3.1. Hydroxyapatite Coating
Figure 3 displays the SEM image of the simultaneously shaped and coated Zr
67Cu
11Ni
10Ti
9Be
3 BMG. The HA coating of about 25.2-µm thick was deposited on the Zr
67Cu
11Ni
10Ti
9Be
3 BMG surface as observed in the cross-sectional SEM micrograph shown in
Figure 3a. The electrodes and the added HA powder are melted at the point of spark during the HAm-EDM process. Thus, some of the melted HA additives and the CpTi electrode material are migrated in elemental or compound form to the BMG surface and solidified by the surrounding dielectric fluid. This produces oxides and carbide coating on the Zr
67Cu
11Ni
10Ti
9Be
3 BMG surface. Oxides are believed biocompatible and apatite inducers which promote biocompatibility of the synthetic tissues [
20]. A thin recast layer of about 5.6-µm thick was also observed. At higher magnification, the HA coating microstructure shows globules nanostructured and nanoporous surface (
Figure 3b). This surface is suitable for tissues in-growth into the nano-size rough and porous structure, thereby enhancing the osteointegration of Zr
67Cu
11Ni
10Ti
9Be
3 BMG as an implant. A very long peak of Ca observed in the EDS spectrum presented in
Figure 4 confirmed the HA deposition on the Zr
67Cu
11Ni
10Ti
9Be
3 BMG workpiece surface. Other peaks representing K and O were also noticed. These elements are the major compositions of HA (Ca
5(PO4)
3OH) powder.
The elements distribution is depicted in the elemental mapping shown in
Figure 5. The elemental analysis validated the presence of the BMG (Zr and Ti) and the HA (Ca, K, and O). The mapping shows a homogeneous distribution of these elements on the coated BMG surface, except for Ca whereby some agglomeration was observed (encircled in red). The agglomeration might result due to non-uniform flushing whereby some HA are higher in some regions.
3.2. Phases Formation
Figure 6 displays the XRD pattern of the HA coated Zr
67Cu
11Ni
10Ti
9Be
3 BMG processed at high discharge energy (P
c = 12 A, D
d = 16 µs) and C = (5, 10, and 20 g/L). The analysis using High score software revealed various crystalline phases, including biomimetic and bioceramic biocompatible oxides (HA, CaZrO
3, and ZrO
2) and hard zirconium carbide. While the oxide induces the apatite formation, the carbides enhance the hardness of the HA coated BMG surface [
21]. Thus, promote the biocompatibility and the strength of the BMG respectively. The oxides are formed due to the reaction of the Ca and O (in the hydroxyapatite) to the Zr (in the Zr
67Cu
11Ni
10Ti
9Be
3 BMG) workpiece material. It is obvious that some detected HA do not decompose, which might be due to poor flushing. The carbon supplied by the hydrocarbon-based dielectric fluid reacted with the Zr in BMG to produce ZrC. In a similar study, Sales et al. [
22] investigated the role of calcium based oxide on the titanium alloy surface. The titanium perovskite (CaTiO
3) formed on the EDMed surface greatly promote the tissues growth and cells adhesion after implantation. Some unknown phases were also detected by the XRD analysis. The XRD peaks of the specimen treated through 5 g/L are sharper and longer than those coated using 10 and 20 g/L HA concentrations. Increasing the HA from 5 to 20 g/L expand the electrodes gap, thereby improving the flushing capacity, stabilizes the HAm-EDM process and hence less machining time for the crystalline peaks formation.
3.4. Screening Results Analysis
EDM machine has a wide range of parameters which might be influential to the responses HDR and SR. Based on the machine capability and literature survey six parameters including P
c, D
d, C, E
p, G
v, and T
off were selected for the screening experiment. A two-level fractional factorial (resolution IV) which gives a total of 16 experiments was used. The values of the HDR and SR were calculated and analysed to determine the most significant factors and the interactions. The normal plot and Pareto chart for the HDR presented in
Figure 8a,b depicted that P
c, D
d, C, and E
p are the most significant factors on the HDR. Moreover, the interactions E
pC and P
cC were also found significant as observed in the Pareto chart for the HDR. Similarly, the factors P
c, D
d, C, E
p, and interaction D
dC were found to have an influence on the SR as observed in the normal plot and Pareto chart of the SR shown in
Figure 8c,d, respectively. Additionally, the factor D
d was found to have a major influence on the SR as confirmed by the Pareto chart. Generally, the factors G
v, T
off and/or their interactions were insignificant on both the HDR and SR. Therefore, these two non-significant factors were dropped, and hence further design of experiment (DOE) will only consider four significant factors including P
c, D
d, C, and E
p. The parameters obtained in this study were in agreement with those achieved when shaping siliconized silicon carbide in an aluminium powder mixed dielectric fluid [
24].
