Sunteți pe pagina 1din 11

Additive Manufacturing 12 (2016) 334–344

Contents lists available at ScienceDirect

Additive Manufacturing
journal homepage: www.elsevier.com/locate/addma

Full length article

Integration of FDM surface quality modeling with process design


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

a r t i c l e i n f o a b s t r a c t

Article history: Fused Deposition Modeling is one of the earliest types of Additive Manufacturing technologies. In this
Received 14 December 2015 process a physical object is fabricated directly from a Computer-Aided Design model using layer-by-
Received in revised form 15 March 2016 layer extrusion of a feedstock plastic filament material through a nozzle. It was originally developed for
Accepted 8 May 2016
design and functional prototype applications but, in the last decade, gained considerable recognition and
Available online 4 June 2016
adoption in industry due to the process simplicity, affordability, and ability to make parts in a range
of common engineering thermoplastics. Thus Fused Deposition Modeling products must comply with
Keywords:
tolerance and roughness requirements in order to satisfy a mechanical coupling or functionality. This
Additive manufacturing
Fused deposition modeling
work strives to develop an integrated methodology able to help the process design with the aim to goal
Roughness prediction the requirements. This method allows knowing in advance the surface quality of the Fused Deposition
Accuracy prediction Modeling product at chosen process parameters. This way an ex ante product check is possible. Moreover
the methodology can find the set of solutions when the requirements are prescribed on the product
features: it is possible to define the technology capability and analyze the proposed fabrication conditions.
This methodology is highly useful in process management also in conjunction with other operations and
can help the computer aided process planning since it can predict admissible scenarios.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction (ABS). Initially FDM were employed to build physical prototype for
design and verification of new products. In the last two decades it
Additive Manufacturing (AM) is a host of technologies in which moves from prototyping aim to end-product fabrication in a wide
the part is fabricated layer by layer with addition of material as range of applications in low volume or customized products becom-
opposed to traditional removal and deformation processes. Since ing one of the most widely used AM process in industry. A number of
the first introduction in late 1980s, the AM processes gained con- applications regard the automotive, the aerospace, the biomedical
siderable recognition and adoption in industry because they do not engineering, the tooling and casting foundry [2–4].
need traditional tooling or computer numerical control program- Notwithstanding the great diffusion, the FDM presents limi-
ming. The prototypes can be fabricated rapidly as compared with tations in accuracy and surface finishing of the fabricated parts.
the time needed by machining or molding. Among the AM technolo- A typical drawback lies in the staircase effect which markedly
gies Fused Deposition Modeling (FDM) is one of the most accepted affects this technology as in the employed thick filament: at present
processes. It fabricates parts by creating layers through a deposition 0.127 mm is the smallest layer thickness (0.254 mm is the most dif-
of a bead of semi-molten material: a feedstock plastic filament is fed fuse) higher than other AM processes. Several problems are related
into a heating head, extruded through a nozzle and deposited onto to thermo-mechanical aspects: since the polymeric material is
a plate or onto a completed layer. Both model and support materi- rapidly cooled, part distortions and residual stresses are introduced.
als are processed for parts with overhanging zones with respect to The strinkage is quite unpredictable thus the process compensation
the stratification direction [1]. by adjusting parameters is needed but the section and the position
The diffusion of this technology is due to the process simplic- of the deposited filament are affected by local conditions due to
ity, affordability, and ability to make parts in a range of common the specific part geometry. Nevertheless the most limiting prob-
engineering thermoplastics such as Acrylonitrile Butadiene Styrene lem is the lack in industry of an accurate prediction of the surface
output both in term of roughness and accuracy. In literature some
attempts have been done in order to solve this problem. Pandey
et al. [5] has been developed a semi-empirical model able to predict
∗ Corresponding author. the surface roughness. This model depends upon the layer thick-
E-mail addresses: alberto.boschetto@uniroma1.it (A. Boschetto), ness and the build orientation and shows a good agreement with
luana.bottini@uniroma1.it (L. Bottini), francesco.veniali@uniroma1.it (F. Veniali).

