Artificial neural network–based and response surface methodology–based predictive models for material removal rate and surface roughness during electro-discharge diamond grinding of Inconel 718

Abstract

Hybrid machining processes growing popularity in the processing of difficult-to-cut materials due to their distinct merits over individual machining processes attributed by an amalgamation of two or more machining mechanisms simultaneously. This research study deals with the response surface methodology and artificial neural network with backpropagation algorithm–based mathematical modeling of electro-discharge diamond grinding of Inconel 718 superalloy. The matrix experiments were designed based on central composite design. The wheel speed, current, pulse-on-time, and duty factor were chosen as control factors, while material removal rate and average surface roughness (R_a) were chosen as performance parameters. The analysis of variance test shows that the wheel speed is the major factor influencing both the material removal rate and the R_a and contributes 89.03% and 79.10% on material removal rate and R_a, respectively, followed by current which contributes 4.43% and 8.38% on material removal rate and R_a, respectively. The modeling and predictive abilities of developed artificial neural network model (4-24-2) were related to the response surface methodology model using root mean square error and absolute standard deviation. The predicted values of material removal rate and R_a by response surface methodology and artificial neural network are in close agreement with the actual experimental results.

Keywords

Electro-discharge diamond grinding hybrid machining artificial neural network response surface method Inconel 718

Introduction

Inconel 718 is extensively used superalloy in aerospace industries due to its high-temperature strength, excellent fatigue, and corrosion resistance. Extremely, though, and thermally resistant nature of Inconel 718 makes it one of the difficult-to-cut materials. Hybrid machining processes (HMPs) can be used to overcome problems of machining of difficult-to-cut materials.¹ HMPs amalgamate two or more machining mechanisms together to take combined advantages and minimize the potential disadvantages found in machining process when applied individually.² As compared to the individual process, the vital improvements in performance characteristics can be observed in a hybrid process.³

Electro-discharge machining (EDM) is non-contactable material subtraction process suitable for machining of difficult-to-machine materials. Although in EDM, discharge energy considerably influences the surface integrity of workpiece. Microcracks induce residual stresses and change the microhardness of the sub-surface, which may occur due to thermal action in EDM.⁴ The electro-discharge diamond grinding (EDDG) is a compound type HMP and is proposed by Koshy et al.^5,6 to overcome surface integrity allied issues in EDM. The EDDG combines electro-discharge erosion action of EDM and mechanical abrasion of diamond grinding (see Figure 1) and significantly improves material removal rate (MRR) as compared to the EDM process.⁷

Figure 1.

Mechanism of material removal in EDDG process.⁸

Koshy et al.^5,6 reported that electrode rotations and efficient flushing attribute to noteworthy enhancement in MRR and surface finish. Furthermore, it is reported that the deposition of recast layer and surface modification during the process may increase the wear resistance of workpiece at low discharge power. Kozak⁹ reported the use of copper-bonded diamond wheel in EDDG of polycrystalline diamond and aluminum carbide and evaluated the effects of speed of the wheel, pulse current, and pulse-on-time on MRR and R_a. In this study, he reported self-dressing phenomena of grinding wheel where the grinding wheel is being dressed continuously without the need of interrupting the machining process. Yadav et al.¹⁰ used copper-bonded diamond wheel and experimentally evaluated the effects of wheel speed, pulse-current, and pulse-on-time on MRR in EDDG of high-speed steel (HSS) and reported that wheel speed and pulse-on-time are significant parameters affecting MRR. Singh et al.^7,8,11 in their recent studies reported bronze-bonded diamond wheel in EDDG of HSS and WC–Co composite. They studied the effects of pulse current, pulse-on-time, duty factor, and wheel revolutions per minute (RPM) on MRR, R_a, and wheel wear rate (WWR) and shown that small current and less wheel speed were not advisable for the EDDG process in machining of the WC–Co composite. They observed that wheel glazing at low current and high wheel speed slumped the MRR capability of the diamond wheel.

