You are on page 1of 14

Integration of Design for

Manufacturing Methods
Rajit Ranjan
Center for Global Design and Manufacturing, With Topology Optimization

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Department of Mechanical and
Materials Engineering,
University of Cincinnati, in Additive Manufacturing
Cincinnati, OH 45221
e-mail: ranjanrt@mail.uc.edu Additive manufacturing (AM) processes are used to fabricate complex geometries using a
layer-by-layer material deposition technique. These processes are recognized for creat-
Rutuja Samant ing complex shapes which are difficult to manufacture otherwise and enable designers to
Center for Global Design and Manufacturing, be more creative with their designs. However, as AM is still in its developing stages, rele-
Department of Mechanical and vant literature with respect to design guidelines for AM is not readily available. This
Materials Engineering, paper proposes a novel design methodology which can assist designers in creating parts
University of Cincinnati, that are friendly to additive manufacturing. The research includes formulation of design
Cincinnati, OH 45221 guidelines by studying the relationship between input part geometry and AM process
e-mail: samantrv@mail.uc.edu, parameters. Two cases are considered for application of the developed design guidelines.
The first case presents a feature graph-based design improvement method in which a pro-
Sam Anand1 ducibility index (PI) concept is introduced to compare AM friendly designs. This method
Center for Global Design and Manufacturing, is useful for performing manufacturing validation of pre-existing designs and modifying
Department of Mechanical and it for better manufacturability through AM processes. The second approach presents a
Materials Engineering, topology optimization-based design methodology which can help designers in creating
University of Cincinnati, entirely new lightweight designs which can be manufactured using AM processes with
Cincinnati, OH 45221 ease. Application of both these methods is presented in the form of case studies depicting
e-mail: sam.anand@uc.edu design evolution for increasing manufacturability and associated producibility index of
the part. [DOI: 10.1115/1.4035216]

Keywords: design for additive manufacturing (DFAM), additive manufacturing (AM),


direct metal laser sintering (DMLS), feature graph, design guidelines, producibility index

1 Introduction property, the DFM concepts for conventional methods cannot be


directly applied to AM, and hence, there is a need for the develop-
Design and manufacturing are two key stages in the develop-
ment of design rules specific to AM processes.
ment of a product. Ideally, an engineer should always consider
Presently, a number of AM processes are being used in the
manufacturability while conceptualizing a design as it has direct
industry such as fused deposition modeling (FDM), stereo lithog-
implications on the design’s feasibility. However, there has
raphy (SLA), selective laser sintering (SLS), direct metal laser
always been a distinct gap between design and manufacturing
sintering (DMLS), electron beam melting (EBM), etc. These proc-
operations in an industry which leads to increase in production
esses are primarily classified based on the method of adhesion
time and cost. The recent trend of outsourcing in industries is
between the layers [5]. Direct metal laser sintering (DMLS) is one
even widening the gap as it implies a cultural separation between
of the leading AM processes which is used by industries to fabri-
designer and manufacturer. The concept of design for manufactur-
cate high-strength metal parts. This process uses a laser power
ing (DFM) and design for assembly (DFA) was developed so that
source for binding metal layers. This process is also particularly
the manufacturability can be incorporated at early design stages.
beneficial while working with materials that are hard to machine
DFM implies that a designer should ensure that their designs are
such as titanium alloys, tool steels, etc. However, due to the inher-
easy to manufacture and overall cost including assembly and
ent nature of DMLS process, there are certain features which are
logistics is minimized [1]. Although generally design guidelines
difficult to manufacture. This paper extends a conference paper by
are formulated specific to manufacturing process, still producibil-
the authors [6] and presents an extensive study of DMLS process
ity of a design depends on several other factors such as geometry,
based on which design guidelines are formulated. Further, a
dimensional tolerances, and raw material [2]. Presently, a consid-
graph-based design improvement methodology is presented which
erable amount of literature is available on DFM and DFA con-
applies the design rules for increasing manufacturability. This
cepts for conventional manufacturing process [3]. This research
method is applicable when a designer intends to redesign a pre-
work aims at developing similar DFM framework and design
existing part for AM.
methodologies for additive manufacturing (AM) processes.
Designers are often provided with a design space and loading
Due to recent technological advancements, AM is gaining pop-
conditions as part specification and there is a need for light
ularity and has the potential to become a mainstream manufactur-
weighting parts while ensuring adequate part strength to take care
ing process [4]. Also, AM processes are capable of fabricating
of applied loads and associated stresses. Such light weight designs
highly complex shapes which are almost impossible to manufac-
are critical to the aerospace and transportation industry and can
ture using conventional methods. The main reason which gives
make full use of the various design structures that can be only pro-
AM this capability is that in this process, designs are sliced and
duced using the additive processes. Topology optimization is an
material is deposited using a layer-by-layer method. Due to this
approach to find the minimum material distribution designs, given
a set of loading conditions [7]. However, topology optimized
1
Corresponding author.
designs may be difficult to manufacture using additive processes
Manuscript received February 6, 2016; final manuscript received November 8, and may result in quality and manufacturability problems. The
2016; published online January 25, 2017. Assoc. Editor: Xiaoping Qian. second part of this paper presents a design methodology which

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-1
C 2017 by ASME
Copyright V
integrates DMLS design guidelines with the topology optimiza- concept of attribute adjacency graph (AAG) which is used to rec-
tion approach to come up with AM process friendly light weight ognize machining features in three-dimensional (3D) geometric
parts. models. Lockett and Marin [27] presented a midsurface-based
approach for feature recognition in molded parts. Gershenson and
Prasad [28] used a component tree diagram to establish the manu-
2 Literature Review facturing modularity of a product. Changchien and Lin [29] used
This section is divided into four subsections. Section 2.1 is a a feature-based method to represent rotational parts with machin-

