The international journal of advanced manufacturing technology, tập 60, số 9 12, 2012

Int J Adv Manuf Technol (2012) 60:841–851
DOI 10.1007/s00170-011-3647-1

ORIGINAL ARTICLE

An image-based methodology to establish correlations
between porosity and cutting force in micromilling
of porous titanium foams
M. Abolghasemi Fakhri & E. V. Bordatchev &
O. R. Tutunea-Fatan

Received: 3 August 2010 / Accepted: 14 September 2011 / Published online: 12 October 2011
# Her Majesty the Queen in Right of Canada 2011

Abstract Porous titanium foam is now a standard material
for various dental and orthopedic applications due to its
light weight, high strength, and full biocompatibility
properties. In practical biomedical applications, outer
surface geometry and porosity topology significantly
influence the adherence between implant and neighboring
bone. New microfabrication technologies, such as micromilling and laser micromachining opened new technological possibilities for shape generation of this class of
products. Besides typical geometric alterations, these manufacturing techniques enable a better control of the surface
roughness that in turn affects to a large extent the friction
between implant and surrounding bone tissue. This paper
proposes an image analysis approach for optical investigation
of the porosity that is tailored to the specifics of micromilling
process, with emphasis on cutting force monitoring. According to this method, the area of porous material removed during
micromilling operation is estimated from optical images of the
micromachined surface, and then the percentage of solid
material cut is calculated for each tool revolution. The

employment of the aforementioned methodology in micromilling of the porous titanium foams revealed reasonable
statistical correlations between porosity and cutting forces,
especially when they were characterized by low-frequency
M. Abolghasemi Fakhri : E. V. Bordatchev :
O. R. Tutunea-Fatan (*)
Department of Mechanical and Materials Engineering,
The University of Western Ontario,
London, ON N6A 5B9, Canada
e-mail:
M. Abolghasemi Fakhri : E. V. Bordatchev
Centre for Automotive Materials and Manufacturing, Industrial
Materials Institute, National Research Council of Canada,
800 Collip Circle,
London, ON N6G 4X8, Canada

variations. The developed procedure unlocks new opportunities in optimization of the implant surface micro-geometry, to
be characterized by an increased roughness with minimal
porosity closures in an attempt to maximize implant fixation
through an appropriate level of bone ingrowth.
Keywords Porous titanium foam . Optical imaging . Image
processing . Porosity . Micromilling . Cutting force

1 Introduction
Porous sintered metals are typically produced by powder
metallurgy and represent a class of relatively new materials
with wide variety of industrial applications, especially
biomedical applications, e.g., dental and orthopedic applications [1–3], due to their light weight, high strength, and
full biocompatibility properties. However, outer surface
geometry and porosity topology for fabricated components,
e.g., implants and prostheses, significantly influence the

adherence of bone cells to implant material. Therefore,
most components fabricated from porous foam metals are
produced in near-net shapes, and still require secondary
machining operations [4] that can provide the desired
roundness, smearing, and other surface quality parameters
which cannot be obtained during sintering [5]. Also, there
are many 3D shapes and geometries that are difficult or
almost impossible to produce by conventional forming
technologies [2] without secondary machining, e.g., slots,
bevels, blind holes, threads, cross-holes, and re-entrants
normal to the pressing directions. At present, about half of
the components produced by powder metallurgy parts
require secondary machining operations [6]. New microfabrication technologies, such as micromilling and laser
micromachining [7], opened new technological possibilities


842

Int J Adv Manuf Technol (2012) 60:841–851

options for characterization and visualization of 3D foam
structures [11]. The micro-CT technique performs segmentation of 3D porosity into a set of 2D CT slice images
providing acceptable image quality, precise density profile,
high spatial resolutions (i.e., below 1 μm), good pore
contrast, and reliable pore shape anisotropy. Micro-CT is
still under development especially in terms of 3D structure
reconstruction and extraction of reliable information on
structural parameters [18, 19].
As mentioned above, metal-foam-based functional components (e.g., biomedical implants and prostheses) require a
rough outer surface with fully open pores in order to attain

high levels of bonding between implant and bone. Among
the options available to generate implant surfaces, material
removal operations have always been well regarded,
especially when moving into the micro-scale domain. In
this sense, the use of small diameter cutting tools (around
25 μm) in micromilling operations is believed to be capable
of significantly enhancing the accuracy of the generated
surfaces. However, when the size of the cutting tools
becomes comparable with pore size, this induces significant
fluctuations in the cutting force [20] leading to excessive
tool wear and/or breakage. Cross-correlation between
micromachining parameters and porosity has been relatively little investigated in the literature and therefore represents
one of the objectives of the current work.
Intuitively, it is easy to understand that each micromilled
slot cut in a porous sample will be characterized by a
unique process signature that is strongly dependent on
porosity distribution. This difference was outlined in the
past [20] by comparing resultant cutting forces during
micromilling of solid and foam Ti. As shown in Fig. 1, the
resultant cutting force amplitude FR is increased (0.52 N vs.
1.05 N) and peak-to-valley size of the cutting force is
reduced (0.4 N vs. 0.1 N) when micromilling porous
compared to solid titanium. The cutting forces were

in porous material geometry modification by allowing a
superior control of surface roughness and even porosity
closure amount. These two surface characteristics influence
to a large extent the friction coefficient between implant
and surrounding bone tissue. However, two main interrelated technical challenges are associated with obtaining the
desired surface geometry and roughness on outer porous

surfaces: accurate characterization of real 2D/3D porosity
values and optimization of micromachining process parameters with respect to the physical porosity amounts. From
this perspective, the present study focuses on optical image
analysis of porosity with respect to material removal
process through micromilling operations.
Physical–mechanical properties of porous metals and
their adherence with bone cells significantly depend on 2D/
3D porosity characterized through parameters like: quantity
of pores (i.e., the fractional porosity), interconnectivity,
size, morphology, permeability, and spatial distribution [8–
11]. Classical non-destructive optical image analysis methodologies have been successfully applied in the past to
analysis of cell morphology and microstructure of porous
metals [8]. In most cases, pore size distribution and shape
analysis was performed by means of commercial image
analysis programs. The main drawback of this approach
resides in destructive techniques involved in preparation of
sample surface because image analysis is significantly
dependent on appropriate differentiation between solid
material and internal cavities. Optical image analysis is
limited to 2D spatial analysis only. Nevertheless, optical
micrographs were used before in the context of porous Ti
materials used in load-bearing implants [12]. Within
machining environment, the most common application of
vision-based methods is represented by tool wear analysis
and monitoring [13–17].
New developments in X-ray-based microcomputer tomography (micro-CT) have opened advanced non-destructive
Fig. 1 Comparison of resultant
cutting forces measured during
micromilling of solid and porous
Ti [20]

1.2

solid Ti

1.0

FR [N]

0.8
porous Ti

0.6
0.4
0.2
0.0
-0 .5

0.0

0.5

1.0

1.5

2.0

l [mm]

2.5

3.0

3.5

4.0


Int J Adv Manuf Technol (2012) 60:841–851

recorded in space domain as a function of cutting tool
center position l.
The present research was primarily focused on the
investigation of the correlations between optically determined
porosity and cutting forces recorded during micromilling of
porous titanium foams. For this purpose, an original imagebased methodology was developed to estimate the percentage
of solid material removed during each revolution of the cutter
along a linear tool path. Once the porosity profile along the
tool path became available, standard statistical measures were
used to determine the amount of correlation between porosity
and cutting force measured dynamically along the intended
tool path. All details of this analysis are presented in the
following sections.

