Artificial neural network-based prediction assessment of wire electric discharge machining parameters for smart manufacturing

1 Introduction

Nickel alloys have good properties such as thermal stability, fatigue strength, corrosion resistance, and high-temperature strength. Industries such as aerospace, petrochemical, marine, food processing, and nuclear have demanded such nickel-based components. Conventional machining of nickel-based alloys leads to many defects in a tool such as flank wear, creator wear, edge chipping, and on machined surfaces such as grooves, surface cavities, crack, and micro-voids [1]. As wire electric discharge machining (WEDM) removes the material by spark erosion technique, it is independent of the hardness of the material. Complex shapes of different materials such as nickel and cobalt superalloys, titanium-based alloys, ceramics, composites, nano-ceramics, and shape memory alloys can be machined using WEDM. The machining of nickel-based superalloys poses a major challenge for engineers due to their unique combination of properties, such as high toughness, heat resistance, hardness, strength-to-weight ratio, chemical reactivity with tool materials, low thermal conductivity, and limited creep resistance. Although these properties are crucial for the intended applications of these materials, the high temperatures and stresses that are generated during machining can lead to suboptimal machining performance and shorter tool life. As a result, unconventional machining methods, such as WEDM, may be required to meet the growing demands of the industry. The WEDM offers exceptional precision and accuracy [2–4]. Mouralova et al. [5] experimentally established that there was an optimum cutting speed from which a good surface quality component was obtained by machining in WEDM. Ming et al. [6] proposed a fusion thermo-physical model based on the finite element method to predict the machining performances of BN-AlN-TiB₂ with different weight proportions and to find the optimal process parameters. The novel model’s accuracy was verified by comparing with experimental results, showing relative errors of 24.56, 16.16, and 1.87% in material removal rate, surface roughness, and kerf width, respectively. Additionally, a series of experiments were conducted to investigate the effects of process parameters on machining performances for WEDM of different BN-AlN-TiB₂ composite ceramics with 6, 8, and 10 wt% TiB2. He et al. [7] machined 2D C/SiC composite using WEDM, where the importance of surface roughness and machining speed was highlighted. Wang et al. [8] proposed a system capable of predicting corner errors and suggesting optimal machining parameters that can result in smaller corner errors and faster machining speeds compared to the original parameters. To validate the effectiveness of the proposed system, cutting experiments were conducted, and the results indicated a 20–39% improvement in corner accuracy. Tapering in WEDM is one of the important operations that help to meet the demand for complex manufacturing components. Kinoshita et al., Martowibowo and Wahyudi, Yan et al., Sanchez et al., and Joy et al. [9–13] have adopted many taper techniques for avoiding the disadvantages of wire break, wire bend, insufficient flushing, guide wear, wire friction, etc. Manoj and Narendranath and Manoj et al. [14,15] examined the variation of profiling speed, surface roughness, micro-hardness, and recast layer for circular profile components at different taper angles. The authors have also investigated the effects of servo voltage, pulse on time, cutting speed override (CO), and pulse off time for the same material. Abyar et al. [16] claimed that 57% of the error was contributed by wire deflection in WEDM during machining. Bisaria and Shandilya [17] have utilized pulse modification techniques to improve the accuracy of corners at a right angle (90°), obtuse angle (120°), and acute angle (60°). Werner [18] experimentally determined optimal machining parameters and tool travel for machining curvilinear profiles by modern computer aided design/computer aided manufacturing systems using WEDM.

Yang et al. [19] used a mechanism for the propagation of acoustic emission (AE) and present a new labeling method along with an effective deep learning dual-input model called batch relevance temporal convolution neural network based on the analysis of collected signals. A relationship between AE and pulse time series was studied. Saha et al. [20] examined the machining of composite material made of tungsten carbide–cobalt where artificial neural network (ANN) models were used to predict the surface roughness and cutting speed. Singh and Misra [21] highlighted that the backpropagation neural network (BPNN) in the ANN toolset was an efficient technique for surface roughness prediction during WEDM. Soni et al. [22] have proved that ANN-predicted surface roughness and material removal rate were closer to the experiments during WEDM of Ti–Ni–Co shape memory alloy. Manoj and Narendranath [23] have used ANN prediction for forecasting the profile areas in slant-type tapering operation during WEDM of Hastelloy-X.

From the literature, it can be concluded that many parameters influence the accuracy of the profile. In the literature, we can observe that most of the experiments overcut surface roughness and cutting speed with traditional parameters such as pulse on, pulse off, wire tension, wire speed, servo voltage, and so on. There are other parameters that influence the overall area; in the present investigation, different effects of parameters such as CO, profile offset (PO), dwell time (DT), and wire distance (WD) on the areas of the holes were analyzed. A fixture is used for machining taper holes at 0°, 15°, and 30° taper angles. It was seen that DT did not affect the areas of the holes. As the WD between guides increased, the area of the holes decreased contrastingly, and the increase in PO increased the area. ANN was used to automate the process by predicting the areas at different taper angles for the various parameters by removing human interference in experimentation and characterization.