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
review of part design features which are difficult to manufacture ing features and established a relationship between them. Huang
using DMLS process. Section 2.2 describes different design for et al. [30] represented design specifications of rotational parts in
additive manufacturing (DFAM) methodologies presented in form of a graph for finding an optimal setup plan. Design graph
recent publications. Section 2.3 presents a review of different presented by Brunetti and Golob [31] included material and geo-
methods for representing a design. Section 2.4 describes research metrical information. Recently, Liu [32] proposed a feature
conducted in the area of topology optimization. graph-based approach for finding relevant features in layer-based
manufacturing. Ranjan et al. [6] presented a feature graph-based
2.1 Part Design Limitations for DMLS. Input design geom- design method for identifying problematic features in additive
etry is one of the important factors that affect the part quality in manufacturing.
DMLS process. This section presents a review of previous
research that correlates various quality governing parameters in 2.4 Topology Optimization. Topology Optimization is the
AM with manufactured part quality. Choi and Samavedam [8] computational method for finding minimum material distribution
studied the staircase error and formulated a mathematical relation scheme for a defined loading condition within a design space. The
for determining cusp height. Cusp height is the part surface error field of topology optimization has been thoroughly investigated
created due to the staircase effect created by the layer-by-layer by researchers for more than two decades. A keynote paper in the
building process. Arni and Gupta [9] correlated cusp height with area is presented by Bendsoe and Kakuchi [33], and a comprehen-
flatness tolerance, while Paul and Anand [10] extended this sive theory for the process is presented in the work by Sigmund
approach to develop a relationship among cylindricity tolerance, and Bendsoe [34]. Several attempts have been made by research-
build orientation, and layer thickness. They also used this relation- ers to include manufacturability constraints within the topology
ship to develop a graph-based approach for finding optimal part optimization process. Zhou et al. [35] introduced casting and
build orientations. Clijsters et al. [11] reported that designs with extrusion manufacturing constraints in the process and presented
sharp corners and thin sections are hard to manufacture as these designs with increased manufacturability. Guest et al. [36] incor-
features influence the thermal behavior of the laser sintering pro- porated minimum length criteria in topology optimization process
cess. Penga et al. [12] studied fabrication of thin-walled metal such that feasible designs can be obtained. Recently, Brackett
parts and reported that accuracy of complex thin-walled metal et al. [37] combined support structure minimization with bi-
parts is compromised while controlling the build height. Soe [13] directional evolutionary structural optimization (BESO) and pre-
performed an experimental study relating part geometry with curl- sented the results for two-dimensional cantilever beam. Gaynor
ing effect in laser sintering and showed that box type geometries et al. [38] implemented an overhang constraint within the topol-
are more susceptible to thermal distortion. Paul et al. [14] have ogy optimization process using a Heaviside projection scheme.
presented effect of thermal distortion on metal parts in DMLS They presented the results for two-dimensional cantilever and
process. The requirement of support structures has been analyzed MBB beam. More recently, Mirzendehdel and Suresh [39] incor-
by multiple authors [5,15,16] and a support minimization method- porated support reduction criteria into the pareto optimal-based
ology is presented by Paul and Anand [17]. The idea of optimizing topology optimization algorithm. Langelaar [40] incorporated an
support volume by using cellular structures has been reported by overhang constraint in the topology optimization algorithm using
Strano et al. [18]. a sensitivity approach. Wang and Qian [41] incorporated a unique
B-Spline-based filtering in their topology optimization method. In
2.2 Design for Additive Manufacturing. The development this paper, a novel mathematical model has been formulated using
of AM technology has motivated researchers toward developing a variable mapping technique for integrating design rules within
new design approaches specific to AM. Vayre et al. [19] presented the topology optimization framework.
a design methodology for AM and illustrated their approach with
redesigning a sample part. Ponche et al. [20] proposed a new 3 Methodology
numerical chain-based design method which can find optimal geo- In this paper, a set of DMLS design guidelines for powder
metries in terms of functionality while considering manufacturing fusion additive manufacturing is formulated based on relevant lit-
process parameters. Klahn et al. [21] presented a list of criteria for erature. Computational algorithms are developed for detecting
redesigning a part for AM and illustrated the redesign process on features which violate design guidelines. A graph-based design
various sample parts. Seepersad et al. [22] fabricated plastic parts feature extraction approach is used in conjunction with the design
using laser sintering (LS) process and formulated a set of design rules to identify and remedy AM design violation problems. This
guidelines for increasing manufacturability. However, the author information is then used to develop a producibility index to com-
stated that the developed design guidelines could not be directly pare designs from AM manufacturability perspective. In a subse-
applied to metal sintering process because of difference in physi- quent step, a topology optimization approach is combined with
cal properties. Adam and Zimmer [23] developed a design rule the design rules to come up with lightweight additive friendly
catalog for laser sintering, laser melting, and fused deposition designs.
modeling (FDM) process. Kerbrat et al. [24] presented a hybrid
manufacturing process that combines machining with additive
manufacturing. The authors also proposed a manufacturability 3.1 Development of Design Guidelines. The DMLS process
evaluation method for subtractive and additive manufacturing. has some inherent limitations which can be used as a basis to for-
Rosen [25] presented a DFAM procedure using cellular structures mulate a knowledge base of design guidelines. Design guidelines
along with the concept of manufacturable elements (MELs). can be categorized into following three categories.
3.1.1 Geometrical Parameters. In additive manufacturing,
2.3 Design Representation. Multiple authors have presented parts are fabricated in the form of 2.5 D layers [5] resulting in
algorithms for decomposing a design into its constitutive elements curved surface being manufactured with a staircase effect. The
for manufacturability analysis. Joshi and Chang [26] presented the concept of cusp height is used to quantify this error. Equation (1)

061007-2 / Vol. 139, JUNE 2017 Transactions of the ASME


by Choi and Samavedam [8] presents a mathematical relation height is presented below. AB and AC represent the cusp height
between cusp error, slice thickness, and the angle between the and slice thickness, respectively. The points A and C are given by
build axis and the facet normal of an stereolithography file format
(STL) facet. The relation was further extended by Paul and Anand
[10] to represent cylindricity tolerance as shown in Eq. (2). We C ¼ ðr  cos b; r  sin bÞ
have developed similar relations between cusp height and slice
thickness for conical and spherical surfaces A ¼ ðr  cos b; r  sin b þ tÞ

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Cusp height ¼ Slice thickness  cos ðøÞ (1) Also, Sin b ¼ 1  ðt=rÞ
where t is slice thickness and r is radius of sphere.
where ø is angle between build axis and facet normal. Thus, maximum cusp height is given as

Cusp height ¼ Slice thickness  sin ðgÞ (2) AB ¼ AO  BO


pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where g is the angle between axis of cylinder and build axis. AB ¼ ðr 2  t2 þ 2tr Þ  r
Figure 1 shows a section of a conical feature. The angle
between build axis (ba) and axis of cone (bo) is h and semi cone In order to keep maximum cusp height less than the given cusp
angle is a. The relationship between cusp height (AC) and the height threshold, i.e.
cone geometry as shown in Eq. (3) is developed using analytical
geometry AB  c

The following relation has been developed:


Cusp heightðACÞ ¼ Slice thicknessðABÞ  sin ðh þ aÞ (3)
t2 þ c2
Figure 2 shows a section of a spherical feature manufactured by r (4)
2ðt  cÞ
AM process. The relationship between radius of sphere and cusp
This suggests that for satisfying a cusp height threshold of “c”
with slice thickness “t,” the radius of a sphere should always be
less than the expression shown in Eq. (4).
Cylindrical, conical, and spherical surfaces of a part are
checked for cusp height threshold violation using Eqs. (2), (3),
and (4), respectively. Nonprimitive features such as extruded,
revolved, or freeform surfaces are approximated using triangular
facets and analyzed using Eq. (1). If any surface of the part viola-
tes the cusp threshold, it is marked as critical. Paul and Anand
[10] proposed that cylindricity tolerance violation can be avoided
by reorienting the part and presented a graph-based method for
finding range of optimal build orientations. However, in the case
of multiple features, sometimes it is not possible to avoid viola-
tions just by reorienting, and thus, design changes may become
necessary.
The shape of laser beam and thermal nature of the sintering pro-
cess can cause manufacturability issues in fabrication of certain
features such as sharp corners and thin sections. Perfect knife
edge features are impossible to fabricate because of the shape of
laser beam. Also, thin openings are often fused while sintering
due to the minimum feature resolution capability of laser sintering
process [22]. Therefore, features forming sharp corners and thin
openings should be avoided at design stage.
In AM, the height of part is directly related to cost as it is pro-
portional to build time. Although, height and volume of a part are
important at the design stage; no explicit recommendation can be
provided as these features often defines functionality. However,
these features are considered while comparing designs based on
the producibility index metric described in Sec. 3.3.
3.1.2 Thermal Parameters. In the laser sintering process, tem-
Fig. 1 Conical feature perature history of part plays an important role in defining the
final part quality. Mercelis and Kruth [42] have used a thermal
gradient mechanism (TGM) for explaining the phenomenon of
residual stresses in selective laser sintering (SLS). Thermal
stresses are one of the most important reasons for introducing
curling in parts which is detrimental for part quality. Box type
structures are most susceptible toward curl and geometries with
high curvature have lower tendency to get warped [13]. Also thin
sections in part’s geometry tend to warp easily [11]. Thus, features
forming box type geometry and thin sections are marked as criti-
cal features from a manufacturability perspective.
Another parameter which is rarely considered is the movement
Fig. 2 Spherical feature of the re-coater arm. The re-coater arm swipes over the part after

