**A Facile Method for the Generation of Fe3C Nanoparticle and Fe–Nx Active Site in Carbon Matrix to Achieve Good Oxygen Reduction Reaction Electrochemical Performances**

#### **Yuzhe Wu, Yuntong Li, Conghui Yuan \* and Lizong Dai \***

Fujian Provincial Key Laboratory of Fire Retardant Materials, College of Materials, Xiamen University, Xiamen 361005, China; 20720150150066@stu.xmu.edu.cn (Y.W.); lyt@stu.xmu.edu.cn (Y.L.) **\*** Correspondence: yuanch@xmu.edu.cn (C.Y.); lzdai@xmu.edu.cn (L.D.); Tel./Fax: +86-592-2186178 (C.Y. & L.D.)

Received: 14 September 2020; Accepted: 23 October 2020; Published: 26 October 2020

**Abstract:** Introduction of both nitrogen and transition metal elements into the carbon materials has demonstrated to be a promising strategy to construct highly active electrode materials for energy shortage. In this work, through the coordination reaction between Fe3<sup>+</sup> and 1,3,5–tris(4–aminophenyl)benzene, metallosupramolecular polymer precursors are designed for the preparation of carbon flakes co-doped with both Fe and N elements. The as-prepared carbon flakes display wrinkled edges and comprise Fe3C nanoparticle and active site of Fe–Nx. These carbon materials exhibit excellent electrocatalytic performance. Towards oxygen reduction reaction (ORR), the optimized sample has Eonset and Ehalf-wave of 0.93 V and 0.83 V in alkaline system, respectively, which are very close to that of Pt/C. This approach may offer a new way to high performance and low-cost electrochemical catalysts.

**Keywords:** coordination; metallosupramolecular polymer; active site; carbon materials; oxygen reduction reaction

#### **1. Introduction**

In recent years, much attention has been focused on the development of facile and applicable methods to fabricate high-activity, low-cost oxygen reduction reaction (ORR) catalysts. This is significant to overcome the challenges in the commercialization and industrialization of hydrogen fuel cells. Nevertheless, Pt/C still acts as a main role in the commercialized ORR catalysts, even though it is relatively expensive [1–4]. Therefore, non-precious metal-based materials with high catalytic performance have attracted great research interest [5–8].

Indeed, transition metal elements have been widely introduced into the carbon materials to achieve high electrochemical performances [9–12]. Incorporation of transition metal elements into the carbon matrix generally relies on the design of composite precursors. For example, by using the reaction between salts (like iron(III) nitrate nonahydrate, nickel(II) nitrate hexahydrate, manganese(II) acetate tetrahydrate, and cobalt(II) nitrate hexahydrate) and graphite oxide, graphite oxide-metal-based precursors can be generated. Subsequently, a thermal procedure leads to the formation of Mn, Fe, Co, and Ni-coped graphene, which exhibited improved ORR performance [13]. Design of metal-organic precursors has been demonstrated to be a controllable method to generate transition metal elements doped carbon materials. It has been reported that precursors derived from the coordination between transition metals salts (MClx, M = Cu, Ni, Co, Fe and Mn) and melamine/aniline, can be used to prepare transition metal doped carbon materials with enhanced ORR properties [14]. More interesting, utilization of metal-organic frameworks as precursors for the fabrication of transition metal-doped carbon materials has attracted great attention, because of the designable composition, pore structure, and tunable metal species [15]. Notably, co-doping of both iron and N elements in the carbon materials is of great advantage for improving the electrochemical properties, due to the generation of iron-nitrogen-carbon (Fe–N–C) active sites (Fe–Nx and Fe–Cx) [16–18]. Usually, Fe–N–C catalysts were prepared by thermal cracking of iron macrocyclic polymers, iron-organic salts, or N-containing compounds [19,20]. The complicated synthetic process and high cost greatly limited their practical application [21,22]. So, it is still desirable to explore simple method to prepare Fe–N–C catalysts from commercial resources.

In this report, we show that a simple ligand-Fe3<sup>+</sup> coordination strategy using commercial resources as starting materials, can create metallosupramolecular polymer precursors for the fabrication of Fe and N elements co-doped carbon materials with high ORR activity. The organic ligand adopted in this research is 1,3,5–tris(4–aminophenyl)benzene (denoted as TA) possessing three amine groups and conjugated structure. Because of the rigid structure of TA, the as-formed TA–Fe coordinative networks display uniform layer structure. After carbonization, carbon flakes with wrinkled edges (denoted as CNSs) can be easily generated. We have found that Fe3C nanoparticle and active site of Fe–Nx are formed in the carbon matrix, which can promote the ORR activity of the carbon materials. In addition, the feature of this research lies in the simple synthesis of CNSs. It can introduce Fe3C nanoparticle and Fe–Nx into the carbon matrix by one step. This synthetic route has certain universality and representativeness. It is also found that the appropriate molar ratio between amino ligand and Fe3<sup>+</sup> was the most important factor that determines the activity of CNSs.

#### **2. Experimental Section**

#### *2.1. Materials*

1,3,5–Tris(4–aminophenyl)benzene, iron(III) chloride hexahydrate and dichloromethane were supplied by Aladdin Company (Shanghai, China) and directly used. Anhydrous ethanol, anhydrous methanol, KOH, and hydrochloric acid were purchased from Shanghai Chemical Reagent Industry (Shanghai, China). Nafion (5 wt%) was supplied by Sigma-Aldrich (Shanghai, China).

#### *2.2. Catalysts Preparation*

TA (0.3 g, 0.85 mmol) was firstly dissolved in 100 mL of dichloromethane with vigorous stirring for 3 h. Then, 1.38, 1.84, and 2.30 mL methanol solutions of Iron(III) chloride hexahydrate (100 mg/mL, 0.37 mmol/mL) were added into the solution drop by drop under vigorous stirring under N2 atmosphere. Thus, the corresponding molar ratios between TA and Fe3<sup>+</sup> were 1.0:0.6, 1.0:0.8, and 1.0:1.0, respectively. The mixtures became yellow, and suspensions of metallosupramolecular polymers were formed after 12 h of reaction. The yellow powders were collected by centrifugation and washing with a mixed solvent of 30 mL dichloromethane and 30 mL anhydrous methanol three times. After drying in vacuum at 50 ◦C overnight, TA–Fe precursors were obtained (denoted as TA–0.6Fe, TA–0.8Fe and TA–1.0Fe). These precursors were then carbonized at 850 ◦C for 2 h in argon gas with a heating rate of 5 ◦C/min to generate the carbon materials. The carbon materials were washed by 6 M HCl for 5 h and 100 mL of ultrapure water three times at room temperature, then dried by freeze-drying overnight. A second carbonization was performed at 850 ◦C for 2 h in argon gas with a heating rate of 10 ◦C/min to get the target carbon materials (denoted as CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe).

#### *2.3. Characterization*

The scanning electron microscopy (SEM) images were observed through an SU-70 microscopy instrument (HITACHI, Tokyo, Japan). The FTIR measurements were tested by an AVATAR 360 FTIR (Nicolet Instrument, Tokyo, Japan) at room temperature. The powder X-ray diffraction (XRD) patterns were measured through a Desktop X-ray Diffractometer ((Rigaku, Tokyo, Japan) using Cu (600 W) Kα radiation. Raman spectra were tested by a Labram HR800 Evolution (Horiba, Lille, France). The X-ray photoelectron spectroscopy (XPS) were tested by PHI Quantum-2000 photoelectron spectrometer

(Physical Electronics, Inc., Chanhassen, MN, USA). The pore volume and Brunauer–Emmett–Teller (BET) were taken through an ASAP 2460 system (Norcross, GA, USA). Electron paramagnetic resonance (EPR) experiments were conducted on an electron spin resonance spectrometer (Bruker EMX-10/12, Bruker UK, Coventry, UK) at 90 K. Transmission electron microscopic (TEM) measurements were performed using a JEM-2100 microscope (JEOL, Tokyo, Japan). The elemental energy-dispersive X-ray spectroscopy (EDX) were obtained by using a FEI TECNAI F20 microscope (Hillsboro, OR, USA).

#### *2.4. Electrochemical Measurements*

The electrochemical experiments were conducted on an electrochemical workstation (CHI 760E, Chenhua, Shanghai, China), by using the typical three-electrode system. A standard rotating disk electrode with a glassy carbon disk (5 mm in diameter) was applied as working electrode. Before test, CNS–0.6Fe, CNS–0.8Fe, and CNS–1.0Fe (5.0 mg) were dispersed in 1.0 mL of homogeneous solvent with 500 μL of anhydrous ethanol, 450 μL of H2O and 50 μL of 5 wt% Nafion. The above newly made slurry (4.5 μL) was carefully dropped onto a glassy carbon electrode as working electrode. The ORR performance was tested in newly made KOH aqueous solution (0.1 mol/L) at room temperature. Pt foil and Ag/AgCl (KCl saturation) electrode were separately applied as counter electrode and reference electrode. The potential in this study was relative to the Ag/AgCl electrode.

#### **3. Results and Discussion**

#### *3.1. Characterization of Metallosupramolecular Polymer Precursors*

The synthetic process of precursor is very simple. As shown in Scheme 1, metallosupramolecular polymer precursors are generated from the direct reaction between commercially available resources. In a dichloromethane solution, the high coordination affinity between TA and Fe3<sup>+</sup> can easily induce the formation precipitates. After a pyrolysis treatment of the precursors, carbon materials comprising both Fe3C nanoparticle and Fe–Nx active site can be fabricated easily.

