Sustainability metrics targeted optimization and electric discharge process modelling by neural networks

Abstract

Aluminium and its alloys, especially Al6061, have gathered significant interest among researchers due to its less density, great durability, and high strength. Due to their lightweight properties, the precise machining of these alloys can become expensive through conventional machining operations for intricate products. Therefore, non-traditional machining such as electric discharge machining (EDM) can potentially be opted for the cutting of Al6061. EDM is often criticized due to its low machining rates, therefore, in the current work,  cryogenic treatment (CT) has been performed on the brass electrode to evaluate the improvement in the machining rates. In addition, kerosene oil (KO) has been engaged in traditional EDM which is replaced with the deionized water (DI) based dielectric as a sustainable alternative. The machining variables such as spark voltage (S_V), pulse-on-time (P_ON), peak current (I_P), and Al₂O₃ powder concentration (C_P) have been chosen to determine the material removal rate (MRR), surface roughness (SR), and specific energy consumption (SEC) while comparing non-treated (NT), and cryogenically treated (CT) brass electrodes during EDM. The results were analyzed through optical micrographs, scanning electron microscopy (SEM) analysis, energy dispersive x-ray (EDX) examination, and 3D surface plots. An artificial neural network (ANN) has been constructed for the better prediction of output responses. Moreover, multi-response optimization through the non-dominated sorting genetic algorithm (NSGA-II) has also been performed. The magnitudes of MRR_CT, SR_CT, and SEC_CT obtained by multi-response optimization are 64.82%, 27.45%, and 46.60% are better than the values obtained by un-optimized settings of CT brass electrodes. However, the optimal magnitudes of processing parameters are I_P = 24.85 A, S_V = 2.18 V, P_ON = 119.11 µs, and C_P = 1.05 g/100 ml.

Introduction

Several industries, including the aviation, marine, automotive, and healthcare sectors, have a strong preference for aluminium alloys, particularly Al6061. This particular aluminium alloy’s exceptional qualities—such as its lightweight, amazing pliability, superior electrical and thermal conductivity, extraordinary ability to be molded, and corrosion resistance—have led to its widespread adoption^1,2. However, conventional machining techniques are difficult to apply to this alloy due to its propensity for chip adherence and edge distortion. This could therefore result in unfavourable effects including high EWR (expensive tooling), less surface finish, and more OC^3,4. SR evaluates the surface’s quality, EWR show the electrodes that are wearing over time, and overcutting guarantees correct dimensional accuracy. Electrically conductive materials are cut using a stochastic method EDM die sinking which does not have mechanical interactions of tooling with the workpiece^5,6,7. In addition to its mechanical qualities, EDM shows promise in producing complex geometries with accurate tolerances and reliable repeatability^8,9,10,11. Die sinking EDM’s basic idea is to remove material from the workpart by creating a series of electric discharges between the tool, which is the cathode, and the specimen, which is the anode^6,12,13. A power supply is what first generates high-voltage pulses. The dielectric fluid deteriorates and sparks between the conducting surfaces when the voltage is beyond a particular threshold, vaporizing minute amounts of material from the workpiece and the electrode^14,15. The dielectric fluid then solidifies these removed particles over the machined cavity. Traditionally, KO is considered as a dielectric medium for a long time, even though it releases dangerous gases, vapours, and fumes that are hurtful to the environment and operator health^16,17. This is explained by the intrinsic qualities of KO, which include low viscosity, high flammability, increased toxicity, and non-biodegradability¹⁸. Several research works have examined the negative consequences linked to the use of paraffin oil. For instance, Ming et al¹⁹. discovered that aerosol by-products and hazardous vapours are released during EDM using KO, affecting the health of the operator and the purity of the air. Hydrocarbon-based dielectrics, such as KO, are known to release hazardous materials such as benzene, dust, aerosols, and volatile particles (Tonshoff et al²⁰).

EDM is criticized because of its limitations, such as less MRR, high EWR, poor surface quality, and more OC^21,22. To fix this, scholars recommend adding powders to dielectric oil^23,24. Nano-powder in the dielectric medium boosts the reactions above and enhances the values of the abovementioned response measures. Nano-additives in dielectric media generate many ionized particles at high voltage, and acceleration and random movement of particles enhance work-electrode distance and electric discharge dispersion. This increased distance causes the current to bridge, stabilizing and speeding up machining^25,26. Researchers also suggest adding surfactant to fix this issue²⁷. Previous research has demonstrated that nano-powder improves EDM’s efficacy. Chaudhari et al.²⁸ improved the response measures for Nitinol shape memory alloy using Al₂O₃powder and EDM fluid. Hosni and Lajis²⁹ studied the surfactant (Span-20) and chromium (Cr) content that affected AISI D2 steel during EDM, emphasizing their importance in improving output characteristics. Abbas et al³⁰. evaluated surface roughness achieved on AISI D2 steel during electric discharge machining process.  A titanium alloy machining through EDM using graphite powder and surfactant was presented by Kolli and Kumar³¹. Researchers argued that surfactant (Span-20) reduced erratic sparks and particle aggregation in the dielectric oil, improving response measures. Paswan et al³². investigated the DI and graphene nano-powder affected Inconel 718 EDM machining performance. The authors also used waveform analysis to study spark phenomena, and employed RSM for optimizing the machine variable as per given factors. Significant improvements in MRR, EWR, and SR values were 20.1%, 2%, and 14%, respectively. In addition, the findings of said analysis showed that discharges in nano-graphene mixed dielectric were more stable than standard EDM.

Cryogenic Treatment (CT), a key material microstructure improvement method, is widely used to improve electro-mechanical characteristics, including durability, grain size, thermal, and electrical conductivities^33,34. Choudhary et al³⁵. applied Taguchi’s design utilizing DCT and NT copper electrodes for EDM on Hastelloy C-4. They examined gap voltage, spark duration, and current in relation to SR and EWR. Compared to NT electrodes, CT electrodes demonstrated reduced EWR and improved SR. Papazoglou et al⁶. evaluated EDM cutting performance in a dielectric medium for Ti-alloy grade 2. Spark-on duration affected EWR during EDM machining, but I_Pprimarily affected MRR. Kumar and Sharma³⁶ employed EN-31 magnetic field-assisted EDM to assess MRR, EWR, and OC. Input parametric settings such as P_ON (10 µs), P_OFF (10 µs), I_P(3.18 A), and magnetic field (0.3 T) yielded maximum MRR and minimal EWR and OC. Meena et al³⁷. optimized multi-response EDM of commercially pure Ti-alloy using GRA. They assessed MRR, EWR, and OC based on current, frequency, and pulse width. For response measures like MRR, EWR, and OC, current is most critical. Ramaswamy and Perumal³⁸ manufactured a hybrid composite LM13 Al-alloy and carried out machining.  Pulse current, P_ON, and P_OFF were adjusted to evaluate MRR, EWR, and OC. The investigation showed that current and P_ON were most relevant for MRR, EWR, and OC.