2 Optical assessment of porosity in context
of micromilling operations
Functional parts and components made from Ti–6Al–4V
foams are characterized by a complex 3D porosity with
significant variations in pore size and morphology as shown

in Fig. 2. For these materials, micromilling with cutting
tools having a diameter smaller than 0.4 mm can be used as
a secondary shaping procedure that is capable of achieving
high surface precision and complex geometries. Since these
procedures are accounted for as finish operations, the axial
cutting depth is typically small, rarely exceeding 20 μm.
This allows treatment of the porosity as a function
dependent on the presence of solid material along the
microtool path trajectory. This study considers only X–Y
(planar/linear) motions, however, the developed method can
be extended also to X–Y–Z (tridimensional) microtool path
trajectories.
During microslot end-micromilling, each cutting tooth
moves along the trochoidal trajectory formed as a superposiFig. 2 Typical sample of a
porous Ti foam (Ti–6Al–4V)
characterized by a complex 3D
porosity structure

843

tion of tooth rotational motion around the cutter axis and linear
translation of tool center along the intended tool path, as shown
in Fig. 3. It is important to emphasize that a clear distinction
has to be made between the volumetric porosity defined as
the fraction of the voids spread throughout the volume of the
porous material (a common physical property of porous
materials) and cutting-force-related porosity defined as the
fraction of the voids encountered by the cutting edge of the
tool along its trochoidal trajectory. Obviously, when analyzing the correlations between porosity and cutting force, only
the distribution of the latter is relevant. As a result, the

remainder of this section focuses on the assessment of
cutting-force-related porosity, essentially derived in a particular manner from generic volumetric porosity.
In this regard, the area swept per ith tool revolution Ai,
represents the geometric difference between two consecutive,
ith and (i−1)-th, cutting tooth trajectories. This area can be
approximated as the difference of two consecutive circles
separated by a feed per tooth value, f (Fig. 3). When
machining solid material, the area Si and the volume of
material removed per tooth revolution Vi, will be always
constant: Si, Vi =const. Moreover, the swept area Ai, and area
of material removed, Si, per tooth revolution are absolutely
identical.
By contrast, the presence of randomly distributed porosity
makes the estimation of the amount of solid material removed
more difficult. In this situation, the area and the corresponding
volume of material removed per tooth revolution exhibit large
variations, Si, Vi =var, although the swept area Ai, remains
unchanged (Fig. 4). Proportion of the material within the
swept area per tooth revolution pi can vary anywhere
between 0 and 1 according to the following relationship:

pi ¼

Si
Ai

ð1Þ
pores
solid Ti

0.2 mm

1x1x1mm


844
Fig. 3 Area removed during
one tooth revolution

Int J Adv Manuf Technol (2012) 60:841–851

trochoidal tooth cutting trajectory

Y

f
i-1
i

X

v

machined
slot
cutting tool
workpiece

In Eq. 1, pi =0 corresponds to the case when swept area
Ai is fully covered by a pore (void), and pi =1 corresponds

to the case when only solid material is removed. Because
cutting force is directly proportional to the volume (area)
of material removed, it is important to recognize that
cutting forces will vary with respect to porosity, specifically the fraction of solid material contained within the
area removed during one complete revolution of the cutter.
This fraction will be quantified herein based on optical
considerations.

Area swept
per tooth
revolution Ai

The absolute novelty of the methodology proposed for
estimation of the porosity fraction per tool revolution
resides in correlation of the optical images of the micromilled surface with micromachining parameters recorded in
real time during cutting (cutting forces and axial positions).
According to the proposed technique, the swept area is
determined by overlapping cutting tooth path trajectories
with optical images acquired for the top surface of the
sample, followed by selection of the “on” pixels located
between two consecutive tooth revolutions. The accuracy of

Fig. 4 Visual appearance of
porosity enclosed within
the area swept in one cutter
revolution

f

i

i-2

porosity

tool diameter

i-1
v

Si-1, Vi-1
Si, Vi

Si-1 ≠ Si
Vi-1 ≠ Vi


Int J Adv Manuf Technol (2012) 60:841–851

845

the method is primarily influenced by optical image
resolution Δh, and spatial sampling period of linear tool
motions Δl. Figure 5 depicts a graphical representation of
the area covered in one tooth revolution. Each tooth path is
offset with a distance equal to feed per tooth f. After swept
area and corresponding pixels for each tooth revolution are
determined, the porosity and the material presence
contained within swept areas of single tool revolutions
can be calculated along the machining direction (tool path

trajectory). If f is comparable to optical image resolution
Δh, calculation of the percentage of solid material
contained within the area swept in one cutter revolution
(2D case) can be reduced to an 1D problem by estimating
the proportion of solid material positioned along the
circumference of circular tooth trajectories. This can be
achieved by counting the “on” pixels that are intersected by
these circular paths. By doing this, the original “static”
image of porosity will be converted into a “dynamic” image
of porosity as observed by the cutting edge of the tool. The
key element of this image conversion resides in tooth path
“linearization”, which essentially means that the semicircular tooth path is unwrapped along a straight line. This
unwrapping is accompanied by an adequate mapping of
pixels encountered along the semicircular tooth trajectory,
as shown in Fig. 5. To preserve the aspect ratio of

“dynamic” image pixels, the semicircular tooth trajectories
have to be sampled at intervals equal to pixel size Δh.
In order to perform the aforementioned conversion,
the image has to be initially pre-processed to a binary
format such that it contains only white (“on”) and black
(“off”) pixels that correspond to solid material and
pores, respectively. As a result, the surface is converted
to an array [Xi, Yi, Zi] defined only through zeros (Zi =0
for solid material) and ones (Zi =1 for voids). After that,
tooth trajectories are calculated based on the position in
real time of the bottom tool center and tool size. The
newly obtained tooth trajectories are then overlapped with
the optical image, such that each tooth trajectory becomes
a vector with {xi, yi, zi} components. The zi values are

initially unknown as they correspond to material presence
for a particular (xi, yi) location. It is also necessary to note
that the sampling period of the tooth trajectory,$l ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxj À xjÀ1 Þ2 þ ðyj À yjÀ1 Þ2 , is set to be equal to the pixel
resolution of acquired optical image Δh. In the next
iteration step, all zi within the swept area are determined as
a subset of the surface matrix [Xi, Yi, Zi] as following:

zi ¼fZi jfxi ; yi g \ ½Xi ; Yi Šg

Fig. 5 Estimation of material
presence within the area swept
in one cutter revolution through
tooth path linearization

f

porosity

material

“on”

machining
direction

Ai-2
Ai-1
Ai

“off”

ð2Þ


846

Int J Adv Manuf Technol (2012) 60:841–851

Based on these definitions, the ratio between the number
ni of “on” pixels (zi =0) and total number of pixels
intersected by particular tooth trajectories Ni will be
fractionally identical to the amount of solid material
distributed within the area swept in one cutter rotation:
pi ¼

ni
Ni

ð3Þ

An example of the pixel quantification procedure
described above is provided in Fig. 6. As it can be noticed,
the circular tooth path has been sampled at distances equal
to pixel size (Δh). Then, the “on” or “off” attribute of each
pixel corresponding to the sampled locations—identifiable
through an adjacent star—has been recorded and eventually
transcribed into a linear column array as part of the
“linearization” procedure involved in cutting-force-related

porosity signature analysis.

3 Experimental verification
Applicability of the developed approach for vision-based
estimation of porosity was verified experimentally in the
context of porous Ti foam micromilling. Real-time measurements of cutting forces and motions were performed
during cutting trials. Then, the actual porosity per single
tooth revolution was estimated and subsequently correlated
with per revolution-averaged cutting force amplitudes.
3.1 Material description
The Ti–6Al–4V foam used in micromilling trials was
manufactured by Industrial Material Institute of the National
Research Council of Canada (Boucherville, Quebec) by

h= l
sampled
tooth path
location

porosity

i

solid
material

means of a patented powder metallurgy procedure [21].
Structural and mechanical properties of the Ti foam are
presented in Table 1. The samples had a cubic shape with an
approximate side length of 13 mm (Fig. 2).

3.2 Experimental setup and methodology for micromilling
Micromilling experiments were carried out on a custom-built
five-axis CNC micromilling system equipped with an airbearing spindle capable of rotational speeds between 5,000
and 100,000 rpm range. The system is capable of providing a
static positional accuracy of 1 μm over a 300-mm maximum
travel range.
Cutting experiments were performed as micromilling of
linear slots along X-axes using a two-flute, uncoated,
tungsten carbide flat-end micromill having a diameter of
800 μm diameter and helix and clearance angles of 25° and
6°, respectively. Linear slots having a width of 800 μm,
depth of 40 μm, and length of 12 mm were machined on
the top face of the foam Ti specimen with a feed rate of
120 mm/min and spindle rotational speed of 10,020 rpm.
An in-house developed real-time data acquisition system
written in Labview was used to measure the X, Y, and Z
components of the cutting force with a Kistler dynamometer (type 9256C2) and a Kistler dual mode charge
amplifier (type 5010B). The X, Y, and Z cutting motions
were tracked based on a signal from position encoders and
three-phase current consumed by spindle drive (Fig. 7).
Spindle current and cutting forces F(t)={FX(t), FY(t),
FZ(t)}, were recorded with a sampling frequency of
25 kHz with a minimum threshold of 0.002 N and were
filtered out of the resonance frequency of the dynamometer
(around 4.8 kHz). Simultaneously, the cutting position l(t)=
{lX(t), lY(t), lZ(t)}, along the tool path trajectory was
recorded with a sampling frequency of 500 Hz. Further, a
mean value of the cutting force in each direction was
calculated for each tooth revolution using spindle current
signal as point of synchronization in space domain. After

that, cutting forces were rearranged as a spatial function of
the cutting tool path trajectory F(l)={FX(l), FY(l), FZ(l)},
using synchronization in time–space domain between
cutting forces and their corresponding axial positions.
Table 1 Structural and mechanical properties of Ti foams [3]

h

analyzed
pixels

Fig. 6 Analyzing pixels and their associated material/void attributes
along circular tooth paths

