Int J Adv Manuf Technol

DOI 10.1007/s00170-016-9699-5

ORIGINAL ARTICLE

Fluidized bed abrasive jet machining (FB-AJM) of K-99 alumina

ceramic using SiC abrasives
B. K. Nanda 1 & A. Mishra 1 & D. Dhupal 1

Received: 27 January 2016 / Accepted: 4 November 2016

# Springer-Verlag London 2016

Abstract This paper is focused on fluidized bed abrasive jet process with high accuracy and economical aspect. AJM is
machining (FB-AJM) of K-99 alumina ceramic using silicon best suitable for cutting, grooving, drilling, deburring, trim-
carbide (SiC) black abrasive with the indigenously fabricated ming, polishing and kerf designing in the hard and brittle
AJM set-up based upon fluidized bed mixing chamber and materials. The kinetic energy of the high-velocity abrasive
pressurized powder feed system. The effects of three machin- particles developed during the flow process creates brittle
ing input process parameters such as pressure (P), nozzle tip fractures on the target material that propagate both in lateral
distance (NTD) and abrasive grain size (GS) on the six re- and longitudinal directions causing material removal.
sponses viz. material removal rate (MRR), surface roughness Machinability in AJM is largely affected by high speed of
(Ra), depth of cut (DOC), taper angle (TA), overcut (OC) and abrasive grits in the jet stream, bed pressure, fluid pressure,
flaring diameter (FD) are analysed. Experiments are per- nozzle tip distance, grain size, hardness of the abrasives and
formed using Box-Behnken response surface design method hardness of the workpiece material.
to investigate the effects of input process parameters on six Many researchers have worked on AJM to ascertain the
responses by developing suitable quadratic regression models, influence of carrier gas, jet velocity, standoff distance, size
and desirability functions are defined accordingly for particle and types of abrasive, nozzle material, nozzle shape and wear
swarm optimization (PSO) to get the optimal parametric con- on overall machining of both brittle and ductile materials by
ditions. The observed optimal process parametric settings are considering some numerical models and statistical approaches
pressure of 5 kgf/cm2, nozzle tip distance of 8 mm and grain with proper design of experiment (DOE). Balasubramanium
size of 260 μm for achieving the optimal responses, and final- et al. [1] used Taguchi orthogonal array design and analysis of
ly, the results are experimentally validated. variance (ANOVA) to investigate the parameters of abrasive
jet deburring process and concluded that burr removal was
Keywords FB-AJM . Alumina ceramic . Desirability affected by jet height and impingement angle. The machining
function of glass workpieces in AJM was performed by Routara et al.
[2] to obtain the optimal combination of the process parame-
ters for the multiresponses, namely material removal rate and
surface roughness by applying Taguchi-based grey relational
1 Introduction analysis. Azmir et al. [3] applied the Taguchi method and
ANOVA to optimize the process parameters of abrasive water
Abrasive jet machining (AJM) is an effective method of ma- jet machining (AWJM) for effective machining of glass fibre
chining hard and brittle materials like glass, ceramics, quartz reinforced epoxy composites. Recently, the new technique of
and other engineering materials due to the ease of machining microabrasive jet machining has been widely used for
micromachining of brittle materials with and without masks.
* D. Dhupal The mask allows the abrasive jet to strike only at the desired
[email protected] portions of the target and blocks it on the other region. Ally
et al. [4] predicted the cross-sectional profile of microchannels
1
Department of Production Engineering, VSSUT, Burla, India by considering the surface evolution models in abrasive jet
 Int J Adv Manuf Technol

