Direct Synthesis of Nanofibrous Nonwoven Carbon Components: Initial Observations, Capabilities, and Challenges

Carbon nanofibers, also known as carbon filaments, have been studied for more than four decades in an effort to understand the mechanisms by which they grow and how to control their properties [1–3]. Vapor-phase synthesis is a scalable method by which to create CNFs from the decomposition of carbon-containing gases using a catalyst of sufficient activity [4]. Catalysts may be pure metals or alloys, with the efficiency being determined by a combination of factors including the choice of gas(es) and reaction temperature. Pure metals known to efficiently catalyze carbon growth include palladium, platinum, nickel, and cobalt [5–7].

Alloys also work well in the synthesis of CNFs and often show higher activity and selectivity. These often contain iron, cobalt, nickel, or copper [8]. Processing conditions dictate maximum deposition kinetics, fiber size, and surface morphology and crystallinity. The numerous governing parameters provide opportunity to tailor the fiber properties.

One method for achieving efficient production and utilization is direct synthesis. By this method, complex materials and components may be created in a single step or as a continuous process and can be designed in a useful, structured way. Using bottom-up architecture, this entails synthesizing the materials into a final (or near final) component in one setup. The work presented here was conducted to determine the feasibility of directly synthesizing carbon nanofibers into useful forms with centimeter dimensions in order to simplify their incorporation into current and future applications.

Carbon nanotubes (CNTs) are produced by methods such as arc discharge, laser ablation, and chemical vapor deposition (CVD) [9–11], and CNFs are often produced by electrospinning [12–14] and carbonization or CVD methods [2,9,15]. The cost, throughput, and selectivity for properties (e.g., chirality, diameter, morphology, etc.) vary for each technique, but generally speaking, manufacturing research is centered on producing the material, not on application-end integration. Of course, application development occurs concomitantly with material synthesis, and as applications drive production, improved production technology enables new applications.

Prior efforts in scaling up the production of carbon nanotubes and nanofibers using a floating catalyst method [16] have allowed them to be produced continuously or semicontinuously (e.g., Ref. [19]). This method employs a growth temperature of 1100–1200 °C and a vapor-phase catalyst (e.g., ferrocene). The catalyst-bearing gas is introduced into a reaction chamber along with the hydrocarbon, hydrogen, and typically a sulfur source [15].

When compared on a per mass basis, the production of CNFs by this method was found to be 13–50 times more energy intensive than common engineering materials such as aluminum and steel [18], although the authors of that study concede that due to limited data on new materials, the numbers are prone to some error. They suggest two key points based on their findings. First, although CNFs may have a larger life cycle environmental impact, products based on these materials may actually be “greener” overall due to improved performance characteristics. Second, and most pertinent to this work, the authors state, “Research efforts are required to increase process yield and optimize the growth of CNFs…” [18].

Carbon nanotubes and nanofibers present an abundance of possibilities in advanced applications such as composites [19,20], sensors [21,22], and much more. Getting the most from these materials necessitates simplification and scaling of production methods to achieve cost competitiveness. While much of the progress has centered on creating the materials, applying those materials readily and reliably still presents challenges.

Most catalytic processes result in loose fibers, which must be contained in some way, usually by a matrix, such as epoxy, to form a composite [19,20]. Electrospinning is able to build up layers from very long fibers which are patterned by electrical field, but the fiber collection is two-dimensional in nature (e.g., fabrics). The carbonization step also increases time and complexity and requires high-temperature (e.g., 1000 °C or more) that precludes the creation of multicomponent devices including materials which are not suitable at that temperature.

A recently developed method for creating bulk materials which are entirely nanostructured has potential to allow finished “components” to be created from a small charge of starting catalyst in a single step. Referred to as the CoFFiN process, this simple technique uses a mold to control the macroscale geometry of the nanoscale fibers [23]. The resulting nonwoven carbon component has been found to be mechanically robust [24] and having properties (such as density and stiffness) which can be controlled. The unique benefit of this process is the ability to create the final geometry during synthesis, akin to weaving traditional carbon fibers directly into the shape of the product, such as an automotive panel. This simplifies post processing and reduces waste. The challenges addressed in this work include understanding the limits of the CoFFiN process and how to extend it to larger volumes while still achieving a mechanically robust, nonwoven material.

