Computer-Aided Design 83 (2017) 64–79

Contents lists available at ScienceDirect

Computer-Aided Design
journal homepage: www.elsevier.com/locate/cad

A survey on 3D CAD model quality assurance and testing tools

Carmen González-Lluch a, *, Pedro Company b , Manuel Contero c , Jorge D. Camba d ,
Raquel Plumed a
a
Department of Mechanical Engineering and Construction, Universitat Jaume I, Av. de Vicent Sos Baynat, s/n 12071, Castellón, Spain
b
Institute of New Imaging Technologies, Universitat Jaume I, Castellón, Spain
c
I3B, Universitat Politècnica de València, Camino de Vera s/n, 46010, Valencia, Spain
d
Gerald D. Hines College of Architecture and Design, University of Houston, 4200 Elgin St, Houston, TX 77204-4000, USA

article info a b s t r a c t
Article history: A new taxonomy of issues related to CAD model quality is presented, which distinguishes between
Received 7 November 2015 explicit and procedural models. For each type of model, morphologic, syntactic, and semantic errors are
Accepted 18 October 2016 characterized. The taxonomy was validated successfully when used to classify quality testing tools, which
are aimed at detecting and repairing data errors that may affect the simplification, interoperability, and
Keywords:
reusability of CAD models.
Quality assurance
The study shows that low semantic level errors that hamper simplification are reasonably covered in
CAD model quality testing
explicit representations, although many CAD quality testers are still unaffordable for Small and Medium
Enterprises, both in terms of cost and training time. Interoperability has been reasonably solved by
standards like STEP AP 203 and AP214, but model reusability is not feasible in explicit representations.
Procedural representations are promising, as interactive modeling editors automatically prevent most
morphologic errors derived from unsuitable modeling strategies. Interoperability problems between
procedural representations are expected to decrease dramatically with STEP AP242. Higher semantic
aspects of quality such as assurance of design intent, however, are hardly supported by current CAD
quality testers.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction where CAD models contain errors or anomalies (in most cases,
unnoticeable to the user) that can hinder simplification, interoper-
According to the SASIG Guidelines for the Global Automotive In- ability, and/or reusability. Although in academic circles, Modeling
dustry, ‘‘good product data quality means providing the right data Quality Testing is considered solved by a number of scholars, it has
to the right people at the right time [1]’’. Because the design field been reported that the implementation and practical application
is so inherently rich in information, it is necessary to implement of these solutions in industrial settings is far from trivial [4].
systems that can efficiently and securely track, control, manage, Therefore, Model Quality Testing remains an open problem.
and share these data. A number of MQT tools such as CAD Checker by Excitech [5,6]
Product data is an umbrella term that includes many different are designed to validate the correctness of 2D CAD drawings. While
types of information. In this paper, we specifically focus on 3D certainly important in industries that have not yet moved to a pa-
CAD models. In this context, CAD model quality is a key concept, perless workflow paradigm based on 3D virtual models, these tools
as the quality of a manufactured product depends on the quality are out of the scope of our study. Instead, we focus on scenarios
of its design processes, which then depend on the quality of their where 3D CAD models are used as the main vehicle for the delivery
data [2]. High quality product data is essential, as low quality often of product information and as the central pieces around which
delays product development, and can negatively impact the overall downstream CAD/CAM/CAE applications are structured. For exam-
quality of the final product [1,3]. ple, design engineers involved in Finite Element Analysis, Rapid
In this paper, we provide an examination of CAD Model Quality prototyping, Numerical Control, and Data Exchange activities often
Testing (MQT) tools and techniques, which are used in situations need complete, correct, and clean CAD models, so derived models
can be efficiently generated while ensuring product quality and
minimizing production costs and time to market.
*Corresponding author.
E-mail addresses: [email protected] (C. González-Lluch), [email protected]
This paper defines a new taxonomy of issues related to CAD
(P. Company), [email protected] (M. Contero), [email protected] (J.D. Camba), model quality. For validation purposes, the taxonomy was used
[email protected] (R. Plumed). to classify currently available MQT tools, thus determining which

http://dx.doi.org/10.1016/j.cad.2016.10.003
0010-4485/© 2016 Elsevier Ltd. All rights reserved.
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 65

aspects of quality are reasonably addressed, and which remain means a clear distinction between the data models (applications
open problems. and logical layer) and the file format for data exchange (physical
The paper is organized as follows: Section 2 describes the layer defined as in part 21) [15]. This is important when consider-
basic terminology used in CAD Model Quality Testing in relation ing downstream applications (such as CAM and CAE systems) that
to the underlying structure of the domain; Section 3 presents a do not support the native modeler [9,16] as their input and require
systematic taxonomy of the techniques, which is connected to the a direct translation or a neutral format. In this context, master
commercially available CAD Model Quality Testing Tools discussed models are also called primary views, and exported o derived mod-
in Section 4. Finally, Section 5 concludes the paper by highlighting els, secondary views [17–19].
open research domains. Strategic CAD knowledge is the ability to recognize that design
The paper is extensive as it includes a survey of quality prob- choices made during the development of the master model may
lems in CAD models, introduces a new taxonomy for classifying drastically determine how easy or how difficult it is to perform sub-
these problems, and also includes a survey of QMT products. sequent design changes to the model [20–22]. Naturally, incorrect
Although this may seem far too much at once, we strongly be- or inefficient master models should always be avoided, but even
lieve that a global analysis was required before subsequent sec- when effective strategies are used, master models may still contain
torial analysis can be efficient. This is because the paper argues errors. It is in this type of situations where Model Quality Testing is
against the general believe that MQT tools are an academically required.
solved problem and only incremental improvements are required. It is usually the secondary models which need cleaning up, as a
Instead, the thesis of this paper is that current MQT tools are result of data transfer problems. But discrepancies between geom-
mostly aimed at homogenizing the vast amount of documents pro- etry and functionality result in errors in the higher levels of quality
duced and shared by large OEM’s, and thus are primarily aimed at of CAD models. It follows if we accept that semantics focuses on the
preventing easily solvable low-semantic level mistakes and inco- relationship between signifiers (elements of geometry in our case),
herencies, while we foresee that they will only become valuable for and what they stand for (the functionality of one particular shape,
other market segments (like SME’s) if document homogenization used to solve a certain design problem). Therefore, master models
ceases to be prevalent over conveying design intent. may also need cleaning up to remove their semantic level errors,
if strategic CAD knowledge is pursued. Thus, we agree that CAD
2. Terminology
Model Quality Testing is an activity that involves identifying ‘‘dirty
clean-up problems’’ in a master CAD model. Model Quality Technol-
Product data refers to all data involved in the design and
ogy enables designers to identify, locate, and even resolve model
manufacturing of a product [7]. Product Data Quality (PDQ) is a
integrity problems before the file leaves the CAD system [2]. Early
measurement of the accuracy and appropriateness of these data
advances in Model Quality Assurance Systems were summarized
combined with the timeliness with which they are provided to the
by [7].
stakeholders who may need them [1,8].
Resolving model integrity problems requires model repair/
A simple product data tool is the native modeler, which is the
healing, which entails making slight adjustments to the geome-
CAD system used to create a native model [9], also called master
try to remove anomalies [7,8,23]. Healing and repairing involves
model by the Automotive Industry Action Group (AIAG) [2].
processing a model with undesirable artifacts and creating a new
A CAD model is a mathematical representation of a product used
version that is free from these errors [24]. There is no general
to explain and predict the behavior of such product before it is
agreement on whether model healing is beneficial as (1) repairing
built. Representations are organized collections of associated data
secondary views without propagating the changes to the primary
elements collected together for specific uses [10]. They are the
data sources for procedures that compute useful properties of view compromises data integrity, and (2) repairing a local problem
objects [11]. A CAD model is stored in a document, which is a fixed in one area of the model may create new local problems elsewhere
and structured amount of information that can be managed and or affect the design intent of the whole part [7,23]. This controversy
interchanged as a unit between users and systems [12]. is out of the scope of our paper, but certainly a topic that needs
CAD models are represented according to languages, which to be considered in related studies. It is generally agreed that
must conform to a certain representation scheme. Two represen- model repair/healing is useful for homogenizing vast amounts of
tation schemes currently dominate most Mechanical CAD systems: documentation used by OEM’s and their network of suppliers.
explicit and procedural. Representations are explicit (also known as However, if focus is on detecting higher level errors to further
declarative or evaluated) when their details are immediately avail- improve quality of CAD models, homogenization becomes just a
able without the need for any calculations. Representations are small part of the problem. In this context, where homogenization
procedural (also known as generative or history-based), if they are is not prevalent over design intent, some repair strategies may be
described in terms of a sequence of procedures (which may include hiding or even inverting the original design intent.
the solution to constraint sets). Finally, hybrid representations In order to describe and classify the types of quality loss that can
result from combining both explicit and procedural representation be automatically detected and repaired, it is important to further
methods [13]. clarify two aspects: the type of representation that describes the
However, representation of geometrical entities is subject to model, and the type of change that causes the loss of quality.
another source of potential discrepancies. Most geometrical en-
tities may be defined by different sets of parameters. This fact is 2.1. Types of representations
acknowledged in Section 4.2.6 of ISO 10303-42:1994 [14] where
in the case that an item of geometry could be defined in more than Two representation schemes currently dominate most Mechan-
one way, a ‘‘preferred form’’ or ‘‘master representation’’ should be ical CAD systems (MCAD): Boundary Representation (B-Rep), and
nominated. It is recommended that parametrization, domain and history-based parametric feature-based models [19,25].
results of evaluation to be derived from the master representation. This distinction is not always accepted as, actually, a feature-
In a note at the end of this section, the standard gives a clear based CAD model is made of two interconnected representations:
warning: ‘‘Any use of the other representations is a compromise B-Rep and feature-based. But we follow the item 3.7.28 of the
for practical considerations’’. standard ISO10303-108:2005(E) [13], which defines procedural
Modern product data standards such as ISO 10303 follow the models ‘‘as opposed to an explicit or evaluated model which cap-
ANSI/SPARC three-layer architecture for database systems. This tures the end result of applying those procedures’’. An explanatory
66 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

