Hybrid Firefly Algorithm-based optimization of reverse EDM for machining of titanium superalloys for high-precision biomedical applications Authors: Renu Kiran Shastri, Chinmaya P Mohanty, Kishore Kumar Mahato, Pravat Ranjan Pati Published: 2026 ABSTRACT AbstractTitanium alloy Grade 5 (Ti-6Al-4V) is widely used for biomedical applications due to its high strength and high corrosion resistance. In this work, the reverse electric discharge machining (R-EDM) for the fabrication of macro-pillared array structures on Ti-6Al-4V work piece is carried out using tungsten, copper and copper-tungsten electrodes. The effect of pulse-on time (Ton), flushing pressure (Fp), peak current (I), voltage (U), duty factor (τ) and electrode material on surface roughness, microhardness, recast layer thickness and surface crack density is investigated. A Box Behnken design of response surface methodology (RSM) is used to assess the significance of the parameters and their interaction effects. A meticulous scanning electron microscopy (SEM) analysis is carried out to evaluate the machined surface quality. Multi-response optimization is carried out using the additive ratio assessment (ARAS) method and is coupled with the Firefly Algorithm (FA) to achieve the optimum results. The optimum machining condition is validated by conducting a confirmative test showing an improvement of 6.86 percentage. The proposed work is useful for selecting ideal process conditions while machining Titanium (Ti-6Al-4V) work piece for biomedical application. Keywords: R-EDM, Ti-6Al-4V, Array structure, SEM, ARAS, Firefly AlgorithmSubject terms: Engineering, Materials science Introduction Titanium alloys, particularly Ti-6Al-4V, extensively used in biomedical applications including replacement of tissues, cardiovascular implants and bone supporting array structures because of their superior corrosion resistance high fatigue strength, good strength-to-weight ratio, and low elastic modulus1. However, extreme hardness, low heat conductivity, and poor machinability of Ti-6Al-4V make it difficult to fabricate intricate arrayed pillared structures by traditional machining techniques, resulting in significant tool wear, poor surface polish, and dimensional errors. To conquer these drawbacks, non-traditional machining procedures, such as electric discharge machining (EDM), have garnered significant interest for machining electrically conductive and difficult-to-machine materials. The absence of direct contact between the tool and workpiece allows the manufacturing of complex forms, with high dimensional accuracy and low mechanical stresses, in EDM2,3. The reverse EDM, a variant of EDM, where the polarity setup between the workpiece and the tool electrode is reversed, is a recently developed promising technique for creating complex cavities and arrayed structures4. In reverse EDM, the shape of the tool electrode is transferred to the workpiece as a reverse profile, which makes it useful for producing precision biomedical microstructures and high aspect ratio features. Recent studies show that the machining of titanium alloys, especially Ti-6Al-4V, is a serious challenge due to their low elastic modulus, high chemical reactivity, work hardening tendency and low thermal conductivity. All of these factors lead to excessive tool wear, poor surface integrity and dimensional errors with traditional machining techniques5. Therefore, non-conventional machining processes such as EDM, micro-EDM and WEDM have been receiving more attention for machining titanium alloys with improved accuracy and lower mechanical stresses6. Recently, significant research has been done on the EDM machinability, surface integrity and optimization of Ti-6Al-4V using advanced electrode materials, intelligent optimization techniques and hybrid machining techniques7. Furthermore, reverse EDM and negative polarity EDM methods have shown promising ability for precision microstructure manufacturing, material deposition and enhanced surface quality promising for biomedical and microengineering applications with complex arrayed structures and high aspect ratio features8. Recent studies have been focused on improving EDM machining performance by process optimization, intelligent modeling and microstructures fabrication for advanced engineering applications. Chaturvedi et al.9 improved the machinability and surface finish using textured tools under MQL conditions. Tamang et al.10 used machine learning techniques in WEDM of Mg-SiC nanocomposites. This study observed an improvement in the quality of machining and the forecast accuracy. Jin et al.11 developed an EDM-based method for fabricating ultra-high aspect ratio microcylindrical electrode arrays with enhanced dimensional accuracy. The proposed approach was successfully used for preparation of stable and precise microelectrode constructs. Pratap and Datta12 investigated reverse micro-EDM for array microelectrode fabrication and found that process optimization improved geometrical precision. The study showed that the size and shape features of microelectrodes could be controlled efficiently. Pratap et al.13 explored material removal and management of recast layer in reverse µEDM to improve surface integrity and sustainably fabricate microelectrode arrays. Under the ideal circumstances, the recast layer thickness and thermal damage were largely reduced. To improve machining performance, Mani et al.14 used reverse sawtooth pulse designs to control reverse micro-EDM parameters for titanium alloy machining. The method enhanced both the machining efficiency and quality of surface. Jin et al.15 explored the effectiveness of EDM on nickel-based superalloys and suggested a self-rounding manufacturing method of micro-cylindrical electrode arrays. The study enhanced the electrode geometry and the machining consistency. Later studies16,17 investigated microstructure fabrication based on reverse EDM and deposition behavior under different machining parameters. The surface integrity and machining performance in EDM of difficult-to-machine materials have also received considerable attention. In EDM of Inconel, steel, tungsten carbide and titanium alloys, the influence of machining parameters and electrode materials on tool wear, surface roughness, surface crack formation and recast layer characteristics have been studied in previous works18–22. Optimization of EDM performance parameters using artificial intelligence-based methods and multi-criteria decision-making (MCDM) has been widely reported in the literature in recent years. Techniques such as Grey relational analysis, Grey-fuzzy optimization, cuckoo search, genetic algorithms, ANFIS and hybrid ANN based models have been successfully used to improve machining responses such as electrode wear rate, surface roughness, overcut, material removal rate, in EDM and WEDM processes23–29. These studies demonstrated how intelligent optimization strategies can enhance machining performance and process dependability. Additive Ratio Assessment (ARAS) is one of the multi-criteria decision making (MCDM) techniques. It has attracted great attention for their simplicity, computational efficiency and ranking of the alternatives with respect to normalized utility values. ARAS is more reliable in decision making and has a simple mathematical implementation as compared to the