The impact of machining Al6061 alloy using NT and CT brass tools, combined with Span-20 and an alumina mixture in DI, is not evaluated. In the present study, four processing parameters such as I_P, S_V, P_ON, and C_P have been used for the evaluation of response measures such as MRR_NT, SR_NT, SEC_NT, MRR_CT, SR_CT, and SEC_CT during the EDM process of Al6061. To further evaluate the process, EDX, SEM, 3D profilometry, and microscopic investigations have all been carried out. Analysis of Variance (ANOVA) has been conducted to recognize the critical input factors that have a direct impact on MRR, SR, and SEC for both types of electrode materials. To achieve the sustainable EDM, carbon footprints have been evaluated in terms of % reduction of CO₂ for DI water based dielectric compared to the kerosene oil based dielectric. Moreover, an ANN has been built for the better prediction of output responses and evaluate the correlation among the predicted and the experimental values. Multi-response optimization using the NSGA-II has also been used. To further highlight the novelty of work, the details are as follows:

• There is no comprehensive work available which explains discrepancy of cutting performance of aerospace alloy considering the process’s physical features.

• The cutting performance of aerospace alloy is not thoroughly optimized for high-speed machining (when compared to conventional process) which advances knowledge towards the key limitation of electric discharge machining systems.

• The use of design of experiments, artificial intelligence and evolutionary intelligence based multi-objective optimization has not been used to explore the novel process characteristics. Considering the multi-physics nature of process involving thermal, electrical, surface melting and solidification characteristics, deionized water mixed with nano-powders and surfactant, the conventional simulations are limited to present a thorough comparison of cutting performance of the above-mentioned aeronautical alloy.

Experimentation characteristics

Materials and procedures

Al6061 alloy of size (60 mm × 60 mm × 5 mm) has been chosen for assessing the machining performance through EDM setup. The measurement of MRR, SR, and SEC was used to evaluate the cutting performance of Al6061. The dielectric fluid, which was housed in a specially made container, was DI. Table 1 presents the elemental composition of Al6061 as determined by spectrometry, and Table 2lists its physical properties as per manufacturer’s quality sheet (also used in³⁹).

Table 1 Elemental percentage of Al6061.

Table 2 The electrical and physical characteristics of Al6061.

An Al6061 alloy was cut utilizing a brass electrode with a 9 mm diameter under both extensive CT and NT conditions to assess cutting efficiency. Specific electro-physical characteristics of the brass tool are given in Table 3⁴⁰. The brass electrode underwent a 24-hour deep cryogenic treatment in liquid nitrogen, gradually reaching a state of -184 °C about 1 °C per minute. This procedure was performed in a cryogenic compartment. The tool was tempered at 150 °C at the same cooling rate (1 °C/min) following the cooling phase until the electrode eventually attained room temperature. The conductivity increased by 10.70% and the grain size of the tool enhanced by 7.89% after deep CT⁴¹. Three different concentrations (0.5, 1, and 1.5 g/100 ml) of alumina (Al₂O₃) were blended to the DI after initial evaluations. A literature search was also carried out in an effort to lessen the restrictions given by low cutting rates in EDM. The physico-chemical attributes of alumina nanopowder (as per manufacturer) are listed in Table 5⁴², whereas the levels of each variable employed in this research are listed in Table 4. Throughout the EDM process, Span-20 was injected at a consistent concentration of 6% by volume to prevent agglomeration.

Table 3 The bodily characteristics of the brass tool.

Table 4 Varieties of parameter input.

Table 5 Alumina possesses electrical and physical characteristics.

The experimental framework adopted in this investigation embraced a Response Surface Methodology (RSM) strategy, specifically the Central Composite Design (CCD). Experiments were conducted utilizing the EDM apparatus,, illustrated in Fig. 1. MRR, quantified in mm³/min, was ascertained by evaluating the variance in weight before and after the EDM operation, and computed using Eq. 1. Conversely, surface roughness was assessed using the 2D roughness meter which slides over the profile in linear fashion recording the movement of pin in the peaks and valleys, and the SEC was assessed using commonly established process energy metric via Eq. 2. To process microscopic images for 3D surface topography investigation, Gwyddion (version 2.63), is utilized.

$$\:MRR=\frac{WB-WA}{tm}$$

(1)

$$\:SEC=\:\frac{Vs\:\times\:\:Is\times\:\:Pon}{{V}_{g}}\times\:0.0027$$

(2)

Within Eq. 1, ‘tm’ denotes the amount of time spent EDM, while ‘W_B’ and ‘W_A’ stand for the workpiece’s weight prior to and following the machining operation, respectively.

Fig. 1

Working of EDM.

Fig. 2

Working schematic of EDM¹⁷ (Licensed Open Access).

A graphic explaining the idea of spark production is shown in Fig. 2. Preliminary experiments were operated to identify the best parameter configurations for maximizing MRR and minimizing SR and SEC. After these first trials, parameters showing increased MRR, and decreased SR and SEC were chosen for follow-up workpiece studies with both type of electrodes. The use of these absolute variables reduced the possibility of unnecessary processing costs or damage to tools and workpieces during EDM operations, in addition to resulting in high MRR and decreased SR and SEC. Determining the proper concentration of surfactant in the DI dielectric was a crucial step in the procedure. The final decision was made based on preliminary experimental testing that met predetermined selection criteria, however, an initial reference was taken from the body of existing literature. Preliminary results show that 6% of surfactant concentration complies with the specifications. To make it easier to combine the surfactant with the DI dielectric, a motorized stirrer tank was created specifically for that purpose. Figure 3 outlines the thorough methods used in this investigation.

Fig. 3

Research methodology of this study.

Analytical computation

An ANN, also known as a “multilayer perceptron,” is a great tool for tying response measures and their predictions to process parameters. Even in systems with inadequate characterizations, the technique mines the complex curvy and quadratic relations in the different input intervals with efficacy⁴³. In the current study, three output variables : material removal rate, surface roughness and specific energy consumption (MRR_NT, SR_NT, SEC_NT, MRR_CT, SR_CT, SEC_CT) have been investigated using NT and CT brass electrodes. The ANN is used to predict the values of these response measures. When it comes to highly non-linear tasks that depend on bias and weight values, an ANN performs similarly to the human brain. The response measures in this work are modelled using the EBPTA⁴⁴. Figure 4 displays the ANN structure used in this investigation.

Fig. 4

The functioning schematic of ANN.

