Journal of

Materials
Processing
Technology
ELSEVIER Journal of Materials Processing Technology 56 (1996) 54-65

OPTIMIZATION OF SURFACE FINISH IN

END MILLING INCONEL 718

M. Alauddin, M.A. E1 Baradie and M.S.J. Hashmi

Advanced Materials Processing Centre, Dublin City University, Ireland

ABSTRACT

This paper presents an approach to optimize the surface finish in end milling Inconel 718 using uncoated
carbide inserts under dry conditions. In view of this, the mathematical models for surface roughness have been
developed in terms of speed and feed by response surface methodology. Response surface contours were
constructed in speed feed planes by computer. From these contours it was possible to select a combination of
cutting speed and feed that reduces machining time without increasing the surface roughness. These were
obtained by superimposing the constant lines of metal removal rates on the surface roughness contours.

1.INTRODUCTION

Inconel 718 is a nickel base superalloy containing a columbium(niobium) age-hardening addition that provides
increased strength without a decrease in ductility. It is oxidation and corrosion resistant and can be used at
temperatures in the range of - 217 to 700C. Due to its good tensile, fatigue, creep and rupture strength, this
material is used in the manufacture of components for liquid rockets, parts for aircraft turbine engines,
cryogenic tankage etc. However, in general this alloy is difficult to machine [1,2] for the following reasons:

- - High work hardening rates at machining, strain rates leading to high cutting forces
-- Abrasiveness
-- Tough, gummy and strong tendency to weld to the tool and to form built up edge
-- Low thermal properties leading to high cutting temperatures

A high surface finish is required because of its use in the design of high strength and reliable components.
This paper presents an approach to develop mathematical models for surface roughness in end milling
Inconel 718 by response surface methodology (RSM) [3] in order to optimize the surface finish of the machined
surface. RSM is a combination of mathematical and statistical techniques used in an empirical study of
relationships and optimization, where several independent variables influence a dependent variable or response.
In applying the RSM, the response or dependent variable is viewed as a surface to which a mathematical model
is fitted. In this study, machining variables like speed and feed which are easily controllable are considered in
building the model. The optimum cutting condition is obtained by constructing contours of constant surface
roughness by computer and used for determining the optimum cutting conditions for a required surface
roughness.

0924-0136/96/$15.00 1996 Elsevier Science S.A. All rights reserved

SSD10924-0136 (95) 01820-5
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65 55

2. WORKPIECE MATERIAL

The Inconel 718 workpiece material used in the machining test was in hot forged and annealed condition. The
chemical composition of the workpiece material conforms to the following specification(%):

C Mn Si Ti A1 Co Mo Cb Fe Cr Ni

0.08 0.35 0.35 0.60 0.80 1.00 3.00 5.00 17.00 19.00 52.82

The hardness of the workpiece material was measured and found to be 260 BHN.

3. SURFACE FINISH IN END MILLING OPERATIONS

The basic geometry of the end milling process is shown in Fig. 1.

Ns
_9
a. 1
I

DI

ar

7
Fig.1 : End milling process

Where
v = cutting speed (peripheral) of the cutter (m/rain)
D = diameter of the cutter (mm)
N s = rotational speed of the cutter (rev/min)
fz = feed per tooth (ram/tooth)
fm = feed per minute (mm/min) or table speed (= fz x z x N~)
z = number of teeth in the cutter
a. = axial depth of cut (mm)
aT = radial depth (width) of cut (mm).
56 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65

In end milling operations, theoretical surface roughness is generally dependent on cutting conditions,
workpiece material, cutting tools etc. as shown in Fig.2. However, the theoretical surface roughness value R~,
for side milling operations by solid end mills is estimated by the following formula [4,5]:

R . --
f, z (1)
32(R+ )

Where Ra - surface roughness CLA Om), z - number of teeth in the cutter, R - radius of the cutter, + ve sign
to up milling and - ve sign to down milling.

