International Journal of Machine Tools & Manufacture 111 (2016) 31–42

Contents lists available at ScienceDirect

International Journal of Machine Tools & Manufacture

journal homepage: www.elsevier.com/locate/ijmactool

Analysis, optimization and accuracy assessment of special-purpose

portable machines by virtual techniques
J. Eguia a,n, L. Uriarte a, A. Lamikiz b
a
IK4 – TEKNIKER, Polo Tecnológico de Eibar, Calle Iñaki Goenaga 5, 20600 Eibar, Spain
b
Departamento de Ingeniería Mecánica, Escuela Técnica Superior de Ingeniería, Universidad del País Vasco UPV/EHU, Alameda de Urquijo s/n, 48013 Bilbao,
Spain

art ic l e i nf o a b s t r a c t

Article history: This paper proposes a streamlined method to model, analyze and assess the overall performance of small
Received 6 July 2016 mobile machines that can move along large parts to perform the required machining operations, con-
Received in revised form ventionally called portable machines. The method is based on virtualization techniques and combines a
14 September 2016
process-force mechanistic model and a reduced machine stiffness model synthesized from virtual and
Accepted 20 September 2016
experimental reduced models of subsystems. The model is used to assess and improve the performance
Available online 21 September 2016
of portable machines by examining the time-domain response of the tool center point in representative
Keywords: operations, rather than limiting the study to the frequency domain. A practical application to a particular
Portable machine portable machine is presented and used to conduct the presentation of the work. With the results of the
Model reduction
analysis, the accuracy of the use of the portable machines is studied. The procedure also proves to be a
Virtual machine
useful tool to optimize the machine design to fit particular applications. The method has been experi-
Optimization
Accuracy mentally evaluated in a conventional three axis milling machine to ensure the accuracy of the simula-
tions.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction: modeling portable machines
Portable machines are small mobile and free machines that can
move along large parts to perform in-situ machining operations
contributing to the manufacturing and maintenance of large parts.
Their advantages and current industrial use have been thoroughly
described by Uriarte et al. [1]. Unlike conventional machines,
portable machines do not show a high flexibility to adapt to dif-
ferent parts, different chip removal processes and changing
machining scenarios [2]. Owing to this, design engineers have
difficulties judging whether a particular portable machine design
is optimum for a given machining task. As a consequence, the
present machine tool industry demands rapid analysis of machine
performance and machine-cutting process interactions in the vir-
tual environment [3]. Usually, machine tools based in workshops
are analyzed searching for improved dynamic stiffness, related to
accurate parts, and maximum stable depths of cut, related to high
productivity. All these machine tool features are variable within
the work volume of the machine. At the same time, machine mass
should be kept low for high speed positioning and high-
n
Corresponding author.
E-mail addresses: josu.eguia@tekniker.es (J. Eguia),
luis.uriarte@tekniker.es (L. Uriarte), aitzol.lamikiz@ehu.eus (A. Lamikiz).
productivity machining. In this context, assessment and optimi-
zation of machine tools needs quick evaluation of different design
alternatives in the virtual environment before physical prototyp-
ing [3].
Usually, the evaluation of machine tools is performed studying
its response in the frequency domain. In the virtual environment,
finite element (FE) models are created, which can efficiently re-
present machine subsystems such as feed drives [4], spindles [5]
and so forth. However, an accurate full representation of a ma-
chine can be computationally costly and require precious time and
resources [6]. Once the model is created, dynamic characteristics
(modes, frequency, tool tip frequency response functions etc.) are
obtained. Depending on the specific requirements on the machine,
further analysis can be focused on stability limits [7], productivity,
dynamics [8] etc.
This situation is worse when position dependent characteristics
need to be studied, so machine studies are performed in average
or “representative” positions. The alternative would be modeling
the position dependency using coupled FE and multibody simu-
lation packages. These methods have been proved useful for rigid-
flexible body analysis but are still problematic for large flexible
machines with bodies subject to relative motion. Certain as-
sumptions and simplifications can give acceptable results but are
only of use under certain conditions. Moreover, the increasing
modular approach of the design process means that individual

