Design of a Low-Cost Electrochemical Reactor System for Graphene Oxide Production

Abstract

Graphene oxide (GO) production via electrochemical exfoliation has attracted a lot of attention because it is a green, safe, and efficient technique with good quality and greater scalability compared to the conventional synthesis routes. In this study, we present the development of a low-cost (ca. US$50), automated electrochemical reactor for the GO synthesis via electrochemical exfoliation of graphite. The system is based on an Arduino-controlled reactor capable of modulating key synthesis parameters, including voltage levels, duration, and the number of cycles, enabling precise process control. A 2^4-1 factorial design was employed to investigate the influence of process variables-particularly the electrolyte concentration, voltage magnitude, electrolysis time, and number of voltage cycles-on the structural, morphological, and colloidal properties of GO. The results demonstrate that optimizing the electrolyte concentration and the reaction time flakes with uniform size distribution (11.38 ± 0.48 μm) were obtained. Increasing the number of cycles enhances exfoliation efficiency, yielding larger (ca. 35 μm) and thinner (3-6 nm) flakes with higher oxidation degrees and improved colloidal stability, as evidenced by zeta potential measurements reaching -49 mV. Temperature monitoring revealed that the cycling strategy helps lower the reaction temperature by 5 ºC, which promotes more controlled oxidation and exfoliation of graphite. This work highlights a versatile and scalable approach to GO production with tunable properties, offering an environmentally friendly and reproducible alternative to conventional chemical oxidation methods.

Keywords:
automation; Arduino; graphene oxide; electrochemical exfoliation

Introduction

Graphene oxide (GO) is a layered carbon nanomaterial with outstanding optical, thermal, and electrical properties.¹,² Due to its hydrophilic nature and the presence of oxygen-containing functional groups-such as hydroxyl, carbonyl, carboxyl, and epoxy-distributed over the basal plane and edges,³ GO exhibits excellent dispersibility in aqueous and polar solvents.⁴,⁵ The specific physical and chemical behavior of this two-dimensional material is strongly influenced by the synthesis method, which governs both the oxidation degree and the defect density in the carbon lattice.⁶ The sp²/sp³ hybridization domains arising from the coexistence of pristine and oxidized carbon structures modulate the electronic properties of GO, allowing its electrical conductivity to span from semiconducting to insulating, depending on the oxygen functionalities.⁷-⁹

GO has attracted growing attention for diverse applications, including drug delivery,¹⁰ lithium-ion batteries,¹¹ supercapacitors,¹² catalysis,¹³-¹⁵ sensors,¹⁶ environmental remediation,¹⁷,¹⁸ and reinforcement in construction materials.¹⁹ The functional groups on the GO surface enable versatile interactions with organic and inorganic compounds, enabling the development of hybrid nanostructures and composites.²⁰,²¹

Traditionally, GO is synthesized using top-down methods, such as chemical oxidation of graphite via the Hummers method or its derivatives.²²-²⁶ Although widely used, these chemical routes involve harsh oxidants, long reaction times, and considerable environmental risks. Bottom-up methods, such as chemical vapor deposition (CVD) and hydrothermal synthesis, provide high crystallinity and scalability but are hindered by high processing temperatures or limited control over product morphology.²⁷-²⁹

In this context, electrochemical exfoliation (EE) of graphite has emerged as a green, efficient, and scalable alternative for GO synthesis.³⁰-³⁵ EE typically involves the intercalation of ionic species-driven by an applied potential-into graphite electrodes, promoting exfoliation through gas evolution and/or Coulombic repulsion. The process can be anodic or cathodic depending on the polarity of the applied voltage and the nature of the electrolyte.⁴,³³,³⁶ Key parameters such as the type and concentration of the electrolyte, applied voltage profile, and number of electrochemical cycles influence the oxidation degree, sheet size, and layer thickness of the resulting graphene or GO.

Recent research³³,³⁷ has demonstrated several strategies for electrochemical exfoliation with different trade-offs in terms of product quality, environmental impact, and scalability. One common approach involves anodic oxidation of graphite in dilute sulfuric acid (typically 0.1 M H₂SO₄). This method yields few-layer graphene with relatively low oxygen content (around 7-10 at%) and high exfoliation efficiency (> 60%), producing flakes of large lateral size (ca.10 μm). It is rapid, scalable, and does not require post-synthesis reduction steps.³⁸ However, it does involve the use of strong acids, and the process generates oxidized by-products that require proper disposal.

An alternative strategy is pulse-voltage cathodic exfoliation using mixed electrolytes (e.g., H₂SO₄ with KOH).³⁹ This approach uses inexpensive carbon sources, such as graphite from pencil leads, and employs cyclic low and high voltage pulses to control intercalation and exfoliation. While this method avoids highly toxic oxidants and uses benign aqueous media, it produces materials with variable sheet quality and requires complex timing control systems for voltage cycling. The flakes often include multilayer structures and some oxidation defects.

A more elaborate technique uses a two-step intercalation-exfoliation process.⁴⁰ In the first step, graphite is intercalated in a nonaqueous electrolyte (e.g., tetrabutylammonium perchlorate in propylene carbonate), which prevents premature water oxidation and creates stress within the graphite layers. In the second step, the material is transferred to an aqueous electrolyte (e.g., (NH₄)₂SO₄) and a high voltage is applied to achieve exfoliation. This method achieves very high yields (up to 85 wt.%) of bilayer GO and offers excellent control over oxidation. However, it requires the use of organic solvents and perchlorate salts, which are more hazardous and increase operational complexity.

Finally, high-voltage cathodic exfoliation using inert electrodes such as boron-doped diamond (BDD) has been employed to obtain pristine graphene sheets.⁴¹ This process applies constant high negative voltages (e.g., -60 V) in organic electrolytes like tetrabutylammonium hexafluorophosphate. It enables the production of large-area, low-defect graphene with minimal oxidation and excellent conductivity, but requires expensive electrodes, extended reaction times (often > 24 h), and specialized high-voltage power supplies.

These various strategies demonstrate the versatility of electrochemical exfoliation as a synthesis method for GO and graphene derivatives. Each technique offers a unique balance between material quality, environmental footprint, scalability, and equipment cost. In this context, the development of instrumentation and its automation applied in GO EE can significantly expand experimental reproducibility, throughput, parameter control with a reduced cost.

Traditional commercial systems, such as high-end potentiostats, while precise and robust, are often prohibitively expensive and inflexible for non-standard applications. As a response to these limitations, the scientific community has increasingly turned to open-source hardware platforms, particularly Arduino and Raspberry microcontrollers, as cost-effective and customizable alternatives for automating laboratory systems.⁴²

Meloni et al.⁴³ have made seminal contributions in this area by demonstrating how Arduino-based microcontrollers and 3D printing can be combined to create sophisticated, low-cost electrochemical instruments. In their 2016 work, Meloni et al.⁴³ introduced a versatile and inexpensive Arduino-powered potentiostat designed for teaching laboratories, capable of performing cyclic voltammetry and chronoamperometry using only off-the-shelf components and open-source software. Building on this foundation, the 2017 follow-up⁴⁴ described the design of a fully functional, microcontroller-based scanning electrochemical microscope (SECM) fabricated with 3D-printed mechanical components and stepper motor control. This custom-built SECM system enabled electrochemical mapping with micrometer precision, offering modularity and performance previously reserved for commercial systems priced in the thousands of dollars. These studies collectively underscore the viability of microcontroller-based systems not only for teaching, but also for research-level electrochemical analysis. Dryden and Wheeler⁴⁵ introduced DStat, an Arduino-based potentiostat for voltammetric measurements, capable of nanovolt resolution at a fraction of the cost of commercial systems.

In the context of material synthesis, especially for GO, similar principles have begun to reshape EE workflows. Salverda et al.⁴⁶ developed an Arduino-controlled exfoliation system in which a microcontroller monitors current via a Hall-effect sensor and triggers stepper-motor-driven electrode immersion based on current thresholds. Their approach ensured continuous exfoliation with minimal operator intervention and high reproducibility in GO quality. Similarly, Meshkov et al.⁴⁷ integrated Arduino modules within a robotic membrane synthesis platform, using stepper motors and custom Arduino code to automate reagent handling and conductivity analysis in GO-polymer membrane fabrication.

Taken together, these studies demonstrate how Arduino-based automation has transitioned from a pedagogical tool to a viable platform for advanced electrochemical systems. By combining affordability, programmability, and sensor integration, such systems offer an accessible pathway to scalable, customizable scientific instrumentation, especially for GO synthesis and electrochemical material processing.