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-3
sintering of every layer to lay the powder for the next layer to be there is no assurance of minimal supports. In this work, a topology
sintered. It has been reported that if the input geometry has a long optimization-based design method is presented which can create
edge parallel to re-coater arm, then it can strike the re-coater arm designs with low support requirement that is additive friendly.
disrupting the functioning of the machine [43]. Also, features that A feature graph-based design improvement method is presented
are curled up due to residual stresses may collide with the re- which can help designers in identifying features requiring sup-
coater arm causing interruption in the movement. So, long edges ports. In this approach, it has been assumed that surfaces making
and thin overhangs facing the re-coater arm are also marked as an angle more than 35 deg with the build axis needs support [16].

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
critical features. Figure 3 depicts the problem of a thin overhang Based on this criterion, features which require support or come in
obstructing the re-coater arm movement. contact with support are marked as critical for designer’s recon-
sideration. Moreover, a percentage value is given to each feature
3.1.3 Support Structures. In AM, parts are built upright in the requiring support so that features that require maximum support
machine chamber along the build axis. Thus, features such as can be modified first. Figure 4 depicts a typical part with support
overhangs, surfaces with negative draft, and undercuts require structures.
support [5]. Also, sometimes supports help in reducing thermal
distortion by increasing heat dissipation [16,44]. However, build-
ing support structures not only increases total sintering time but 3.2 Computational Algorithms for Feature Detection. This
also requires additional effort in removal. Generally, supports are section presents two computational algorithms for detecting fea-
removed using conventional processes and support removal can tures which can cause problems related to manufacturability. The
be detrimental to the surface finish of the areas where support first algorithm performs a slice-by-slice analysis in order to iden-
comes in contact with part. So, in general, minimum requirement tify features such as sharp corners, thin sections, and fusible open-
of supports should be incorporated at design stage. Researchers ings. The second algorithm uses a planar triangular approximation
have suggested partial, cellular [45], and surface inclination-based of the surfaces of the CAD model in order to identify support
[44] supports for surfaces. Performing topology optimization needs and calculate cusp error for a given design.
based on the loading presents optimum lightweight designs, but 3.2.1 Slice Contour Offset Algorithm for Identification of
Sharp Corners, Small Holes, and Thin Regions. The CAD design
of the part is converted into its STL format, and slicing operation

Fig. 5 (a) Internal angle and offsetting, (b) thin region detec-
Fig. 3 Re-coater arm movement
tion, and (c) detection of thin openings

Fig. 4 Sample bracket with supports Fig. 6 Flowchart for contour offset algorithm

061007-4 / Vol. 139, JUNE 2017 Transactions of the ASME


in a slice contour (Fig 5(c)). The flowchart for contour-based off-
set algorithm is presented in Fig. 6.
3.2.2 Triangular Facet-Based Algorithm for Cusp Violations
and Detection of Support Requirements. The part surfaces are
approximated as triangular facets within the modeling environ-
ment using the application programming interface (API) function-

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
ality of Siemens NX modeler. Cusp height for each facet is
calculated and summed up using Eq. (1) [8]. Mean cusp height is
thus obtained as shown in Eq. (5).
Fig. 7 Hollow cylinder: (a) facet requiring support and (b) facet P
n
touching support t  cos h
i¼1
Mean Cusp Height ¼ (5)
is performed using the algorithm described by Topcu et.al. [46]. n
Points obtained for every slice contour are arranged in a sequen-
tial order and connected using line segments. The internal angle where n is number of facets, t is slice thickness, and h is angle
(/) between two consecutive line segments is computed and between facet normal and build axis. Part surfaces are checked
stored. The number of sharp corners with interior angles less than and cusp violations are marked using Eqs. (1)–(4) presented in
a predefined threshold of 20 deg in any slice is counted and recog- Sec. 3.1.1.
nized as sharp corners. Next, an inward offset of the slice contour For calculation of support volume and contact area, the
is constructed using the reverse STL facet normal and internal normal vector of each NX model facet is computed and the angle
angle information as shown in Fig. 5(a). All offset polygons between the normal and build axis is calculated. If the angle is
(ABCD) are examined for intersection with any other offset poly- greater than 125 deg (90 deg þ 35 deg), then that facet is marked
gons. If the offset polygons intersect with each other a thin region as a support facet. Supports are manually created for marked fac-
is recognized in the slice (Fig 5(b)). Similarly, when the offsetting ets using the modeling tool of NX and volume of support is
is performed in the direction of STL facet normal, the algorithm stored.
provides information about the number of small openings or holes Further, for calculation of support contact area, a facet projec-
tion algorithm is used. First, area of every triangular support facet

Fig. 8 Flowchart for triangular facet algorithm

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-5
in NX is computed using triangle area formula. Next, the facets
which are not marked for support but lie directly beneath a sup-
port facet are identified and marked as contact facets. The total
contact area is given by sum of the areas of both support and con-
tact facets. Figures 7(a) and 7(b) highlight the support and contact
facets, respectively, for a cylindrical part with build axis perpen-
dicular to the axis of cylinder. The detailed algorithm is described

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
in the flowchart shown in Fig. 8.