http://dx.doi.org/10.1016/j.addma.2016.05.008
2214-8604/© 2016 Elsevier B.V. All rights reserved.
A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344 335

method [17,18] and fuzzy modeling [19] highlighted how the accu-
Nomenclature racy can be predicted by considering the extruding parameters and
the part orientation. For industrial prototyping systems the extrud-
r radius of the theoretical FDM profile peak ing parameters are rarely modified and depend upon the material,
f spacing of the theoretical FDM profile while the part orientation modifies the deposition angles of the part
L layer thickness surfaces. This factor is considered in the work of [20] which pro-
b̂ stratification direction vided a semi-empirical model of the filament profile able to predict
n̂ surface normal the FDM dimensional deviation. This formulation demonstrates
n̂i normal of the generic i-th triangle that the deviations are only marginal affected by the material and
␣ deposition angle by the prototyping system and depend upon the layer thickness and
˛i deposition angle of the generic i-th triangle the deposition angle: the choice of these factors is critical during
Ra average roughness the FDM process design to obtain a specific part quality. A comple-
h dimensional deviation from the nominal value tion of this model has been proposed in [21] in order to implement
 rotated x coordinate this formulation in the STL processing.
 rotated y coordinate The need to predict in advance how the obtainable quality varies
 rotated z coordinate according to the process parameters change is important to estab-
ϕx rotation angle around the axis x lish if the technology permits to comply with the requirements and
ϕy rotation angle around the axis y which parameters levels allow to obtain the desired surface quality.
(i)
Pk vertex of the i-th triangle for k = 1, 2, 3 Since the FDM fabricates geometries without complexity limita-
tion, the part orientation at a specific layer thickness is a hard task
for a single purpose (roughness or dimensional deviation require-
ments) and becomes much more difficult for multiple specifications
experiments and literature in the range 10◦ –45◦ of build orienta- over the part surfaces. The aim of this work is to provide a complete
tion. In [6] a theoretical model considering the surface angle and the description of the attainable surface part quality when specific pro-
layer thickness has been proposed in order to predict the average cess parameters have been set and to develop a methodology able to
roughness. The surface angle has been calculated over each facet identify the domain of possible solutions: this domain is the basis in
obtained by reading the STL model file. A model able to predict the which operates further choices without losing the complying with
geometrical profile of FDM fabrication has been developed in [7]. the requirements.
A full factorial plan demonstrated the capability to predict several
parameters in a wide range of the deposition angle and for two layer
thicknesses. An extension of this model has been provided in [8] 2. Methodology and tool development
introducing the measuring direction as a variable and allowing the
prediction of the spacing and the hybrid roughness parameters. The The proposed methodology is described by the flow diagram
model reliability falls in the range of the deposition angle between reported in Fig. 1. Firstly the virtual model is handled and pro-
25◦ and 155◦ and for measurement direction with measuring angle cessed in order to organize data related to facets information. Then
smaller than 80◦ . In order to overcome this limitation a refinement a crossroads allows opting for a direct or indirect method. In the
has been introduced in [9]: by using neural network approach a former the process parameters can be chosen and the correspond-
new model depending upon the layer thickness and the deposi- ing prediction of the quality results are provided in form of 3D
tion angle has been found; it is effective all over the range of the mapping superimposed to virtual model: a manual check of the
deposition angle. In the same work experiments have been carried requirements can help to decide about the process planning in
out to demonstrate that the use of different materials, machines, case of positive or negative results. In the latter method the surface
filling and support strategies does not affect the agreement of the requirements, by means of tolerances and minimum roughness for
proposed formulation with the experimental data. This ultimate the selected surfaces, are provided in input and a subsequent anal-
model provides a prediction within few percent prediction errors ysis allows to generate the domain of process parameters solution
and it is ready for industrial process implementation. to this specific problem. If the domain is empty the process plan
As regards the dimensional accuracy many studies have been should be revised, else it is possible to choose a couple of process
conducted in order to investigate the FDM process parameters parameters levels within the not-empty domain and display the
which influence the deviation from the nominal value. Some resulting 3D representations of this possible solution by following
authors focused their attention on benchmarking studies: they the direct method.
consist in the design and the experimentation of test parts char-
acterized by specific geometrical features in order to compare
different AM technologies. With regards to the FDM process some 2.1. Roughness and dimensional deviation models
benchmarking studies [10–14] found deviations from the nomi-
nal value ranging between 0.1 mm and 0.7 mm in disagreement As indicated in the Introduction this methodology employs two
with the Stratasys Ltd (the company developer of the technology) models with the aim to predict the surface quality obtainable by a
which assures a part accuracy in x-y-z direction following the rule particular FDM parameters set. These formulations originate from
±0.127 mm or ±0.0015 mm per mm whichever is greater. In [15] a a deposited filament model described in [7]. According to this
methodology for improving the accuracy of the distance between approach the profile of a FDM part surface is composed by a series
parallel faces concerning FDM manufactured prisms have been pro- of circular arcs, r in radius and spaced by f, as shown in Fig. 2: L is
posed. For this purpose, a model predicting the dimensions of the the layer thickness, the versor b̂ is the stratification direction, the
fabricated parts have been found by Neural Network approach com- versor n̂ is the normal to the surface and ␣ is the deposition angle,
bined with an optimization algorithm. Positive deviations of FDM i.e. the angle between the versors b̂ and n̂. Thus the profile can be
parts are observed by Sood et al. [16]; adopting the Grey Taguchi’s defined as [7]:
method they obtained the optimum level of the process parame-
ters to simultaneously minimize the percentage change in length, 
width and thickness. Other experimentations based on Taguchi f (x) = r2 − x 2 (1)
336 A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344

Fig. 1. Flow diagram of the methodology.