The surface finish produced in face milling operations [6,7] by end mill inserts can be expressed as:

f~ (2)
R a - 32r

Where r - nose radius of the end mill insert.

The above formulae for surface roughness are a function of feed per tooth and tool geometry. The actual
surface roughness is usually larger than the theoreti_cal surface roughness values obtained by these formulae,
because, these formulae do not take into account built up edge formation, deflection and vibration which are
usually a function of cutting conditions like speed and axial depth of cut.

4. DEVELOPMENT OF THE MATHEMATICAL MODEL BY RSM

4.1 Postulation of the mathematical model

Factors which affect the surface finish in end milling are shown in Fig.2. However, for a particular work-tool
geometry, the surface roughness in end milling is assumed to be a function of cutting conditions. Moreover,
in a rigid work-tool system, a low value of axial depth of cut has a negligible effect on the surface roughness
in end milling [8]. Therefore, in this study, the surface roughness in end milling is assumed-to be a function
of cutting speed (v) and feed (f,). The multiplicative model [9] for the predictive surface roughness (response)
in end milling in terms of the investigated independent variables can be expressed as:

Ra = C vk fz I (3)

Where I~, - predictive surface roughness CLA Oam), v - cutting speed (m/min), fz - feed per tooth (ram/tooth),
and C, k, 1 - model parameters to be estimated.
Taking natural logarithm converts the above intrinsically linear type nonlinear model into the first-order
polynomial as:
= boxo + b l X l + b2x2 (4)

The second-order model can be extended from the first-order model's equation as:

(5)

Where ~, - predictive response (roughness) on natural logarithmic scale, while xo = 1 (a dummy variable) and
x~, x 2 - the coded value ( logarithmic transformations) of v and fz respectively and b's - model parameters to
be estimated using experimentally measured surface roughness (Ra) data.
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 5 4 ~ 5 57

Input ~ Cutting process~ Output -I

(Controllable indepcndent variahh.s) DelJt'mlent var,ahle~; J

m m
Speed
Feed

Axial depth o f cut

Machine
Cutting f l u i d

Rigidity

Type of materia-i- Machina bility

Surface]
parameter
( Mierostruc#ure
Compnsition I

Workpiece Properties
~[ End milling H
t Ilard.css) .__..~ operations finish
S i d e milling Roughness
Tool geometry Fare milling WaiRes~
SI.I millinx Eeeurs of form

-- Tool material --
flies .Carbide ,~ialon,
Tool geometry
'Rake,lleliz &clearance
a n l i e , aase eadius.No.
I Tool o~ Itelh & Diametee
Cutter runout
Tool wear
- - Milling mode m

Fig.2: Factors influencing surface finish in end milling processes

4.2 Experimental Design

A well designed series of experiments can substantially reduce the total number of experiments often carried
out randomly. In order to determine the equation of the response surface, several experimental designs have
been developed which attempt to approximate the equation using the smallest number of experiments possible.
The most widely preferred classes of response surface design are the orthogonal fu'st-order and central
composite second-order design. The orthogonal first-order design (with two factors) consisting of nine
experiments has been used to develop the postulated f'Lrst-order model.These 9 tests consist of four points ( 22
) located at the comers of the square and a centre point repeated five times as shown in Fig.3. As the first-order
model is accepted only over a narrow range of variables, the experiments were extended to obtain a second
order model. The central composite rotatable design for the second-order response surface consists of an
additional 4 axial points called augments (Fig.3).

4.3 Coding of independent Variables

The independent variables were coded taking into consideration the limitation and capacity of the milling
machine. The coded values of the variables shown in Table i for use in equations (4) and (5) were obtained
from the following transforming equations:

lnv - In18.37 (6)

xl = In24.10-In18.37
Inf - In 0 . 0 7 5 <7)
-- m 0 . 0 9 8 - tn 0 . 0 7 5
58 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65

X2

C')

C 03

Fig.3: Orthogonal first-order design (Trial nos. 1,2 ...... 9) and Central composite design (Trial
nos. 1,2 ...... 13) for 2 factors

Where x, - the coded valtie of the cutting speed corresponding to its natural value v and x2 - the coded value
of the feed per tooth corresponding to its natural value f,.

