ARTICLE IN PRESS

International Journal of Machine Tools & Manufacture 47 (2007) 236246

www.elsevier.com/locate/ijmactool

Experimental analysis of dimensional error vs. cycle time

in high-speed milling of aluminium alloy
M. Albert a, J. Ciuranaa,, C.A. Rodriguezb
a

Department of Mechanical Engineering and Industrial Construction, University of Girona, Av. Llus Santalo s/n 17071 Girona, Spain
b
Centro de Innovacion en Diseno y Tecnologa, Instituto Tecnologico y de Estudios Superiores de Monterrey, Av.
Eugenio Garza Sada #2501, Monterrey, NL 64849, Mexico
Received 12 October 2005; received in revised form 4 April 2006; accepted 25 April 2006
Available online 9 June 2006

Abstract
Manufacturers of machined aluminium parts are among the principal users of high-speed milling. The excellent machinability of
aluminium allows this technology to be used with severe cutting conditions, and at the same time permits the machining of complex
shapes. There are many factors inuencing the quality of manufactured aluminium parts and the economics of the manufacturing
process, but little corresponding data is available, making process planning enormously difcult.
The aim of the work presented in this article is to experimentally analyse the inuence of some of these factors, specically the feed
rate, the type of interpolation and the toolholder, on the dimensional accuracy of the product and the cycle time. Design of experiments
(DoE) is used to determine which experiments have to be conducted to obtain a mathematical model that relates the mentioned factors
with the responses.
The results show that the toolholder has considerable inuence over dimensional accuracy and that the type of interpolation
appreciably affects the cycle time. Details of the rst-order interactions between factors have been included.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: High speed machining; Dimensional error; Cycle time; Tool path

1. Introduction
The machining of aluminium alloys is one of the largest
elds of high-speed machining applications. The sectors
most commonly employing this technology are the aeronautic sector and the moulds and dies industry, especially
in the manufacturing of blow moulds which, being more
and more demanding and competitive, require greater
dimensional accuracy and surface nish and, at the same
time, a reduction in costs and in manufacturing time [1].
Such a requirement is normal if we consider how the
effectiveness of mould manufacturing affects the entire
development cycle of new products and the technology
used to manufacture the moulds is therefore an essential
link in the production line [2].

Corresponding author. Tel.: +34 972 419822; fax: +34 972 418098.

E-mail address: quim.ciurana@udg.es (J. Ciurana).

0890-6955/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijmachtools.2006.04.009

Numerous and varied factors of high-speed milling

inuence the quality of the nal part and its manufacturing
economy [3]. Among these factors are the part and tool
materials, the shape of the tool and the toolholder, the
cutting conditions, the behaviour of the machine tool and
the control performance, the type of interpolation used in
generating CNC programmes and the use of refrigerants.
Individually analysing the effect of each of these factors
on the nal result has generated much interest. Various
research projects have recently been conducted in which
one of the previously mentioned factors has been
correlated with mould surface roughness: Vivancos et al.
[3] and Lopez de Lacalle et al. [4] for steel machining, and
Coelho et al. [5] for aluminium alloys machining. However,
only a few research projects have analysed the relationship
between these factors and dimensional accuracy [6], and
the inuence of these same factors on cycle time has been
analysed in very few cases [7], almost all of which have been
focussed on steel machining. Another key objective of

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

recent research has been the optimisation of cutting

conditions in high-speed machining, and in this eld a
great variety of work can be found: Kaldos et al. [8], based
on aluminium alloys machining, and Chien and Tsai [9],
Juan et al. [10] and Lopez de Lacalle et al. [11], based on
steel machining.
The purpose of this article is to analyse the inuence of a
set of factors (including some of those mentioned
previously) on two of the parameters, which although
considered important by mould manufacturers [2], have
been researched far less: dimensional error and cycle time.

2. Manufacturing conditions
In order to minimise the cutting time of the manufacturing processes and obtain dimensional accuracy in accordance with product specications, the most suitable
manufacturing conditions for each operation must be
carefully selected [12].
While, generally, only the cutting conditions (feed rate,
depth of cut, cutting speed) are taken into account, each
element involved in the machining process has some
inuence on the nal result of that process.
The machine, and in our case the machining centre,
is the basic element of the process. There are many
types of machining centres with a variety of characteristics:
power, workspace dimensions, spindle speed, travel speed
on axes, number of axes and tool magazine capacity,
among others.
However, other aspects related to machining centre
behaviour are not as easily quantiable and yet are
also very important for the nal result: the dynamic
behaviour of the machine and the thermal expansion
of the spindle. The latter, unavoidable due to the rising
temperature of the functioning machine, can lead to
dimensional errors in the part. To control it, some machine
tools have temperature-controlled spindles, but warm-up is
recommended in all cases to assure that expansion occurs
before the manufacturing process begins, not during it, so
that the entire production takes place under the same
conditions.
The cutting tool is another key machining element. The
term high-speed milling is generally used to describe end
milling with a small diameter tool, less than or equal to
10 mm, at a high rotation speed, superior to 10,000 rpm
[13]. Under these conditions, and given the high propensity
for vibration, tool rigidity is a critical issue. This problem
can be solved or minimised using a suitable toolholder. It is
also important to control the tool wear and to replace the
tool when the parameter to measure that wear (VB, for
example) exceeds a specic value.
Numerical control is another important element in
machining. Control features provide options for accurate
machining and functions like look-ahead or feed-forward,
considered very important in high-speed machining [14].
With regard to the machine tool and control, the type of