The drawback of this model is that it does not well work for
deposition angles less than 30◦ and greater than 150◦ . In order to
overcome this problem, an extension of it has been developed using
Neural Network approach [9]. This new formulation provides a full
effective prediction all over the deposition angles:

Ra = −6118 + 1766Tanh (15.67 + 0.016˛ − 0.049L)

− 200.4Tanh (11.54 − 0.08˛ − 0.045L)

− 24.87Tanh (17.13 − 0.044˛ − 0.042L)

−1131Tanh (10.33 + 0.019˛ − 0.034L)

Fig. 2. Deposited filament scheme. +5501Tanh (20.84 − 0.056˛ − 0.02L)

−195.5Tanh (4.559 + 0.062˛ − 0.018L)


In order to evaluate the symbolic integral of the average rough-
ness, the Taylor series expansion of the Eq.1 limited to the second +101.3Tanh (20.09 − 0.088˛ − 0.018L)
order term has been considered: +116.1Tanh (21.66 − 0.105˛ − 0.013L) (6)
x2
f (x) ∼
=r− (2)
2r
Since this model allows predicting with a great precision (less
The profile has been compensated by the mean line as following: than 4% mean absolute percentage error for ABS and less than 9%
 f⁄2 f2
for other materials) the obtainable average roughness for a chosen
1 − 24r + f·r layer thickness L and a deposition angle ˛, it is employed in the
c= f(x)dx = (3)
f −f⁄2 f methodology developed in this paper.
As regards the dimensional deviation, the model developed in
The compensated height of the profile is: [20,21] takes into account the shape of the surface profile which
deviates from nominal one when the deposition angle is differ-
f2 − 12x2
fc = f (x) − c = (4) ent from 90◦ : the empirical determination of f and r expressions
24r allowed to gain two separate formulations for each investigated
Using the values of f and r found by theoretical and experimental layer:
formulations and integrating the absolute values of the compen-
sated height of the profile, it is possible to obtain the prediction of ∼ 2
h(˛)| = 0.19 Cos (˛) (7)
Ra as function of L and ␣ [7,8]: L = 0.254
 f ⁄2
1 f2 Lcsc˛ ∼ 2
Ra = |fc |dx = √ = √ (5) h(˛)| = 0.36 Cos (˛) (8)
f −f ⁄2 18 3r 9 3 L = 0.331
A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344 337

Fig. 3. Schematization of 3D rotation applied to solid model.

These equations are not in a convenient form for the implemen- Fig. 4. Solid model of the object employed to evaluate the methodology processing
tation of the methodology so they are joined in a unique expression time.
as a function of both L and ˛:
h(˛, L) = 2.21(L − 0.168) Cos2 (˛) (9) By using Eq. (6) the average roughness value of each triangle can
be calculated. An association between this value and an adequate
The presentation of the results gained through Eq. (6) and (9) indexed color palette has been built. A direct mapping has been
can be visualized via color shading applied on the 3D model: this applied to 3D model by coloring each triangle with the correspon-
way all roughness and deviation values can be displayed by rotat- dent palette element. The same procedure has been employed to
ing and zooming the geometry in two distinct representations and develop the dimensional accuracy representation using Eq. (9) and
the areas of interest can be accessed allowing the verification of considering that this formulation is affected by an error of 0.05 mm,
the requirements. A processing of the STL model in a specific con- i.e. the experimental error of the model calculated as the standard
figuration is needed. As said, the parameters are the layer thickness deviation of the residuals. These quantitative results allow verifying
and the deposition angle. The former can be chosen in a dedicated whether specific zones of the component comply with the design
window while the latter needs a STL model orientation which is requirements. The process planner can manually search for a com-
typically employed in the CAM environment. In this methodology bination of part orientation and layer thickness which meets the
the part orientation with respect to the stratification direction is part specifications.
allowed according to the following procedure. The STL model is As described in the flow diagram, the indirect method allows
loaded by means of vertices data and facets interconnections and defining the requirements of some specified surfaces. This is typi-
aligned to the reference system of the machine: the plane x-y is cal because a part does not need the same restrictive tolerances and
coincident with the fabrication table and the z axis is the stratifi- surface roughness all over its surfaces. The proposed methodology
cation direction b̂, as shown in Fig. 3. The rotated STL model can be permits the selection of one or more zones and the definition of tol-
obtained by a rigid-body homogeneous transformation as follows: erance and surface roughness requirements for each of them. Then
⎡ ⎤
⎡ ⎤ Cos[ϕy ] 0 0 0 ⎡ ⎤ the STL model in the original orientation is analyzedby means of the

