1b shows the Thermogravimetric analysis (TGA) analysis of the Fe₃O₄-SiO₂/Al₂O₃ monometallic catalyst, illustrating its weight loss. It reveals an initial weight loss between 100 and 150℃, attributed to the removal of physically adsorbed water [36]. Beyond this region, the catalyst exhibits less than 4% weight loss up to 800℃, demonstrating the catalyst’s excellent thermal stability for high-temperature reactions.

Figure 1c shows the X-ray diffraction (XRD) patterns of fresh and used Fe₃O₄-SiO₂/Al₂O₃ catalysts. The broad diffraction peak observed in both spectra, centered at about 2θ = 25° and extending to near 40°, indicates the predominantly amorphous nature silica-alumina structure [37]. The weak diffraction peaks observed at 2θ of 35.6° and 62.5° are indexed to the (311) and (440) lattice planes of Fe₃O₄, confirming the presence of magnetite nanoparticles [17]. The relatively low intensity of these peaks suggests high dispersion of Fe₃O₄ particles on the silica-alumina support. In the used catalyst, the overall diffraction pattern remains largely unchanged, indicating that the structural integrity is preserved during the reaction.

As shown in Figure 2a and Figure 2b, the adsorption isotherms for both fresh and used catalysts displayed typical type IV isotherms with H3-type hysteresis loops, characteristic of mesoporous materials with slit-like pores formed by aggregated particles [36]. The fresh catalyst (Figure 2a) shows a gradual increase in nitrogen uptake at low relative pressure (P/P₀ < 0.3), corresponding to monolayer–multilayer adsorption, followed by a pronounced capillary condensation step at intermediate relative pressures (P/P₀ ≈ 0.4–0.9), confirming the presence of well-developed mesoporosity. In contrast, the used catalyst (Figure 2b) exhibits a significantly reduced adsorption capacity across the entire relative pressure range, indicating a substantial decrease in accessible surface area and pore volume. The hysteresis loop remains present but is less pronounced, suggesting partial pore blockage and reduced pore accessibility. This reduction in textural properties is attributed to the deposition of carbonaceous byproducts (humin) during the dehydration reaction, which obstructs pore channels and limits nitrogen diffusion. Additionally, the slight shift toward larger average pore diameter observed after use suggests preferential blockage of smaller mesopores, leaving wider pores relatively accessible [38]. The fresh catalyst achieved a high surface area of 426.935 m²/g, an average pore size of 6.1400 nm, and a pore volume of 0.608 cc/g. After reaction, the surface area decreases significantly to 175.139 m²/g, accompanied by a reduction in pore volume (0.476 cm³ g⁻¹) and an increase in average pore diameter (10.45 nm). The pore size distribution curves further confirm that both catalysts possess mesopores in the 2–50 nm range, consistent with BJH analysis. The preservation of mesoporosity after the reaction indicates that, despite partial deactivation, the catalyst’s overall pore structure remains intact.

SEM examines the morphology, particle distribution, and surface structure of a material, while EDX is employed to identify and quantify elements. Figure (3a-f) shows the SEM-EDX of the fresh Fe₃O₄-SiO₂/Al₂O₃ monometallic catalyst. Figure 3a displays the morphology of a fresh catalyst surface, which appears rough, relatively clean, and porous surface, composed of quasi-spherical particles, indicative of high surface area and abundant active sites. EDX analysis (Figure 3b) confirms the presence of Si, Al, O, and Fe, verifying the successful incorporation of iron oxides into the silica-alumina matrix. Elemental mapping (Figures c-f) confirms that Si, Al, and O are evenly spread across the surface. Notably, the Fe mapping reveals a fairly uniform distribution, suggesting that iron oxide is well dispersed on the support without forming significant agglomerates. Figure 3g displays the surface morphology of the used catalyst, where particles appear partial agglomeration and some pores are blockage, probably due to deposition of reaction byproducts. A slight structural collapse is also observed, which may contribute to the reduced surface area. The corresponding EDX spectrum (Figure 3h) still confirms the presence of all constituent elements, Slight variations in elemental composition are observed, particularly a relative decrease in Fe content, which may indicate minor leaching or surface coverage of active sites. Nevertheless, elemental mapping (Figure 3i-l) indicates that Fe remains relatively well dispersed after use.