micromachining of glass polymers with 50-mm aluminium optimization techniques to predict the optimal parametric data
abrasive powder and also extended the experiment for alumin- for the responses using Taguchi-based grey relational analysis
ium 6061-T6, Ti-6Al-4V alloy and 316L stainless steel for [1–3], response surface methodology (RSM)-based
establishing the relationship between erosion rate and inclina- desirability-based optimization [11], argentinaical methods
tion angle. The shot blasting phenomenon for [17], artificial neural networks and particle swarm optimiza-
micromachining of ceramic materials was experimentally per- tion (PSO). Majumdar et al. [18] predicted the optimum pa-
formed [5] to reveal that radial cracks did not propagate down- rameters for material removal rate and electrode wear ratio
ward due to particle impact. Zhang et al. [6] applied the con- during electro discharge machining of AISI 316L stainless
cept of microabrasive intermittent jet machining (MAIJM) steel using RSM-based multiobjective PSO.
process to blow away the abrasives accumulated inside the The present research work has focused on multiobjective
machined hole and performed the multivariable linear regres- optimization of process parameters of FB-AJM of K-99 alu-
sion analysis using Taguchi orthogonal array. The implemen- mina ceramic using various grades of silicon carbide (SiC).
tation of the thick SU-8 layer of mask in the microabrasive jet The influence of three process parameters such as pressure
machining process to cut arc-shaped microchannels for (P), nozzle tip distance (NTD) and abrasive grain size (GS)
microfluidic applications and fabrication of the passive on the six responses, namely material removal rate (MRR),
micromixer with 3D feature using microabrasive jet machin- surface roughness (Ra), depth of cut (DOC), taper angle (TA),
ing process was successfully performed by Saragih et al. [7, overcut (OC) and flaring diameter (FD) are discussed with
8]. The technique of multiphase microabrasive jet machining ANOVA and regression analysis using Box-Behnken design
was implemented by Su et al. [9] to improve the tribological of RSM. Then, multiobjective desirability-based PSO pro-
and sealing properties of the mechanical seals. Kim et al. [10] gram is developed and run in MATLAB interface with initial
introduced the 3D CAD mask modelling algorithm to fabri- population of 50 for individual input parameters to carry out
cate the 3D m odel for the mask structure using 200 numbers of iterations for predicting the optimized values,
microstereolithography and implemented the technique in and finally, the results are experimentally confirmed.
AJM for all general planar and non-planar workpieces.
Sooraj et al. [11] applied the simple and cost-effective method
of generating ultra-fine finish on the internal surfaces of tubu- 2 Experimental details
lar specimens using elastic abrasives by conducting experi-
ments according to central composite design of response sur- 2.1 Experimental set-up and procedure
face methodology and investigated about the effect of axial
pressure, abrasive grain size and longitudinal stroke velocity An indigenously designed FB-AJMM set-up having different
on the surface finish. fixtures is fabricated as shown in Fig. 1. A double-acting sin-
The concept of fluidized bed is used in AJM and gle cylinder, air compressor of 1 hp and electric motor with
microabrasive jet machining for uniform mixing of the abra- working pressure of 10 kgf/cm2 is connected for supplying
sives with the carrier gas to produce damage-free machined continuous high-pressure air to act as working fluid medium
surfaces with proper cooling by the carrier gas acting as a for FB-AJM. The compressed air then passed through the air
lubricant. Barletta et al. [12–16] used the technique of fluid- filter-regulator-lubricator (FRL) or dehumidifier unit to pre-
ized bed-assisted abrasive jet machining (FB-AJM) to finish vent any agglomeration and clogging of abrasives at the exit
the internal surfaces of the high-resistance stainless steel nar- of the nozzle. The nozzle is properly designed and
row and long tubes, tubular components of Inconel 718, high- manufactured by the high carbon and high chromium-
strength aluminium alloy (AA 6082T6) using the systematic contained D2 steel possessing with high wear and abrasion
design of experiments to measure surface roughness and ma- resistance. Assumptions, i.e. (i) uniform mixing of abrasives
terial removal rate due to the impact of different input process with the compressed air inside the fluidized bed, (ii) availabil-
parameters. They also studied the effect of operational param- ity of homogeneous mixture at the nozzle exit, (iii) normal
eters, namely abrasive size, jet pressure and processing time, position of the nozzle with respect to workpiece to reduce
by using the full factorial design of experiments for the easy- nozzle wear, (iv) each abrasive particle is a multipoint cutting
to-automate finishing of atmospheric plasma spraying (APS) tool with sharp edges having excellent flow characteristics, (v)
TiO2 coatings on AISI 1040 steel substrates with the devel- restriction of reuse of the abrasives and (vi) negligible effect of
oped technique of fast regime-fluidized bed machining (FR- machine vibration, are considered in this experimental work.
FBM).
Different design of experiments (DOE) has been explained 2.2 Specification of the workpiece material and abrasives
[17] to obtain the suitable responses with minimum number of
experiments for saving the machining cost and time. Also, K-99 alumina ceramic material of specimen specification
researchers have applied the different multiobjective 25 × 25 × 3 mm (with ±1% dimensional accuracy) is chosen
Int J Adv Manuf Technol

Fig. 1 Schematic diagram of FB-AJM set-up

as workpiece for experimentation as shown in Fig. 2. K-99 2.3 Selection of input process parameters and responses
alumina ceramic includes material possessing 99.8% Al2O3
and less percentage of SiO2, Fe2O, Na2O, MgO with CaO. P, NTD and GS are selected as the input process parameters
The physical properties like melting point 2000 °C, porosity for this FB-AJM process to measure the six responses, namely
0.1% and density 3.90 g/cm3 with gas-tight structure also MRR, Ra, DOC, TA, FD and OC.
indicate about its hard-to-machine characteristics. The ma-
chining operation is conducted with black-coloured SiC of
average grain sizes 260, 525 and 745 μm as in Fig. 3. SiC 2.3.1 Measurement of material removal rate
abrasive has hardness (MOH) 9.4, bulk density (as per size)
1.4–1.53 g/cm3 with SiC 97.04%. SiC is selected as the suit- Material removal rate is calculated on mass basis (g/s) as
able abrasive for the K-99 alumina ceramic material to obtain MRR = (W1 − W2) / Δt, where W1 and W2 are the weights
the desired surface finish and material removal rate. of the specimen before and after machining with a time span
of Δt. Here, W1 and W2 are measured by using the weighing
machine of least count (LC) 0.0001.

