**Study on ZrB2-Based Ceramics Reinforced with SiC Fibers or Whiskers Machined by Micro-Electrical Discharge Machining**

#### **Mariangela Quarto 1,\* , Giuliano Bissacco <sup>2</sup> and Gianluca D'Urso <sup>1</sup>**


Received: 1 October 2020; Accepted: 26 October 2020; Published: 26 October 2020

**Abstract:** The effects of different reinforcement shapes on stability and repeatability of micro electrical discharge machining were experimentally investigated for ultra-high-temperature ceramics based on zirconium diboride (ZrB2) doped by SiC. Two reinforcement shapes, namely SiC short fibers and SiC whiskers were selected in accordance with their potential effects on mechanical properties and oxidation performance. Specific sets of process parameters were defined minimizing the short circuits in order to identify the best combination for different pulse types. The obtained results were then correlated with the energy per single discharge and the discharges occurred for all the combinations of material and pulse type. The pulse characterization was performed by recording pulses data by means of an oscilloscope, while the surface characteristics were defined by a 3D reconstruction. The results indicated how reinforcement shapes affect the energy efficiency of the process and change the surface aspect.

**Keywords:** micro-EDM; Zirconium Boride; silicon carbide fibers; silicon carbide whiskers; advanced material

#### **1. Introduction**

Among the advanced ceramic materials, ultra-high-temperature ceramics (UHTCs) are characterized by excellent performances in extreme environments. This family of materials is based on borides (ZrB2, HfB2), carbides (ZrC, HfC, TaC), and nitrides (HfN), which are marked by high melting point, high hardness and good resistance to oxidation. In particular, ZrB2-based materials are of particular interest because of their suitable properties combination and are considered attractive in applications, such as a component for the re-entry vehicles and devices [1–4].

The relative density of the base material ZrB<sup>2</sup> is usually about 85% because of the high level of porosity of the structure, furthermore, in the last years, researchers are focused on fabricating high-density composites characterized by good strength (500–1000 MPa). For these reasons, the use of single-phase materials is not sufficient for high-temperature structural applications. Many efforts have been done on ZrB2-based composites in order to improve the mechanical properties, oxidation performances, and fracture toughness; however, the low fracture toughness remains one of the greatest limitations for the application of these materials under severe conditions [5–10]. Usually, the fracture toughness of ceramic materials can be improved by incorporating appropriate additives that activated toughening mechanisms such as phase transformation, crack pinning, and deflection. An example is the addition of SiC, in fact, it has been widely proved that its addition improves the fracture strength and the oxidation resistance of ZrB2-based materials due to the grain

refinement and the formation of a protective silica-based layer. Based on these aspects, recent works have focused on ZrB2-based composites behavior generated by the addition of SiC with different shapes (e.g., whiskers or fibers). Specifically, it has been reported that the addition of whiskers or fibers gives promising results, improving the fracture toughness and this improvement could be justified by crack deflection [5,6,9,11,12]. The critical aspect of the reinforced process was the reaction or the degeneration of the reinforcement during the sintering process [13,14].

Despite all the studies that aim to improve mechanical properties and resistance, this group of materials is very difficult to machine by the traditional technologies, because of their high hardness and fragility. Only two groups of processes are effective in processing them: On one side the abrasive processes as grinding, ultrasonic machining, and waterjet, on the other side the thermal processes as laser and electrical discharges machining (EDM) [15–18].

In this work, ZrB<sup>2</sup> materials containing 20% vol. SiC whiskers or fibers produced by hot pressing were machined by the micro-EDM process; in particular, it would investigate the effect of the non-reactive additive shapes on the process performances, verifying if the process is stable and repeatable for advanced ceramics, and in particular for materials characterized by different geometry of the additive. The choice of 20% vol. is related to the evidence reported in some previous works [14,18,19], in which it has been shown that this fraction of additive has allowed generating the best combination of oxidation resistance and mechanical characteristics useful for obtaining better results in terms of process performances and dimensional accuracy for features machined by micro-EDM.

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

The following ZrB2-based composites, provided by ISTEC-CNR of Faenza (Consiglio Nazionale delle Ricerche—Istituto di Scienza e Tecnologia dei Materiali Ceramici, Faenza (RA)—Italy), have been selected for evaluating the influence of additive shape on the process performances and geometrical of micro-slots machined by micro-EDM technology:


Such as reported in [14], commercial powders were used to prepare the ceramic composites: ZrB<sup>2</sup> Grade B (H.C. Starck, Goslar, Germany), SiC HI Nicalon-chopped short fibers, Si:C:O = 62:37:0.5, characterized by 15 µm diameter and 300 µm length or SiC whiskers characterized by average diameter 1 µm and average length 30 µm.

The powder mixtures were ball milled for 24 h in pure ethanol using silicon carbide media. Subsequently, the slurries have been dried in a rotary evaporator. Hot-pressing cycles were conducted in low vacuum (100 Pa) using an induction-heated graphite die with a uniaxial pressure of 30 MPa during the heating and were increased up to 50 MPa at 1700 ◦C (TMAX), for the material containing fibers, and at 1650 ◦C (TMAX) for the composites with whiskers. The maximum sintering temperature was set based on the shrinkage curve. Free cooling followed. Details about the sintering runs are reported in Table 1, where TON identify the temperature at which the shrinkage started. Density was estimated by the Archimedes method.



The raw materials were analyzed by Scanning Electron Microscopy (SEM) (Figure 1). The samples reinforced by the fibers show a very clear separation between the base matrix and the non-reactive additive. Fibers dispersion into the matrix was homogenous, as no agglomeration was observed in the sintered body. However, some porosity was retained in the microstructure. The fibers showed a tendency to align their long-axis perpendicular to the direction of applied pressure. It is apparent that their length was significantly reduced compared with the starting dimensions, as the maximum observed length was about 300 µm. For the sample containing whiskers reinforcement, a dense microstructure was observed and the whiskers are generally well dispersed into the matrix. The mean grain size of ZrB<sup>2</sup> grains was slightly lower (2.1 ± 0.2 µm) than the base material (3.0 ± 0.5 µm); furthermore, the whiskers showed a tendency to form large bundles. The addition of SiC whiskers promoted both strengthening and toughening compared with the base material ZrB2. The maximum toughness increase (5.7 MPa·m1/<sup>2</sup> ) was of the order of 50% when whiskers were added. In the sample containing fibers, the increase in fracture toughness (5.5 MPa·m1/<sup>2</sup> ) corresponds to a strength reduction. A direct observation of the crack morphology and the comparison with theoretical models demonstrates that, besides a residual stress contribution, the toughness increase was almost entirely explained in terms of crack deflection in the whiskers and in terms of crack bowing in the fibers. No crack bridging was obtained as the reinforcement pullout was hindered by the formation of interphases or intergranular wetting phases, which promoted a strong bonding between matrix and reinforcement. The values of fracture toughness of the ZrB20f and ZrB20w were close to those of hot-pressed composites previously. This indicates that, even if a more efficient thermal treatment is used, the nature of the matrix/reinforcement interface did not change notably [2].

**Figure 1.** SEM backscattered images of typical appearance of ZrB20f (**a**) and ZrB20w (**b**).

#### **3. Experimental Section**

#### *3.1. Experimental Set-Up*

A simple circular pocket having a diameter equal to 1 mm and a depth of about 200 µm was selected as the test feature for machining experiments. These micro-features had been processed on the SARIX® SX-200 machine (Sarix, Sant'Antonino, Switzerland) by micro-EDM milling. Solid tungsten carbide electrode with a diameter equal to 300 µm was used as a tool; while the dielectric fluid was formed by hydrocarbon oil. The experiments were performed for three different process parameters settings, corresponding to different pulse shapes. It is essential to remark that the machine used for the experimental tests expresses some process parameters as indexes (e.g., peak current, width). The instantaneous values cannot be set, because the machine presents an autoregulating system. Thus, the characterization of electrical discharges population is very important not only to assess the real value of process parameters but, most of all, to evaluate the stability and repeatability of the process. Due to this characteristic of the machine, an acquisition system was developed to cover the gap. In particular, a current monitor and a voltage probe were connected to the EDM machine and

to a programmable counter and a digital oscilloscope. These connections allow to acquire the current waveforms and count the discharges occurred during the process. Specifically, a current monitor with a bandwidth of 200 MHz and a Rohde & Schwarz RTO1014 oscilloscope were used. The counter has been set once the trigger value was established to avoid recording and counting the background noise.

Preliminary tests were performed to define the optimal process parameters for each combination of material and pulse type. The experimental campaign was based on a general full factorial design, featured by two factors: The additive shape, defined by two levels, and the pulse type, defined by three levels. Different levels of pulse types identify the different duration of the discharges, in particular, level A is referred to long pulses while level C identify the short pulses. Three repetitions were performed for each run.

#### *3.2. Discharges Population Characterization*

For each pulse types, combined with both materials, discharge populations have been characterized by repeated waveform samples of current and voltage signals. The current and voltage probes were connected to the digital oscilloscope having a real-time sample rate of 40 MSa/s. The trigger level of the current signal was set to 0.5 A in order to acquire all the effective discharges. The acquired waveform samples were stored in the oscilloscope buffer and then transferred to a computer to be processed by a Matlab code, written by the authors. The Matlab code was used to evaluate the numbers of electrical discharge, the current peak values, the duration (width), the voltage and to estimate the energy content in each discharge. Finally, the average value of energy per discharge (*E*) was estimated by integrating the instantaneous value of the power, calculated as the product of the instantaneous values of current (*i*(*t*)) and voltage (*v*(*t*)), with respect to the time (Equation (1)).

$$E = \int\_0^T v(t) \cdot i(t)dt\tag{1}$$

To show the discharge population distribution for all pulse types and for both additive shapes, the peak current distribution histograms are plotted, showing the number of observed discharges, with peak current within discrete intervals, for all the (Figure 2). Histograms show good reproducibility and stability of the process, providing information regarding the frequency waveforms with different peaks of current. The discharge samples are well described by a normal distribution, which is characterized by a good reproducibility suggesting a stable process. Considering both additive shapes, the values of peak current are included in a similar range for pulse type A and B, while, for pulse type C can be identified a great difference; in fact, for whiskers, the range of variation of current peaks is smaller than for short fibers.

Thanks to the discharge characterization it was possible to define the real value of the process parameters. The pulse characteristics for the parameters setting estimated by the data elaboration with Matlab are reported in Table 2.

In terms of peak current and voltage, the differences between the two materials, machined by the same pulse type, were really tiny. This suggested that EDM machining on ZrB2-based composites reinforced by whiskers would have been characterized by higher machining speed.

#### *3.3. Characterization Procedure*

A 3D reconstruction of micro-slots was performed by means of a confocal laser scanning microscope (Olympus LEXT, Southend, Essex, UK) with a magnification of 20×. This microscope recognizes the peaks of the reflected light intensities of multiple layers and, setting each layer as the focal point, makes it possible to analyze and measure each layer. Then, the images were analyzed with an image processor software (SPIP 6.7.3, Image Metrology, Lyngby, Denmark), firstly performing a plane correction on all the images to level the surfaces and to remove primary profiles, then the surface roughness (Sa) was assessed, on the base of the international standard UNI EN ISO 25178:2017 by the

real-topography method. The process performances were evaluated through the estimation of three indicators: The material removal per discharge (MRD), the tool wear per discharge (TWD), and the tool wear ratio (TWR).

