Journal of Computational Design and Engineering 5 (2018) 409–418

Contents lists available at ScienceDirect

Journal of Computational Design and Engineering

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

Visualization of potential sink marks using thickness analysis of finely

tessellated solid model
Masatomo Inui, Shunsuke Onishi, Nobuyuki Umezu ⇑
Department of Intelligent Systems Engineering, Ibaraki University, Hitachi, Japan

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

Article history: Sink marks are unwanted shallow depressions on the molded plastic surface caused by localized shrink-
Received 30 September 2017 age during the hardening process of injection molding. Sink marks appearing in the exterior impair the
Received in revised form 25 December 2017 aesthetic quality of the product. In this study, a novel method for extracting potential sink marks that
Accepted 12 February 2018
can occur on the part surface is proposed. The thicker portion of the part shrinks with a greater amount
Available online 21 February 2018
than that of the thinner portion. This difference in the shrinkage amount is the main cause of the sink
mark. In the plastic part design practice, engineers often check the thickness distribution to predict
Keywords:
potential sink marks in the part surface. Our method can be considered as an automated technique of
Sink marks
Shrinkage analysis
such manual inspection task. A polyhedral solid model of the part with sufficiently small triangles of
Sphere method nearly the same size is prepared. The amount of shrinkage at each polygon is estimated based on its
Solid modeling thickness and the shrinkage ratio of the part. The developed algorithm extracts the potential sink marks
CAD by analyzing the shrinkage distribution on the part surface.
Ó 2018 Society for Computational Design and Engineering. Publishing Services by Elsevier. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction appearing in the exterior impair the aesthetic quality of the pro-
duct and require costly redesigning when detected in the final trial
Injection molding is a popular manufacturing process to fabri- production stage.
cate plastic parts for consumer products such as digital cameras, A plastic part is typically designed to form a shell shape with a
printers, and other mechano-electrical appliances. In this process, constant thickness all over the surface. Designers often add rib
the plastic material is injected into a heated barrel where it is structures in the inner side of the part to increase the strength of
mixed and forced into a mold cavity to cool, harden, and form the part. Certain boss shapes with holes are also placed to connect
the part shape. Injection molding is considered as a complex tech- to the pins of other parts. While placing ribs and bosses, the
nology that can possibly involve many production problems. Thus, designer must consider the change in thickness at the connection
the part designers must consider not only the function and aes- between the ribs or bosses and the shell shape. Fig. 1 illustrates a
thetic properties of the product but also the various physical con- section of a boss connection. The thicker sections are sandwiched
ditions of the molding process to avoid the occurrence of molding between thinner sections at the connection region. They pull the
defects that can negatively impact the quality of the products. surface of the connection region inward during the hardening pro-
‘‘Sink marks”, which appear on the outside surface of the part, cess generating sink marks.
are one form of such molding defects. When the plastic part Reducing the number of parts is a basic design strategy to lower
shrinks during the hardening process of the injection molding, the production cost. Accordingly, a single part is expected to serve
the thicker portion of the part shrinks with a greater amount than multiple functions leading to more complex design configurations
that of the thinner portion. Thus, the surface of the thicker portion that often have many bosses and ribs. Detecting the occurrence of
surrounded by thinner portions often has a localized depression potential sink marks on the surface of such complex parts is a dif-
due to the difference in the shrinkage amount. Sink marks ficult task even for experienced designers. In this study, we pro-
pose a novel method for visualizing the potential sink marks on
the surface of a solid plastic part model. The surface of the model
Peer review under responsibility of Society for Computational Design and
is assumed to be finely tessellated with sufficiently small polygons.
Engineering. The proposed method calculates the amount of shrinkage of each
⇑ Corresponding author. polygon. Then, the potential sink marks are extracted by analyzing
E-mail addresses: [email protected] (M. Inui), 17nm905l@ the shrinkage distribution on the part surface.
vc.ibaraki.ac.jp (S. Onishi), [email protected] (N. Umezu).

https://doi.org/10.1016/j.jcde.2018.02.003
2288-4300/Ó 2018 Society for Computational Design and Engineering. Publishing Services by Elsevier.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
410 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418

Fig. 1. Thickness distribution at the boss connection part and potential sink marks.