**Scheme 1.** Synthetic process of the CNSs.

The typical SEM image of TA–0.8Fe is displayed in Figure 1a, from which sheet-like morphology can be observed. The coordination reaction between TA and Fe3<sup>+</sup> was verified by FTIR as shown in Figure 1b. The peaks located at 3350–3343 cm−<sup>1</sup> correspond to the characteristic signals of amino groups of TA, and the peaks at 845, 685, and 600 cm−<sup>1</sup> are derived from iron(III) chloride hexahydrate. Comparing the spectra of precursors with TA, the characteristic peak of amino groups shifts and becomes broaden. Also, the characteristic peaks of Fe3<sup>+</sup> in the precursors were evidently weakened. These results indicate the coordination between Fe3<sup>+</sup> and TA [23].

In the EPR spectra (Figure 2a), all samples show a small radical signal of amino group at *g'* = 2.00. For TA–0.6Fe, TA–0.8Fe, and TA–1.0Fe, the representative signal of Fe3<sup>+</sup> at *g'* = 4.25 can be observed, indicating the presence of Fe3<sup>+</sup> in the precursor [24]. The XPS survey spectra of TA, TA–0.6Fe, TA–0.8Fe, and TA–1.0Fe are displayed in Figure 2b. The representative signals of Fe 2p locate at 714.4 ± 0.1 and 725.4 <sup>±</sup> 0.1 eV, which can be attributed to the binding energies of 2p3/<sup>2</sup> and 2p1/<sup>2</sup> orbitals of Fe3<sup>+</sup>, respectively. Figure 2c–f shows the high-resolution XPS spectra of N 1s of TA and the precursors. A signal at 399.4 ± 0.1 eV is attributed to the amino group. In the case of TA–0.6Fe, TA–0.8Fe,

and TA–1.0Fe, a peak at 401.6 <sup>±</sup> 0.1 eV is attributed to amino group perturbed by Fe3<sup>+</sup> [25], which is helpful for the formation of Fe–N–C active site during carbonization. As showed in Table S1, TA–0.8Fe has higher content of Fe–NH2 than TA–0.6Fe and TA–1.0Fe, which may result in more content of Fe–Nx active site after carbonization.

**Figure 1.** (**a**) SEM image of TA–0.8Fe. (**b**) FTIR spectra of TA, iron(III) chloride hexahydrate and TA–0.6Fe, TA–0.8Fe, and TA–1.0Fe.

**Figure 2.** (**a**) EPR spectra of TA, TA–0.6Fe, TA–0.8Fe and TA–1.0Fe. (**b**) XPS survey spectra of TA, TA–0.6Fe, TA–0.8Fe and TA–1.0Fe. High-resolution XPS spectra of N 1s of TA (**c**), TA–0.6Fe (**d**), TA–0.8Fe (**e**) and TA–1.0Fe (**f**).

#### *3.2. Structure and Composition of CNSs*

After twice carbonization at 850 ◦C, the as obtained CNSs can maintain the lamellar structure, but the edges become wrinkled (Figure 3a,b and Figure S1a,b). This structure may be helpful for the direct contact between the active sites with the oxygen, thus improving the electrocatalytic activity of carbon materials. Notably, CNS–0.8Fe possesses the most uniform lamellar morphology (Figure 3b). The high-resolution TEM images of CNS–0.8Fe show clear inter-planar distance of 0.201 nm derived from the (031) plan of Fe3C nanoparticle (Figure 3c,d). The outer carbon coating on the Fe3C nanoparticle has a good lattice structure with a spacing of 0.34 nm, corresponding to the (002) plan of graphitic carbon (Figure 3c,d). The outer layer of graphitized carbon on Fe3C nanoparticles has good electrical conductivity. During the ORR catalytic process, Fe3C nanoparticles may not contact with electrolyte directly, but can play the catalytic role indirectly through the outer layer of graphitized carbon to improve the catalytic activity. The Fe3C nanoparticle generated in CNSs can improve the ORR activity of carbon materials, which was confirmed already [26]. Figure 3e–i gives the dark-field TEM image and EDX mapping of CNS–0.8Fe. Obviously, elements of C, N, Fe, and O are homogeneously dispersed all through the carbon materials.

**Figure 3.** SEM image of CNS–0.8Fe (**a**). TEM image of CNS–0.8Fe (**b**). High-resolution TEM images of CNS–0.8Fe (**c**,**d**). Dark-field image and EDX mappings C, N, Fe, and O elements (**e**–**i**) of CNS–0.8Fe.

Figure 4a illustrates the Raman spectra of the CNSs. For CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe, the intensity ratios between D band (1340 cm−1) derived from disordered graphitic structure and G band (1571 cm−1) derived from ordered carbon structure are calculated to be 0.97, 0.95, and 1.04, respectively. This result indicates that the graphitic degree of CNS–0.8Fe is higher than that of the other samples. The crystalline structures of CNSs were evaluated by the XRD. As displayed in Figure 4b, a broad diffraction peak at about 25◦ is attributed to the (002) plane of ordered graphitic structure. Moreover, the XRD results clearly confirm that the iron element is remained in the carbon matrix as a Fe3C form. All the diffraction peaks are in good agreement with that of the Fe3C (JCPDS Card No.65−2413). These results, in combination with the high-resolution TEM images, prove the presence of Fe3C nanoparticle in the CNSs catalysts, which may promote the ORR activity of carbon materials [27].

**Figure 4.** (**a**) Raman spectra of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe catalysts. (**b**) XRD patterns of CNS–0.6Fe, CNS–0.8Fe, and CNS–1.0Fe catalysts.

The pore character of CNSs was characterized through the physisorption of nitrogen at 77 K. As shown in Figure 5, all samples show well-developed micro-pore and mesoporous-pore structures. Table 1 shows the corresponding information about BET surface area as well as total pore volumes of CNSs. The surface areas of CNS–0.6Fe, CNS–0.8Fe, and CNS–1.0Fe are 167.86, 196.20, and 145.20 m2·g<sup>−</sup>1, with relevant pore volumes of 0.16, 0.17, and 0.17 cm3·g−1, respectively. Obviously, surface areas of CNSs are resulted from both micro-pore and mesoporous-pore structures. The CNS–0.8Fe has a higher BET surface area than CNS–0.6Fe and CNS–1.0Fe. We consider that the large surface area of CNS–0.8Fe is attributed to the moderate crosslinking degree of the metallosupramolecular polymer networks.

**Figure 5.** (**a**) N2 adsorption and desorption isotherms of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe. (**b**) Pore size distribution of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe.


**Table 1.** Surface area, porosity of CNSs.

<sup>a</sup> Specific surface area derived from BET. <sup>b</sup> Surface area about micropores calculated through the t-plot method. <sup>c</sup> Surface area of mesopores and macropores calculated through the t-plot method. <sup>d</sup> Total pore volume.

The XPS survey spectra of the CNSs are shown in Figure 6a. Also, the high-resolution XPS spectra of C 1s, N 1s, and Fe 2p of CNS–0.8Fe are displayed Figure 6b–d. The C 1s signal is split into three representative peaks at 284.4 ± 0.1 (C=C, C–C), 285.4 ± 0.1 (C–O, C–N), and 288.2 ± 0.1 eV (C=O). Four peaks of N 1s signal at 398.5 ± 0.1, 399.5 ± 0.1, 401.2 ± 0.1, and 404.5 ± 0.1 eV are respectively belong to the pyridinic N, Fe–Nx, graphitic N, and oxidized N. Notably, Fe–Nx and pyridinic N are recognized to be promising for the improvement of ORR activity [28,29]. The N 1s spectra of CNS–0.6Fe and CNS–1.0Fe are also showed in Figure S2. For the Fe 2p spectrum, the peak located at 725.4 <sup>±</sup> 0.1 eV is assigned to the binding energy of Fe3<sup>+</sup> for the 2p1/<sup>2</sup> band, and the peak of Fe2<sup>+</sup> is detected at 723.2 ± 0.1 eV for the 2p1/<sup>2</sup> band. Another two peaks at 714.4 ± 0.1 and 710.5 ± 0.1 eV can be respectively attributed to the binding energies of 2p3/<sup>2</sup> orbitals of Fe3<sup>+</sup> as well as Fe2<sup>+</sup> species. The last signal at 719.6 ± 0.1 eV is the satellite peak. These XPS results, in combination with the XRD results, clearly confirm that the presence of Fe3C nanoparticle, and active sites of Fe–Nx and pyridinic N in the carbon matrix of CNSs. Moreover, as listed in Table 2, the pyridinic N and Fe–Nx contents of CNS–0.8Fe are 0.28 and 0.43 at.%, respectively, which are much higher than that of CNS–0.6Fe (0.17 and 0.20 at.%) and CNS–1.0Fe (0.25 and 0.22 at.%). This result indicates that CNS–0.8Fe may have more active sites towards ORR. In summary, the Fe and N elements co-doping effect leads to the generation of both Fe–Nx active site and Fe3C nanoparticle when carbonization, which can greatly improve the catalytic activity towards ORR. That is the main mechanism and contribution of the co-doping effect of Fe and N elements in CNSs.

**Figure 6.** (**a**) XPS survey spectra of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe. High-resolution XPS spectra of (**b**) C 1s, (**c**) N 1s, (**d**) Fe 2p of CNS–0.8Fe.


**Table 2.** Contents (at.%) of nitrogen with different chemical environments calculated from the N1s XPS spectra.

<sup>a</sup> The different contents of nitrogen (at.%) calculated by the analysis of the peak area as for pyridinic N, Fe–Nx, graphitic N and oxidized N.