In this work, we developed a microcontroller-driven electrochemical exfoliation reactor for GO synthesis that automates voltage cycling, thermal monitoring, and process timing using components that cost under US$50. This system incorporates relay-based modulation and pulse width modulation (PWM) control to finely tune intercalation and oxidation stages, with real-time thermal feedback from DS18B20 sensors. The reactor was validated and the influence of process variables-particularly the electrolyte concentration, the magnitude and duration of the applied voltage, and the number of voltage cycles-on the structural, morphological, and colloidal properties of GO was investigated employing a systematic factorial design. By modulating the number of applied electrochemical cycles (N = 1 and N = 10), we studied the impact on structural quality, defect density, and sheet size. This reactor design offers simplicity, reproducibility, and flexibility for optimizing GO production using low-cost graphite sources, including recycled materials.

Experimental

Construction of the automated electrochemical reactor

The construction of the automated electrochemical reactor employed the following materials: Arduino UNO, 12 V power supply model S-60-12, single relay by TONGLING (model JOC-3FF-S-Z), and double relay by SONGLE (model SRD-05V-DC-SL-C). Figure 1 presents the electrical circuit diagram, where a 1 kOhm resistor simulates the electrodes. The relays allow voltage switching in the system and control the reaction time via the Arduino microcontroller. A diode positioned near the low-voltage power source serves as a safety device, protecting the Arduino from potential backflow currents.

Figure 1
Theoretical circuit for the power supply built with Arduino and peripherals for the electrochemical reactor. (b) Simulation on the Tinkercad platform is used to verify the voltage applied to the electric circuit.

Low-voltage application was achieved using a digital pin from the Arduino, providing up to 5 V through PWM. The “analogWrite” function enables analog simulation to control the output voltage.

An initial prototype was assembled on a prototype board. However, to improve component protection and organization, a custom support structure was designed and 3D-printed using a Creality CR5 printer. Polylactic acid (PLA) filament was used, printed at a nozzle temperature of 205 ºC and bed temperature between 60 and 70 ºC. The reactor was designed with a fixed electrode spacing of 1.5 cm. Its removable lid allows integration of external sensors, such as for temperature, liquid level, pH, and others. Electrical components (wires, resistors, boards, etc.) were soldered and connected to a 16 × 2 liquid crystal display (LCD) interfaced with the Arduino UNO.

The automated system was programmed in C++ in the Arduino IDE to accept user-defined parameters: low voltage (in volts), low-voltage duration (in seconds), high-voltage duration (in seconds, fixed at 12 V), and number of voltage cycles. The control program is available in the Open Science Framework (OSF) repository.⁴⁸

Finally, a DS18B20 temperature sensor (operating range: -55 to 150 ºC; sensitivity: 10 mV per ºC) was connected to the Arduino and inserted into the reactor to monitor temperature variations during synthesis.

Synthesis of graphene oxide nanosheets

Electrochemical exfoliation followed the process illustrated in Figure 2. A two-electrode system was employed, using a graphite working electrode (anode) and an inert tungsten counter-electrode (cathode). A (NH₄)₂SO₄ solution was prepared as the electrolyte. Rectangular sections (4 × 1 cm) of graphite foil served as the working electrodes, and tungsten wires were used as counter-electrode. First, an ultrasound washing process of electrodes is performed where the graphite foil was sonicated for 10 min in deionized water and ethyl alcohol. The tungsten wire was washed with HCl (5%) and sonicated in acetone for 10 min. A 100 mL aliquot of the electrolyte was added to the reactor, and the electrodes were mounted onto the custom support. Synthesis parameters were entered into the Arduino interface. The Arduino was connected to a computer via USB A/B cable. The software CoolTerm version 2.4.0.1425 (Roger Meier, 2025) was used for serial communication and to monitor temperature in real time.

Figure 2
Schematic representation of the electrochemical synthesis and purification workflow for graphene oxide (GO) nanosheets. The process begins with electrochemical exfoliation of graphite in (NH₄)₂SO₄ electrolyte under low (1-2 V) and high (12 V) voltage stages, followed by filtration and ultrasound to remove residual electrolyte and facilitate exfoliation. GO flakes are then separated by centrifugation at increasing speeds (1000-3000 rpm) and durations (30 min each), yielding progressively thinner fractions. The final product (GO nanosheet dispersion) is obtained from the 3000 rpm supernatant. Key synthesis parameters-electrolyte concentration (EC), low-voltage amplitude (LV), low-voltage duration (LVt), and high-voltage duration (HVt) are also indicated.

The reaction proceeds in two distinct stages. In the first stage, a low voltage ranging from 1 to 2 V is applied for up to 10 min to promote ion intercalation. In the second stage, the applied voltage is increased to 12 V for a maximum duration of 30 min, initiating the exfoliation process. At the end of the synthesis, the material that remained suspended was filtered using filter paper with an average pore size of 28 µm and washed five times with deionized water to remove adsorbed electrolyte. The retained material was resuspended in 200 mL of deionized water and subjected to sonication for 30 min in an Ultronique ultrasonic bath (200 W) to further exfoliate the material. Immediately after sonication, precipitates at the bottom of the beaker were discarded. The supernatant was then fractionated by centrifugation using a Nova Técnica NT810 centrifuge at 1000, 2000, and 3000 rpm for 30 min. Each synthesis produced three sediment fractions and one supernatant (Figure 2). At lower speeds (1000 rpm), amorphous carbon and poorly exfoliated GO (with many layers) precipitate due to higher density. At higher speeds, GO flakes with varied size and thickness were separated.

In order to screen the most relevant variables of the system, such as electrolyte concentration, applied potential difference, and the duration of both high and low voltage stages, a factorial design was implemented. This approach enables the identification of optimal parameters for the experimental conditions. For this study, as shown in Table 1, a 2^4-1 factorial design was chosen, as this type of experimental planning is particularly suitable when multiple variables must be evaluated simultaneously. After the 2^4-1 factorial design the best condition was selected and the effect of the number of cycles was evaluated accordingly to Table 2.

Table 1
Factorial design matrix (2^4-1) used to investigate the influence of key synthesis parameters on the electrochemical exfoliation of graphene oxide. The variables include electrolyte concentration (EC), low-voltage amplitude (LV), low-voltage duration (LVt), and high-voltage duration (HVt), each coded at two levels (-1 and +1) and the middle point (MP) This design enables the identification of the most significant factors affecting GO sheet quality and yield

		Level (-)	Middle point (MP)	Level (+)
1 - Electrolyte concentration (EC) / (mol L-1)		0.1	0.5	1.0
2 - Low voltage amplitude (LV) / V		1.0	1.5	2.0
3 - LV time duration (LVt) / s		60	330	600
4 - High voltage time duration (HVt) / s		600	1200	1800
Factorial design matrix 24-1
Synthesis number	1	2	3	4
1	-	-	-	+
2	+	-	-	+
3	-	+	-	+
4	+	+	-	+
5	-	-	+	+
6	+	-	+	+
7	-	+	+	+
8	+	+	+	+
9	MP	MP	MP	MP

Table 2
Synthesis parameters of GO synthesis including electrolyte concentration, low-voltage amplitude, low-voltage duration, high-voltage duration, and the number of cycles

Number of cycles	01	10
Electrolyte concentration (EC) / (mol L-1)	0.5	0.5
Low voltage (LV) / V	1.5	1.5
LV time duration (LVt) / s	900	90
High voltage time duration (HVt) / s	900	90

Statistical modeling and visualization

To evaluate the influence of experimental parameters on the lateral size of GO nanosheets according to the factorial design of experiments, the resulting data were analyzed using multiple linear regression in a Python based code. The code is available in the OSF repository.⁴⁸

A coded design matrix was constructed to represent the four independent variables: electrolyte concentration (EC), low-voltage amplitude (LV), low-voltage duration (LVt), and high-voltage duration (HVt), each coded at two levels (-1 and +1), with an additional central point (0) included for validation purposes, as shown in Table 1. The GO sheet size obtained at 3000 rpm was used as the response variable.

A second-order linear regression model was fitted to the data, including all main effects and two-way interaction terms. The model was implemented using the ols function from the statsmodels library in Python with the following formulation:

(1) $$\text{formula}=\text{'GO\_size}\sim (\mathrm{EC}+\mathrm{LV}+\mathrm{LVt}+\mathrm{HVt})\text{** 2'}$$

This allowed quantifying not only the individual contribution of each factor but also their pairwise interactions. The statistical significance and strength of each effect were assessed through the model summary and p-values, while the model’s overall fit was evaluated using the coefficient of determination (R²) and the F-statistic.

To visualize the magnitude of each term in the regression model, a Pareto plot was generated by extracting the absolute values of the standardized regression coefficients (excluding the intercept), sorting them in descending order, and displaying them as a bar chart. This helped identify the most influential parameters governing GO sheet size.

