Experimental analysis on the exploding wire process for nanopowder production: Influence of initial energy and exploding atmosphere

Metallic nanopowder is employed in several applications for its flexibility of use, thus its production is growing over the years. Exploding wire technique can produce a wide range of nanoparticles from electrically conducting samples, and it has a potential for high efficiency and productivity, due to the strongly localized power dissipation. On the other hand, the phenomena involved during the explosion are complex, and phenomenological relationships are usually considered for controlling the output product characteristics. In this study, we investigated the effect of the input energy and the medium in which the explosion takes place on the final dimensional distribution of the nanopowder. Utilizing a proper power supply and an exploding system developed on purpose, we applied up to 10 GA/m 2 to copper wire samples immersed in air or water. We collected the products for morphological analysis, we carried out a shape and dimensional characterization based on a statistical analysis of the obtained particles, and we compared and discussed the results concerning the adopted initial conditions and the most recent literature data.


INTRODUCTION
Metallic-based nanoparticles are of great interest due to their versatility of use. Indeed, they can be deposited over or embedded into support materials to vary the overall physical and chemical properties. For instance, nanoparticles are used as surface coating agents to control the surface electrical and thermal conductivity, to improve its mechanical resistance, to increase its performance against chemical agents, and to control its optical characteristics [1]- [3]. Moreover, nanoparticles can be used in combination with different substrates, such as polymers [4], thus controlling the mean properties of the resulting medium, and allowing the development of novel materials [5]. Therefore, nanoparticles find a wide application in several sectors, from nano-biomedicine to sensors and transducer manufacturing. Indeed, nanoparticles can be used to boost the sensing performance of several devices, such as environmental sensors and bio-sensors [6]- [8], or the efficiency of batteries [9] among the others.
Nanoparticle morphology has a large impact on the resulting application effectiveness; indeed, equivalent diameter, shape, and chemical composition are of great relevance to this purpose. These values mainly vary as a function of the materials involved, and the production process itself. Literature describes several techniques for nanoparticle production. For instance, lasergrowth allows depositing particles with a diameter in the order of the laser wavelength but results in several thermal issues which can affect the particle dimension control [10]. Moreover, the operation of lasers, both for growing or ablating-based techniques, can generate unwanted and/or toxic secondary products [11]. Mechanical methods, such as grinding and milling, can result in time-consuming or energy-expensive processes [12], [13]. Chemical deposition and similar bottom-up approaches allow the growth of metal nanoparticles with satisfactory control of their morphology; however, the main disadvantage lies in the ABSTRACT Metallic nanopowder is employed in several applications for its flexibility of use, thus its production is growing over the years. Exploding wire technique can produce a wide range of nanoparticles from electrically conducting samples, and it has a potential for high efficiency and productivity, due to the strongly localized power dissipation. On the other hand, the phenomena involved during the explosion are complex, and phenomenological relationships are usually considered for controlling the output product characteristics. In this study, we investigated the effect of the input energy and the medium in which the explosion takes place on the final dimensional distribution of the nanopowder. Utilizing a proper power supply and an exploding system developed on purpose, we applied up to 10 GA/m 2 to copper wire samples immersed in air or water. We collected the products for morphological analysis, we carried out a shape and dimensional characterization based on a statistical analysis of the obtained particles, and we compared and discussed the results concerning the adopted initial conditions and the most recent literature data.
use of potentially hazardous and environmentally toxic chemicals, thus reducing the sustainability of this method [14].
Among the others, wire explosion is a promising technique [15] used to produce nanoparticles from wires based on conducting materials.
For this purpose, a current of magnitude 10 7 -10 9 A/m 2 [16] flows throughout the sample, and a quick joule adiabatic heating and a further explosion produce a mixture of liquid-gas metal clusters and ionized plasma [17]- [18]. Therefore, the vapours undergo rapid nucleation to form the desired nanoparticles [17]. Concerning other strategies, the use of harmful chemicals is avoided, and the very localized power dissipation limits the process duration, reduces the thermal losses, and possibly enhances both energy efficiency and productivity [19]. Moreover, this method is flexible and can produce multicomponent nanostructures [18], even if immiscible with standard techniques [20].