#### *3.3. ORR Performance of CNSs*

The CV curves of CNSs were tested in both Ar or O2-saturated 0.1 M KOH solution (Figure 7a). The samples show no reduction peak in Ar-saturated solution but show a typical reduction peak when changed into O2-saturated solution. The double-layer capacitance of the three samples was researched, which are showed in Figure S3. The CV area of CNS–0.8Fe in Ar is larger than CNS–0.6Fe and CNS–1.0Fe. This indicates that CNS–0.8Fe owns larger electrochemically active surface area than the other two samples, which is helpful for improving the catalytic activity of ORR. Figure 7b list the LSV curves of CNSs. The onset potential (Eonset) of CNS–0.6Fe, CNS–0.8Fe, and CNS–1.0Fe for ORR are 0.89, 0.93, and 0.88 V vs. RHE, separately. The half-wave (Ehalf-wave) of the CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe are 0.78, 0.83, and 0.80 V vs. RHE, separately. The above results were confirmed by LSV tests in the newly made O2-saturated solution at the scanned rate of 10 mV/s with the fixed rotation speed of 1600 rpm. As a control experiment, the ORR activity of Pt/C catalyst was also tested with the same experimental condition (Figure 7b). Apparently, the Eonset (0.93 V) and Ehalf-wave (0.83 V) values of CNS–0.8Fe are very close to Pt/C catalyst (Eonset = 0.95 V and Ehalf-wave = 0.85 V). The Tafel plots of CNSs were tested (Figure S4). The Tafel slopes of CNS–0.6Fe and CNS–1.0Fe are 89 and 84 mV·dec<sup>−</sup>1, respectively. However, a Tafel slope of 73 mV·dec−<sup>1</sup> is detected for CNS–0.8Fe, which is lower than that of Pt/C (78 mV·dec<sup>−</sup>1) and directly indicating that CNS–0.8Fe owns faster ORR kinetics.

**Figure 7.** (**a**) CV curves of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe in Ar or O2, scan rate: 50 V/s. (**b**) LSV curves of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe at 1600 rpm with a scan rate of 10 mV/s.

The evidently improved electrocatalytic performance of CNS–0.8Fe among the three samples can be explained by the following four reasons. First, CNS–0.8Fe has relative higher specific surface area in comparison with CNS–0.6Fe and CNS–1.0Fe, thus resulting in the expose of more active sites. Second, CNS–0.8Fe possesses a better developed lamellar structure than CNS–0.6Fe and CNS–1.0Fe as indicated by the TEM images, which is beneficial for the contact between active sites and oxygen molecules during ORR process. Third, the calculated ID/IG value testified that the graphitic degree of CNS–0.8Fe is higher than the others, thus endowing this sample with a better electrical conductivity. Fourth and most importantly, CNS–0.8Fe possesses higher pyridinic N and Fe–Nx contents than

CNS–0.6Fe and CNS–1.0Fe, which can provide more active sites towards ORR, thus greatly enhancing the catalytic activity.

The LSV curves of CNSs were also collected with different rotation rates. The current density of CNS–0.6Fe, CNS–0.8Fe, and CNS–1.0Fe increase gradually when consecutively changing the rotation speeds from 400 to 1600 rpm, as listed in Figure S5. Probably, the shortening of the diffusion distance directly leads to this regular phenomenon. To further explore the reaction kinetics of the ORR process, rotating ring disk electrode (RRDE) experiments were performed to calculate the generation of HO2 − also with electron transfer numbers (n) values. CNS–0.6Fe, CNS–0.8Fe, CNS–1.0Fe, and the commercial Pt/C displayed higher disk current but minor ring current, as shown in Figure 8a. With the increase of potential from 0.2 to 0.5 V, the corresponding HO2 − yield ranges of CNS–0.6Fe, CNS–0.8Fe, CNS–1.0Fe, and Pt/C catalyst are 10.57 to 11.90%, 3.99 to 6.22%, 4.66 to 5.93%, and 1.81 to 2.71%, as shown in Figure 8b. Also, the corresponding n values of CNS–0.6Fe, CNS–0.8Fe, CNS–1.0Fe, and Pt/C catalyst are 3.67 to 3.78, 3.87 to 3.92, 3.88 to 3.90, and 3.94 to 3.96 as shown in Figure 8c. So, these results just could indicate that CNSs catalyze ORR by the typical dominant four-electron transfer pathway [30–33].

**Figure 8.** (**a**) Rotating ring disk electrode (RRDE) tests of CNS–0.6Fe, CNS–0.8Fe, CNS–1.0Fe and Pt/C at 1600 rpm. (**b**) The HO2 − yields, (**c**) electron transfer number.

Taking CNS–0.8Fe as an example, the durability of CNSs was evaluated at 1600 rpm with consecutive 1000 cycles of CV scan in O2-saturated 0.1 M KOH solution (Figure S6). The decrease of the onset and half-wave potentials is not evident. The LSV curves recorded before and after 1000 cycles reveal negative shifts of Ehalf-wave of 7 mV for CNS–0.8Fe, which is lower than that of Pt/C (12 mV) as reported [21]. This result indicates that CNS–0.8Fe is relatively stable for ORR. The relevant crossover effects tests were conducted by taking CNS–0.8Fe as an example through chronoamperometric measurement to evaluate the catalytic selectivity of the catalysts. The methanol oxidation reaction resulted that the current density of Pt/C catalyst directly decreased at once when scrupulously adding 3.0 M methanol, as shown in Figure 9. However, this was not the case for CNS–0.8Fe, as the current density of that did not show evident change (Figure 9). These results confirmed that CNSs have good catalytic selectivity for ORR [34–36].

**Figure 9.** Methanol crossover tests of CNS-0.8Fe and Pt/C at 1600 rpm.

#### **4. Conclusions**

In summary, we prepared a new type of Fe and N co-doped carbon materials through a simple and effective method in one step. Direct coordination between amino ligand and Fe3<sup>+</sup> could easily afford metallosupramolecular polymer precursors. After two carbonization processes, carbon flakes with wrinkled edges and active site of Fe–Nx and Fe3C nanoparticle were fabricated. The catalytic activity of the carbon materials towards ORR were detailed investigated. The carbon material of CNS–0.8Fe possessed Eonset = 0.93 V and Ehalf-wave = 0.83 V vs. RHE in alkaline system, which were comparable to Pt/C catalyst. The ligand TA and Fe3<sup>+</sup> could generate more content of Fe–NH2 in the precursor at a proper proportion through the coordination reaction and further led to the generation of more content of Fe–Nx active site when carbonization. So, the appropriate molar ratio between amino ligand and Fe3<sup>+</sup> was the most important factor that determined the activity of CNSs. We considered that this simple method and conclusion might be of practical interest for the exploration of electrocatalysts with excellent ORR activity.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/1996-1944/13/21/4779/s1, Figure S1: TEM images of CNS–0.6Fe (**a**) and CNS–1.0Fe (**b**), Figure S2: High-resolution XPS spectra of N 1s for CNS–0.6Fe (**a**) and CNS–1.0Fe (**b**), Figure S3: CV curves of CNS–0.6Fe, CNS–0.8Fe and CNS–1.0Fe in Ar-saturated 0.1 M KOH solution, scan rate: 50 V/s, Figure S4: Tafel slopes derived from the LSV curves of CNS–0.6Fe, CNS–0.8Fe CNS–1.0Fe and Pt/C, Figure S5: LSV curves of (**a**) CNS–0.6Fe, (**b**) CNS–0.8Fe and (**c**) CNS–1.0Fe at 400–1600 rpm with a scan rate of 10 mV/s in O2-saturated 0.1 M KOH solution, Figure S6: LSV curves of CNS–0.8Fe at 1600 rpm before and after durability test in O2-saturated 0.1 M KOH solution, scan rate: 50 mV/s, Table S1: Contents (atomic %) of N element with different chemical environments calculated from the N 1s XPS spectrum.

**Author Contributions:** The experiments were designed by Y.W.; the experiment was carried out and the manuscript was written by Y.W., Y.L., C.Y. and L.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** The National Natural Science Foundation of China (51673161, 51773172), Scientific and Technological Innovation Platform of Fujian Province (2014H2006), National Science and Technology Ministry (2014BAF08B03) supported our work.

**Conflicts of Interest:** There are no conflicts to declare.

#### **References**


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

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

### *Article* **Effects of Boron Carbide on Coking Behavior and Chemical Structure of High Volatile Coking Coal during Carbonization**

**Qiang Wu, Can Sun, Zi-Zong Zhu \*, Ying-Dong Wang and Chong-Yuan Zhang**

College of Materials Science and Engineering, Chongqing University, Chongqing 400044, China; qiangwucqu@163.com (Q.W.); suncan2230@163.com (C.S.); 18243934868@163.com (Y.-D.W.); zhangchongyuan0811@163.com (C.-Y.Z.)

**\*** Correspondence: zzzhu666@cqu.edu.cn; Tel.: +86-139-8327-0208

**Abstract:** Modified cokes with improved resistance to CO2 reaction were produced from a high volatile coking coal (HVC) and different concentrations of boron carbide (B4C) in a laboratory scale coking furnace. This paper focuses on modification mechanism about the influence of B4C on coking behavior and chemical structure during HVC carbonization. The former was studied by using a thermo-gravimetric analyzer. For the latter, four semi-cokes prepared from carbonization tests for HVC with or without B4C at 450 ◦C and 750 ◦C, respectively, were analyzed by using Fourier transform infrared spectrum and high-resolution transmission electron microscopy technologies. It was found that B4C will retard extensive condensation and crosslinking reactions by reducing the amount of active oxygen obtained from thermally produced free radicals and increase secondary cracking reactions, resulting in increasing size of aromatic layer and anisotropic degree in coke structure, which eventually improves the coke quality.

**Keywords:** high volatile coking coal; boron carbide; coking behavior; chemical structure; coke quality

#### **1. Introduction**

Adding cheap materials into coal blends to produce metallurgical coke has been extensively researched, due to the gradual rise of coking coal price and inadequate supply of the prime-coking coals with medium volatility. The most studied method is harnessing non-coking coal to replace part of coking coals, but the caking property of coal blends will deteriorate in such a situation. To improve the caking property, on the one hand, various high bond-ability substances, such as pitches, coal tar pitches, coal extracts, and solvent-refined coals, have been used in the carbonization process of coal blending with low bond-ability [1–7]. On the other hand, non-coking coal was pretreated by thermal treatments [8], hydrothermal treatments [9,10], and steam treatments [11,12]. Additionally, to reduce the costs of coal blends and the amount of CO2 emission, adding small amount of biomass material into coal blends to produce metallurgical coke has also been suggested [13–15]. Although these methods can broaden coking coal resources, few industrial applications about the above studies are successful because of factors such as coke quality, cost, production conditions, etc.