note follows: ‘‘NOTE. Although procedural models are outside the (made of patches), and meshes (made of smaller homogeneous
scope of this part of ISO 10303, they are defined here to make faces). Subdivision surfacing is a method where a coarse piecewise-
an important distinction between two fundamentally different linear polygon mesh is recursively refined by subdividing each
modeling approaches. The present resource is intended to be com- polygonal face into smaller faces that better approximate the de-
patible with ISO 10303-55, which provides representations for sired smooth surface. Alternatively, mesh reduction techniques are
the exchange of procedurally defined models’’. Furthermore, when applied to dense triangle meshes to build simplified approxima-
explicit models are defined (item 3.7.16) another explanatory note tions while retaining important topological and geometric char-
is added: ‘‘In the case of product shape models, an explicit (or acteristics of the model. The reduced mesh is interpolated with
evaluated) shape model is a fully detailed model of the boundary piecewise-algebraic surface patches which approximate the origi-
representation or some related type, as defined in ISO 10303-42. nal points [30].
More specifically, an explicit model is a model that is not of the Solid models are defined by the closed shell that encloses its
procedural or hybrid type, which may contain little or no explicit material. In this context, the term closed means that the surface
geometry’’. Finally, the item 3.7.19 of the standard ISO10303- of the model divides the entire space into disjoint internal and
108:2005(E) defines hybrid models, which are useful because one external volumes [36]. Closed B-Rep and meshed models provide
of their representations is better for some actions (like speeding a complete representation of a solid shape, but do not save the
model render), while the other is better for other purposes (like details on how the shape was created [37].
answering geometrical queries posed by the user). Alternatively, procedural modeling techniques such as history-
B-Rep representation is a particular type of explicit representa- based parametric feature-based modeling create 3D models from
tion where suitable sets of connected geometric elements are used sets of rules [38]. Procedural models have the advantages of cap-
to represent the vertices, edges, and faces that define the boundary turing all or part of the design intent and being easy to edit [39].
of a solid [26]. Geometric information defines the exact shape and Procedural approaches to create 3D CAD models include feature-
spatial position of the elements, whereas topological information based design (FBD) and constraint-based modeling [40,41]. The
defines the links between the elements [27]. Some authors have recent emergence of feature-based downstream applications such
further classified B-Rep representations as B-Rep, faceted B-Rep, as Numerical Control (NC) machining [42] is favoring the already
and boundary curve-based [28]. dominant position of feature-based master models.
Generally speaking, B-Rep models consist of patches. B-Rep We observe that the representations described in this section
patches are portions of curved surfaces bounded by curved con- may result from interaction with the user or from automatic algo-
tours, which are pieces of curves delimited by vertices (which lie rithms that shift from a low semantic representation up to a higher
at points). There seems to be a terminological confusion in the one. Reverse engineering (which is aimed at extracting knowledge
literature, as patches are sometimes called faces. Some authors or design information from actual products in order to produce
consider a face as any bounded portion of a surface (it may be pla- their CAD models) is a complex problem by itself and, as such, is
nar, cylindrical or adopt any curved form) while other researchers out of the scope of this paper. Only the final output is classified in
claim that faces must always be planar (they are only faces in the this study.
particular case where a flat contour delimits a region of a plane).
For example, a can is modeled by three patches: one cylindrical 2.2. Types of changes
patch and two faces delimited by circles. Then, the shell is a set
of connected patches/faces. But the literature also differentiates There is general consensus on the types of changes that can be
shells as piecewise-algebraic surfaces (made of algebraic pieces) performed to a master model: simplification, interoperability, and
or piecewise-linear surfaces (made of linear pieces). reuse.
On the other hand, there is a problem regarding the sizes of Simplification means converting accurate, highly complex mod-
patches/faces. B-Rep patches are common but inefficient to render els into simpler models that retain the important details and
to a computer screen and difficult to process in physical simula- eliminate the irrelevant ones [16]. Reducing the geometric com-
tions [29]. First, each shape requires its own particular algorithm to plexity of the master model at various levels of detail (LODs) is
simulate its behavior, and second, this information is unavailable in desirable and useful so valid simulations can run at a reasonable
models generated from point clouds [30] or sets of scanned images cost. Feature-based multi-resolution solid modeling strategies are
[31]. In other words, there is a need to convert the trimmed sur- helpful for creating secondary views (i.e. different versions for
faces of B-reps into a form that removes the gaps between adjacent different needs [43,44]). In some cases, users may want to remove
faces and allows a homogeneous representation of shape [32]. In the ‘‘know-how’’ from simplified models used as final deliverables
these cases, the best solution involves discretizing the model into for fear of losing intellectual property (as argued, for instance by
a mesh of smaller elementary mesh cells of mesh elements (usually [45]). However, it would be inefficient to remove the ‘‘know-how’’
triangles) that approximates a geometric domain [33]. According entirely from all simplified models created for internal use.
to [34], a mesh is an arrangement of cells with connectivity be- Interoperability refers to the level up to which a master model
tween the cells defined by the possession of common cell faces can be accurately transferred from one modeler to another [2,9].
or cell edges. A cell is a manifold of dimensionality one or higher Interoperability is concerned with the resolution of semantic dif-
that is a part (or the entirety) of a mesh. In this paper, we use ferences between similar constructs in different representations
the term face to refer to a mesh cell. Thus, the cylindrical patch of and the inherently different tolerances within the geometric al-
the can from the earlier example may be subdivided into a mesh gorithms used at the core of different CAD modelers [46,47]. In-
of an adjacent slender quadrilateral faces (that approximate the teroperability encompasses the exchange of CAD models between
cylindrical surface as an n-sided prism). Computer generated data different CAD systems and to downstream applications [48]. Early
produces tessellated meshes, while scanned data produces digitized attempts to solve product information exchange problems were
meshes [35]. described by [49,50]. Interoperability is particularly important in
A recent survey by Shimada discusses current trends and issues collaborative design environments, where CAD/CAM/CAE [51] and
in meshing and geometric processing for computational engineer- information and knowledge sharing tools are commonplace [52].
ing analyses [31]. Researchers Attene et al. analyzed common mesh Model reuse involves modifying a master model in the native
defects, and surveyed existing algorithms to process, repair, and modeler so it can be utilized in other situations. Reuse can be
improve meshes [35]. In our paper, we distinguish between B-Reps performed at different levels ranging from selecting library items
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 67

visualization or Finite Elements Analysis), and thus may introduce

quality errors in some critical CAD/CAM properties, like small
variations in the exact position of the holes of a flange.

3.1. Morphologic errors in explicit CAD models

The morphological level of quality for B-Rep and mesh repre-

sentations has been widely studied, as morphologic errors may
Fig. 1. General levels of CAD Model Quality. affect simplification, which is critical for downstream applica-
tions. For example, authors Bøhn and Wozny considered punctures
(which produce open shells), inconsistent facet-orientations (which
to adapting existing designs [53]. In general, two types of reusabil- create uncertainty in terms of what side of the shell contains
ity can be defined: (1) editing a model to produce a redesigned material), and internal features (non-manifold elements within the
version of the current object (instantiating), and (2) cannibaliz- shell that create ambiguities when solidifying) [59].
ing old designs to use ‘‘spare parts’’ (version modeling) [54,55]. Gu et al. distinguished geometric errors (those that produce
Modifications are frequent, even for instantiated models, so the solids that are topologically valid, but potentially difficult to sim-
quality of reusable models must be particularly high, as they need plify by downstream applications) from topological errors (those
to reliably allow for modifications while maintaining the original that result in invalid topologies) [9]. By observing the effect of
design intent. A related problem, model retrieval for reusabil- errors on a model, the authors concluded that the former type
ity, has also received some attention by the research community (topological) includes realism errors, including: (1) singularities
[56–58]. characterized by sudden changes in normals or tangents, (2) ab-
To note that the coupling of the three different types of changes normally small or uncommon volumes or holes, and (3) abnor-
may influence on the quality of the CAD model [45]. However, mally small distances between edges or vertices. They subdivided
the current classification is limited to consider the impact of each topological errors as: (1) Accuracy errors that consist of excessive
action separately, which is still a highly challenging goal. geometric gaps between topologically connected elements (like
one vertex that falls outside a tolerance volume which surrounds
3. Taxonomy of CAD model quality the intersection between the edges that should share it), and
(2) Structure errors that occur when any topological element of a
In order to understand CAD model quality issues, Pratt [23] model, including vertices, edges, faces, and shells, is undefined or
emphasized the importance of the semantic study. By further de- incorrectly linked [9].
veloping this idea, authors Contero et al. [18] defined three levels Researchers Yang et al. [28] identified six types of CAD model
of quality to classify CAD models: morphologic, which relates to the errors in explicit representations: tiny faces, narrow regions, non-
geometric and topological correctness of the CAD model; syntac- tangent faces, narrow steps, sharp face angles, and narrow spaces.
tic, which assesses the proper use of modeling conventions, and The catalog of error types was updated by Chong et al. [60], who
semantic/pragmatic, which focuses on the CAD model’s ability for used the source of the errors to identify six types of geometric
modification and reuse. Other authors classified quality issues by errors (non-manifold mesh, small gaps and overlaps, large gaps and
the type of representation [37,42]. overlaps, duplicated or undesired elements, surface orientation
The taxonomy presented in this paper defines a ‘‘frame’’ with and sliver surfaces), and two types of topological errors (holes, and
three levels of CAD quality for each type of representation, and missing parts/surfaces).
then identifies the artifacts that may hinder simplification, in- More recently, Attene et al. [35], prioritized geometric realiza-
teroperability, or reuse of the model. The general levels of our tion to distinguish between geometric issues (which, in addition
taxonomy are shown in Fig. 1. In the next sections, we argue to punctures, include intersections, near degeneracies, noise, and
that each semantic level can be mapped (to a certain extent) to a feature chamfering/aliasing), local topology flaws (non-manifold
particular type of change: morphologic to simplification, syntactic mesh), and global topology flaws (inconsistent orientation and
to interoperability, and semantic to reuse. wrong genus).
Fig. 1 highlights that classifying only by representation type Several error classifications have been proposed. The main
or only by the semantic level of the failures are both incomplete difference is whether errors should be considered geometric or
views. However, it omits the big differences inside each category – topological, which depends on whether the classification considers
for instance, B-rep and mesh models are very different explicit what caused the error (source), how the error affects the whole
models – as adding more details (like sub-categories of explicit or object (scope), or how the error manifests (visibility). Gaps and
procedural models) would mask the main proposal of a ‘‘frame’’ holes break the connectivity (i.e., they change the topology), but
that merges the two relevant dimensions. they usually come from imperfect geometric calculations (source),
In the following sections, each of the six types defined in Fig. 1 and are usually detected by counting connections (visibility). A
is further subdivided into sub-types (represented by horizontal similar problem applies to the distinction between local and global
boxes with horizontal text). For each subtype, some of the most errors. For instance, a local displacement of a vertex may cause a
representative errors found in the literature are included to better global loss of symmetry.
illustrate the scope of the sub-type (vertical boxes with vertical Our classification targets the source of the artifacts and their
text). scopes, while paying less attention to their visibility (which is
We note that some readers may be reluctant to perceive some of oriented at defining procedures and algorithms that detect and
the following errors, since it is quite commonly accepted that some repair the errors). Assuming this approach, errors that violate the
models have a very concise, simple, common data structure, where design intent but affect to isolated elements are local. They are
it seems that no room for errors is left. However, all languages may topological, if the isolated elements are the incorrect ones; and
be used to convey the wrong message, or to incorrectly convey geometrical, if the isolated elements interact with other elements
the right message. For instance, syntactic and semantic errors in (or with references) in the wrong manner. For instance, parallelism
mesh models may be hardly perceived. But mesh models usually is a geometrical property, thus, two faces that fail to be parallel
prioritize regularity of the mesh (as they are usually aimed at determine a geometrical problem. If the lack of parallelism of those
68 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