237

position sensor must also be taken into account, given its

signicant inuence on the quality of the nal part.
Another element closely related to control is the NC
programme. The type of interpolation used for tool path
generation introduces variety, and although linear interpolation is used in most situations, when milling sculptured
surfaces, the possibility of using circular or polynomial
interpolations should be considered to reduce the size of
the programmes and improve the dimensional accuracy.
Not all controls are capable of processing polynomials,
however, so in many cases the dilemma is limited to linear
and circular interpolation. Between the two options, the
solution most often adopted is the rst.
Crucial to the creation of the NC programme is the
CAM system. Currently, the majority of commercial CAM
systems have high-speed machining options, which provide
more suitable machining strategies for this technology.
However, some research has also allowed the cutting
parameters to be calculated [15], as the majority of CAM
systems are based on programming the geometry and not
on selecting the cutting parameter, which requires a larger
amount of information about materials and tools.
This list of aspects capable of varying the results is far
from exhaustive. A much longer list could be drawn up,
taking into account all the factors involved in manufacturing, from the expertise of the machinist to the quality of the
CAD model or the maintenance of the machine. Although
it would be a very interesting starting point, drawing up
such an exhaustive, comprehensive list is not one of the
aims of this research.
It is obviously difcult to analyse all the factors
mentioned, and for this reason the present research is
limited to those factors that, according to various authors
[16,17], are most representative and have the greatest effect
on cycle time and dimensional accuracy. The factors
considered in the analysis are:





The type of toolholder, since good tool balance is critical

to avoiding premature failure of the tool and to
obtaining a good surface nish.
The cutting conditions and, specically, the programmed feed rate, which is the parameter that, in this
case, ensures that the process is high-speed milling.
The type of interpolation used to create the programmes, distinguishing specically between linear and
circular interpolation.

In addition to these variable factors, there are some

other factors that can be considered as xed factors by
using the same conguration in all the experiments, for
instance, the tool used, the execution mode, the controller
type, the thermal expansion of the spindle (controlled
through warm-up) and tool wear (for which a maximum
value is established to indicate when it should be replaced).
The effect of tool vibration can also be minimised by using
a not very long tool with a large enough diameter to assure
rigidity.

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

238

The strategy used for the analysis consists of machining

the same shape, varying the manufacturing conditions and
repeating the tests in different machines.

The minimum value of the feed rate corresponds to 70%

of the maximum value. This value has been selected
because the difference between both levels was large
enough to perceive signicant variations in the responses,

3. Design of the experiment/experimental details

Design of experiments (DoE) is an efcient experiment
planning process that allows the data obtained to be
analysed, valid conclusions to be drawn and objectives to
be set ([18], w3). DoE has been used to determine the
appropriate number of tests and the experimental conditions necessary to obtain the desired goal of analysing
which factors of the process inuence the response
variables.
Factorial design has been used in this research to
optimise resource use and study the effects of the main
factors. A 23 total factorial design, with each of the
variables at two levels, has been chosen. The most common
design consists of running a test with all the possible
combinations of variables at each of two levels, thereby
obtaining most of the information required for a multilevel
experiment, if necessary [19]. In that way, the factorial
design does an excellent job relating the experimental effort
to the information obtained [20]. Table 1 presents the
variable factors considered in the experiment and the
selected levels. Two of the factors are qualitative, so one
level has been arbitrarily assigned a value of 1 and the
other a value of +1.
The specic values for the programmed feed rate are
relatively low with respect to those normally used in
aluminium mould machining because the shape used is very
complex and requires continual accelerations and decelerations of the headstock, making it difcult to reach the
programmed feed rate. Carrying out the experiments at a
very high feed rate would mean never arriving at the
programmed value and, as a result, the data obtained
would not be useful from the analytical point of view
presented in this research.
The procedure followed to determine the maximum
value of the feed rate consists of test running the
programme, starting at a very high programmed feed rate
and decreasing it little by little. This feed rate progression is
monitored in the control for each one of the runs, and it
continues decreasing until the actual feed rate corresponds
to the programmed rate in approximately 30% of the cycle.
A programmed feed rate of 5 m/min meets this condition.