http://dx.doi.org/10.1016/j.ijmachtools.2016.09.006
0890-6955/& 2016 Elsevier Ltd. All rights reserved.
32 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42
subsystems have to be modeled and evaluated separately, creating
complex situations during the synthesis process due to varying
mesh resolutions at the interface, compatibility conditions etc.
Besides this, the most demanding scenario would be having to
simulate relevant or typical manufacturing operations in the time
domain. For this, an accurate process force model is needed, which
must be coupled to the FE-multibody system. Process forces can be
predicted with different accuracy levels employing a wide variety
of analytical, numerical or mechanistic models. These three ap-
proaches show different computational costs and must be con-
nected somehow to the FE-multibody solver. For a large machine,
with an amount of nodes leading to more than a million degrees of
freedom, this approach is simply unfeasible for its computational
cost and complexity [6].
However, portable machines are usually very small in size, with
minor structures and working spaces [9] and therefore they can be
accurately represented by a mid-sized model with sufficient ac-
curacy. As a consequence, the described approach would not be
extremely computationally costly and FE-multibody systems re-
main a valid tool for complex and complete simulations.
In spite of this, the portable-machine concept is finding diffi-
culties to enter the industrial use. Technicians are not confident
about their true potential and how portable machines will even-
tually behave on site under variable conditions: e.g. portable ma-
chines are affected by changing gravity directions depending on
their position and orientation in space and with respect to the
work piece. The changing gravity direction may affect the ma-
chines in different ways.
Some recent works have already started to explore the limits:
virtual techniques were used by Olarra [10] to develop methods
for direct tool center point position identification with respect to
the machined feature. Law eta al. [11,12] proposed a framework to
predict the dynamic response of a given mobile machine with
parallel kinematics with varying clamping solutions based on
model-based substructure analysis and receptance coupling, and
limiting the study to the effect at the tool-center-point (TCP) in the
frequency domain. And recently, Eguia et al.. performed the ana-
lysis of the error budget and the uncertainty of the use of portable
machines equipped with triangulation line laser scanner for ma-
chine-part inter-referencing [13].
To facilitate the evaluation of portable machines, this paper
proposes an integrated virtual method based on the time-domain
analysis of a computationally manageable multibody dynamic
model of a portable machine. The method responds to the need for
a streamlined, fast and sufficiently accurate method to assess the
performance of machine tools under different circumstances. A
substructural modeling and analysis approach is used for the
Fig. 1. Concept design of portable machine.
machine, models which are then reduced to position independent
light models. The machine model is then obtained from the
synthesis of these reduced models and coupled to a base me-
chanistic force calculation method for the time domain analysis.
This machine representation is used to study and optimize some
components limiting machine accuracy and target requirements.
In parallel, representative machining operations are represented
and their forces calculated with a mechanistic model. These force
values are then coupled to the machine model along the ma-
chining trajectories. The analysis is performed tracking the TCP
displacements along meaningful machining operations. The ob-
tained deflections and TCP vibrations are then employed to feed an
error budget model to assess the accuracy of the machine [14], as
it is done in non-conventional machine systems [15,16]. Successive
improvements in subsystem or component levels are also possible
following the diagram depicted in Fig. 1. All this is performed
before the actual machine is built, to maximize the impact of the
virtual model in the design phase of portable machines.
The effect of this method on the virtual design and optimiza-
tion of portable machines is shown on a particular machine al-
ready presented by the authors [17], leading to breakthrough
characteristics in stiffness/weight ratios and part accuracies. To
ensure that the conclusions obtained from the model of the por-
table machine are accurate, the whole virtual method is experi-
mentally validated on a conventional three axis milling machines,
which is modeled exactly like the portable machine. On this
conventional machine, machining tests are performed to prove
that the simulated results are acceptable.
The paper is organized as follows: the target machine and its
functional requirements are presented in Section 2; in-depth in-
sight on the virtual machine model is given in Section 3; in Section
4, the machine performance is analyzed in the time domain and
optimized to meet the requirements of the application, followed
by the experimental evaluation in a conventional machine in
Section 5 and conclusions in Section 6.
2. A portable machine for large vacuum vessels
In response to the requirements of its application described in
[18], authors developed a new portable milling machine [16], a
five-axis, miniature milling machine based on a serial kinematic
architecture. This portable machine can perform both mid-duty
milling and drilling operations in a five axis configuration. To
achieve this, three stacked linear axes are included, which carry a
two axis rotary-head holding the spindle, in a compact machine
envelope of 1200
1200
1200 mm. The work volume of the
 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42
Fig. 2. Concept design of portable machine.
machine is 340
300
220 mm with spindle rotary axes C (rota-
tion around Z axis) and A (rotation around X axis) ranging from
40 to 100° and 0–420°, respectively. The machine is equipped
with a lines laser scanning system for machine-piece inter-refer-
encing operations, a systems described in [17] (Fig. 2).
3. Streamlined substructural formulation for time-domain
assessment
The method proposed has been applied to the design process of
the portable machine presented. The context of this development
is a broader activity to make serial-kinematic portable machines
competitive against the better-established parallel-kinematic
portable machines. Hence, the method proposed in this paper is a
tool to define the limits and functional capabilities of new state-of-
the-art portable machines.
3.1. Substructure models
The time domain evaluation of the machine is modeled using a
substructural assembly approach. Early during the design phase,
each of the substructures of the machine (lower frame, upper,
frame, body and ram) are represented in a FE package and reduced
independently. FE models for the structural substructures were