3.3 Producibility Index. To assess the extent of part’s con-


formance to the AM design rules, a producibility index metric has
been developed in this work. Producibility index (PI) is the mea-
sure of goodness of a design’s manufacturability in a particular
build orientation. The concept is formulated to quantify and com-
Fig. 9 Algorithm for making feature graph
pare designs. The following eight factors are included in calcula-
tion of PI:
the weights remain unity. In this study, equal weights of 0.125 are
(1) Total number of sharp corners in all the slice contours
given to each parameter. Designs are compared by assessing the
(2) Number of small holes/small openings in all slice contours
producibility index metric; the higher the value of the index, the
(3) Number of thin regions
better the design is from an additive manufacturing perspective.
(4) Mean cusp height (mm)
(5) Surface area contacting support (mm2)
(6) Volume of support structure needed (mm3) 3.4 Design Methodology. This section presents two
(7) Height of the part (mm) approaches for optimally designing additive manufactured parts.
(8) Volume of the part (mm3) In the first method (section 3.4.1), a feature graph-based design
improvement scheme is proposed that can be used for improving
The first three parameters are calculated using the contour off-
manufacturability of designs. The second approach (section 3.4.2)
set algorithm described in Sec. 3.2.1, while parameters 4–6 are
integrates topology optimization with design for additive rules
calculated using the triangular facet algorithm described in Sec.
described in the previous section to produce lightweight designs
3.2.2. The height of the part can be calculated based on the height
that are manufacturable using additive processes.
of the bounding box of the part model and is obtained directly
from the NX CAD model. Based on the triangular approximation
3.4.1 Feature Graph-Based Design Improvement Method. In
of the surfaces of the CAD model, the volume of the part can be
this section, a method of depicting a design in the form of feature
calculated and is obtained directly using the NX application pro-
graph with inter related part features is presented. The feature
gramming interface (API).
graph is an extension of the AAG graph presented in Ref. [26] and
3.3.1 Mathematical Definition of Producibility Index (PI). All is constructed based on the face adjacency of neighboring fea-
the eight factors considered are used to calculate part producibility tures. Features and their relations are then verified against design
index (PI). Since all the factors considered for PI have a different guidelines developed for the DMLS process. Computational algo-
order of magnitude, they have to be normalized to the same scale rithms described in Sec. 3.2 are used to mark design guideline vio-
for comparison. Suppose S is the set of “n” designs to be lations on the graph and this information can then be used for
compared further design improvement. Functional features are identified and
any violations related to them are ignored because changing the
S(D) ¼ {D1, D2, D3,……., Dn} functionality of a part would defeat the purpose of design. The
flowchart for constructing a feature graph for manufacturability
and C is the matrix of “m” factors considered for n designs
analysis is presented in Fig. 9.
C ¼ {C1(D1) C1(D2) C1(D3)……….…. C1 (Dn), All functional features are marked on the central node of the
C2(D1) C2(D2) C2(D3)…………. C2 (Dn), graph. A nonfunctional feature is arranged in the adjoining node
.…………………………………………., based on its relation with other features. Relationship between
Cm(D1) Cm(D2) Cm(D3)………….Cm (Dn)} two features is marked as an attribute on the connecting link.
Guideline violations can occur either because of features (nodes)
Then, PI for a design (Dj) is given by Eq. (6) or relations (links). As shown in Fig. 10, a part is broken down
into features, namely, F1, F2, F3, F4, and F5. F3 and F4 are identi-
X
r
Wi  Ci ðDj Þ
PIðDj Þ ¼ fied as functional features, while the rest of the features are
i¼1
MaxfCi ðD1 Þ; Ci ðD2 Þ… Ci ðDn Þg arranged as nodes. Relations (R1, R2, and R3) are marked as
X
s attributes on the links. Violations are recognized based on design
Wk  MaxfCk ðD1 Þ; Ck ðD2 Þ:::Ck ðDn Þg guidelines and the corresponding element (node or link) that is
þ (6)
k¼1
Ck ðDj Þ marked.
In order to construct the feature graph, first a feature relation
where Wi and Wk are weights for different factors: i ¼ 1 to r repre- table is constructed within the Siemens NX CAD environment
sents favorable criteria for design, k ¼ 1 to s represents unfavora- using NX application programming interface (API). All the surfa-
ble criteria for design ces of a 3D CAD model are analyzed and their topology informa-
Such that tion is stored in a (N þ 2) x (N þ 2) matrix, where N is the number
of surfaces. The adjacency relation between these surfaces is
rþs¼m stored and marked on the feature table. The feature relation table
of a sample part shown in Fig. 11 is presented in Table 1. In the
where larger values of favorable criteria and smaller values of table shown, the first two columns represent the CAD surface ID
unfavorable criteria are preferred from a design for AM perspec- and surface type, respectively. The first two rows are transpose of
tive. Here, Cm (m ¼ 8) represents the eight factors described in the first two columns. Also, “1” in the table represents the adjacency
section of producibility index and the weights are decided based status between a row and column surface similar to the AAG
on designer’s preference for these factors such that the sum of all graph presented by Joshi and Chang [26]. Using the information

061007-6 / Vol. 139, JUNE 2017 Transactions of the ASME


Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Fig. 10 Representative feature graph

3.4.2 DFAM Constrained Topology Optimization-Based


Design. Topology optimization is a mathematical approach that
optimizes material layout for given design space and loading con-
ditions [34]. The most common application of the optimization
process is for minimum compliance or maximum stiffness in
which an iterative finite-element analysis (FEA) is conducted to
find material distribution for minimum deformation. The material
density (Xph) is allowed to vary between zero and one where den-
sity value of one represents material and zero represents a void.
Solid isotropic material with penalization (SIMP) method pre-
sented by Ref. [47] is typically used to penalize intermediate den-
Fig. 11 Sample part for feature graph sity values and drives the final result toward a 1/0 material
configuration. This research extends the topology optimization
method presented by Liu and Tovar [7] and reformulates it into a
presented in the feature relation table, a graph is constructed and new DFAM constrained topology optimization-based design
analyzed. Algorithms described in Sec. 3.2 are used to find critical approach by incorporating additive manufacturing constraints.
features and manufacturability problems are marked on the feature As the topology optimization process is computationally inten-
graph. As complex parts may have a large number of features to sive, there are certain numerical difficulties such as mesh depend-
be analyzed, populating the feature relation table helps in con- ency, local minima, and checkerboard pattern that have to be
structing the feature graph by providing information about inter overcome. Many researchers have suggested the use of a density
feature relationships. The feature graph with design guideline vio- filter in order to reduce the effect of such computational complex-
lations for the sample part is shown in Fig. 12. ities. The density filter suggested by Ref. [7] presents a relation

Table 1 Feature relation table

Face ID 38,805 40,793 40,729 40,744 40,961 38,840 38,841 38,842 38,843 38,844

Type 16 16 16 22 16 22 22 22 16 22

38,805 16 1 1 1 1
40,793 16 1 1 1 1
40,729 16 1 1 1
40,744 22 1 1 1
40,961 16 1 1
38,840 22 1 1 1 1
38,841 22 1 1 1 1
38,842 22 1 1 1 1 1
38,843 16 1 1
38,844 22 1 1 1 1 1

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-7
Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Fig. 12 Feature graph for sample part

between design variable and physical density. In this paper, a new number of elements in each direction (X, Y, and Z) is a user
intermediate constraint variable (Xa) is introduced linking the defined variable. The Z direction is assumed to be the build direc-
design space and physical density to redefine the mapping. tion for AM and constraints are applied such that elements are
As discussed in the section on design guidelines, fabrication checked for support only along the negative Z direction. A cubical
and removal of support structures is one of the most important element can have maximum nine elements lying just below it as
factors affecting the manufacturability of a design. Thus, an addi- shown in Fig. 13(a). For boundary elements shown in Figs 13(b)
tional constraint of developing designs with minimum supports is and 13(c), six or four elements, respectively, may lie below it
introduced within the topology optimization algorithm. A mathe- depending on element’s location. It has been assumed that if an
matical formulation is developed which performs a support check element has a physical density greater than 0.5, it represents mate-
for every element after the FEA process. The density of a properly rial and vice versa. Also, using this assumption an element is said
supported element remains unchanged; however, density of to be properly supported if the element lying directly below, it has
unsupported element is divided by a factor proportional to number a physical density greater than 0.5.
of void elements lying just below it. Due to density reduction after The minimum compliance problem that minimizes deformation
every FEA iteration, unsupported elements tend to diminish and of the structure for given support and loading conditions has been
subsequently a design with a lower support requirement is modified from Ref. [7] and defined as follows:
generated.
The process starts by dividing the entire design space into cubi- Find, design variable (X) ¼ [X1, X2, X3,……, Xe,…., Xn]T
cal elements (voxel). The extent of discretization specified by the Minimize: C (Xph) ¼ FTU(Xph)