The above relationships were obtained from the following transforming equation:

- ,x.o
x = (8)
1 - I- x.o

Where x - the coded value of any factor corresponding to its natural value X,, X,a - the natural value of the
factor at the +1 level and X.o - the natural value of the factor corresponding to the base or zero level [10].

Table 1: Levels of Independent Variables and Coding Identification

Levels in .Coded form

Independent Variables
-q2 (lowest) -1 (low) 0 (centre) +1 (high) +42 (highest)

v , m/min ( x I ) 13.12 14.08 18.37 24.10 27.0

fz, mm/tooth ( x 2 ) 0.053 0.057 0.075 0.098 0.107

4.4 Experiment
The process utilized for surface roughness was a slot milling operation, performed on a Bridgeport vertical
milling machine. The cutting tests were carried out under .dry conditions using an end mill with uncoated
tungsten carbide inserts( =ISO K20 grade, top rake angle y. = +5 , helix angle ~ = 5 , right hand helix, number
of inserts in the cutter z = 2, nose radius of the insert r = 0.80 mm, diameter of the end mill D = 25 mm ). The
bottom surface roughness (R,) produced in the slot was measured by a stylus type instrument (Taylor Hobson's
Talysurf 5-120). Table 2 shows the cutting conditions and measured surface roughness results. The measured
(experimental) surface roughness was obtained by averaging the surface roughness values at a minimum of three
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65 59

location points on the bottom surface of the slot. A cut-off value of 0.8 mm was used when measuring the
surface roughness of the machined surface.

4.5 Estimation of the model parameters

The regression parameters of the postulated model were estimated by the method of least squares, using the
following basic formula:
b = ( X r X ) -1X r Y (9)
Where
b = the matrix of parameter estimates
X = the matrix of independent variables or design matrix
X T = the transpose of the matrix X
Y = the matrix of logarithm of the measured surface roughnesses (responses).

4.6 Analysis of results

4.6.1 Development of the f'trst-order model

The first-order model for surface roughness was developed using 9 tests of the first-order orthogonal design,
the parameters in equation (4) were estimated, yielding the surface roughness predicting equations as:

)3 = - 0.6078 - 0.1206x 1 + 0.4609x z 0o)

Table 2: Experimental Conditions and Results

Trial No. Speed Feed Experimental average

v (m/min) fz (turn/tooth) Coding surface roughness
R, (pm)

1 14.08 0.057 -1 -1 0.40

2 24.10 0.057 1 -1 0.35
3 14.08 0.098 -1 1 1.12
4 24.10 0.098 1 1 0.79
5 18.37 0.075 0 0 0.53
6 18.37 0.075 0 0 0.49
7 18.37 0.075 0 0 0.47
8 18.37 0.075 0 0 0.48
9 18.37 0.075 0 0 0.58
10 13.12 0.075 -~/2 o 0.62
11 27 0.075 ,12 0 0.51
12 18.37 0.053 o -42 0.37
13 18.37 0.107 o q2 0.64
60 M. Alauddin et al. /Journal of Materials Processing Technology 56 (1996) 54-65

Equation (10) can be transformed using equations (6) and (7) to provide the surface roughness (CLA, ~m) as
a function of the cutting speed, v (m/min) and feed, fz (mm/tooth) as follows:
/?a = 172.12 V-0"44fz1"72 (11)

Equation (11) indicates that an increase in the cutting speed decreases the surface roughness while an
increase in the feed increases the surface roughness. This equation is valid for slot milling Inconel 718 using
an end mill with carbide inserts under dry conditions and the following ranges of cutting speed(v) and feed per
tooth(f,):