conventional methods like TOPSIS and MOORA30. Similarly, Firefly Algorithm (FA), a nature-inspired metaheuristic optimization technique proposed by Yang, has shown better exploration and exploitation capability with faster convergence and less chances of trapping in local optima than Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). FA is particularly well-suited for nonlinear and multi-objective optimization problems due to its automatic subdivision and adaptive search mechanism31,32. There is huge amount of research published on conventional EDM, WEDM and micro-EDM processes, however the work on reverse EDM processing of Ti-6Al-4V is still limited as indicated in Table 1. Reverse EDM studies available until now are mostly concerning microstructure fabrication, deposition mechanisms and machining of materials except titanium alloys33–35. Additionally, relatively little research has been done on the simultaneous assessment of surface integrity characteristics in reverse EDM of Ti-6Al-4V, such as surface roughness, microhardness, recast layer thickness, and surface crack density. Moreover, the comparison of different electrode materials and hybrid multi-response optimization methods for biomedical applications is still rarely reported. Table 1Comparison of previous EDM/reverse EDM studies with present work.Ref.Material/processFocusLimitation 11–13 Reverse EDM/µEDMMicroelectrode fabrication and recast layer controlNo Ti-6Al-4V application or hybrid optimization 14 Titanium alloy/reverse micro-EDMMachining performance optimizationLimited surface integrity analysis 15–17 EDM/reverse EDMSurface fabrication and modificationNo biomedical application 18–21 EDMSurface integrity, SCD, and recast layer analysisNo reverse EDM investigation 23–29 EDM/WEDMMulti-response optimization using AI and MCDM methodsNo ARAS–Firefly optimization for reverse EDM of Ti-6Al-4Vpresent workTi-6Al-4V/reverse EDMSurface roughness, microhardness, SCD, and recast layer optimization using ARAS–FireflyAddresses identified research gap Comparison of previous EDM/reverse EDM studies with present work. In this respect, a great scientific gap exists in understanding and optimization of reverse EDM machinability of Ti-6Al-4V for manufacturing of precision structures with potential biomedical applications. The present study seeks to bridge this gap by systematically examining the influence of copper, tungsten and copper-tungsten electrodes on the reverse EDM performance of Ti-6Al-4V. Surface characterization is performed by SEM for detailed analysis of the influence of pulse-on-time (Ton), peak current (I), voltage (U), duty factor (τ), flushing pressure (Fp) and electrode material on recast layer thickness, surface crack density, microhardness, and surface roughness. Moreover, a hybrid optimization framework based on additive ratio assessment (ARAS) and Firefly Algorithm is applied to find the optimal machining conditions for better surface integrity performances and the optimized results are validated experimentally. The Additive Ratio Assessment (ARAS) optimization approach A multi-criteria optimization technique is utilized to identify the most optimized function from a set of functions concurrently. The performance of specified parameters is assessed using Additive Ratio Assessment (ARAS), a multi-criteria decision-making (MCDM) technique. It distinguishes the scores of these parameters in comparison to the optimal parameter. The ARAS technique, initially introduced by EK Zavadskas and Turskis Z30, has been effectively applied in previous works within the domains of science and technology35,36. ARAS procedure Step 1: Developing the initial decision matrix(X). The decision matrix obtained from the experiment can be organized into a matrix format, where m(rows) represents the number of alternatives and n(columns) represent the number of criteria. The following equation no. 1 shows the decision matrix. Step 2: Generation of optimal value of j criterion and adding it to the matrix. In this experimental work surface crack density, surface roughness, micro hardness, recast layer thickness is considered as the non-beneficial criteria terms. Step 3: Formation of normalized decision matrix (r). Normally, the criteria have various dimensions. The main objective of this level is to generate dimensionless weighted values from the comparative criteria. Optimal values were utilized to eliminate the difficulties caused by various dimensions. For beneficial criteria, whose values should be preferably maximum is calculate by equation no. 4. The normalization will be applied in two phases for the non-beneficial criteria whose preferable values should be minimum and it is given by equation no.3. Hence . Step 4: Calculation of weighted normalized decision matrix (w). The normalized values of the data are multiplied to the weights of the criteria in order to influence the outcome of the problem. The weights of the surface crack density, surface roughness, micro hardness, and recast layer thickness are considered as 0.25 individually so the total sum is 1. Step 5: Determine the optimality function (Si) for ith alternative by following equation no.4. The higher the Si value, the better the alternative. The values in the choice matrix and the weights of the criteria are directly and proportionally related to the optimality function Si. Step 6: Calculate the utility degree (Ki) for each alternative by comparing it with the most efficient one (S0). Use equation no. 5 to determine the utility degree (Ki). The alternatives have utility values between 0% and 100%. The best choice among the alternativesis the alternative with the largest utility value (Kmax). Experimental design and planning Response Surface Methodology (RSM) encompasses a collection of statistical methods employed to model and investigate scenarios where several factors impact a specific response, with the ultimate objective of optimizing that response. One of the most popular techniques under the umbrella of Design of Experiments (DOE) is the Box-Behnken design. This design is chosen because it can provide optimum estimations for all the response variables by strategically distributing each numeric value to three different levels. The Box-Behnken design has limitations in covering the corners of nonlinear design spaces. However, it is preferred over alternatives like three-level full factorial design, Doehlert design and central composite design (CCD) due to its ability to provide more accurate design data. Design Expert software was used to generate the Box-Behnken design. The software generates a detailed three-level design by specifying the input machining parameters such as pulse-on-time (Ton), flushing pressure (Fp), peak current (I), voltage (V), duty factor (τ), and electrode material. This enhances the accuracy and efficiency of the experimentation process as shown in Table 2. Table 2Process parameters and their levels.Process parametersUnitSymbolsLevels−101Pulse on time (Ton)µsA300400500Flushing pressure (Fp)kg/cm2B0.20.30.4Peak current (I)AC101520Voltage (V)VD100200300Duty factor (τ)%E808590ElectrodeFTungstenCopper tungstenCopper Process parameters and their levels. Experimental planning The workpiece material of each type of electrode was titanium alloy Grade 5 (Ti-6Al-4V). The tool electrodes are made of copper, tungsten and