2.3.2 Measurement of surface roughness

The surface roughness of the secondary etched surface on the

ceramic specimen under the same machining conditions is
measured with MITUTOYO 400 having least count of
0.001 μm, sampling length of 0.75 mm and an evaluation
length of 3 mm. Surface roughness of each surface is mea-
sured thrice at different positions, and their average value is
Fig. 2 Machined view of K-99 ceramic workpiece taken as the final reading for better accuracy.
 Int J Adv Manuf Technol

Fig. 4 Geometrical representation of hole diameter and flaring diameter

ImageJ software to measure OC as the difference between the

FD and nozzle diameter as in Fig. 4.

2.3.4 Measurement of depth of cut

Generally, tapered holes are produced in AJM process, so

DOC (or h) is obtained by using the geometrical formulae
for truncated cone as
πh
2 
Volume of truncated cone : V ¼ r þ r1  r2 þ r22 ð1Þ
3 1

Density of specimen : ρ ¼ W V ð2Þ

Here, the internal radius (r1) and external radius (r2) are
obtained from the microscope image using ImageJ software.

Fig. 3 a SEM of abrasive grains of average size 260 μm. b SEM of

abrasive grains of average size 525 μm. c SEM of abrasive grains of
average size 745 μm

2.3.3 Measurement of flaring diameter and overcut

The images of the machined specimens are taken by an optical

microscope of ×10 magnification, and then, the original hole
diameter and flaring diameter are determined with the help of Fig. 5 Taper angle in cut profile
Int J Adv Manuf Technol

2.3.5 Measurement of taper angle design [17]. Then, Design Expert software is used to obtain
the design matrix, statistical calculations and surface plots.
The type of hole cut in the specimen is similar to a truncated The experiments are performed on the indigenously devel-
cone with larger radius (r1) at surface and smaller radius (r2) oped FB-AJM set-up utilizing the specified input parameters
inside the hole at depth h. Then, TA or α, in Fig. 5, is as per the design matrix, and six responses viz. MRR, Ra,
DOC, TA, OC and FD are measured accordingly. The input
ðr1 −r2 Þ
α¼ tan−1 ð3Þ parameters with their levels and the coded design matrix ac-
h cording to Box-Behnken design method are given in Tables 1,
2 and 3 respectively.
2.4 Experimental planning

RSM of DOE is used to perform the experiments for the three

quantitative input factors, namely P, NTD and GS. RSM is the 3 Experimental results and discussions
collection of techniques, based on mathematical and statistical
concepts that analyse a response being affected by multiple The 6 numbers of responses are measured by experimenting
variable factors to get an optimized model. Generally, the 15 numbers of trials, based on Box-Behnken RSM design
response surface is described by a second-order polynomial matrix as shown below in Table 3. The results along with
mathematical model as the three levels of process parameters are put in the Design
Expert software for detailed analysis. The microscopic images
X
q
X
q
X
q
X
q
of the specimens after machining are given in Fig. 6, clearly
E ðyÞ ¼ β 0 þ βi X i þ βij X j þ γ i X 2i þ ε ð4Þ indicating the FD.
i¼1 i¼1 i< j i¼1

where y represents the response, X1 and X2 show the input 3.1 Adequacy test of responses
process parameters, βs are the unknown parameters estimated
from the data in the experiment, βij indicates the second order The data obtained from the measured specimens are put in the
or quadratric effect and ε is the random error term [17]. Box- Design Expert software, and ANOVA is obtained to find out
Behnken design (BBD) of RSM is selected to conduct the the most significant variables affecting the responses, i.e.
non-sequential experiments in a safe operating limit consider- MRR, Ra, DOC, TA, OC and FD. Regression analysis also
ing P, NTD and GS as the three input parameters to affect the performed for each of these responses, and the following qua-
machining process, each with three levels, i.e. low, mid and dratic models are obtained given in Eqs. 5, 6, 7, 8, 9 and 10
high, along with three centre points to develop the required respectively.

MRR ¼ 6:97  10−3 þ 2:587  10−3 P þ 1:934  10−3 NTD−9:0  10−6 GS−2:53  10−4 P2 −9:1  10−5 NTD2
ð5Þ
þ1:0  10−6 GS2 −6:1  10−5 P  NTD−1:0  10−6 P  GS−1:0  10−6 NTD  GS

Ra ¼ 1:83−0:347 P−0:107 NTD þ 9:0  10−5 GS þ 6:65  10−2 P2 −1:75  10−2 NTD2 þ 1:0  10−5 GS2
ð6Þ
−0:033 P  NTD þ 6:1  10−5 P  GS þ 2:6  10−6 NTD  GS

DOC ¼ 0:733 þ 0:017 P þ 0:145 NTD−6:9  10−4 GS þ 0:001 P2 −1:438  10−2 NTD2 þ 1:0  10−6 GS2
ð7Þ
þ0:0281 P  NTD−6:3  10−5 P  GS−1:01  10−4 NTD  GS