In the plastic part design practice, engineers often check the

thickness distribution to predict potential sink marks in the part
surface. The developed method can be considered as an automated
technique of such manual inspection task. This method is not accu-
rate; however, it is simple and fast, and can be used as a supporting
tool in the early design stages. The software using this method can
visualize the result in a shorter time; thus, the designers can
invoke the software at any stage of the shape design process to
detect any potential sink marks. Unlike the manual methods that
rely on the experience, proposed method does not overlook the Fig. 2. Thickness computation using sphere method.
surface portions with potential sink marks. The visualization result
can be useful in efficiently improving the part shape. which each voxel records the distance between its center and the
In the next section, some related studies on sink mark analysis object boundary (Jones, Bærentzen, & Sramek, 2006). Lu et al. pro-
and thickness computation are briefly reviewed. In Section 3, the posed a distance-field-based thickness evaluation method to visu-
developed sink mark extraction algorithm is illustrated. The exper- alize thicker sections of 3D objects (Lu, Rebello, Miller, Kinzel, &
imental computation results are discussed in Section 4, and the Yagel, 1997). Further, a thickness visualization system for a solid
conclusions drawn are summarized in Section 5. model was developed in a previous study (Inui, Umezu,
Wakasaki, & Sato, 2015; Inui, Umezu, & Kobayashi, 2014), whereby
2. Related studies the thickness at a point on the object surface is determined using
the sphere method and a distance field.
Many research studies have been conducted to understand the Some commercial CAD systems (3DS [Internet], 2017;
mechanism of the sink mark generation and to determine the loca- SOLIDWORKS [Internet], 2017) provide thickness visualization
tion where sink marks may potentially occur on the part surface functions. They utilize the thickness evaluation software GeomCa-
before the actual production. For example, Shi and Gupta studied liper (GeomCaliper [Internet], 2017) as an add-on function. Accord-
a method to predict the sink mark depth by analyzing the localized ing to the developer’s report (Sinha, 2007), GeomCaliper employs a
shrinkage using finite element method for flow simulation (Shi & uniform grid and k-d tree to calculate the effective thickness. Inui
Gupta, 1998). Some commercial software that simulates the flow et al. developed an iterative algorithm, namely, the shrinking
process in injection molding provides a sink mark estimation func- sphere algorithm, to calculate the MIS at a specific surface point
tion, such as Moldflow (Autodesk [Internet], 2017); however, their of a polyhedral object (Inui, Umezu, & Shimane, 2015).
technical details have not been published. These flow-simulation-
based methods take significant time to perform the analysis, typi- 3. Sink mark visualization using shrinkage distribution analysis
cally more than an hour. Beiter and Ishii statistically analyzed the
visibility of the sink marks based on field studies, and proposed In this study, a novel method to visualize potential sink marks
guidelines for the appropriate texture depth required to conceal on the plastic part surface is proposed. This method uses the
the sink marks (Hayden, 2006). Gruber et al. developed a method shrinkage distribution analysis to determine the locations where
to rapidly evaluate the visual perceptibility of the sink marks sink marks can potentially form on the part surface. When consid-
(Gruber et al., 2014). ering a small surface region of a plastic part, sink marks are likely
The method proposed in this study predicts the location and to develop if the surface region is in rather flat surface and the
depth of the potential sink marks based on analyzing the thickness amount of the shrinkage at the region is larger than that of the
of the section. There are two main methods to define the thickness nearby surface area. For example, a surface region P is a part of a
of a 3D object, namely the ray method and the sphere method sink mark if its shrinkage is larger than the shrinkage of the nearby
(Sinha, 2007; Cocks, 1983; Beiter & Ishii, 1991). In the sphere regions Q and Q0 , as shown in Fig. 3. Shrinkage distribution analysis
method, which is the most widely used method, the thickness t is applied to detect sink marks on such regions of the plastic part
at point p on the object surface is given by the diameter t of the surface.
maximum inscribed sphere (MIS) S touching p (see Fig. 2) Sinha, Fig. 4 shows a processing flow of our sink mark visualization
2007. A distance field is a voxel-based shape representation, in algorithm. The input data of the proposed algorithm is a polyhedral
 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418 411