copper-tungsten alloy. The tool as well as workpiece electrodes were standardized to 25 × 25 mm2 dimensions. Table 3 is a ranking of the electrodes by their thermal conductivity, with copper being the highest, followed by copper-tungsten and tungsten. The tool electrodes (copper, tungsten and copper-tungsten) were pre-drilled by conventional drilling method to make holes of ⌀ 3 mm diameter and 2 mm depth, spaced 4 mm apart as shown in Fig. 2d–f. The experimental setup for the fabrication of pattern array structures is prepared with a die sinking spark electric discharge machine named RATHNAPAKARTHI EDM 5530. Figure 1. Shows the schematic diagram of EDM setup. During this process, the workpiece material was the anode, and thus it was subjected to positive voltage, while the tool electrode was the cathode and was supplied with negative voltage. The process was conducted using grade 3 hydrocarbon EDM oil with kinematic viscosity of 4.0 CST and dielectric strength of 45 kV. Table 3Thermal conductivity of the tool and work electrodes.CopperTungstenCopper-tungstenTi-6Al-4V413 W/mk197 W/mk238 W/mk24 W/mk Thermal conductivity of the tool and work electrodes. (a–c) Ti-6Al-4V, (d) copper, (e) tungsten, (f) copper-tungsten. Schematic of EDM setup. The tool electrode cube moves systematically against the work piece material at a pre-decided rate as per the input parameters in the Box-Behnken design33,34. At the breakdown voltage, a distinct phenomenon of material removal takes place on the surface of the workpiece. The removal is exactly in the shape of the tool electrode, leading to the formation of elevated arrayed structures on the Ti-6Al-4V workpiece. The procedure is repeated for 54 runs of 5 min each. Figure 2a–c visually displays Ti-6Al-4V workpieces depicting the corresponding high array structures after the machining with copper, tungsten and copper-tungsten electrodes. The recorded output responses following each run included surface crack density, surface roughness, micro hardness and recast layer thickness at the specified input levels. After 54 runs, three workpieces were fabricated, which showed the array features of pillared structures as illustrated in figure. Calculation of output responses Micro hardness (Hv) ATSM E384 test method is used to determine the micro hardness. Vickers hardness test equipment is utilized for the process. The principle involves is applying the load to work piece at an intended area to find the hardness value. F is set to 30Kgf. Surface roughness (µm) The “surface roughness” means the deviation of the surface from the ideal surface. Before machining the surface is measured for its roughness by using a Mitutoyo (make) Surftest SJ-210. This is obtained by averaging five measurements at different locations on the machined surface for each parameter. Surface crack density (µm2/µm) Surface crack density is characterized as the number of cracks per unit volume. Cracks will inevitably occur during the EDM process. This can be explained by the cooling of the work surface by the dielectric fluid after strong heating, which creates thermal tensions that contribute to the formation of small craters and microcracks. To assess and quantify these cracks, scanning electron microscopy is employed to capture an image area on the workpiece, with the image size fixed at 12,400 μm2. Subsequently, ImageJ software is utilized to measure the length of each crack within the specified image area. The initial length is predetermined in the software based on the prescribed length from the SEM micrograph. Recast layer thickness (µm) The recast layer is known as the molten particles that have been re-deposited on the surface of the workpiece during a pulse of time. The recast layer length is measured using ImageJ, with the micrograph area set to 8099 μm2 in the SEM. Equation 7 represents the calculation for RCT. Results and discussion The Box-Behnken design was used in the experimental runs to evaluate the machinability of the Ti-6Al-4V alloy utilizing copper, tungsten, and copper-tungsten tool electrode materials. The process parameters chosen, their levels, and the need to assess the interaction effects and quadratic behavior of the created models were taken into consideration while determining the overall number of experiments. Design Expert software is used to generate Analysis of Variance (ANOVA) for examining the output responses that correlate to each process parameter. The ANOVA table was made for the results after irrelevant parameters were removed from the data at a significance level of 0.05. Selected tests were conducted three times in order to improve measurement reliability, and average data were obtained for analysis. Standard deviation was used to compute surface roughness, microhardness, surface fracture density, and recast layer thickness. The repeatability of the experimental process was confirmed by the variation in the repeated trials falling within the allowed range. Dielectric conditions were kept constant and the electrodes were properly aligned. The machining conditions were controlled during the experimentation to minimize the experimental uncertainty. The resultant R-squared and adjacent R-squared values for all four outcome responses exceeded ninety and eighty%, respectively, with the lack of fit deemed as not significant. The Cor Total (corrected total sum of squares) in the ANOVA table serves as a valuable metric for discerning the percentage contribution of individual machining parameters to the outcome responses. The percentage influence of each parameter can be computed by dividing its sum of squares by the Cor Total. The rank of utility degree as observed empirically is shown in Table 4. The ranks in Table 4 are determined using the ARAS (Additive Ratio Assessment) technique, which takes into account each alternative’s normalized weighted performance score in respect to the ideal answer. Options with higher overall utility values are ranked higher. Table 4Experimental design along with the outcome response measures.RunABCDEFSRMHSCDRCTKiRANK11010− 108.154090.02552.130.52605320100− 117.754040.00845.830.6945233− 1001017.554030.01133.710.69792140− 1− 100− 17.714120.00944.350.674327500− 11− 107.374050.00742.480.7441106100− 1017.954060.01650.380.572948701001− 17.454110.00539.420.8465280000007.654050.00745.320.72741690− 100− 117.834060.0146.370.64913010100− 10− 17.924100.01649.630.57404511110− 1007.914080.01446.880.597939121001017.894050.01746.610.579744130− 100− 1− 17.734100.00943.250.678725140000007.484050.00742.240.74241215− 10− 10− 107.494030.00631.750.83673160000007.654060.00742.240.7382141700− 11107.494050.00742.240.74221318− 110− 1007.324040.00731.750.801151901− 100− 17.294090.00640.210.7928620001− 1107.854060.02146.320.562649210011− 107.644060.01946.880.5736462201100− 17.554090.01741.310.6046382300− 1− 1107.714050.00946.320.67052824− 10− 10107.514030.00730.360.80714250110017.834060.02245.990.5605502610010− 17.814100.01246.60.6197362710− 10− 107.774070.00942.360.68242428− 100− 10− 17.334080.00836.130.74321129− 1010− 107.424040.01141.190.66302930− 1− 10− 1007.74070.01341.310.63193231− 1010107.514040.01231.210.702519320000007.634050.00744.160.731915330− 110017.894060.01953.940.548851340000007.624060.00844.270.701920351− 10− 1008.284110.02355.130.521154360− 1100− 17.964090.01350.360.59524037001− 1− 