1 4 . 0 8 < v 524.10 m/min

0.057 < fz < 0.098 mm/tooth

4.6.2 Development of the second-order model

The second-order model was developed to extend the variable range to describe adequately the relationship
between the slot milling output (surface roughness) and the investigated independent variables. The model was
developed utilizing the central composite rotatable design as shown in Fig.3 and the model parameters of
equation (5) have been estimated using equation (9). Hence the second-order surface roughness predictive
equation is:

= - 0.674 - 0.094x 1 0.32671 x 2 0.083x 2 + 0.0076x~ - 0.053xlx 2 (12)

4.6.3 Adequacy of the predictive models

The analysis of variance (ANOVA) [11,12] was used to check the adequacy of the developed models. As per
this technique, the F-ratio of the model is calculated and compared with the standard tabulated value of the F-
ratio for a specific level of confidence. If the calculated value of the F-ratio does not exceed the tabulated value,
then with the corresponding confidence probability, the model may be considered adequate. The results of the
analysis of variance of the predictive models are shown in Table 3 and 4.

Table 3: ANOVA of the First-order model

Source Sum of degree of Mean Frat(cal) Frataab) Remarks

Square(SS) freedom(d0 Square(MS)

Zero order 3.32501 1 3.32501

terms
First order 0.90789 2 0.45395
terms
Lack of fit 0.06450 2 0.03225 4.31 6.94 * Adequate
Pure error 0.02992 4 0.00748
Residual 0.09442 6 0.01574
Total 4.32732 9
* 95% confidence limit for F2,4
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 5 4 ~ 5 61

Table 4: ANOVA of the second-order model

Sources SS df Ms Frat(cal) Fr.t(~b) Remarks

Zero-order terms 5.00014 1

First-order terms 0.90789 2

Second-order terms 0.02210 3

Lack of fit 0.17981 3 0.05993 8.01 16.69 Adequate

Pure error 0.02992 4 0.00748

Residual 0.20973 7

Total 6.13986 13
* 99% confidence limit for F3,4

4.6.3 Significance Testing of the individual variables of the predictive models

The sum of squares for any variable x i adjusted for all other variables in the predictive models is available
by taking b2~ / % [13]. This amount is the sum of the squares and is added to a model containing all the
variables except this particular one. For this purpose, the variance ratio:

(13)
F - ca
S2

for a particular variable x i, adjusted for all other variables, is compared with a standard F value. Where bi -
coefficients of the predictive models, cii - the elements of matrix ( X r X) ~ and s - standard deviation
The results of the significance testing for individual variables of the predictive models are shown in Tables
5&6.

Table 5: Significance testing for individual variables (First-order Model)

Sources SS df MS Fret(cat) Fr~,~b) Remarks

x t (v) 0.05818 1 0.05818 3.70 3.78" almost significant

x2 ( f z ) 0.84972 1 0.84972 150.25 " Significant
Residual 0.09442 6 0.01574

4.6.3 Confidence interval (Precision of prediction)

Due to the experimental error, the estimated parameters and hence the estimated surface roughness ~ are
subject to uncertainty. Hence, the precision of these quantities has been estimated by calculating appropriate
confidence intervals on the surface roughness. The confidence intervals for the predictive responses ~ are given
by (.9 + Ag), where:

A? = tar, ,s2~(V(?)
62 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65

o o

0.091 / ~ ~ ~ ~ , ~ 0.098
,.- '

0.075

0,057 0.057 ~30 ern~/rnin

0,053. 0.052
13.12 14.08 18.37 24.10 16.22 13.12 14.08 18.37 24.10 26.22
Speed.v m/rain Speed v, m/mln
Fig.6: Surface roughness contours (Second-order Fig.7: Contours of surface roughness (second-order
Model) for Inconel 718 in speed-feed planes Model) and metal removal rates for Inconel
at a selected level of axial depth 718 in speed-feed planes at 0.50 mm axial
(a, 0.50 mm)
= depth of cut

6. CONCLUSIONS

6.1 Reliable surface roughness models have been developed and utilized to enhance the efficiency of slot
milling Inconel 718 when using an end mill with carbide inserts.

