Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101

Contents lists available at ScienceDirect

Robotics and Computer-Integrated Manufacturing

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

Finishing of Fused Deposition Modeling parts by CNC machining

Alberto Boschetto, Luana Bottini n, Francesco Veniali
Department of Mechanical and Aerospace Engineering, University of Rome La Sapienza, Via Eudossiana 18, 00184 Rome, Italy

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

Article history: Fused Deposition Modeling is a filament extrusion-base Additive Manufacturing process that integrates
Received 14 October 2015 Computer Aided Design system, material science, Computer Numerical Control and the extrusion process
Received in revised form to fabricate physical parts without geometrical limitations. Notwithstanding the wide industrial diffu-
28 January 2016
sion, one of the most limiting aspects of this technology is the obtainable surface roughness. This lim-
Accepted 17 March 2016
itation implies that secondary finishing operations are necessary in order to comply with the design
requirements. Several efforts have been done in this field but a lack exists about Computer Numerical
Keywords: Control machining: the Fused Deposition Modeling mesostructure and the anisotropic surface mor-
Additive Manufacturing phology, which strongly depends upon the deposition angle, make very difficult the determination of the
Fused Deposition Modeling
cutting process parameters. The aim of this work is to develop a methodology able to unlock the pos-
Roughness improvement
sibility to finish Fused Deposition Modeling parts by Computer Numerical Control machining. A variable
Computer Numerical Control machining
cutting depth has been considered to avoid inner defects arising and to eliminate initial surface mor-
phology. An experimental campaign allowed to determine how cutting depth should be set as a function
of deposition angle. A particular virtual model offset permitted to generate in Computer Aided Manu-
facturing the machine code. A case study characterized by functional surfaces confirmed the applicability
of the method to complex geometry: a great reduction of average roughness and a reliable uniformity of
finished surfaces have been obtained.
& 2016 Published by Elsevier Ltd.

1. Introduction the STL file the better the quality of the model. The data is
transferred to the prototyping system and the process parameters
Additive Manufacturing (AM) is a host of technologies which are tailored within a Computer Aided Manufacturing (CAM) en-
build physical parts by adding material, layer by layer, directly vironment: the layer thickness, the part orientation, the interior
from a virtual model [1]. This offers advantages in many applica- filling method, the support generation methodologies are set. The
tions because parts can be fabricated with any geometric com- next step involves the slicing of the model into thin layers and the
plexity without the need for elaborate machine setup or final as- generation of the toolpath. Once the data have been sent to the
sembly. A variety of AM processes have been developed: they machine an autonomous building takes place layer by layer. After
differ in employed materials and in the manner of creating and the fabrication, the part is removed and post processed: the left-
depositing layers leading to different product quality, accuracy and over powder or resin or the support structures are eliminated.
build speed [2]. One of the most diffused AM technology is the Fused Deposi-
An AM process involves a number of stages from the virtual tion Modeling (FDM). The process fabrication consists in the de-
model creation to the fabrication of the physical component. The position of a thermoplastic polymer in form of filament. It is drawn
into a liquefier head and heated to a molten state. Then it is ex-
first step is the model creation typically in Computer Aided Design
truded through a nozzle and deposited on a plate one layer at a
(CAD) environment or by Reverse Engineering techniques. This
time by a Computer Numerical Control (CNC) system. The exiting
model is then converted into a file format that can be understood
material quickly solidifies since the air into the chamber is main-
by the AM machine software. Although the ISO 10303 promotes
tained to a temperature below the polymer melting point but
the International Standard for the Exchange of Product (STEP)
higher than room temperature in order to reduce residual stresses
model data, the informal standard tessellation language STL for-
[4]. A variety of polymers can be employed by the FDM process.
mat is the most employed in industry [3]. This format converts the
Current industrial machines use Acrylonitrile Butadiene Styrene
original model into a triangulated one; the higher the resolution of (ABS), nylon, polyethylene, polypropylene, polycarbonate and
other engineered thermoplastics: these polymers are character-
n
Corresponding author. ized by additives such as fillers, plasticizers, stabilizers and lu-
E-mail address: luana.bottini@uniroma1.it (L. Bottini). bricants. These materials do not cover the world of polymers and

http://dx.doi.org/10.1016/j.rcim.2016.03.004
0736-5845/& 2016 Published by Elsevier Ltd.
 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101 93