The process behaviour depends on complex interactions among thermal, mechanical, and electromagnetic phenomena into play, and a consistent model of this process is still lacking in literature [20]. Therefore, a phenomenological approach is often pursued to study the relationship between the input process parameters and the resulting nanoparticle characteristics.
For instance, the exploding atmosphere has a large influence in terms of product chemistry, as showed in [21]- [22], but also affects the product nucleation behaviour and the final morphology. Moreover, exploding energy and timing are crucial to control the final dimensions of the nanoparticles [23].
In this work, we used a wire explosion technique to produce copper-based nanoparticles from a copper wire sample. We provide experimental results utilizing different current densities and different mediums in which the explosion takes place. Therefore, we characterized the morphology of the products by measuring and describing the granular distribution and the particle shape, thus investigating the relationship between the product characteristics, and the operating conditions employed.

EXPERIMENTAL SETUP
As an exploding sample, we used specimens made of 4 twisted 0.1 mm diameter copper wires having a length of 40 mm circa. The wires come from commercial electrical wiring (i.e., standard 16 A cable with a 4 mm2 cross-section). We prepared a set of 30 samples, from which we randomly selected the ones for the tests (i.e., 2 samples per test case).
The outer diameter of the specimens resulted in about 0.4 mm, and their mass in about 0.14 g. We housed the wire between two aluminium electrodes with 3 mm thickness with dedicated screw connectors. The length of the electrodes, constant in the tests, determines the active length of the specimen (i.e., about 30 mm) which is the distance between the aluminium plates. Therefore, we measured the resistance between the electrodes for each tested sample as 10 ± 2 mΩ. We utilized the same initial configuration for all the test, thus limiting the variability on the results due to the setup. We placed the sample and the electrodes inside a cylindrical Nylon vessel with a 100 mm internal diameter and 100 mm height. We used a holed Nylon cap fixed with stainless-steel screws to enclose the environment and to retain most of the produced powder within the vessel. We connected the electrodes to a power coaxial cable via custom electrical connectors. A calibrated Rogowski coil [24] is wrapped around one electrode for current measurement purposes. Figure 1 shows the main components of the explosion setup.
The coaxial cable is connected to a power system made by a 765 µF capacitance, with a voltage limit of 10 kV. A high-voltage DC power supply allows to charge the capacitor to the desired voltage V, and an electronic control system allows a fast discharge thanks to an optical fiber trigger. We monitor the voltage using a 1:1000 dedicated probe. Depending on the system resulting impedance, we achieve a 10 10 A/m 2 current density. The main contributions to the system inductance are the connections between the electrodes and the sample, the electrode geometry, and the plasma formed during the explosion process. Figure 2 shows a picture and the sketch of the power circuit.
We acquired signals from the Rogowski coil and the voltage probe by employing a Lecroy W wave Pro 725Zi (equipped with four 8-bit boards and 2.5 GHz bandwidth) and elaborated through a Matlab post-processing routine. When provided, the expanded uncertainties are calculated according to the GUM and with a coverage factor k = 2.

Operating parameters
We provide a simplified scheme of the exploding wire process in the following paragraphs. First, a current flow passes through the sample, thus producing adiabatic heating due to the Joule effect. Indeed, the fast nature of the heating process (involving thousands of watts in tens of nanoseconds) does not allow a significant heat exchange with the surrounding medium. Therefore, the products (mainly vapours and ionized gas) rapidly expand while interacting with the explosion environment, thus nucleating and forming nanoparticles clusters. Generally, agglomeration occurs while the clusters are cooling down before their stabilization, thus producing macro clusters of nanoparticles. The initial energy, the flowing current, and the medium in which the explosion occurs have the main influence on the process behaviour. In particular, the initial energy E depends on the capacitor voltage V and capacitance C according to equation (1) We set the initial capacitor voltage at 800 V, 1 kV, and 1.5 kV, hence we use an initial energy of 244 J, 382 J, and 860 J, respectively. By considering the system resistance R concentrated on the wire sample resistance , the exploding power can be evaluated as a function of the circulating current according to equation (2) The time behaviour of depends on the inductance of the system, which is a function of the system geometry, of the involved materials, and of the plasma arc formed during the process. Apart from the stochastic nature of the plasma arc, the other impedance components are constant, and the current profile varies with the initial energy .