Fig. 1. Prole used in the experiment.

Table 2
Design matrix
Test

F (m/min)

1
2
3
4
5
6
7
8

Mechanical
Mechanical
Mechanical
Mechanical
Thermal
Thermal
Thermal
Thermal

Linear
Linear
Circular
Circular
Linear
Linear
Circular
Circular

3.5
5
3.5
5
3.5
5
3.5
5

Table 3
Technical features of the machining centres and controls used in the
experiment

Table
Travels

Spindle

Feed rate

Main drive
Positioning

Table 1
Selected factors and levels
Factors

Acronym

T
F
I

Mechanical
3.5
Linear

Maximum (1)
Thermal
5
Circular

Size
Block
processing
time

MC2

1.120  600
1.000
600
510
HSK-63-A
24.000

+500
500
420
380
HSK-63-A
18.000

17.5

25.0

No
030.000

Yes
020.000

30.000

50.000

Unknown

1.0

AC
Encoder

AC
Encoder and
scales
0.008
(according to
VDI/DGQ
3441)
0.001
0.080
3.6

Accuracy (mm)

0.010
(according to
ISO 230-2)

Resolution (mm)
Work space (m3)
Time for 3D linear
(ms)

0.001
0.306
Unknown

Levels
Minimum ( 1)

Toolholder
Feed rate (m/min)
Interpolation

Work surface (mm)

X (mm)
Y (mm)
Z (mm)
Cone
Range of speeds
(rpm)
Headstock power
(kW)
Temperature control
Working feed rate
(mm/min)
Rapid feed rate
(mm/min)
Maximum
acceleration (g)
Spindle-motor
Positioning feedback

MC1

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

while maintaining a feed rate close to the one generally

used in industrial applications.
Several researchers have analysed which geometry types
are most suitable for the testing of machine tools and have
designed test parts that contain such designs. Lopez de
Lacalle et al. [21] present some of those tests. One of the
most often used is the Mercedes test, but there are many
more. In most situations, large parts have their surfaces
sculptured in three dimensions and a large investment of
resources is required to perform the experiments.
For the present analysis, geometries meeting the following requirements have been sought:

239

initial size of the part (20 mm) has been enlarged by 90% so
that it can be machined with an 8 mm diameter tool. In this
way, the tool is sufciently rigid to make the effect of
vibration on the result negligible.
The experiment consisted of a total of 8 tests, presented
in Table 2. It was repeated in two separate high-velocity
machining centres, which we will call MC1 and MC2.
These centres have different features, and the advantages
supposed by these features can be seen by comparing the
results of one with the other. Table 3 shows the technical
features of each of the machining centres.
4. Experimental procedure

(a) containing sculptured surfaces,

(b) machinable with an 8 mm or larger diameter tool, to
minimise the vibration effects,
(c) containing critical points to study the dynamic
behaviour of the machine centres: sharp edges, smallradius concave and convex curves, and inection
points,
(d) being a test part recognised by researchers or manufacturers of machine tools,
(e) being machinable with 3-axis machine centres.
The simple test part considered most suitable for this
study meets most of the requirements. It is the prole
shown in Fig. 1, proposed by MECOF, a manufacturer of
machine tools. Formed by curves of both large and small
radiuses in relation to the diameter of the tool, it also has
both concave and convex zones. To achieve full compliance
with all of the specications, some changes have been made
to the original prole: a sharp edge has been added and the

Fig. 2. Tool used in the experiment.

Table 4
Dimensions of the tool used in the experiment
80
16
35
8
8
7.4

l1
l2
l3
d1 f8
d2 h6
d3

The tool used was especially designed for machining

aluminium with a silicon content of less than 12%. It is a
long hard metal milling cutter, with a TiAlN multilayered
coating and 12% Co/CCm-2 micrograin material, with a
spherical head, a helix angle of 301 and two cutting edges.
The tool and its dimensions are shown in Fig. 2 and in
Table 4, respectively.
The machined material was AA 5083, commonly used in
prototyping because its low silicon content allows the tool
to have a reasonably long life. The chemical composition of
this material is presented in Table 5.
PowerShape and PowerMill software was used to
generate the CAD and CAM models, respectively. The
machining parameters used in the CAM are presented in
Table 6.
A roughing operation is carried out on the part until
leftover material is left at a constant depth of 0.3 mm,
which will be removed according to the conditions of the
experiment.
The experimental protocol used consists rst of a warmup stage, consistent with a progressive increase of the
revolution velocity of the spindle, from 2500 to 18,000 rpm,
increasing the velocity by 2500 rpm every 400 s. Once this
has been done, the taper is cleaned and the tool to be used
is mounted. Then the runout is checked in order to ensure
that it is lower than 5 mm, so that it can be considered
irrelevant. Finally the previously randomised tests are
carried out. Refrigerant is used during all the tests to avoid
the built-up edge that usually results from machining
aluminium. All the experiments are done at 18,000 rpm
spindle speed.
The machining is carried out in one direction, beginning
with 151 tool cuts. The eight tests corresponding to each
machine are done on the same block of material, separated
by grooves in which the same prole has been machined
but at a lower Z dimension. Fig. 3(a and b) show the blocks