TA ¼ 42:9−0:49 P−7:41 NTD−0:0225 GS þ 0:557 P2 þ 0:555 NTD2 þ 4:0  10−5 GS2

ð8Þ
−0:537 P  NTD−9:31  10−3 P  GS þ 6:1  10−3 NTD  GS

OC ¼ −1:21 þ 0:47 P þ 0:275 NTD þ 2:02  10−3 GS−0:113 P2 −0:0179 NTD2 þ 1:0  10−5 GS2
ð9Þ
þ0:1072 P  NTD þ 2:02  10−4 P  GS−3:95  10−4 NTD  GS

FD ¼ 1:27 þ 0:658 P þ 0:586 NTD þ 1:81  10−3 GS−0:0872 P2 −0:0068 NTD2 þ 1:0  10−5 GS2
ð10Þ
þ0:052 P  NTD−5:4  10−5 P  GS−1:5  10−4 NTD  GS
 Int J Adv Manuf Technol

Table 1 Input parameters with

their coded value at different Serial no. Input parameter Low level Mid-level High level
levels (−1) (0) (+1)

1 Pressure (P) 3 4 5
2 Nozzle tip distance (NTD) 4 6 8
3 Grain size (GS) 260 525 745

The p values, lack of fit, R-squared and adj R-squared with increase in GS up to certain value, and then, it increases
values of the response models are shown in Table 4. It is seen with increase of GS as in Fig. 7b, c.
that the p values and model F values are significant and R- The surface plot analysis of Ra is given in Fig. 8a–c respec-
squared values are above 90%, whereas the lack of fit values tively. Figure 8a shows that the surface roughness is minimum
are insignificant for all responses [17]. when both the pressure and nozzle tip distance are at their mid-
levels and maximum at low nozzle tip distance with high pres-
sure. So, the mid-level pressure and nozzle tip distance should
3.2 Surface plot analysis of responses be used for better surface roughness. Initially, velocity of the
particles leaving the nozzle increases with increase in NTD due
Variations of material removal rate on surface plots are shown to acceleration of the abrasives that causes increase in MRR
in Fig. 7a–c respectively. It is seen from Fig. 7a that for higher and decrease in Ra, but further increase in NTD decreases
material removal rate, pressure is high and nozzle tip distance velocity and increases Ra as shown in Fig. 8a. It is seen from
plays an important role up to some extent. Small MRR at low Fig. 8b that the surface roughness is almost directly propor-
NTD is due to reduction in nozzle pressure with decrease in tional to grain size and pressure has less significant role. In
distance, but drop in MRR at high NTD is due to reduction in Fig. 8b, Ra decreases with increase in P for a certain value
jet velocity with increase in distance. The material removal because initially, MRR increases with increase in P causing
rate is high at high pressure with larger grain size that is evi- Ra to decrease. Then, Ra increases as MRR decreases because
dent from Fig. 7b. Again, high material removal rate is of choking and agglomeration of the abrasives occurring inside
achieved at higher nozzle tip distance and low grain size as the nozzle. Figure 8c gives the evidence that the surface rough-
indicated in Fig. 7c. The tendency of one variable plot be- ness is almost directly proportional to grain size and nozzle tip
tween MRR and GS shows that MRR increases with increase distance has a less significant role in determination of surface
in GS up to certain value, but after that, MRR decreases with roughness. Ra decreases with increase in NTD because of high
GS when P and NTD remain unchanged. Here, MRR is an MRR due to acceleration of the abrasive particles, but with
interactive response of the simultaneous action of three vari- further increase in NTD, MRR decreases causing Ra to in-
ables P, GS and NTD because of which MRR first decreases crease as the machining area increases because of flaring of
the particles as in Fig. 8c.
Table 2 Box-Behnken RSM based coded experimental design matrix The surface plot analysis of DOC is indicated in Fig. 9a–c
Sl. no. Pressure Nozzle tip distance Grain size respectively. Figure 9a shows that depth of cut is almost direct-
ly proportional to pressure and nozzle tip distance has a less
1 −1 −1 0 significant role. Figure 9b reveals that maximum depth of cut is
2 1 0 1 achieved at high levels of pressure and grain size. Figure 9c
3 0 1 1 indicates that the depth of cut is almost directly proportional to
4 0 1 −1 grain size and nozzle tip distance has a less significant role.
5 1 0 −1 The surface plot variations of TA are plotted in Fig. 10a–c.
6 1 1 0 It is clearly seen from Fig. 10a that the taper angle varies
7 −1 1 0 linearly with pressure and nozzle tip distance has a little effect
8 0 0 0 on it. When P increases, then velocity of the jet stream in-
9 1 −1 0 creases, and it maintains the cylindrical path of the stream with
10 0 −1 −1 less flaring as given in Fig. 10a causing TA to decrease.
11 −1 0 1 Figure 10b gives the evidence that taper angle is maximum at
12 0 −1 1 high levels of grain size and pressure. Figure 10c reveals that
13 −1 0 −1 the maximum taper angle is achieved at high grain size and low
14 0 0 0 nozzle tip distance, but taper angle increases at low grain size
15 0 0 0 and high nozzle tip distance. Within the limits of the parame-
ters taken in the DOE, TA increases with GS because higher
Int J Adv Manuf Technol