Property

Value

Density
Porosity
Pore size
Surface area
Compressive yield strength
Compressive elastic modulus

1.6–2.25 g/cm3
50–65%
50–400 μm
0.05 m2/g
25–125 MPa

5–20 GPa


Int J Adv Manuf Technol (2012) 60:841–851

847

I(t)

Tool outline

Tool path

t
FR(t)
Microslot
profile

FRi = FR(ti)

l

li
Tooth engagement
area

ti

t

FR(l)

FRi

l(t)
li = l(ti)

li

l
space domain

ti-1

ti

ti+1

t

time domain

Fig. 7 Achieving space domain synchronization of porosity and cutting force through concurrent time domain representations

3.3 Experimental setup and methodology for image analysis
of porosity
Optical image acquisition of micrographs was performed
after micromachining by using a CANON EPS 450D
camera with a 12.2-megapixel image sensor equipped with
an EF 100 mm macro lens (f/2.8 Macro USM) that provides

a spatial image resolution Δh of approximately 6 μm/pixel.
A custom indirect LED light was used to reduce light
reflections and enhance picture contrast for better porosity
assessments. In this study, the length of the machined slot
was 12.56 mm and the feed per tooth f was 0.012 mm. Due
to the comparable values of Δh and f, estimations of the
solid material fraction located along the circular tooth
trajectory can be performed accurately through the proposed 1D approach as described in Section 2, above. It will
be assumed here that the porosity has a minimal cross
sectional variation over the investigated axial depth of cut
(20–40 μm) that is characteristic to finish operations. As a
result, the porosity amount measured on top of the sample
surface before and after the microslot cutting operation will
be assumed to be approximately identical.
The general methodology for vision-based estimation of
porosity involves the following steps:
a. Acquire and convert the optical image of the micromilled surface to the binary format
b. Synchronize tooth position with respect to space
domain and superimpose its trajectory on the optical
image of the surface

c. Perform circular to linear tooth path conversions (e.g.,
“linearizations”)
d. Calculate the total number of optical pixels intersected
by a singular tooth path Ni
e. Calculate the total number of “on” pixels ni (zi =0)
along the investigated tooth path
f. Establish the fraction of solid material removed pi
during each revolution of the tool (Eq. 3)
g. Track the amount of solid material removed with

respect of axial position of the tool p(l)
In step (a), gray-scale micrographs of the sample surface
captured by a high-resolution camera are converted to
binary format by using imaging threshold set by Otsu’s
method [22] to clearly delimit the porosity (black) and solid
material (white). This value of threshold ensures accurate
conversions of voids into black, especially when considering pore boundaries (Fig. 8). Any dark spot covering less
than 10 pixels was considered noise (light reflected on nonplanar foam inclusions) and therefore was filtered out. The
accuracy of thresholding procedure used was validated by
comparing void dimensions against those obtained with a
Wyko interferometric microscope.
Figure 9 illustrates the steps (b) to (f) involved in visionbased estimation of the porosity. First, circular trajectories
of the cutting tooth rotation along tool path motion are
superimposed on the black and white micrographs according to step (b). Then, in step (c), the circular tooth trajectory
is “linearized”. After performing the quantitative assessments
involved in steps (d) through (f), the outcome of step (g) is a


848

Int J Adv Manuf Technol (2012) 60:841–851

Fig. 8 Converting gray porosity
micrographs (top) into binary
format (bottom) through
thresholding

solid
material

porosity

Fig. 9 Optical image
processing and porosity
assessment

Circular tooth trajectory

Binary micrograph
of micromilled
porous surface

2r

A

machining
direction
A’

Tooth path conversion

A
Porosity
cross-section
πr

“Linearization” of
tooth trajectory

Instantaneous
amount of solid
material removed
along tool path

Percentage of material p(l), %

A’

Calculation of Ni, ni, pi,

100

p(l)
50
p(A)

0

A
Position of cutting tool center l, mm


Int J Adv Manuf Technol (2012) 60:841–851
100

849
2.5

p(l )

90
80

2.0

60

1.5

50
40

FR [N]

p(l) [%]

70

1.0
FR(l)

30
20

0.5
Stable cutting

10
0

-1

0

1

2

3

4

5 6 7
l [mm]

8

9

0.0
10 11 12 13

Fig. 10 Comparison between optically determined porosity and
cutting force in space domain

function p(l) expressing the fraction of solid material
removed for any position of the tool along the intended tool
path. As noticeable here, the cutting-force-related porosity
(Fig. 9b) whose spatial variation along the tool path is plotted
in Fig. 9c represents nothing else but a transformed (mapped)

form of the real surface porosity (Fig. 9a) as observed at the
top of the surface to be micromilled. Evidently, the
transformation function between the two types of porosities
depends primarily on the size and cutting regime of the
micromilling operation and it relies on the “linearization”
process described above.
The real surface porosity constitutes in fact a discrete value
of the volumetric porosity as measured at a certain depth within
the sample. Conversely, the volumetric porosity averages all
real surface porosities measured throughout the height of the
sample. However, other than on the originally exposed exterior
surfaces of the raw sample, noninvasive accurate determinations of the real surface porosity are relatively difficult at the
current development level of technology.

3.4 Preliminary experimental analysis of correlation
between porosity and cutting forces
To analyze the interplay between porosity and cutting
forces from a statistical standpoint, they should be both
dependent on the same variable. For this application, the
most rational choice for the common variable would be the
space domain, specifically the axial position of the micromill along its trajectory l(t)={lX(t), lY(t), lZ(t)}. As a result,
both porosity and cutting forces were expressed as a
function of l, respectively. Figure 10 depicts a space
domain comparison between the amount of solid material
removed and the corresponding signature of the resultant
cutting force. A simple visual inspection of the graph reveals
that cutting force variations follow closely the percentage of
solid material removed in each tool revolution.

1

Coherence value, dimensionless

Fig. 11 Coherence function between proportion of material and
resultant cutting force

While correlations between cutting forces and real top
surface porosity can be established in an absolutely similar
manner as those between cutting forces and linearized/
transformed porosity, they will be always lower and thereby
less convincing and/or useful. Cutting-force-related porosity
reflects in a more accurate manner material discontinuities as
perceived by the cutting edge of the tool. However, as Fig. 9
itself shows, when material voids are large enough, then
significant drops will occur on both transformed and actual
porosities plots, and they will propagate further on cutting
force signature. To summarize these observations, while
linearized porosity represents a more accurate instrument for
prediction of the cutting force variation, the real top surface
porosity enables quick visual estimations of the correlations
between material discontinuities and cutting forces pattern.
These visual correlations tend to be more obvious in case of
larger material discontinuities characterized by more pronounced effects on cutting force signature.

0.8

0.6

0.4

0.2
1.6 mm-1

0

0

1

2

3

4

5

6

Spatial frequency, mm

7
-1

8

9

10

850

Cross-correlation between the fraction of material removed per revolution p(l) and resultant cutting force
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
FR ðlÞ ¼ FX2 ðlÞ þ FY2 ðlÞ was analyzed further by means
of traditional statistical characteristics, such as correlation
coefficient and coherence function. While the reader is
referred to a specialized literature [23] for detailed
mathematical formulations of the two aforementioned
statistical metrics, it will be briefly reminded here that
when analyzing the similarities/differences between two
random signals, there are few possible assessment criteria.
For instance, correlation coefficient represents a good
measure of the strength of linear dependence between the
two analyzed variables. While it is relatively difficult to
draw “hard” numerical boundaries between strong and
weak correlations, those with correlation coefficients above
0.7 are generally considered as reasonably strong, since
more than 70% of the investigated output (cutting force
signature, in this case) is linearly dependent on the analyzed
input (cutting-force-related porosity).
Unlike correlation coefficient, coherence is calculated
in frequency domain and it is capable to quantify the
amount of power transferred between input and output.
Elevated coherence values (typically over 0.7) are
usually regarded as a good indication of the fact that
at most of the cutting-force-related porosity frequencies,
more than 70% of the cutting force variations are
caused by material discontinuities. Extremely important,

if the relationship between input and output is nonlinear
or noncausal, then coherence values can be erroneous.
This practically explains the complementarity of correlation coefficient and coherence analyses and provides
basis for their combined calculations.
It was found that resultant cutting force is dependent
on solid material fraction with a correlation coefficient
of 75% during stable micromilling that excludes
entering and exiting cutting stages. Coherence function
between p(l) and FR(l) displayed in Fig. 11 reveals that a
significant correlation exists between them in lowfrequency area, e.g., below 1.6 mm−1, that is characterized
by high coherence values (> 0.7). While quantitative
translations of this frequency threshold value back into
the physical domain are difficult, it can be asserted in a
more qualitative manner that this constitutes an indication
that the dependence between cutting forces and material
discontinuities is more pronounced when pore sizes are
relatively large compared to those of the cutting edge/tool.
When pore dimensions are small, they can be neglected,
such that other input variables (process parameters) will
acquire a more prominent role on cutting force pattern.
The explicit influence of micromilling process parameters
on porosity/cutting force interplay has been analyzed
elsewhere [24].