generated from their CAD models using Tet10 elements with the
following material properties: modulus of elasticity of 169 GPa;
density of 7800 kg/m3; Poisson's ratio of 0.25. The elements
modeled in this way include the lower frame, the upper frame, the
body and the moving ram. For the whole structure, a damping
Fig. 3. Experimental receptance of the bi-rotary system in free-free configuration (5 points).
33
coefficient of 0.02 has been considered, a bit lower than other
sound references [8] but in tune with others [7].
The subassembly composed by the bi-rotary head, the ma-
chining spindle, the tool holder and the tools has been experi-
mentally characterized in free-free configuration to obtain their
frequency response under impulse forces located at the TCP (Point
100 of Fig. 3). Fig. 3 shows the actual subassembly and the re-
sponse amplitude of the acceleration as a function of the fre-
quency of the exciting force of five different points of the sub-
system 1–5) during the experiment. The response of the TCP is not
shown for scale purposes since its values would be much higher.
With those six points it is possible to create a reduced model
which is accurate enough and light in order to keep computational
costs low.
The portable machine modeled employs ball-screw plus nut
technology as a solution for feed drives. Previous works [19,20]
prove that axial modes and torsion modes can not be neglected in
a faithful and precise representation of the feed drive, whereas
that the influence rate of the bending effect of the ball-screw
system is not significant regarding overall dynamics of the ma-
chine. Therefore, this paper employs a model where only axial and
torsional dynamics have been represented accurately. The model is
based on spring elements to represent ball screw drives, as a
sound trade-off between accuracy and computational cost. The
details of the model can be seen in the following Fig. 4. The ball
screw is divided into two elements defined by the position of the
nut, which divides the ball-screw in two. In that context, x is the
position of the nut (or the position of the moving element), s is the
stroke of the moving axis. Thus L(x) is the distance between one of
the supports of the ball screw to the nut (a function of the axis
34 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42
Fig. 4. Model for the ball screw, nuts and supports.
coordinate x) whereas L(x-s) is the distance from the nut to the
other end.
In the FEM model, the two elements are represented by dif-
ferent axial and torsional stiffness values. In fact, the stiffness
values considered for the model vary with the actual position of
the nut and carriage. If the nut and the carriage are located at a
given position x along the stroke s of the ball screw, the stiffness of
the support element near the origin of x would follow the fol-
lowing expressions:
EA JG
KL = Kθ =
L( x ) L( x )
whereas the other support element would show the following
values:
EA JG
KL = Kθ =
L( s − x ) L( s − x )
In those equations, E is the young modulus of the material, A an
equivalent section of the ball screw, J is the torsional constant and
G is the modulus of rigidity (shear modulus) of the material.
3.2. Substructure reduction and synthesis
Once the different subsystems have been modeled either ex-
perimentally or through simplified FEM models, the machine has
been assembled. To faithfully represent the machine behavior
while keeping the amount of modes and nodes in reasonable
figures, the Craig-Bampton (CB) method (Fig. 5) has been used to
assemble the machine from its individual subsystems [21].
This technique is based on the selection of some nodes and
degrees of freedom (DOF) which will not be subject to mode
synthesis. These degrees of freedom are the boundary degrees of
freedom, linked to boundary nodes whereas the rest are named
interior nodes and degrees of freedom. In this way, structural
modes can be divided in two groups:
 Constraint modes: These modes are static shapes obtained by
giving each boundary DOF a unit displacement while holding all
other boundary DOF fixed. The basis of constraint modes com-
pletely spans all possible motions of the boundary DOFs, with a
one-to-one correspondence between the modal coordinates of
the constraint modes and the displacement in the correspond-
ing boundary DOF.
 Fixed-boundary normal modes: These modes are obtained by
fixing the boundary DOF and computing an eigensolution. There
are as many fixed-boundary normal modes as the user desires.
These modes define the modal expansion of the interior DOF.
The quality of this modal expansion is proportional to the
Subsystem 3
Interior nodes
boundary
Subsystem 2 nodes
Subsystem 1
Interior nodes
Fig. 5. Interior and boundary nodes according to CB-method.
number of modes retained by the user.
However, the raw Craig-Bampton modal basis has one defi-
ciency that makes it unsuitable for direct use in a dynamic system
simulation in Adams, which is the package used in the reported
investigation: embedded in the Craig-Bampton constraint modes
are 6 rigid body DOF which must be eliminated before the Adams
analysis because Adams provides its own large-motion rigid body
(1) DOF.
This problem with the raw Craig-Bampton modal basis is re-
solved by applying a simple mathematical operation on the Craig-
Bampton modes. Since the Craig-Bampton modes are not an or-
thogonal set of modes, by solving an eigenvalue problem we ob-
(2) tain eigenvectors that we arrange in a transformation matrix,
which transforms the Craig-Bampton modal basis to an equivalent,
orthogonal basis with modal coordinates:
The orthogonalized Craig-Bampton modes are not eigenvectors
of the original system. They are eigenvectors of the Craig-Bampton
representation of the system and as such have a natural frequency
associated with them. A physical description of these modes is
difficult, but in general the following is observed: Fixed-boundary
normal modes are replaced with an approximation of the eigen-
vectors of the unconstrained body. This is an approximation be-
cause it is based only on the Craig-Bampton modes. Out of these
modes, 6 modes are usually the rigid body modes, which can now
be disabled.
From there on two bodies A and B can be coupled easily by
identifying their coincident boundary degrees of freedom.
x Ab = x Bb → p Ab = p Bb
In that equation, b stands for boundary, x is the vector con-
taining the degrees of freedom and p is the vector containing
modal coordinates achieved from the explained process. If q in-
dicates a fixed boundary mode, to couple the two structures:
⎡ pA ⎤
⎢ q ⎥ ⎡ Iq 0 0 ⎤⎡ A ⎤
⎢ ⎥⎢ p q ⎥
⎡ pA⎤ ⎢ pA ⎥ ⎢ 0 Ib 0 ⎥⎢ A ⎥
⎢ ⎥ = ⎢ b⎥ = ⎢ ⎥ p
⎢⎣ p B ⎥⎦ ⎢ p B ⎥ ⎢ 0 0 Iq ⎥⎢ b ⎥
⎢ q⎥ ⎢ ⎥
⎢ B ⎥ ⎢⎣ 0 ⎥ pB
I 0 ⎦⎣ q ⎦
⎣ pb ⎦ b
That way, the modal vector of the coupled substructures can be
obtained. That way, the system can be represented in the usual
manner, once the model has been constrained in the way that best
represents the physical behavior of the system:
 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42 35