**Figure 2.** Examples of frequency distribution histograms for pulses occurred during ZrB20 machining.


**Table 2.** Mean values of pulse type characteristics.

MRD (Equation (2)) was calculated as the ratio of material removed from the workpiece (MRW (mm<sup>3</sup> )), estimated through SPIP, and the number of discharges (N) recorded by the programmable counter.

$$\text{MRD} = \frac{\text{MRW}}{\text{N}} \tag{2}$$

Since this kind of materials is characterized by a high level of porosity, to get the actual values of MRD, the volume of the micro-slots was adjusted considering the relative density (δ—Table 1). for compensating the presence of porosity in the sample structure. The MRD is calculated as reported in Equation (3).

$$\text{MRD}\_{\delta} = \frac{\text{MRW} \cdot \delta}{\text{N}} = \text{MRD} \cdot \delta \tag{3}$$

TWD (Equation (4)) was estimated as the ratio between the material removed from the tool electrode (MRT (mm<sup>3</sup> )) and the number of discharges [N] recorded by the programmable counter. Tool wear was measured as the difference between the length of the electrode before and after the single milling machining. The length was measured through a touching procedure executed in a reference position. The electrode wear volume was estimated starting from the length of the tool wear and considering the tool such as a cylindrical part.

$$\text{TWD} = \frac{\text{MRT}}{\text{N}} \tag{4}$$

Tool Wear Ratio (Equation (5)) was calculated as the ratio between the previous performances' indicators considering the relative density of the workpiece material (TWRδ).

$$\text{TWR}\_{\delta} = \frac{\text{TWD}}{\text{MRD}} = \frac{\text{MRT}}{\text{MRW} \cdot \delta} \tag{5}$$

#### **4. Results and Discussion**

During the analysis, the energy efficiency of a single discharge was evaluated. Figure 3 shows the tool wear per discharge divided by the energy of single discharge (TWDE), as a function of the additive fraction and the pulse type. The plot shows a lower energy efficiency for pulse type A and this is a positive aspect because it indicates less impact on the electrode wear. For both additive shapes, the medium pulses are characterized by a higher impact on the tool wear.

**Figure 3.** Average ratio between tool wear per discharge (TWD) and energy per discharge as a function of the additive shape and pulse type.

To evaluate the energy efficiency from the material removed point of view, as has already been done in the previous plot, also the material removal per single discharge was evaluate as the ratio with the energy for single discharge. In this case, the best results were obtained for pulse type C where a single discharge is characterized by lower duration and energy but and higher efficiency, such as illustrated in Figure 4.

Figure 5 shows that the TWR for all pulse type is lower for specimens containing whiskers reinforcement. In particular, the whiskers allow to reduce the tool wear and increasing the material removal rate efficiency probably thanks to the dense microstructure and the reduction of the grains size.

**Figure 4.** Average ratio between material removal per discharge (MRD) and energy per discharge estimated considering the relative density of the samples as a function of the additive shape and pulse type.

**Figure 5.** Average tool wear ratio (TWR) as a function of the additive shape and pulse type.

In general, for pulse type A the energy per discharge is higher in comparison to others pulse types but for both Energy efficiency of TWD (TWDE) and Energy efficiency of MRD<sup>δ</sup> (MRDδ/E), the energy efficiency is lower (positive aspect from TWD<sup>E</sup> point of view). This can suggest a greater energy dispersion in machining performed with pulse type A than for others; in fact, energy per discharge of pulse type C is 40–50 times smaller than pulse type A but the energy efficiency is higher.

The factorial design was analyzed in order to comprehend which factors and interactions are statistically significant for the performance indicators and surface roughness. Table 3 shows the average results obtained from the experiments reporting the average and the standard deviation of MRDδ, TWD, TWRδ, and Sa for each level of the experimental design. A general linear model was used to perform a univariate analysis of variance, including all the main factors and their interactions.


**Table 3.** Results of the experimental campaign for process performances, where µ and σ identifies the average value and the standard deviation respectively.

The Analysis of Variance (ANOVA) results of the experimental plan are reported in Table 4 omitting the values related to the pulse type, since all the *p*-values results to be very close to 0.00. The parameters are statistically significant for the process when the p-value is less than 0.05 since a confidence interval of 95% is applied. As a general remark, all the indicators resulted to be influenced by the additive shape and the pulse type. In some cases (for MRDδ and TWD), also the interaction showed an effect in terms of ANOVA. This aspect suggested that the interaction of factors is relevant for indicators that, in some way, can be correlated to the machining duration.



\* indicate the interaction between Pulse Type and Additive Shape

Main effects plots (Figure 6) show that indicators are mainly influenced by the pulse type that establishes the range in which process parameters can vary, and in particular, the characteristics of the pulses. For all indicators, reduction in pulse duration and in peak current intensity generate the lower value of MRDδ, TWD, and TWRδ. At the same time, tests with whiskers additive generate increment in MRDδ and in surface roughness, and a reduction in TWD and TWRδ. By increasing the pulse duration and peak current intensity from type C to type A, the machining speed (MRDδ) is 10 times greater, but the surface quality decreases by −60%. For MRD<sup>δ</sup> and TWD, also the interaction between pulse type and additive shape influence the results. The interaction plots (Figure 7) give more information than the *p*-value showing that the weight of the interaction is slight. What it is possible to notice is that for the samples that are machined by pulse type C, the effect on MRDδ is more evident, while from the TWD point of view it is more evident for pulse type A). In general, tests performed on materials with whiskers additive are characterized by better results, both in terms of process performances and surface finishing. In fact, optimal performance for ED-machining are characterized by a higher level of MRDδ, to perform a fast machining, and lower TWD, to reduce the waste of material related to the tool wear as a function of material removed from the workpiece.

**Figure 6.** Main effects plot for indicators affected by pulse type and additive shape.

**Figure 7.** Interaction plot for MRDδ and TWD.

Figures 8–10 represent each indicator as a function of the energy per discharge and, regardless of the additive shape, it is possible to observe that each indicator is well represented (R<sup>2</sup> <sup>≈</sup> 1) by the same type of regression equation for both additive shapes. TWD shows lower values of R<sup>2</sup> . Specifically, the MRDδ is well described by a logarithmic regression equation (Equation (6) and Equation (7)), while the TWD and the Sa are well described by a power regression equation (Equations (8)–(11)). In all regression equation reported in the plots, y-axis is referred to the indicator, while x-axis is referred to the energy generated by a single discharge (E). For MRDδ and TWD the differences as a function of additive shape are very small, in fact, despite they cover different ranges, the values are very close to each other. The situation is different for Sa, which is characterized by completely different ranges without overlap.

= 0.3322

0.3614

= 0.1791

= 0.1359

$$\text{MRD}\_{\delta\_{-}f} = 13.934 \ln(\text{E}) - 32.056 \qquad \text{R}^2 = 0.9965 \tag{6}$$

$$\text{MRD}\_{\text{\ $}\_{\text{\$ }}\text{W}} = 14.28 \ln(\text{E}) - 32.328 \qquad \text{R}^2 = 0.9939 \tag{7}$$

$$\text{TWD}\_{\text{f}} = 0.2569 \text{E}^{0.7868} \qquad \qquad \text{R}^2 = 0.9727 \tag{8}$$

$$\text{TWD}\_{\text{W}} = 0.3322 \text{E}^{0.7243} \qquad \text{ } \quad \text{R}^2 = 0.9892 \tag{9}$$

$$\text{Sa}\_{\text{f}} = 0.1791 \text{E}^{0.3614} \qquad \text{R}^2 = 0.9999 \tag{10}$$

$$\text{Sa}\_{\text{W}} = 0.1359 \text{E}^{0.3784} \qquad \text{R}^2 = 0.9987 \tag{11}$$

**Figure 8.** Material removal per discharge as a function of energy per discharge for both additive shapes and their regression equation.

**Figure 9.** Tool wear per discharge as a function of energy per discharge for both additive shapes and their regression equation.

**Figure 10.** Surface roughness as a function of energy per discharge for both additive shapes and their regression equation.

A 3D reconstruction of an ED-machined surfaces detail scanned by the confocal laser scanning microscope with a magnification of about 100× is reported as an example in Figure 11. In particular, Figure 11a represents a portion of the machined area on ZrB20f. Here it is possible to identify very clearly, a sort of "protrusion" in correspondence of the fibers. Such as reported in a previous work [20], this aspect is probably related to incomplete machining because of SiC low electrical conductivity characteristic and the great extension of the area of the fibers. Figure 11b represents the ED-machined surface for the sample containing SiC whiskers. In this case, the surface appears uniform and homogeneous because the SiC particles, such as reported in the materials section, are better dispersed in the base matrix. These aspects justified the different results obtained in terms of surface quality and, in general, these different textures can be considered a starting point for further studies about the material removal mechanism occurred on UHTCs, in particular when there is a low-electrically conductive parts in the structure. From the 3D reconstruction it is possible to observe that the surfaces are not characterized by the typical aspect of the ED-machined surfaces which present a texture well-described by the presence of craters. The different aspect has already been presented in a previous study [20]. In his specific case, the UHTC is different, but the same considerations can be done. A sort of craters can be observed by means of a SEM (Figure 12).

**Figure 11.** Details of machined surfaces for ZrB20f (**a**) and ZrB20w (**b**) machined by pulse type A.

**Figure 12.** SEM backscatter images of the machined surface.

Machined surfaces on specimens doped with SiC whiskers are characterized by higher fragmentation of the recast layer. This aspect is particularly evident on surfaces machined by long pulses; in fact, for pulse type A and B, the surface is for the major part covered by recast materials for both additive shapes, but the specimens doped with whiskers presented smaller extensions of the single "crater" of recast materials.

Different behavior can be observed for surfaces machined by the short pulses. In this case, for specimens contained the SiC fibers can be observed a smaller area of recast material, probably because of the union of the greater dimensions of the fibers in comparison to the whiskers, the lower energy per discharge for short pulses and the low electrical conductibility of the SiC. In particular, the fibers have a bigger surface and it needs to work hard to remove the entire parts. For this reason, the surfaces contained fibers presented in their correspondence a sort of protrusion.

It is very clear that the whiskers affect in a significant way the micro-EDM process improving the energy efficiency and the machining performances increasing the machining speed and reducing the tool wear. Furthermore, the geometry and the behavior of whiskers during the samples preparation generate a homogenous surface improving the removal rate and reducing the risk of leave machining witness (as the fibers) on the surfaces which would generate better surface finishing. These considerations allow to suppose a different behavior of the discharges path as a function of the kind of material met and the distribution of the reinforcement.

#### **5. Conclusions**

An evaluation of the machinability of ZrB2-based composites hot-pressed with different shapes non-reactive additive (SiC) was performed in this work. Stability and repeatability of the micro-EDM were evaluated to identify the effects of the additive shapes. The analysis taking into account the process performances and surface finishing.

First of all, a discharge characterization was performed to feature the different pulse type used during the machining. In general, the discharge characterization and the performances indicators allowed to identify a stable and repeatable process with a faster material removal for samples doped with whiskers.

The analysis of variance showed that both factors, pulse type, and additive shape, are statistically significant for the indicators selected in the process evaluation and in general, the use of whiskers improve the material removal rate generating lower tool wear. Furthermore, the interaction between

the two factors turns out to be influential only for MRD and TWD, which are indirectly related to the machining duration, since the number of discharges occurred during the machining was considered in their estimation.

This investigation shows that the specimens having a 20 vol.% of additive in form of whiskers results to be the best solution in terms of machinability by EDM process not only for the better process performances, but also for the higher level of surface quality which is one of the essential criteria for making a proper decision for industrial application.

**Author Contributions:** Conceptualization, M.Q., G.B., and G.D.; methodology, M.Q., G.B. and G.D.; software, M.Q., G.B.; formal analysis, M.Q.; data curation, M.Q., G.B.; writing—original draft preparation, M.Q., writing—review and editing, G.B., G.D.; supervision, G.B., G.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors would like to thank Diletta Sciti and Laura Silvestroni from ISTEC—CNR of Faenza (Italy) for the production and supply of the materials used in this study.

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

#### **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* **Investigation of a Liquid-Phase Electrode for Micro-Electro-Discharge Machining**

#### **Ruining Huang 1,\* , Ying Yi <sup>2</sup> , Erlei Zhu <sup>1</sup> and Xiaogang Xiong <sup>1</sup>**


Received: 18 September 2020; Accepted: 12 October 2020; Published: 14 October 2020

**Abstract:** Micro-electro-discharge machining (µEDM) plays a significant role in miniaturization. Complex electrode manufacturing and a high wear ratio are bottlenecks for µEDM and seriously restrict the manufacturing of microcomponents. To solve the electrode problems in traditional EDM, a µEDM method using liquid metal as the machining electrode was developed. Briefly, a liquid-metal tip was suspended at the end of a capillary nozzle and used as the discharge electrode for sparking the workpiece and removing workpiece material. During discharge, the liquid electrode was continuously supplied to the nozzle to eliminate the effects of liquid consumption on the erosion process. The forming process of a liquid-metal electrode tip and the influence of an applied external pressure and electric field on the electrode shape were theoretically analyzed. The effects of external pressure and electric field on the material removal rate (MRR), liquid-metal consumption rate (LMCR), and groove width were experimentally analyzed. Simulation results showed that the external pressure and electric field had a large influence on the electrode shape. Experimental results showed that the geometry and shape of the liquid-metal electrode could be controlled and constrained; furthermore, liquid consumption could be well compensated, which was very suitable for µEDM.