Int J Adv Manuf Technol (2012) 60:841–851

4 Summary and conclusions
This study proposes an original image analysis approach for
optical determination of porosity in context of micromilling
operation. The developed technique extracts from an optical

image (micrograph) the area of material removed through
micromilling and then calculates the fraction of solid
material removed in each tool revolution. The proposed
methodology was experimentally tested by micromilling a
linear slot in porous Ti foam (Ti–6Al–4V). The resultant
cutting force is dependent on the proportion of solid
material with 75% correlation during stable machining
and a significant correlation exist between them particularly
at low frequencies of the cutting force. From the results
obtained, the following conclusions can be drawn:
1. A new innovative methodology for optical image analysis
of porous surfaces and their associated parameters was
developed. The effectiveness of the proposed method was
validated in the context of finish micromilling operations
performed on porous titanium samples.
2. The developed approach involves pixel level quantifications to determine the fraction of solid material
removed during each cutter revolution. However, pixel
counts might not be accurate enough at low optical
image resolutions. More accurate estimation of the area
swept per tooth revolution can be achieved by applying
an alternative approach where the partial area of each
pixel intersected by a tooth trajectory will have to be
estimated (2D approach).
3. It was proved experimentally that the resultant cutting
force is dependent on the proportion of material with a
correlation coefficient of 75% during stable machining.
This correlation is valid only for low-frequency cutting
force variations. In the high-frequency domain, it is
expected that the dynamics of the cutting process exerts
a predominant influence on variation of the cutting forces.

The proposed approach opens up new opportunities in
analysis, monitoring, optimization, and control of the
cutting process during finish micromilling of porous foams
aiming generation of optimized microstructural surface
geometries in terms of both surface roughness and porosity
closures.

Acknowledgments This paper is the result of collaboration between
The University of Western Ontario and NRC-IMI-CAMM. The
authors thank Louis-Philippe Lefebvre, Maxime Gauthier, Eric Baril,
and Sylvain Pelletier, from NRC-IMI, and Hugo Reshef, Mike
Meinert, and Suwas Nikumb from NRC-IMI-CAMM for their
continued technical and partial financial support in this work. The
financial support was also provided in part by Natural Sciences and
Engineering Research Council of Canada.


Int J Adv Manuf Technol (2012) 60:841–851

References
1. Banhart J, Weaire D (2002) On the road again: metal foams find
favor. Phys Today 55:37–42
2. Ryan G, Pandit A, Apatsidis DP (2006) Fabrication methods of
porous metals for use in orthopaedic applications. Biomaterials
27:2651–2670
3. Baril E, Lefebvre L-P, Thomas Y, Ilinca F (2008) Foam-coated
mim gives new edge to titanium implants. Met Powder Rep
63:46–50
4. Smith GT (1998) Getting the measure of pm machinability. Met
Powder Rep 53:31–35

5. Eisenmann M (2000) Porous P/M technology, ASM handbook,
vol 7. ASM International, New York
6. Alizadeh E (2008) Factors influencing the machinability of
sintered steels. Powder Metall Met Ceram 47:304–315
7. Bordatchev E, Nikumb S (2008) Fabrication of moulds and dies
using precision laser micromachining and micromilling technologies.
J LMN 3:175–181
8. Banhart J (2001) Manufacture, characterisation and application of
cellular metals and metal foams. Prog Mater Sci 46:559–U553
9. Medraj M, Baril E, Loya V, Lefebvre LP (2007) The effect of
microstructure on the permeability of metallic foams. J Mater Sci
42:4372–4383
10. Baril E, Mostafid A, Lefebvre LP, Medraj M (2008) Experimental
demonstration of entrance/exit effects on the permeability measurements of porous materials. Adv Eng Mater 10:889–894
11. Lefebvre LP, Banhart J, Dunand DC (2008) Porous metals and
metallic foams: current status and recent developments. Adv Eng
Mater 10:775–787
12. Thelen S, Barthelat F, Brinson LC (2004) Mechanics considerations for microporous titanium as an orthopedic implant
material. J Biomed Mater Res A 69A:601–610

851
13. Zhang YH, Zhang YC, Tang HQ, Wang LH (2010) Images
acquisition of a high-speed boring cutter for tool condition
monitoring purposes. Int J Adv Manuf Technol 48:455–460
14. Kerr D, Pengilley J, Garwood R (2006) Assessment and visualisation of machine tool wear using computer vision. Int J Adv
Manuf Technol 28:781–791
15. Wang W, Wong YS, Hong GS (2005) Flank wear measurement by
successive image analysis. Comput Ind 56:816–830
16. Wang WH, Hong GS, Wong YS (2006) Flank wear measurement
by a threshold independent method with sub-pixel accuracy. Int J

Mach Tools Manufact 46:199–207
17. Atli AV, Urhan O, Erturk S, Sonmez M (2006) A computer visionbased fast approach to drilling tool condition monitoring. Proc
IME B J Eng Manufact 220:1409–1415
18. Malcolm AA, Leong HY, Spowage AC, Shacklock AP (2007)
Image segmentation and analysis for porosity measurement. J
Mater Process Technol 192:391–396
19. Prokop J, Sveda L, Jancarek A, Pina L (2009) Porosity
measurement method by X-ray computed tomography. Fractography of Advanced Ceramics Iii 409:402–405
20. Tutunea-Fatan OR, Abolghasemi Fakhri M, Bordatchev EV
(2010) Porosity and cutting forces: from macroscale to microscale machining correlations. Proc IME B J Eng Manufact
225:619–630
21. Lefebvre LP, Thomas Y (2003) Method of making open cell
material, United States Patent No. 6660224
22. Otsu N (1979) A threshold selection method from gray-level
histograms. IEEE Trans Syst Man Cybern 9:62–66
23. Bendat JS, Piersol AG (1993) Engineering applications of
correlation and spectral analysis. Wiley, New York
24. Abolghasemi Fakhri M., Bordatchev EV, Tutunea-Fatan, OR
(2011) Evaluation of the relative effect of process parameters on
porosity-cutting force correlation in micromilling of porous
titanium, under revision for Machining Science and Technology.


Int J Adv Manuf Technol (2012) 60:853–863
DOI 10.1007/s00170-011-3656-0

ORIGINAL ARTICLE

Feasibility of mild hard turning of stainless steel using coated
carbide tool

M. Y. Noordin & D. Kurniawan & Y. C. Tang &
K. Muniswaran

Received: 9 January 2011 / Accepted: 19 September 2011 / Published online: 6 October 2011
# Springer-Verlag London Limited 2011

Abstract Hard turning has been explored as an alternative
to the traditional processing technique used to manufacture
parts made of hardened steels. However, advanced cutting
tool materials for hard turning applications are relatively
expensive. The continuous developments in carbide tool
material and its coating technology have offered inexpensive cutting tool alternatives for a mild range of hard
turning operations. Commercially available TiAlN-coated
carbide tool is utilized in this study to perform hard turning
of stainless steel within the mild range (47–48 HRC) at
various cutting parameters, i.e., cutting speed and feed.
Empirical models to measure its performance by quantifying the effect of the cutting parameters to the tool’s service
lifetime and the machined workpiece’s surface roughness
are developed. The coated carbide tool performed hard
turning with fair tool life and fine surface finish, especially
at low cutting parameters as shown by the models’
solutions for the optimized input selection.
Keywords Hard turning . Coated carbide . Tool life .
Surface roughness . Design of experiments

1 Introduction
The need for high performance components has been
responded partially by the use of hardened steels (above
45 HRC) as the appropriate raw materials. Hardened steel’s
properties of high wear resistance and compressive strength