107.844070.02152.130.54565238− 100− 1017.564040.01232.170.694722390100− 1− 17.494070.00840.750.71801840 0 0 0 0 0 0 7.65 405 0.004 47.23 0.8975 1 411101007.644060.01443.420.61643742− 1− 101007.594040.0132.540.719017430011107.684060.0243.420.580243440− 1− 10017.874050.01744.640.58674245− 10010− 17.254070.00831.720.772474610− 10107.274070.01442.150.62943347− 1101007.154030.0129.350.75319480− 100117.884060.01644.590.5923414900− 1− 1− 107.754050.01342.390.627335501− 101007.984080.01944.560.5733475101− 10017.514040.00645.990.76798521010108.064080.01733.640.632631530− 1001− 17.814070.01244.280.628734540100117.714050.00944.640.676226 Experimental design along with the outcome response measures. Table 5 depicts the surface roughness ANOVA table and its correlation with each input parameter. Based on the ANOVA table, the pulse on time is the most crucial process parameter, which contributes 38.98% of the variation in surface roughness. Then the order of influence of flushing pressure, peak current, voltage, electrode type and duty factor is descending. Table 5ANOVA outcomes for surface roughness.SourceSum of squaresdfMean squareF− valuep− valuePercentage contributionModel2.6290.290637.87< 0.000188.81SignificantA-A1.1511.15149.69< 0.000138.98B-B0.549010.549071.56< 0.000118.61C-C0.290410.290437.85< 0.00019.84D-D0.180310.180323.50< 0.00016.11E-E0.003810.00380.48880.48820.13F-F0.153610.153620.02< 0.00015.21AC0.192210.192225.05< 0.00016.52AE0.061210.06127.980.00712.07BF0.036110.03614.710.03551.22Residual0.3376440.0077Lack of fit0.3154390.00811.830.2607Not significantPure error0.022150.0044Cor total2.9553 ANOVA outcomes for surface roughness. ANOVA analysis for micro hardness showing the percentage influence of each process parameter (Table 6). The ANOVA results for micro hardness depicts that electrode type is the most crucial parameter which contributes 38.18% to the micro hardness. Next up are pulse-on time, flushing pressure, voltage and peak current. It is observed that duty factor does not affect micro hardness significantly. Table 6ANOVA outcomes for micro hardness.SourceSum of squaresdfMean SquareF-valuep-valuePercentage contributionModel230.851023.0831.86< 0.000188.11SignificantA-A70.04170.0496.67< 0.000126.73B-B9.3819.3812.940.00083.58C-C4.1714.175.750.02091.59D-D7.0417.049.720.00322.69E-E0.000010.00000.00001.00000.00F-F100.041100.04138.08< 0.000138.18BE8.0018.0011.040.00183.05CF4.5014.506.210.01661.72B26.0016.008.290.00622.29F210.80110.8014.910.00044.12Residual31.15430.7245Lack of fit29.82380.78482.940.1145Not significantPure error1.3350.2667Cor total262.0053 ANOVA outcomes for micro hardness. In the same way, the ANOVA table for surface crack density shows that peak current has the highest influence (33.33%), followed by pulse-on time, flushing pressure, electrode type, and voltage. Duty factor has a minimal impact on surface crack density. Similarly, the ANOVA table for recast layer thickness indicates that pulse-on time stands out as the most influential process parameter, accounting for 45.15%, followed by flushing pressure, voltage, peak current, duty factor, and electrode type. Error source analysis and model reliability In the present study on reverse electric discharge machining (R-EDM) of Ti-6Al-4V alloy, the experimental errors were classified as systematic and random errors. Fluctuations of the discharge energy, differences in the conditions of the dielectric flushing, electrode misalignment, thermal instability, drift of the machine calibration and errors in the measurement instruments cause systematic errors. These errors were minimized by calibration of EDM setup before experimentation, maintaining constant dielectric conditions, proper alignment of electrodes and all experiments were performed under controlled laboratory conditions with the same batch of workpiece material and electrodes. Experiments were carried out in a randomized order. Repeated measurements were made for each response parameter to minimize random errors due to spark instability, thermal fluctuations, vibration and measurement variability. Average values of repeated measurements were taken for further statistical analysis. The model adequacy and reliability of the experiment were checked by evaluation of statistical indicators like standard deviation, coefficient of variation (CV%), coefficient of determination (R2), adjusted R2, predicted R2 and adequate precision. Table 7 indicates the coefficient of variance and standard deviation for the Surface roughness. The CV% for each response was less than 10% and standard deviation values were also low. Regression equations are as well generated for each response and the model values are compared with the experimental data and the error percentage is found to be less than 10%. This indicates the high repeatability and precision of the experiment. The validity and sufficiency of the experimental data obtained for this study are therefore confirmed by the statistical indicators. Table 7Statistical table for surface roughness.Std. Dev.0.0876 R 2 0.8857Mean7.67Adjusted R20.8623C.V. %1.14Predicted R20.8000Adeq Precision27.6024 Statistical table for surface roughness. Figure 3 shows the predicted vs. actual plots for (a) surface roughness (b) microhardness (c) surface crack density and (d) recast layer thickness. There is a strong correlation between the experimental and projected findings, as evidenced by the experimental points’ near concentration around the reference line. The small deviation between the predicted and actual values confirms the accuracy, reliability and adequacy of the formulated regression models for prediction of machining responses in the R-EDM process. Furthermore Normal probability and Residual plots are generated for each regression models and are displayed in Fig. 4 (a) surface roughness (b) microhardness (c) surface crack density and (d) recast layer thickness. The residuals are tightly distributed along the straight reference line indicating normal distribution of errors. Predicted vs. actual values plot for (a) Surface roughness (b) Microhardness (c) Surface crack density (SCD) (d) Recast layer thickness (RCT). Residual diagnostic plots for (a) Surface roughness (b) Microhardness (c) Surface crack density (SCD) (d) Recast layer thickness (RCT). Parametric study of the outcome responses Parametric study on the surface roughness Surface roughness, a key characteristic of surface integrity, can impact the performance of biomedical components, affecting protein adsorption, cell adhesion and proliferation. Reverse electrical discharge machining (reverse EDM) provides a precise control of the surface morphology by thermal material removal. Surface plots are created using the Design Expert software to demonstrate the effect of all process parameters on surface roughness. Figure 5a illustrates how peak current and pulse-on time affect surface roughness. As the pulse duration increases, the extended machining time among the tool and the workpiece increases, which increases the material removal rate, causing deeper craters on the work surface, thus leading to higher surface roughness. The surface roughness increases with an increase in the peak current which is due to the increased intensity of the spark energy resulting in an increased Material Removal Rate (MRR). The increase in surface roughness with increase in discharge