It is widely known that the reserve of high volatile coking coal (HVC) occupies about half of all coking coal reserves, and the price of HVC is relatively cheap. Therefore, increasing the usage amount of HVC in coal blending to produce metallurgical coke is also an effective solution to reduce the cost of coke and maintain the sustainable development of the coke industry. However, the usage amount of HVC in coal blending is usually limited to 20–30% in the producing process of metallurgical coke [5,12], which is extremely disproportionate to the reserve of HVC. This is because HVC is a low proportion of metamorphism coking coal and possesses plenty of oxygen-containing groups as well as aliphatic side chains, which results in the rapid formation of large quantities of gases, free radicals, and plastic mass with high fluidity during HVC carbonization, subsequently

Wang, Y.-D.; Zhang, C.-Y. Effects of Boron Carbide on Coking Behavior and Chemical Structure of High Volatile Coking Coal during Carbonization. *Materials* **2021**, *14*, 302. https://doi.org/10.3390/ma14020302

**Citation:** Wu, Q.; Sun, C.; Zhu, Z.-Z.;

Received: 8 December 2020 Accepted: 5 January 2021 Published: 8 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional clai-ms in published maps and institutio-nal affiliations.

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

leaving a weak coke with a thin-walled porous structure [13,14,16–19]. In such a case, Qian et al found that the addition of pitches could contribute to developing intermediate texture during HVC carbonization [20]. This improves the coke quality, but the complex operation process and plugging problem limit its industrial application. Vega et.al., asserted that mild oxidation was an effective method to improve the coking performance of HVC with low oxygen content (less than 5%) [19]. However, HVC usually contains high oxygen content (more than 10%) so that mild oxidation will lead to excessive oxygen in HVC, which worsens its thermoplasticity and coking performance.

Admittedly, exploring a new method which is able to improve the coking performance of HVC, finally increasing its usage in coal blends, is urgent and essential. Previous studies have indicated that during resin pyrolysis, B4C reacts with oxygen-containing fragments released from resin to generate boric oxide, whose formation further affects the viscosity of resin at high temperature [21,22]. The compositions of the oxygen-containing fragments obtained from the pyrolysis process of resin and HVC are partly similar; that B4C may react with the oxygen-containing fragments derived from the plastic zone of HVC. In such a case, there are more indigenous donor hydrogen stabilizing free radicals due to the decrease in reaction between oxygen-containing fragments and transferable hydrogen. Consequently, B4C may be a promising additive to improve HVC's coking performance and coke quality.

Initially, this work aims to investigate whether adding B4C can improve the quality of coke obtained from HVC. Five carbonization tests were carried out in a laboratorial scale coking furnace, and the quality of resulting coke was reflected by coke reactivity towards CO2 (CRI) and coke strength after reaction (CSR) indexes. Secondly, one of the important tasks of this work was to acquire the mechanism of improvement of coke quality in detail. The influence of B4C on coking behavior during HVC carbonization was evaluated by using a thermo-gravimetric analyzer (TG). Based on the coke quality data and TG analysis, four semi-cokes with or without B4C were manufactured under two characteristic temperatures and were analyzed by Fourier transform infrared spectrum (FTIR) and high-resolution transmission electron microscopy (HRTEM) techniques to investigate the effects of B4C on chemical structure during HVC carbonization. Having a better understanding of the interactional mechanism between HVC and B4C contributes to increasing the proportion of low-rank coking coal (such as HVC) in coal blending, especially in the case of adding B4C, to produce low-cost metallurgical coke. Simultaneously, this method is easy to apply in industry.

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

#### *2.1. Samples and Carbonization Experiments of Coke*

The HVC used in the current study was collected from Kubai coal field of Xinjiang province, located in northwest China. The boron carbide (B4C) powder with particle diameter <58 μm had a purity of 95%. The coke carbonization experiments between HVC and B4C were carried out in a laboratorial scale coking furnace, and experimental schemes are shown in Table 1.

The coking furnace temperature was controlled automatically by a programmable controller and was heated by resistance wire. The density and moisture of coal blends were limited to 0.95 g/cm−<sup>3</sup> and 10 wt.%, respectively. Approximately 2 kg of the coal blend (particle size of HVC less than 3 mm) was mixed evenly and placed in a coking retort with an internal diameter of 100 mm and a length of 500 mm. Next, this coking retort was put into the coking furnace after the experimental temperature reached 700 ◦C, combining with a heating rate 10 ◦C/min. After that, it was consecutively heated to 1000 ◦C at the rate of 5 ◦C/min, and the target temperature was maintained for 5 h, finally cooling the retort to room temperature in the atmosphere. The GB1997-89 standard was applied to produce coke samples whose CRI and CSR indexes were measured using GB/T4000-1996 standard. These indexes are shown as the average value of three trials in a later context.


**Table 1.** Carbonization schemes for coke and semi-coke.

#### *2.2. TG Measurements*

Generally, the carbonization process of coal is similar to its pyrolysis process under an inert gas. In order to simulate and evaluate the influence of adding B4C on coking behavior in the carbonization process of HVC, two groups of thermo-gravimetric analysis experiments for HVC (particle size of <74 μm in diameter) with and without 0.5 wt.% B4C were carried out on a NETZSCH STA 449 C analyzer (Nestal, Selbu, Germany). About 10 mg of coal blend was placed in an alumina cell and heated from ambient temperature to 1000 ◦C at a rate of 10 ◦C/min under a continuous argon atmosphere with flow rate of 50 mL/min.

#### *2.3. Carbonization Experiments of Semi-Coke*

To investigate the influence of B4C on the chemical structure in the carbonization process of HVC, the semi-coke carbonization experiment was carried out in an electrically heated oven using a 200 mL crucible and corresponding schemes, based on the coke quality analysis and characteristic temperatures analysis from TG, are listed in Table 1. Approximately 100 g coal blending (particle size of HVC less than 1 mm) was loaded into the crucible. The crucible was placed in the oven's chamber filled with inert atmosphere, heated at the rate of 10 ◦C/min to 450 ◦C and 750 ◦C, respectively, held at the target temperature for 5 min, and then removed and cooled to room temperature under an N2 atmosphere.

#### *2.4. Preparation of Demineralized Samples*

Raw coal samples and semi-coke samples were ground and sieved to obtain particles of <74 μm in diameter. To eliminate the potential effects of minerals on the FTIR and HRTEM analyses, these samples were acid-washed using HCl and HF solution at room temperature as described in a previous study [23]. Generally, acid treatment under such conditions does not cause significant structural changes [24]. These demineralized samples were dried for 12 h in vacuum room at 60 ◦C and stored under nitrogen atmosphere. Ultimate analyses of these samples were determined according to GB/T 476-208 criterion and the analytical results are listed in Table 2.

**Table 2.** Elemental analyses of studied samples (wt.%, daf).


<sup>a</sup> By difference; wt.%, daf: weight percentage of various elements on a dry and ash-free basis.

#### *2.5. FTIR Measurements*

The FTIR spectra of the demineralized samples were recorded on a Thermo-Nicolet iS5 FTIR spectrometer (Thermo Fisher Scientific, MMAS, Waltham, MA, USA). All samples for the FTIR measurement were prepared by mixing the investigated sample with dried KBr powder, and the mixture was pressed to form a pellet under 12 MPa for 2 min. All spectra were obtained within the 400–4000 cm−<sup>1</sup> wave number range at a resolution of 4 cm<sup>−</sup>1, and 32 scans per spectrum were performed.

#### *2.6. HRTEM Measurements*

The influence of adding B4C on lattice fringes in the carbonization process of HVC was investigated by using HRTEM. The demineralized samples were grounded in ethanol and sonicated in an ultrasonic washer for 10 min, and then sprayed on a copper microgrid as HRTEM specimens. The HRTEM images of the samples were acquired from a 200 kV transmission electron microscope (JEM-2100F, JEOL, Tokyo, Japan). Detailed procedures and conditions of HRTEM analysis were derived from a previous thesis [25].

#### **3. Results and Discussion**

#### *3.1. Coke Quality Analysis*

In order to easily observe the changing trend of strength, two smooth curves in Figure 1 were automatically performed by linking data points based on the B-spline mode of software Origin 9.1. As shown in Figure 1, the CRI index of coke decreased by 26.3%, while CSR index of coke increased by 18.5% when 0.25 wt.% B4C was added into HVC, indicating that the coke quality is improved distinctly through the addition of B4C. In addition, these thermal strength indexes were further promoted when B4C content increased to 0.50 wt.%. However, continuing to increase the content of B4C to 0.75 wt.% even 1 wt.% led to little change of these indexes. These results indicate that adding 0.5 wt.% B4C into HVC to improve the coke quality is sufficient. Therefore, the addition amount of B4C was set as 0.5 wt.% in the following research on the modification mechanism.

**Figure 1.** The relationship between thermal strength indexes of coke and the concentration of B4C in coal blending. CRI: coke reactivity towards CO2; CSR: coke strength after reaction.

#### *3.2. Ultimate Analyses*

As listed in Table 2, raw coal has a high oxygen content (16.41 wt.%, daf) and low carbon content (76.93 wt.%, daf), suggesting that raw coal is a low coalification coking coal with abundant oxygen-containing groups. In the carbonization process, the contents of hydrogen and oxygen decrease in semi-cokes while the content of carbon increases with the increase in temperature (Table 2), which is largely caused by releasing small molecular weight gases, such as CO2, CO, H2O, H2, etc. It is worth noting that some differences on element content and H/C ratio were observed in semi-cokes under the same temperature, indicating that B4C causes a change of carbonization process of HVC.