Fig. 8. Feed, lateral and vertical milling forces of the rough operation.

Fig. 6. Components and final look of the multibody model.

Table 1
A+ B ¨ A+ B A+ B A+ B A+ B Parameters and main results of the machining operation.
M p +K p =f
Machining parameters Steel
This allows different combinations to be modeled in-
dependently and coupled subsequently without having to re- vc [m/min] 200
generate machine models. Fig. 6 shows the structure of the model D [mm] 12
generated and the position of the supports were interfacing ball N [rpm] 5305
ae [mm] 16
screws have been clamped.
z - N° dientes 3
fz nominal [mm/diente] 0,1
3.3. Process force model F [mm/min] 1061
λs hélix angle [°] 30
Κr position angle (°) 90
The described virtual machine model is completed with a
Kt Tangential specific coeficcient [N/mm2] 2400
sufficiently accurate process force prediction model which can Kr Radial specific coeficcient [N/mm2] 1680
deliver instantaneous process force values at the tool tip during a Ka Axial specific coeficcient [N/mm2] 600
given operation. This means that a “typical” manufacturing op- ap [mm] 1
Maximum Mz [Nm] 1,91
eration must be considered during the virtualization process. The
Maximum Fx[N]  21
portable machine modeled in this paper is designed for both fin- Maximum Fy[N] 266
ishing contouring operations and rough milling operations, so one Maximum Fz[N]  52
regular process of a kind have been considered covering two dif- P[kW] 0,8
ferent conventional materials as work piece materials.
As a rough milling operation, a grooving operation has been
selected on steel, with the following process parameters and
conditions (Fig. 7 and 8). 10791-7). This standard and the part described are used for com-
To predict the cutting forces of said operation, a base me- parison of three-axis-milling machines in contouring operations
chanistic model developed by Gonzalo et al. [22] has been im- and are well suited for our purpose. The footprint of the piece is
plemented and fed by the process defined before. The calculated 160 mmx160 mm, which fits inside the workspace of most por-
process forces are shown in the next figure and Table 1. table machines and can in fact be moved inside said workspace.
As a finishing operation, the final contouring operation to The finishing operation described in that standard is free from
manufacture the standard NAS work piece has been considered. stability or regenerative vibrations. In general, attempts are made
Both the part geometry and the final machining operation are to choose process conditions in which stable cutting is ensured,
defined in UNE 15450-7 (equivalent to the international ISO with no regenerative chatter. The cutting forces of said operation
have been estimated using the same methodology by Gonzalo
et al. [22], following the process conditions found in the standard.
The calculated process forces are shown in the next figure (Fig. 9).

4. Performance analysis and optimization of portable

machines

4.1. Requirements of the portable machine

ITER (International Thermonuclear Experimental Reactor'') is

an international nuclear fusion research and engineering project
Fig. 7. Grooving operation considered in the characterization. which aims to make transition from experimental studies of
36 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42

Fig. 9. Feed and lateral milling forces of the finishing operation.

plasma physics to full-scale electricity-producing fusion power

plants. Its core is an experimental machine called tokamak. The
vacuum vessel of ITER is a paradigmatic example of a gargantuan
system that can only be processed in-situ and from the inside [18].
Its assembly implies performing at least the following operations:
(i) measurement of the existing space between two consecutive
sectors. (ii) positioning of the splice plates between the sectors
(approximate weight 150 kg), (iii) welding the splice plates to the
sector shells, (iv) repair operations, including machining of weld-
ing seams, (vi) leak tests of the welding seams, (vi) any other
contingency operation not planned. These operations should be
possible in any configuration, upside-down and rotating 360°
following the internal surface of the vacuum vessel. According to
Wu et al. [16], the accuracy requirement for the systems to per-
form the different tasks described earlier is 70.1 mm. The por-
table machine modeled in this research is the authors’ response to
these requirements.