Fig. 13 Support elements for (a) nonboundary element, (b) boundary element with six bot-
tom elements, and (c) boundary element with four bottom elements

061007-8 / Vol. 139, JUNE 2017 Transactions of the ASME


Subject to: Vol (Xph) ¼ (Xph)T.v - v  0 reduced. The role of transformation w(x) is to discretize the vary-
where v is vector of element volume, v is maximum allowable ing physical density to 0/1 condition. Repeating the process itera-
volume, C(Xph) is compliance, F is applied load, and U(Xph) is tively reduces the probability of material growth at elements
nodal displacement vector. which are not properly supported. An extra “1” has been added to
The problem formulation and detailed solution for topology the denominator just to avoid an indeterminate condition of 0 deg.
optimization is found in Ref. [7], where physical density (Xph) is Another term, Cdeg, in support coefficient (re ) definition basically
computed directly from design variable (X). In this research, com- controls the intensity of support constraint. Higher values of Cdeg

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
putation of constraint variable (Xa) is added in the process where mean that nonsupported elements will be more severely penalized.
physical density (Xph) is a function of constraint variable (Xa) However, increasing Cdeg more than a certain limit makes the pro-
which, in turn, is a function of design variable (X). The relation cess unstable and produces trivial results.
between design variable (X) and constraint variable (Xa) is It is observed that adding the new support constraint with the
adopted from Ref. [7] and shown in Eq. (7). topology optimization process makes it in general hard to con-
The relation between constraint variable (Xa) and physical den- verge. Thus, a penalty variation scheme has been introduced to
sity (Xph) is presented by Eq. (9), which defines the support reduc- increase or decrease the SIMP’s penalty depending on support
tion criteria. The relation uses a transformation scheme (w(x)), availability for elements. An exponential function defined by
which basically converts material presence into a 0/1 condition. Eq. (11) is used to assign penalty for each element depending on
The mathematical definition of w(x) presented by Eq. (10) is such its support coefficient.
that it returns a value of one for density less than 0.5 and vice
HL
versa. Equation (7) from Ref. [7] represents the physical density Penalty ¼ H  (11)
(Xeph Þ for eth element, where Nes represents the support neighbor- eSðre 1Þ þ 1
hood for eth element. Support neighborhood Nes is defined such
that it contains all the bottom neighbors of eth element except the where H is high-penalty threshold, L is low-penalty threshold, S is
element lying directly below it. Every element is identified in 3D slope constant, and re is support coefficient
space according to its X, Y, and Z coordinates. If eth element has The penalty function is designed to allocate a high-penalty
coordinates (i,j,k), then support neighborhood Nes will have eight threshold, H, if support coefficient ðre Þ is greater than one and a
members with coordinates as (i  1,j,k  1), (i þ 1,j,k  1), lower penalty threshold, L, if support coefficient ðre Þ is less than
(i,j  1,k  1), (i,j þ 1,k  1), (i  1,j  1,k  1), (i þ 1,j  1,k  1), one. Figure 14 shows the plot of penalty function. Basically, the
(i  1,j þ 1,k  1), and (i þ 1,j þ 1,k  1). Equation (8) presents developed function signifies that if an element is not properly sup-
the formulation for support coefficient (re Þ, which serves as a ported and has support coefficient greater than one, then penalty
division factor for unsupported elements value for that element is increased and thus reduces the possibility
X of material growth for that element. The modified
xh  vh  Heh Karush–Kuhn–Tucker conditions for the optimization are given
hNe by Eqs. (12)–(14). A comprehensive flowchart representing the
Xea ¼ X (7) DFAM-constrained topology optimization is shown in Fig. 15
Heh  vh
hNe

  X  uðXbottom
ph
Þ @cðXph Þ @volðXph Þ
re ¼ Cdeg  uðXmph Þ þ 1 (8) þk ¼0 (12)
mNes
@x @x

Xea Xea @cðXph Þ @cðXph Þ @Xph @Xa


Xeph ¼ ¼ (9) ¼   (13)
re   X    uð ph
Xbottom Þ @x @Xph @ Xa @x
Cdeg  u Xmph þ 1
mNes @volðXph Þ @volðXph Þ @Xph @Xa
¼   (14)
@x @Xph @ Xa @x
where Xeph is the physical density for eth element, Ne is neighbor-
hood of an element Xe with volume; ve and its definition is
described in Ref. [7], Heh is the weight factor defined as Heh ¼ R–
dist(e, h), Nes is the support neighborhood defined such that it con-
tains all bottom neighbors of eth element except the element that
lies directly below it in the negative Z direction, Cdeg is degree of
ph
support filter, re is support coefficient, Xbottom represents the phys-
ical density of element that lies directly below eth element in the
negative Z direction,
and the transformation w(x) is defined as-

0 x  0:5
wðxÞ ¼ (10)
1 x < 0:5
The definition of support coefficient (re ) shown in Eq. (8) is
such that it gives a definite control to the element lying directly
below eth (i, j, and k) element by keeping its transformed physical
density as exponent of the denominator. The coordinates of such
element is (i, j, k  1). Support coefficient will become 1 in case
the element (i, j, k  1) has a physical density greater than 0.5,
i.e., element (i, j, and k) is properly supported. This signifies that
in the case of proper support, element density will remain
unchanged and material growth is not affected. However, if ele-
ment (i, j, and k) is not properly supported, then the support coeffi-
cient (re ) becomes proportional to sum of all the void elements in
neighborhood Nes and the resulting physical density (Xeph ) is Fig. 14 Plot for penalty function

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-9
Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Fig. 15 Flowchart for DFAM topology optimization

Fig. 16 Process of design evolution

Fig. 17 Design 1 for heat exchanger


4 Results
This section presents the application of described design meth- all the knife edges as critical features because they form sharp cor-
odologies to the design evolution process. Four case studies are ners in the slice contours. The “slice contour offset” algorithm
used to depict the improvement in the manufacturability of a described in Sec. 3.2.1 is used to successfully detect all these
design based on the design rules formulated in this work. Figure 16 sharp corners. A uniform layer thickness of 50 lm is considered
depicts the proposed design improvement cycle with increasing for this analysis. In the second iteration, the design has been modi-
producibility index with every design iteration. As the value of fied to eliminate features that form the sharp corners while
index increases, the designs become more amenable to manufac- increasing the surface area for better heat transfer.
turing with reduced cost. Figure 18 shows final iteration of the design modification pro-
The first two case studies present the design improvement cess of a heat sink. It represents a modified design for a radial heat
method that described in Sec. 3.4.1 and the increase in manufac- sink with twice the surface area of design 1. It is clear that from
turability as observed by comparing the producibility index (PI) the point of view of conventional manufacturing methods, design
for designs. The last two case studies implement the topology 1 is easier to manufacture. But in the case of AM, design 2 shows
optimization-based design method described in Sec. 3.4.2 and higher PI and thus has better manufacturability due to elimination
compare the support reduction achieved in the designs. The algo- of thin regions and sharp corners. Table 2 shows the computation
rithms described in Secs. 3.2.1 and 3.2.2 are used for detection of of PI and the progression of design iterations for this case study.
manufacturability features and take about 15–20 s of run time on a
core i7 computer with 16GB of RAM for the part models
described in case studies 1 and 2 below. 4.2 Case Study 2. In this case study, the design of a simple
bicycle pedal is considered for manufacturability validation.
Figure 19(a) shows the first iteration of design with a central
4.1 Case Study 1. In this study, the design of a heat sink is cylindrical feature as the main functional feature. The build orien-
considered. The first design shown in Fig. 17 presents a type of tation is decided based on the fact that best build quality for cylin-
heat sink that has fins with knife edges where surface area is a ders is achieved when they are built upright with the build axis
measure of efficiency. The feature graph of this design highlights aligned with the cylinder axis. The feature graph of this design is