XPS was employed to investigate the surface composition and oxidation states of a catalyst (Figure 4). The survey spectrum (Figure 4a) confirms the presence of Fe, O, Si, and Al, with characteristic binding energies at 710.86 eV, 532.29 eV, 104.11 eV, and 74.76 eV for Fe 2p, O 1s, Si 2p, and Al 2p, respectively, indicating successful incorporation of iron oxide into the silica–alumina matrix without detectable impurities. The high-resolution Fe 2p spectrum (Figure 4b) exhibits characteristic features of mixed-valence iron oxide. The deconvoluted peaks at 711.1 eV (Fe 2p₃/₂) and 724.4 eV (Fe 2p₁/₂), along with the satellite peak at 719.7 eV, confirm the coexistence of Fe²⁺ and Fe³⁺ species, which is a hallmark of Fe₃O₄ [39]. The presence of these mixed oxidation states is critical, as it enhances redox flexibility and may contribute to catalytic activity. The Si 2p spectrum shows a dominant peak at 103.56 eV, which can be attributed to Si⁴⁺ in SiO₂, confirming that silicon is predominantly present in an oxidized silica framework, as depicted in Figure 4c. Similarly, the Al 2p region in Figure 4d displays a peak at 73.98 eV, characteristic of Al³⁺ in the aluminosilicate structure, indicating the preservation of the support framework [40]. The O1s spectrum (Figure 4e) can be deconvoluted into three distinct components: (i) lattice oxygen associated with metal oxides (Fe–O) at 530.0 eV, (ii) bridging oxygen in Si–O–Si and Si–O–Al linkages at 532.5 eV, and (iii) surface hydroxyl groups (Si–OH/Al–OH) at 533.9 eV. The presence of these hydroxyl groups suggests the availability of Brønsted acid sites, while metal–oxygen bonds may contribute to Lewis acidity. Overall, the XPS results confirm that the catalyst surface comprises well-dispersed Fe₃O₄ species embedded within a stable silica–alumina framework, with both acidic and redox-active sites. This combination is expected to facilitate glucose dehydration by promoting protonation, stabilizing intermediates, and suppressing side reactions.

3.2. Statistical Analysis and Model Adequacy

Table 3 summarizes the experimental and model-predicted HMF yields for the 15 runs of the Box-Behnken design, together with the corresponding residuals. Across all factor combinations, the predicted values are in close agreement with the actual experimental values, with residuals generally within ±2, indicating that the quadratic model describes the response very well across the entire design space. A high HMF yield of 60.13% was obtained at 180℃, 14h, and 50 wt% catalyst loading (run 10), with a small residual of 1.42, demonstrating the reliability of the model under optimal conditions. In contrast, significantly lower yields were observed under milder conditions, such as low temperature and catalyst loading (run 9) or short reaction time at low temperature (run 13), highlighting the sensitivity of the system to operating parameters. The absence of any systematic trend in the residuals with respect to run order or factor levels suggests no obvious outliers or unexplained curvature beyond the fitted model, supporting the statistical adequacy of the RSM model for predicting HMF yield under the conditions studied. The graphs of predicted vs. actual yield and the normal plot of residuals are shown in Figure S2.

3.3. ANOVA Table for the Quadratic Model

The ANOVA results in Table 4 indicate that the overall quadratic model is highly significant, with a model p-value of 0.001 and an F-value of 27.68, confirming that the model effectively captures the variability in HMF yield. Within linear terms, Temperature (A) is highly significant with p-value 0.0002 and very high F-value 92.06. This indicates that temperature strongly governs the dehydration kinetics and HMF selectivity. Whereas reaction time (B) shows a weaker but noticeable effect (p-value = 0.0584), while catalyst loading (C) is also significant with p-value 0.0138, demonstrating that the availability of active acid sites contributes meaningfully to HMF formation. Regarding the interaction effects, the AB term is statistically significant (p-value 0.0342), indicating that the effect of temperature on HMF yield depends on reaction time and vice versa. In contrast, AC and BC are not significant, suggesting limited synergistic effects between catalyst loading and the other variables. The quadratic terms A², B², and C² are all highly significant (p-value ≤ 0.0112), supporting the strong curvature observed in the response surface and justifying the use of a second-order polynomial model to capture the presence of an optimum region. The model exhibits a high coefficient of determination (R² = 0.9803), indicating that 98.03% of the variance in HMF yield is explained by the selected variables. The adjusted R² (0.9449) indicates that the model explains most of the experimental variance with only limited overfitting, while predicted R² (0.7657) remains reasonably close to adjusted R², suggesting good predictive ability for new experimental points. Furthermore, the lack-of-fit test is not significant (p-value = 0.4112), confirming that the quadratic model is adequate and no higher-order cubic terms are needed to describe the data within the experimental domain. In Table 5, the sequential model sum of squares and p-values further support this conclusion. The quadratic model is highly significant (p-value 0.0006) and markedly improves fit compared with the linear and 2FI models, whereas the cubic model is aliased and not statistically justified. The effect of all factors on HMF yield is depicted in Figure S3.