The basic ANN structure has three layers: input, hidden, and output. Figure 4 shows that this study uses three ANN layers. Since the research has been conducted using four process parameters (I_P, S_V, P_ON, and C_P), these parameters comprise the opening layer, or input layer. Input variables typically range from 1x to 2.5x neurons^45,46. Consequently, the learning rate (E^-4 to E^-1) is compared to a maximal neuron combination set. The hidden layer in this investigation is represented by the 3–10 neurons in the second/middle layer. The performance of ANN models is enhanced with increased model complexity. More neurons in the hidden layer improve the model’s performance and curve spots. Generalization of a model may be constrained by its complex design. The lowest number within a range might lead to data fit issues and mistakes, just like the maximum number of neurons. Thus, the optimal number of neurons is selected with care. For NT and CT brass electrodes, the third and final output layer is made up of six response measurements (MRR, SR, SEC). Nevertheless, only 10% of the data—80% of which was used for training—was utilized for testing and validating the ANN model. The R² and RMSE formulas are:

$$\:{R}^{2}=1-\:\frac{{\sum\:_{i}^{N}({y}_{i}-\:{\widehat{y}}_{i})}^{2}}{{\sum\:_{i}^{N}({y}_{i}-\:{\stackrel{-}{y}}_{i})}^{2}}$$

(3)

$$\:RMSE=\sqrt{\frac{1}{N}\sum\:_{i=1}^{N}{\left({\widehat{y}}_{i}-{y}_{i}\right)}^{2}}$$

(4)

Here, $\:{y}_{i}$ represents the original input value, $\:{\widehat{y}}_{i}$ the projected value, and $\:{\stackrel{-}{y}}_{i}$ the average, with $\:i$ = 1, 2, 3…, and N the total number of observations. $\:{R}^{2}$ measures accuracy and ranges from 0 to 1.

Results and discussion

The design of experiments for the current study has been compiled in Table 6. The whole discussion and key findings are elaborated with the help of microscopic, SEM images, and EDX profiles. For statistical analysis, ANOVA was also carried out to check which parameter(s) have significance.

Table 6 The design of experiments for response extraction.

Parametric effect analysis

In order to predict the impacts of both type of brass electrode (NT and CT) on the responses obtained during EDM of Al-alloy, parametric effects were examined in this study. Three-dimensional surface plots against the four design factors (I_P, S_V, C_P, and P_ON) were created for this purpose. It is important to note that two variables are shown to have an influence within their designated ranges in the 3D surface plots, while the other factors are kept at their median values.

Process productivity in terms of material removal rate

In the EDM of Al6061, the MRR was compared to various input variables using NT and CT brass electrodes. Figure 5(a) demonstrates the combined influence of I_P and S_V on the MRR for the NT brass tool. Figure 5(a) illustrates the removal of material from the specimen as the values of both variables rise. The primary cause of this is the strong discharge energy at high I_P and S_V values, which causes more material from the specimen to melt and evaporate. As shown in Fig. 5(b), a similar result is also visible when employing the CT brass electrode for Al6061 machining using EDM setup. The CT brass tool’s superior atomic packing and grain refinement led to better wear characteristics, the MRR obtained with it is better than achieved with the NT brass tool. It is further evident from Fig. 6(a) that 33.3% and 90.4% improvements have been predicted in the value of MRR when CT electrode is engaged as compared with the NT electrode at higher state (25 A, 6 V) and lower state (5 A, 2 V), respectively.

The influence of I_P and P_ON on the MRR with NT brass was also assessed for Al6061 alloy, as presented in Fig. 5(c). The surface plot highlights that these factors have a direct impact on the MRR. It means that the MRR raises with the increase of both I_P and P_ON collectively. The reason lies in the formation of spark density in the plasma channel for a longer period, causing the removal of significant material from the workpiece. This material removal is further enhanced when the CT brass electrode is employed due to its improved mechanical and metallurgical properties, as depicted in Fig. 5(d). Hence, the CT brass electrode can be an alternative choice for achieving exceptional MRR values when considering the EDM of Al6061. According to Fig. 6(b), the bar graph shows an increase in MRR of about 2 times at 150µs, 25 A, and 3.2 times at 50µs, 5 A when comparing the CT electrode to the NT electrode.

Figures 5(e) and 5(f) represent the effect of C_P and I_P on the MRR for NT brass and CT brass, respectively. As long as discussed for the NT brass electrode, the effect of said parameters on MRR shows an increasing trend, and then at higher levels of both parameters, the curve slightly decreases. It specifies that MRR upsurges with the regular change of I_P, but depreciation can be noticed in the case of C_P after 1 g/100 ml. This is attributed to the generation of agglomerated particles in the dielectric, which reduces the thermal energy over the workpiece, resulting in a drop in the MRR. It is interesting to note that the same trend appears for the CT brass electrode, as shown in Fig. 5(f). However, the value of MRR even at maximum parametric levels is greater than the value found with the NT brass tool, as far as EDM is operated for the cutting of Al6061. This is mainly due to the enhanced surface properties of the CT brass electrode, which generate discharge more uniformly, resulting in the accomplishment of a high MRR. In addition, MRR has significantly improved by 80.8% at the lower state (0.5 g/100 ml, 5 A) and 43.7% at the higher state (1.5 g/100 ml, 25 A) when comparing the CT electrode to the NT electrode, as witnessed in Fig. 6(c).

Fig. 5

Visualizing MRR across varying processing factors with (a, c, e) NT brass tool and (b, d, f) CT brass tool through 3D surface plots.

Fig. 6

Bar graphs represent the effect of different parametric combinations on MRR for NT and CT, such as (a) 5 A, 2 V and 25 A, 6 V, (b) 50 µs, 5 A and 150 µs, 25 A, (c) 0.5 g/100 ml, 5 A and 1.5 g/100 ml, 25 A.

Part quality in terms of surface roughness

Surface roughness (SR) is another crucial response variable illustrated herein against the given parameters during the EDM of Al6061, as depicted in Fig. 7. Initially, for the NT brass tool, the simultaneous impact of I_P and S_V on SR is evaluated in a 3D surface plot, as shown in Fig. 7(a). It can be noticed that SR is compromised when either I_P or S_V increases. The intense heat input available in the plasma at high current and voltage values is the key reason for the machined impression becoming rough. Similarly, for the CT brass electrode, the same trend is noticed. However, for comparison purposes, the CT brass electrode in the EDM of Al6061 provided less SR compared to the NT brass, as evidenced in Fig. 7(b). This is due to the regular discharge phenomenon even at large magnitude of variables, thereby improving surface quality. Mathematical point of view, SR is reduced by 11.49% at 5 A, 2 V and 25.8% at 25 A, 6 V when machining results of CT is compared with the findings of NT electrode, such as displayed in Fig. 8(a).