Table 5
Chemical composition of AA5083 (%)
Si

Fe

Cu

Mn

Mg

Zn

Ti

Pb

Cr

Others

Al

0.40

0.40

0.10

0.401.00

4..004.9

0.25

+Zr 0.20

0.050.25

0.15

Balance

ARTICLE IN PRESS
240

M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

of aluminium in which the tests were made corresponding

to the two machining centres, MC1 and MC2.
Of the two response variables considered in the experiment, cycle time and dimensional accuracy, the rst can be
taken directly from the information provided by the
control system during the machining. The second, however,
requires that each of the machined proles be measured
with a 3-coordinate measuring machine. 50 points have
been measured for each experiment, giving a total of 800
points selected by identifying the critical zones (sharp edges
and areas with especially important curves) and uniformly
distributing the points in these zones. Apart from these
points in critical zones, some points have also been selected
in the remaining zones to provide measurement continuity.
Fig. 4 shows the distribution of the measured points. The
measurements have been made at each point in a normal
direction to the surface. Once these data are obtained, they
are compared with theoretical values extracted from the
CAD model and tolerance rankings are established for
Table 6
Machining parameters programmed in the CAM
Rth
ae
ap

N
Vfr
Vfp
fz

Crest (mm)
Radial depth (of cut) (mm)
Axial depth (mm)
Tolerance (mm)
Arc t
Spindle speed (rpm)
Speed of rapid feed rate (m/min)
Speed of working feed rate (m/min)
Working feed rate (mm)
Flood coolant

0.1
0.17
0.3
0.001
1
18,000
24
3.5/5
0.10/0.14

each point. Fig. 5 shows the measuring of the blocks in the

3-coordinate machine.
5. Results and discussion
Table 7 presents the results obtained in the MC1 and
MC2 experiments. These data have been analysed using the
STATGRAPHICSs software version 5.0. The following
sections present the results obtained for each one of the
response variables: cycle time and dimensional error.
The cycle time is obtained from measurements made
with a chronometer and has been corroborated with the
value obtained from the control system itself.
The dimensional accuracy data have been entered into
the Reproducibility Testr programme developed by Traian
Onaciu of the ASCAMM Foundation to analyse the results
obtained in the prole measuring. This programme allows
upper or lower tolerance intervals to be assigned, and for
each case determines which of the measured points are
found outside this interval. Similarly, the opposite process
can be carried out and the minimum tolerance for which all
the points are found within the interval can be determined.
The values appearing in the Upper Tolerance and Lower
Tolerance columns of Table 7 have been determined using
this reverse procedure.
5.1. Cycle time analysis
In order to analyse the relationship between cycle time
and factors F (programmed feed rate) and I (interpolation
type), the main effects of each factor, the rst-order

Fig. 3. (a) Machined proles in MC1. (b) Blocks corresponding to the tests made in the two machining centres.

Fig. 4. Distribution of the points to measure over the prole.

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

interactions and the regression equations of the model,

have been calculated. The T factor (type of toolholder) has
been rejected in this analysis because it has no effect on
cycle time.
The experiment can be considered as a factorial design
with two variables at two levels replicated twice. Each
replication corresponds to a block.
Calculating the main effects and the rst-order interactions for each one of the factors leads to the results
presented in Table 8. A graphic representation of the main
effects is provided in Fig. 6.
The results obtained for the two machining centres are
similar, yet slightly higher for MC1. The variable most
affecting the cycle time is the programmed feed rate.
However, the interpolation type also inuences the

241

duration of the cycle. Specically, the use of circular

interpolation reverts to a reduction of the cycle time.
As shown in Table 9, the experiments carried out with
linear interpolation are 17% longer than those done at
the same feed rate but with circular interpolation. For the
highest feed rate (5 m/min) the difference between the
length of the cycles in linear and circular interpolation is
greater than it is for the lowest feed rate (3 m/min). In the
rst case, the linear interpolation means a maximum
increase of approximately 4%, while in the second case
the increase is greater than 7%.
Regarding the variation of the cycle time according to
the feed rate, there must obviously be a reduction of the
cycle time as the programmed feed rate is increased.
Nevertheless, this reduction is less than what was expected:
a 30% increase in the feed rate only results in a reduction in
cycle time oscillating between approximately 14% and
18%. This difference is due to the geometric complexity of
the part and the difcult dynamic demands that that
entails. During an important part of the cycle, the average
actual feed rate is less than the programmed one, and
therefore the time reduction is not proportional to the
increase in the feed rate. This analysis, presented in greater
detail in Table 10, allows us to conclude that the dynamic
response of MC2 is greater than that of MC1.
Table 8
Main effects on the cycle time of the factors analysed

Fig. 5. Measuring the proles in the 3 coordinate machine.