Table 3 Experimental data

Sl. no. Specimen no. Parameter 1 Parameter 2 Parameter 3 Response 1 Response 2 Response 3 Response 4 Response 5 Response 6
P (kgf/cm2) NTD (mm) GS (μm) MRR (g/min) Ra (μm) DOC (mm) TA, α (°) OC (mm) FD (mm)

1 1 3 4 525 0.001735 1.228 0.8751 15.92 1.902 5.902

2 2 5 6 745 0.004298 1.525 1.0946 12.82 2.891 7.651
3 3 4 8 745 0.003821 1.521 0.8796 21.09 3.147 8.675
4 4 4 8 260 0.004612 0.921 1.5115 8.83 3.424 8.602
5 19 5 6 260 0.004125 0.926 1.6151 7.47 2.642 7.344
6 6 5 8 525 0.003621 1.217 1.4621 10.12 3.412 8.773
7 7 3 8 525 0.002696 1.264 0.9663 21.05 2.264 8.101
8 8 4 6 525 0.003461 0.966 1.1791 11.60 2.676 7.491
9 9 5 4 525 0.003151 1.445 1.1458 9.29 2.192 6.158
10 10 4 4 260 0.002845 0.978 1.3445 15.63 1.576 5.792
11 26 3 6 745 0.003212 1.513 0.8666 24.64 2.261 7.161
12 16 4 4 745 0.003489 1.522 0.9216 16.37 2.004 6.133
13 13 3 6 260 0.003575 0.964 1.3151 10.35 2.192 6.806
14 14 4 6 525 0.003465 1.242 1.1466 11.75 2.724 7.209
15 15 4 6 525 0.003325 1.248 1.1808 10.60 2.482 7.344

GS reduce the velocity of the jet stream, causing deviation of obtained from Fig. 11c. The maximum overcut is achieved
the particles from the cylindrical path and hence TA increases. at low level of grain size and high level of nozzle tip distance,
The surface plot descriptions of OC are indicated in and minimum overcut is obtained at low levels of nozzle tip
Fig. 11a–c. It is seen from Fig. 11a that overcut varies linearly distance and grain size.
with nozzle tip distance and pressure has less impact on it, but The surface plots of FD are shown in Fig. 12a–c respec-
maximum overcut is achieved at high levels of both pressure tively to show the variations. Figure 12a shows that maximum
and nozzle tip distance. Figure 11b gives the evidence that flaring diameter is achieved at high levels of nozzle tip dis-
there is very less significant change in overcut with change tance and pressure. Again, Fig. 12b gives the evidence that the
in grain size, whereas the overcut varies linearly with pressure, flaring diameter is maximum at high levels of both grain size
but maximum overcut is achieved at high levels of both grain and pressure. It is seen from Fig. 12c that flaring diameters are
size and pressure. The changes in the nozzle tip distance and maximum at maximum nozzle tip distance due to reduction in
grain size significantly change the amount of overcut as velocity because of atmospheric drag.

Fig. 6 Microscopic images of the machined specimens at ×10

 Int J Adv Manuf Technol

Table 4 ANOVA analysis data of responses where di(ŷi) = individual desirability for ith response,
Response p value Lack of fit R-squared Adj R-squared ŷi(x) = response value at ith experiment and Ui, Li and
Ti are the upper, lower and targeted values for a response
MRR 0.0050 0.1809 0.984869 0.957633 respectively. The powers ‘s’ and ‘t’ are the weights (wj),
Ra 0.0243 0.9593 0.924205 0.787773 set by the analyst to determine how important it is for ŷi to
DOC <0.0001 0.0590 0.955143 0.921500 be close to the optimum.
TA 0.0039 0.0893 0.965104 0.902292 (ii) Determination of global desirability index (D) by
OC 0.0182 0.1733 0.933017 0.812447 combining the individual desirability indices using
FD <0.0001 0.06346 0.994068 0.983389 the equation:
1
=X

 Wj
D¼ d r11  d r22  d r33  …:  d rnm ð12Þ
4 Optimization using PSO with desirability function
where wj denotes the individual weight of the re-
Candioti et al. [19] optimized the multiple responses by de- sponses and n is the total number of responses.
veloping the desirability function according to the solutions of
Derringer and Suich to obtain the operating conditions in com-
pliance with the criteria of all responses and provided the best
value of compromise in the desirable joint response. The op- 4.1 Individual desirability function (Yi) with weight
timization process is achieved by converting all individual
responses into a single composite function known as global All responses are required to be optimized simulta-
desirability function, and then, it is optimized. The steps in- neously using desirability-based PSO for which the na-
volved in formation of composite function are ture of optimization of each response is represented in
Table 5 along with their individual desirability func-
(i) Calculation of individual desirability (di) for each of the tions. The desirability index of each response is based
responses from the experimental data with the following on its optimization objective for the quadratic models of
equation: the input parameters. Then, each response is assigned
2 3 with some weight according to its importance for the char-
0 ! if ^yi ðxÞ < Li acterization of the process in a scale from 1 to 5 (where 1 is
6 7
6 ^yi ðxÞ−Li 7 for the least and 5 for the highest importance) as given in
6 if Li ≤^yi ðxÞ < T i 7
6 T i −Li 7 Table 5.
 6
6
7
7
d i ^yi ðxÞ 6
6 1 ^yi ðxÞ ¼ T i 7 7 ð11Þ
6 ! if 7
6 ^yi ðxÞ−U i 7 4.2 Global desirability function (DFi) and fitness
6 if T i ≤^yi ðxÞ ≤U 7
6 T i −U i 7 function (Y)
4 5
0 if ^yi ðxÞ > U i
The global desirability function (DFi) is defined in Eq. (13) as