faces causes another error to appear elsewhere (i.e. propagates), automatically repair the remaining errors. For example, ap-
then the original error is global. Otherwise, it is local, despite the proaches exist to heal meshes with holes [33,71]. Gaps can be
size of the faces. Therefore, we distinguish: stitched by physically moving vertices (i.e., changing their coor-
dinates), although this change introduces an error in the order of
• Local geometric issues are generally due to isolated accuracy the resolution tolerance [72]. Geometric hashing finds particular
errors, including punctures and overlaps. instances of objects in a given scene, even if they are partially
• Global geometric issues are noise caused by repetitive accu- occluded [73]. Hashing techniques heal gaps by assuming that
racy errors. one border is a slightly deformed sub-curve of the other. Then,
• Local topology flaws include non-manifold meshes, unreal- they find the translation and rotation of the sub-curve that yield
istic shapes, and abnormally small or uncommon elements. the best least-squares fit to the appropriate portion of the larger
• Global flaws in topology include inconsistent orientation curve, and construct new patches that fill the remaining holes
and wrong genus. (typical patches are triangles obtained by the 3D minimum-area
triangulation technique) [74]. Alternative approaches, classified
The length to width aspect ratio plays an important role when as ‘‘loyal’’ to the input model, have also been described in the
examining punctures. Length and width are similar in large gaps literature [75].
or holes (those having three or more borders). Small gaps are Local flaws in topology can be solved by most commercial
slender and include only two borders [61]. We consider T-joints as applications. These flaws are detected through indirect cues like
a particular type of gap where the number of edges and vertices on counting connections, or using angles and sizes to diagnose various
opposite borders are different [60,62,63]. Transversal penetrations artifacts [9].
(or clashes) and intersections are the complementary of holes and Wrong genus can be solved by Topology Denoise Technolo-
gaps. They can be classified according to the type of contact curve: gies [76]. However, manual healing is often required to ensure
closed (for interpenetrations) and open (for intersections). Finally, anisotropy and directionality, which are not effectively handled by
both small gaps/overlaps and rendering inaccuracies may cause commercial mesh generators [31].
visual artifacts (known as cracks) between adjacent faces, which
are not considered in this paper, as they affect the visualization 3.2. Syntactic errors in explicit CAD models
quality of the model [64,65].
The most common global geometric issue is the repetitive Syntactic errors are often the source of interoperability prob-
noise that results from data corruption when processing digitized lems that appear because the native model and the downstream
meshes. This noise distorts most of the elements of the mesh, applications use different data structures.
producing undulations, such as ripples and creases. The problem is Although there are different strategies to address interoper-
common in digitized meshes, but not as common in B-Rep models. ability [77], the most practical solution requires mapping, which
In order to classify local topological errors, it is important to re- involves univocally associating every geometric description from
view some concepts. A mesh is two-manifold if it is homeomorphic the original representation to a fully equivalent geometric de-
to a disk in the neighborhood of every point (i.e., resembles a two scription in the image representation [78]. Mapping between data
dimensional Euclidean space near each point). We can further sub- structures may be direct, or use a neutral exchange format [9].
divide non-manifold models as those produced by under-connected Direct Translation is the ability to read one CAD format and write
elements (also called isolated, dangling, or naked elements), or the information to a different format. The fundamentals of the
over-connected elements (singular elements). For instance, when direct mapping style are available in [79]. Most of the solutions in
more than two polygons share a common edge, then such edge is this field are practical, but commercially protected (although some
said to be singular, complex, or non-manifold. became publicly available [80]). The alternative is saving CAD data
According to Gu et al. [9], unrealistic shapes are linked to sudden to a neutral format, so a second CAD system can read it.
changes in normals or tangents, and can be further subdivided Direct translation is advantageous when systems share the
into appendage-volumes (tiny features that protrude), knife-edges same kernel (the geometric engine that stores and organizes the ba-
(two faces that join at a sharp edge), and sliver-cracks (sharp sic geometric shapes and model topologies), as these CAD systems
wedge-shaped cracks between features). Abnormally small or un- can directly share modeling data recorded in the kernel’s native
common volumes or holes include sliver-faces, micro-slots, etc. A format [46]. Nevertheless, kernel-level data exchange is limited
sliver face is a face with a high aspect ratio, whereas a spike is a by the different ‘‘flavors’’ developed by manufacturers, which are
region with a high aspect ratio inside a face [66]. Abnormally small often protected for commercial reasons [81].
distances between edges or vertices can also be the cause of errors. Translation problems due to neutral files can be subdivided into
Global flaws in topology include inconsistent orientation and inaccuracy, inconsistency, and loss of semantics [46]. The former two
wrong genus. The latter is also called topological noise [67] and problems are explained in the next paragraphs. Loss of semantics
is due to the finite precision of digitization tools that alters the is described in the next section.
number of handles and tunnels in the geometry [68]. A model that is inaccurately translated becomes disintegrated,
Anisotropy and directionality are particularly important prop- as its entities become disconnected or fall outside the required
erties in engineering simulations [31]. They are indirectly imposed tolerance (as a result of moving them from one system to another
by Delaunay meshes (which maximize the minimum angle of that uses tolerances that are too small or too large) [46,82–84]. The
all the angles of the triangles in the triangulation), or sliver-free underlying problem is the inability to maintain complete internal
meshes (a sliver is a nearly flat tetrahedron in a tetrahedral 3D consistency between topological and geometric information. This
mesh [69]). Consequently, their absence is detected as a potential problem is known as the geometric accuracy problem [85,86] and is
error by some MQT tools. caused by computational errors in geometric calculations [23]. It
Morphologic errors in explicit models are summarized in Fig. 2. has been stated that precision is particularly necessary for NURBS
Horizontal boxes represent the realm of the errors. Vertical boxes surfaces in large models with fine details [87]. Yang et al. [88] iden-
identify common errors linked to those realms. tified disintegration errors in the IGES and STEP neutral formats.
A general technique to prevent geometric errors is robust com- For the sake of brevity this paper does not consider the inconsis-
putation, which introduces knowledge at every possible level to tencies caused by the lack of standardization. However, (a) failure
guide the computation [70]. Various strategies can be used to to meet standards may result in some data not being visible or
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 69

Fig. 2. Morphologic errors in explicit models.

being incorrectly interpreted [3], and (b) there are mapping inco-
herencies due to the shift from old technical drawing standards
[89] to 3D model-based standards such as ASME Y14.41 [90]. The
only standardization topic we consider is that standardizing neutral
formats helps prevent inconsistent mappings of 3D CAD models. In
this regard, the most relevant standard neutral formats are IGES
and STEP.
‘‘Initial Graphics Exchange Specification’’ (IGES) was first pub-
lished in 1980 as ‘‘Digital Representation for Communication of
Product Definition Data’’ by the US National Bureau of Standards
(NBSIR 80-1978). STEP (Standard for the Exchange of Product
model data, ISO 10303) [91,92] includes two standard formats that
are widely implemented: AP 203, maintained by PDES, Inc. and AP
214, maintained by ProSTEP iViP and SASIG2.
Some inconsistencies have survived standardization. First,
some standards are sectoral to certain industries. For example,
AP 203 can be used to transfer purely geometric models with a
high degree of success [37], but it is primarily supported by the
aerospace and defense industry, while AP 214 is the preferred for- Fig. 3. Syntactic errors in explicit models.
mat in the automotive industry [93]. Since standards are a subset
of all the information that is relevant to a diverse number of CAD
systems, they can only represent the geometric information that is
common to all systems, [94]. available AP214 translators address only the AP203 ‘‘look alike’’
Second, standardization may be unintentionally open to inter- conformance classes (i.e., AP214 cc’s 1 & 2) [46].
pretation. For instance, IGES translators never became fully stan- Syntactic errors in explicit models are summarized in Fig. 3.
dardized, and thus their rules vary greatly from vendor to vendor, Horizontal boxes represent the realm of the errors. Vertical boxes
which produces inconsistencies [46]. identify common errors linked to those realms.
With regard to mapping scope, it is incomplete as it fails to sup- Full implementations of neutral formats are still required to
port legacy and domain. The strategy to support legacy is durability, prevent miscommunication from incomplete domain mappings.
which requires compatibility (consistency along time) between Adoption of neutral formats by Small and Medium Enterprises is
CAD data of different versions of the same system [84]. Domain also pending, although neutral formats for explicit CAD models lack
mismatches are failures to support the full domains from differ- the semantic level that would motivate those enterprises to adopt
ent understandings and assumptions. The most common domain them, as explained in the next section. The problem is particularly
mismatch is incomplete domain. The most acclaimed STEP AP214 aggravated for SME’s that are not connected to OEM’s, because they
translators have only implemented conformance classes 1 and/or lack the motivation to adopt those strategies, and the stimulus of
2, which are essentially identical to AP203. Most commercially long production series to offset the added costs.
70 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