current and pulse on time is due to development of larger plasma channel and higher discharge energy during spark erosion. The increase in thermal energy results in a higher melting and evaporation rate of the workpiece material, deeper discharge craters and larger molten pools on the machined surface. The irregular shape of the surface and its increased surface roughness result from the dielectric breakdown and the rapid solidification of the melted material. Similar plots are reported in37. Figure 5b shows that the surface roughness is reduced with increasing flushing pressure. This is because higher flushing force helps better removal of debris particles from surface of work piece and hence better surface finish. Figure 5c indicates that the use of copper as the tool electrode is associated with higher surface roughness. Copper has high electrical and thermal conductivity which results in intense spark energy and fast heat transfer which causes aggressive material removal and formation of deep craters, which result in poor surface finish. However, tungsten has low thermal conductivity and high melting point. This limits the amount of energy transferred during each spark. Thus it removes less material and leaves shallower craters – therefore a better surface finish. Copper-tungsten is a composite which offers an intermediate behavior by balancing thermal conductivity and erosion resistance, giving surface quality values between pure copper and pure tungsten. The increase in surface roughness is shown in Fig. 5d with the increase of voltage. Because of the higher voltage and the increased temperature of the spark it is a little higher. Also, the surface roughness increases with the increase in duty factor due to the prolonged machining time and higher Material Removal Rate (MRR). Surface roughness versus (a) peak current and pulse on time (b) peak current and flushing pressure (c) peak current and electrode type (d) peak current and voltage. Parametric study on the micro hardness Microhardness is a key factor that controls the functional performance of biomedical components, and therefore their wear, deformation and fatigue resistance in physiological environments. In applications such as surgical instruments, prosthetic joints and components in contact with biological tissues or fluids, higher micro hardness can improve surface durability and longevity. However, the increased hardness caused by a thick re-solidified layer can result in brittleness or influence the surface properties for biomedical applications. Therefore, optimal microhardness is a prerequisite for mechanical reliability and favorable biological interactions, especially for biomedical devices subjected to load or significant motion. The rapid thermal cycles in reverse EDM lead to the formation of a re-solidified layer on the machined surface, called the white layer. This layer is different in composition and structure from the bulk material due to the formation of oxides, carbides and intermetallic compounds during spark erosion. The thickness and characteristics of this layer are strongly influenced by process parameters such as pulse-on time, peak current and electrode material. Generally, the micro hardness of the thicker white layer is increased, improving the surface durability, but it has to be controlled not to be too brittle or lose the biocompatibility. From Fig. 6a it is observed that the micro hardness increases with increase in pulse on time and peak current. The recast layer increases with the current and pulse-on time. This is because more material is solidified on the base material which increases the micro hardness. This trend agrees with the results reported by other researchers in different studies18,19,37. The very high local discharge temperature in reverse EDM favors the decomposition of dielectric hydrocarbons and the migration of tool electrode material toward the molten surface of the workpiece. This encourages the formation of carbide phases and re-solidified white layer structures enriched in carbon and tool material elements. The rapid quenching effect induced by dielectric cooling also contributes to phase transformation and increased surface hardening, resulting in an increase in the measured micro-hardness values. As depicted in Fig. 6b, the use of tungsten and copper-tungsten electrodes results in higher micro hardness compared to copper. This is attributed to the formation of titanium carbides and tungsten carbides during the machining process. Similar carbide formation, white layer evolution and residual stress behavior during EDM are also reported by previous researchers18–21. Conversely, flushing pressure and duty factor exhibit minimal influence on micro hardness as shown in Fig. 6c and d. The flushing pressure is primarily for flushing the chips from the cutting zone and has little effect on the thermal energy transferred to the workpiece surface. Similarly, the change of duty factor did not generate enough change in heat accumulation or re-solidification behavior in the selected experimental range to significantly change the micro-hardness values. Micro hardness versus (a) peak current and pulse on time (b) peak current and electrode type (c) duty factor and peak current (d) flushing pressure and peak current. Parametric study on the surface crack density Microcracks formed during reverse EDM due to rapid thermal cycling can significantly affect the surface integrity of biomedical components. In applications involving contact with biological tissues or fluids, microcracks may serve as initiation sites for mechanical failure under cyclic loading. Surface crack density is an important metric to characterize severity of these defects. Thus, minimizing microcracks is very important to ensure structural reliability, to avoid adverse biological reactions and to maintain the functional safety of components used in biomedical environments. Figure 7a shows that the spark energy increases with the increasing pulse-on time and peak current, which results in the increase of thermal stresses and the maximization of the surface crack density. Surface crack density (a) peak current and pulse on time (b) peak current and electrode type (c) peak current and flushing pressure d) peak current and duty factor. For copper electrode the Surface Crack Density (SCD) is maximum as compared to the other two tool materials. This is because the copper has a high Material Removal Rate (MRR) which causes intense heating in a localized area and rapid melting and vaporization of the workpiece surface. These extreme thermal stresses lead to the formation of more microcracks, thereby increasing SCD. In contrast, the tungsten electrode generates the lowest SCD, owing to its lower MRR and better thermal resistance, which minimizes thermal shock. The copper-tungsten electrode yields an intermediate SCD value, as its thermal and electrical properties balance the aggressive nature of copper and the stability of tungsten. When tungsten or copper-tungsten is employed, the formation of metal carbides, such as tungsten carbides and titanium carbides, on the Ti-6Al-4V work material surface enhances ultimate strength, thereby minimizing microcracks (see Fig. 7b). Voltage and flushing pressure exhibit minimal influence on surface crack density as shown in Fig. 7c. This is due to the