4.2. Quick check and optimization

According to the requirements of the application reported in Fig. 10. Optimized components – geometry suggested by optimization software
[18], the moving ram moves out of the body to reach the inner wall and final shapes.
of the ITER tokamak. Therefore, the machine movements are
performed with the ram in the extreme lower position. From here
on, all the machine analysis are performed in this position.
From the general requirements, authors further investigated
the application in [18] and updated the requirements. Among
them, the minimum static stiffness values were fixed to a value of
10 N/mm, with desired stiffness values closer to 15 N/mm. A simple
static stiffness analysis of the designed machine shows stiffness
values below this requested figure. With the ram in the extreme
low position, and the X and Y axes in the mid position of the
strokes, the static stiffness values of the presented design are:
Kxx-9.8 N/mm, Kyy-13.7 N/mm and Kzz¼ 73.8 N/mm [17].
Therefore, a frequency response analysis or a time domain analysis
of the TCP response is not necessary to conclude that the machine
does not meet the requirements. As a response, following the flow
diagram of Fig. 1, the overall stiffness of the system is increased
through structural optimization to achieve the target machining
accuracy. Since the dominant modes refer to the body and the Fig. 11. Frequency response at the tool center point and main modes affecting
moving ram, these two elements are optimized using Altairs machining accuracy.
OptiStructs, a commercial optimization and structural analysis
solver by Hyperworkss (Fig. 10). The design space for both ele- 863.8 kg [16]. This optimization process took 19 h plus 8 additional
ments is defined as all the space other than the interfaces with hours of refinement and changes to ensure manufacturability.
other elements. The optimization process leads to increased stiff- After this design iteration, the TCP frequency response is ana-
ness values Kxx-12.7 N/mm (29% increase), Kyy-18.6 N/mm (35% lyzed as the flow diagram in Fig. 1 calls for. Fig. 11 shows the mode
increase) and Kzz ¼73.8 N/mm (11% increase), with simultaneous shapes with the post-optimization frequency values and the dy-
improvements in the weight of the machine from 932.7 to namic amplitude of the TCP response when subject to unitary
 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42
Fig. 12. Simulated TCP trajectory and error close up in of the sides.
loads in a given frequency. The traditional design procedure would
end at this stage, followed by a physical prototype where machine
accuracy and other functional parameters such as stability would
be evaluated [3]. In the case of portable machines, their tiny size
makes it possible to perform further analysis in the virtual en-
vironment. More specifically, the work in this paper performs the
virtual accuracy evaluation of the machine through the analysis of
the TCP response in the time domain.
4.3. Response in the time domain in representative operations
In the proposed methodology, the multibody model is used to
obtain an estimate of the actual TCP trajectory during machining
operations to confirm the behavior of the machine in the time
domain. This is more profitable and interesting when trying to
analyze precision finishing operations rather than rough milling
operations because the information obtained can be used to pre-
dict the geometry of the machined operation.
The coordinates of the TCP of the calculated/simulated dy-
namics can then be used to draw the estimated trajectory. Fig. 12
shows the simulated trajectory of the TCP when trying to manu-
facture the test work piece. From one of the sides, a close-up is
presented where the difference between the theoretical trajectory
and the simulated trajectory has been depicted focusing on the
error in the lateral direction. Seemingly, the error in the feed di-
rection can be shown. Lateral direction errors are important since
they create differences between the planned part shape and the
final geometry achieved by the machining operation. Different
sides and different errors could be drawn equally, but only one
side and one error have been shown for simplicity.
From this process force and machine structure interaction, the
tool is expected to bend considerably under cutting forces, which
affects the geometrical precision of the work piece machined. The
tool vibrates under process loads, which also affects the surface
roughness obtained, the tool wear and the tool life. To estimate
tool vibration, calculated instantaneous process forces for both
roughing and finishing operations are introduced in the machine
multibody model while machine bodies move in accordance with
the tool path defined by the manufacturing operations.
During the rough milling operations, typical process
37
parameters result in a cutting force with peak values of around
270 N. Subject to this process forces, the virtual model delivers
information about tool deflection and vibration simultaneously.
For the rough milling operation in a central section of the grooving
operation (far from edges or insertion/extraction areas). Fig. 13
shows the result of the displacement of the tool center point in the
feed and lateral directions for the rough milling operation in a
central section of the grooving operation, far from edges or in-
sertion/extraction areas.
In the feed direction, simulations show deflection values (in-
cluding vibration) of around 23 mm, with 9 mm in the lateral
direction.
During finishing operations, process forces are much lower, in
the range of 60 N (Fig. 9). Introduced in the virtual model, these
instantaneous forces generate TCP displacement levels as shown
in Fig. 14.
In the feed direction, simulations show deflection values (in-
cluding vibration) of around 10 mm, with 3 mm in the lateral
direction.
4.4. Accuracy assessment
To assess the accuracy of the use of the portable machine, the
simulated tool deflection and vibration values are combined with
several other error sources to establish the error budget and per-
form a complete uncertainty analysis, as defined by Slocum [14].
The budget includes the most significant elements that affect the
final accuracy of the work piece, i.e. the machine, the process,
auxiliary equipment and the interactions between them [23]. The
error sources considered were summarized by Lamikiz et al. [24].
From all the studied errors, thermal errors are very seldom re-
levant in portable machines, since they operate in very short
periods of time over large parts to perform usually small ma-
chining operations inside a small work volume in comparison with
the whole part. Usually, feed drives and tools are stopped most of
the time, a time mostly devoted to clamp, translate and reference
the machine. As a consequence, heat generation and power con-
sumption occur in a small fraction of total time while the part
represents a massive heat sink. Therefore, thermal deformations
can be neglected without introducing significant errors in the
38 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42
Fig. 13. Displacements of the TCP in the rough grooving operation.
error budget.
However, the ambient temperature can have a remarkable ef-
fect on the machine-work piece combination. First, temperature
changes can create deformations and distortions of the machine
structure, guideways [25] and spindle [26]. In a conventional large
machine with dimensions in the range of meters, these thermal
perturbations would mean that accuracy is lost and positioning
errors would increase. However, once again, the studied portable
machine shows small dimensions and short guideways (the
longest axis stroke is 300 mm), which ensures that externally in-
duced thermal deformations are very low. As such, neglecting
these deformations is also a safe assumption.
Table 2 shows different error sources, their estimated impact as
an expanded uncertainty (K ¼2–95% coverage) and their combined
effect. An in-depth analysis of the estimates of different errors is
beyond the scope of this paper and can be found in [13]. Therefore,
Table 2 is shown as an example of how the results of the virtual
model can be integrated in a broader accuracy analysis of portable
machines. For a given error source, when two values are given, the
upper refers to the rough milling operation, with the lower being
related to the finishing operation. The combined result shows that
the machine analyzed and optimized in this paper can effectively
meet the 200-mm accuracy requirements of the different tasks of
the post-assembly operations of ITER [17,18].
Fig. 14. Displacements of the TCP in the finishing operation.
5. Validation of the virtual method
To validate the virtual method experimentally, the same proce-
dure has been applied to the BIZKOR machine, a conventional three-
axis milling machine (workspace:580 mmx500 mmx360 mm)
owned by IK4-TEKNIKER. The described process leads to the multi-
body model depicted in Fig. 15.
The reduced multibody model was used to perform an analysis
of the dynamic stiffness that could be compared with the ex-
perimental and FEM results of the same machine reported in [6].
For the analysis and comparison, the TCP is located in a centered
position in X and Y machine axes, whereas the ram is placed in its
lower Z position. The frequency response of the machine calcu-
lated with the reduced model is shown in Fig. 16, where the main
modes affecting machining accuracy have also been shown.
As reported in [6], the frequency of the calculated modes show
a maximum difference of 6% with respect to the modes calculated
using a heavy FEM model. This frequency values had themselves a
1% difference with respect to the experimental modes [6]. There-
fore, it is safe to conclude that the reduced model is remarkably
accurate when trying to anticipate the dynamic behavior of the
Bizkor machine.
This reduced multibody model can also be used to obtain an
assessment of the accuracy of the simulations to predict TCP tra-
jectory during a machining operation, by comparing a real and
 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42 39