As stated in [25], during the process onset the expansion rate is the highest, and ions are dispersed with kinetic energy proportional to . Therefore, these ions are possibly responsible for nucleation centres. Moreover, the vapor supersaturation is high, hence condensation starts at the very beginning of the expansion and at the maximum rate. Therefore, a higher causes a larger ion dispersion, the nucleation occurs on a less dense areas, and smaller particle diameter can be obtained. The medium in which the explosion occurs has a key role in the cooling and nucleation process. In particular, we infer that a denser fluid with larger heat capacity and thermal conductivity should increase the cooling rate, promoting a faster stabilization of the clusters and mitigating the agglomeration phenomena. On the other hand, a denser fluid can increase the particle-to-fluid drag, leading to a comprehensive negative effect, due to reduced ion dispersion. For comparison, we produced nanoparticles with air and distilled water as mediums in the same initial energy conditions, and we studied the resulting dimensional characteristics of the produced nanoparticles. Both the air and the water are utilized with initial temperature of 25 °C (i.e., environmental temperature).

Collections of the samples
We collected samples generated in air and water to analyse the products with a Coxem EM-30 scanning electron microscope (SEM). The collected samples are deposited over a 10 mm x 10 mm ultra-smooth pure silicon sample. In the case of tests with air, the silicon sample is housed at the bottom of the nylon vessel and directly exposed to the exploding wire (Figure 3a). In the case of tests carried out in water, the copper wire is immersed in the medium. The explosion causes the products to disperse in water; therefore, we collect the liquid (Figure 3b), and further evaporate it while the silicon sample is immersed with a dedicated system consisting in a custom static oven. Its temperature control is operated by means of 3 Pt100 RTD sensors and a PID controller operated by a LabView software made on purpose. This system can keep the desired temperature setpoint within 2 °C.
During the evaporation of water, small particles can be carried out by the vapor thus altering the final measured dimensional distribution. On the other hand, the sedimentation allows the particles to stick together due to cohesion, thus forming larger clusters of nanoparticles and reducing their mobility. High evaporation rate and temperatures can promote internal recirculation of liquid and nanoparticles, limiting the sedimentation rate. However, a high evaporation rate can also reduce the drying time, increasing the productivity of the process. Therefore, we evaporated the water at 40 °C as a trade-off between the minimum temperature and the minimum time for drying completion. We remark that the optimization of this procedure is out of the purpose of the present study and that higher drying temperatures could be successfully employed but after a proper assessment of its impact on the final products. From the sample images, we expect to find particles ranging from tens to hundreds of nanometers. In order to carry out a statistical analysis on the product diameter, we require a significant number of samples per image. However, we restricted the measurement to the elements above 10 nm size, due to the reliability of both the measurement method and the SEM collected images. We found a satisfactory result using 5000× magnification during image acquisition.