Figure 7(c) demonstrates the examination of SR by changing the states of I_P and P_ON employing the NT brass tool while EDM of given alloy. It can be highlighted from Fig. 7(c) that SR initially increases with the rise of P_ON, then it slightly decreases when P_ON is shifted to 150 µsec, even at all I_Plevels. The initial increase in SR is due to prolonged sparking occurring over the target surface. After that, a slight decrease in SR is observed because of the presence of a large number of ionized particles, which hinder discharges and lead to a better surface finish⁴⁷. As far as I_P is taken into account, the value of SR continuously increases with the change of I_P, as depicted in Fig. 7(c). On the contrary, for the CT brass electrode, SR increases with the increase of both I_P and P_ON values, as shown in Fig. 7(d). However, SR attained with CT brass is better than that achieved with NT brass due to its better grain structure and prevention of irregular discharges. It is highlighted from Fig. 8(b) that surface finish is enhanced by 11.9% and 33.6% at lower setting (50 µs, 5 A) and higher setting (150 µs, 25 A), respectively, as long as compared the cutting performance of CT with the NT electrode.

Figure 7(d) also demonstrates the impact of two input factors (I_P and C_P) on the SR in the EDM of Al6061. For both parameters, SR increases with their values. The reason behind this lies in the intensification (due to current) and spreading of sparks (due to C_P causing agglomeration) at higher values, which alter the surface micro-impressions and result in a high SR value. However, the 3D surface plot shows different behavior when C_P is set to 1.5 g/100 ml and I_P changes to 5–25 A. It means that surface finish decreases when I_P changes from 5 A to 10 A, then improves again when I_P is set at 15 A. Conversely, it has been observed that SR is tremendously improved at higher levels of both I_P and C_P when the CT brass is employed for the EDM of selected alloy, owing to the high grain refinement. The bar graph in Fig. 8(c) displays the comparative assessment, showing 17.8% and 33.6% reduction in SR at corresponding lower (0.5 g/100 ml, 5 A) and higher (1.5 g/100 ml, 25 A) combinations, when CT electrode is treated in EDM process.

Fig. 7

Visualizing SR across varying processing factors with (a, c, e) NT brass tool and (b, d, f) CT brass tool through 3D surface plots.

Fig. 8

Bar graphs portray the influence of varied parametric conditions on SR for NT and CT, such as (a) 5 A, 2 V and 25 A, 6 V, (b) 50 µs, 5 A and 150 µs, 25 A, (c) 0.5 g/100ml, 5 A and 1.5 g/100ml, 25 A.

Environmental burden in terms of specific energy consumption

Specific energy consumption (SEC) is thoroughly investigated against machining parameters employing NT brass and CT brass as electrodes, as depicted in Fig. 9. Primarily, the combined effect of S_V and I_P on SEC for NT brass is explained in Fig. 9(a). According to Fig. 9(a), SEC decreases with the increase in I_P. However, SEC slightly deviates downward with the increase of spark voltage. For the combined influence, the magnitude of SEC sharply lowers at large values of both input parameters. This mainly occurs because of the high exclusion of material from the specimen, resulting in the decline of SEC. For the CT brass electrode, the behavior of SEC against the same considered variables is presented in Fig. 9(b). SEC reduces with the increase of I_P value, whereas SEC initially decreases and then increases when S_V rises to its maximum value. However, their combined interaction yields a low SEC magnitude. For a comparative viewpoint, the SEC from Fig. 9(b) (CT brass electrode) is lower than the value noticed in Fig. 9(a) (NT brass tool) when EDM of Al6061 is taken into account. From Fig. 10(a), it can be predicted that SEC is improved by 27.2% and 18.6% at their respective states, such as lower (5 A, 2 V) and higher (25 A, 6 V), when CT is used during EDM of Al6061.

Figure 9(c) is developed for studying the effect of P_ON and I_P on the SEC when NT brass is utilized in the EDM of the Al6061 process. It can be clearly seen from Fig. 9(c) that the value of SEC decreases with the increase of both aforementioned parameters, even the combined effect of the two parameters also lowers the magnitude of SEC. This is mainly because of proper heat distribution over the machined cavity due to the high-temperature gradient and appropriate gap of the plasma channel. The same behavior can also be noticed for the CT brass electrode in Fig. 9(d), as long as the EDM of Al6061 is taken into account. However, the value of SEC with the CT brass is far less than the value achieved with the NT brass electrode. The primary reason for this shift is the appropriate grain size and structure of CT brass, which effectively utilizes the heat input in the plasma space. Hence, a higher magnitude of input parameters is ideal for getting low SEC while EDM of Al6061. The bar graph in Fig. 10(b) also shows the significance of CT over NT electrode. Mathematically, CT electrode has provided 14.5% and 16.16% lower value of SEC at their respective conditions, such as lower (50 µs, 5 A) and higher (150 µs, 25 A), when comparison is developed with NT brass tool.

The combined effect of C_P and I_P on SEC against NT and CT brass tools is shown in Fig. 9(e) and 9(f), respectively. A similar trend of SEC can be observed in the 3D surface plots. For instance, in Fig. 9(e), the value of SEC decreases with increasing I_P but remains almost unchanged when C_P increases to its maximum level. It is well-known that an increase in powder concentration results in agglomerated material, which blocks discharges and leads to inappropriate machining, thereby increasing SEC. However, an increase in I_P also intensifies the spark phenomenon, which increases the MRR, consequently reducing SEC. The same trend can be seen for CT brass during EDM of Al6061 alloy. From the comparison, the values of SEC are higher in the case of NT brass than that of CT. Thus, CT brass is preferred over NT brass electrodes to get low SEC. Figure 10(c) indicates that the CT electrode is more effective during the EDM of Al6061, providing lower values of SEC compared to the NT electrode, with reductions of 28.79% at C_P: 0.5 g/100 ml, I_P: 5 A, and 22.79% at C_P: 1.5 g/100 ml, I_P: 25 A.

Fig. 9

Visualizing SEC across varying processing variables with (a, c, e) NT brass tool and (b, d, f) CT brass tool through 3D surface plots.

Fig. 10

Bar plots highlight the change of parametric combinations on SEC for NT and CT, such as (a) 5 A, 2 V and 25 A, 6 V, (b) 50 µs, 5 A and 150 µs, 25 A, (c) 0.5 g/100 ml, 5 A and 1.5 g/100 ml, 25 A.

Topographical analysis

Surface topography analysis has also been performed on the EDMed surfaces of Al6061 against both types of brass electrodes. The impact of changing parametric levels on responses have been investigated by looking at the surface quality of the cut profiles using optical microscopic, SEM, and 3D roughness plots.