Equations 3 and 4, in terms of coded and actual factors, respectively, are used to predict the response for the given levels of each factor.

$$HMFYield=58.71+7.125\ast A+1.815\ast B-2.76\ast C-3.035\ast AB-1.7\ast AC-0.38\ast BC-11.3225\ast A^2-4.3975\ast B^2-4.2825\ast C^2$$

(3)

$$HMFYield=-\mathrm{4,312.38}+43.6345\ast A+54.9875\ast B+1.89\ast C-0.15175\ast AB-0.0068\ast AC-0.0076\ast BC-0.113225\ast A^2-1.09938\ast B^2-0.006852\ast C^2$$

(4)

3.4. Regression Coefficient and Factor Effects

Table 6 shows that temperature has a strong positive regression coefficient, indicating that increasing temperature from low to mid-levels strongly promotes HMF formation, consistent with the endothermic nature of the dehydration step. In contrast, the negative quadratic coefficient (A²) indicates that excessive temperature eventually reduces HMF yield, consistent with the observed decline at the extreme high-temperature region of the response surface due to HMF decomposition and humin formation. Reaction time shows a moderate positive linear coefficient, suggesting that longer residence time initially enhances HMF formation, whereas its negative quadratic term (B²) again indicates an optimum beyond which prolonged heating causes side reactions and product loss. Catalyst loading shows a negative linear coefficient under the coded conditions, implying that increasing the catalyst amount beyond a certain level may favor side reactions (e.g., formation of levulinic acid, formic acid, and humin) rather than selectively promoting HMF generation. This trend is reinforced by the significant negative quadratic term C². The negative coefficient of the AB interaction term indicates that the beneficial effect of increasing temperature is more pronounced at shorter reaction times, while at longer times, the same temperature rise becomes less favorable or even detrimental to HMF yield. The relatively narrow 95% confidence intervals of the significant coefficients indicate good precision in estimating the model parameters. Finally, all variance inflation factors (VIFs) are close to 1, confirming the absence of multicollinearity among the coded variables and supporting the statistical robustness of the fitted response surface model. Diagnostic plots, including Box-Cox transformation, perturbation, and cube plots, are shown in Figure S4 and Figure S5, respectively, further validating the statistical reliability and predictive capability of the developed model.

3.5. 3D Surface and Contour Plots

The three-dimensional (3D) response surfaces and corresponding contour plots (Figure 5) demonstrate in the combined effects of temperature (A), reaction time (B), and catalyst loading (C) on HMF yield within the Box-Behnken design domain. At a fixed catalyst loading of 50wt% (Figure 5a), the temperature-time surface displays a distinct dome-shaped profile. As temperature and reaction time increase from their lower levels, HMF yield steadily rises until reaching a clear optimum in the mid-to-high range of both variables. Beyond this optimum point, further increases lead to a decline in predicted yield, indicating HMF degradation and side reactions occurring under excessively harsh conditions. Yang Liu et al. reported that glucose conversion steadily increased over time when using a catalyst of nano-Al₂O₃ and HCl, ultimately achieving full glucose conversion. The HMF yield initially rose, peaking at 69.1% at 180 °C after 30 min, then decreased as the reaction continued up to 90 min [34]. Xiangbo et al. also observed a significant dependence on temperature and time; at lower temperatures (140-150 °C), the HMF yield gradually increased with reaction time, whereas at 170 °C, the yield peaked within 1h and then decreased with longer reaction times [41]. Arumugam Ramesh and colleagues investigated H-SO₄^2-/Ti-Al₂O₃ catalyst with a 200 mg loading, achieving a 65% HMF selectivity. Their results demonstrated that HMF selectivity increased with increasing reaction time from 1 to 4h, but declined after 4h due to increased oligomer and humin formation. Temperature screening from 130 °C to 190 °C identified 170 °C as the optimal temperature for HMF selectivity, whereas further increasing the temperature to 190 °C led to lower HMF yield and more side reactions [42]. In another study, Wenze Guo et al. reported comparable temperature-dependent behavior, noting that P-3Ti/SBA-15 catalyst achieved a maximum HMF yield of 71% at 160℃ in 150 min, which declined to 55.4% at 180℃ due to the degradation of glucose and HMF into humin and formic acid [43]. The corresponding contour plot (Figure 5b) shows closed, elliptical contours confirming a strong quadratic relationship and a significant interaction between temperature and time, as equal-yield lines bend toward an optimum ridge rather than remaining parallel.