⎢ ⎥ x deposition angles and for a chosen layer thickness and the dimen-
⎢⎥ ⎢0 Cos [ϕx ] 0 0 ⎥ ⎢y ⎥ sional deviation and the average roughness values are calculated
⎢ ⎥=⎢ ⎥ · ⎢ ⎥ (10)
⎣ ⎦ ⎢0 ⎥
Cos [ϕx ] Cos[ϕy ] 0 ⎦ ⎣ z ⎦
for each triangle. A routine checks if all the triangles complies with
⎣ 0
the requirements. Then the model is rotated and the verification
1 1
0 0 0 1 is repeated. The rotations cover the range {−90◦ ,90◦ } for both the
angles ϕx and ϕy in two phases in order to reduce the computational
where ϕx and ϕy are the rotation angles along the axes x and y
time: the former is characterized by a coarse discretization of the
respectively.
solid angle and the set of solutions are collected; the latter checks
This operation affects the STL facets orientation. The normal of
(i) (i) (i) the angles in a fine discretization around the previous found solu-
the i-th triangle defined by the vertices P1 , P2 , P3 can be calculated
tions. The object of Fig. 4 is characterized by 249805 vertices and
considering the cross product between the two vectors of the plane
87234 triangles. An average roughness under 40 ␮m is required on
containing the triangle itself, e.g. two triangle sides, as reported in
the red surface and a tolerance of 0.1 mm is assigned to the conical
Eq. (11):
surface colored in blue.
(i) (i) (i) (i) The code has been implemented in Mathematica 10.2 consider-
→P P ×→P P
n̂i =
1 2 3 2

(11) ing a step of 4.5◦ which is fine enough for the FDM part orientation

→ P (i) P (i) × → P (i) P (i)

1 2 3 2 purpose. The computation has been run on a laptop computer with


Since the deposition angle ˛i is defined as the angle between a Intel(R) Core(TM) i7-6500U CPU @ 2.50 GHz and required 2 s, 5 s
the normal n̂i and the stratification direction b̂, the scalar product and 2 s for the tolerance, the roughness and the final refinement
between these two versors is the cosine of ˛i thus it can be obtained routines respectively. The total time of 9 s shows the methodology
as following: is easily employable in a commercial personal computer. A code
improvement will further improve this computational time. This
˛i = arccos(n̂i · b̂) (12) way the method allows knowing if it is possible to satisfy the prod-
338 A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344

uct requirements and, if some solutions exist, it provides a set of


possible part orientations for a chosen layer thickness. The attain-
able surface quality can be displayed by applying the direct method
of the methodology.

3. Case study

In this section the methodology described in the Section 2 is


applied to a functional prototype: this component is character-
ized by design specifications such as the average roughness and the
shape tolerance in specific zones in order to permit its functionality.
For this case study the application of the methodology, the choice of
a part orientation that permits to comply with the design require-
ments, the experimental validation of a chosen configuration are
provided.

3.1. Application of the methodology Fig. 5. Door handle and its requirements.

The considered case study is a door handle: its virtual model and
design specifications are reported in Fig. 5. The joint pin (in blue) of obtainable average roughness and dimensional accuracy. This
is a functional element because it allows the rotation of the door latter part is important for the methodology because it permits to
handle; thus this cylindrical element (nominal diameter 12 mm) check in real time as the obtainable surface quality changes when
has been characterized by a circularity tolerance of 0.2 mm. The the process parameters are modified. As concern the dimensional
surface colored in red, for aesthetic and ergonomic purposes, must accuracy, since the chromatic map permits to extend the control
have a surface roughness better than other elements: in this specific on the entire surface, the method can be employed to evaluate the
case the requirement is an average roughness Ra lower than 30 ␮m. shape deviation of the fabricated part from the nominal one.
The component has been processed according to the methodol- In Fig. 6 a sketch of the tool interface for the door handle in its
ogy. In order to facilitate the application of the method a tool with original positioning is showed.
an operative interface has been developed in Wolfram Mathemat- In this positioning and for 0.254 mm layer thickness, the door
ica 10.2. This interface is composed by three parts: the first one handle shows the best obtainable surface quality for vertical sur-
gives information about the loaded STL file (number of vertices, faces: thus the joint pin, the coaxial cylindrical surface and the
number of triangles, model volume, file dimension) and shows the lateral sides of the component will be fabricated with an average
rendered and the facetted STL models; the second one is an oper- roughness of about 16 ␮m and a dimensional accuracy of 0.05 mm.
ative panel that permits selecting the layer thickness and the part Although this configuration allows the joint pin component to com-
orientation by the choice of the angles ϕx and ϕy ; the third part ply with the shape tolerance, the upper curved surface shows a
shows, through two 3D chromatic maps, the surface quality in term marked disagreement with the roughness requirement. Thus other

Fig. 6. Operational and output panel for the door handle in the original orientation (ϕx = 0◦ , ϕy = 0◦ ).
A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344 339

Fig. 7. Operational and output panel for the door handle in the ϕx = 90◦ , ϕy = 0◦ orientation.