6.2 The first-order prediction model is valid within the speed range of .14.08 - 24.10 m/min and feed range of
0.057 - 0.098 ram/tooth

6.3 By utilizing the second-order model, it is possible to extend the variable range and the second-order
prediction model is valid within the speed range of 13.12 - 27 rn/min and feed range of 0.053 - 0.107
ram/tooth

6.4 The feed effect is very significant and dominant in both the first and second-order models. The speed effect
is almost significant in the first-order model while in the second-order model it is not so significant.

6.5 An increase in the feed increases the surface roughness while an increase in the cutting speed decreases
the surface roughness

6.6 Contours of the surface roughness outputs were constructed and utilized to select the proper combination
of the cutting speed and feed to increase the metal removal rate without sacrificing the quality of the
produced surface.

REFERENCES

1. N. Richard and Aspinwal,"Use of ceramic tools for machining nickel-based alloys",Int.J.Mach.Tool

Manufact., Vol.29,No.4, pp.575-588,1989.
2. M.C. Shaw,"Metal Cutting Principles", 2nd edition, London, Oxford University Press,1986.
3. D.C. Montgomery, "Design and Analysis of Experiments",2nd edition, New York, John Wiley and
Sons,1976.
4. D. Montgomery and Y. Altintas,"Mechanism of cutting forces and surface generation in dynamic
milling",ASME J.Eng.Ind.,Vol. 113,pp. 160-168,1991.
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65 63

For a specified axial depth of cut ( a~ = 0.50 mm) for a cutter with the number of teeth (z) = 2 and diameter
of 25 mm ( i,e., % = 25 for slot milling) and using the transformation equations (6) and (7), equation (15)
becomes:

lnQ = - 0.8256 + 0.2715x 1 + 0.2676x 2 (16)

For a constant rate of metal removal, equation (16) can be represented by a straight line as illustrated in Fig.5.
This figure was obtained by superimposing the constant Q lines on the surface roughness contours in speed-
feed plane (e.g., Fig.4). Fig. 5 can be used to increase the efficiency of the machining processes as follows.
Consider that a surface roughness of 0.60 pm is desired; however, cutting conditions that provide a higher
rate of material removal must be selected. It is shown that the selection of cutting conditions represented by
the point "B" is better than those represented by point "A". This 50% increase in material removal rate is
obtained without any sacrifice in the quality of the produced surface.

0.098

0.098

E 0.07:

0.075

0.05"
14.08 1837 24.10
Speed v, ndmin 0.0571 , \ I
14.08 18.37 24.10
Speed v, n~/min
Fig.4: Surface roughness contours (First-order Model) Fig.5: Contours of surface roughness (First-order
for lnconel 718 in speed-feed planes at 0.50 Model) and metal removal rates for Inconel
mm axial depth of cut 718 in speed-feed planes at 0.50 mm axial
depth of cut

5.2 Utilization of the second-order surface roughness model

Equation 12 ( i.e. the second-order model ) is plotted in Fig.6 as contours of each of the response surfaces
at a selected level of axial depth of cut (0.50 turn). These contours were also obtained by utilizing a "Matlab"
computer package. It can also be seen from Fig.6 that surface roughness decreases with an increase in cutting
speed, and increases as feed increases. Hence, a better surface finish is obtained at a combination of high speed
and low feed. However, in order to reduce machining time, both the feed rate and cutting speed should be high
as possible.