Table 2
Error-budget of the studied portable machine.

Source of error Technique for estimates Values Type

U (xi) (mm)

Machine errors
Positioning error (including reversal of linear movements) Experiment on equivalent subsystem (machine) 16.38 A

Machine trajectory errors Experiment on equivalent subsystem (machine) 12.6 B

0
Errors in the spindle, including the spindle-shank, shank-collet and collet-tool interfaces From scientific literature and references 12.6 A

Combined machine errors 24.18

20.66

Errors due to process

Tool deflection & vibration Analysis on multibody flexible model of machine 28.3 B
10.44
Tool wear Analysis on multibody flexible model of machine – B
Burr formation Analysis on multibody flexible model of machine 2 B

Combined process errors 14.31

5.59

Auxiliary alignment systems and reference set-up

Uncertainty of the reference position Experiment on subsystem (inter-referencing system) 43 A

Total errors 57.03

48.99

simulated machining operation. As a test machining operation, the

finishing operation of the aluminum part described earlier has
been considered. From the process parameters and the part di-
mensions, the theoretical trajectories of the TCP can de drawn
(Fig. 17).
These coordinates are introduced in the multibody model and
synchronized with the process forces. The lateral displacement
during ten seconds of the machining of the CD side of the piece is
shown in the Fig. 18. In the lateral direction, deflection values
(including vibration) reach values around 0.4 mm.
These figures are used to complete an error budget table for the
three axis milling machine, equivalent to that presented for the
portable machine. However, there are some differences worth
mentioning. During the manufacturing of the work piece, tactile
Fig. 16. Frequency response of the Bizkor Machine and main modes affecting part
probes are used to establish the reference planes and coordinate accuracy.
systems. These work piece alignment systems ensure positioning
repeatability 73 mm. However, these probes are removed to in- of 8.2 mm.
troduce the tool holder and tool, so the alignment error must be Table 3 Shows the different error sources, their estimated im-
added to the repeatability error of the shank, 3 mm, and the tool pact as an expanded uncertainty (K ¼2–95% coverage) and their
holder and collet, 5 mm, giving a maximum expanded uncertainty

Fig. 15. (Left) The Bizkor Machine (Right) Multibody model of the Bizkor Machine.
40 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42

Fig. 17. View of the starting geometry and the nominal TCP trajectories.

combined effect. interval, expanded uncertainty U ¼2 mm):

In order to demonstrate the capability of the virtual model to
predict the result of the manufacturing process, machining trials
were carried out to generate test pieces in the same conditions as
those used during simulations, which come from the ISO 10791-7
standard. A continuous cutting tool with a diameter of 16 mm and
3 cutting edges was used and the process parameters were the
following: ap ¼4.5 mm, ae ¼0.2 mm, Vc ¼300 m/min, fz ¼0.07 mm/
tooth, Vf ¼ 1260 m/min. The extra material in the initial piece
geometry made it possible to repeat this specified machining
process two times before the nominal dimension of 160 mm was
achieved.
Following each machining operation, the test piece has been
quantitatively assessed in terms of geometrical accuracy. The op-
tical observations of the machined surfaces did not show any
evidence of chatter marks proving that the micro-milling process
is dynamically stable within the tested domain.
The results obtained from the subsequent highly accurate CMM
evaluation (ZEISS PRISMO Navigator CMM – Serial number 98619-
7110) can be summarized as follows (expressed at 95% confidence
 Straightness deviation of the sides of the square (AB &
CD):4 mm.
 Straightness deviation of the sides of the square (BC &
DA):3 mm.
 Perpendicularity deviations of two adjacent sides of the square
(ABC and CDA):3 mm.
 Dimension deviations between two parallel sides (AB to CD &
BC to DA):9 mm.
 Parallelism (AB to CD & BC to DA):6 mm.
From these results, it can be noted that, in general, the virtual
model gave higher error values than those in the experimental
machining tests. Authors are inclined to think that this is mainly
due to some overestimates of two error sources: the machine
trajectory errors and the spindle-shank, shank-collet and collet-
tool errors. The figures used have been obtained from the scientific
literature and often represent worst-case-scenarios that may be
easily improved with a careful set-up of the process. If these error

Fig. 18. Displacement of the TCP of the Bizkor machine during the manufacturing of the NAS part.
 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42 41

Table 3
Error-budget of the Bizkor machine.