Kumar and Choudhury¹² applied the artificial neural network (ANN) technique to predict wheel wear and R_a during EDDG of HSS and compared predicted results with regression models. They reported that the prediction ability of the regression model was poorer than that of the ANN model. Yadav and Yadava¹³ reported the application of slotted EDDG in face grinding mode to overcome drawbacks in EDDG such as the gathering of debris into the gap, aberrant arcing, and large wheel wear. A joint approach of ANN and non-dominated sorting genetic algorithm-II optimizes the process parameters of surface electro-discharge diamond grinding (S-EDDG) of metal matrix composite proposed by Yadav and Yadava.¹⁴ ANN, genetic algorithm, and grey relational analysis methods have been suggested by Shrivastava and Dubey¹⁵ for multi-objective optimization of the EDDG process performance during grinding of the copper–iron–graphite composite.

From the literature review, it is observed that no plausible studies were conducted on machining of Inconel 718 superalloy using EDDG process. In this study, response surface methodology (RSM) and ANN models were established to predict the effect of several machining parameters on MRR and R_a using experimental results obtained based on the central composite design (CCD) design. Analysis of variance (ANOVA) test performed to evaluate significant factors in EDDG of Inconel 718. The effectiveness of developed models in predicting machining performance is compared at last.

Experimentation

Surface grinding setup was fabricated on ENC-35 EDM machine that was used to perform experiments. Figure 2 shows actual photograph in-house fabricated EDDG setup. This setup consists of bronze–diamond abrasive grinding wheel, direct current motor (HP: 2, RPM: 4600 max), horizontal shaft (diameter, 32 mm), V-pulley (diameter, 32 mm) and bearing, and alternating current reversible synchronous motor (RPM: 60, 50 Hz).

Figure 2.

Photograph of EDDG setup.

The relative motion between grinding wheel and material to be machined has been attained by reciprocating the machine bed with the help of an automatic feed arrangement consisting of a reversible AC synchronous motor control circuit, a relay switch, two limit switches, and regulated power supply. The workpiece was fixed on machine bed using the fixture and then dipped in the dielectric fluid (EDM oil) which majorly consists of paraffinic hydrocarbon oil. The bronze–diamond wheel was attached to the ram of the machine through special attachment, and the servo system of the Z-axis numeric controlled (ZNC) EDM machine was used to uphold the predetermined gap between the two electrodes, that is, the diamond wheel and the flat workpiece. The details of metal-bonded diamond wheel are given in Table 1. The surface grinding on workpiece was achieved by rotating the grinding wheel about the horizontal axis perpendicular to the downward motion of servo system of EDM machine with the help of variable speed permanent magnet direct current (PMDC) motor through a V-belt and pulley arrangement. The speed of the PMDC motor has been controlled with the help of a DC drive.

Table 1.

Details of metal-bonded grinding wheel.

Description	Specification
Diamond wheel diameter	100 mm
Thickness	10 mm
Abrasive used	Diamond
Bore diameter	32 mm
Bonding material	Bronze
Concentration	75%
Grit size	80/100
Depth of abrasive	3 mm

All the experiments were carried on the flat rectangular surface (60 mm × 10 mm) of Inconel 718 in surface grinding mode of EDDG setup. The composition of Inconel 718 has been found after conducting the optical emission scanning (OES) test and is shown in Table 2.The wheel speed (S), pulse-current (C), pulse-on-time (P), and duty factor (DF) were selected as control factors, while MRR and R_a were taken as performance parameters. The effective range of each factor was chosen on the basis of pilot experimentation. Table 3 shows the actual and coded values of control factors selected for the experiments. In the present case, three levels of four control factors with seven central point of standard CCD matrix have been selected, which require total 31 experimental runs with 2^p factorial runs, 2^p axial runs, and seven center points (where p is the number of control factor). This design is usually used for fitting the second-order model. It is essential to second-order model to deliver good prediction throughout the region of interest. The experimental matrix (coded variables) and experimental results are shown in Table 4. High precision electronic digital balance (WENSAR HPB-310 Model) with the readability of 1 mg has been used to observe the weight difference in workpiece before and after machining. The MRR was calculated by the difference between initial weight and final weight of the workpiece over a period of machining time. To measure the R_a value of machined surface after each experiment surface roughness tester (Taylor Hobson Surtronic-25) has been used. Surface roughness is measured perpendicular to grinding direction with 4-mm evaluation length. The grinding wheel speed was measured by a digital tachometer in RPM.

Table 2.

Chemical composition of Inconel 718 based on OES test.