several choices must be performed starting from these limitations. networks were also employed to increase the complete dimen-
Other problems spring from thermo-mechanical aspects: the rapid sional accuracy. Software error compensation technique has been
cooling introduces distortions and stresses and the shrinkage is applied in [21] by considering correction of both the STL and slice
quite unpredictable. These aspects forced, in the past, the choice of files. Other experimentations based on Taguchi method [22,23]
a circular section for the nozzle. Therefore the part surface fin- and fuzzy modeling [24] highlighted how the accuracy can be
ishing is strongly affected by the rounded shape of the filament predicted by considering the extruding parameters and the part
and by the staircase effect [5]. Moreover other deterministic pro- orientation. This last factor is considered in the work of [25] which
blems arise from the deposition angle, i.e. the part surface slope provided a semiempirical model of the filament profile able to
with respect to the stratification direction, since the deposition predict the FDM dimensional deviation. This formulation demon-
conditions change. These occurrences force users to compensate strates that the deviations are only marginal affected by the ma-
for them by adjusting the process parameters through practical terial and by the prototyping system and depend upon the layer
experiences. This has direct impact on post processing cost and thickness and the deposition angle: the vertical walls are very
functionality of final part [6,7] because it limits both accuracy and close to the nominal dimension, while quite horizontal walls de-
surface finish obtainable by the technology. viate several tenths of millimeter.
Considerable works have been undertaken to investigate and Both the surface roughness weakness and the dimensional in-
improve the part quality. FDM process is characterized by a accuracy lead to the need of secondary finishing operations when
number of parameters; therefore researchers focused the attention the functional requirements are not satisfied.
on the estimation of the effect on quality output. Great effort has Galantucci et al. [26] performed a surface finishing by chemical
been paid to the surface roughness investigation. Anitha et al. [8] dipping based on the immersion in a dimethyl ketone -water so-
investigated the critical parameters affecting the surface finishing lution. The observations concluded that the chemical post treat-
by implementing Taguchi technique. It has been found that the ment led to an improvement in surface finishing between 2 mm
layer thickness is one of the most important factors. A semi-em- and 9 mm with a change in the prototype size under one tenth of
pirical model able to predict the surface roughness has been de- millimeter. In [27] a similar dimethyl ketone -based method has
veloped by Pandey et al. [9]; this model depends upon the layer been proposed: the aim was to seal the surfaces and make the
thickness and the build orientation and shows a good agreement FDM suitable for microdevices which need the hydrophilicity and
with experiments and literature in the range 10°–45° of build or- the biocompatibility. Chemical methods present the drawback that
ientation. A theoretical model has been proposed in [10] con- they need a deep knowledge of the materials and, at present, they
sidering the surface angle and the layer thickness to predict the have been applied only to ABS polymer. Industry and AM services
average roughness. The surface angle is calculated over each facet typically employ technologies less sensitive to the material such as
obtained by reading the STL model file. This approach is an ex- mass finishing operations. In [28] the barrel finishing operation
tension of the work of [11] and allows visualizing the part surface has been investigated because the main advantage to fabricate
finishing. Armillotta [12] investigated the capability of the FDM complex objects in a manageable and straightforward way is
technique to reproduce fine details and texturing of the parts. A maintained: the parts do not need to be clamped and process
model able to predict the geometrical profile of FDM fabrication parameters slightly depend upon the component shape com-
has been developed in [13]. A full factorial plan demonstrated the plexity. The experimental campaign showed that the material re-
capability to predict several parameters in a wide range of the moval rate is different as the deposition angle changes thus in-
deposition angle and for two layer thicknesses; in two critical fluencing abrading action and final quality results: attainable
zones, namely the near horizontal walls, other parameters, such as average roughness ranges from 3 mm to 7 mm.
the interior filling and the support structure type, interact with the Chemical and mass finishing have the benefit to condition the
main factors. An extension of this model has been provided in [14] entire part and to improve surface roughness without the need of
introducing the measuring direction as a factor and thus allowing a particular setup. On the other hand the obtained surface fin-
the prediction of the spacing and the hybrid parameters. The ishing and accuracy show a scattering due to FDM initial varia-
model reliability falls in the range of the deposition angle between bility. Moreover not all the part surfaces are required to be con-
25° and 155° and for measurement direction with measuring angle ditioned in order to comply with the specifications. Thus some
smaller than 80°. A refinement has been proposed by using neural efforts have been made in the direction of employing traditional
network approach making it effective all over the range of the machining to improve those zones with functional requirements.
deposition angle [15]. In the same work experiments have been Manufacturing constraints coming from subtractive and additive
undertaken to demonstrate that the use of different materials, processes have been considered in the design of the combined
machines, filling and support strategies does not affect the operation by Kebrat [29,30] highlighting how geometry complex-
agreement of the proposed formulation with the experimental ity takes a central role in the process planning. Part specific ana-
data. lysis must be performed in order to integrate so different opera-
As regard the dimensional accuracy many researchers have tions [31–33] but these approaches do not mention problems re-
carried out studies about the investigation of the FDM process lated to the toolpath generation and the part surface aligning. In
parameters which influence the part deviations and the attainable [32] a hybrid layered manufacturing machine combines arc
tolerances. Some authors have devoted efforts to the accuracy welding unit with a CNC milling machine: in this process the
benchmarking which consists of the design and the experi- shape of the object is firstly built using weld-deposition and the
mentation of test parts characterized by particular geometrical near-net shape is finished by machining subsequently. Also Kulk-
features in order to compare different technologies. Several stu- arni and Dutta [34] performed a study on the integration of milling
dies [3,16–19] found deviations ranging between 0.1 mm and with layer manufacturing focusing on the toolpath generation.
0.7 mm in disagreement with the Stratasys Ltd, the developer of They did not resolve the alignment between the cutter and the
the technology, which assures a part accuracy in x–y–z direction parts. In [35] the authors integrated five-axis machining and FDM
following the rule 70.127 mm or 7 0.0015 mm per mm which- reducing the difficulty for the cutter to align with the part fabri-
ever is greater. Sood et al. in [20] observed positive deviations of cated layer by layer: the results showed an improved efficiency
FDM and adopted Grey Taguchi's method; they gained optimum with respect to a low cost FDM machine but an uncertainty about
level of the process parameters to simultaneously minimize the the obtainable accuracy and a low surface finish have been de-
percentage change in length, width and thickness; artificial neural tected. Pandey et al. [9] developed a simple material removal
94 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101

Fig. 1. Macroscopic structure scheme of an FDM part (a); voids and aggregate voids (b, c); extruding aperture (d, e).