2 Materials

Nickelvac-HX is a type of nickel superalloy that boasts exceptional mechanical properties, making it a popular choice for a wide range of applications. This alloy is commonly used in the manufacture of combustor cans, spray bars, flame holders, afterburners, tailpipes, dyes, and metal stampings, among other uses. The alloy was heated to 2,150°F (1,177°C) and rapidly cooled as heat treatment (The solution annealing of 1 hr per inch of the section was followed) [24]. After the heat treatment process, the plate was machined to 260 mm × 22 mm × 10 mm dimensions. The measurement of the elements and their percentage in the material using energy dispersive spectroscopy (EDS) technique is shown in Figure 1. This was then fixed to the fixture for machining as shown in Figure 2. Similar experiments with fixture for tapering were also conducted in Manoj et al. [14,15,23].

Figure 1

EDS graph of the Nickelvac-HX.

Figure 2

(a) Slant fixture on WEDM table at 0°, 15°, and 30° angular positions and components after machining. (b) Machined components.

3 Experimental setup and design

The “ELPULS 15 CNC WEDM” from Electronica, Pune, was used to machine Nickelvac-HX. Throughout the experiment, the dielectric fluid was deionized water and the electrode was a zinc-coated copper wire of 0.25 mm in diameter. The circular hole was programmed using numerically controlled codes for different PO. These are converted into WC files for the necessary machining conditions by the computer numerical control (CNC) profiling software called ELCAM. The WC files were instructions for the machine for profiling with specific conditions such as shape, distance, offset, and curvature. This WC file is loaded to the CNC-controlled WEDM. The slant-type taper fixture was fixed to the WEDM bed where the workpiece was fixed to the fixture as shown in Figure 2(a). It also shows the movement of the angular plate to achieve the required angle during machining and different workpieces after machining. Different dimensions of slant holes namely 1, 3, and 5 mm were machined as shown in Figure 2(b). As WEDM is a complex process, among different parameters, a suitable parameter setting has to be found. The machine parameters were fixed during the experimentation so that machining occurs at all the taper angles: Pulse-off time = 44 µs, corner control = 3%, wire speed = 6 m/min, servo feed = 20 mm/min, servo voltage = 40 V, pulse-on time = 115 µs, and flushing pressure = 0 kg/cm². Table 1 indicates the profiling parameters. These machining parameters were selected based on the preliminary experiments, machining range, and fixture angles so that the profiling can be easily carried out.

4 Characterization process

The machined components were measured by “Hitachi SU 3500” and “JEO JSM-6368OLA” scanning electron microscope (SEM) and “TESA VISIO 200”-made coordinate measuring machine (CMM). The machined holes were characterized differently as indicated in Figure 3. The 1 mm hole images were taken using SEM. Furthermore, the SEM images were imported into ImageJ software for measuring the diameter and calculating the areas. The 3 and 5 mm holes were measured using CMM. As the hole dimension changes in different taper angles, we have taken the areas of the hole as output parameters. The areas of taper profiles were calculated, and a similar characterization was followed for all the holes. The 3 mm profile was neglected as it showed the same effect as 1 mm and 5 mm. ANN toolbox was used for prediction using MATLAB software.

Figure 3

Measurement of the different holes in the component.

5 Results and discussion

The areas of the hole were machined for various parameters at different taper angles as shown in Table 2. It can be seen that the highest areas of the hole were found at 100% PO. Although the cutting parameters were the same, the difference in profiling parameters leads to variations in areas.

5.1 Analysis of variance (ANOVA) and main effect plot

The Variation of areas of holes at different taper angles is as shown in Table 2. Table 3 shows the ANOVA, and Figures 4–6 show the main effect plots. From the table and figures, it can be noticed that WD and PO were the most influencing factors in the area of the hole. The PO parameter has the highest % contribution of 67.2–78.2%, making it the most important parameter. Furthermore, WD parameter has a % contribution of 18.1–23.3%. It was then followed by the CO parameter, making it the least effective parameter. It can also be seen that WD and PO were the only significant factors compared to other parameters. The main effect plot and ANOVA indicate that the DT parameter has no role in influencing the areas of the hole. As the DT parameter is the dwell time parameter, it gets activated at the sharp edges. This parameter stops in an edge coordinate before the next command [14,24]. This reduces the wire bend errors, especially the corner errors. In the profile, there are no edges as it is a hole, so the DT parameter becomes the least effective. The DT parameter depends on the geometry of the profile also. Small variations such as increase and decrease were spotted in the main effect plot due to the vibrations, as stated by Habib [25]. Similar results were obtained by Manoj and Narendranath and Soni et al. [14,22] for circular profile areas. So the DT parameter is neglected for further investigation.