Fig. 3. Typical shrinkage distribution to form sink marks.

temperature and pressure during the injection molding hardening

process influence the shrinkage ratio. In the following computa-
tion, the mid value of the range shown in the table is used as the
representative shrinkage ratio for each material, for example r =
1.75% for poly propylene (PP). Sink mark visualization result
obtained by our method is not accurate because a fixed value is
used as the shrinking ratio and the temperature and pressure
impacts are ignored in this computation model. However, this
method is simple and fast compared to conventional flow-
simulation-based analysis. Plastic part designer can use the
method frequently in the design process so that the plastic part
optimized in a producibility view point can be generated rapidly.

3.1. Conversion to a model with small surface triangles (step 1)

The developed algorithm analyzes the distribution of the

shrinkage using the shrinkage values associating with polygons.
Thus, better visualization results of the sink mark can generally
be obtained when part models with finely tessellated polygons
are used. Most commercial CAD systems provide a function to out-
Fig. 4. Processing flow of our sink mark visualization algorithm. put a solid model as a group of triangular polygons with a specified
polygon size. For example, computer aided three-dimensional
CAD model of a plastic part, such as the model in the STL format. In interactive application (CATIA) system (3DS [Internet], 2017) users
the first step of the algorithm, the surface polygons of the model can control the polygon size of the output tessellated model using
are converted into sufficiently small triangles with almost the a parameter referred to as ‘‘step.” The width of sink marks typically
same size. In the second step of the algorithm, the thickness of ranges between 1 and 5 mm, as shown in Fig. 3. Therefore, the step
each triangle is computed at its center of gravity. The amount of size of the surface polygon of the part must be smaller than this
shrinkage at a surface point is basically proportional to the thick- size. A step size between 0.1 mm and 0.5 mm is used in the current
ness at that point. The proportional factor between the amount implementation.
of the shrinkage and the unit length, which can be defined with To accurately detect the potential location of sink marks on a
each material, is referred to as the shrinkage ratio. The amount model with large polygons, a software was developed to finely sub-
of shrinkage of a triangle is thus determined based on the thickness divide the part surface into a set of small triangles with almost the
of the section and the shrinkage ratio. In the final step of the algo- same size. Fig. 5 illustrates the basic processing steps of the devel-
rithm, the potential sink marks are extracted by analyzing the oped method using a subdivision example of a two-dimensional
shrinkage distribution on the part surface. The extraction result polygon. An axis-aligned bounding box holding the part shape
is given with the surface polygons corresponding to the detected within is prepared in the first step. A spatial grid with axis-
sink mark location painted by a unique color. aligned lines in a predetermined interval is generated in the box,
Table 1 represents the shrinkage ratio for various plastic mate- as shown in Fig. 5(a). The surface points of the part are sampled
rials (MiSUMi [Internet], 2017). As shown in the table, the shrink- as the intersection points between the grid lines and the surface
age ratio is not a single value but a range of values. The of the shape in the second step (see Fig. 5(b)). The final shape of
the finely tessellated object is obtained by properly connecting
Table 1 the sampled points (see Fig. 5(c)).
Shrinking ratio r of plastic materials (MiSUMi [Internet], 2017).
The axis-aligned spatial grid with equal interval allows organiz-
Plastic material Shrinking ratio (%) ing the cell structure with cubic cells. Such a cubic cell is illustrated
Poly propylene (PP) 1.0–2.5 in Fig. 6. As the cell is located in the middle of a model surface, four
Acrylonitrile Butadiene Styrene polymer (ABS) 0.4–0.9 segments of the cell have intersection points with the surface. The
Polycarbonate (PC) 0.5–0.7 small black points shown in the Fig. 6 denote the intersection
Poly ethylene terephthalate (PET) 0.2–0.4
points. The small triangles with almost the same size, which
412 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418

Fig. 5. Processing steps of the developed subdivision algorithm.