Carbon nanofibers can serve in a variety of applications, since they may have a variety of graphitic structures (not strictly concentric tubes, like multiwall CNTs). These unique conformations include “stacked cup” [25,26] and “herringbone” [1] structures as well as graphitic planes oriented perpendicular to the longitudinal axis of the fiber [27]. In addition to the graphitic orientations, there are a variety of morphological possibilities as well. Nanofibers grown through vapor deposition can be straight and smooth, but they are more often tortuous as the growth direction continuously varies during deposition. The fibers may be further characterized by twisted, helical, branched, or irregular geometries. The origin of these differences has been linked to the catalyst conformation [1,27] and is a function of temperature, gas chemistry, and catalyst composition.

The understanding and optimization of growth conditions have practical applications to reducing cost and complexity. Control over the atomic and morphological structure allows one to tailor the fiber properties, and by extension, the applications which incorporate them. For instance, fibers with a rough or irregular surface morphology may provide enhanced composite reinforcement [28], or collections of fibers which are either highly crystalline or highly amorphous may be used for electrical or thermal management such as appropriate for carbon foams [29,30]. This ability to achieve specific properties in the as-synthesized state is quite useful for any application, but when incorporating these carbon nanostructures into finished products during direct synthesis, it is especially important. By controlling fiber characteristics through catalyst choice and reaction conditions, specific application requirements can be achieved during material production.

An alloy comprised of nickel with 30 at. % copper was used as the catalyst. It was prepared by mechanically alloying elemental powders of nickel (99.8%, <1 μm) and copper (99%, <75 μm) using a Spex 8000 M high-energy ball mill with a ball-to-powder mass ratio of 5:1. Each alloy was milled with 1 wt.% stearic acid (reagent grade, 95%) to prevent cold welding during the 4 h milling duration (see Ref. [31], for more information on this preparation method). All starting materials were obtained from Sigma-Aldrich and used without modification. Each milling run consisted of 10 g of powder in a 7:3 atomic ratio of Ni:Cu. All powders were handled and stored in a glove box under constant purge with ultrahigh purity (UHP) nitrogen. Catalysts were weighed in the glove box and transferred to molds immediately before the reaction.

Reaction gases used in this work were ethylene (chemically pure) and forming gas (5% H₂ in Ar). Each reaction began with ethylene (C₂H₄) and H₂ flowing in a 4:1 ratio controlled using digitally programmed MKS G-series mass flow controllers. Constrained reactions were performed using a 15 cm diameter, three-zone Lindberg/Blue M tube furnace equipped with a 7.5 cm outside diameter (OD) quartz tube and matching ceramic vestibule adapters. Molds were plumbed separately within the quartz tube. Reactions were conducted at 550 °C until the inlet pressure limit of 2 psig was achieved, which indicated the mold was filled. This setup is shown schematically in Fig. 1. All reactions were allowed to run overnight until the maximum process pressure of 2 psig was reached (as controlled by a low-pressure regulator positioned ahead of the inlet). Nanofibers were analyzed by scanning electron microscopy (SEM) with a Zeiss Auriga 60 CrossBeam microscope operated at 7 kV, and X-ray diffraction (XRD) was performed with a PANanalytical X'Pert Pro MRD using Cu K_α radiation (λ = 0.15418 nm).

SAE/AISI type 303 and type 304 stainless steels were used to construct molds used as reaction vessels. A rectangular mold with dimensions of 0.9 × 3.3 × 5.4 cm (H × W × L) was machined from billet and sealed by polishing mating surfaces and using a distribution of stainless steel fasteners (1/4-20 UNC) along the perimeter of the mold. Catalyst positioning was studied in this mold by varying the distribution of powder along the bottom. The catalyst was evenly distributed and preferentially positioned along the sides, as depicted in Fig. 2.

Cylindrical molds were produced by welding billet end caps onto tubing of equivalent diameter. Round tube with outside diameters (ODs) of 2.5 cm and 5.1 cm were used. Inside diameters (IDs) of the tubing were 2.4 cm and 4.8 cm, respectively. Gas inlet and exhaust tubing was 0.6 cm OD (0.4 cm ID). All permanent joining operations were performed using gas tungsten arc welding (or GTAW), also referred to as tungsten inert gas (TIG) welding, with argon as the cover gas.