3.3. Semantic errors in explicit CAD models

As long as semantics focuses on the relationship between sig-

nifiers (geometry elements in our case), and what they stand for
(the functionality of one particular shape, used to solve a certain
design problem), discrepancies between geometry and functional-
ity result in semantic level errors in CAD models. Clearly, there is
no single source for those errors. When and how they come to be
introduced in the model depends on both the type of modeling and
the flow of the design data.
Reusability implies transferring design intent and manipulating
the geometry of an existing CAD model, so it can be used in new
situations. Reusability refers to the ability to make changes as if
the model was created in the receiving feature-based CAD system.
Efficient and adaptive reuse is only feasible if design intent and
design rationale are available to a certain extent, as it occurs in
procedural models [95]. While it might be easy to detect problems
in B-Reps or meshes, it is often impossible to automatically solve
them if the designer’s intent is lost [51,59].
When reuse of explicit models is the only option, reverse en-
gineering strategies (also called refeaturing) must be used to find
features [96]. In this process, ‘‘plain’’ B-Rep models are ‘‘enriched’’
Fig. 4. Semantic errors in explicit models.
by shifting to ‘‘hybrid’’ models that include part of the design in-
tent. Different strategies have been used to enrich explicit models.
Beautification is perhaps the most successful. A beautified geomet-
ric model is a modification of the input model that incorporates ap- models may produce different analysis results from those ob-
propriate symmetries and regularities [47,97,98]. For instance, the tained with the original fully-featured model, which may result
Finite Element (FE) mesh of a symmetric part will not inadvertently in defeaturing-induced analysis error and sensitivity modification
be made non-symmetric during editing, if a particular approach for [111–114]. In front of the argument that there are no defeaturing
detecting design intent has enriched the model [99]. Declarative semantic errors (since results of a deliberate defeaturing can hardly
feature recognizers are the most recent advance for refeaturing be called errors, while if the process is done incorrectly, that is
[100]. just a software bug, not a semantic error), we argue that from the
Refeaturing may include restoring the face structure of a B-Rep point of view of a computer scientist (a programmer), producing
from a triangular mesh [101,102], recovering features [103,104], the wrong model is a software bug. But, from the point of view
and finding geometric constraints [105]. Currently, expert user of a product engineer (a user), the imperfect software (the only
guidance is required (setting tolerances to define ‘‘likelihoods’’), available) outputs a semantic error in the defeatured model.
not all types of features can be detected (free-form surfaces and Semantic errors in explicit models are summarized in Fig. 4.
general sweeps are excluded), and some are prone to errors (due to Horizontal boxes describe the realm of the errors. Vertical boxes
the noise and incompleteness of measured data, and the numerical identify common errors linked to those realms.
nature of the subsequent algorithmic phases) [105].
The sampling process (also called acquisition or re-meshing) 3.4. Morphologic errors in procedural CAD models
may produce a particular type of noise that affects tessellated or
digitized meshes. It produces irregularly triangulated chamfers The morphological level is not critical in procedural models,
(aliasing artifacts that appear when restrictions in the location as it is automatically tested by most CAD applications (the term
of vertices prevent them from coinciding with sharp edges and ‘‘not critical’’ means important while also solvable to a great ex-
corners in the model), thus forcing sharp edges and corners to be tent). These systems permanently check the correctness of the
restored by feature sharpening approaches [35]. model trees (including constrained profiles, sweeping operations,
The lack of suitable methods to detect reference systems, as well features and datums), and warn the user about any incoherencies
as suitable datums to skeletonize the model, may prevent finding (although careless users may certainly ignore these messages).
constraints (both directly or as a result of misaligned reference In this context, Company et al. developed a classification that
systems and undetected datums). Some steps have been taken in includes six dimensions of quality in CAD models: validity, com-
this direction [99,106]. pleteness, consistency, conciseness, simplicity, and conveying de-
There are geometric simplifications that need to be performed sign intent [115,116]. Validity relates to the morphological level. A
in preparation for Computer-Aided Engineering (CAE) processes: model is valid when it can be retrieved (it is not missing) and used
geometric feature removal and dimensional reduction [31]. The (it is free of errors).
former is also known as defeaturing, as some features that con- Sources of morphologic errors include kernel malfunction [70],
vey functionality (fillets, chamfers, holes, fasteners, etc.) are re- and inappropriate modeling practices, which may result in subtle
moved from the geometry to analyze certain behaviors of the part variations in shape and size, or inconsistencies that do not cause
[107,108]. A common example of the latter type is converting a errors or warning signals in the original model, but may hinder
thin-walled solid into a shell by extracting its mid-surface and reusability. Legal, and apparently simple, modifications may cer-
meshing it [109]. tainly produce unexpected crashes.
Dimensional reduction usually requires expert manual adjust- Researchers Yang et al. [28] advocate for a hybrid method
ments. Additionally, boundary conditions which are critical for to solve these failures. They initially identified six types of CAD
assembly analysis are frequently lost [31]. model errors in explicit representations (by using the approaches
Defeaturing a B-Rep model produces a ‘‘wound’’, or hole that described in Section 3.1): tiny faces, narrow regions, non-tangent
needs to be covered by a replacement surface [110]. Defeatured faces, narrow steps, sharp face angles, and narrow spaces, and
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 71

used procedural information in the model tree to trace these errors

back to the modeling operations. The operations are then reformu-
lated until the errors disappear. The interdependency of feature
commands is analyzed to find the best reformulation. However,
the shape may occasionally become distorted or even collapse if
only the sequence of feature commands is modified. Therefore,
the authors also analyzed parametric data to find mutual depen-
dencies between parameters and constraints of different feature
commands.
Huang et al. detected problems that affect the NC machining
of a part [42]. These failures are similar to the unrealistic shapes
described in Fig. 2 and fall in the range of the completeness dimen-
sion by Company et al. [115], which measures how well the model
replicates the actual shape and size of the part. Errors include
missing elements as well as those elements that are incorrectly
created as a result of kernel malfunction or inappropriate modeling
strategies.
Errors caused by ambiguous definitions of procedural models
have also been studied. References to entities defined during the
design process may be incorrectly reevaluated. Ambiguously de-
Fig. 5. Morphologic errors in procedural models.
fined datums – mainly those defined implicitly (‘‘on the fly’’) – may
inadvertently ‘‘switch’’ after editing the model (e.g. a blend applied
to one of the two edges that result from cutting a round slot across
a rectangular block [117]). This problem is known as topological Direct mapping through geometric modeling kernel-level data
naming (when names use only topological information) or persis- exchange (which is a particular type of direct translation that
tent naming (when other types of information are used) [118–122]. occurs between two different CAD applications that share the same
Furthermore, Marcheix distinguished between persistent naming kernel, like ACIS by Dassault Systèmes or Parasolid by Siemens)
of atomic entities (such as vertices, edges or faces) and aggregates continues to be an interesting approach for the semantic interop-
(such as sets of faces) [123]. The evolution of the approaches used erability of procedural representations [125], but its use is limited
to solve this problem was summarized by [28] and [123], while to certain types of model exchanges because of its high cost and
authors [122] included a detailed taxonomy of persistent naming. proprietary nature (the latter is described as a potential drawback
Seemingly well-constrained profiles may sometimes be com- in [78]).
patible with different geometric or topological solutions, which Mapping through neutral representations is a more efficient
are not necessarily compatible with design intent. They appear alternative, as procedural models with parameters, features and
because most of the problems of the modeling with constraints constraints can be processed.
technique described by Anderl and Mendgen [124] are still un- Although geometric constraint solving has undergone substan-
solved. For instance, an arc at a corner of a rounded polygon that tial progress in terms of the types of objects and constraints that
produces a blend as a result of sweeping a profile should be tangent can be handled robustly, parametric operations have largely re-
to the two lines converging at the corner. However, after a number mained within the same conceptualization and are now beginning
to limit the flexibility of CAD systems [126]. We can classify these
of changes, internally tangent arcs may be incorrectly selected by
limitations through the quality dimensions of consistency and
the kernel as equally valid solutions [118]. This type of failure may
conciseness [115].
also occur between a modeling operation and its parent modeling
A neutral model is consistent if it is simultaneously flexible (to
operation, when the child operation is relocated to a new location
enable re-design) and robust (to prevent undesired changes and
in the design tree (from the standpoint of the Boolean rules that
failures during edition). A neutral model is concise if it does not con-
govern Constructive Solid Geometry) and the resulting shape that
tain repetitive or fragmented constraints, modeling operations or
does not maintain the desired design intent. We refer to this failure
datums, as those used in poor translations when direct mappings
as ‘‘reversal constraining’’.
are unavailable. A model must also be represented by high level
Morphologic errors in procedural models are summarized in modeling operations (contrary to what happens when they are
Fig. 5. replaced by low level operations to solve incomplete equivalences
between representations).
3.5. Syntactic errors in procedural CAD models Mapping cannot maintain attributes that are missing in the
original model. For example, buried features are those completely
Researchers Tessier and Wang identified two types of data in- included within other features [46]. In the best case scenario, they
teroperability incompatibilities in procedural CAD systems: Struc- are an example of overlapping that cannot be mapped as a concise
tural heterogeneity (caused by the use of different data structures; image. In the worst case scenario, they may result in altered models
such as a CAD system that defines a fillet by the removed edge (unexpected geometry and/or topology).
and the radius, while other requires the tangent edges instead of Some authors have used explicit ontologies (a formal represen-
the original edge), and semantic heterogeneity (caused by naming tation of a set of concepts, their properties, and the relationships
and terminology differences, such as a CAD system using the term between those concepts within a given domain) for the semantic
fillet for a feature that other systems recognize as round) [78]. interoperability of CAD systems [78,94,127,128]. Explicit ontolo-
While structural heterogeneity is the main cause of morphologic gies define representations that are application-independent, ex-
errors (and requires the mapping of features that are equivalent pressive and unambiguous [94].
but defined with different data structures), the ultimate goal is the Explicit ontologies are converging with the ISO application
‘‘semantic interoperability’’ of procedural representations, where protocol for managed model-based 3D engineering, which was
the term semantic can be broadly defined as the meaning associ- recently published as API 242 [129] after an extensive development
ated with a terminology in a particular context [94]. period [37,93,130].
72 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

Fig. 6. Syntactic errors in procedural models.

Dialects or ‘‘flavors’’ are an open problem, as there are no A model is simple if its model tree is clear and understand-
commonly accepted standards on how to solve certain complex able, which means that modeling operations in the modeling tree
geometry that results from combining multiple shapes. A common must be labeled to emphasize their function (instead of how they
example involves how a constant-radius blend in an edge should were built), and related modeling operations must be grouped to
end at complex vertices [78,118]. emphasize functional parent–child relationships. A model is also
An interesting new technique to tackle the mapping type prob- simple if it uses compatible and standard modeling operations. In fact,
lem is the application of the dual model strategy as follows: the pri- standard modeling commands have been proposed as mechanisms
mary procedural model is associated to a secondary B-Rep model to exchange design intent [130].
used by the receiving system to check the validity of the model It has been confirmed that relatively simpler features, the use
transfer [48]. This method provides a sort of ground truth, which of reference geometry, and the correct feature sequence are posi-
can be useful. However, there is a large collection of errors that this tively correlated with design intent proxy ratings [132]. Nonethe-
type of ground truth may contain, as discussed previously. We note less, the qualities of a model that more directly convey design
that this was not the original aim of dual models, whose origins and intent are described as follows [115]:
current use can be traced back to the work by Kim et al. [131].
Domain problems in procedural models are similar to those de- (1) The modeling process must effectively convey the right
scribed in Section 3.2 for explicit representations. Legacy problems information about function. Geometric constraints must
will likely be solved once the new STEP standard becomes fully highlight functional relationships, models must use feature-
available and widespread. based operations that convey the functionality of the parts
Syntactic errors in procedural models are summarized in Fig. 6. [134], datums must convey the skeleton or scaffold of the
Horizontal boxes describe the realm of the errors. Vertical boxes model [135], and functional patterns (regularities, symme-
describe common errors linked to those realms. tries and repetitive patterns) must be explicitly included in
the model tree [133].
3.6. Semantic errors in procedural CAD models (2) The modeling process must be efficacious. It must be free of
(1) fragmented operations that mask the final part function,
Company et al. [115] described two fundamental qualities of (2) overlapping operations that do not change the model
a semantic error-free procedural CAD model: the model must be geometry, and (3) overlapping operations that mask pre-
simple and convey design intent. Poorly perceived models require vious operations. The model must also be free of generic
more time to alter [132]. In this context, the use of formal CAD (unspecific) feature types (i.e., suitable features are not re-
modeling strategies and best practices for history-based paramet- placed by merely similar ones), and shapes. Also unspecific
ric design is growing interest as a method to improve the semantic patterns should be avoided (i.e. those that simply reduce the
quality of CAD models. A recent contribution by Camba et al. [133] modeling effort, but do not convey design intent).
on approaches specifically designed to emphasize CAD reusability (3) The model must be efficient. This means that: (1) The model
(Delphi’s horizontal modeling, explicit reference modeling, and tree distinguishes between datums, scaffolds, core, detail,
resilient modeling) reveals significant advantages of formal mod- replication and cosmetic operations; (2) Replication opera-
eling methodologies, particularly resilient techniques, over non- tions (based on patterns) are used to convey functionality in
structured approaches, as well as the unexpected problems of the the model tree, and (3) Design decisions are traceable within
horizontal strategy in numerous modeling situations. the model tree.
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 73