061007-10 / Vol. 139, JUNE 2017 Transactions of the ASME


Table 3 Calculation of producibility index for case study 2

S.No Factor Design 1 Design 2 Design 3

1. Mean cusp height 0.024771 0.029488 0.03543


2. No. of sharp corners 24 0 0
3. No. of thin regions 0 0 0
4. No. of fusible contours 4 0 0

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
5. Build height (mm) 130 128 127.5
6. Volume of support (mm3) 297312.51 231100.92 154655.8739
7. Volume of part (mm3) 86043.0272 81026.32 87925.4944
8. Area touching support (mm2) 14556.77 9063.9824 1214.2
Producibility index (PI) 7.336 10.763 26.15

Fig. 18 Design 2 for heat exchanger

Table 2 Calculation of producibility index for case study 1

S.No Factor Design 1 Design 2

1. Mean cusp height (mm) 0.03607 0.043281


2. No. of sharp corners per layer 24 0 Fig. 20 Problem definition for cantilever beam
3. No. of thin regions per layer 24 0
4. No. of fusible contours 0 0
5. Build height (mm) 4 3 Table 4 Process specifications for topology optimization
6. Volume of support (mm3) 0 0
7. Volume of part (mm3) 64.37 61.06 Standard Constrained
8. Area touching support (mm2) 0 0 topology topology
Producibility index (PI) 2.95 12.21 optimization optimization
S.No Factor process process

1. No. of elements in X direction, Nx 50 50


2. No. of elements in X direction, Ny 20 20
3. No. of elements in X direction, Nz 10 10
4. Initial density 0.3 0.3
5. Density filter threshold 2 2
62
6. Penalty 4 6
e60ðre 1Þþ1

Fig. 19 Bicycle pedal: (a) design iteration1, (b) design itera-


tion2, and (c) design iteration3

constructed and violations associated with every face are marked


on the feature graph. It is observed that one particular face con-
tributes to almost the entire support requirement and the same
face is also responsible for problems related to re-coater move- Fig. 21 (a) Topology optimization result, (b) design for cantile-
ment. Thus, modification of this particular feature is the main ver beam, and (c) design with support
focus in the second iteration.
In the second iteration, the design is changed by keeping the
central cylindrical feature unchanged and modifying the topmost this change, producibility of design 3 is approximately 3.5 times
planar face to a curved feature as shown in Fig 19(b). A major more than the initial design. Table 3 presents the calculation of PI
reduction in support volume and support contact area is achieved and shows the progression of the designs.
in this iteration. Also, tilted cylinders were made horizontal so
that the number of sharp corners is reduced. The effects of these
changes are reflected on the increased value of PI, with the PI for 4.3 Case Study 3. This case study presents the application of
design 2 being 1.4 times than that of design 1. Further, using the topology optimization process for generating designs with mini-
feature graph, it is observed that there is additional scope for mal support requirements. The design of a cantilever beam is con-
reduction of support volume. So, the third iteration focuses on sidered here where the mathematical formulation presented in
modifying features such that support need is further minimized. Sec. 3.4.2 will be applied. Figure 20 shows the problem formula-
In the third design iteration shown in Fig. 19(c), the side pedal tion in which loads are applied at the end of the design space and
members were made slightly oblique and the top curvature was the other end is fixed and constrained from any movement. Both
adjusted such that support needs are further reduced. Owing to standard and DFAM-constrained topology optimization processes

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-11
Fig. 25 (a) Results for standard topology optimization pro-

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
cess, (b) 3D model for design, and (c) support structures
Fig. 22 (a) Results for constrained topology optimization pro-
cess, (b) 3D model for design, and (c) support structures

Table 5 Design comparison table for case study 3

Conventional top opt Top opt with DFAM

Build time: height of part (mm) 20 18.1


Area requiring support (mm2) 216.34 149.32
Volume (mm3) 3416 4014 Fig. 26 (a) Results for constrained topology optimization pro-
Support volume (mm3) 2909.7 767.66 cess, (b) 3D model for design, and (c) support structures
Total sintering volume (mm3) 6325.7 4781.66

Table 7 Design comparison table for case study 4

Table 6 Validation FEA results for case study 3 Conventional top opt Top opt with DFAM

Results Max. deformation (mm) Max. stress (Von-Mises) Pa Build time: height of part (mm) 10 10
Area requiring support (mm2) 232.36 190.1
3 7
Top. opt. 4.02  10 4.53  10 Volume (mm3) 9784.74 8998.10
Top. opt. with DFAM 6.25  103 3.01  107 Support volume (mm3) 1730.5 948.21
Total sintering volume (mm3) 11515.24 9946.31

Fig. 27 Deformation results for (a) standard top. opt. and (b)
constrained top. opt.

Fig. 23 Deformation results for (a) standard top. opt. and (b)
constrained top. opt. requirement for this design are shown in Figs. 22(b) and 22(c),
respectively. It is observed that design 1 shown in Fig 21(b) has a
top flat feature which is responsible for most of the support. How-
ever, for design 2 (Fig. 22(b)), material distribution has been
changed to avoid such long, flat feature, with a resulting total sup-
port volume reduction of 73.6%. Table 5 presents the comparison
of both designs, and it can be observed that a 24% reduction in
total sintering volume is obtained.
A final FEA validation of both designs is conducted using ANSYS
workbench. It is found that the new design obtained by DFAM-
constrained topology optimization is well within the limits and is
capable of sustaining the applied loads. The maximum stresses are
Fig. 24 Design space and loading conditions for case study 4 found to be less than the yield strength of specified material, i.e.,
2.5  108 Pa. Table 6 presents maximum stress and deformation
values for both designs, and Fig. 23 shows deformation results.
are applied for finding cantilever design and results are presented.
Table 4 shows the specifications used for running the topology 4.4 Case Study 4. Generally in AM, build orientation is typi-
optimization process. Figure 21(a) shows the output design of cally decided based on build time and critical quality of critical
standard topology optimization process. The voxel-based output features. For example, it is advised to fabricate cylindrical fea-
of optimization process was used as a basis for developing a 3D tures with high tolerances such that it is manufactured with the
model for the cantilever beam using NX model and is depicted in build axis along the cylindrical feature axis. The process of
Fig. 21(b). Support requirements for the design are identified and DFAM-constrained topology optimization becomes more relevant
support structures are generated using the 35 deg angle-based cri- in such conditions, because in this process, build orientation is
teria. Figure 21(c) shows the design with required support defined and fixed before beginning the optimization. This case
structures. study applies the DFAM-constrained topology optimization pro-
The result for DFAM constrained topology optimization is cess for a design space which has a predefined cylindrical hole. In
shown in Fig. 22(a). The NX-based 3D model and support this situation, the axis of hole is defined as the best build