M. Y. Noordin (*) : D. Kurniawan : Y. C. Tang : K. Muniswaran
Department of Manufacturing and Industrial Engineering,
Universiti Teknologi Malaysia,
81310 UTM Skudai,
Johor Bahru, Johor, Malaysia
e-mail:

meet the demands in automotive and tool and die industries
[1]. The usual technique to manufacture hardened parts
involves three sequential steps, i.e., rough machining of
unhardened steel, heat treating the steel to the required
hardness, and finish machining to the required dimensional
accuracy. The introduction of hard turning using tools with
high hot hardness (e.g., crystalline cubic boron nitride/cBN
and ceramic) has simplified the process flow by allowing
the steel blank to be machined to its final dimension in the
hardened state [2]. This technique offers a profitable
alternative for finish machining. High material removal
rate and relatively low tool cost compared to the incumbent
grinding as the finishing operation are some of the
economical benefits. Additionally, hard turning has been
successfully performed in dry conditions [3].
Despite its significant advantages, the lack of data
concerning surface quality and tool wear for the many
combinations of workpiece and cutting tool impedes the
acceptance of hard turning by the manufacturing industry
[4]. Moreover, the common tools used in hard turning are
relatively high in unit price.
Some applications in the mold and die industry have
been identified to require parts made of hardened steels

within the mild range of hard turning (45–50 HRC). Using
advanced and expensive cutting tools for these applications
may hinder the economical benefit of hard turning.
Coated carbide tool is the proposed cutting tool
alternative. Cemented carbide is the most popular cutting
tool; in the global market, it makes 75% of indexable insert
[5], and it has been significantly developed over the years
to improve its toughness and hardness combination.
Evidenced by our previous hard turning of stainless steel
(43–45 HRC) using coated carbide tools [6], it is likely that
coated carbide has the potential to machine steels of even
higher hardness within the mild range of hard turning. Fine


854

grained substrate, better binder that optimizes strength
and toughness, and improved coating using the physical
vapor deposition (PVD) technique are some developments in fortifying coated carbide tools currently available
commercially [7].
To encourage machine shops to fully adopt hard turning,
clarification on the aspects of the tool life and machined
surface’s quality should be addressed. This also applies to
the strategy proposed in this study, which is to conduct mild
hard turning without cutting fluid using a coated carbide
tool. The machining cost per part is a function of tool life
and, hence, the demand for long tool life from machine
shops. On the other hand, fine surface finish should be
produced through finish machining as requested by the
users of the machined parts to meet the specific requirements of certain applications [8].

In this study, empirical models which quantify the
influences of cutting parameters, i.e., cutting speed and
feed, on the machining responses, i.e., tool life and surface
roughness, are developed. Hwang and Lee [9] and Noordin
et al. [10] have shown that the developed empirical models
using the design of experiment approach are adequate to
represent the performance of a coated carbide tool.
Therefore, the method was adapted for this experiment so
that the resulting data could be objectively analyzed.
Considering the extreme task the coated carbide tool is
subjected to, the empirical models are used to determine at
which cutting parameters it can perform and to what extent.

Int J Adv Manuf Technol (2012) 60:853–863

chemical composition of 0.38% C, 0.9% Si, 0.5% Mn,
13.6% Cr, 0.3% V, and balance Fe, as disclosed by the
manufacturer (Assab Steel).
The cutting tool is a tool with fine grained WC-6 wt.%
Co substrate covered with 3.0 to 3.5 μm TiAlN coating
(Fig. 1a). This Kennametal coated carbide tool has an ISO
designation of CNMG 120408. The tool was mounted on a
holder with an ISO designation of MCLNL 1616-H12,
giving the 10° rake angle, 5° relief angle, and −5° side
cutting edge angle. The tool holder actually positions
the cutting tool at −5° angles as hard turning is
commonly set at the negative rake angle. However,
since the tool has 10° chip breaker profile (Fig. 1b), the
effective rake angle was positive. The tool wear was
measured according to ANSI/ASME B94.55M-1985 standard, subjected to the maximum flank wear width

(VBmax) within the nose radius of the tool (zone C).
The tool life criteria were set at maximum flank wear
width of 0.14 mm or when the tool was severely broken.
For the purpose of achieving finish machining results, an
additional tool life criterion was also set at the machined
surface roughness beyond 1.6 μm. Force measures were
taken by using a three-component dynamometer system.
Radial force (Fr), cutting force (Fc), and feed force (Ff)
were the output of the dynamometer system. Image
capturing was conducted by utilizing an optical microscope and a scanning electron microscope.
2.2 Cutting parameters

2 Experimental details
2.1 Cutting conditions
The hard turning tests were performed on a two-axes CNC
lathe machine. The workpiece material is stainless steel
with thorough hardness of 47–48 HRC. The stainless steel
is an AISI 420 (modified) martensitic type with the

Fig. 1 Cross-section images of
the coated carbide tool: a
coating thickness of 3–3.5 μm
and b positive rake angle
of 15°

The range of cutting parameters was within those of finish
machining values provided by the tool’s manufacturer.
Also, the values of the feed were selected such that,
theoretically, the resulting surface roughness would fall
below 1.6 μm.

In order to determine if there was a relationship between
the input (cutting parameters) and the response (tool life
and surface roughness) variables investigated, the data


Int J Adv Manuf Technol (2012) 60:853–863

855

collected must be analyzed using regression. A regression
was performed whereby an observed, empirical variable
(response) is approximated based on a functional relationship between the estimated variable, yest, and one or more
input variables, x1 and x2. In the case where a nonlinear
relationship existed between a particular response and two
input variables, a quadratic equation,

Table 1 Design layout of the experiment

yest ¼ b0 þ b1 x1 þ b2 x2 þ b3 x1 x2 þ b4 x21 þ b5 x22 þ error

ð1Þ
may be used to describe this functional relationship. The
least square technique was being used to fit a model
equation containing the said regressors or input variables by
minimizing the residual error measured by the sum of
square deviations between the actual and the estimated
responses. This involved the calculation of estimates for the
regression coefficients, i.e., the coefficients of the model
variables including the intercept or constant term, for
statistical significance test.

The lower and upper limit values of the input variables
have been selected according to the recommendation by the
tool manufacturer for cutting parameters for martensitic
stainless steel with hardness of up to 48 HRC. Considering
that the carbide tool was designed for finish machining and
that the depth of cut was kept constant at 0.4 mm, the
selected lower and upper limit values for feed were 0.1 and
0.16 mm/rev, respectively, and for cutting speed, they were
100 and 170 m/min, respectively. The values selected for
the middle limit were 130 m/min for cutting speed and
0.125 mm/rev for feed.
A commercial statistical analysis software (Design
Expert 6.0 of Stat-Ease Inc.) was used for the convenience
of designing the experiments and analyzing the results.
Using a three-level factorial design with two factors as the
input and repeated twice at the center point, the design
requires a total of 11 runs (Table 1).

Standard

Cutting speed

Feed

Coded form

(m/min)

(mm/rev)

x1

1

100

0.1

−1

−1

2
3

130
170

0.1
0.1

0
1

−1
−1

4

100

0.125

−1

0

5
6

130
170

0.125
0.125

0
1

0
0

7

100

0.16

−1

1

8
9

130
170

0.16
0.16

0
1

1
1

10

130

0.125

0

0

11

130

0.125

0

0

The ratio of the maximum tool life to the minimum one
is 12.22. A ratio of more than 3 indicates that power
transformation is required to develop the empirical model.
One way to define which power transformation is appropriate is by diagnosing the Box–Cox plot of the tool life
data. The lowest point of the plot that might result in the
minimum residual sum of square in the transformed model
(λ value) was 0.13. Logarithmic transformation (λ=0) was
selected as the closest value to the actual one and is deemed
as the most appropriate transformation.
Fit summary output to determine the most suitable
regression model was then evaluated. Having the least
probabilistic value, Prob>F, and the most insignificant lack
of fit, a quadratic model was selected to represent the tool
life data.
Analysis of variance was then performed to test the
significance of the regression model and the individual
coefficients of the quadratic model. The maximum probabilistic value of 5% was set for the model and its
coefficients to be considered significant. The quadratic

3 Results and discussion
3.1 Tool life
The experimental results for tool life are presented in Fig. 2.
The tool life shortens with increasing cutting speed and

feed. The maximum tool life of 30.5 min was obtained by
selecting the low cutting speed–low feed combination
whereas the minimum tool life of 2.5 min was achieved
when the high cutting speed–high feed combination was
chosen. The ability of the coated carbide tool to withstand
over 10 min of service life indicates that coated carbide
performs well for materials with hardness up to 48 HRC
within the selected cutting parameters.