#### *3.3. TG Analysis*

Figure 2 shows the curves of mass loss (TG) and their derivatives (DTG) for raw coal and modified coal, respectively. According to previous research [26], seven characteristic temperatures derived from the TG and DTG curves are defined and the two curves are divided into five stages as shown in Figure 2. For the drying stage (room temperature—Ti) and slow pyrolysis stage (Ti–Tm; the process with a low reaction rate—Figure 2), there is little difference in both coals, indicating that the B4C has little effect on the dehydration process, the releasing process of gas soaked in the pores, and the decomposition process of unstable functional groups below 400 ◦C.

**Figure 2.** Thermo-gravimetric (TG) and DTG curves of raw coal and modified coal (addition of 0.5 wt.% B4C into raw coal) at a heating rate of 10 ◦C/min. T\*,r: represents the characteristic temperatures of raw coal; T\*,m: represents the characteristic temperatures of modified coal; Tc: moisture loss peak temperature; Ti: pyrolysis initial temperature; Tm: initial temperature of the pyrolysis process with a fast reaction rate; Tmax: highest pyrolysis peak temperature; Tn: finish temperature of the pyrolysis process with a fast reaction rate; Tc: second cracking peak temperature; Tf: pyrolysis finish temperature.

For the fast pyrolysis stage (Tm–Tn), the process with a fast reaction rate, a weight loss peak with a high mass loss rate (2.53 %/min) was observed in the DTG curve of raw coal, indicating that many considerable reactions occurred, and these reactions caused the formation of abundant volatile matters, free radicals, and molecular fragments in this stage. Compared with the weight loss peak, however, the corresponding peak in modified coal showed a lower mass loss rate (1.98 %/min), indicating that the reaction rate of pyrolysis reactions in the fast pyrolysis stage are slowed down by adding B4C. This may be because partial active oxygen obtained from thermally produced molecular fragments are combined with B4C to form boron oxide [21,27–29], which decreases the consumption of transferable hydrogen used to stabilize reactive oxygen substances so that more transferable hydrogen can be used to stabilize the free radicals. In such a case, on the one hand, more stable free radicals can be arranged in an ordered structure, resulting in the development of anisotropic mesophase structures in semi-coke [30]. On the other hand, the reaction rate of condensation and crosslinking reactions will be reduced in modified coal, which is consistent with a lower mass loss rate peak in the fast pyrolysis stage.

For the fast polycondensation stage (Tn–Tf), the peak at Tp,m is more intensive than the peak at Tp,r, indicating that the intensity of secondary cracking reactions in modified coal is higher than that in raw coal. This is because, with the development of mesophase structures in modified coal, more macromolecular weight polymers will form and participate in secondary cracking reactions [26].

For the slow polycondensation stage (Tf–1000 ◦C), there were no obvious differences in the transformation process from char to coke for both coals, but the weight of pyrolytic residues of modified coal was higher than that of raw coal after the pyrolysis finished. This is attributed not only to the residue of boron oxide, but also the lower release of small molecular weight gases in the fast pyrolysis stage.

#### *3.4. FTIR Analysis*

Figure 3a shows the FTIR spectra of raw coal and the four semi-coke samples, which exhibit similar absorption bands primarily consisting of oxygen-containing groups, aliphatic C–H groups, aromatic nucleus C=C, and substituted aromatic rings, while their intensities of absorption bands vary considerably. For further observing the changes of significant functional groups in the FTIR spectra, the baselines of zone representing aliphatic C–H groups from 3000 to 2800 cm−1, zones representing oxygen-containing functional groups from 1800 to 1500 cm−<sup>1</sup> and 1350 to 1000 cm<sup>−</sup>1, and zone representing aromatic C–H groups from 900 to 700 cm−<sup>1</sup> were corrected, as shown in Figure 3b–e, respectively. According to the literature [23,24,26,31–33], band assignments are shown in Table 3.


**Table 3.** Band assignments derived from FTIR spectra.

As shown in Figure 3a,b, the intensities of peaks at 2955 cm<sup>−</sup>1, 2920 cm−1, 2852 cm−1, and 1465 cm−<sup>1</sup> for aliphatic C–H functional groups steadily decrease with increasing temperatures due to the thermal cleavage of aliphatic chains. Besides, these fractured aliphatic segments will participate in polycondensation reactions during aromatization to form larger condensed aromatic nucleus polymers. It is worth noting that the peak intensity at 3000–2800 cm−<sup>1</sup> of C450M is greater than that of C450, indicating that C450M has a higher concentration of aliphatic C–H groups as well as a lower condensation degree of aromatic nuclei. Conversely, the content of aliphatic C–H groups of C750M are slightly lower than that of C750, suggesting that the condensation degree of aromatic nuclei in C750M increases with the addition of B4C.

As can be seen in Figure 3a,c,d, the intensities of peaks at 3400 cm−1, 1702 cm−1, 1262 cm<sup>−</sup>1, 1097 cm−1, 1032 cm−1, and 1010 cm−<sup>1</sup> for oxygen-containing functional groups monotonously decrease with the increasing temperatures, even the four peaks at 1702 cm<sup>−</sup>1, 1165 cm−1, 1032 cm−1, and 1010 cm−<sup>1</sup> disappear in semi-cokes at 750 ◦C. This is because these –COOH, O–H, and C–O bands are gradually broken or destroyed in the HVC carbonization process and released through small molecules gases [17,18]. It is interesting that the intensity of peak at 1610 cm−<sup>1</sup> for C=C stretching band also declines with the increase in temperature. Ideally, the C=C stretching band representing the condensation

degree of aromatic nuclei will increase with increasing carbonization temperature, but the presence of plenty of phenolic groups and COO– groups in low-rank HVC (C 76.93 wt.% in Table 1) is likely to increase intensity of the 1610 cm−<sup>1</sup> band [34]. Therefore, the decrease in intensity of the peak at 1610 cm−<sup>1</sup> is mainly due to the decomposition of oxygen-containing groups in the carbonization process of HVC.

**Figure 3.** FTIR spectra of raw coal and four semi-cokes: (**a**) FTIR spectra from the 4000–400 cm−<sup>1</sup> zone; (**b**) FTIR spectra after subtracting the baseline from the 3000–2800 cm−<sup>1</sup> zone; (**c**) FTIR spectra after subtracting the baseline from the 1800–1500 cm−<sup>1</sup> zone; (**d**) FTIR spectra after subtracting the baseline from the 1350–1000 cm−<sup>1</sup> zone; (**e**) FTIR spectra after subtracting the baseline from the 900–700 cm−<sup>1</sup> zone.

It is worth noting that a new peak at 1121 cm−<sup>1</sup> (in Figure 3d) attributed to a B–O bond is observed in C450M [29,35,36], indicating that partial active oxygen obtained from oxygen-containing fragments were consumed by reaction with B4C to form the boron oxide in the plastic zone of HVC. In addition, by comparing the peak strength of oxygencontaining functional groups at the same position in C450 and C450M, the strength of all oxygen-containing peaks was increased by adding B4C. One of hypotheses was that substitution reactions between boron compound and oxygen-containing functional groups occur, which leads to the formation of organically bound B–O groups with higher bond energies. Simultaneously, a shift of the C=C stretching band from 1630 cm−<sup>1</sup> (in C750) to 1634 cm−<sup>1</sup> (in C750M) and a new band at around 1219 cm−<sup>1</sup> representing B–C stretching vibration are

observed. Generally, an increase in wavelength of the C=C stretching band is attributed to the generation of the B–C band. Meanwhile, boron atoms should be incorporated in the sp<sup>2</sup> C networks in coke structure. These two explanations are clarified in related articles [29,36]. However, compared with conditions in these articles, the experimental temperature in this work was lower. Therefore, whether the explanations introduced in the above articles are suitable to support the phenomena in this work still needs to be illustrated.

As shown in Figure 3e, these fingerprint absorption peaks at 900–700 cm−<sup>1</sup> are caused by aromatic C–H out-of-plane bending vibrations. It is generally accepted that the degree of aromatic substitution and condensation of aromatic nuclei rely on the number of adjacent hydrogens per ring [24]. The intensities of peaks at 812 cm−<sup>1</sup> and 750 cm−<sup>1</sup> decrease gradually and the intensity of peaks at 870 cm−<sup>1</sup> increases gradually with increasing temperature, indicating that the concentrations of highly-substituted aromatic rings and aromatic structures with 1–2 rings decrease in the carbonization process and the size of condensed aromatic nuclei becomes larger. It is noteworthy that the peak intensity at 876 cm−<sup>1</sup> of C450M is lower than that of C450, but the peak intensity at 876 cm−<sup>1</sup> of C750M is higher than that of C750 when the temperature arrives at 750 ◦C. Simultaneously, both peak intensities at 801 cm−<sup>1</sup> and 750 cm−<sup>1</sup> of C450M are higher than those of C450, but both peak intensities of C750M are lower than those of C750. These results show that the addition of B4C can retard condensation reactions in the plastic zone and increase the secondary cracking reactions in the fast polycondensation zone of HVC, which will lead to the formation of larger sized condensed aromatic nuclei in C750M.