Main effect of the factor

MC1

MC2

Average
I+block
F+block
IF+block
Block

949.12571.8186
19.7573.67319
155.7573.67319
1.7573.67319
0.75

819.62571.23111
40.7572.46221
117.7572.46221
12.7572.46221
5.2572.46221

Table 7
Results of the response variables in the experiments carried out in MC1 and MC2
Test

T (type of toolholder)

I (interpolation)

F (feed) (m/min)

Cycle time (s)

Upper tolerance (mm)

MC1
1
2
3
4
5
6
7
8

Mechanical
Mechanical
Mechanical
Mechanical
Thermal
Thermal
Thermal
Thermal

Circular
Linear
Linear
Circular
Linear
Linear
Circular
Circular

3.5
5
3.5
5
3.5
5
3.5
5

1020
885
1031
862
1041
879
1016
859

0.200
0.180
0.175
0.165
0.160
0.140
0.130
0.110

0.155
0.080
0.095
0.105
0.085
0.120
0.140
0.155

MC2
1
2
3
4
5
6
7
8

Mechanical
Mechanical
Mechanical
Mechanical
Thermal
Thermal
Thermal
Thermal

Linear
Circular
Circular
Linear
Linear
Circular
Linear
Circular

3.5
3.5
5
5
5
5
3.5
3.5

887
865
731
785
790
737
898
864

0.220
0.210
0.210
0.205
0.220
0.220
0.220
0.220

0.050
0.055
0.055
0.050
0.045
0.045
0.045
0.040

Lower tolerance (mm)

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

242

Main Effects Plot for Cycle time

Main Effects Plot for Cycle time

880

1030

Cycle time

Cycle time

860
990
950
910

840
820
800
780
760

870
Linear
Circular
Interpolation

3.5
5
Feed (m/min)

Linear
Circular
Interpolation

3.5
5
Feed (m/min)

Fig. 6. Main effects plot for cycle time. Results corresponding to MC1(left) and MC2 (right).

Table 9
Percentage of the increase of the length of the cycle time of the
experiments carried out with linear interpolation compared to those done
with circular interpolation for MC1 and MC2
F (Feed) (m/min)

MC1

MC2

3.5
3.5
5
5

1.08%
2.46%
2.67%
2.33%

2.54%
3.94%
7.39%
7.19%

Table 10
Percentage increase of the cycle time length of the experiments carried out
with the minimum programmed feed rate with respect to those carried out
at the maximum feed rate, for MC1 and MC2
I (interpolation type)

MC1 (%)

MC2 (%)

Linear
Circular
Linear
Circular

16.50
18.33
18.43
18.28

12.99
18.33
13.67
17.23

Fig. 7 shows the interaction plots for the experimental

design factors. In all cases the cycle time in MC2 has been
less than in MC1, given that the latter has less dynamic
features.
The results obtained in the rst-order interactions for
MC1 and MC2 are similar, since the interaction between
the different factors is not inuenced by the characteristics
of the machine.
In conclusion, the lowest cycle time is obtained from the
highest values for feed rate and circular interpolation.
In order to compare different machining centres, there
are other parameters that should be taken into account:
maximum acceleration and maximum jerk, and the
interaction between them. Different authors, as Erkorkmaz
and Altintas [22], Yong and Narayanaswami [23] and
Geldart et al. [24], consider that those parameters have a
signicant inuence on cycle time. The introduction of
those parameters in the analysis of cycle time will be
considered by the authors in future work.

5.2. Dimensional error analysis

The Reproducibility Testr has been used to calculate the
values of the Upper Tolerance and Lower Tolerance
columns that appear in Table 7, as well as to study which
points are most critical. This analysis started with a large
tolerance interval, for both the upper part and the lower
part, so that all the points were included within said
interval. After that, the tolerance was reduced so that some
points remained outside the interval. In this way it was
possible to observe which type of geometry is most critical
in each case. For most of the experiments, the critical
points are practically the same. Nevertheless, in certain
combinations of factors the following characteristics can be
observed:




The experiments performed with a thermal toolholder

have less dimensional error than those carried out using
a mechanical toolholder. The difference is especially
obvious in the machining of sharp edges.
More error is produced in the convex curves using
circular interpolation than when linear interpolation is
used.
No signicant variation of the accuracy is observed
when the feed rate is increased.