**Keywords:** Micro-electro-discharge machining (µEDM); liquid-metal electrode; Galinstan

#### **1. Introduction**

Micro-electro-discharge machining (µEDM) is a powerful micromachining technique with various advantages resulting from it being a noncontact and thermal process; thus, µEDM is applicable to any electrically conductive material regardless of the mechanical properties of the material. The machining process utilizes the thermal erosion of the material caused by pulses of electrical discharge generated between a microscopic electrode and the workpiece in the presence of a dielectric fluid for the removal of the workpiece material. The µEDM technique is capable of producing microholes, microchannels, and real three-dimensional (3D) microstructures. These attractive features have been leveraged for producing micromechanical components, as well as for prototyping various micro-electro-mechanical systems (MEMS) and devices. Furthermore, µEDM has been widely used in the aerospace, die, mold, and biomedical industries for machining small cavities [1–3].

It is well known that electrode miniaturization is critical for µEDM because it determines the precision of the process to manufacture smaller and more precise parts [4]. Many methods have been proposed for fabricating microelectrodes. For example, Masuzawa et al. were the first to develop wire electrode discharge grinding (WEDG) technology to make a microelectrode with a diameter of Φ2.5 µm [5]. WEDG is one of the most widely used methods to fabricate microelectrodes, and it offers the benefit of sustaining good grinding accuracy by utilizing fresh wire during the whole fabrication

process. According to this method, some evolutions occurred in different ways. Egashira et al. made use of WEDG and electrochemical machining to produce a microelectrode with a diameter of Φ0.3 µm [6]. Zhang et al. proposed a tangential feed WEDG (TF-WEDG) method to improve the microelectrode accuracy [7]. Lim et al. utilized a rotating disc instead of a moving wire to grind the microelectrode [8]. Other methods reported for µEDM electrode fabrication include the single-side block electrode discharge grinding method (BEDG) [9], bilateral BEDG method [10], scanning discharge method [11], single-pulse discharge method [12], self-drilled hole reverse-EDM method [13], electroforming method [14], Lithographie, Galvanoformung and Abformung (LIGA) process method [15], and electrostatic ejection method [16]. The goal of all these methods is the same: to improve the processing efficiency and quality, reduce the processing difficulty, and make smaller and better microelectrodes. However, these methods either are complex processes or require additional devices, leading to the process being cumbersome and inefficient. It is difficult to control the size and precision of the microelectrode. Although high-quality electrodes can sometimes be obtained, the consistency is poor, and it is difficult to obtain the same size electrode again.

In addition, the mechanism of µEDM is an electrothermal physical process that removes material by repeated spark discharges, whereby the workpiece is eroded at high temperature, and the electrode itself will also wear down. Therefore, the fundamental theory determines the unavoidability of wear with the electrode tool. Due to the area effect, with the decrease in electrode diameter, the relative electrode wear rate becomes more serious [17]. Some research work has focused on electrode wear issues during the µEDM process to obtain improved machining accuracy. For example, Bissacco et al. analyzed the electrode wear rate at different energy levels in detail [18]. Wang et al. carried out quantitative research on the electrode wear amount of positive and negative pulses [19]. Tsai et al. conducted a detailed investigation on the wear rate of electrodes of different materials [20]. All these investigations indicate that electrode wear will have a serious impact on the subsequent processing performance, resulting in a decrease in machining accuracy. To address a variety of problems associated with electrode wear, many explorations on preventing and compensating electrode wear have been performed. The measures reported for preventing µEDM wear include the ultrasonic-assisted debris removal method [2], coating electrode method [21], special material electrode method [22], and discharge in gas method [23]. The methods reported for µEDM electrode compensation include the electrode uniform wear compensation method [24], electrode fixed length linear compensation method [25], effective discharge pulse monitoring compensation method [26], and prediction electrode compensation method [27]. However, the above methods are only suitable for specific occasions and have certain limitations, which lead to imperfect final results and difficulty in the promotion of these methods in industrial fields. Moreover, it is difficult to measure the actual wear of microelectrodes online, and the consistency of the online electrode is difficult to guarantee. Therefore, electrode compensation has always been one of the difficult problems in µEDM, and the various electrode compensation methods have not completely solved the problem caused by electrode wear. Thus, it is urgent to find a new processing method to solve the issues of electrode wear.

To address the wear-related problems associated with µEDM, Huang et al. proposed a novel µEDM method that uses a liquid alloy as the machining electrode instead of traditional electrodes [17,28], in which the liquid metal consumed in the process can be compensated over time, and the capillary containing the liquid metal does not participate in the discharge; thus, its shape remains unchanged and solves the problem of electrode wear. Nevertheless, these references reported the experimental characterizations and demonstration of the process on the basis of a preliminary study and lacked an analysis of the liquid-metal electrode morphology, which will affect the machining accuracy. Therefore, this study specifically focuses on analyzing the influence of the external pressure and electric field on the shape of liquid electrodes, as well as the influence of different electrode shapes on machining characteristics. The characterization of the process with varying discharge parameters is discussed, along with the arbitrary patterning of silicon substrates using the developed method.

#### **2. Method**

The principle of conducting µEDM with a liquid-metal electrode is illustrated in Figure 1. A liquid metal is held through a metallic capillary nozzle coated with a dielectric film so that the liquid protrudes on the nozzle tip and forms a droplet. Under the action of an appropriate pressure and electric field, the droplet suspended at the tip will countervail part of the surface tension and become a conical tip that serves as a microdischarge electrode. A pulse generator is applied between the liquid metal and the workpiece. Moving the workpiece toward the liquid electrode initiates a spark discharge when the gap meets the breakdown condition and produces a high temperature to melt and erode the workpiece material. Similar to conventional µEDM, the liquid-metal electrode material will also be consumed, but liquid metal is continuously supplied to the nozzle tip to compensate for the consumption at the right pressure, thereby eliminating its impact on the removal process. In the current method, the system is configured so that the discharge pulses are generated only between the liquid electrode and workpiece through the controlled supply of liquid, and the capillary itself does not participate in the discharge; therefore, it remains intact. Thus, this method uses a liquid metal tip as a microdischarge electrode, which can be automatically compensated. Furthermore, its geometric shape can be controlled and constrained, which is highly desirable for microsized thin-walled flexible devices, thin-film sensors, or workpiece surface etching. μ μ

**Figure 1.** Principle of micro-electro-discharge machining (µEDM) using a liquid-metal electrode.

μ

#### **3. Shape of the Liquid-Phase Electrode**

#### *3.1. Mathematical Model*

To obtain a liquid-metal electrode that meets the discharge demand, the formation conditions and morphology of the electrode are extremely critical.

We know that the surface tension at any point on the plane of the static liquid surface is the same in all directions, counteracting each other. In other words, the pressure just outside the surface *P*<sup>o</sup> and just inside the surface *P<sup>i</sup>* is equal. Therefore, there is no additional pressure on the plane of the static liquid surface. However, if the pressure inside the surface *P<sup>i</sup>* is greater than the pressure outside the surface *P*o, an additional pressure *P<sup>f</sup>* will be generated and induce a curved liquid surface (if the drop is small, the effect of gravity may be neglected and the shape may be assumed to be spherical), and vice versa. Figure 2 shows that the tip of the liquid-metal electrode appears to be a convex sphere projected over the flat bottom of the coated needle (the liquid metal does not wet the coated needle).

**Figure 2.** Tip shape of the liquid-metal electrode.

From the geometric relationship of the solid–liquid contact surface shown in Figure 2, it can be found that the relationship between θ and α is

$$
\theta = \frac{\pi}{2} + \alpha,\tag{1}
$$

2 *θ α* where θ is the liquid–solid contact angle (◦ ), and α is the angle between the boundary radius of the curved liquid surface and the axis of the capillary (◦ ).

The additional pressure at the tip of the liquid electrode is as follows [29]:

*θ α*

$$P\_f = \frac{2\pi r^2 \delta}{\mathcal{R}} / \pi r^2 = \frac{2\delta}{\mathcal{R}} = \frac{2\delta \sin\left(\theta - \frac{\pi}{2}\right)}{r} \tag{2}$$
 
$$\dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots$$

2 2 2 2 =/ = where δ is the coefficient of surface tension (N/m), *r* is the inner radius of the capillary (m), and *R* is the radius of the curved liquid surface (m).

δ *θ θ* π *θ* π There is a trigonometric relationship between the contact angle θ and the additional pressure *P<sup>f</sup>* produced by the curved liquid surface. When the contact angle is θ = π/2, the additional pressure is *P<sup>f</sup>* = 0 Pa. When the contact angle is θ < π/2, the additional pressure is *P<sup>f</sup>* < 0 Pa. This result indicates a negative pressure. The liquid surface at the tip of the liquid metal is concave, and the direction of additional pressure is downward, which will not happen in our research case. When the contact angle is θ = π, the additional pressure *P<sup>f</sup>* is at the maximum, and this state usually leads to a liquid flow or the ejection of liquid.

*θ* π However, under natural conditions, the *P<sup>f</sup>* is too small, and the radius *R* of the convex sphere suspended at the end of the coated needle tip is too large. This result leads to the sagittal *h* (the height of a segment) being too small; thus, the protruding convex sphere used as an electrode cannot meet the requirement for the µEDM process. Therefore, extra pressure is applied to the needle. The pressure on the tip of the liquid-metal electrode is shown in Figure 3.

μ From Figure 3, the liquid metal remains stationary inside the needle at equilibrium, the same as the tip of the liquid metal. The additional pressure *P<sup>f</sup>* generated on the convex liquid surface of the liquid metal is

$$P\_f = P\_1 + P\_{l1} - P\_{l2} \tag{3}$$

where *P*<sup>1</sup> is the extra applied pressure (Pa), *P*l1 is the liquid-metal gravity (Pa), and *P*l2 is the dielectric-liquid gravity (Pa).

1

l1

**Figure 3.** Pressures of the liquid-metal electrode on the nozzle tip.

Hence, the tip of the liquid-metal electrode will maintain its shape in the equilibrium state. Then, the contact angle θ of the liquid-metal electrode tip is *θ*

$$\theta = \frac{\pi}{2} + \arcsin\left[\frac{(P\_1 + P\_{11} - P\_{12}) \cdot r}{2\delta}\right].\tag{4}$$

*θ* 1 l1 l2 + During the EDM process, a pulse generator is applied to the electrode and workpiece. This applied voltage will generate an electric field between the liquid-metal electrode and the workpiece. The electric field will exert an electric field force at the tip of the liquid-metal electrode, which will affect or even change the tip shape of the liquid-metal electrode. At the same time, the change in the tip shape of the liquid-metal electrode will change the electric field distribution between the electrode and the workpiece, and the electric field force on the tip of the electrode will also change.

1 l1 l2

= +arcsin 2 2 In the case without an electric field force, as shown in Figure 4, the tip shape of the liquid-metal electrode will form a spherical shape due to surface tension, as discussed before.

θ α **Figure 4.** Electric field force of the liquid-metal electrode tip.

β

F

1

A 2 0 β β Suppose that the liquid metal is the ideal conductor (its resistivity is assumed to be 0 (as conductivity approaches infinity), which makes calculations easy to perform), the surface of the liquid metal is smooth, and the electric field is an irrotational field; then, the electric field line at point A follows arc

G D B

E

C

3

4

1

⌢ *AB* and line BC. Suppose line segment DA is equal to *r*0; then, according to the geometric relationship in Figure 4, the length of line segment BC can be obtained as follows:

$$d\_1 = \, d\_0 - R(1 - \cos a). \tag{5}$$

The length of arc ⌢ *AB* can be calculated as follows:

$$
\sigma\_0 \beta = \ R \tan(\frac{\beta}{2}) \beta = \beta \mathbb{R} \frac{1 - \cos \beta}{\sin \beta}.\tag{6}
$$

The relationship between the electric potential *U* at point A and the electric field intensity *E* can be expressed as

$$\mathcal{U} = \int \begin{array}{c} \stackrel{\rightarrow}{E} \stackrel{\rightarrow}{d} \stackrel{\rightarrow}{l} = E(r\_0 \beta + d\_1) = E(\frac{\beta \mathcal{R} (1 - \cos \beta) + d\_1 \sin \beta}{\sin \beta}). \tag{7}$$

Then, the average electric field intensity *E* on line ABC can be obtained as follows:

$$E = \frac{U\sin\beta}{\beta R(1-\cos\beta) + d\_1\sin\beta}.\tag{8}$$

In the spherical coordinate system with point *O* as the center of the ball and *R* as the radius, by taking an area differential element *ds* near point A, the charge of point A in this differential element can be calculated as

$$d\eta = \varepsilon Ed = \frac{\varepsilon Ul \sin \beta}{\beta R (1 - \cos \beta) + d\_1 \sin \beta} R^2 \sin \beta d\beta d\gamma,\tag{9}$$

where γ is the azimuth in the coordinate system with a range of 0–2π, and ε is the dielectric constant.