#### *3.5. HRTEM Analysis*

The lattice fringes cannot be observed directly from HRTEM images, so these original images were organized to acquire the lattice fringe images and every step is briefly described as follows: firstly, using Fourier transformation to cope with original images obtains frequency domain images; secondly, using rounded filtering eliminates the disordered part in these images; thirdly, using inverse Fourier transformation transfers the images obtained after the second step to new images; then, using threshold segmentation to handle the new images obtains black-and-white binary images that can present microcrystalline fringe; next, initially etch, then expend, and finally skeleton process the black-and-white binary images to obtain lattice fringe images [37]. HRTEM images and the corresponding lattice fringe images of raw coal and four semi-coke samples are shown in Figure 4. According to Figure 4, these lattice fringe images show a striking difference in the shape, size, and orientation of the layers. For the raw coal, the majority of layers are small, twisted, and lack orientation; that few stacks can be observed in Figure 4b, which is consistent with its low coalification. With the elevation of carbonization temperature, aromatic layers' stacking-number and their size increase while they become better-orientated in semi-cokes, as shown in Figure 4d,f. In addition, these changes of aromatic layers become more evident in semi-cokes at 750 ◦C (in Figure 4h,j). These results show that the crystallinity of semi-coke increases with the increase in carbonization temperature.

According to the literature [25,37], the number of aromatic carbon atoms can be determined by the lattice fringe length of HRTEM image; therefore, the extracted lattice fringe images were analyzed by image processing software (ImageJ V1.47) to calculate the aromatic fringe size. The relationships between lattice fringe length and aromatic sheet assignments are listed in Table 4, and the distribution frequency of lattice fringe length in the samples is also shown in Table 4 through the form of the average value based on the examination of three different regions containing 1000 aromatic fringes.

Table 4 shows a summary of the classification of the aromatic fringes by fringe lengths and their frequency of occurrence in the aromatic fringe population. For raw coal, the concentrations of benzene and parallelogram-shape aromatic structures of (<4 × 4) occupy 17.16% and 80.09%, respectively, while the ratio of parallelogram-shape aromatic structures of (>3 × 3) are only 2.75%. This result indicates that HVC contains large quantities of small size condensed aromatic structure, which are responsible for its weak thermal stability and the formation of plenty of free radicals in the plastic zone [13,16,26,38]. With increasing carbonization temperature, the concentrations of parallelogram-shape aromatic structures of (>3 × 3) increase significantly in semi-cokes, as shown in Table 4, suggesting that large quantities of condensation and repolymerization reactions occur in semi-cokes at 450 ◦C and 750 ◦C. It is noticeable that the concentration of parallelogram-shape aromatic structures of (>3 × 3) in C450M is lower than that in C450 whereas the concentration of the above structures in C750M is higher than that in C750, indicating that adding B4C contribute to constraining condensation reactions in the plastic zone and promoting repolymerization reactions in the fast polycondensation zone.

**Figure 4.** High-resolution transmission electron microscopy (HRTEM) images and the corresponding extracted lattice fringe images of coal and semi-coke samples: (**a**,**b**) raw coal; (**c**,**d**) C450; (**e**,**f**) C450M; (**g**,**h**) C750; (**i**,**j**) C750M.

**Table 4.** Aromatic fringe assignments and distribution frequency based on the analysis of HRTEM fringe images.


#### *3.6. Role of the B4C in the Improvement of Coke Quality*

B4C reacts with active oxygen obtained from oxygen-containing compounds in the plastic zone of HVC, which leads to reducing the reactions between active oxygen and transferable hydrogen. Therefore, more transferable hydrogens are available to stabilize free radicals, reducing the condensation and crosslinking reactions of free radicals. In this case, the anisotropic mesophase structures develop in the plastic zone of modified coal, which contributes to increasing the degree of anisotropy in coke. Simultaneously, more stable free radical fragments will be involved in the polymerization reaction under higher temperature, increasing the size of the aromatic sheet and condensed degree in semi-coke, as shown in C750M. Consequently, with further elevating temperature, the size of aromatic layer will increase in coke structures by adding B4C. In summary, adding B4C into HVC to produce coke has the result of increasing the size of the aromatic layer and anisotropic degree in the coke structure, which contributes to enhancing the coke resistance to CO2 reaction, ultimately resulting in the significant improvement of coke quality.

#### **4. Conclusions**

In this work, in addition to a new additive (B4C) that is primarily introduced, the modification mechanisms of coke quality were analyzed by using TG, FTIR, and HRTEM techniques. The main conclusions are summarized as follows:


**Author Contributions:** Conceptualization, Q.W. and Z.-Z.Z.; software, Q.W. and C.-Y.Z.; validation, Q.W., C.S. and Y.-D.W.; formal analysis, Q.W. and C.S.; investigation, Q.W.; resources, Z.-Z.Z.; data curation, Q.W.; writing—original draft preparation, Q.W.; writing—review and editing, Q.W. and C.S.; visualization, Q.W.; supervision, Z.-Z.Z. and C.-Y.Z.; project administration, Q.W.; funding acquisition, Z.-Z.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Key Project of Science and Technology of Chongqing, grant number CSTS.2010AB4084.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy issues.

**Acknowledgments:** The authors gratefully appreciate the National Natural Science Foundation of China, grant number 51044005 for financial support.

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

#### **References**


### *Article* **Crumpled Graphene-Storage Media for Hydrogen and Metal Nanoclusters**

**Liliya R. Safina 1,\*, Karina A. Krylova 2,3, Ramil T. Murzaev 2, Julia A. Baimova 2,3 and Radik R. Mulyukov 1,2**


**Abstract:** Understanding the structural behavior of graphene flake, which is the structural unit of bulk crumpled graphene, is of high importance, especially when it is in contact with the other types of atoms. In the present work, crumpled graphene is considered as storage media for two types of nanoclusters—nickel and hydrogen. Crumpled graphene consists of crumpled graphene flakes bonded by weak van der Waals forces and can be considered an excellent container for different atoms. Molecular dynamics simulation is used to study the behavior of the graphene flake filled with the nickel nanocluster or hydrogen molecules. The simulation results reveal that graphene flake can be considered a perfect container for metal nanocluster since graphene can easily cover it. Hydrogen molecules can be stored on graphene flake at 77 K, however, the amount of hydrogen is low. Thus, additional treatment is required to increase the amount of stored hydrogen. Remarkably, the size dependence of the structural behavior of the graphene flake filled with both nickel and hydrogen atoms is found. The size of the filling cluster should be chosen in comparison with the specific surface area of graphene flake.

**Keywords:** crumpled graphene; Ni-graphene composite; hydrogen; molecular dynamics; storage media

#### **1. Introduction**

The transformation of nanoscale structural elements into three-dimensional (3D) complex architecture is currently an important task of materials science. Since the discovery of fullerenes in 1985, a lot of new carbon structures have been proposed. Carbon polymorphs can be used to obtain nanostructures with unique mechanical and physical properties applicable in nanoelectronics, energy storage devices, sensors, supercapacitors, Li-ion batteries, etc. [1–7].

Graphene is capable of enhancing the performance, functionality as well as durability of many applications. The one-atom thin structure can serve as the platform for other materials especially since the graphene layer can be bent or crumpled. Extensive studies have been carried out in recent decades to investigate the crumpling behavior of thin sheets like graphene, by both theoretical and experimental methods [8–19]. It was shown that crumpled structures can have excellent compression and aggregation-resistant properties [5,6,9,10,20–22]. Crumpled graphene can be used for the production of new supercapacitors [19]. However, one of their new and important applications is that of a natural container for other atoms.

It is known that coupling metal nanoclusters with carbon materials can efficiently promote catalysis and electrocatalytic activities [23,24] or prevent the corrosion of metals [25]. redAmong metals, some can be easily attached to its surface, while others even repulse from the surface. The bigger the carbon solubility of metal, the more amounts of carbon can be attracted to the metal surface. Metals such as Ni, Cu, Pt, or Au are the most preferable

**Citation:** Safina, L.R.; Krylova, K.A.; Murzaev, R.T.; Baimova, J.A.; Mulyukov, R.R. Crumpled Graphene-Storage Media for Hydrogen and Metal Nanoclusters. *Materials* **2021**, *14*, 2098. https:// doi.org/10.3390/ma14092098

Academic Editor: Gueorgui Gueorguiev

Received: 17 March 2021 Accepted: 14 April 2021 Published: 21 April 2021

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

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

as catalysts for the growth of graphene [26–28]. However, nickel is one of the most broadly used metals for a range of graphene applications, including the growth of carbon nanotubes (CNTs) and graphene [29–31]. Lattice mismatch is one of the important factors for choosing the metal for metal–graphene interface and Ni(111) surface is the closest matched interface with respect to the graphene of all transition metals [26]. In [29], a simple and scalable method for the synthesis of hollow graphene balls using a Ni nanocluster template was developed. The interface between graphene and Ni attracts considerable interest due to the possibility of the synthesis of large-area graphene on metal substrate [32–36]. Since the fabrication of graphene layers on the metal substrate was very successful, the idea of the fabrication of metal–graphene layered composites was raised by [37–40]. Graphene added into the copper and nickel matrix by chemical vapor deposition can considerably improve the strength of metal [38]. The overview of graphene–nickel composites can be found in [39]. Both the theoretical and experimental works on Ni-graphene layered composites showed that the mechanical properties of metal-matrix composite reinforced with the graphene layer considerably improved. In the present work, the other approach to obtain the Ni-graphene composite was applied: introducing metal nanoparticles to the crumpled graphene matrix. In [41–44] it was shown that such composites can be obtained through compression combined with high temperatures.