Source of error Technique for estimates Values Type

U (xi) (mm)

Machine errors
Positioning error (including reversal of linear movements) Experiment on equivalent subsystem (machine) 0.8 A

Machine trajectory errors Experiment on equivalent subsystem (machine) 12.6 B

0
Errors in the spindle, including the spindle-shank, shank-collet and collet-tool interfaces From scientific literature and references 8.2 A

Combined machine errors 15.9

8.2

Errors due to process

Tool deflection & vibration Analysis on multibody flexible model of machine B
0.4
Tool wear Analysis on multibody flexible model of machine – B
Burr formation Analysis on multibody flexible model of machine 2 B

Combined process errors

2.03

Auxiliary alignment systems and reference set-up

Uncertainty of the reference position Experiment on subsystem 8.2 A

Total errors 18
11.7

values are thoroughly studied and more accurate (and probably
lower) figures are determined, the virtual model gives similar
piece geometry error values to those detected during the experi-
mental campaign. As a result, the experimental validation has
shown that the virtual model is an accurate tool to predict the
performance of a particular machine and its use.
Therefore, it is safe to state that the accuracy assessment of the
special portable machine examined in the paper is also accurate or,
in the worst case, errors are slightly overestimated. To extend the
experimental assessment, the method presented is under study to
be used to analyze already existing-portable machines, with
completely different architectures, functions and auxiliary systems
[27].
Even if the method is presented for a specific machine and
certain machining operations, the methodology presented is
generic. Taking into account that it could be used with any ma-
chine tool, the time domain response approach is particularly re-
levant for small machines, so authors conclude that the impact of
this work is particularly important for small-sized special ma-
chines like portable machines. Even more, the method proposed
can be further enhanced to include virtual models of control
strategies, CNC and others to boost its application during early
designs stages.
Acknowledgments

The authors thank the Industry Department of the Basque

6. Conclusions
This paper has showed the feasibility of virtual engineering in
the design process of state-of-the-art portable machines with
stringent accuracy requirements and process and material flex-
ibility demands. Following the procedure demonstrated in this
work, portable machines are assessed based on their time domain
response in machining operations representative of the in-situ
work, by taking into account the position dependency of the ma-
chine coupled to a cost-effective process-force model. Indeed, a
position dependent, substructurally synthesized reduced order
machine model is presented and coupled with a base mechanistic
model to obtain the machine response. The model allows quick
investigation of process-machine interaction errors without hav-
ing to perform experimental tests. Applied to a particular machine,
equipped with given feed drive solutions and specific clamping
and referencing elements, the obtained machine response and tool
deflection and vibration errors can be combined with other error
sources for a general machine assessment.
It is also proved that the reduced order models of substructures
can be used to quickly analyze and redesign components which
limit the accuracy and other performance characteristics (e.g. the
stiffness weigh ratios). With this, the functional capabilities of
serial kinematic portable machines can be enhanced. As a con-
sequence, the method can also be employed to explore functional
capability limits of special machines.
Government for the financial support of this work, which is part of
ACTION IV of the Profuture project (grant number E10-271-2010)
developed under the ETORTEK program.
References
[1] L. Uriarte, M. Zatarain, D. Axinte, J. Yagüe-Fabra, S. Ihlenfeldt, J. Eguia, A. Olarra,
Machine tools for large parts, CIRP Ann.-Manuf. Technol. 62 (2) (2013)
731–750.
[2] J. Allen, D. Axinte, P. Roberts, A review of recent developments in the design of
special-purpose machine tools with a view to identification of solutions for
portable in situ-machining systems, Int. J. Adv. Manuf. Technol. 50 (2010)
843–857.
[3] Y. Altintas, C. Brecher, M. Weck, S. Witt, Virtual machine tool, CIRP Ann.-
Manuf. Technol. 54 (2) (2005) 115–138.
[4] M.F. Zaeh, T. Oertli, J. Milberg, Finite element modelling of ball screw feed
drive systems, Ann. CIRP 53 (2004) 289–292.
[5] Y. Cao, Y. Altintas, A general method for the modelling of spindle-bearing
systems, ASME J. Mech. Des. 126 (2004) 1089–1104.
[6] M. Zatarain, E. Lejardi, F. Egana, Modular synthesis of machine tools, Ann. CIRP
47 (1998) 333–336.
[7] M. Law, Y. Altintas, A. SrikanthaPhani, Rapid evaluation and optimization of
machine tools with position-dependent stability, Int. J. Mach. Tools Manuf. 68
(2013) 81–90.
[8] M. Law, Evaluating mobile machine tool dynamics by substructure synthesis,
Adv. Mater. Res. 1 (1018) (2014) 373–380.
[9] R. Neugebauer, M. Wabner, H. Rentzsch, s. Ihlenfeldt. Structure principles of
energy efficient machine tool, CIRP J. Manuf. Sci. Technol. 4 (2) (2011) 136–147.
[10] A. Olarra, J.M. Allen, D.A. Axinte, Experimental evaluation of a special purpose
miniature machine tool with parallel kinematics architecture: free leg hex-
apod, Precis. Eng. 38 (3) (2014) 589–604.
42 J. Eguia et al. / International Journal of Machine Tools & Manufacture 111 (2016) 31–42