Integrating the entire convex spherical surface, the charge of the entire spherical surface can be obtained as follows:

$$\begin{array}{ll} q & = \int\_{0}^{2\pi} \int\_{0}^{\alpha} \frac{\varepsilon l I \sin \beta}{\beta \mathbb{R} (1 - \cos \beta) + d\_1 \sin \beta} \mathbb{R}^2 \sin \beta d\beta d\gamma \\ & = 2\pi \varepsilon l I \mathbb{R}^2 \int\_{0}^{\alpha} \frac{\sin^2 \beta}{\beta \mathbb{R} (1 - \cos \beta) + d\_1 \sin \beta} d\beta. \end{array} \tag{10}$$

According to the principle of virtual displacement, the electric field force received by the convex spherical surface is given as

$$\begin{split} \text{Fe} \quad &= \frac{1}{2} \frac{\partial (q/\mathcal{U})}{\partial d\_1} \mathcal{U}^2 \\ &= \pi \varepsilon \mathcal{U}^2 \mathcal{R}^2 \int\_0^\alpha \frac{\sin^3 \beta}{\left[\beta \mathcal{R} (1 - \cos \beta) + d\_1 \sin \beta\right]^2} d\beta. \end{split} \tag{11}$$

Since β < π/2, β can be approximately replaced by 2 sin(β/2), substituting this into Equation (11) provides the following:

$$\begin{array}{rcl} \text{Fe} &=& \frac{\pi \epsilon \mathcal{U}^2}{2} \Bigg[ \frac{2(d\_1 + R)}{d\_1} + \frac{2 \mathcal{R}(d\_1 - 4\mathcal{R})}{16 \mathcal{R}^2 + d\_1^2} + \ln \frac{d\_1}{2 \mathcal{R} \sin^2(a/2) + d\_1 \cos(a/2)} \\ &+ \frac{4 \mathcal{R}(8\mathcal{R}^2 + d\_1^2) + 2d\_1(12\mathcal{R}^2 + d\_1^2) \cos(a/2)}{[\mathcal{R}(\cos a - 1) - d\_1 \cos(a/2)] \Big(16 \mathcal{R}^2 + d\_1^2\big)} \\ &- \frac{d\_1(24\mathcal{R}^2 + d\_1^2)}{\left(16 \mathcal{R}^2 + d\_1^2\right)^{3/2}} \ln \frac{\left[\left(\sqrt{16 \mathcal{R}^2 + d\_1^2} + 4\mathcal{R}\right) [-\cos(a/2)] + d\_1[1 + \cos(a/2)]\right]^2}{2\mathcal{R}\sin^2(a/2) + d\_1 \cos(a/2)}. \end{array} \tag{12}$$

Equation (12) shows the electric field force of the entire liquid-metal convex sphere. This is to analyze the entire spherical surface as a whole. However, to analyze the influence of the electric field force on the tip shape of the liquid-metal electrode, it is necessary to divide the spherical surface of 2

 

1

e

the liquid-metal tip into an infinite number of microelements to analyze the force situation of each microelement. The electric field force near any point A on the spherical surface is

11 1

1 1

2 2 2 2 1 1

22 22 1 1 1

2 si 16

4 (8 ) 2 (12 ) cos( / 2) cos 1 cos( / 2) 16

2( + ) 2 ( 4 ) = +ln

1 3/2 2 2 1

22 2 1 1 1

2 16 2 sin ( / 2) cos( / 2)

2 2

 2

n ( / 2) cos( / 2) <sup>1</sup>

16 4 1 cos( / 2) 1 cos( / 2) (24 ) ln

$$dF\_{\varepsilon} = \frac{1}{2} \frac{\varepsilon \mathcal{U}^2 \mathbb{R}^2 \sin^3 \beta}{\left[\beta \mathbb{R}(1 - \cos \beta) + d\_1 \sin \beta\right]^2} d\beta d\gamma. \tag{13}$$

2

If *r* = 10 µm, *d*<sup>0</sup> = 100 µm, and *U* = 100 V, the value range of β is [−π/2–π/2], and *d*β and *d*γ are constant at 1; then, the curve of the electric field force *dFe* around β at any point A on the spherical surface is as shown in Figure 5. μ μ *β* −π π *β γ β*

*β* **Figure 5.** Relationship between the electric field force *dFe* and β at any point A on the spherical surface.

As shown in Figure 5, the electric field force on the spherical surface is not evenly distributed. The electric field force is the largest at the lowest point G of the spherical surface, and, with point G as the center, the electric field force gradually decreases toward point E, which is similar to point F. This nonuniform distribution of the electric field force may cause the tip shape of the liquid-metal electrode to change from a spherical shape to a Taylor cone.

#### *3.2. Simulation*

#### 3.2.1. Geometric Model and Boundary Condition

μ Figure 6 shows the geometric size and boundary conditions of the model used for simulation. The size of the simulation area is 0.4 × 0.6 mm, the length of the capillary nozzle is 0.5 mm, and the distance between the nozzle tip and the workpiece is 0.1 mm. Zone A is the liquid metal inside the electrode, zone B is stainless steel, zone C is the EDM oil, and zone D is a parylene C coating with a thickness of 20 µm, as shown in Table 1, where *E* the electrostatic field on the boundary when a voltage is applied between liquid-metal electrode and workpiece.

At present, most metals or alloys are in the solid state at room temperature. Exceptions include francium, cesium, rubidium, mercury, sodium–potassium alloys, and gallium-based alloys, which can be defined as liquid metals. Their melting points are either lower than or close to room temperature, which enable them to remain in the liquid state at room temperature. Unfortunately, the intrinsic radioactivity of cesium, extreme instability of francium and rubidium, flammability and corrosivity of sodium–potassium alloys, and toxicity of mercury limit their applications to certain specific areas. On the other hand, gallium-based alloys, such as Galinstan (68.5 wt.% gallium, 21.5 wt.% indium, and 10 wt.% tin), a commercially available eutectic liquid alloy, is a low-activity and nontoxic liquid with a low melting point (−19 ◦C) and low viscosity (0.0024 Pa·s at 20 ◦C) that allows it to be easily transferred through microscale nozzles. Its high electrical conductivity (~3.5 <sup>×</sup> <sup>10</sup><sup>6</sup> <sup>S</sup>/m at 20 ◦C), high

thermal conductivity (16.5 W·M−<sup>1</sup> ·K−<sup>1</sup> ), and high boiling point (>1300 ◦C) are desirable features for a µEDM electrode application. Therefore, Galinstan is used as the liquid electrode in this study.

**Figure 6.** Geometric size and the boundary conditions of the model.



3.2.2. Tip Shape of the Liquid Electrode at Different Extra Pressures

− − − μ *θ* From the theoretical analysis in Section 2, it is known that the extra pressure is an important factor affecting the contact angle θ of the liquid-metal electrode. To evaluate this effect in detail, the influence of extra pressure on the tip shape of the liquid-metal electrode was simulated by varying extra pressures. The simulation results are shown in Figure 7. When the extra pressure *P*<sup>1</sup> is 0.1 atm, the liquid metal cannot be extruded from the needle (Figure 7a). The reason may be that the extra pressure is too small to overcome the internal frictional resistance caused by Galinstan's viscosity. As the extra pressure increases, the liquid-metal electrode tip gradually becomes a cone (Figure 7b,c). A larger extra pressure results in a faster change in the tip shape of the liquid-metal electrode. The tip shape of the liquid-metal electrode gradually changes into a spherical shape under the action of pressure. The liquid metal gradually extends to the parylene coating at the bottom of the needle, making the radius of the liquid-metal electrode tip larger than the inner diameter of the needle. This result implies that the width of the groove patterning by liquid-metal electrode µEDM may be larger than the inner diameter of the needle. Moreover, as the extra pressure increases, the sagitta of the tip of the liquid-metal electrode grows longer (Figure 7d). This result verifies the analysis results of Section 2. It is worth noting that, for an extra pressure over 2 atm, the liquid metal will promptly eject and touch the workpiece (Figure 7e). Therefore, it is necessary to set an appropriate pressure to obtain an electrode that meets processing needs.

**Figure 7.** Simulation results at different extra pressures. (**a**) 0.1 atm. (**b**) 0.8 atm. (**c**) 1.2 atm. (**d**) 1.4 atm. (**e**) 2 atm.

#### 3.2.3. Tip Shape of the Liquid Electrode at Different Voltages

Figure 8 shows the simulation result of the tip shape of the liquid-metal electrode at different voltages (tested up to 2 kV, with an extra pressure of *P*<sup>1</sup> = 1.2 atm). As shown in Figure 8a,b, it can be seen that the tips of the liquid-metal electrode are almost the same in the case of low voltage. This outcome is reasonable because the electric force at low voltage is not high enough to affect the tip of the liquid-metal electrode.

**Figure 8.** Simulation results at different voltages. (**a**) 50 V. (**b**) 100 V. (**c**) 0.5 kV. (**d**) 1 kV. (**e**) 1.5 kV. (**f**) 2 kV.

When the voltage is 0.5 kV, as shown in Figure 8c, the tip of the liquid-metal electrode is gradually stretched into an ellipsoidal shape under the combined action of the electric field force and the extra pressure. Compared with Figure 8a or 8b, the sagitta of the tip shape of the liquid-metal electrode is even longer and changes more rapidly over time. This may be attributed to the tensile effect of the electric field force generated by the high voltage on the liquid metal. As the voltage continues to increase, the tip of the liquid-metal electrode will clearly change. As seen from Figure 8d,e, the tip of the liquid-metal electrode changes over time and is gradually stretched into an inverted cone. Compared with Figure 8c, the tip of the liquid-metal electrode is now more of a conical shape, with a smaller diameter. The electric field force generated by the voltage constrains the tip shape of the liquid-metal electrode, and the constraint becomes stronger with increasing voltage. When the electric field force

generated by the additional high voltage is sufficiently large, the liquid-metal electrode terminal can be stretched into a reverse cone.

It is also worth noting that for a voltage greater than 2 kV, the tip of the liquid-metal electrode is gradually stretched into an inverted cone and then ejected to the workpiece (as shown in Figure 8f). This result can be explained as the stretching effect of the liquid metal under the extra pressure and electric field force. This explanation is reasonable because the electric field force is too strong.

From the simulation, it can be seen that the effect of a low voltage on the electrode tip is not obvious, but the effect of a high electric field force is very obvious. Therefore, the external electric field may be an appealing way to restrain the tip of the liquid-metal electrode in future research.

#### **4. Experimental Set-Up and Procedure**

≥

μ

μ

The schematic in Figure 9 illustrates the µEDM process experiments. The set-up has a servo-controlled three-axis stage with a 100 nm positioning resolution, comprising an *XY* stage with the holder on which the work tank is held and a *Z* stage to vertically position the syringe containing the needle with a liquid-metal electrode. The work tank is configured to have a sample holder made of plastic, in which the workpieces are immersed in a dielectric EDM fluid, fixed and electrically coupled with the discharge circuit using conductive adhesive tape. The nozzle is connected to the syringe that stores Galinstan, and a pressurizing unit is used to apply pressure to the syringe to feed Galinstan to the needle. To prevent the oxidation of Galinstan, a low-concentration (2%) sulfuric acid solution is added to the syringe and floats on top of Galinstan. Furthermore, the H2SO<sup>4</sup> solution keeps the liquid metal clean to avoid clogging the needle. Galinstan is electrically coupled with the discharge circuit using a conductive copper wire immersed in it and sealed with glue on the syringe wall.