The EDMed surface profile using NT brass and CT brass electrodes has been depicted in the form of microscopy and 3D profilometry analysis as shown in Figs. 11(a, b) and 11(c, d), respectively. The NT brass electrode provided a lower MRR due to uneven and non-uniform sparking, whereas the CT brass electrode yielded a higher MRR due to improved grain sizes, as shown by the optimal microscopic image in Fig. 11(c, d). For further analysis, SEM images were also collected with both types of electrodes, as represented in Fig. 12. According to Fig. 11(a, b), and Fig. 12(a), high redeposited melt is found on the machined profile. This is because the NT brass did not properly remove the debris around the machined cavity, which was governed by the uneven sparking and poor flushing. In addition, ionization of dielectric also takes place during sparking, which generates multitude carbon particles, depositing over the machined surface during pulse-off time. The 3D surface topography analysis also revealed the material erosion areas by peaks and valleys. Peaks are highlighted by blue colour and valleys are indicated by red colour. The 3D surface topography analysis (corresponding to Fig. 11(a, b)) has revealed the small heightened peaks and valleys which is the indication of low material erosion for NT brass electrode. An EDX image shown in Fig. 13 highlights the presence of carbon atoms in the workpiece’s surface. Conversely, in Fig. 12(b), less redeposited melt is found with small-size craters, which is an indication of a good flushing phenomenon occurring during machining due to the high grain structure of the CT brass electrode. Moreover, the 3D surface topography analysis for CT brass electrode as shown in the correspondence of Fig. 11(c, d) revealed that deeper, wider, and larger craters have been formed on the surface of Al6061, with smaller amount of melt redeposited.

Fig. 11

Microstructural surface texture analysis with (a, b) NT brass tool and (c, d) CT brass tool.

Fig. 12

SEM images showing surface micro-texture with (a) NT brass tool: (b) CT brass tool.

Fig. 13

EDX plot of machined surface.

In Fig. 14(a), shallow craters are evident on the machined surface when the NT brass electrode is used, at low level of process parameters. Moreover, the 3D surface profile depicts a higher blue portion, indicating increased SR. This is mainly due to arcing phenomena and improper flushing effects in the work-electrode zone. Consequently, the surface finish is compromised. The SEM image represented in Fig. 15(a) also indicates the presence of shallow craters with greater redeposited melt. However, when the values of control parameters are changed from small to high levels, no significant change in the magnitude of SR is detected. However, variation in the machined surface topology can be observed in Figs. 14(a) and (b). In both figures, tiny craters with little resolidified melt are found on the machined profile. This occurs due to the widening of the plasma zone, which improves the stability of sparking and reduces the arcing phenomenon, resulting in a somewhat improved surface finish (Fig. 15(b)). The peaks and valleys formed on the surface of Al6061 at the low processing parameters indicating the smaller material erosion, whereas the peaks and valleys formation on the surface of the Al6061 illustrates the larger amount of material erosion during the EDM operation of Al6061.

Fig. 14

Microstructural surface texture analysis at (a, b) I_P = 5 A, S_V = 2 V, P_ON = 50 µs, C_P = 0.5 g/100 ml; (c, d) I_P = 25 A, S_V = 6 V, P_ON = 150 µs, C_P = 1.5 g/100 ml.

Fig. 15

SEM of machined specimen for SR_NT at (a) I_P = 5 A, S_V = 2 V, P_ON = 50 µs, C_P = 0.5 g/100 ml; (b) I_P = 25 A, S_V = 6 V, P_ON = 150 µs, C_P = 1.5 g/100 ml.

The micrograph depicted in Fig. 16(a) illustrates shallow and greater redeposited melt on the machined cavity using CT brass electrode for cutting Al6061. This micrograph was captured at low levels of input parameters. Furthermore, the 3D plot exhibits a low peak-to-valley distance, indicating a low SR value. This occurs due to the low discharge density, which partially melts the material from the workpiece. The SEM image highlighted in Fig. 17(a) also reveals the presence of shallow-sized craters. Conversely, Fig. 16(b) displays tiny craters on the machined profile when the process parameters are elevated, while still utilizing the CT brass electrode. Additionally, the surface topography in Fig. 16(b) shows a high peak-to-valley distance, indicating a high SR compared to the one depicted in Fig. 17(a). The generation of high peaks and valleys is primarily caused by the expansion of the plasma channel, delivering poor surface quality (Fig. 17(b)).

Fig. 16

Surface topography with CT brass tool at (a) I_P = 5 A, S_V = 2 V, P_ON = 50 µs, C_P = 0.5 g/100 ml; (b) I_P = 25 A, S_V = 6 V, P_ON = 150 µs, C_P = 1.5 g/100 ml.

Fig. 17

SEM of machined specimen for SR_CT at (a) I_P = 5 A, S_V = 2 V, P_ON = 50 µs, C_P = 0.5 g/100 ml; (b) I_P = 25 A, S_V = 6 V, P_ON = 150 µs, C_P = 1.5 g/100 ml.

For the NT brass electrode, very little material re-solidification is noticed during EDM of Al6061, as depicted in Fig. 18(a). On the opposite side, the impression made with the CT brass tool gives more redeposited melt over the specimen surface, as presented in Fig. 18(b). The 3D surface topography analysis (corresponding to the micrographs (Fig. 18(a, b))) revealed the same scenario of smaller peaks and valleys which is in the favour of smaller material erosion during the machining process. It has already been studied that NT brass provides high SEC due to the arcing effect and uneven distribution of discharges. However, CT brass, owing to better metallurgical properties, provides a minimum value of SEC.

Fig. 18

Microscopic images of machined profile for SEC utilizing (a) NT brass tool, (b) CT brass tool.

Parametric significance of control variables

The formulated designs have been tested by ANOVA. It is usually applied for identifying the significant parameter and studying the interactions among input variables and individual output. The parametric significance and involvement percentage were examined by p-values and f-values, respectively. It also assists us in determining the statistical validation of the proposed models by computing R², adjusted R², and predicted R² magnitudes.

Material removal rate

The ANOVA results for MRR with both types of electrodes (NT brass and CT brass) are compiled in Table 7. It can be inferred from Table 7 that all four parameters have significance, with p-values less than 0.05. Furthermore, among various combined interactions, I_P×S_V and S_V×P_ON have p-values smaller than 0.05, showing their importance as model terms. Regarding quadratic effects, all parameters have been found significant except I_P². Similarly, for MRR computed against CT brass, all input parameters and their combined interactions, except P_ON×C_P, have p-values smaller than 0.05, portraying their importance as model value. However, in the case of MRR calculated with CT brass, C_P² is the only dominant factor among all the given quadratic terms. The R² values for both types of electrodes are provided in Table 7 to evaluate the reliability of the model. It can be observed that the R² values for NT brass and CT brass are greater than 0.9, confirming a high level of reliability in the dataset. The predicted models for MRR against NT brass and CT brass are presented in Eqs. 3 and 4, respectively.