method, namely Hot Cutter Machining, to finish FDM prototypes. successfully attempted and rarely assists the industry and the AM
This operation showed a marked improvement of the surface services in complying with requirements for FDM products. This
quality but it was only applied to flat parts and the blade-like work strives to provide a methodology able to overcome these
cutter cannot be used on the freeform surfaces with convex or limits and allow the CNC finishing of an FDM part.
intricate elements.
The abovementioned literature works evidenced several pro-
blems arising from the CNC machining of FDM parts notwith- 2. Methodology
standing the machining of polymers is well known. The main
factors that influence the quality of a machined part are the As highlighted in the Introduction the machining of polymers is
temperature stability, the slide straightness, the tool design and well known and, generally, the tool manufacturers provide the
the machining conditions such as the rake angle, the tip radius, the process parameters for the roughing and the finishing of these
depth of cut, the feed rate and the cutting speed, the physical and materials. The cutting speed, the feed rate and the cutting depth
chemical properties of the stock material [36]. Milling of ther- are given in form of range assuring a good reproducibility of the
moplastics requires no special techniques and a normal cutter operation with predictable wear of the tool.
with a down milling is recommended to avoid burning. In [37] the In the case of FDM parts two issues must be considered. Firstly,
machinability of thermoplastic and thermosetting polymers has these engineering thermoplastic are modified by particular fillers
been investigated by means of the surface integrity, the chip for- and plasticizers in order to increase the mechanical properties and
mation and the machining forces. The results provided the ma- no data are present for their machining, thus the process para-
chining conditions that must be chosen to avoid material visco- meters must be found. The second issue is related to the macro-
plastic scaling/tearing and brittle cracking. Keresztes et al. found a scopic structure of a FDM part. It is composed by an intricate net of
correlation between the specific cutting resistance of several filaments which unavoidably presents voids and anisotropies. The
polymers and their yield strength [38]. This relation can ease the unidirectional problem has been investigated by [40] which
cutting forces calculations since the resistance values of the en- quantified the obtainable range of voids for different process
gineering polymers are practically not available. In fact only the conditions. The structure can be evaluated considering the sche-
mechanical properties are provided in material producers’ data matic of the deposited filament section reported in Fig. 1a. A single
sheets, together the instructions and some advices, in form of filament can be described by a cylinder with an elliptical cross
guides, about milling parameters. They include the material tool, section and the part surface is the envelope of the external fila-
the feed rate, the cutting speed, the rake angles. For the thermo- ments. The normal to the surface is inclined with respect to the
^
plastics the best surface finishing is obtained with sharp tools, direction of the stratification b by the so-called deposition angle α.
high speeds and slow feed rates [39]. Both the machining speed Since the section is curved [41] some voids take place between
and the feed rate should be uniform and uninterrupted. Cooling adjacent and stacked filaments. This is confirmed by the physical
allows higher cutting speeds but liquids could be avoided for those sections reported in Fig. 1b obtained by a metallographic cut. It can
polymers which can absorb moisture and swell. The tool must be be noticed that the voids are characterized by shape and dimen-
designed with adequate clearance for the chips in order to reduce sion variability: this is due to the distortion which takes place
problems with clogging and interference with the cutting opera- during the cooling, the stacking and the overlapping of the fila-
tion. This knowledge makes the CNC milling of polymer very in- ments which changes with the considered part geometry and or-
teresting for FDM prototypes but its implementation is prevented ientation. Somewhere the voids join in an aggregate as shown in
by several problems coming from the abovementioned weak- Fig. 1c. Another aspect which can be considered is the profile error
nesses of this AM technology: the filament section cannot be in- coming from the toolpath generation strategies, namely the ex-
terested by a severe reduction, otherwise a surface damaging oc- truding aperture [42]. This error takes place when the model fill-
curs; the deviations affecting the fabrication are of the order of ing path encounters the contour and a filament bending is re-
tenths of millimeter; surface quality strongly depends upon de- quired originating a void (Fig. 1d). Otherwise some wider voids
position angle. For these reasons CNC machining has not been occur when there is not enough space for a further path, as shown
 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101 95

Fig. 2. FDM roughness profiles related to the deposition angles 0° (a), 20° (b) and 90° (c).

in Fig. 1e. As a consequence of these problems, particular attention

must be paid in the choice of the cutting depth in order to avoid
that these internal defects come out the finished surface.
As regards the FDM part surface quality a marked irregularity is
observed: for the most diffused layer thickness of 0.254 mm, the
average roughness Ra varies between 16 mm and 60 mm depending
upon the deposition angle [13]. Hence, in order to obtain a regular
surface a adequate cutting depth is required. For example in the
case of 0° deposition angle (Fig. 2a) the profile is composed by only
interior filling with a high degree of periodicity: it is characterized
by a spacing pair to 0.75 mm and a peak to valley height of 180 mm,
highlighting a flattened filament section. It is well evident that this
surface needs a minimum cutting depth of 0.18 mm in order to
remove the initial surface morphology. For 20° deposition angle
the profile, shown in Fig. 2b, is less regular: this is due to the
presence of both the contour model and the interior filling. The
profile is characterized by a peak to valley height (about 400 mm)
bigger than the previous case, with a spacing of 0.82 mm, resulting
in an elongated filament section. In this case the minimal cutting
depth must be 0.4 mm. For a deposition angle pair to 90°, the fi-
laments are perfectly stacked over each other: the profile is
composed only by contour model spaced by the layer thickness
(0.254 mm). This case is characterized by the smallest total
roughness (100 mm) and a very low cutting depth is required. Fig. 3. Block diagram of the proposed methodology.
The abovementioned aspects underline that the cutting depth
has a limited range: it cannot exceed the upper limit related to the upon the deposition angle. The dimensional deviation has been
underlying voids and to the internal defects, and at the same time modeled as a function of the deposition angle [25]:
it cannot be less than the total roughness in order to delete the ∆h (α ) ≅ 0. 19 Cos2 (α ) (1)
initial morphology. Since all these issues depend upon the de-
position angle, the cutting depth must be a function of this process This expression allows determining the actual position of a
parameter. This function will be developed in the next Section. surface produced by the FDM. This way it is possible to provide to
The proposed methodology is described in the block diagram of the CNC CAM the actual geometry as the starting stock.
Fig. 3. The virtual model used in this analysis is an STL file: it is
The virtual model, employed to fabricate the FDM physical composed by triangular facets which envelope a closed volume.
prototype, is analyzed to calculate the dimensional deviation from For each facet, the deposition angle can be calculated according to
the nominal value and to evaluate the cutting depth depending the following equation:
96 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101