fact that the surface crack formation in EDM is mainly associated with the discharge energy and rapid thermal cycling, which is more affected by the peak current and the pulse-on time. The variation of voltage in the chosen range of experiments has not caused any appreciable variation in the intensity of the spark or the generation of the thermal stress to affect the formation of the crack appreciably. Similarly, flushing pressure is effective only for debris removal and dielectric circulation, with little impact on thermal gradients and development of residual stress that lead to crack initiation. The variation in duty factor in Fig. 7d shows an initial decrease and a slight subsequent increase in surface crack density. At moderate duty factor values, sufficient pulse-off time allows improved cooling and flushing of molten material, thereby reducing thermal stress accumulation and minimizing crack formation. However, at higher duty factors, the reduction in cooling duration increases heat accumulation and thermal loading on the machined surface, leading to higher residual stresses and slight increase in surface crack density. Surface crack density of Ti-6Al-4V with tungsten as tool electrode Surface crack density of Ti-6Al-4V with copper as tool electrode Surface crack density of Ti-6Al-4V with copper-tungsten as tool electrode Figures 8, 9, 10, 11, 12 and 13 displays the SEM images of surface cracks on Ti-6Al-4V workpiece at 10 A and 20 A current for tungsten, copper and copper-tungsten tool electrodes. The images show increase in surface crack density (SCD) with rising peak current. As the peak current increase, the spark energy between the electrode gap also rises, generating extremely high temperatures. Following this intense heating, a dielectric fluid stream cools the machined surface, leading to rapid thermal changes. These thermal cycles generate surface stresses, which when the ultimate strength of the material is exceeded, result in cracking. Finally, the recast layer fails and invades the base metal leading to visible micro-cracks on the surface which increase the SCD. Using a copper electrode yields the highest SCD due to its higher melting and vaporization rates, leading to more surface cracks38. In contrast, tungsten electrodes produce the lowest SCD among the materials tested. Copper-tungsten electrodes generate SCD values that fall between those of the pure copper and tungsten electrodes. SEM image of SCD at run number 4 = 118.886/12,400 = 0.009 (µm/µm2). SEM image of SCD at run number 22 = 205.068/12,400 = 0.017 (µm/µm2). SEM image for SCD at run number 44 = 210.8/12,400 = 0.017 (µm/µm2). SEM image for SCD at run number 25 = 272.8/12,400 = 0.022 (µm/µm2). SEM image for SCD at run number 46 = 173.6/12,400 = 0.014 (µm/µm2). SEM image for SCD at run number 21 = 235.6/12,400 = 0.019 (µm/µm2). Parametric study on the recast layer thickness In the EDM process, the sparking occurs between the tool electrode and work piece. Thus, the temperature is high. The high temperature causes melting and evaporation of small particles from the tool electrode and from the surface of the workpiece. During the pulse-off time molten and debris particles are flushed away by the introduction of dielectric fluid, and some resilient un-flushed molten particles may persist and combine with the dielectric liquid, resulting in the formation of a recast layer on the surface of the work piece. As shown in Fig. 14a, the increase in the peak current rises the recast layer thickness due to higher wear rates of the material and tool. The recast layer initially increases and then stabilizes with a higher pulse-on time. The rise in the peak current results in an increment in recast layer thickness because raised discharge energy, increased material melting and increased tool wear rates. At higher current, larger molten pools are formed which are not fully flushed away during the pulse-off time. This causes deposition and re-solidification of molten material on the machined surface. With higher pulse-on time, the recast layer increases initially and then stabilizes as the longer spark duration increases the heat input and molten material formation but at a certain limit the flushing action and thermal equilibrium reduce further significant increase in recast layer thickness. Contrarily, Fig. 14d demonstrates that voltage and duty factor do not follow a similar trend, likely due to irregularities in Recast Layer Thickness (RCT) with increased fluctuation in shape. The change of voltage does not significantly affect the thermal energy transferred to the workpiece surface in the given experimental range. Likewise, the duty factor variation created a fluctuating heating and cooling cycle, which caused the unstable molten material deposition and irregular re-solidification behavior. Hence, the evolution of the recast layer thickness showed oscillatory variations instead of monotonic increase/decrease. According to Fig. 14b, an increase in flushing pressure results in a slight reduction in recast layer thickness. This is attributed to higher pressure flushing away more molten debris particles with the dielectric fluid, thereby forming a thinner recast layer. The copper tool electrode gives a thicker recast layer than tungsten and copper-tungsten, displayed in Fig. 14c. This is because of the high thermal conductivity of copper which results in higher Material Removal Rate (MRR) and more material deposition on the Ti-6Al-4V surface. Recast layer thickness (a) peak current and pulse on time (b) peak current and flushing pressure (c) peak current and electrode type (d) peak current and duty factor. Recast layer thickness of Ti-6Al-4V with tungsten as tool electrode Recast layer thickness of Ti-6Al-4V with copper as tool electrode Recast layer thickness of Ti-6Al-4V with copper-tungsten as tool electrode Figures 15, 16, 17, 18 and 19, and 20 present SEM images illustrating the recast layer thickness (RCT) on a Ti-6Al-4V workpiece at current settings of 10 A and 20 A for tungsten, copper, and copper-tungsten electrodes. The images indicate that RCT increases with higher current values. It is important to note that the copper electrode has the highest RCT out of the three materials, which also has higher material removal rates (MRR) than the other two materials39. The increased RCT is due to the increased retention of the material on the machined surface during flushing. However, the tungsten electrode has the least RCT due to the lower MRR. The RCT from the copper-tungsten electrode is in an intermediate range in relation to the other two electrodes. SEM image for RCT at run number 4 = 8099/240.151 = 33.71 μm. SEM image for RCT at run number 22 = 8099/205.430 = 39.42 μm. SEM image for RCT at run number 44 = 8099/182.620 = 44.35 μm. SEM image for RCT at run number 25 = 8099/176.672 = 45.83 μm. SEM image for RCT at run number 46 = 8099/190.654 = 42.48 μm. SEM image for RCT at run number 21 = 8099/178.706 = 45.32 μm. Hybrid optimization strategy In this paper, the ARAS method is utilized to synthesize multiple responses into one equivalent response by computing the degree of utility. Using this utility degree, we could rank the alternatives from the most favorable to the least favorable. As shown in Table 3, experiment no.40 has the maximum utility degree among the fifty-four