Semiconstrained growth was performed using a single-zone, 5 cm diameter, split-hinge tube furnace (Across International). Catalyst was placed at the center of a quartz tube (5 cm OD), and the reaction was run under the same conditions described above. The length of the tube (70 cm) was much greater than the heated zone (20 cm), and although the tube diameter provided constraint, no longitudinal restriction was imposed on the fiber growth (hence the “semiconstrained” terminology).

In a final configuration, carbon was grown using 150 mg of catalyst in the same furnace tube, while constraining the growth using stainless steel screens spaced 5 cm apart to allow gas to flow freely but restrict carbon growth. As-milled catalyst was used and compared to preform carbon created from the same catalyst loading in a mold 1.2 cm wide, 1.9 cm tall, and 10 cm long. This created a rectangular carbon monolith which was chopped into various sized blocks and also crushed into a powder to determine the benefit to filling the given volume.

There has been extensive work undertaken to understand, control, and optimize fiber growth (e.g., Refs. [27] and [32]). The Ni 30 at. % Cu catalyst was chosen for its ease of processing, low cost, and high selectivity toward nanofiber growth. In kinetic studies [31], it resulted in a deposition rate of ∼5000% mass gain per hour of reaction and yields a fiber mass dominated by nanofibers approximately 100 nm in diameter with a high degree of integration, even during unconstrained growth. The growth rate observed in this work was ∼ 2100% mass gain per hour of reaction (e.g., 20 mg of catalyst yields 2 g of carbon over about 5 h). Constrained fiber formation is lower than unconstrained growth, since the gas flow rate drops as the mold fills at the low pressure used here.

The carbon deposited after 1 h of growth was analyzed by XRD to assess the crystallinity of the carbon, as shown in Fig. 3. Peak positions for graphite and the face-centered cubic structure of the Ni–Cu catalyst overlap in places, but the distinct phases are identifiable as given in Fig. 3. The character of the carbon is turbostratic with no significant long-range order, as indicated by the considerable breadth of the carbon peaks their shift to lower angles than “ideal” for graphite. After 1 h of growth, the catalyst concentration was approximately 3% by volume (13% of the final mass), but after 12 h of growth the catalyst concentration is only 0.2% by volume (1% by mass). If the remaining catalyst is undesirable, the component may be demineralized by dissolving the metal in a suitable solution (e.g., HNO₃).

Scanning electron microscopy was used to examine the fibers produced without external constraint (see Fig. 4). Figure 4(a) shows the general appearance of the fiber growth before reaching a mold boundary. Figure 4(b) reveals the fibers are relatively uniform in diameter and possess a smooth morphology. These fibers are the sole structural constituents of bulk masses, so the characteristics of the individual fibers are also of interest to the macroscale characteristics of the fiber collection.

A rectangular mold was constructed to create a simple method for constraining growth and removing the resulting nonwoven carbon. Given the large volume relative to the initial catalyst loading, the placement of the catalyst is very important. Each catalyst particle serves a discrete site for fiber growth, and the overall distribution of carbon is linked to the initial distribution of the catalyst particles.

Through a series of experiments, it was determined that the best overall integration of carbon, resulting in a structurally sound nonwoven carbon material, was achieved by distributing the majority of catalyst (∼80%) evenly along the edges parallel to the gas flow with the remainder being placed along the center of the mold, also parallel to the gas flow. When catalyst is distributed evenly, the gas flow is blocked by the CNFs before significant ingrowth has occurred, and the resulting density and rigidity are lower. If the catalyst is distributed along the edges only, the two progressing fronts do not integrate well at the center before flow is blocked.

If the entirety of the catalyst is distributed along a single side, ingrowth outpaces expansion into the mold, and the flow may never be blocked, but the highest rigidity is produced. These results are summarized schematically in Fig. 2. It is important to note that the results for catalyst placement assume a fixed catalyst loading. Less catalyst results in a higher density/rigidity, and more catalyst yields lower density/rigidity. The general trend is that longer reaction durations before the mold is blocked will result in a material with higher density and rigidity. Also, lateral carbon growth is proportional to initial vertical space as illustrated by semiconstrained growth discussed below.