**Figure 9.** Experimental set-up for liquid-metal electrode µEDM.

The system employs a relaxation-type (resistor–capacitor, or RC) pulse generator with a variable direct current (DC) voltage source. In this type of pulse generation circuit, the loop-voltage equation can be expressed as

$$\mathcal{U}\_{\mathfrak{C}} = E(1 - e^{-\overleftarrow{\tilde{\tau}}^{t}}),\tag{14}$$

μ μ

μ

μ

μ

(1 ) where *U*<sup>c</sup> is the voltage on the capacitor C (V), *E* is the voltage of the DC source S (V), and *T* = *RC*. In this case, the discharge energy is stored in the capacitor. When the voltage of the capacitor meets the condition of *U<sup>c</sup>* ≥ *U<sup>d</sup>* (*U*<sup>d</sup> is the breakdown voltage of the discharge gap), the gap discharges and instantly releases the energy forming a pulse current. The generated discharge pulses are monitored

using a detection circuit, and the readouts of the signals are obtained via a General Purpose Interface Bus (GPIB) interface connected to a computer; moreover, a probe is coupled with the discharge circuit and connected to an oscilloscope for visual observations.

A stainless-steel needle (outer diameter *D*<sup>O</sup> = 240 µm, inner diameter *D<sup>I</sup>* = 100 µm, length *L* = 13 mm) is used as the needle in the set-up. The needle is coated with a dielectric film, parylene C (thickness, 20 µm), to eliminate the effects of discharge on the needle surface. The coating consequently modifies the *D*<sup>O</sup> and *D<sup>I</sup>* of the needle to 280 and 60 µm, respectively. Although the distance, *D*, between the workpiece surface and the needle tip cannot be directly determined via electrical surface detection, it can make use of a nozzle with the presence of Galinstan at its tip. Briefly, a low pressure (less than 0.5 atm) is applied via a precision pressure device to ensure that the liquid metal in the needle is just flush with the tip of the needle. Then, using a small voltage (10 V) between them electrical surface detection is carried out using the *Z*-axis of the stage at a low speed, and retracting the needle (50 µm) to determine the gap, as shown in Figure 10a–c. To decrease the measuring error, electrical surface detection both toward and away from the gap is repeated continuously three times, and the average is taken as the gap position. The µEDM process with this setting is performed in the die-sinking mode, and the workpiece is electrically coupled with the discharge circuit cathode while the electrode is connected to the anode. After applying a normal pressure at the nozzle and applying a machining voltage between the nozzle and workpiece, the gap *d* between the liquid metal and the workpiece is adaptively determined during discharging, with horizontal scanning of the workpieces using the *XY* stage with respect to the liquid electrode having a constant gap distance *D* for the lateral removal of the material using the liquid electrode, as shown in Figure 10d. The experimental conditions are outlined in Table 2.

μ μ **Figure 10.** Electrical surface detection and scanning erosion. A low pressure (0.1 atm) is applied via a precision pressure device to ensure that the liquid metal in the needle is just flush with the tip of the needle; then, a small voltage (10 V) is used between them to carry out electrical surface detection using the *Z*-axis of the stage at a low speed from (**a**) *D* = 0 µm (**b**) to *D* = 50 µm (retracting the needle) to determine the gap. (**c**) The above electrical surface detection is repeated three times, and the average is taken as the gap position. Additionally, the gap *d* between the liquid metal and the workpiece is adaptively determined during discharging. (**d**) Horizontal scanning of the workpieces using the *XY* stage with respect to the liquid electrode at a constant gap distance *D* for the lateral removal of the material using the liquid electrode.

μ

Ω

μ

Ω

μ


**Table 2.** Experimental conditions.

Machining characteristics such as the material removal rate (MRR) (mm<sup>3</sup> /min) and liquid-metal consumption rate (LMCR) (mm<sup>3</sup> /min) are adopted to evaluate the effects of the machining parameters on the liquid-metal electrode µEDM processes. The MRR is computed as the ratio of the material removed from the workpiece (approximated as the volume of the frustum of the cone) to the recorded machining time of the EDM system. The LMCR is calculated as the ratio of the liquid metal consumed to the machining time.

#### **5. Experimental Results and Discussion**

#### *5.1. Microgroove and Arbitrary Patterning*

Microgroove erosion was performed through programming the motions of the three-axis stage. An example of a microgroove pattern using the coated needle is shown in Figure 11. The electrical conditions were *C* = 27 nF, *R* = 1 kΩ, *V<sup>s</sup>* = 100 V, and *P*<sup>1</sup> = 1.2 atm. The tool scanning path in the *XY* plane was repeated for 200 scans at scanning speeds of up to 1.0 mm/s with a constant gap distance of *D* = 100 µm. A pressure of 1.2 atm was applied to the syringe. The profile of the produced groove was analyzed using a laser confocal microscope, and the sample results are included in Figure 12. The linewidth and depth of the produced groove were measured to be approximately 280 and 90 µm, respectively. The cross-sectional profile of the groove was similar to the shape of the liquid electrode tip. This result is reasonable because EDM is a copying process mode, and the groove will match the electrode. In addition, randomly distributed pits appeared at the edge of the processed groove in Figure 11a. The reason for the micropits near the edges of the produced groove is that some pulses were likely generated between the suspended liquid microelectrode and debris comprising the removed material of the workpiece and the liquid metal that had already dripped on the workpiece surface. By programming the contour of the nozzle motion with respect to the *XY* plane (with the same *D* value), arbitrary patterning was successfully demonstrated on the silicon sample (Figure 12).

**Figure 11.** Linear groove pattern. (**a**) Sample of the microgroove and (**b**) profile of the microgroove.

Ω

**Figure 12.** Sample result of arbitrary scanning-mode patterning performed using the developed process: (**a**) convex circular pillar and (**b**) "HIT" characters.

#### *5.2. Process Characterizations*

#### 5.2.1. Dependence of Machining Characteristics on the Pressure

Ω The widths of the linear patterns produced by lateral scanning of the Galinstan electrode were characterized while varying the pressure. The line widths of the produced patterns were measured using an optical measuring microscope. Figure 13 shows the relationship between the measured line widths and pressure (with *V<sup>s</sup>* = 100 V, *C* = 100 nF, *R* = 0.5 kΩ, as shown in Table 2). As shown in the graph, an optimal pressure value for the minimum line width was found under the same discharge conditions.

**Figure 13.** Effect of pressure *P*<sup>1</sup> on the line widths.

*θ θ θ* When the applied pressure was small, the liquid-solid contact angle θ was also small; hence, the tip of the liquid electrode was large. With an increase in applied pressure, the contact angle θ gradually increased, and the radius *r* of the curved liquid surface gradually decreased. That is, the end of the electrode correspondingly decreased; thus, the width of the machined groove decreased directly. When the applied pressure continued to increase, although the contact angle θ gradually increased, the radius *r* of the curved liquid surface gradually decreased, forming an electrode with a finer end; however, the width of the groove gradually increased. The reason for this result may be that when the applied pressure is greater, the liquid metal at the tip is more likely to fall onto the workpiece, which will affect the processing and cause discharge instability. Moreover, we also found that, when the external pressure was greater than 2 atm, the liquid metal ejected like a continuous stream of water; thus, it was impossible to carry out EDM.

In this connection, it is clear that there was a similar trend in the relationship between the MRR and pressure *P*1, as shown in Figure 14. During the discharge process, when the pressure was small, the consumed Galinstan could not be compensated in time, leading to an open-circuit state most of the time; this resulted in the small value of the MRR. The MRR increases with increasing pressure because the frequency of discharge increases with an increasing Galinstan compensation. After the optimal pressure value, as the pressure increases, excess Galinstan is suspended on the needle tip, and this droplet easily falls due to the discharge and electrostatic forces and causes unstable discharge, thus resulting in the MRR decreasing.

*θ*

*θ*

*θ*

**Figure 14.** Effect of pressure *P*<sup>1</sup> on the material removal rate (MRR).

Figure 15 shows the relationship between the LMCR and pressure *P*1. As shown in the graph, the LMCR increased with increasing pressure in a consistent manner. This outcome is reasonable because the flow rate, which determines the consumption of Galinstan based on the velocity of the flow, increased with increasing pressure. It is worth noting that, for pressures over 1.3 atm, the amount of fed Galinstan was much greater than the amount of Galinstan consumed by discharge at an excessive flow velocity. This result is disadvantageous for both the MRR and LMCR. Therefore, the feeding of consumed Galinstan is very important for high efficiency and stable processing, which largely depends on the applied pressure.

**Figure 15.** Effect of pressure *P*<sup>1</sup> on the liquid-metal consumption rate (LMCR).

#### 5.2.2. Dependence of Machining Characteristics on the Voltage

Ω Figure 16 plots the measured line widths as a function of the voltage (tested up to 220 V, with constant *C* = 100 nF, *R* = 0.5 kΩ, and *P*<sup>1</sup> = 1.3 atm, as shown in Table 2), and the figure clearly shows that the line width increased with the applied voltage in a consistent manner. This result is reasonable as the discharge energy, which determines the amount of material removed by a single pulse, increased with the voltage. In this regard, as shown in Figures 17 and 18, a similar trend was obtained for both the MRR and LMCR by varying the voltage *V*s; the trends were consistent given the dependence of the discharge energy on the voltage. Increasing the discharge energy caused a greater spark intensity, which generated more melted material in the spark region. Thus, the workpiece and the electrode material were both subjected to an increase in sparks, which increased both the MRR and the LMCR. As a result, the MRR and LMCR substantially increased with increasing discharge energy.

Ω

Ω

**Figure 16.** Effect of voltage *V<sup>s</sup>* on the line widths.

**Figure 17.** Effect of open voltage *V<sup>s</sup>* on the MRR.

**Figure 18.** Effect of voltage *V<sup>s</sup>* on the LMCR.

*θ* It can be seen from the simulation that a higher voltage resulted in a greater electric field, whereas, a larger contact angle θ resulted in a smaller radius *r* of the curved liquid surface. That is, the electrode tip was correspondingly smaller, and the widths of the linear patterns produced by the lateral scanning of the Galinstan electrode should be smaller as the voltage increases. However, the line width showed the opposite trend in this experiment, in that the line width increased with the voltage. There may be two reasons. First, the material removal depends on the energy of a single electric discharge, *E*SED, which is expressed as

$$E\_{\rm SED} = \frac{1}{2} \mathcal{C} \mathcal{U}^2. \tag{15}$$

μ μ

1 <sup>2</sup> 2 where *C* is the capacitance of the RC circuit and *U* is the machining voltage. From Equation (15), it can be seen that, as the voltage increases, the discharge energy of a single pulse increases squarely with the voltage; thus, the amount of removed material increases. The effect of increasing the voltage on the energy for material removal is much greater than that on the size of the electrode.

μ

μ

Second, the applied voltage for µEDM is relatively small, and the voltage changes within a small range of 50 to 220 V. The size of the electrode tip does not change significantly at low voltages. Only when the voltage is over 1 kV will a change in the liquid electrode tip be apparent. However, it is impossible to use such a high voltage for µEDM.

Therefore, changing the size of the electrode by increasing the voltage shows a much smaller effect on groove width than the effect of increased energy by increasing the voltage. Notably, from an energy point of view, an increase in voltage is beneficial to the MRR.

#### **6. Conclusions**

In this paper, the shape of the electrode, which is used in liquid-electrode µEDM processing for the purpose of resolving the problems related to electrode wear in traditional µEDM, was analyzed and simulated. Moreover, experiments were performed for verification. Arbitrary patterns on silicon samples using a liquid-metal electrode were well presented. The influence of extra pressure and open voltage *V<sup>s</sup>* on the line widths, MRR, and LMCR were also investigated. The external pressure had a significant impact on the tip shape of the liquid electrode and its compensation. There was an optimum pressure value for the minimum line widths and maximum MRR. The LMCR increased with pressure; hence, an appropriate pressure was also beneficial to the LMCR. Although the voltage also influenced the tip shape of the liquid electrode, it was negligible at low voltage. However, at low voltages, the line widths, MRR, and LMCR consistently exhibited the same trend as the voltage. That is, they increased with voltage. Thus, the single-pulse discharge energy, which determines the amount of material removed by a single pulse, increased with the voltage.