x2

Fig. 2 Tool life at various cutting speeds and feeds


856
Table 2 Analysis of variance
(partial sum of squares)
for reduced quadratic model
of the tool life as the response

Int J Adv Manuf Technol (2012) 60:853–863

Source

Sum of
square

Degrees of
freedom

Mean
square

F value

Prob>F

Model

5.929

3

1.976

119.008

<0.0001

x1

5.083
0.544
0.617

1
1
1

5.083

0.544
0.617

306.056
32.732
37.147

<0.0001
0.0007
0.0005

0.116
0.065

7
5

0.017
0.013

0.507

0.7663

Pure error

0.051

2

0.026

Cor total

6.045

10
R2
Adequate precision

x2
x12
Residual
Lack of fit

model has probabilistic value well below 0.05, but not all
its coefficients do. The square of feed, having a probabilistic value of 0.65, and the cutting speed–feed interaction,
having a probabilistic value of 0.43, were considered not
significant and thus removed to improve the model. The
reduced quadratic model was then retested (Table 2). The
reduced model’s coefficient of determination, R2, is now
0.98. This very close to unity R2 indicates that the model
closely approximates the tool life data. The model’s
adequate precision ratio, which compares the range of the
predicted values at the design points to the average
prediction error, is well beyond the minimum adequacy
limit of 4.
The obtained final equation of the model could then be
presented in terms of coded factors as:
ln T ¼ 1:63 À 0:92x1 À 0:30x2 þ 0:49x21

ð2Þ

where T is the tool life and x1 and x2 are the cutting speed
and feed in coded factors, respectively, or in terms of actual

Significant

Not significant

0.981
31.389

factors, the tool life of the coated carbide tool can be
predicted using the equation
ln T ¼ 13:737 À 0:134v À 9:975f þ 0:0004v2

ð3Þ

where v is cutting speed (meters per minute) and f is feed
(millimeters per revolution).
Inspection to some diagnostic plots of the model was
done to test the statistical validity of the model. The
residuals could be said to follow a straight line in the
normal plot of residuals implying that errors were distributed normally and were randomly scattered within constant
variance across the residuals versus the predicted plot
(Fig. 3a, b).
In order to verify the adequacy of the developed model,
another trial was conducted. Selecting cutting speed of
130 m/min and feed of 0.125 mm/rev, the resulting tool life

was 5.68 min. The predicted tool life for this cutting
parameter combination was 6.23 min, and with confidence
interval of 95%, the tool life should range between 5.39 and

Fig. 3 Diagnostic plots of the tool life model: a normal plot of residuals and b residuals vs. predicted


Int J Adv Manuf Technol (2012) 60:853–863

857

Fig. 4 Response surface graphs of a contours and b 3D surface for tool life

7.09 min. This test verified that the model is sufficient to
represent the tool life data for this particular hard turning
operation.
The obtained final equation is represented by graphs of
contours and 3D surface (Fig. 4a, b) for ease of analyzing
the influence of cutting speed and feed to the life of the
coated carbide tool. Both graphs show that longer tool life
is obtained when lower cutting speed and feed are selected.
Most of the combinations would result in tool life of more
than 3 min and could even go beyond 20 min at low range
of cutting speed and feed.

Since the tool life is also affected by the interaction
between cutting speed and feed, additional information
could also be obtained by analyzing the interaction graph
(Fig. 5). At low cutting speed, the difference in tool life
between high and low feed is high. Yet, when the cutting

speed increases, this feed difference does not affect much
on the resulting tool life.
After knowing that both cutting speed and feed are
inversely proportional to tool life, statistical analysis
was also used to determine which factor is more
influential. For this purpose, the graph of perturbation

Fig. 5 Relation between tool life and cutting speed with the presence
of interaction between cutting speed and feed

Fig. 6 Perturbation plot for tool life where x1 and x2 are cutting speed
and feed in coded factors, respectively


858
Table 3 Cutting forces at
various cutting
parameters

Int J Adv Manuf Technol (2012) 60:853–863

Feed (mm/rev)

Fr (N)

Fc (N)

Ff (N)

100

100

0.1
0.16

103
127

80
124

39
40

Depth of cut=0.4 mm

170

0.1

96

73

35

Fr radial force, Fc cutting force,
Ff feed force

170

0.16

114

118

33

Cutting speed (m/min)

(Fig. 6) was utilized. Using the cutting speed of 135 m/min
and feed of 0.125 mm/rev as the reference point (0.0),
reducing either cutting speed or feed (moving more
negative from the reference point) results in longer tool
life and vice versa. However, negative movement of the
cutting speed factor from the reference point associates
with even longer tool life than the same movement of
the feed factor and positive movement of the cutting
speed factor associates with even shorter tool life than
the same movement of the feed factor. This suggests
that cutting speed has more influence to the resulting
tool life than feed does.

Fig. 7 Cross-section images of the generated chips

These statistical analysis results support the remark that
longer tool life is obtained when lower cutting parameters
are selected. Thus, in order to ensure that the cutting tool
lasts long enough, it is suggested that performing hard

turning of stainless steel using coated carbide tool should
apply low cutting parameters.
Based on the information provided by the tool’s
manufacturer, coating the carbide substrate with TiAlN
was actually intended to enable the tool to be operated at
higher cutting speed. However, the intention was obstructed
by the hardness of the workpiece since both the cutting
speed and the workpiece’s hardness relate proportionally to


Int J Adv Manuf Technol (2012) 60:853–863

859

Fig. 8 Surface roughness
at various cutting speed and
feed

cutting temperature [11] while carbide’s hot hardness is
relatively low. The coated carbide tool withstood the
workpiece’s high hardness but, consequently, was unable
to perform effectively at high cutting speeds.
The force measures at some of the cutting parameter
combinations presented in Table 3 support this remark to
some extent. The measured cutting force (Fc) increased as
the cutting speed was reduced from 170 to 100 m/min. The
higher temperature involved at high cutting speed influenced the workpiece’s thermal properties, and thus, the
recorded cutting force was lower [12]. Nevertheless, at this
cutting speed reduction from high to low at the same feed
of 0.1 mm/rev, there was chip morphology transition from

segmented to continuous type (Fig. 7). The segmented chip
type was said to drive regenerative chatter [12], which
might deteriorate the tool’s lifetime. Feed was also reported
to be proportional to cutting temperature [13] and is
responsible to the more rapid deterioration of tool life at
high feeds. This was in agreement with the measured
cutting force that increased as the feed was raised from 0.1
Table 4 Analysis of variance
(partial sum of squares)
for linear model of the surface
roughness as the response

Source

Sum of
square

Model
x1
x2
Residual
Lack of fit
Pure error
Cor total

0.190
2
0.072
1
0.118

1
0.085
8
0.030
6
0.055
2
0.275
10
R2
Adequate precision

to 0.16 mm/rev. However, it should be noted that hard
turning involves the interaction between thermal and
mechanical loadings. When transition from low to high
cutting speed was involved, indicated by chip type
transition from continuous to segmented, these types of
loading could not simply be differentiated without considering the high temperature generated. Therefore, it was
inappropriate to compare directly between the cutting
forces at low cutting speed–low feed combination with that
of the high cutting speed–high feed combination.
3.2 Surface finish
The generated surface roughness is less than 1.6 μm, as
expected (Fig. 8). This means that the coated carbide tool
could generate surface finish at tight tolerance range of
finish machining as they are intended to. Some of the
results were even one level better by being less than 0.8 μm
in surface roughness. The decrease in feed improves the Ra.
Cutting speed was found to be proportional to the achieved

Degrees of
freedom

Mean
square

F value

Prob>F

0.095
0.072
0.118
0.011
0.005
0.028

8.941
6.801
11.130

0.0091
0.0312
0.0103

Significant

0.178

0.9577

Not significant

0.691
9.242


860

Int J Adv Manuf Technol (2012) 60:853–863

Fig. 9 Diagnostic plots of the surface roughness model: a normal plot of residuals and b residuals vs. predicted

Ra. Using similar steps to develop and analyze the tool life
model, the model of surface roughness data from the hard
turning tests was developed and validated.
Preliminary diagnosis to determine the appropriate power
transformation was conducted. Having the maximum to
minimum ratio of less than 3, applying any power transformation would have little effect, and thus, no power transformation was set (λ=1). The next step was to determine the
suitable regression model. The probabilistic value, Prob>F,
and the lack of fit of each model were calculated and the
linear model was selected for having the least probabilistic
value and the most insignificant lack of fit.
The analysis of variance was then performed to test the
significance of the selected regression model and its
coefficients (Table 4). As before, the maximum probabilis-

tic values of 5% were set for the model and its coefficients
to be considered significant. The linear model and its
coefficients, having probabilistic value of less than 0.05,

were considered significant.
The obtained final equation of the model in terms of
coded factors is
Ra ¼ 0:66 À 0:11x1 þ 0:14x2

ð4Þ

where x1 and x2 are cutting speed and feed in coded factors,
respectively. In actual values, the Ra resulted by the coated
carbide tool is
Ra ¼ 0:479 À 0:003v þ 4:651f

ð5Þ

where v is the cutting speed (meters per minute) and f is the
feed (millimeters per revolution).