Table 3

ANOVA

Sl. no.	Parameters	Degree of freedom	Areas 1 mm² hole		Areas 5 mm² hole
			Adjacent sum of squares	Contribution in %	Adjacent sum of squares	Contribution in %
0° Taper angle
1	WD	3	0.263	18.119	53.431	19.705
2	DT	3	0.005	0.367	1.860	0.686
3	PO	3	1.137	78.273	203.659	75.109
4	CO	3	0.039	2.663	9.403	3.468
5	Error	3	0.008	0.577	2.797	1.032
15° Taper angle
1	WD	3	0.236	20.738	7.279	24.309
2	DT	3	0.006	0.534	0.053	0.176
3	PO	3	0.843	73.993	20.129	67.219
4	CO	3	0.042	3.650	1.589	5.306
5	Error	3	0.012	1.085	0.895	2.990
30° Taper angle
1	WD	3	0.263	18.119	53.431	19.705
2	DT	3	0.005	0.367	1.860	0.686
3	PO	3	1.137	78.273	203.659	75.109
4	CO	3	0.039	2.663	9.403	3.468
5	Error	3	0.008	0.577	2.797	1.032

The bold values indicates that WD is the least contribution and PO is the highest contribution.

Figure 4

Effect plot at 0° for (a) 1 mm and (b) 5 mm holes.

Figure 5

Effect plot at 15° for (a) 1 mm and (b) 5 mm holes.

Figure 6

: Effect plot at 30° for (a) 1 mm and (b) 5 mm holes.

5.1.1 Influence of WD parameter on the area of the hole

The next significant factor affecting the areas of the holes is the WD parameter. It controls the wire distance between the two guides during machining. It becomes important when cutting taper complex profiles. From the effect plot, we can see that as the WD parameter escalates, the area of the hole decreases. However, as the length of the wire increases during machining, it decreases the tension in the wire, which induces wire bending. The bending of this wire causes a lag affecting the area of the holes machined [26]. Figures 5(a) and 6(b) show a decrease in areas of the profile as the WD parameter increases. It was noticed in the remaining graphs that there were small variations in decrease because of the wire vibration. Chaudhary et al. [27] reported that as the wire length escalates, the tension in the wire decreases; this decrease in the tension of the wire leads to wire vibration.

5.1.2 Influence of CO parameter on the area of the hole

The CO is an online parameter that controls the cutting speed during machining of the hole. This CO parameter aids in the machining of complex geometrical profiles at different taper angles. It wheels the cutting speed by controlling the discharge energy generated during machining. This is also called an online parameter as it alters the discharge energy instantaneously during machining based on the geometry of the profile and machining conditions. It avoids wire breaks during the complex machining process [14,24]. As the CO parameter increases, it was seen that the area of the hole decreases, as observed clearly in Figures 4(a), 5(a) and 6(a and b). This is because as the CO increases, the cutting speed also increases. Higher cutting speed results in wire lag as the wire does not travel accurately to the specific coordinates. So this lag in the wire induced by the cutting speed decreases the areas of the holes. Manoj and Narendranath and Soni et al. [14,22] also observed a similar phenomenon in their study. There were small decreases in main effect graphs due to the vibrations caused by instantaneous changes in cutting speed [28].

5.2 Variation of taper areas of the hole at different taper angles

Figure 7 shows the variation in the area of holes at different taper angles. As the taper angle escalates, the area of the hole also increases. This trend was also observed in 1, 3, and 5 mm holes. It can be seen at various profiling parameters (experimental trials) that at 30° taper angles the areas were the highest, and at 0° taper angles, the areas were the lowest. The 15° taper angles always remained in between them. This phenomenon was noticed because of the taper provided by the fixture during machining as shown in Figure 2(b). The material available at taper or slant (15° and 30°) machining is higher compared to horizontal machining (0°). It can be seen that as the taper angle increases, the workpiece also tilts with respect to the wire. The wire path which is programmed remains the same, and the material is given an angle with the help of fixture. This increases the material available for machining which in turn increases the surface area. As the material availability increases, the areas of the hole also increase. Manoj and Narendranath and Soni et al. [14,22] also noted similar results during slant profiling.

Figure 7

Areas of the holes at different taper angles.

6 Conclusion

The parametric variation was outlined by machining holes at different taper angles on Nickelvac-HX. The ANN, which is one of the AI techniques, acts as a predictor and an automation tool for the set of parameters. Here, it is used as an automation tool as it gives the area of the hole for a defined set of parameters without experimentation and avoiding human intervention. The following conclusions were drawn:

• As the PO parameter increases from 0 to 40 µm, the area of the holes increases from 3.32 to 12.3% which proves that it is the most influential on the area of the hole. The DT parameter has no significant effect on the areas.

• The WD is the next significant factor; as it increases, the areas increase from 2.04 to 6.73%, and the CO parameter is seen to affect the areas of the hole to the least extent as % contribution also varies from 2.66 to 5.30%.

• As the taper angle escalates from 0° to 30°, the areas of the holes also increase from 9.70 to 82.41%.

• The ANN model showed that errors are ranging up to 8% in prediction during validation by experimentation.

• Similar research could be carried out for several additional materials and manufacturing processes as future work. Furthermore, as AI models can improve over time by being trained on new datasets, such algorithms can be used for additional prediction and process control.