Table 1
Document checked by different Model Quality Testing tools.

Legend: Y for available document type and void for unavailable.

To a certain extent, some of the previous practices are subjec-

tive; sometimes, even contradictory. Therefore, controls are not
easily implementable. Nevertheless, some of the recommenda-
tions that result in clear, specific, and objective tasks could be easily
detected by existing CAD packages. Unfortunately, there is a lack
of checking tools to confirm that such rules have been followed.
For example, similar to the way most CAD applications highlight
insufficiently constrained sketches, they could also highlight de-
fault named features in the model tree to encourage users to apply
proper naming conventions.
In addition, techniques to simplify secondary models can shed
light on the features that should be modeled separately or last, in
order to facilitate its removal [16]. However, those techniques have
not yet been used to check the quality of the model sequence or the
grouping of operations that may be functionally related.
A model is ineffective if its design intent is not maintained
when the model is altered. In this regard, knowing the range where
topology does not change is important. Authors van der Meiden et
al. presented a method to compute the critical values when a single
Fig. 7. Semantic errors in procedural models.
parameter of a model is modified, (i.e. the parameter values for
which the topology of the model changes) [136]. Accordingly, MQT
tools could simply highlight those parameters that are particularly
sensitive to changes, making explicit where a model is robust but The scopes of the documents generated by CAD applications that
barely flexible. can be tested by the various MQT tools available in the market
Similar tools such as a consistency checker between the de- are shown in Table 1. The types of CAD documents that can be
clared typology of the features and its actual nature after editing typically tested by commercial Quality Testing Tools include mod-
could also be implemented (e.g., a hole declared as blind at creation els, assemblies, and drawings. Those tools that were unable to test
that becomes a thru hole after some editing) [137]. Enriching the models have been excluded from our study. In our sample, all MQT
model with annotations to communicate geometric design intent tools that test models also test assemblies, and most can also test
explicitly is also a related active field of study [138]. drawings.
Semantic errors in procedural models are summarized in Fig. 7. Table 1 also shows whether these MQT tools are linked to or
embedded into any particular 3D CAD Design software. 36% of the
4. Commercial CAD model quality testing tools MQT tools studied are embedded.
Average prices of MQT are summarized in Table 2. The table in-
A preliminary analysis of current commercially available MQT’s cludes three types of MQT tools: embedded systems that obviously
has been summarized in a series of tables, provided in this section. require a particular CAD system to be installed; non-embedded
74 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

Table 2
List prices of the different Model Quality Testing groups.
Type Average MQT price one license (e) Average annual maintenance MQT (e)
Embedded and linked 1450 225
Not-embedded but linked 6900 1150
Neither embedded nor linked 15500 3500

Table 3
Formats supported by model quality testers.

Legends: Y for available format and ‘‘blank’’ for unavailable. (1) CATIA V4 system interface, (2) CATIA V5 system interface,
(3)
CATIA V4/V5 system interface.

systems that are linked to a particular CAD package, and systems check. As an example, the mapping for the module Build Checks
that are neither embedded nor linked to any CAD package. of SolidWorks Design Checker – a tool to set the requirements
Average prices were calculated from list prices that were com- for evaluation – is shown in Table 4. The requirements that this
piled at the time of writing this paper. They are meant to be module is able to check are grouped in up to 7 different categories:
indicative, as prices can vary rapidly. Furthermore, list prices may document, annotations, dimension, drawing document, part doc-
differ from final prices, as sellers often negotiate custom rates ument, assembly document and feature. Obviously, requirements
with each client. Nevertheless, three price ranges typically stand for groups drawing document and assembly document are not re-
out: embedded systems which require a particular CAD package lated to model quality, so these requirements were excluded from
to be installed are in the $1,000 to $2,000 range, plus $200–$500 the mapping. Alternatively, part document and feature checks are
of maintenance; non-embedded systems that are linked to a clearly related. In addition, some document, annotations and dimen-
particular CAD package are in the $5,000–$10,000 range plus sion checks were also included in the mapping, as they are more
$1,000–$3,000 of maintenance, and standalone systems (not
or less transversal to models, drawings and assemblies. Certain
linked to any particular CAD package) are in the $10,000–$20,000
requirements of those groups, however, were excluded as they
range plus $2,000–$5,000 of maintenance.
are limited to drawings (Table font, Balloon font, References up to
We conclude that the estimated costs (exact prices are usually
date, etc.). Finally, some requirements were only indirectly linked
unavailable) are high enough to prevent SME’s from adopting MQT
to the taxonomy. Material property is used for drawings’ hatch
tools.
Interoperability of MQT tools is illustrated in Table 3. Embedded sections and to calculate mass properties. Therefore, an incorrect
systems (Design Checker, ModelCHECK, NX Check-Mate and Q- material selection may result in a non-standard hatching, which,
Checker) support less formats than standalone systems, which is in turn, results in an inconsistent standardization. Furthermore,
logical, as priority is always given to models and assemblies created the selection can also produce an incorrect calculation of mass
with their own software. While CATIA files are supported by 71% of properties, which may misguide a geometrical accuracy problem. A
the MQT tools, most formats are supported by less than 8% of the missing blend table results in undetectable blend constraints that,
MQT tools. in turn, prevent its re-featuring.
We have reviewed the model quality tools listed in Tables 1 As a result of the mappings described earlier, a summary of the
and 3 in light of the criteria defined in the taxonomy (this is, performance levels of each MQT tool is provided in Table 5 for each
those defined in Figs. 2 to 7). The procedure consisted in mapping of the six topics of the taxonomy defined in Section 3. Preliminary
the taxonomy criteria against the requirements that the tools can conclusions include:
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 75

Table 4
Model quality criteria supported by the module Build Checks of SolidWorks Design Checker.

• The morphological level is reasonably well covered in ex- - Two criteria were considered to evaluate coverage of seman-
plicit representations, but there is still room for improve- tic parameters in explicit and procedural representations. First,
ment. whether the MQT repaired a detected error (and whether this
• The syntactic quality in explicit representations can be bet- procedure was manual or automatic); and second, the amount of
ter improved by using efficient modeling strategies, rather errors detected was used to determine a qualitative score of high,
than by healing poorly mapped translations. average, or low coverage.
• Procedural representations have not yet been thoroughly Our tables are only intended to provide a broad picture of
considered by MQT’s. the current state-of-the-art in MQT systems. Although conclusions
must be validated through quantitative studies, we can report that
The tables included in this section summarize our qualitative
even though a number of commercial MQT tools are available, the
evaluation of the information that is currently available for com-
technology is not widespread. Reasons include: the proprietary
mercial MQT tools. The limited availability of information and the
nature of most tools (they only work as part of a particular CAD
lack of homogeneity in terms of the advertised details of each tool
application), their high cost, and the fact that they are not com-
may have an effect on the results. In addition, identical parameters
prehensive (only a fraction of the inconsistencies, inaccuracies and
or tools are called differently depending on the system. To the
best of our knowledge, a quantitative evaluation that compares failures of CAD models can be detected).
each MQT tool against a common benchmark has not yet been
conducted. The metrics used in our approximation are defined as 5. Open problems
follows:
- The available information on geometric or topological de- By analyzing current MQT tools in the context of our taxonomy,
tections for each MQT was used to determine the coverage of we identified some aspects that require further study and devel-
morphological failures in explicit and procedural representations. opment.
- The main factor to determine syntactic failures in explicit and Different types of quality loss are associated with master model
procedural representations was the number of supported formats, changes, and also with its representation type. In short, simplifi-
as indicated in the product’s documentation. IGES and STEP were cation may produce non representative secondary models if de-
included as complementary formats. sign intent – which should guide the simplification – is absent
76 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

Table 5
Semantic levels supported by the Model Quality Testers.