Fig. 10 Response surface graphs of a contours and b 3D surface for surface roughness


Int J Adv Manuf Technol (2012) 60:853–863

861

The statistical validity of the model was evaluated by
inspecting the normal plot of residuals and residuals
versus predicted plots. The residuals could be said to
follow a straight line in the normal plot of residuals and
were randomly scattered within constant variance across
the residuals versus the predicted plot (Fig. 9a, b). The

model’s quiet high coefficient of determination, R2, of
0.678 and its adequate precision of 9 suggested that the
model is adequately representative.

When tested at cutting speed of 130 m/min and feed of
0.125 mm/rev, the resulting Ra was 0.67 μm, which is very
close to the predicted Ra value at this cutting parameter
combination of 0.65 μm and within the range of 95%
confidence interval of 0.58–0.73 μm. This verifies the
adequacy represented by the model.
The final surface roughness model equation is represented by graphs of contours and 3D surface in a and b of
Fig. 10, respectively. Both graphs show that lower surface
roughness was obtained by selecting lower feed and higher
cutting speed.
The fact that cutting speed and feed have opposite
effect to the surface roughness was confirmed by the
graph of perturbation (Fig. 11). Using the cutting speed
of 135 m/min and feed of 0.125 mm/rev as the reference
point (0.0), reducing the cutting speed (moving more
negative from the reference point) associates with rougher
surface and vice versa. On the contrary, reducing the feed
associates with finer surface and vice versa.
The ability of the coated carbide tool to achieve finish
machining criterion for surface roughness is partially
contributed by the sharp edge radius provided by the PVD
coating technique and positive rake angle of the tool.
Commonly, hard turning is set at negative rake angle to
prevent the chipping of the cutting edge [14]. The coated
carbide tool, however, was able to withstand the subjected
loading from causing severe chipping even when positive

rake angle was set. The finer surface finish obtained at high
cutting speed is in agreement with those reported by Chen
[15] although no satisfactory explanation on this phenomenon was available.

Fig. 12 Overlay plot of the input factors for the predetermined
response criteria of minimum 10-min tool life and of maximum 0.8μm surface roughness

Fig. 13 Desirability plot of the input factors to obtain the maximum
tool life and minimum surface roughness

Fig. 11 Perturbation plot for surface roughness where x1 and x2 are
cutting speed and feed in coded factors, respectively


862

Int J Adv Manuf Technol (2012) 60:853–863

Fig. 14 Microstructures of the a
machined surface and b chip
generated at the combination of
100 m/min cutting speed and
0.1 mm/rev feed

3.3 Optimum cutting parameters
In general, the coated carbide tool is usable to perform mild
hard turning of stainless steel (47–48 HRC). The developed
empirical models reflect that the finest surface finish would
be obtained by operating at low feed and high cutting speed
combination. Yet, the coated carbide tool could not perform

well at high cutting speed. It is also suggested that the
selected cutting parameters should be as low as possible to
keep the coated carbide tool to operate long enough.
This difference in machining results is unfavorable in
selecting the cutting parameters to meet the ideal requirements. Practically, hard turning as a finishing operation
should be able to generate a fine surface finish to meet the
costumer’s demand of geometrical accuracy of the machined component. On the other hand, machine shops will
be more efficient when the cutting tool lasts longer.
Therefore, a compromised solution seems appropriate to
select the cutting parameters.
Being able to quantify the effect of each input factor to
the machining responses, the models offer an optimizing
option to select the range of cutting speed and feed that
would result in predetermined criteria of the tool life and
the surface roughness values. It is suggested that a coated
carbide tool should last at least 10 min to be convenient to
the machine shops. The coated carbide tool should also be
able to generate surface roughness of 0.8 μm at maximum
to meet the costumer’s strict tolerance. These criteria would
be met when the cutting speed and feed combination are
within the gray area of the overlay plot (Fig. 12). The
solution is the intersection between the solutions for the
tool life criteria (area on the left of the tool life contour of
10 min) and the solutions for surface roughness criteria
(area below the Ra contour of 0.8 μm).
The solution could even be more detailed by setting
sharp response criteria. By setting that the tool life should
be maximum and that the surface roughness should be
minimum, the desirability value in Fig. 13 is higher towards
lower cutting speed and lower feed and being maximum at

the cutting speed of 100 m/min and the feed of 0.1 mm/rev,
i.e., the low cutting speed–low feed combination.

It is within interest that the low cutting speed–low
feed combination is the most effective cutting parameter
in terms of tool life and surface finish. The generated
chip has a continuous type and the machined surface
has minimum microstructure alteration with no white
layer (Fig. 14). The chip of continuous type is preferred
for indicating the absence of deteriorating regenerative
chatter. The chip microstructure where its grains are
visible, albeit elongated due to compression, also indicated that mechanical action was more dominant than the
thermal action. On the machined surface microstructure,
the outermost grains are visible, contrary to the featureless
layer shown when white layer is present as often the case
in machining of steels [16, 17]. White layer was found
disadvantageous to fatigue life, stress corrosion, and wear
resistance [17, 18]. When present, white layer requires
secondary finishing process for its removal. These features
suggest that at this particular cutting parameters combination, machined surface deterioration related to phase
transformation due to excessively high temperature is
minimal. Thus, this remarks positively to the surface
quality of the resulting component.

4 Conclusions
The coated carbide tool fairly performs when performing
mild hard turning of stainless steel (47–48 HRC) in dry
conditions at the cutting parameters selected. The tool life is
proportional to both cutting speed and feed. Cutting speed
is the more influencing factor to tool life. The measured

cutting force reduces with increasing cutting speed and
rises with increasing feed. The obtained surface roughness
is proportional to feed and inversely proportional to cutting
speed. Considering the tool life and the surface finish, low
cutting parameters are the optimum solutions to make the
tool last long enough and generate a fine surface finish with
good surface quality.
Acknowledgment Financial support from the Ministry of Science,
Technology and Innovation, Malaysia is acknowledged.


Int J Adv Manuf Technol (2012) 60:853–863

References
1. Tönshoff HK, Arendt C, Ben Armor R (2000) Cutting of
hardened steel. Ann CIRP 49:547–566
2. Poulachon G, Bandyopadhyay BP, Jawahir IS, Pheulpin S, Seguin
E (2003) The influence of the microstructure of hardened tool
steel workpiece on the wear of PCBN cutting tools. Int J Machine
Tools Manufacture 43:139–144
3. Mamalis AG, Kundrak J, Gyani K (2002) On the dry machining
of steel surfaces using superhard tools. Int J Adv Manufacturing
Tech 19:157–162
4. Pavel R, Marinescu I, Deis M, Pillar J (2005) Effect of tool wear
on surface finish for a case of continuous and interrupted hard
turning. J Mater Processing Tech 170:341–349
5. Hallberg L (2010) Recycling on the rise. Cutting Tool Eng 62
(10):60–64
6. Noordin MY, Venkatesh VC, Sharif S (2007) Dry turning of
tempered martensitic stainless tool steel using coated cermet and

coated carbide tools. J Mater Processing Tech 185:83–90
7. Jindal PC, Santhanam AT, Schleinkofer U, Shuster AF (1999)
Performance of PVD TiN, TiCN, and TiAlN coated cemented
carbide tools in turning. Int J Refrac Metals Hard Mater 17:163–170
8. Gillibrand D, Sarwar M, Pierce CT (1996) The economic benefit
of finish turning with coated carbide. Surface Coatings Tech 86–
87:809–813
9. Hwang YK, Lee CM (2010) Surface roughness and cutting force
prediction in MQL and wet turning process of AISI 1045 using
design of experiments. J Mech Sci Tech 24:1669–1677