Further, to explore the interaction effects and response behavior in multidimensional parameter space, several three-dimensional response surface plots were generated. These plots represent model-predicted GO sizes across two varying factors, while the remaining two were held constant at their central coded values (0). Specifically, the following pairwise response surfaces were constructed:

(2) $$ECvs.\mathrm{LV}(\mathrm{LVt}=\mathrm{HVt}=0)$$

(3) $$LVtvs.\mathrm{HVt}(\mathrm{EC}=\mathrm{LV}=0)$$

(4) $$ECvs.\mathrm{LVt}(\mathrm{HVt}=\mathrm{LV}=0)$$

(5) $$ECvs.\mathrm{HVt}(\mathrm{LVt}=\mathrm{LV}=0)$$

For each case, a meshgrid was created across the two varying factors, and predicted response values were computed using the fitted regression model. The response surfaces were plotted using Matplotlib’s 3D plotting toolkit (mpl_toolkits.mplot3d), with the colormap encoding the magnitude of the GO sheet size response. All plots were exported as high-resolution images (300 dpi) for inclusion in the Supplementary Information (SI) section.

This statistical modeling approach enabled the identification of significant process variables and interactions affecting GO production, and the visualizations provided intuitive guidance for optimizing the electrochemical exfoliation conditions.

Characterization techniques

Optical microscopy

A ZEISS Axioplan optical microscope equipped with 10×, 40×, and 50× objectives was used to verify the presence of GO sheets. The fractions obtained from the graphene oxide synthesis were diluted in deionized water, and a small aliquot (15 µL) was deposited onto glass slides with a surface area of 1 cm². After drying the solution on a heating plate, the samples were analyzed.

Raman spectroscopy

A StellarNet Raman Hyper-Nova HR Raman spectrophotometer with an excitation wavelength of 532 nm and a fixed power of 100 mW was used. Aliquots of the GO water suspensions were separated for quantification in a tube. The exposure time was varied from a minimum of 2 s to up to 10 s, while the number of accumulations varied from 1 to 8.

Scanning electron microscopy (SEM) and atomic force microscopy (AFM)

For the scanning electron microscopy (SEM) analysis, a Phenom XL G2 benchtop SEM was used, operating at an accelerating voltage of 10 kV and using a back-scattered electron detector. The solutions were diluted in deionized water to the point where they were almost colorless to avoid aggregation of the nanosheets after drying on the substrate. A sample volume of 15 μL was deposited on a copper foil. The samples were dried under a vacuum and subsequently analyzed. For atomic force microscope imaging the samples were deposited on a Si/SiO₂ wafer and dried in vacuum. The AFM images were obtained in NT-MDT Solver NEXT equipment in non-contact mode using a high accuracy non-contact silicon tip with cantilever length and width of 225 and 34 μm, respectively, force constant of 3.5 N m^-1 and resonance frequency of 77 kHz.

Zeta potential

The quantification of the zeta potential was performed using the Malvern ZetaSizer. The solutions obtained were diluted to the concentration of 0.3 mg mL^-1, pipetted, and placed in the cuvette for analysis.

Thermogravimetric analysis (TGA)

The thermal stability was evaluated using a Shimadzu TGA-50 thermogravimetric analyzer, from room temperature to 1000 °C, at a heating rate of 5 °C min^-1, under a nitrogen atmosphere (100 mL min^-1).

FTIR spectra

Fourier-transformed infrared (FTIR) spectra (650 4000 cm^-1) were collected using a Bruker Tensor 27 FT-IR spectrometer equipped with a Platinum-ATR (attenuated total reflectance) Universal sampling module at a resolution of 4 cm^-1, with an average of 32 scans per spectrum at room temperature.

XPS measurements

X-ray photoelectron spectroscopy (XPS) measurements were conducted in a high-resolution Thermo Scientific ESCALAB 250Xi spectrometer equipped with an electron energy hemispherical analyzer and using monochromatized Al Kα line (1486.6 eV) excitation. The spectra were energy referenced to the C1s signal of aliphatic C atoms at the binding energy of 284.8 eV. XPS spectra were collected using X-ray beam spot size = 650 μm with an emission angle of 90^o with respect to the sample surface. High resolution spectra were acquired with 25 eV pass energy.

All data extraction and analysis were done using Python-based codes developed for this work. All codes are available in the OSF repository.⁴⁸

Results and Discussion

Construction of the automated electrochemical reactor

The electrochemical reactor to perform the graphite exfoliation was built using one Arduino UNO board, a 12 V direct current power source, and two relays such as displayed in Figure 1. Furthermore, the synthesis process is fully automated by the Arduino, with the voltage and timing parameters adjustable by the user through modifications to the control script prior to execution. Figure 1a shows the diagram of the automated system in which the 1 kOhm resistor represents the electrochemical cell. The function of the relays is to allow the change of voltage in the system in addition to controlling the reaction time using the Arduino microcontroller. The diode, close to the low-voltage source, acts as a safety device. To apply a low voltage, an Arduino digital port was used that provides up to 5 V using the PWM.

Preliminary analyses of the electrical system’s operation were necessary. The initial scope was simulated using the Tinkercad platform,⁴⁹ which allows feasibility and functionality tests of the circuit to be carried out virtually. The simulations confirmed that the system operated correctly, as it was able to deliver a low-voltage output (Figure S1, SI section), followed by a high-voltage output using relay modules that activated a lamp, representing the application of voltage to the electrodes (Figure S1). The full list of components and their approximate costs is provided in the SI section. The total cost of the system is approximately US$ 48.50, based on commonly available components.

The first physical version of the system was assembled on a prototype-board (Figure S1). The experimental tests demonstrated satisfactory performance, with the system successfully delivering the expected voltages during both low and high-voltage stages. Ultimately, the power source met the required specifications and automated the voltage switching process according to the defined timing. Figure S2 (SI section) presents two images of the experimental setup, illustrating the prototype in operation. The presence of gas bubbles on the electrode surface confirms the occurrence of electrochemical reactions. Therefore, the reactor is suitable for application in the electrochemical exfoliation of graphite for GO production.

To enable a user-friendly system, the electronic components were soldered and arranged in printed circuit board. A support case was 3D-printed and designed in a compartmentalized manner, allowing the reaction section to be separated from the electrical system, as shown in Figure 3. All the 3D-printer STL files are available in the SI section (Figure S3). Additionally, an LCD system was incorporated, featuring buttons that allow adjustment of parameters without the need to change the Arduino code manually. The automated system was programmed to receive as parameters from the users the intensity of the low voltage (in V), the low voltage time (in s), the high voltage time (in s), keeping the high voltage value fixed at 12 V, and the different cycles, according to the program shown in the SI section.

Figure 3
Photographic images of the designed electrochemical reactor.

The system’s lid isolates the reaction chamber. It was printed with four holes: two for the electrodes, keeping a fixed distance of 1.5 cm, and two redundant ones, which can be sealed or used for sensor insertion, such as a thermometer, level sensor, pH, and others. Here, one of them was used for a temperature sensor for dynamic temperature measurement. A function to export temperature values over time was included in the script and the read values were recorded using the CoolTerm software.

Validation of the low-cost electrochemical reactor

In order to validate the reactor developed and optimize the most relevant variables of the system, such as electrolyte concentration, applied potential difference, and the durations of both high and low voltage stages, a factorial design was implemented. This approach enables the identification of optimal parameters for the experimental conditions. For this study, as shown in Table 1, a 2^4-1 factorial design was chosen. The initial characterization of GO sheet was performed using optical microscopy to evaluate the average size and their optical contrast, that is related to the sheet thickness. For each synthesis corresponding to the factorial design tests described in Table 1, analyses were conducted using an optical microscope with both transmission and reflection modes to obtain a large sample size, approximately 100 GO sheets per test, ensuring reliable average size measurements, shown in Table 3. All optical images and the corresponding tables with extracted sheet sizes are available in the OSF repository. Representative images from each test are shown in the SI section (Figures S4 to S11) along with the complete data set for all GO sheet measured. Figure 4 shows images for test 4 (Table 1) that used 1 mol L^-1 of electrolyte, 2 V for the low voltage during 60 s and 1800 s for the high voltage time duration. Figures 4a, 4b, and 4c show the precipitate of the first, second and third centrifugation at 1000, 2000 and 3000 rpm, respectively. In all fractions it is possible to visualize the GO sheets with different optical contrasts.