Legend: *** high coverage, ** average coverage, * poor coverage, and ‘‘blank’’ for no coverage.

in the master model. Mapping procedures aimed at guaranteeing Virtual Prototyping for optimizing processes, small and medium-
interoperability will always be prone to data corruption. Finally, sized enterprises (SME) frequently fail because of the lack of nec-
CAD model reuse is particularly sensitive to hidden errors and essary preconditions [4]. Commercially available MQT tools are
anomalies. Design intent is still poorly addressed by MQT tools, expensive, and thus rarely adopted by SME’s.
but there are other bottlenecks for different stakeholders (Original
Equipment Manufacturers (OEM), lower tier suppliers, and SME’s). 6. Conclusions
Dominant OEM’s force top-down interoperability with their
suppliers, which results in ‘‘defensive’’ or ‘‘conservative’’ designs, Quality testing tools for model verification, validation, and com-
which are robust but hardly creative. Interoperability is a main parison are essential, as exporting CAD models that contain errors
concern for OEM’s, as reusability is guaranteed by the best practices or anomalies to different downstream applications is prone to data
they impose, whereas simplification tasks are transferred to sup- corruption, which typically requires the models to be reworked by
pliers. A hidden problem that hinders interoperability is the lack the downstream user.
of proven modeling guidelines [125]. Best practices are checked In this paper, a new taxonomy of issues related to CAD model
by MQT tools but also imposed and tuned by the OEM, whose quality was defined. It was validated by using it to classify the cur-
current goal involves improving interoperability by abandoning rently available CAD quality assurance tools and determine which
explicit representations and adopting STEP AP 242. Validation rules aspects of quality are reasonably addressed, and which remain
require setup, and although quantitative metrics for shape errors open problems.
already exist, they are context dependent and are governed by The new taxonomy is based on the assumption that classifying
computational threshold values that are different for each MQT only by representation type or only by the semantic level of the
failures are both incomplete views. The proposed taxonomy distin-
tool [8]. Better quantitative metrics are still required.
guishes between explicit and procedural models, and, for each type
Lower tier suppliers are mainly involved with CAM/CAE tasks.
of model, morphologic, syntactic, and semantic errors are char-
Therefore, they need to improve simplification to reduce costs.
acterized. Additionally, each semantic level has been paired to a
Therefore, they must introduce strategic modeling by distinguish-
particular type of change: morphologic to simplification, syntactic
ing main and secondary features. They also need to remove inter-
to interoperability, and semantic to reuse. This pairing is not strict,
mediate formats (for example, by producing NC directly from the
as some crossed relations still occur.
master model), or even eliminate secondary views entirely [158],
We hypothesize that taxonomies of such a complex and evolv-
which would increase the need for a good quality master model
ing subject will always be incomplete, but useful. By using our
even further.
resulting frame, we have shed light on the different states-of-the-
For SME’s, the demand for interoperability is low. Usually, there
art in explicit and procedural models. For explicit models, we have
is direct contact between the original designer and the user in need argued the following:
of model reuse, and suppliers do not send back their secondary
views to the main SME. Alternatively, SME’s need to improve (1) Morphological correctness of explicit models can be evalu-
reusability. They must be taught how to strategically use CAD, as ated with ad hoc software.
they do not learn it through best practices (best practices are not (2) Interoperability is minimally supported by suitable stan-
imposed by any OEM). SME’s also require low cost MQT tools (both dards such as STEP AP 203.
in terms of acquisition and maintenance), which should not be (3) Efficient and adaptive model reuse is unfeasible for purely
tuned by or linked to alien best practices. While automotive and explicit representations, but limited for enriched explicit
aerospace companies are successfully implementing CAD-based representations.
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 77

With regards to procedural models, we have argued that: [5] Lion R. Excitech’s CAD checker: Simplifying the burden of CAD standards.
Excitech Comput Ltd, Des Product J 1999;1(3):28–9 URL: http://www.
(1) Interactive modeling editors prevent most morphologic er- excitech.co.uk/resources/sublibrary.asp?vol=1&iss=3.
rors, and MQT tools can handle the remaining reasonably [6] Bates T. Excitech’s CAD checker: Quality assurance for dwgs. Excitech Com-
puters Limited, Design Productivity J 2001;2(3):30–2 URL: http://www.
well.
excitech.co.uk/resources/sublibrary.asp?vol=2&iss=3.
(2) Interoperability problems between procedural representa- [7] Yang J, Han S, Kang H, Kim J. Product data quality assurance for
tions are likely to improve drastically with the development e-manufacturing in the automotive industry. Int J Comput Integr Manuf
of STEP AP242 (although implementation and adoption have 2006;19(2):136–47.
been slow). [8] Son S, Na S, Kim K. Product data quality validation system for product devel-
opment processes in high-tech industry. Int J Prod Res 2011;49(12):3751–66.
(3) Higher semantic aspects of quality – such as assurance of the
[9] Gu H, Chase TR, Cheney DC, Bailey T, Johnson D. Identifying, correcting, and
design intent embedded in the master model – are hardly avoiding errors in computer-aided design models which affect interoperabil-
addressed by current CAD quality testers. ity. J Comput Inf Sci Eng 2001;1(2):156–66.
[10] ISO 10303-43. Industrial automation systems and integration—Product data
Although the scope and level of detail of the comparison of representation and exchange. Integrated generic resource: Representation
current MQT tools limits our study, two preliminary conclusions structures. Geneva (Switzerland): International Organization for Standard-
ization; 2011.
arise:
[11] Requicha A. Representations for rigid solids: Theory, methods, and systems.
ACM Comput Surv 1980;12(4):438–64.
(1) MQT tools are mostly aimed at homogenizing the vast [12] ISO 11442. Technical product documentation—Document management.
amount of documents produced and shared by large OEM’s, Geneva (Switzerland): International Organization for Standardization; 2006.
and thus are primarily aimed at preventing easily solvable [13] ISO 10303-108. Industrial automation systems and integration—Product
low-semantic level mistakes and incoherencies. data representation and exchange: Integrated application resource: Param-
(2) MQT tools are still unaffordable for many Small and Medium eterization and constraints for explicit geometric product models. Geneva
(Switzerland): International Organization for Standardization; 2005.
Enterprises, as they are expensive both in terms of cost [14] ISO 10303-42. Industrial automation systems and integration—Product data
and training time. We presume that they will only become representation and exchange: Integrated generic resource: Geometric and
valuable for this particular market segment if document ho- topological representation. Geneva (Switzerland): International Organiza-
mogenization ceases to be prevalent over conveying design tion for Standardization; 1994.
intent. [15] Fowler J. STEP for data management, exchange and sharing. Twickenham, UK:
Technology Appraisals; 1995.
[16] Thakur A, Banerjee AG, Gupta SK. A survey of CAD model simplification
Although valid as plausible hypotheses, it can be argued that an
techniques for physics-based simulation applications. Comput Aided Des
in-depth analysis of all commercially available MQT tools would be 2009;41(2):65–80.
necessary to fully support our conclusions. Nonetheless, the most [17] Hoffmann CM, Joan-Arinyo R. Distributed maintenance of multiple product
significant contribution of this work is the value and effectiveness views. Comput Aided Des 2000;32(7):421–31.
of the taxonomy, as demonstrated by the classification of strengths [18] Contero M, Company P, Vila C, Aleixos N. Product data quality and collabora-
tive engineering. IEEE Comput Graphics Appl 2002;22:32–42.
and weaknesses of MQT tools.
[19] Hoffmann CM. Constraint-based CAD. J Comput Inf Sci Eng 2005;5:182–7.
As a future development, a more comprehensive theoretical [20] Bhavnani SK, John BE, Flemming U. The strategic use of CAD: an empirically
basis is required to define quantitative metrics for complex quality inspired theory-based course. In: CHI’99 proceedings of the SIGCHI confer-
requirements. Although some metrics for shape errors already ence on human factors in computing systems. Pittsburgh (USA, NewYork):
exist (such as those considered in Part 42 of STEP), further im- ACM Press; 1999. p. 183–90.
[21] Allsop CVP. The development of three-dimensional Computer Aided Design
provement is required, as these metrics are context dependent and
(CAD) modeling strategies and an investigation into their impact on novice
require tuning by expert users. users [Ph.D.dissertation], Loughborough University; 2009.
Finally, higher level semantics in the master model are re- [22] Toto R, Colledge T, Frederick D, Pung WH. Instructional strategies to pro-
quired to guarantee that design intent is made explicit and eas- mote student strategic thinking when using solidWorks. Adv Eng Educ
ily understandable by all users involved in creating downstream 2014;4(1):1–37.
[23] Pratt MJ. Geometric Modelling: Lessons Learned From The ‘Step’ Stan-
models (for example, was a symmetric part inadvertently made
dard. Series IFIP—The International Federation for Information Processing.
non-symmetric during the simplification process required to cre- 2001;75:130–46.
ate a FE mesh?). [24] Nooruddin FS, Turk G. Simplification and repair of polygonal models using
volumetric techniques. IEEE Trans Vis Comput Graphics 2003;9(2):191–205.
Acknowledgments [25] Shah JJ. Assessment of features technology. Comput Aided Des 1991
23(5):331–43.
[26] Braid I. The synthesis of solids bounded by many faces. Comm ACM
This work was supported by the Spanish Ministry of Economy 1975;18(4):209–16.
and Competitiveness and the European Regional Development [27] Qian G, Tong R. Automatical reconstruction of deficient CAD model. Adv
Fund, through the ANNOTA project (Ref. TIN2013-46036-C3-1-R). Mater Res 2011;186:241–5.
The authors also wish to thank the editor and reviewers for their [28] Yang J, Han S. Repairing CAD model errors based on the design history.
Comput Aided Des 2006;38(6):627–40.
valuable comments and suggestions that helped us improve the
[29] Foucault G, Cuillière JC, François V, Léon JC, Maranzana R. Adaptation
quality of the paper. of CAD model topology for finite element analysis. Comput Aided Des
2008;40(2):176–96.
References [30] Bernardini F, Bajaj CL, Chen J, Schikore DR. Automatic reconstruction of 3D
CAD models from digital scans. Internat J Comput Geom Appl 1999;9(4–5)
[1] Strategic Automotive product data Standards Industry Group (SASIG), Prod- 327–69.
uct Data Quality Workgroup (PDQ). SASIG–Product Data Quality Guidelines [31] Shimada K. Current issues and trends in meshing and geometric process-
for the Global Automotive Industry, Guideline version 2.1 (STEP Part 59. ing for computational engineering analyses. ASME J Comput Inf Sci Eng
ISO/PAS 26183:2006); 2005 [retrieved 04-11-2015]. Available from URL: 2011;11(2):021008–021008–13.
http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?cs [32] Kosinka J, Cashman TJ. Watertight conversion of trimmed CAD surfaces to
number=43436. Clough–Tocher splines. Comput Aided Geom Design 2015;37:25–41.
[2] McKenney D. Model quality: the key to CAD/CAM/CAE interoperability. [33] Bischoff S, Kobbelt L. Structure preserving CAD model repair. Comput Graph-
In: Proceedings of the 1998 MSC software Americas users conference. CA: ics Forum 2005;24(3):527–36.
Universal City; 1998. [34] ISO 10303-52. Industrial automation systems and integration—Product data
[3] Horwood M, Kulkarni S. CAD data quality. Eng Des 2005;31(3):14–6. representation and exchange: Integrated generic resource: Mesh-based
[4] Danjou S, Koehler P. Challenges for design management. Comput Aided topology. Geneva (Switzerland): International Organization for Standardiza-
Design Appl 2007;4(1–6):109–16. tion; 2011.
78 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79