**Author Contributions:** Conceptualization, R.H. and Y.Y.; methodology, R.H.; software, E.Z.; validation, R.H., Y.Y., E.Z. and X.X.; formal analysis, E.Z. and R.H; investigation, R.H. and E.Z.; resources, R.H.; data curation, E.Z.; writing—original draft preparation, R.H.; writing—review and editing, R.H. and X.X.; visualization, X.X.; supervision, R.H.; project administration, R.H.; funding acquisition, R.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the National Natural Science Foundation of China (No. 51975155, No. 51475107) and the Shenzhen Basic Research Program (No. JCYJ20170811160440239, No. JCYJ20200824082533001).

**Acknowledgments:** The author would like to thank Professor K. Takahata from the University of British Columbia for providing technical support in this research.

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

#### **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* **Pulse-Type Influence on the Micro-EDM Milling Machinability of Si3N4–TiN Workpieces**

#### **Valeria Marrocco 1,\* , Francesco Modica <sup>1</sup> , Vincenzo Bellantone <sup>1</sup> , Valentina Medri <sup>2</sup> and Irene Fassi <sup>3</sup>**


Received: 15 September 2020; Accepted: 12 October 2020; Published: 13 October 2020

**Abstract:** In this paper, the effect of the micro-electro discharge machining (EDM) milling machinability of Si3N4–TiN workpieces was investigated. The material removal rate (MRR) and tool wear rate (TWR) were analyzed in relation to discharge pulse types in order to evaluate how the different pulse shapes impact on such micro-EDM performance indicators. Voltage and current pulse waveforms were acquired during micro-EDM trials, scheduled according to a Design of Experiment (DOE); then, a pulse discrimination algorithm was used to post-process the data off-line and discriminate the pulse types as short, arc, delayed, or normal. The analysis showed that, for the considered process parameter combinations, MRR was sensitive only to normal pulses, while the other pulse types had no remarkable effect on it. On the contrary, TWR was affected by normal pulses, but the occurrence of arcs and delayed pulses induced unexpected improvements in tool wear. Those results suggest that micro-EDM manufacturing of Si3N4–TiN workpiece is relevantly different from the micro-EDM process performed on metal workpieces such as steel. Additionally, the inspection of the Si3N4–TiN micro-EDM surface, performed by SEM and EDS analyses, showed the presence of re-solidified droplets and micro-cracks, which modified the chemical composition and the consequent surface quality of the machined micro-features.

**Keywords:** ceramic composite; micro-EDM milling; pulse discrimination

#### **1. Introduction**

Silicon nitride-based ceramics constitute a class of structural materials characterized by high strength, fracture toughness, thermal shock resistance, wear resistance, low coefficient of friction, and hardness. These ceramics can often be sintered by adding electrically conductive reinforcements, such WC, MoSi2, TiN, TiC, TiCN, TiB<sup>2</sup> and ZrN, with variable content, thus obtaining electro-conductive ceramic composites [1–3]. This feature allows for the manufacture of these materials via non-contact technologies such as electrical discharge machining (EDM). In particular, EDM has shown feasibility for the realization of free form features and shapes required in high-temperature and aggressive environments for the production of various heating elements (e.g., ceramic glow plugs, igniters, ceramic heaters [4]. Moreover, due to the fact of their biocompatibility, some of these ceramic composites were investigated for the realization of load-bearing prostheses and bone mini-fixation devices [3–6]. In this field, micro-EDM displayed a high potential for the fabrication of micro-free-form features in Si3N4–TiN ceramic composite workpieces needed to realize such medical components [7].

Issues related to micro-EDM manufacturing of Si3N4–TiN ceramic composite have been already explored by the research community. Indeed, machining performance and surface quality were analyzed by Liu et al. [8] in relation to different machining regimes and to the manufacturing of a mesoscopic gas turbine. From the process viewpoint, the main results demonstrated that when a relaxation-type pulse generator was chosen for the micro-EDM process, faster machining and optimized material removal rate (MRR) and low tool wear rate (TWR) could be accomplished at the expense of surface quality and roughness. In particular, surface quality inspection showed that chemical reactions, triggered by the spark occurrence, induced a modification of the composite structure and also took part in the material removal mechanism. Subsequently, Liu et al. [9] presented an analysis on the micro-EDM capability of manufacturing Si3N4–TiN ceramic composites, considering different machining regimes. In that work, the authors presented evidence of the variability ascribed to the pulse shape and duration of the discharge waveforms, not entirely due to the process setting. Therefore, also in this case, Si3N4–TiN material seemed to have played an active role in the material removal process, almost independently of the machining regimes considered for the analyses. It was evident that the optimization of micro-EDM related to ceramic composites deserved further investigation, since the relatively low value of the electrical conductivity of such material, the chemical composition of the matrix, and the consequent intrinsic inhomogeneity of the composite introduced several challenges. Very recently, a work by Selvarajan et al. [10] pointed out the effect of micro-EDM process parameters (voltage, current, pulse on time, and pulse off time) on MRR and surface quality related to two Si3N4-based composites. The results showed once more the challenge of assessing optimized process parameters and accomplishing the required feature accuracy.

In the last decade, several authors have proposed alternative approaches to investigate micro-EDM process more thoroughly. In particular, the monitoring of voltage (gap and/or open) and discharge current waveforms became established to improve the general understanding of the micro-EDM process. However, the majority of the papers dealing with the micro-EDM monitoring approach considered metals as workpieces, and their main goal was the assessment of tool wear. For instance, the authors in Reference [11] exploited discharge monitoring during micro-EDM milling to measure tool wear per discharge and material removal per discharge. A method based on pulse counting was proposed in Reference [12] to estimate the total energy associated to discharge pulses and investigate material removal and tool wear characteristic for different micro-EDM machining types (shape-up and flat-head) and conditions (spindle rotation and tool electrode vibration). The discrimination of positive and negative parts of voltage and current waveforms was proposed in Reference [13] with the goal of estimating tool wear and tool wear error. However, the reported results held validity for relatively stable micro-manufacturing and features having small depths.

A different use of micro-EDM monitoring was proposed in Reference [14]. In this work, a first classification of pulses in relation to micro-EDM milling and wire processes was presented, and four main pulse categories were identified: normal, effective arc, transient short circuit, and complex pulses. A pulse-type discrimination strategy was also implemented and applied to micro-EDM milling of hardened steel in order to investigate the influence of process parameters on discharge shapes and process performance [15]; the study emphasized the importance of sparking gap and feed rate for the process stability and identified that tool wear increase could be associated to the increase of arc number. Recently, numerical simulations and experiments were used to evaluate the occurrence of pulse types in dependence of debris accumulation [16,17]. The studies were carried out considering reverse micro-EDM (RMEDM) and confirmed that the increase of debris quantity induced a modification of pulse shapes. In particular, the authors found different pulse orders, where the first one was identified as a normal pulse, and the other two pulse types (second and higher orders, characterized by lower voltage values) occurred more frequently as the concentration of debris increased within the sparking gap. In order to evaluate the differences among pulse types and shapes, a different approach based on the analysis of power spectral density (PSD) applied to the micro-EDM milling process was proposed considering two different workpiece materials: hardened steel [18] and Si3N4–TiN [19]. The

μ

results underlined that especially for Si3N4–TiN, the number of discharges per acquisition windows underwent a huge variability, and the discharge probability was very low compared to steel workpiece manufacturing. Moreover, different energetic contributions to the material removal process from normal pulses was observed for the micro-EDM milling of Si3N4–TiN workpieces independent of the energy settings.

Although relevant investigations of the manufacturing of Si3N4–TiN ceramic composites are currently available in the literature, the influence of discharge pulse shapes on micro-EDM performance indicators has not been fully assessed. Therefore, in the present paper, micro-EDM milling of Si3N4–TiN ceramic composite was analyzed by means of pulse-type characterization and MRR and TWR evaluation. The experiments were performed by implementing a design of experiment (DOE): the finishing regime was applied, thus implying the use of relaxation-type generator producing short pulses. The voltage and current waveforms were acquired during 54 trials resulting from the DOE, where process parameters, frequency (F) and gap and pulse width (W), were varied. The pulse classification was obtained by an off-line pulse discrimination strategy. The post-processed data related to pulse types and distribution were then put in relation to MRR and TWR. Finally, chemical and surface characterization of the Si3N4–TiN ceramic composite workpiece, before and after the micro-EDM process, is also reported and discussed.

#### **2. Material, Micro-EDM Settings, and Design of Experiment (DOE)**

The workpiece considered in this study was a sintered billet (25 mm in diameter) of Si3N4–TiN composed by a mixture of commercial raw powders. A TiN (grade C, HC Starck Ltd., Munich, Germany) 35% in volume was used as secondary electro-conductive phase and Y2O<sup>3</sup> (grade C, HC Starck Ltd., Munich, Germany) and Al2O<sup>3</sup> (Ultra-High Purity, Baikowski Chemie SA, Poisy, France) were used as sintering aids. The electrical resistivity was equal to 5.88 <sup>×</sup> <sup>10</sup>−<sup>4</sup> <sup>Ω</sup> cm. Figure <sup>1</sup> reports the SEM micrograph of the mirror polished and plasma-etched surface of the Si3N4–TiN-35 vol% composite: the typical microstructure of silicon nitride, composed of elongated silicon nitride grains in a matrix of nearly equiaxed sub-micrometric grains, was highlighted. <sup>−</sup> Ω

**Figure 1.** SEM micrograph of a mirror polished and plasma-etched surface of the Si3N<sup>4</sup> -35 vol% composite: dark grains are Si3N<sup>4</sup> particles, while coarser, light grey grains are TiN particles.

μ μ μ As highlighted in Figure 1, TiN particles are coarser (from 1 to 5 µm) than the silicon nitride grains. The grain boundary phase, consisting of silicates and oxynitrides of the cations from sintering aids, locates preferentially at triple points, and a continuous film of amorphous phase is present at the interfaces between Si3N<sup>4</sup> grains. Furthermore, after cutting the sintered billet by diamond tool machining (DTM), two different Ra were measured: a longitudinal surface roughness equal to Ra = 0.16 µm and a transversal surface roughness equal to Ra = 0.30 µm.

The basic feature used to implement the experimental plan was a micro-channel 5 mm long, 50 µm deep, and 0.42 mm wide, manufactured onto the sintered Si3N4–TiN workpiece. The micro-EDM machine was the Sarix SX 200. The tool was a tungsten carbide (WC) cylindrical rod having a nominal diameter of 0.4 mm; hydrocarbon oil was used as dielectric fluid in the experiments. All technological parameters adopted for the trials are reported in Table 1.


**Table 1.** The micro-electro discharge machining (EDM) milling process parameter settings.

For Sarix SX 200, E is an index identifying the pulse generator type selected for the machining. In this case, E110 indicates finishing regime actuated by RC relaxation-type generator capable of producing short pulses. The OCV is the open voltage value, i.e., the maximum voltage achievable before discharge. Since the micro-EDM milling approach is implemented via a layer-by-layer strategy, a layer thickness (LT) of 1 µm was set. Frequency (F) indicates the frequency of the micro-EDM pulse generator, while pulse width (W) indicates the time interval in which the micro-EDM generator is disconnected from the tool and the workpiece and the discharge occurs within the sparking gap. Finally, Gap is the index related to the servo control loop responsible for the sparking gap stability (i.e., the distance between the tool and the workpiece). In order to estimate the statistical impact of the parameters' variability on machining performance and pulse distribution, a general full factorial Design of Experiment (DOE) was implemented: the varied process parameters were frequency (F) and pulse width (W) and Gap, which were grouped in 18 sets of parameters (SoP); three replicas of the experimental plan were performed, resulting in a total of 54 trials. μ

#### **3. Micro-EDM Monitoring Setup and Pulse Discrimination Strategy**

The sketch of the micro-EDM monitoring setup, along with the actual machining setup, is reported in Figure 2.

**Figure 2.** (**a**) Sketch of the micro-EDM monitoring setup; (**b**) actual monitoring setup connected to the Sarix SX 200 via voltage and current probes; (**c**) voltage and current probes mounted on the Sarix SX 200.

The acquisition of the voltage and current waveforms was done by means of two probes—a standard voltage probe and a Tektronix TCP312 current probe with a bandwidth of 100 MHz. The voltage probe was placed close to the tool tip in order to record voltage values (OCV, gap voltage). As the current probe works through Hall effect, it was hooked to the cable supplying the tool. In order to acquire and record the waveforms, both probes were connected to the oscilloscope—Tektronix MSO4054. The pulse generator was not connected to the oscilloscope.