Table 3
Average lateral size of graphene oxide (GO) nanosheets obtained from each synthesis condition defined in the factorial design. The values represent the mean lateral size ± standard deviation for GO flakes collected after centrifugation at 1000, 2000, and 3000 rpm. The results highlight how synthesis parameters-including electrolyte concentration, applied voltage, and voltage durations-affect the size distribution of GO flakes. The data also reflect the role of centrifugation speed in separating flakes by size and thickness

Synthesis number	Average lateral size of GO sheets / μm
	Precipitated at 1000 rpm	Precipitated at 2000 rpm	Precipitated at 3000 rpm
1	6.98 ± 0.39	4.72 ± 0.21	4.82 ± 0.22
2	5.68 ± 0.34	11.12 ± 0.35	9.57 ± 0.60
3	7.16 ± 0.74	9.21 ± 0.40	5.44 ± 0.24
4	13.94 ± 1.35	5.45 ± 0.21	11.38 ± 0.48
5	20.30 ± 1.00	8.69 ± 0.44	7.27 ± 0.33
6	6.74 ± 0.44	9.30 ± 0.37	4.84 ± 0.36
7	9.08 ± 0.48	9.25 ± 0.42	8.53 ± 0.41
8	6.84 ± 0.35	9.16 ± 0.48	6.39 ± 0.30
9	7.20 ± 0.29	6.41 ± 0.30	8.19 ± 0.34

Figure 4
Optical and AFM characterization of GO nanosheets obtained in synthesis test 4. (a-c) Optical microscopy images of GO flakes collected after centrifugation at (a) 1000 rpm, (b) 2000 rpm, and (c) 3000 rpm, showing flakes with varying thickness and lateral sizes. (d) AFM topography image of representative GO flakes from the 3000 rpm fraction. (e) and (f) Height profiles extracted from the AFM image, indicating nanosheet thicknesses of 7.34 and 4.38 nm, respectively. The results confirm the successful formation of few-layer to multilayer graphene oxide with size and thickness distribution dependent on centrifugation speed.

Fractional centrifugation separates GO flakes based on density, thickness, lateral size, and surface chemistry.⁵⁰,⁵¹ This technique exploits differences in sedimentation coefficients arising from variations in particles density and hydrodynamic radius. Thus, larger flakes sediment at lower speeds, while smaller flakes require higher speeds. As shown in Figure 4b, the flakes expected to exhibit a larger average size than those in (Figure 4c). However, Table 3 data indicate that centrifugation at 3000 rpm yields slightly larger flakes, likely due to differences in layer thickness. Multilayer GO flakes possess higher molecular weight and consequently precipitate at lower centrifugation speeds. Thus, despite their larger lateral dimensions, flakes isolated at higher speeds (e.g., in Figure 4c) likely feature fewer layers, as evidenced by their greater transparency. Figure 4d presents an AFM topography image of representative GO flakes from test 4, precipitated at 3000 rpm, confirming the successful formation of GO sheets with thicknesses of 7.34 nm (Figure 4e) and 4.38 nm (Figure 4f). AFM analysis of multiple flakes indicates that the thickness of the GO obtained in test 4 ranges from approximately 4 to 15 nm (Figures 4 and S12-S17), further supporting the presence of few-layered to multilayered GO sheets.

The multiple linear regression model developed to describe the lateral size of GO sheets under varying electrochemical exfoliation conditions presented an excellent statistical fit, with an R² of 0.998 and an adjusted R² of 0.980. These values indicate that the model explains nearly all the variability observed in the experimental data. However, despite this high goodness-of-fit, caution is warranted in interpreting the results due to the very limited number of degrees of freedom available (only one residual degree of freedom), which undermines the statistical reliability of the coefficient estimates and p-values.

Among the main effects and two-way interactions considered in the model, only the interaction between EC and LVt was statistically significant (p = 0.038), as shown in the Pareto chart in Figure 5a. This term exhibited a strong negative coefficient (-1.91), suggesting a substantial synergistic effect whereby specific combinations of EC and LVt lead to a pronounced reduction in GO sheet size. All other terms, including the main effects of EC, LV, LVt, and HVt, as well as their interactions, showed p-values above 0.05 and were therefore not statistically significant. However, several of these terms (e.g., EC, LV, and EC:HVt) exhibited borderline significance (p ≈ 0.09 0.13), suggesting they may become significant with a larger dataset.

Figure 5
(a) Pareto plot of main factors and two-way interactions affecting GO lateral size at 3000 rpm, showing EC × LVt as the dominant effect, followed by HVt. (b) Response surface of EC vs. LVt (LV, HVt fixed at 0) reveals a strong compensatory interaction: large GO sizes occur with high EC + short LVt or low EC + long LVt. (c) Response surface of EC vs. HVt (LV, LVt fixed at 0) shows high EC and short HVt favor larger GO flakes. (d) Response surface of LVt vs. HVt (EC, LV fixed at 0) indicates longer LVt and shorter HVt enhance flake growth.

The response surface depicting the effects of EC and LVt reveals a notable interaction between these two parameters (Figure 5b). Rather than exerting independent, additive effects on the GO sheet size, EC and LVt demonstrate a compensatory relationship. At high EC values, even short low-voltage durations lead to the formation of large GO flakes, likely due to enhanced ionic conductivity and more effective intercalation. In contrast, when EC is low, longer LVt appears to mitigate the reduced ionic strength by allowing sufficient time for electrochemical processes to proceed. Interestingly, when both EC and LVt are either simultaneously high or simultaneously low, the GO lateral size is minimized, suggesting suboptimal exfoliation due to either over-treatment or inefficiencies in ion transport. This saddle-shaped surface highlights the importance of balancing chemical and electrical factors in the exfoliation process, and confirms that optimized GO production cannot be achieved by tuning each parameter in isolation.

The response surface depicted in Figure 5c illustrates the combined influence of EC and HVt on the lateral size of GO sheets, with low-voltage amplitude (LV) and low-voltage exfoliation time (LVt) fixed at their central coded levels. The model reveals a clear interaction between EC and HVt, as evidenced by the curved and tilted shape of the response surface. At low electrolyte concentrations (EC = -1), increasing the high-voltage time leads to a marked reduction in GO sheet size, suggesting that prolonged high-voltage exposure under insufficient ionic strength may cause structural damage or over-fragmentation of the exfoliated flakes. Conversely, when the electrolyte concentration is high (EC = +1), the GO size remains relatively large even as HVt increases, indicating that a higher ionic environment may buffer the negative effects of extended high-voltage pulses.

The response surface generated for the interaction between LVt and Hvt in Figure 5d provides further insight into the dynamic balance required for effective electrochemical exfoliation of graphite. When electrolyte concentration and low-voltage amplitude are held constant, it becomes evident that a longer LVt combined with a shorter HVt promotes the formation of larger GO sheets. This suggests that sufficient time under low-voltage conditions is critical to facilitate intercalation and swelling of the graphite structure, effectively priming the material for exfoliation. In contrast, excessively long high-voltage pulses appear detrimental, likely due to mechanical or oxidative degradation that fragments the exfoliated sheets.

Overall, the most favorable condition for producing large GO sheets appears to be the combination of high electrolyte concentration and short high-voltage duration. This implies that while a strong ionic medium facilitates effective intercalation and delamination, excessive energy input from long high-voltage pulses may be detrimental to the integrity of the exfoliated layers, especially in less conductive solutions. The observed trend emphasizes that GO size is maximized not by aggressive exfoliation but by a delicate sequence, intercalation followed by a brief, controlled exfoliation pulse. Thus, the optimization of voltage staging and timing emerges as a key factor for tailoring GO morphology in electrochemical synthesis.

Despite the promising R² values, the high condition number of the model (ca. 1.7 × 10¹⁶) signals severe multicollinearity among the predictors, likely due to the factorial design and limited sample size (n = 9). This multicollinearity inflates the variance of the coefficient estimates and further reduces the reliability of individual p-values. Moreover, the residual analysis revealed moderate non-normality and heteroscedasticity, as indicated by the Omnibus and Jarque-Bera tests, residual analysis showed a random distribution with no evident patterns (Figure S18, SI section), suggesting that measurement variability, while not explicitly isolated, was limited and not systematic. The Durbin-Watson (ca. 2.6) statistic further indicated no autocorrelation in the residuals, which supports the assumption of independent errors.

The residual standard deviation of the regression model was calculated as 0.1125 µm, reflecting the unexplained variation in GO sheet size after accounting for the experimental factors and their interactions. To determine whether instrumental noise significantly contributes to this variability, we compared it with the estimated instrumental error, determined as ± 0.038 µm based on the pooled standard error from 100 flake measurements per synthesis, see SI section. By squaring and comparing their variances, we estimate that instrumental noise accounts for approximately 11% of the total residual variance: (0.038)²/(0.1125)² ≈ 0.114

This indicates that while the measurement system introduces some uncertainty, it is not the dominant source of variability in the experimental data. Most of the residual variation likely arises from other factors such as local sample heterogeneity, minor process fluctuations, or unmodeled higher-order effects. These results support the reliability of the measurement system and confirm that the observed trends in GO sheet size are primarily driven by the experimental conditions rather than instrumentation limitations.