Element	Mo	Co	Al	C, 165	Si, 251	Mn, 403	P, 185	Cr, 267	Fe, 406	V, 411	Nb, 405	W, 429	B, 182	Ti, 337	Al, 308	Zr, 339	Ni, 216
Percentage (%)	10.8	0.19	0.16	0.034	0.105	0.031	0.004	23.5	4.2	0.054	3.04	0.2	0.001	0.699	0.185	0.099	57

Table 3.

Experimental input parameter and their levels.

Input factors			Coded level
Description	Unit	Symbol	−1	0	1
Wheel speed	RPM	S	600	900	1200
Current	A	C	3	6	9
Pulse-on-time	µs	P	50	100	150
Duty factor	%	DF	0.58	0.63	0.70

Table 4.

Experimental layout as per CCD showing coded input parameters and responses along with comparison of actual and predicted values by ANN and RSM of MRR and R_a.

Run order	S	C	P	DF	MRR (mm³/min)			R_a (µm)
Run order	S	C	P	DF	Experimental	ANN	RSM	Experimental	ANN	RSM
1	−1	0	0	0	5.64	6.18	4.61	2.82	3.36	3.34
2	0	0	−1	0	11.54	12.04	11.25	6.91	5.67	6.01
3	−1	1	1	−1	8.25	8.68	7.97	5.21	5.00	4.21
4	−1	−1	1	−1	5.03	5.10	5.01	2.52	3.00	2.46
5	0	0	0	1	10.86	11.89	11.25	5.43	5.74	6.01
6	1	−1	−1	1	18.95	15.95	18.31	8.20	7.05	7.82
7	1	−1	1	1	19.25	16.58	18.31	7.09	7.80	7.82
8	−1	−1	−1	−1	4.27	4.47	5.01	2.14	2.25	2.46
9	0	0	0	0	10.95	12.35	11.25	5.98	6.04	6.01
10	1	−1	1	−1	19.45	17.43	18.31	9.73	8.37	7.82
11	0	0	0	0	12.25	12.35	11.25	6.13	6.04	6.01
12	1	1	−1	−1	22.37	20.39	21.27	11.19	9.62	9.56
13	1	1	1	−1	21.45	21.02	21.27	10.24	10.37	9.56
14	−1	−1	1	1	4.87	4.24	5.01	2.44	2.42	2.46
15	0	0	0	0	11.75	12.35	11.25	6.10	6.04	6.01
16	0	0	0	0	11.65	12.35	11.25	5.98	6.04	6.01
17	0	0	0	0	11.24	12.35	11.25	5.88	6.04	6.01
18	0	0	1	0	11.54	12.66	11.25	5.77	6.42	6.01
19	−1	1	−1	−1	9.09	8.06	7.97	4.55	4.25	4.21
20	0	0	0	0	11.16	12.35	11.25	5.98	6.04	6.01
21	0	1	0	0	12.09	14.14	14.62	5.35	7.04	6.89
22	1	1	−1	1	20.24	19.54	21.27	8.78	9.05	9.56
23	0	0	0	−1	10.43	12.74	11.25	5.22	6.31	6.01
24	−1	1	−1	1	6.64	7.20	7.97	3.32	3.67	4.21
25	1	1	1	1	22.75	20.16	21.27	9.42	9.80	9.56
26	0	−1	0	0	12.83	10.56	11.66	6.42	5.04	5.14
27	1	−1	−1	−1	15.74	16.81	18.31	6.94	7.62	7.82
28	−1	−1	−1	1	4.53	3.62	5.01	2.27	1.67	2.46
29	1	0	0	0	16.54	18.52	17.90	7.25	8.73	8.69
30	0	0	0	0	10.76	12.35	11.25	5.87	6.04	6.01
31	−1	1	1	1	8.73	7.83	7.97	5.37	4.42	4.21

ANN: artificial neural network; RSM: response surface methodology; MRR: material removal rate.

Results and discussion

ANOVA

ANOVA technique was performed to test the adequacy of the established models, and the outcomes of response surface model fitting are given in Tables 5 and 6. The test for significance of the regression models, the test for significance on individual model coefficients, and the lack-of-fit test were performed using the Minitab 16 software package. The backward elimination was performed to eliminate the insignificant model terms, and the resulting ANOVA (Tables 5 and 6) for the response surface quadratic models summarizes the ANOVA of each response and shows the significant model terms. The adequacy measures R², adjusted R², and predicted R² are also shown in the same tables. The goodness of fit of the model can be perceived by the coefficient of determination R². In this case, all the values of coefficient of determination R² are closer to 1. Clearly, we must have 0 ≤ R² ≤ 1, with larger values being more desirable. The adjusted coefficient of determination R² is a variation of the ordinary R² statistic that reflects the number of factors in the model. The entire adequacy measures are closer to 1, which is in reasonable agreement and indicate adequate models.