Once optimal catalyst placement is determined, carbon monoliths can be reliably generated. The carbon is readily removed from the mold as a single piece with good replication of geometry and can be employed in the as-grown condition or it can be subdivided before use (see Figs. 5(a) and 5(b)). The material is mechanically robust, and it can be compressed, bent, and cut without significant degradation. This allows postprocessing to be applied without loss of integrity. Figure 5 shows the rectangular mold and resulting carbon after growth. Also shown are cubes that were cut from a carbon monolith using a razor blade.

To explore larger volumes, cylindrical molds were used. This allowed for simple scale-up by fabricating cylindrical molds with varying lengths and diameters. Although catalyst loading per volume was constant, at the point where molds became blocked, the smaller cylindrical mold (2.36 cm ID) was evenly filled by fiber growth throughout the length (15.0 cm), while the larger mold (4.75 cm ID) was filled in the center (lengthwise), but the ends were not entirely integrated. The resulting growth is shown in Fig. 6. Although the carbon was well-integrated within the cylindrical molds, the inherently destructive nature of opening the molds made it untenable to remove the carbon as a single cylindrical component.

To assist in tracking growth behavior in large volumes, catalyst was placed at the center of a quartz tube housed in a split-hinge tube furnace. No constraint was present to stop longitudinal growth, so the fibers were only semiconstrained. This approximates the early stages of any constrained growth in a large mold, and in cylinders, this represents growth before the end caps are reached. It can be seen that fiber growth does not follow a uniform point source approximation of growth radiating outward from the catalyst, and horizontal expansion outpaces the vertical integration. These results are shown in Fig. 7.

To constrain growth in the quartz tube, stainless steel mesh screens were introduced at a distance of 50 cm. This restricted fiber expansion without blocking gas flow. Several modifications were tested with this setup. First, bare catalyst powder was introduced and reacted. While this mostly filled the volume, the carbon was not well-integrated. In an effort to traverse some of the volume initially, the same catalyst loading was used to pregrow carbon monoliths, as described in the Experimental section. This preform carbon was then divided up (similar to Fig. 5(b)) and introduced and processed in the quartz tube as the bare catalyst powder was before. It was found that when divided into larger pieces, the percent mass gain per hour (% mass/h) was less than the bare catalyst. The growth rate increased as the carbon preform size decreased, but the growth amount (mg/h) was relatively constant for all circumstances. All used the same initial catalyst loading, so it is expected that the percent growth rate would appear lower for preforms, since noncontributing carbon mass is considered in the initial mass. By also considering the mass gained, however, it is found that there is little dependence on initial mass or form (cube or crushed), and there is also little change in deposition rate after 1 h or after 12 h. This indicates the carbon preforms contained comparable catalytic activity to the bare catalyst powder, and that catalytic activity was sustained for the 12 h maximum reaction time used. That is, the amount of carbon added was independent of initial size, and the rate of deposition was independent of time (see Fig. 8).

Although deposition was fairly constant across all forms, the cubed preforms tended to poorly integrate and did not fill the volume as effectively as the bare catalyst or the crushed fibers, as shown in Fig. 9. The crushed fibers did provide a small benefit over bare catalyst powder, and it is expected that using pregrown fibers containing sufficient catalyst is a viable way to extend the growth volume, but they have similar issues in filling the volume initially.

The goal of this work was to build on previous findings related to the CoFFiN process [23], primarily with respect to determining the scalability of it. In that initial work, reactions were conducted in a rectangular mold 2.5 cm × 5.0 cm × 0.5 cm (6.25 cm³ volume). Here, mold dimensions were increased by 257% for the new rectangular mold (3.3 cm × 5.4 cm × 0.9 cm, or a 16.0 cm³ volume), more than 1000% for the smaller tubing (2.36 cm ID, 15 cm long, or a volume of 65.6 cm³) and more than 2000% for the larger tubing (4.75 cm ID, 7.5 cm long, or a volume of 133 cm³).