The oscilloscope sampling frequency (Fs = 50 MHz) and the number of samples for each waveform (Ns = 10<sup>6</sup> ) were set considering the bandwidth of signals to be acquired and the duration of the phenomenon. With this setting, the duration of each acquisition window was equal to Tacq = 20 ms. The acquired data were delivered to a PC through the oscilloscope USB port: due to the USB port speed, an elapsed time of 0.8 s between two subsequent windows in each trial must be considered. The acquired waveforms were then supplied as input to the off-line classification algorithm which provided the number of each pulse type as an output. The pulse types were gathered in four classes: short, arc, delayed, and normal [14,15,18,19]. The discharge events occur within the sparking gap during the time interval equal to T–W, being T = 1/F, but only if the spark condition is fulfilled. When the discharge happen, a drop of OCV to its minimum value is observed. In this case, a normal pulse is detected. If the spark condition is not favorable, OCV keeps its constant value, as no discharge occurs. Nonetheless, local machining conditions, such as the presence of unremoved debris in the sparking gap, can promote discharge events independently of the pulse generator; in this case, two types of pulses can be observed: arcs and delayed. The distinction between these two categories can be made based on the maximum voltage value. Generally, arcs have lower Vmax than delayed. Short pulses happen when both tool and workpiece electrodes come into contact due to the instantaneous local debris concentration, and they are characterized by low voltage values approaching the ground level. Short pulses do not provide any material removal process and generally produce a waste of time.

The flow chart of the pulse discrimination strategy is reported in Figure 3; all routines were implemented in MATLAB®. The beginning of the algorithm foresees an initialization phase, where all pulse type counters—N<sup>p</sup> (total), N<sup>s</sup> (shorts), N<sup>a</sup> (arcs), N<sup>d</sup> (delayed), N<sup>n</sup> (normal)—are reset. Subsequently, the acquisition windows of one trial are analyzed one by one via the algorithm, to assess the presence and type of pulses. The first iteration comprises the computation of the time derivative of the V waveform (dV); then, this value is compared to a set threshold (dVth). If dV < dVth, one pulse is identified, and the total counter N<sup>p</sup> is incremented by one (N<sup>p</sup> + 1). If this condition is not satisfied, the program starts searching again for a pulse. If a pulse is found, the algorithm proceeds to its identification, starting from shorts: in this case, a comparison among mean values of V and I and defined thresholds VthrShort and IthrShort is performed. If this check succeeds, the counter N<sup>s</sup> is incremented by one (N<sup>s</sup> + 1). If this check fails, then the algorithm searches for delayed pulse. A delayed pulse is identified by comparing the maximum I recorded during the discharge with the defined threshold IthrDel. If this check succeeds the counter N<sup>d</sup> is incremented by one (N<sup>d</sup> + 1), otherwise the algorithm searches for arcs. The criterion for this comparison is the same as before, along with the comparison of the mean V with a programmable threshold VthrArc. If this check succeeds the counter N<sup>a</sup> is incremented by one (N<sup>a</sup> + 1), otherwise the pulse is classified as normal and the counter N<sup>n</sup> is incremented by one (N<sup>n</sup> + 1). At the end of the classification, the index of the derivative of the voltage array is incremented of a programmable amount (about a half of the pulse width) in order to speed up the next pulse search, and the loop starts over again to analyze the subsequent acquired observation window.

**Figure 3.** Flow chart of the discrimination strategy algorithm applied to all acquisition windows at each trial.

The defined current and voltage threshold values shown in Table 2.

**Table 2.** I and V threshold settings.


#### **4. Results**

In this section, pulse-type distribution, statistical analysis, and the results regarding the influence of pulse-type number and occurrence on MRR and TWR are presented and discussed.

#### *4.1. Pulse-Type Distribution*

μ μ The monitoring of each trial required a certain number of acquisition windows, approximately 250 for each process. The monitored trials related to all micro-EDM processes displayed relevant variability of discharge occurrence. In order to explain this fact, we reported the count of all pulse types recorded in the last 40 acquisition windows (i.e., the last part of the machining) and pertaining two distinct trials: F = 120 kHz, G = 60, W = 3 µs and F = 120 kHz, G = 60, W = 1 µs. As shown in Figure 4, the total number of all discharges occurring in each acquisition window was very variable. This fact confirms issues about the repeatability of the micro-EDM process on this kind of ceramic composite, mainly ascribable to the workpiece itself rather than to the parameter settings. Nonetheless, it is evident that in all acquired subsequent widows, the number of normal pulses always exceeds the number of arcs, delayed, and shorts.

**F = 120 KHz, W = 3 us, Gap = 60 Number of Pulses**

μ μ **Figure 4.** Number of normal pulses counted in the latest 40 acquisition windows for two different trials: (**a**) F = 120 kHz, W = 3 µs, G = 60; (**b**) F = 120 kHz, W = 1 µs, G = 60. μ μ

Figure 5 shows the average distribution in percentage of all pulse types for all trials calculated during the last part of the micro-EDM manufacturing. The histograms highlight that normal pulses are the most numerous, indicating the general stability experienced during all micro-EDM trials. This fact is also confirmed by the negligible number of shorts, arcs, and delayed pulses. It is worth stressing that an anomalous behavior was found in trial 45, which was repeated at the end of all trials.

#### *4.2. Analysis of Variance: Pulse Type and Process Parameters*

Before proceeding with the statistical analysis of pulse types, the analysis of variance (ANOVA) was carried out to assess the relation between the selected process parameters (F, W and Gap) and the performance indicators, MRR, and TWR. The coefficients of determination *R* 2 , which defines how much variation in the response is explained by the model, were 86.7% for MRR and 37.4% for TWR. These values suggested that the used model described quite properly the relation among F, W and Gap, and MRR, but it was not accurate in relation to TWR. It is worth stressing that the tool wear lengths measured by control touch procedure during micro-EDM milling process were generally very low

(ranging between 3.5–6 µm) and this could have prevented the model from describing the statistical relation properly.

μ

The analysis of variance was applied to the pulse types in order to identify the statistical relation with the process parameters. The coefficients of determination *R* 2 for normal, arc, delayed, and short pulses were 81.5%, 85.7%, 85.2%, and 68.8%, respectively. In the following analysis, short pulses were not considered; since their number was low, they were not statistically relevant, as indicated by the lowest coefficient of determination, and they did not take part as a material removal mechanism.

The first step of the analysis comprised the examination of residuals for normal, arc, and delayed pulses to assess the validity of the ANOVA method. The Pareto charts are reported in Figure 6. For all cases, the Pareto showed normal distribution and homogeneity of variances, thus confirming the validity of the assumptions for the ANOVA method application.

**Figure 6.** Pareto charts for (**a**) normal, (**b**) arcs, and (**c**) delayed pulses.

As evident from the charts, all pulses were influenced by combinations of the chosen process parameters. Only delayed pulses displayed less sensitivity to gap, while it was mainly influenced by W and F. All regression equations obtained from the statistical analysis and related to normal, arc, and delayed pulses are reported in Appendix A.

μ Replica diagrams and main effects plots were computed and reported in Figures 7–9. Figure 7a,b refer to normal pulses; the graphs shows that their number increased when W was equal to 3 µs. In this case, the generator was completely recharged within W, so that the rest of the period t–W was reserved for discharge events. As one can notice, in the considered range of values, parameter F had almost no effects on pulse occurrence, whereas higher gap values (farther distances between tool and workpiece) favored normal pulses.

**Figure 7.** (**a**) Replica diagrams and (**b**) main effects plot related to normal pulses.

**Figure 8.** (**a**) Replica diagrams and (**b**) main effects plot related to arc pulses.

**Figure 9.** (**a**) Replica diagrams and (**b**) main effects plot related to delayed pulses.

Figure 8a,b refer to arcs, where the minimum number of such a pulse type was detected in correspondence of W = 3 µs and Gap = 80, thus confirming that these parameter values could actually provide a stable process. Nonetheless, also in this case, F seemed to have negligible impact on arcs. It is worth noting that normal and arc pulses exhibited a dual response in relation to the process parameter effects, i.e., an increase in normal pulses should correspond a decrease in arcs.

The plots related to delayed pulses are reported in Figure 10a,b.

μ

**Figure 10.** MRR versus (**a**) normal, (**b**) arc, and (**c**) delayed pulse density: blue dots refer to gap = 40, red dots to gap = 60, and green dots to gap = 80.

4 4.5 5 5.5x 10-3 TWR vs. Normal Pulse Density TWR Delayed pulses have different dependences on W, F and Gap compared to the previous pulse types. First, they exhibit a strong linear dependence on W. Moreover, if F is reduced, the discharge probability time t = T–W increases slightly, thus promoting the probability for a delayed to happen. Furthermore, for a higher gap value, the delayed pulse occurrence is certainly reduced. In other words, when the tool and the workpiece are farther from each other, the required energy to allow spark ignition is the highest possible, i.e., set OCV. However, when maximum voltage is applied to the load, it is more likely to trigger normal pulses rather than delayed, which are characterized by lower OCV values. The result is that delayed number decreases. Furthermore, the plots show quite clearly that the delayed–Gap relation is similar to the arcs.

#### *4.3. Relation between Pulse Types and Machining Performance*

3 3.5

2 2.5 3 3.5 4 4.5 5

TWR

2.5 (a) In previous works [15,18], the monitoring, pulse discrimination, and statistical analysis were performed on hardened steel workpieces with similar process parameter settings. In particular, those

2 3 4 5 6 7 8 9 10

Arc Pulse Density (1/s)

x 10-3

(b)

6 6.5 7

7

previous results suggested that a higher number of normal pulses had a positive effect on process stability and machining speed, so that MRR results would be good, while TWR increased its value, thus following a dual trend with respect to MRR. It was also found that as arcs increased in number and occurrence, MRR had a beneficial turn, and TWR worsened. Considering this roadmap, similar results were partly expected in the present analysis, as the role of the ceramic workpiece composition would certainly affect the final results. In order to represent MRR and TWR as a function of pulse type number and occurrence, a normalized pulse density (NPD) was defined according to the following equation: 2 3 4 5 6 7 8 9 10 x 10-3 4 4.5 5 5.5 Arc Pulse Density (1/s) MRR (mm 3/min)(b) 7.5 x 10-4 MRR vs. Delayed Pulse Density

7.5 x 10-4 MRR vs. Arc Pulse Density

$$\text{NPD} = \text{average} \left( \text{Npulse} / \text{T}\_{\text{acq}} \right) \tag{1}$$

where NPulseType is the number of each pulse type (normal, arc, short or delayed) recorded in the last 25 µm of micro-channel manufacturing, and Tacq is the duration of the acquisition window. Figures 10 and 11 depict the corresponding average values of MRR and TWR. The diversity in the dots color indicates the process parameter chosen to discriminate the behavior, in this case, Gap. 5.5 6 6.5 MRR (mm 3/min)

Figure 10a shows that MRR increases as the occurrence of normal pulses increases as well. When Gap is higher, a lower number of overall pulses is expected as the tool and workpiece are farther from each other. Moreover, in this condition, the erosion process was slower thus decreasing MRR. On the contrary, arcs (Figure 10b) did not induce any variability on MRR, while a slight decrease of it was observed when a delayed occurrence was higher (Figure 10c). Reasonably, by considering the nature of delayed pulses, the increase in the machining times could explain this trend. Nonetheless, these results reveal that arcs and delayed did not affect MRR significantly; so, even though these pulse types are capable of removing material, their occurrence was neither problematic nor positive to the process, as long as their percentage was kept smaller than normal pulses. 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 x 10-3 4 4.5 5 Delayed Pulse Density (1/s) (c)

**Figure 11.** *Cont*.

**Figure 11.** TWR versus (**a**) normal, (**b**) arc, and (**c**) delayed pulse density: blue dots refer to gap = 40, red dots to gap = 60, and green dots to gap = 80.

Figure 11 reports TWR behavior in the function of normal, arc, and delayed normalized pulse density.