 1
DF ¼ Y 1 W 1  Y 2 W 2  Y 3 W 3  Y 4 W 4  Y 5 W 5  Y 6 W 6 ðW 1 þW 2 þW 3 þW 4 þW 5 þW 6 Þ ð13Þ

1
And fitness function : Y ¼ ð14Þ 4.3 Implementation of particle swarm optimization
1 þ DF
The highest quality characteristics are obtained when The logic behind the PSO technique is the interaction
the goal has to predict the optimal experimental condi- among the group members (or particles) to carry out a
tions of the FB-AJM that will give the maximum value group activity in which each particle informs about in-
of the desirability function (DF) by minimizing the fit- dividual time duration to complete a work. Then, other
ness function (Y). particles update their paths with that particle to have the
Int J Adv Manuf Technol

Fig. 7 a Surface plot of MRR

with nozzle tip distance and
pressure. b Surface plot of MRR
with grain size and pressure. c
Surface plot of MRR with nozzle
tip distance and grain size

least time duration to complete the work, i.e. target is cluster of particles from the minimum and maximum
achieved in minimum duration of time. PSO algorithm value of the input factors and their velocity profiles,
is performed with the following steps: (a) generating a (b) updating velocity of each particle with respect to
 Int J Adv Manuf Technol

Fig. 8 a Surface plot of surface

roughness with pressure and
nozzle tip distance. b Surface plot
of surface roughness with
pressure and grain size. c Surface
plot of surface roughness with
grain size and nozzle tip distance

its position in every iteration, (c) evaluating the fitness minimal optimal value and the corresponding input fac-
function value, (d) updating the personal best (pbest) and tors as the optimal conditions (according to the desir-
the global best (gbest) for the particles, (e) selecting the ability functions) at the end of the number of iterations
Int J Adv Manuf Technol

Fig. 9 a Surface plot of depth of

cut with pressure and nozzle tip
distance. b Surface plot of depth
of cut with pressure and grain
size. c Surface plot of depth of cut
with grain size and nozzle tip
distance

and (f) finally, multiobjective desirability-based PSO At the beginning of the iteration process, Xio and Vio are
code is developed and run in MATLAB interface. taken as the position and velocity of the ith particle. The initial
 Int J Adv Manuf Technol

Fig. 10 a Surface plot of taper

angle with pressure and nozzle tip
distance. b Surface plot of taper
angle with pressure and grain
size. c Surface plot of taper angle
with grain size and nozzle tip
distance
Int J Adv Manuf Technol

Fig. 11 a Surface plot of overcut

with pressure and nozzle tip
distance. b Surface plot of overcut
with pressure and grain size. c
Surface plot of overcut with grain
size and nozzle tip distance
 Int J Adv Manuf Technol

Fig. 12 a Surface plot of flaring

diameter with pressure and nozzle
tip distance. b Surface plot of
flaring diameter with grain size
and pressure. c Surface plot of
flaring diameter with grain size
and nozzle tip distance
Int J Adv Manuf Technol

Table 5 Optimization objective of responses with weight distribution

Sl. no Response Optimization objective Desirability function (Di) Weight (wi)

1 Material removal rate Maximization MRR−MRRmin 5

D1 = MRRmax −MRRmin
2 Surface roughness Minimization Ramax −Ra 5
D2 = Ramax −Ramin
3 Depth of cut Maximization DOC−DOCmin 3
D3 = DOCmax −DOCmin
4 Taper angle Minimization TAmax −TA 3
D4 = TAmax −TAmin
5 Overcut Minimization OCmax −OC 5
D5 = OCmax −OCmin
6 Flaring diameter Minimization FDmax −FD 5
D6 = FDmax −FDmin