Fig. 8. Set of part orientations that satisfies surface roughness specifications (red arrows), shape tolerance(green arrows) and both the requirements (black arrows) for
0.254 mm (a) and 0.331 mm (b) layer thickness. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

part orientations must be investigated in order to comply with all tem capability to satisfy the required specifications. For example,
the specifications. If the door handle is positioned so that the curved Fig. 8a is related to a layer thickness of 0.254 mm: as shown, the
surface is vertical (ϕx = 90◦ , ϕy = 0◦ ), the roughness requirement 2 sets of solutions have an intersection, represented by the black
would be satisfied but the joint pin would be fabricated horizontally arrows, demonstrating that the system can fabricate a part satis-
with several zones exceeding the admissible maximum deviation fying the requirements. Conversely, for 0.331 mm layer thickness
(Fig. 7). (Fig. 8b), the intersection between the red and the green arrows
Since this problem has contrasting issues, a configuration allow- does not exist: thus this layer thickness cannot be employed in this
ing the conformity of both the requirements must be searched. This case. From this representation is well evident that the solutions are
aim could be obtained by a number of attempts suggested by the not simple to manually deduce. In fact while the positioning of a
representation output but it is not a simple task. Otherwise the cylindrical surface is expected to give results within a cone, for the
indirect method could be employed to display all the possible sat- curved surface the solutions are not symmetrical.
isfying configurations. According to this methodology, the surfaces The methodology provides a set of solutions by means of the
interested by the specifications have been selected and provided rotation angles couple ϕx and ϕy , as reported in Table 1. The time
to the processing system together with the roughness and toler- needed to process this geometry has been 11 s on the same CPU
ance limits. The outputs of this elaboration are the graphs reported mentioned in the section Methodology. At this stage, it is possible
in Fig. 8. The results are described by a set of arrows representing to choose a solution in the provided set and simulate the obtainable
the part orientations related to each specification: the solutions for surface part quality.
the surface roughness and the shape tolerance requirements are Even if the considered surfaces are apparently easy to manage
reported in red and in green respectively. A first result is the sys- with a single rotation, the methodology found solutions with rota-
340 A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344

Table 1
Set of solutions, for 0.254 mm layer thickness, by means of rotation angles ϕx and ϕy , complying with the door handle specifications.

␸x ␸y ␸x ␸y ␸x ␸y ␸x ␸y ␸x ␸y ␸x ␸y ␸x ␸y ␸x ␸y

−45.5 −9.5 −41.5 −7.5 −43.5 3.5 −43.5 6.5 40.5 −9.5 44.5 −7.5 42.5 3.5 40.5 12.5
−45.5 −7.5 −41.5 −5.5 −41.5 −4.5 −43.5 8.5 40.5 −7.5 44.5 −5.5 44.5 −4.5 42.5 4.5
−45.5 −5.5 −45.5 −4.5 −41.5 −2.5 −43.5 10.5 40.5 −5.5 40.5 −4.5 44.5 −2.5 42.5 6.5
−43.5 −13.5 −45.5 −2.5 −41.5 −0.5 −43.5 12.5 42.5 −13.5 40.5 −2.5 44.5 −0.5 42.5 8.5
−43.5 −11.5 −45.5 −0.5 −41.5 1.5 −41.5 4.5 42.5 −11.5 40.5 −0.5 44.5 1.5 42.5 10.5
−43.5 −9.5 −45.5 1.5 −41.5 3.5 −41.5 6.5 42.5 −9.5 40.5 1.5 44.5 3.5 42.5 12.5
−43.5 −7.5 −45.5 3.5 −45.5 4.5 −41.5 8.5 42.5 −7.5 40.5 3.5 46.5 −0.5 44.5 4.5
−43.5 −5.5 −43.5 −4.5 −45.5 6.5 −41.5 10.5 42.5 −5.5 42.5 −4.5 40.5 4.5 44.5 6.5
−41.5 −13.5 −43.5 −2.5 −45.5 8.5 −41.5 12.5 44.5 −13.5 42.5 −2.5 40.5 6.5 44.5 8.5
−41.5 −11.5 −43.5 −0.5 −45.5 10.5 40.5 −13.5 44.5 −11.5 42.5 −0.5 40.5 8.5 44.5 10.5
−41.5 −9.5 −43.5 1.5 −43.5 4.5 40.5 −11.5 44.5 −9.5 42.5 1.5 40.5 10.5 44.5 12.5