^ The obtained parts have been machined by a 3 axis CNC milling

αj = Arc cos (n^j ⋅b ) (2)
system equipped with an 24,000 rpm ELTE spindle. A Gühring two
^
where n^j is the normal of the generic facet j and the versor b is the flute solid carbide ball-end mill 6 mm in diameter has been used.
stratification direction. This angle is necessary to determine the According to the manufacturer and calculating the engaged ef-
fective diameter, the cutting speed ranged between 2.6 and 6 m/s
dimensional deviation according to Eq. (1) and the cutting depth,
within the optimum range for the thermoplastic material. The feed
which will be evaluated in the next Section. The final model can be
and the cross feed have been set at 0.05 mm/round and 0.25 mm
obtained by subtracting the cutting depth to the physical proto-
respectively. Ten cutting depths, ranging from 0.05 mm to 0.5 mm
type which has dimensional values pair to the nominal value plus
with step of 0.05 mm, have been investigated.
the deviation introduced by the technology. It corresponds to
A Taylor Hobson Form Talyprofile Plus with a 120 mm pro-
apply to each triangle an outward offset by ∆h (α ) and an inward
longed stylus has been used for the profile measurements. The
offset by the cutting depth p (α ) as following:
vertical range and the declared accuracy are 2200 μm and 16 nm
T offset =Tj+∆h ( αj )−p ( αj ) respectively. The roughness profiles and parameters have been
j (3)
obtained according to [44] choosing appropriate exploratory
Operatively this is possible employing the methodology de- length and cut-off [45] and filtered by a corrected Gaussian filter
veloped in [43]. This method operates an offset of the triangular [46]. Each specimen has been characterized by six measurements
mesh with a variable radius without changing the interconnec- covering the entire machined surface in order to describe the
tions and generating a modified model in a stable and robust way. uniformity by means of a statistical approach.
In fact no intersection and gap between two or more neighboring
triangles can occur.
The obtained model is imported in CAM environment and used 4. Results
to generate the toolpath for a finishing operation according to
most widely employed strategies such as parallel finishing, hor- In this Section the results of the experimentation are presented.
izontal finishing, scallop finishing, radial finishing, spiral finishing. The most critical parameter, i.e. the cutting depth, has been locally
After the generation of the toolpath, the machine code can be identified taking into account the surface integrity and the surface
transferred to the CNC system. The physical part can be aligned finishing.
considering the dimensions of the model obtained by the outward In Fig. 5a a 3D map of a partially machined FDM surface,
offset by ∆h (α ) and the machining can take place. characterized by 60° deposition angle, is reported. In this case the
chosen cutting depth is 0.1 mm: a marked reduction in peaks
height is observed but the original valleys are not eliminated, as
confirmed by the profile of Fig. 5b. In this case the initial average
3. Experimentation
roughness is 17 mm and, after the finishing, it becomes 2.5 mm: a
little increment in cutting depth improves this result. The same
In order to consider all the above mentioned observations, the
cutting parameters have been employed to machine a surface
methodology needs the definition of the cutting depth through a
characterized by 9° deposition angle: since this surface shows a
function of the deposition angle. An experimentation designed for
spacing and peak to valley height greater than the previous case,
the purpose has been carried out. The specimen geometry is a
the chosen cutting depth leaves more pronounced valleys as
square parallelepiped 20  20  10 mm3.
highlighted in Fig. 5c. The profile reported in Fig. 5d is character-
They have been fabricated with seven different inclinations in
ized by two unmachined zones 50 mm in Ra. The finishing opera-
order to obtain surfaces characterized by the deposition angles 0°, tion lowers this value to 20 mm highlighting that a marked incre-
3°, 9°, 15°, 30°, 60°, 90°. ment in cutting depth is needed. These simple examples confirm
For the fabrication a Stratasys Dimension Bst 768 has been used that particular attention must be paid in the choice of the cutting
employing ABS-P400 material. The chosen fabrication parameters depth.
have been: 0.254 mm layer thickness, solid model filling, break- In order to locally identify this parameter, the experimentation
away support generation method. In Fig. 4 the fabricated speci- described in the previous section has been performed. The results
mens are shown. for different deposition angles are summarized in Figs. 6 and 7.
For each cutting depth six profilometer measurements have
been carried out; the corresponding average roughness values
have been graphed by a box and whiskers plot diagram. A mac-
rography of the skin section, i.e. the section of the prototype ex-
ternal filaments, has been added for each deposition angle in order
to measure the skin thickness.
In Fig. 6 the case of 90° deposition angle is analyzed. It is well
evident that for low values of cutting depth (0.05 mm and 0.1 mm)
the average roughness is very high because the original mor-
phology has been not completely deleted. For example, the profile
machined with a cutting depth of 0.05 mm reported in Fig. 6b
shows that the valleys are not conditioned by the operation. At
0.15 mm cutting depth the Ra is very low (about 1 mm): in this
case, as evidenced by the profile (Fig. 6b), no morphology referable
to the original surface is present since this cutting depth is higher
than the peak to valley height. When the cutting depth increases
(less than 0.35 mm) the Ra maintains below 1.4 mm and the cutting
depth is smaller than the skin thickness (shown in Fig. 6a) pair to
0.47 mm. For higher values the skin is completely cut away, a
Fig. 4. Fabricated specimens. damaged surface is expected and the abovementioned internal
 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101 97