position and velocity for the cluster of particles are generated where wmax and wmin are the first and last values of the
by using Eqs (15) and (16) with the help of upper limit of the inertia weight respectively and imax is the maximum
input factor Ui and lower limit of the input factor Li with number of iterations used in PSO. (Commonly, wmax
design space time Δt. and wmin are taken as 0.9 and 0.4 respectively). Here,
the initial swarm size is set at 50 with maximum num-
X i0 ¼ Li þ randðU i −Li Þ ð15Þ ber of iteration of 200. The constants C1 and C2 are
U i þ randðU i −Li Þ Position of particle taken as 2 with initial inertia weight of 0.4. The itera-
V i0 ¼ ¼ ð16Þ tion is carried out, and the output is given in Table 6. It
Δt Designed space time
is clear from Fig. 13 that the optimal condition is ob-
The next iteration is to evaluate the values of the fitness tained in lesser number of iterations by applying
functions, i.e. the personal best pbest and the global best desirability-based PSO method. An experiment is carried
gbest. The velocity profile of the particles is given by the out utilizing the predicted conditions to validate the
following equation: predicted responses at optimal experimental conditions.
The validated experiment of the workpiece after ma-
0 1
vi ðt þ 1Þ chining is shown in Fig. 14. The deviations among the
@ A predicted optimal response values and experimental re-
ðvelocity of particle i at time t þ 1Þ sponse values are shown in Table 7, and it is seen that
0 1 the errors obtained are less and within the acceptable
wvi ðt Þ
range. Hence, the optimality of the machining condition
¼@ A
is validated.
ðcurrent motionÞ
The SEM micrograph analysis of the machined ceramic
!
pbesti ðt Þ−xi ðt Þ surface is given in Fig. 15a, b. The eroded surface is very
þ C 1 rand Δt rough, and non-uniformity of material removal is visible giv-
ðparticle memory influenceÞ ing that the erodent particles are not in uniform size and shape
!
gbesti ðt Þ−xi ðt Þ as in Fig. 15a. As the cutting edges of the abrasive grits used
þ C 2 rand ð17Þ are different, the material removal throughout the plane is also
Δt
ðswarm influenceÞ different. The white patches indicate the removal of material,
and at the left side, a void of the ceramic material is clearly
X i ¼ X i ð t Þ þ X i ð t þ 1Þ ð18Þ
seen. Again, plastic deformation is clearly visible by the white
and lines, and a major white line in the left side indicates the
  initiation of crack. Some abrasive particles attached to the
wmax −wmin eroded surface may be due to high temperature or secondary
w ðiÞ ¼ wmax − ð19Þ
imax bombardment as revealed from Fig. 15b. These dust particles

Table 6 Predicted optimal conditions

Process parameters Response values

P (kgf/cm2) NTD (mm) GS (μm) MRR (g/min) Ra (μm) DOC (mm) TA, α (°) OC (mm) FD (mm)
5 8 260 0.003800 0.925880 1.800560 6.574000 3.623600 8.801200
 Int J Adv Manuf Technol

Fig. 13 Graph between

desirability function value vs.
number of iterations

are nothing but the erodent which is grinded to smaller sizes experimental analysis on FB-AJM operation, the following
after the bombardment. The SEM micrograph indicates a void outcomes can be drawn:
fraction of approximately 2%. It also gives that maximum
erosion occurs at the centre. (i) The dry filtered compressed air with SiC abrasive is im-
pinged on the K-99 alumina ceramic workpiece material
for successfully producing precision holes with the indig-
enously build FB-AJM set-up.
5 Conclusions (ii) From the surface plot, it is profound that better
MRR is achieved at parametric combinations of
Experiments have been performed successfully to produce high pressure with 745 μm grain size and high
drilled holes on rectangular K-99 workpiece on the indige- nozzle tip distance with 260 μm grain size respec-
nously developed FB-AJM set-up. The experiments are suit- tively. At moderate levels of pressure with nozzle
ably conducted with the predetermined input parameters ac- tip distance of 6 mm, it gives a better surface fin-
cording to the DOE to obtain the responses viz. MRR, Ra, ish, and at the same time, surface roughness in-
DOC, TA, OC and FD using the SiC abrasives. From the creases with increase in grain size. In the case of

Fig. 14 Machined view of

specimen and its microscopic
image after confirmation test
Int J Adv Manuf Technol

Table 7 Deviation among the

predicted optimal response vales Type of Process parameters Response values
and experimental response values model
P (kgf/ NTD GS MRR Ra DOC TA, α OC FD
cm2) (mm) (μm) (g/min) (μm) (mm) (°) (mm) (mm)

Predicted 5 8 260 0.0038 0.9260 1.8006 6.574 3.6236 8.8012

Experimental 5 8 260 0.0036 0.8797 1.6745 6.048 3.5148 8.3611
Error % 4 5 7 8 3 5

DOC, maximum levels of pressure with 745 μm tip distance with grain size of 260 μm produces mini-
grain size give a better effect. Surface plot predicts mum overcut. FD changes linearly with change in grain
that the depth of cut is almost directly proportional size and pressure, but minimum flaring diameter is
to grain size where nozzle tip distance has a less achieved at a minimum nozzle tip distance irrespective
significant role. of the grain size.
(iii) From the analysis of individual responses, it can be pre- (iv) The PSO-predicted optimal combination of parameter
dicted that TA varies linearly with pressure and nozzle setting is pressure of 5 kgf/cm2, nozzle tip distance of
tip distance has little effect. Surface plot also reveals that 8 mm and grain size of 260 μm for achieving the opti-
OC varies linearly with nozzle tip distance and pressure mal responses such as material removal rate of 0.0038 g/
has less impact. In addition to this, low level of nozzle s, surface roughness of 0.9260 μm, depth of cut of
1.8006 mm, taper angle of 6.574°, overcut of
3.6236 mm and flaring diameter of 8.8012 mm.
(v) Experiment is performed at predicted optimal conditions
to validate the optimality of the model, and the measured
experimental response values obtained are material re-
moval rate of 0.0036 g/s, surface roughness of
0.8797 μm, depth of cut of 1.6745 mm, taper angle of
6.048°, overcut of 3.5148 mm and flaring diameter of
8.3611 mm. The errors between the predicted
models and experimental-validated models are in
specified range.