Table 7 ANOVA for MRR_NT and MRR_CT.

$$\begin{aligned}\:{MRR}_{NT}&=\:-1.46289+0.312404\times\:\:{I}_{P}\:-0.160609\times\:{S}_{V}\:-0.002255\times\:{P}_{ON}+5.62701\times\:{C}_{P}-0.017000\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}-0.000211\times\:{I}_{P}\times\:{P}_{ON}-0.039750\times\:{I}_{P}\times\:\:{C}_{P}+0.005600\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.161875\times\:\:{S}_{V}\times\:\:{C}_{P}-0.010850\times\:{P}_{ON}\\& \times\:{C}_{P}-0.001916\times\:{{I}_{P}}^{2}+0.142105\times\:{{S}_{V}}^{2}+0.000193\times\:{{P}_{ON}}^{2}-2.70632\times\:{{C}_{P}}^{2}\end{aligned}$$

(5)

$$\begin{aligned}\:{MRR}_{CT}&=\:-1.72470+0.320069\times\:\:{I}_{P}+0.446025\times\:{S}_{V}\:+0.031883\times\:{P}_{ON}+5.67716\times\:{C}_{P}-0.019906\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}-0.000568\times\:{I}_{P}\times\:{P}_{ON}+0.007375\times\:{I}_{P}\times\:\:{C}_{P}-0.003469\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.267500\times\:\:{S}_{V}\times\:\:{C}_{P}+0.000200\times\:{P}_{ON}\\& \times\:{C}_{P}-0.000942\times\:{{I}_{P}}^{2}+0.098947\times\:{{S}_{V}}^{2}+0.000082\times\:{{P}_{ON}}^{2}-2.31684\times\:{{C}_{P}}^{2}\end{aligned}$$

(6)

Surface roughness

The ANOVA findings for SR with both NT and CT brass within a predetermined range of input variables are shown in Table 8. All major effect variables for SRNT have p-values smaller than 0.05, demonstrating the importance of these variables as model terms. Furthermore, the significance of these model components is shown by the p-values for interaction parameters, which include I_P×S_V, I_P×C_P, S_V×C_P, and P_ON×C_P, all of which are below 0.05. Only two factors (I_P² and P_ON²) are determined to be significant in the case of quadratic effects, with p-values less than 0.05. For SR_CT, all main effect factors and quadratic effect parameters are significant except I_P², with p-values less than 0.0001 and less than 0.05, respectively. Conversely, significance has been observed only in the three combined interactions (I_P×C_P, S_V×P_ON, and S_V×C_P). The R² values for both types of electrodes demonstrate greater reliability in the proposed models. The adequacy precision values for SR_NT and SR_CT are 37.4301 and 36.4146, respectively. The mathematical models for SR against NT and CT brass electrodes are highlighted in Eq. 5 and Eq. 6, respectively.

Table 8 ANOVA for SR_NT and SR_CT.

$$\begin{aligned}\:{SR}_{NT}&=\:-2.21096+0.064810\times\:\:{I}_{P}\:+0.634265\times\:{S}_{V}\:+0.056732\times\:{P}_{ON}+1.41762\times\:{C}_{P}-0.006875\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}-0.000025\times\:{I}_{P}\times\:{P}_{ON}-0.062500\times\:{I}_{P}\times\:\:{C}_{P}-1.95314\text{E}-18\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.275000\times\:\:{S}_{V}\times\:\:{C}_{P}+0.006000\times\:{P}_{ON}\\& \times\:{C}_{P}+0.002219\times\:{{I}_{P}}^{2}+0.005482\times\:{{S}_{V}}^{2}-0.000251\times\:{{P}_{ON}}^{2}+0.087719\times\:{{C}_{P}}^{2}\end{aligned}$$

(7)

$$\begin{aligned}\:{SR}_{CT}&=\:-2.65406+0.116455\times\:\:{I}_{P}\:+0.239501\times\:{S}_{V}\:+0.033691\times\:{P}_{ON}+4.93856\times\:{C}_{P}-0.003438\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}-0.000138\times\:{I}_{P}\times\:{P}_{ON}-0.088750\times\:{I}_{P}\times\:\:{C}_{P}-0.001813\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.206250\times\:\:{S}_{V}\times\:\:{C}_{P}-0.001250\times\:{P}_{ON}\\& \times\:{C}_{P}+0.000623\times\:{{I}_{P}}^{2}+0.053070\times\:{{S}_{V}}^{2}-0.000075\times\:{{P}_{ON}}^{2}-1.35088\times\:{{C}_{P}}^{2} \end{aligned}$$

(8)

Specific energy consumption

The ANOVA findings for SEC_NT and SEC_CT are tabulated in Table 9. Concerning NT brass, all main effect parameters except S_v are found to be significant. However, only the interactions I_P×C_P, S_V×C_P, and P_ON×C_P are deemed significant with p-values smaller than 0.05. Among the quadratic effects, only I_p² exhibits a p-value less than 0.05, showing its significance in the model. For SEC_CT, all main effect parameters and combined parameters except I_P×C_P and S_V×C_P demonstrate significance with p-values less than 0.05, highlighting their importance in the model. In addition, I_P², S_V², and P_ON² show significant effects, while C_P² has a p-value very close to 0.05, suggesting some significance. In terms of model reliability, the R² values for both SEC_NT and SEC_CT exceed 0.9, indicating high reliability. The adequacy precision for SEC_NT and SEC_CT is 16.6711 and 24.2616, respectively. The mathematical models for SEC for NT and CT brass electrodes are given in Eqs. (7) and (8).

Table 9 ANOVA for SEC_NT and SEC_CT.

$$\begin{aligned}\:{SEC}_{NT}&=\:+6.96115+0.067744\times\:\:{I}_{P}\:-0.157111\times\:{S}_{V}\:-0.038580\times\:{P}_{ON}+5.09558\times\:{C}_{P}+0.006187\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}-0.000385\times\:{I}_{P}\times\:{P}_{ON}+0.082000\times\:{I}_{P}\times\:\:{C}_{P}+0.001900\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.340000\times\:\:{S}_{V}\times\:\:{C}_{P}+0.013650\times\:{P}_{ON}\\& \times\:{C}_{P}-0.007470\times\:{{I}_{P}}^{2}+0.018246\times\:{{S}_{V}}^{2}+0.000045\times\:{{P}_{ON}}^{2}-2.44807\times\:{{C}_{P}}^{2} \end{aligned}$$

(9)

$$\begin{aligned}\:{SEC}_{CT}&=\:+9.07871-0.045330\times\:\:{I}_{P}\:-1.60848\times\:{S}_{V}\:+0.002022\times\:{P}_{ON}+0.505954\times\:{C}_{P}+0.008906\times\:\:{I}_{P}\:\\& \times\:\:{S}_{V}+0.000386\times\:{I}_{P}\times\:{P}_{ON}+0.027375\times\:{I}_{P}\times\:\:{C}_{P}+0.003806\times\:{S}_{V}\times\:\:{P}_{ON}\:-0.093125\times\:\:{S}_{V}\times\:\:{C}_{P}+0.014975\times\:{P}_{ON}\\& \times\:{C}_{P}-0.003939\times\:{{I}_{P}}^{2}+0.135263\times\:{{S}_{V}}^{2}-0.000244\times\:{{P}_{ON}}^{2}-0.475789\times\:{{C}_{P}}^{2}\end{aligned}$$

(10)

Sustainability and process modelling

Over time, EDM emissions create dangerous compounds that contaminate the air and lead to respiratory illnesses and air pollution. To lessen these effects on the environment and public health, industries must adopt cleaner technologies, strict emission regulations, and ethical waste management.