Figure 5c shows the interaction between temperature and catalyst loading at a constant reaction time of 12h. A similar trend is observed, with moderate increases in temperature and catalyst loading enhancing HMF yield up to an optimum, followed by a decline at higher catalyst loadings and temperatures. This trend suggests that excessive acidity promotes undesired side reactions, leading to the formation of levulinic acid, formic acid, and humin, as shown in Figure S6. Thawanrat Kobkeatthawin and colleagues enhanced commercial HZSM-5 through nitric acid treatment to boost Brønsted acidity, then incorporated iron modification to add Lewis acidity sites. The optimized Fe-modified zeolite achieved a 67.5% HMF yield at 170℃. When varying Fe loading from 0.1 to 0.5 wt% on dealuminated HZSM-5, they observed that 0.25 wt% Fe loading produced a 61.1% HMF yield with 86.3% glucose conversion. Higher Fe loadings decreased both HMF yield and glucose conversion due to excessive Fe₂O₃ coverage, thereby reducing the number of accessible acid sites. HMF yield increased with rising temperature up to 170℃, reaching 50.3%, while at a temperature of 190℃, humin formation became dominant, leading to 60% humin yield [39]. Yang Liu et al. emphasized that an appropriate balance between Lewis acidic Al₂O₃ and Brønsted acidic HCl was essential. Using Al₂O₃ alone yielded only 20.8% HMF with 79.4% glucose conversion, whereas adding 0.06% HCl significantly increased the yield to 69.1%. Further increasing of HCl concentration to 0.4% reduced the HMF yield, indicating that excessive Brønsted acidity promotes side reactions and HMF degradation [34]. The contours in this plot (Figure 5d) again form closed ovals, indicating curvature and a nonlinear dependence on both factors, with a relatively narrow high-yield region along the catalyst-loading axis, suggesting that HMF formation is particularly sensitive to overdosing than to moderate shifts in temperature.

Figure 5e illustrates the interaction between reaction time and catalyst loading at a fixed temperature of 180℃. Compared to the temperature-dependent surfaces, HMF yield is less sensitive to variations in reaction time near the center point, while changes in catalyst loading produce more pronounced effects on HMF yield. Nevertheless, at prolonged reaction times and high catalyst loadings, a clear decrease in yield is observed, again indicating the dominance of HMF secondary degradation pathways. The contour plot (Figure 5f) shows broad, concentric regions with the highest yields occurring at intermediate reaction times and catalyst levels, further confirming that optimal performance requires a careful balance between sufficient residence time and controlled catalyst dosage to maximize dehydration while minimizing subsequent decomposition and polymerization pathways. Jingjing Wang et al. reported that increasing the catalyst-to-glucose ratio in a 20wt% SnO₂/MCM system enhanced HMF yield from 55.3% (1:8) to an optimum at 1:2, beyond which further increases led to a decline due to the promotion of secondary reactions [44]. Similarly, Xiangbo et al. demonstrated that Sn loading in SAPO-34 catalysts strongly influences product distribution: a 5 wt% Sn loading yielded 51.7% HMF, while increasing the loading to 10 wt% and 15 wt% reduced the yield to 41.7% and 34.8%, respectively, accompanied by increased levulinic acid formation. These findings highlight the critical role of acid-site balance, where Lewis acidity facilitates glucose isomerization to fructose, and Brønsted acidity promotes subsequent dehydration to HMF [41]. Excessive acidity, however, accelerates side reactions such as rehydration and polymerization, leading to byproduct formation. Overall, the trends observed in the present study are in strong agreement with previous reports, confirming that optimal HMF production requires a careful balance between reaction conditions and catalyst properties.

Collectively, the three sets of contour and 3D surface plots reveal a well-defined optimum region within the experimental domain where the maximum HMF yield of 60.13% is achieved at approximately 180-185℃, 11-12h, and 50wt% catalyst loading. These results indicate that moderately high temperature combined with intermediate reaction times and controlled catalyst loading favor HMF formation while limiting its subsequent degradation. The graphical analysis is fully consistent with the statistical model and ANOVA results, providing a clear visualization of the optimization window for maximizing HMF yield.