Step 2.2: As an initial candidate for the target sphere, a sphere

with a sufficiently large radius touching p is defined. The MIS
touching p is obtained by repeatedly shrinking the initial sphere
in the following manner.

 The candidate sphere has some inclusions and/or intersections

with the surface polygons of the object. The developed algo-
rithm recognizes any case in which the sphere touches or con-
tains a surface polygon as an intersection.
 For each intersecting polygon fi except f, a point pi on the poly-
gon that is closest to the center point c of the sphere is com-
puted. In the case shown in Fig. 7(a), the candidate sphere
Fig. 6. Placement of small triangles in a cubic cell using marching cubes algorithm.
intersects with five polygons. Five points, p0, p1, p2, p3, and
p4, are computed as the closest points on each of the intersect-
represent a fragment of the model surface, are obtained in the cell ing polygons. These computations for each intersecting polygon
by properly connecting the intersection points with the marching are mutually independent; therefore, they can be performed
cubes algorithm (Lorensen & Cline, 1987). using the graphics processing unit (GPU) in a parallel manner
In this method, the triangles size of the result shape is con- (NVIDIA: CUDA Compute Unified Device Architecture, 2007).
trolled by changing the resolution of the spatial grid or the interval  The next candidate sphere touching p is obtained based on the
distance between the grid lines. In our study, the interval of the computed points. For each pi, a new sphere whose center point
grid lines is set to control the size of the result triangles at less than is on the ray and whose surface passes through both p and pi is
0.5 mm (approximately 0.2 to 0.3 mm). As shown in Fig. 5(c), the computed (see Fig. 7(b)). From the computed spheres, the
developed tessellation algorithm cannot maintain the sharp cor- sphere with the smallest radius (bold sphere passing through
ners in the original model, thus small triangles (small segments p4 in the figure) is selected as the next candidate. Hence, the
in the figure) are often inserted in the corners. As the sink marks new sphere must have a smaller radius than the previous
typically appear on a flat surface, not at the sharp corner, the sphere.
inserted triangles do not affect the sink mark extraction results.
The sphere-shrinking process described above is iterated until
the diameter change between the previous sphere and the new
3.2. Computation of thickness (step 2)
sphere becomes sufficiently small. In the current implementation,
the process terminates when the diameter change becomes smal-
Our shrinking sphere algorithm (Inui et al., 2015) is used in this
ler than 1/100 of the model size. The obtained sphere corresponds
system to calculate the thickness of the polyhedral object. Thick-
to the MIS touching p. The diameter of the sphere is considered
ness computation is achieved using the following three-step
equal to the thickness of polygon f at p. See the original paper
manner:
(Inui et al., 2015) for the implementation details. After calculating
Step 1: For each polygon f, select a point on the polygon where
the thickness, the shrinkage of the polygon is computed using the
the thickness of the object is measured. The center of gravity of the
shrinking ratio and thickness of the polygon.
polygon is selected as the point for measuring the thickness in our
current implementation.
Step 2: The thickness of all polygons is computed using the 3.3. Extraction of sink marks (step 3)
sphere method. The shrinking sphere algorithm is adopted in the
computation. In the final step of the algorithm, the distribution of the shrink-
Step 3: The calculated thickness is stored in association with its age at the surface polygons are analyzed and a set of polygons cor-
corresponding polygon f. responding to sink marks are extracted. Such polygons are painted
As Steps 1 and 3 involve straightforward processes, only the with a unique color and the potential sink marks locations on the
shrinking sphere algorithm used in Step 2 is explained in details. part surface are visualized. A sink mark is a shallow depression
In this step, the following processing is repeated for a point p on or dimple appearing on a rather flat surface. It is generated when
each surface polygon f of the object to obtain the MIS touching p. a small surface portion with large shrinkage is sandwiched or sur-
Step 2.1: Cast a ray from p along the opposite direction of the rounded by surface areas with small shrinkage as shown in Fig. 3.
surface normal at p. The center points of all spheres touching p Thus, a polygon fi is considered in a sink mark if it satisfies the fol-
must be located on the ray. lowing three conditions:
 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418 413

Fig. 7. Sphere shrinking process (Inui et al., 2015).