Sustainable EDM practices with high MRR, low SR, SEC, and CO₂ emissions can be achieved through advanced EDM assistive technologies such as, novel electrodes for stable pulse generation establishing controllable erosion, environmentally friendly dielectric fluids and process parametric control etc. These factors determine the effectiveness, economy, and environmental impact of EDM operations. High MRR affects profitability, SR regulates surface quality, SEC limits power consumption during machining, and CO₂ emissions guarantee environmentally responsible operation. These emissions are caused by non-biodegradable dielectric fluids, hence sustainable EDM approaches need to reduce them. By reducing their environmental effect and achieving sustainability targets for waste reduction, energy efficiency, and environmental preservation, these techniques enable EDM operations to operate as efficiently as possible. Figure 19 illustrates an overview of sustainability for EDM. However, the scope in this study is emissions based on the consumed electricity.

Fig. 19

Sustainable approaches in EDM⁴⁸ (Licensed Open Access and reused with the permission of author).

When comparing the SEC of KO dielectric with that of DI dielectric used in EDM, significant differences in energy consumption become apparent. KO has an SEC index of 102.523 kJ/mm³, however its DI cousin has significantly lower SEC parameters. Using DI dielectric in EDM operations has the potential to reduce CO₂emissions while also producing energy savings. This mitigation effort is evaluated based on adherence to the IPCC’s guidelines⁴⁹. The proportional decrease in CO₂ emissions for both KO and DI dielectrics is shown in Fig. 20. Using DI results in an average CO₂ reduction of around 94.81 ± 2.58% for the NT brass electrode and 95.56 ± 1.01% for the CT brass electrode. As a result, in the context of the EDM framework, DI dielectric becomes the model substitute for KO, supporting sustainability goals and providing more control over carbon emissions.

Fig. 20

% Reduction in CO₂ emissions for different electrodes.

Figure 21 illustrates the relationship between the goal and output values with a total ANN modelling regression coefficient R value of 0.95105. The two primary measures for assessing the effectiveness of machine learning models are R²and RMSE⁵⁰.

Fig. 21

The ANN provided forecasts for MRR, SR, and SEC for the training, testing, and validation datasets.

The results from the experiment indicated that the ANN model predicts response measures and that the R² for each response measure is more than 0.98 with an exception of few. In Fig. 22(a-c), contour graphs are made using NT brass electrode between the number of neuron layers (3 to 10), R²_train, R²_test, and R²_val, and RMSE_train, RMSE_test, and RMSE_val. Figure 22(a) shows MRR_NT with R²_train > 0.99 and RMSE_val = 0.09 mm³/min for 8 neuron layer and E^-4 learning rate. Figure 22(b) shows that the R²_train value for SR_NT is > 0.99 and the RMSE_val is 0.032 μm for the 9 neuron layer and E^-1 learning rate. As shown in Fig. 22(c), the best SEC_NT model has 8 neurons, E^-3 learning rate, R²_train > 0.99, and RMSE_val = 0.085 kJ/mm³. In the same way, In Fig. 20(a-c), contour graphs are made using CT brass electrode between the number of neuron layers (3 to 10), R²_train, R²_test, and R²_val, and RMSE_train, RMSE_test, and RMSE_val. Figure 23(a) shows MRR_CT with R²_train > 0.99 and RMSE_val = 0.07 mm³/min for 9 neuron layer and E^-2 learning rate. Figure 23(b) shows that the R²_train value for SR_CT is also more than 0.99 and the RMSE_val is 0.042 μm for the 5 neuron layer and E^-1 learning rate. As shown in Fig. 23(c), the best SEC_CT model has 7 neurons, E^-4 learning rate, R²_train > 0.99, and RMSE_val = 0.11 kJ/mm³. Thus, the ANN model estimates response measure values better than other models. Table 10 summarizes all ANN model results. However, there are still challenges about the generalization and reproducibility of similar performance metric because of inherent limitations of ANN and sensitivity towards data categorization. 

Fig. 22

Construction of an ANN model for the NT brass electrode, involving hidden layer neurons for (a) MRR, (b) SR, and (c) SEC, is currently undergoing training, testing, and validation periods.

Fig. 23

Construction of an ANN model for the CT brass electrode, involving hidden layer neurons for (a) MRR, (b) SR, and (c) SEC, is currently undergoing training, testing, and validation periods.

Table 10 Results of the ANN model.

Sustainability targeted process optimization

NSGA-II is an evolutionary algorithm which finds a compromise between multiple objectives.  Using heuristic techniques or random selection, it first creates a population of viable solutions. Subsequently, every solution is evaluated by means of the established objective functions that encompass the several optimization criteria. Individuals are sorted into several fronts according to their dominance relationship through a process known as non-dominant sorting, in which a person (as an example) performs better than another in one or more objectives without degrading in any other area. The population is thus sorted into fronts, the first of which is made up of people who are independent of all others. The density of surrounding solutions in the objective space is then measured by assigning a crowding distance to each individual within a front. In order to preserve diversity along the Pareto front and favour solutions with larger crowding distances, NSGA-II strikes a balance between exploration and exploitation throughout the selection process by taking into account both non-dominant sorting and crowding distance. To create children for the following generation, a select few undertake genetic procedures like crossover and mutation, which introduce new variations. The best solutions from the previous generation are preserved in the following population through the use of elitism. Decision-makers are presented with a variety of Pareto-optimal solutions that provide trade-off possibilities for multi-objective optimization issues by means of this iterative procedure, which continues until a termination requirement is satisfied. Figure 24 depicts the working schematic of NSGA-II.

Fig. 24

NSGA-II working schematic⁵¹.