061007-12 / Vol. 139, JUNE 2017 Transactions of the ASME


Table 8 Validation FEA results for case study 4 product development as well. Computational algorithms presented
here can be enhanced to detect other guideline violation such as
Max. Max. stress Compliance re-coater movement and specific features prone to thermal defor-
deformation (mm) (Von-Mises) Pa (%) mations. Also, DFAM-constrained topology optimization-based
method only focuses on support structure reduction. A more inte-
Top. opt. 6.408  104 2.837  106 100
Top. opt. with DFAM 7.31  104 4.005  106 127
grated topology optimization approach combining other important
DFAM factors can be developed. The effect of build orientation is

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
a critical factor in AM. Often guideline violations can be avoided
orientation. Figure 24 presents the problem formulation where a just by changing the build orientation. Moreover, topology optimi-
square design space is loaded with uniform vertical force distribu- zation for different build orientations can be considered and the
tion applied along its edges. The dimensions for design space are best topologically optimized design that is manufacturable can be
defined as 50 units  50 units  10 units with a five units radius selected out of obtained results. Thus, analyzing orientation
hole in the center. All other parameters are same as shown in effects on manufacturability can be another avenue for future
Table 4. Due to symmetrical design space geometry, only a quar- research.
ter of design space is considered for optimization process and
obtained results are extrapolated. Figure 25(a) shows the result for References
standard topology optimization process. The NX 3D model is cre- [1] Rosen, I. G., and Stucker, B., 2010, Additive Manufacturing Technologies,
ated using the voxelized output of MATLAB code, and supports are Springer, New York.
generated in NX using the angle-based criteria. Figures 25(b) and [2] Ponche, R., Hascoet, J. Y., Kerbrat, O., and Mognol, P., 2012, “A New Global
25(c) present the NX model and support structures, respectively. Approach to Design for Additive Manufacturing,” Virtual Phys. Prototyping,
7(2), pp. 93–105.
The DFAM-constrained topology optimization process is [3] Chiu, M.-C., and Okudan, G., 2010, “Evolution of Design for X Tools Applicable
applied on same design problem and the generated output is to Design Stages: A Literature Review,” ASME Paper No. DETC2010-29091.
shown in Fig. 26(a). The 3D model based on the output is con- [4] Hague, R., Mansour, S., and Saleh, N., 2004, “Material and Design Considera-
structed using NX which is shown in Fig. 26(b). Support struc- tions for Rapid Manufacturing,” Int. J. Prod. Res., 42(22), pp. 4691–4708.
[5] Kulkarni, P., Marsan, A., and Dutta, D., 2000, “A Review of Process Planning
tures are created by identifying regions requiring supports and Techniques in Layered Manufacturing,” Rapid Prototyping J., 6(1), pp. 18–35.
support volumes for both designs are compared. In the case of [6] Ranjan, R., Samant, R., and Anand, S., 2015, “Design for Manufacturability in
DFAM-constrained topology optimization process, the need for Additive Manufacturing Using a Graph Based Approach,” ASME Paper No.
support is reduced by 45% resulting in a reduction of total sinter- MSEC2015-9448.
[7] Liu, K., and Tovar, A., 2014, “An Efficient 3D Topology Optimization Code
ing volume by 13.6%. Figure 26(c) presents the design with Written in MATLAB,” Struct. Multidiscip. Optim., 50(6), pp. 1175–1196.
required supports, and Table 7 shows the reduction in support vol- [8] Choi, S. H., and Samavedam, S., 2002, “Modeling and Optimization of Rapid
ume and contact area. Figure 27 shows the deformation results for Prototyping,” Comput. Ind., 47(1), pp. 39–53.
both the designs, and Table 8 presents the FEA validation results. [9] Arni, R., and Gupta, S. K., 2001, “Manufacturability Analysis of Flatness Toler-
ances in Solid Freeform Fabrication,” ASME J. Mech. Des., 123(1),
A final FEA of both designs is performed as it is observed that pp. 148–56.
compliance value of DFAM constrained topology optimization is [10] Paul, R., and Anand, S., 2011, “Optimal Part Orientation in Rapid Manufactur-
higher than that of normal topology optimization. The support crite- ing Process for Achieving Geometric Tolerances,” J. Manuf. Syst., 30(4),
ria integrated in the DFAM topology optimization is responsible for pp. 214–222.
[11] Clijsters, S., Craeghs, T., Moesen, M., and Kruth, J.-P., 2012, “Optimization of
making the process more difficult to converge, and thus, increasing Thin Wall Structures in SLM,” Fraunhofer Additive Manufacturing Alliance,
the compliance value. The final FEA validation results verify that Direct Digital Manufacturing Conference, Berlin, March 14–15.
both the designs are suitable for sustaining the specified loads. [12] Penga, L., Shengqinb, J., Xiaoyanb, Z., Qianwub, H., and Weihaoc, Z., 2007,
“Direct Laser Fabrication of Thin-Walled Metal Parts Under Open-Loop Con-
trol,” Int. J. Mach. Tools Manuf., 47(6), pp. 996–1002.
[13] Shwe, P. S., 2012, “Quantitative Analysis on SLS Part Curling Using EOS
5 Conclusions and Future Work P700 Machine,” J. Mater. Process. Technol., 212(12), pp. 2433–2442.
In this paper, the DMLS additive manufacturing process is ana- [14] Paul, R., Anand, S., and Gerner, F., 2014, “Effect of Thermal Deformation on
Part Errors in Metal Powder Based Additive Manufacturing Processes,” ASME
lyzed and sources of error in part fabrication are identified. Using J. Manuf. Sci. Eng., 136(3), p. 031009.
this analysis, design guidelines are formulated and two design [15] Allen, S., and Dutta, D., 1995, “Determination and Evaluation of Support Struc-
approaches are presented, which focuse on increasing part manu- tures in Layered Manufacturing,” J. Des. Manuf., 5(3), pp. 153–162.
facturability. In the first method, a feature graph-based design [16] Cloots, M., Spierings, A., and Wegener, K., 2013, “Assessing New Support
Minimizing Stratergies for the Additive Manufacturing Technology SLM,”
improvement technique which applies developed design guide- Solid Freeform Fabrication Symposium, Austin, Texas, pp. 631–643.
lines for creating more feasible designs is presented. The concept [17] Paul, R., and Anand, S., 2014, “Optimization of Layered Manufacturing Pro-
of feature graph is presented which can assist a designer to iden- cess for Reducing Form Errors With Minimal Support Structures,” J. Manuf.
tify features that may hinder manufacturability. In the second Syst., 36, pp. 231–243.
[18] Strano, G., Hao, L., Everson, R. M., and Evans, K. E., 2013, “A New Approach
method, a DFAM-constrained topology optimization-based design to the Design and Optimisation of Support Structures in Additive Man-
which integrates DFAM design rules with topology optimization ufacturing,” Int. J. Adv. Manuf. Technol., 66(9), pp. 1247–1254.
process is presented. The design methods presented in this paper [19] Vayre, B., Vignat, F., and Villeneuve, F., 2012, “Designing for Additive Man-
are not only useful for a novice designer who may not be aware of ufacturing,” Procedia CIRP, 3, pp. 632–637.
[20] Ponche, R., Kerbrat, O., Mognol, P., and Hascoet, J.-Y., 2014, “A Novel Meth-
the nuances of the DMLS process but also beneficial for experi- odology of Design for Additive Manufacturing Applied to Additive Laser Man-
enced designers who may find it difficult to integrate all the con- ufacturing Process,” Rob. Comput.-Integr. Manuf., 30(4), pp. 389–398.
straints into a unified design process. Moreover, there are features [21] Klahn, C., Leutenecker, B., and Meboldt, M., 2014, “Design for Additive
which may require tradeoffs and decisions from a designer based Manufacturing-Supporting the Substitution of Components in Series Products,”
Procedia CIRP, 21, pp. 138–143.
on functionality of the product. The proposed method can be [22] Seepersad, C. C., Govett, T., Kim, K., Lundin, M., and Pinero, D., 2012, “A
applied to determine these features and increase manufacturability Designer’s Guide for Dimensioning and Tolerancing SLS Parts,” 23rd Annual
while reducing costs. Also, a scoring scheme is presented which International Solid Freeform Fabrication Symposium, Austin, TX, pp. 921–931.
can quantify a design and its improvement from the viewpoint of [23] Adam, G. A. O., and Zimmer, D., 2014, “Design for Additive Manufacturing—
Element Transitions and Aggregated Structures,” CIRP J. Manuf. Sci. Technol.,
manufacturability. The methodology is demonstrated using case 7(1), pp. 20–28.
studies where improvement in manufacturability is observed. [24] Kerbrat, O., Mognol, P., and Hascoet, J.-Y., 2011, “A New DFM Approach to
Future work in this area includes further investigation of laser Combine Machining and Additive Manufacturing,” Comput. Ind., 62(7),
sintering process and development of more design guidelines. pp. 684–692.
[25] Rosen, D. W., 2007, “Computer-Aided Design for Additive Manufacturing of
Here, feature graph-based design improvement method considers Cellular Structures,” Comput.-Aided Des. Appl., 4(5), pp. 585–594.
only manufacturability issues based on part geometric analysis. [26] Joshi, S., and Chang, T. C., 1998, “Graph Based Heuristics for Recognition of
Moreover, there could be other design and topology constraints in Machined Features From 3D Solid Model,” Comput.-Aided Des., 20(2), pp. 58–66.