1. The amount of the shrinkage of fi is larger than the amount of mentation, h should be larger than 165.0° (see Fig. 8(b)). This
the shrinkage of the two polygons fj and fk adjacent to fi (see condition is used to select a polygon fi that sandwiched or sur-
Fig. 8(a)). More precisely, the difference between the shrinkage rounded by two adjacent polygons fj and fk.
value of fi and the value of fj (or fk) is larger than a predefined 3. The normal vectors of three polygons fi, fj, and fk are almost par-
threshold value. In our current implementation, 0.001 mm is allel, therefore they are in rather flat surface. In the current
used as the threshold value. This value is variable according implementation, the angular difference of the three normal vec-
to the sink mark detection purpose. When a larger value is spec- tors ni, nj, and nk of fi, fj, and fk, respectively, must be less than
ified, deep marks are only detected. Very shallow marks can be 25.0° (see Fig. 8(c)).
recognized when a smaller value is given. The adjacent range
(red dotted circle in the figure) is defined by the distance from In this method, polygons adjacent to fi must be selected before
fi. In the current implementation, 1–2 mm is used as a distance assessing the conditions. As the STL models do not record adja-
for defining the adjacency range. cency information between polygons, the adjacent polygons must
2. A vector vij from the center of gravity of polygon fi to the center be extracted from a number of the triangle data in that file. In
of gravity of fj, and another vector vik from the center of gravity the current implementation, triangle fj is considered as an adjacent
of fi to the center of gravity of fk are considered. The angle h triangle to fi if the distance between fi and fj is less than or equal to
between vij and vik must be close to 180°. In the current imple- the defined adjacency range d.

Fig. 8. Three conditions for detecting polygons fi corresponding to sink marks.

414 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418

To formulate the appropriate data that can efficiently detect the  If node nd is a leaf node in the tree, a set of polygons within
adjacent triangles around a specific triangle fi, the surface polygons AABB of the node is selected as candidates of the adjacent poly-
of the input model are classified into small groups according to gons of fi.
their proximity. This classification proceeds according to the hier-  Otherwise, traverse the child nodes of nd with the AABB of fi.
archical structure of the axis-aligned bounding box (AABB)
(Ericson, 2004; Moller & Haines, 1999). When n polygons are con- After the tree traversal, the adjacency relationship between fi
sidered to form the model surface, an AABB that tightly confines and obtained candidate polygons is checked and the polygons
the polygons is defined by measuring the coordinate ranges of whose distance to fi is less than or equal to d are selected.
the polygons in the x-, y-, and z-directions. One root AABB is
defined to hold all the polygons of the model. The polygons in 4. Computational experiments
the AABB are sorted and classified into two groups. For each group,
a smaller AABB is formed and registered as a descendant of the Using the algorithms explained in Section 3, we implemented a
original AABB. The process of defining descendant AABBs is iter- potential sink mark visualization system based on Microsoft Visual
ated until the number of polygons in a group becomes less than C++ and CUDA 8.0. The conducted experiments employed a Win-
or equal to a predetermined number nmax, and a binary AABB tree dows 8.1 64-bit PC (Intel Core-i7-5960X 3.0 GHz CPU, 32 GB main
is obtained. nmax is set to be 4 in the current implementation. This memory, NVIDIA GTX-980 GPU).
number is determined based on numerical experiments. The developed system was applied to 5 plastic part models,
Polygons whose distance to a given polygon fi are less than or namely, Models A, B, C, D, and E, as shown in Figs. 9–13. Each figure
equal to d are obtained by traversing the hierarchical AABB tree illustrates a part with complex rib and/or boss structure in (a). A
in a depth-first manner. An AABB tightly contains fi within is color map shown in (b) represents the thickness computation
defined. The box is expanded by an adjacency distance d by shifting results of the reversed side of the rib structure. The red curves
the six rectangles in the box to their outward directions by d. This appearing in (c) represent the potential sink marks extracted by
box is given to the root node of the AABB tree. At each node nd in the algorithm. In the color map, red is assigned to polygons with
the tree, the AABB containing fi and an AABB corresponding to nd is zero thickness and blue is assigned to polygons with the maximum
compared. If they do not overlap, the polygons contained within thickness of the model. Faint color differences appear in figures (b),
the AABB at the node are not in the adjacent range of fi and any fur- such as the faint yellow lines in Fig. 10(b), which typically corre-
ther processing from the node is stopped. If they have potential spond to the thickness difference in the connecting part of the
overlap, then; rib structure on the other side of the part. As such connecting parts