Fig. 5. 3D maps and profiles of machined surfaces related to 60° (a, b) and 9° (c, d) deposition angles.

voids and defects could be uncovered leading to an increasing of The minimum Ra is found at 0.30 mm which is the closest value
the average roughness. An example of this behavior is shown by above the peak to valley height, as expected.
the profile related to 0.45 mm cutting depth reported in Fig. 6b. The surface inclined by 9° deposition angle is analyzed in
By repeating this analysis for all the considered deposition Fig. 7c. Similarly to the previous case at 0.05 mm cutting depth, a
angles, the graphs of Fig. 7 have been obtained. In the case of very high Ra (35.5 mm) is observed with a relatively small scat-
deposition angle pair to 0°, the skin thickness is measured along tering (1.05 mm) and at 0.10 mm the Ra lowers to 20 mm with a
the stratification direction and thus it is equal to 0.254 mm, i.e. the standard deviation of 5.65 mm. This is due to the peak to valley
layer thickness. A cutting depth pair to 0.05 mm is not enough to height (220 mm) very far from the set cutting depth. As expected,
eliminate the original morphology since the peak to valley height the Ra lowers below 2 mm at 0.25 mm. In this condition a very
is bigger. If 0.10 mm cutting depth is set, the roughness is under small scattering is observed: the standard deviation is reduced to
1 mm and tends to increase with the cutting depth. When this 0.08 mm.
machining parameter is set at 0.25 mm, the Ra markedly worsens Also in the case of 15° deposition angle a high value of the
since the skin is completely cut away. At greater cutting depth the cutting depth is needed because the peak to valley height is re-
behavior starts again with a low value of Ra. markable (about 250 mm). At 0.05 mm and 0.1 mm cutting depth
In the case of 3° deposition angle, the staircase effect is re- the Ra is above 20 mm while at 0.15 mm drastically lowers to
markable and the peak to valley height is about 270 mm. A cutting 5.97 mm but the standard deviation is still greater than 2 mm. The
depth, greater than this value, is necessary to delete the original minimum (1.12 7 0.21 mm) is reached at 0.35 mm cutting depth
morphology. At 0.05 mm cutting depth the Ra is very high: the but a reliable result can be achieved at 0.30 mm (1.687 0.06 mm).
mean value is 16.9 mm with a standard deviation of 0.57 mm which At 0.40 mm the machined surfaces show some defects which
is few less than the original surface. At 0.10 mm a 28% reduction in make the Ra to increase with a large scattering.
Ra and a greater scattering is observed. As the cutting depth in- When the deposition angle is 30° the profile roughness be-
creases the Ra decreases and, at 0.20 mm, it lowers below 2 mm. comes more regular and the original morphology is deleted

Fig. 6. Box and whiskers plot diagrams for different cutting depths and skin thickness macrography of a surface characterized by 90° deposition angle (a); some machined
profiles (b).
98 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101

Fig. 7. Box and whiskers plot diagrams for the investigated cutting depths and skin thickness macrographys of different deposition angles.

already at 0.15 mm cutting depth. After this value the Ra tends to constant low cutting depth the obtainable surface roughness is
increase: since the skin thickness is 0.4 mm, the worse result is coarse for every deposition angle. By imposing a higher cutting
reached at cutting depth around this value. depth, e.g. 0.20 mm, it is possible to obtain good surface roughness
The specimen fabricated with 60° deposition angle shows a only for deposition angles below 5° and greater than 30°. For a
very low initial roughness: by setting the cutting depth at cutting depth about 0.30 mm it is possible to have the most of
0.05 mm about 4 mm Ra can be obtained. A very good result has results less than 2 mm, but it implies the removal of entire layers
been achieved at 0.15 mm cutting depth: the average roughness for low deposition angle surfaces. The necessity to consider low
shows a minimum value of 0.97 mm with a standard deviation of cutting depths emerges from this observation. In order to solve
0.022 mm. This surface is characterized by a skin thickness of this scope a multi-objective optimization has been employed.
0.44 mm: a big scattering of data is observed for a cutting depth For each deposition angle a function Raα (p) which relates the
set nearby this value. obtainable roughness to the cutting depth p has been found by a
polynomial interpolation fit. The other objective function is the
cutting depth, thus it is the identity function. This optimization
5. The variable cutting depth modeling can be described in mathematical terms as follows:

All the achieved data have been structured in order to obtain ⎧ R aα (p)
⎪
the contour plot of the Ra as function of the cutting depth and the min ⎨ p ∀ α ∈ [0°, 90°]
deposition angle reported in Fig. 8. This graph permits to navigate ⎪
⎩ 0.05 ≤ p ≤ 0.5 (4)
the map of the attainable average roughness on a surface char-
acterized by a specific deposition angle when it is machined with a By using the scalarization technique, this problem can be
particular cutting depth. It is well evident that, by setting a solved by combining its multiple objectives in one single objective
 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101 99

areas employ efficient sources by cheap and effective micro-hy-

dropower plants [48,49]. As a consequence an increased effort is
recently addressed to the design of such small turbines with the
aim of the efficiency and cheapness. This way in [50] FDM tech-
nique has been employed to have two advantages: design defi-
ciencies and failures can be identified at early stages; water ex-
periment validation can be provided by using a fast and cheap way
with respect to traditional waterwheel fabrication.
The bucket is double hemispherical in shape and typically it is
detachable from the wheel. Since the water strikes the bucket in
the center and flows out at both sides making a U turn, the surface
inside the bucket is polished and smooth to reduce hydraulic
losses [51]. For this reason the finishing has been performed on
this surface characterized by all the deposition angles belonging to
the range 0–90°. In order to operate a comparison before and after
the machining only half bucket has been finished.
The steps of the methodology application are reported in Fig. 9.
The bucket prototype has been fabricated by a Stratasys Dimension
Bst 768 using the ABS P-400 and choosing the same process
parameters of the previous experimentation (0.254 mm layer
thickness, solid model filling, break-away support generation
Fig. 8. Contour plot of the average roughness as function of the cutting depth and
method). The prototype is not detached from the fabrication plate
the deposition angle and the found cutting depth function (dashed black line).
and positioned onto the milling machine table. The part has been
aligned considering the information provided by Eq. (3). The
scalar function. In detail this approach minimizes a positively
milling tool and the chosen cutting parameters are the same re-
weighted convex sum of the objectives as follows:
ported in Section 3. The employed CAM software is MecSoft Rhi-
min ⎡⎣ w1R aα (p) + w2 p⎤⎦ ∀ α ∈ [0, 90] noCAM 3.0.0.56: the virtual model of the finished part used to plan
2 the machining has been obtained applying the offset methodology
∑ wi = 1 described in Section 2.
i=1 For the validation some roughness measurements have been
wi ≥ 0 i = 1, 2 carried out in the zones indicated by capital letters in Fig. 10. These
0.05 ≤ p ≤ 0.5 (5) zones have been chosen in order to investigate the critical con-
ditions related to both the FDM fabrication and the chipping op-
The solutions found by Eq. (5) have been fitted by a rational eration. The zone A is characterized by a slight curvature with a
interpolation. This type of model has been chosen because it has mean deposition angle pair to 0°. The initial roughness is more
the potential to take a wide range of shapes and it is smoother and than 20 mm, as expected [15], and it is lowered to 1.83 mm. The
less oscillatory than polynomial models even if occasional roughness profile of the machined surface shows a periodicity
asymptotes can occur. The obtained rational polynomial is: with the same spacing of the original one but with a decimated
12785+27447α−1670. 2α 2+30. 12α 3−0. 0905α 4 amplitude. The original surface related to the zone B presents a
p (α )= marked staircase effect, thus the average roughness is the worst.
1282. 9+897. 72α−57. 681α 2 + α 3 (6)
The roughness obtained by the machining is below 2 mm, in ac-
It is represented in Fig. 8 by the dashed line. cordance with the expected value. As the deposition angle in-
creases (zone C) the initial profile becomes more regular with a
reduced average roughness. Also in this case the finishing opera-
6. Case study tion leads to a very good finishing result. The zone D is char-
acterized by a marked curvature and a deposition angle of 42°. The
The proposed methodology has been applied to a case study machined profile shows some small defects with high spatial
characterized by a complex geometry. The chosen component is frequency notwithstanding the resulting Ra is low. The measure
the bucket of a Pelton wheel. It is an impulse hydraulic turbine related to the zone E is taken nearby the exit border of the bucket.
commonly used as high and medium heads and, recently it has This zone has a deposition angle pair to 67°: the correspondent
been applied for lower head micro-sites becoming an accepted initial roughness is at lowest level for this layer thickness [15]. In
alternative practice in many countries [47]. Moreover most rural this case the machining leads to a finished surface characterized

Fig. 9. Block diagram of the developed methodology applied to the case study.
100 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101

Fig. 10. Prototype measurement zones and relating roughness profiles before and after the finishing operation.

by irregular peaks and valleys with an average roughness near to the mesostructure and the surfaces morphology of FDM parts. The
2 mm. The last investigated zone is the central edge of the bucket: former depends upon the deposition process and the geometry of
this surface is flat and its deposition angle is 78°. Differently from the deposited filament: it is anisotropic in terms of mechanical
the previous cases, this element is a thin wall with problems that properties and presents some internal voids and defects. The latter
concerns both the FDM fabrication and the cutting forces. The is strongly affected by a FDM fabrication parameter, namely the
initial surface is characterized by the best quality but it is affected deposition angle. It is well known that the surface morphology
by some geometrical problems [25]. However the machined pro- deeply changes with its slope: this requires different machining
file is not markedly affected by such a causes. conditions for a prototype characterized by surfaces with different
slopes. This preliminary analysis has highlighted the necessity of an
appropriate choice of the cutting depth. For this reason an experi-
7. Conclusions mental campaign has been designed for the purpose to determine
the cutting depth as a function of the deposition angle. The ob-
In this work a methodology addressed to the CNC finishing of tained formulation has been integrated in a methodology that,
FDM parts has been developed. The lack in literature of similar starting from the original STL file employed for the FDM fabrication,
methods derives from to the difficulty to determine the correct permits to obtain the modified virtual model providing the CNC
machining parameters both for the obtainable surface quality and CAM with a variable cutting depth. The method has been applied to
its integrity. Preliminary observations have been conducted about some surfaces of a functional prototype leading to very good results.
 A. Boschetto et al. / Robotics and Computer-Integrated Manufacturing 41 (2016) 92–101 101