Table 5.

ANOVA results for MRR model (only significant terms).

Source	Degree of freedom (df)	Sum of squares (SS)	Mean square (MS)	F value	p value probability > F	Percentage contribution
Model	3	862.30	287.43	245.51	0.00	96.46
Linear	2	835.45	417.72	356.80	0.00	93.46
S	1	795.87	795.87	679.80	0.00	89.03
C	1	39.58	39.58	33.80	0.00	4.43
Square	1	26.85	26.85	22.93	0.00	3.00
C × C	1	26.85	26.85	22.93	0.00	3.00
Residual error	27	31.61	1.17
Lack of fit	5	12.09	2.418	2.72	0.046
Pure error	22	19.52	0.887
Total	30	893.91
Standard deviation = 1.08			R ² = 96.46%
PRESS = 43.10			Adjusted R² = 96.07%
			Predicted R² = 95.18%

PRESS: prediction error sum of squares.

Table 6.

ANOVA results for R_a model (only significant terms).

Source	Degree of freedom (df)	Sum of squares (SS)	Mean square (MS)	F value	p value probability > F	Percentage contribution
Model	2.00	142.84	71.42	97.90	0.00	87.49
Linear	2.00	142.84	71.42	97.90	0.00	87.49
S	1.00	129.15	129.15	177.04	0.00	79.10
C	1.00	13.69	13.69	18.77	0.00	8.38
Residual error	28.00	20.43	0.73
Lack of fit	6.00	7.69	1.28	2.22	0.08
Pure error	22.00	12.73	0.58
Total	30.00	163.27
Standard deviation = 0.854111			R ² = 87.49%
PRESS = 26.2583			Adjusted R² = 86.6%
			Predicted R² = 83.92%

PRESS: prediction error sum of squares.

The value of the determination coefficient (R²= 96.46%)_MRR and ${(R^{2} = 87.49 %)}_{R_{a}}$ indicates that only less than 3.54% and 12.51% of the total variations in MRR and R_a, respectively, are not clarified by the models. The large value of the adjusted determination coefficient (adjusted R²= 96.07%)_MRR and ${(R^{2} = 86.6 %)}_{R_{a}}$ assures significance of the models. Predicted R² is also in conformity with the adjusted R² in both cases. Lack of fit is non-significant as required. The normal distribution of residuals (see Figure 3) confirms there are no obvious patterns indicating an outstanding adequacy of the regression model. The ANOVA (Tables 5 and 6) also shows the model terms std. dev. and PRESS. Standard deviation (“Std. Dev.”) is a square root of the error mean square, and “PRESS” stands for “prediction error sum of squares,” and it is a measure of how well the model for the experiment is likely to predict the responses in a new experiment. Small values of PRESS are desirable. In all the cases, the values of PRESS are considerably low.

Figure 3.

(a) Normal probability plot of residual for MRR and (b) normal probability plot of residual for R_a.

ANOVA test recognizes the significant control factors that affect the MRR and R_a of Inconel 718 in EDDG in surface grinding mode at 95% confidence level. Statistically, the F test delivers a choice at a 95% confidence level to assess the significant parameters on deciding the performance characteristics. The results of ANOVA after backward elimination method for the machining performance are presented in Tables 5 and 6. From the results of ANOVA, it is observed that the wheel speed (S) is the major affecting factor to decide both the MRR and the R_a and contributes 89.03% and 79.10% on MRR and R_a, respectively, whereas current (C) contributes 4.43% and 8.38% on MRR and R_a, respectively. The similar observations were also reported by Yadav et al.¹⁰ and Singh et al.⁷ Pulse-on-time (P) and duty factor (DF) have less significance on the MRR and R_a of Inconel 718 in EDDG process.