Fig. 9. Sample plastic part Model A with rib structure (a), thickness color map of the reversed side of the part (b), and extracted possible sink marks in the reversed side (c).
 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418 415

Fig. 10. Sample plastic part Model B with rib structure (a), thickness color map of the reversed side of the part (b), and extracted possible sink marks in the reversed side (c).

Fig. 11. Sample plastic part Model C with rib structure (a), thickness color map of the reversed side of the part (b), and extracted possible sink marks in the reversed side (c).

are thicker than their adjacent part, these lines usually correspond tic part, positive impressions were received as they considered the
to the potential sink marks shown in (c). When these visualization developed visualization software to be useful in minimizing injec-
results were presented to engineers involved in the design of plas- tion mold design mistakes.
416 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418

Fig. 12. Sample plastic part Model D with rib and boss structure (a), thickness color map of the reversed side of the part (b), and extracted possible sink marks in the reversed
side (c).

Fig. 13. Sample plastic part Model E with boss structure (a), thickness color map of the reversed side of part (b), and extracted possible sink marks in the reversed side (c).

Fig. 14 shows a comparison between a sink mark appearing in model of the part in (c). Fig. 14(b) shows a color map representing
an actual plastic part (an area surrounded by a dotted rectangle the thickness distribution of the part. A red line appearing in the
in (a)) and our sink mark visualization result with a polyhedral part surface in (c) represents a potential sink mark extracted by
 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418 417

Fig. 14. A sink mark appearing in a plastic part and our analysis results.

Table 2
Required computation time for extracting potential sink marks.

Model Number of polygons of Required time for Number of polygons obtained by Required time for computing Required time for extracting
original model converting model (s) the conversion thickness (s) sink marks (s)
A 2,364 2.07 2,166,732 100.10 38.96
B 1,708,000 N/A 1,708,000 206.96 29.39
C 1,794 6.58 4,541,942 704.33 86.09
D 3,048 9.29 3,736,372 1216.02 65.79
E 15,700 8.05 2,861,140 691.03 53.35

our software. As shown in the figure, our software can properly In this study, a novel method for extracting possible sink marks
extract the surface area corresponding to the sink mark. on the part surface is proposed. The input data in this method is a
The required computation time for extracting and visualizing polyhedral model of the part. Surface polygons of the model are
sink marks for each model are listed in Table 2. The required time converted into a set of sufficiently small triangles of nearly the
for tessellating the input model is given in the third column. The same size. The amount of shrinkage is directly proportional to
required time for Model B tessellation is canceled because a finely the thickness of the part. The developed algorithm uses the sphere
tessellated model is given as input model for this case. As shown in method to calculate the thickness of all polygons on the part sur-
Table 2, the thickness computation consumed significantly much face, and it then extracts the potential sink marks by analyzing
time in the current implementation, as the thickness computation the shrinkage distribution on the part surface. Three conditions
of some parts required 10–20 min. Even though this computation related to the adjacent polygons are evaluated in the analysis. To
time is much shorter compared to that required by the flow- accelerate the selection of the adjacent polygons, hierarchical tree
simulation-based analysis, the developed system is still not fast structure of AABBs holding close polygons is utilized.
enough to be used interactively during the part design process. The proposed system is in the experimental stage. The system
Improving the thickness computation speed is indispensable. will be field tested in an actual plastic part design process. The
future studies will focus on further improving the developed sys-
5. Conclusions tem based on the comments and requests to be received from
the designers.
Injection molding is a complex technology with many potential
production problems. Part designers must consider not only the References
function and aesthetic properties of the product but also the vari-
ous physical conditions of the molding process to avoid molding Autodesk [Internet]. San Rafael, CA [cited 2017 Sep 20]. Available from: <https://
www.autodesk.com/products/moldflow/overview>.
defects that can impact the quality of the produced parts. Sink
Beiter, K. A., & Ishii, K. (1991). Geometry-Based Index for Predicting Sink Mark in
marks, which have been considered as unwanted shallow depres- Plastic Parts, Technical Report, ERC/NSM-P-91-61, Engineering Research Center
sions or dimples, are caused by the difference in the amount of for Net Shape Manufacturing, The Ohio State University.
the shrinkage during the hardening process of the injection mold- Cocks, D. (1983). Die cast components: Aid to efficient design. London: Zinc
Development Association.
ing. Costly redesigning is required when they are detected in the DS [Internet]. Dassault Systemes, France [cited 2017 Sep 20]. Available from:
trial production stage. <http://www.3ds.com>.
418 M. Inui et al. / Journal of Computational Design and Engineering 5 (2018) 409–418