utility degree values which is the best option. Further we propose to integrate the Firefly Algorithm (FA) for enhancing the efficiency of ARAS approach. To this end, we propose a hybrid optimization model for the linkage between ARAS and Firefly algorithm. This model is based on a mathematical formulation obtained by regression analysis relating to the parameters of the process and the degree of utility. This relationship is then used as an objective function in the Firefly algorithm. The quadratic model in the regression analysis was given by the following equation no.8. Where, A represents pulse on time, B indicates flushing pressure, C indicates peak current, D represents voltage, E indicates duty factor and F denotes electrode. The developed quadratic regression model establishes the relation between the machining parameters and the utility degree obtained by ARAS approach. This regression equation is used as the objective function to be optimized by Firefly algorithm. Firefly Algorithm (FA) The Firefly Algorithm (FA) is an optimization technique inspired by nature, which applies the flashing behavior of fireflies to find the optimal solutions for optimization problems. It was proposed by Xin-She Yang in 2008 and has become popular for its simplicity and efficiency in solving a wide range of optimization problems. In Firefly Algorithm (FA), the fireflies of the initial population are randomly distributed in the limits of the process parameters. These parameters are updated at each iteration according to the randomness (a), light absorption (b) and the distance between the fireflies (r). Consequently, these parameters evolve and are continually assessed using the objective function. The optimization process aims to correlate the target function value with the previous iteration’s outcome, iterating until an optimal performance parameter is achieved. Control over the search process is managed by setting a maximum number of iterations (tmax)39,40. The steps of Firefly algorithm are given below: Initialize the positions of the fireflies randomly within the defined limits of the problem variables and set the control parameters for the Firefly Algorithm (FA). Initialize the positions of the fireflies randomly within the defined limits of the problem variables and set the control parameters for the Firefly Algorithm (FA). Specify the objective function and define the boundaries for the variables pertinent to the given optimization problem. Specify the objective function and define the boundaries for the variables pertinent to the given optimization problem. Evaluate the brightness (or light intensity) of each firefly in the population, which represents the value of the objective function. Evaluate the brightness (or light intensity) of each firefly in the population, which represents the value of the objective function. Identify the best-performing firefly with the highest intensity value. Identify the best-performing firefly with the highest intensity value. Calculate the distance between each firefly and the best firefly, then update the position of each firefly accordingly. Calculate the distance between each firefly and the best firefly, then update the position of each firefly accordingly. Re-evaluate the intensities of the fireflies based on their updated positions. Re-evaluate the intensities of the fireflies based on their updated positions. Arrange and rank the fireflies according to their intensities and positions. Arrange and rank the fireflies according to their intensities and positions. Choose the best firefly for the current iteration and replace the previous iteration’s best firefly if the current one has a higher intensity. If not, keep the previous solution. Choose the best firefly for the current iteration and replace the previous iteration’s best firefly if the current one has a higher intensity. If not, keep the previous solution. Update the results. If the maximum iteration limit is reached, proceed to step 10; if not, return to step 5. Update the results. If the maximum iteration limit is reached, proceed to step 10; if not, return to step 5. The intensity value of the firefly at the end of the iterations represents the optimal solution to the optimization problem. The intensity value of the firefly at the end of the iterations represents the optimal solution to the optimization problem. In the Firefly Algorithm (FA), the intensity (I) serves as an indicator of the solution’s fitness function (f), evolving according to Equation no. 939. Here, I0 represents the initial light intensity and c stands for the light absorption coefficient. The attractiveness (b) of fireflies correlates with their light intensities (I). Therefore, formulation akin to equation describes the attractiveness (b) as follows in equation no. 10. In equation b0 denotes the attractiveness when the distance r is zero. The spatial separation between fireflies ‘i’ and ‘j’ positioned at si and sj respectively is calculated through the Euclidean distance, as depicted in Equation no. 11. Here, n represents the model’s dimensionality. Less attractive fireflies (indexed as ‘i’) gravitate towards the more attractive ones (indexed as ‘j’). This directional movement updates the FA parameters in accordance with the equation no. 12. Here, ei stands for random number, and the update involves three terms: the current position of the ith firefly, the attractiveness towards another more attractive firefly, and a randomization constraint (a) along with the random number ei. Table 8. Shows the parameters selected for Firefly algorithm. The parameters of the Firefly Algorithm were chosen after several trial runs to ensure a stable convergence and better optimization performance. The population size and maximum iterations were selected to compromise between computational efficiency and solution accuracy. Randomness coefficient (α) described the ability of exploration of algorithms. Attractiveness coefficient (β₀) and light absorption coefficient (γ) described the movement and convergence behavior of fireflies during optimization. Table 8Parameters selected in Firefly Algorithm.ParameterSymbolValuePopulation sizen20Maximum iterationstmax100Randomness coefficientα0.25Attractiveness coefficientβ₀1Light absorption coefficientγ1Dimension of problemd6Stopping criteria–Maximum iteration/convergence Parameters selected in Firefly Algorithm. ARAS based Firefly Algorithm The utility degree values from the ARAS approach are correlated with process parameters via non-linear regression (Eq. 8). A hybrid ARAS-based Firefly Algorithm is developed to optimize the utility degree, treating electrode material as a qualitative parameter and voltage, peak current, pulse-on-time, flushing pressure, and duty factor as quantitative parameters within the range [−1, 1]. For electrode material, values from − 1 to 0.29 represent tungsten, −0.3 to 0.59 represent copper-tungsten, and 0.6 to 1 represent copper. Subsequently, Eq. 8 is employed as the objective function within the Firefly Algorithm, implemented in Matlab. The optimization problem is carried out on a Ryzen processor in an HP laptop to evaluate the effectiveness of the Firefly Algorithm. After tuning the algorithm’s parameters, the solution converges, and the optimal solution is obtained with parameters set to A = 400, B = 0.3, C = 10, D = 200, E = 