The research findings along with various empirical rela-

tionships will provide effective guidelines to select parameter
settings for achieving desired machining during abrasive jet
machining process. The present research work on FB-
AJM of rectangular workpieces of ceramic material will
be useful for the modern engineering industries that are
working in the field of machining and fabrication of
ceramic parts and components.

Acknowledgements The authors would like to acknowledge the

TEQIP-II (WB), VSSUT, Burla, Odisha, for providing all the assistances
to carry out the research work inside the Non-Traditional Machining
(NTM) Lab.

References

1. Balasubramaniam R, Krishnan J, Ramakrishnan N (2000) An em-

pirical study on the generation of an edge radius in abrasive jet
external deburring (AJED). J Mater Process Technol 99(1):49–53
Fig. 15 a SEM of eroded surface showing crack lines. b SEM of 2. Routara BC, Nanda BK, Sahoo AK, Thatoi DN, Nayak BB (2011)
embedded grains inside the machined surface Optimisation of multiple performance characteristics in abrasive jet
 Int J Adv Manuf Technol

machining using grey relational analysis. Int J Manuf Technol 12. Barletta M, Tagliaferri V (2006) Development of an abrasive jet
Manag 24(1–4):4–22 machining system assisted by two fluidized beds for internal
3. Azmir MA, Ahsan AK (2008) Investigation on glass/epoxy com- polishing of circular tubes. Int J Mach Tools Manuf 46(3):271–283
posite surfaces machined by abrasive water jet machining. J Mater 13. Barletta M, Guarino S, Rubino G, Tagliaferri V (2007) Progress in
Process Technol 198(1):122–128 fluidized bed assisted abrasive jet machining (FB-AJM): internal
4. Ally S, Spelt JK, Papini M (2012) Prediction of machined surface polishing of aluminium tubes. Int J Mach Tools Manuf 47(3):483–495
evolution in the abrasive jet micro-machining of metals. Wear 292: 14. Barletta M, Ceccarelli D, Guarino S, Tagliaferri V (2007) Fluidized
89–99 bed assisted abrasive jet machining (FB-AJM): precision internal
5. Wakuda M, Yamauchi Y, Kanzaki S (2003) Material response to finishing of Inconel 718 components. J Manuf Sci Eng 129(6):
particle impact during abrasive jet machining of alumina ceramics. 1045–1059
J Mater Process Technol 132(1):177–183 15. Barletta M (2006) A new technology in surface finishing: fluidized
6. Zhang L, Kuriyagawa T, Yasutomi Y, Zhao J (2005) Investigation bed machining (FBM) of aluminium alloys. J Mater Process
into micro abrasive intermittent jet machining. Int J Mach Tools Technol 173(2):157–165
Manuf 48(7):932–945 16. Barletta M, Rubino G, Guarino S, Bolelli G, Lusvarghi L, Gisario A
7. Saragih AS, Ko TJ (2009) A thick SU-8 mask for microabrasive jet (2008) Fast regime-fluidized bed machining (FR-FBM) of atmo-
machining on glass. Int J Adv Manuf Technol 41(7–8):734–740 spheric plasma spraying (aps) TiO2 coatings. Surf Coat Technol
8. Saragih AS, Ko TJ (2009) Fabrication of passive glass micromixer 203(5):855–861
with third-dimensional feature by employing SU8 mask on micro- 17. Montgomery DC (2002) Design and analysis of experiments.
abrasive jet machining. Int. J Adv Manuf Technol 42(5–6):474–481 Wiley, New York
9. Su X, Shi L, Huang W, Wang X (2016) A multi-phase micro-abra- 18. Majumder A, Das PK, Majumder A, Debnath M (2014) An ap-
sive jet machining technique for the surface texturing of mechanical proach to optimize the EDM process parameters using
seals. Int J AdvManufTechnol 1–8 desirability-based multi-objective PSO. Prod Manuf Res 2(1):
10. Kim HC, Lee IH, Ko TJ (2012) Direct 3D mask modeling for 228–240
nonplanarworkpieces in microabrasive jet machining. Int J Adv 19. Candioti LV, De Zan MM, Camara MS, Goicoechea HC (2014)
Manuf Technol 58(1–4):175–186 Experimental design and multiple response optimization: using
11. Sooraj VS, Radhakrishnan V (2014) Fine finishing of internal sur- the desirability function in analytical methods development.
faces using elastic abrasives. Int J Mach Tools Manuf 78:30–40 Talanta 124:123–138