Despite the increase in volume, the catalyst loading (per volume) actually decreased by approximately 75% (mass per mold volume). A reduction in catalyst can be achieved by using a more effective catalyst or increasing the growth duration. In initial studies, submicron palladium powder was used as the catalyst. Here, a Ni–Cu catalyst was created by mechanical alloying [33] using ball milling as described by Guevara et al. [31] and presents a 97% reduction in catalyst cost (over Pd) while providing comparable deposition kinetics.

Even as a rudimentary, laboratory scale process, the Ni–Cu catalyst can be produced at a rate of 250 mg/min or more. The molds used here typically employed between 25 and 100 mg of catalyst, making the catalyst production rate suitable for running more and larger molds continuously. Considering that ball milling is currently employed at an industrial scale, catalyst preparation in this manner is both possible and may in fact be preferable for its increased throughput and unique capabilities when compared to more common coprecipitation methods commonly used to prepare catalysts [32,34,35].

Although fiber growth from a single catalyst particle may be approximated as a point source over small (micron) distances, the behavior of a collection of catalyst particles cannot be accurately predicted in that manner. The fiber growth will follow the path of least resistance, but the source of “resistance” will change over time. There is competition between intrinsic (self-generated) and extrinsic (externally imposed) factors. At first, fiber growth will occur in an unconstrained fashion, and if sufficient space is given, gravity will be the primary factor controlling bulk growth direction. Depending on the mold geometry, there may be very little effect of gravity.

Thin cross sections will be dominated by extrinsic constraint, but what is considered “thin” can be appreciable. For example, even though the rectangular mold has a significant height (0.9 cm) when compared to fiber size, the vertical growth is constrained by the mold lid and not gravity. In many respects, this type of growth is a two-dimensional problem. This is evidenced by the fact catalyst distribution along the bottom of the mold is sufficient to address quality issues in the three-dimensional material produced.

In very large volumes, where the catalyst is placed away from boundaries, the growth will respond in a three-dimensional manner. Vertical growth becomes unstable before the upper boundary of the mold is reached, and horizontal spreading begins to dominate. Controlling fiber growth in three dimensions is a challenge that must be addressed when creating materials with features in all three dimensions.

One method for achieving this is to control the initial distribution of catalyst in three dimensions. The catalyst may be suspended in a liquid, such as ethanol, and dispersed on mold surfaces to ensure the growth begins from multiple directions, or physical vapor deposition could be used to coat the mold interior with a catalyst. Although there are means for distributing the catalyst initially, the growth itself must be anticipated as well, because the catalyst is spread during growth. The original catalyst particle undergoes surface break-up [36,37] during initial fiber formation, and the catalyst begins dispersing away from the original source. Those smaller catalyst particles can continue to divide leading to further dispersion. Fiber growth is continually changing during processing, and predicting behavior is a multifaceted problem.

Larger volumes are important for creating bulk components for use in applications such as filtration, composites, or thermal management. In large volumes, such as the 5 cm quartz tube used in this work, it was observed that fiber growth is not of uniform density, even without reaching boundaries which might affect the growth characteristics. In long-range, unconstrained growth, some sections of carbon will achieve significantly higher stiffness and integration. These volumes appear to be the result of self-constraint, likely due to the ingrowth of fibers occurring more rapidly than the expansion of the fiber mass. That is, fiber growth was observed to result in increasing density without utilizing external constraint. This behavior seems to impose an upper limit on the free volume nanofiber growth can traverse while still having properties dictated by the mold boundaries.

Filling larger volumes quickly and completely can be achieved by simply using a larger catalyst loading. It is known, however, that this results in a lower density material [23]. Where a low density and lower stiffness are not desirable, the catalyst must be distributed within the open volume of the mold. Using a sacrificial template or adding more catalyst after growth has ceased expanding may be additional techniques for extending volume and placing catalyst where it is needed most.

Any method which requires adding or modifying the setup midprocess requires a mold design which is easy to access and which can be quickly disassembled and reassembled. This would likely require the mold to be cooled between steps, which reduces the efficiency of the process. Only the crushed, pregrown fibers can be done with immediate efficiency, since large quantities of catalyst can be reacted to create the catalyst-rich carbon, which can later be subdivided for use in the constrained growth processing. Just as with filling large volumes directly, preventing intrinsic constraint and directing uniform mold filling still requires careful forethought. It is quite possible, however, that the mold could be nearly filled with the pregrowth material, and the majority of processing time used to integrate the disparate fibers.