In summary, while the model provides an excellent descriptive fit to the experimental data, its predictive reliability is compromised by overfitting and multicollinearity. The significant EC:LVt interaction indicates a potentially important synergistic effect worth exploring further, but additional experiments with increased sample size are essential to validate this finding and allow more robust inference about the influence of other synthesis parameters.

Effect of the number of cycles

The factorial design showed that GO size is maximized by staged intercalation followed by brief, controlled exfoliation, highlighting voltage timing as key to tailoring morphology. Therefore, we chose to investigate the effect of the number of voltage cycles on the electrochemical exfoliation of graphite to produce GO due to its critical role in modulating the physicochemical properties of the resulting material. The electrochemical exfoliation process involves alternating phases of low and high voltage, which drive ionic intercalation and subsequent exfoliation through gas evolution and electrochemical reactions.³³,³⁴,⁵² The number of applied cycles directly influences the extent of intercalation, oxidation, and exfoliation efficiency.³⁷,⁵² Therefore, to investigate the influence of cycle number on electrochemical exfoliation, we used the parameters of the middle point of the factorial design in Table 1 comparing two distinct cycle regimes: a single-cycle protocol (N = 1), consisting of a low voltage cycle and a high voltage cycle both with 900 s duration, and a multi-cycle protocol (N = 10), where each cycle consisted of a low-voltage stage (2 V, 90 s), followed by a high-voltage stage (12 V, 90 s). In both cases the total duration of the electrolysis is the same.

During synthesis, the temperature evolution as a function of reaction time was recorded (Figure 6). A DS18B20 temperature sensor (range -55 to 150 °C; sensitivity ca. 10 mV °C^-1) connected to the Arduino monitored thermal variations within the reactor. In the dynamic temperature profiles for a single cycle (N = 1), a steady linear temperature increase was observed, rising from ca. 25 to ca. 45 °C over 30 min. This gradual rise reflects continuous energy input from the applied current, which does not allow the system to reach thermal equilibrium.

Figure 6
Temperature profiles recorded during the electrochemical exfoliation of graphite for graphene oxide synthesis using N = 1 and N = 10 cycle protocols. The single-cycle process (N = 1) exhibits a continuous linear increase in temperature due to sustained energy input, reaching approximately 45 °C. In contrast, the ten-cycle process (N = 10) shows a stepwise temperature pattern, with stable temperatures during low-voltage stages and sharp increases during high-voltage stages, followed by gradual cooling. This behavior highlights the influence of cycling strategy on thermal dynamics during the exfoliation process.

In contrast, the ten-cycle protocol (N = 10) featured alternating low and high voltage stages, with low voltage active for roughly 50% of the total time. During the low-voltage phases, the temperature remained relatively constant, whereas the onset of high voltage triggered a rapid temperature increase due to an influx of electrical energy. As shown in Figure 6b, this resulted in a stepwise temperature profile that reflects the alternation between voltage stages, followed by gradual cooling during transitions as a result of convective heat exchange with the ambient environment.

The C 1s spectrum of sample N = 1 reveals clear signatures of uncontrolled oxidation (Figure 7b). The main sp² carbon peak at ca. 284.94 eV, associated with C-C bonds in aromatic domains, is strongly suppressed relative to the oxygenated components. The C-OH component at ca. 285.7 eV is dominant, indicating a high density of hydroxyl groups. Additional oxygenated species, including C-O (epoxy/ether) at ca. 286.9 eV, C=O at ca. 288.65 eV, and O-C=O at ca. 290.32 eV, are also present with substantial intensity. The shake-up satellite (π-π*) in the 291-292 eV range is barely visible, consistent with disruption of the conjugated sp² network. This combination of features points to the breaking of C=C bonds and the transformation of the carbon framework into a more amorphous structure dominated by hydroxyl functionalities. The O 1s spectrum further in Figure 7c supports this interpretation, showing a strong contribution from C-OH/C-O species, consistent with extensive hydroxylation. The data suggest that GO produced with one cycle protocol underwent a non selective, high-rate oxidation process, likely driven by rapid temperature increase and water oxidation, leading to over-oxidation and loss of graphitic character.

Figure 7
XPS C 1s and O 1s spectra of samples N1 and N10. Sample N1 shows suppressed sp² C-C (ca. 284.94 eV) relative to oxygenated species, with C-OH (ca. 285.7 eV) as the dominant component, weak π-π* satellite, and an O 1s spectrum rich in C-OH/C-O, consistent with uncontrolled, high-rate oxidation and loss of graphitic order. In contrast, N10 exhibits a strong sp² peak, lower relative oxygenated components, visible π-π* satellite, and reduced hydroxyl content in O 1s, indicating controlled oxidation and preservation of the conjugated carbon framework.

In contrast, the C 1s spectrum of sample N = 10 shows characteristics of a better-preserved graphitic structure (Figure 7e). The sp² peak at ca. 284.94 eV remains the most intense component, surpassing the combined intensity of the oxygenated species. The C-OH and other oxygenated components (C-O, C=O, O-C=O) are present but in significantly lower relative proportions compared to N = 1. Moreover, the π-π* shake-up satellite around 291-292 eV is clearly visible, reflecting a higher degree of aromatic conjugation and lower disruption of the sp² lattice. The O 1s spectrum of N = 10 also exhibits a reduced predominance of hydroxyl/ether groups relative to N = 1, suggesting a lower oxygen content overall (Figure 7f). These spectral features are consistent with controlled oxidation, preserving a substantial portion of the conjugated carbon framework and minimizing the formation of highly oxidized functionalities.

Taken together, the comparison between N = 1 and N = 10 highlights the impact of oxidation rate and selectivity on the chemical state of carbon. N = 1 represents a case of over-oxidation with extensive hydroxylation and structural disorder, while N = 10 retains a predominantly graphitic character with a balanced and controlled incorporation of oxygen-containing groups.

FTIR analysis (Figure S19, SI section) supports the XPS findings for both samples. N1 shows a broad, intense O-H band (3373-3230 cm^-1), strong C=O (1734 cm^-1), and intense C-O-C/C-O absorptions (1215/1060 cm^-1), alongside a weakened aromatic C=C (ca. 1639 cm^-1), indicating extensive hydroxylation and oxidation-induced disruption of the sp² network. In contrast, N = 10 exhibits weaker O-H and carbonyl bands, reduced C-O-C/C-O intensities, and a more pronounced C=C feature, consistent with lower oxygen content and better preservation of the graphitic structure. Together, these results confirm that N = 1 underwent uncontrolled over-oxidation, whereas N = 10 experienced controlled oxidation with retention of sp² domains.

TGA (Figure S20, SI section) further supports the structural differences observed by XPS and FTIR. Sample N1 exhibits a large initial mass loss (ca. 44%) below 150 °C, attributed to the release of physisorbed water and moisture bound to abundant hydrophilic oxygen-containing groups. Subsequent stepwise losses between 150-400 °C correspond to the decomposition of hydroxyl, epoxy, and ether functionalities, while a gradual decline up to 900 °C reflects the breakdown of more stable carbonyl and defect-rich domains. N = 10 also shows a high initial weight loss due to physisorbed water; however, shows a minimal decomposition at temperatures between 150 and 350 ºC, indicating lower oxygen content and a more thermally stable, graphitic framework. These results are consistent with the XPS and FTIR data, which reveal that N1 is over-oxidized and rich in C-OH/C-O species, whereas N = 10 retains a higher fraction of sp² carbon with fewer oxygenated defects.

Therefore, the single-cycle protocol (N = 1) reached elevated temperatures, and this correlated with an increase in defect density compared to the ten-cycle protocol (N = 10), as confirmed by zeta potential and Raman spectroscopy. In the Raman spectra of GO, the two most prominent features are the D and G bands (Figure S21, SI section). The G band, typically observed around ca. 1580 cm^-1, arises from the first-order scattering of the E_2g phonon mode at the Brillouin zone center and is associated with the in-plane vibration of sp²-bonded carbon atoms in a hexagonal lattice. Its position and intensity reflect the degree of graphitization and the preservation of the conjugated sp² network. The D band, located near ca. 1350 cm^-1, is a disorder-induced feature that becomes Raman active in the presence of defects, edges, or disruptions in the sp² lattice, such as those introduced by oxygen-containing functional groups in GO. The intensity ratio I_D/I_G is commonly used to assess the defect density and the average size of the sp² domains: higher ratios correspond to increased disorder and smaller sp² domain sizes, while lower ratios indicate a more ordered graphitic structure. In GO, the presence of a strong D band relative to the G band reflects significant structural defects and oxidation-induced disruption of the carbon lattice. The observed I_D/I_G ratios were 0.86 for N = 1 and 0.76 for N = 10, indicating smaller structural disorder with increased cycling (Figure S21). These results align with the XPS and FTIR data, which show that N = 1 underwent uncontrolled over-oxidation with extensive hydroxylation and disruption of the sp² network, whereas N = 10 retained more graphitic character.