Image analysis
We carried out image analysis utilizing a custom Matlab script. The SEM provide 8-bit grayscale images. The brightest pixel indicates the location of the nanoparticles, while the darkest pixel indicates the sample silicon background. First, we crop the image to the particles' area, and we filtered the pixel whose value is equal to or below the background one (i.e., background removal). Then, we increase the image contrast by exponentiating the pixel brightness values with a factor within 1.4 and 1.5 (i.e., contrast improvement), thus obtaining a steep gradient in the particle's proximity and located in a few pixels across the particles. Therefore, we binarize the data utilizing a threshold based on the pixel brightness fixed at 30% of the maximum value. By varying this value from 20% to 40% we obtain a negligible influence on the results. Finally, we removed small or poorly connected pixel clusters by using the "bwareaopen" function [26], and we use a routine based on the "regionprop" function [26] to get information from each isolated cluster. We considered each isolated cluster as a single particle, and the software provides the pixel area px and the perimeter p of each cluster. Therefore, we calculate the equivalent diameter eq and the circularity Γ as: It is worth noting that Γ = 1 indicates a circular shape, while Γ = 0 possibly indicates elongated shapes or elements with jagged perimeter. Figure 4 shows a scheme of the main image processing steps. Figure 5 shows a qualitative example of the algorithm operating steps.
The equivalent diameter measured by the aforementioned procedure is converted into S.I. units by means of the scale provided by SEM images. For each single batch with the same operating condition, several images are analysed by the SEM analysis over different positions of the silicon sample, to maximize the statistical population (i.e., more than 600 particles are detected for each test). In particular, we were able to count 30 to 70 particles per image at 5k magnification.
According to the current literature [23], [27], [28], and as explained by the model used in [29], the particles produced via exploding wire recall a lognormal dimensional probability distribution. In this study, we assume the lognormal model to statistically describe the results, as suggested in literature. In particular, the lognormal distribution of eq is modelled as The parameters and in (5) are calculated by fitting logn( eq ) and eq ; therefore, the mean of eq with the lognormal distribution hypothesis (i.e., eq * ) can be calculated as [30] eq * = + 2 2 .
Moreover, we can calculate the fraction of particles below a threshold diameter t , by using the cumulative distribution of logn( eq ) Usually, nanoparticles are in the dimensional range of 10 to 100 nm. On the other hand, in some applications the size of the nanoparticles can reach up to 250 nm, as reported in [31]. Therefore, we use D t = 250 nm to compute for comparison purpose among the different operating conditions. Finally, we compare eq * (obtained by considering a lognormal distribution) with the mean of the measured data eq . Figure 6 provides a comparison of the current profiles carried out from different test. Current data are calculated by integrating the voltage signal from the calibrated Rogowski coil. In particular, Figure 6a shows the comparison between air-based explosions at different initial voltage, and Figure 6b between air and water-based explosions at the same initial voltage.  According to Figure 6a, the profile structure results in two different peaks at approximately 50 µs and 70 µs. This structure is consistent with literature results. For example, [32] reports that the first current drop is mostly due to the vaporization of the wire, which increases the resulting resistance. In a second phase, a plasma arc is formed, and the current starts to increase again (i.e., the resistance decays) until the capacitor voltage supports the discharge process. The first peak time positioning delays when we adopt a lower initial voltage. Indeed, the lower injected energy leads to a slower current increase. Figure 6b shows good accordance between repetitions made in the same operating conditions, despite small differences in the transient current profiles that can occur due to possibly small variations of the samples and capacitor initial voltage. As expected, we obtained the same peak current of 23 kA by using the same input voltage. In the case of water, the first current peak anticipates, hence the wire explosion occurs at a lower current threshold. This fact depends on the complex exploding wire dynamic. We infer that the higher water density limits the initial expansion of the wire during the adiabatic heating process, thus increasing the sample's internal energy and allowing a quicker explosion behaviour. In the end, we calculate the maximum elapsed power at the first peak in the range between 2 to 3 kW. In Figure 7 we report the qualitative results from the exploding wire process, highlighting the different powder sample large-scale differences concerning the exploding atmosphere and input energy.

RESULTS AND DISCUSSION
From a qualitative analysis, the explosion atmosphere significantly affects the obtained result. Figure 7a and Figure 7b show particles with different shape; indeed, in the case of water the clusters borders are more jagged, and significantly more dispersed as well. This fact can be caused by the larger heat exchange possibly obtained in water, which induced cluster solidification before the surface tension allows the circular shape formation. Figure 7c and Figure 7d suggest that high input energy can cause liquid particles to impact the silicon sample surface, possibly introducing a bias on the nanopowder sampling process. As per the case of droplets formation due to shock waves [33], the larger the inertial forces concerning the surface tension, the more cluster fragmentation is achieved, and we assume that the higher the energy is released through the wire, the higher is the initial speed of the particles. Therefore, the high particle speed led to the impact of melted clusters on the silicon surface.