Fig. 9. Operational and output panel for the door handle for the ␸x = 45.5◦ and ϕy = −9.5◦ . orientation.

tions around both the axes: this underlines the usefulness of finding The joint pin, which is characterized by a circularity tolerance,
the parameters’ set all over the solid angle. For example let’s take has been measured using the digital image analysis. This shape
the solution ␸x = −45.5◦ and ϕy = −9.5◦ . The Fig. 9 shows the sim- requirement means that the circular cross section of this element
ulation for this part orientation. As expected, the surface roughness must lay between two concentric circles. This way the cylinder sec-
is less than 30 ␮m upon the curved surface, and the dimensional tion has been acquired by a Canon Canoscan Lide 90 flatbed scanner
deviation of the joint pin diameter is less than 0.1 mm. with a not interpolated resolution of 1200 dpi. The image process-
ing consisted of 4 steps: the binarization, the image polishing, the
edge detection and the data analysis. In order to separate the object
3.2. Experimental validation from the background, a selection has been performed by thresh-
olding brightness values according to the Otsu’s cluster variance
In order to validate the methodology results, the component maximization method [22]. This way the obtained binary image has
has been fabricated in the abovementioned three configurations: been polished from the noises coming from the source variation and
the original positioning ␸x = 0◦ , ␸y = 0◦ the configuration related the image debris. In this case the scanned image is affected by the
to ␸x = 90◦ , ␸y = 0◦ , and the orientation defined by ␸x = −45.5◦ , interference between the camera scan and flickering of the fluores-
␸y − 9.5◦ . The employed machine is a Stratasys Dimension bst 768 cent tube in a way different from the typical Gaussian noise [23].
and ABS has been used. The virtual model has been loaded within In order to preserve the feature intensity values a spatial operator
the machine CAM environment, i.e. the Catalyst Ex, and oriented has been employed to directly clean the defects: a morphological
according to the chosen configurations. closing deleted all the objects smaller than the structuring element,
The fabricated components have been measured in order to set at a dimension of 0.02 mm. The border of the cylinder has been
compare the simulated configurations and to evaluate the satis- obtained by the Canny detection algorithm, based on the first order
faction of the surface quality requirements as predicted by the directional Gaussian derivatives [24]. Finally the point cloud has
methodology. been referred to the least square circle in order to define the out-
A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344 341

Fig. 10. Digital image analysis outputs (red line) aligned with the nominal section (dashed black line) for the configurations ␸x = 0◦ , ␸y = 0◦ (a), ␸x = 90◦ , ␸y = 0◦ (b),
␸x = −45.5◦ , ␸y = −9.5◦ (c).

of-roundness as the radial peak to valley distance from this circle; over the entire surface and having care to match the STL position
then the convex hull method has been employed. This method cal- when reported in the graphical representation.
culates the smallest convex set that contains the acquired points
in the Euclidean plane [25] and it is useful when considering geo-
metrical aspects and determining whether a shaft and a bore can be 3.2.1. Results
assembled together [26]. The calibration through a set of Johansson The results of shape deviations measured on the join pin are now
gauges of this measurement chain led to an uncertainty of 21 ␮m. discussed. In Fig. 10 the output of the digital image analysis (in red)
The roughness measurements upon the upper surface of the has been aligned with the nominal section (in black) in order to
door handle has been performed by a Taylor-Hobson Form Taly- evaluate how the fabricated shape deviates from original one. It
surf Plus profilometer with a declared height accuracy of 16 nm. is well evident that the configuration ␸x = 0◦ , ␸y = 0◦ (Fig. 10a) is
The measurement length is 12.5 mm; the data have been filtered by characterized by a very good accordance with the desired shape:
a corrected Gaussian filter [27] characterized by a cutoff of 2.5 mm all the magnified zones are close to the nominal profile with a very
according to [28]. The measures, spaced by 10 mm, have been taken little deviation. This is expected because the considered section
is entirely contained in the same layer, i.e. the cylindrical shape
342 A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344

Fig. 11. Box and whiskers plot of the measured contact points radial distance of the joint pin and out-of-roundness values for the three considered configurations.