Journal of Manufacturing Science and Engineering JUNE 2017, Vol. 139 / 061007-13
[27] Lockett, H. L., and Guenov, M. D., 2005, “Graph-Based Feature Recognition [37] Brackett, D., Ashcroft, I., and Hague, R., 2011, “Topology Optimization For
for Injection Moulding Based on a Mid-Surface Approach,” Comput.-Aided Additive Manufacturing,” 24th Solid Freeform Fabrication Symposium, pp. 6–8.
Des., 37(2), pp. 251–262. [38] Gaynor, T. A., Meisel, N. A., Williams, C. B., and Guest, J. K., 2014,
[28] Gershenson, J., and Prasad, G., 1997, “Modularity in Product Design for Man- “Topology Optimization for Additive Manufacturing: Considering Maximum
ufacturability,” Int. J. Agile Manuf., 1(1), pp. 99–110. Overhang Constraint,” 15th AIAA/ISSMO Multidisciplinary Analysis and Opti-
[29] Changchien, S., and Lin, L., 1996, “A Knowledge-Based Design Critique Sys- mization Conference, Atlanta, GA, Jne 16–20, Paper No. AIAA 2014-2036.
tem for Manufacture and Assembly of Rotational Machined Parts in Concurrent [39] Mirzendehdel, A. M., and Suresh, K., 2016, “Support Structure Constrained
Engineering,” Comput. Ind., 32(2), pp. 117–140. Topology Optimization for Additive Manufacturing,” Comput. Aided Des., 81,
[30] Huang, S., Zhang, H., and Oldham, W., 1997, “Tolerance Analysis for Setup pp. 1–13.

Downloaded from https://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/139/6/061007/6273437/manu_139_06_061007.pdf by Velammal College OF Engineering & Techology user on 21 December 2019
Planning: A Graph Theoretical Approach,” Int. J. Prod. Res., 35(4), [40] Langelaar, M., 2016, “Topology Optimization of 3D Self-Supporting Structures
pp. 1107–1124. for Additive Manufacturing,” Addit. Manuf., 12(Part A), pp. 60–70.
[31] Brunetti, G., and Golob, B., 2000, “A Feature-Based Approach Towards an [41] Wang, M., and Qian, X., 2015, “Efficient Filtering in Topology Optimization
Integrated Product Model Including Conceptual Design Information,” Comput.- Via B-Splines,” ASME J. Mech. Des., 137(3), p. 031402.
Aided Des., 32(14), pp. 877–887. [42] Mercelis, P., and Kruth, J.-P., 2006, “Residual Stresses in Selective Laser Sin-
[32] Liu, X.-J., 2012, “Modeling of Additive Manufacturing Process Relevant Fea- tering and Selective Laser Melting,” Rapid Prototyping J., 12(5), pp. 254–265.
ture in Layer Based Manufacturing Process Planning,” J. Shanghai Jiaotong [43] Jacobson, D. M., and Bennett, G., 2006, “Practical Issues in the Application of
Univ. (Sci.), 17(2), pp. 241–244. Direct Metal Laser Sintering,” Solid Freeform Fabrication Symposium, Austin,
[33] Bendsøe, M. P., and Kikuchi, N., 1988, “Generating Optimal Topologies in TX, Aug. 14–16, pp. 728–739.
Structural Design Using a Homogenization Method,” Comput. Methods Appl. [44] Hussein, A., Hao, L., Yan, C., Everson, R., and Young, P., 2013, “Advanced
Mech. Eng., 71(2), pp. 197–224. Lattice Support Structures for Metal Additive Manufacturing,” J. Mater. Pro-
[34] Bendsøe, M. P., and Sigmund, O., 2003, Topology Optimization: Theory, cess. Technol., 213(7), pp. 1019–1026.
Method and Applications, Springer-Verlag, Berlin/Heidelberg. [45] Krol, T. A., Zaehb, M. F., and Seidela, C., 2012, “Optimization of Supports in
[35] Zhou, M., Fleury, R., Shyy, Y., Thomas, H., and Brennan, J., 2002, “Progress Metal-Based Additive Manufacturing by Means of Finite Element Models,”
in Topology Optimization With Manufacturing Constraints,” 9th AIAA/ISSMO Solid Freeform Fabrication Symposium, Austin, TX, pp. 707–718.
Symposium on Multidisciplinary Analysis and Optimization, Atlanta, GA, [46] Topçu, O., Taşcıo €
glu, Y., and Unver, € 2011, “A Method for Slicing CAD
H. O.,
Sept. 4–6, Paper No. AIAA 2002-5614. Models in Binary STL Format,” 6th International Advanced Technologies Sym-
[36] Guest, J., Prevost, J., and Belytschko, T., 2004, “Achieving posium, Elazıg, Turkey, May, 16–18, pp. 141–145.
Minimum Length Scale in Topology Optimization Using Nodal Design Varia- [47] Zhou, M., and Rozvany, G., 1991, “The Algorithm, C. O. C.,—Part II: Topolo-
bles and Projection Functions,”. Int. J. Numer. Methods Eng., 61(2), gical,Geometrical and Generalized Shape Optimization,” Comput. Methods
pp. 238–254. Appl. Mech. Eng., 89(1–3), pp. 309–336.

061007-14 / Vol. 139, JUNE 2017 Transactions of the ASME

You might also like