Zeta potential measurements reveal that samples produced with N = 10 cycles exhibited values around -49 mV, whereas those from N = 1 cycle hovered near -30 mV. A zeta potential below -30 mV is commonly considered the threshold for good colloidal stability, with more negative values indicating stronger electrostatic repulsion and, hence, enhanced dispersion stability. In our recent work,⁵³ we demonstrated that the zeta potential of GO suspensions depends on both the degree of oxidation and the concentration. We showed that concentrations around 100 mg L^-1 resulted in smaller agglomerates, more negative zeta potential values, and consequently more stable suspensions. Therefore, the zeta potential reflects not only the overall oxygen content but also the spatial distribution of functional groups and their exposure to the solution. In sample N = 1, the predominance of hydroxyl groups and the presence of disrupted domains likely reduce the effective surface charge density, leading to agglomeration and a decrease in the absolute value of the zeta potential due to limited exposure of charged groups. In contrast, the controlled incorporation of oxygenated groups in N = 10 produces a more homogeneous surface charge distribution and enhanced colloidal stability, with greater exposure of oxygen-containing groups to the solution.

These results indicate that controlling the temperature and the electrochemical cycling can effectively reduce structural defects, making it a viable strategy for controlling the oxidation degree of GO during the electrochemical exfoliation process. However, when reaction time is extended or current density is increased, for example, by using electrodes with a larger surface area, active temperature control becomes crucial to maintain product quality within the desired specifications. In fact, preliminary experiments using larger electrodes (2 cm × 5 cm and 5 cm × 5 cm) showed that the N = 1 cycle condition led to temperatures reaching approximately 60 °C, highlighting the importance of thermal regulation in scaled-up or intensified electrochemical exfoliation setups.

The morphological characterization of the samples was performed by SEM and AFM microscopy (Figure 8). SEM images of GO sheets deposited on copper substrate obtained for N = 1 and N = 10 cycles reveal the effect of cycle numbers during the synthesis process in the shape, dimensions, and quality of formed GO. For N = 1, low density of irregular GO sheets and a high number of carbon/graphite particles are observed (Figure S20, SI section), corroborating the XPS results. With an increase in the number of cycles (N = 10), larger (mean size of 35 μm, Figure 8b) and more defined GO sheets were obtained (Figure 8a). Figure 8c presents an AFM topography image of representative GO flakes exfoliated with ten cycles, confirming the successful formation of GO sheets with thicknesses between 3 and 6 nm (Figures 8d-8g), supporting the presence of few-layered GO sheets.

Figure 8
SEM and AFM characterization of graphene oxide (GO) nanosheets synthesized with N = 10 electrochemical cycles. (a) SEM image showing large, well-defined GO sheets with reduced presence of residual graphite particles compared to the N = 1 condition. (b) Histogram of GO sheet lateral size distribution obtained from SEM image analysis. (c) AFM topography image of a representative GO sheet confirming successful exfoliation. (d-g) AFM height profiles extracted from (c), indicating nanosheet thicknesses ranging from approximately 3 to 6 nm, consistent with few-layer GO.

We performed three additional synthesis batches of GO with N = 10, under identical experimental conditions, and characterized each batch using Raman spectroscopy to assess reproducibility (Figure 9). To quantify the degree of similarity between replicates, we developed a Python-based statistical analysis workflow. This code imports each Raman spectrum, interpolates them onto a common wavenumber grid, and calculates (i) a pairwise correlation matrix to assess spectral shape similarity, (ii) the relative standard deviation (RSD) across wavenumbers to quantify intensity variability, and (iii) a principal component analysis (PCA) to visualize and interpret variance among replicates in reduced dimensional space. The python code and all the Raman spectra used are available in the OSF repository.

Figure 9
(a) Mean Raman spectrum (solid line) of graphene oxide synthesized, with shaded area representing the standard deviation across replicates, illustrating the high consistency in spectral shape. (b) Principal component analysis (PCA) plot showing the distribution of replicates in PC1-PC2 space, where the tight clustering confirms strong reproducibility of the synthesis, with only minor variations attributable to experimental factors such as laser alignment or sample positioning.

The statistical results reveal an extremely high level of spectral agreement. The pairwise correlation coefficients between all spectra are ≥ 0.9977, indicating that the spectral profiles-peak positions, relative intensities, and baseline shapes-are almost identical across all replicates. Such high correlations strongly suggest that the Raman-active structural features of the GO are highly reproducible under the applied synthesis conditions.

The average RSD across wavenumbers was 38.10%, which is higher than the < 10-20% typically expected. While still acceptable for carbon-based materials, where microscopic heterogeneity and flake-to-flake variability are expected, this elevated RSD likely reflects small experimental factors, such as minor variations in laser spot alignment, sample positioning, or laser coupling efficiency during acquisition. These effects influence overall intensity but do not substantially alter the spectral fingerprint.

PCA results further support these findings (Figure 9b). The spectra form a tight cluster in the PC1-PC2 space, with only small separations between replicates. PC1 accounts for the dominant variance, likely corresponding to overall intensity scaling or subtle shape changes, while PC2 captures minor variations, possibly related to local baseline shifts or noise. The compact distribution of points indicates that no major structural differences exist among the replicates, and the observed spread is consistent with minor experimental variability.

Overall, the combination of near-unity spectral correlations, PCA clustering, and RSD analysis demonstrates that the GO synthesis method delivers highly reproducible Raman spectral features. The slight intensity variability observed does not reflect chemical or structural changes, but rather small, unavoidable experimental fluctuations during measurement.

By varying the number of cycles, it is possible to control the degree of oxidation, the thickness of the GO sheets, and the defect density in the carbon lattice. A lower number of cycles tends to result in uncontrolled oxidation and exfoliation, yielding thicker, multilayered flakes and more amorphous carbon with more defects. Conversely, a higher number of cycles increases the ion intercalation in the graphite layers, promoting more controllable exfoliation and oxidation. This can lead to thinner sheets with higher aromatic content. Understanding this relationship is essential for optimizing the synthesis parameters to tailor the GO properties for specific applications, whether aiming for high conductivity, better dispersibility, or mechanical reinforcement.

The electrochemically exfoliated GO exhibited an average lateral size of between 10 and 40 μm and flake thickness ranging from 3 to 6 nm, depending on the number of applied voltage cycles. Raman spectroscopy revealed I_D/I_G ratios between 0.76 and 0.86, indicating moderate structural disorder consistent with partial oxidation. In contrast, in our previous work,⁵³ the GO produced via the modified Hummers method showed broader lateral size distributions (1-100 μm) and more intense D-bands in Raman spectra, suggesting a higher defect density associated with extensive basal-plane oxidation, especially for highly oxidized GO, that showed I_D/I_G values close to 1. Additionally, Nagaoka et al.³ reported a high-quality GO synthesized using a refined Hummers method, resulting in flakes with broad lateral size distribution (500 nm-50 μm), thicknesses of 2-4 nm, and I_D/I_G ratios below 0.5, indicating lower defect densities and greater structural preservation.

While the Hummers method can yield thinner and less defective flakes under optimized conditions, the size distribution is higher and it involves harsh chemical oxidants (e.g., KMnO₄, concentrated acids), generates hazardous waste. In contrast, the electrochemical approach developed here provides programmable control over the intercalation and exfoliation stages through voltage cycling, improving the size distribution, operates in an environmentally friendly manner without the use of strong oxidants or acids, and enables scalable and cost-effective implementation, with the entire system constructed for under US$50.

Challenges in scaling up the system

Although graphene and its derivatives present a wide range of applications, certain factors have hindered the widespread adoption of related technologies. Among these factors are unit price, productivity, imperfections, contamination, non-uniform size and thickness, nanoproduct stability, and associated pollution. The choice of method for producing GO depends on a trade-off between these factors.

While liquid-phase methods, such as Hummers method or electrochemical exfoliation, produce material at a lower cost and offer a variety of methodological options, they tend to yield products with low durability, poor uniformity, and the potential for pollution due to the large volume of waste generated.