Fig. 12. Predicted (continuous line) and measured (dots) average roughness values for the three considered configurations.

is fabricated vertically. Conversely, in the configuration related to As regards the roughness measurements the predicted values of
␸x = 90◦ , ␸y = 0◦ (Fig. 10b), the cylinder axis is perpendicular to the average roughness have been displayed together the measured
the stratification direction, i.e. it is fabricated horizontally. This way ones and reported in Fig. 12. For the sake of clarity the upper sur-
the section is characterized by the most marked staircase effect and face has been superimposed to the reported graph to identify the
by wide deviations nearby the top and bottom zones: the oscilla- position of each measure. The first configuration (in black) is char-
tions related to the deposited filaments are well evident. The lateral acterized by the best positioning of the surface leading to a constant
zone shows a less pronounced variation, as it is fabricated vertically. average roughness of about 16 ␮m. The experimental data shows
The third configuration, characterized by ␸x = −45.5◦ , ␸y = −9.5◦ a slight positive deviation with respect to this prediction and it
(Fig. 10c), points out a behavior similar to the first case with a little is plentifully within the requirement. The configuration colored in
deviation also at the top and the bottom zones. blue is related to the surface positioned horizontally and thus char-
By applying the convex hull it is possible to select the probable acterized by several deposition angles: as expected the predicted
point of contact of this section. These points have been analyzed trend oscillates between 20 ␮m and 55 ␮m and the experimental
by means of a statistical approach and graphed by the box and values confirm this behavior; this case is not acceptable for the
whiskers plots reported in Fig. 11 for a comparative purpose. required surface finish. The configuration proposed by the method-
The first configuration is in full accordance with the prediction ology orients the surface in a particular 3D rotation making the
model developed in [20] because the distribution is centered on predicted trend to vary between 19 ␮m and 27 ␮m. Also in this
the nominal value and falls within ± 0.1 mm; the out-of-roundness case the measured data confirms the modeled values and points
is 0.13 mm complying with the tolerance requirement. As regards out that the part complies with the surface specification.
the second configuration, the surface is characterized by a number In order to visually compare the measured and predicted rough-
of deposition angles and thus it is affected by a larger distribution: ness values, the 3D colored maps of the measured surfaces have
this corresponds to an out-of-roundness of 0.27 mm which exceeds been constructed and reported in Fig. 13 together the predicted
the tolerance limits. In the configuration found by the proposed ones. It is evident a very marked correspondence between the
methodology, the surface inclination is nearby 90◦ with respect to predicted and fabricated surface morphology for each considered
the stratification direction: thus in this case the tolerance limit is configuration: only very slight deviations are observed confirm-
satisfied. ing the good capability of the employed method to predict the
obtainable surface finishing.
A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344 343

Fig. 13. Comparison between the predicted and the measured average roughness maps for the three considered configurations.