To facilitate dimensional error analysis, the Error

parameter has been calculated for each experiment as the
difference between the values of the Upper Tolerance and
Lower Tolerance columns in Table 7. The Error value
represents the value of the minimum tolerance interval for
which all the measured points are found within the interval.
The experiment is analysed by zones to obtain greater
representativity of the results. The zones to be analysed are
those labelled A and B in Fig. 1.
The analysis of the experiment gives the main effects of
the three analysed factors presented in Tables 11 and 12. A
graphic representation of these effects in zones MC1 and
MC2 is presented in Figs. 8 and 9, respectively.
Unlike what has been observed in the cycle time analysis,
in this case there are some important divergences between
the Error parameter obtained in the two machining centres.

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

This parameter has greater values in the tests of MC1 than

in those of MC2, especially in Zone B, indicating that a
larger dimensional error was produced in the tests of the
rst centre.
The most important difference refers to the toolholder.
According to the results obtained in MC1, more accuracy
is obtained with the mechanical toolholder than with the
thermal one, while the results obtained in MC2 indicate the

opposite is true. This divergence might be due to the fact

that only one of the two machining centres (MC2) has a
spindle temperature sensor. In addition, MC2 has a much
smaller volume of work and tends to be more accurate.
Finally, it has to be taken into account that MC2 uses
linear encoders as position sensors. This might be a
reection of the different behaviours when each centre is
confronted with error, since not having temperature

Interaction Plot for Cycle time

Interaction Plot for Cycle time

1040

Feed=-1.0

880

980

Cycle time

Cycle time

910

Feed=-1.0

1010

950
920
890

Feed=-1.0
Feed=-1.0

850
820
790

Feed=1.0

760

Feed=1.0
Feed=1.0
Linear
Circular
Interpolation

860

243

Feed=1.0
Linear
Circular
Interpolation

730

Fig. 7. Interaction plots for the cycle time with results corresponding to MC1 (left) and MC2 (right).

Table 11
Results of the analysis of the dimensional error by zones in the experiments done in MC1 and MC2
Test

T (type of
toolholder)

I (interpolation)

F (feed)
(m/min)

Upper
tolerance
zone A (mm)

Lower
tolerance
zone A (mm)

Error zone
A (mm)

Upper
tolerance
zone B (mm)

Lower
tolerance
zone B (mm)

Error zone
B (mm)

MC1
1
2
3
4
5
6
7
8

1
1
1
1
1
1
1
1

1
1
1
1
1
1
1
1

1
1
1
1
1
1
1
1

0.025
0.030
0.035
0.045
0.025
0.025
0.030
0.035

0.040
0.075
0.085
0.100
0.075
0.120
0.125
0.155

0.065
0.105
0.12
0.145
0.1
0.145
0.155
0.19

0.025
0.030
0.035
0.045
0.025
0.025
0.030
0.035

0.055
0.080
0.095
0.105
0.085
0.120
0.140
0.155

0.08
0.11
0.13
0.15
0.11
0.145
0.17
0.19

MC2
1
2
3
4
5
6
7
8

1
1
1
1
1
1
1
1

1
1
1
1
1
1
1
1

1
1
1
1
1
1
1
1

0.025
0.025
0.025
0.025
0.030
0.030
0.030
0.030

0.050
0.055
0.055
0.050
0.045
0.045
0.045
0.040

0.075
0.08
0.08
0.075
0.075
0.075
0.075
0.07

0.015
0.020
0.025
0.010
0.015
0.030
0.015
0.030

0.030
0.040
0.035
0.030
0.030
0.030
0.025
0.025

0.045
0.06
0.06
0.04
0.045
0.06
0.04
0.055

Table 12
Main effects on tolerance of the analysed factors
Main effect of the factor

Zone A
MC1

T (type of toolholder)
I (type of interpolation)
F (feed)

0.0387570.02625
0.0212570.02625
0.0362370.02625

Zone B
MC2
0.0037570.00125
0.0012570.00125
0.0012570.00125

MC1
0.0362570.02625
0.0237570.00125
0.0262670.00125

MC2
0.0012570.00125
0.0162570.00125
0.0012570.00125

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

244

Main Effects Plot for Error

0.16

0.14

0.15

0.13

0.14

Error

Error

Main Effects Plot for Error

0.15

0.12
0.11

0.13
0.12

0.1

0.11
Mechan. Thermal Linear Circular
Toolholder
Interpolation

3.5
5
Feed

Mechan. Thermal Linear Circular

Toolholder
Interpolation

3.5
5
Feed

Fig. 8. Main effects plot for tolerance. Results corresponding to Zone A (left) and Zone B (right) for MC1.