3.6. Mechanism of Glucose Dehydration to HMF

A plausible reaction mechanism for the conversion of glucose to HMF over the Fe₃O₄-SiO₂/Al₂O₃ catalyst is illustrated in Figure 6. This process is well known to involve two key steps: (i) isomerization of glucose to fructose, and (ii) dehydration of fructose to HMF. The isomerization step is generally considered rate-limiting and is catalyzed by Lewis acid sites, whereas the dehydration step is promoted by Brønsted acidity. In the present catalytic system, silica-alumina support supplies Brønsted acid sites, while Fe₃O₄ offers Lewis acid sites. The Brønsted acidity of silica-alumina arises from Si-OH-Al bridging hydroxyl groups, also known as protonated silanol-aluminum sites [45]. These groups are highly polarized because the electron density is drawn toward the Al³⁺ center, weakening the O-H bond and promoting proton release. The reaction is initiated by protonation of cyclic glucose, followed by ring opening to form the corresponding linear intermediate under Brønsted acid catalysis. Subsequently, Lewis acidic Fe centers coordinate with the oxygen atoms of the open-chain glucose, enabling a 1,2-hydride shift that converts glucose to fructose. In biphasic aqueous–organic systems, this pathway predominantly proceeds via an acyclic mechanism, which stabilizes intermediates and promotes efficient extraction of HMF into the organic phase, thereby suppressing undesired side reactions [10]. In the final step, fructose undergoes sequential protonation and elimination of three water molecules over Brønsted acid sites to form HMF. The cooperative interaction between Lewis and Brønsted acid sites is therefore essential, as it facilitates both isomerization and dehydration steps while enhancing selectivity toward HMF. This synergistic effect is consistent with the observed catalytic performance and aligns with literature reports on bifunctional acid catalysts.

3.7. Catalyst Reusability

The reusability of the Fe₃O₄-SiO₂/Al₂O₃ catalyst was evaluated over four consecutive reaction cycles under the optimized conditions (180 °C, 14 h, and 50 wt% catalyst loading). As shown in Figure 7, the catalyst’s reactivity slightly decreased, with glucose conversion decreasing from 100% to 93.5% and HMF yield from 63.9% to 49.5% after four cycles. This deactivation is primarily attributed to the deposition of carbonaceous byproducts (humin and oligomers) on the catalyst surface and within its pore structure, leading to partial blockage of active sites and reduced accessibility. These findings are consistent with the BET and SEM analyses, which revealed a decrease in surface area and evidence of pore obstruction after use [46,47]. Cycle 5 shows regenerated catalyst results that indicate conversion increased to 95.93% and HMF yield increased to 51.29%. After each cycle, the catalyst was recovered by centrifugation, thoroughly washed with deionized water, and dried prior to reuse. Although a moderate loss of activity is observed, the catalyst retains a significant portion of its performance, demonstrating reasonable stability under the reaction conditions used.

4. Conclusions

A novel Fe₃O₄-SiO₂/Al₂O₃ monometallic catalyst was successfully synthesized via a simple, cost-effective co-precipitation method for the efficient conversion of glucose into HMF. The catalyst exhibits bifunctional acidity, combining Lewis acid sites (Fe₃O₄, Al₂O₃) with Brønsted acid sites originating from the silica–alumina framework. The optimized balance between these acid functionalities, achieved through iron incorporation, plays a crucial role in enhancing catalytic performance. A plausible reaction mechanism was proposed, involving Lewis acid-catalyzed isomerization of glucose to fructose, followed by Brønsted acid-driven dehydration to HMF. The use of a biphasic water/methyl isobutyl ketone (MIBK) system (1:4, v/v) facilitated in situ extraction of HMF into the organic phase, thereby suppressing side reactions and improving selectivity. Reaction parameters, including temperature, time, and catalyst loading, were systematically optimized using RSM-BBD study. The maximum HMF yield of 63.91% with 99.22% glucose conversion was achieved under optimal conditions (180℃, 14h, 50wt% catalyst loading). Complete glucose conversion was obtained at 190℃ under all tested conditions, although with reduced selectivity due to side reactions. Statistical analysis confirmed that temperature is the most influential factor governing HMF yield, followed by catalyst loading and time. The developed quadratic model showed excellent agreement with experimental data (R² = 0.9803). Comprehensive characterization techniques of the catalyst confirmed the successful formation of a thermally stable, mesoporous catalyst with a high specific surface area (426.9 m² g⁻¹) and well-dispersed Fe₃O₄ species. Reusability studies demonstrated that the catalyst retains reasonable activity over multiple cycles, with only a moderate decrease in HMF yield attributed to pore blockage and active site coverage by humin and other carbonaceous byproducts. Overall, this study presents an economically viable and environmentally benign approach for converting abundant glucose into HMF using a robust bifunctional catalyst system. The insights gained into the role of acid-site balance and reaction optimization provide valuable guidance for the rational design of advanced catalysts for biomass valorization.