Using NT and CT brass electrodes, the input variable magnitudes are optimized in this study using NSGA-II to find the optimal MRR_NT, SR_NT, SEC_NT, MRR_CT, SR_CT, and SEC_CT solutions. Equation 11 demonstrates the objective function that was used to maximize the values of the input variables, but Eq. 12 illustrates the objective function’s limits.

$$\:\text{S}\text{u}\text{s}\text{t}\text{a}\text{i}\text{n}\text{a}\text{b}\text{i}\text{l}\text{i}\text{t}\text{y}\:\text{b}\text{a}\text{s}\text{e}\text{d}\:\text{o}\text{b}\text{j}\text{e}\text{c}\text{t}\text{i}\text{v}\text{e}\:\text{f}\text{u}\text{n}\text{c}\text{t}\text{i}\text{o}\text{n}\::\:f=\:\left\{\begin{array}{c}Maximize\:MRR\_NT\\\:Minimize\:SR\_NT\\\:Minimize\:SEC\_NT\\\:Maximize\:MRR\_CT\\\:Minimize\:SR\_CT\\\:Minimize\:SEC\_CT\end{array}\right.$$

(11)

$$\:\text{C}\text{o}\text{n}\text{s}\text{t}\text{r}\text{a}\text{i}\text{n}\text{t}\text{s}:\:\left\{\begin{array}{c}\:5\le\:{I}_{P}\le\:25\\\:\:2\le\:{S}_{V}\le\:6\\\:\:50\le\:{P}_{ON}\le\:150\\\:0.5\le\:{C}_{P}\le\:1.5\end{array}\right.$$

(12)

Upon successfully executing NSGA-II within the designated restrictions, various solutions for the input variables and output answers are achieved. The process parameters, i.e., I_P, S_V, P_ON, and C_P values of 24.85 (A), 2.18 (V), 119.11 (µs), and 1.05 (g/100 ml), respectively, should be used, the algorithm concluded, in order to optimize MRR_NT, and MRR_CT and minimize SR_NT, SR_CT, SEC_NT and SEC_CT. Figure 25(a) illustrates the MRR_NT of 8.10 mm³/min, SR_NT of 4.16 μm, and SEC_NT of 4.86 kJ/mm³, and the MRR_CT of 11.60 mm³/min, SR_CT of 4.08 μm, and SEC_CT of 3.70 kJ/mm³ as a consequence of the multi-objective function (Fig. 25(b)). Additionally, the ideal parametric configuration recommended by NSGA-II is provided in Table 11. Confirmatory runs have verified the ideal configuration proposed by NSGA-II; the findings are collected in Table 12. NSGA-II was run twice more and confirmatory runs were performed to prove the process’s efficacy. Table 12 also includes the outcomes of these validation runs.

Fig. 25

NSGA-II contours’ plots showing the best possible solution for response measurements.

Table 11 The ideal parametric configuration recommended by NSGA-II.

Table 12 Verified experimental outcomes and enhanced data.

The important improvements made achievable by including NSGA-II in the confirmation investigations are shown in Table 12. In particular, the maximum percentages for MRR increased by 77.75%, SR fell by 31.80%, and SEC improved by roughly 43.02% for the NT brass electrode in the confirmatory experiment. In the experiment, MRR increased by 64.82%, SR lowered by 27.45%, and SEC improved by roughly 46.60% for the CT brass electrode, respectively, for the greatest percentages.

Conclusions

This article reports the multi-response examination of Al6061 machined via EDM against different sets of control parameters. The findings have been thoroughly illustrated with optical micrographs, SEM, EDX, 3D surface plots, and bar graphs. ANOVA has also been applied to determine the significance of parameters. Multi-response optimization through the NSGA-II has been performed for the set of processing parameters, and response measures. The key points of this study have been concluded below:

• The maximum MRR (10.42 mm³/min) was achieved with the CT brass at 25 A, 2 V, 150 µs, 0.5 g/100 ml, which is almost 25% higher than the MRR (8.32 mm³/min) observed with the NT brass tool. This improvement is primarily due to the enhanced characteristics of the CT electrode. On average, the maximum MRR (10.08 mm³/min) was achieved with the CT electrode at a combination of 150 µs and 25 A. Compared to the NT brass tool, this MRR value is approximately twice as high under the same parametric conditions.

• MRR is found to be higher when I_P, S_V, and P_ON are increased to 25 A, 6 V, and 150 µs due to the intensification of sparks at these higher parameter values, resulting in greater material erosion from the specimen’s surface. Additionally, MRR is significantly influenced, with an increase of 80.8% at low conditions (C_P: 0.5 g/100 ml, I_P: 5 A) and 43.7% at high conditions (C_P: 1.5 g/100 ml, I_P: 25 A) when comparing CT performance with the NT electrode.

• According to the findings, SR is reduced by 11.49% at low settings (5 A, 2 V) and 25.8% at high settings (25 A, 6 V) with the CT brass electrode compared to NT brass, due to more consistent discharge phenomena and better surface quality even at higher current and voltage levels. CT brass also improves SR by 17.8% at low (0.5 g/100 ml, 5 A) and 33.6% at high (1.5 g/100 ml, 25 A) settings, owing to its refined grain structure, which minimizes irregular discharges and enhances surface finish in the EDM process. Moreover, the CT brass electrode enhanced surface finish by 11.9% and 33.6% at 50 µs, 5 A and 150 µs, 25 A, respectively.

• In the case of SEC, the minimum value (2.25 kJ/mm³) is achieved with the CT electrode at the parametric setting of (25 A, 2 V, 150 µs, 0.5 g/100 ml) during the EDM of Al6061. This value is approximately 1.07 times lower than the Sect. (2.41 kJ/mm³) obtained with the NT brass electrode. On average, the minimum Sect. (3.34 kJ/mm³) is observed with the CT electrode at the parametric combination of (150 µs and 25 A). This improvement is due to proper heat distribution over the machined cavity, attributed to the high-temperature gradient and the appropriate gap of the plasma channel. Additionally, at this combination, the SEC is 16.16% lower compared to the Sect. (3.88 kJ/mm³) achieved with the NT brass tool.

• Surface morphology obtained with CT brass is somewhat better than that obtained with NT electrode. The improved grain refinement in the CT brass is the key reason for providing improved microstructure. However, the surface topology provided by NT brass is due to the unstable arcing effect, causing large peak-to-valley height.

• The ANOVA analysis revealed that all the parametric effects are significant in the case of CT brass, showing their p-value less than 0.05, whereas only S_V remains insignificant when NT brass is utilized during EDM of Al6061. Moreover, the magnitude of R² in each response was larger than 0.9, which highlights a close connection between the actual value and the proposed value.

• An ANN has been constructed for the better forecast of output responses. It has been found that the R² for almost each response measure is greater than 0.99 which shows the close relationship among the predicted and the experimental values. Moreover, the respective RMSE magnitudes are lower, which is in the favour that ANN has better predict the values and predict the close relationship between the experimental and predict values. The values given by the ANN models for different response measures for RMSE are within the desired range for the EDM setup.

• Multi-response optimization has been performed through the NSGA-II for each response measure. It has been found that the NSGA-II has recorded the better results for all the response measures. The magnitudes of MRR_CT, SR_CT, and SEC_CT obtained by multi-response optimization are 64.82%, 27.45%, and 46.60% are better than the values obtained by un-optimized settings of CT brass electrodes. However, the optimal magnitudes of processing parameters are I_P = 24.85 A, S_V = 2.18 V, P_ON = 119.11 µs, and C_P = 1.05 g/100 ml.