The focus of scaling up the production of nonwoven CNF structures is to make them easier to incorporate into products and processes. Targeted applications [38] include bulk composites, surface sensitive processes (e.g., battery and fuel cell electrodes and catalyst support), and thermal management, to name a few. These applications require a means to effectively incorporate nanomaterials into commercial production. The ideal case would include a cost-effective, repeatable process that produces a material which can be used without significant modification. This can provide a realizable impact on current technology with minimal complexity.

There are two main aspects demonstrated here which can be exploited to this end: the creation of freestanding structures, such as in Fig. 2(b), and the integration of fibers into other structures, as shown in Figs. 5(a) and 5(b). The first requires a process which results in a durable, well-integrated material that can be easily extracted from the mold it is grown in. The ability to tailor these structures in three dimensions is limited by common challenges in other molding processes. These include uneven section thicknesses, reproducing fine detail and creating internal detail, and/or placing inserts into the material.

While many of these considerations are beyond the scope of the present work, the foundational understanding necessary to advance those aspects has been illuminated. With respect to self-supporting structures, simple shapes can be created and reshaped after molding. As with any molding operation, the mold geometry can be designed to accommodate issues such as dimensional change, and the properties can be controlled by placing catalyst in a careful manner (e.g., Fig. 3).

A unique aspect of this nonwoven material is that fiber orientation is not constricted to two dimensions as in woven fibrous materials. Since growth direction is mostly regulated by catalyst irregularities and fiber–fiber interactions, the material is expected to be highly isotropic. Because the fibers can traverse considerable thickness (compared to woven fabrics), composite components may be generated with near-net shape and resin infiltration performed on a single, fully integrated “layer.” Alternatively, fiber growth may be completed in a mold which is later used for infiltration, such as by resin transfer molding, thereby reducing setup time.

In a similar manner, items may be incorporated into the fiber growth. Any material which can withstand the processing temperature, time, and process gases can be incorporated into a mold. For instance, tubing may be held in the mold, very much like a core would be used in casting for later removal. This feature could be removed to create an internal feature, or it may be left inserted in the fibrous material, such as a thermal management application which might utilize the fibers for heat retention or rejection, depending on the thermal properties of the fibers individually or as a collection. Instead of a sacrificial template for catalyst distribution, reinforcement or other structures may serve dual proposes to three-dimensionally distribute catalyst. Indeed, a great number of applications may be satisfied using various combinations of macroscale and nanoscale manufacturing controls.

It has been demonstrated that nonwoven components consisting entirely of carbon nanofibers can be created in bulk form (cm dimensions) at low cost using a variety of mold configurations. Stainless steel molds were constructed with rectangular and circular cross sections, and the volume of the molds varied from 16 cm³ to 133 cm³. Important factors were found to include the placement and amount of catalyst (Ni 30 at. % Cu) and the proximity of the mold boundaries to the catalyst. Fiber growth was fully constrained and well-integrated over the 0.9 cm height of the rectangular mold, but semiconstrained growth within the 5 cm quartz tube revealed that growth was vertically limited by gravity (spreading horizontally) and the fiber growth was dominated by self-constraint as the fibers entangled with one another and prevented expansion of the fiber mass before reaching the mold boundaries.

Self-supporting monoliths were created by removing the carbon from the mold (demonstrated in the rectangular mold), but fibers can also be synthesized within a mold that may later be an integral part of a larger system (e.g., tubing that is connected inline for filtration), as demonstrated with cylindrical molds.

The greatest challenge is controlling the growth over large volumes. Where the growth originates and how it progresses over large volumes may be addressed through judicious placement of catalyst in three dimensions or by using a multistep growth process. The use of pregrown fiber bundles which still contain viable catalyst may be a good choice for maintaining the efficiency of the process and enabling its use as a manufacturing technique to create nanostructured carbon materials cost-effectively.

This work was supported by the National Science Foundation, Award No. 1436444. X-ray diffraction performed by the Pennsylvania State University Materials Characterization Laboratory.