85, and F = tungsten tool. It is to be noted that the utility degree has shown a major enhancement. The hybrid method (ARAS-based Firefly Algorithm) has predicted a value of 0.9021 which is higher than the previous best value obtained using ARAS method. Figure 19 shows the convergence plot of this procedure. The convergence characteristic of ARAS based Firefly Algorithm in the optimization is shown in Fig. 21. It can be observed that the utility degree was gradually improved with the iterations and slowly converged to the stable state after several iterations, which demonstrates the effective convergence and stability of the proposed optimization framework. The convergence characteristics show the capability of the Firefly Algorithm to efficiently search the global optimum solution of the multi-response optimization problem. Convergence plot for utility degree. The results obtained by the ARAS-based Firefly Algorithm were verified by a confirmatory test under the optimized machining conditions, and the results revealed an overall improvement of 6.86% in terms of surface roughness (8.89%), micro hardness (3.33%), SCD (8.75%), and RCT (6.46%). Detailed results are provided in Table 9. Table 9Confirmative test results and comparisons of responses generated by ARAS and ARAS based Firefly Algorithm.MethodAµsBCAmpDVE%FSurface roughnessMicro hardnessSCDRCTUtility degreeOverall improvementARAS4000.31520085Copper-tungsten7.654050.00447.230.8975ARAS based Firefly Algorithm4000.31020085Tungsten6.974350.0036544.170.9021% improvement8.89%3.33%8.75%6.46%6.86% Confirmative test results and comparisons of responses generated by ARAS and ARAS based Firefly Algorithm. The proposed ARAS-based Firefly Algorithm combines the ranking capability of the ARAS method and the strong global search capability of the Firefly Algorithm for effective multi-response optimization. The proposed hybrid approach shows better convergence, less computational complexity and better handling of conflicting machining responses as compared to the conventional optimization methods such as Grey Relational Analysis (GRA), Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and NSGA-II as reported in the previous EDM studies. The application of Firefly Algorithm improves the exploration and exploitation capacity of the optimization process and enhances the utility degree obtained from ARAS, unlike the MCDM approaches used individually. The convergence stability exhibited in Fig. 21 further validates the proposed hybrid optimization framework’s robustness and effectiveness for reverse EDM of Ti-6Al-4V alloy. Conclusions This work shows the effective application of reverse electrical discharge machining (R-EDM) process for machining on the Ti-6Al-4V work piece and producing complex surface features useful for biomedical usages. A novel hybrid optimization framework is used for combining response surface methodology (RSM), additive ratio assessment (ARAS) and fire fly algorithm and is experimentally validated. The proposed approach achieved and overall performance enhancement of 6.86 percentages confirming its effectiveness in enhancing its machining results and process efficiency. The major findings are highlighted below: Tungsten electrodes showed the least surface roughness, which indicates better machining ability for high-precision biomedical components. Tungsten electrodes showed the least surface roughness, which indicates better machining ability for high-precision biomedical components. The improved surface smoothness achieved under the optimized machining conditions is beneficial for the surface characteristics associated with biomedical applications, and also help minimize tissue irritation in the implant related applications. The improved surface smoothness achieved under the optimized machining conditions is beneficial for the surface characteristics associated with biomedical applications, and also help minimize tissue irritation in the implant related applications. Tungsten electrodes resulted in the thinnest thickness of the recast layer, hence reducing the thermal damage and improving the surface integrity. Tungsten electrodes resulted in the thinnest thickness of the recast layer, hence reducing the thermal damage and improving the surface integrity. Tungsten electrodes showed lower crack density on the machined surface due to their higher melting point and stable discharge characteristics. Tungsten electrodes showed lower crack density on the machined surface due to their higher melting point and stable discharge characteristics. Tungsten electrodes resulted in the maximum micro-hardness values, which indicates better surface strengthening, and wear resistance. Tungsten electrodes resulted in the maximum micro-hardness values, which indicates better surface strengthening, and wear resistance. Electrode material was identified as the most influential parameter on micro-hardness while pulse-on-time and peak current had influencing effect on the surface roughness and thermal damage characteristics. Electrode material was identified as the most influential parameter on micro-hardness while pulse-on-time and peak current had influencing effect on the surface roughness and thermal damage characteristics. The machined surfaces are more suitable for orthopedic and dental implant applications by the improved micro-hardness and reduced surface defects under the optimized conditions. The machined surfaces are more suitable for orthopedic and dental implant applications by the improved micro-hardness and reduced surface defects under the optimized conditions. The convergence stability and enhanced utility degree achieved with the ARAS-Firefly framework confirm the effectiveness, robustness and superior optimization ability of the proposed hybrid approach over traditional standalone optimization methods. The convergence stability and enhanced utility degree achieved with the ARAS-Firefly framework confirm the effectiveness, robustness and superior optimization ability of the proposed hybrid approach over traditional standalone optimization methods. Future scope The present study lays the groundwork for optimizing R-EDM of Ti-6Al-4V, with future prospects including hybrid techniques (U-EDM, L-EDM), advanced electrode materials, and micro/nano-scale applications. Emphasis on sustainable machining using eco-friendly dielectrics and additive-manufactured electrodes can further enhance process efficiency and surface integrity. The present study has shown good surface integrity properties that are of interest for biomedical applications but further studies on corrosion resistance, wettability, cytotoxicity, cell adhesion behavior and detailed biocompatibility analysis are needed to fully establish the biomedical performance of the machined Ti-6Al-4V surfaces. Author contributions Renu Kiran Shastri – Conducted the experimental work and drafted the manuscript.Chinmaya P. Mohanty – Conceived the idea, supervised the research, and critically reviewed the manuscript.Kishore Kumar Mahato – Performed the optimization studies related to the Reverse EDM (REDM) process.Pravat Ranjan Pati – Analysed the experimental work and formatted the final version of the manuscript. Funding Open access funding provided by Vellore Institute of Technology. Data availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.