For the utilization of the second-order model, Fig.7 was constructed, using a procedure similar to that of Fig.5.
From Fig.7 it can be seen that the rate of metal removal can be doubled without increasing the surface
roughness, for example, the selection of the cutting conditions represented by point B ( at R~ = 0.60 pm) is
better than that represented by point A (also at 0.60 pro). This 100% increase in metal removal rate is obtained
without any sacrifice in quality of the produced surface.
64 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65

Table 6: Significance testing for the individual variables (Second-order Model)

Sources SS df MS E,t(,t) F,,t~b) Remarks

xt 0.07084 0.07084 2.36 3.59" not so significant

x2 0.85334 28.50 28.50 " Significant

xl 2 0.04746 0.04746 1.58 " Insignificant
x22 0.00040 0.00040 0.01 " Insignificant
x~x2 0.01103 0.01103 0.37 " Insignificant
Residual 0.20973 0.02996 0.02996

Where t is the value of the horizontal coordinate of the t-distribution corresponding to the specified degrees of
freedom (dO and level of confidence (ix). The 95% confidence intervals for ~ values of the first-order model
and 99% confidence intervals for S' values of the second-order model were found to be quite satisfactory. The
precision of prediction of the predictive models ( i.e. A~, value)) is shown in Table 7.

Table 7: Assessment of precision of prediction of the predictive models

Model Standard deviation A~ values

S
Comer points Central points Axial points
First-order 0.1254459 0.2400 0.1023
Second-order 0.1730936 0.4931 0.2708 0.4789

5. OPTIMIZATION OF SURFACE FINISH

5.1 Utilization of the first-order surface roughness model

Equation (10) (i.e.first-order model) has been plotted in Fig.4 as contours (sections) for each of the response
surfaces at a selected level of axial depth of cut (0.50 mm). These contours were obtained by utilizing a
"Matlab" computer package. It can be seen from Fig.4 that surface roughness decreases with an increase in
cutting speed, and increases as feed increases. Hence, a better surface finish is obtained by a combination of
high speed and low feed. However, in order to reduce machining time ( i.e. to achieve higher production rates)
the feed rate and cutting speed should be as high as possible.
From the contours shown in Fig.4 it is possible to select a combination of feed and cutting speed that reduces
machining time without increasing the surface roughness, since there are a large number of combinations of
cutting speed and feed that produce the same surface roughness. This can be illustrated by further utilization
of the model to include the rate of metal removed. The rate of metal removed Q (cm3/min) is given by:

Q=f~zxN,aaa, (14)
1000

Equation (14) can be written as

ln Q = ln f z + I n z + ln N , + l n a,, + l n a r - In1000 (15)
 M. Alauddin et al. / Journal of Materials Processing Technology 56 (1996) 54-65 65

5. M.Martelotti,"Analysis of the milling process", Trans.ASME,Vol.63,pp.667-700,1941.

6. B.L. Juneja and G.S. Sekhon,"Fundamentals of Metal Cutting and Machine Tools", 1st edition, New
Delhi,India, Wiley Eastern Limited,1987.
7. National Twist Drill and Co.,"Accuracy of Milled Surfaces, Part 1: Effect of cutter diameter, number of
teeth, runout and feed",Metal Cuttings,Vol.9,No. 1,pp.2-11,1961.
8. M. Alauddin and M.A. E1 Baradie and M.S.J. Hashmi,"Surface roughness model for end milling metal
Matrix composite(Al/SiC)", Proc..30th Int.MATADOR Conf., Manchester,pp.135-142,1993.
9. P.Balakrishnan and M.F, DeVries,"Sequential Estimation of Machinability parameters for adaptive
optimization of machinability data base systems",ASME J.Eng. Ind., Vol.107,pp.159-166,1985.
10. K.C. Lo and N.N.S. Chen,"Prediction of tool life in hot machining of alloy
steel",Int.J.Prod.Res.,Vol.15,No.l,pp.47-63,1977
11. K.K Padmanabhan and A.S. Murty,"Evaluation of Frictional Damping by Response Surface Methodology",
Int.J.Mach.Tool Manufact, Vol.31,No. 1,pp.95-105,1991.
12 N.R. Drapper and H. Smith, "Applied Regression Analysis",2nd edition, New York,John Wiley and
sons,1966.
13. Joginder Sing and M.K Khare,"A mathematical model for tool wear weight using response surface
methodology", 12th AIMTDR Conf., IIT Delhi,pp.256-260, Tata McGraw-Hill Pub.Co.,New Delhi, India,
1986.