Among the key limitations for scaling up the production of graphene and its derivatives, several technical and economic challenges stand out. One of the primary concerns is the high cost of production. For instance, the price per gram of graphene produced via chemical vapor deposition (CVD) can range from approximately $ 50.00 to over $ 2,000.00, making it economically costly for large scale applications. In addition to cost, low productivity is another constraint of the CVD method, as it typically enables the deposition of only monolayer or few-layer crystals, thus limiting its use to specific technological sectors.

Furthermore, imperfections and contamination during the CVD process can negatively affect the quality of the final product. The use of high temperatures restricts its integration with silicon-based devices, and the transfer steps may introduce contaminants or cause mechanical damage. The method also involves numerous variables, making it difficult to reproduce results consistently.

Liquid-phase methods, while offering lower costs and greater flexibility, also face significant challenges. One such issue is the non-uniform size and thickness of the resulting flakes. The crystallinity of the starting graphite material and the complexity of the exfoliation process contribute to difficulties in standardizing the lateral dimensions and thicknesses of the GO sheets. Additionally, low durability and instability are concerns, as the presence of structural defects and insufficient oxidation can compromise the long-term performance of the GO produced by these methods.

Lastly, environmental concerns must be considered. Liquid-phase synthesis often involves the use of strong oxidizing agents and acids (as in the Hummers method), or requires large quantities of water and energy (as in electrochemical methods), leading to potential pollution and waste management issues.

It is important to note that while both CVD and liquid-phase methods have their drawbacks, liquid-phase approaches can yield sufficiently high-quality material for various applications when proper control of synthesis parameters is maintained.

In the context of our low-cost electrochemical system, scaling up would require increasing the electrode surface area and electrolyte volume while maintaining uniform current distribution and thermal management. In our preliminary trials, the use of larger inert electrodes led to uncontrolled heating, darker dispersions, and likely over-exfoliation, suggesting that uncontrolled scale-up could compromise product quality. These results highlight the need for future studies to optimize reactor geometry, mixing efficiency, and heat dissipation for larger batch volumes.

The robustness of the system to variations in graphite feedstock is another key factor for industrial implementation. Liquid-phase methods, including our electrochemical approach, are particularly sensitive to the crystallinity, particle size, and purity of the graphite precursor. These factors affect the intercalation dynamics, oxidation level, and exfoliation behavior, which in turn influence the morphology and surface chemistry of the resulting GO sheets. While our reactor demonstrated reproducible performance with high-purity natural graphite flakes, we acknowledge that variations in feedstock could introduce inconsistencies.

Despite these limitations, our system offers a promising foundation for scalable GO production, with advantages in cost, modularity, and programmability. With targeted engineering improvements and process control strategies, the reactor can be adapted for higher throughput while maintaining product quality and environmental sustainability.

Conclusions

In this work, we developed and validated a low-cost, fully automated electrochemical reactor for the GO synthesis via electrochemical exfoliation of graphite. The system, controlled by an Arduino-based potentiostat, allows precise adjustment of synthesis parameters such as applied voltage, voltage duration, and number of cycles. This flexibility enables systematic studies of how these parameters influence the properties of the produced GO.

Our investigation demonstrated that the number of voltage cycles plays a crucial role in controlling the structural and morphological characteristics of GO. Increasing the number of cycles from N = 1 to N = 10 led to more effective exfoliation, producing larger, thinner nanosheets with smaller degrees of oxidation and enhanced colloidal stability, as confirmed by SEM, AFM, XPS, Raman spectroscopy, and zeta potential analysis. Conversely, lower cycle numbers resulted in thicker, multilayered GO sheets with higher defect density and higher amorphous carbon content.

Temperature monitoring revealed that the thermal profile of the process is directly influenced by the cycling strategy, highlighting the importance of temperature control, especially for scaled-up or prolonged syntheses.

Overall, this reactor design provides a versatile platform for the scalable and sustainable production of graphene oxide, offering significant advantages over conventional methods. Its modular architecture allows for the integration of additional sensors-such as for pH, conductivity, and pressure-enabling real-time monitoring and process optimization. This approach opens pathways for further studies on the synthesis of graphene derivatives and the development of customized nanomaterials tailored for specific applications in energy storage, catalysis, sensing, and composite materials.

Supplementary Information

Supplementary Information

Supplementary information (additional experimental details, Arduino source code, optical microscopy images, AFM topographies, and Raman spectra supporting the findings of this study) is available free of charge at http://jbcs.sbq.org.br as PDF file.

Data Availability Statement

All data supporting the findings of this study are publicly available and have been deposited in the Open Science Framework (OSF) under the Creative Commons CC-BY license, ensuring open access and proper attribution. The data repository includes raw experimental data, Arduino source code, optical microscopy images, AFM measurements, Raman spectra, and all supplementary materials associated with this article. The dataset can be accessed via the following [Link].

Any additional information or clarification can be obtained from the corresponding author upon reasonable request.

Acknowledgments

This work was supported by the Brazilian Nanocarbon Institute of Science and Technology (INCT/Nanocarbono), Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES). DG acknowledges financial support from Rio de Janeiro State Foundation (FAPERJ - grant Nos. E-26/210.296/2022, E-26/211.464/2021, and E-26/201.254/2022), Serrapilheira Institute (grant No. Serra - R-2012-37959), and Financiadora de Estudos e Projetos - Finep grant number 01.22.0208.00.