(X 0.001)

Main Effects Plot for Error

Main Effects Plot for Error

(X 0,001)

78
77

57

76

54
Error

Error

60

75

51
48

74

45

73

42
Mechan. Thermal Linear Circular
Toolholder

Interpolation

3.5

Feed

Mechan. Thermal Linear Circular

Toolholder

Interpolation

3.5

Feed

Fig. 9. Main effects plot for tolerance. Results corresponding to Zone A (left) and Zone B (right) for MC2.

control increases error in the vertical direction (Z-axis) to a

great extent. This difference might also explain, in part, the
superiority of the error values of MC1 over those of MC2.
In addition, it is important to point out that the
Tolerance value obtained is less when circular interpolation
is used and therefore greater accuracy is obtained. The
main effects graph also shows that the three factors
analysed affect dimensional accuracy, a result, which
maintains interest in this study high.
The interaction plot of the factors for the tolerance
response is shown in Figs. 10 and 11. It is important to pay
close attention to the scale difference between the two
graphs, since it allows us to see that the values for the
Tolerance variable in MC1 are greater than in MC2 and,
therefore, more dimensional error is produced.
The graphs do not present parallel lines, thereby
indicating strong interaction between all the pairs of
factors in each of the cases (MC1 and MC2). However,
the interactions obtained for one or another centre are
noticeably different. The differences obtained in the
interactions involving the variable T (corresponding to A
in the graph) might be due to the mechanical and thermal
toolholders used not being exactly the same for the two
centres.

6. Conclusions and future work

DoE is a very useful tool when modelling the responses
of different variables against a set of factors. In this study,
this technique has allowed the variation of the cycle time
and the dimensional accuracy to be studied with respect to
the programmed feed rate, the type of interpolation used
and the toolholder.
It has been demonstrated that circular interpolation,
which is used very little in the industry, allows more
precise parts to be obtained more efciently than linear
interpolation. It has also been shown that the toolholder
has a signicant inuence over dimensional accuracy.
With regard to the feed rate, a substantial increase in its
value causes a decrease in the cycle time, but not at the
same rate, given that in complex shapes like those used in
the experiment, the effect of the accelerations and
decelerations is so important that it distorts the nal
results.
In the future this work could be extended by adding
other factors, such as the guide systems mentioned
previously, to analyse the same two responses, cycle time
and dimensional accuracy, or other parameters of interest,
such as surface nish.

ARTICLE IN PRESS
M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

(X 0.001)

Interaction Plot for Error

+
-

92
-1.0

1.0

-1.0

AB

0.16
Error

Error

152
132

172

112

Interaction Plot for Error

+

0.18

192

245

0.14
0.12

1.0

-1.0

-1.0

1.0

+
-

0.1
AC

1.0

-1.0

1.0

-1.0

AC

AB

BC

1.0
BC

Fig. 10. Interaction plot for tolerance. Results corresponding to Zone A (left) and Zone B (right) for MC1. Note: T corresponds to the type of toolholder
( 1: mechanical, 1:thermal), I corresponds to the type of interpolation ( 1: linear, 1: circular) and F corresponds to the variable feed ( 1:3.5 m/min,
1:5 m/min).

(X 0.001)

Interaction Plot for Error

60
+

76
-

74

Interaction Plot for Error

+
+

57

78
Error

(X 0.001)

54
Error

80

51

48

45
-

72
-1.0

1.0
AB

-1.0

1.0
AC

42
-1.0

1.0

-1.0

BC

1.0
AB

-1.0

1.0
AC

-1.0

1.0
BC

Fig. 11. Interaction plot for tolerance. Results corresponding to Zone A (left) and Zone B (right) for MC2. Note: T corresponds to the type of toolholder
( 1: mechanical, 1:thermalr), I corresponds to the type of interpolation ( 1: linear, 1: circular) and F corresponds to the variable feed ( 1:3.5 m/min,
1: 5 m/min).

Another possible extension of this work might be

repeating the experiments in different machining centres,
with the goal of quantifying the improvements obtained in
the part thanks to certain features of the machine tools. It
would be especially interesting to use machines equipped
with temperature sensors for a better comparison of the
results obtained in MC2.
Acknowledgements
The authors would like to express their gratitude to the
ASCAMM Foundation, and would especially like to thank
Mr. Xavier Planta`, Mr. A`lex Crehuet and Mr. Traian
Onaciu for the facilities they have blessed us with. This
work was carried out with support of the ITESM through
their Research Chair in Mechatronics.
References
[1] CIRP Working Group Modelling of Machining Operations: C.A.
Van Luttervelt, T.H.C. Childs, L.S. Jawahir, F. Klocke, P.K.
Venuvinod, with contributions from: Y. Altintas, E. Armarego,
I. Grabec, J. Leopold, B. Lindstrom, D. Lucca, T. Obicawa,

[2]
[3]

[4]

[5]

[6]

[7]

[8]