In Figure 8 we report the result of the statistical analysis on the particles in terms of equivalent diameter and circularity distribution concerning equations (3) and (4). Figure 8a and Figure 8b suggest the particle distribution shift towards smaller when we utilize a higher initial voltage, independently by the exploding medium.
On the other hand, Figure 8c and Figure 8d show that the exploding medium has the larger influence on the particles shape, thus influencing the circularity of the particles. Indeed, the circularity for water-based explosion peaks at around 0.4, and the one for air-based explosion at around 0.7. On the other hand, has a minor influence on Γ, affecting the circularity to a minor extent concerning the exploding medium. We use the data presented in Figure 8 to calculate eq * and as shown in Eqs. 6 and 7, thus obtaining a quantitative estimation of the dimensional characteristics of the products. The parameters associated with the obtained data are resumed and reported in Table 1. Table 1 shows a satisfactory accordance between eq * , calculated by considering a lognormal distribution of the equivalent diameter, and the mean of the data regarding the equivalent diameter eq . In Figure 9 we report the behaviour of the lognormal distributions characteristics as a function of .  highlights that by utilizing a higher input energy, we eventually obtain a larger fraction of smaller particles in the case of air as medium in which the explosion takes place. We notice that the distribution obtained in the case of air and 800 V has a double-peak behavior, hence the use of lognormal model has some major reliability concerns, and Figure 9 shows a not monotone behavior. In the case of water as medium, an increasing E causes the average equivalent diameter to grow, and this is the opposite of the attended result.
It is worth noting that the in-water explosions create clusters and nanoparticles with lower circularity (Figure 8). On one hand, this can trivially indicate that the nanoparticles have a larger and their shape is less regular due to the interaction with the surrounding media; on the other hand, the same result could be obtained for a larger aggregation of smaller particles, thus resulting in larger and less regular clusters. Indeed, with the used magnification and SEM detector, we cannot distinguish the aforementioned occurrences. A preliminary analysis on the same samples at larger magnification (i.e., 100k) showed that the bigger particles detected with the method described in this study are indeed larger clusters of smaller particles, as showed in Figure 10.
It appears that the impurities contained in the used water could have promoted the aggregation of nanoparticles, thus introducing a bias in the estimation of the equivalent diameter distribution.   Therefore, a deeper study based on back-scattered electrons and Energy Dispersive X-ray Spectroscopy is mandatory to shed light on the product morphology and chemistry as a function of the input energy and the medium in which the explosion takes place.
According to the present finding, further investigation is required on a wider range of operating conditions with the purpose of developing a phenomenological model of the process. Moreover, the development of a novel and comprehensive method for the analysis of the nanoparticles is crucial to achieve satisfactory reliability of the results.

CONCLUSIONS
In the present work, we developed a setup to produce nanopowder from electrically conductive wires via wire explosion technique. In particular, we produced powder from copper wires with nanometer scale diameter by using current densities in the order of 10 9 -10 10 A/m 2 , and different explosion atmospheres (i.e., water and air). We collected powder's samples on silicon surfaces, and we further analyze the products morphology by means of a SEM. We used an image analysis tool based on Matlab software to analyze the samples, by measuring the equivalent diameter and circularity distribution of the powder. By using air as medium in which the explosion takes place, it turns out that the dimensional distribution of the nanoparticles is shifted towards smaller equivalent diameter when higher input energy is used. However, the data obtained from the in-water experiments show an inverted trend. A deeper analysis showed that the larger detected clusters are indeed large aggregation of small particles. Therefore, we infer that the higher input energy contributes to significantly decrease the equivalent diameter of the products, but significant efforts are still required to shed light to this occurrence.