[35] Attene M, Campen M, Kobbelt L. Polygon mesh repairing: An application [69] Chew LP. Guaranteed-quality delaunay meshing in 3D. In: Proceedings of the
perspective. ACM Comput Surv 2013;45(2):article 15. 13th ACM symposium on computational geometry; 1997. p. 391–3.
[36] Ju T. Robust repair of polygonal models. ACM Trans Graphics 2004;23(3); [70] Piegl LA. Knowledge-guided computation for robust CAD. Comput Aided
888–95. Design Appl 2005;2(5):685–95.
[37] Pratt MJ, Anderson BD, Ranger T. Towards the standardized exchange of [71] Zhao W, Gao S, Lin H. A robust hole-filling algorithm for triangular mesh. Vis
parameterized feature-based CAD models. Comput Aided Des 2005;37(12); Comput 2007;23(12):987–97.
1251–65. [72] Patel PS, Marcum DL, Remotigue MG. Stitching and filling: Creating confor-
[38] Shah JJ, Mantyla M. Parametric and feature-based CAD/CAM. New York: John mal faceted geometry. In: Proceedings of the 14th international meshing
Wiley and Sons, Inc.; 1995. roundtable, IMR 2005, p. 239–56.
[39] Bianconi F. Towards a procedural CAD model for data exchange: prob- [73] Lamdan Y, Wolfson HJ, Geometric hashing: A general and efficient model-
lems and perspectives. In: Proc. of the Joint international congress 17th based recognition scheme. In: Second international conference on computer
INGEGRAF—15th ADM, Seville (Spain); 2005. vision; 1988. p. 238–49.
[40] Salomons OW, van Houten FJAM, Kals HJJ. Review of research in feature- [74] Barequet G. Using geometric hashing to repair CAD objects. IEEE Comput Sci
based design. J Manuf Syst 1993;12(2):113–32. Eng 1997;4(4):22–8.
[41] Shahin TMM. Feature-based design—an overview. Comput Aided Design Appl [75] Chen J, Cao B, Zheng Y, Xie L, Li C, Xiao Z. Automatic surface repairing,
2008;5:639–53. defeaturing and meshing algorithms based on an extended B.-Rep. Adv Eng
[42] Huang B, Xu C, Huang R, Zhang S. An automatic 3D CAD model errors detec- Softw 2015;86:55–69.
tion method of aircraft structural part for NC machining. J Comput Design Eng
[76] Ning Z, Tengfei Y, Yu W, Li M, Li X, A survey of topology denoise technolo-
2015;2(4):253–60.
gies. In: ICCSE 2011-6th international conference on computer science and
[43] Lee SH. Feature-based multiresolution modeling of solids. ACM Trans Graph-
education; 2011. p. 1390–5.
ics 2005;24(4):1417–41.
[77] Rowell A. The challenge of CAD interoperability. Comput Graphics World
[44] Lee SH. A CAD-CAE integration approach using feature-based multi-resolution
1997;20(6):57.
and multi-abstraction modeling techniques. Comput Aided Des 2005;37(9);
[78] Tessier S, Wang Y. Ontology-based feature mapping and verification between
941–55.
CAD systems. Adv Eng Inf 2013;27(1):76–92.
[45] Kwon S, Kim BC, Hwang H, Mun D, Han S. Simplification of feature-based
[79] Rappoport A. An architecture for universal CAD data exchange. In: SM’03 Pro-
3D CAD assembly data of ship and offshore equipment using quantitative
ceedings of the eighth ACM symposium on solid modeling and applications;
evaluation metrics. Comput Aided Des 2015;59:140–54.
2003. p. 266–9.
[46] Gerbino S. Tools for interoperability among CAD systems. XIII ADM—XV
INGEGRAF International conference on tools and methods evolution in en- [80] Spitz S, Rappoport A. Integrated feature-based and geometric CAD data
gineering design; 2003. exchange. In: SM ’04 proceedings of the ninth ACM symposium on solid
[47] Gao CH, Langbein FC, Marshall AD, Martin RR. Local topological beautification modeling and applications; 2004. p. 183–190.
of reverse engineered models. Comput Aided Des 2004;36(13):1337–55. [81] Szykman S, Fenves S, Keirouz W, Shooter S. A foundation for interoperabil-
[48] Kim J, Pratt MJ, Iyer RG, Sriram RD. Standardized data exchange of CAD ity in next-generation product development systems. Comput Aided Des
models with design intent. Comput Aided Des 2008;40(7):760–77. 2001;33(7):545–59.
[49] Butlin G, Stops C. CAD data repair. In: Proc. fifth int’l meshing roundtable; [82] Chinn A. Why won’t it work? CAD User magazine, 2002.
1996: p. 7–12. [83] Kasik DJ, Buxton W, Ferguson DR. Ten CAD challenges. IEEE Comput Graphics
[50] Fenves SJ, Sriram RD, Subrahmanian E, Rachuri S. Product information ex- Appl 2005;25(2):81–92.
change: Practices and standards. J Comput Inf Sci Eng 2005;5(3):238–46. [84] Matsuki N. Problems in current CAD systems. Int J Prod Lifecycle Manag
[51] Bianconi F, Conti P, Di Angelo L. Interoperability among CAD/CAM/CAE 2010;4(4):326–30.
systems: A review of current research trends. In: Proceedings of geomet- [85] Hoffmann CM. The problem of accuracy and robustness in geometric com-
ric modeling and imaging—New trends (GMAI06) art. No 1648749; 2006. putation. In: Computer science technical reports, paper 660. Purdue e-Pubs.
p. 83–9. Report number 88–771; 1997 [retrieved 04.11.2015]. http://docs.lib.purdue.
[52] Tornincasa S, DiMonaco F. The future and the evolution of CAD. TMT 2010, edu/cgi/viewcontent.cgi?article=1659&context=cstech.
In: Proceedings of the 14th international research/expert conference: Trends [86] Hoffmann CM. In: Geometric and solid modeling. The Morgan Kaufmann
in the development of machinery and associated technology; 2010. p. 11–8. series in computer graphics and geometric modeling; 1989 [retrieved
[53] Altmeyer J, Ohnsorge S, Schuermann B. Reuse of design objects in CAD 04.11.2015].https://www.cs.purdue.edu/homes/cmh/distribution/books/geo
frameworks. IEEE Trans Comput-Aided Des Integr Circuits Syst 1994;754–61. .html.
[54] Katz RH. Towards a unified framework for version modeling in engineering [87] Yares E. CAD interoperability today. Design World; 2012 [retrieved
databases. ACM Comput Surv 1990;22(4):375–408. 04.11.2015].http://www.designworldonline.com/cad-interoperability-today
[55] Bai J, Gao S, Tang W, Liu Y, Guo S. Design reuse oriented partial retrieval of /#.
CAD models. Comput Aided Des 2010;42(12):1069–84. [88] Yang J, Han S, Park S. A method for verification of computer-aided design
[56] Chu CH, Hsu YC. Similarity assessment of 3D mechanical components for model errors. J Eng Des 2005;16(3):337–52.
design reuse. Rob Comput Integr Manuf 2006;22(4):332–41. [89] Lieblich JH, Fischer BR. Drawing requirements manual: Guide for digital mod-
[57] Hong T, Lee K, Kim S. Similarity comparison of mechanical parts to reuse els, digital and traditional drawings, and technical data packages for com-
existing designs. Comput Aided Des 2006;38(9):973–84. mercial and military applications. 11th ed., Colo: Englewood; 2008 Global
[58] Li M, Zhang YF, Fuh JYH, Qiu ZM. Design reusability assessment for effective Engineering Documents.
CAD model retrieval and reuse. Int J Comput Appl Technol 2011;40(1/2); [90] Srinivasan V. Standardizing the specification, verification, and exchange
3–12. of product geometry: Research, status and trends. Comput Aided Des
[59] Bøhn JH, Wozny MJ. Automatic CAD-model repair: Shell-closure. In: Proc. 2008;40(7):738–49.
symp. on solid freeform fabrication; 1992. p. 86–94. [91] International Organization for Standardization. International standard ISO
[60] Chong CS, Kumar AS, Lee HP. Automatic mesh-healing technique for model 10303-1, 1994: Industrial Automation Systems and Integration—Product
repair and finite element. Finite Elem Anal Des 2007;43:1109–19.
Data Representation and Exchange—Part 1: Overview and Fundamental
[61] Barequet G, Sharir M. Filling gaps in the boundary of a polyhedron. Comput
Principles; 1994 [retrieved 04.11.2015]. http://www.iso.org/iso/catalogue_
Aided Geom Design 1995;12(2):207–29.
detail?csnumber=20579.
[62] Barequet G, Kumar S. Repairing CAD models. IEEE Vis 1997;363–70.
[92] Gu P, Chan K. Product modelling using STEP. Comput Aided Des 1995;27(3);
[63] Barequet G, Duncan CA, Kumar S. RSVP: A geometric toolkit for controlled
163–79.
repair of solid models. IEEE Trans Vis Comput Graphics 1998;4(2):162–77.
[93] STEP AP242 Project. Whitepaper. Development of a convergent modular
[64] Nielson G, Holliday D, Roxborough T, Cracking the cracking problem with
STEP application protocol based on AP 203 and AP 214: STEP AP 242—
Coons patches. In: Proceedings of the IEEE visualization conference; 1999.
Managed Model Based 3D Engineering. Version 1.0, 2009-11-05; 2009
p. 285–90.
[retrieved 04.11.2015]. http://www.ap242.org/c/document_library/get_file?
[65] Claux F, Barthe L, Vanderhaeghe D, Jessel JP, Paulin M. Crack-free ren-
uuid=157a9a8f-6888-4589-906d-71d09e75878c&groupId=52520.
dering of dynamically tesselated B.-Rep models. Comput Graphics Forum
2014;33(2):263–72. [94] Patil L, Dutta D, Sriram R. Ontology-based exchange of product data seman-
[66] Olasz L. Working with imported CAD designs. COMSOL Blog 2014. URL: tics. IEEE Trans Autom Sci Eng 2005;2(3):213–24.
http://www.comsol.com/blogs/working-imported-cad-designs/. [95] Andrews PTJ, Shahin TMM, Sivaloganathan S. Design reuse in a CAD environ-
[67] Guskov I, Wood ZJ. Topological noise removal [Proceedings of graphics inter- ment - four case studies. Comput Ind Eng 1999;37(1):105–9.
face GI’01]; 2001. p. 19–26. [96] Bénière R, Subsol G, Gesquière G, Le Breton F, Puech W. A comprehensive
[68] Dey TK, Li K, Sun J. On computing handle and tunnel loops. IEEE Proc process of reverse engineering from 3D meshes to CAD models. Comput
NASAGEM 07; 2007. p. 357–66. Aided Des 2013;45(11):1382–93.
 C. González-Lluch et al. / Computer-Aided Design 83 (2017) 64–79 79