Ericson, C. (2004). Real-time collision detection. Morgan-Kaufmann. Lorensen, W. E., & Cline, H. E. (1987). Marching cubes: A high resolution 3D surface
GeomCaliper [Internet]. Geometric [cited 2017 Sep 20]. Available from: construction algorithm. Computer Graphics (Proceedings of ACM SIGGRAPH), 21
<http://geomcaliper.geometricglobal.com>. (4), 163–169.
Gruber, D. P., Macher, J., Haba, D., Berger, G. R., Pacher, G., & Friesenbichler, W. Lu, S. C., Rebello, A. B., Miller, R. A., Kinzel, G. L., & Yagel, R. (1997). A simple
(2014). Measurement of the visual perceptibility of sink marks on injection visualization tool to support concurrent engineering design. Computer-Aided
molding parts by a new fast processing model. Polymer Testing, 33, 7–12. Design, 29(10), 727–735.
Hayden, K. F. (2006). Determining the probability of the visual detection of sink marks MiSUMi [Internet]. Tokyo [cited 2017 Sep 20]. Available from: <http://www.
on differently textured injection molded products PhD Thesis. Western Michigan misumi-techcentral.com/tt/en/mold/2010/02/028-molding-shrinkage-ratios-
University. of-major-plastic-materials.html>.
Inui, M., Umezu, N., Kobayashi, K. (2014). Parallel distance field computation with Moller, T., Haines, E. (1999). Real-Time Rendering, AK Peters.
GPU and its application for evaluating part thickness. In Proc of ISCIE/ASME NVIDIA: CUDA Compute Unified Device Architecture, Programming Guide (2007).
International Symposium on Flexible Automation. ISFA2014-40. Shi, L., & Gupta, M. (1998). A localized shrinkage analysis for predicting sink marks
Inui, M., Umezu, N., & Shimane, S. (2015). Shrinking sphere: A parallel algorithm for in injection-molded plastic parts. Journal of Injection Molding Technology, 2(4),
computing the thickness of 3D objects. Computer-Aided Design and Applications, 149–155.
13(2), 199–207. Sinha, B. (2007). Efficient Wall Thickness Analysis Methods for Optimal Design of
Inui, M., Umezu, N., Wakasaki, K., & Sato, S. (2015). Thickness and clearance Casting Parts, Engineering Design. Available from: <http://geomcaliper.
visualization based on distance field of 3D objects. Journal of Computational geometricglobal.com/images/file/EfficientWallThicknessAnalysis GeomCaliper.
Design and Engineering, 2(3), 183–194. pdf>.
Jones, M. W., Bærentzen, J. A., & Sramek, M. (2006). 3D distance fields: A survey of SOLIDWORKS [Internet]. Waltham, MA [cited 2017 Sep 20]. Available from: <http://
techniques and applications. IEEE Transaction on Visualization and Computer www.solidworks.com>.
Graphics, 12(4). 881-599.