In EDDG, the wheel speed is observed to be major factor influencing both the MRR and the R_a due to the fact that due to rotation of grinding wheel, the number of diamond abrasive particle comes in contact with the workpiece, thereby abrading more volume of the workpiece. Thus, both MRR and R_a will get largely influenced by the wheel speed in comparison to current, pulse-on-time, and duty factor. The rotations of the wheel also improve the flushing conditions at inter-electrode gap (IEG) and effectively ejecting the molten material and debris accumulation in IEG.^14,16 At high current, more discharge energy will go into the workpiece and melting more material from the workpiece and forming larger craters.¹⁶ Thus, current is also an influencing factor on both the MRR and the R_a. Therefore, the wheel speed and current are observed to be a factor influencing the MRR and R_a.

Prediction of MRR and R_a in EDDG

ANN

ANN is one of the widely used dominant computational tools and used to explain complex functions in numerous applications. ANN has been efficaciously adopted for solving various problems in areas like fault diagnosis, process identification, property estimation, data smoothing and error filtering, product design and development, optimization, and estimation of activity coefficients.¹⁷ The neural network is an immensely parallel distributed processing method consisting of greatly interconnected neural computing elements having the capability to learn, obtain information, and make it available for use.¹⁸ Therefore, by appreciating the computational capability of a multilayered neural network, it has been adopted in the modeling of the EDDG process and used to predict the MRR and R_a in machining of Inconel 718.

In this study, the effects of wheel speed (S), current, pulse-on-time (P), and duty factor (DF) on MRR and R_a were statistically investigated in EDDG of Inconel 718. The best ANN architecture was designed by MATLAB software. The neuron in the input layer corresponds to wheel speed (S), current, pulse-on-time (P), and duty factor (DF). The output layer corresponds with MRR and R_a. In this model, the inputs are fully connected to the hidden layer and the hidden layer neurons are fully connected to the outputs. After exhaustive trails, ANN model with one hidden layer comprising 24 neurons and 4 input and 2 output neurons as shown in Figure 4 was developed on the basis of performance of the network. The training of network was done using Levenberg–Marquardt (LM) backpropagation neural network (BPNN) algorithm for fast supervised learning, and 70% of data have been used for training, whereas 15% of data were used for testing and 15% for validation each. The details of the ANN model developed are shown in Table 7. Mean square error (MSE) determined to evaluate the performance of the network which can be calculated as

MSE = \frac{1}{M \times N} \sum_{i = 1}^{N} \sum_{j = 1}^{M} (d_{j} - y_{j})

(1)

where d_j is the desired output of the jth neuron, y_j is the predicted value of jth neuron, N is the total number of training pattern, and M is the number of output nodes.¹⁹

Figure 4.

ANN structure 4-24-2.

Table 7.

ANN model parameters.

		Learning conditions
Object model	MRR, R_a	Learning scheme	Supervised learning
Input neurons	S, C, P, and DF	Learning rule	Gradient descent rule
Output neuron	MRR, R_a	Hidden neurons	24
Network structure		Output neurons	Two
Network type	Feed-forward back-propagation	Sample pattern vector	70% (train), 15% validation, and 15% testing
Transfer function	Purelin	Learning rate	α: 0.0001
Training function	TRAINLM	Marquardt adjustment parameter	Mu: 0.05
Learning function	LEARNGDM	Maximum number of epochs to train	500
		Goal	0.0001

MRR: material removal rate.

Response surface method

RSM commonly used for establishing the relationship between control factors and response variables. To fit the second-order polynomial equation to the experimental data and to recognize the relevant model terms for MRR and R_a, a backward elimination method was used in the Minitab 16 software. By performing the backward elimination method, the insignificant model terms were eliminated, and a reduced model for each response having only the significant model terms was obtained. The CCD matrix was used to obtain MRR and R_a responses in EDDG of Inconel 718 after 31 data measurement from machining conditions. Codings +1 and −1 were used as the upper limit and lower limit of input factors, respectively. RSM fits polynomial models for the existing data into equation

y = β_{0} + \sum_{i = 1}^{k} β_{i} x_{i} + \sum_{i = 1}^{k} β_{ii} x_{i}^{2} + \sum_{i} \sum_{j} β_{ij} x_{i} x_{j}

(2)

where y is the predicted response, β₀ is a constant, β_i is the linear coefficient, β_ii is the squared coefficient, β_ij is the cross product coefficient, and k is the number of factors.¹⁹

The final models were developed using only significant factors through the backward elimination method and were given by equations (3) and (4), respectively, as

MRR = 11.255 + 6.65 S + 1.48 C + 1.88 C^{2}

(3)

R_{a} = 6.014 + 2.68 S + 0.87 C

(4)

The experimental results compared with the ANN and RSM predicted results (see Table 4). It was observed that the predictions by ANN as well as RSM have a good agreement with the experimental results.