[97] Langbein FC, Gao CH, Mills BI, Marshall AD, Martin RR. Topological and [128] Abdul-Ghafour S, Ghodous P, Shariat B, Perna E, Khosrowshahi F. Semantic
geometric beautification of reverse engineered geometric models. In: Pro- interoperability of knowledge in feature-based CAD models. Comput Aided
ceedings of the ninth ACM symposium on solid modeling and applications; Des 2014;56:45–57.
2004. p. 255-60. [129] International Organization for Standardization. International standard ISO
[98] Langbein FC, Marshall AD, Martin RR. Choosing consistent constraints for 10303-242: Industrial automation systems and integration—Product data
beautification of reverse engineered geometric models. Comput Aided Des representation and exchange—Part 242: application protocol: managed
2004;36(3):261–78. model-based 3D engineering; 2014 [Retrieved 04.11.15]. http://www.iso.
[99] Li M, Langbein FC, Martin RR. Detecting design intent in approximate CAD org/iso/catalogue_detail.htm?csnumber=57620.
models using symmetry. Comput Aided Des 2010;42(3):183–201. [130] Mun D, Han S, Kim J, Oh Y. A set of standard modeling commands for the
[100] Niu Z, Martin RR, Langbein FC, Sabin MA. Rapidly finding CAD features using history-based parametric approach. Comput Aided Des 2003;35:1171–9.
database optimization. Comput Aided Des 2015;69:35–50. [131] Kim BC, Mun D, Han S, Pratt MJ. A method to exchange procedurally rep-
[101] Lavoué G, Dupont F, Baskurt A. A new CAD mesh segmentation method, based resented 2D CAD model data using ISO 10303 STEP. Comput Aided Des
on curvature tensor analysis. Comput Aided Des 2005;37(10):975–87. 2011;43(12):1717–28.
[102] Yi B, Liu Z, Tan J, Cheng F, Duan G, Liu L. Shape recognition of CAD models via [132] Diwakaran RP, Johnson MD. Analyzing the effect of alternative goals and
iterative slippage analysis. Comput Aided Des 2014;55:13–25. model attributes on CAD model creation and alteration. Comput Aided Des
[103] Thompson WB, Owen JC, St. Germain HJ, Stark Jr SR, Henderson TC. Feature- 2012;44:343–53.
based reverse engineering of mechanical parts. IEEE Trans Rob Autom [133] Camba JD, Contero M, Company P. Parametric CAD modeling: An analysis of
1999;15(1):57–66. strategies for design reusability. Comput Aided Des 2016;74:18–31.
[104] Agathos A, Pratikakis I, Perantonis S, Sapidis N, Azariadis P. 3D mesh seg- [134] Hounsell MS, Case K. Representation validation in feature-based modelling:
mentation methodologies for CAD applications. Comput Aided Design Appl a framework for design correctness analysis and assurance. In: Advances
2007;4(1–6):827–41. in manufacturing technology X. The proceedings of the twelfth national
[105] Kovács I, Várady T, Salvi P. Applying geometric constraints for perfecting CAD conference on manufacturing research. UK: University of Bath; 1996. p. 256–
models in reverse engineering. Graph Models 2015;82:44–57. 60.
[106] Aleixos N, Company P, Contero M. Integrated modeling with top-down ap- [135] Bodein Y, Rose B, Caillaud E. Explicit reference modeling methodology in
proach in subsidiary industries. Comput Ind 2004;53(1):97–116. parametric CAD system. Comput Ind 2014;65:136–47.
[107] Gao SM, Zhao W, Lin HW, Yang FQ, Chen X. Feature suppression based CAD [136] van der Meiden HA, Bronsvoort WF. Tracking topological changes in para-
mesh model simplification. Comput Aided Des 2010;42(12):1178–88. metric models. Comput Aided Geom Design 2010;27(3):281–93.
[108] Quadros WR, Owen SJ. Defeaturing CAD models using a geometry-based [137] Bidarra R, Bronsvoort WF. Semantic feature modelling. Comput Aided Des
size field and facet-based reduction operators. Eng Comput 2012;28(3); 2000;32(3):201–25.
211–24. [138] Camba J, Contero M, Johnson M, Company P. Extended 3D annotations as
[109] Nolan DC, Tierney CM, Armstrong CG, Robinson TT, Makem JE. Automatic a new mechanism to explicitly communicate geometric design intent and
dimensional reduction and meshing of stiffened thin-wall structures. Eng increase CAD model reusability. Comput Aided Des 2014;57:61–73.
Comput 2013;30(4):689–701. [139] ITI Transcendata. CADfix; 2015 [Retrieved 04.11.15]. http://www.transcendata.
[110] Wang X, Liu X, Lu L, Li B, Yin B, Shi X. Automatic hole-filling of CAD models com/products/cadfix.
with feature-preserving. Comput Graphics 2012;36(2):101–10. [140] ITI Transcendata. CADfix 6.0 Introductory User Tutorial; 2004.
[111] Li M, Gao S, Martin RR. Estimating the effects of removing negative features [141] CADInterop. CADfix; 2015 [Retrieved 04.11.15]. https://www.cadinterop.
on engineering analysis. Comput Aided Des 2011;43(11):1402–12. com/en/cadfix.html.
[112] Li M, Gao S. Estimating defeaturing-induced engineering analysis errors for [142] Cheney D. 3D CAD model validation. In: proceedings 3D collaboration and
arbitrary 3D features. Comput Aided Des 2011;43(12):1587–97. interoperability congress (CIC 2008).
[113] Li M, Gao S, Zhang K. A goal-oriented error estimator for the analysis of [143] ITI Transcendata. CADIQ; 2015 [Retrieved 04.11.15]. http://www.transcendata.
simplified designs. Comput Methods Appl Mech Engrg 2013;255:89–103. com/products/cadiq.
[114] Tang J, Gao S, Li M. Evaluating defeaturing-induced impact on model analysis. [144] CADInterop. CADIQ; 2015 [Retrieved 04.11.15]. https://www.cadinterop.
Math Comput Modelling 2013;57(3–4):413–24. com/en/cadiq.html.
[115] Company P, Contero M, Otey J, Plumed R. Approach for developing coordi- [145] Capvidia. 3DTransVidia Translates and Repairs Common Errors Created in
nated rubrics to convey quality criteria in CAD training. Comput Aided Des Different CAD Systems; 2015 [Retrieved 04.11.15]. http://www.capvidia.
2015;63:101–17. com/capvidia-products/3d-transvidia-cad-data-translation-repair.
[116] Company P, Contero M, Otey J, Camba JD, Agost MJ, Pérez-López DC. Web- [146] DAIMLER. Quick reference. Heidelberg CAx Quality Manager; 2013 [Retrieved
based system for adaptable rubrics: case study on CAD assessment. J Educ 04.11.15]. https://swan-dtna.i.daimler-trucksnorthamerica.com/dtna/downl
Technol Soc 2016 [In press]. oad/Quality/2407883_Tipsheet_QUALITY_Heidelberg_Quality_Manager.pdf.
[117] Capoyleas V, Chen X, Hoffmann CM. Generic naming in generative, constraint- [147] DASSAULT SYSTEMES. Solidworks Help. Welcome to Solidworks Design
based design. Comput Aided Des 2006;28(1):17–28. Checker; 2015 [Retrieved 04.11.15]. http://help.solidworks.com/2015/english/
[118] Hoffmann CM. Semantic problems of generative, constraint based design. solidworks/solidworks_design_checker/c_welcome_design_checker.htm.
Purdue e-Pubs: Purdue University; 1993 Report Number: 93-062. [148] PrescientQA. DesignQA; 2015 [Retrieved 04.11.15]. http://www.prescientqa.
[119] Wu J, Zhang T, Zhang X, Zhou J. A face based mechanism for naming, record- com/designqa.aspx.
ing and retrieving topological entities. Comput Aided Des 2001;33(10); [149] PrescientQA. GeometryQA; 2015 [Retrieved 04.11.15]. http://www.prescientqa.
687–98. com/geometryqa.aspx.
[120] Marcheix D, Pierra G. A survey of the persistent naming problem. In: SMA [150] PrescientQA. PrescientQA; 2015 [Retrieved 04.11.15]. http://www.prescientqa.
’02 proceedings of the seventh ACM symposium on solid modeling and com/pqav5.aspx.
applications; 2002. p. 13–22. [151] Tata Technologies. iCHECK IT; 2015 [Retrieved 04.11.15]. http://myicheckit.
[121] Mun D, Han S. Identification of topological entities and naming map- com/products/catia-v5.
ping based on IGM for parametric CAD model exchanges. Int J CAD/CAM [152] CATIA. Knowledge Advisor; 2015 [Retrieved 04.11.15]. http://catiadesign.
2005;5(1):69–81. org/_doc/v5r14/catpdfkwrug_C2/kwrug.pdf.
[122] Cheon SU, Mun D, Han S, Kim BC. Name matching method using topology [153] CATIA. Knowledge Expert; 2015 [Retrieved 04.11.15]. http://www.catia.com.
merging and splitting history for exchange of feature-based CAD models. J pl/tutorial/z2/expert_knowledge.pdf.
Mech Sci Technol 2012;26(10):3201–12. [154] PTC CREO ParametricTM. FIND ANSWERS; 2015 [Retrieved 04.11.15]. htt
[123] Marcheix D. A persistent naming of shells. In: CADCG ’05: proceedings of p://help.ptc.com/creo_hc/creo30_pma_hc/usascii/#page/pma/model_analys
the ninth international conference on computer aided design and computer is/ModelCHECK_Overview.html.
graphics; 2005. p. 259–68. [155] SIEMENS. NX Check-Mate. Knowledge-driven validation enhances product
[124] Anderl R, Mendgen R. Modelling with constraints: theoretical foundation and development processes and quality; 2015 [Retrieved 04.11.15]. https://m.
application. Comput Aided Des 1996;28(3):155–68. plm.automation.siemens.com/en_us/Images/2504_tcm1224-11882.pdf.
[125] Gerbino S, Brondi A. Interoperability issues among CAD systems: a bench- [156] TRANSCAT A Dassault Systemes Company, Q-Checker V1.20.1 for CATIA⃝ R
V5.
marking study of 7 commercial MCAD software. In: International design Users Manual; 2015.
conference-DESIGN: Dubrovnik; 2004. p. 617–26. [157] TRANSCAT A Dassault Systemes Company. Q-Checker; 2015 [Retrieved
[126] Bettig B, Hoffman C. Geometric constraint solving in parametric CAD. J Com- 04.11.15]. https://www.transcat-plm.com/en/software/transcat-software/q-
put Inf Sci Eng 2011;11(2):021001. checker.html.
[127] Dartigues C, Ghodous P, Gruninger M, Pallez D, Sriram R. CAD/CAPP integra- [158] Ruz̧arovský R. Direct production from CAD models considering on integration
tion using feature ontology. Concurr Eng Res Appl 2007;15(2):237–49. with CIM flexible production system. Appl Mech Mater 2014;474:103–18.