Acknowledgment accuracy in fused deposition modelling: a fuzzy logic approach, Int. J. Prod.
Qual. Manag. 7 (1) (2001) 22–43.
[25] A. Boschetto, L. Bottini, Accuracy prediction in fused deposition modeling, Int.
The authors wish to thanks Eng. Matteo La Barbera for its J. Adv. Manuf. Technol. 73 (2014) 913–928.
precious help in experimental campaign. [26] L.M. Galantucci, F. Lavecchia, G. Percoco, Experimental study aiming to en-
hance the surface finish of fused deposition modeled parts, CIRP Ann. Manuf.
Technol. 58 (1) (2009) 189–192.
[27] E.J. McCullough, V.K. Yadavalli, Surface modification of fused deposition
References modelling ABS to enable rapid prototyping of biomedical microdevices, J.
Mater. Process. Technol. 213 (2013) 947–954.
[28] A. Boschetto, L. Bottini, Roughness prediction in coupled operations of fused
[1] I. Gibson, D.W. Rosen, B. Stucker, Additive Manufacturing Technologies, Rapid
deposition modeling and barrel finishing, J. Mater. Process. Technol. 219
Prototyping to Direct Digital Manufacturing, Springer, New York, 2010.
[2] N. Hopkinson, R. Hague, P. Dickens, Rapid Manufacturing: An Industrial Re- (2015) 181–192.
volution for the Digital Age, John Wiley & Sons, Chichester, 2005. [29] O. Kerbrat, P. Mognol, J.Y. Hascoët, Manufacturability analysis to combine
[3] C.K. Chua, K.F. Leong, C.S. Lim, Rapid Prototyping: Principles and Applications, additive and subtractive processes, Rapid Prototyp. J 16 (1) (2010) 63–72.
World Scientific, River Edge, 2010. [30] O. Kerbrat, P. Mognol, J.Y. Hascoët, A new DFM approach to combine ma-
[4] S.H. Masood, Advances in fused deposition modeling, in: S. Hashmi (Ed.), chining and additive manufacturing, Comput. Ind. 62 (7) (2011) 684–692.
Comprehensive Materials Processing, Elsevier, Amsterdam, 2014, pp. 69–91. [31] Z. Zhu, V. Dhokia, S.T. Newman, The development of a novel process planning
[5] A. Boschetto, L. Bottini, Surface characterization of fused deposition modeling, algorithm for an unconstrained hybrid manufacturing process, J. Manuf. Pro-
in: L. Santo, J.P. Davim (Eds.), Surface Engineering Techniques and Applica- cess. 15 (4) (2013) 404–413.
tions: Research Advancements, IGI Global, Hershey, 2014, pp. 249–280. [32] K.P. Karunakaran, S. Suryakumar, V. Pushpa, Akula Sreenathbabu, Low cost
[6] B. Vasudevarao, D.P. Natarajan, M. Henderson, A. Razdan. Sensitivity of RP integration of additive and subtractive processes for hybrid layered manu-
surface finish to process parameter variation, in: Proceedings of the Solid Free facturing, Robot. Comput. Integr. Manuf. 26 (5) (2010) 490–499.
Form Fabrication Symposium, Austin, 2000, p. 252–258. [33] J. Hur, K. Lee, Zhu-hu, J. Kim, Hybrid rapid prototyping system using ma-
[7] D.K. Ahn, H. Kim, S. Lee, Fabrication direction optimization to minimize post- chining and deposition, Comput. Aided Des. 34 (2002) 741–754.
machining in layered manufacturing, Int. J Mach. Tools Manuf. 47 (3–4) (2007) [34] P. Kulkarni, D. Dutta, On the integration of layered manufacturing and material
593–606. removal processes, J. Manuf. Sci. E-T ASME 122 (2000) 100–108.
[8] R. Anitha, S. Arunachalam, P. Radhakrishnan, Critical parameters influencing [35] W.C. Lee, C.C. Wei, S.C. Chung, Development of a hybrid rapid prototyping
the quality of prototypes in fused deposition modelling, J. Mater. Process. system using low-cost fused deposition modeling and five-axis machining, J.
Technol. 118 (2001) 385–388. Mater. Process. Technol. 214 (11) (2014) 2366–2374.
[9] P.M. Pandey, N.V. Reddy, S.G. Dhande, Improvement of surface finish by [36] M. Alauddin, I.A. Choudhury, M.A. El Baradie, M.S.J. Hashmi, Plastics and their
staircase machining in fused deposition modelling, J Mater. Process. Technol. 1 machining: a review, J. Mater. Process. Technol. 54 (1995) 40–46.
(1-3) (2003) 323–331. [37] K.Q. Xiao, L.C. Zhang, The role of viscous deformation in the machining of
[10] D. Ahn, H. Kim, S. Lee, Surface roughness prediction using measured data and polymers, Int. J. Mech. Sci. 44 (2002) 2317–2336.
interpolation in layered manufacturing, J. Mater. Process. Technol. 209 (2009) [38] R. Keresztes, G. Kalácska, L. Zsidai, Z. Dobrocsi, Machinability of engineering
664–671. polymers, Sustain. Constr. Des. (2011) 106–114.