[11] M. Law, H. Rentzsch, S. Ihlenfeldt, Development of a dynamic substructuring Springer 2009, pp. 2019–2260, ISBN: 978-1-84800-379-8 (print), 978-1-
framework to facilitate in-situ machining solutions using mobile machine 84800-380-4 (online).
tools, in: Proceedings of the 43rd North American Manufacturing Research, [25] J. Mayr, J. Jedrzejewski, E. Uhlmann, M. Alkan Donmez, W. Knapp, F. Härtig,
Procedia Manuf., vol. 1, 2015, 756–767. K. Wendt, T. Moriwaki, P. Shore, R. Schmitt, C. Brecher, T. Würz, K. Wegener,
[12] M. Law, H. Rentzsch, S. Ihlenfeldt, Predicting mobile machine tool dynamics by Thermal issues in machine tools, CIRP Ann. Manuf. Technol. 61 (2) (2012)
experimental dynamic substructuring, Int J. Mach. Tools Manuf. 108 (2016) 771–791.
127–134. [26] S. Yang, K. Kim, Y. Park, Measurement of spindle thermal errors in machine
[13] J. Eguia, A. Lamikiz, L. Uriarte, Error budget and uncertainty analysis of por- tool using hemispherical ball-bar test, Int J. Mach. Tools Manuf. 44 (2–3)
table machines by mixed experimental and virtual techniques, Precis. Eng. (2004) 333–340.
(2016), http://dx.doi.org/10.1016/j.precisioneng.2016.07.001. [27] 〈http://www.tekniker.es/en/self-portable-machine-tool〉 (Last accessed
[14] A. Slocum, Precision Machine Design, Prentice Hall, Englewood Cliffs, NJ, USA, 30.08.16).
1992.
[15] L. Uriarte, A. Herrero, M. Zatarain, G. Santiso, L.N. López de Lacalle, A. Lamikiz,
J. Albizuri, Error budget and stiffness chain assessment in a micromilling
machine equipped with tools less than 0.3 mm in diameter, Precis. Eng. 31
(2007) 1–12.
Mr. Josu Eguia is MSc. in Mechanical Engineering and a researcher in precision
[16] D.A. Axinte, S. Abdul Shukor, A.T. Bozdana, An analysis of the functional cap-
machinery at the Mechanical Engineering Department of IK-4 TEKNIKER. His re-
ability of an in-house developed miniature 4-axis machine tool, Int J. Mach.
search interests include machine tool modeling and accuracy enhancement.
Tools Manuf. 50 (2) (2010) 191–203.
[17] J. Eguia, L. Uriarte, A. Lamikiz, A. Fernandez, Main challenges in the develop-
ment of a serial kinematic portable machine for the ITER vacuum vessel, Fu-
sion Eng. Des. 92 (2015) 16–28.
[18] H. Wu, H. Handroos, J. Kovanen, A. Rouvinen, P. Hannukainen, T. Saira, et al.,
Design of a parallel intersector weld/cut robot for machining processes in ITER Dr. Luis Uriarte is Deputy General Manager and Business Director at IK4-TEKNIKER.
vacuum vessel, Fusion Eng. Des. 69 (2003) 327–331. Industrial engineer, he holds a Phd. in Mechanical Engineering from the University
[19] P. Albertelli, M. Goletti, M. Monnoa, A new receptance coupling substructure of the Basque Country (2008). He is a regular contributor at CIRP with more than 10
analysis methodology to improve chatter free cutting conditions prediction, indexed papers.
Int J. Mach. Tools Manuf. 72 (2013) 16–24.
[20] T.L. Schmitz, Predicting high-speed machining dynamics by substructure
analysis, Ann. CIRP 49 (2000) 303–308.
[21] M.C.C. Bampton, J.R.R. Craig, Coupling of substructures for dynamic analyses,
AIAA J. 6 (1968) 1313–1319.
[22] O. Gonzalo, J. Beristain, H. Jauregi, C. Sanz, A method for the identification of Dr. Aitzol Lamikiz is associate professor (Manufacturing Technologies and Machine
the specific force coefficients for mechanistic milling simulation, Int J. Mach. Technologies) of the High Performance Machining Group at the Department of
Tools Manuf. 50 (2010) 765–774. Mechanical Engineering of the Faculty of Engineering of Bilbao (University of the
[23] T. Treib, E. Matthias, Error budgeting—applied to the calculation and optimi- Basque Country). Industrial engineer (1998) and PhD in Mechanical Engineering
zation of the volumetric error field of multiaxis systems, Ann. CIRP 36 (1) (2003) by the University of the Basque Country, he is co-author of 55 JCR-indexed
(1987) 365–368. papers and 3 books. Since March-2014 he is Director of External Relations of the
[24] A. Lamikiz, L.N. López de Lacalle, A. Celaya, Machine Tool Performance and Campus of Biscay of the University of the Basque Country (UPV/EHU).
Precision, Machine Tools for High Performance Machining (Chapter 6),