Surface plots for MRR and R_a in three dimensions (3D) by varying the two factors wheel speed and current at constant pulse-on-time and duty factor for EDDG of Inconel-718 are as shown in Figure 5(a) and (b). These figures indicate that the MRR, as well as R_a, increases with the increase in wheel speed and current.

Figure 5.

Surface plots of (a) MRR versus wheel speed (S) and current (C) and (b) surface plot of R_a versus wheel speed (S) and current (C).

Five confirmation experiments that have been executed to inquire the validation of the experimental results at diverse machining conditions are shown in Figure 6(a) and (b) keeping pulse-on-time and duty factor constant. The levels (coded) of control factors, viz., wheel speed and current chosen for confirmation experiments are shown in Figure 6(a) and (b). The experimental results and predicted values by both RSM and ANN models have less error, suggesting that established models are appropriate and reliable in the prediction of MRR and R_a.

Figure 6.

(a) Comparison of the experimental versus predicted values by ANN and RSM for MRR for validation runs. (b) Comparison of the experimental versus predicted values by ANN and RSM for R_a for validation runs.

Comparison of predicted MRR and R_a by ANN and RSM

The predicted values of MRR and R_a by both ANN and RSM models are listed in Table 4. The prediction performance of ANN and RSM models in terms of absolute standard deviation (ABSD) and percentage root mean square error (RMSE) is shown in Table 8. Figures 7 and 8 demonstrate a correlation between experimental responses and predicted results from RSM and ANN. The comparative analysis demonstrates that both models implemented well offering stable responses in predicting the independent factors with respect to the response. However, as compared to RSM method, the ANN technique offered superior fitting to the output responses. Table 8 shows the percentage of RMSE and the ABSD for models developed by both methods, and again ANN modeling of EDDG process of Inconel 718 for predicting responses is highly rationalized. Both constructions of RSM and an ANN model involve less computing time and money with the aid of Minitab 16 evaluation software packages and MATLAB software, respectively. The ANN model involves some iterative computations, and correct choice of the ANN structure results in the accurate results.

Table 8.

Comparison of prediction between RSM and ANN.

Model summary and prediction error	RSM		ANN
Model summary and prediction error	MRR (mm³/min)	R_a (µm)	MRR (mm³/min)	R_a (µm)
Absolute standard deviation (ABSD)	0.068	0.068	0.033	0.020
Percentage of root mean square error (RMSE)	1.009	0.785	1.009	0.785

ANN: artificial neural network; RSM: response surface methodology; MRR: material removal rate.

Figure 7.

Comparison of experimental MRR with predicted results from RSM and ANN.

Figure 8.

Comparison of experimental R_a with predicted results from RSM and ANN.

Conclusion

In this article, in-house fabricated EDDG experimental setup was used in surface grinding mode to evaluate the MRR and R_a in machining of Inconel 718, and CCD experimental design was employed to investigate the influence of various input factors on the performance characteristics. Two models were developed for predicting MRR and R_a of EDDG of Inconel 718 using RSM and ANN. The following vital conclusions are derived from the investigations:

The experimental study was executed using the CCD design based on RSM. The developed mathematical model based on experimental data for MRR and R_a was found to give satisfactory results in prediction of the EDDG process performance, that is, these models predicted the MRR and R_a which are in close agreement with the experimentally obtained results.

Also, a multi-layered feed-forward backpropagation neural network was used to construct an EDDG process model, and based on performance a model with 4-24-2 structure was selected for modeling of EDDG.

The confirmation experiments performed for validation of both RSM and ANN models show that both techniques found to give appropriate and reliable predictions of MRR and R_a.

From the ANOVA results, the wheel speed (S) is found to be the key factor having an effect on the MRR and R_a in EDDG of Inconel 718 and contributes 89.03% and 79.10%, respectively. The increase in the wheel speed (S) results in increased MRR and R_a. The current (C) is the next most significant factor contributing 4.43% and 8.38% on the MRR and R_a respectively.