4. Conclusions References

In this paper a methodology able to integrate the prediction [1] I. Gibson, D. Rosen, B. Stucker, Additive Manufacturing Technologies, Rapid
Prototyping to Direct Digital Manufacturing, Springer Verlag New York, 2010.
models of the surface quality obtainable by the FDM technology [2] N. Hopkinson, R. Hague, P. Dickens, Rapid Manufacturing: An Industrial
and the process design have been developed. The used formula- Revolution for the Digital Age, John Wiley & Sons, Chichester, 2005.
tions permit to foresee the dimensional accuracy and the average [3] F.W. Liou, Rapid prototyping and engineering applications, in: A Toolbox for
Prototype Development, CRC Press, New York, 2007.
roughness as a function of two FDM process parameters: the [4] S.H. Masood, Advances in fused deposition modelling, in: S. Hashmi (Ed.),
layer thickness and the deposition angle. These formulations have Comprehensive Materials Processing, Elsevier, Amsterdam, 2014, pp. 69–91.
been integrated in a designed for the purpose tool, equipped by [5] P.M. Pandey, N.V. Reddy, S.G. Dhande, Improvement of surface finish by
staircase machining in fused deposition modelling, J. Mater. Process. Technol.
a simple operator interface, which allows the modification of the 1 (1–3) (2003) 323–331.
abovementioned parameters and the visualization of the surface [6] D. Ahn, H. Kim, S. Lee, Surface roughness prediction using measured data and
quality changes by colored mapping of the 3D model. Moreover interpolation in layered manufacturing, J. Mater. Process. Technol. 209 (2009)
664–671.
the methodology permits to select specific zones, characterized by
[7] A. Boschetto, V. Giordano, F. Veniali, Modelling micro geometrical profile in
tolerance and/or roughness requirements, and provides, if exists, fused deposition process, Int. J. Adv. Manuf. Technol. 61 (9–12) (2011)
a set of solutions represented by the layer thickness and a cou- 945–956.
ple of rotation angles that define the part orientation. In order [8] A. Boschetto, V. Giordano, F. Veniali, 3D roughness profile model in fused
deposition modelling, Rapid Prototyp. J. 19 (4) (2013) 240–252.
to check one or more solutions the suggested process parame- [9] A. Boschetto, V. Giordano, F. Veniali, Surface roughness prediction in fused
ters can be simulated through the generation of the 3D colored deposition modelling by neural networks, Int. J. Adv. Manuf. Technol. 67
representations. A case study has been proposed to validate the (9–12) (2013) 2727–2742.
[10] C.K. Chua, K.F. Leong, C.S. Lim, Rapid Prototyping: Principles and Applications,
method. The considered component is a door handle characterized World Scientific River Edge, 2010.
by both shape tolerance and roughness requirements in two dif- [11] R. Ippolito, L. Iuliano, A. Gatto, A benchmarking of rapid prototyping
ferent zones. The method application supplied a set of solutions techniques in terms of dimensional accuracy and surface finish, CIRP Ann. 44
(1) (1995) 157–160.
for the layer thickness of 0.254 mm, while claimed no solutions [12] M. Mahesh, Y.S. Wong, J.Y.H. Fuh, H.T. Loh, Benchmarking for comparative
for the layer 0.331 mm. One of the found solutions and two rea- evaluation of RP systems and processes, Rapid Prototyp. J. 10 (2) (2004)
sonable orientations, not belonging to the solutions’ set, have been 123–135.
[13] I. El-Katatny, S.H. Masood, Y.S. Morsi, Error analysis of FDM fabricated
experimentally verified. The measurements performed by the digi- medical replicas, Rapid Prototyp. J. 16 (1) (2010) 36–43.
tal image analysis, for the shape tolerances, and by the profilometric [14] R. Singh, Some investigations for small-sized product fabrication with FDM
acquisitions, for the average roughness, highlighted that the found for plastic components, Rapid Prototyp. J. 19 (1) (2013) 58–63.
[15] A. Noriega, D. Blanco, B.J. Alvarez, A. Garcia, Dimensional accuracy
solution permits to satisfy the design requirements while the other
improvement of FDM square cross-section parts using artificial neural
configurations do not, claiming the reliability of the simulation. networks and an optimization algorithm, Int. J. Adv. Manuf. Technol. 69
This methodology fulfills the industrial need of the FDM implemen- (2013) 2301–2313.
tation within the process planning since it simulates in advance [16] A.K. Sood, R.K. Ohdar, S.S. Mahapatra, Improving dimensional accuracy of
fused deposition modelling processed part using grey Taguchi method, Mater.
an entire set of possibilities for particular product geometry and Des. 30 (10) (2009) 4243–4252.
requirements and allows to determine the capability to fabricate
good parts.
344 A. Boschetto et al. / Additive Manufacturing 12 (2016) 334–344

[17] Y.R. Kumar, C.S.P. Rao, T.A.J. Reddy, A robust process optimisation for fused [22] M. Sezgin, B. Sankur, Survey over image thresholding techniques and
deposition modelling, Int. J. Manuf. Technol. Manage. 14 (1–2) (2008) quantitative performance evaluation, J. Electron. Imaging 13 (1) (2004)
228–245. 146–165.
[18] D.Y. Chang, B.H. Huang, Studies on profile error and extruding aperture for the [23] J.C. Russ, The Image Processing Handbook, CRC Press, Boca Raton, 2011.
RP parts using the fused deposition modeling process, Int. J. Adv. Manuf. [24] J.F. Canny, A computational approach to edge detection, IEEE Trans. Pattern
Technol. 53 (9–12) (2011) 1027–1037. Anal. 8 (6) (1986) 679–698.
[19] A. Equbal, A.K. Sood, R.K. Ohdar, S.S. Mahapatra, Prediction of dimensional [25] R. Featherstone, Contact and impact, in: R. Featherstone (Ed.), Robot
accuracy in fused deposition modelling: a fuzzy logic approach, Int. J. Prod. Dynamics Algorithms, Springer US, New York, 2008, pp. 213–239.
Qual. Manage. 7 (1) (2001) 22–43. [26] L. Xiuming, S. Zhaoyao, Application of convex hull in the assessment of
[20] A. Boschetto, L. Bottini, Accuracy prediction in fused deposition modeling, Int. roundness error, Int. J. Mach. Tool Manu. 48 (2008) 711–714.
J. Adv. Manuf. Technol. 73 (2014) 913–928. [27] ISO 16610-22, Geometrical product specifications (GPS) – Filtration – Part 22:
[21] A. Boschetto, L. Bottini, Triangular mesh offset aiming to enhance Fused Linear profile filters: Spline filters, 2015.
Deposition Modeling accuracy, Int. J. Adv. Manuf. Technol. 80 (2015) [28] ISO 4287, Geometrical Product Specifications (GPS) – Surface texture: Profile
99–111. method – Terms, definitions and surface texture parameters, 1997.

S-ar putea să vă placă și