The other important advantages of such structures are a large specific surface area (SSA) and, respectively, a high rate of gas adsorption, which makes it possible to predict their use in hydrogen storage [45–53]. Various structural parameters can affect the degree of hydrogen accumulation [45]. For example, the amount of adsorbed hydrogen increases with an increase in the CNT diameter, since this increases the surface area on which hydrogen can be adsorbed. An increase in the distance between graphene sheets in the structure also leads to an increase in the gravimetric hydrogen absorption density. Another promising carbon structure is two layers of graphene connected by short CNTs with graphene cones added on top of the structure. The idea of such a structure appeared when it was shown that the hydrogenation of graphene can lead to the buckling of the graphene sheet [54]: the amount of accumulated hydrogen increased from only 3 wt% for the simple graphene plane to 20 wt% for graphene plane with graphene cone. Carbon nanostructures doped with alkali metals (Li and K) can adsorb even more hydrogen [48–53]: 14 wt% (Li) and 20 wt% (K) of hydrogen at moderate conditions, in contradiction with the lower values reported later [50]. The Li-doped activated carbon [55] can store from 2.1 to 2.6 wt% of H2 at 77 K and at 2 MPa, which shows that at given pressure–temperature conditions the amount of stored hydrogen can be increased. It should be mentioned that the required amount of stored hydrogen for carbon nanostructures was still not achieved. Thus, an active search for new materials and structures for hydrogen storage and transportation is of high importance.

In the present work, two types of nanofillers for crumpled graphene flake were considered—Ni and hydrogen nanoclusters. Thus, the idea of using crumpled graphene flake for the storage of different atoms using Ni and hydrogen clusters as the example was realized. It is important to understand the dynamics of interaction between single graphene flake and a metal or non-metal nanocluster before the study of storage in three-dimensional crumpled graphene. The behavior of the hybrid structure is studied by atomistic simulation at different temperatures. For the carbon–nickel system, room temperature (300 K) and temperature close to the melting point of Ni (1000 K) were chosen. However, for the carbon–hydrogen system, the temperature of liquid nitrogen (77 K) and room temperature were chosen. The dynamics of the structure were studied during exposure for 20 ps.

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

To study the interaction between graphene flake (GF) and atomic nanocluster, an initial structure composed of the GF of one size (*NC* = 252) and atomic clusters of different sizes (*N* = 21; 38; 47; 66; 78) inside the flake was considered (see Figure 1a). One graphene flake was a small CNT(11,11) 1.3 nm-long with two atomic rows deleted along the axis of the nanotube. Since the size of this small sample of graphene along three dimensions is very small (1.3 nm along the *z* axis, 3.6 nm along the *y* axis and one-atom-thick along the *x* axis) it is called "flake" rather than "nanoribbon". This GF was filled with nanoclusters of two types: (i) nickel nanocluster and (ii) hydrogen nanocluster.

**Figure 1.** (**a**) Initial structure: graphene flake with the diameter *D* filled with nanocluster with the diameter *d*; (**b**) atomic and molecular hydrogen inside the graphene flake.

The maximum size of the nanocluster was chosen to almost completely fill the cavity of the graphene flake. The size of the structural elements is shown in Table 1. The distance between the edge of the nanocluster and the side of the flake can be defined as *a* = (*D* − *d*)/2. However, these parameters affect the structural transformation just in the case of metal nanocluster since it is a solid particle. a hydrogen cluster is gas and small H atoms can easily spread inside GF.


**Table 1.** The size of the structural elements.

It should be mentioned, that nanocluster Ni78 was considered here as nanocluster of critical size for the chosen size of the GF. However, for the metal cluster, this size was too big and the sides of GF were closer than it should have been in a real system. To eliminate the negative effect of the overly sized nanocluster, nanocluster Ni66 was considered the characteristic for the case of metal–graphene interaction. While for the case of hydrogen– graphene interaction, an even bigger size of the nanocluster could be considered since the size of the hydrogen atoms was much smaller than the size of Ni atoms.

The simulation was conducted using a large-scale atomic/molecular massively parallel simulator (LAMMPS). Equations of motion for the atoms were integrated numerically using the fourth-order Verlet method with the time step of 0.2 fs. The Nose–Hoover thermostat was used to control the system temperature. The periodic boundary conditions were applied in all directions, however, the simulation box is much bigger than the size

of GF filled with nanocluster. The adaptive intermolecular reactive empirical bond order potential (AIREBO) [56] was used to describe the interatomic interactions between carbon atoms, including both covalent bonds in the basal plane of graphene and van der Waals interactions between GF and nanocluster. The simulation configurations were visualized by Visual Molecular Dynamics (VMD) Software [57].

#### *2.1. Graphene with Nickel Nanocluster*

Graphene flake with the nanocluster was exposed at 300 K and 1000 K for 20 ps to study the dynamics of nanocluster inside graphene flake. Previously [41], it was shown that the melting temperature of the Ni nanocluster was about 1300 K. Thus, in the present work, the highest considered temperature was 1000 K. At this temperature, Ni nanocluster was close to melting but not melted yet, even for the smallest nanocluster, which consisted of 21 atoms. Initially, Ni atoms were packed into the face-centered cubic lattice.

To describe the interatomic interaction between Ni–Ni and Ni–C, Morse interatomic potential was used with the parameters *De* = 0.4205 eV, *Re* = 2.78 Å and *β* = 1.4199 1/Å for Ni–Ni [58]; and *De* = 0.433 eV, *Re* = 2.316 Å, *β* = 3.244 1/Å for Ni–C. The parameters of the Morse potential for describing the interaction of nickel and carbon atoms were recently proposed using *ab-initio* simulation [59,60].

#### *2.2. Graphene with Hydrogen Nanocluster*

Graphene flake with hydrogen nanocluster was exposed at 77 K and 300 K for 20 ps. The temperature of 77 K was chosen since it was previously shown that better sorption of hydrogen can be found at 77 K [61]. The temperature of 300 K was chosen to study the dynamics at room temperature, however, it is known that this temperature is too high for hydrogen storage [46].

To describe the interatomic interaction between C and H, AIREBO interatomic potential was used.

It should be mentioned that the initially obtained structure contains atomic hydrogen, however, H atoms transform to H2 molecules and just several H atoms remain single (see Figure 1b). As it was shown, the lowest binding energy between graphene and H2 was observed when the distance between C and H2 in the range from 2.9 to 3.2 Å [61]. Hydrogen molecules can be bonded by van der Waals interaction when they move close enough to the side of GF. Single hydrogen atoms can be chemically adsorbed on graphene by covalent bonding.

#### **3. Results**

#### *3.1. Graphene with Ni Nanocluster*

At first, the evolution of the potential energy of the system during the crumpling process for graphene flake filled within spherical the Ni nanocluster was analyzed. In Figure 2, potential energy as the function of exposure time at 300 K (a) and 1000 K (b) for five types of nanoclusters was shown. It was found that the total potential energy of the system was saturated to a practically constant value at the end of the equilibration process, indicating that the system reached equilibrium and a stable state. All the changes in the energy curves correspond to some structural changes.

The longest time of stabilization at 300 K was found for Ni21 (5 ps). Curves for nanoclusters of close diameter are almost the same. Graphene flake with Ni38 and Ni47 reach the equilibrium state at about 4 ps of exposure, while the GF with Ni66 and Ni78 reach equilibrium state at 3.2 ps. The bigger the diameter of the nanocluster, the less the equilibration time. This can be explained by the mutual arrangement of the nanocluster and GF. For a small nanocluster, the distance *a* is two times higher than for the biggest one which means that the time required to attach the nanocluster by the graphene flake is longer. For nanocluster Ni78, 3 ps is enough for GF to fully cover the nanocluster, while for Ni21, not only coverage took place, but also the further crumpling of the flake with the changing of the round shape of nanocluster.

At higher temperature (1000 K), again a strong correlation between equilibration time and the size of the nanocluster was found. For GF with Ni21 nanocluster, the transformation was fast since the temperature was close to the melting point and the nanocluster can be easily destroyed. Temperature fluctuations facilitate the crumpling process and the appearance of new bonds between the edges of GF. For bigger nanoclusters, the temperature slightly affects the time of equilibration. Again, the same values of equilibration time were obtained for Ni38 and Ni47 (about 4 ps) and for Ni66 and Ni78 (about 3.2 ps). However, temperature decreases the total potential energy of the system.

**Figure 2.** Potential energy as the function of exposure time at 300 K (**a**) and 1000 K (**b**) for five types of nanoclusters.

In Figure 3, the process of crumpling for nanoclusters Ni21 (a), Ni47 (b) and Ni66 (c) was shown in details after exposure at 300 K (blue line) and 1000 K (red line). At the beginning of the crumpling process, GF startED to change its round shape, and the fcc crystalline order of Ni nanocluster WAs destroyed. Metal atoms are attracted by the graphene surface and tend to occupy equilibrium positions above the center of the carbon hexagon, which was also mentioned in [41]. Ni atoms WEre interacting with graphene hexagons by van der Waals forces. The edges of GF can be bonded during exposure.

At first, consider GF with Ni21 (see Figure 3a, lower line of snapshots). At 300 K, GF lost its round shape at about 0.5 ps, edges of GF move towards each other, and at *t* = 1.5 ps one covalent bond appeared between two edges. Nanocluster disturbed by the temperature and several atoms attach to the graphene plane (state II). This leads to the spreading of the Ni atoms over the graphene plane (state III). Simultaneously, such a small nanocluster allowed the graphene flake to easily bend and the structural unit transforms to the bi-layer graphene with Ni atoms spread inside (state IV). Several more covalent bonds appeared on the side edges of GF. At 300 K, not many covalent bonds between the edges of GF can appear, since the temperature is far from the melting temperature of graphene [62,63].

Increase in the exposure temperature to 1000 K facilitates the transformation. At 300 K, equilibration time is equal to 5 ps, while at 1000 K, it is equal to 3.5 ps. Subsquently, the structure came to a stable state and no further changes were observed despite slight thermal fluctuations. At 1000 K, more carbon atoms on the edges of GF found neighbors since the structure is disturbed by thermal fluctuations.

**Figure 3.** Potential energy as the function of exposure time for Ni21 (**a**), Ni47 (**b**) and Ni66 (**c**). Corresponding snapshots of the structure of crumpled graphene flake filled with nanocluster during exposure at 300 K (bottom line of snapshots) and 1000 K (upper line of snapshots). Carbon atoms are shown by violet and nickel atoms are shown by green.