863
10. Noordin MY, Venkatesh VC, Sharif S, Elting S, Abdullah A
(2004) Application of response surface methodology in describing
the performance of coated carbide tools when turning AISI 1045
steel. J Mater Processing Tech 145:46–58
11. Ueda T, Al Huda M, Yamada K, Nakayama K (1999) Temperature
measurement of CBN tool in turning of high hardness steel. Ann
CIRP 48:63–66
12. Trent EM, Wright PK (2000) Metal cutting. ButterworthHeinemann, Boston
13. El Wardany TI, Mohammed E, Elbestawi MA (1996) Cutting
temperature of ceramic tools in high speed machining of difficultto-cut materials. Int J Machine Tools Manufacture 36:611–
634
14. Nakayama K, Arai M, Kanda T (1988) Machining characteristics
of hard materials. Ann CIRP 37:89–92
15. Chen W (2000) Cutting forces and surface finish when machining
medium hardness steel using CBN tools. Int J Machine Tools
Manufacture 40:455–466
16. Han S, Melkote SN, Haluska MS, Watkins TR (2008) White layer
formation due to phase transformation in orthogonal machining of

AISI 1045 annealed steel. Mater Sci Eng, A 488:195–204
17. Umbrello D, Jawahir IS (2009) Numerical modeling of the
influence of process parameters and workpiece hardness on white
layer formation in AISI 52100 steel. Int J Adv Manuf Technol
44:955–968
18. Aramcharoen A, Mativenga PT (2008) White layer formation and
hardening effects in hard turning of H13 tool steel with CrTiAlN
and CrTiAlN/MoST-coated carbide tools. Int J Adv Manuf
Technol 36:650–657


Int J Adv Manuf Technol (2012) 60:865–882
DOI 10.1007/s00170-011-3677-8

ORIGINAL ARTICLE

Prediction of depth of cut for single-pass laser micro-milling
process using semi-analytical, ANN and GP approaches
Chinmay K. Desai & Abdulhafiz Shaikh

Received: 13 February 2011 / Accepted: 26 September 2011 / Published online: 22 October 2011
# Springer-Verlag London Limited 2011

Abstract The present study is aimed to investigate micromilling performance of thermoplastics with different
parameters, namely laser beam absorptivity, latent heat of
vaporization, laser power and cutting speed. The 25-W CO2
(CW) laser engraving machine is used for the investigation.
In total 50 different combinations of laser power and
cutting speed with four categories of thermoplastics,
namely poly-methyl-methacrylate, poly-propylene, acrylonitrile butadiene styrene and nylon 6, are used in this study.

Experimental results suggest that laser beam absorptivity,
cutting power and cutting speed are the major influencing
parameters on depth of cut. Theoretical model for the
prediction of depth of cut in terms of material properties,
cutting power and cutting speed has been proposed. Two
correction parameters have been introduced in this analysis
using non-linear regression method to improve the theoretical model. Comparison has been made between prediction
capabilities of theoretical model, model based on multigene genetic programming and artificial neural network.
The comparison clearly indicates that all the three models
provide accurate prediction of depth of cut. The details of
experimentation, model development, testing and the
performance comparison are presented in this paper.

C. K. Desai (*)
Department of Mechanical Engineering,
C.G. Patel Institute of Technology,
Bardoli, India
e-mail:
A. Shaikh
Department of Mechanical Engineering,
S.V. National Institute of Technology,
Surat, India
e-mail:

Keywords Micro-milling . Non-linear regression . Multigene genetic programming . Artificial neural network

1 Introduction
Laser beam cutting is gaining popularity nowadays due to
superior cut quality, reliability and suitability to cut all
categories of material including plastics [1]. General

research work in laser cutting whether experimental or
theoretical assumes a given depth of cut usually identical to
the work piece thickness and attempts to determine a
cutting speed which optimizes particular criteria like cutting
rate and surface quality [2, 3]. However, in ‘blind’ cutting
(grooving, engraving, milling and turning), the depth is not
a set parameter but a critical variable which depends upon
material properties, cutting power and speed [4]. This
creates necessity of an accurate model to predict the depth
of cut for a specific combination of laser cutting process
parameters.
In the past, the laser beam cutting mechanism and its
affecting factors were studied by several researchers on
different materials [5–7]. These investigations revealed the
significance of laser power, cutting speed, cutting material’s
composition as well as its thickness and laser beam mode on
laser cutting performance. The study carried on CO2 laser
cutting of mild steel focused on the effects of high-pressure
assistant gas flow on cutting quality [8]. The study on the
laser cutting of ceramic plates emphasizes the importance of
cutting power, the feed-rate of the specimens and the
material properties, namely specific heat, conductivity and
thermal expansion coefficient in controlling the formation of
cracks that result from the thermal stresses induced [9].
It had been suggested by Zhou and Mahdavian [10] that
low-power 60-W CO2 laser could be useful for cutting non-


866

Int J Adv Manuf Technol (2012) 60:865–882

metallic materials and plastic board. Many researchers had
studied the usefulness of different types of mathematical
models to predict laser beam ‘kerf’ profile. These include
analytical models [11–14], numerical models [15, 16],
models based on statistical design of experiments [2, 17,
18] for analysing mechanism behind ‘kerf’ formation and
effects of process parameters on the cut quality.
Recently, some researchers [19–23] have carried out
investigation on micro-machining (micro-grooving, micromilling, micro-turning) of ceramics/difficult to cut material
for studying the effect of process parameters on cutting rate
and surface roughness. Their studies involve multiple pass
operations to achieve desired depth of cut. The parametric
analysis carried out by Dhupal et al. [19] using pulse Nd:
YAG laser for micro-grooving operation on aluminium
titanate ceramics focused on study of lamp current, pulse
frequency, pulse width, assist air pressure and cutting speed
of laser beam on groove profile. The mathematical model
based on response surface methodology for correlating the
deviation on half taper angle of groove and depth of microgroove was proposed.
Kibria et al. [20] have carried out experimental investigation on Nd:YAG laser micro-turning of aluminium
ceramic. The objective of their study was to assess the
effect of laser turning parameters such as lamp current,
pulse frequency and laser beam scanning speed on depth of
cut and surface roughness. Their study revealed that laser
lamp current and scanning speed have great effect on both
depth of cut as well as surface roughness.
The work reported by Orazi et al. [21] was focused on
laser surface micro-manufacturing using 20-W IPG pulse

ytterbium fibre laser. They proposed automatic process
planning system based on multiple regression method to
predict material removal rate for multi-pass square pocket
milling on AA7075 aluminium alloy.
The investigation carried out by Saklakoglu and Kasman
[22] was focused on micro-milling performance of the AISI
H13 using 30-W fibre laser. Their study involved the effect
of laser power, scan speed, frequency and fill spacing on
surface roughness and milling depth during multi-pass
operation. The second-order regression model was proposed to predict milling depth and surface roughness.
Snakenborg et al. [23] had reported the use of commercial
CO2 laser for fabrication of micro-fluidic systems in polymers.

We have noticed from our local industry practices and
literature surveys that low-power lasers are widely used for
engraving gift articles made from plastics, cutting thin plastic
sheets, marking and cutting decorative cards and milling of
plastic moulds for casting polymer resins and rubbers.
The common technique for obtaining the proper depth
and its associated laser parameters during engraving/milling
is an iterative-based process in which the laser process
parameters are changed individually and the resulting
engraving/milling is examined to determine the accurate
depth. This is a trial-and-error approach which can take
several days or weeks. At the same time, there is a need for
analytical/empirical tools, choosing the optimum working
conditions on a scientifically sound basis.
For that reason, the aim of the present research is to
device simple yet comprehensive techniques to predict
depth of cut for single-pass micro-milling operation for

plastic materials. Three widely used modelling techniques,
namely semi-analytical method, genetic programming (GP)
and artificial neural network (ANN), are compared to
predict depth of cut for single-pass circular pocketing
operation during micro-milling of thermoplastics.

2 Experimentation and measurements
2.1 Work piece materials
Four engineering plastics, namely poly-methyl-methacrylate
(PMMA), poly-propylene (PP), acrylonitrile butadiene styrene (ABS) and nylon 6, were chosen in this research based on
specific manufacturing requirements in terms of product and
process in local industry. Thermal properties were evaluated
using thermal gravimetric analysis. Absorptivity of laser beam
for all polymeric materials was evaluated using methodology
explained in [24]. Test specimens for all polymeric materials
were moulded from virgin plastic granules with dimensions
10×5×5 cm3. The properties of the polymers are given in
Table 1.
2.2 Laser machine
For all micro-milling experiments, the laser beam is
supplied by a 25-W continuous laser (Synrad, Inc., mod-

Table 1 Measured thermal and optical properties for thermoplastics
Sr. no.

Name of the material

1
2
3

4

Poly-methyl-methacrylate (acrylic)
Poly-propylene
Nylon 6
Acrylonitrile butadiene styrene

Glass transition temperature (°C)

Specific energy (J/m3)

Absorptivity (a)

256
163
220
368

24.20
73.9
139.44
53.24

0.9990
0.9309
0.9203
0.9028