References

• ¹ Pei, S.; Wei, Q.; Huang, K.; Cheng, H.-M.; Ren, W.; Nat. Commun. 2018, 9, 145. [Crossref]
  » Crossref
• ² Allen, M. J.; Tung, V. C.; Kaner, R. B.; Chem. Rev. 2010, 110, 132. [Crossref]
  » Crossref
• ³ Nagaoka, D. A.; Grasseschi, D.; Domingues, S. H.; Diamond Relat. Mater. 2021, 120, 108616. [Crossref]
  » Crossref
• ⁴ Liu, W.-W.; Aziz, A.; ACS Omega 2022, 7, 33719. [Crossref]
  » Crossref
• ⁵ Yang, D.; Velamakanni, A.; Bozoklu, G.; Park, S.; Stoller, M.; Piner, R. D.; Stankovich, S.; Jung, I.; Field, D. A.; Ventrice, C. A.; Ruoff, R. S.; Carbon 2009, 47, 145. [Crossref]
  » Crossref
• ⁶ Gupta, V.; Sharma, N.; Singh, U.; Arif, Mohd.; Singh, A.; Optik 2017, 143, 115. [Crossref]
  » Crossref
• ⁷ Carvalho, A.; Costa, M. C. F.; Marangoni, V. S.; Ng, P. R.; Nguyen, T. L. H.; Castro Neto, A. H.; Nanomaterials 2021, 11, 560. [Crossref]
  » Crossref
• ⁸ Gómez-Navarro, C.; Meyer, J. C.; Sundaram, R. S.; Chuvilin, A.; Kurasch, S.; Burghard, M.; Kern, K.; Kaiser, U.; Nano Lett. 2010, 10, 1144. [Crossref]
  » Crossref
• ⁹ Sharma, N.; Tomar, S.; Shkir, M.; Kant Choubey, R.; Singh, A.; Mater. Today: Proc. 2021, 36, 730. [Crossref]
  » Crossref
• ¹⁰ Bao, H.; Pan, Y.; Ping, Y.; Sahoo, N. G.; Wu, T.; Li, L.; Li, J.; Gan, L. H.; Small 2011, 7, 1569. [Crossref]
  » Crossref
• ¹¹ Kalyan, S.; Bhosale, A.; Patil, P. D.; Bahadure, N. B.; Appl. Surf. Sci. Adv. 2022, 11, 100285. [Crossref]
  » Crossref
• ¹² Bressi, A. C.; Dallinger, A.; Steksova, Y.; Greco, F.; ACS Appl. Mater. Interfaces 2023, 15, 35788. [Crossref]
  » Crossref
• ¹³ Raghavendra, P.; Vishwakshan Reddy, G.; Sivasubramanian, R.; Sri Chandana, P.; Subramanyam Sarma, L.; Int. J. Hydrog. Energy 2018, 43, 4125. [Crossref]
  » Crossref
• ¹⁴ Kumar, D.; Lee, A.; Lee, T.; Lim, M.; Lim, D.-K.; Nano Lett. 2016, 16, 1760. [Crossref]
  » Crossref
• ¹⁵ Darabdhara, G.; Boruah, P. K.; Borthakur, P.; Hussain, N.; Das, M. R.; Ahamad, T.; Alshehri, S. M.; Malgras, V.; Wu, K. C.-W.; Yamauchi, Y.; Nanoscale 2016, 8, 8276. [Crossref]
  » Crossref
• ¹⁶ Vivaldi, F. M.; Dallinger, A.; Bonini, A.; Poma, N.; Sembranti, L.; Biagini, D.; Salvo, P.; Greco, F.; Di Francesco, F.; ACS Appl. Mater. Interfaces 2021, 13, 30245. [Crossref]
  » Crossref
• ¹⁷ Wang, J.; Zhang, J.; Han, L.; Wang, J.; Zhu, L.; Zeng, H.; Adv. Colloid Interface Sci. 2021, 289, 102360. [Crossref]
  » Crossref
• ¹⁸ Anegbe, B.; Ifijen, I. H.; Maliki, M.; Uwidia, I. E.; Aigbodion, A. I.; Environ. Sci. Eur. 2024, 36, 15. [Crossref]
  » Crossref
• ¹⁹ Shamsaei, E.; de Souza, F. B.; Yao, X.; Benhelal, E.; Akbari, A.; Duan, W.; Constr. Build. Mater. 2018, 183, 642. [Crossref]
  » Crossref
• ²⁰ Asthana, N.; Pal, K. In Nanofabrication for Smart Nanosensor Applications; Pal, K.; Gomes, F., eds; Elsevier, 2020, p. 349. [Link]
  » Link
• ²¹ Eda, G.; Emrah Unalan, H.; Rupesinghe, N.; Amaratunga, G. A. J.; Chhowalla, M.; Appl. Phys. Lett. 2008, 93, 233502. [Crossref]
  » Crossref
• ²² Hummers Jr., W. S.; Offeman, R. E.; J. Am. Chem. Soc. 1958, 80, 1339. [Crossref]
  » Crossref
• ²³ Kang, J. H.; Kim, T.; Choi, J.; Park, J.; Kim, Y. S.; Chang, M. S.; Jung, H.; Park, K. T.; Yang, S. J.; Park, C. R.; Chem. Mater. 2016, 28, 756. [Crossref]
  » Crossref
• ²⁴ Muzyka, R.; Kwoka, M.; Smędowski, Ł.; Díez, N.; Gryglewicz, G.; New Carbon Mater. 2017, 32, 15. [Crossref]
  » Crossref
• ²⁵ Yoo, M. J.; Park, H. B.; Carbon 2019, 141, 515. [Crossref]
  » Crossref
• ²⁶ Yu, H.; Zhang, B.; Bulin, C.; Li, R.; Xing, R.; Sci. Rep. 2016, 6, 36143. [Crossref]
  » Crossref
• ²⁷ Mei, X.; Meng, X.; Wu, F.; Phys. E 2015, 68, 81. [Crossref]
  » Crossref
• ²⁸ Poniatowska, A.; Trzaskowski, M.; Ciach, T.; Colloids Surf., A 2019, 561, 315. [Crossref]
  » Crossref
• ²⁹ Li, X.; Cai, W.; An, J.; Kim, S.; Nah, J.; Yang, D.; Piner, R.; Velamakanni, A.; Jung, I.; Tutuc, E.; Banerjee, S. K.; Colombo, L.; Ruoff, R. S.; Science 2009, 324, 1312. [Crossref]
  » Crossref
• ³⁰ Yu, P.; Lowe, S. E.; Simon, G. P.; Zhong, Y. L.; Curr. Opin. Colloid Interface Sci. 2015, 20, 329. [Crossref]
  » Crossref
• ³¹ Abbas, Q.; Shinde, P. A.; Abdelkareem, M. A.; Alami, A. H.; Mirzaeian, M.; Yadav, A.; Olabi, A. G.; Materials 2022, 15, 7804. [Crossref]
  » Crossref
• ³² Yu, Q.; Wei, L.; Yang, X.; Wang, C.; Chen, J.; Du, H.; Shen, W.; Kang, F.; Huang, Z.-H.; Appl. Surf. Sci. 2022, 598, 153788. [Crossref]
  » Crossref
• ³³ Liu, F.; Wang, C.; Sui, X.; Riaz, M. A.; Xu, M.; Wei, L.; Chen, Y.; Carbon Energy 2019, 1, 173. [Crossref]
  » Crossref
• ³⁴ Achee, T. C.; Sun, W.; Hope, J. T.; Quitzau, S. G.; Sweeney, C. B.; Shah, S. A.; Habib, T.; Green, M. J.; Sci. Rep. 2018, 8, 14525. [Crossref]
  » Crossref
• ³⁵ Mag-isa, A. E.; Kim, J.; Lee, H.; Oh, C.; 2D Mater. 2015, 2, 34017. [Crossref]
  » Crossref
• ³⁶ Parvez, K.; Wu, Z.-S.; Li, R.; Liu, X.; Graf, R.; Feng, X.; Müllen, K.; J. Am. Chem. Soc. 2014, 136, 6083. [Crossref]
  » Crossref
• ³⁷ Ren, H.; Xia, X.; Sun, Y.; Zhai, Y.; Zhang, Z.; Wu, J.; Li, J.; Liu, M.; J. Mater. Chem. A 2024, 12, 23416. [Crossref]
  » Crossref
• ³⁸ Parvez, K.; Li, R.; Puniredd, S. R.; Hernandez, Y.; Hinkel, F.; Wang, S.; Feng, X.; Müllen, K.; ACS Nano 2013, 7, 3598. [Crossref]
  » Crossref
• ³⁹ Jiwarawat, N.; Leukulwatanachai, T.; Subhakornphichan, K.; Limwathanagura, S.; Wanotayan, S.; Atthi, N.; Pankiew, A.; Pungetmongkol, P.; Sci. Rep. 2024, 14, 15892. [Crossref]
  » Crossref
• ⁴⁰ Zhang, D.; Sasidharan, S.; Shi, J.; Sasikala Devi, A. A.; Su, J.; Huang, J.; Xia, Z.; ACS Appl. Nano Mater. 2023, 6, 19639. [Crossref]
  » Crossref
• ⁴¹ Roscher, S.; Hoffmann, R.; Prescher, M.; Knittel, P.; Ambacher, O.; RSC Adv. 2019, 9, 29305. [Crossref]
  » Crossref
• ⁴² Prabhu, G. R. D.; Urban, P. L.; Chem. Rev. 2020, 120, 9482. [Crossref]
  » Crossref
• ⁴³ Meloni, G. N.; J. Chem. Educ. 2016, 93, 1320. [Crossref]
  » Crossref
• ⁴⁴ Meloni, G. N.; Anal. Chem. 2017, 89, 8643. [Crossref]
  » Crossref
• ⁴⁵ Dryden, M. D. M.; Wheeler, A. R.; PLos One 2015, 10, e0140349. [Crossref]
  » Crossref
• ⁴⁶ Salverda, M.; Thiruppathi, A. R.; Pakravan, F.; Wood, P. C.; Chen, A.; Molecules 2022, 27, 8643. [Crossref]
  » Crossref
• ⁴⁷ Meshkov, A. V.; Nikitina, A. A.; Aliev, T. A.; Gromov, V. S.; Chen, S.; Yang, K.; Wang, Q.; Novoselov, K. S.; Andreeva, D. V.; Skorb, E. V.; Adv. Intell. Syst. 2024, 6, 2300655. [Crossref]
  » Crossref
• ⁴⁸ https://doi.org/10.17605/OSF.IO/J5BS3, accessed in September 2025.
  » https://doi.org/10.17605/OSF.IO/J5BS3
• ⁴⁹ AutoDesk, Tinkercad, https://www.tinkercad.com/, accessed in October 2025.
  » https://www.tinkercad.com/
• ⁵⁰ Timochenco, L.; Costa-Almeida, R.; Bogas, D.; Silva, F. A. L. S.; Silva, J.; Pereira, A.; Magalhães, F. D.; Pinto, A. M.; Materials 2021, 14, 1916. [Crossref]
  » Crossref
• ⁵¹ Gacka, E.; Majchrzycki, Ł.; Marciniak, B.; Lewandowska-Andralojc, A.; Sci. Rep. 2021, 11, 15969. [Crossref]
  » Crossref
• ⁵² Gutiérrez-Pineda, E.; Subrati, A.; Rodríguez-Presa, M. J.; Gervasi, C. A.; Moya, S. E.; Chem.- Eur. J. 2023, 29, e202302450. [Crossref]
  » Crossref
• ⁵³ Maia, K. C. B.; Francisco, A. D. D. S.; do Nascimento, A. S.; da Cruz, G. C.; Macedo, F. A. M. M.; Raitz, C.; Grasseschi, D.; J. Appl. Polym. Sci. 2025, e57758. [Crossref]
  » Crossref

• Editor handled this article:
  Cristiane Luísa Jost (Associate)

Publication Dates

• Publication in this collection
  28 Nov 2025
• Date of issue
  2025

History

• Received
  24 June 2025
• Published
  09 Oct 2025