Shirakashi, S.H, Present Situation and Future Trends in Modelling

of Machining Operations Progress Report of the CIRP Working
Group Modelling of Machining Operations, Annals of the CIRP
47/2, 587-626, 1998.
T. Altan, B. Lilly, Y.C. Yen, Manufacturing of dies and molds,
Annals of the CIRP 50 (2001) 405423.
J. Vivancos, C.J. Luis, L. Costa, J.A. Ort z, Optimal machining
parameters selection in high speed milling of hardened steels for
injection moulds, Journal of Materials Processing Technology
155156 (2004) 15051512.
L.N. Lopez de Lacalle, A. Lamikiz, J.A. Sanchez, J.L. Arana,
Improving the surface nish in high speed milling of stamping dies,
Journal of Materials Processing Technology 123 (2002) 292302.
R.T. Coelho, L.R. Silva, A. Braghini, A.A. Bezerra, Some effects of
cutting edge preparation and geometric modications when turning
INCONEL 718TM at high cutting speeds, Journal of Materials
Processing Technology 148 (2004) 147153.
M. Terrier, A. Dugas, J.Y. Hascoet, Qualication of parallel
kinematics machines in high-speed milling on free form surfaces,
International Journal of Machine Tools & Manufacture 44 (2004)
865877.
M. Monreal, C.A. Rodriguez, Inuence of tool path strategy on the
cycle time of high-speed milling, Computer-Aided Design 35 (2003)
395401.
A. Kaldos, I.F. Dagiloke, A. Boyle, Computer aided cutting process
parameter selection for high speed milling, Journal of Materials
Processing Technology 61 (1996) 219224.

ARTICLE IN PRESS
246

M. Albert et al. / International Journal of Machine Tools & Manufacture 47 (2007) 236246

[9] W.T. Chien, C.S. Tsai, The investigation on the prediction of tool
wear and the determination of optimum cutting conditions in
machining 17-4PH stainless steel, Journal of Materials Processing
Technology 140 (2003) 340345.
[10] H. Juan, S.F. Yu, B.Y. Lee, The optimal cutting-parameter selection
of production cost in HSM for SKD61 tool steels, International
Journal of Machine Tools & Manufacture. Design, Research and
Application 43 (2003) 679686.
[11] L.N. Lopez de Lacalle, A. Lamikiz, M.A. Salgado, S. Herranz, A.
Rivero, Process planning for reliable high-speed machining of
moulds, International Journal of Production Research 40 (12)
(2002) 27892809.
[12] M. Arnone, Mecanizado de Alta Velocidad y Gran Precision, Ed. El
Mercado Tecnico, Bilbao, 2000.
[13] C.K. Toh, Design, evaluation and optimisation of cutter path
strategies when high speed machining hardened mould and die
materials, Materials and Design 26 (6) (2005) 517533.
[14] J. Aranceta, R. Uribe-Etxeberria, J.M. Penagaricano, X. Sabalza,
J. Hernandez, Controles numericos de nueva generacion, Informacion de maquinas-herramienta, equipos y accesorios 251B (1999).
[15] M. Albert , J. de Ciurana, M. Casadesus, A system for optimising
cutting parameters when planning milling operations in high speed
machining, Journal of Materials Processing Technology, in press.
[16] I.F. Dagiloke, A. Kaldos, S. Douglas, B. Mills, High-speed
machining: an approach to process analysis, Journal of Materials
Processing Technology 54 (1995) 8287.

[17] T. Raj Aggarwal, General Theory and its application in the highspeed milling of aluminium, in: R.I. King (Ed.), Handbook of HighSpeed Machining Technology, Chapman & Hall, New York, 1985,
pp. 197240.
[18] NIST/SEMATECH e-Handbook of Statistical Methods, http://
www.itl.nist.gov/div898/handbook/ (accessed 06.04.04).
[19] D.C. Montgomery, Diseno y Analisis de Experimentos, Grupo
Editorial Iberoamerica, Mexico, 1991.
[20] J. Antony, Design of Experiments for Engineers and Scientists,
Butterworth Heinemann, London, 2003.
[21] L.N. Lopez de Lacalle, A. Arteta, A. Lopez, J. Aranceta, Piezas de
ensayos para mecanizado a alta velocidad, Informacion de maquinasherramienta, equipos y accesorios 251B (1999).
[22] K. Erkorkmaz, Y. Altintas, High speed CNC system design. Prt I:
jerk limited trajectory generation and quintic spline interpolation,
International Journal of Machine Tools & Manufacture 41 (2001)
13231345.
[23] T. Yong, R. Narayanaswami, A parametric interpolator with
conned chord errors, acceleration and deceleration for NC
machining, Computer-Aided Design 35 (2003) 12491259.
[24] M. Geldart, P. Webb, H. Larsson, M. Backstrom, N. Gindy,
K. Rask, A direct comparison of the machininh performance of a
variax 5 axis parallel kinetic machining centre with conventional 3
and 5 axis machine tools, International Journal of Machine Tools &
Manufacture 43 (2003) 11071116.