[11] R.I. Campbell, M. Martorelli, H.S. Lee, Surface roughness visualization for rapid [39] Machining guidelines for semi-finished plastics, 〈www.ensinger-online.com〉,
prototyping models, Comput. Aided Des. 34 (2002) 717–725. 2010.
[12] A. Armillotta, Assessment of surface quality on textured FDM prototypes, [40] J.F. Rodriguez, J.P. Thomas, J.E. Renaud, Characterization of the mesostructure
Rapid Prototyp. J. 12 (1) (2006) 35–41. of fused-deposition acrylonitrile-butadiene styrene materials, Rapid Prototyp.
[13] A. Boschetto, V. Giordano, F. Veniali, Modelling micro geometrical profile in J. 6 (3) (2000) 175–186.
fused deposition process, Int. J. Adv. Manuf. Technol. 61 (9–12) (2011) [41] D. Ahn, J.H. Kweon, S. Kwon, J. Song, S. Lee, Representation of surface
945–956. roughness in fused deposition modeling, J. Mater. Process. Technol. 209 (2009)
[14] A. Boschetto, V. Giordano, F. Veniali, 3D roughness profile model in fused 5593–5600.
deposition modelling, Rapid Prototyp. J. 19 (4) (2013) 240–252. [42] D.Y. Chang, B.H. Huang, Studies on profile error and extruding aperture for the
[15] A. Boschetto, V. Giordano, F. Veniali, Surface roughness prediction in fused
RP parts using the fused deposition modeling process, Int. J. Adv. Manuf.
deposition modelling by neural networks, Int. J. Adv. Manuf. Technol. 67 (9–
Technol. 53 (2011) 1027–1037.
12) (2013) 2727–2742.
[43] A. Boschetto, L. Bottini, Triangular mesh offset aiming to enhance Fused De-
[16] R. Ippolito, L. Iuliano, A. Gatto, A benchmarking of rapid prototyping techni-
position Modeling accuracy, Int. J. Adv. Manuf. Technol. 80 (2015) 99–111.
ques in terms of dimensional accuracy and surface finish, CIRP Ann. 44 (1)
[44] ISO 4287, Geometrical product specifications (GPS) – surface texture: profile
(1995) 157–160.
method—terms, definitions and surface texture parameters, Technical Report
[17] M. Mahesh, Y.S. Wong, J.Y.H. Fuh, H.T. Loh, Benchmarking for comparative
ISO, 1997, p. 1–25.
evaluation of RP systems and processes, Rapid Prototyp. J. 10 (2) (2004)
[45] ISO 4288, Geometrical Product Specifications (GPS) – Surface texture: Profile
123–135.
[18] I. El-Katatny, S.H. Masood, Y.S. Morsi, Error analysis of FDM fabricated medical method – Rules and procedures for the assessment of surface texture, Tech-
replicas, Rapid Prototyp. J. 16 (1) (2010) 36–43. nical Report ISO, 1997, p. 1–8.
[19] R. Singh, Some investigations for small-sized product fabrication with FDM for [46] ISO 16610-21, Geometrical product specifications (GPS) – Filtration – Part 21:
plastic components, Rapid Prototyp. .J 19 (1) (2013) 58–63. Linear profile filters: Gaussian filters, Technical Report ISO, 2011, p. 1–23.
[20] A.K. Sood, R.K. Ohdar, S.S. Mahapatra, Improving dimensional accuracy of [47] A.R. Wallace, H.W. Whittington, Performance Prediction of Standardized Im-
fused deposition modelling processed part using grey Taguchi method, Mater. pulse Turbines for Micro Hydro, Elsevier B.V. Sutton., International. Water
Des. 30 (10) (2009) 4243–4252. Power & Dam Construction, U.K., 2008.
[21] K. Tong, S. Joshi, E. Lehtihet, Amine error compensation for fused deposition [48] A.H. Elbatran, O.B. Yaakob, Y.M. Ahmed, H.M. Shabara, Operation, performance
modeling (FDM) machine by correcting slice files, Rapid Prototyp. J. 14 (1) and economic analysis of low head micro-hydropower turbines for rural and
(2008) 4–14. remote areas: a review, Renew. Sustain. Energy Rev. 43 (2015) 40–50.
[22] Y.R. Kumar, C.S.P. Rao, T.A.J. Reddy, A robust process optimisation for fused [49] L. Barelli, L. Liucci, A. Ottaviano, D. Valigi, Mini-hydro: a design approach in
deposition modelling, Int. J. Manuf. Technol. Manag. 14 (1–2) (2008) 228–245. case of torrential rivers, Energy 58 (2013) 695–706.
[23] D.Y. Chang, B.H. Huang, Studies on profile error and extruding aperture for the [50] M. Takagi, Y. Watanabe, S. Ikematsu, T. Hayashi, T. Fujimoto, Y. Shimatani,
RP parts using the fused deposition modeling process, Int. J. Adv. Manuf. Energy Procedia 61 (2014) 1593–1596.
Technol. 53 (9–12) (2011) 1027–1037. [51] Z. Hussian, M. Abdullah, Z. Alimuddin, Pelton Turbine, in: Basic Fluid Me-
[24] A. Equbal, A.K. Sood, R.K. Ohdar, S.S. Mahapatra, Prediction of dimensional chanics and Hydraulic Machines, BS Publication, Hyderabad, 2009, pp. 99–126.