For structures with Ni47 and Ni66, the behavior is qualitatively close. An almost round Ni nanocluster inside the graphene flake was observed at the initial state at both 300 K and 1000 K (as can be seen in Figure 3b,c). The edges of GF can attach to each other with the formation of new covalent bonds. During exposure, graphene flake transforms into a capsule containing nickel nanocluster. The nickel nanocluster almost completely fills the graphene flake, in comparison with Ni21. As a result, it is difficult to deform such a structure. However, at 1000 K, the nanocluster is more disturbed and the stable state (state III, the top line of snapshots) reached at a lower equilibration time. The nickel nanocluster became more planar since GF is rigid and can bend the soft metal nanocluster. In the case of Ni47, GF transforms the "bag" for nickel, while Ni66 nanocluster is too big and edges on both sides cannot be attached. In the case of Ni66, GF just covers the Ni nanocluster as much is possible.

A graphene flake with Ni38 behaves in a similar way to Ni47, while GF with Ni78 behaves similarly to Ni66. Note that GF always tends to wrap the metal cluster.

One of the important applications of such structural elements is the fabrication of composites. In Figure 4, the initial structure (a, a') was composed of graphene flake filled with Ni nanocluster Ni21 and Ni78 correspondingly. The initial structure is compressed at 1000 K to obtain graphene–nickel composite material (c, c'). Composite can be fabricated under hydrostatic compression at high temperatures [41–44].

**Figure 4.** Snapshots of the structure of crumpled graphene filled with (**a**–**c**) Ni21 and (**a'**–**c'**) Ni78 nanocluster. Initial structure was obtained by annealing at 300 K, while composite was obtained by hydrostatic compression at 1000 K. **Ni** atoms are shown by purple and **C** atoms are shown by green color.

#### *3.2. Graphene with Hydrogen Nanocluster*

In Figure 5, the potential energy of GF filled with hydrogen cluster during exposure at 77 K (a) and 300 K (b) is presented as the function of equilibration time. Similar to GF with the Ni nanocluster, five structural units were divided into three groups—GF with 21 hydrogen atoms; GF with 38 and 47 H atoms; and GF with 66 and 78 H atoms. Despite that the hydrogen atoms transform into hydrogen molecules, the initial number of atoms was also used, since the number of H2 molecules and single hydrogen atoms can change from one simulation run to another which depends on thermal fluctuations.

At 77 K (Figure 5a), the equilibrium state is the state when hydrogen atoms found their sits and GF change the cylinder shape to the one with minimal energy. At 300 K (Figure 5a), the equilibrium state is the state when all hydrogen atoms disrobed from the side of GF or even fully leave the cavity of GF. This would be discussed later together with the description of corresponding snapshots. The biggest drop of the energy value took place during the first picosecond and connected with the change of initial nonequilibrium configuration of the cluster and slightly with the change of shape of the GF.

At 77 K, the time of equilibration was the longest (3 ps) for hydrogen clusters consist of 77, 66, and 38 atoms, and the shortest (1.5 ps) is for 21 and 47 atoms. At that temperature, the size of the cluster plays quite an important role which is connected with the number of sites for hydrogen on the side of GF. All hydrogen molecules and atoms can easily find sites on graphene for an initial number of atoms less than 47, while for bigger clusters, the number of sites is not enough and some molecules and atoms will move outside GF or seek better sits near GF.

**Figure 5.** Potential energy as the function of exposure time at 77 K (**a**) and 300 K (**b**) for five types of hydrogen nanoclusters.

At 300 K, the main point is the process of dehydrogenation which is quite quick at such a high temperature, and for clusters with 38–78 atoms, about 1.5 ps is enough for all molecules to detach the side of GF and move outside. The long time of equilibration for cluster H21 can be explained simply: there are about nine molecules and three atoms and all the time spent for the slow movement of this hydrogen. A large number of atoms in the cluster pushes the sides of GF and opens it much faster.

To understand the dynamics of the hydrogen sorption/desorption, snapshots of the structural units altogether with their potential energy are presented in Figure 6 for three groups. If the hydrogen cluster is quite small (21 atoms), there are a lot of vacant places on the graphene plane to attach hydrogen. Only several atoms or molecules can move outside GF at a low temperature equal to 77 K. Most of the hydrogen molecules attached by van der Waals force to graphene and slightly moving along graphene flake.

At 300 K, hydrogen molecules and atoms have no chance to form even a weak bond to graphene. At 300 K, all hydrogen atoms move outside GF which is quite understandable. Numerous theoretical and experimental works confirm the sorption of molecular hydrogen on carbon nanostructures [50,64–66] in a specific temperature range (50–200 K). Here, such a high temperature was used to analyze how hydrogen will leave the cavity of GF. As it can be seen from Figure 6a, this process is very fast and at 4 ps, only two molecules stay inside GF (stage IV, upper line).

A large number of hydrogen atoms almost completely fills the cavity of the GF. At 77 K, hydrogen atoms are easily converted into hydrogen molecules and under the action of chemical attraction or van der Waals forces, attach to the walls of the graphene flake. Mainly, hydrogen, which is located in the center of the flake under the influence of temperature, tends to leave the graphene cavity. Therefore, in structures with 66 and 78 hydrogen atoms, the stabilization of the structure takes longer (3 ps), in contrast to small hydrogen clusters, such as 21 and 47 (1.5 ps).

**Figure 6.** Potential energy as the function of exposure time for three types of hydrogen nanoclusters: (**a**) 21 atoms; (**b**) 47 atoms; and (**c**) 78 atoms. Corresponding snapshots of graphene flake filled with nanocluster during exposure at 77 K (bottom line of snapshots) and 300 K (upper line of snapshots). Carbon atoms are shown by violet and hydrogen atoms are shown by blue.

For a bigger nanocluster, energy curves almost coincide for two temperatures, since hydrogen molecules at 77 K can find the places for sorption just after the first steps. In this case, hydrogen atoms are placed near graphene and link it as far as the van der Waals radius of the molecule reaches. Thus, the hydrogen adsorption capacities depend considerably on the initial size of the hydrogen cluster. The SSA of the GF should be big enough for a chosen number of H atoms and H2 molecules. Here, SSA is equal to 1153.9 m/g2, which is enough to settle down 21–38 H atoms, but not enough for a bigger number of H atoms.

In Figure 7, snapshots of GF with 78 hydrogen atoms during exposure at 77 K for 20 ps are presented. As it can be seen, some hydrogen molecules attach the opposite side of GF since there were no sites inside the cavity. During exposure, even at 77 K, GF opens and then closes again. If the exposure time would be increased to even 100 ps, GF will move like the wings of a flying bird with adsorbed hydrogen atoms. This state is equilibrium and can be preserved for a long time.

As well as for a metal nanocluster, the three-dimensional structure of crumpled graphene filled with hydrogen is presented in Figure 8. The corresponding structure was considered in [46,47], where the effect of hydrostatic compression on the possibility of hydrogen storage was studied. In Figure 8a, the initial structure of crumpled graphene was presented. As it can be seen, initially there are too many channels for hydrogen to move out of the structure. Thus, additional treatment was required to save hydrogen inside the pores of crumple graphene. In Figure 8b, the structure after 40% of hydrostatic compression was presented. In such a compressed structure, hydrogen can be stored much more effectively than in undeformed crumpled graphene [46,47]. It can be concluded that crumpled graphene is an effective storage media for hydrogen. However, a search for the improvement of the quantity of stored hydrogen should be found, for instance, in the introduction of metal atoms. From this point of view, understanding the interaction between hydrogen and metal nanoclusters and GF is of high importance.

**Figure 7.** Snapshots of graphene flake with 78 hydrogen atoms during exposure at 77 K for 20 ps. Colors are as in Figure 6.

**Figure 8.** Snapshots of crumpled graphene filled with hydrogen atoms before (**a**) and after (**b**) hydrostatic compression. Colors are as in Figure 6.

#### **4. Conclusions**

Molecular dynamics simulation is used to study the dynamics of graphene flakes filled with the nanoclusters of two different types: metal and non-metal. The obtained results can shed the light on understanding the possibility of using crumpled graphene as a storage media from the point of single flake behavior.

It was found that cavities of crumpled graphene (the structure consists of crumpled graphene flakes) can be used as containers for metal nanoclusters—for instance, for Ni. A nanocluster consisting of 21 to 78 Ni atoms was considered. A graphene flake can easily cover nanocluster, however, the dynamics of the interaction strongly depend on the nanocluster size. Small nanoclusters can be easily bent by rigid graphene flake, while the biggest conserve the shape. Such a structure, composed of Ni nanoclusters and graphene flakes, can be further used to obtain composite material with improved mechanical properties.

The problem of hydrogen storage has been of high importance for decades and the application of carbon nanostructures from this point of view also looks promising. Crumpled graphene has a high specific surface area, light weight and a lot of pores and cavities which can be filled with hydrogen. It was shown that at 77 K, hydrogen molecules and atoms can be absorbed by both sides of graphene flakes. However, a single flake cannot store enough hydrogen for practical application. However, when graphene flakes composed in another structure and with additional treatment like hydrostatic compression [46,47], it can be successfully used for hydrogen storage and transportation.

**Author Contributions:** Software and initial simulation setup were conducted by L.R.S. and R.T.M. The molecular dynamics study and drafting the manuscript were conducted by J.A.B., K.A.K. and R.R.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** Part of this research was supported by the grant of Russian Science Foundation (No. 20-72-10112). Work of R.R.M. and R.T.M. supported by the program of fundamental researches of Government Academy of Sciences of IMSP RAS.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

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

#### **Abbreviations**

The following abbreviations are used in this manuscript:


#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Materials* Editorial Office E-mail: materials@mdpi.com www.mdpi.com/journal/materials

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com

ISBN 978-3-0365-3440-4