First, it must be pointed out that the overall recorded TWR values are very small, in particular if these results are compared to those obtained with hardened steel (TWR is one order of magnitude higher for steel, and it ranged between 0.02–0.04 [15,18]). However, the relation between TWR and normal pulses confirms the expected trend, i.e., higher pulse occurrence leads to higher tool consumption. Conversely, a peculiar trend can be observed in Figure 11b: when arcs increase, TWR decrease, and this is in contrast with the expected results. Indeed, arcs are capable of removing material from both workpieces and tools, and an increase of TWR should have actually been measured. It was also verified that the tool wear volume was actually lower. Two reasons may explain these results: the first relies on the energy density of arcs, which was lower than normal pulses, and so less material could have been removed from both electrodes. Furthermore, a protective layer, generated by the melting, decomposition, and evaporation of Si3N<sup>4</sup> particles during the micro-EDM process, was deposited on the tool surface and prevented the tool from more severe wearing. Finally, Figure 11c displays the trend for delayed pulses which is akin to that reported for arcs, thus suggesting a similarity of the nature for such pulses.

#### *4.4. Surface Evaluation of Si3N4–TiN Workpiece before and after the Micro-EDM Process*

The SEM and EDS analyses were also performed to evaluate the surface quality after micro-EDM milling of the Si3N4–TiN ceramic composite. Figure 12 evidences a foamy and porous structure, as also observed in Reference [8] with many spark-induced craters and melt-formation droplets appearing on the composite surface after micro-EDM.

The super-positioning of craters with varying diameters, which is not clearly distinguishable, derives from the material removal mechanism, linked to the interaction between the materials characteristics and chemical reactions triggered by the erosion process. The quantity of material that solidified and adhered to the surface is a function of the composition and microstructure of the starting ceramic materials, whereas the re-solidified droplets are almost entirely due to the TiN particles; conversely, Si3N<sup>4</sup> particles are removed by evaporation. We also observed that, for a length of about 1 mm, the tool tip was covered by a protective layer, typically composed of Si3N<sup>4</sup> particles. The presence of micro-cracks was due to the thermal expansion mismatch between silicon nitride, titanium nitride (3.0·10-6 ◦C-1 and 9.4·10-6 ◦C-1, respectively) and re-solidified particles induced by the plasma breakdown. It can be also noticed that, since the material exhibited a degree of porosity, this could have affected the discharge probability, thus reducing the number of total discharges. For sake of completeness, surface roughness Ra of the machined micro-channels was also measured after micro-EDM processes: the Ra values ranged between 0.77–0.98 µm.

2 2.5 3 3.5 4 4.5 5

TWR

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Delayed Pulse Density (1/s)

5.5 x 10-3 TWR vs. Delayed Pulse Density

x 10-3

(c)

**Figure 12.** SEM micrograph of the EDM surface (**a**) and magnified (**b**).

Elemental composition was checked by EDS and reported in Figure 13, where DTM and micro-EDM surfaces are compared. The element concentration is explicated in Table 3.

From the analysis of the micro-EDM surfaces, the presence of C and W belonging to the tool was detected. Moreover, a marked decrease of N due to the fact of its evaporation during the erosion process was evident with respect to the DTM surface (Table 3). The elemental composition was checked on the micro-EDM surfaces also by point EDS analysis on a merging particle. Besides the presence of C and W, Co was also detected due to the contribution of the tool (Co is commonly used in WC composition), while N was absent. It can be noticed that the Si/Ti atomic ratio was not constant, as it depends on the dimension of the checked area and on the surface modification due to the preferential ablation/oxidation of Si3N<sup>4</sup> or TiN during the micro-EDM process. μ

**Figure 13.** SEM-EDS characterization of Diamond Tool Machining (DTM) (**a**) and micro-EDM (**b**) surfaces.


**Table 3.** EDS element concentration.

#### **5. Discussion**

The experimental results highlighted that the micro-EDM machinability of Si3N4–TiN ceramic composite workpieces differ drastically from experienced practice on metal workpieces. In particular, the statistical analysis of the pulse types in relation to process parameters showed that the triggering of normal, arc, and delayed pulses cannot be simply controlled through the setting of a single process parameter, but rather through a combination of them, such as pulse width W, frequency F and Gap. Moreover, when the pulse type occurrence is put in relation to micro-EDM performance indicators, it is clear that the MRR results were reasonably influenced by normal pulses, while arcs and delayed had a negligible effect on it. On the contrary, TWR exhibited very low values and did not display the common response expected from the different pulse types; in particular, arcs and delayed pulses did not affect TWR negatively.

#### **6. Conclusions**

The micro-EDM machinability of Si3N4–TiN ceramic composite workpieces was investigated by putting in relation pulse-type distribution with process parameters and performance indicators. To this aim, a Design of Experiment was implemented by considering the variation of pulse width W, frequency F and Gap. During the experiments, voltage and discharge current waveforms were acquired: the data were then post-processed to discern pulse type (i.e., normal, arcs, delayed, and shorts) via the pulse discrimination strategy. First, the pulse distribution showed that normal pulses were the most numerous, whereas other pulse types occurred very sporadically. The collected data were then statistically evaluated to elicit the relation existing among process parameter combinations and pulse-type distribution. The statistical analysis proved statistical reliability only for normal, arc, and delayed pulses, whereas short pulses were neglected in the discussion due to the numerical and statistical irrelevance. Pareto charts, regression equations, replica diagrams, and main effects plots revealed that all pulses were influenced by combinations of all selected process parameters.

The final analysis of the influence between pulse type and process performance indicators (MRR and TWR) showed that:


Finally, it was shown that the Si3N4–TiN workpiece features intrinsic degrees of electrical-conductivity inhomogeneity before machining as assessed by SEM and EDS evaluation. Additionally, same analyses performed on the micro-EDM surface displayed re-solidified droplets and micro-cracks, induced by chemical mechanism, which witnessed a surface roughness increase with

respect to un-machined surface. Indeed, the surface roughness Ra measured after micro-EDM ranges between 0.77–0.98 µm, whereas the starting Ra of the workpiece ranges between 0.16 and 0.30 µm.

**Author Contributions:** Conceptualization, V.M. (Valeria Marrocco) and F.M.; methodology, V.M. (Valeria Marrocco), F.M. and V.B.; software, V.M. (Valeria Marrocco); validation, V.M. (Valeria Marrocco), F.M., V.B. and V.M. (Valentina Medri); formal analysis, V.M. (Valeria Marrocco) and V.B.; investigation, V.M. (Valeria Marrocco) and V.M. (Valentina Medri); data curation, V.M. (Valeria Marrocco), V.B. and V.M. (Valentina Medri); writing—original draft preparation, V.M. (Valeria Marrocco); writing—review and editing, V.M. (Valeria Marrocco) and V.M. (Valentina Medri); supervision, I.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

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

#### **Appendix A**

Since two of the considered factors (W and Gap) had three levels, a general full factorial design was planned for the experiments. The corresponding regression equations obtained by the statistical analyses for normal, arcs, and delayed pulses are reported below:

W = Width, G = Gap, F = Frequency

Normal = 87.379 − 1292 W\_1 + 1667 W\_3 − 0.375 W\_5 − 0.042 F\_120 + 0.042 F\_160 − 0.198 G\_40 − 0.305 G\_60 + 0.503 G\_80 − 1.455 W × F\_1 120 + 1.455 W × F\_1 160 + 0.677 W × F\_3 120 − 0.677 W × F\_3 160 + 0.779 W × F\_5 120 − 0.779 W × F\_5 160 + 0.495 W × G\_1 40 + 0.055 W × G\_1 60 − 0.550 W × G\_1 80 + 0.141 W × G\_3 40 − 0.015 W × G\_3 60 − 0.126 W × G\_3 80 − 0.636 W × G\_5 40 − 0.039 W × G\_5 60 + 0.676 W × G\_5 80 − 0.968 F × G\_120 40 + 0.840 F × G\_120 60 + 0.128 F × G\_120 80 + 0.968 F × G\_160 40 − 0.840 F × G\_160 60 − 0.128 F × G\_160 80 − 0.858 W × F × G\_1 120 40 + 0.761 W × F × G\_1 120 60 + 0.097 W × F × G\_1 120 80 + 0.858 W × F × G\_1 160 40 − 0.761 W × F × G\_1 160 60 − 0.097 W × F × G\_1 160 80 − 0.203 W × F × G\_3 120 40 − 0.238 W × F × G\_3 120 60 + 0.441 W × F × G\_3 120 80 + 0.203 W × F × G\_3 160 40 + 0.238 W × F × G\_3 160 60 − 0.441 W × F × G\_3 160 80 + 1.061 W × F × G\_5 120 40 − 0.523 W × F × G\_5 120 60 − 0.538 W × F × G\_5 120 80 − 1.061 W × F × G\_5 160 40 + 0.523 W × F × G\_5 160 60 + 0.538 W × F × G\_5 160 80 (A1)

Arc = 9.963 + 0.848 W\_1 − 1.643 W\_3 + 0.795 W\_5 + 0.051 F\_120 − 0.051 F\_160 + 0.136 G\_40 + 0.181 G\_60 − 0.318 G\_80 + 1.169 W × F\_1 120 − 1.169 W × F\_1 160 − 0.580 W × F\_3 120 + 0.580 W × F\_3 160 − 0.589 W × F\_5 120 + 0.589 W × F\_5 160 − 0.298 W × G\_1 40 − 0.170 W × G\_1 60 + 0.468 W × G\_1 80 − 0.141 W × G\_3 40 + 0.022 W × G\_3 60 + 0.120 W × G\_3 80 + 0.439 W × G\_5 40 + 0.148 W × G\_5 60 − 0.587 W × G\_5 80 + 0.721 F × G\_120 40 − 0.632 F × G\_120 60 − 0.089 F × G\_120 80 − 0.721 F × G\_160 40 + 0.632 F × G\_160 60 + 0.089 F × G\_160 80 + 0.668 W × F × G\_1 120 40 − 0.632 W × F × G\_1 120 60 − 0.036 W × F × G\_1 120 80 − 0.668 W × F × G\_1 160 40 + 0.632 W × F × G\_1 160 60 + 0.036 W × F × G\_1 160 80 + 0.016 W × F × G\_3 120 40 + 0.274 W × F × G\_3 120 60 − 0.290 W × F × G\_3 120 80 − 0.016 W × F × G\_3 160 40 − 0.274 W × F × G\_3 160 60 + 0.290 W × F × G\_3 160 80 − 0.684 W × F × G\_5 120 40 + 0.358 W × F × G\_5 120 60 + 0.326 W × F × G\_5 120 80 + 0.684 W × F × G\_5 160 40 − 0.358 W × F × G\_5 160 60 − 0.326 W × F × G\_5 160 80 (A2)

Delayed = 1.7033 + 0.4235 W\_1 − 0.0001 W\_3 − 0.4234 W\_5 + 0.0827 F\_120 − 0.0827 F\_160 + 0.0260 G\_40 + 0.0799 G\_60 − 0.1059 G\_80 + 0.1428 W × F\_1 120 − 0.1428 W × F\_1 160 − 0.0747 W × F\_3 120 + 0.0747 W × F\_3 160 − 0.0681 W × F\_5 120 + 0.0681 W × F\_5 160 − 0.0924 W × G\_1 40 +0.0834 W × G\_1 60 + 0.0090 W × G\_1 80 + 0.0165 W × G\_3 40 + 0.0237 W × G\_3 60 − 0.0403 W × G\_3 80 + 0.0759 W × G\_5 40 − 0.1071 W × G\_5 60 + 0.0312 W × G\_5 80 + 0.1045 F × G\_120 40 − 0.0603 F × G\_120 60 − 0.0441 F × G\_120 80 − 0.1045 F × G\_160 40 + 0.0603 F × G\_160 60 + 0.0441 F × G\_160 80 + 0.0708 W × F × G\_1 120 40 − 0.0340 W × F × G\_1 120 60 − 0.0368 W × F × G\_1 120 80 − 0.0708 W × F × G\_1 160 40 + 0.0340 W × F × G\_1 160 60 + 0.0368 W × F × G\_1 160 80 + 0.1106 W × F × G\_3 120 40 − 0.0560 W × F × G\_3 120 60 − 0.0545 W × F × G\_3 120 80 − 0.1106 W × F × G\_3 160 40 + 0.0560 W × F × G\_3 160 60 + 0.0545 W × F × G\_3 160 80 − 0.1814 W × F × G\_5 120 40 + 0.0900 W × F × G\_5 120 60 + 0.0913 W × F × G\_5 120 80 + 0.1814 W × F × G\_5 160 40 − 0.0900 W × F × G\_5 160 60 − 0.0913 W × F × G\_5 160 80. (A3)

#### **References**


© 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*