Developed ANN and RSM models found to be adequate for accurate and robust predictions of MRR and R_a during machining of Inconel 718 in EDDG process and are well in agreement with the experimental results. However, as compared to RSM method, the ANN technique offered superior fitting to the output responses.

Footnotes

Appendix 1 Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Feyzi

Safavi

SM.

Improving machinability of Inconel 718 with a new hybrid machining technique. Int J Adv Manuf Tech 2012; 66: 1025–1030.

Unune

Mali

HS.

Current status and applications of hybrid micro-machining processes: a review. Proc IMechE, Part B: J Engineering Manufacture. Epub ahead of print 2 September 2014. DOI: 10.1177/095440541454614.

Lauwers

Surface integrity in hybrid machining processes. Proc Eng 2011; 19: 241–251.

Rajurkar

Sundaram

Malshe

AP.

Review of electrochemical and electrodischarge machining. Proc CIRP 2013; 6: 13–26.

Koshy

Jain

Lal

GK.

Mechanism of material removal in electrical discharge diamond grinding. Int J Mach Tool Manu 1996; 36: 1173–1185.

Koshy

Jain

Lal

GK.

Grinding of cemented carbide with electrical spark assistance. J Mater Process Tech 1997; 72: 61–68.

Singh

Yadava

Kumar

Diamond face grinding of WC-Co composite with spark assistance: experimental study and parameter optimization. Int J Precis Eng Man 2010; 11: 509–518.

Singh

Yadava

Kumar

Multiresponse optimization of electro-discharge diamond face grinding process using robust design of experiments. Mater Manuf Process 2010; 25: 851–856.

Kozak

Abrasive electro discharge grinding (AEDG) of advanced materials. Arch Civil Mech Eng 2002; 2: 83–101.

10.

Yadav

SKS

Yadava

Narayana

VL.

Experimental study and parameter design of electro-discharge diamond grinding. Int J Adv Manuf Tech 2008; 36: 34–42.

11.

Singh

Yadava

Kumar

Robust parameter design and multi-objective optimisation of electro-discharge diamond face grinding of HSS. Int J Mach Mach Mater 2012; 11: 429–433.

12.

Kumar

Choudhury

SK.

Prediction of wear and surface roughness in electro-discharge diamond grinding. J Mater Process Tech 2007; 191: 206–209.

13.

Yadav

Yadava

Machining performance of slotted-electrical discharge diamond face grinding of Al/SiC/Gr composite. Mater Manuf Process 2014; 29: 585–592.

14.

Yadav

Yadava

Multiobjective optimization of slotted electrical discharge abrasive grinding of metal matrix composite using artificial neural network and nondominated sorting genetic algorithm. Proc IMechE, Part B: J Engineering Manufacture 2013; 227: 1442–1452.

15.

Shrivastava

Dubey

AK.

Intelligent modeling and multiobjective optimization of electric discharge diamond grinding. Mater Manuf Process 2012; 28: 1036–1041.

16.

Yadav

Yadava

Intelligent modelling and prediction of slotted-electrical discharge diamond grinding of aluminium-silicon carbide-graphite composite. Int J Abras Technol 2013; 6: 93.

17.

Ranganathan

Senthilvelan

Sriram

Evaluation of machining parameters of hot turning of stainless steel (type 316) by applying ANN and RSM. Mater Manuf Process 2010; 25: 1131–1141.

18.

Bhattacharyya

Pradhan

BB.

Modelling of micro-electrodischarge machining during machining of titanium alloy Ti–6Al–4V using response surface methodology and artificial neural network algorithm. Proc IMechE, Part B: J Engineering Manufacture 2009; 223: 683–693.

19.

Youssefi

Emam-Djomeh

Mousavi

SM.

Comparison of artificial neural network (ANN) and response surface methodology (RSM) in the prediction of quality parameters of spray-dried pomegranate juice. Drying Technol 2009; 27: 910–917.

Artificial neural network–based and response surface methodology–based predictive models for material removal rate and surface roughness during electro-discharge diamond grinding of Inconel 718

Abstract

Keywords

Introduction

Experimentation

Results and discussion

ANOVA

Prediction of MRR and Ra in EDDG

ANN

Response surface method

Comparison of predicted MRR and Ra by ANN and RSM

Conclusion

Footnotes

Appendix 1

Declaration of Conflicting Interests

Funding

References

Prediction of MRR and R_a in EDDG

Comparison of predicted MRR and R_a by ANN and RSM