3.5. Modelling and Optimization of HDR and SR
D-optimal design is a response surface methodology which generates a design that best estimate the influence of the process parameters particularly suited for screening studies [
18]. Four factors, including P
c, D
d, C, and E
p and mixed level was used to achieve a total of 24 experiments.
Table 2 displays the DOE matrix and the data for the selected responses, HDR and SR.
The contribution of each parameter on the HDR is indicated in the ANOVA shown in
Table 3. The model value of 0.0001 (“Prob > F”) justifies its significance. The table also shows that the factors P
c, D
d, C, E
p, and interactions AB, AD, BC, CD, and B
2 with “Prob > F” values less than 0.05 are significant terms. The lack of fit “Prob > F” of 0.2170 indicates its insignificance. The model is required to fit the response (HDR), therefore it is good for the lack of fit to be non-significance.
The analysis of variance (ANOVA) for the SR is depicted in
Table 4. The regression model, significant factors, interactions and lack of fit tests are summarized by the ANOVA table. Model and the model terms with less than 0.05 “prob > F” are regarded as significant terms 18. Therefore, from the ANOVA table the model and the factors P
c, D
d, C, and E
p are considered the significant terms of the SR. The model is expected to fit, because the lack of fit “prob > F” value of 0.2213 indicates its non-significance.
The normal probability of the residuals and the residuals versus run plots for the HDR and SR are depicted in
Figure 9. The normal plots for both HDR and SR shows that the data lies within the straight line as seen in
Figure 9a,b respectively. This confirmed that the selected terms are only the significant factors and the errors are normally distributed. In the residuals versus run plots for HDR and SR presented in
Figure 9c,d respectively, all the studentized residuals of regression lie within the required limit (±3 sigma), and not any outliers observed. This confirmed the capability of the model to predict the responses.
A curved plot 3D surface plots for the HDR and SR are presented in
Figure 10. The curves surface observed in the HDR plots (
Figure 10a–c) indicated the significance of the curvature and a quadratic model. The highest HDR is observed at 10 A and 8 µs parameters setting in the current versus discharge time 3D surface plot shown in
Figure 10a. Moreover, the C versus P
c plot shows highest HDR of 0.002 g/min at C = 20 g/L and P
c = 5 A as observed
Figure 10b. A similar HDR is achieved when C = 20 g/L and D
d = 8 µs parameters setting was employed as shown in
Figure 10c. The influence of P
c and D
d on the SR is presented in the 3D surface plot shown in
Figure 10d. A flat surface which indicated insignificant curvature is observed. From this plot, it can be noticed that the SR increases with increases in both P
c and D
d. Therefore, higher SR can be observed at P
c = 12 A and D
d = 16 µs while the lowest SR can be achieved at P
c = 5 A and D
d = 4µs parameters setting.
The ANOVA table presented produces the equation which relates the responses HDR and SR to the input parameters. The final predicted model equations in terms of the actual factors for HDR and SR are presented in Equations (3)–(6).
Table 5 shows the model summary statistic for HDR. The software recommended a quadratic model for the HDR, due to its least standard deviation (0.23), largest statistic R-square (0.7710), and lowest Predicated Error Sum of Squares (PRESS) (4.20) compared to the linear model.
The model summary statistic for the SR is shown in
Table 6. It could be observed that the linear model for SR is suggested by the design expert software for the SR. This is due to its least standard deviation of 0.41, largest statistic (predicted) R-square of 0.7000 and lower Predicated Error Sum of Squares (PRESS) of 5.36 as required.
The response optimization can be directly obtained from the design expert software by considering the response value and desirability. The software suggests the optimum parameters combination for maximizing HDR and minimizing SR. The solutions suggested by the software for maximizing HDR and minimizing SR are presented in
Table 7 and
Table 8.
Following a successful selection of the input parameters’ optimum level, a confirmatory test was conducted to validate the numerical model developed for the responses HDR and SR. In this study, the optimum parameters recommended by the design software was used to conduct the confirmation test. Thus, it will be possible to establish the adequacy of the achieved numerical models. Optimum parameters setting plays a significant role not only in improving the quality but also in the industries by reducing parts production time and cost. This study planned to achieve a maximum rate of hydroxyapatite deposition and minimum surface roughness (nanostructured surface). Using the optimum parameters conditions generated by the software, five sets of confirmatory tests were carried out. The results of both responses are presented in
Table 9.
The five sets of results for the HDR and SR provided by the software (predicated) and those conducted at optimum parameters setting (actual experiment) were compared in
Table 10.
To enable accurate estimation of the predicted models, prediction error (PE) of each response was determined. The average error of the five results was also calculated to achieve the average PE (APE). The PE is calculated through Equation (1). The APE for HDR and SR was found to be 4.94% and 5.09%, respectively. A study conducted by Mir et al. [
25] and Aliyu et al. [
26] reported that an APE of less than 10% confirmed the excellent reproducibility of the experimental conclusions. Therefore, the APE of 4.95% (HDR) and 5.09% (SR) in this study can be considered to be within the acceptable limit.